xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision c8e7f78a3d28ff6e6223ed136ada8e1e2f34965e)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2011 Chelsio Communications, Inc.
5  * All rights reserved.
6  * Written by: Navdeep Parhar <np@FreeBSD.org>
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 #include "opt_ddb.h"
32 #include "opt_inet.h"
33 #include "opt_inet6.h"
34 #include "opt_kern_tls.h"
35 #include "opt_ratelimit.h"
36 #include "opt_rss.h"
37 
38 #include <sys/param.h>
39 #include <sys/conf.h>
40 #include <sys/priv.h>
41 #include <sys/kernel.h>
42 #include <sys/bus.h>
43 #include <sys/eventhandler.h>
44 #include <sys/module.h>
45 #include <sys/malloc.h>
46 #include <sys/queue.h>
47 #include <sys/taskqueue.h>
48 #include <sys/pciio.h>
49 #include <dev/pci/pcireg.h>
50 #include <dev/pci/pcivar.h>
51 #include <dev/pci/pci_private.h>
52 #include <sys/firmware.h>
53 #include <sys/sbuf.h>
54 #include <sys/smp.h>
55 #include <sys/socket.h>
56 #include <sys/sockio.h>
57 #include <sys/sysctl.h>
58 #include <net/ethernet.h>
59 #include <net/if.h>
60 #include <net/if_types.h>
61 #include <net/if_dl.h>
62 #include <net/if_vlan_var.h>
63 #ifdef RSS
64 #include <net/rss_config.h>
65 #endif
66 #include <netinet/in.h>
67 #include <netinet/ip.h>
68 #ifdef KERN_TLS
69 #include <netinet/tcp_seq.h>
70 #endif
71 #if defined(__i386__) || defined(__amd64__)
72 #include <machine/md_var.h>
73 #include <machine/cputypes.h>
74 #include <vm/vm.h>
75 #include <vm/pmap.h>
76 #endif
77 #ifdef DDB
78 #include <ddb/ddb.h>
79 #include <ddb/db_lex.h>
80 #endif
81 
82 #include "common/common.h"
83 #include "common/t4_msg.h"
84 #include "common/t4_regs.h"
85 #include "common/t4_regs_values.h"
86 #include "cudbg/cudbg.h"
87 #include "t4_clip.h"
88 #include "t4_ioctl.h"
89 #include "t4_l2t.h"
90 #include "t4_mp_ring.h"
91 #include "t4_if.h"
92 #include "t4_smt.h"
93 
94 /* T4 bus driver interface */
95 static int t4_probe(device_t);
96 static int t4_attach(device_t);
97 static int t4_detach(device_t);
98 static int t4_child_location(device_t, device_t, struct sbuf *);
99 static int t4_ready(device_t);
100 static int t4_read_port_device(device_t, int, device_t *);
101 static int t4_suspend(device_t);
102 static int t4_resume(device_t);
103 static int t4_reset_prepare(device_t, device_t);
104 static int t4_reset_post(device_t, device_t);
105 static device_method_t t4_methods[] = {
106 	DEVMETHOD(device_probe,		t4_probe),
107 	DEVMETHOD(device_attach,	t4_attach),
108 	DEVMETHOD(device_detach,	t4_detach),
109 	DEVMETHOD(device_suspend,	t4_suspend),
110 	DEVMETHOD(device_resume,	t4_resume),
111 
112 	DEVMETHOD(bus_child_location,	t4_child_location),
113 	DEVMETHOD(bus_reset_prepare,	t4_reset_prepare),
114 	DEVMETHOD(bus_reset_post,	t4_reset_post),
115 
116 	DEVMETHOD(t4_is_main_ready,	t4_ready),
117 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
118 
119 	DEVMETHOD_END
120 };
121 static driver_t t4_driver = {
122 	"t4nex",
123 	t4_methods,
124 	sizeof(struct adapter)
125 };
126 
127 
128 /* T4 port (cxgbe) interface */
129 static int cxgbe_probe(device_t);
130 static int cxgbe_attach(device_t);
131 static int cxgbe_detach(device_t);
132 device_method_t cxgbe_methods[] = {
133 	DEVMETHOD(device_probe,		cxgbe_probe),
134 	DEVMETHOD(device_attach,	cxgbe_attach),
135 	DEVMETHOD(device_detach,	cxgbe_detach),
136 	{ 0, 0 }
137 };
138 static driver_t cxgbe_driver = {
139 	"cxgbe",
140 	cxgbe_methods,
141 	sizeof(struct port_info)
142 };
143 
144 /* T4 VI (vcxgbe) interface */
145 static int vcxgbe_probe(device_t);
146 static int vcxgbe_attach(device_t);
147 static int vcxgbe_detach(device_t);
148 static device_method_t vcxgbe_methods[] = {
149 	DEVMETHOD(device_probe,		vcxgbe_probe),
150 	DEVMETHOD(device_attach,	vcxgbe_attach),
151 	DEVMETHOD(device_detach,	vcxgbe_detach),
152 	{ 0, 0 }
153 };
154 static driver_t vcxgbe_driver = {
155 	"vcxgbe",
156 	vcxgbe_methods,
157 	sizeof(struct vi_info)
158 };
159 
160 static d_ioctl_t t4_ioctl;
161 
162 static struct cdevsw t4_cdevsw = {
163        .d_version = D_VERSION,
164        .d_ioctl = t4_ioctl,
165        .d_name = "t4nex",
166 };
167 
168 /* T5 bus driver interface */
169 static int t5_probe(device_t);
170 static device_method_t t5_methods[] = {
171 	DEVMETHOD(device_probe,		t5_probe),
172 	DEVMETHOD(device_attach,	t4_attach),
173 	DEVMETHOD(device_detach,	t4_detach),
174 	DEVMETHOD(device_suspend,	t4_suspend),
175 	DEVMETHOD(device_resume,	t4_resume),
176 
177 	DEVMETHOD(bus_child_location,	t4_child_location),
178 	DEVMETHOD(bus_reset_prepare,	t4_reset_prepare),
179 	DEVMETHOD(bus_reset_post,	t4_reset_post),
180 
181 	DEVMETHOD(t4_is_main_ready,	t4_ready),
182 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
183 
184 	DEVMETHOD_END
185 };
186 static driver_t t5_driver = {
187 	"t5nex",
188 	t5_methods,
189 	sizeof(struct adapter)
190 };
191 
192 
193 /* T5 port (cxl) interface */
194 static driver_t cxl_driver = {
195 	"cxl",
196 	cxgbe_methods,
197 	sizeof(struct port_info)
198 };
199 
200 /* T5 VI (vcxl) interface */
201 static driver_t vcxl_driver = {
202 	"vcxl",
203 	vcxgbe_methods,
204 	sizeof(struct vi_info)
205 };
206 
207 /* T6 bus driver interface */
208 static int t6_probe(device_t);
209 static device_method_t t6_methods[] = {
210 	DEVMETHOD(device_probe,		t6_probe),
211 	DEVMETHOD(device_attach,	t4_attach),
212 	DEVMETHOD(device_detach,	t4_detach),
213 	DEVMETHOD(device_suspend,	t4_suspend),
214 	DEVMETHOD(device_resume,	t4_resume),
215 
216 	DEVMETHOD(bus_child_location,	t4_child_location),
217 	DEVMETHOD(bus_reset_prepare,	t4_reset_prepare),
218 	DEVMETHOD(bus_reset_post,	t4_reset_post),
219 
220 	DEVMETHOD(t4_is_main_ready,	t4_ready),
221 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
222 
223 	DEVMETHOD_END
224 };
225 static driver_t t6_driver = {
226 	"t6nex",
227 	t6_methods,
228 	sizeof(struct adapter)
229 };
230 
231 
232 /* T6 port (cc) interface */
233 static driver_t cc_driver = {
234 	"cc",
235 	cxgbe_methods,
236 	sizeof(struct port_info)
237 };
238 
239 /* T6 VI (vcc) interface */
240 static driver_t vcc_driver = {
241 	"vcc",
242 	vcxgbe_methods,
243 	sizeof(struct vi_info)
244 };
245 
246 /* ifnet interface */
247 static void cxgbe_init(void *);
248 static int cxgbe_ioctl(if_t, unsigned long, caddr_t);
249 static int cxgbe_transmit(if_t, struct mbuf *);
250 static void cxgbe_qflush(if_t);
251 #if defined(KERN_TLS) || defined(RATELIMIT)
252 static int cxgbe_snd_tag_alloc(if_t, union if_snd_tag_alloc_params *,
253     struct m_snd_tag **);
254 #endif
255 
256 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
257 
258 /*
259  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
260  * then ADAPTER_LOCK, then t4_uld_list_lock.
261  */
262 static struct sx t4_list_lock;
263 SLIST_HEAD(, adapter) t4_list;
264 #ifdef TCP_OFFLOAD
265 static struct sx t4_uld_list_lock;
266 SLIST_HEAD(, uld_info) t4_uld_list;
267 #endif
268 
269 /*
270  * Tunables.  See tweak_tunables() too.
271  *
272  * Each tunable is set to a default value here if it's known at compile-time.
273  * Otherwise it is set to -n as an indication to tweak_tunables() that it should
274  * provide a reasonable default (upto n) when the driver is loaded.
275  *
276  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
277  * T5 are under hw.cxl.
278  */
279 SYSCTL_NODE(_hw, OID_AUTO, cxgbe, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
280     "cxgbe(4) parameters");
281 SYSCTL_NODE(_hw, OID_AUTO, cxl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
282     "cxgbe(4) T5+ parameters");
283 SYSCTL_NODE(_hw_cxgbe, OID_AUTO, toe, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
284     "cxgbe(4) TOE parameters");
285 
286 /*
287  * Number of queues for tx and rx, NIC and offload.
288  */
289 #define NTXQ 16
290 int t4_ntxq = -NTXQ;
291 SYSCTL_INT(_hw_cxgbe, OID_AUTO, ntxq, CTLFLAG_RDTUN, &t4_ntxq, 0,
292     "Number of TX queues per port");
293 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq);	/* Old name, undocumented */
294 
295 #define NRXQ 8
296 int t4_nrxq = -NRXQ;
297 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nrxq, CTLFLAG_RDTUN, &t4_nrxq, 0,
298     "Number of RX queues per port");
299 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq);	/* Old name, undocumented */
300 
301 #define NTXQ_VI 1
302 static int t4_ntxq_vi = -NTXQ_VI;
303 SYSCTL_INT(_hw_cxgbe, OID_AUTO, ntxq_vi, CTLFLAG_RDTUN, &t4_ntxq_vi, 0,
304     "Number of TX queues per VI");
305 
306 #define NRXQ_VI 1
307 static int t4_nrxq_vi = -NRXQ_VI;
308 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nrxq_vi, CTLFLAG_RDTUN, &t4_nrxq_vi, 0,
309     "Number of RX queues per VI");
310 
311 static int t4_rsrv_noflowq = 0;
312 SYSCTL_INT(_hw_cxgbe, OID_AUTO, rsrv_noflowq, CTLFLAG_RDTUN, &t4_rsrv_noflowq,
313     0, "Reserve TX queue 0 of each VI for non-flowid packets");
314 
315 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
316 #define NOFLDTXQ 8
317 static int t4_nofldtxq = -NOFLDTXQ;
318 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldtxq, CTLFLAG_RDTUN, &t4_nofldtxq, 0,
319     "Number of offload TX queues per port");
320 
321 #define NOFLDRXQ 2
322 static int t4_nofldrxq = -NOFLDRXQ;
323 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldrxq, CTLFLAG_RDTUN, &t4_nofldrxq, 0,
324     "Number of offload RX queues per port");
325 
326 #define NOFLDTXQ_VI 1
327 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
328 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldtxq_vi, CTLFLAG_RDTUN, &t4_nofldtxq_vi, 0,
329     "Number of offload TX queues per VI");
330 
331 #define NOFLDRXQ_VI 1
332 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
333 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldrxq_vi, CTLFLAG_RDTUN, &t4_nofldrxq_vi, 0,
334     "Number of offload RX queues per VI");
335 
336 #define TMR_IDX_OFLD 1
337 int t4_tmr_idx_ofld = TMR_IDX_OFLD;
338 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_timer_idx_ofld, CTLFLAG_RDTUN,
339     &t4_tmr_idx_ofld, 0, "Holdoff timer index for offload queues");
340 
341 #define PKTC_IDX_OFLD (-1)
342 int t4_pktc_idx_ofld = PKTC_IDX_OFLD;
343 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_pktc_idx_ofld, CTLFLAG_RDTUN,
344     &t4_pktc_idx_ofld, 0, "holdoff packet counter index for offload queues");
345 
346 /* 0 means chip/fw default, non-zero number is value in microseconds */
347 static u_long t4_toe_keepalive_idle = 0;
348 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, keepalive_idle, CTLFLAG_RDTUN,
349     &t4_toe_keepalive_idle, 0, "TOE keepalive idle timer (us)");
350 
351 /* 0 means chip/fw default, non-zero number is value in microseconds */
352 static u_long t4_toe_keepalive_interval = 0;
353 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, keepalive_interval, CTLFLAG_RDTUN,
354     &t4_toe_keepalive_interval, 0, "TOE keepalive interval timer (us)");
355 
356 /* 0 means chip/fw default, non-zero number is # of keepalives before abort */
357 static int t4_toe_keepalive_count = 0;
358 SYSCTL_INT(_hw_cxgbe_toe, OID_AUTO, keepalive_count, CTLFLAG_RDTUN,
359     &t4_toe_keepalive_count, 0, "Number of TOE keepalive probes before abort");
360 
361 /* 0 means chip/fw default, non-zero number is value in microseconds */
362 static u_long t4_toe_rexmt_min = 0;
363 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, rexmt_min, CTLFLAG_RDTUN,
364     &t4_toe_rexmt_min, 0, "Minimum TOE retransmit interval (us)");
365 
366 /* 0 means chip/fw default, non-zero number is value in microseconds */
367 static u_long t4_toe_rexmt_max = 0;
368 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, rexmt_max, CTLFLAG_RDTUN,
369     &t4_toe_rexmt_max, 0, "Maximum TOE retransmit interval (us)");
370 
371 /* 0 means chip/fw default, non-zero number is # of rexmt before abort */
372 static int t4_toe_rexmt_count = 0;
373 SYSCTL_INT(_hw_cxgbe_toe, OID_AUTO, rexmt_count, CTLFLAG_RDTUN,
374     &t4_toe_rexmt_count, 0, "Number of TOE retransmissions before abort");
375 
376 /* -1 means chip/fw default, other values are raw backoff values to use */
377 static int t4_toe_rexmt_backoff[16] = {
378 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
379 };
380 SYSCTL_NODE(_hw_cxgbe_toe, OID_AUTO, rexmt_backoff,
381     CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
382     "cxgbe(4) TOE retransmit backoff values");
383 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 0, CTLFLAG_RDTUN,
384     &t4_toe_rexmt_backoff[0], 0, "");
385 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 1, CTLFLAG_RDTUN,
386     &t4_toe_rexmt_backoff[1], 0, "");
387 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 2, CTLFLAG_RDTUN,
388     &t4_toe_rexmt_backoff[2], 0, "");
389 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 3, CTLFLAG_RDTUN,
390     &t4_toe_rexmt_backoff[3], 0, "");
391 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 4, CTLFLAG_RDTUN,
392     &t4_toe_rexmt_backoff[4], 0, "");
393 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 5, CTLFLAG_RDTUN,
394     &t4_toe_rexmt_backoff[5], 0, "");
395 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 6, CTLFLAG_RDTUN,
396     &t4_toe_rexmt_backoff[6], 0, "");
397 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 7, CTLFLAG_RDTUN,
398     &t4_toe_rexmt_backoff[7], 0, "");
399 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 8, CTLFLAG_RDTUN,
400     &t4_toe_rexmt_backoff[8], 0, "");
401 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 9, CTLFLAG_RDTUN,
402     &t4_toe_rexmt_backoff[9], 0, "");
403 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 10, CTLFLAG_RDTUN,
404     &t4_toe_rexmt_backoff[10], 0, "");
405 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 11, CTLFLAG_RDTUN,
406     &t4_toe_rexmt_backoff[11], 0, "");
407 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 12, CTLFLAG_RDTUN,
408     &t4_toe_rexmt_backoff[12], 0, "");
409 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 13, CTLFLAG_RDTUN,
410     &t4_toe_rexmt_backoff[13], 0, "");
411 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 14, CTLFLAG_RDTUN,
412     &t4_toe_rexmt_backoff[14], 0, "");
413 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 15, CTLFLAG_RDTUN,
414     &t4_toe_rexmt_backoff[15], 0, "");
415 #endif
416 
417 #ifdef DEV_NETMAP
418 #define NN_MAIN_VI	(1 << 0)	/* Native netmap on the main VI */
419 #define NN_EXTRA_VI	(1 << 1)	/* Native netmap on the extra VI(s) */
420 static int t4_native_netmap = NN_EXTRA_VI;
421 SYSCTL_INT(_hw_cxgbe, OID_AUTO, native_netmap, CTLFLAG_RDTUN, &t4_native_netmap,
422     0, "Native netmap support.  bit 0 = main VI, bit 1 = extra VIs");
423 
424 #define NNMTXQ 8
425 static int t4_nnmtxq = -NNMTXQ;
426 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmtxq, CTLFLAG_RDTUN, &t4_nnmtxq, 0,
427     "Number of netmap TX queues");
428 
429 #define NNMRXQ 8
430 static int t4_nnmrxq = -NNMRXQ;
431 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmrxq, CTLFLAG_RDTUN, &t4_nnmrxq, 0,
432     "Number of netmap RX queues");
433 
434 #define NNMTXQ_VI 2
435 static int t4_nnmtxq_vi = -NNMTXQ_VI;
436 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmtxq_vi, CTLFLAG_RDTUN, &t4_nnmtxq_vi, 0,
437     "Number of netmap TX queues per VI");
438 
439 #define NNMRXQ_VI 2
440 static int t4_nnmrxq_vi = -NNMRXQ_VI;
441 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmrxq_vi, CTLFLAG_RDTUN, &t4_nnmrxq_vi, 0,
442     "Number of netmap RX queues per VI");
443 #endif
444 
445 /*
446  * Holdoff parameters for ports.
447  */
448 #define TMR_IDX 1
449 int t4_tmr_idx = TMR_IDX;
450 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_timer_idx, CTLFLAG_RDTUN, &t4_tmr_idx,
451     0, "Holdoff timer index");
452 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx);	/* Old name */
453 
454 #define PKTC_IDX (-1)
455 int t4_pktc_idx = PKTC_IDX;
456 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_pktc_idx, CTLFLAG_RDTUN, &t4_pktc_idx,
457     0, "Holdoff packet counter index");
458 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx);	/* Old name */
459 
460 /*
461  * Size (# of entries) of each tx and rx queue.
462  */
463 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
464 SYSCTL_INT(_hw_cxgbe, OID_AUTO, qsize_txq, CTLFLAG_RDTUN, &t4_qsize_txq, 0,
465     "Number of descriptors in each TX queue");
466 
467 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
468 SYSCTL_INT(_hw_cxgbe, OID_AUTO, qsize_rxq, CTLFLAG_RDTUN, &t4_qsize_rxq, 0,
469     "Number of descriptors in each RX queue");
470 
471 /*
472  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
473  */
474 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
475 SYSCTL_INT(_hw_cxgbe, OID_AUTO, interrupt_types, CTLFLAG_RDTUN, &t4_intr_types,
476     0, "Interrupt types allowed (bit 0 = INTx, 1 = MSI, 2 = MSI-X)");
477 
478 /*
479  * Configuration file.  All the _CF names here are special.
480  */
481 #define DEFAULT_CF	"default"
482 #define BUILTIN_CF	"built-in"
483 #define FLASH_CF	"flash"
484 #define UWIRE_CF	"uwire"
485 #define FPGA_CF		"fpga"
486 static char t4_cfg_file[32] = DEFAULT_CF;
487 SYSCTL_STRING(_hw_cxgbe, OID_AUTO, config_file, CTLFLAG_RDTUN, t4_cfg_file,
488     sizeof(t4_cfg_file), "Firmware configuration file");
489 
490 /*
491  * PAUSE settings (bit 0, 1, 2 = rx_pause, tx_pause, pause_autoneg respectively).
492  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
493  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
494  *            mark or when signalled to do so, 0 to never emit PAUSE.
495  * pause_autoneg = 1 means PAUSE will be negotiated if possible and the
496  *                 negotiated settings will override rx_pause/tx_pause.
497  *                 Otherwise rx_pause/tx_pause are applied forcibly.
498  */
499 static int t4_pause_settings = PAUSE_RX | PAUSE_TX | PAUSE_AUTONEG;
500 SYSCTL_INT(_hw_cxgbe, OID_AUTO, pause_settings, CTLFLAG_RDTUN,
501     &t4_pause_settings, 0,
502     "PAUSE settings (bit 0 = rx_pause, 1 = tx_pause, 2 = pause_autoneg)");
503 
504 /*
505  * Forward Error Correction settings (bit 0, 1 = RS, BASER respectively).
506  * -1 to run with the firmware default.  Same as FEC_AUTO (bit 5)
507  *  0 to disable FEC.
508  */
509 static int t4_fec = -1;
510 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fec, CTLFLAG_RDTUN, &t4_fec, 0,
511     "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
512 
513 /*
514  * Controls when the driver sets the FORCE_FEC bit in the L1_CFG32 that it
515  * issues to the firmware.  If the firmware doesn't support FORCE_FEC then the
516  * driver runs as if this is set to 0.
517  * -1 to set FORCE_FEC iff requested_fec != AUTO. Multiple FEC bits are okay.
518  *  0 to never set FORCE_FEC. requested_fec = AUTO means use the hint from the
519  *    transceiver. Multiple FEC bits may not be okay but will be passed on to
520  *    the firmware anyway (may result in l1cfg errors with old firmwares).
521  *  1 to always set FORCE_FEC. Multiple FEC bits are okay. requested_fec = AUTO
522  *    means set all FEC bits that are valid for the speed.
523  */
524 static int t4_force_fec = -1;
525 SYSCTL_INT(_hw_cxgbe, OID_AUTO, force_fec, CTLFLAG_RDTUN, &t4_force_fec, 0,
526     "Controls the use of FORCE_FEC bit in L1 configuration.");
527 
528 /*
529  * Link autonegotiation.
530  * -1 to run with the firmware default.
531  *  0 to disable.
532  *  1 to enable.
533  */
534 static int t4_autoneg = -1;
535 SYSCTL_INT(_hw_cxgbe, OID_AUTO, autoneg, CTLFLAG_RDTUN, &t4_autoneg, 0,
536     "Link autonegotiation");
537 
538 /*
539  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
540  * encouraged respectively).  '-n' is the same as 'n' except the firmware
541  * version used in the checks is read from the firmware bundled with the driver.
542  */
543 static int t4_fw_install = 1;
544 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fw_install, CTLFLAG_RDTUN, &t4_fw_install, 0,
545     "Firmware auto-install (0 = prohibited, 1 = allowed, 2 = encouraged)");
546 
547 /*
548  * ASIC features that will be used.  Disable the ones you don't want so that the
549  * chip resources aren't wasted on features that will not be used.
550  */
551 static int t4_nbmcaps_allowed = 0;
552 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nbmcaps_allowed, CTLFLAG_RDTUN,
553     &t4_nbmcaps_allowed, 0, "Default NBM capabilities");
554 
555 static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
556 SYSCTL_INT(_hw_cxgbe, OID_AUTO, linkcaps_allowed, CTLFLAG_RDTUN,
557     &t4_linkcaps_allowed, 0, "Default link capabilities");
558 
559 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
560     FW_CAPS_CONFIG_SWITCH_EGRESS;
561 SYSCTL_INT(_hw_cxgbe, OID_AUTO, switchcaps_allowed, CTLFLAG_RDTUN,
562     &t4_switchcaps_allowed, 0, "Default switch capabilities");
563 
564 #ifdef RATELIMIT
565 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
566 	FW_CAPS_CONFIG_NIC_HASHFILTER | FW_CAPS_CONFIG_NIC_ETHOFLD;
567 #else
568 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
569 	FW_CAPS_CONFIG_NIC_HASHFILTER;
570 #endif
571 SYSCTL_INT(_hw_cxgbe, OID_AUTO, niccaps_allowed, CTLFLAG_RDTUN,
572     &t4_niccaps_allowed, 0, "Default NIC capabilities");
573 
574 static int t4_toecaps_allowed = -1;
575 SYSCTL_INT(_hw_cxgbe, OID_AUTO, toecaps_allowed, CTLFLAG_RDTUN,
576     &t4_toecaps_allowed, 0, "Default TCP offload capabilities");
577 
578 static int t4_rdmacaps_allowed = -1;
579 SYSCTL_INT(_hw_cxgbe, OID_AUTO, rdmacaps_allowed, CTLFLAG_RDTUN,
580     &t4_rdmacaps_allowed, 0, "Default RDMA capabilities");
581 
582 static int t4_cryptocaps_allowed = -1;
583 SYSCTL_INT(_hw_cxgbe, OID_AUTO, cryptocaps_allowed, CTLFLAG_RDTUN,
584     &t4_cryptocaps_allowed, 0, "Default crypto capabilities");
585 
586 static int t4_iscsicaps_allowed = -1;
587 SYSCTL_INT(_hw_cxgbe, OID_AUTO, iscsicaps_allowed, CTLFLAG_RDTUN,
588     &t4_iscsicaps_allowed, 0, "Default iSCSI capabilities");
589 
590 static int t4_fcoecaps_allowed = 0;
591 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fcoecaps_allowed, CTLFLAG_RDTUN,
592     &t4_fcoecaps_allowed, 0, "Default FCoE capabilities");
593 
594 static int t5_write_combine = 0;
595 SYSCTL_INT(_hw_cxl, OID_AUTO, write_combine, CTLFLAG_RDTUN, &t5_write_combine,
596     0, "Use WC instead of UC for BAR2");
597 
598 static int t4_num_vis = 1;
599 SYSCTL_INT(_hw_cxgbe, OID_AUTO, num_vis, CTLFLAG_RDTUN, &t4_num_vis, 0,
600     "Number of VIs per port");
601 
602 /*
603  * PCIe Relaxed Ordering.
604  * -1: driver should figure out a good value.
605  * 0: disable RO.
606  * 1: enable RO.
607  * 2: leave RO alone.
608  */
609 static int pcie_relaxed_ordering = -1;
610 SYSCTL_INT(_hw_cxgbe, OID_AUTO, pcie_relaxed_ordering, CTLFLAG_RDTUN,
611     &pcie_relaxed_ordering, 0,
612     "PCIe Relaxed Ordering: 0 = disable, 1 = enable, 2 = leave alone");
613 
614 static int t4_panic_on_fatal_err = 0;
615 SYSCTL_INT(_hw_cxgbe, OID_AUTO, panic_on_fatal_err, CTLFLAG_RWTUN,
616     &t4_panic_on_fatal_err, 0, "panic on fatal errors");
617 
618 static int t4_reset_on_fatal_err = 0;
619 SYSCTL_INT(_hw_cxgbe, OID_AUTO, reset_on_fatal_err, CTLFLAG_RWTUN,
620     &t4_reset_on_fatal_err, 0, "reset adapter on fatal errors");
621 
622 static int t4_clock_gate_on_suspend = 0;
623 SYSCTL_INT(_hw_cxgbe, OID_AUTO, clock_gate_on_suspend, CTLFLAG_RWTUN,
624     &t4_clock_gate_on_suspend, 0, "gate the clock on suspend");
625 
626 static int t4_tx_vm_wr = 0;
627 SYSCTL_INT(_hw_cxgbe, OID_AUTO, tx_vm_wr, CTLFLAG_RWTUN, &t4_tx_vm_wr, 0,
628     "Use VM work requests to transmit packets.");
629 
630 /*
631  * Set to non-zero to enable the attack filter.  A packet that matches any of
632  * these conditions will get dropped on ingress:
633  * 1) IP && source address == destination address.
634  * 2) TCP/IP && source address is not a unicast address.
635  * 3) TCP/IP && destination address is not a unicast address.
636  * 4) IP && source address is loopback (127.x.y.z).
637  * 5) IP && destination address is loopback (127.x.y.z).
638  * 6) IPv6 && source address == destination address.
639  * 7) IPv6 && source address is not a unicast address.
640  * 8) IPv6 && source address is loopback (::1/128).
641  * 9) IPv6 && destination address is loopback (::1/128).
642  * 10) IPv6 && source address is unspecified (::/128).
643  * 11) IPv6 && destination address is unspecified (::/128).
644  * 12) TCP/IPv6 && source address is multicast (ff00::/8).
645  * 13) TCP/IPv6 && destination address is multicast (ff00::/8).
646  */
647 static int t4_attack_filter = 0;
648 SYSCTL_INT(_hw_cxgbe, OID_AUTO, attack_filter, CTLFLAG_RDTUN,
649     &t4_attack_filter, 0, "Drop suspicious traffic");
650 
651 static int t4_drop_ip_fragments = 0;
652 SYSCTL_INT(_hw_cxgbe, OID_AUTO, drop_ip_fragments, CTLFLAG_RDTUN,
653     &t4_drop_ip_fragments, 0, "Drop IP fragments");
654 
655 static int t4_drop_pkts_with_l2_errors = 1;
656 SYSCTL_INT(_hw_cxgbe, OID_AUTO, drop_pkts_with_l2_errors, CTLFLAG_RDTUN,
657     &t4_drop_pkts_with_l2_errors, 0,
658     "Drop all frames with Layer 2 length or checksum errors");
659 
660 static int t4_drop_pkts_with_l3_errors = 0;
661 SYSCTL_INT(_hw_cxgbe, OID_AUTO, drop_pkts_with_l3_errors, CTLFLAG_RDTUN,
662     &t4_drop_pkts_with_l3_errors, 0,
663     "Drop all frames with IP version, length, or checksum errors");
664 
665 static int t4_drop_pkts_with_l4_errors = 0;
666 SYSCTL_INT(_hw_cxgbe, OID_AUTO, drop_pkts_with_l4_errors, CTLFLAG_RDTUN,
667     &t4_drop_pkts_with_l4_errors, 0,
668     "Drop all frames with Layer 4 length, checksum, or other errors");
669 
670 #ifdef TCP_OFFLOAD
671 /*
672  * TOE tunables.
673  */
674 static int t4_cop_managed_offloading = 0;
675 SYSCTL_INT(_hw_cxgbe, OID_AUTO, cop_managed_offloading, CTLFLAG_RDTUN,
676     &t4_cop_managed_offloading, 0,
677     "COP (Connection Offload Policy) controls all TOE offload");
678 #endif
679 
680 #ifdef KERN_TLS
681 /*
682  * This enables KERN_TLS for all adapters if set.
683  */
684 static int t4_kern_tls = 0;
685 SYSCTL_INT(_hw_cxgbe, OID_AUTO, kern_tls, CTLFLAG_RDTUN, &t4_kern_tls, 0,
686     "Enable KERN_TLS mode for T6 adapters");
687 
688 SYSCTL_NODE(_hw_cxgbe, OID_AUTO, tls, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
689     "cxgbe(4) KERN_TLS parameters");
690 
691 static int t4_tls_inline_keys = 0;
692 SYSCTL_INT(_hw_cxgbe_tls, OID_AUTO, inline_keys, CTLFLAG_RDTUN,
693     &t4_tls_inline_keys, 0,
694     "Always pass TLS keys in work requests (1) or attempt to store TLS keys "
695     "in card memory.");
696 
697 static int t4_tls_combo_wrs = 0;
698 SYSCTL_INT(_hw_cxgbe_tls, OID_AUTO, combo_wrs, CTLFLAG_RDTUN, &t4_tls_combo_wrs,
699     0, "Attempt to combine TCB field updates with TLS record work requests.");
700 #endif
701 
702 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
703 static int vi_mac_funcs[] = {
704 	FW_VI_FUNC_ETH,
705 	FW_VI_FUNC_OFLD,
706 	FW_VI_FUNC_IWARP,
707 	FW_VI_FUNC_OPENISCSI,
708 	FW_VI_FUNC_OPENFCOE,
709 	FW_VI_FUNC_FOISCSI,
710 	FW_VI_FUNC_FOFCOE,
711 };
712 
713 struct intrs_and_queues {
714 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
715 	uint16_t num_vis;	/* number of VIs for each port */
716 	uint16_t nirq;		/* Total # of vectors */
717 	uint16_t ntxq;		/* # of NIC txq's for each port */
718 	uint16_t nrxq;		/* # of NIC rxq's for each port */
719 	uint16_t nofldtxq;	/* # of TOE/ETHOFLD txq's for each port */
720 	uint16_t nofldrxq;	/* # of TOE rxq's for each port */
721 	uint16_t nnmtxq;	/* # of netmap txq's */
722 	uint16_t nnmrxq;	/* # of netmap rxq's */
723 
724 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
725 	uint16_t ntxq_vi;	/* # of NIC txq's */
726 	uint16_t nrxq_vi;	/* # of NIC rxq's */
727 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
728 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
729 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
730 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
731 };
732 
733 static void setup_memwin(struct adapter *);
734 static void position_memwin(struct adapter *, int, uint32_t);
735 static int validate_mem_range(struct adapter *, uint32_t, uint32_t);
736 static int fwmtype_to_hwmtype(int);
737 static int validate_mt_off_len(struct adapter *, int, uint32_t, uint32_t,
738     uint32_t *);
739 static int fixup_devlog_params(struct adapter *);
740 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
741 static int contact_firmware(struct adapter *);
742 static int partition_resources(struct adapter *);
743 static int get_params__pre_init(struct adapter *);
744 static int set_params__pre_init(struct adapter *);
745 static int get_params__post_init(struct adapter *);
746 static int set_params__post_init(struct adapter *);
747 static void t4_set_desc(struct adapter *);
748 static bool fixed_ifmedia(struct port_info *);
749 static void build_medialist(struct port_info *);
750 static void init_link_config(struct port_info *);
751 static int fixup_link_config(struct port_info *);
752 static int apply_link_config(struct port_info *);
753 static int cxgbe_init_synchronized(struct vi_info *);
754 static int cxgbe_uninit_synchronized(struct vi_info *);
755 static int adapter_full_init(struct adapter *);
756 static void adapter_full_uninit(struct adapter *);
757 static int vi_full_init(struct vi_info *);
758 static void vi_full_uninit(struct vi_info *);
759 static int alloc_extra_vi(struct adapter *, struct port_info *, struct vi_info *);
760 static void quiesce_txq(struct sge_txq *);
761 static void quiesce_wrq(struct sge_wrq *);
762 static void quiesce_iq_fl(struct adapter *, struct sge_iq *, struct sge_fl *);
763 static void quiesce_vi(struct vi_info *);
764 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
765     driver_intr_t *, void *, char *);
766 static int t4_free_irq(struct adapter *, struct irq *);
767 static void t4_init_atid_table(struct adapter *);
768 static void t4_free_atid_table(struct adapter *);
769 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
770 static void vi_refresh_stats(struct vi_info *);
771 static void cxgbe_refresh_stats(struct vi_info *);
772 static void cxgbe_tick(void *);
773 static void vi_tick(void *);
774 static void cxgbe_sysctls(struct port_info *);
775 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
776 static int sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS);
777 static int sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS);
778 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
779 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
780 static int sysctl_tx_vm_wr(SYSCTL_HANDLER_ARGS);
781 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
782 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
783 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
784 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
785 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
786 static int sysctl_link_fec(SYSCTL_HANDLER_ARGS);
787 static int sysctl_requested_fec(SYSCTL_HANDLER_ARGS);
788 static int sysctl_module_fec(SYSCTL_HANDLER_ARGS);
789 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
790 static int sysctl_force_fec(SYSCTL_HANDLER_ARGS);
791 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
792 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
793 static int sysctl_vdd(SYSCTL_HANDLER_ARGS);
794 static int sysctl_reset_sensor(SYSCTL_HANDLER_ARGS);
795 static int sysctl_loadavg(SYSCTL_HANDLER_ARGS);
796 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
797 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
798 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
799 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
800 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
801 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
802 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
803 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
804 static int sysctl_tid_stats(SYSCTL_HANDLER_ARGS);
805 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
806 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
807 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
808 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
809 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
810 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
811 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
812 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
813 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
814 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
815 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
816 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
817 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
818 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
819 static int sysctl_tnl_stats(SYSCTL_HANDLER_ARGS);
820 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
821 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
822 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
823 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
824 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
825 static int sysctl_cpus(SYSCTL_HANDLER_ARGS);
826 static int sysctl_reset(SYSCTL_HANDLER_ARGS);
827 #ifdef TCP_OFFLOAD
828 static int sysctl_tls(SYSCTL_HANDLER_ARGS);
829 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
830 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
831 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
832 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
833 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
834 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
835 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
836 #endif
837 static int get_sge_context(struct adapter *, struct t4_sge_context *);
838 static int load_fw(struct adapter *, struct t4_data *);
839 static int load_cfg(struct adapter *, struct t4_data *);
840 static int load_boot(struct adapter *, struct t4_bootrom *);
841 static int load_bootcfg(struct adapter *, struct t4_data *);
842 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
843 static void free_offload_policy(struct t4_offload_policy *);
844 static int set_offload_policy(struct adapter *, struct t4_offload_policy *);
845 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
846 static int read_i2c(struct adapter *, struct t4_i2c_data *);
847 static int clear_stats(struct adapter *, u_int);
848 static int hold_clip_addr(struct adapter *, struct t4_clip_addr *);
849 static int release_clip_addr(struct adapter *, struct t4_clip_addr *);
850 #ifdef TCP_OFFLOAD
851 static int toe_capability(struct vi_info *, bool);
852 static int t4_deactivate_all_uld(struct adapter *);
853 static void t4_async_event(struct adapter *);
854 #endif
855 #ifdef KERN_TLS
856 static int ktls_capability(struct adapter *, bool);
857 #endif
858 static int mod_event(module_t, int, void *);
859 static int notify_siblings(device_t, int);
860 static uint64_t vi_get_counter(if_t, ift_counter);
861 static uint64_t cxgbe_get_counter(if_t, ift_counter);
862 static void enable_vxlan_rx(struct adapter *);
863 static void reset_adapter_task(void *, int);
864 static void fatal_error_task(void *, int);
865 static void dump_devlog(struct adapter *);
866 static void dump_cim_regs(struct adapter *);
867 static void dump_cimla(struct adapter *);
868 
869 struct {
870 	uint16_t device;
871 	char *desc;
872 } t4_pciids[] = {
873 	{0xa000, "Chelsio Terminator 4 FPGA"},
874 	{0x4400, "Chelsio T440-dbg"},
875 	{0x4401, "Chelsio T420-CR"},
876 	{0x4402, "Chelsio T422-CR"},
877 	{0x4403, "Chelsio T440-CR"},
878 	{0x4404, "Chelsio T420-BCH"},
879 	{0x4405, "Chelsio T440-BCH"},
880 	{0x4406, "Chelsio T440-CH"},
881 	{0x4407, "Chelsio T420-SO"},
882 	{0x4408, "Chelsio T420-CX"},
883 	{0x4409, "Chelsio T420-BT"},
884 	{0x440a, "Chelsio T404-BT"},
885 	{0x440e, "Chelsio T440-LP-CR"},
886 }, t5_pciids[] = {
887 	{0xb000, "Chelsio Terminator 5 FPGA"},
888 	{0x5400, "Chelsio T580-dbg"},
889 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
890 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
891 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
892 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
893 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
894 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
895 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
896 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
897 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
898 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
899 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
900 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
901 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
902 	{0x5418,  "Chelsio T540-BT"},		/* 4 x 10GBaseT */
903 	{0x5419,  "Chelsio T540-LP-BT"},	/* 4 x 10GBaseT */
904 	{0x541a,  "Chelsio T540-SO-BT"},	/* 4 x 10GBaseT, nomem */
905 	{0x541b,  "Chelsio T540-SO-CR"},	/* 4 x 10G, nomem */
906 
907 	/* Custom */
908 	{0x5483, "Custom T540-CR"},
909 	{0x5484, "Custom T540-BT"},
910 }, t6_pciids[] = {
911 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
912 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
913 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
914 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
915 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
916 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
917 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
918 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
919 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
920 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
921 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
922 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
923 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
924 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
925 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
926 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
927 
928 	/* Custom */
929 	{0x6480, "Custom T6225-CR"},
930 	{0x6481, "Custom T62100-CR"},
931 	{0x6482, "Custom T6225-CR"},
932 	{0x6483, "Custom T62100-CR"},
933 	{0x6484, "Custom T64100-CR"},
934 	{0x6485, "Custom T6240-SO"},
935 	{0x6486, "Custom T6225-SO-CR"},
936 	{0x6487, "Custom T6225-CR"},
937 };
938 
939 #ifdef TCP_OFFLOAD
940 /*
941  * service_iq_fl() has an iq and needs the fl.  Offset of fl from the iq should
942  * be exactly the same for both rxq and ofld_rxq.
943  */
944 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
945 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
946 #endif
947 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
948 
949 static int
950 t4_probe(device_t dev)
951 {
952 	int i;
953 	uint16_t v = pci_get_vendor(dev);
954 	uint16_t d = pci_get_device(dev);
955 	uint8_t f = pci_get_function(dev);
956 
957 	if (v != PCI_VENDOR_ID_CHELSIO)
958 		return (ENXIO);
959 
960 	/* Attach only to PF0 of the FPGA */
961 	if (d == 0xa000 && f != 0)
962 		return (ENXIO);
963 
964 	for (i = 0; i < nitems(t4_pciids); i++) {
965 		if (d == t4_pciids[i].device) {
966 			device_set_desc(dev, t4_pciids[i].desc);
967 			return (BUS_PROBE_DEFAULT);
968 		}
969 	}
970 
971 	return (ENXIO);
972 }
973 
974 static int
975 t5_probe(device_t dev)
976 {
977 	int i;
978 	uint16_t v = pci_get_vendor(dev);
979 	uint16_t d = pci_get_device(dev);
980 	uint8_t f = pci_get_function(dev);
981 
982 	if (v != PCI_VENDOR_ID_CHELSIO)
983 		return (ENXIO);
984 
985 	/* Attach only to PF0 of the FPGA */
986 	if (d == 0xb000 && f != 0)
987 		return (ENXIO);
988 
989 	for (i = 0; i < nitems(t5_pciids); i++) {
990 		if (d == t5_pciids[i].device) {
991 			device_set_desc(dev, t5_pciids[i].desc);
992 			return (BUS_PROBE_DEFAULT);
993 		}
994 	}
995 
996 	return (ENXIO);
997 }
998 
999 static int
1000 t6_probe(device_t dev)
1001 {
1002 	int i;
1003 	uint16_t v = pci_get_vendor(dev);
1004 	uint16_t d = pci_get_device(dev);
1005 
1006 	if (v != PCI_VENDOR_ID_CHELSIO)
1007 		return (ENXIO);
1008 
1009 	for (i = 0; i < nitems(t6_pciids); i++) {
1010 		if (d == t6_pciids[i].device) {
1011 			device_set_desc(dev, t6_pciids[i].desc);
1012 			return (BUS_PROBE_DEFAULT);
1013 		}
1014 	}
1015 
1016 	return (ENXIO);
1017 }
1018 
1019 static void
1020 t5_attribute_workaround(device_t dev)
1021 {
1022 	device_t root_port;
1023 	uint32_t v;
1024 
1025 	/*
1026 	 * The T5 chips do not properly echo the No Snoop and Relaxed
1027 	 * Ordering attributes when replying to a TLP from a Root
1028 	 * Port.  As a workaround, find the parent Root Port and
1029 	 * disable No Snoop and Relaxed Ordering.  Note that this
1030 	 * affects all devices under this root port.
1031 	 */
1032 	root_port = pci_find_pcie_root_port(dev);
1033 	if (root_port == NULL) {
1034 		device_printf(dev, "Unable to find parent root port\n");
1035 		return;
1036 	}
1037 
1038 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
1039 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
1040 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
1041 	    0)
1042 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
1043 		    device_get_nameunit(root_port));
1044 }
1045 
1046 static const struct devnames devnames[] = {
1047 	{
1048 		.nexus_name = "t4nex",
1049 		.ifnet_name = "cxgbe",
1050 		.vi_ifnet_name = "vcxgbe",
1051 		.pf03_drv_name = "t4iov",
1052 		.vf_nexus_name = "t4vf",
1053 		.vf_ifnet_name = "cxgbev"
1054 	}, {
1055 		.nexus_name = "t5nex",
1056 		.ifnet_name = "cxl",
1057 		.vi_ifnet_name = "vcxl",
1058 		.pf03_drv_name = "t5iov",
1059 		.vf_nexus_name = "t5vf",
1060 		.vf_ifnet_name = "cxlv"
1061 	}, {
1062 		.nexus_name = "t6nex",
1063 		.ifnet_name = "cc",
1064 		.vi_ifnet_name = "vcc",
1065 		.pf03_drv_name = "t6iov",
1066 		.vf_nexus_name = "t6vf",
1067 		.vf_ifnet_name = "ccv"
1068 	}
1069 };
1070 
1071 void
1072 t4_init_devnames(struct adapter *sc)
1073 {
1074 	int id;
1075 
1076 	id = chip_id(sc);
1077 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
1078 		sc->names = &devnames[id - CHELSIO_T4];
1079 	else {
1080 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
1081 		sc->names = NULL;
1082 	}
1083 }
1084 
1085 static int
1086 t4_ifnet_unit(struct adapter *sc, struct port_info *pi)
1087 {
1088 	const char *parent, *name;
1089 	long value;
1090 	int line, unit;
1091 
1092 	line = 0;
1093 	parent = device_get_nameunit(sc->dev);
1094 	name = sc->names->ifnet_name;
1095 	while (resource_find_dev(&line, name, &unit, "at", parent) == 0) {
1096 		if (resource_long_value(name, unit, "port", &value) == 0 &&
1097 		    value == pi->port_id)
1098 			return (unit);
1099 	}
1100 	return (-1);
1101 }
1102 
1103 static void
1104 t4_calibration(void *arg)
1105 {
1106 	struct adapter *sc;
1107 	struct clock_sync *cur, *nex;
1108 	uint64_t hw;
1109 	sbintime_t sbt;
1110 	int next_up;
1111 
1112 	sc = (struct adapter *)arg;
1113 
1114 	KASSERT((hw_off_limits(sc) == 0), ("hw_off_limits at t4_calibration"));
1115 	hw = t4_read_reg64(sc, A_SGE_TIMESTAMP_LO);
1116 	sbt = sbinuptime();
1117 
1118 	cur = &sc->cal_info[sc->cal_current];
1119 	next_up = (sc->cal_current + 1) % CNT_CAL_INFO;
1120 	nex = &sc->cal_info[next_up];
1121 	if (__predict_false(sc->cal_count == 0)) {
1122 		/* First time in, just get the values in */
1123 		cur->hw_cur = hw;
1124 		cur->sbt_cur = sbt;
1125 		sc->cal_count++;
1126 		goto done;
1127 	}
1128 
1129 	if (cur->hw_cur == hw) {
1130 		/* The clock is not advancing? */
1131 		sc->cal_count = 0;
1132 		atomic_store_rel_int(&cur->gen, 0);
1133 		goto done;
1134 	}
1135 
1136 	seqc_write_begin(&nex->gen);
1137 	nex->hw_prev = cur->hw_cur;
1138 	nex->sbt_prev = cur->sbt_cur;
1139 	nex->hw_cur = hw;
1140 	nex->sbt_cur = sbt;
1141 	seqc_write_end(&nex->gen);
1142 	sc->cal_current = next_up;
1143 done:
1144 	callout_reset_sbt_curcpu(&sc->cal_callout, SBT_1S, 0, t4_calibration,
1145 	    sc, C_DIRECT_EXEC);
1146 }
1147 
1148 static void
1149 t4_calibration_start(struct adapter *sc)
1150 {
1151 	/*
1152 	 * Here if we have not done a calibration
1153 	 * then do so otherwise start the appropriate
1154 	 * timer.
1155 	 */
1156 	int i;
1157 
1158 	for (i = 0; i < CNT_CAL_INFO; i++) {
1159 		sc->cal_info[i].gen = 0;
1160 	}
1161 	sc->cal_current = 0;
1162 	sc->cal_count = 0;
1163 	sc->cal_gen = 0;
1164 	t4_calibration(sc);
1165 }
1166 
1167 static int
1168 t4_attach(device_t dev)
1169 {
1170 	struct adapter *sc;
1171 	int rc = 0, i, j, rqidx, tqidx, nports;
1172 	struct make_dev_args mda;
1173 	struct intrs_and_queues iaq;
1174 	struct sge *s;
1175 	uint32_t *buf;
1176 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1177 	int ofld_tqidx;
1178 #endif
1179 #ifdef TCP_OFFLOAD
1180 	int ofld_rqidx;
1181 #endif
1182 #ifdef DEV_NETMAP
1183 	int nm_rqidx, nm_tqidx;
1184 #endif
1185 	int num_vis;
1186 
1187 	sc = device_get_softc(dev);
1188 	sc->dev = dev;
1189 	sysctl_ctx_init(&sc->ctx);
1190 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
1191 
1192 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
1193 		t5_attribute_workaround(dev);
1194 	pci_enable_busmaster(dev);
1195 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
1196 		uint32_t v;
1197 
1198 		pci_set_max_read_req(dev, 4096);
1199 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
1200 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
1201 		if (pcie_relaxed_ordering == 0 &&
1202 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
1203 			v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
1204 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
1205 		} else if (pcie_relaxed_ordering == 1 &&
1206 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
1207 			v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
1208 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
1209 		}
1210 	}
1211 
1212 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
1213 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
1214 	sc->traceq = -1;
1215 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
1216 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
1217 	    device_get_nameunit(dev));
1218 
1219 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
1220 	    device_get_nameunit(dev));
1221 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
1222 	t4_add_adapter(sc);
1223 
1224 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
1225 	TAILQ_INIT(&sc->sfl);
1226 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
1227 
1228 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
1229 
1230 	sc->policy = NULL;
1231 	rw_init(&sc->policy_lock, "connection offload policy");
1232 
1233 	callout_init(&sc->ktls_tick, 1);
1234 
1235 	callout_init(&sc->cal_callout, 1);
1236 
1237 	refcount_init(&sc->vxlan_refcount, 0);
1238 
1239 	TASK_INIT(&sc->reset_task, 0, reset_adapter_task, sc);
1240 	TASK_INIT(&sc->fatal_error_task, 0, fatal_error_task, sc);
1241 
1242 	sc->ctrlq_oid = SYSCTL_ADD_NODE(&sc->ctx,
1243 	    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->dev)), OID_AUTO, "ctrlq",
1244 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "control queues");
1245 	sc->fwq_oid = SYSCTL_ADD_NODE(&sc->ctx,
1246 	    SYSCTL_CHILDREN(device_get_sysctl_tree(sc->dev)), OID_AUTO, "fwq",
1247 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "firmware event queue");
1248 
1249 	rc = t4_map_bars_0_and_4(sc);
1250 	if (rc != 0)
1251 		goto done; /* error message displayed already */
1252 
1253 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
1254 
1255 	/* Prepare the adapter for operation. */
1256 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
1257 	rc = -t4_prep_adapter(sc, buf);
1258 	free(buf, M_CXGBE);
1259 	if (rc != 0) {
1260 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
1261 		goto done;
1262 	}
1263 
1264 	/*
1265 	 * This is the real PF# to which we're attaching.  Works from within PCI
1266 	 * passthrough environments too, where pci_get_function() could return a
1267 	 * different PF# depending on the passthrough configuration.  We need to
1268 	 * use the real PF# in all our communication with the firmware.
1269 	 */
1270 	j = t4_read_reg(sc, A_PL_WHOAMI);
1271 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
1272 	sc->mbox = sc->pf;
1273 
1274 	t4_init_devnames(sc);
1275 	if (sc->names == NULL) {
1276 		rc = ENOTSUP;
1277 		goto done; /* error message displayed already */
1278 	}
1279 
1280 	/*
1281 	 * Do this really early, with the memory windows set up even before the
1282 	 * character device.  The userland tool's register i/o and mem read
1283 	 * will work even in "recovery mode".
1284 	 */
1285 	setup_memwin(sc);
1286 	if (t4_init_devlog_params(sc, 0) == 0)
1287 		fixup_devlog_params(sc);
1288 	make_dev_args_init(&mda);
1289 	mda.mda_devsw = &t4_cdevsw;
1290 	mda.mda_uid = UID_ROOT;
1291 	mda.mda_gid = GID_WHEEL;
1292 	mda.mda_mode = 0600;
1293 	mda.mda_si_drv1 = sc;
1294 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
1295 	if (rc != 0)
1296 		device_printf(dev, "failed to create nexus char device: %d.\n",
1297 		    rc);
1298 
1299 	/* Go no further if recovery mode has been requested. */
1300 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
1301 		device_printf(dev, "recovery mode.\n");
1302 		goto done;
1303 	}
1304 
1305 #if defined(__i386__)
1306 	if ((cpu_feature & CPUID_CX8) == 0) {
1307 		device_printf(dev, "64 bit atomics not available.\n");
1308 		rc = ENOTSUP;
1309 		goto done;
1310 	}
1311 #endif
1312 
1313 	/* Contact the firmware and try to become the master driver. */
1314 	rc = contact_firmware(sc);
1315 	if (rc != 0)
1316 		goto done; /* error message displayed already */
1317 	MPASS(sc->flags & FW_OK);
1318 
1319 	rc = get_params__pre_init(sc);
1320 	if (rc != 0)
1321 		goto done; /* error message displayed already */
1322 
1323 	if (sc->flags & MASTER_PF) {
1324 		rc = partition_resources(sc);
1325 		if (rc != 0)
1326 			goto done; /* error message displayed already */
1327 		t4_intr_clear(sc);
1328 	}
1329 
1330 	rc = get_params__post_init(sc);
1331 	if (rc != 0)
1332 		goto done; /* error message displayed already */
1333 
1334 	rc = set_params__post_init(sc);
1335 	if (rc != 0)
1336 		goto done; /* error message displayed already */
1337 
1338 	rc = t4_map_bar_2(sc);
1339 	if (rc != 0)
1340 		goto done; /* error message displayed already */
1341 
1342 	rc = t4_create_dma_tag(sc);
1343 	if (rc != 0)
1344 		goto done; /* error message displayed already */
1345 
1346 	/*
1347 	 * First pass over all the ports - allocate VIs and initialize some
1348 	 * basic parameters like mac address, port type, etc.
1349 	 */
1350 	for_each_port(sc, i) {
1351 		struct port_info *pi;
1352 
1353 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
1354 		sc->port[i] = pi;
1355 
1356 		/* These must be set before t4_port_init */
1357 		pi->adapter = sc;
1358 		pi->port_id = i;
1359 		/*
1360 		 * XXX: vi[0] is special so we can't delay this allocation until
1361 		 * pi->nvi's final value is known.
1362 		 */
1363 		pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
1364 		    M_ZERO | M_WAITOK);
1365 
1366 		/*
1367 		 * Allocate the "main" VI and initialize parameters
1368 		 * like mac addr.
1369 		 */
1370 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1371 		if (rc != 0) {
1372 			device_printf(dev, "unable to initialize port %d: %d\n",
1373 			    i, rc);
1374 			free(pi->vi, M_CXGBE);
1375 			free(pi, M_CXGBE);
1376 			sc->port[i] = NULL;
1377 			goto done;
1378 		}
1379 
1380 		if (is_bt(pi->port_type))
1381 			setbit(&sc->bt_map, pi->tx_chan);
1382 		else
1383 			MPASS(!isset(&sc->bt_map, pi->tx_chan));
1384 
1385 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1386 		    device_get_nameunit(dev), i);
1387 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1388 		sc->chan_map[pi->tx_chan] = i;
1389 
1390 		/*
1391 		 * The MPS counter for FCS errors doesn't work correctly on the
1392 		 * T6 so we use the MAC counter here.  Which MAC is in use
1393 		 * depends on the link settings which will be known when the
1394 		 * link comes up.
1395 		 */
1396 		if (is_t6(sc)) {
1397 			pi->fcs_reg = -1;
1398 		} else if (is_t4(sc)) {
1399 			pi->fcs_reg = PORT_REG(pi->tx_chan,
1400 			    A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L);
1401 		} else {
1402 			pi->fcs_reg = T5_PORT_REG(pi->tx_chan,
1403 			    A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L);
1404 		}
1405 		pi->fcs_base = 0;
1406 
1407 		/* All VIs on this port share this media. */
1408 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1409 		    cxgbe_media_status);
1410 
1411 		PORT_LOCK(pi);
1412 		init_link_config(pi);
1413 		fixup_link_config(pi);
1414 		build_medialist(pi);
1415 		if (fixed_ifmedia(pi))
1416 			pi->flags |= FIXED_IFMEDIA;
1417 		PORT_UNLOCK(pi);
1418 
1419 		pi->dev = device_add_child(dev, sc->names->ifnet_name,
1420 		    t4_ifnet_unit(sc, pi));
1421 		if (pi->dev == NULL) {
1422 			device_printf(dev,
1423 			    "failed to add device for port %d.\n", i);
1424 			rc = ENXIO;
1425 			goto done;
1426 		}
1427 		pi->vi[0].dev = pi->dev;
1428 		device_set_softc(pi->dev, pi);
1429 	}
1430 
1431 	/*
1432 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1433 	 */
1434 	nports = sc->params.nports;
1435 	rc = cfg_itype_and_nqueues(sc, &iaq);
1436 	if (rc != 0)
1437 		goto done; /* error message displayed already */
1438 
1439 	num_vis = iaq.num_vis;
1440 	sc->intr_type = iaq.intr_type;
1441 	sc->intr_count = iaq.nirq;
1442 
1443 	s = &sc->sge;
1444 	s->nrxq = nports * iaq.nrxq;
1445 	s->ntxq = nports * iaq.ntxq;
1446 	if (num_vis > 1) {
1447 		s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1448 		s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1449 	}
1450 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1451 	s->neq += nports;		/* ctrl queues: 1 per port */
1452 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1453 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1454 	if (is_offload(sc) || is_ethoffload(sc)) {
1455 		s->nofldtxq = nports * iaq.nofldtxq;
1456 		if (num_vis > 1)
1457 			s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1458 		s->neq += s->nofldtxq;
1459 
1460 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_ofld_txq),
1461 		    M_CXGBE, M_ZERO | M_WAITOK);
1462 	}
1463 #endif
1464 #ifdef TCP_OFFLOAD
1465 	if (is_offload(sc)) {
1466 		s->nofldrxq = nports * iaq.nofldrxq;
1467 		if (num_vis > 1)
1468 			s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1469 		s->neq += s->nofldrxq;	/* free list */
1470 		s->niq += s->nofldrxq;
1471 
1472 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1473 		    M_CXGBE, M_ZERO | M_WAITOK);
1474 	}
1475 #endif
1476 #ifdef DEV_NETMAP
1477 	s->nnmrxq = 0;
1478 	s->nnmtxq = 0;
1479 	if (t4_native_netmap & NN_MAIN_VI) {
1480 		s->nnmrxq += nports * iaq.nnmrxq;
1481 		s->nnmtxq += nports * iaq.nnmtxq;
1482 	}
1483 	if (num_vis > 1 && t4_native_netmap & NN_EXTRA_VI) {
1484 		s->nnmrxq += nports * (num_vis - 1) * iaq.nnmrxq_vi;
1485 		s->nnmtxq += nports * (num_vis - 1) * iaq.nnmtxq_vi;
1486 	}
1487 	s->neq += s->nnmtxq + s->nnmrxq;
1488 	s->niq += s->nnmrxq;
1489 
1490 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1491 	    M_CXGBE, M_ZERO | M_WAITOK);
1492 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1493 	    M_CXGBE, M_ZERO | M_WAITOK);
1494 #endif
1495 	MPASS(s->niq <= s->iqmap_sz);
1496 	MPASS(s->neq <= s->eqmap_sz);
1497 
1498 	s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1499 	    M_ZERO | M_WAITOK);
1500 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1501 	    M_ZERO | M_WAITOK);
1502 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1503 	    M_ZERO | M_WAITOK);
1504 	s->iqmap = malloc(s->iqmap_sz * sizeof(struct sge_iq *), M_CXGBE,
1505 	    M_ZERO | M_WAITOK);
1506 	s->eqmap = malloc(s->eqmap_sz * sizeof(struct sge_eq *), M_CXGBE,
1507 	    M_ZERO | M_WAITOK);
1508 
1509 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1510 	    M_ZERO | M_WAITOK);
1511 
1512 	t4_init_l2t(sc, M_WAITOK);
1513 	t4_init_smt(sc, M_WAITOK);
1514 	t4_init_tx_sched(sc);
1515 	t4_init_atid_table(sc);
1516 #ifdef RATELIMIT
1517 	t4_init_etid_table(sc);
1518 #endif
1519 #ifdef INET6
1520 	t4_init_clip_table(sc);
1521 #endif
1522 	if (sc->vres.key.size != 0)
1523 		sc->key_map = vmem_create("T4TLS key map", sc->vres.key.start,
1524 		    sc->vres.key.size, 32, 0, M_FIRSTFIT | M_WAITOK);
1525 
1526 	/*
1527 	 * Second pass over the ports.  This time we know the number of rx and
1528 	 * tx queues that each port should get.
1529 	 */
1530 	rqidx = tqidx = 0;
1531 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1532 	ofld_tqidx = 0;
1533 #endif
1534 #ifdef TCP_OFFLOAD
1535 	ofld_rqidx = 0;
1536 #endif
1537 #ifdef DEV_NETMAP
1538 	nm_rqidx = nm_tqidx = 0;
1539 #endif
1540 	for_each_port(sc, i) {
1541 		struct port_info *pi = sc->port[i];
1542 		struct vi_info *vi;
1543 
1544 		if (pi == NULL)
1545 			continue;
1546 
1547 		pi->nvi = num_vis;
1548 		for_each_vi(pi, j, vi) {
1549 			vi->pi = pi;
1550 			vi->adapter = sc;
1551 			vi->first_intr = -1;
1552 			vi->qsize_rxq = t4_qsize_rxq;
1553 			vi->qsize_txq = t4_qsize_txq;
1554 
1555 			vi->first_rxq = rqidx;
1556 			vi->first_txq = tqidx;
1557 			vi->tmr_idx = t4_tmr_idx;
1558 			vi->pktc_idx = t4_pktc_idx;
1559 			vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1560 			vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1561 
1562 			rqidx += vi->nrxq;
1563 			tqidx += vi->ntxq;
1564 
1565 			if (j == 0 && vi->ntxq > 1)
1566 				vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1567 			else
1568 				vi->rsrv_noflowq = 0;
1569 
1570 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1571 			vi->first_ofld_txq = ofld_tqidx;
1572 			vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1573 			ofld_tqidx += vi->nofldtxq;
1574 #endif
1575 #ifdef TCP_OFFLOAD
1576 			vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1577 			vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1578 			vi->first_ofld_rxq = ofld_rqidx;
1579 			vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1580 
1581 			ofld_rqidx += vi->nofldrxq;
1582 #endif
1583 #ifdef DEV_NETMAP
1584 			vi->first_nm_rxq = nm_rqidx;
1585 			vi->first_nm_txq = nm_tqidx;
1586 			if (j == 0) {
1587 				vi->nnmrxq = iaq.nnmrxq;
1588 				vi->nnmtxq = iaq.nnmtxq;
1589 			} else {
1590 				vi->nnmrxq = iaq.nnmrxq_vi;
1591 				vi->nnmtxq = iaq.nnmtxq_vi;
1592 			}
1593 			nm_rqidx += vi->nnmrxq;
1594 			nm_tqidx += vi->nnmtxq;
1595 #endif
1596 		}
1597 	}
1598 
1599 	rc = t4_setup_intr_handlers(sc);
1600 	if (rc != 0) {
1601 		device_printf(dev,
1602 		    "failed to setup interrupt handlers: %d\n", rc);
1603 		goto done;
1604 	}
1605 
1606 	rc = bus_generic_probe(dev);
1607 	if (rc != 0) {
1608 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1609 		goto done;
1610 	}
1611 
1612 	/*
1613 	 * Ensure thread-safe mailbox access (in debug builds).
1614 	 *
1615 	 * So far this was the only thread accessing the mailbox but various
1616 	 * ifnets and sysctls are about to be created and their handlers/ioctls
1617 	 * will access the mailbox from different threads.
1618 	 */
1619 	sc->flags |= CHK_MBOX_ACCESS;
1620 
1621 	rc = bus_generic_attach(dev);
1622 	if (rc != 0) {
1623 		device_printf(dev,
1624 		    "failed to attach all child ports: %d\n", rc);
1625 		goto done;
1626 	}
1627 	t4_calibration_start(sc);
1628 
1629 	device_printf(dev,
1630 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1631 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1632 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1633 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1634 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1635 
1636 	t4_set_desc(sc);
1637 
1638 	notify_siblings(dev, 0);
1639 
1640 done:
1641 	if (rc != 0 && sc->cdev) {
1642 		/* cdev was created and so cxgbetool works; recover that way. */
1643 		device_printf(dev,
1644 		    "error during attach, adapter is now in recovery mode.\n");
1645 		rc = 0;
1646 	}
1647 
1648 	if (rc != 0)
1649 		t4_detach_common(dev);
1650 	else
1651 		t4_sysctls(sc);
1652 
1653 	return (rc);
1654 }
1655 
1656 static int
1657 t4_child_location(device_t bus, device_t dev, struct sbuf *sb)
1658 {
1659 	struct adapter *sc;
1660 	struct port_info *pi;
1661 	int i;
1662 
1663 	sc = device_get_softc(bus);
1664 	for_each_port(sc, i) {
1665 		pi = sc->port[i];
1666 		if (pi != NULL && pi->dev == dev) {
1667 			sbuf_printf(sb, "port=%d", pi->port_id);
1668 			break;
1669 		}
1670 	}
1671 	return (0);
1672 }
1673 
1674 static int
1675 t4_ready(device_t dev)
1676 {
1677 	struct adapter *sc;
1678 
1679 	sc = device_get_softc(dev);
1680 	if (sc->flags & FW_OK)
1681 		return (0);
1682 	return (ENXIO);
1683 }
1684 
1685 static int
1686 t4_read_port_device(device_t dev, int port, device_t *child)
1687 {
1688 	struct adapter *sc;
1689 	struct port_info *pi;
1690 
1691 	sc = device_get_softc(dev);
1692 	if (port < 0 || port >= MAX_NPORTS)
1693 		return (EINVAL);
1694 	pi = sc->port[port];
1695 	if (pi == NULL || pi->dev == NULL)
1696 		return (ENXIO);
1697 	*child = pi->dev;
1698 	return (0);
1699 }
1700 
1701 static int
1702 notify_siblings(device_t dev, int detaching)
1703 {
1704 	device_t sibling;
1705 	int error, i;
1706 
1707 	error = 0;
1708 	for (i = 0; i < PCI_FUNCMAX; i++) {
1709 		if (i == pci_get_function(dev))
1710 			continue;
1711 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1712 		    pci_get_slot(dev), i);
1713 		if (sibling == NULL || !device_is_attached(sibling))
1714 			continue;
1715 		if (detaching)
1716 			error = T4_DETACH_CHILD(sibling);
1717 		else
1718 			(void)T4_ATTACH_CHILD(sibling);
1719 		if (error)
1720 			break;
1721 	}
1722 	return (error);
1723 }
1724 
1725 /*
1726  * Idempotent
1727  */
1728 static int
1729 t4_detach(device_t dev)
1730 {
1731 	int rc;
1732 
1733 	rc = notify_siblings(dev, 1);
1734 	if (rc) {
1735 		device_printf(dev,
1736 		    "failed to detach sibling devices: %d\n", rc);
1737 		return (rc);
1738 	}
1739 
1740 	return (t4_detach_common(dev));
1741 }
1742 
1743 int
1744 t4_detach_common(device_t dev)
1745 {
1746 	struct adapter *sc;
1747 	struct port_info *pi;
1748 	int i, rc;
1749 
1750 	sc = device_get_softc(dev);
1751 
1752 #ifdef TCP_OFFLOAD
1753 	rc = t4_deactivate_all_uld(sc);
1754 	if (rc) {
1755 		device_printf(dev,
1756 		    "failed to detach upper layer drivers: %d\n", rc);
1757 		return (rc);
1758 	}
1759 #endif
1760 
1761 	if (sc->cdev) {
1762 		destroy_dev(sc->cdev);
1763 		sc->cdev = NULL;
1764 	}
1765 
1766 	sx_xlock(&t4_list_lock);
1767 	SLIST_REMOVE(&t4_list, sc, adapter, link);
1768 	sx_xunlock(&t4_list_lock);
1769 
1770 	sc->flags &= ~CHK_MBOX_ACCESS;
1771 	if (sc->flags & FULL_INIT_DONE) {
1772 		if (!(sc->flags & IS_VF))
1773 			t4_intr_disable(sc);
1774 	}
1775 
1776 	if (device_is_attached(dev)) {
1777 		rc = bus_generic_detach(dev);
1778 		if (rc) {
1779 			device_printf(dev,
1780 			    "failed to detach child devices: %d\n", rc);
1781 			return (rc);
1782 		}
1783 	}
1784 
1785 	for (i = 0; i < sc->intr_count; i++)
1786 		t4_free_irq(sc, &sc->irq[i]);
1787 
1788 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1789 		t4_free_tx_sched(sc);
1790 
1791 	for (i = 0; i < MAX_NPORTS; i++) {
1792 		pi = sc->port[i];
1793 		if (pi) {
1794 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1795 			if (pi->dev)
1796 				device_delete_child(dev, pi->dev);
1797 
1798 			mtx_destroy(&pi->pi_lock);
1799 			free(pi->vi, M_CXGBE);
1800 			free(pi, M_CXGBE);
1801 		}
1802 	}
1803 	callout_stop(&sc->cal_callout);
1804 	callout_drain(&sc->cal_callout);
1805 	device_delete_children(dev);
1806 	sysctl_ctx_free(&sc->ctx);
1807 	adapter_full_uninit(sc);
1808 
1809 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1810 		t4_fw_bye(sc, sc->mbox);
1811 
1812 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1813 		pci_release_msi(dev);
1814 
1815 	if (sc->regs_res)
1816 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1817 		    sc->regs_res);
1818 
1819 	if (sc->udbs_res)
1820 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1821 		    sc->udbs_res);
1822 
1823 	if (sc->msix_res)
1824 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1825 		    sc->msix_res);
1826 
1827 	if (sc->l2t)
1828 		t4_free_l2t(sc->l2t);
1829 	if (sc->smt)
1830 		t4_free_smt(sc->smt);
1831 	t4_free_atid_table(sc);
1832 #ifdef RATELIMIT
1833 	t4_free_etid_table(sc);
1834 #endif
1835 	if (sc->key_map)
1836 		vmem_destroy(sc->key_map);
1837 #ifdef INET6
1838 	t4_destroy_clip_table(sc);
1839 #endif
1840 
1841 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1842 	free(sc->sge.ofld_txq, M_CXGBE);
1843 #endif
1844 #ifdef TCP_OFFLOAD
1845 	free(sc->sge.ofld_rxq, M_CXGBE);
1846 #endif
1847 #ifdef DEV_NETMAP
1848 	free(sc->sge.nm_rxq, M_CXGBE);
1849 	free(sc->sge.nm_txq, M_CXGBE);
1850 #endif
1851 	free(sc->irq, M_CXGBE);
1852 	free(sc->sge.rxq, M_CXGBE);
1853 	free(sc->sge.txq, M_CXGBE);
1854 	free(sc->sge.ctrlq, M_CXGBE);
1855 	free(sc->sge.iqmap, M_CXGBE);
1856 	free(sc->sge.eqmap, M_CXGBE);
1857 	free(sc->tids.ftid_tab, M_CXGBE);
1858 	free(sc->tids.hpftid_tab, M_CXGBE);
1859 	free_hftid_hash(&sc->tids);
1860 	free(sc->tids.tid_tab, M_CXGBE);
1861 	t4_destroy_dma_tag(sc);
1862 
1863 	callout_drain(&sc->ktls_tick);
1864 	callout_drain(&sc->sfl_callout);
1865 	if (mtx_initialized(&sc->tids.ftid_lock)) {
1866 		mtx_destroy(&sc->tids.ftid_lock);
1867 		cv_destroy(&sc->tids.ftid_cv);
1868 	}
1869 	if (mtx_initialized(&sc->tids.atid_lock))
1870 		mtx_destroy(&sc->tids.atid_lock);
1871 	if (mtx_initialized(&sc->ifp_lock))
1872 		mtx_destroy(&sc->ifp_lock);
1873 
1874 	if (rw_initialized(&sc->policy_lock)) {
1875 		rw_destroy(&sc->policy_lock);
1876 #ifdef TCP_OFFLOAD
1877 		if (sc->policy != NULL)
1878 			free_offload_policy(sc->policy);
1879 #endif
1880 	}
1881 
1882 	for (i = 0; i < NUM_MEMWIN; i++) {
1883 		struct memwin *mw = &sc->memwin[i];
1884 
1885 		if (rw_initialized(&mw->mw_lock))
1886 			rw_destroy(&mw->mw_lock);
1887 	}
1888 
1889 	mtx_destroy(&sc->sfl_lock);
1890 	mtx_destroy(&sc->reg_lock);
1891 	mtx_destroy(&sc->sc_lock);
1892 
1893 	bzero(sc, sizeof(*sc));
1894 
1895 	return (0);
1896 }
1897 
1898 static inline bool
1899 ok_to_reset(struct adapter *sc)
1900 {
1901 	struct tid_info *t = &sc->tids;
1902 	struct port_info *pi;
1903 	struct vi_info *vi;
1904 	int i, j;
1905 	int caps = IFCAP_TOE | IFCAP_NETMAP | IFCAP_TXRTLMT;
1906 
1907 	if (is_t6(sc))
1908 		caps |= IFCAP_TXTLS;
1909 
1910 	ASSERT_SYNCHRONIZED_OP(sc);
1911 	MPASS(!(sc->flags & IS_VF));
1912 
1913 	for_each_port(sc, i) {
1914 		pi = sc->port[i];
1915 		for_each_vi(pi, j, vi) {
1916 			if (if_getcapenable(vi->ifp) & caps)
1917 				return (false);
1918 		}
1919 	}
1920 
1921 	if (atomic_load_int(&t->tids_in_use) > 0)
1922 		return (false);
1923 	if (atomic_load_int(&t->stids_in_use) > 0)
1924 		return (false);
1925 	if (atomic_load_int(&t->atids_in_use) > 0)
1926 		return (false);
1927 	if (atomic_load_int(&t->ftids_in_use) > 0)
1928 		return (false);
1929 	if (atomic_load_int(&t->hpftids_in_use) > 0)
1930 		return (false);
1931 	if (atomic_load_int(&t->etids_in_use) > 0)
1932 		return (false);
1933 
1934 	return (true);
1935 }
1936 
1937 static inline int
1938 stop_adapter(struct adapter *sc)
1939 {
1940 	if (atomic_testandset_int(&sc->error_flags, ilog2(ADAP_STOPPED)))
1941 		return (1);		/* Already stopped. */
1942 	return (t4_shutdown_adapter(sc));
1943 }
1944 
1945 static int
1946 t4_suspend(device_t dev)
1947 {
1948 	struct adapter *sc = device_get_softc(dev);
1949 	struct port_info *pi;
1950 	struct vi_info *vi;
1951 	if_t ifp;
1952 	struct sge_rxq *rxq;
1953 	struct sge_txq *txq;
1954 	struct sge_wrq *wrq;
1955 #ifdef TCP_OFFLOAD
1956 	struct sge_ofld_rxq *ofld_rxq;
1957 #endif
1958 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1959 	struct sge_ofld_txq *ofld_txq;
1960 #endif
1961 	int rc, i, j, k;
1962 
1963 	CH_ALERT(sc, "suspend requested\n");
1964 
1965 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK, "t4sus");
1966 	if (rc != 0)
1967 		return (ENXIO);
1968 
1969 	/* XXX: Can the kernel call suspend repeatedly without resume? */
1970 	MPASS(!hw_off_limits(sc));
1971 
1972 	if (!ok_to_reset(sc)) {
1973 		/* XXX: should list what resource is preventing suspend. */
1974 		CH_ERR(sc, "not safe to suspend.\n");
1975 		rc = EBUSY;
1976 		goto done;
1977 	}
1978 
1979 	/* No more DMA or interrupts. */
1980 	stop_adapter(sc);
1981 
1982 	/* Quiesce all activity. */
1983 	for_each_port(sc, i) {
1984 		pi = sc->port[i];
1985 		pi->vxlan_tcam_entry = false;
1986 
1987 		PORT_LOCK(pi);
1988 		if (pi->up_vis > 0) {
1989 			/*
1990 			 * t4_shutdown_adapter has already shut down all the
1991 			 * PHYs but it also disables interrupts and DMA so there
1992 			 * won't be a link interrupt.  So we update the state
1993 			 * manually and inform the kernel.
1994 			 */
1995 			pi->link_cfg.link_ok = false;
1996 			t4_os_link_changed(pi);
1997 		}
1998 		PORT_UNLOCK(pi);
1999 
2000 		for_each_vi(pi, j, vi) {
2001 			vi->xact_addr_filt = -1;
2002 			mtx_lock(&vi->tick_mtx);
2003 			vi->flags |= VI_SKIP_STATS;
2004 			mtx_unlock(&vi->tick_mtx);
2005 			if (!(vi->flags & VI_INIT_DONE))
2006 				continue;
2007 
2008 			ifp = vi->ifp;
2009 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
2010 				mtx_lock(&vi->tick_mtx);
2011 				callout_stop(&vi->tick);
2012 				mtx_unlock(&vi->tick_mtx);
2013 				callout_drain(&vi->tick);
2014 			}
2015 
2016 			/*
2017 			 * Note that the HW is not available.
2018 			 */
2019 			for_each_txq(vi, k, txq) {
2020 				TXQ_LOCK(txq);
2021 				txq->eq.flags &= ~(EQ_ENABLED | EQ_HW_ALLOCATED);
2022 				TXQ_UNLOCK(txq);
2023 			}
2024 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
2025 			for_each_ofld_txq(vi, k, ofld_txq) {
2026 				ofld_txq->wrq.eq.flags &= ~EQ_HW_ALLOCATED;
2027 			}
2028 #endif
2029 			for_each_rxq(vi, k, rxq) {
2030 				rxq->iq.flags &= ~IQ_HW_ALLOCATED;
2031 			}
2032 #if defined(TCP_OFFLOAD)
2033 			for_each_ofld_rxq(vi, k, ofld_rxq) {
2034 				ofld_rxq->iq.flags &= ~IQ_HW_ALLOCATED;
2035 			}
2036 #endif
2037 
2038 			quiesce_vi(vi);
2039 		}
2040 
2041 		if (sc->flags & FULL_INIT_DONE) {
2042 			/* Control queue */
2043 			wrq = &sc->sge.ctrlq[i];
2044 			wrq->eq.flags &= ~EQ_HW_ALLOCATED;
2045 			quiesce_wrq(wrq);
2046 		}
2047 	}
2048 	if (sc->flags & FULL_INIT_DONE) {
2049 		/* Firmware event queue */
2050 		sc->sge.fwq.flags &= ~IQ_HW_ALLOCATED;
2051 		quiesce_iq_fl(sc, &sc->sge.fwq, NULL);
2052 	}
2053 
2054 	/* Stop calibration */
2055 	callout_stop(&sc->cal_callout);
2056 	callout_drain(&sc->cal_callout);
2057 
2058 	/* Mark the adapter totally off limits. */
2059 	mtx_lock(&sc->reg_lock);
2060 	atomic_set_int(&sc->error_flags, HW_OFF_LIMITS);
2061 	sc->flags &= ~(FW_OK | MASTER_PF);
2062 	sc->reset_thread = NULL;
2063 	mtx_unlock(&sc->reg_lock);
2064 
2065 	if (t4_clock_gate_on_suspend) {
2066 		t4_set_reg_field(sc, A_PMU_PART_CG_PWRMODE, F_MA_PART_CGEN |
2067 		    F_LE_PART_CGEN | F_EDC1_PART_CGEN | F_EDC0_PART_CGEN |
2068 		    F_TP_PART_CGEN | F_PDP_PART_CGEN | F_SGE_PART_CGEN, 0);
2069 	}
2070 
2071 	CH_ALERT(sc, "suspend completed.\n");
2072 done:
2073 	end_synchronized_op(sc, 0);
2074 	return (rc);
2075 }
2076 
2077 struct adapter_pre_reset_state {
2078 	u_int flags;
2079 	uint16_t nbmcaps;
2080 	uint16_t linkcaps;
2081 	uint16_t switchcaps;
2082 	uint16_t niccaps;
2083 	uint16_t toecaps;
2084 	uint16_t rdmacaps;
2085 	uint16_t cryptocaps;
2086 	uint16_t iscsicaps;
2087 	uint16_t fcoecaps;
2088 
2089 	u_int cfcsum;
2090 	char cfg_file[32];
2091 
2092 	struct adapter_params params;
2093 	struct t4_virt_res vres;
2094 	struct tid_info tids;
2095 	struct sge sge;
2096 
2097 	int rawf_base;
2098 	int nrawf;
2099 
2100 };
2101 
2102 static void
2103 save_caps_and_params(struct adapter *sc, struct adapter_pre_reset_state *o)
2104 {
2105 
2106 	ASSERT_SYNCHRONIZED_OP(sc);
2107 
2108 	o->flags = sc->flags;
2109 
2110 	o->nbmcaps =  sc->nbmcaps;
2111 	o->linkcaps = sc->linkcaps;
2112 	o->switchcaps = sc->switchcaps;
2113 	o->niccaps = sc->niccaps;
2114 	o->toecaps = sc->toecaps;
2115 	o->rdmacaps = sc->rdmacaps;
2116 	o->cryptocaps = sc->cryptocaps;
2117 	o->iscsicaps = sc->iscsicaps;
2118 	o->fcoecaps = sc->fcoecaps;
2119 
2120 	o->cfcsum = sc->cfcsum;
2121 	MPASS(sizeof(o->cfg_file) == sizeof(sc->cfg_file));
2122 	memcpy(o->cfg_file, sc->cfg_file, sizeof(o->cfg_file));
2123 
2124 	o->params = sc->params;
2125 	o->vres = sc->vres;
2126 	o->tids = sc->tids;
2127 	o->sge = sc->sge;
2128 
2129 	o->rawf_base = sc->rawf_base;
2130 	o->nrawf = sc->nrawf;
2131 }
2132 
2133 static int
2134 compare_caps_and_params(struct adapter *sc, struct adapter_pre_reset_state *o)
2135 {
2136 	int rc = 0;
2137 
2138 	ASSERT_SYNCHRONIZED_OP(sc);
2139 
2140 	/* Capabilities */
2141 #define COMPARE_CAPS(c) do { \
2142 	if (o->c##caps != sc->c##caps) { \
2143 		CH_ERR(sc, "%scaps 0x%04x -> 0x%04x.\n", #c, o->c##caps, \
2144 		    sc->c##caps); \
2145 		rc = EINVAL; \
2146 	} \
2147 } while (0)
2148 	COMPARE_CAPS(nbm);
2149 	COMPARE_CAPS(link);
2150 	COMPARE_CAPS(switch);
2151 	COMPARE_CAPS(nic);
2152 	COMPARE_CAPS(toe);
2153 	COMPARE_CAPS(rdma);
2154 	COMPARE_CAPS(crypto);
2155 	COMPARE_CAPS(iscsi);
2156 	COMPARE_CAPS(fcoe);
2157 #undef COMPARE_CAPS
2158 
2159 	/* Firmware config file */
2160 	if (o->cfcsum != sc->cfcsum) {
2161 		CH_ERR(sc, "config file %s (0x%x) -> %s (0x%x)\n", o->cfg_file,
2162 		    o->cfcsum, sc->cfg_file, sc->cfcsum);
2163 		rc = EINVAL;
2164 	}
2165 
2166 #define COMPARE_PARAM(p, name) do { \
2167 	if (o->p != sc->p) { \
2168 		CH_ERR(sc, #name " %d -> %d\n", o->p, sc->p); \
2169 		rc = EINVAL; \
2170 	} \
2171 } while (0)
2172 	COMPARE_PARAM(sge.iq_start, iq_start);
2173 	COMPARE_PARAM(sge.eq_start, eq_start);
2174 	COMPARE_PARAM(tids.ftid_base, ftid_base);
2175 	COMPARE_PARAM(tids.ftid_end, ftid_end);
2176 	COMPARE_PARAM(tids.nftids, nftids);
2177 	COMPARE_PARAM(vres.l2t.start, l2t_start);
2178 	COMPARE_PARAM(vres.l2t.size, l2t_size);
2179 	COMPARE_PARAM(sge.iqmap_sz, iqmap_sz);
2180 	COMPARE_PARAM(sge.eqmap_sz, eqmap_sz);
2181 	COMPARE_PARAM(tids.tid_base, tid_base);
2182 	COMPARE_PARAM(tids.hpftid_base, hpftid_base);
2183 	COMPARE_PARAM(tids.hpftid_end, hpftid_end);
2184 	COMPARE_PARAM(tids.nhpftids, nhpftids);
2185 	COMPARE_PARAM(rawf_base, rawf_base);
2186 	COMPARE_PARAM(nrawf, nrawf);
2187 	COMPARE_PARAM(params.mps_bg_map, mps_bg_map);
2188 	COMPARE_PARAM(params.filter2_wr_support, filter2_wr_support);
2189 	COMPARE_PARAM(params.ulptx_memwrite_dsgl, ulptx_memwrite_dsgl);
2190 	COMPARE_PARAM(params.fr_nsmr_tpte_wr_support, fr_nsmr_tpte_wr_support);
2191 	COMPARE_PARAM(params.max_pkts_per_eth_tx_pkts_wr, max_pkts_per_eth_tx_pkts_wr);
2192 	COMPARE_PARAM(tids.ntids, ntids);
2193 	COMPARE_PARAM(tids.etid_base, etid_base);
2194 	COMPARE_PARAM(tids.etid_end, etid_end);
2195 	COMPARE_PARAM(tids.netids, netids);
2196 	COMPARE_PARAM(params.eo_wr_cred, eo_wr_cred);
2197 	COMPARE_PARAM(params.ethoffload, ethoffload);
2198 	COMPARE_PARAM(tids.natids, natids);
2199 	COMPARE_PARAM(tids.stid_base, stid_base);
2200 	COMPARE_PARAM(vres.ddp.start, ddp_start);
2201 	COMPARE_PARAM(vres.ddp.size, ddp_size);
2202 	COMPARE_PARAM(params.ofldq_wr_cred, ofldq_wr_cred);
2203 	COMPARE_PARAM(vres.stag.start, stag_start);
2204 	COMPARE_PARAM(vres.stag.size, stag_size);
2205 	COMPARE_PARAM(vres.rq.start, rq_start);
2206 	COMPARE_PARAM(vres.rq.size, rq_size);
2207 	COMPARE_PARAM(vres.pbl.start, pbl_start);
2208 	COMPARE_PARAM(vres.pbl.size, pbl_size);
2209 	COMPARE_PARAM(vres.qp.start, qp_start);
2210 	COMPARE_PARAM(vres.qp.size, qp_size);
2211 	COMPARE_PARAM(vres.cq.start, cq_start);
2212 	COMPARE_PARAM(vres.cq.size, cq_size);
2213 	COMPARE_PARAM(vres.ocq.start, ocq_start);
2214 	COMPARE_PARAM(vres.ocq.size, ocq_size);
2215 	COMPARE_PARAM(vres.srq.start, srq_start);
2216 	COMPARE_PARAM(vres.srq.size, srq_size);
2217 	COMPARE_PARAM(params.max_ordird_qp, max_ordird_qp);
2218 	COMPARE_PARAM(params.max_ird_adapter, max_ird_adapter);
2219 	COMPARE_PARAM(vres.iscsi.start, iscsi_start);
2220 	COMPARE_PARAM(vres.iscsi.size, iscsi_size);
2221 	COMPARE_PARAM(vres.key.start, key_start);
2222 	COMPARE_PARAM(vres.key.size, key_size);
2223 #undef COMPARE_PARAM
2224 
2225 	return (rc);
2226 }
2227 
2228 static int
2229 t4_resume(device_t dev)
2230 {
2231 	struct adapter *sc = device_get_softc(dev);
2232 	struct adapter_pre_reset_state *old_state = NULL;
2233 	struct port_info *pi;
2234 	struct vi_info *vi;
2235 	if_t ifp;
2236 	struct sge_txq *txq;
2237 	int rc, i, j, k;
2238 
2239 	CH_ALERT(sc, "resume requested.\n");
2240 
2241 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK, "t4res");
2242 	if (rc != 0)
2243 		return (ENXIO);
2244 	MPASS(hw_off_limits(sc));
2245 	MPASS((sc->flags & FW_OK) == 0);
2246 	MPASS((sc->flags & MASTER_PF) == 0);
2247 	MPASS(sc->reset_thread == NULL);
2248 	sc->reset_thread = curthread;
2249 
2250 	/* Register access is expected to work by the time we're here. */
2251 	if (t4_read_reg(sc, A_PL_WHOAMI) == 0xffffffff) {
2252 		CH_ERR(sc, "%s: can't read device registers\n", __func__);
2253 		rc = ENXIO;
2254 		goto done;
2255 	}
2256 
2257 	/* Note that HW_OFF_LIMITS is cleared a bit later. */
2258 	atomic_clear_int(&sc->error_flags, ADAP_FATAL_ERR | ADAP_STOPPED);
2259 
2260 	/* Restore memory window. */
2261 	setup_memwin(sc);
2262 
2263 	/* Go no further if recovery mode has been requested. */
2264 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
2265 		CH_ALERT(sc, "recovery mode on resume.\n");
2266 		rc = 0;
2267 		mtx_lock(&sc->reg_lock);
2268 		atomic_clear_int(&sc->error_flags, HW_OFF_LIMITS);
2269 		mtx_unlock(&sc->reg_lock);
2270 		goto done;
2271 	}
2272 
2273 	old_state = malloc(sizeof(*old_state), M_CXGBE, M_ZERO | M_WAITOK);
2274 	save_caps_and_params(sc, old_state);
2275 
2276 	/* Reestablish contact with firmware and become the primary PF. */
2277 	rc = contact_firmware(sc);
2278 	if (rc != 0)
2279 		goto done; /* error message displayed already */
2280 	MPASS(sc->flags & FW_OK);
2281 
2282 	if (sc->flags & MASTER_PF) {
2283 		rc = partition_resources(sc);
2284 		if (rc != 0)
2285 			goto done; /* error message displayed already */
2286 		t4_intr_clear(sc);
2287 	}
2288 
2289 	rc = get_params__post_init(sc);
2290 	if (rc != 0)
2291 		goto done; /* error message displayed already */
2292 
2293 	rc = set_params__post_init(sc);
2294 	if (rc != 0)
2295 		goto done; /* error message displayed already */
2296 
2297 	rc = compare_caps_and_params(sc, old_state);
2298 	if (rc != 0)
2299 		goto done; /* error message displayed already */
2300 
2301 	for_each_port(sc, i) {
2302 		pi = sc->port[i];
2303 		MPASS(pi != NULL);
2304 		MPASS(pi->vi != NULL);
2305 		MPASS(pi->vi[0].dev == pi->dev);
2306 
2307 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
2308 		if (rc != 0) {
2309 			CH_ERR(sc,
2310 			    "failed to re-initialize port %d: %d\n", i, rc);
2311 			goto done;
2312 		}
2313 		MPASS(sc->chan_map[pi->tx_chan] == i);
2314 
2315 		PORT_LOCK(pi);
2316 		fixup_link_config(pi);
2317 		build_medialist(pi);
2318 		PORT_UNLOCK(pi);
2319 		for_each_vi(pi, j, vi) {
2320 			if (IS_MAIN_VI(vi))
2321 				continue;
2322 			rc = alloc_extra_vi(sc, pi, vi);
2323 			if (rc != 0) {
2324 				CH_ERR(vi,
2325 				    "failed to re-allocate extra VI: %d\n", rc);
2326 				goto done;
2327 			}
2328 		}
2329 	}
2330 
2331 	/*
2332 	 * Interrupts and queues are about to be enabled and other threads will
2333 	 * want to access the hardware too.  It is safe to do so.  Note that
2334 	 * this thread is still in the middle of a synchronized_op.
2335 	 */
2336 	mtx_lock(&sc->reg_lock);
2337 	atomic_clear_int(&sc->error_flags, HW_OFF_LIMITS);
2338 	mtx_unlock(&sc->reg_lock);
2339 
2340 	if (sc->flags & FULL_INIT_DONE) {
2341 		rc = adapter_full_init(sc);
2342 		if (rc != 0) {
2343 			CH_ERR(sc, "failed to re-initialize adapter: %d\n", rc);
2344 			goto done;
2345 		}
2346 
2347 		if (sc->vxlan_refcount > 0)
2348 			enable_vxlan_rx(sc);
2349 
2350 		for_each_port(sc, i) {
2351 			pi = sc->port[i];
2352 			for_each_vi(pi, j, vi) {
2353 				mtx_lock(&vi->tick_mtx);
2354 				vi->flags &= ~VI_SKIP_STATS;
2355 				mtx_unlock(&vi->tick_mtx);
2356 				if (!(vi->flags & VI_INIT_DONE))
2357 					continue;
2358 				rc = vi_full_init(vi);
2359 				if (rc != 0) {
2360 					CH_ERR(vi, "failed to re-initialize "
2361 					    "interface: %d\n", rc);
2362 					goto done;
2363 				}
2364 
2365 				ifp = vi->ifp;
2366 				if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING))
2367 					continue;
2368 				/*
2369 				 * Note that we do not setup multicast addresses
2370 				 * in the first pass.  This ensures that the
2371 				 * unicast DMACs for all VIs on all ports get an
2372 				 * MPS TCAM entry.
2373 				 */
2374 				rc = update_mac_settings(ifp, XGMAC_ALL &
2375 				    ~XGMAC_MCADDRS);
2376 				if (rc != 0) {
2377 					CH_ERR(vi, "failed to re-configure MAC: %d\n", rc);
2378 					goto done;
2379 				}
2380 				rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true,
2381 				    true);
2382 				if (rc != 0) {
2383 					CH_ERR(vi, "failed to re-enable VI: %d\n", rc);
2384 					goto done;
2385 				}
2386 				for_each_txq(vi, k, txq) {
2387 					TXQ_LOCK(txq);
2388 					txq->eq.flags |= EQ_ENABLED;
2389 					TXQ_UNLOCK(txq);
2390 				}
2391 				mtx_lock(&vi->tick_mtx);
2392 				callout_schedule(&vi->tick, hz);
2393 				mtx_unlock(&vi->tick_mtx);
2394 			}
2395 			PORT_LOCK(pi);
2396 			if (pi->up_vis > 0) {
2397 				t4_update_port_info(pi);
2398 				fixup_link_config(pi);
2399 				build_medialist(pi);
2400 				apply_link_config(pi);
2401 				if (pi->link_cfg.link_ok)
2402 					t4_os_link_changed(pi);
2403 			}
2404 			PORT_UNLOCK(pi);
2405 		}
2406 
2407 		/* Now reprogram the L2 multicast addresses. */
2408 		for_each_port(sc, i) {
2409 			pi = sc->port[i];
2410 			for_each_vi(pi, j, vi) {
2411 				if (!(vi->flags & VI_INIT_DONE))
2412 					continue;
2413 				ifp = vi->ifp;
2414 				if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING))
2415 					continue;
2416 				rc = update_mac_settings(ifp, XGMAC_MCADDRS);
2417 				if (rc != 0) {
2418 					CH_ERR(vi, "failed to re-configure MCAST MACs: %d\n", rc);
2419 					rc = 0;	/* carry on */
2420 				}
2421 			}
2422 		}
2423 	}
2424 
2425 	/* Reset all calibration */
2426 	t4_calibration_start(sc);
2427 
2428 done:
2429 	if (rc == 0) {
2430 		sc->incarnation++;
2431 		CH_ALERT(sc, "resume completed.\n");
2432 	}
2433 	end_synchronized_op(sc, 0);
2434 	free(old_state, M_CXGBE);
2435 	return (rc);
2436 }
2437 
2438 static int
2439 t4_reset_prepare(device_t dev, device_t child)
2440 {
2441 	struct adapter *sc = device_get_softc(dev);
2442 
2443 	CH_ALERT(sc, "reset_prepare.\n");
2444 	return (0);
2445 }
2446 
2447 static int
2448 t4_reset_post(device_t dev, device_t child)
2449 {
2450 	struct adapter *sc = device_get_softc(dev);
2451 
2452 	CH_ALERT(sc, "reset_post.\n");
2453 	return (0);
2454 }
2455 
2456 static int
2457 reset_adapter(struct adapter *sc)
2458 {
2459 	int rc, oldinc, error_flags;
2460 
2461 	CH_ALERT(sc, "reset requested.\n");
2462 
2463 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK, "t4rst1");
2464 	if (rc != 0)
2465 		return (EBUSY);
2466 
2467 	if (hw_off_limits(sc)) {
2468 		CH_ERR(sc, "adapter is suspended, use resume (not reset).\n");
2469 		rc = ENXIO;
2470 		goto done;
2471 	}
2472 
2473 	if (!ok_to_reset(sc)) {
2474 		/* XXX: should list what resource is preventing reset. */
2475 		CH_ERR(sc, "not safe to reset.\n");
2476 		rc = EBUSY;
2477 		goto done;
2478 	}
2479 
2480 done:
2481 	oldinc = sc->incarnation;
2482 	end_synchronized_op(sc, 0);
2483 	if (rc != 0)
2484 		return (rc);	/* Error logged already. */
2485 
2486 	atomic_add_int(&sc->num_resets, 1);
2487 	mtx_lock(&Giant);
2488 	rc = BUS_RESET_CHILD(device_get_parent(sc->dev), sc->dev, 0);
2489 	mtx_unlock(&Giant);
2490 	if (rc != 0)
2491 		CH_ERR(sc, "bus_reset_child failed: %d.\n", rc);
2492 	else {
2493 		rc = begin_synchronized_op(sc, NULL, SLEEP_OK, "t4rst2");
2494 		if (rc != 0)
2495 			return (EBUSY);
2496 		error_flags = atomic_load_int(&sc->error_flags);
2497 		if (sc->incarnation > oldinc && error_flags == 0) {
2498 			CH_ALERT(sc, "bus_reset_child succeeded.\n");
2499 		} else {
2500 			CH_ERR(sc, "adapter did not reset properly, flags "
2501 			    "0x%08x, error_flags 0x%08x.\n", sc->flags,
2502 			    error_flags);
2503 			rc = ENXIO;
2504 		}
2505 		end_synchronized_op(sc, 0);
2506 	}
2507 
2508 	return (rc);
2509 }
2510 
2511 static void
2512 reset_adapter_task(void *arg, int pending)
2513 {
2514 	/* XXX: t4_async_event here? */
2515 	reset_adapter(arg);
2516 }
2517 
2518 static int
2519 cxgbe_probe(device_t dev)
2520 {
2521 	char buf[128];
2522 	struct port_info *pi = device_get_softc(dev);
2523 
2524 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
2525 	device_set_desc_copy(dev, buf);
2526 
2527 	return (BUS_PROBE_DEFAULT);
2528 }
2529 
2530 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
2531     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
2532     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS | \
2533     IFCAP_HWRXTSTMP | IFCAP_MEXTPG)
2534 #define T4_CAP_ENABLE (T4_CAP)
2535 
2536 static int
2537 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
2538 {
2539 	if_t ifp;
2540 	struct sbuf *sb;
2541 	struct sysctl_ctx_list *ctx = &vi->ctx;
2542 	struct sysctl_oid_list *children;
2543 	struct pfil_head_args pa;
2544 	struct adapter *sc = vi->adapter;
2545 
2546 	sysctl_ctx_init(ctx);
2547 	children = SYSCTL_CHILDREN(device_get_sysctl_tree(vi->dev));
2548 	vi->rxq_oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rxq",
2549 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "NIC rx queues");
2550 	vi->txq_oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "txq",
2551 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "NIC tx queues");
2552 #ifdef DEV_NETMAP
2553 	vi->nm_rxq_oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "nm_rxq",
2554 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "netmap rx queues");
2555 	vi->nm_txq_oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "nm_txq",
2556 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "netmap tx queues");
2557 #endif
2558 #ifdef TCP_OFFLOAD
2559 	vi->ofld_rxq_oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "ofld_rxq",
2560 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "TOE rx queues");
2561 #endif
2562 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
2563 	vi->ofld_txq_oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "ofld_txq",
2564 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "TOE/ETHOFLD tx queues");
2565 #endif
2566 
2567 	vi->xact_addr_filt = -1;
2568 	mtx_init(&vi->tick_mtx, "vi tick", NULL, MTX_DEF);
2569 	callout_init_mtx(&vi->tick, &vi->tick_mtx, 0);
2570 	if (sc->flags & IS_VF || t4_tx_vm_wr != 0)
2571 		vi->flags |= TX_USES_VM_WR;
2572 
2573 	/* Allocate an ifnet and set it up */
2574 	ifp = if_alloc_dev(IFT_ETHER, dev);
2575 	if (ifp == NULL) {
2576 		device_printf(dev, "Cannot allocate ifnet\n");
2577 		return (ENOMEM);
2578 	}
2579 	vi->ifp = ifp;
2580 	if_setsoftc(ifp, vi);
2581 
2582 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
2583 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
2584 
2585 	if_setinitfn(ifp, cxgbe_init);
2586 	if_setioctlfn(ifp, cxgbe_ioctl);
2587 	if_settransmitfn(ifp, cxgbe_transmit);
2588 	if_setqflushfn(ifp, cxgbe_qflush);
2589 	if (vi->pi->nvi > 1 || sc->flags & IS_VF)
2590 		if_setgetcounterfn(ifp, vi_get_counter);
2591 	else
2592 		if_setgetcounterfn(ifp, cxgbe_get_counter);
2593 #if defined(KERN_TLS) || defined(RATELIMIT)
2594 	if_setsndtagallocfn(ifp, cxgbe_snd_tag_alloc);
2595 #endif
2596 #ifdef RATELIMIT
2597 	if_setratelimitqueryfn(ifp, cxgbe_ratelimit_query);
2598 #endif
2599 
2600 	if_setcapabilities(ifp, T4_CAP);
2601 	if_setcapenable(ifp, T4_CAP_ENABLE);
2602 	if_sethwassist(ifp, CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
2603 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
2604 	if (chip_id(sc) >= CHELSIO_T6) {
2605 		if_setcapabilitiesbit(ifp, IFCAP_VXLAN_HWCSUM | IFCAP_VXLAN_HWTSO, 0);
2606 		if_setcapenablebit(ifp, IFCAP_VXLAN_HWCSUM | IFCAP_VXLAN_HWTSO, 0);
2607 		if_sethwassistbits(ifp, CSUM_INNER_IP6_UDP | CSUM_INNER_IP6_TCP |
2608 		    CSUM_INNER_IP6_TSO | CSUM_INNER_IP | CSUM_INNER_IP_UDP |
2609 		    CSUM_INNER_IP_TCP | CSUM_INNER_IP_TSO | CSUM_ENCAP_VXLAN, 0);
2610 	}
2611 
2612 #ifdef TCP_OFFLOAD
2613 	if (vi->nofldrxq != 0)
2614 		if_setcapabilitiesbit(ifp, IFCAP_TOE, 0);
2615 #endif
2616 #ifdef RATELIMIT
2617 	if (is_ethoffload(sc) && vi->nofldtxq != 0) {
2618 		if_setcapabilitiesbit(ifp, IFCAP_TXRTLMT, 0);
2619 		if_setcapenablebit(ifp, IFCAP_TXRTLMT, 0);
2620 	}
2621 #endif
2622 
2623 	if_sethwtsomax(ifp, IP_MAXPACKET);
2624 	if (vi->flags & TX_USES_VM_WR)
2625 		if_sethwtsomaxsegcount(ifp, TX_SGL_SEGS_VM_TSO);
2626 	else
2627 		if_sethwtsomaxsegcount(ifp, TX_SGL_SEGS_TSO);
2628 #ifdef RATELIMIT
2629 	if (is_ethoffload(sc) && vi->nofldtxq != 0)
2630 		if_sethwtsomaxsegcount(ifp, TX_SGL_SEGS_EO_TSO);
2631 #endif
2632 	if_sethwtsomaxsegsize(ifp, 65536);
2633 #ifdef KERN_TLS
2634 	if (is_ktls(sc)) {
2635 		if_setcapabilitiesbit(ifp, IFCAP_TXTLS, 0);
2636 		if (sc->flags & KERN_TLS_ON || !is_t6(sc))
2637 			if_setcapenablebit(ifp, IFCAP_TXTLS, 0);
2638 	}
2639 #endif
2640 
2641 	ether_ifattach(ifp, vi->hw_addr);
2642 #ifdef DEV_NETMAP
2643 	if (vi->nnmrxq != 0)
2644 		cxgbe_nm_attach(vi);
2645 #endif
2646 	sb = sbuf_new_auto();
2647 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
2648 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
2649 	switch (if_getcapabilities(ifp) & (IFCAP_TOE | IFCAP_TXRTLMT)) {
2650 	case IFCAP_TOE:
2651 		sbuf_printf(sb, "; %d txq (TOE)", vi->nofldtxq);
2652 		break;
2653 	case IFCAP_TOE | IFCAP_TXRTLMT:
2654 		sbuf_printf(sb, "; %d txq (TOE/ETHOFLD)", vi->nofldtxq);
2655 		break;
2656 	case IFCAP_TXRTLMT:
2657 		sbuf_printf(sb, "; %d txq (ETHOFLD)", vi->nofldtxq);
2658 		break;
2659 	}
2660 #endif
2661 #ifdef TCP_OFFLOAD
2662 	if (if_getcapabilities(ifp) & IFCAP_TOE)
2663 		sbuf_printf(sb, ", %d rxq (TOE)", vi->nofldrxq);
2664 #endif
2665 #ifdef DEV_NETMAP
2666 	if (if_getcapabilities(ifp) & IFCAP_NETMAP)
2667 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
2668 		    vi->nnmtxq, vi->nnmrxq);
2669 #endif
2670 	sbuf_finish(sb);
2671 	device_printf(dev, "%s\n", sbuf_data(sb));
2672 	sbuf_delete(sb);
2673 
2674 	vi_sysctls(vi);
2675 
2676 	pa.pa_version = PFIL_VERSION;
2677 	pa.pa_flags = PFIL_IN;
2678 	pa.pa_type = PFIL_TYPE_ETHERNET;
2679 	pa.pa_headname = if_name(ifp);
2680 	vi->pfil = pfil_head_register(&pa);
2681 
2682 	return (0);
2683 }
2684 
2685 static int
2686 cxgbe_attach(device_t dev)
2687 {
2688 	struct port_info *pi = device_get_softc(dev);
2689 	struct adapter *sc = pi->adapter;
2690 	struct vi_info *vi;
2691 	int i, rc;
2692 
2693 	sysctl_ctx_init(&pi->ctx);
2694 
2695 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
2696 	if (rc)
2697 		return (rc);
2698 
2699 	for_each_vi(pi, i, vi) {
2700 		if (i == 0)
2701 			continue;
2702 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
2703 		if (vi->dev == NULL) {
2704 			device_printf(dev, "failed to add VI %d\n", i);
2705 			continue;
2706 		}
2707 		device_set_softc(vi->dev, vi);
2708 	}
2709 
2710 	cxgbe_sysctls(pi);
2711 
2712 	bus_generic_attach(dev);
2713 
2714 	return (0);
2715 }
2716 
2717 static void
2718 cxgbe_vi_detach(struct vi_info *vi)
2719 {
2720 	if_t ifp = vi->ifp;
2721 
2722 	if (vi->pfil != NULL) {
2723 		pfil_head_unregister(vi->pfil);
2724 		vi->pfil = NULL;
2725 	}
2726 
2727 	ether_ifdetach(ifp);
2728 
2729 	/* Let detach proceed even if these fail. */
2730 #ifdef DEV_NETMAP
2731 	if (if_getcapabilities(ifp) & IFCAP_NETMAP)
2732 		cxgbe_nm_detach(vi);
2733 #endif
2734 	cxgbe_uninit_synchronized(vi);
2735 	callout_drain(&vi->tick);
2736 	mtx_destroy(&vi->tick_mtx);
2737 	sysctl_ctx_free(&vi->ctx);
2738 	vi_full_uninit(vi);
2739 
2740 	if_free(vi->ifp);
2741 	vi->ifp = NULL;
2742 }
2743 
2744 static int
2745 cxgbe_detach(device_t dev)
2746 {
2747 	struct port_info *pi = device_get_softc(dev);
2748 	struct adapter *sc = pi->adapter;
2749 	int rc;
2750 
2751 	/* Detach the extra VIs first. */
2752 	rc = bus_generic_detach(dev);
2753 	if (rc)
2754 		return (rc);
2755 	device_delete_children(dev);
2756 
2757 	sysctl_ctx_free(&pi->ctx);
2758 	begin_vi_detach(sc, &pi->vi[0]);
2759 	if (pi->flags & HAS_TRACEQ) {
2760 		sc->traceq = -1;	/* cloner should not create ifnet */
2761 		t4_tracer_port_detach(sc);
2762 	}
2763 	cxgbe_vi_detach(&pi->vi[0]);
2764 	ifmedia_removeall(&pi->media);
2765 	end_vi_detach(sc, &pi->vi[0]);
2766 
2767 	return (0);
2768 }
2769 
2770 static void
2771 cxgbe_init(void *arg)
2772 {
2773 	struct vi_info *vi = arg;
2774 	struct adapter *sc = vi->adapter;
2775 
2776 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
2777 		return;
2778 	cxgbe_init_synchronized(vi);
2779 	end_synchronized_op(sc, 0);
2780 }
2781 
2782 static int
2783 cxgbe_ioctl(if_t ifp, unsigned long cmd, caddr_t data)
2784 {
2785 	int rc = 0, mtu, flags;
2786 	struct vi_info *vi = if_getsoftc(ifp);
2787 	struct port_info *pi = vi->pi;
2788 	struct adapter *sc = pi->adapter;
2789 	struct ifreq *ifr = (struct ifreq *)data;
2790 	uint32_t mask;
2791 
2792 	switch (cmd) {
2793 	case SIOCSIFMTU:
2794 		mtu = ifr->ifr_mtu;
2795 		if (mtu < ETHERMIN || mtu > MAX_MTU)
2796 			return (EINVAL);
2797 
2798 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
2799 		if (rc)
2800 			return (rc);
2801 		if_setmtu(ifp, mtu);
2802 		if (vi->flags & VI_INIT_DONE) {
2803 			t4_update_fl_bufsize(ifp);
2804 			if (!hw_off_limits(sc) &&
2805 			    if_getdrvflags(ifp) & IFF_DRV_RUNNING)
2806 				rc = update_mac_settings(ifp, XGMAC_MTU);
2807 		}
2808 		end_synchronized_op(sc, 0);
2809 		break;
2810 
2811 	case SIOCSIFFLAGS:
2812 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4flg");
2813 		if (rc)
2814 			return (rc);
2815 
2816 		if (hw_off_limits(sc)) {
2817 			rc = ENXIO;
2818 			goto fail;
2819 		}
2820 
2821 		if (if_getflags(ifp) & IFF_UP) {
2822 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
2823 				flags = vi->if_flags;
2824 				if ((if_getflags(ifp) ^ flags) &
2825 				    (IFF_PROMISC | IFF_ALLMULTI)) {
2826 					rc = update_mac_settings(ifp,
2827 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
2828 				}
2829 			} else {
2830 				rc = cxgbe_init_synchronized(vi);
2831 			}
2832 			vi->if_flags = if_getflags(ifp);
2833 		} else if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
2834 			rc = cxgbe_uninit_synchronized(vi);
2835 		}
2836 		end_synchronized_op(sc, 0);
2837 		break;
2838 
2839 	case SIOCADDMULTI:
2840 	case SIOCDELMULTI:
2841 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4multi");
2842 		if (rc)
2843 			return (rc);
2844 		if (!hw_off_limits(sc) && if_getdrvflags(ifp) & IFF_DRV_RUNNING)
2845 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
2846 		end_synchronized_op(sc, 0);
2847 		break;
2848 
2849 	case SIOCSIFCAP:
2850 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
2851 		if (rc)
2852 			return (rc);
2853 
2854 		mask = ifr->ifr_reqcap ^ if_getcapenable(ifp);
2855 		if (mask & IFCAP_TXCSUM) {
2856 			if_togglecapenable(ifp, IFCAP_TXCSUM);
2857 			if_togglehwassist(ifp, CSUM_TCP | CSUM_UDP | CSUM_IP);
2858 
2859 			if (IFCAP_TSO4 & if_getcapenable(ifp) &&
2860 			    !(IFCAP_TXCSUM & if_getcapenable(ifp))) {
2861 				mask &= ~IFCAP_TSO4;
2862 				if_setcapenablebit(ifp, 0, IFCAP_TSO4);
2863 				if_printf(ifp,
2864 				    "tso4 disabled due to -txcsum.\n");
2865 			}
2866 		}
2867 		if (mask & IFCAP_TXCSUM_IPV6) {
2868 			if_togglecapenable(ifp, IFCAP_TXCSUM_IPV6);
2869 			if_togglehwassist(ifp, CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
2870 
2871 			if (IFCAP_TSO6 & if_getcapenable(ifp) &&
2872 			    !(IFCAP_TXCSUM_IPV6 & if_getcapenable(ifp))) {
2873 				mask &= ~IFCAP_TSO6;
2874 				if_setcapenablebit(ifp, 0, IFCAP_TSO6);
2875 				if_printf(ifp,
2876 				    "tso6 disabled due to -txcsum6.\n");
2877 			}
2878 		}
2879 		if (mask & IFCAP_RXCSUM)
2880 			if_togglecapenable(ifp, IFCAP_RXCSUM);
2881 		if (mask & IFCAP_RXCSUM_IPV6)
2882 			if_togglecapenable(ifp, IFCAP_RXCSUM_IPV6);
2883 
2884 		/*
2885 		 * Note that we leave CSUM_TSO alone (it is always set).  The
2886 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
2887 		 * sending a TSO request our way, so it's sufficient to toggle
2888 		 * IFCAP_TSOx only.
2889 		 */
2890 		if (mask & IFCAP_TSO4) {
2891 			if (!(IFCAP_TSO4 & if_getcapenable(ifp)) &&
2892 			    !(IFCAP_TXCSUM & if_getcapenable(ifp))) {
2893 				if_printf(ifp, "enable txcsum first.\n");
2894 				rc = EAGAIN;
2895 				goto fail;
2896 			}
2897 			if_togglecapenable(ifp, IFCAP_TSO4);
2898 		}
2899 		if (mask & IFCAP_TSO6) {
2900 			if (!(IFCAP_TSO6 & if_getcapenable(ifp)) &&
2901 			    !(IFCAP_TXCSUM_IPV6 & if_getcapenable(ifp))) {
2902 				if_printf(ifp, "enable txcsum6 first.\n");
2903 				rc = EAGAIN;
2904 				goto fail;
2905 			}
2906 			if_togglecapenable(ifp, IFCAP_TSO6);
2907 		}
2908 		if (mask & IFCAP_LRO) {
2909 #if defined(INET) || defined(INET6)
2910 			int i;
2911 			struct sge_rxq *rxq;
2912 
2913 			if_togglecapenable(ifp, IFCAP_LRO);
2914 			for_each_rxq(vi, i, rxq) {
2915 				if (if_getcapenable(ifp) & IFCAP_LRO)
2916 					rxq->iq.flags |= IQ_LRO_ENABLED;
2917 				else
2918 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
2919 			}
2920 #endif
2921 		}
2922 #ifdef TCP_OFFLOAD
2923 		if (mask & IFCAP_TOE) {
2924 			int enable = (if_getcapenable(ifp) ^ mask) & IFCAP_TOE;
2925 
2926 			rc = toe_capability(vi, enable);
2927 			if (rc != 0)
2928 				goto fail;
2929 
2930 			if_togglecapenable(ifp, mask);
2931 		}
2932 #endif
2933 		if (mask & IFCAP_VLAN_HWTAGGING) {
2934 			if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
2935 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
2936 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
2937 		}
2938 		if (mask & IFCAP_VLAN_MTU) {
2939 			if_togglecapenable(ifp, IFCAP_VLAN_MTU);
2940 
2941 			/* Need to find out how to disable auto-mtu-inflation */
2942 		}
2943 		if (mask & IFCAP_VLAN_HWTSO)
2944 			if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
2945 		if (mask & IFCAP_VLAN_HWCSUM)
2946 			if_togglecapenable(ifp, IFCAP_VLAN_HWCSUM);
2947 #ifdef RATELIMIT
2948 		if (mask & IFCAP_TXRTLMT)
2949 			if_togglecapenable(ifp, IFCAP_TXRTLMT);
2950 #endif
2951 		if (mask & IFCAP_HWRXTSTMP) {
2952 			int i;
2953 			struct sge_rxq *rxq;
2954 
2955 			if_togglecapenable(ifp, IFCAP_HWRXTSTMP);
2956 			for_each_rxq(vi, i, rxq) {
2957 				if (if_getcapenable(ifp) & IFCAP_HWRXTSTMP)
2958 					rxq->iq.flags |= IQ_RX_TIMESTAMP;
2959 				else
2960 					rxq->iq.flags &= ~IQ_RX_TIMESTAMP;
2961 			}
2962 		}
2963 		if (mask & IFCAP_MEXTPG)
2964 			if_togglecapenable(ifp, IFCAP_MEXTPG);
2965 
2966 #ifdef KERN_TLS
2967 		if (mask & IFCAP_TXTLS) {
2968 			int enable = (if_getcapenable(ifp) ^ mask) & IFCAP_TXTLS;
2969 
2970 			rc = ktls_capability(sc, enable);
2971 			if (rc != 0)
2972 				goto fail;
2973 
2974 			if_togglecapenable(ifp, mask & IFCAP_TXTLS);
2975 		}
2976 #endif
2977 		if (mask & IFCAP_VXLAN_HWCSUM) {
2978 			if_togglecapenable(ifp, IFCAP_VXLAN_HWCSUM);
2979 			if_togglehwassist(ifp, CSUM_INNER_IP6_UDP |
2980 			    CSUM_INNER_IP6_TCP | CSUM_INNER_IP |
2981 			    CSUM_INNER_IP_UDP | CSUM_INNER_IP_TCP);
2982 		}
2983 		if (mask & IFCAP_VXLAN_HWTSO) {
2984 			if_togglecapenable(ifp, IFCAP_VXLAN_HWTSO);
2985 			if_togglehwassist(ifp, CSUM_INNER_IP6_TSO |
2986 			    CSUM_INNER_IP_TSO);
2987 		}
2988 
2989 #ifdef VLAN_CAPABILITIES
2990 		VLAN_CAPABILITIES(ifp);
2991 #endif
2992 fail:
2993 		end_synchronized_op(sc, 0);
2994 		break;
2995 
2996 	case SIOCSIFMEDIA:
2997 	case SIOCGIFMEDIA:
2998 	case SIOCGIFXMEDIA:
2999 		rc = ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
3000 		break;
3001 
3002 	case SIOCGI2C: {
3003 		struct ifi2creq i2c;
3004 
3005 		rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
3006 		if (rc != 0)
3007 			break;
3008 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
3009 			rc = EPERM;
3010 			break;
3011 		}
3012 		if (i2c.len > sizeof(i2c.data)) {
3013 			rc = EINVAL;
3014 			break;
3015 		}
3016 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
3017 		if (rc)
3018 			return (rc);
3019 		if (hw_off_limits(sc))
3020 			rc = ENXIO;
3021 		else
3022 			rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
3023 			    i2c.offset, i2c.len, &i2c.data[0]);
3024 		end_synchronized_op(sc, 0);
3025 		if (rc == 0)
3026 			rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
3027 		break;
3028 	}
3029 
3030 	default:
3031 		rc = ether_ioctl(ifp, cmd, data);
3032 	}
3033 
3034 	return (rc);
3035 }
3036 
3037 static int
3038 cxgbe_transmit(if_t ifp, struct mbuf *m)
3039 {
3040 	struct vi_info *vi = if_getsoftc(ifp);
3041 	struct port_info *pi = vi->pi;
3042 	struct adapter *sc;
3043 	struct sge_txq *txq;
3044 	void *items[1];
3045 	int rc;
3046 
3047 	M_ASSERTPKTHDR(m);
3048 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
3049 #if defined(KERN_TLS) || defined(RATELIMIT)
3050 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
3051 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
3052 #endif
3053 
3054 	if (__predict_false(pi->link_cfg.link_ok == false)) {
3055 		m_freem(m);
3056 		return (ENETDOWN);
3057 	}
3058 
3059 	rc = parse_pkt(&m, vi->flags & TX_USES_VM_WR);
3060 	if (__predict_false(rc != 0)) {
3061 		if (__predict_true(rc == EINPROGRESS)) {
3062 			/* queued by parse_pkt */
3063 			MPASS(m != NULL);
3064 			return (0);
3065 		}
3066 
3067 		MPASS(m == NULL);			/* was freed already */
3068 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
3069 		return (rc);
3070 	}
3071 
3072 	/* Select a txq. */
3073 	sc = vi->adapter;
3074 	txq = &sc->sge.txq[vi->first_txq];
3075 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
3076 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
3077 		    vi->rsrv_noflowq);
3078 
3079 	items[0] = m;
3080 	rc = mp_ring_enqueue(txq->r, items, 1, 256);
3081 	if (__predict_false(rc != 0))
3082 		m_freem(m);
3083 
3084 	return (rc);
3085 }
3086 
3087 static void
3088 cxgbe_qflush(if_t ifp)
3089 {
3090 	struct vi_info *vi = if_getsoftc(ifp);
3091 	struct sge_txq *txq;
3092 	int i;
3093 
3094 	/* queues do not exist if !VI_INIT_DONE. */
3095 	if (vi->flags & VI_INIT_DONE) {
3096 		for_each_txq(vi, i, txq) {
3097 			TXQ_LOCK(txq);
3098 			txq->eq.flags |= EQ_QFLUSH;
3099 			TXQ_UNLOCK(txq);
3100 			while (!mp_ring_is_idle(txq->r)) {
3101 				mp_ring_check_drainage(txq->r, 4096);
3102 				pause("qflush", 1);
3103 			}
3104 			TXQ_LOCK(txq);
3105 			txq->eq.flags &= ~EQ_QFLUSH;
3106 			TXQ_UNLOCK(txq);
3107 		}
3108 	}
3109 	if_qflush(ifp);
3110 }
3111 
3112 static uint64_t
3113 vi_get_counter(if_t ifp, ift_counter c)
3114 {
3115 	struct vi_info *vi = if_getsoftc(ifp);
3116 	struct fw_vi_stats_vf *s = &vi->stats;
3117 
3118 	mtx_lock(&vi->tick_mtx);
3119 	vi_refresh_stats(vi);
3120 	mtx_unlock(&vi->tick_mtx);
3121 
3122 	switch (c) {
3123 	case IFCOUNTER_IPACKETS:
3124 		return (s->rx_bcast_frames + s->rx_mcast_frames +
3125 		    s->rx_ucast_frames);
3126 	case IFCOUNTER_IERRORS:
3127 		return (s->rx_err_frames);
3128 	case IFCOUNTER_OPACKETS:
3129 		return (s->tx_bcast_frames + s->tx_mcast_frames +
3130 		    s->tx_ucast_frames + s->tx_offload_frames);
3131 	case IFCOUNTER_OERRORS:
3132 		return (s->tx_drop_frames);
3133 	case IFCOUNTER_IBYTES:
3134 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
3135 		    s->rx_ucast_bytes);
3136 	case IFCOUNTER_OBYTES:
3137 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
3138 		    s->tx_ucast_bytes + s->tx_offload_bytes);
3139 	case IFCOUNTER_IMCASTS:
3140 		return (s->rx_mcast_frames);
3141 	case IFCOUNTER_OMCASTS:
3142 		return (s->tx_mcast_frames);
3143 	case IFCOUNTER_OQDROPS: {
3144 		uint64_t drops;
3145 
3146 		drops = 0;
3147 		if (vi->flags & VI_INIT_DONE) {
3148 			int i;
3149 			struct sge_txq *txq;
3150 
3151 			for_each_txq(vi, i, txq)
3152 				drops += counter_u64_fetch(txq->r->dropped);
3153 		}
3154 
3155 		return (drops);
3156 
3157 	}
3158 
3159 	default:
3160 		return (if_get_counter_default(ifp, c));
3161 	}
3162 }
3163 
3164 static uint64_t
3165 cxgbe_get_counter(if_t ifp, ift_counter c)
3166 {
3167 	struct vi_info *vi = if_getsoftc(ifp);
3168 	struct port_info *pi = vi->pi;
3169 	struct port_stats *s = &pi->stats;
3170 
3171 	mtx_lock(&vi->tick_mtx);
3172 	cxgbe_refresh_stats(vi);
3173 	mtx_unlock(&vi->tick_mtx);
3174 
3175 	switch (c) {
3176 	case IFCOUNTER_IPACKETS:
3177 		return (s->rx_frames);
3178 
3179 	case IFCOUNTER_IERRORS:
3180 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
3181 		    s->rx_fcs_err + s->rx_len_err);
3182 
3183 	case IFCOUNTER_OPACKETS:
3184 		return (s->tx_frames);
3185 
3186 	case IFCOUNTER_OERRORS:
3187 		return (s->tx_error_frames);
3188 
3189 	case IFCOUNTER_IBYTES:
3190 		return (s->rx_octets);
3191 
3192 	case IFCOUNTER_OBYTES:
3193 		return (s->tx_octets);
3194 
3195 	case IFCOUNTER_IMCASTS:
3196 		return (s->rx_mcast_frames);
3197 
3198 	case IFCOUNTER_OMCASTS:
3199 		return (s->tx_mcast_frames);
3200 
3201 	case IFCOUNTER_IQDROPS:
3202 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
3203 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
3204 		    s->rx_trunc3 + pi->tnl_cong_drops);
3205 
3206 	case IFCOUNTER_OQDROPS: {
3207 		uint64_t drops;
3208 
3209 		drops = s->tx_drop;
3210 		if (vi->flags & VI_INIT_DONE) {
3211 			int i;
3212 			struct sge_txq *txq;
3213 
3214 			for_each_txq(vi, i, txq)
3215 				drops += counter_u64_fetch(txq->r->dropped);
3216 		}
3217 
3218 		return (drops);
3219 
3220 	}
3221 
3222 	default:
3223 		return (if_get_counter_default(ifp, c));
3224 	}
3225 }
3226 
3227 #if defined(KERN_TLS) || defined(RATELIMIT)
3228 static int
3229 cxgbe_snd_tag_alloc(if_t ifp, union if_snd_tag_alloc_params *params,
3230     struct m_snd_tag **pt)
3231 {
3232 	int error;
3233 
3234 	switch (params->hdr.type) {
3235 #ifdef RATELIMIT
3236 	case IF_SND_TAG_TYPE_RATE_LIMIT:
3237 		error = cxgbe_rate_tag_alloc(ifp, params, pt);
3238 		break;
3239 #endif
3240 #ifdef KERN_TLS
3241 	case IF_SND_TAG_TYPE_TLS:
3242 	{
3243 		struct vi_info *vi = if_getsoftc(ifp);
3244 
3245 		if (is_t6(vi->pi->adapter))
3246 			error = t6_tls_tag_alloc(ifp, params, pt);
3247 		else
3248 			error = EOPNOTSUPP;
3249 		break;
3250 	}
3251 #endif
3252 	default:
3253 		error = EOPNOTSUPP;
3254 	}
3255 	return (error);
3256 }
3257 #endif
3258 
3259 /*
3260  * The kernel picks a media from the list we had provided but we still validate
3261  * the requeste.
3262  */
3263 int
3264 cxgbe_media_change(if_t ifp)
3265 {
3266 	struct vi_info *vi = if_getsoftc(ifp);
3267 	struct port_info *pi = vi->pi;
3268 	struct ifmedia *ifm = &pi->media;
3269 	struct link_config *lc = &pi->link_cfg;
3270 	struct adapter *sc = pi->adapter;
3271 	int rc;
3272 
3273 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mec");
3274 	if (rc != 0)
3275 		return (rc);
3276 	PORT_LOCK(pi);
3277 	if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
3278 		/* ifconfig .. media autoselect */
3279 		if (!(lc->pcaps & FW_PORT_CAP32_ANEG)) {
3280 			rc = ENOTSUP; /* AN not supported by transceiver */
3281 			goto done;
3282 		}
3283 		lc->requested_aneg = AUTONEG_ENABLE;
3284 		lc->requested_speed = 0;
3285 		lc->requested_fc |= PAUSE_AUTONEG;
3286 	} else {
3287 		lc->requested_aneg = AUTONEG_DISABLE;
3288 		lc->requested_speed =
3289 		    ifmedia_baudrate(ifm->ifm_media) / 1000000;
3290 		lc->requested_fc = 0;
3291 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
3292 			lc->requested_fc |= PAUSE_RX;
3293 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
3294 			lc->requested_fc |= PAUSE_TX;
3295 	}
3296 	if (pi->up_vis > 0 && !hw_off_limits(sc)) {
3297 		fixup_link_config(pi);
3298 		rc = apply_link_config(pi);
3299 	}
3300 done:
3301 	PORT_UNLOCK(pi);
3302 	end_synchronized_op(sc, 0);
3303 	return (rc);
3304 }
3305 
3306 /*
3307  * Base media word (without ETHER, pause, link active, etc.) for the port at the
3308  * given speed.
3309  */
3310 static int
3311 port_mword(struct port_info *pi, uint32_t speed)
3312 {
3313 
3314 	MPASS(speed & M_FW_PORT_CAP32_SPEED);
3315 	MPASS(powerof2(speed));
3316 
3317 	switch(pi->port_type) {
3318 	case FW_PORT_TYPE_BT_SGMII:
3319 	case FW_PORT_TYPE_BT_XFI:
3320 	case FW_PORT_TYPE_BT_XAUI:
3321 		/* BaseT */
3322 		switch (speed) {
3323 		case FW_PORT_CAP32_SPEED_100M:
3324 			return (IFM_100_T);
3325 		case FW_PORT_CAP32_SPEED_1G:
3326 			return (IFM_1000_T);
3327 		case FW_PORT_CAP32_SPEED_10G:
3328 			return (IFM_10G_T);
3329 		}
3330 		break;
3331 	case FW_PORT_TYPE_KX4:
3332 		if (speed == FW_PORT_CAP32_SPEED_10G)
3333 			return (IFM_10G_KX4);
3334 		break;
3335 	case FW_PORT_TYPE_CX4:
3336 		if (speed == FW_PORT_CAP32_SPEED_10G)
3337 			return (IFM_10G_CX4);
3338 		break;
3339 	case FW_PORT_TYPE_KX:
3340 		if (speed == FW_PORT_CAP32_SPEED_1G)
3341 			return (IFM_1000_KX);
3342 		break;
3343 	case FW_PORT_TYPE_KR:
3344 	case FW_PORT_TYPE_BP_AP:
3345 	case FW_PORT_TYPE_BP4_AP:
3346 	case FW_PORT_TYPE_BP40_BA:
3347 	case FW_PORT_TYPE_KR4_100G:
3348 	case FW_PORT_TYPE_KR_SFP28:
3349 	case FW_PORT_TYPE_KR_XLAUI:
3350 		switch (speed) {
3351 		case FW_PORT_CAP32_SPEED_1G:
3352 			return (IFM_1000_KX);
3353 		case FW_PORT_CAP32_SPEED_10G:
3354 			return (IFM_10G_KR);
3355 		case FW_PORT_CAP32_SPEED_25G:
3356 			return (IFM_25G_KR);
3357 		case FW_PORT_CAP32_SPEED_40G:
3358 			return (IFM_40G_KR4);
3359 		case FW_PORT_CAP32_SPEED_50G:
3360 			return (IFM_50G_KR2);
3361 		case FW_PORT_CAP32_SPEED_100G:
3362 			return (IFM_100G_KR4);
3363 		}
3364 		break;
3365 	case FW_PORT_TYPE_FIBER_XFI:
3366 	case FW_PORT_TYPE_FIBER_XAUI:
3367 	case FW_PORT_TYPE_SFP:
3368 	case FW_PORT_TYPE_QSFP_10G:
3369 	case FW_PORT_TYPE_QSA:
3370 	case FW_PORT_TYPE_QSFP:
3371 	case FW_PORT_TYPE_CR4_QSFP:
3372 	case FW_PORT_TYPE_CR_QSFP:
3373 	case FW_PORT_TYPE_CR2_QSFP:
3374 	case FW_PORT_TYPE_SFP28:
3375 		/* Pluggable transceiver */
3376 		switch (pi->mod_type) {
3377 		case FW_PORT_MOD_TYPE_LR:
3378 			switch (speed) {
3379 			case FW_PORT_CAP32_SPEED_1G:
3380 				return (IFM_1000_LX);
3381 			case FW_PORT_CAP32_SPEED_10G:
3382 				return (IFM_10G_LR);
3383 			case FW_PORT_CAP32_SPEED_25G:
3384 				return (IFM_25G_LR);
3385 			case FW_PORT_CAP32_SPEED_40G:
3386 				return (IFM_40G_LR4);
3387 			case FW_PORT_CAP32_SPEED_50G:
3388 				return (IFM_50G_LR2);
3389 			case FW_PORT_CAP32_SPEED_100G:
3390 				return (IFM_100G_LR4);
3391 			}
3392 			break;
3393 		case FW_PORT_MOD_TYPE_SR:
3394 			switch (speed) {
3395 			case FW_PORT_CAP32_SPEED_1G:
3396 				return (IFM_1000_SX);
3397 			case FW_PORT_CAP32_SPEED_10G:
3398 				return (IFM_10G_SR);
3399 			case FW_PORT_CAP32_SPEED_25G:
3400 				return (IFM_25G_SR);
3401 			case FW_PORT_CAP32_SPEED_40G:
3402 				return (IFM_40G_SR4);
3403 			case FW_PORT_CAP32_SPEED_50G:
3404 				return (IFM_50G_SR2);
3405 			case FW_PORT_CAP32_SPEED_100G:
3406 				return (IFM_100G_SR4);
3407 			}
3408 			break;
3409 		case FW_PORT_MOD_TYPE_ER:
3410 			if (speed == FW_PORT_CAP32_SPEED_10G)
3411 				return (IFM_10G_ER);
3412 			break;
3413 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3414 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3415 			switch (speed) {
3416 			case FW_PORT_CAP32_SPEED_1G:
3417 				return (IFM_1000_CX);
3418 			case FW_PORT_CAP32_SPEED_10G:
3419 				return (IFM_10G_TWINAX);
3420 			case FW_PORT_CAP32_SPEED_25G:
3421 				return (IFM_25G_CR);
3422 			case FW_PORT_CAP32_SPEED_40G:
3423 				return (IFM_40G_CR4);
3424 			case FW_PORT_CAP32_SPEED_50G:
3425 				return (IFM_50G_CR2);
3426 			case FW_PORT_CAP32_SPEED_100G:
3427 				return (IFM_100G_CR4);
3428 			}
3429 			break;
3430 		case FW_PORT_MOD_TYPE_LRM:
3431 			if (speed == FW_PORT_CAP32_SPEED_10G)
3432 				return (IFM_10G_LRM);
3433 			break;
3434 		case FW_PORT_MOD_TYPE_NA:
3435 			MPASS(0);	/* Not pluggable? */
3436 			/* fall throough */
3437 		case FW_PORT_MOD_TYPE_ERROR:
3438 		case FW_PORT_MOD_TYPE_UNKNOWN:
3439 		case FW_PORT_MOD_TYPE_NOTSUPPORTED:
3440 			break;
3441 		case FW_PORT_MOD_TYPE_NONE:
3442 			return (IFM_NONE);
3443 		}
3444 		break;
3445 	case FW_PORT_TYPE_NONE:
3446 		return (IFM_NONE);
3447 	}
3448 
3449 	return (IFM_UNKNOWN);
3450 }
3451 
3452 void
3453 cxgbe_media_status(if_t ifp, struct ifmediareq *ifmr)
3454 {
3455 	struct vi_info *vi = if_getsoftc(ifp);
3456 	struct port_info *pi = vi->pi;
3457 	struct adapter *sc = pi->adapter;
3458 	struct link_config *lc = &pi->link_cfg;
3459 
3460 	if (begin_synchronized_op(sc, vi , SLEEP_OK | INTR_OK, "t4med") != 0)
3461 		return;
3462 	PORT_LOCK(pi);
3463 
3464 	if (pi->up_vis == 0 && !hw_off_limits(sc)) {
3465 		/*
3466 		 * If all the interfaces are administratively down the firmware
3467 		 * does not report transceiver changes.  Refresh port info here
3468 		 * so that ifconfig displays accurate ifmedia at all times.
3469 		 * This is the only reason we have a synchronized op in this
3470 		 * function.  Just PORT_LOCK would have been enough otherwise.
3471 		 */
3472 		t4_update_port_info(pi);
3473 		build_medialist(pi);
3474 	}
3475 
3476 	/* ifm_status */
3477 	ifmr->ifm_status = IFM_AVALID;
3478 	if (lc->link_ok == false)
3479 		goto done;
3480 	ifmr->ifm_status |= IFM_ACTIVE;
3481 
3482 	/* ifm_active */
3483 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
3484 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
3485 	if (lc->fc & PAUSE_RX)
3486 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
3487 	if (lc->fc & PAUSE_TX)
3488 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
3489 	ifmr->ifm_active |= port_mword(pi, speed_to_fwcap(lc->speed));
3490 done:
3491 	PORT_UNLOCK(pi);
3492 	end_synchronized_op(sc, 0);
3493 }
3494 
3495 static int
3496 vcxgbe_probe(device_t dev)
3497 {
3498 	char buf[128];
3499 	struct vi_info *vi = device_get_softc(dev);
3500 
3501 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
3502 	    vi - vi->pi->vi);
3503 	device_set_desc_copy(dev, buf);
3504 
3505 	return (BUS_PROBE_DEFAULT);
3506 }
3507 
3508 static int
3509 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
3510 {
3511 	int func, index, rc;
3512 	uint32_t param, val;
3513 
3514 	ASSERT_SYNCHRONIZED_OP(sc);
3515 
3516 	index = vi - pi->vi;
3517 	MPASS(index > 0);	/* This function deals with _extra_ VIs only */
3518 	KASSERT(index < nitems(vi_mac_funcs),
3519 	    ("%s: VI %s doesn't have a MAC func", __func__,
3520 	    device_get_nameunit(vi->dev)));
3521 	func = vi_mac_funcs[index];
3522 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
3523 	    vi->hw_addr, &vi->rss_size, &vi->vfvld, &vi->vin, func, 0);
3524 	if (rc < 0) {
3525 		CH_ERR(vi, "failed to allocate virtual interface %d"
3526 		    "for port %d: %d\n", index, pi->port_id, -rc);
3527 		return (-rc);
3528 	}
3529 	vi->viid = rc;
3530 
3531 	if (vi->rss_size == 1) {
3532 		/*
3533 		 * This VI didn't get a slice of the RSS table.  Reduce the
3534 		 * number of VIs being created (hw.cxgbe.num_vis) or modify the
3535 		 * configuration file (nvi, rssnvi for this PF) if this is a
3536 		 * problem.
3537 		 */
3538 		device_printf(vi->dev, "RSS table not available.\n");
3539 		vi->rss_base = 0xffff;
3540 
3541 		return (0);
3542 	}
3543 
3544 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
3545 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
3546 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
3547 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3548 	if (rc)
3549 		vi->rss_base = 0xffff;
3550 	else {
3551 		MPASS((val >> 16) == vi->rss_size);
3552 		vi->rss_base = val & 0xffff;
3553 	}
3554 
3555 	return (0);
3556 }
3557 
3558 static int
3559 vcxgbe_attach(device_t dev)
3560 {
3561 	struct vi_info *vi;
3562 	struct port_info *pi;
3563 	struct adapter *sc;
3564 	int rc;
3565 
3566 	vi = device_get_softc(dev);
3567 	pi = vi->pi;
3568 	sc = pi->adapter;
3569 
3570 	rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
3571 	if (rc)
3572 		return (rc);
3573 	rc = alloc_extra_vi(sc, pi, vi);
3574 	end_synchronized_op(sc, 0);
3575 	if (rc)
3576 		return (rc);
3577 
3578 	rc = cxgbe_vi_attach(dev, vi);
3579 	if (rc) {
3580 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
3581 		return (rc);
3582 	}
3583 	return (0);
3584 }
3585 
3586 static int
3587 vcxgbe_detach(device_t dev)
3588 {
3589 	struct vi_info *vi;
3590 	struct adapter *sc;
3591 
3592 	vi = device_get_softc(dev);
3593 	sc = vi->adapter;
3594 
3595 	begin_vi_detach(sc, vi);
3596 	cxgbe_vi_detach(vi);
3597 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
3598 	end_vi_detach(sc, vi);
3599 
3600 	return (0);
3601 }
3602 
3603 static struct callout fatal_callout;
3604 static struct taskqueue *reset_tq;
3605 
3606 static void
3607 delayed_panic(void *arg)
3608 {
3609 	struct adapter *sc = arg;
3610 
3611 	panic("%s: panic on fatal error", device_get_nameunit(sc->dev));
3612 }
3613 
3614 static void
3615 fatal_error_task(void *arg, int pending)
3616 {
3617 	struct adapter *sc = arg;
3618 	int rc;
3619 
3620 #ifdef TCP_OFFLOAD
3621 	t4_async_event(sc);
3622 #endif
3623 	if (atomic_testandclear_int(&sc->error_flags, ilog2(ADAP_CIM_ERR))) {
3624 		dump_cim_regs(sc);
3625 		dump_cimla(sc);
3626 		dump_devlog(sc);
3627 	}
3628 
3629 	if (t4_reset_on_fatal_err) {
3630 		CH_ALERT(sc, "resetting on fatal error.\n");
3631 		rc = reset_adapter(sc);
3632 		if (rc == 0 && t4_panic_on_fatal_err) {
3633 			CH_ALERT(sc, "reset was successful, "
3634 			    "system will NOT panic.\n");
3635 			return;
3636 		}
3637 	}
3638 
3639 	if (t4_panic_on_fatal_err) {
3640 		CH_ALERT(sc, "panicking on fatal error (after 30s).\n");
3641 		callout_reset(&fatal_callout, hz * 30, delayed_panic, sc);
3642 	}
3643 }
3644 
3645 void
3646 t4_fatal_err(struct adapter *sc, bool fw_error)
3647 {
3648 	const bool verbose = (sc->debug_flags & DF_VERBOSE_SLOWINTR) != 0;
3649 
3650 	stop_adapter(sc);
3651 	if (atomic_testandset_int(&sc->error_flags, ilog2(ADAP_FATAL_ERR)))
3652 		return;
3653 	if (fw_error) {
3654 		/*
3655 		 * We are here because of a firmware error/timeout and not
3656 		 * because of a hardware interrupt.  It is possible (although
3657 		 * not very likely) that an error interrupt was also raised but
3658 		 * this thread ran first and inhibited t4_intr_err.  We walk the
3659 		 * main INT_CAUSE registers here to make sure we haven't missed
3660 		 * anything interesting.
3661 		 */
3662 		t4_slow_intr_handler(sc, verbose);
3663 		atomic_set_int(&sc->error_flags, ADAP_CIM_ERR);
3664 	}
3665 	t4_report_fw_error(sc);
3666 	log(LOG_ALERT, "%s: encountered fatal error, adapter stopped (%d).\n",
3667 	    device_get_nameunit(sc->dev), fw_error);
3668 	taskqueue_enqueue(reset_tq, &sc->fatal_error_task);
3669 }
3670 
3671 void
3672 t4_add_adapter(struct adapter *sc)
3673 {
3674 	sx_xlock(&t4_list_lock);
3675 	SLIST_INSERT_HEAD(&t4_list, sc, link);
3676 	sx_xunlock(&t4_list_lock);
3677 }
3678 
3679 int
3680 t4_map_bars_0_and_4(struct adapter *sc)
3681 {
3682 	sc->regs_rid = PCIR_BAR(0);
3683 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
3684 	    &sc->regs_rid, RF_ACTIVE);
3685 	if (sc->regs_res == NULL) {
3686 		device_printf(sc->dev, "cannot map registers.\n");
3687 		return (ENXIO);
3688 	}
3689 	sc->bt = rman_get_bustag(sc->regs_res);
3690 	sc->bh = rman_get_bushandle(sc->regs_res);
3691 	sc->mmio_len = rman_get_size(sc->regs_res);
3692 	setbit(&sc->doorbells, DOORBELL_KDB);
3693 
3694 	sc->msix_rid = PCIR_BAR(4);
3695 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
3696 	    &sc->msix_rid, RF_ACTIVE);
3697 	if (sc->msix_res == NULL) {
3698 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
3699 		return (ENXIO);
3700 	}
3701 
3702 	return (0);
3703 }
3704 
3705 int
3706 t4_map_bar_2(struct adapter *sc)
3707 {
3708 
3709 	/*
3710 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
3711 	 * to map it if RDMA is disabled.
3712 	 */
3713 	if (is_t4(sc) && sc->rdmacaps == 0)
3714 		return (0);
3715 
3716 	sc->udbs_rid = PCIR_BAR(2);
3717 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
3718 	    &sc->udbs_rid, RF_ACTIVE);
3719 	if (sc->udbs_res == NULL) {
3720 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
3721 		return (ENXIO);
3722 	}
3723 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
3724 
3725 	if (chip_id(sc) >= CHELSIO_T5) {
3726 		setbit(&sc->doorbells, DOORBELL_UDB);
3727 #if defined(__i386__) || defined(__amd64__)
3728 		if (t5_write_combine) {
3729 			int rc, mode;
3730 
3731 			/*
3732 			 * Enable write combining on BAR2.  This is the
3733 			 * userspace doorbell BAR and is split into 128B
3734 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
3735 			 * with an egress queue.  The first 64B has the doorbell
3736 			 * and the second 64B can be used to submit a tx work
3737 			 * request with an implicit doorbell.
3738 			 */
3739 
3740 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
3741 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
3742 			if (rc == 0) {
3743 				clrbit(&sc->doorbells, DOORBELL_UDB);
3744 				setbit(&sc->doorbells, DOORBELL_WCWR);
3745 				setbit(&sc->doorbells, DOORBELL_UDBWC);
3746 			} else {
3747 				device_printf(sc->dev,
3748 				    "couldn't enable write combining: %d\n",
3749 				    rc);
3750 			}
3751 
3752 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
3753 			t4_write_reg(sc, A_SGE_STAT_CFG,
3754 			    V_STATSOURCE_T5(7) | mode);
3755 		}
3756 #endif
3757 	}
3758 	sc->iwt.wc_en = isset(&sc->doorbells, DOORBELL_UDBWC) ? 1 : 0;
3759 
3760 	return (0);
3761 }
3762 
3763 struct memwin_init {
3764 	uint32_t base;
3765 	uint32_t aperture;
3766 };
3767 
3768 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
3769 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
3770 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
3771 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
3772 };
3773 
3774 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
3775 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
3776 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
3777 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
3778 };
3779 
3780 static void
3781 setup_memwin(struct adapter *sc)
3782 {
3783 	const struct memwin_init *mw_init;
3784 	struct memwin *mw;
3785 	int i;
3786 	uint32_t bar0;
3787 
3788 	if (is_t4(sc)) {
3789 		/*
3790 		 * Read low 32b of bar0 indirectly via the hardware backdoor
3791 		 * mechanism.  Works from within PCI passthrough environments
3792 		 * too, where rman_get_start() can return a different value.  We
3793 		 * need to program the T4 memory window decoders with the actual
3794 		 * addresses that will be coming across the PCIe link.
3795 		 */
3796 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
3797 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
3798 
3799 		mw_init = &t4_memwin[0];
3800 	} else {
3801 		/* T5+ use the relative offset inside the PCIe BAR */
3802 		bar0 = 0;
3803 
3804 		mw_init = &t5_memwin[0];
3805 	}
3806 
3807 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
3808 		if (!rw_initialized(&mw->mw_lock)) {
3809 			rw_init(&mw->mw_lock, "memory window access");
3810 			mw->mw_base = mw_init->base;
3811 			mw->mw_aperture = mw_init->aperture;
3812 			mw->mw_curpos = 0;
3813 		}
3814 		t4_write_reg(sc,
3815 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
3816 		    (mw->mw_base + bar0) | V_BIR(0) |
3817 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
3818 		rw_wlock(&mw->mw_lock);
3819 		position_memwin(sc, i, mw->mw_curpos);
3820 		rw_wunlock(&mw->mw_lock);
3821 	}
3822 
3823 	/* flush */
3824 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
3825 }
3826 
3827 /*
3828  * Positions the memory window at the given address in the card's address space.
3829  * There are some alignment requirements and the actual position may be at an
3830  * address prior to the requested address.  mw->mw_curpos always has the actual
3831  * position of the window.
3832  */
3833 static void
3834 position_memwin(struct adapter *sc, int idx, uint32_t addr)
3835 {
3836 	struct memwin *mw;
3837 	uint32_t pf;
3838 	uint32_t reg;
3839 
3840 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
3841 	mw = &sc->memwin[idx];
3842 	rw_assert(&mw->mw_lock, RA_WLOCKED);
3843 
3844 	if (is_t4(sc)) {
3845 		pf = 0;
3846 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
3847 	} else {
3848 		pf = V_PFNUM(sc->pf);
3849 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
3850 	}
3851 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
3852 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
3853 	t4_read_reg(sc, reg);	/* flush */
3854 }
3855 
3856 int
3857 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
3858     int len, int rw)
3859 {
3860 	struct memwin *mw;
3861 	uint32_t mw_end, v;
3862 
3863 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
3864 
3865 	/* Memory can only be accessed in naturally aligned 4 byte units */
3866 	if (addr & 3 || len & 3 || len <= 0)
3867 		return (EINVAL);
3868 
3869 	mw = &sc->memwin[idx];
3870 	while (len > 0) {
3871 		rw_rlock(&mw->mw_lock);
3872 		mw_end = mw->mw_curpos + mw->mw_aperture;
3873 		if (addr >= mw_end || addr < mw->mw_curpos) {
3874 			/* Will need to reposition the window */
3875 			if (!rw_try_upgrade(&mw->mw_lock)) {
3876 				rw_runlock(&mw->mw_lock);
3877 				rw_wlock(&mw->mw_lock);
3878 			}
3879 			rw_assert(&mw->mw_lock, RA_WLOCKED);
3880 			position_memwin(sc, idx, addr);
3881 			rw_downgrade(&mw->mw_lock);
3882 			mw_end = mw->mw_curpos + mw->mw_aperture;
3883 		}
3884 		rw_assert(&mw->mw_lock, RA_RLOCKED);
3885 		while (addr < mw_end && len > 0) {
3886 			if (rw == 0) {
3887 				v = t4_read_reg(sc, mw->mw_base + addr -
3888 				    mw->mw_curpos);
3889 				*val++ = le32toh(v);
3890 			} else {
3891 				v = *val++;
3892 				t4_write_reg(sc, mw->mw_base + addr -
3893 				    mw->mw_curpos, htole32(v));
3894 			}
3895 			addr += 4;
3896 			len -= 4;
3897 		}
3898 		rw_runlock(&mw->mw_lock);
3899 	}
3900 
3901 	return (0);
3902 }
3903 
3904 static void
3905 t4_init_atid_table(struct adapter *sc)
3906 {
3907 	struct tid_info *t;
3908 	int i;
3909 
3910 	t = &sc->tids;
3911 	if (t->natids == 0)
3912 		return;
3913 
3914 	MPASS(t->atid_tab == NULL);
3915 
3916 	t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
3917 	    M_ZERO | M_WAITOK);
3918 	mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
3919 	t->afree = t->atid_tab;
3920 	t->atids_in_use = 0;
3921 	for (i = 1; i < t->natids; i++)
3922 		t->atid_tab[i - 1].next = &t->atid_tab[i];
3923 	t->atid_tab[t->natids - 1].next = NULL;
3924 }
3925 
3926 static void
3927 t4_free_atid_table(struct adapter *sc)
3928 {
3929 	struct tid_info *t;
3930 
3931 	t = &sc->tids;
3932 
3933 	KASSERT(t->atids_in_use == 0,
3934 	    ("%s: %d atids still in use.", __func__, t->atids_in_use));
3935 
3936 	if (mtx_initialized(&t->atid_lock))
3937 		mtx_destroy(&t->atid_lock);
3938 	free(t->atid_tab, M_CXGBE);
3939 	t->atid_tab = NULL;
3940 }
3941 
3942 int
3943 alloc_atid(struct adapter *sc, void *ctx)
3944 {
3945 	struct tid_info *t = &sc->tids;
3946 	int atid = -1;
3947 
3948 	mtx_lock(&t->atid_lock);
3949 	if (t->afree) {
3950 		union aopen_entry *p = t->afree;
3951 
3952 		atid = p - t->atid_tab;
3953 		MPASS(atid <= M_TID_TID);
3954 		t->afree = p->next;
3955 		p->data = ctx;
3956 		t->atids_in_use++;
3957 	}
3958 	mtx_unlock(&t->atid_lock);
3959 	return (atid);
3960 }
3961 
3962 void *
3963 lookup_atid(struct adapter *sc, int atid)
3964 {
3965 	struct tid_info *t = &sc->tids;
3966 
3967 	return (t->atid_tab[atid].data);
3968 }
3969 
3970 void
3971 free_atid(struct adapter *sc, int atid)
3972 {
3973 	struct tid_info *t = &sc->tids;
3974 	union aopen_entry *p = &t->atid_tab[atid];
3975 
3976 	mtx_lock(&t->atid_lock);
3977 	p->next = t->afree;
3978 	t->afree = p;
3979 	t->atids_in_use--;
3980 	mtx_unlock(&t->atid_lock);
3981 }
3982 
3983 static void
3984 queue_tid_release(struct adapter *sc, int tid)
3985 {
3986 
3987 	CXGBE_UNIMPLEMENTED("deferred tid release");
3988 }
3989 
3990 void
3991 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
3992 {
3993 	struct wrqe *wr;
3994 	struct cpl_tid_release *req;
3995 
3996 	wr = alloc_wrqe(sizeof(*req), ctrlq);
3997 	if (wr == NULL) {
3998 		queue_tid_release(sc, tid);	/* defer */
3999 		return;
4000 	}
4001 	req = wrtod(wr);
4002 
4003 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
4004 
4005 	t4_wrq_tx(sc, wr);
4006 }
4007 
4008 static int
4009 t4_range_cmp(const void *a, const void *b)
4010 {
4011 	return ((const struct t4_range *)a)->start -
4012 	       ((const struct t4_range *)b)->start;
4013 }
4014 
4015 /*
4016  * Verify that the memory range specified by the addr/len pair is valid within
4017  * the card's address space.
4018  */
4019 static int
4020 validate_mem_range(struct adapter *sc, uint32_t addr, uint32_t len)
4021 {
4022 	struct t4_range mem_ranges[4], *r, *next;
4023 	uint32_t em, addr_len;
4024 	int i, n, remaining;
4025 
4026 	/* Memory can only be accessed in naturally aligned 4 byte units */
4027 	if (addr & 3 || len & 3 || len == 0)
4028 		return (EINVAL);
4029 
4030 	/* Enabled memories */
4031 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
4032 
4033 	r = &mem_ranges[0];
4034 	n = 0;
4035 	bzero(r, sizeof(mem_ranges));
4036 	if (em & F_EDRAM0_ENABLE) {
4037 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
4038 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
4039 		if (r->size > 0) {
4040 			r->start = G_EDRAM0_BASE(addr_len) << 20;
4041 			if (addr >= r->start &&
4042 			    addr + len <= r->start + r->size)
4043 				return (0);
4044 			r++;
4045 			n++;
4046 		}
4047 	}
4048 	if (em & F_EDRAM1_ENABLE) {
4049 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
4050 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
4051 		if (r->size > 0) {
4052 			r->start = G_EDRAM1_BASE(addr_len) << 20;
4053 			if (addr >= r->start &&
4054 			    addr + len <= r->start + r->size)
4055 				return (0);
4056 			r++;
4057 			n++;
4058 		}
4059 	}
4060 	if (em & F_EXT_MEM_ENABLE) {
4061 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
4062 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
4063 		if (r->size > 0) {
4064 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
4065 			if (addr >= r->start &&
4066 			    addr + len <= r->start + r->size)
4067 				return (0);
4068 			r++;
4069 			n++;
4070 		}
4071 	}
4072 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
4073 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
4074 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
4075 		if (r->size > 0) {
4076 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
4077 			if (addr >= r->start &&
4078 			    addr + len <= r->start + r->size)
4079 				return (0);
4080 			r++;
4081 			n++;
4082 		}
4083 	}
4084 	MPASS(n <= nitems(mem_ranges));
4085 
4086 	if (n > 1) {
4087 		/* Sort and merge the ranges. */
4088 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
4089 
4090 		/* Start from index 0 and examine the next n - 1 entries. */
4091 		r = &mem_ranges[0];
4092 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
4093 
4094 			MPASS(r->size > 0);	/* r is a valid entry. */
4095 			next = r + 1;
4096 			MPASS(next->size > 0);	/* and so is the next one. */
4097 
4098 			while (r->start + r->size >= next->start) {
4099 				/* Merge the next one into the current entry. */
4100 				r->size = max(r->start + r->size,
4101 				    next->start + next->size) - r->start;
4102 				n--;	/* One fewer entry in total. */
4103 				if (--remaining == 0)
4104 					goto done;	/* short circuit */
4105 				next++;
4106 			}
4107 			if (next != r + 1) {
4108 				/*
4109 				 * Some entries were merged into r and next
4110 				 * points to the first valid entry that couldn't
4111 				 * be merged.
4112 				 */
4113 				MPASS(next->size > 0);	/* must be valid */
4114 				memcpy(r + 1, next, remaining * sizeof(*r));
4115 #ifdef INVARIANTS
4116 				/*
4117 				 * This so that the foo->size assertion in the
4118 				 * next iteration of the loop do the right
4119 				 * thing for entries that were pulled up and are
4120 				 * no longer valid.
4121 				 */
4122 				MPASS(n < nitems(mem_ranges));
4123 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
4124 				    sizeof(struct t4_range));
4125 #endif
4126 			}
4127 		}
4128 done:
4129 		/* Done merging the ranges. */
4130 		MPASS(n > 0);
4131 		r = &mem_ranges[0];
4132 		for (i = 0; i < n; i++, r++) {
4133 			if (addr >= r->start &&
4134 			    addr + len <= r->start + r->size)
4135 				return (0);
4136 		}
4137 	}
4138 
4139 	return (EFAULT);
4140 }
4141 
4142 static int
4143 fwmtype_to_hwmtype(int mtype)
4144 {
4145 
4146 	switch (mtype) {
4147 	case FW_MEMTYPE_EDC0:
4148 		return (MEM_EDC0);
4149 	case FW_MEMTYPE_EDC1:
4150 		return (MEM_EDC1);
4151 	case FW_MEMTYPE_EXTMEM:
4152 		return (MEM_MC0);
4153 	case FW_MEMTYPE_EXTMEM1:
4154 		return (MEM_MC1);
4155 	default:
4156 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
4157 	}
4158 }
4159 
4160 /*
4161  * Verify that the memory range specified by the memtype/offset/len pair is
4162  * valid and lies entirely within the memtype specified.  The global address of
4163  * the start of the range is returned in addr.
4164  */
4165 static int
4166 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, uint32_t len,
4167     uint32_t *addr)
4168 {
4169 	uint32_t em, addr_len, maddr;
4170 
4171 	/* Memory can only be accessed in naturally aligned 4 byte units */
4172 	if (off & 3 || len & 3 || len == 0)
4173 		return (EINVAL);
4174 
4175 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
4176 	switch (fwmtype_to_hwmtype(mtype)) {
4177 	case MEM_EDC0:
4178 		if (!(em & F_EDRAM0_ENABLE))
4179 			return (EINVAL);
4180 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
4181 		maddr = G_EDRAM0_BASE(addr_len) << 20;
4182 		break;
4183 	case MEM_EDC1:
4184 		if (!(em & F_EDRAM1_ENABLE))
4185 			return (EINVAL);
4186 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
4187 		maddr = G_EDRAM1_BASE(addr_len) << 20;
4188 		break;
4189 	case MEM_MC:
4190 		if (!(em & F_EXT_MEM_ENABLE))
4191 			return (EINVAL);
4192 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
4193 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
4194 		break;
4195 	case MEM_MC1:
4196 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
4197 			return (EINVAL);
4198 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
4199 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
4200 		break;
4201 	default:
4202 		return (EINVAL);
4203 	}
4204 
4205 	*addr = maddr + off;	/* global address */
4206 	return (validate_mem_range(sc, *addr, len));
4207 }
4208 
4209 static int
4210 fixup_devlog_params(struct adapter *sc)
4211 {
4212 	struct devlog_params *dparams = &sc->params.devlog;
4213 	int rc;
4214 
4215 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
4216 	    dparams->size, &dparams->addr);
4217 
4218 	return (rc);
4219 }
4220 
4221 static void
4222 update_nirq(struct intrs_and_queues *iaq, int nports)
4223 {
4224 
4225 	iaq->nirq = T4_EXTRA_INTR;
4226 	iaq->nirq += nports * max(iaq->nrxq, iaq->nnmrxq);
4227 	iaq->nirq += nports * iaq->nofldrxq;
4228 	iaq->nirq += nports * (iaq->num_vis - 1) *
4229 	    max(iaq->nrxq_vi, iaq->nnmrxq_vi);
4230 	iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
4231 }
4232 
4233 /*
4234  * Adjust requirements to fit the number of interrupts available.
4235  */
4236 static void
4237 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
4238     int navail)
4239 {
4240 	int old_nirq;
4241 	const int nports = sc->params.nports;
4242 
4243 	MPASS(nports > 0);
4244 	MPASS(navail > 0);
4245 
4246 	bzero(iaq, sizeof(*iaq));
4247 	iaq->intr_type = itype;
4248 	iaq->num_vis = t4_num_vis;
4249 	iaq->ntxq = t4_ntxq;
4250 	iaq->ntxq_vi = t4_ntxq_vi;
4251 	iaq->nrxq = t4_nrxq;
4252 	iaq->nrxq_vi = t4_nrxq_vi;
4253 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
4254 	if (is_offload(sc) || is_ethoffload(sc)) {
4255 		iaq->nofldtxq = t4_nofldtxq;
4256 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
4257 	}
4258 #endif
4259 #ifdef TCP_OFFLOAD
4260 	if (is_offload(sc)) {
4261 		iaq->nofldrxq = t4_nofldrxq;
4262 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
4263 	}
4264 #endif
4265 #ifdef DEV_NETMAP
4266 	if (t4_native_netmap & NN_MAIN_VI) {
4267 		iaq->nnmtxq = t4_nnmtxq;
4268 		iaq->nnmrxq = t4_nnmrxq;
4269 	}
4270 	if (t4_native_netmap & NN_EXTRA_VI) {
4271 		iaq->nnmtxq_vi = t4_nnmtxq_vi;
4272 		iaq->nnmrxq_vi = t4_nnmrxq_vi;
4273 	}
4274 #endif
4275 
4276 	update_nirq(iaq, nports);
4277 	if (iaq->nirq <= navail &&
4278 	    (itype != INTR_MSI || powerof2(iaq->nirq))) {
4279 		/*
4280 		 * This is the normal case -- there are enough interrupts for
4281 		 * everything.
4282 		 */
4283 		goto done;
4284 	}
4285 
4286 	/*
4287 	 * If extra VIs have been configured try reducing their count and see if
4288 	 * that works.
4289 	 */
4290 	while (iaq->num_vis > 1) {
4291 		iaq->num_vis--;
4292 		update_nirq(iaq, nports);
4293 		if (iaq->nirq <= navail &&
4294 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
4295 			device_printf(sc->dev, "virtual interfaces per port "
4296 			    "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
4297 			    "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
4298 			    "itype %d, navail %u, nirq %d.\n",
4299 			    iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
4300 			    iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
4301 			    itype, navail, iaq->nirq);
4302 			goto done;
4303 		}
4304 	}
4305 
4306 	/*
4307 	 * Extra VIs will not be created.  Log a message if they were requested.
4308 	 */
4309 	MPASS(iaq->num_vis == 1);
4310 	iaq->ntxq_vi = iaq->nrxq_vi = 0;
4311 	iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
4312 	iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
4313 	if (iaq->num_vis != t4_num_vis) {
4314 		device_printf(sc->dev, "extra virtual interfaces disabled.  "
4315 		    "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
4316 		    "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
4317 		    iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
4318 		    iaq->nnmrxq_vi, itype, navail, iaq->nirq);
4319 	}
4320 
4321 	/*
4322 	 * Keep reducing the number of NIC rx queues to the next lower power of
4323 	 * 2 (for even RSS distribution) and halving the TOE rx queues and see
4324 	 * if that works.
4325 	 */
4326 	do {
4327 		if (iaq->nrxq > 1) {
4328 			do {
4329 				iaq->nrxq--;
4330 			} while (!powerof2(iaq->nrxq));
4331 			if (iaq->nnmrxq > iaq->nrxq)
4332 				iaq->nnmrxq = iaq->nrxq;
4333 		}
4334 		if (iaq->nofldrxq > 1)
4335 			iaq->nofldrxq >>= 1;
4336 
4337 		old_nirq = iaq->nirq;
4338 		update_nirq(iaq, nports);
4339 		if (iaq->nirq <= navail &&
4340 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
4341 			device_printf(sc->dev, "running with reduced number of "
4342 			    "rx queues because of shortage of interrupts.  "
4343 			    "nrxq=%u, nofldrxq=%u.  "
4344 			    "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
4345 			    iaq->nofldrxq, itype, navail, iaq->nirq);
4346 			goto done;
4347 		}
4348 	} while (old_nirq != iaq->nirq);
4349 
4350 	/* One interrupt for everything.  Ugh. */
4351 	device_printf(sc->dev, "running with minimal number of queues.  "
4352 	    "itype %d, navail %u.\n", itype, navail);
4353 	iaq->nirq = 1;
4354 	iaq->nrxq = 1;
4355 	iaq->ntxq = 1;
4356 	if (iaq->nofldrxq > 0) {
4357 		iaq->nofldrxq = 1;
4358 		iaq->nofldtxq = 1;
4359 	}
4360 	iaq->nnmtxq = 0;
4361 	iaq->nnmrxq = 0;
4362 done:
4363 	MPASS(iaq->num_vis > 0);
4364 	if (iaq->num_vis > 1) {
4365 		MPASS(iaq->nrxq_vi > 0);
4366 		MPASS(iaq->ntxq_vi > 0);
4367 	}
4368 	MPASS(iaq->nirq > 0);
4369 	MPASS(iaq->nrxq > 0);
4370 	MPASS(iaq->ntxq > 0);
4371 	if (itype == INTR_MSI) {
4372 		MPASS(powerof2(iaq->nirq));
4373 	}
4374 }
4375 
4376 static int
4377 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
4378 {
4379 	int rc, itype, navail, nalloc;
4380 
4381 	for (itype = INTR_MSIX; itype; itype >>= 1) {
4382 
4383 		if ((itype & t4_intr_types) == 0)
4384 			continue;	/* not allowed */
4385 
4386 		if (itype == INTR_MSIX)
4387 			navail = pci_msix_count(sc->dev);
4388 		else if (itype == INTR_MSI)
4389 			navail = pci_msi_count(sc->dev);
4390 		else
4391 			navail = 1;
4392 restart:
4393 		if (navail == 0)
4394 			continue;
4395 
4396 		calculate_iaq(sc, iaq, itype, navail);
4397 		nalloc = iaq->nirq;
4398 		rc = 0;
4399 		if (itype == INTR_MSIX)
4400 			rc = pci_alloc_msix(sc->dev, &nalloc);
4401 		else if (itype == INTR_MSI)
4402 			rc = pci_alloc_msi(sc->dev, &nalloc);
4403 
4404 		if (rc == 0 && nalloc > 0) {
4405 			if (nalloc == iaq->nirq)
4406 				return (0);
4407 
4408 			/*
4409 			 * Didn't get the number requested.  Use whatever number
4410 			 * the kernel is willing to allocate.
4411 			 */
4412 			device_printf(sc->dev, "fewer vectors than requested, "
4413 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
4414 			    itype, iaq->nirq, nalloc);
4415 			pci_release_msi(sc->dev);
4416 			navail = nalloc;
4417 			goto restart;
4418 		}
4419 
4420 		device_printf(sc->dev,
4421 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
4422 		    itype, rc, iaq->nirq, nalloc);
4423 	}
4424 
4425 	device_printf(sc->dev,
4426 	    "failed to find a usable interrupt type.  "
4427 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
4428 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
4429 
4430 	return (ENXIO);
4431 }
4432 
4433 #define FW_VERSION(chip) ( \
4434     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
4435     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
4436     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
4437     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
4438 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
4439 
4440 /* Just enough of fw_hdr to cover all version info. */
4441 struct fw_h {
4442 	__u8	ver;
4443 	__u8	chip;
4444 	__be16	len512;
4445 	__be32	fw_ver;
4446 	__be32	tp_microcode_ver;
4447 	__u8	intfver_nic;
4448 	__u8	intfver_vnic;
4449 	__u8	intfver_ofld;
4450 	__u8	intfver_ri;
4451 	__u8	intfver_iscsipdu;
4452 	__u8	intfver_iscsi;
4453 	__u8	intfver_fcoepdu;
4454 	__u8	intfver_fcoe;
4455 };
4456 /* Spot check a couple of fields. */
4457 CTASSERT(offsetof(struct fw_h, fw_ver) == offsetof(struct fw_hdr, fw_ver));
4458 CTASSERT(offsetof(struct fw_h, intfver_nic) == offsetof(struct fw_hdr, intfver_nic));
4459 CTASSERT(offsetof(struct fw_h, intfver_fcoe) == offsetof(struct fw_hdr, intfver_fcoe));
4460 
4461 struct fw_info {
4462 	uint8_t chip;
4463 	char *kld_name;
4464 	char *fw_mod_name;
4465 	struct fw_h fw_h;
4466 } fw_info[] = {
4467 	{
4468 		.chip = CHELSIO_T4,
4469 		.kld_name = "t4fw_cfg",
4470 		.fw_mod_name = "t4fw",
4471 		.fw_h = {
4472 			.chip = FW_HDR_CHIP_T4,
4473 			.fw_ver = htobe32(FW_VERSION(T4)),
4474 			.intfver_nic = FW_INTFVER(T4, NIC),
4475 			.intfver_vnic = FW_INTFVER(T4, VNIC),
4476 			.intfver_ofld = FW_INTFVER(T4, OFLD),
4477 			.intfver_ri = FW_INTFVER(T4, RI),
4478 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
4479 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
4480 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
4481 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
4482 		},
4483 	}, {
4484 		.chip = CHELSIO_T5,
4485 		.kld_name = "t5fw_cfg",
4486 		.fw_mod_name = "t5fw",
4487 		.fw_h = {
4488 			.chip = FW_HDR_CHIP_T5,
4489 			.fw_ver = htobe32(FW_VERSION(T5)),
4490 			.intfver_nic = FW_INTFVER(T5, NIC),
4491 			.intfver_vnic = FW_INTFVER(T5, VNIC),
4492 			.intfver_ofld = FW_INTFVER(T5, OFLD),
4493 			.intfver_ri = FW_INTFVER(T5, RI),
4494 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
4495 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
4496 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
4497 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
4498 		},
4499 	}, {
4500 		.chip = CHELSIO_T6,
4501 		.kld_name = "t6fw_cfg",
4502 		.fw_mod_name = "t6fw",
4503 		.fw_h = {
4504 			.chip = FW_HDR_CHIP_T6,
4505 			.fw_ver = htobe32(FW_VERSION(T6)),
4506 			.intfver_nic = FW_INTFVER(T6, NIC),
4507 			.intfver_vnic = FW_INTFVER(T6, VNIC),
4508 			.intfver_ofld = FW_INTFVER(T6, OFLD),
4509 			.intfver_ri = FW_INTFVER(T6, RI),
4510 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
4511 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
4512 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
4513 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
4514 		},
4515 	}
4516 };
4517 
4518 static struct fw_info *
4519 find_fw_info(int chip)
4520 {
4521 	int i;
4522 
4523 	for (i = 0; i < nitems(fw_info); i++) {
4524 		if (fw_info[i].chip == chip)
4525 			return (&fw_info[i]);
4526 	}
4527 	return (NULL);
4528 }
4529 
4530 /*
4531  * Is the given firmware API compatible with the one the driver was compiled
4532  * with?
4533  */
4534 static int
4535 fw_compatible(const struct fw_h *hdr1, const struct fw_h *hdr2)
4536 {
4537 
4538 	/* short circuit if it's the exact same firmware version */
4539 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
4540 		return (1);
4541 
4542 	/*
4543 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
4544 	 * features that are supported in the driver.
4545 	 */
4546 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
4547 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
4548 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
4549 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
4550 		return (1);
4551 #undef SAME_INTF
4552 
4553 	return (0);
4554 }
4555 
4556 static int
4557 load_fw_module(struct adapter *sc, const struct firmware **dcfg,
4558     const struct firmware **fw)
4559 {
4560 	struct fw_info *fw_info;
4561 
4562 	*dcfg = NULL;
4563 	if (fw != NULL)
4564 		*fw = NULL;
4565 
4566 	fw_info = find_fw_info(chip_id(sc));
4567 	if (fw_info == NULL) {
4568 		device_printf(sc->dev,
4569 		    "unable to look up firmware information for chip %d.\n",
4570 		    chip_id(sc));
4571 		return (EINVAL);
4572 	}
4573 
4574 	*dcfg = firmware_get(fw_info->kld_name);
4575 	if (*dcfg != NULL) {
4576 		if (fw != NULL)
4577 			*fw = firmware_get(fw_info->fw_mod_name);
4578 		return (0);
4579 	}
4580 
4581 	return (ENOENT);
4582 }
4583 
4584 static void
4585 unload_fw_module(struct adapter *sc, const struct firmware *dcfg,
4586     const struct firmware *fw)
4587 {
4588 
4589 	if (fw != NULL)
4590 		firmware_put(fw, FIRMWARE_UNLOAD);
4591 	if (dcfg != NULL)
4592 		firmware_put(dcfg, FIRMWARE_UNLOAD);
4593 }
4594 
4595 /*
4596  * Return values:
4597  * 0 means no firmware install attempted.
4598  * ERESTART means a firmware install was attempted and was successful.
4599  * +ve errno means a firmware install was attempted but failed.
4600  */
4601 static int
4602 install_kld_firmware(struct adapter *sc, struct fw_h *card_fw,
4603     const struct fw_h *drv_fw, const char *reason, int *already)
4604 {
4605 	const struct firmware *cfg, *fw;
4606 	const uint32_t c = be32toh(card_fw->fw_ver);
4607 	uint32_t d, k;
4608 	int rc, fw_install;
4609 	struct fw_h bundled_fw;
4610 	bool load_attempted;
4611 
4612 	cfg = fw = NULL;
4613 	load_attempted = false;
4614 	fw_install = t4_fw_install < 0 ? -t4_fw_install : t4_fw_install;
4615 
4616 	memcpy(&bundled_fw, drv_fw, sizeof(bundled_fw));
4617 	if (t4_fw_install < 0) {
4618 		rc = load_fw_module(sc, &cfg, &fw);
4619 		if (rc != 0 || fw == NULL) {
4620 			device_printf(sc->dev,
4621 			    "failed to load firmware module: %d. cfg %p, fw %p;"
4622 			    " will use compiled-in firmware version for"
4623 			    "hw.cxgbe.fw_install checks.\n",
4624 			    rc, cfg, fw);
4625 		} else {
4626 			memcpy(&bundled_fw, fw->data, sizeof(bundled_fw));
4627 		}
4628 		load_attempted = true;
4629 	}
4630 	d = be32toh(bundled_fw.fw_ver);
4631 
4632 	if (reason != NULL)
4633 		goto install;
4634 
4635 	if ((sc->flags & FW_OK) == 0) {
4636 
4637 		if (c == 0xffffffff) {
4638 			reason = "missing";
4639 			goto install;
4640 		}
4641 
4642 		rc = 0;
4643 		goto done;
4644 	}
4645 
4646 	if (!fw_compatible(card_fw, &bundled_fw)) {
4647 		reason = "incompatible or unusable";
4648 		goto install;
4649 	}
4650 
4651 	if (d > c) {
4652 		reason = "older than the version bundled with this driver";
4653 		goto install;
4654 	}
4655 
4656 	if (fw_install == 2 && d != c) {
4657 		reason = "different than the version bundled with this driver";
4658 		goto install;
4659 	}
4660 
4661 	/* No reason to do anything to the firmware already on the card. */
4662 	rc = 0;
4663 	goto done;
4664 
4665 install:
4666 	rc = 0;
4667 	if ((*already)++)
4668 		goto done;
4669 
4670 	if (fw_install == 0) {
4671 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
4672 		    "but the driver is prohibited from installing a firmware "
4673 		    "on the card.\n",
4674 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
4675 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
4676 
4677 		goto done;
4678 	}
4679 
4680 	/*
4681 	 * We'll attempt to install a firmware.  Load the module first (if it
4682 	 * hasn't been loaded already).
4683 	 */
4684 	if (!load_attempted) {
4685 		rc = load_fw_module(sc, &cfg, &fw);
4686 		if (rc != 0 || fw == NULL) {
4687 			device_printf(sc->dev,
4688 			    "failed to load firmware module: %d. cfg %p, fw %p\n",
4689 			    rc, cfg, fw);
4690 			/* carry on */
4691 		}
4692 	}
4693 	if (fw == NULL) {
4694 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
4695 		    "but the driver cannot take corrective action because it "
4696 		    "is unable to load the firmware module.\n",
4697 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
4698 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
4699 		rc = sc->flags & FW_OK ? 0 : ENOENT;
4700 		goto done;
4701 	}
4702 	k = be32toh(((const struct fw_hdr *)fw->data)->fw_ver);
4703 	if (k != d) {
4704 		MPASS(t4_fw_install > 0);
4705 		device_printf(sc->dev,
4706 		    "firmware in KLD (%u.%u.%u.%u) is not what the driver was "
4707 		    "expecting (%u.%u.%u.%u) and will not be used.\n",
4708 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
4709 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k),
4710 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
4711 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
4712 		rc = sc->flags & FW_OK ? 0 : EINVAL;
4713 		goto done;
4714 	}
4715 
4716 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
4717 	    "installing firmware %u.%u.%u.%u on card.\n",
4718 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
4719 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
4720 	    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
4721 	    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
4722 
4723 	rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
4724 	if (rc != 0) {
4725 		device_printf(sc->dev, "failed to install firmware: %d\n", rc);
4726 	} else {
4727 		/* Installed successfully, update the cached header too. */
4728 		rc = ERESTART;
4729 		memcpy(card_fw, fw->data, sizeof(*card_fw));
4730 	}
4731 done:
4732 	unload_fw_module(sc, cfg, fw);
4733 
4734 	return (rc);
4735 }
4736 
4737 /*
4738  * Establish contact with the firmware and attempt to become the master driver.
4739  *
4740  * A firmware will be installed to the card if needed (if the driver is allowed
4741  * to do so).
4742  */
4743 static int
4744 contact_firmware(struct adapter *sc)
4745 {
4746 	int rc, already = 0;
4747 	enum dev_state state;
4748 	struct fw_info *fw_info;
4749 	struct fw_hdr *card_fw;		/* fw on the card */
4750 	const struct fw_h *drv_fw;
4751 
4752 	fw_info = find_fw_info(chip_id(sc));
4753 	if (fw_info == NULL) {
4754 		device_printf(sc->dev,
4755 		    "unable to look up firmware information for chip %d.\n",
4756 		    chip_id(sc));
4757 		return (EINVAL);
4758 	}
4759 	drv_fw = &fw_info->fw_h;
4760 
4761 	/* Read the header of the firmware on the card */
4762 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
4763 restart:
4764 	rc = -t4_get_fw_hdr(sc, card_fw);
4765 	if (rc != 0) {
4766 		device_printf(sc->dev,
4767 		    "unable to read firmware header from card's flash: %d\n",
4768 		    rc);
4769 		goto done;
4770 	}
4771 
4772 	rc = install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw, NULL,
4773 	    &already);
4774 	if (rc == ERESTART)
4775 		goto restart;
4776 	if (rc != 0)
4777 		goto done;
4778 
4779 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
4780 	if (rc < 0 || state == DEV_STATE_ERR) {
4781 		rc = -rc;
4782 		device_printf(sc->dev,
4783 		    "failed to connect to the firmware: %d, %d.  "
4784 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
4785 #if 0
4786 		if (install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw,
4787 		    "not responding properly to HELLO", &already) == ERESTART)
4788 			goto restart;
4789 #endif
4790 		goto done;
4791 	}
4792 	MPASS(be32toh(card_fw->flags) & FW_HDR_FLAGS_RESET_HALT);
4793 	sc->flags |= FW_OK;	/* The firmware responded to the FW_HELLO. */
4794 
4795 	if (rc == sc->pf) {
4796 		sc->flags |= MASTER_PF;
4797 		rc = install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw,
4798 		    NULL, &already);
4799 		if (rc == ERESTART)
4800 			rc = 0;
4801 		else if (rc != 0)
4802 			goto done;
4803 	} else if (state == DEV_STATE_UNINIT) {
4804 		/*
4805 		 * We didn't get to be the master so we definitely won't be
4806 		 * configuring the chip.  It's a bug if someone else hasn't
4807 		 * configured it already.
4808 		 */
4809 		device_printf(sc->dev, "couldn't be master(%d), "
4810 		    "device not already initialized either(%d).  "
4811 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
4812 		rc = EPROTO;
4813 		goto done;
4814 	} else {
4815 		/*
4816 		 * Some other PF is the master and has configured the chip.
4817 		 * This is allowed but untested.
4818 		 */
4819 		device_printf(sc->dev, "PF%d is master, device state %d.  "
4820 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
4821 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", rc);
4822 		sc->cfcsum = 0;
4823 		rc = 0;
4824 	}
4825 done:
4826 	if (rc != 0 && sc->flags & FW_OK) {
4827 		t4_fw_bye(sc, sc->mbox);
4828 		sc->flags &= ~FW_OK;
4829 	}
4830 	free(card_fw, M_CXGBE);
4831 	return (rc);
4832 }
4833 
4834 static int
4835 copy_cfg_file_to_card(struct adapter *sc, char *cfg_file,
4836     uint32_t mtype, uint32_t moff)
4837 {
4838 	struct fw_info *fw_info;
4839 	const struct firmware *dcfg, *rcfg = NULL;
4840 	const uint32_t *cfdata;
4841 	uint32_t cflen, addr;
4842 	int rc;
4843 
4844 	load_fw_module(sc, &dcfg, NULL);
4845 
4846 	/* Card specific interpretation of "default". */
4847 	if (strncmp(cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
4848 		if (pci_get_device(sc->dev) == 0x440a)
4849 			snprintf(cfg_file, sizeof(t4_cfg_file), UWIRE_CF);
4850 		if (is_fpga(sc))
4851 			snprintf(cfg_file, sizeof(t4_cfg_file), FPGA_CF);
4852 	}
4853 
4854 	if (strncmp(cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
4855 		if (dcfg == NULL) {
4856 			device_printf(sc->dev,
4857 			    "KLD with default config is not available.\n");
4858 			rc = ENOENT;
4859 			goto done;
4860 		}
4861 		cfdata = dcfg->data;
4862 		cflen = dcfg->datasize & ~3;
4863 	} else {
4864 		char s[32];
4865 
4866 		fw_info = find_fw_info(chip_id(sc));
4867 		if (fw_info == NULL) {
4868 			device_printf(sc->dev,
4869 			    "unable to look up firmware information for chip %d.\n",
4870 			    chip_id(sc));
4871 			rc = EINVAL;
4872 			goto done;
4873 		}
4874 		snprintf(s, sizeof(s), "%s_%s", fw_info->kld_name, cfg_file);
4875 
4876 		rcfg = firmware_get(s);
4877 		if (rcfg == NULL) {
4878 			device_printf(sc->dev,
4879 			    "unable to load module \"%s\" for configuration "
4880 			    "profile \"%s\".\n", s, cfg_file);
4881 			rc = ENOENT;
4882 			goto done;
4883 		}
4884 		cfdata = rcfg->data;
4885 		cflen = rcfg->datasize & ~3;
4886 	}
4887 
4888 	if (cflen > FLASH_CFG_MAX_SIZE) {
4889 		device_printf(sc->dev,
4890 		    "config file too long (%d, max allowed is %d).\n",
4891 		    cflen, FLASH_CFG_MAX_SIZE);
4892 		rc = EINVAL;
4893 		goto done;
4894 	}
4895 
4896 	rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
4897 	if (rc != 0) {
4898 		device_printf(sc->dev,
4899 		    "%s: addr (%d/0x%x) or len %d is not valid: %d.\n",
4900 		    __func__, mtype, moff, cflen, rc);
4901 		rc = EINVAL;
4902 		goto done;
4903 	}
4904 	write_via_memwin(sc, 2, addr, cfdata, cflen);
4905 done:
4906 	if (rcfg != NULL)
4907 		firmware_put(rcfg, FIRMWARE_UNLOAD);
4908 	unload_fw_module(sc, dcfg, NULL);
4909 	return (rc);
4910 }
4911 
4912 struct caps_allowed {
4913 	uint16_t nbmcaps;
4914 	uint16_t linkcaps;
4915 	uint16_t switchcaps;
4916 	uint16_t niccaps;
4917 	uint16_t toecaps;
4918 	uint16_t rdmacaps;
4919 	uint16_t cryptocaps;
4920 	uint16_t iscsicaps;
4921 	uint16_t fcoecaps;
4922 };
4923 
4924 #define FW_PARAM_DEV(param) \
4925 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
4926 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
4927 #define FW_PARAM_PFVF(param) \
4928 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
4929 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
4930 
4931 /*
4932  * Provide a configuration profile to the firmware and have it initialize the
4933  * chip accordingly.  This may involve uploading a configuration file to the
4934  * card.
4935  */
4936 static int
4937 apply_cfg_and_initialize(struct adapter *sc, char *cfg_file,
4938     const struct caps_allowed *caps_allowed)
4939 {
4940 	int rc;
4941 	struct fw_caps_config_cmd caps;
4942 	uint32_t mtype, moff, finicsum, cfcsum, param, val;
4943 
4944 	rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
4945 	if (rc != 0) {
4946 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
4947 		return (rc);
4948 	}
4949 
4950 	bzero(&caps, sizeof(caps));
4951 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4952 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
4953 	if (strncmp(cfg_file, BUILTIN_CF, sizeof(t4_cfg_file)) == 0) {
4954 		mtype = 0;
4955 		moff = 0;
4956 		caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
4957 	} else if (strncmp(cfg_file, FLASH_CF, sizeof(t4_cfg_file)) == 0) {
4958 		mtype = FW_MEMTYPE_FLASH;
4959 		moff = t4_flash_cfg_addr(sc);
4960 		caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
4961 		    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
4962 		    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) |
4963 		    FW_LEN16(caps));
4964 	} else {
4965 		/*
4966 		 * Ask the firmware where it wants us to upload the config file.
4967 		 */
4968 		param = FW_PARAM_DEV(CF);
4969 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4970 		if (rc != 0) {
4971 			/* No support for config file?  Shouldn't happen. */
4972 			device_printf(sc->dev,
4973 			    "failed to query config file location: %d.\n", rc);
4974 			goto done;
4975 		}
4976 		mtype = G_FW_PARAMS_PARAM_Y(val);
4977 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
4978 		caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
4979 		    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
4980 		    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) |
4981 		    FW_LEN16(caps));
4982 
4983 		rc = copy_cfg_file_to_card(sc, cfg_file, mtype, moff);
4984 		if (rc != 0) {
4985 			device_printf(sc->dev,
4986 			    "failed to upload config file to card: %d.\n", rc);
4987 			goto done;
4988 		}
4989 	}
4990 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
4991 	if (rc != 0) {
4992 		device_printf(sc->dev, "failed to pre-process config file: %d "
4993 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
4994 		goto done;
4995 	}
4996 
4997 	finicsum = be32toh(caps.finicsum);
4998 	cfcsum = be32toh(caps.cfcsum);	/* actual */
4999 	if (finicsum != cfcsum) {
5000 		device_printf(sc->dev,
5001 		    "WARNING: config file checksum mismatch: %08x %08x\n",
5002 		    finicsum, cfcsum);
5003 	}
5004 	sc->cfcsum = cfcsum;
5005 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", cfg_file);
5006 
5007 	/*
5008 	 * Let the firmware know what features will (not) be used so it can tune
5009 	 * things accordingly.
5010 	 */
5011 #define LIMIT_CAPS(x) do { \
5012 	caps.x##caps &= htobe16(caps_allowed->x##caps); \
5013 } while (0)
5014 	LIMIT_CAPS(nbm);
5015 	LIMIT_CAPS(link);
5016 	LIMIT_CAPS(switch);
5017 	LIMIT_CAPS(nic);
5018 	LIMIT_CAPS(toe);
5019 	LIMIT_CAPS(rdma);
5020 	LIMIT_CAPS(crypto);
5021 	LIMIT_CAPS(iscsi);
5022 	LIMIT_CAPS(fcoe);
5023 #undef LIMIT_CAPS
5024 	if (caps.niccaps & htobe16(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
5025 		/*
5026 		 * TOE and hashfilters are mutually exclusive.  It is a config
5027 		 * file or firmware bug if both are reported as available.  Try
5028 		 * to cope with the situation in non-debug builds by disabling
5029 		 * TOE.
5030 		 */
5031 		MPASS(caps.toecaps == 0);
5032 
5033 		caps.toecaps = 0;
5034 		caps.rdmacaps = 0;
5035 		caps.iscsicaps = 0;
5036 	}
5037 
5038 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
5039 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
5040 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
5041 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
5042 	if (rc != 0) {
5043 		device_printf(sc->dev,
5044 		    "failed to process config file: %d.\n", rc);
5045 		goto done;
5046 	}
5047 
5048 	t4_tweak_chip_settings(sc);
5049 	set_params__pre_init(sc);
5050 
5051 	/* get basic stuff going */
5052 	rc = -t4_fw_initialize(sc, sc->mbox);
5053 	if (rc != 0) {
5054 		device_printf(sc->dev, "fw_initialize failed: %d.\n", rc);
5055 		goto done;
5056 	}
5057 done:
5058 	return (rc);
5059 }
5060 
5061 /*
5062  * Partition chip resources for use between various PFs, VFs, etc.
5063  */
5064 static int
5065 partition_resources(struct adapter *sc)
5066 {
5067 	char cfg_file[sizeof(t4_cfg_file)];
5068 	struct caps_allowed caps_allowed;
5069 	int rc;
5070 	bool fallback;
5071 
5072 	/* Only the master driver gets to configure the chip resources. */
5073 	MPASS(sc->flags & MASTER_PF);
5074 
5075 #define COPY_CAPS(x) do { \
5076 	caps_allowed.x##caps = t4_##x##caps_allowed; \
5077 } while (0)
5078 	bzero(&caps_allowed, sizeof(caps_allowed));
5079 	COPY_CAPS(nbm);
5080 	COPY_CAPS(link);
5081 	COPY_CAPS(switch);
5082 	COPY_CAPS(nic);
5083 	COPY_CAPS(toe);
5084 	COPY_CAPS(rdma);
5085 	COPY_CAPS(crypto);
5086 	COPY_CAPS(iscsi);
5087 	COPY_CAPS(fcoe);
5088 	fallback = sc->debug_flags & DF_DISABLE_CFG_RETRY ? false : true;
5089 	snprintf(cfg_file, sizeof(cfg_file), "%s", t4_cfg_file);
5090 retry:
5091 	rc = apply_cfg_and_initialize(sc, cfg_file, &caps_allowed);
5092 	if (rc != 0 && fallback) {
5093 		dump_devlog(sc);
5094 		device_printf(sc->dev,
5095 		    "failed (%d) to configure card with \"%s\" profile, "
5096 		    "will fall back to a basic configuration and retry.\n",
5097 		    rc, cfg_file);
5098 		snprintf(cfg_file, sizeof(cfg_file), "%s", BUILTIN_CF);
5099 		bzero(&caps_allowed, sizeof(caps_allowed));
5100 		COPY_CAPS(switch);
5101 		caps_allowed.niccaps = FW_CAPS_CONFIG_NIC;
5102 		fallback = false;
5103 		goto retry;
5104 	}
5105 #undef COPY_CAPS
5106 	return (rc);
5107 }
5108 
5109 /*
5110  * Retrieve parameters that are needed (or nice to have) very early.
5111  */
5112 static int
5113 get_params__pre_init(struct adapter *sc)
5114 {
5115 	int rc;
5116 	uint32_t param[2], val[2];
5117 
5118 	t4_get_version_info(sc);
5119 
5120 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
5121 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
5122 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
5123 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
5124 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
5125 
5126 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
5127 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
5128 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
5129 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
5130 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
5131 
5132 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
5133 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
5134 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
5135 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
5136 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
5137 
5138 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
5139 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
5140 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
5141 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
5142 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
5143 
5144 	param[0] = FW_PARAM_DEV(PORTVEC);
5145 	param[1] = FW_PARAM_DEV(CCLK);
5146 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
5147 	if (rc != 0) {
5148 		device_printf(sc->dev,
5149 		    "failed to query parameters (pre_init): %d.\n", rc);
5150 		return (rc);
5151 	}
5152 
5153 	sc->params.portvec = val[0];
5154 	sc->params.nports = bitcount32(val[0]);
5155 	sc->params.vpd.cclk = val[1];
5156 
5157 	/* Read device log parameters. */
5158 	rc = -t4_init_devlog_params(sc, 1);
5159 	if (rc == 0)
5160 		fixup_devlog_params(sc);
5161 	else {
5162 		device_printf(sc->dev,
5163 		    "failed to get devlog parameters: %d.\n", rc);
5164 		rc = 0;	/* devlog isn't critical for device operation */
5165 	}
5166 
5167 	return (rc);
5168 }
5169 
5170 /*
5171  * Any params that need to be set before FW_INITIALIZE.
5172  */
5173 static int
5174 set_params__pre_init(struct adapter *sc)
5175 {
5176 	int rc = 0;
5177 	uint32_t param, val;
5178 
5179 	if (chip_id(sc) >= CHELSIO_T6) {
5180 		param = FW_PARAM_DEV(HPFILTER_REGION_SUPPORT);
5181 		val = 1;
5182 		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
5183 		/* firmwares < 1.20.1.0 do not have this param. */
5184 		if (rc == FW_EINVAL &&
5185 		    sc->params.fw_vers < FW_VERSION32(1, 20, 1, 0)) {
5186 			rc = 0;
5187 		}
5188 		if (rc != 0) {
5189 			device_printf(sc->dev,
5190 			    "failed to enable high priority filters :%d.\n",
5191 			    rc);
5192 		}
5193 
5194 		param = FW_PARAM_DEV(PPOD_EDRAM);
5195 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
5196 		if (rc == 0 && val == 1) {
5197 			rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param,
5198 			    &val);
5199 			if (rc != 0) {
5200 				device_printf(sc->dev,
5201 				    "failed to set PPOD_EDRAM: %d.\n", rc);
5202 			}
5203 		}
5204 	}
5205 
5206 	/* Enable opaque VIIDs with firmwares that support it. */
5207 	param = FW_PARAM_DEV(OPAQUE_VIID_SMT_EXTN);
5208 	val = 1;
5209 	rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
5210 	if (rc == 0 && val == 1)
5211 		sc->params.viid_smt_extn_support = true;
5212 	else
5213 		sc->params.viid_smt_extn_support = false;
5214 
5215 	return (rc);
5216 }
5217 
5218 /*
5219  * Retrieve various parameters that are of interest to the driver.  The device
5220  * has been initialized by the firmware at this point.
5221  */
5222 static int
5223 get_params__post_init(struct adapter *sc)
5224 {
5225 	int rc;
5226 	uint32_t param[7], val[7];
5227 	struct fw_caps_config_cmd caps;
5228 
5229 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
5230 	param[1] = FW_PARAM_PFVF(EQ_START);
5231 	param[2] = FW_PARAM_PFVF(FILTER_START);
5232 	param[3] = FW_PARAM_PFVF(FILTER_END);
5233 	param[4] = FW_PARAM_PFVF(L2T_START);
5234 	param[5] = FW_PARAM_PFVF(L2T_END);
5235 	param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5236 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
5237 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
5238 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
5239 	if (rc != 0) {
5240 		device_printf(sc->dev,
5241 		    "failed to query parameters (post_init): %d.\n", rc);
5242 		return (rc);
5243 	}
5244 
5245 	sc->sge.iq_start = val[0];
5246 	sc->sge.eq_start = val[1];
5247 	if ((int)val[3] > (int)val[2]) {
5248 		sc->tids.ftid_base = val[2];
5249 		sc->tids.ftid_end = val[3];
5250 		sc->tids.nftids = val[3] - val[2] + 1;
5251 	}
5252 	sc->vres.l2t.start = val[4];
5253 	sc->vres.l2t.size = val[5] - val[4] + 1;
5254 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
5255 	    ("%s: L2 table size (%u) larger than expected (%u)",
5256 	    __func__, sc->vres.l2t.size, L2T_SIZE));
5257 	sc->params.core_vdd = val[6];
5258 
5259 	param[0] = FW_PARAM_PFVF(IQFLINT_END);
5260 	param[1] = FW_PARAM_PFVF(EQ_END);
5261 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
5262 	if (rc != 0) {
5263 		device_printf(sc->dev,
5264 		    "failed to query parameters (post_init2): %d.\n", rc);
5265 		return (rc);
5266 	}
5267 	MPASS((int)val[0] >= sc->sge.iq_start);
5268 	sc->sge.iqmap_sz = val[0] - sc->sge.iq_start + 1;
5269 	MPASS((int)val[1] >= sc->sge.eq_start);
5270 	sc->sge.eqmap_sz = val[1] - sc->sge.eq_start + 1;
5271 
5272 	if (chip_id(sc) >= CHELSIO_T6) {
5273 
5274 		sc->tids.tid_base = t4_read_reg(sc,
5275 		    A_LE_DB_ACTIVE_TABLE_START_INDEX);
5276 
5277 		param[0] = FW_PARAM_PFVF(HPFILTER_START);
5278 		param[1] = FW_PARAM_PFVF(HPFILTER_END);
5279 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
5280 		if (rc != 0) {
5281 			device_printf(sc->dev,
5282 			   "failed to query hpfilter parameters: %d.\n", rc);
5283 			return (rc);
5284 		}
5285 		if ((int)val[1] > (int)val[0]) {
5286 			sc->tids.hpftid_base = val[0];
5287 			sc->tids.hpftid_end = val[1];
5288 			sc->tids.nhpftids = val[1] - val[0] + 1;
5289 
5290 			/*
5291 			 * These should go off if the layout changes and the
5292 			 * driver needs to catch up.
5293 			 */
5294 			MPASS(sc->tids.hpftid_base == 0);
5295 			MPASS(sc->tids.tid_base == sc->tids.nhpftids);
5296 		}
5297 
5298 		param[0] = FW_PARAM_PFVF(RAWF_START);
5299 		param[1] = FW_PARAM_PFVF(RAWF_END);
5300 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
5301 		if (rc != 0) {
5302 			device_printf(sc->dev,
5303 			   "failed to query rawf parameters: %d.\n", rc);
5304 			return (rc);
5305 		}
5306 		if ((int)val[1] > (int)val[0]) {
5307 			sc->rawf_base = val[0];
5308 			sc->nrawf = val[1] - val[0] + 1;
5309 		}
5310 	}
5311 
5312 	/*
5313 	 * MPSBGMAP is queried separately because only recent firmwares support
5314 	 * it as a parameter and we don't want the compound query above to fail
5315 	 * on older firmwares.
5316 	 */
5317 	param[0] = FW_PARAM_DEV(MPSBGMAP);
5318 	val[0] = 0;
5319 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
5320 	if (rc == 0)
5321 		sc->params.mps_bg_map = val[0];
5322 	else
5323 		sc->params.mps_bg_map = 0;
5324 
5325 	/*
5326 	 * Determine whether the firmware supports the filter2 work request.
5327 	 * This is queried separately for the same reason as MPSBGMAP above.
5328 	 */
5329 	param[0] = FW_PARAM_DEV(FILTER2_WR);
5330 	val[0] = 0;
5331 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
5332 	if (rc == 0)
5333 		sc->params.filter2_wr_support = val[0] != 0;
5334 	else
5335 		sc->params.filter2_wr_support = 0;
5336 
5337 	/*
5338 	 * Find out whether we're allowed to use the ULPTX MEMWRITE DSGL.
5339 	 * This is queried separately for the same reason as other params above.
5340 	 */
5341 	param[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
5342 	val[0] = 0;
5343 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
5344 	if (rc == 0)
5345 		sc->params.ulptx_memwrite_dsgl = val[0] != 0;
5346 	else
5347 		sc->params.ulptx_memwrite_dsgl = false;
5348 
5349 	/* FW_RI_FR_NSMR_TPTE_WR support */
5350 	param[0] = FW_PARAM_DEV(RI_FR_NSMR_TPTE_WR);
5351 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
5352 	if (rc == 0)
5353 		sc->params.fr_nsmr_tpte_wr_support = val[0] != 0;
5354 	else
5355 		sc->params.fr_nsmr_tpte_wr_support = false;
5356 
5357 	/* Support for 512 SGL entries per FR MR. */
5358 	param[0] = FW_PARAM_DEV(DEV_512SGL_MR);
5359 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
5360 	if (rc == 0)
5361 		sc->params.dev_512sgl_mr = val[0] != 0;
5362 	else
5363 		sc->params.dev_512sgl_mr = false;
5364 
5365 	param[0] = FW_PARAM_PFVF(MAX_PKTS_PER_ETH_TX_PKTS_WR);
5366 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
5367 	if (rc == 0)
5368 		sc->params.max_pkts_per_eth_tx_pkts_wr = val[0];
5369 	else
5370 		sc->params.max_pkts_per_eth_tx_pkts_wr = 15;
5371 
5372 	param[0] = FW_PARAM_DEV(NUM_TM_CLASS);
5373 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
5374 	if (rc == 0) {
5375 		MPASS(val[0] > 0 && val[0] < 256);	/* nsched_cls is 8b */
5376 		sc->params.nsched_cls = val[0];
5377 	} else
5378 		sc->params.nsched_cls = sc->chip_params->nsched_cls;
5379 
5380 	/* get capabilites */
5381 	bzero(&caps, sizeof(caps));
5382 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
5383 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
5384 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
5385 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
5386 	if (rc != 0) {
5387 		device_printf(sc->dev,
5388 		    "failed to get card capabilities: %d.\n", rc);
5389 		return (rc);
5390 	}
5391 
5392 #define READ_CAPS(x) do { \
5393 	sc->x = htobe16(caps.x); \
5394 } while (0)
5395 	READ_CAPS(nbmcaps);
5396 	READ_CAPS(linkcaps);
5397 	READ_CAPS(switchcaps);
5398 	READ_CAPS(niccaps);
5399 	READ_CAPS(toecaps);
5400 	READ_CAPS(rdmacaps);
5401 	READ_CAPS(cryptocaps);
5402 	READ_CAPS(iscsicaps);
5403 	READ_CAPS(fcoecaps);
5404 
5405 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_HASHFILTER) {
5406 		MPASS(chip_id(sc) > CHELSIO_T4);
5407 		MPASS(sc->toecaps == 0);
5408 		sc->toecaps = 0;
5409 
5410 		param[0] = FW_PARAM_DEV(NTID);
5411 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
5412 		if (rc != 0) {
5413 			device_printf(sc->dev,
5414 			    "failed to query HASHFILTER parameters: %d.\n", rc);
5415 			return (rc);
5416 		}
5417 		sc->tids.ntids = val[0];
5418 		if (sc->params.fw_vers < FW_VERSION32(1, 20, 5, 0)) {
5419 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
5420 			sc->tids.ntids -= sc->tids.nhpftids;
5421 		}
5422 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
5423 		sc->params.hash_filter = 1;
5424 	}
5425 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
5426 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
5427 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
5428 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
5429 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
5430 		if (rc != 0) {
5431 			device_printf(sc->dev,
5432 			    "failed to query NIC parameters: %d.\n", rc);
5433 			return (rc);
5434 		}
5435 		if ((int)val[1] > (int)val[0]) {
5436 			sc->tids.etid_base = val[0];
5437 			sc->tids.etid_end = val[1];
5438 			sc->tids.netids = val[1] - val[0] + 1;
5439 			sc->params.eo_wr_cred = val[2];
5440 			sc->params.ethoffload = 1;
5441 		}
5442 	}
5443 	if (sc->toecaps) {
5444 		/* query offload-related parameters */
5445 		param[0] = FW_PARAM_DEV(NTID);
5446 		param[1] = FW_PARAM_PFVF(SERVER_START);
5447 		param[2] = FW_PARAM_PFVF(SERVER_END);
5448 		param[3] = FW_PARAM_PFVF(TDDP_START);
5449 		param[4] = FW_PARAM_PFVF(TDDP_END);
5450 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
5451 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
5452 		if (rc != 0) {
5453 			device_printf(sc->dev,
5454 			    "failed to query TOE parameters: %d.\n", rc);
5455 			return (rc);
5456 		}
5457 		sc->tids.ntids = val[0];
5458 		if (sc->params.fw_vers < FW_VERSION32(1, 20, 5, 0)) {
5459 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
5460 			sc->tids.ntids -= sc->tids.nhpftids;
5461 		}
5462 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
5463 		if ((int)val[2] > (int)val[1]) {
5464 			sc->tids.stid_base = val[1];
5465 			sc->tids.nstids = val[2] - val[1] + 1;
5466 		}
5467 		sc->vres.ddp.start = val[3];
5468 		sc->vres.ddp.size = val[4] - val[3] + 1;
5469 		sc->params.ofldq_wr_cred = val[5];
5470 		sc->params.offload = 1;
5471 	} else {
5472 		/*
5473 		 * The firmware attempts memfree TOE configuration for -SO cards
5474 		 * and will report toecaps=0 if it runs out of resources (this
5475 		 * depends on the config file).  It may not report 0 for other
5476 		 * capabilities dependent on the TOE in this case.  Set them to
5477 		 * 0 here so that the driver doesn't bother tracking resources
5478 		 * that will never be used.
5479 		 */
5480 		sc->iscsicaps = 0;
5481 		sc->rdmacaps = 0;
5482 	}
5483 	if (sc->rdmacaps) {
5484 		param[0] = FW_PARAM_PFVF(STAG_START);
5485 		param[1] = FW_PARAM_PFVF(STAG_END);
5486 		param[2] = FW_PARAM_PFVF(RQ_START);
5487 		param[3] = FW_PARAM_PFVF(RQ_END);
5488 		param[4] = FW_PARAM_PFVF(PBL_START);
5489 		param[5] = FW_PARAM_PFVF(PBL_END);
5490 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
5491 		if (rc != 0) {
5492 			device_printf(sc->dev,
5493 			    "failed to query RDMA parameters(1): %d.\n", rc);
5494 			return (rc);
5495 		}
5496 		sc->vres.stag.start = val[0];
5497 		sc->vres.stag.size = val[1] - val[0] + 1;
5498 		sc->vres.rq.start = val[2];
5499 		sc->vres.rq.size = val[3] - val[2] + 1;
5500 		sc->vres.pbl.start = val[4];
5501 		sc->vres.pbl.size = val[5] - val[4] + 1;
5502 
5503 		param[0] = FW_PARAM_PFVF(SQRQ_START);
5504 		param[1] = FW_PARAM_PFVF(SQRQ_END);
5505 		param[2] = FW_PARAM_PFVF(CQ_START);
5506 		param[3] = FW_PARAM_PFVF(CQ_END);
5507 		param[4] = FW_PARAM_PFVF(OCQ_START);
5508 		param[5] = FW_PARAM_PFVF(OCQ_END);
5509 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
5510 		if (rc != 0) {
5511 			device_printf(sc->dev,
5512 			    "failed to query RDMA parameters(2): %d.\n", rc);
5513 			return (rc);
5514 		}
5515 		sc->vres.qp.start = val[0];
5516 		sc->vres.qp.size = val[1] - val[0] + 1;
5517 		sc->vres.cq.start = val[2];
5518 		sc->vres.cq.size = val[3] - val[2] + 1;
5519 		sc->vres.ocq.start = val[4];
5520 		sc->vres.ocq.size = val[5] - val[4] + 1;
5521 
5522 		param[0] = FW_PARAM_PFVF(SRQ_START);
5523 		param[1] = FW_PARAM_PFVF(SRQ_END);
5524 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
5525 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
5526 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
5527 		if (rc != 0) {
5528 			device_printf(sc->dev,
5529 			    "failed to query RDMA parameters(3): %d.\n", rc);
5530 			return (rc);
5531 		}
5532 		sc->vres.srq.start = val[0];
5533 		sc->vres.srq.size = val[1] - val[0] + 1;
5534 		sc->params.max_ordird_qp = val[2];
5535 		sc->params.max_ird_adapter = val[3];
5536 	}
5537 	if (sc->iscsicaps) {
5538 		param[0] = FW_PARAM_PFVF(ISCSI_START);
5539 		param[1] = FW_PARAM_PFVF(ISCSI_END);
5540 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
5541 		if (rc != 0) {
5542 			device_printf(sc->dev,
5543 			    "failed to query iSCSI parameters: %d.\n", rc);
5544 			return (rc);
5545 		}
5546 		sc->vres.iscsi.start = val[0];
5547 		sc->vres.iscsi.size = val[1] - val[0] + 1;
5548 	}
5549 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
5550 		param[0] = FW_PARAM_PFVF(TLS_START);
5551 		param[1] = FW_PARAM_PFVF(TLS_END);
5552 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
5553 		if (rc != 0) {
5554 			device_printf(sc->dev,
5555 			    "failed to query TLS parameters: %d.\n", rc);
5556 			return (rc);
5557 		}
5558 		sc->vres.key.start = val[0];
5559 		sc->vres.key.size = val[1] - val[0] + 1;
5560 	}
5561 
5562 	/*
5563 	 * We've got the params we wanted to query directly from the firmware.
5564 	 * Grab some others via other means.
5565 	 */
5566 	t4_init_sge_params(sc);
5567 	t4_init_tp_params(sc);
5568 	t4_read_mtu_tbl(sc, sc->params.mtus, NULL);
5569 	t4_load_mtus(sc, sc->params.mtus, sc->params.a_wnd, sc->params.b_wnd);
5570 
5571 	rc = t4_verify_chip_settings(sc);
5572 	if (rc != 0)
5573 		return (rc);
5574 	t4_init_rx_buf_info(sc);
5575 
5576 	return (rc);
5577 }
5578 
5579 #ifdef KERN_TLS
5580 static void
5581 ktls_tick(void *arg)
5582 {
5583 	struct adapter *sc;
5584 	uint32_t tstamp;
5585 
5586 	sc = arg;
5587 	tstamp = tcp_ts_getticks();
5588 	t4_write_reg(sc, A_TP_SYNC_TIME_HI, tstamp >> 1);
5589 	t4_write_reg(sc, A_TP_SYNC_TIME_LO, tstamp << 31);
5590 	callout_schedule_sbt(&sc->ktls_tick, SBT_1MS, 0, C_HARDCLOCK);
5591 }
5592 
5593 static int
5594 t6_config_kern_tls(struct adapter *sc, bool enable)
5595 {
5596 	int rc;
5597 	uint32_t param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5598 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_KTLS_HW) |
5599 	    V_FW_PARAMS_PARAM_Y(enable ? 1 : 0) |
5600 	    V_FW_PARAMS_PARAM_Z(FW_PARAMS_PARAM_DEV_KTLS_HW_USER_ENABLE);
5601 
5602 	rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &param);
5603 	if (rc != 0) {
5604 		CH_ERR(sc, "failed to %s NIC TLS: %d\n",
5605 		    enable ?  "enable" : "disable", rc);
5606 		return (rc);
5607 	}
5608 
5609 	if (enable) {
5610 		sc->flags |= KERN_TLS_ON;
5611 		callout_reset_sbt(&sc->ktls_tick, SBT_1MS, 0, ktls_tick, sc,
5612 		    C_HARDCLOCK);
5613 	} else {
5614 		sc->flags &= ~KERN_TLS_ON;
5615 		callout_stop(&sc->ktls_tick);
5616 	}
5617 
5618 	return (rc);
5619 }
5620 #endif
5621 
5622 static int
5623 set_params__post_init(struct adapter *sc)
5624 {
5625 	uint32_t mask, param, val;
5626 #ifdef TCP_OFFLOAD
5627 	int i, v, shift;
5628 #endif
5629 
5630 	/* ask for encapsulated CPLs */
5631 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
5632 	val = 1;
5633 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
5634 
5635 	/* Enable 32b port caps if the firmware supports it. */
5636 	param = FW_PARAM_PFVF(PORT_CAPS32);
5637 	val = 1;
5638 	if (t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val) == 0)
5639 		sc->params.port_caps32 = 1;
5640 
5641 	/* Let filter + maskhash steer to a part of the VI's RSS region. */
5642 	val = 1 << (G_MASKSIZE(t4_read_reg(sc, A_TP_RSS_CONFIG_TNL)) - 1);
5643 	t4_set_reg_field(sc, A_TP_RSS_CONFIG_TNL, V_MASKFILTER(M_MASKFILTER),
5644 	    V_MASKFILTER(val - 1));
5645 
5646 	mask = F_DROPERRORANY | F_DROPERRORMAC | F_DROPERRORIPVER |
5647 	    F_DROPERRORFRAG | F_DROPERRORATTACK | F_DROPERRORETHHDRLEN |
5648 	    F_DROPERRORIPHDRLEN | F_DROPERRORTCPHDRLEN | F_DROPERRORPKTLEN |
5649 	    F_DROPERRORTCPOPT | F_DROPERRORCSUMIP | F_DROPERRORCSUM;
5650 	val = 0;
5651 	if (chip_id(sc) < CHELSIO_T6 && t4_attack_filter != 0) {
5652 		t4_set_reg_field(sc, A_TP_GLOBAL_CONFIG, F_ATTACKFILTERENABLE,
5653 		    F_ATTACKFILTERENABLE);
5654 		val |= F_DROPERRORATTACK;
5655 	}
5656 	if (t4_drop_ip_fragments != 0) {
5657 		t4_set_reg_field(sc, A_TP_GLOBAL_CONFIG, F_FRAGMENTDROP,
5658 		    F_FRAGMENTDROP);
5659 		val |= F_DROPERRORFRAG;
5660 	}
5661 	if (t4_drop_pkts_with_l2_errors != 0)
5662 		val |= F_DROPERRORMAC | F_DROPERRORETHHDRLEN;
5663 	if (t4_drop_pkts_with_l3_errors != 0) {
5664 		val |= F_DROPERRORIPVER | F_DROPERRORIPHDRLEN |
5665 		    F_DROPERRORCSUMIP;
5666 	}
5667 	if (t4_drop_pkts_with_l4_errors != 0) {
5668 		val |= F_DROPERRORTCPHDRLEN | F_DROPERRORPKTLEN |
5669 		    F_DROPERRORTCPOPT | F_DROPERRORCSUM;
5670 	}
5671 	t4_set_reg_field(sc, A_TP_ERR_CONFIG, mask, val);
5672 
5673 #ifdef TCP_OFFLOAD
5674 	/*
5675 	 * Override the TOE timers with user provided tunables.  This is not the
5676 	 * recommended way to change the timers (the firmware config file is) so
5677 	 * these tunables are not documented.
5678 	 *
5679 	 * All the timer tunables are in microseconds.
5680 	 */
5681 	if (t4_toe_keepalive_idle != 0) {
5682 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
5683 		v &= M_KEEPALIVEIDLE;
5684 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
5685 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
5686 	}
5687 	if (t4_toe_keepalive_interval != 0) {
5688 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
5689 		v &= M_KEEPALIVEINTVL;
5690 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
5691 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
5692 	}
5693 	if (t4_toe_keepalive_count != 0) {
5694 		v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
5695 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
5696 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
5697 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
5698 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
5699 	}
5700 	if (t4_toe_rexmt_min != 0) {
5701 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
5702 		v &= M_RXTMIN;
5703 		t4_set_reg_field(sc, A_TP_RXT_MIN,
5704 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
5705 	}
5706 	if (t4_toe_rexmt_max != 0) {
5707 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
5708 		v &= M_RXTMAX;
5709 		t4_set_reg_field(sc, A_TP_RXT_MAX,
5710 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
5711 	}
5712 	if (t4_toe_rexmt_count != 0) {
5713 		v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
5714 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
5715 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
5716 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
5717 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
5718 	}
5719 	for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
5720 		if (t4_toe_rexmt_backoff[i] != -1) {
5721 			v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
5722 			shift = (i & 3) << 3;
5723 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
5724 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
5725 		}
5726 	}
5727 #endif
5728 
5729 #ifdef KERN_TLS
5730 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS &&
5731 	    sc->toecaps & FW_CAPS_CONFIG_TOE) {
5732 		/*
5733 		 * Limit TOE connections to 2 reassembly "islands".
5734 		 * This is required to permit migrating TOE
5735 		 * connections to UPL_MODE_TLS.
5736 		 */
5737 		t4_tp_wr_bits_indirect(sc, A_TP_FRAG_CONFIG,
5738 		    V_PASSMODE(M_PASSMODE), V_PASSMODE(2));
5739 	}
5740 
5741 	if (is_ktls(sc)) {
5742 		sc->tlst.inline_keys = t4_tls_inline_keys;
5743 		sc->tlst.combo_wrs = t4_tls_combo_wrs;
5744 		if (t4_kern_tls != 0 && is_t6(sc))
5745 			t6_config_kern_tls(sc, true);
5746 	}
5747 #endif
5748 	return (0);
5749 }
5750 
5751 #undef FW_PARAM_PFVF
5752 #undef FW_PARAM_DEV
5753 
5754 static void
5755 t4_set_desc(struct adapter *sc)
5756 {
5757 	char buf[128];
5758 	struct adapter_params *p = &sc->params;
5759 
5760 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
5761 
5762 	device_set_desc_copy(sc->dev, buf);
5763 }
5764 
5765 static inline void
5766 ifmedia_add4(struct ifmedia *ifm, int m)
5767 {
5768 
5769 	ifmedia_add(ifm, m, 0, NULL);
5770 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE, 0, NULL);
5771 	ifmedia_add(ifm, m | IFM_ETH_RXPAUSE, 0, NULL);
5772 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE, 0, NULL);
5773 }
5774 
5775 /*
5776  * This is the selected media, which is not quite the same as the active media.
5777  * The media line in ifconfig is "media: Ethernet selected (active)" if selected
5778  * and active are not the same, and "media: Ethernet selected" otherwise.
5779  */
5780 static void
5781 set_current_media(struct port_info *pi)
5782 {
5783 	struct link_config *lc;
5784 	struct ifmedia *ifm;
5785 	int mword;
5786 	u_int speed;
5787 
5788 	PORT_LOCK_ASSERT_OWNED(pi);
5789 
5790 	/* Leave current media alone if it's already set to IFM_NONE. */
5791 	ifm = &pi->media;
5792 	if (ifm->ifm_cur != NULL &&
5793 	    IFM_SUBTYPE(ifm->ifm_cur->ifm_media) == IFM_NONE)
5794 		return;
5795 
5796 	lc = &pi->link_cfg;
5797 	if (lc->requested_aneg != AUTONEG_DISABLE &&
5798 	    lc->pcaps & FW_PORT_CAP32_ANEG) {
5799 		ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
5800 		return;
5801 	}
5802 	mword = IFM_ETHER | IFM_FDX;
5803 	if (lc->requested_fc & PAUSE_TX)
5804 		mword |= IFM_ETH_TXPAUSE;
5805 	if (lc->requested_fc & PAUSE_RX)
5806 		mword |= IFM_ETH_RXPAUSE;
5807 	if (lc->requested_speed == 0)
5808 		speed = port_top_speed(pi) * 1000;	/* Gbps -> Mbps */
5809 	else
5810 		speed = lc->requested_speed;
5811 	mword |= port_mword(pi, speed_to_fwcap(speed));
5812 	ifmedia_set(ifm, mword);
5813 }
5814 
5815 /*
5816  * Returns true if the ifmedia list for the port cannot change.
5817  */
5818 static bool
5819 fixed_ifmedia(struct port_info *pi)
5820 {
5821 
5822 	return (pi->port_type == FW_PORT_TYPE_BT_SGMII ||
5823 	    pi->port_type == FW_PORT_TYPE_BT_XFI ||
5824 	    pi->port_type == FW_PORT_TYPE_BT_XAUI ||
5825 	    pi->port_type == FW_PORT_TYPE_KX4 ||
5826 	    pi->port_type == FW_PORT_TYPE_KX ||
5827 	    pi->port_type == FW_PORT_TYPE_KR ||
5828 	    pi->port_type == FW_PORT_TYPE_BP_AP ||
5829 	    pi->port_type == FW_PORT_TYPE_BP4_AP ||
5830 	    pi->port_type == FW_PORT_TYPE_BP40_BA ||
5831 	    pi->port_type == FW_PORT_TYPE_KR4_100G ||
5832 	    pi->port_type == FW_PORT_TYPE_KR_SFP28 ||
5833 	    pi->port_type == FW_PORT_TYPE_KR_XLAUI);
5834 }
5835 
5836 static void
5837 build_medialist(struct port_info *pi)
5838 {
5839 	uint32_t ss, speed;
5840 	int unknown, mword, bit;
5841 	struct link_config *lc;
5842 	struct ifmedia *ifm;
5843 
5844 	PORT_LOCK_ASSERT_OWNED(pi);
5845 
5846 	if (pi->flags & FIXED_IFMEDIA)
5847 		return;
5848 
5849 	/*
5850 	 * Rebuild the ifmedia list.
5851 	 */
5852 	ifm = &pi->media;
5853 	ifmedia_removeall(ifm);
5854 	lc = &pi->link_cfg;
5855 	ss = G_FW_PORT_CAP32_SPEED(lc->pcaps); /* Supported Speeds */
5856 	if (__predict_false(ss == 0)) {	/* not supposed to happen. */
5857 		MPASS(ss != 0);
5858 no_media:
5859 		MPASS(LIST_EMPTY(&ifm->ifm_list));
5860 		ifmedia_add(ifm, IFM_ETHER | IFM_NONE, 0, NULL);
5861 		ifmedia_set(ifm, IFM_ETHER | IFM_NONE);
5862 		return;
5863 	}
5864 
5865 	unknown = 0;
5866 	for (bit = S_FW_PORT_CAP32_SPEED; bit < fls(ss); bit++) {
5867 		speed = 1 << bit;
5868 		MPASS(speed & M_FW_PORT_CAP32_SPEED);
5869 		if (ss & speed) {
5870 			mword = port_mword(pi, speed);
5871 			if (mword == IFM_NONE) {
5872 				goto no_media;
5873 			} else if (mword == IFM_UNKNOWN)
5874 				unknown++;
5875 			else
5876 				ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | mword);
5877 		}
5878 	}
5879 	if (unknown > 0) /* Add one unknown for all unknown media types. */
5880 		ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | IFM_UNKNOWN);
5881 	if (lc->pcaps & FW_PORT_CAP32_ANEG)
5882 		ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
5883 
5884 	set_current_media(pi);
5885 }
5886 
5887 /*
5888  * Initialize the requested fields in the link config based on driver tunables.
5889  */
5890 static void
5891 init_link_config(struct port_info *pi)
5892 {
5893 	struct link_config *lc = &pi->link_cfg;
5894 
5895 	PORT_LOCK_ASSERT_OWNED(pi);
5896 
5897 	lc->requested_caps = 0;
5898 	lc->requested_speed = 0;
5899 
5900 	if (t4_autoneg == 0)
5901 		lc->requested_aneg = AUTONEG_DISABLE;
5902 	else if (t4_autoneg == 1)
5903 		lc->requested_aneg = AUTONEG_ENABLE;
5904 	else
5905 		lc->requested_aneg = AUTONEG_AUTO;
5906 
5907 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX |
5908 	    PAUSE_AUTONEG);
5909 
5910 	if (t4_fec & FEC_AUTO)
5911 		lc->requested_fec = FEC_AUTO;
5912 	else if (t4_fec == 0)
5913 		lc->requested_fec = FEC_NONE;
5914 	else {
5915 		/* -1 is handled by the FEC_AUTO block above and not here. */
5916 		lc->requested_fec = t4_fec &
5917 		    (FEC_RS | FEC_BASER_RS | FEC_NONE | FEC_MODULE);
5918 		if (lc->requested_fec == 0)
5919 			lc->requested_fec = FEC_AUTO;
5920 	}
5921 	if (t4_force_fec < 0)
5922 		lc->force_fec = -1;
5923 	else if (t4_force_fec > 0)
5924 		lc->force_fec = 1;
5925 	else
5926 		lc->force_fec = 0;
5927 }
5928 
5929 /*
5930  * Makes sure that all requested settings comply with what's supported by the
5931  * port.  Returns the number of settings that were invalid and had to be fixed.
5932  */
5933 static int
5934 fixup_link_config(struct port_info *pi)
5935 {
5936 	int n = 0;
5937 	struct link_config *lc = &pi->link_cfg;
5938 	uint32_t fwspeed;
5939 
5940 	PORT_LOCK_ASSERT_OWNED(pi);
5941 
5942 	/* Speed (when not autonegotiating) */
5943 	if (lc->requested_speed != 0) {
5944 		fwspeed = speed_to_fwcap(lc->requested_speed);
5945 		if ((fwspeed & lc->pcaps) == 0) {
5946 			n++;
5947 			lc->requested_speed = 0;
5948 		}
5949 	}
5950 
5951 	/* Link autonegotiation */
5952 	MPASS(lc->requested_aneg == AUTONEG_ENABLE ||
5953 	    lc->requested_aneg == AUTONEG_DISABLE ||
5954 	    lc->requested_aneg == AUTONEG_AUTO);
5955 	if (lc->requested_aneg == AUTONEG_ENABLE &&
5956 	    !(lc->pcaps & FW_PORT_CAP32_ANEG)) {
5957 		n++;
5958 		lc->requested_aneg = AUTONEG_AUTO;
5959 	}
5960 
5961 	/* Flow control */
5962 	MPASS((lc->requested_fc & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG)) == 0);
5963 	if (lc->requested_fc & PAUSE_TX &&
5964 	    !(lc->pcaps & FW_PORT_CAP32_FC_TX)) {
5965 		n++;
5966 		lc->requested_fc &= ~PAUSE_TX;
5967 	}
5968 	if (lc->requested_fc & PAUSE_RX &&
5969 	    !(lc->pcaps & FW_PORT_CAP32_FC_RX)) {
5970 		n++;
5971 		lc->requested_fc &= ~PAUSE_RX;
5972 	}
5973 	if (!(lc->requested_fc & PAUSE_AUTONEG) &&
5974 	    !(lc->pcaps & FW_PORT_CAP32_FORCE_PAUSE)) {
5975 		n++;
5976 		lc->requested_fc |= PAUSE_AUTONEG;
5977 	}
5978 
5979 	/* FEC */
5980 	if ((lc->requested_fec & FEC_RS &&
5981 	    !(lc->pcaps & FW_PORT_CAP32_FEC_RS)) ||
5982 	    (lc->requested_fec & FEC_BASER_RS &&
5983 	    !(lc->pcaps & FW_PORT_CAP32_FEC_BASER_RS))) {
5984 		n++;
5985 		lc->requested_fec = FEC_AUTO;
5986 	}
5987 
5988 	return (n);
5989 }
5990 
5991 /*
5992  * Apply the requested L1 settings, which are expected to be valid, to the
5993  * hardware.
5994  */
5995 static int
5996 apply_link_config(struct port_info *pi)
5997 {
5998 	struct adapter *sc = pi->adapter;
5999 	struct link_config *lc = &pi->link_cfg;
6000 	int rc;
6001 
6002 #ifdef INVARIANTS
6003 	ASSERT_SYNCHRONIZED_OP(sc);
6004 	PORT_LOCK_ASSERT_OWNED(pi);
6005 
6006 	if (lc->requested_aneg == AUTONEG_ENABLE)
6007 		MPASS(lc->pcaps & FW_PORT_CAP32_ANEG);
6008 	if (!(lc->requested_fc & PAUSE_AUTONEG))
6009 		MPASS(lc->pcaps & FW_PORT_CAP32_FORCE_PAUSE);
6010 	if (lc->requested_fc & PAUSE_TX)
6011 		MPASS(lc->pcaps & FW_PORT_CAP32_FC_TX);
6012 	if (lc->requested_fc & PAUSE_RX)
6013 		MPASS(lc->pcaps & FW_PORT_CAP32_FC_RX);
6014 	if (lc->requested_fec & FEC_RS)
6015 		MPASS(lc->pcaps & FW_PORT_CAP32_FEC_RS);
6016 	if (lc->requested_fec & FEC_BASER_RS)
6017 		MPASS(lc->pcaps & FW_PORT_CAP32_FEC_BASER_RS);
6018 #endif
6019 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6020 	if (rc != 0) {
6021 		/* Don't complain if the VF driver gets back an EPERM. */
6022 		if (!(sc->flags & IS_VF) || rc != FW_EPERM)
6023 			device_printf(pi->dev, "l1cfg failed: %d\n", rc);
6024 	} else {
6025 		/*
6026 		 * An L1_CFG will almost always result in a link-change event if
6027 		 * the link is up, and the driver will refresh the actual
6028 		 * fec/fc/etc. when the notification is processed.  If the link
6029 		 * is down then the actual settings are meaningless.
6030 		 *
6031 		 * This takes care of the case where a change in the L1 settings
6032 		 * may not result in a notification.
6033 		 */
6034 		if (lc->link_ok && !(lc->requested_fc & PAUSE_AUTONEG))
6035 			lc->fc = lc->requested_fc & (PAUSE_TX | PAUSE_RX);
6036 	}
6037 	return (rc);
6038 }
6039 
6040 #define FW_MAC_EXACT_CHUNK	7
6041 struct mcaddr_ctx {
6042 	if_t ifp;
6043 	const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
6044 	uint64_t hash;
6045 	int i;
6046 	int del;
6047 	int rc;
6048 };
6049 
6050 static u_int
6051 add_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
6052 {
6053 	struct mcaddr_ctx *ctx = arg;
6054 	struct vi_info *vi = if_getsoftc(ctx->ifp);
6055 	struct port_info *pi = vi->pi;
6056 	struct adapter *sc = pi->adapter;
6057 
6058 	if (ctx->rc < 0)
6059 		return (0);
6060 
6061 	ctx->mcaddr[ctx->i] = LLADDR(sdl);
6062 	MPASS(ETHER_IS_MULTICAST(ctx->mcaddr[ctx->i]));
6063 	ctx->i++;
6064 
6065 	if (ctx->i == FW_MAC_EXACT_CHUNK) {
6066 		ctx->rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, ctx->del,
6067 		    ctx->i, ctx->mcaddr, NULL, &ctx->hash, 0);
6068 		if (ctx->rc < 0) {
6069 			int j;
6070 
6071 			for (j = 0; j < ctx->i; j++) {
6072 				if_printf(ctx->ifp,
6073 				    "failed to add mc address"
6074 				    " %02x:%02x:%02x:"
6075 				    "%02x:%02x:%02x rc=%d\n",
6076 				    ctx->mcaddr[j][0], ctx->mcaddr[j][1],
6077 				    ctx->mcaddr[j][2], ctx->mcaddr[j][3],
6078 				    ctx->mcaddr[j][4], ctx->mcaddr[j][5],
6079 				    -ctx->rc);
6080 			}
6081 			return (0);
6082 		}
6083 		ctx->del = 0;
6084 		ctx->i = 0;
6085 	}
6086 
6087 	return (1);
6088 }
6089 
6090 /*
6091  * Program the port's XGMAC based on parameters in ifnet.  The caller also
6092  * indicates which parameters should be programmed (the rest are left alone).
6093  */
6094 int
6095 update_mac_settings(if_t ifp, int flags)
6096 {
6097 	int rc = 0;
6098 	struct vi_info *vi = if_getsoftc(ifp);
6099 	struct port_info *pi = vi->pi;
6100 	struct adapter *sc = pi->adapter;
6101 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
6102 	uint8_t match_all_mac[ETHER_ADDR_LEN] = {0};
6103 
6104 	ASSERT_SYNCHRONIZED_OP(sc);
6105 	KASSERT(flags, ("%s: not told what to update.", __func__));
6106 
6107 	if (flags & XGMAC_MTU)
6108 		mtu = if_getmtu(ifp);
6109 
6110 	if (flags & XGMAC_PROMISC)
6111 		promisc = if_getflags(ifp) & IFF_PROMISC ? 1 : 0;
6112 
6113 	if (flags & XGMAC_ALLMULTI)
6114 		allmulti = if_getflags(ifp) & IFF_ALLMULTI ? 1 : 0;
6115 
6116 	if (flags & XGMAC_VLANEX)
6117 		vlanex = if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING ? 1 : 0;
6118 
6119 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
6120 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
6121 		    allmulti, 1, vlanex, false);
6122 		if (rc) {
6123 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
6124 			    rc);
6125 			return (rc);
6126 		}
6127 	}
6128 
6129 	if (flags & XGMAC_UCADDR) {
6130 		uint8_t ucaddr[ETHER_ADDR_LEN];
6131 
6132 		bcopy(if_getlladdr(ifp), ucaddr, sizeof(ucaddr));
6133 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
6134 		    ucaddr, true, &vi->smt_idx);
6135 		if (rc < 0) {
6136 			rc = -rc;
6137 			if_printf(ifp, "change_mac failed: %d\n", rc);
6138 			return (rc);
6139 		} else {
6140 			vi->xact_addr_filt = rc;
6141 			rc = 0;
6142 		}
6143 	}
6144 
6145 	if (flags & XGMAC_MCADDRS) {
6146 		struct epoch_tracker et;
6147 		struct mcaddr_ctx ctx;
6148 		int j;
6149 
6150 		ctx.ifp = ifp;
6151 		ctx.hash = 0;
6152 		ctx.i = 0;
6153 		ctx.del = 1;
6154 		ctx.rc = 0;
6155 		/*
6156 		 * Unlike other drivers, we accumulate list of pointers into
6157 		 * interface address lists and we need to keep it safe even
6158 		 * after if_foreach_llmaddr() returns, thus we must enter the
6159 		 * network epoch.
6160 		 */
6161 		NET_EPOCH_ENTER(et);
6162 		if_foreach_llmaddr(ifp, add_maddr, &ctx);
6163 		if (ctx.rc < 0) {
6164 			NET_EPOCH_EXIT(et);
6165 			rc = -ctx.rc;
6166 			return (rc);
6167 		}
6168 		if (ctx.i > 0) {
6169 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
6170 			    ctx.del, ctx.i, ctx.mcaddr, NULL, &ctx.hash, 0);
6171 			NET_EPOCH_EXIT(et);
6172 			if (rc < 0) {
6173 				rc = -rc;
6174 				for (j = 0; j < ctx.i; j++) {
6175 					if_printf(ifp,
6176 					    "failed to add mcast address"
6177 					    " %02x:%02x:%02x:"
6178 					    "%02x:%02x:%02x rc=%d\n",
6179 					    ctx.mcaddr[j][0], ctx.mcaddr[j][1],
6180 					    ctx.mcaddr[j][2], ctx.mcaddr[j][3],
6181 					    ctx.mcaddr[j][4], ctx.mcaddr[j][5],
6182 					    rc);
6183 				}
6184 				return (rc);
6185 			}
6186 			ctx.del = 0;
6187 		} else
6188 			NET_EPOCH_EXIT(et);
6189 
6190 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, ctx.hash, 0);
6191 		if (rc != 0)
6192 			if_printf(ifp, "failed to set mcast address hash: %d\n",
6193 			    rc);
6194 		if (ctx.del == 0) {
6195 			/* We clobbered the VXLAN entry if there was one. */
6196 			pi->vxlan_tcam_entry = false;
6197 		}
6198 	}
6199 
6200 	if (IS_MAIN_VI(vi) && sc->vxlan_refcount > 0 &&
6201 	    pi->vxlan_tcam_entry == false) {
6202 		rc = t4_alloc_raw_mac_filt(sc, vi->viid, match_all_mac,
6203 		    match_all_mac, sc->rawf_base + pi->port_id, 1, pi->port_id,
6204 		    true);
6205 		if (rc < 0) {
6206 			rc = -rc;
6207 			if_printf(ifp, "failed to add VXLAN TCAM entry: %d.\n",
6208 			    rc);
6209 		} else {
6210 			MPASS(rc == sc->rawf_base + pi->port_id);
6211 			rc = 0;
6212 			pi->vxlan_tcam_entry = true;
6213 		}
6214 	}
6215 
6216 	return (rc);
6217 }
6218 
6219 /*
6220  * {begin|end}_synchronized_op must be called from the same thread.
6221  */
6222 int
6223 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
6224     char *wmesg)
6225 {
6226 	int rc, pri;
6227 
6228 #ifdef WITNESS
6229 	/* the caller thinks it's ok to sleep, but is it really? */
6230 	if (flags & SLEEP_OK)
6231 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
6232 		    "begin_synchronized_op");
6233 #endif
6234 
6235 	if (INTR_OK)
6236 		pri = PCATCH;
6237 	else
6238 		pri = 0;
6239 
6240 	ADAPTER_LOCK(sc);
6241 	for (;;) {
6242 
6243 		if (vi && IS_DETACHING(vi)) {
6244 			rc = ENXIO;
6245 			goto done;
6246 		}
6247 
6248 		if (!IS_BUSY(sc)) {
6249 			rc = 0;
6250 			break;
6251 		}
6252 
6253 		if (!(flags & SLEEP_OK)) {
6254 			rc = EBUSY;
6255 			goto done;
6256 		}
6257 
6258 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
6259 			rc = EINTR;
6260 			goto done;
6261 		}
6262 	}
6263 
6264 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
6265 	SET_BUSY(sc);
6266 #ifdef INVARIANTS
6267 	sc->last_op = wmesg;
6268 	sc->last_op_thr = curthread;
6269 	sc->last_op_flags = flags;
6270 #endif
6271 
6272 done:
6273 	if (!(flags & HOLD_LOCK) || rc)
6274 		ADAPTER_UNLOCK(sc);
6275 
6276 	return (rc);
6277 }
6278 
6279 /*
6280  * Tell if_ioctl and if_init that the VI is going away.  This is
6281  * special variant of begin_synchronized_op and must be paired with a
6282  * call to end_vi_detach.
6283  */
6284 void
6285 begin_vi_detach(struct adapter *sc, struct vi_info *vi)
6286 {
6287 	ADAPTER_LOCK(sc);
6288 	SET_DETACHING(vi);
6289 	wakeup(&sc->flags);
6290 	while (IS_BUSY(sc))
6291 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
6292 	SET_BUSY(sc);
6293 #ifdef INVARIANTS
6294 	sc->last_op = "t4detach";
6295 	sc->last_op_thr = curthread;
6296 	sc->last_op_flags = 0;
6297 #endif
6298 	ADAPTER_UNLOCK(sc);
6299 }
6300 
6301 void
6302 end_vi_detach(struct adapter *sc, struct vi_info *vi)
6303 {
6304 	ADAPTER_LOCK(sc);
6305 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
6306 	CLR_BUSY(sc);
6307 	CLR_DETACHING(vi);
6308 	wakeup(&sc->flags);
6309 	ADAPTER_UNLOCK(sc);
6310 }
6311 
6312 /*
6313  * {begin|end}_synchronized_op must be called from the same thread.
6314  */
6315 void
6316 end_synchronized_op(struct adapter *sc, int flags)
6317 {
6318 
6319 	if (flags & LOCK_HELD)
6320 		ADAPTER_LOCK_ASSERT_OWNED(sc);
6321 	else
6322 		ADAPTER_LOCK(sc);
6323 
6324 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
6325 	CLR_BUSY(sc);
6326 	wakeup(&sc->flags);
6327 	ADAPTER_UNLOCK(sc);
6328 }
6329 
6330 static int
6331 cxgbe_init_synchronized(struct vi_info *vi)
6332 {
6333 	struct port_info *pi = vi->pi;
6334 	struct adapter *sc = pi->adapter;
6335 	if_t ifp = vi->ifp;
6336 	int rc = 0, i;
6337 	struct sge_txq *txq;
6338 
6339 	ASSERT_SYNCHRONIZED_OP(sc);
6340 
6341 	if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
6342 		return (0);	/* already running */
6343 
6344 	if (!(sc->flags & FULL_INIT_DONE) && ((rc = adapter_init(sc)) != 0))
6345 		return (rc);	/* error message displayed already */
6346 
6347 	if (!(vi->flags & VI_INIT_DONE) && ((rc = vi_init(vi)) != 0))
6348 		return (rc); /* error message displayed already */
6349 
6350 	rc = update_mac_settings(ifp, XGMAC_ALL);
6351 	if (rc)
6352 		goto done;	/* error message displayed already */
6353 
6354 	PORT_LOCK(pi);
6355 	if (pi->up_vis == 0) {
6356 		t4_update_port_info(pi);
6357 		fixup_link_config(pi);
6358 		build_medialist(pi);
6359 		apply_link_config(pi);
6360 	}
6361 
6362 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
6363 	if (rc != 0) {
6364 		if_printf(ifp, "enable_vi failed: %d\n", rc);
6365 		PORT_UNLOCK(pi);
6366 		goto done;
6367 	}
6368 
6369 	/*
6370 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
6371 	 * if this changes.
6372 	 */
6373 
6374 	for_each_txq(vi, i, txq) {
6375 		TXQ_LOCK(txq);
6376 		txq->eq.flags |= EQ_ENABLED;
6377 		TXQ_UNLOCK(txq);
6378 	}
6379 
6380 	/*
6381 	 * The first iq of the first port to come up is used for tracing.
6382 	 */
6383 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
6384 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
6385 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
6386 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
6387 		    V_QUEUENUMBER(sc->traceq));
6388 		pi->flags |= HAS_TRACEQ;
6389 	}
6390 
6391 	/* all ok */
6392 	pi->up_vis++;
6393 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
6394 	if (pi->link_cfg.link_ok)
6395 		t4_os_link_changed(pi);
6396 	PORT_UNLOCK(pi);
6397 
6398 	mtx_lock(&vi->tick_mtx);
6399 	if (vi->pi->nvi > 1 || sc->flags & IS_VF)
6400 		callout_reset(&vi->tick, hz, vi_tick, vi);
6401 	else
6402 		callout_reset(&vi->tick, hz, cxgbe_tick, vi);
6403 	mtx_unlock(&vi->tick_mtx);
6404 done:
6405 	if (rc != 0)
6406 		cxgbe_uninit_synchronized(vi);
6407 
6408 	return (rc);
6409 }
6410 
6411 /*
6412  * Idempotent.
6413  */
6414 static int
6415 cxgbe_uninit_synchronized(struct vi_info *vi)
6416 {
6417 	struct port_info *pi = vi->pi;
6418 	struct adapter *sc = pi->adapter;
6419 	if_t ifp = vi->ifp;
6420 	int rc, i;
6421 	struct sge_txq *txq;
6422 
6423 	ASSERT_SYNCHRONIZED_OP(sc);
6424 
6425 	if (!(vi->flags & VI_INIT_DONE)) {
6426 		if (__predict_false(if_getdrvflags(ifp) & IFF_DRV_RUNNING)) {
6427 			KASSERT(0, ("uninited VI is running"));
6428 			if_printf(ifp, "uninited VI with running ifnet.  "
6429 			    "vi->flags 0x%016lx, if_flags 0x%08x, "
6430 			    "if_drv_flags 0x%08x\n", vi->flags, if_getflags(ifp),
6431 			    if_getdrvflags(ifp));
6432 		}
6433 		return (0);
6434 	}
6435 
6436 	/*
6437 	 * Disable the VI so that all its data in either direction is discarded
6438 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
6439 	 * tick) intact as the TP can deliver negative advice or data that it's
6440 	 * holding in its RAM (for an offloaded connection) even after the VI is
6441 	 * disabled.
6442 	 */
6443 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
6444 	if (rc) {
6445 		if_printf(ifp, "disable_vi failed: %d\n", rc);
6446 		return (rc);
6447 	}
6448 
6449 	for_each_txq(vi, i, txq) {
6450 		TXQ_LOCK(txq);
6451 		txq->eq.flags &= ~EQ_ENABLED;
6452 		TXQ_UNLOCK(txq);
6453 	}
6454 
6455 	mtx_lock(&vi->tick_mtx);
6456 	callout_stop(&vi->tick);
6457 	mtx_unlock(&vi->tick_mtx);
6458 
6459 	PORT_LOCK(pi);
6460 	if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING)) {
6461 		PORT_UNLOCK(pi);
6462 		return (0);
6463 	}
6464 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
6465 	pi->up_vis--;
6466 	if (pi->up_vis > 0) {
6467 		PORT_UNLOCK(pi);
6468 		return (0);
6469 	}
6470 
6471 	pi->link_cfg.link_ok = false;
6472 	pi->link_cfg.speed = 0;
6473 	pi->link_cfg.link_down_rc = 255;
6474 	t4_os_link_changed(pi);
6475 	PORT_UNLOCK(pi);
6476 
6477 	return (0);
6478 }
6479 
6480 /*
6481  * It is ok for this function to fail midway and return right away.  t4_detach
6482  * will walk the entire sc->irq list and clean up whatever is valid.
6483  */
6484 int
6485 t4_setup_intr_handlers(struct adapter *sc)
6486 {
6487 	int rc, rid, p, q, v;
6488 	char s[8];
6489 	struct irq *irq;
6490 	struct port_info *pi;
6491 	struct vi_info *vi;
6492 	struct sge *sge = &sc->sge;
6493 	struct sge_rxq *rxq;
6494 #ifdef TCP_OFFLOAD
6495 	struct sge_ofld_rxq *ofld_rxq;
6496 #endif
6497 #ifdef DEV_NETMAP
6498 	struct sge_nm_rxq *nm_rxq;
6499 #endif
6500 #ifdef RSS
6501 	int nbuckets = rss_getnumbuckets();
6502 #endif
6503 
6504 	/*
6505 	 * Setup interrupts.
6506 	 */
6507 	irq = &sc->irq[0];
6508 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
6509 	if (forwarding_intr_to_fwq(sc))
6510 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
6511 
6512 	/* Multiple interrupts. */
6513 	if (sc->flags & IS_VF)
6514 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
6515 		    ("%s: too few intr.", __func__));
6516 	else
6517 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
6518 		    ("%s: too few intr.", __func__));
6519 
6520 	/* The first one is always error intr on PFs */
6521 	if (!(sc->flags & IS_VF)) {
6522 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
6523 		if (rc != 0)
6524 			return (rc);
6525 		irq++;
6526 		rid++;
6527 	}
6528 
6529 	/* The second one is always the firmware event queue (first on VFs) */
6530 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
6531 	if (rc != 0)
6532 		return (rc);
6533 	irq++;
6534 	rid++;
6535 
6536 	for_each_port(sc, p) {
6537 		pi = sc->port[p];
6538 		for_each_vi(pi, v, vi) {
6539 			vi->first_intr = rid - 1;
6540 
6541 			if (vi->nnmrxq > 0) {
6542 				int n = max(vi->nrxq, vi->nnmrxq);
6543 
6544 				rxq = &sge->rxq[vi->first_rxq];
6545 #ifdef DEV_NETMAP
6546 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
6547 #endif
6548 				for (q = 0; q < n; q++) {
6549 					snprintf(s, sizeof(s), "%x%c%x", p,
6550 					    'a' + v, q);
6551 					if (q < vi->nrxq)
6552 						irq->rxq = rxq++;
6553 #ifdef DEV_NETMAP
6554 					if (q < vi->nnmrxq)
6555 						irq->nm_rxq = nm_rxq++;
6556 
6557 					if (irq->nm_rxq != NULL &&
6558 					    irq->rxq == NULL) {
6559 						/* Netmap rx only */
6560 						rc = t4_alloc_irq(sc, irq, rid,
6561 						    t4_nm_intr, irq->nm_rxq, s);
6562 					}
6563 					if (irq->nm_rxq != NULL &&
6564 					    irq->rxq != NULL) {
6565 						/* NIC and Netmap rx */
6566 						rc = t4_alloc_irq(sc, irq, rid,
6567 						    t4_vi_intr, irq, s);
6568 					}
6569 #endif
6570 					if (irq->rxq != NULL &&
6571 					    irq->nm_rxq == NULL) {
6572 						/* NIC rx only */
6573 						rc = t4_alloc_irq(sc, irq, rid,
6574 						    t4_intr, irq->rxq, s);
6575 					}
6576 					if (rc != 0)
6577 						return (rc);
6578 #ifdef RSS
6579 					if (q < vi->nrxq) {
6580 						bus_bind_intr(sc->dev, irq->res,
6581 						    rss_getcpu(q % nbuckets));
6582 					}
6583 #endif
6584 					irq++;
6585 					rid++;
6586 					vi->nintr++;
6587 				}
6588 			} else {
6589 				for_each_rxq(vi, q, rxq) {
6590 					snprintf(s, sizeof(s), "%x%c%x", p,
6591 					    'a' + v, q);
6592 					rc = t4_alloc_irq(sc, irq, rid,
6593 					    t4_intr, rxq, s);
6594 					if (rc != 0)
6595 						return (rc);
6596 #ifdef RSS
6597 					bus_bind_intr(sc->dev, irq->res,
6598 					    rss_getcpu(q % nbuckets));
6599 #endif
6600 					irq++;
6601 					rid++;
6602 					vi->nintr++;
6603 				}
6604 			}
6605 #ifdef TCP_OFFLOAD
6606 			for_each_ofld_rxq(vi, q, ofld_rxq) {
6607 				snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
6608 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
6609 				    ofld_rxq, s);
6610 				if (rc != 0)
6611 					return (rc);
6612 				irq++;
6613 				rid++;
6614 				vi->nintr++;
6615 			}
6616 #endif
6617 		}
6618 	}
6619 	MPASS(irq == &sc->irq[sc->intr_count]);
6620 
6621 	return (0);
6622 }
6623 
6624 static void
6625 write_global_rss_key(struct adapter *sc)
6626 {
6627 #ifdef RSS
6628 	int i;
6629 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
6630 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
6631 
6632 	CTASSERT(RSS_KEYSIZE == 40);
6633 
6634 	rss_getkey((void *)&raw_rss_key[0]);
6635 	for (i = 0; i < nitems(rss_key); i++) {
6636 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
6637 	}
6638 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
6639 #endif
6640 }
6641 
6642 /*
6643  * Idempotent.
6644  */
6645 static int
6646 adapter_full_init(struct adapter *sc)
6647 {
6648 	int rc, i;
6649 
6650 	ASSERT_SYNCHRONIZED_OP(sc);
6651 
6652 	/*
6653 	 * queues that belong to the adapter (not any particular port).
6654 	 */
6655 	rc = t4_setup_adapter_queues(sc);
6656 	if (rc != 0)
6657 		return (rc);
6658 
6659 	for (i = 0; i < nitems(sc->tq); i++) {
6660 		if (sc->tq[i] != NULL)
6661 			continue;
6662 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
6663 		    taskqueue_thread_enqueue, &sc->tq[i]);
6664 		if (sc->tq[i] == NULL) {
6665 			CH_ERR(sc, "failed to allocate task queue %d\n", i);
6666 			return (ENOMEM);
6667 		}
6668 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
6669 		    device_get_nameunit(sc->dev), i);
6670 	}
6671 
6672 	if (!(sc->flags & IS_VF)) {
6673 		write_global_rss_key(sc);
6674 		t4_intr_enable(sc);
6675 	}
6676 	return (0);
6677 }
6678 
6679 int
6680 adapter_init(struct adapter *sc)
6681 {
6682 	int rc;
6683 
6684 	ASSERT_SYNCHRONIZED_OP(sc);
6685 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
6686 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
6687 	    ("%s: FULL_INIT_DONE already", __func__));
6688 
6689 	rc = adapter_full_init(sc);
6690 	if (rc != 0)
6691 		adapter_full_uninit(sc);
6692 	else
6693 		sc->flags |= FULL_INIT_DONE;
6694 
6695 	return (rc);
6696 }
6697 
6698 /*
6699  * Idempotent.
6700  */
6701 static void
6702 adapter_full_uninit(struct adapter *sc)
6703 {
6704 	int i;
6705 
6706 	t4_teardown_adapter_queues(sc);
6707 
6708 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
6709 		taskqueue_free(sc->tq[i]);
6710 		sc->tq[i] = NULL;
6711 	}
6712 
6713 	sc->flags &= ~FULL_INIT_DONE;
6714 }
6715 
6716 #ifdef RSS
6717 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
6718     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
6719     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
6720     RSS_HASHTYPE_RSS_UDP_IPV6)
6721 
6722 /* Translates kernel hash types to hardware. */
6723 static int
6724 hashconfig_to_hashen(int hashconfig)
6725 {
6726 	int hashen = 0;
6727 
6728 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
6729 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
6730 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
6731 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
6732 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
6733 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
6734 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
6735 	}
6736 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
6737 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
6738 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
6739 	}
6740 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
6741 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
6742 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
6743 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
6744 
6745 	return (hashen);
6746 }
6747 
6748 /* Translates hardware hash types to kernel. */
6749 static int
6750 hashen_to_hashconfig(int hashen)
6751 {
6752 	int hashconfig = 0;
6753 
6754 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
6755 		/*
6756 		 * If UDP hashing was enabled it must have been enabled for
6757 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
6758 		 * enabling any 4-tuple hash is nonsense configuration.
6759 		 */
6760 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
6761 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
6762 
6763 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
6764 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
6765 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
6766 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
6767 	}
6768 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
6769 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
6770 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
6771 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
6772 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
6773 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
6774 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
6775 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
6776 
6777 	return (hashconfig);
6778 }
6779 #endif
6780 
6781 /*
6782  * Idempotent.
6783  */
6784 static int
6785 vi_full_init(struct vi_info *vi)
6786 {
6787 	struct adapter *sc = vi->adapter;
6788 	struct sge_rxq *rxq;
6789 	int rc, i, j;
6790 #ifdef RSS
6791 	int nbuckets = rss_getnumbuckets();
6792 	int hashconfig = rss_gethashconfig();
6793 	int extra;
6794 #endif
6795 
6796 	ASSERT_SYNCHRONIZED_OP(sc);
6797 
6798 	/*
6799 	 * Allocate tx/rx/fl queues for this VI.
6800 	 */
6801 	rc = t4_setup_vi_queues(vi);
6802 	if (rc != 0)
6803 		return (rc);
6804 
6805 	/*
6806 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
6807 	 */
6808 	if (vi->nrxq > vi->rss_size) {
6809 		CH_ALERT(vi, "nrxq (%d) > hw RSS table size (%d); "
6810 		    "some queues will never receive traffic.\n", vi->nrxq,
6811 		    vi->rss_size);
6812 	} else if (vi->rss_size % vi->nrxq) {
6813 		CH_ALERT(vi, "nrxq (%d), hw RSS table size (%d); "
6814 		    "expect uneven traffic distribution.\n", vi->nrxq,
6815 		    vi->rss_size);
6816 	}
6817 #ifdef RSS
6818 	if (vi->nrxq != nbuckets) {
6819 		CH_ALERT(vi, "nrxq (%d) != kernel RSS buckets (%d);"
6820 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
6821 	}
6822 #endif
6823 	if (vi->rss == NULL)
6824 		vi->rss = malloc(vi->rss_size * sizeof (*vi->rss), M_CXGBE,
6825 		    M_ZERO | M_WAITOK);
6826 	for (i = 0; i < vi->rss_size;) {
6827 #ifdef RSS
6828 		j = rss_get_indirection_to_bucket(i);
6829 		j %= vi->nrxq;
6830 		rxq = &sc->sge.rxq[vi->first_rxq + j];
6831 		vi->rss[i++] = rxq->iq.abs_id;
6832 #else
6833 		for_each_rxq(vi, j, rxq) {
6834 			vi->rss[i++] = rxq->iq.abs_id;
6835 			if (i == vi->rss_size)
6836 				break;
6837 		}
6838 #endif
6839 	}
6840 
6841 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size,
6842 	    vi->rss, vi->rss_size);
6843 	if (rc != 0) {
6844 		CH_ERR(vi, "rss_config failed: %d\n", rc);
6845 		return (rc);
6846 	}
6847 
6848 #ifdef RSS
6849 	vi->hashen = hashconfig_to_hashen(hashconfig);
6850 
6851 	/*
6852 	 * We may have had to enable some hashes even though the global config
6853 	 * wants them disabled.  This is a potential problem that must be
6854 	 * reported to the user.
6855 	 */
6856 	extra = hashen_to_hashconfig(vi->hashen) ^ hashconfig;
6857 
6858 	/*
6859 	 * If we consider only the supported hash types, then the enabled hashes
6860 	 * are a superset of the requested hashes.  In other words, there cannot
6861 	 * be any supported hash that was requested but not enabled, but there
6862 	 * can be hashes that were not requested but had to be enabled.
6863 	 */
6864 	extra &= SUPPORTED_RSS_HASHTYPES;
6865 	MPASS((extra & hashconfig) == 0);
6866 
6867 	if (extra) {
6868 		CH_ALERT(vi,
6869 		    "global RSS config (0x%x) cannot be accommodated.\n",
6870 		    hashconfig);
6871 	}
6872 	if (extra & RSS_HASHTYPE_RSS_IPV4)
6873 		CH_ALERT(vi, "IPv4 2-tuple hashing forced on.\n");
6874 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
6875 		CH_ALERT(vi, "TCP/IPv4 4-tuple hashing forced on.\n");
6876 	if (extra & RSS_HASHTYPE_RSS_IPV6)
6877 		CH_ALERT(vi, "IPv6 2-tuple hashing forced on.\n");
6878 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
6879 		CH_ALERT(vi, "TCP/IPv6 4-tuple hashing forced on.\n");
6880 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
6881 		CH_ALERT(vi, "UDP/IPv4 4-tuple hashing forced on.\n");
6882 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
6883 		CH_ALERT(vi, "UDP/IPv6 4-tuple hashing forced on.\n");
6884 #else
6885 	vi->hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
6886 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
6887 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
6888 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
6889 #endif
6890 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, vi->hashen, vi->rss[0],
6891 	    0, 0);
6892 	if (rc != 0) {
6893 		CH_ERR(vi, "rss hash/defaultq config failed: %d\n", rc);
6894 		return (rc);
6895 	}
6896 
6897 	return (0);
6898 }
6899 
6900 int
6901 vi_init(struct vi_info *vi)
6902 {
6903 	int rc;
6904 
6905 	ASSERT_SYNCHRONIZED_OP(vi->adapter);
6906 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
6907 	    ("%s: VI_INIT_DONE already", __func__));
6908 
6909 	rc = vi_full_init(vi);
6910 	if (rc != 0)
6911 		vi_full_uninit(vi);
6912 	else
6913 		vi->flags |= VI_INIT_DONE;
6914 
6915 	return (rc);
6916 }
6917 
6918 /*
6919  * Idempotent.
6920  */
6921 static void
6922 vi_full_uninit(struct vi_info *vi)
6923 {
6924 
6925 	if (vi->flags & VI_INIT_DONE) {
6926 		quiesce_vi(vi);
6927 		free(vi->rss, M_CXGBE);
6928 		free(vi->nm_rss, M_CXGBE);
6929 	}
6930 
6931 	t4_teardown_vi_queues(vi);
6932 	vi->flags &= ~VI_INIT_DONE;
6933 }
6934 
6935 static void
6936 quiesce_txq(struct sge_txq *txq)
6937 {
6938 	struct sge_eq *eq = &txq->eq;
6939 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
6940 
6941 	MPASS(eq->flags & EQ_SW_ALLOCATED);
6942 	MPASS(!(eq->flags & EQ_ENABLED));
6943 
6944 	/* Wait for the mp_ring to empty. */
6945 	while (!mp_ring_is_idle(txq->r)) {
6946 		mp_ring_check_drainage(txq->r, 4096);
6947 		pause("rquiesce", 1);
6948 	}
6949 	MPASS(txq->txp.npkt == 0);
6950 
6951 	if (eq->flags & EQ_HW_ALLOCATED) {
6952 		/*
6953 		 * Hardware is alive and working normally.  Wait for it to
6954 		 * finish and then wait for the driver to catch up and reclaim
6955 		 * all descriptors.
6956 		 */
6957 		while (spg->cidx != htobe16(eq->pidx))
6958 			pause("equiesce", 1);
6959 		while (eq->cidx != eq->pidx)
6960 			pause("dquiesce", 1);
6961 	} else {
6962 		/*
6963 		 * Hardware is unavailable.  Discard all pending tx and reclaim
6964 		 * descriptors directly.
6965 		 */
6966 		TXQ_LOCK(txq);
6967 		while (eq->cidx != eq->pidx) {
6968 			struct mbuf *m, *nextpkt;
6969 			struct tx_sdesc *txsd;
6970 
6971 			txsd = &txq->sdesc[eq->cidx];
6972 			for (m = txsd->m; m != NULL; m = nextpkt) {
6973 				nextpkt = m->m_nextpkt;
6974 				m->m_nextpkt = NULL;
6975 				m_freem(m);
6976 			}
6977 			IDXINCR(eq->cidx, txsd->desc_used, eq->sidx);
6978 		}
6979 		spg->pidx = spg->cidx = htobe16(eq->cidx);
6980 		TXQ_UNLOCK(txq);
6981 	}
6982 }
6983 
6984 static void
6985 quiesce_wrq(struct sge_wrq *wrq)
6986 {
6987 
6988 	/* XXXTX */
6989 }
6990 
6991 static void
6992 quiesce_iq_fl(struct adapter *sc, struct sge_iq *iq, struct sge_fl *fl)
6993 {
6994 	/* Synchronize with the interrupt handler */
6995 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
6996 		pause("iqfree", 1);
6997 
6998 	if (fl != NULL) {
6999 		MPASS(iq->flags & IQ_HAS_FL);
7000 
7001 		mtx_lock(&sc->sfl_lock);
7002 		FL_LOCK(fl);
7003 		fl->flags |= FL_DOOMED;
7004 		FL_UNLOCK(fl);
7005 		callout_stop(&sc->sfl_callout);
7006 		mtx_unlock(&sc->sfl_lock);
7007 
7008 		KASSERT((fl->flags & FL_STARVING) == 0,
7009 		    ("%s: still starving", __func__));
7010 
7011 		/* Release all buffers if hardware is no longer available. */
7012 		if (!(iq->flags & IQ_HW_ALLOCATED))
7013 			free_fl_buffers(sc, fl);
7014 	}
7015 }
7016 
7017 /*
7018  * Wait for all activity on all the queues of the VI to complete.  It is assumed
7019  * that no new work is being enqueued by the hardware or the driver.  That part
7020  * should be arranged before calling this function.
7021  */
7022 static void
7023 quiesce_vi(struct vi_info *vi)
7024 {
7025 	int i;
7026 	struct adapter *sc = vi->adapter;
7027 	struct sge_rxq *rxq;
7028 	struct sge_txq *txq;
7029 #ifdef TCP_OFFLOAD
7030 	struct sge_ofld_rxq *ofld_rxq;
7031 #endif
7032 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
7033 	struct sge_ofld_txq *ofld_txq;
7034 #endif
7035 
7036 	if (!(vi->flags & VI_INIT_DONE))
7037 		return;
7038 
7039 	for_each_txq(vi, i, txq) {
7040 		quiesce_txq(txq);
7041 	}
7042 
7043 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
7044 	for_each_ofld_txq(vi, i, ofld_txq) {
7045 		quiesce_wrq(&ofld_txq->wrq);
7046 	}
7047 #endif
7048 
7049 	for_each_rxq(vi, i, rxq) {
7050 		quiesce_iq_fl(sc, &rxq->iq, &rxq->fl);
7051 	}
7052 
7053 #ifdef TCP_OFFLOAD
7054 	for_each_ofld_rxq(vi, i, ofld_rxq) {
7055 		quiesce_iq_fl(sc, &ofld_rxq->iq, &ofld_rxq->fl);
7056 	}
7057 #endif
7058 }
7059 
7060 static int
7061 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
7062     driver_intr_t *handler, void *arg, char *name)
7063 {
7064 	int rc;
7065 
7066 	irq->rid = rid;
7067 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
7068 	    RF_SHAREABLE | RF_ACTIVE);
7069 	if (irq->res == NULL) {
7070 		device_printf(sc->dev,
7071 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
7072 		return (ENOMEM);
7073 	}
7074 
7075 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
7076 	    NULL, handler, arg, &irq->tag);
7077 	if (rc != 0) {
7078 		device_printf(sc->dev,
7079 		    "failed to setup interrupt for rid %d, name %s: %d\n",
7080 		    rid, name, rc);
7081 	} else if (name)
7082 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
7083 
7084 	return (rc);
7085 }
7086 
7087 static int
7088 t4_free_irq(struct adapter *sc, struct irq *irq)
7089 {
7090 	if (irq->tag)
7091 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
7092 	if (irq->res)
7093 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
7094 
7095 	bzero(irq, sizeof(*irq));
7096 
7097 	return (0);
7098 }
7099 
7100 static void
7101 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
7102 {
7103 
7104 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
7105 	t4_get_regs(sc, buf, regs->len);
7106 }
7107 
7108 #define	A_PL_INDIR_CMD	0x1f8
7109 
7110 #define	S_PL_AUTOINC	31
7111 #define	M_PL_AUTOINC	0x1U
7112 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
7113 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
7114 
7115 #define	S_PL_VFID	20
7116 #define	M_PL_VFID	0xffU
7117 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
7118 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
7119 
7120 #define	S_PL_ADDR	0
7121 #define	M_PL_ADDR	0xfffffU
7122 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
7123 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
7124 
7125 #define	A_PL_INDIR_DATA	0x1fc
7126 
7127 static uint64_t
7128 read_vf_stat(struct adapter *sc, u_int vin, int reg)
7129 {
7130 	u32 stats[2];
7131 
7132 	if (sc->flags & IS_VF) {
7133 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
7134 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
7135 	} else {
7136 		mtx_assert(&sc->reg_lock, MA_OWNED);
7137 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
7138 		    V_PL_VFID(vin) | V_PL_ADDR(VF_MPS_REG(reg)));
7139 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
7140 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
7141 	}
7142 	return (((uint64_t)stats[1]) << 32 | stats[0]);
7143 }
7144 
7145 static void
7146 t4_get_vi_stats(struct adapter *sc, u_int vin, struct fw_vi_stats_vf *stats)
7147 {
7148 
7149 #define GET_STAT(name) \
7150 	read_vf_stat(sc, vin, A_MPS_VF_STAT_##name##_L)
7151 
7152 	if (!(sc->flags & IS_VF))
7153 		mtx_lock(&sc->reg_lock);
7154 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
7155 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
7156 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
7157 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
7158 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
7159 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
7160 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
7161 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
7162 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
7163 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
7164 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
7165 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
7166 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
7167 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
7168 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
7169 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
7170 	if (!(sc->flags & IS_VF))
7171 		mtx_unlock(&sc->reg_lock);
7172 
7173 #undef GET_STAT
7174 }
7175 
7176 static void
7177 t4_clr_vi_stats(struct adapter *sc, u_int vin)
7178 {
7179 	int reg;
7180 
7181 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) | V_PL_VFID(vin) |
7182 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
7183 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
7184 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
7185 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
7186 }
7187 
7188 static void
7189 vi_refresh_stats(struct vi_info *vi)
7190 {
7191 	struct timeval tv;
7192 	const struct timeval interval = {0, 250000};	/* 250ms */
7193 
7194 	mtx_assert(&vi->tick_mtx, MA_OWNED);
7195 
7196 	if (vi->flags & VI_SKIP_STATS)
7197 		return;
7198 
7199 	getmicrotime(&tv);
7200 	timevalsub(&tv, &interval);
7201 	if (timevalcmp(&tv, &vi->last_refreshed, <))
7202 		return;
7203 
7204 	t4_get_vi_stats(vi->adapter, vi->vin, &vi->stats);
7205 	getmicrotime(&vi->last_refreshed);
7206 }
7207 
7208 static void
7209 cxgbe_refresh_stats(struct vi_info *vi)
7210 {
7211 	u_int i, v, tnl_cong_drops, chan_map;
7212 	struct timeval tv;
7213 	const struct timeval interval = {0, 250000};	/* 250ms */
7214 	struct port_info *pi;
7215 	struct adapter *sc;
7216 
7217 	mtx_assert(&vi->tick_mtx, MA_OWNED);
7218 
7219 	if (vi->flags & VI_SKIP_STATS)
7220 		return;
7221 
7222 	getmicrotime(&tv);
7223 	timevalsub(&tv, &interval);
7224 	if (timevalcmp(&tv, &vi->last_refreshed, <))
7225 		return;
7226 
7227 	pi = vi->pi;
7228 	sc = vi->adapter;
7229 	tnl_cong_drops = 0;
7230 	t4_get_port_stats(sc, pi->port_id, &pi->stats);
7231 	chan_map = pi->rx_e_chan_map;
7232 	while (chan_map) {
7233 		i = ffs(chan_map) - 1;
7234 		mtx_lock(&sc->reg_lock);
7235 		t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
7236 		    A_TP_MIB_TNL_CNG_DROP_0 + i);
7237 		mtx_unlock(&sc->reg_lock);
7238 		tnl_cong_drops += v;
7239 		chan_map &= ~(1 << i);
7240 	}
7241 	pi->tnl_cong_drops = tnl_cong_drops;
7242 	getmicrotime(&vi->last_refreshed);
7243 }
7244 
7245 static void
7246 cxgbe_tick(void *arg)
7247 {
7248 	struct vi_info *vi = arg;
7249 
7250 	MPASS(IS_MAIN_VI(vi));
7251 	mtx_assert(&vi->tick_mtx, MA_OWNED);
7252 
7253 	cxgbe_refresh_stats(vi);
7254 	callout_schedule(&vi->tick, hz);
7255 }
7256 
7257 static void
7258 vi_tick(void *arg)
7259 {
7260 	struct vi_info *vi = arg;
7261 
7262 	mtx_assert(&vi->tick_mtx, MA_OWNED);
7263 
7264 	vi_refresh_stats(vi);
7265 	callout_schedule(&vi->tick, hz);
7266 }
7267 
7268 /*
7269  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
7270  */
7271 static char *caps_decoder[] = {
7272 	"\20\001IPMI\002NCSI",				/* 0: NBM */
7273 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
7274 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
7275 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
7276 	    "\006HASHFILTER\007ETHOFLD",
7277 	"\20\001TOE",					/* 4: TOE */
7278 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
7279 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
7280 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
7281 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
7282 	    "\007T10DIF"
7283 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
7284 	"\20\001LOOKASIDE\002TLSKEYS\003IPSEC_INLINE"	/* 7: Crypto */
7285 	    "\004TLS_HW",
7286 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
7287 		    "\004PO_INITIATOR\005PO_TARGET",
7288 };
7289 
7290 void
7291 t4_sysctls(struct adapter *sc)
7292 {
7293 	struct sysctl_ctx_list *ctx = &sc->ctx;
7294 	struct sysctl_oid *oid;
7295 	struct sysctl_oid_list *children, *c0;
7296 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
7297 
7298 	/*
7299 	 * dev.t4nex.X.
7300 	 */
7301 	oid = device_get_sysctl_tree(sc->dev);
7302 	c0 = children = SYSCTL_CHILDREN(oid);
7303 
7304 	sc->sc_do_rxcopy = 1;
7305 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
7306 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
7307 
7308 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
7309 	    sc->params.nports, "# of ports");
7310 
7311 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
7312 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, doorbells,
7313 	    (uintptr_t)&sc->doorbells, sysctl_bitfield_8b, "A",
7314 	    "available doorbells");
7315 
7316 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
7317 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
7318 
7319 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
7320 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
7321 	    sc->params.sge.timer_val, sizeof(sc->params.sge.timer_val),
7322 	    sysctl_int_array, "A", "interrupt holdoff timer values (us)");
7323 
7324 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
7325 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
7326 	    sc->params.sge.counter_val, sizeof(sc->params.sge.counter_val),
7327 	    sysctl_int_array, "A", "interrupt holdoff packet counter values");
7328 
7329 	t4_sge_sysctls(sc, ctx, children);
7330 
7331 	sc->lro_timeout = 100;
7332 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
7333 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
7334 
7335 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
7336 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
7337 
7338 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
7339 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
7340 
7341 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
7342 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
7343 
7344 	if (sc->flags & IS_VF)
7345 		return;
7346 
7347 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
7348 	    NULL, chip_rev(sc), "chip hardware revision");
7349 
7350 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
7351 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
7352 
7353 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
7354 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
7355 
7356 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
7357 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
7358 
7359 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
7360 	    CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
7361 
7362 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
7363 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
7364 
7365 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
7366 	    sc->er_version, 0, "expansion ROM version");
7367 
7368 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
7369 	    sc->bs_version, 0, "bootstrap firmware version");
7370 
7371 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
7372 	    NULL, sc->params.scfg_vers, "serial config version");
7373 
7374 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
7375 	    NULL, sc->params.vpd_vers, "VPD version");
7376 
7377 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
7378 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
7379 
7380 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
7381 	    sc->cfcsum, "config file checksum");
7382 
7383 #define SYSCTL_CAP(name, n, text) \
7384 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
7385 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, caps_decoder[n], \
7386 	    (uintptr_t)&sc->name, sysctl_bitfield_16b, "A", \
7387 	    "available " text " capabilities")
7388 
7389 	SYSCTL_CAP(nbmcaps, 0, "NBM");
7390 	SYSCTL_CAP(linkcaps, 1, "link");
7391 	SYSCTL_CAP(switchcaps, 2, "switch");
7392 	SYSCTL_CAP(niccaps, 3, "NIC");
7393 	SYSCTL_CAP(toecaps, 4, "TCP offload");
7394 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
7395 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
7396 	SYSCTL_CAP(cryptocaps, 7, "crypto");
7397 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
7398 #undef SYSCTL_CAP
7399 
7400 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
7401 	    NULL, sc->tids.nftids, "number of filters");
7402 
7403 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
7404 	    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7405 	    sysctl_temperature, "I", "chip temperature (in Celsius)");
7406 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "reset_sensor",
7407 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
7408 	    sysctl_reset_sensor, "I", "reset the chip's temperature sensor.");
7409 
7410 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "loadavg",
7411 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7412 	    sysctl_loadavg, "A",
7413 	    "microprocessor load averages (debug firmwares only)");
7414 
7415 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "core_vdd",
7416 	    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0, sysctl_vdd,
7417 	    "I", "core Vdd (in mV)");
7418 
7419 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "local_cpus",
7420 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, LOCAL_CPUS,
7421 	    sysctl_cpus, "A", "local CPUs");
7422 
7423 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "intr_cpus",
7424 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, INTR_CPUS,
7425 	    sysctl_cpus, "A", "preferred CPUs for interrupts");
7426 
7427 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "swintr", CTLFLAG_RW,
7428 	    &sc->swintr, 0, "software triggered interrupts");
7429 
7430 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "reset",
7431 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_reset, "I",
7432 	    "1 = reset adapter, 0 = zero reset counter");
7433 
7434 	/*
7435 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
7436 	 */
7437 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
7438 	    CTLFLAG_RD | CTLFLAG_SKIP | CTLFLAG_MPSAFE, NULL,
7439 	    "logs and miscellaneous information");
7440 	children = SYSCTL_CHILDREN(oid);
7441 
7442 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
7443 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7444 	    sysctl_cctrl, "A", "congestion control");
7445 
7446 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
7447 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7448 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
7449 
7450 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
7451 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 1,
7452 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
7453 
7454 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
7455 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 2,
7456 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
7457 
7458 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
7459 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 3,
7460 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
7461 
7462 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
7463 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 4,
7464 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
7465 
7466 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
7467 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 5,
7468 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
7469 
7470 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
7471 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7472 	    sysctl_cim_la, "A", "CIM logic analyzer");
7473 
7474 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
7475 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7476 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
7477 
7478 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
7479 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7480 	    0 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
7481 
7482 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
7483 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7484 	    1 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
7485 
7486 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
7487 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7488 	    2 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
7489 
7490 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
7491 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7492 	    3 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
7493 
7494 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
7495 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7496 	    4 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
7497 
7498 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
7499 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7500 	    5 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
7501 
7502 	if (chip_id(sc) > CHELSIO_T4) {
7503 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
7504 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7505 		    6 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A",
7506 		    "CIM OBQ 6 (SGE0-RX)");
7507 
7508 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
7509 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7510 		    7 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A",
7511 		    "CIM OBQ 7 (SGE1-RX)");
7512 	}
7513 
7514 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
7515 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7516 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
7517 
7518 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
7519 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7520 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
7521 
7522 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
7523 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7524 	    sysctl_cpl_stats, "A", "CPL statistics");
7525 
7526 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
7527 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7528 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
7529 
7530 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tid_stats",
7531 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7532 	    sysctl_tid_stats, "A", "tid stats");
7533 
7534 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
7535 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7536 	    sysctl_devlog, "A", "firmware's device log");
7537 
7538 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
7539 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7540 	    sysctl_fcoe_stats, "A", "FCoE statistics");
7541 
7542 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
7543 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7544 	    sysctl_hw_sched, "A", "hardware scheduler ");
7545 
7546 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
7547 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7548 	    sysctl_l2t, "A", "hardware L2 table");
7549 
7550 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "smt",
7551 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7552 	    sysctl_smt, "A", "hardware source MAC table");
7553 
7554 #ifdef INET6
7555 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "clip",
7556 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7557 	    sysctl_clip, "A", "active CLIP table entries");
7558 #endif
7559 
7560 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
7561 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7562 	    sysctl_lb_stats, "A", "loopback statistics");
7563 
7564 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
7565 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7566 	    sysctl_meminfo, "A", "memory regions");
7567 
7568 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
7569 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7570 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
7571 	    "A", "MPS TCAM entries");
7572 
7573 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
7574 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7575 	    sysctl_path_mtus, "A", "path MTUs");
7576 
7577 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
7578 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7579 	    sysctl_pm_stats, "A", "PM statistics");
7580 
7581 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
7582 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7583 	    sysctl_rdma_stats, "A", "RDMA statistics");
7584 
7585 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
7586 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7587 	    sysctl_tcp_stats, "A", "TCP statistics");
7588 
7589 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
7590 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7591 	    sysctl_tids, "A", "TID information");
7592 
7593 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
7594 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7595 	    sysctl_tp_err_stats, "A", "TP error statistics");
7596 
7597 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tnl_stats",
7598 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7599 	    sysctl_tnl_stats, "A", "TP tunnel statistics");
7600 
7601 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
7602 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
7603 	    sysctl_tp_la_mask, "I", "TP logic analyzer event capture mask");
7604 
7605 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
7606 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7607 	    sysctl_tp_la, "A", "TP logic analyzer");
7608 
7609 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
7610 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7611 	    sysctl_tx_rate, "A", "Tx rate");
7612 
7613 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
7614 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7615 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
7616 
7617 	if (chip_id(sc) >= CHELSIO_T5) {
7618 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
7619 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7620 		    sysctl_wcwr_stats, "A", "write combined work requests");
7621 	}
7622 
7623 #ifdef KERN_TLS
7624 	if (is_ktls(sc)) {
7625 		/*
7626 		 * dev.t4nex.0.tls.
7627 		 */
7628 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "tls",
7629 		    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "KERN_TLS parameters");
7630 		children = SYSCTL_CHILDREN(oid);
7631 
7632 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "inline_keys",
7633 		    CTLFLAG_RW, &sc->tlst.inline_keys, 0, "Always pass TLS "
7634 		    "keys in work requests (1) or attempt to store TLS keys "
7635 		    "in card memory.");
7636 
7637 		if (is_t6(sc))
7638 			SYSCTL_ADD_INT(ctx, children, OID_AUTO, "combo_wrs",
7639 			    CTLFLAG_RW, &sc->tlst.combo_wrs, 0, "Attempt to "
7640 			    "combine TCB field updates with TLS record work "
7641 			    "requests.");
7642 	}
7643 #endif
7644 
7645 #ifdef TCP_OFFLOAD
7646 	if (is_offload(sc)) {
7647 		int i;
7648 		char s[4];
7649 
7650 		/*
7651 		 * dev.t4nex.X.toe.
7652 		 */
7653 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe",
7654 		    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "TOE parameters");
7655 		children = SYSCTL_CHILDREN(oid);
7656 
7657 		sc->tt.cong_algorithm = -1;
7658 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
7659 		    CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
7660 		    "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
7661 		    "3 = highspeed)");
7662 
7663 		sc->tt.sndbuf = -1;
7664 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
7665 		    &sc->tt.sndbuf, 0, "hardware send buffer");
7666 
7667 		sc->tt.ddp = 0;
7668 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp",
7669 		    CTLFLAG_RW | CTLFLAG_SKIP, &sc->tt.ddp, 0, "");
7670 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_zcopy", CTLFLAG_RW,
7671 		    &sc->tt.ddp, 0, "Enable zero-copy aio_read(2)");
7672 
7673 		sc->tt.rx_coalesce = -1;
7674 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
7675 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
7676 
7677 		sc->tt.tls = 0;
7678 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls", CTLTYPE_INT |
7679 		    CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0, sysctl_tls, "I",
7680 		    "Inline TLS allowed");
7681 
7682 		sc->tt.tx_align = -1;
7683 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
7684 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
7685 
7686 		sc->tt.tx_zcopy = 0;
7687 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
7688 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
7689 		    "Enable zero-copy aio_write(2)");
7690 
7691 		sc->tt.cop_managed_offloading = !!t4_cop_managed_offloading;
7692 		SYSCTL_ADD_INT(ctx, children, OID_AUTO,
7693 		    "cop_managed_offloading", CTLFLAG_RW,
7694 		    &sc->tt.cop_managed_offloading, 0,
7695 		    "COP (Connection Offload Policy) controls all TOE offload");
7696 
7697 		sc->tt.autorcvbuf_inc = 16 * 1024;
7698 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "autorcvbuf_inc",
7699 		    CTLFLAG_RW, &sc->tt.autorcvbuf_inc, 0,
7700 		    "autorcvbuf increment");
7701 
7702 		sc->tt.update_hc_on_pmtu_change = 1;
7703 		SYSCTL_ADD_INT(ctx, children, OID_AUTO,
7704 		    "update_hc_on_pmtu_change", CTLFLAG_RW,
7705 		    &sc->tt.update_hc_on_pmtu_change, 0,
7706 		    "Update hostcache entry if the PMTU changes");
7707 
7708 		sc->tt.iso = 1;
7709 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "iso", CTLFLAG_RW,
7710 		    &sc->tt.iso, 0, "Enable iSCSI segmentation offload");
7711 
7712 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
7713 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7714 		    sysctl_tp_tick, "A", "TP timer tick (us)");
7715 
7716 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
7717 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 1,
7718 		    sysctl_tp_tick, "A", "TCP timestamp tick (us)");
7719 
7720 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
7721 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 2,
7722 		    sysctl_tp_tick, "A", "DACK tick (us)");
7723 
7724 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
7725 		    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
7726 		    sysctl_tp_dack_timer, "IU", "DACK timer (us)");
7727 
7728 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
7729 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7730 		    A_TP_RXT_MIN, sysctl_tp_timer, "LU",
7731 		    "Minimum retransmit interval (us)");
7732 
7733 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
7734 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7735 		    A_TP_RXT_MAX, sysctl_tp_timer, "LU",
7736 		    "Maximum retransmit interval (us)");
7737 
7738 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
7739 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7740 		    A_TP_PERS_MIN, sysctl_tp_timer, "LU",
7741 		    "Persist timer min (us)");
7742 
7743 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
7744 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7745 		    A_TP_PERS_MAX, sysctl_tp_timer, "LU",
7746 		    "Persist timer max (us)");
7747 
7748 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
7749 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7750 		    A_TP_KEEP_IDLE, sysctl_tp_timer, "LU",
7751 		    "Keepalive idle timer (us)");
7752 
7753 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
7754 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7755 		    A_TP_KEEP_INTVL, sysctl_tp_timer, "LU",
7756 		    "Keepalive interval timer (us)");
7757 
7758 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
7759 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7760 		    A_TP_INIT_SRTT, sysctl_tp_timer, "LU", "Initial SRTT (us)");
7761 
7762 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
7763 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7764 		    A_TP_FINWAIT2_TIMER, sysctl_tp_timer, "LU",
7765 		    "FINWAIT2 timer (us)");
7766 
7767 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
7768 		    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7769 		    S_SYNSHIFTMAX, sysctl_tp_shift_cnt, "IU",
7770 		    "Number of SYN retransmissions before abort");
7771 
7772 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
7773 		    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7774 		    S_RXTSHIFTMAXR2, sysctl_tp_shift_cnt, "IU",
7775 		    "Number of retransmissions before abort");
7776 
7777 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
7778 		    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7779 		    S_KEEPALIVEMAXR2, sysctl_tp_shift_cnt, "IU",
7780 		    "Number of keepalive probes before abort");
7781 
7782 		oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
7783 		    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
7784 		    "TOE retransmit backoffs");
7785 		children = SYSCTL_CHILDREN(oid);
7786 		for (i = 0; i < 16; i++) {
7787 			snprintf(s, sizeof(s), "%u", i);
7788 			SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
7789 			    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
7790 			    i, sysctl_tp_backoff, "IU",
7791 			    "TOE retransmit backoff");
7792 		}
7793 	}
7794 #endif
7795 }
7796 
7797 void
7798 vi_sysctls(struct vi_info *vi)
7799 {
7800 	struct sysctl_ctx_list *ctx = &vi->ctx;
7801 	struct sysctl_oid *oid;
7802 	struct sysctl_oid_list *children;
7803 
7804 	/*
7805 	 * dev.v?(cxgbe|cxl).X.
7806 	 */
7807 	oid = device_get_sysctl_tree(vi->dev);
7808 	children = SYSCTL_CHILDREN(oid);
7809 
7810 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
7811 	    vi->viid, "VI identifer");
7812 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
7813 	    &vi->nrxq, 0, "# of rx queues");
7814 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
7815 	    &vi->ntxq, 0, "# of tx queues");
7816 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
7817 	    &vi->first_rxq, 0, "index of first rx queue");
7818 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
7819 	    &vi->first_txq, 0, "index of first tx queue");
7820 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_base", CTLFLAG_RD, NULL,
7821 	    vi->rss_base, "start of RSS indirection table");
7822 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
7823 	    vi->rss_size, "size of RSS indirection table");
7824 
7825 	if (IS_MAIN_VI(vi)) {
7826 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
7827 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, vi, 0,
7828 		    sysctl_noflowq, "IU",
7829 		    "Reserve queue 0 for non-flowid packets");
7830 	}
7831 
7832 	if (vi->adapter->flags & IS_VF) {
7833 		MPASS(vi->flags & TX_USES_VM_WR);
7834 		SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_vm_wr", CTLFLAG_RD,
7835 		    NULL, 1, "use VM work requests for transmit");
7836 	} else {
7837 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_vm_wr",
7838 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, vi, 0,
7839 		    sysctl_tx_vm_wr, "I", "use VM work requestes for transmit");
7840 	}
7841 
7842 #ifdef TCP_OFFLOAD
7843 	if (vi->nofldrxq != 0) {
7844 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
7845 		    &vi->nofldrxq, 0,
7846 		    "# of rx queues for offloaded TCP connections");
7847 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
7848 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
7849 		    "index of first TOE rx queue");
7850 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
7851 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, vi, 0,
7852 		    sysctl_holdoff_tmr_idx_ofld, "I",
7853 		    "holdoff timer index for TOE queues");
7854 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
7855 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, vi, 0,
7856 		    sysctl_holdoff_pktc_idx_ofld, "I",
7857 		    "holdoff packet counter index for TOE queues");
7858 	}
7859 #endif
7860 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
7861 	if (vi->nofldtxq != 0) {
7862 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
7863 		    &vi->nofldtxq, 0,
7864 		    "# of tx queues for TOE/ETHOFLD");
7865 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
7866 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
7867 		    "index of first TOE/ETHOFLD tx queue");
7868 	}
7869 #endif
7870 #ifdef DEV_NETMAP
7871 	if (vi->nnmrxq != 0) {
7872 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
7873 		    &vi->nnmrxq, 0, "# of netmap rx queues");
7874 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
7875 		    &vi->nnmtxq, 0, "# of netmap tx queues");
7876 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
7877 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
7878 		    "index of first netmap rx queue");
7879 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
7880 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
7881 		    "index of first netmap tx queue");
7882 	}
7883 #endif
7884 
7885 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
7886 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, vi, 0,
7887 	    sysctl_holdoff_tmr_idx, "I", "holdoff timer index");
7888 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
7889 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, vi, 0,
7890 	    sysctl_holdoff_pktc_idx, "I", "holdoff packet counter index");
7891 
7892 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
7893 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, vi, 0,
7894 	    sysctl_qsize_rxq, "I", "rx queue size");
7895 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
7896 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, vi, 0,
7897 	    sysctl_qsize_txq, "I", "tx queue size");
7898 }
7899 
7900 static void
7901 cxgbe_sysctls(struct port_info *pi)
7902 {
7903 	struct sysctl_ctx_list *ctx = &pi->ctx;
7904 	struct sysctl_oid *oid;
7905 	struct sysctl_oid_list *children, *children2;
7906 	struct adapter *sc = pi->adapter;
7907 	int i;
7908 	char name[16];
7909 	static char *tc_flags = {"\20\1USER"};
7910 
7911 	/*
7912 	 * dev.cxgbe.X.
7913 	 */
7914 	oid = device_get_sysctl_tree(pi->dev);
7915 	children = SYSCTL_CHILDREN(oid);
7916 
7917 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc",
7918 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, pi, 0,
7919 	    sysctl_linkdnrc, "A", "reason why link is down");
7920 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
7921 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
7922 		    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, pi, 0,
7923 		    sysctl_btphy, "I", "PHY temperature (in Celsius)");
7924 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
7925 		    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, pi, 1,
7926 		    sysctl_btphy, "I", "PHY firmware version");
7927 	}
7928 
7929 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
7930 	    CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE, pi, 0,
7931 	    sysctl_pause_settings, "A",
7932 	    "PAUSE settings (bit 0 = rx_pause, 1 = tx_pause, 2 = pause_autoneg)");
7933 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "link_fec",
7934 	    CTLTYPE_STRING | CTLFLAG_MPSAFE, pi, 0, sysctl_link_fec, "A",
7935 	    "FEC in use on the link");
7936 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "requested_fec",
7937 	    CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE, pi, 0,
7938 	    sysctl_requested_fec, "A",
7939 	    "FECs to use (bit 0 = RS, 1 = FC, 2 = none, 5 = auto, 6 = module)");
7940 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "module_fec",
7941 	    CTLTYPE_STRING | CTLFLAG_MPSAFE, pi, 0, sysctl_module_fec, "A",
7942 	    "FEC recommended by the cable/transceiver");
7943 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
7944 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, pi, 0,
7945 	    sysctl_autoneg, "I",
7946 	    "autonegotiation (-1 = not supported)");
7947 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "force_fec",
7948 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, pi, 0,
7949 	    sysctl_force_fec, "I", "when to use FORCE_FEC bit for link config");
7950 
7951 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rcaps", CTLFLAG_RD,
7952 	    &pi->link_cfg.requested_caps, 0, "L1 config requested by driver");
7953 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "pcaps", CTLFLAG_RD,
7954 	    &pi->link_cfg.pcaps, 0, "port capabilities");
7955 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "acaps", CTLFLAG_RD,
7956 	    &pi->link_cfg.acaps, 0, "advertised capabilities");
7957 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lpacaps", CTLFLAG_RD,
7958 	    &pi->link_cfg.lpacaps, 0, "link partner advertised capabilities");
7959 
7960 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
7961 	    port_top_speed(pi), "max speed (in Gbps)");
7962 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
7963 	    pi->mps_bg_map, "MPS buffer group map");
7964 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
7965 	    NULL, pi->rx_e_chan_map, "TP rx e-channel map");
7966 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_c_chan", CTLFLAG_RD, NULL,
7967 	    pi->rx_c_chan, "TP rx c-channel");
7968 
7969 	if (sc->flags & IS_VF)
7970 		return;
7971 
7972 	/*
7973 	 * dev.(cxgbe|cxl).X.tc.
7974 	 */
7975 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc",
7976 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
7977 	    "Tx scheduler traffic classes (cl_rl)");
7978 	children2 = SYSCTL_CHILDREN(oid);
7979 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "pktsize",
7980 	    CTLFLAG_RW, &pi->sched_params->pktsize, 0,
7981 	    "pktsize for per-flow cl-rl (0 means up to the driver )");
7982 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "burstsize",
7983 	    CTLFLAG_RW, &pi->sched_params->burstsize, 0,
7984 	    "burstsize for per-flow cl-rl (0 means up to the driver)");
7985 	for (i = 0; i < sc->params.nsched_cls; i++) {
7986 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
7987 
7988 		snprintf(name, sizeof(name), "%d", i);
7989 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
7990 		    SYSCTL_CHILDREN(oid), OID_AUTO, name,
7991 		    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "traffic class"));
7992 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "state",
7993 		    CTLFLAG_RD, &tc->state, 0, "current state");
7994 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "flags",
7995 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, tc_flags,
7996 		    (uintptr_t)&tc->flags, sysctl_bitfield_8b, "A", "flags");
7997 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
7998 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
7999 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
8000 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc,
8001 		    (pi->port_id << 16) | i, sysctl_tc_params, "A",
8002 		    "traffic class parameters");
8003 	}
8004 
8005 	/*
8006 	 * dev.cxgbe.X.stats.
8007 	 */
8008 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats",
8009 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "port statistics");
8010 	children = SYSCTL_CHILDREN(oid);
8011 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
8012 	    &pi->tx_parse_error, 0,
8013 	    "# of tx packets with invalid length or # of segments");
8014 
8015 #define T4_REGSTAT(name, stat, desc) \
8016     SYSCTL_ADD_OID(ctx, children, OID_AUTO, #name, \
8017         CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, \
8018 	(is_t4(sc) ? PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_##stat##_L) : \
8019 	T5_PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_##stat##_L)), \
8020         sysctl_handle_t4_reg64, "QU", desc)
8021 
8022 /* We get these from port_stats and they may be stale by up to 1s */
8023 #define T4_PORTSTAT(name, desc) \
8024 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
8025 	    &pi->stats.name, desc)
8026 
8027 	T4_REGSTAT(tx_octets, TX_PORT_BYTES, "# of octets in good frames");
8028 	T4_REGSTAT(tx_frames, TX_PORT_FRAMES, "total # of good frames");
8029 	T4_REGSTAT(tx_bcast_frames, TX_PORT_BCAST, "# of broadcast frames");
8030 	T4_REGSTAT(tx_mcast_frames, TX_PORT_MCAST, "# of multicast frames");
8031 	T4_REGSTAT(tx_ucast_frames, TX_PORT_UCAST, "# of unicast frames");
8032 	T4_REGSTAT(tx_error_frames, TX_PORT_ERROR, "# of error frames");
8033 	T4_REGSTAT(tx_frames_64, TX_PORT_64B, "# of tx frames in this range");
8034 	T4_REGSTAT(tx_frames_65_127, TX_PORT_65B_127B, "# of tx frames in this range");
8035 	T4_REGSTAT(tx_frames_128_255, TX_PORT_128B_255B, "# of tx frames in this range");
8036 	T4_REGSTAT(tx_frames_256_511, TX_PORT_256B_511B, "# of tx frames in this range");
8037 	T4_REGSTAT(tx_frames_512_1023, TX_PORT_512B_1023B, "# of tx frames in this range");
8038 	T4_REGSTAT(tx_frames_1024_1518, TX_PORT_1024B_1518B, "# of tx frames in this range");
8039 	T4_REGSTAT(tx_frames_1519_max, TX_PORT_1519B_MAX, "# of tx frames in this range");
8040 	T4_REGSTAT(tx_drop, TX_PORT_DROP, "# of dropped tx frames");
8041 	T4_REGSTAT(tx_pause, TX_PORT_PAUSE, "# of pause frames transmitted");
8042 	T4_REGSTAT(tx_ppp0, TX_PORT_PPP0, "# of PPP prio 0 frames transmitted");
8043 	T4_REGSTAT(tx_ppp1, TX_PORT_PPP1, "# of PPP prio 1 frames transmitted");
8044 	T4_REGSTAT(tx_ppp2, TX_PORT_PPP2, "# of PPP prio 2 frames transmitted");
8045 	T4_REGSTAT(tx_ppp3, TX_PORT_PPP3, "# of PPP prio 3 frames transmitted");
8046 	T4_REGSTAT(tx_ppp4, TX_PORT_PPP4, "# of PPP prio 4 frames transmitted");
8047 	T4_REGSTAT(tx_ppp5, TX_PORT_PPP5, "# of PPP prio 5 frames transmitted");
8048 	T4_REGSTAT(tx_ppp6, TX_PORT_PPP6, "# of PPP prio 6 frames transmitted");
8049 	T4_REGSTAT(tx_ppp7, TX_PORT_PPP7, "# of PPP prio 7 frames transmitted");
8050 
8051 	T4_REGSTAT(rx_octets, RX_PORT_BYTES, "# of octets in good frames");
8052 	T4_REGSTAT(rx_frames, RX_PORT_FRAMES, "total # of good frames");
8053 	T4_REGSTAT(rx_bcast_frames, RX_PORT_BCAST, "# of broadcast frames");
8054 	T4_REGSTAT(rx_mcast_frames, RX_PORT_MCAST, "# of multicast frames");
8055 	T4_REGSTAT(rx_ucast_frames, RX_PORT_UCAST, "# of unicast frames");
8056 	T4_REGSTAT(rx_too_long, RX_PORT_MTU_ERROR, "# of frames exceeding MTU");
8057 	T4_REGSTAT(rx_jabber, RX_PORT_MTU_CRC_ERROR, "# of jabber frames");
8058 	if (is_t6(sc)) {
8059 		T4_PORTSTAT(rx_fcs_err,
8060 		    "# of frames received with bad FCS since last link up");
8061 	} else {
8062 		T4_REGSTAT(rx_fcs_err, RX_PORT_CRC_ERROR,
8063 		    "# of frames received with bad FCS");
8064 	}
8065 	T4_REGSTAT(rx_len_err, RX_PORT_LEN_ERROR, "# of frames received with length error");
8066 	T4_REGSTAT(rx_symbol_err, RX_PORT_SYM_ERROR, "symbol errors");
8067 	T4_REGSTAT(rx_runt, RX_PORT_LESS_64B, "# of short frames received");
8068 	T4_REGSTAT(rx_frames_64, RX_PORT_64B, "# of rx frames in this range");
8069 	T4_REGSTAT(rx_frames_65_127, RX_PORT_65B_127B, "# of rx frames in this range");
8070 	T4_REGSTAT(rx_frames_128_255, RX_PORT_128B_255B, "# of rx frames in this range");
8071 	T4_REGSTAT(rx_frames_256_511, RX_PORT_256B_511B, "# of rx frames in this range");
8072 	T4_REGSTAT(rx_frames_512_1023, RX_PORT_512B_1023B, "# of rx frames in this range");
8073 	T4_REGSTAT(rx_frames_1024_1518, RX_PORT_1024B_1518B, "# of rx frames in this range");
8074 	T4_REGSTAT(rx_frames_1519_max, RX_PORT_1519B_MAX, "# of rx frames in this range");
8075 	T4_REGSTAT(rx_pause, RX_PORT_PAUSE, "# of pause frames received");
8076 	T4_REGSTAT(rx_ppp0, RX_PORT_PPP0, "# of PPP prio 0 frames received");
8077 	T4_REGSTAT(rx_ppp1, RX_PORT_PPP1, "# of PPP prio 1 frames received");
8078 	T4_REGSTAT(rx_ppp2, RX_PORT_PPP2, "# of PPP prio 2 frames received");
8079 	T4_REGSTAT(rx_ppp3, RX_PORT_PPP3, "# of PPP prio 3 frames received");
8080 	T4_REGSTAT(rx_ppp4, RX_PORT_PPP4, "# of PPP prio 4 frames received");
8081 	T4_REGSTAT(rx_ppp5, RX_PORT_PPP5, "# of PPP prio 5 frames received");
8082 	T4_REGSTAT(rx_ppp6, RX_PORT_PPP6, "# of PPP prio 6 frames received");
8083 	T4_REGSTAT(rx_ppp7, RX_PORT_PPP7, "# of PPP prio 7 frames received");
8084 
8085 	T4_PORTSTAT(rx_ovflow0, "# drops due to buffer-group 0 overflows");
8086 	T4_PORTSTAT(rx_ovflow1, "# drops due to buffer-group 1 overflows");
8087 	T4_PORTSTAT(rx_ovflow2, "# drops due to buffer-group 2 overflows");
8088 	T4_PORTSTAT(rx_ovflow3, "# drops due to buffer-group 3 overflows");
8089 	T4_PORTSTAT(rx_trunc0, "# of buffer-group 0 truncated packets");
8090 	T4_PORTSTAT(rx_trunc1, "# of buffer-group 1 truncated packets");
8091 	T4_PORTSTAT(rx_trunc2, "# of buffer-group 2 truncated packets");
8092 	T4_PORTSTAT(rx_trunc3, "# of buffer-group 3 truncated packets");
8093 
8094 #undef T4_REGSTAT
8095 #undef T4_PORTSTAT
8096 }
8097 
8098 static int
8099 sysctl_int_array(SYSCTL_HANDLER_ARGS)
8100 {
8101 	int rc, *i, space = 0;
8102 	struct sbuf sb;
8103 
8104 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
8105 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
8106 		if (space)
8107 			sbuf_printf(&sb, " ");
8108 		sbuf_printf(&sb, "%d", *i);
8109 		space = 1;
8110 	}
8111 	rc = sbuf_finish(&sb);
8112 	sbuf_delete(&sb);
8113 	return (rc);
8114 }
8115 
8116 static int
8117 sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS)
8118 {
8119 	int rc;
8120 	struct sbuf *sb;
8121 
8122 	rc = sysctl_wire_old_buffer(req, 0);
8123 	if (rc != 0)
8124 		return(rc);
8125 
8126 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
8127 	if (sb == NULL)
8128 		return (ENOMEM);
8129 
8130 	sbuf_printf(sb, "%b", *(uint8_t *)(uintptr_t)arg2, (char *)arg1);
8131 	rc = sbuf_finish(sb);
8132 	sbuf_delete(sb);
8133 
8134 	return (rc);
8135 }
8136 
8137 static int
8138 sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS)
8139 {
8140 	int rc;
8141 	struct sbuf *sb;
8142 
8143 	rc = sysctl_wire_old_buffer(req, 0);
8144 	if (rc != 0)
8145 		return(rc);
8146 
8147 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
8148 	if (sb == NULL)
8149 		return (ENOMEM);
8150 
8151 	sbuf_printf(sb, "%b", *(uint16_t *)(uintptr_t)arg2, (char *)arg1);
8152 	rc = sbuf_finish(sb);
8153 	sbuf_delete(sb);
8154 
8155 	return (rc);
8156 }
8157 
8158 static int
8159 sysctl_btphy(SYSCTL_HANDLER_ARGS)
8160 {
8161 	struct port_info *pi = arg1;
8162 	int op = arg2;
8163 	struct adapter *sc = pi->adapter;
8164 	u_int v;
8165 	int rc;
8166 
8167 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
8168 	if (rc)
8169 		return (rc);
8170 	if (hw_off_limits(sc))
8171 		rc = ENXIO;
8172 	else {
8173 		/* XXX: magic numbers */
8174 		rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e,
8175 		    op ? 0x20 : 0xc820, &v);
8176 	}
8177 	end_synchronized_op(sc, 0);
8178 	if (rc)
8179 		return (rc);
8180 	if (op == 0)
8181 		v /= 256;
8182 
8183 	rc = sysctl_handle_int(oidp, &v, 0, req);
8184 	return (rc);
8185 }
8186 
8187 static int
8188 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
8189 {
8190 	struct vi_info *vi = arg1;
8191 	int rc, val;
8192 
8193 	val = vi->rsrv_noflowq;
8194 	rc = sysctl_handle_int(oidp, &val, 0, req);
8195 	if (rc != 0 || req->newptr == NULL)
8196 		return (rc);
8197 
8198 	if ((val >= 1) && (vi->ntxq > 1))
8199 		vi->rsrv_noflowq = 1;
8200 	else
8201 		vi->rsrv_noflowq = 0;
8202 
8203 	return (rc);
8204 }
8205 
8206 static int
8207 sysctl_tx_vm_wr(SYSCTL_HANDLER_ARGS)
8208 {
8209 	struct vi_info *vi = arg1;
8210 	struct adapter *sc = vi->adapter;
8211 	int rc, val, i;
8212 
8213 	MPASS(!(sc->flags & IS_VF));
8214 
8215 	val = vi->flags & TX_USES_VM_WR ? 1 : 0;
8216 	rc = sysctl_handle_int(oidp, &val, 0, req);
8217 	if (rc != 0 || req->newptr == NULL)
8218 		return (rc);
8219 
8220 	if (val != 0 && val != 1)
8221 		return (EINVAL);
8222 
8223 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8224 	    "t4txvm");
8225 	if (rc)
8226 		return (rc);
8227 	if (hw_off_limits(sc))
8228 		rc = ENXIO;
8229 	else if (if_getdrvflags(vi->ifp) & IFF_DRV_RUNNING) {
8230 		/*
8231 		 * We don't want parse_pkt to run with one setting (VF or PF)
8232 		 * and then eth_tx to see a different setting but still use
8233 		 * stale information calculated by parse_pkt.
8234 		 */
8235 		rc = EBUSY;
8236 	} else {
8237 		struct port_info *pi = vi->pi;
8238 		struct sge_txq *txq;
8239 		uint32_t ctrl0;
8240 		uint8_t npkt = sc->params.max_pkts_per_eth_tx_pkts_wr;
8241 
8242 		if (val) {
8243 			vi->flags |= TX_USES_VM_WR;
8244 			if_sethwtsomaxsegcount(vi->ifp, TX_SGL_SEGS_VM_TSO);
8245 			ctrl0 = htobe32(V_TXPKT_OPCODE(CPL_TX_PKT_XT) |
8246 			    V_TXPKT_INTF(pi->tx_chan));
8247 			if (!(sc->flags & IS_VF))
8248 				npkt--;
8249 		} else {
8250 			vi->flags &= ~TX_USES_VM_WR;
8251 			if_sethwtsomaxsegcount(vi->ifp, TX_SGL_SEGS_TSO);
8252 			ctrl0 = htobe32(V_TXPKT_OPCODE(CPL_TX_PKT_XT) |
8253 			    V_TXPKT_INTF(pi->tx_chan) | V_TXPKT_PF(sc->pf) |
8254 			    V_TXPKT_VF(vi->vin) | V_TXPKT_VF_VLD(vi->vfvld));
8255 		}
8256 		for_each_txq(vi, i, txq) {
8257 			txq->cpl_ctrl0 = ctrl0;
8258 			txq->txp.max_npkt = npkt;
8259 		}
8260 	}
8261 	end_synchronized_op(sc, LOCK_HELD);
8262 	return (rc);
8263 }
8264 
8265 static int
8266 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
8267 {
8268 	struct vi_info *vi = arg1;
8269 	struct adapter *sc = vi->adapter;
8270 	int idx, rc, i;
8271 	struct sge_rxq *rxq;
8272 	uint8_t v;
8273 
8274 	idx = vi->tmr_idx;
8275 
8276 	rc = sysctl_handle_int(oidp, &idx, 0, req);
8277 	if (rc != 0 || req->newptr == NULL)
8278 		return (rc);
8279 
8280 	if (idx < 0 || idx >= SGE_NTIMERS)
8281 		return (EINVAL);
8282 
8283 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8284 	    "t4tmr");
8285 	if (rc)
8286 		return (rc);
8287 
8288 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
8289 	for_each_rxq(vi, i, rxq) {
8290 #ifdef atomic_store_rel_8
8291 		atomic_store_rel_8(&rxq->iq.intr_params, v);
8292 #else
8293 		rxq->iq.intr_params = v;
8294 #endif
8295 	}
8296 	vi->tmr_idx = idx;
8297 
8298 	end_synchronized_op(sc, LOCK_HELD);
8299 	return (0);
8300 }
8301 
8302 static int
8303 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
8304 {
8305 	struct vi_info *vi = arg1;
8306 	struct adapter *sc = vi->adapter;
8307 	int idx, rc;
8308 
8309 	idx = vi->pktc_idx;
8310 
8311 	rc = sysctl_handle_int(oidp, &idx, 0, req);
8312 	if (rc != 0 || req->newptr == NULL)
8313 		return (rc);
8314 
8315 	if (idx < -1 || idx >= SGE_NCOUNTERS)
8316 		return (EINVAL);
8317 
8318 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8319 	    "t4pktc");
8320 	if (rc)
8321 		return (rc);
8322 
8323 	if (vi->flags & VI_INIT_DONE)
8324 		rc = EBUSY; /* cannot be changed once the queues are created */
8325 	else
8326 		vi->pktc_idx = idx;
8327 
8328 	end_synchronized_op(sc, LOCK_HELD);
8329 	return (rc);
8330 }
8331 
8332 static int
8333 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
8334 {
8335 	struct vi_info *vi = arg1;
8336 	struct adapter *sc = vi->adapter;
8337 	int qsize, rc;
8338 
8339 	qsize = vi->qsize_rxq;
8340 
8341 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
8342 	if (rc != 0 || req->newptr == NULL)
8343 		return (rc);
8344 
8345 	if (qsize < 128 || (qsize & 7))
8346 		return (EINVAL);
8347 
8348 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8349 	    "t4rxqs");
8350 	if (rc)
8351 		return (rc);
8352 
8353 	if (vi->flags & VI_INIT_DONE)
8354 		rc = EBUSY; /* cannot be changed once the queues are created */
8355 	else
8356 		vi->qsize_rxq = qsize;
8357 
8358 	end_synchronized_op(sc, LOCK_HELD);
8359 	return (rc);
8360 }
8361 
8362 static int
8363 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
8364 {
8365 	struct vi_info *vi = arg1;
8366 	struct adapter *sc = vi->adapter;
8367 	int qsize, rc;
8368 
8369 	qsize = vi->qsize_txq;
8370 
8371 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
8372 	if (rc != 0 || req->newptr == NULL)
8373 		return (rc);
8374 
8375 	if (qsize < 128 || qsize > 65536)
8376 		return (EINVAL);
8377 
8378 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8379 	    "t4txqs");
8380 	if (rc)
8381 		return (rc);
8382 
8383 	if (vi->flags & VI_INIT_DONE)
8384 		rc = EBUSY; /* cannot be changed once the queues are created */
8385 	else
8386 		vi->qsize_txq = qsize;
8387 
8388 	end_synchronized_op(sc, LOCK_HELD);
8389 	return (rc);
8390 }
8391 
8392 static int
8393 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
8394 {
8395 	struct port_info *pi = arg1;
8396 	struct adapter *sc = pi->adapter;
8397 	struct link_config *lc = &pi->link_cfg;
8398 	int rc;
8399 
8400 	if (req->newptr == NULL) {
8401 		struct sbuf *sb;
8402 		static char *bits = "\20\1RX\2TX\3AUTO";
8403 
8404 		rc = sysctl_wire_old_buffer(req, 0);
8405 		if (rc != 0)
8406 			return(rc);
8407 
8408 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
8409 		if (sb == NULL)
8410 			return (ENOMEM);
8411 
8412 		if (lc->link_ok) {
8413 			sbuf_printf(sb, "%b", (lc->fc & (PAUSE_TX | PAUSE_RX)) |
8414 			    (lc->requested_fc & PAUSE_AUTONEG), bits);
8415 		} else {
8416 			sbuf_printf(sb, "%b", lc->requested_fc & (PAUSE_TX |
8417 			    PAUSE_RX | PAUSE_AUTONEG), bits);
8418 		}
8419 		rc = sbuf_finish(sb);
8420 		sbuf_delete(sb);
8421 	} else {
8422 		char s[2];
8423 		int n;
8424 
8425 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX |
8426 		    PAUSE_AUTONEG));
8427 		s[1] = 0;
8428 
8429 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
8430 		if (rc != 0)
8431 			return(rc);
8432 
8433 		if (s[1] != 0)
8434 			return (EINVAL);
8435 		if (s[0] < '0' || s[0] > '9')
8436 			return (EINVAL);	/* not a number */
8437 		n = s[0] - '0';
8438 		if (n & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG))
8439 			return (EINVAL);	/* some other bit is set too */
8440 
8441 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
8442 		    "t4PAUSE");
8443 		if (rc)
8444 			return (rc);
8445 		if (!hw_off_limits(sc)) {
8446 			PORT_LOCK(pi);
8447 			lc->requested_fc = n;
8448 			fixup_link_config(pi);
8449 			if (pi->up_vis > 0)
8450 				rc = apply_link_config(pi);
8451 			set_current_media(pi);
8452 			PORT_UNLOCK(pi);
8453 		}
8454 		end_synchronized_op(sc, 0);
8455 	}
8456 
8457 	return (rc);
8458 }
8459 
8460 static int
8461 sysctl_link_fec(SYSCTL_HANDLER_ARGS)
8462 {
8463 	struct port_info *pi = arg1;
8464 	struct link_config *lc = &pi->link_cfg;
8465 	int rc;
8466 	struct sbuf *sb;
8467 	static char *bits = "\20\1RS-FEC\2FC-FEC\3NO-FEC\4RSVD1\5RSVD2";
8468 
8469 	rc = sysctl_wire_old_buffer(req, 0);
8470 	if (rc != 0)
8471 		return(rc);
8472 
8473 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
8474 	if (sb == NULL)
8475 		return (ENOMEM);
8476 	if (lc->link_ok)
8477 		sbuf_printf(sb, "%b", lc->fec, bits);
8478 	else
8479 		sbuf_printf(sb, "no link");
8480 	rc = sbuf_finish(sb);
8481 	sbuf_delete(sb);
8482 
8483 	return (rc);
8484 }
8485 
8486 static int
8487 sysctl_requested_fec(SYSCTL_HANDLER_ARGS)
8488 {
8489 	struct port_info *pi = arg1;
8490 	struct adapter *sc = pi->adapter;
8491 	struct link_config *lc = &pi->link_cfg;
8492 	int rc;
8493 	int8_t old;
8494 
8495 	if (req->newptr == NULL) {
8496 		struct sbuf *sb;
8497 		static char *bits = "\20\1RS-FEC\2FC-FEC\3NO-FEC\4RSVD2"
8498 		    "\5RSVD3\6auto\7module";
8499 
8500 		rc = sysctl_wire_old_buffer(req, 0);
8501 		if (rc != 0)
8502 			return(rc);
8503 
8504 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
8505 		if (sb == NULL)
8506 			return (ENOMEM);
8507 
8508 		sbuf_printf(sb, "%b", lc->requested_fec, bits);
8509 		rc = sbuf_finish(sb);
8510 		sbuf_delete(sb);
8511 	} else {
8512 		char s[8];
8513 		int n;
8514 
8515 		snprintf(s, sizeof(s), "%d",
8516 		    lc->requested_fec == FEC_AUTO ? -1 :
8517 		    lc->requested_fec & (M_FW_PORT_CAP32_FEC | FEC_MODULE));
8518 
8519 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
8520 		if (rc != 0)
8521 			return(rc);
8522 
8523 		n = strtol(&s[0], NULL, 0);
8524 		if (n < 0 || n & FEC_AUTO)
8525 			n = FEC_AUTO;
8526 		else if (n & ~(M_FW_PORT_CAP32_FEC | FEC_MODULE))
8527 			return (EINVAL);/* some other bit is set too */
8528 
8529 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
8530 		    "t4reqf");
8531 		if (rc)
8532 			return (rc);
8533 		PORT_LOCK(pi);
8534 		old = lc->requested_fec;
8535 		if (n == FEC_AUTO)
8536 			lc->requested_fec = FEC_AUTO;
8537 		else if (n == 0 || n == FEC_NONE)
8538 			lc->requested_fec = FEC_NONE;
8539 		else {
8540 			if ((lc->pcaps |
8541 			    V_FW_PORT_CAP32_FEC(n & M_FW_PORT_CAP32_FEC)) !=
8542 			    lc->pcaps) {
8543 				rc = ENOTSUP;
8544 				goto done;
8545 			}
8546 			lc->requested_fec = n & (M_FW_PORT_CAP32_FEC |
8547 			    FEC_MODULE);
8548 		}
8549 		if (!hw_off_limits(sc)) {
8550 			fixup_link_config(pi);
8551 			if (pi->up_vis > 0) {
8552 				rc = apply_link_config(pi);
8553 				if (rc != 0) {
8554 					lc->requested_fec = old;
8555 					if (rc == FW_EPROTO)
8556 						rc = ENOTSUP;
8557 				}
8558 			}
8559 		}
8560 done:
8561 		PORT_UNLOCK(pi);
8562 		end_synchronized_op(sc, 0);
8563 	}
8564 
8565 	return (rc);
8566 }
8567 
8568 static int
8569 sysctl_module_fec(SYSCTL_HANDLER_ARGS)
8570 {
8571 	struct port_info *pi = arg1;
8572 	struct adapter *sc = pi->adapter;
8573 	struct link_config *lc = &pi->link_cfg;
8574 	int rc;
8575 	int8_t fec;
8576 	struct sbuf *sb;
8577 	static char *bits = "\20\1RS-FEC\2FC-FEC\3NO-FEC\4RSVD2\5RSVD3";
8578 
8579 	rc = sysctl_wire_old_buffer(req, 0);
8580 	if (rc != 0)
8581 		return (rc);
8582 
8583 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
8584 	if (sb == NULL)
8585 		return (ENOMEM);
8586 
8587 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mfec") != 0) {
8588 		rc = EBUSY;
8589 		goto done;
8590 	}
8591 	if (hw_off_limits(sc)) {
8592 		rc = ENXIO;
8593 		goto done;
8594 	}
8595 	PORT_LOCK(pi);
8596 	if (pi->up_vis == 0) {
8597 		/*
8598 		 * If all the interfaces are administratively down the firmware
8599 		 * does not report transceiver changes.  Refresh port info here.
8600 		 * This is the only reason we have a synchronized op in this
8601 		 * function.  Just PORT_LOCK would have been enough otherwise.
8602 		 */
8603 		t4_update_port_info(pi);
8604 	}
8605 
8606 	fec = lc->fec_hint;
8607 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE ||
8608 	    !fec_supported(lc->pcaps)) {
8609 		sbuf_printf(sb, "n/a");
8610 	} else {
8611 		if (fec == 0)
8612 			fec = FEC_NONE;
8613 		sbuf_printf(sb, "%b", fec & M_FW_PORT_CAP32_FEC, bits);
8614 	}
8615 	rc = sbuf_finish(sb);
8616 	PORT_UNLOCK(pi);
8617 done:
8618 	sbuf_delete(sb);
8619 	end_synchronized_op(sc, 0);
8620 
8621 	return (rc);
8622 }
8623 
8624 static int
8625 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
8626 {
8627 	struct port_info *pi = arg1;
8628 	struct adapter *sc = pi->adapter;
8629 	struct link_config *lc = &pi->link_cfg;
8630 	int rc, val;
8631 
8632 	if (lc->pcaps & FW_PORT_CAP32_ANEG)
8633 		val = lc->requested_aneg == AUTONEG_DISABLE ? 0 : 1;
8634 	else
8635 		val = -1;
8636 	rc = sysctl_handle_int(oidp, &val, 0, req);
8637 	if (rc != 0 || req->newptr == NULL)
8638 		return (rc);
8639 	if (val == 0)
8640 		val = AUTONEG_DISABLE;
8641 	else if (val == 1)
8642 		val = AUTONEG_ENABLE;
8643 	else
8644 		val = AUTONEG_AUTO;
8645 
8646 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
8647 	    "t4aneg");
8648 	if (rc)
8649 		return (rc);
8650 	PORT_LOCK(pi);
8651 	if (val == AUTONEG_ENABLE && !(lc->pcaps & FW_PORT_CAP32_ANEG)) {
8652 		rc = ENOTSUP;
8653 		goto done;
8654 	}
8655 	lc->requested_aneg = val;
8656 	if (!hw_off_limits(sc)) {
8657 		fixup_link_config(pi);
8658 		if (pi->up_vis > 0)
8659 			rc = apply_link_config(pi);
8660 		set_current_media(pi);
8661 	}
8662 done:
8663 	PORT_UNLOCK(pi);
8664 	end_synchronized_op(sc, 0);
8665 	return (rc);
8666 }
8667 
8668 static int
8669 sysctl_force_fec(SYSCTL_HANDLER_ARGS)
8670 {
8671 	struct port_info *pi = arg1;
8672 	struct adapter *sc = pi->adapter;
8673 	struct link_config *lc = &pi->link_cfg;
8674 	int rc, val;
8675 
8676 	val = lc->force_fec;
8677 	MPASS(val >= -1 && val <= 1);
8678 	rc = sysctl_handle_int(oidp, &val, 0, req);
8679 	if (rc != 0 || req->newptr == NULL)
8680 		return (rc);
8681 	if (!(lc->pcaps & FW_PORT_CAP32_FORCE_FEC))
8682 		return (ENOTSUP);
8683 	if (val < -1 || val > 1)
8684 		return (EINVAL);
8685 
8686 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4ff");
8687 	if (rc)
8688 		return (rc);
8689 	PORT_LOCK(pi);
8690 	lc->force_fec = val;
8691 	if (!hw_off_limits(sc)) {
8692 		fixup_link_config(pi);
8693 		if (pi->up_vis > 0)
8694 			rc = apply_link_config(pi);
8695 	}
8696 	PORT_UNLOCK(pi);
8697 	end_synchronized_op(sc, 0);
8698 	return (rc);
8699 }
8700 
8701 static int
8702 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
8703 {
8704 	struct adapter *sc = arg1;
8705 	int rc, reg = arg2;
8706 	uint64_t val;
8707 
8708 	mtx_lock(&sc->reg_lock);
8709 	if (hw_off_limits(sc))
8710 		rc = ENXIO;
8711 	else {
8712 		rc = 0;
8713 		val = t4_read_reg64(sc, reg);
8714 	}
8715 	mtx_unlock(&sc->reg_lock);
8716 	if (rc == 0)
8717 		rc = sysctl_handle_64(oidp, &val, 0, req);
8718 	return (rc);
8719 }
8720 
8721 static int
8722 sysctl_temperature(SYSCTL_HANDLER_ARGS)
8723 {
8724 	struct adapter *sc = arg1;
8725 	int rc, t;
8726 	uint32_t param, val;
8727 
8728 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
8729 	if (rc)
8730 		return (rc);
8731 	if (hw_off_limits(sc))
8732 		rc = ENXIO;
8733 	else {
8734 		param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
8735 		    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
8736 		    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
8737 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
8738 	}
8739 	end_synchronized_op(sc, 0);
8740 	if (rc)
8741 		return (rc);
8742 
8743 	/* unknown is returned as 0 but we display -1 in that case */
8744 	t = val == 0 ? -1 : val;
8745 
8746 	rc = sysctl_handle_int(oidp, &t, 0, req);
8747 	return (rc);
8748 }
8749 
8750 static int
8751 sysctl_vdd(SYSCTL_HANDLER_ARGS)
8752 {
8753 	struct adapter *sc = arg1;
8754 	int rc;
8755 	uint32_t param, val;
8756 
8757 	if (sc->params.core_vdd == 0) {
8758 		rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8759 		    "t4vdd");
8760 		if (rc)
8761 			return (rc);
8762 		if (hw_off_limits(sc))
8763 			rc = ENXIO;
8764 		else {
8765 			param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
8766 			    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
8767 			    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
8768 			rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1,
8769 			    &param, &val);
8770 		}
8771 		end_synchronized_op(sc, 0);
8772 		if (rc)
8773 			return (rc);
8774 		sc->params.core_vdd = val;
8775 	}
8776 
8777 	return (sysctl_handle_int(oidp, &sc->params.core_vdd, 0, req));
8778 }
8779 
8780 static int
8781 sysctl_reset_sensor(SYSCTL_HANDLER_ARGS)
8782 {
8783 	struct adapter *sc = arg1;
8784 	int rc, v;
8785 	uint32_t param, val;
8786 
8787 	v = sc->sensor_resets;
8788 	rc = sysctl_handle_int(oidp, &v, 0, req);
8789 	if (rc != 0 || req->newptr == NULL || v <= 0)
8790 		return (rc);
8791 
8792 	if (sc->params.fw_vers < FW_VERSION32(1, 24, 7, 0) ||
8793 	    chip_id(sc) < CHELSIO_T5)
8794 		return (ENOTSUP);
8795 
8796 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4srst");
8797 	if (rc)
8798 		return (rc);
8799 	if (hw_off_limits(sc))
8800 		rc = ENXIO;
8801 	else {
8802 		param = (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
8803 		    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
8804 		    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_RESET_TMP_SENSOR));
8805 		val = 1;
8806 		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
8807 	}
8808 	end_synchronized_op(sc, 0);
8809 	if (rc == 0)
8810 		sc->sensor_resets++;
8811 	return (rc);
8812 }
8813 
8814 static int
8815 sysctl_loadavg(SYSCTL_HANDLER_ARGS)
8816 {
8817 	struct adapter *sc = arg1;
8818 	struct sbuf *sb;
8819 	int rc;
8820 	uint32_t param, val;
8821 
8822 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4lavg");
8823 	if (rc)
8824 		return (rc);
8825 	if (hw_off_limits(sc))
8826 		rc = ENXIO;
8827 	else {
8828 		param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
8829 		    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_LOAD);
8830 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
8831 	}
8832 	end_synchronized_op(sc, 0);
8833 	if (rc)
8834 		return (rc);
8835 
8836 	rc = sysctl_wire_old_buffer(req, 0);
8837 	if (rc != 0)
8838 		return (rc);
8839 
8840 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8841 	if (sb == NULL)
8842 		return (ENOMEM);
8843 
8844 	if (val == 0xffffffff) {
8845 		/* Only debug and custom firmwares report load averages. */
8846 		sbuf_printf(sb, "not available");
8847 	} else {
8848 		sbuf_printf(sb, "%d %d %d", val & 0xff, (val >> 8) & 0xff,
8849 		    (val >> 16) & 0xff);
8850 	}
8851 	rc = sbuf_finish(sb);
8852 	sbuf_delete(sb);
8853 
8854 	return (rc);
8855 }
8856 
8857 static int
8858 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
8859 {
8860 	struct adapter *sc = arg1;
8861 	struct sbuf *sb;
8862 	int rc, i;
8863 	uint16_t incr[NMTUS][NCCTRL_WIN];
8864 	static const char *dec_fac[] = {
8865 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
8866 		"0.9375"
8867 	};
8868 
8869 	rc = sysctl_wire_old_buffer(req, 0);
8870 	if (rc != 0)
8871 		return (rc);
8872 
8873 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8874 	if (sb == NULL)
8875 		return (ENOMEM);
8876 
8877 	mtx_lock(&sc->reg_lock);
8878 	if (hw_off_limits(sc))
8879 		rc = ENXIO;
8880 	else
8881 		t4_read_cong_tbl(sc, incr);
8882 	mtx_unlock(&sc->reg_lock);
8883 	if (rc)
8884 		goto done;
8885 
8886 	for (i = 0; i < NCCTRL_WIN; ++i) {
8887 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
8888 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
8889 		    incr[5][i], incr[6][i], incr[7][i]);
8890 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
8891 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
8892 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
8893 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
8894 	}
8895 
8896 	rc = sbuf_finish(sb);
8897 done:
8898 	sbuf_delete(sb);
8899 	return (rc);
8900 }
8901 
8902 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
8903 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
8904 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
8905 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
8906 };
8907 
8908 static int
8909 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
8910 {
8911 	struct adapter *sc = arg1;
8912 	struct sbuf *sb;
8913 	int rc, i, n, qid = arg2;
8914 	uint32_t *buf, *p;
8915 	char *qtype;
8916 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
8917 
8918 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
8919 	    ("%s: bad qid %d\n", __func__, qid));
8920 
8921 	if (qid < CIM_NUM_IBQ) {
8922 		/* inbound queue */
8923 		qtype = "IBQ";
8924 		n = 4 * CIM_IBQ_SIZE;
8925 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
8926 		mtx_lock(&sc->reg_lock);
8927 		if (hw_off_limits(sc))
8928 			rc = -ENXIO;
8929 		else
8930 			rc = t4_read_cim_ibq(sc, qid, buf, n);
8931 		mtx_unlock(&sc->reg_lock);
8932 	} else {
8933 		/* outbound queue */
8934 		qtype = "OBQ";
8935 		qid -= CIM_NUM_IBQ;
8936 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
8937 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
8938 		mtx_lock(&sc->reg_lock);
8939 		if (hw_off_limits(sc))
8940 			rc = -ENXIO;
8941 		else
8942 			rc = t4_read_cim_obq(sc, qid, buf, n);
8943 		mtx_unlock(&sc->reg_lock);
8944 	}
8945 
8946 	if (rc < 0) {
8947 		rc = -rc;
8948 		goto done;
8949 	}
8950 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
8951 
8952 	rc = sysctl_wire_old_buffer(req, 0);
8953 	if (rc != 0)
8954 		goto done;
8955 
8956 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
8957 	if (sb == NULL) {
8958 		rc = ENOMEM;
8959 		goto done;
8960 	}
8961 
8962 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
8963 	for (i = 0, p = buf; i < n; i += 16, p += 4)
8964 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
8965 		    p[2], p[3]);
8966 
8967 	rc = sbuf_finish(sb);
8968 	sbuf_delete(sb);
8969 done:
8970 	free(buf, M_CXGBE);
8971 	return (rc);
8972 }
8973 
8974 static void
8975 sbuf_cim_la4(struct adapter *sc, struct sbuf *sb, uint32_t *buf, uint32_t cfg)
8976 {
8977 	uint32_t *p;
8978 
8979 	sbuf_printf(sb, "Status   Data      PC%s",
8980 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
8981 	    "     LS0Stat  LS0Addr             LS0Data");
8982 
8983 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
8984 		if (cfg & F_UPDBGLACAPTPCONLY) {
8985 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
8986 			    p[6], p[7]);
8987 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
8988 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
8989 			    p[4] & 0xff, p[5] >> 8);
8990 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
8991 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
8992 			    p[1] & 0xf, p[2] >> 4);
8993 		} else {
8994 			sbuf_printf(sb,
8995 			    "\n  %02x   %x%07x %x%07x %08x %08x "
8996 			    "%08x%08x%08x%08x",
8997 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
8998 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
8999 			    p[6], p[7]);
9000 		}
9001 	}
9002 }
9003 
9004 static void
9005 sbuf_cim_la6(struct adapter *sc, struct sbuf *sb, uint32_t *buf, uint32_t cfg)
9006 {
9007 	uint32_t *p;
9008 
9009 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
9010 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
9011 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
9012 
9013 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
9014 		if (cfg & F_UPDBGLACAPTPCONLY) {
9015 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
9016 			    p[3] & 0xff, p[2], p[1], p[0]);
9017 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
9018 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
9019 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
9020 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
9021 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
9022 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
9023 			    p[6] >> 16);
9024 		} else {
9025 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
9026 			    "%08x %08x %08x %08x %08x %08x",
9027 			    (p[9] >> 16) & 0xff,
9028 			    p[9] & 0xffff, p[8] >> 16,
9029 			    p[8] & 0xffff, p[7] >> 16,
9030 			    p[7] & 0xffff, p[6] >> 16,
9031 			    p[2], p[1], p[0], p[5], p[4], p[3]);
9032 		}
9033 	}
9034 }
9035 
9036 static int
9037 sbuf_cim_la(struct adapter *sc, struct sbuf *sb, int flags)
9038 {
9039 	uint32_t cfg, *buf;
9040 	int rc;
9041 
9042 	MPASS(flags == M_WAITOK || flags == M_NOWAIT);
9043 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
9044 	    M_ZERO | flags);
9045 	if (buf == NULL)
9046 		return (ENOMEM);
9047 
9048 	mtx_lock(&sc->reg_lock);
9049 	if (hw_off_limits(sc))
9050 		rc = ENXIO;
9051 	else {
9052 		rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
9053 		if (rc == 0)
9054 			rc = -t4_cim_read_la(sc, buf, NULL);
9055 	}
9056 	mtx_unlock(&sc->reg_lock);
9057 	if (rc == 0) {
9058 		if (chip_id(sc) < CHELSIO_T6)
9059 			sbuf_cim_la4(sc, sb, buf, cfg);
9060 		else
9061 			sbuf_cim_la6(sc, sb, buf, cfg);
9062 	}
9063 	free(buf, M_CXGBE);
9064 	return (rc);
9065 }
9066 
9067 static int
9068 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
9069 {
9070 	struct adapter *sc = arg1;
9071 	struct sbuf *sb;
9072 	int rc;
9073 
9074 	rc = sysctl_wire_old_buffer(req, 0);
9075 	if (rc != 0)
9076 		return (rc);
9077 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9078 	if (sb == NULL)
9079 		return (ENOMEM);
9080 
9081 	rc = sbuf_cim_la(sc, sb, M_WAITOK);
9082 	if (rc == 0)
9083 		rc = sbuf_finish(sb);
9084 	sbuf_delete(sb);
9085 	return (rc);
9086 }
9087 
9088 static void
9089 dump_cim_regs(struct adapter *sc)
9090 {
9091 	log(LOG_DEBUG, "%s: CIM debug regs1 %08x %08x %08x %08x %08x\n",
9092 	    device_get_nameunit(sc->dev),
9093 	    t4_read_reg(sc, A_EDC_H_BIST_USER_WDATA0),
9094 	    t4_read_reg(sc, A_EDC_H_BIST_USER_WDATA1),
9095 	    t4_read_reg(sc, A_EDC_H_BIST_USER_WDATA2),
9096 	    t4_read_reg(sc, A_EDC_H_BIST_DATA_PATTERN),
9097 	    t4_read_reg(sc, A_EDC_H_BIST_STATUS_RDATA));
9098 	log(LOG_DEBUG, "%s: CIM debug regs2 %08x %08x %08x %08x %08x\n",
9099 	    device_get_nameunit(sc->dev),
9100 	    t4_read_reg(sc, A_EDC_H_BIST_USER_WDATA0),
9101 	    t4_read_reg(sc, A_EDC_H_BIST_USER_WDATA1),
9102 	    t4_read_reg(sc, A_EDC_H_BIST_USER_WDATA0 + 0x800),
9103 	    t4_read_reg(sc, A_EDC_H_BIST_USER_WDATA1 + 0x800),
9104 	    t4_read_reg(sc, A_EDC_H_BIST_CMD_LEN));
9105 }
9106 
9107 static void
9108 dump_cimla(struct adapter *sc)
9109 {
9110 	struct sbuf sb;
9111 	int rc;
9112 
9113 	if (sbuf_new(&sb, NULL, 4096, SBUF_AUTOEXTEND) != &sb) {
9114 		log(LOG_DEBUG, "%s: failed to generate CIM LA dump.\n",
9115 		    device_get_nameunit(sc->dev));
9116 		return;
9117 	}
9118 	rc = sbuf_cim_la(sc, &sb, M_WAITOK);
9119 	if (rc == 0) {
9120 		rc = sbuf_finish(&sb);
9121 		if (rc == 0) {
9122 			log(LOG_DEBUG, "%s: CIM LA dump follows.\n%s\n",
9123 			    device_get_nameunit(sc->dev), sbuf_data(&sb));
9124 		}
9125 	}
9126 	sbuf_delete(&sb);
9127 }
9128 
9129 void
9130 t4_os_cim_err(struct adapter *sc)
9131 {
9132 	atomic_set_int(&sc->error_flags, ADAP_CIM_ERR);
9133 }
9134 
9135 static int
9136 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
9137 {
9138 	struct adapter *sc = arg1;
9139 	u_int i;
9140 	struct sbuf *sb;
9141 	uint32_t *buf, *p;
9142 	int rc;
9143 
9144 	rc = sysctl_wire_old_buffer(req, 0);
9145 	if (rc != 0)
9146 		return (rc);
9147 
9148 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9149 	if (sb == NULL)
9150 		return (ENOMEM);
9151 
9152 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
9153 	    M_ZERO | M_WAITOK);
9154 
9155 	mtx_lock(&sc->reg_lock);
9156 	if (hw_off_limits(sc))
9157 		rc = ENXIO;
9158 	else
9159 		t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
9160 	mtx_unlock(&sc->reg_lock);
9161 	if (rc)
9162 		goto done;
9163 
9164 	p = buf;
9165 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
9166 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
9167 		    p[1], p[0]);
9168 	}
9169 
9170 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
9171 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
9172 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
9173 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
9174 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
9175 		    (p[1] >> 2) | ((p[2] & 3) << 30),
9176 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
9177 		    p[0] & 1);
9178 	}
9179 	rc = sbuf_finish(sb);
9180 done:
9181 	sbuf_delete(sb);
9182 	free(buf, M_CXGBE);
9183 	return (rc);
9184 }
9185 
9186 static int
9187 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
9188 {
9189 	struct adapter *sc = arg1;
9190 	u_int i;
9191 	struct sbuf *sb;
9192 	uint32_t *buf, *p;
9193 	int rc;
9194 
9195 	rc = sysctl_wire_old_buffer(req, 0);
9196 	if (rc != 0)
9197 		return (rc);
9198 
9199 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9200 	if (sb == NULL)
9201 		return (ENOMEM);
9202 
9203 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
9204 	    M_ZERO | M_WAITOK);
9205 
9206 	mtx_lock(&sc->reg_lock);
9207 	if (hw_off_limits(sc))
9208 		rc = ENXIO;
9209 	else
9210 		t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
9211 	mtx_unlock(&sc->reg_lock);
9212 	if (rc)
9213 		goto done;
9214 
9215 	p = buf;
9216 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
9217 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
9218 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
9219 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
9220 		    p[4], p[3], p[2], p[1], p[0]);
9221 	}
9222 
9223 	sbuf_printf(sb, "\n\nCntl ID               Data");
9224 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
9225 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
9226 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
9227 	}
9228 
9229 	rc = sbuf_finish(sb);
9230 done:
9231 	sbuf_delete(sb);
9232 	free(buf, M_CXGBE);
9233 	return (rc);
9234 }
9235 
9236 static int
9237 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
9238 {
9239 	struct adapter *sc = arg1;
9240 	struct sbuf *sb;
9241 	int rc, i;
9242 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
9243 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
9244 	uint16_t thres[CIM_NUM_IBQ];
9245 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
9246 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
9247 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
9248 
9249 	cim_num_obq = sc->chip_params->cim_num_obq;
9250 	if (is_t4(sc)) {
9251 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
9252 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
9253 	} else {
9254 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
9255 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
9256 	}
9257 	nq = CIM_NUM_IBQ + cim_num_obq;
9258 
9259 	mtx_lock(&sc->reg_lock);
9260 	if (hw_off_limits(sc))
9261 		rc = ENXIO;
9262 	else {
9263 		rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
9264 		if (rc == 0) {
9265 			rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq,
9266 			    obq_wr);
9267 			if (rc == 0)
9268 				t4_read_cimq_cfg(sc, base, size, thres);
9269 		}
9270 	}
9271 	mtx_unlock(&sc->reg_lock);
9272 	if (rc)
9273 		return (rc);
9274 
9275 	rc = sysctl_wire_old_buffer(req, 0);
9276 	if (rc != 0)
9277 		return (rc);
9278 
9279 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
9280 	if (sb == NULL)
9281 		return (ENOMEM);
9282 
9283 	sbuf_printf(sb,
9284 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
9285 
9286 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
9287 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
9288 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
9289 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
9290 		    G_QUEREMFLITS(p[2]) * 16);
9291 	for ( ; i < nq; i++, p += 4, wr += 2)
9292 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
9293 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
9294 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
9295 		    G_QUEREMFLITS(p[2]) * 16);
9296 
9297 	rc = sbuf_finish(sb);
9298 	sbuf_delete(sb);
9299 
9300 	return (rc);
9301 }
9302 
9303 static int
9304 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
9305 {
9306 	struct adapter *sc = arg1;
9307 	struct sbuf *sb;
9308 	int rc;
9309 	struct tp_cpl_stats stats;
9310 
9311 	rc = sysctl_wire_old_buffer(req, 0);
9312 	if (rc != 0)
9313 		return (rc);
9314 
9315 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
9316 	if (sb == NULL)
9317 		return (ENOMEM);
9318 
9319 	mtx_lock(&sc->reg_lock);
9320 	if (hw_off_limits(sc))
9321 		rc = ENXIO;
9322 	else
9323 		t4_tp_get_cpl_stats(sc, &stats, 0);
9324 	mtx_unlock(&sc->reg_lock);
9325 	if (rc)
9326 		goto done;
9327 
9328 	if (sc->chip_params->nchan > 2) {
9329 		sbuf_printf(sb, "                 channel 0  channel 1"
9330 		    "  channel 2  channel 3");
9331 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
9332 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
9333 		sbuf_printf(sb, "\nCPL responses:  %10u %10u %10u %10u",
9334 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
9335 	} else {
9336 		sbuf_printf(sb, "                 channel 0  channel 1");
9337 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
9338 		    stats.req[0], stats.req[1]);
9339 		sbuf_printf(sb, "\nCPL responses:  %10u %10u",
9340 		    stats.rsp[0], stats.rsp[1]);
9341 	}
9342 
9343 	rc = sbuf_finish(sb);
9344 done:
9345 	sbuf_delete(sb);
9346 	return (rc);
9347 }
9348 
9349 static int
9350 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
9351 {
9352 	struct adapter *sc = arg1;
9353 	struct sbuf *sb;
9354 	int rc;
9355 	struct tp_usm_stats stats;
9356 
9357 	rc = sysctl_wire_old_buffer(req, 0);
9358 	if (rc != 0)
9359 		return(rc);
9360 
9361 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
9362 	if (sb == NULL)
9363 		return (ENOMEM);
9364 
9365 	mtx_lock(&sc->reg_lock);
9366 	if (hw_off_limits(sc))
9367 		rc = ENXIO;
9368 	else
9369 		t4_get_usm_stats(sc, &stats, 1);
9370 	mtx_unlock(&sc->reg_lock);
9371 	if (rc == 0) {
9372 		sbuf_printf(sb, "Frames: %u\n", stats.frames);
9373 		sbuf_printf(sb, "Octets: %ju\n", stats.octets);
9374 		sbuf_printf(sb, "Drops:  %u", stats.drops);
9375 		rc = sbuf_finish(sb);
9376 	}
9377 	sbuf_delete(sb);
9378 
9379 	return (rc);
9380 }
9381 
9382 static int
9383 sysctl_tid_stats(SYSCTL_HANDLER_ARGS)
9384 {
9385 	struct adapter *sc = arg1;
9386 	struct sbuf *sb;
9387 	int rc;
9388 	struct tp_tid_stats stats;
9389 
9390 	rc = sysctl_wire_old_buffer(req, 0);
9391 	if (rc != 0)
9392 		return(rc);
9393 
9394 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
9395 	if (sb == NULL)
9396 		return (ENOMEM);
9397 
9398 	mtx_lock(&sc->reg_lock);
9399 	if (hw_off_limits(sc))
9400 		rc = ENXIO;
9401 	else
9402 		t4_tp_get_tid_stats(sc, &stats, 1);
9403 	mtx_unlock(&sc->reg_lock);
9404 	if (rc == 0) {
9405 		sbuf_printf(sb, "Delete:     %u\n", stats.del);
9406 		sbuf_printf(sb, "Invalidate: %u\n", stats.inv);
9407 		sbuf_printf(sb, "Active:     %u\n", stats.act);
9408 		sbuf_printf(sb, "Passive:    %u", stats.pas);
9409 		rc = sbuf_finish(sb);
9410 	}
9411 	sbuf_delete(sb);
9412 
9413 	return (rc);
9414 }
9415 
9416 static const char * const devlog_level_strings[] = {
9417 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
9418 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
9419 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
9420 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
9421 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
9422 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
9423 };
9424 
9425 static const char * const devlog_facility_strings[] = {
9426 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
9427 	[FW_DEVLOG_FACILITY_CF]		= "CF",
9428 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
9429 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
9430 	[FW_DEVLOG_FACILITY_RES]	= "RES",
9431 	[FW_DEVLOG_FACILITY_HW]		= "HW",
9432 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
9433 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
9434 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
9435 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
9436 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
9437 	[FW_DEVLOG_FACILITY_VI]		= "VI",
9438 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
9439 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
9440 	[FW_DEVLOG_FACILITY_TM]		= "TM",
9441 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
9442 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
9443 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
9444 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
9445 	[FW_DEVLOG_FACILITY_RI]		= "RI",
9446 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
9447 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
9448 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
9449 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
9450 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
9451 };
9452 
9453 static int
9454 sbuf_devlog(struct adapter *sc, struct sbuf *sb, int flags)
9455 {
9456 	int i, j, rc, nentries, first = 0;
9457 	struct devlog_params *dparams = &sc->params.devlog;
9458 	struct fw_devlog_e *buf, *e;
9459 	uint64_t ftstamp = UINT64_MAX;
9460 
9461 	if (dparams->addr == 0)
9462 		return (ENXIO);
9463 
9464 	MPASS(flags == M_WAITOK || flags == M_NOWAIT);
9465 	buf = malloc(dparams->size, M_CXGBE, M_ZERO | flags);
9466 	if (buf == NULL)
9467 		return (ENOMEM);
9468 
9469 	mtx_lock(&sc->reg_lock);
9470 	if (hw_off_limits(sc))
9471 		rc = ENXIO;
9472 	else
9473 		rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf,
9474 		    dparams->size);
9475 	mtx_unlock(&sc->reg_lock);
9476 	if (rc != 0)
9477 		goto done;
9478 
9479 	nentries = dparams->size / sizeof(struct fw_devlog_e);
9480 	for (i = 0; i < nentries; i++) {
9481 		e = &buf[i];
9482 
9483 		if (e->timestamp == 0)
9484 			break;	/* end */
9485 
9486 		e->timestamp = be64toh(e->timestamp);
9487 		e->seqno = be32toh(e->seqno);
9488 		for (j = 0; j < 8; j++)
9489 			e->params[j] = be32toh(e->params[j]);
9490 
9491 		if (e->timestamp < ftstamp) {
9492 			ftstamp = e->timestamp;
9493 			first = i;
9494 		}
9495 	}
9496 
9497 	if (buf[first].timestamp == 0)
9498 		goto done;	/* nothing in the log */
9499 
9500 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
9501 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
9502 
9503 	i = first;
9504 	do {
9505 		e = &buf[i];
9506 		if (e->timestamp == 0)
9507 			break;	/* end */
9508 
9509 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
9510 		    e->seqno, e->timestamp,
9511 		    (e->level < nitems(devlog_level_strings) ?
9512 			devlog_level_strings[e->level] : "UNKNOWN"),
9513 		    (e->facility < nitems(devlog_facility_strings) ?
9514 			devlog_facility_strings[e->facility] : "UNKNOWN"));
9515 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
9516 		    e->params[2], e->params[3], e->params[4],
9517 		    e->params[5], e->params[6], e->params[7]);
9518 
9519 		if (++i == nentries)
9520 			i = 0;
9521 	} while (i != first);
9522 done:
9523 	free(buf, M_CXGBE);
9524 	return (rc);
9525 }
9526 
9527 static int
9528 sysctl_devlog(SYSCTL_HANDLER_ARGS)
9529 {
9530 	struct adapter *sc = arg1;
9531 	int rc;
9532 	struct sbuf *sb;
9533 
9534 	rc = sysctl_wire_old_buffer(req, 0);
9535 	if (rc != 0)
9536 		return (rc);
9537 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9538 	if (sb == NULL)
9539 		return (ENOMEM);
9540 
9541 	rc = sbuf_devlog(sc, sb, M_WAITOK);
9542 	if (rc == 0)
9543 		rc = sbuf_finish(sb);
9544 	sbuf_delete(sb);
9545 	return (rc);
9546 }
9547 
9548 static void
9549 dump_devlog(struct adapter *sc)
9550 {
9551 	int rc;
9552 	struct sbuf sb;
9553 
9554 	if (sbuf_new(&sb, NULL, 4096, SBUF_AUTOEXTEND) != &sb) {
9555 		log(LOG_DEBUG, "%s: failed to generate devlog dump.\n",
9556 		    device_get_nameunit(sc->dev));
9557 		return;
9558 	}
9559 	rc = sbuf_devlog(sc, &sb, M_WAITOK);
9560 	if (rc == 0) {
9561 		rc = sbuf_finish(&sb);
9562 		if (rc == 0) {
9563 			log(LOG_DEBUG, "%s: device log follows.\n%s",
9564 			    device_get_nameunit(sc->dev), sbuf_data(&sb));
9565 		}
9566 	}
9567 	sbuf_delete(&sb);
9568 }
9569 
9570 static int
9571 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
9572 {
9573 	struct adapter *sc = arg1;
9574 	struct sbuf *sb;
9575 	int rc;
9576 	struct tp_fcoe_stats stats[MAX_NCHAN];
9577 	int i, nchan = sc->chip_params->nchan;
9578 
9579 	rc = sysctl_wire_old_buffer(req, 0);
9580 	if (rc != 0)
9581 		return (rc);
9582 
9583 	mtx_lock(&sc->reg_lock);
9584 	if (hw_off_limits(sc))
9585 		rc = ENXIO;
9586 	else {
9587 		for (i = 0; i < nchan; i++)
9588 			t4_get_fcoe_stats(sc, i, &stats[i], 1);
9589 	}
9590 	mtx_unlock(&sc->reg_lock);
9591 	if (rc != 0)
9592 		return (rc);
9593 
9594 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
9595 	if (sb == NULL)
9596 		return (ENOMEM);
9597 
9598 	if (nchan > 2) {
9599 		sbuf_printf(sb, "                   channel 0        channel 1"
9600 		    "        channel 2        channel 3");
9601 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
9602 		    stats[0].octets_ddp, stats[1].octets_ddp,
9603 		    stats[2].octets_ddp, stats[3].octets_ddp);
9604 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
9605 		    stats[0].frames_ddp, stats[1].frames_ddp,
9606 		    stats[2].frames_ddp, stats[3].frames_ddp);
9607 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
9608 		    stats[0].frames_drop, stats[1].frames_drop,
9609 		    stats[2].frames_drop, stats[3].frames_drop);
9610 	} else {
9611 		sbuf_printf(sb, "                   channel 0        channel 1");
9612 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
9613 		    stats[0].octets_ddp, stats[1].octets_ddp);
9614 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
9615 		    stats[0].frames_ddp, stats[1].frames_ddp);
9616 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
9617 		    stats[0].frames_drop, stats[1].frames_drop);
9618 	}
9619 
9620 	rc = sbuf_finish(sb);
9621 	sbuf_delete(sb);
9622 
9623 	return (rc);
9624 }
9625 
9626 static int
9627 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
9628 {
9629 	struct adapter *sc = arg1;
9630 	struct sbuf *sb;
9631 	int rc, i;
9632 	unsigned int map, kbps, ipg, mode;
9633 	unsigned int pace_tab[NTX_SCHED];
9634 
9635 	rc = sysctl_wire_old_buffer(req, 0);
9636 	if (rc != 0)
9637 		return (rc);
9638 
9639 	sb = sbuf_new_for_sysctl(NULL, NULL, 512, req);
9640 	if (sb == NULL)
9641 		return (ENOMEM);
9642 
9643 	mtx_lock(&sc->reg_lock);
9644 	if (hw_off_limits(sc)) {
9645 		rc = ENXIO;
9646 		goto done;
9647 	}
9648 
9649 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
9650 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
9651 	t4_read_pace_tbl(sc, pace_tab);
9652 
9653 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
9654 	    "Class IPG (0.1 ns)   Flow IPG (us)");
9655 
9656 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
9657 		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
9658 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
9659 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
9660 		if (kbps)
9661 			sbuf_printf(sb, "%9u     ", kbps);
9662 		else
9663 			sbuf_printf(sb, " disabled     ");
9664 
9665 		if (ipg)
9666 			sbuf_printf(sb, "%13u        ", ipg);
9667 		else
9668 			sbuf_printf(sb, "     disabled        ");
9669 
9670 		if (pace_tab[i])
9671 			sbuf_printf(sb, "%10u", pace_tab[i]);
9672 		else
9673 			sbuf_printf(sb, "  disabled");
9674 	}
9675 	rc = sbuf_finish(sb);
9676 done:
9677 	mtx_unlock(&sc->reg_lock);
9678 	sbuf_delete(sb);
9679 	return (rc);
9680 }
9681 
9682 static int
9683 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
9684 {
9685 	struct adapter *sc = arg1;
9686 	struct sbuf *sb;
9687 	int rc, i, j;
9688 	uint64_t *p0, *p1;
9689 	struct lb_port_stats s[2];
9690 	static const char *stat_name[] = {
9691 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
9692 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
9693 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
9694 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
9695 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
9696 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
9697 		"BG2FramesTrunc:", "BG3FramesTrunc:"
9698 	};
9699 
9700 	rc = sysctl_wire_old_buffer(req, 0);
9701 	if (rc != 0)
9702 		return (rc);
9703 
9704 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9705 	if (sb == NULL)
9706 		return (ENOMEM);
9707 
9708 	memset(s, 0, sizeof(s));
9709 
9710 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
9711 		mtx_lock(&sc->reg_lock);
9712 		if (hw_off_limits(sc))
9713 			rc = ENXIO;
9714 		else {
9715 			t4_get_lb_stats(sc, i, &s[0]);
9716 			t4_get_lb_stats(sc, i + 1, &s[1]);
9717 		}
9718 		mtx_unlock(&sc->reg_lock);
9719 		if (rc != 0)
9720 			break;
9721 
9722 		p0 = &s[0].octets;
9723 		p1 = &s[1].octets;
9724 		sbuf_printf(sb, "%s                       Loopback %u"
9725 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
9726 
9727 		for (j = 0; j < nitems(stat_name); j++)
9728 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
9729 				   *p0++, *p1++);
9730 	}
9731 
9732 	rc = sbuf_finish(sb);
9733 	sbuf_delete(sb);
9734 
9735 	return (rc);
9736 }
9737 
9738 static int
9739 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
9740 {
9741 	int rc = 0;
9742 	struct port_info *pi = arg1;
9743 	struct link_config *lc = &pi->link_cfg;
9744 	struct sbuf *sb;
9745 
9746 	rc = sysctl_wire_old_buffer(req, 0);
9747 	if (rc != 0)
9748 		return(rc);
9749 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
9750 	if (sb == NULL)
9751 		return (ENOMEM);
9752 
9753 	if (lc->link_ok || lc->link_down_rc == 255)
9754 		sbuf_printf(sb, "n/a");
9755 	else
9756 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
9757 
9758 	rc = sbuf_finish(sb);
9759 	sbuf_delete(sb);
9760 
9761 	return (rc);
9762 }
9763 
9764 struct mem_desc {
9765 	u_int base;
9766 	u_int limit;
9767 	u_int idx;
9768 };
9769 
9770 static int
9771 mem_desc_cmp(const void *a, const void *b)
9772 {
9773 	const u_int v1 = ((const struct mem_desc *)a)->base;
9774 	const u_int v2 = ((const struct mem_desc *)b)->base;
9775 
9776 	if (v1 < v2)
9777 		return (-1);
9778 	else if (v1 > v2)
9779 		return (1);
9780 
9781 	return (0);
9782 }
9783 
9784 static void
9785 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
9786     unsigned int to)
9787 {
9788 	unsigned int size;
9789 
9790 	if (from == to)
9791 		return;
9792 
9793 	size = to - from + 1;
9794 	if (size == 0)
9795 		return;
9796 
9797 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
9798 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
9799 }
9800 
9801 static int
9802 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
9803 {
9804 	struct adapter *sc = arg1;
9805 	struct sbuf *sb;
9806 	int rc, i, n;
9807 	uint32_t lo, hi, used, free, alloc;
9808 	static const char *memory[] = {
9809 		"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:", "HMA:"
9810 	};
9811 	static const char *region[] = {
9812 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
9813 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
9814 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
9815 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
9816 		"RQUDP region:", "PBL region:", "TXPBL region:",
9817 		"TLSKey region:", "DBVFIFO region:", "ULPRX state:",
9818 		"ULPTX state:", "On-chip queues:",
9819 	};
9820 	struct mem_desc avail[4];
9821 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
9822 	struct mem_desc *md = mem;
9823 
9824 	rc = sysctl_wire_old_buffer(req, 0);
9825 	if (rc != 0)
9826 		return (rc);
9827 
9828 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9829 	if (sb == NULL)
9830 		return (ENOMEM);
9831 
9832 	for (i = 0; i < nitems(mem); i++) {
9833 		mem[i].limit = 0;
9834 		mem[i].idx = i;
9835 	}
9836 
9837 	mtx_lock(&sc->reg_lock);
9838 	if (hw_off_limits(sc)) {
9839 		rc = ENXIO;
9840 		goto done;
9841 	}
9842 
9843 	/* Find and sort the populated memory ranges */
9844 	i = 0;
9845 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
9846 	if (lo & F_EDRAM0_ENABLE) {
9847 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
9848 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
9849 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
9850 		avail[i].idx = 0;
9851 		i++;
9852 	}
9853 	if (lo & F_EDRAM1_ENABLE) {
9854 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
9855 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
9856 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
9857 		avail[i].idx = 1;
9858 		i++;
9859 	}
9860 	if (lo & F_EXT_MEM_ENABLE) {
9861 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
9862 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
9863 		avail[i].limit = avail[i].base + (G_EXT_MEM_SIZE(hi) << 20);
9864 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
9865 		i++;
9866 	}
9867 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
9868 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
9869 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
9870 		avail[i].limit = avail[i].base + (G_EXT_MEM1_SIZE(hi) << 20);
9871 		avail[i].idx = 4;
9872 		i++;
9873 	}
9874 	if (is_t6(sc) && lo & F_HMA_MUX) {
9875 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
9876 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
9877 		avail[i].limit = avail[i].base + (G_EXT_MEM1_SIZE(hi) << 20);
9878 		avail[i].idx = 5;
9879 		i++;
9880 	}
9881 	MPASS(i <= nitems(avail));
9882 	if (!i)                                    /* no memory available */
9883 		goto done;
9884 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
9885 
9886 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
9887 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
9888 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
9889 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
9890 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
9891 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
9892 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
9893 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
9894 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
9895 
9896 	/* the next few have explicit upper bounds */
9897 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
9898 	md->limit = md->base - 1 +
9899 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
9900 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
9901 	md++;
9902 
9903 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
9904 	md->limit = md->base - 1 +
9905 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
9906 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
9907 	md++;
9908 
9909 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
9910 		if (chip_id(sc) <= CHELSIO_T5)
9911 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
9912 		else
9913 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
9914 		md->limit = 0;
9915 	} else {
9916 		md->base = 0;
9917 		md->idx = nitems(region);  /* hide it */
9918 	}
9919 	md++;
9920 
9921 #define ulp_region(reg) \
9922 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
9923 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
9924 
9925 	ulp_region(RX_ISCSI);
9926 	ulp_region(RX_TDDP);
9927 	ulp_region(TX_TPT);
9928 	ulp_region(RX_STAG);
9929 	ulp_region(RX_RQ);
9930 	ulp_region(RX_RQUDP);
9931 	ulp_region(RX_PBL);
9932 	ulp_region(TX_PBL);
9933 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
9934 		ulp_region(RX_TLS_KEY);
9935 	}
9936 #undef ulp_region
9937 
9938 	md->base = 0;
9939 	if (is_t4(sc))
9940 		md->idx = nitems(region);
9941 	else {
9942 		uint32_t size = 0;
9943 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
9944 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
9945 
9946 		if (is_t5(sc)) {
9947 			if (sge_ctrl & F_VFIFO_ENABLE)
9948 				size = fifo_size << 2;
9949 		} else
9950 			size = G_T6_DBVFIFO_SIZE(fifo_size) << 6;
9951 
9952 		if (size) {
9953 			md->base = t4_read_reg(sc, A_SGE_DBVFIFO_BADDR);
9954 			md->limit = md->base + size - 1;
9955 		} else
9956 			md->idx = nitems(region);
9957 	}
9958 	md++;
9959 
9960 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
9961 	md->limit = 0;
9962 	md++;
9963 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
9964 	md->limit = 0;
9965 	md++;
9966 
9967 	md->base = sc->vres.ocq.start;
9968 	if (sc->vres.ocq.size)
9969 		md->limit = md->base + sc->vres.ocq.size - 1;
9970 	else
9971 		md->idx = nitems(region);  /* hide it */
9972 	md++;
9973 
9974 	/* add any address-space holes, there can be up to 3 */
9975 	for (n = 0; n < i - 1; n++)
9976 		if (avail[n].limit < avail[n + 1].base)
9977 			(md++)->base = avail[n].limit;
9978 	if (avail[n].limit)
9979 		(md++)->base = avail[n].limit;
9980 
9981 	n = md - mem;
9982 	MPASS(n <= nitems(mem));
9983 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
9984 
9985 	for (lo = 0; lo < i; lo++)
9986 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
9987 				avail[lo].limit - 1);
9988 
9989 	sbuf_printf(sb, "\n");
9990 	for (i = 0; i < n; i++) {
9991 		if (mem[i].idx >= nitems(region))
9992 			continue;                        /* skip holes */
9993 		if (!mem[i].limit)
9994 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
9995 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
9996 				mem[i].limit);
9997 	}
9998 
9999 	sbuf_printf(sb, "\n");
10000 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
10001 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
10002 	mem_region_show(sb, "uP RAM:", lo, hi);
10003 
10004 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
10005 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
10006 	mem_region_show(sb, "uP Extmem2:", lo, hi);
10007 
10008 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
10009 	for (i = 0, free = 0; i < 2; i++)
10010 		free += G_FREERXPAGECOUNT(t4_read_reg(sc, A_TP_FLM_FREE_RX_CNT));
10011 	sbuf_printf(sb, "\n%u Rx pages (%u free) of size %uKiB for %u channels\n",
10012 		   G_PMRXMAXPAGE(lo), free,
10013 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
10014 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
10015 
10016 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
10017 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
10018 	for (i = 0, free = 0; i < 4; i++)
10019 		free += G_FREETXPAGECOUNT(t4_read_reg(sc, A_TP_FLM_FREE_TX_CNT));
10020 	sbuf_printf(sb, "%u Tx pages (%u free) of size %u%ciB for %u channels\n",
10021 		   G_PMTXMAXPAGE(lo), free,
10022 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
10023 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
10024 	sbuf_printf(sb, "%u p-structs (%u free)\n",
10025 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT),
10026 		   G_FREEPSTRUCTCOUNT(t4_read_reg(sc, A_TP_FLM_FREE_PS_CNT)));
10027 
10028 	for (i = 0; i < 4; i++) {
10029 		if (chip_id(sc) > CHELSIO_T5)
10030 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
10031 		else
10032 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
10033 		if (is_t5(sc)) {
10034 			used = G_T5_USED(lo);
10035 			alloc = G_T5_ALLOC(lo);
10036 		} else {
10037 			used = G_USED(lo);
10038 			alloc = G_ALLOC(lo);
10039 		}
10040 		/* For T6 these are MAC buffer groups */
10041 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
10042 		    i, used, alloc);
10043 	}
10044 	for (i = 0; i < sc->chip_params->nchan; i++) {
10045 		if (chip_id(sc) > CHELSIO_T5)
10046 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
10047 		else
10048 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
10049 		if (is_t5(sc)) {
10050 			used = G_T5_USED(lo);
10051 			alloc = G_T5_ALLOC(lo);
10052 		} else {
10053 			used = G_USED(lo);
10054 			alloc = G_ALLOC(lo);
10055 		}
10056 		/* For T6 these are MAC buffer groups */
10057 		sbuf_printf(sb,
10058 		    "\nLoopback %d using %u pages out of %u allocated",
10059 		    i, used, alloc);
10060 	}
10061 done:
10062 	mtx_unlock(&sc->reg_lock);
10063 	if (rc == 0)
10064 		rc = sbuf_finish(sb);
10065 	sbuf_delete(sb);
10066 	return (rc);
10067 }
10068 
10069 static inline void
10070 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
10071 {
10072 	*mask = x | y;
10073 	y = htobe64(y);
10074 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
10075 }
10076 
10077 static int
10078 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
10079 {
10080 	struct adapter *sc = arg1;
10081 	struct sbuf *sb;
10082 	int rc, i;
10083 
10084 	MPASS(chip_id(sc) <= CHELSIO_T5);
10085 
10086 	rc = sysctl_wire_old_buffer(req, 0);
10087 	if (rc != 0)
10088 		return (rc);
10089 
10090 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
10091 	if (sb == NULL)
10092 		return (ENOMEM);
10093 
10094 	sbuf_printf(sb,
10095 	    "Idx  Ethernet address     Mask     Vld Ports PF"
10096 	    "  VF              Replication             P0 P1 P2 P3  ML");
10097 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
10098 		uint64_t tcamx, tcamy, mask;
10099 		uint32_t cls_lo, cls_hi;
10100 		uint8_t addr[ETHER_ADDR_LEN];
10101 
10102 		mtx_lock(&sc->reg_lock);
10103 		if (hw_off_limits(sc))
10104 			rc = ENXIO;
10105 		else {
10106 			tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
10107 			tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
10108 		}
10109 		mtx_unlock(&sc->reg_lock);
10110 		if (rc != 0)
10111 			break;
10112 		if (tcamx & tcamy)
10113 			continue;
10114 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
10115 		mtx_lock(&sc->reg_lock);
10116 		if (hw_off_limits(sc))
10117 			rc = ENXIO;
10118 		else {
10119 			cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
10120 			cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
10121 		}
10122 		mtx_unlock(&sc->reg_lock);
10123 		if (rc != 0)
10124 			break;
10125 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
10126 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
10127 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
10128 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
10129 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
10130 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
10131 
10132 		if (cls_lo & F_REPLICATE) {
10133 			struct fw_ldst_cmd ldst_cmd;
10134 
10135 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
10136 			ldst_cmd.op_to_addrspace =
10137 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
10138 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
10139 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
10140 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
10141 			ldst_cmd.u.mps.rplc.fid_idx =
10142 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
10143 				V_FW_LDST_CMD_IDX(i));
10144 
10145 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
10146 			    "t4mps");
10147 			if (rc)
10148 				break;
10149 			if (hw_off_limits(sc))
10150 				rc = ENXIO;
10151 			else
10152 				rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
10153 				    sizeof(ldst_cmd), &ldst_cmd);
10154 			end_synchronized_op(sc, 0);
10155 			if (rc != 0)
10156 				break;
10157 			else {
10158 				sbuf_printf(sb, " %08x %08x %08x %08x",
10159 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
10160 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
10161 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
10162 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
10163 			}
10164 		} else
10165 			sbuf_printf(sb, "%36s", "");
10166 
10167 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
10168 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
10169 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
10170 	}
10171 
10172 	if (rc)
10173 		(void) sbuf_finish(sb);
10174 	else
10175 		rc = sbuf_finish(sb);
10176 	sbuf_delete(sb);
10177 
10178 	return (rc);
10179 }
10180 
10181 static int
10182 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
10183 {
10184 	struct adapter *sc = arg1;
10185 	struct sbuf *sb;
10186 	int rc, i;
10187 
10188 	MPASS(chip_id(sc) > CHELSIO_T5);
10189 
10190 	rc = sysctl_wire_old_buffer(req, 0);
10191 	if (rc != 0)
10192 		return (rc);
10193 
10194 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
10195 	if (sb == NULL)
10196 		return (ENOMEM);
10197 
10198 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
10199 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
10200 	    "                           Replication"
10201 	    "                                    P0 P1 P2 P3  ML\n");
10202 
10203 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
10204 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
10205 		uint16_t ivlan;
10206 		uint64_t tcamx, tcamy, val, mask;
10207 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
10208 		uint8_t addr[ETHER_ADDR_LEN];
10209 
10210 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
10211 		if (i < 256)
10212 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
10213 		else
10214 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
10215 		mtx_lock(&sc->reg_lock);
10216 		if (hw_off_limits(sc))
10217 			rc = ENXIO;
10218 		else {
10219 			t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
10220 			val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
10221 			tcamy = G_DMACH(val) << 32;
10222 			tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
10223 			data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
10224 		}
10225 		mtx_unlock(&sc->reg_lock);
10226 		if (rc != 0)
10227 			break;
10228 
10229 		lookup_type = G_DATALKPTYPE(data2);
10230 		port_num = G_DATAPORTNUM(data2);
10231 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
10232 			/* Inner header VNI */
10233 			vniy = ((data2 & F_DATAVIDH2) << 23) |
10234 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
10235 			dip_hit = data2 & F_DATADIPHIT;
10236 			vlan_vld = 0;
10237 		} else {
10238 			vniy = 0;
10239 			dip_hit = 0;
10240 			vlan_vld = data2 & F_DATAVIDH2;
10241 			ivlan = G_VIDL(val);
10242 		}
10243 
10244 		ctl |= V_CTLXYBITSEL(1);
10245 		mtx_lock(&sc->reg_lock);
10246 		if (hw_off_limits(sc))
10247 			rc = ENXIO;
10248 		else {
10249 			t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
10250 			val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
10251 			tcamx = G_DMACH(val) << 32;
10252 			tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
10253 			data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
10254 		}
10255 		mtx_unlock(&sc->reg_lock);
10256 		if (rc != 0)
10257 			break;
10258 
10259 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
10260 			/* Inner header VNI mask */
10261 			vnix = ((data2 & F_DATAVIDH2) << 23) |
10262 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
10263 		} else
10264 			vnix = 0;
10265 
10266 		if (tcamx & tcamy)
10267 			continue;
10268 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
10269 
10270 		mtx_lock(&sc->reg_lock);
10271 		if (hw_off_limits(sc))
10272 			rc = ENXIO;
10273 		else {
10274 			cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
10275 			cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
10276 		}
10277 		mtx_unlock(&sc->reg_lock);
10278 		if (rc != 0)
10279 			break;
10280 
10281 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
10282 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
10283 			    "%012jx %06x %06x    -    -   %3c"
10284 			    "        I  %4x   %3c   %#x%4u%4d", i, addr[0],
10285 			    addr[1], addr[2], addr[3], addr[4], addr[5],
10286 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
10287 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
10288 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
10289 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
10290 		} else {
10291 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
10292 			    "%012jx    -       -   ", i, addr[0], addr[1],
10293 			    addr[2], addr[3], addr[4], addr[5],
10294 			    (uintmax_t)mask);
10295 
10296 			if (vlan_vld)
10297 				sbuf_printf(sb, "%4u   Y     ", ivlan);
10298 			else
10299 				sbuf_printf(sb, "  -    N     ");
10300 
10301 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
10302 			    lookup_type ? 'I' : 'O', port_num,
10303 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
10304 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
10305 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
10306 		}
10307 
10308 
10309 		if (cls_lo & F_T6_REPLICATE) {
10310 			struct fw_ldst_cmd ldst_cmd;
10311 
10312 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
10313 			ldst_cmd.op_to_addrspace =
10314 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
10315 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
10316 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
10317 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
10318 			ldst_cmd.u.mps.rplc.fid_idx =
10319 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
10320 				V_FW_LDST_CMD_IDX(i));
10321 
10322 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
10323 			    "t6mps");
10324 			if (rc)
10325 				break;
10326 			if (hw_off_limits(sc))
10327 				rc = ENXIO;
10328 			else
10329 				rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
10330 				    sizeof(ldst_cmd), &ldst_cmd);
10331 			end_synchronized_op(sc, 0);
10332 			if (rc != 0)
10333 				break;
10334 			else {
10335 				sbuf_printf(sb, " %08x %08x %08x %08x"
10336 				    " %08x %08x %08x %08x",
10337 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
10338 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
10339 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
10340 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
10341 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
10342 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
10343 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
10344 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
10345 			}
10346 		} else
10347 			sbuf_printf(sb, "%72s", "");
10348 
10349 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
10350 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
10351 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
10352 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
10353 	}
10354 
10355 	if (rc)
10356 		(void) sbuf_finish(sb);
10357 	else
10358 		rc = sbuf_finish(sb);
10359 	sbuf_delete(sb);
10360 
10361 	return (rc);
10362 }
10363 
10364 static int
10365 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
10366 {
10367 	struct adapter *sc = arg1;
10368 	struct sbuf *sb;
10369 	int rc;
10370 	uint16_t mtus[NMTUS];
10371 
10372 	rc = sysctl_wire_old_buffer(req, 0);
10373 	if (rc != 0)
10374 		return (rc);
10375 
10376 	mtx_lock(&sc->reg_lock);
10377 	if (hw_off_limits(sc))
10378 		rc = ENXIO;
10379 	else
10380 		t4_read_mtu_tbl(sc, mtus, NULL);
10381 	mtx_unlock(&sc->reg_lock);
10382 	if (rc != 0)
10383 		return (rc);
10384 
10385 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
10386 	if (sb == NULL)
10387 		return (ENOMEM);
10388 
10389 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
10390 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
10391 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
10392 	    mtus[14], mtus[15]);
10393 
10394 	rc = sbuf_finish(sb);
10395 	sbuf_delete(sb);
10396 
10397 	return (rc);
10398 }
10399 
10400 static int
10401 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
10402 {
10403 	struct adapter *sc = arg1;
10404 	struct sbuf *sb;
10405 	int rc, i;
10406 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
10407 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
10408 	static const char *tx_stats[MAX_PM_NSTATS] = {
10409 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
10410 		"Tx FIFO wait", NULL, "Tx latency"
10411 	};
10412 	static const char *rx_stats[MAX_PM_NSTATS] = {
10413 		"Read:", "Write bypass:", "Write mem:", "Flush:",
10414 		"Rx FIFO wait", NULL, "Rx latency"
10415 	};
10416 
10417 	rc = sysctl_wire_old_buffer(req, 0);
10418 	if (rc != 0)
10419 		return (rc);
10420 
10421 	mtx_lock(&sc->reg_lock);
10422 	if (hw_off_limits(sc))
10423 		rc = ENXIO;
10424 	else {
10425 		t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
10426 		t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
10427 	}
10428 	mtx_unlock(&sc->reg_lock);
10429 	if (rc != 0)
10430 		return (rc);
10431 
10432 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
10433 	if (sb == NULL)
10434 		return (ENOMEM);
10435 
10436 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
10437 	for (i = 0; i < 4; i++) {
10438 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
10439 		    tx_cyc[i]);
10440 	}
10441 
10442 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
10443 	for (i = 0; i < 4; i++) {
10444 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
10445 		    rx_cyc[i]);
10446 	}
10447 
10448 	if (chip_id(sc) > CHELSIO_T5) {
10449 		sbuf_printf(sb,
10450 		    "\n              Total wait      Total occupancy");
10451 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
10452 		    tx_cyc[i]);
10453 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
10454 		    rx_cyc[i]);
10455 
10456 		i += 2;
10457 		MPASS(i < nitems(tx_stats));
10458 
10459 		sbuf_printf(sb,
10460 		    "\n                   Reads           Total wait");
10461 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
10462 		    tx_cyc[i]);
10463 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
10464 		    rx_cyc[i]);
10465 	}
10466 
10467 	rc = sbuf_finish(sb);
10468 	sbuf_delete(sb);
10469 
10470 	return (rc);
10471 }
10472 
10473 static int
10474 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
10475 {
10476 	struct adapter *sc = arg1;
10477 	struct sbuf *sb;
10478 	int rc;
10479 	struct tp_rdma_stats stats;
10480 
10481 	rc = sysctl_wire_old_buffer(req, 0);
10482 	if (rc != 0)
10483 		return (rc);
10484 
10485 	mtx_lock(&sc->reg_lock);
10486 	if (hw_off_limits(sc))
10487 		rc = ENXIO;
10488 	else
10489 		t4_tp_get_rdma_stats(sc, &stats, 0);
10490 	mtx_unlock(&sc->reg_lock);
10491 	if (rc != 0)
10492 		return (rc);
10493 
10494 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
10495 	if (sb == NULL)
10496 		return (ENOMEM);
10497 
10498 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
10499 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
10500 
10501 	rc = sbuf_finish(sb);
10502 	sbuf_delete(sb);
10503 
10504 	return (rc);
10505 }
10506 
10507 static int
10508 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
10509 {
10510 	struct adapter *sc = arg1;
10511 	struct sbuf *sb;
10512 	int rc;
10513 	struct tp_tcp_stats v4, v6;
10514 
10515 	rc = sysctl_wire_old_buffer(req, 0);
10516 	if (rc != 0)
10517 		return (rc);
10518 
10519 	mtx_lock(&sc->reg_lock);
10520 	if (hw_off_limits(sc))
10521 		rc = ENXIO;
10522 	else
10523 		t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
10524 	mtx_unlock(&sc->reg_lock);
10525 	if (rc != 0)
10526 		return (rc);
10527 
10528 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
10529 	if (sb == NULL)
10530 		return (ENOMEM);
10531 
10532 	sbuf_printf(sb,
10533 	    "                                IP                 IPv6\n");
10534 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
10535 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
10536 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
10537 	    v4.tcp_in_segs, v6.tcp_in_segs);
10538 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
10539 	    v4.tcp_out_segs, v6.tcp_out_segs);
10540 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
10541 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
10542 
10543 	rc = sbuf_finish(sb);
10544 	sbuf_delete(sb);
10545 
10546 	return (rc);
10547 }
10548 
10549 static int
10550 sysctl_tids(SYSCTL_HANDLER_ARGS)
10551 {
10552 	struct adapter *sc = arg1;
10553 	struct sbuf *sb;
10554 	int rc;
10555 	uint32_t x, y;
10556 	struct tid_info *t = &sc->tids;
10557 
10558 	rc = sysctl_wire_old_buffer(req, 0);
10559 	if (rc != 0)
10560 		return (rc);
10561 
10562 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
10563 	if (sb == NULL)
10564 		return (ENOMEM);
10565 
10566 	if (t->natids) {
10567 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
10568 		    t->atids_in_use);
10569 	}
10570 
10571 	if (t->nhpftids) {
10572 		sbuf_printf(sb, "HPFTID range: %u-%u, in use: %u\n",
10573 		    t->hpftid_base, t->hpftid_end, t->hpftids_in_use);
10574 	}
10575 
10576 	if (t->ntids) {
10577 		bool hashen = false;
10578 
10579 		mtx_lock(&sc->reg_lock);
10580 		if (hw_off_limits(sc))
10581 			rc = ENXIO;
10582 		else if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
10583 			hashen = true;
10584 			if (chip_id(sc) <= CHELSIO_T5) {
10585 				x = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
10586 				y = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
10587 			} else {
10588 				x = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
10589 				y = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
10590 			}
10591 		}
10592 		mtx_unlock(&sc->reg_lock);
10593 		if (rc != 0)
10594 			goto done;
10595 
10596 		sbuf_printf(sb, "TID range: ");
10597 		if (hashen) {
10598 			if (x)
10599 				sbuf_printf(sb, "%u-%u, ", t->tid_base, x - 1);
10600 			sbuf_printf(sb, "%u-%u", y, t->ntids - 1);
10601 		} else {
10602 			sbuf_printf(sb, "%u-%u", t->tid_base, t->tid_base +
10603 			    t->ntids - 1);
10604 		}
10605 		sbuf_printf(sb, ", in use: %u\n",
10606 		    atomic_load_acq_int(&t->tids_in_use));
10607 	}
10608 
10609 	if (t->nstids) {
10610 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
10611 		    t->stid_base + t->nstids - 1, t->stids_in_use);
10612 	}
10613 
10614 	if (t->nftids) {
10615 		sbuf_printf(sb, "FTID range: %u-%u, in use: %u\n", t->ftid_base,
10616 		    t->ftid_end, t->ftids_in_use);
10617 	}
10618 
10619 	if (t->netids) {
10620 		sbuf_printf(sb, "ETID range: %u-%u, in use: %u\n", t->etid_base,
10621 		    t->etid_base + t->netids - 1, t->etids_in_use);
10622 	}
10623 
10624 	mtx_lock(&sc->reg_lock);
10625 	if (hw_off_limits(sc))
10626 		rc = ENXIO;
10627 	else {
10628 		x = t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4);
10629 		y = t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6);
10630 	}
10631 	mtx_unlock(&sc->reg_lock);
10632 	if (rc != 0)
10633 		goto done;
10634 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users", x, y);
10635 done:
10636 	if (rc == 0)
10637 		rc = sbuf_finish(sb);
10638 	else
10639 		(void)sbuf_finish(sb);
10640 	sbuf_delete(sb);
10641 
10642 	return (rc);
10643 }
10644 
10645 static int
10646 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
10647 {
10648 	struct adapter *sc = arg1;
10649 	struct sbuf *sb;
10650 	int rc;
10651 	struct tp_err_stats stats;
10652 
10653 	rc = sysctl_wire_old_buffer(req, 0);
10654 	if (rc != 0)
10655 		return (rc);
10656 
10657 	mtx_lock(&sc->reg_lock);
10658 	if (hw_off_limits(sc))
10659 		rc = ENXIO;
10660 	else
10661 		t4_tp_get_err_stats(sc, &stats, 0);
10662 	mtx_unlock(&sc->reg_lock);
10663 	if (rc != 0)
10664 		return (rc);
10665 
10666 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
10667 	if (sb == NULL)
10668 		return (ENOMEM);
10669 
10670 	if (sc->chip_params->nchan > 2) {
10671 		sbuf_printf(sb, "                 channel 0  channel 1"
10672 		    "  channel 2  channel 3\n");
10673 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
10674 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
10675 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
10676 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
10677 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
10678 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
10679 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
10680 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
10681 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
10682 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
10683 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
10684 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
10685 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
10686 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
10687 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
10688 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
10689 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
10690 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
10691 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
10692 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
10693 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
10694 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
10695 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
10696 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
10697 	} else {
10698 		sbuf_printf(sb, "                 channel 0  channel 1\n");
10699 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
10700 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
10701 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
10702 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
10703 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
10704 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
10705 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
10706 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
10707 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
10708 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
10709 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
10710 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
10711 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
10712 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
10713 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
10714 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
10715 	}
10716 
10717 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
10718 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
10719 
10720 	rc = sbuf_finish(sb);
10721 	sbuf_delete(sb);
10722 
10723 	return (rc);
10724 }
10725 
10726 static int
10727 sysctl_tnl_stats(SYSCTL_HANDLER_ARGS)
10728 {
10729 	struct adapter *sc = arg1;
10730 	struct sbuf *sb;
10731 	int rc;
10732 	struct tp_tnl_stats stats;
10733 
10734 	rc = sysctl_wire_old_buffer(req, 0);
10735 	if (rc != 0)
10736 		return(rc);
10737 
10738 	mtx_lock(&sc->reg_lock);
10739 	if (hw_off_limits(sc))
10740 		rc = ENXIO;
10741 	else
10742 		t4_tp_get_tnl_stats(sc, &stats, 1);
10743 	mtx_unlock(&sc->reg_lock);
10744 	if (rc != 0)
10745 		return (rc);
10746 
10747 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
10748 	if (sb == NULL)
10749 		return (ENOMEM);
10750 
10751 	if (sc->chip_params->nchan > 2) {
10752 		sbuf_printf(sb, "           channel 0  channel 1"
10753 		    "  channel 2  channel 3\n");
10754 		sbuf_printf(sb, "OutPkts:  %10u %10u %10u %10u\n",
10755 		    stats.out_pkt[0], stats.out_pkt[1],
10756 		    stats.out_pkt[2], stats.out_pkt[3]);
10757 		sbuf_printf(sb, "InPkts:   %10u %10u %10u %10u",
10758 		    stats.in_pkt[0], stats.in_pkt[1],
10759 		    stats.in_pkt[2], stats.in_pkt[3]);
10760 	} else {
10761 		sbuf_printf(sb, "           channel 0  channel 1\n");
10762 		sbuf_printf(sb, "OutPkts:  %10u %10u\n",
10763 		    stats.out_pkt[0], stats.out_pkt[1]);
10764 		sbuf_printf(sb, "InPkts:   %10u %10u",
10765 		    stats.in_pkt[0], stats.in_pkt[1]);
10766 	}
10767 
10768 	rc = sbuf_finish(sb);
10769 	sbuf_delete(sb);
10770 
10771 	return (rc);
10772 }
10773 
10774 static int
10775 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
10776 {
10777 	struct adapter *sc = arg1;
10778 	struct tp_params *tpp = &sc->params.tp;
10779 	u_int mask;
10780 	int rc;
10781 
10782 	mask = tpp->la_mask >> 16;
10783 	rc = sysctl_handle_int(oidp, &mask, 0, req);
10784 	if (rc != 0 || req->newptr == NULL)
10785 		return (rc);
10786 	if (mask > 0xffff)
10787 		return (EINVAL);
10788 	mtx_lock(&sc->reg_lock);
10789 	if (hw_off_limits(sc))
10790 		rc = ENXIO;
10791 	else {
10792 		tpp->la_mask = mask << 16;
10793 		t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U,
10794 		    tpp->la_mask);
10795 	}
10796 	mtx_unlock(&sc->reg_lock);
10797 
10798 	return (rc);
10799 }
10800 
10801 struct field_desc {
10802 	const char *name;
10803 	u_int start;
10804 	u_int width;
10805 };
10806 
10807 static void
10808 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
10809 {
10810 	char buf[32];
10811 	int line_size = 0;
10812 
10813 	while (f->name) {
10814 		uint64_t mask = (1ULL << f->width) - 1;
10815 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
10816 		    ((uintmax_t)v >> f->start) & mask);
10817 
10818 		if (line_size + len >= 79) {
10819 			line_size = 8;
10820 			sbuf_printf(sb, "\n        ");
10821 		}
10822 		sbuf_printf(sb, "%s ", buf);
10823 		line_size += len + 1;
10824 		f++;
10825 	}
10826 	sbuf_printf(sb, "\n");
10827 }
10828 
10829 static const struct field_desc tp_la0[] = {
10830 	{ "RcfOpCodeOut", 60, 4 },
10831 	{ "State", 56, 4 },
10832 	{ "WcfState", 52, 4 },
10833 	{ "RcfOpcSrcOut", 50, 2 },
10834 	{ "CRxError", 49, 1 },
10835 	{ "ERxError", 48, 1 },
10836 	{ "SanityFailed", 47, 1 },
10837 	{ "SpuriousMsg", 46, 1 },
10838 	{ "FlushInputMsg", 45, 1 },
10839 	{ "FlushInputCpl", 44, 1 },
10840 	{ "RssUpBit", 43, 1 },
10841 	{ "RssFilterHit", 42, 1 },
10842 	{ "Tid", 32, 10 },
10843 	{ "InitTcb", 31, 1 },
10844 	{ "LineNumber", 24, 7 },
10845 	{ "Emsg", 23, 1 },
10846 	{ "EdataOut", 22, 1 },
10847 	{ "Cmsg", 21, 1 },
10848 	{ "CdataOut", 20, 1 },
10849 	{ "EreadPdu", 19, 1 },
10850 	{ "CreadPdu", 18, 1 },
10851 	{ "TunnelPkt", 17, 1 },
10852 	{ "RcfPeerFin", 16, 1 },
10853 	{ "RcfReasonOut", 12, 4 },
10854 	{ "TxCchannel", 10, 2 },
10855 	{ "RcfTxChannel", 8, 2 },
10856 	{ "RxEchannel", 6, 2 },
10857 	{ "RcfRxChannel", 5, 1 },
10858 	{ "RcfDataOutSrdy", 4, 1 },
10859 	{ "RxDvld", 3, 1 },
10860 	{ "RxOoDvld", 2, 1 },
10861 	{ "RxCongestion", 1, 1 },
10862 	{ "TxCongestion", 0, 1 },
10863 	{ NULL }
10864 };
10865 
10866 static const struct field_desc tp_la1[] = {
10867 	{ "CplCmdIn", 56, 8 },
10868 	{ "CplCmdOut", 48, 8 },
10869 	{ "ESynOut", 47, 1 },
10870 	{ "EAckOut", 46, 1 },
10871 	{ "EFinOut", 45, 1 },
10872 	{ "ERstOut", 44, 1 },
10873 	{ "SynIn", 43, 1 },
10874 	{ "AckIn", 42, 1 },
10875 	{ "FinIn", 41, 1 },
10876 	{ "RstIn", 40, 1 },
10877 	{ "DataIn", 39, 1 },
10878 	{ "DataInVld", 38, 1 },
10879 	{ "PadIn", 37, 1 },
10880 	{ "RxBufEmpty", 36, 1 },
10881 	{ "RxDdp", 35, 1 },
10882 	{ "RxFbCongestion", 34, 1 },
10883 	{ "TxFbCongestion", 33, 1 },
10884 	{ "TxPktSumSrdy", 32, 1 },
10885 	{ "RcfUlpType", 28, 4 },
10886 	{ "Eread", 27, 1 },
10887 	{ "Ebypass", 26, 1 },
10888 	{ "Esave", 25, 1 },
10889 	{ "Static0", 24, 1 },
10890 	{ "Cread", 23, 1 },
10891 	{ "Cbypass", 22, 1 },
10892 	{ "Csave", 21, 1 },
10893 	{ "CPktOut", 20, 1 },
10894 	{ "RxPagePoolFull", 18, 2 },
10895 	{ "RxLpbkPkt", 17, 1 },
10896 	{ "TxLpbkPkt", 16, 1 },
10897 	{ "RxVfValid", 15, 1 },
10898 	{ "SynLearned", 14, 1 },
10899 	{ "SetDelEntry", 13, 1 },
10900 	{ "SetInvEntry", 12, 1 },
10901 	{ "CpcmdDvld", 11, 1 },
10902 	{ "CpcmdSave", 10, 1 },
10903 	{ "RxPstructsFull", 8, 2 },
10904 	{ "EpcmdDvld", 7, 1 },
10905 	{ "EpcmdFlush", 6, 1 },
10906 	{ "EpcmdTrimPrefix", 5, 1 },
10907 	{ "EpcmdTrimPostfix", 4, 1 },
10908 	{ "ERssIp4Pkt", 3, 1 },
10909 	{ "ERssIp6Pkt", 2, 1 },
10910 	{ "ERssTcpUdpPkt", 1, 1 },
10911 	{ "ERssFceFipPkt", 0, 1 },
10912 	{ NULL }
10913 };
10914 
10915 static const struct field_desc tp_la2[] = {
10916 	{ "CplCmdIn", 56, 8 },
10917 	{ "MpsVfVld", 55, 1 },
10918 	{ "MpsPf", 52, 3 },
10919 	{ "MpsVf", 44, 8 },
10920 	{ "SynIn", 43, 1 },
10921 	{ "AckIn", 42, 1 },
10922 	{ "FinIn", 41, 1 },
10923 	{ "RstIn", 40, 1 },
10924 	{ "DataIn", 39, 1 },
10925 	{ "DataInVld", 38, 1 },
10926 	{ "PadIn", 37, 1 },
10927 	{ "RxBufEmpty", 36, 1 },
10928 	{ "RxDdp", 35, 1 },
10929 	{ "RxFbCongestion", 34, 1 },
10930 	{ "TxFbCongestion", 33, 1 },
10931 	{ "TxPktSumSrdy", 32, 1 },
10932 	{ "RcfUlpType", 28, 4 },
10933 	{ "Eread", 27, 1 },
10934 	{ "Ebypass", 26, 1 },
10935 	{ "Esave", 25, 1 },
10936 	{ "Static0", 24, 1 },
10937 	{ "Cread", 23, 1 },
10938 	{ "Cbypass", 22, 1 },
10939 	{ "Csave", 21, 1 },
10940 	{ "CPktOut", 20, 1 },
10941 	{ "RxPagePoolFull", 18, 2 },
10942 	{ "RxLpbkPkt", 17, 1 },
10943 	{ "TxLpbkPkt", 16, 1 },
10944 	{ "RxVfValid", 15, 1 },
10945 	{ "SynLearned", 14, 1 },
10946 	{ "SetDelEntry", 13, 1 },
10947 	{ "SetInvEntry", 12, 1 },
10948 	{ "CpcmdDvld", 11, 1 },
10949 	{ "CpcmdSave", 10, 1 },
10950 	{ "RxPstructsFull", 8, 2 },
10951 	{ "EpcmdDvld", 7, 1 },
10952 	{ "EpcmdFlush", 6, 1 },
10953 	{ "EpcmdTrimPrefix", 5, 1 },
10954 	{ "EpcmdTrimPostfix", 4, 1 },
10955 	{ "ERssIp4Pkt", 3, 1 },
10956 	{ "ERssIp6Pkt", 2, 1 },
10957 	{ "ERssTcpUdpPkt", 1, 1 },
10958 	{ "ERssFceFipPkt", 0, 1 },
10959 	{ NULL }
10960 };
10961 
10962 static void
10963 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
10964 {
10965 
10966 	field_desc_show(sb, *p, tp_la0);
10967 }
10968 
10969 static void
10970 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
10971 {
10972 
10973 	if (idx)
10974 		sbuf_printf(sb, "\n");
10975 	field_desc_show(sb, p[0], tp_la0);
10976 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
10977 		field_desc_show(sb, p[1], tp_la0);
10978 }
10979 
10980 static void
10981 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
10982 {
10983 
10984 	if (idx)
10985 		sbuf_printf(sb, "\n");
10986 	field_desc_show(sb, p[0], tp_la0);
10987 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
10988 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
10989 }
10990 
10991 static int
10992 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
10993 {
10994 	struct adapter *sc = arg1;
10995 	struct sbuf *sb;
10996 	uint64_t *buf, *p;
10997 	int rc;
10998 	u_int i, inc;
10999 	void (*show_func)(struct sbuf *, uint64_t *, int);
11000 
11001 	rc = sysctl_wire_old_buffer(req, 0);
11002 	if (rc != 0)
11003 		return (rc);
11004 
11005 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
11006 	if (sb == NULL)
11007 		return (ENOMEM);
11008 
11009 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
11010 
11011 	mtx_lock(&sc->reg_lock);
11012 	if (hw_off_limits(sc))
11013 		rc = ENXIO;
11014 	else {
11015 		t4_tp_read_la(sc, buf, NULL);
11016 		switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
11017 		case 2:
11018 			inc = 2;
11019 			show_func = tp_la_show2;
11020 			break;
11021 		case 3:
11022 			inc = 2;
11023 			show_func = tp_la_show3;
11024 			break;
11025 		default:
11026 			inc = 1;
11027 			show_func = tp_la_show;
11028 		}
11029 	}
11030 	mtx_unlock(&sc->reg_lock);
11031 	if (rc != 0)
11032 		goto done;
11033 
11034 	p = buf;
11035 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
11036 		(*show_func)(sb, p, i);
11037 	rc = sbuf_finish(sb);
11038 done:
11039 	sbuf_delete(sb);
11040 	free(buf, M_CXGBE);
11041 	return (rc);
11042 }
11043 
11044 static int
11045 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
11046 {
11047 	struct adapter *sc = arg1;
11048 	struct sbuf *sb;
11049 	int rc;
11050 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
11051 
11052 	rc = sysctl_wire_old_buffer(req, 0);
11053 	if (rc != 0)
11054 		return (rc);
11055 
11056 	mtx_lock(&sc->reg_lock);
11057 	if (hw_off_limits(sc))
11058 		rc = ENXIO;
11059 	else
11060 		t4_get_chan_txrate(sc, nrate, orate);
11061 	mtx_unlock(&sc->reg_lock);
11062 	if (rc != 0)
11063 		return (rc);
11064 
11065 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
11066 	if (sb == NULL)
11067 		return (ENOMEM);
11068 
11069 	if (sc->chip_params->nchan > 2) {
11070 		sbuf_printf(sb, "              channel 0   channel 1"
11071 		    "   channel 2   channel 3\n");
11072 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
11073 		    nrate[0], nrate[1], nrate[2], nrate[3]);
11074 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
11075 		    orate[0], orate[1], orate[2], orate[3]);
11076 	} else {
11077 		sbuf_printf(sb, "              channel 0   channel 1\n");
11078 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
11079 		    nrate[0], nrate[1]);
11080 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
11081 		    orate[0], orate[1]);
11082 	}
11083 
11084 	rc = sbuf_finish(sb);
11085 	sbuf_delete(sb);
11086 
11087 	return (rc);
11088 }
11089 
11090 static int
11091 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
11092 {
11093 	struct adapter *sc = arg1;
11094 	struct sbuf *sb;
11095 	uint32_t *buf, *p;
11096 	int rc, i;
11097 
11098 	rc = sysctl_wire_old_buffer(req, 0);
11099 	if (rc != 0)
11100 		return (rc);
11101 
11102 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
11103 	if (sb == NULL)
11104 		return (ENOMEM);
11105 
11106 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
11107 	    M_ZERO | M_WAITOK);
11108 
11109 	mtx_lock(&sc->reg_lock);
11110 	if (hw_off_limits(sc))
11111 		rc = ENXIO;
11112 	else
11113 		t4_ulprx_read_la(sc, buf);
11114 	mtx_unlock(&sc->reg_lock);
11115 	if (rc != 0)
11116 		goto done;
11117 
11118 	p = buf;
11119 	sbuf_printf(sb, "      Pcmd        Type   Message"
11120 	    "                Data");
11121 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
11122 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
11123 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
11124 	}
11125 	rc = sbuf_finish(sb);
11126 done:
11127 	sbuf_delete(sb);
11128 	free(buf, M_CXGBE);
11129 	return (rc);
11130 }
11131 
11132 static int
11133 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
11134 {
11135 	struct adapter *sc = arg1;
11136 	struct sbuf *sb;
11137 	int rc;
11138 	uint32_t cfg, s1, s2;
11139 
11140 	MPASS(chip_id(sc) >= CHELSIO_T5);
11141 
11142 	rc = sysctl_wire_old_buffer(req, 0);
11143 	if (rc != 0)
11144 		return (rc);
11145 
11146 	mtx_lock(&sc->reg_lock);
11147 	if (hw_off_limits(sc))
11148 		rc = ENXIO;
11149 	else {
11150 		cfg = t4_read_reg(sc, A_SGE_STAT_CFG);
11151 		s1 = t4_read_reg(sc, A_SGE_STAT_TOTAL);
11152 		s2 = t4_read_reg(sc, A_SGE_STAT_MATCH);
11153 	}
11154 	mtx_unlock(&sc->reg_lock);
11155 	if (rc != 0)
11156 		return (rc);
11157 
11158 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
11159 	if (sb == NULL)
11160 		return (ENOMEM);
11161 
11162 	if (G_STATSOURCE_T5(cfg) == 7) {
11163 		int mode;
11164 
11165 		mode = is_t5(sc) ? G_STATMODE(cfg) : G_T6_STATMODE(cfg);
11166 		if (mode == 0)
11167 			sbuf_printf(sb, "total %d, incomplete %d", s1, s2);
11168 		else if (mode == 1)
11169 			sbuf_printf(sb, "total %d, data overflow %d", s1, s2);
11170 		else
11171 			sbuf_printf(sb, "unknown mode %d", mode);
11172 	}
11173 	rc = sbuf_finish(sb);
11174 	sbuf_delete(sb);
11175 
11176 	return (rc);
11177 }
11178 
11179 static int
11180 sysctl_cpus(SYSCTL_HANDLER_ARGS)
11181 {
11182 	struct adapter *sc = arg1;
11183 	enum cpu_sets op = arg2;
11184 	cpuset_t cpuset;
11185 	struct sbuf *sb;
11186 	int i, rc;
11187 
11188 	MPASS(op == LOCAL_CPUS || op == INTR_CPUS);
11189 
11190 	CPU_ZERO(&cpuset);
11191 	rc = bus_get_cpus(sc->dev, op, sizeof(cpuset), &cpuset);
11192 	if (rc != 0)
11193 		return (rc);
11194 
11195 	rc = sysctl_wire_old_buffer(req, 0);
11196 	if (rc != 0)
11197 		return (rc);
11198 
11199 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
11200 	if (sb == NULL)
11201 		return (ENOMEM);
11202 
11203 	CPU_FOREACH(i)
11204 		sbuf_printf(sb, "%d ", i);
11205 	rc = sbuf_finish(sb);
11206 	sbuf_delete(sb);
11207 
11208 	return (rc);
11209 }
11210 
11211 static int
11212 sysctl_reset(SYSCTL_HANDLER_ARGS)
11213 {
11214 	struct adapter *sc = arg1;
11215 	u_int val;
11216 	int rc;
11217 
11218 	val = atomic_load_int(&sc->num_resets);
11219 	rc = sysctl_handle_int(oidp, &val, 0, req);
11220 	if (rc != 0 || req->newptr == NULL)
11221 		return (rc);
11222 
11223 	if (val == 0) {
11224 		/* Zero out the counter that tracks reset. */
11225 		atomic_store_int(&sc->num_resets, 0);
11226 		return (0);
11227 	}
11228 
11229 	if (val != 1)
11230 		return (EINVAL);	/* 0 or 1 are the only legal values */
11231 
11232 	if (hw_off_limits(sc))		/* harmless race */
11233 		return (EALREADY);
11234 
11235 	taskqueue_enqueue(reset_tq, &sc->reset_task);
11236 	return (0);
11237 }
11238 
11239 #ifdef TCP_OFFLOAD
11240 static int
11241 sysctl_tls(SYSCTL_HANDLER_ARGS)
11242 {
11243 	struct adapter *sc = arg1;
11244 	int i, j, v, rc;
11245 	struct vi_info *vi;
11246 
11247 	v = sc->tt.tls;
11248 	rc = sysctl_handle_int(oidp, &v, 0, req);
11249 	if (rc != 0 || req->newptr == NULL)
11250 		return (rc);
11251 
11252 	if (v != 0 && !(sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS))
11253 		return (ENOTSUP);
11254 
11255 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4stls");
11256 	if (rc)
11257 		return (rc);
11258 	if (hw_off_limits(sc))
11259 		rc = ENXIO;
11260 	else {
11261 		sc->tt.tls = !!v;
11262 		for_each_port(sc, i) {
11263 			for_each_vi(sc->port[i], j, vi) {
11264 				if (vi->flags & VI_INIT_DONE)
11265 					t4_update_fl_bufsize(vi->ifp);
11266 			}
11267 		}
11268 	}
11269 	end_synchronized_op(sc, 0);
11270 
11271 	return (rc);
11272 
11273 }
11274 
11275 static void
11276 unit_conv(char *buf, size_t len, u_int val, u_int factor)
11277 {
11278 	u_int rem = val % factor;
11279 
11280 	if (rem == 0)
11281 		snprintf(buf, len, "%u", val / factor);
11282 	else {
11283 		while (rem % 10 == 0)
11284 			rem /= 10;
11285 		snprintf(buf, len, "%u.%u", val / factor, rem);
11286 	}
11287 }
11288 
11289 static int
11290 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
11291 {
11292 	struct adapter *sc = arg1;
11293 	char buf[16];
11294 	u_int res, re;
11295 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
11296 
11297 	mtx_lock(&sc->reg_lock);
11298 	if (hw_off_limits(sc))
11299 		res = (u_int)-1;
11300 	else
11301 		res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
11302 	mtx_unlock(&sc->reg_lock);
11303 	if (res == (u_int)-1)
11304 		return (ENXIO);
11305 
11306 	switch (arg2) {
11307 	case 0:
11308 		/* timer_tick */
11309 		re = G_TIMERRESOLUTION(res);
11310 		break;
11311 	case 1:
11312 		/* TCP timestamp tick */
11313 		re = G_TIMESTAMPRESOLUTION(res);
11314 		break;
11315 	case 2:
11316 		/* DACK tick */
11317 		re = G_DELAYEDACKRESOLUTION(res);
11318 		break;
11319 	default:
11320 		return (EDOOFUS);
11321 	}
11322 
11323 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
11324 
11325 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
11326 }
11327 
11328 static int
11329 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
11330 {
11331 	struct adapter *sc = arg1;
11332 	int rc;
11333 	u_int dack_tmr, dack_re, v;
11334 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
11335 
11336 	mtx_lock(&sc->reg_lock);
11337 	if (hw_off_limits(sc))
11338 		rc = ENXIO;
11339 	else {
11340 		rc = 0;
11341 		dack_re = G_DELAYEDACKRESOLUTION(t4_read_reg(sc,
11342 		    A_TP_TIMER_RESOLUTION));
11343 		dack_tmr = t4_read_reg(sc, A_TP_DACK_TIMER);
11344 	}
11345 	mtx_unlock(&sc->reg_lock);
11346 	if (rc != 0)
11347 		return (rc);
11348 
11349 	v = ((cclk_ps << dack_re) / 1000000) * dack_tmr;
11350 
11351 	return (sysctl_handle_int(oidp, &v, 0, req));
11352 }
11353 
11354 static int
11355 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
11356 {
11357 	struct adapter *sc = arg1;
11358 	int rc, reg = arg2;
11359 	u_int tre;
11360 	u_long tp_tick_us, v;
11361 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
11362 
11363 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
11364 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
11365 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
11366 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
11367 
11368 	mtx_lock(&sc->reg_lock);
11369 	if (hw_off_limits(sc))
11370 		rc = ENXIO;
11371 	else {
11372 		rc = 0;
11373 		tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
11374 		tp_tick_us = (cclk_ps << tre) / 1000000;
11375 		if (reg == A_TP_INIT_SRTT)
11376 			v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
11377 		else
11378 			v = tp_tick_us * t4_read_reg(sc, reg);
11379 	}
11380 	mtx_unlock(&sc->reg_lock);
11381 	if (rc != 0)
11382 		return (rc);
11383 	else
11384 		return (sysctl_handle_long(oidp, &v, 0, req));
11385 }
11386 
11387 /*
11388  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
11389  * passed to this function.
11390  */
11391 static int
11392 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
11393 {
11394 	struct adapter *sc = arg1;
11395 	int rc, idx = arg2;
11396 	u_int v;
11397 
11398 	MPASS(idx >= 0 && idx <= 24);
11399 
11400 	mtx_lock(&sc->reg_lock);
11401 	if (hw_off_limits(sc))
11402 		rc = ENXIO;
11403 	else {
11404 		rc = 0;
11405 		v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
11406 	}
11407 	mtx_unlock(&sc->reg_lock);
11408 	if (rc != 0)
11409 		return (rc);
11410 	else
11411 		return (sysctl_handle_int(oidp, &v, 0, req));
11412 }
11413 
11414 static int
11415 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
11416 {
11417 	struct adapter *sc = arg1;
11418 	int rc, idx = arg2;
11419 	u_int shift, v, r;
11420 
11421 	MPASS(idx >= 0 && idx < 16);
11422 
11423 	r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
11424 	shift = (idx & 3) << 3;
11425 	mtx_lock(&sc->reg_lock);
11426 	if (hw_off_limits(sc))
11427 		rc = ENXIO;
11428 	else {
11429 		rc = 0;
11430 		v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
11431 	}
11432 	mtx_unlock(&sc->reg_lock);
11433 	if (rc != 0)
11434 		return (rc);
11435 	else
11436 		return (sysctl_handle_int(oidp, &v, 0, req));
11437 }
11438 
11439 static int
11440 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
11441 {
11442 	struct vi_info *vi = arg1;
11443 	struct adapter *sc = vi->adapter;
11444 	int idx, rc, i;
11445 	struct sge_ofld_rxq *ofld_rxq;
11446 	uint8_t v;
11447 
11448 	idx = vi->ofld_tmr_idx;
11449 
11450 	rc = sysctl_handle_int(oidp, &idx, 0, req);
11451 	if (rc != 0 || req->newptr == NULL)
11452 		return (rc);
11453 
11454 	if (idx < 0 || idx >= SGE_NTIMERS)
11455 		return (EINVAL);
11456 
11457 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
11458 	    "t4otmr");
11459 	if (rc)
11460 		return (rc);
11461 
11462 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
11463 	for_each_ofld_rxq(vi, i, ofld_rxq) {
11464 #ifdef atomic_store_rel_8
11465 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
11466 #else
11467 		ofld_rxq->iq.intr_params = v;
11468 #endif
11469 	}
11470 	vi->ofld_tmr_idx = idx;
11471 
11472 	end_synchronized_op(sc, LOCK_HELD);
11473 	return (0);
11474 }
11475 
11476 static int
11477 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
11478 {
11479 	struct vi_info *vi = arg1;
11480 	struct adapter *sc = vi->adapter;
11481 	int idx, rc;
11482 
11483 	idx = vi->ofld_pktc_idx;
11484 
11485 	rc = sysctl_handle_int(oidp, &idx, 0, req);
11486 	if (rc != 0 || req->newptr == NULL)
11487 		return (rc);
11488 
11489 	if (idx < -1 || idx >= SGE_NCOUNTERS)
11490 		return (EINVAL);
11491 
11492 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
11493 	    "t4opktc");
11494 	if (rc)
11495 		return (rc);
11496 
11497 	if (vi->flags & VI_INIT_DONE)
11498 		rc = EBUSY; /* cannot be changed once the queues are created */
11499 	else
11500 		vi->ofld_pktc_idx = idx;
11501 
11502 	end_synchronized_op(sc, LOCK_HELD);
11503 	return (rc);
11504 }
11505 #endif
11506 
11507 static int
11508 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
11509 {
11510 	int rc;
11511 
11512 	if (cntxt->cid > M_CTXTQID)
11513 		return (EINVAL);
11514 
11515 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
11516 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
11517 		return (EINVAL);
11518 
11519 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
11520 	if (rc)
11521 		return (rc);
11522 
11523 	if (hw_off_limits(sc)) {
11524 		rc = ENXIO;
11525 		goto done;
11526 	}
11527 
11528 	if (sc->flags & FW_OK) {
11529 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
11530 		    &cntxt->data[0]);
11531 		if (rc == 0)
11532 			goto done;
11533 	}
11534 
11535 	/*
11536 	 * Read via firmware failed or wasn't even attempted.  Read directly via
11537 	 * the backdoor.
11538 	 */
11539 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
11540 done:
11541 	end_synchronized_op(sc, 0);
11542 	return (rc);
11543 }
11544 
11545 static int
11546 load_fw(struct adapter *sc, struct t4_data *fw)
11547 {
11548 	int rc;
11549 	uint8_t *fw_data;
11550 
11551 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
11552 	if (rc)
11553 		return (rc);
11554 
11555 	if (hw_off_limits(sc)) {
11556 		rc = ENXIO;
11557 		goto done;
11558 	}
11559 
11560 	/*
11561 	 * The firmware, with the sole exception of the memory parity error
11562 	 * handler, runs from memory and not flash.  It is almost always safe to
11563 	 * install a new firmware on a running system.  Just set bit 1 in
11564 	 * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
11565 	 */
11566 	if (sc->flags & FULL_INIT_DONE &&
11567 	    (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
11568 		rc = EBUSY;
11569 		goto done;
11570 	}
11571 
11572 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
11573 
11574 	rc = copyin(fw->data, fw_data, fw->len);
11575 	if (rc == 0)
11576 		rc = -t4_load_fw(sc, fw_data, fw->len);
11577 
11578 	free(fw_data, M_CXGBE);
11579 done:
11580 	end_synchronized_op(sc, 0);
11581 	return (rc);
11582 }
11583 
11584 static int
11585 load_cfg(struct adapter *sc, struct t4_data *cfg)
11586 {
11587 	int rc;
11588 	uint8_t *cfg_data = NULL;
11589 
11590 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
11591 	if (rc)
11592 		return (rc);
11593 
11594 	if (hw_off_limits(sc)) {
11595 		rc = ENXIO;
11596 		goto done;
11597 	}
11598 
11599 	if (cfg->len == 0) {
11600 		/* clear */
11601 		rc = -t4_load_cfg(sc, NULL, 0);
11602 		goto done;
11603 	}
11604 
11605 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
11606 
11607 	rc = copyin(cfg->data, cfg_data, cfg->len);
11608 	if (rc == 0)
11609 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
11610 
11611 	free(cfg_data, M_CXGBE);
11612 done:
11613 	end_synchronized_op(sc, 0);
11614 	return (rc);
11615 }
11616 
11617 static int
11618 load_boot(struct adapter *sc, struct t4_bootrom *br)
11619 {
11620 	int rc;
11621 	uint8_t *br_data = NULL;
11622 	u_int offset;
11623 
11624 	if (br->len > 1024 * 1024)
11625 		return (EFBIG);
11626 
11627 	if (br->pf_offset == 0) {
11628 		/* pfidx */
11629 		if (br->pfidx_addr > 7)
11630 			return (EINVAL);
11631 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
11632 		    A_PCIE_PF_EXPROM_OFST)));
11633 	} else if (br->pf_offset == 1) {
11634 		/* offset */
11635 		offset = G_OFFSET(br->pfidx_addr);
11636 	} else {
11637 		return (EINVAL);
11638 	}
11639 
11640 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
11641 	if (rc)
11642 		return (rc);
11643 
11644 	if (hw_off_limits(sc)) {
11645 		rc = ENXIO;
11646 		goto done;
11647 	}
11648 
11649 	if (br->len == 0) {
11650 		/* clear */
11651 		rc = -t4_load_boot(sc, NULL, offset, 0);
11652 		goto done;
11653 	}
11654 
11655 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
11656 
11657 	rc = copyin(br->data, br_data, br->len);
11658 	if (rc == 0)
11659 		rc = -t4_load_boot(sc, br_data, offset, br->len);
11660 
11661 	free(br_data, M_CXGBE);
11662 done:
11663 	end_synchronized_op(sc, 0);
11664 	return (rc);
11665 }
11666 
11667 static int
11668 load_bootcfg(struct adapter *sc, struct t4_data *bc)
11669 {
11670 	int rc;
11671 	uint8_t *bc_data = NULL;
11672 
11673 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
11674 	if (rc)
11675 		return (rc);
11676 
11677 	if (hw_off_limits(sc)) {
11678 		rc = ENXIO;
11679 		goto done;
11680 	}
11681 
11682 	if (bc->len == 0) {
11683 		/* clear */
11684 		rc = -t4_load_bootcfg(sc, NULL, 0);
11685 		goto done;
11686 	}
11687 
11688 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
11689 
11690 	rc = copyin(bc->data, bc_data, bc->len);
11691 	if (rc == 0)
11692 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
11693 
11694 	free(bc_data, M_CXGBE);
11695 done:
11696 	end_synchronized_op(sc, 0);
11697 	return (rc);
11698 }
11699 
11700 static int
11701 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
11702 {
11703 	int rc;
11704 	struct cudbg_init *cudbg;
11705 	void *handle, *buf;
11706 
11707 	/* buf is large, don't block if no memory is available */
11708 	buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
11709 	if (buf == NULL)
11710 		return (ENOMEM);
11711 
11712 	handle = cudbg_alloc_handle();
11713 	if (handle == NULL) {
11714 		rc = ENOMEM;
11715 		goto done;
11716 	}
11717 
11718 	cudbg = cudbg_get_init(handle);
11719 	cudbg->adap = sc;
11720 	cudbg->print = (cudbg_print_cb)printf;
11721 
11722 #ifndef notyet
11723 	device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
11724 	    __func__, dump->wr_flash, dump->len, dump->data);
11725 #endif
11726 
11727 	if (dump->wr_flash)
11728 		cudbg->use_flash = 1;
11729 	MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
11730 	memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
11731 
11732 	rc = cudbg_collect(handle, buf, &dump->len);
11733 	if (rc != 0)
11734 		goto done;
11735 
11736 	rc = copyout(buf, dump->data, dump->len);
11737 done:
11738 	cudbg_free_handle(handle);
11739 	free(buf, M_CXGBE);
11740 	return (rc);
11741 }
11742 
11743 static void
11744 free_offload_policy(struct t4_offload_policy *op)
11745 {
11746 	struct offload_rule *r;
11747 	int i;
11748 
11749 	if (op == NULL)
11750 		return;
11751 
11752 	r = &op->rule[0];
11753 	for (i = 0; i < op->nrules; i++, r++) {
11754 		free(r->bpf_prog.bf_insns, M_CXGBE);
11755 	}
11756 	free(op->rule, M_CXGBE);
11757 	free(op, M_CXGBE);
11758 }
11759 
11760 static int
11761 set_offload_policy(struct adapter *sc, struct t4_offload_policy *uop)
11762 {
11763 	int i, rc, len;
11764 	struct t4_offload_policy *op, *old;
11765 	struct bpf_program *bf;
11766 	const struct offload_settings *s;
11767 	struct offload_rule *r;
11768 	void *u;
11769 
11770 	if (!is_offload(sc))
11771 		return (ENODEV);
11772 
11773 	if (uop->nrules == 0) {
11774 		/* Delete installed policies. */
11775 		op = NULL;
11776 		goto set_policy;
11777 	} else if (uop->nrules > 256) { /* arbitrary */
11778 		return (E2BIG);
11779 	}
11780 
11781 	/* Copy userspace offload policy to kernel */
11782 	op = malloc(sizeof(*op), M_CXGBE, M_ZERO | M_WAITOK);
11783 	op->nrules = uop->nrules;
11784 	len = op->nrules * sizeof(struct offload_rule);
11785 	op->rule = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
11786 	rc = copyin(uop->rule, op->rule, len);
11787 	if (rc) {
11788 		free(op->rule, M_CXGBE);
11789 		free(op, M_CXGBE);
11790 		return (rc);
11791 	}
11792 
11793 	r = &op->rule[0];
11794 	for (i = 0; i < op->nrules; i++, r++) {
11795 
11796 		/* Validate open_type */
11797 		if (r->open_type != OPEN_TYPE_LISTEN &&
11798 		    r->open_type != OPEN_TYPE_ACTIVE &&
11799 		    r->open_type != OPEN_TYPE_PASSIVE &&
11800 		    r->open_type != OPEN_TYPE_DONTCARE) {
11801 error:
11802 			/*
11803 			 * Rules 0 to i have malloc'd filters that need to be
11804 			 * freed.  Rules i+1 to nrules have userspace pointers
11805 			 * and should be left alone.
11806 			 */
11807 			op->nrules = i;
11808 			free_offload_policy(op);
11809 			return (rc);
11810 		}
11811 
11812 		/* Validate settings */
11813 		s = &r->settings;
11814 		if ((s->offload != 0 && s->offload != 1) ||
11815 		    s->cong_algo < -1 || s->cong_algo > CONG_ALG_HIGHSPEED ||
11816 		    s->sched_class < -1 ||
11817 		    s->sched_class >= sc->params.nsched_cls) {
11818 			rc = EINVAL;
11819 			goto error;
11820 		}
11821 
11822 		bf = &r->bpf_prog;
11823 		u = bf->bf_insns;	/* userspace ptr */
11824 		bf->bf_insns = NULL;
11825 		if (bf->bf_len == 0) {
11826 			/* legal, matches everything */
11827 			continue;
11828 		}
11829 		len = bf->bf_len * sizeof(*bf->bf_insns);
11830 		bf->bf_insns = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
11831 		rc = copyin(u, bf->bf_insns, len);
11832 		if (rc != 0)
11833 			goto error;
11834 
11835 		if (!bpf_validate(bf->bf_insns, bf->bf_len)) {
11836 			rc = EINVAL;
11837 			goto error;
11838 		}
11839 	}
11840 set_policy:
11841 	rw_wlock(&sc->policy_lock);
11842 	old = sc->policy;
11843 	sc->policy = op;
11844 	rw_wunlock(&sc->policy_lock);
11845 	free_offload_policy(old);
11846 
11847 	return (0);
11848 }
11849 
11850 #define MAX_READ_BUF_SIZE (128 * 1024)
11851 static int
11852 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
11853 {
11854 	uint32_t addr, remaining, n;
11855 	uint32_t *buf;
11856 	int rc;
11857 	uint8_t *dst;
11858 
11859 	mtx_lock(&sc->reg_lock);
11860 	if (hw_off_limits(sc))
11861 		rc = ENXIO;
11862 	else
11863 		rc = validate_mem_range(sc, mr->addr, mr->len);
11864 	mtx_unlock(&sc->reg_lock);
11865 	if (rc != 0)
11866 		return (rc);
11867 
11868 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
11869 	addr = mr->addr;
11870 	remaining = mr->len;
11871 	dst = (void *)mr->data;
11872 
11873 	while (remaining) {
11874 		n = min(remaining, MAX_READ_BUF_SIZE);
11875 		mtx_lock(&sc->reg_lock);
11876 		if (hw_off_limits(sc))
11877 			rc = ENXIO;
11878 		else
11879 			read_via_memwin(sc, 2, addr, buf, n);
11880 		mtx_unlock(&sc->reg_lock);
11881 		if (rc != 0)
11882 			break;
11883 
11884 		rc = copyout(buf, dst, n);
11885 		if (rc != 0)
11886 			break;
11887 
11888 		dst += n;
11889 		remaining -= n;
11890 		addr += n;
11891 	}
11892 
11893 	free(buf, M_CXGBE);
11894 	return (rc);
11895 }
11896 #undef MAX_READ_BUF_SIZE
11897 
11898 static int
11899 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
11900 {
11901 	int rc;
11902 
11903 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
11904 		return (EINVAL);
11905 
11906 	if (i2cd->len > sizeof(i2cd->data))
11907 		return (EFBIG);
11908 
11909 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
11910 	if (rc)
11911 		return (rc);
11912 	if (hw_off_limits(sc))
11913 		rc = ENXIO;
11914 	else
11915 		rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
11916 		    i2cd->offset, i2cd->len, &i2cd->data[0]);
11917 	end_synchronized_op(sc, 0);
11918 
11919 	return (rc);
11920 }
11921 
11922 static int
11923 clear_stats(struct adapter *sc, u_int port_id)
11924 {
11925 	int i, v, chan_map;
11926 	struct port_info *pi;
11927 	struct vi_info *vi;
11928 	struct sge_rxq *rxq;
11929 	struct sge_txq *txq;
11930 	struct sge_wrq *wrq;
11931 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
11932 	struct sge_ofld_txq *ofld_txq;
11933 #endif
11934 #ifdef TCP_OFFLOAD
11935 	struct sge_ofld_rxq *ofld_rxq;
11936 #endif
11937 
11938 	if (port_id >= sc->params.nports)
11939 		return (EINVAL);
11940 	pi = sc->port[port_id];
11941 	if (pi == NULL)
11942 		return (EIO);
11943 
11944 	mtx_lock(&sc->reg_lock);
11945 	if (!hw_off_limits(sc)) {
11946 		/* MAC stats */
11947 		t4_clr_port_stats(sc, pi->tx_chan);
11948 		if (is_t6(sc)) {
11949 			if (pi->fcs_reg != -1)
11950 				pi->fcs_base = t4_read_reg64(sc, pi->fcs_reg);
11951 			else
11952 				pi->stats.rx_fcs_err = 0;
11953 		}
11954 		for_each_vi(pi, v, vi) {
11955 			if (vi->flags & VI_INIT_DONE)
11956 				t4_clr_vi_stats(sc, vi->vin);
11957 		}
11958 		chan_map = pi->rx_e_chan_map;
11959 		v = 0;	/* reuse */
11960 		while (chan_map) {
11961 			i = ffs(chan_map) - 1;
11962 			t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
11963 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
11964 			chan_map &= ~(1 << i);
11965 		}
11966 	}
11967 	mtx_unlock(&sc->reg_lock);
11968 	pi->tx_parse_error = 0;
11969 	pi->tnl_cong_drops = 0;
11970 
11971 	/*
11972 	 * Since this command accepts a port, clear stats for
11973 	 * all VIs on this port.
11974 	 */
11975 	for_each_vi(pi, v, vi) {
11976 		if (vi->flags & VI_INIT_DONE) {
11977 
11978 			for_each_rxq(vi, i, rxq) {
11979 #if defined(INET) || defined(INET6)
11980 				rxq->lro.lro_queued = 0;
11981 				rxq->lro.lro_flushed = 0;
11982 #endif
11983 				rxq->rxcsum = 0;
11984 				rxq->vlan_extraction = 0;
11985 				rxq->vxlan_rxcsum = 0;
11986 
11987 				rxq->fl.cl_allocated = 0;
11988 				rxq->fl.cl_recycled = 0;
11989 				rxq->fl.cl_fast_recycled = 0;
11990 			}
11991 
11992 			for_each_txq(vi, i, txq) {
11993 				txq->txcsum = 0;
11994 				txq->tso_wrs = 0;
11995 				txq->vlan_insertion = 0;
11996 				txq->imm_wrs = 0;
11997 				txq->sgl_wrs = 0;
11998 				txq->txpkt_wrs = 0;
11999 				txq->txpkts0_wrs = 0;
12000 				txq->txpkts1_wrs = 0;
12001 				txq->txpkts0_pkts = 0;
12002 				txq->txpkts1_pkts = 0;
12003 				txq->txpkts_flush = 0;
12004 				txq->raw_wrs = 0;
12005 				txq->vxlan_tso_wrs = 0;
12006 				txq->vxlan_txcsum = 0;
12007 				txq->kern_tls_records = 0;
12008 				txq->kern_tls_short = 0;
12009 				txq->kern_tls_partial = 0;
12010 				txq->kern_tls_full = 0;
12011 				txq->kern_tls_octets = 0;
12012 				txq->kern_tls_waste = 0;
12013 				txq->kern_tls_options = 0;
12014 				txq->kern_tls_header = 0;
12015 				txq->kern_tls_fin = 0;
12016 				txq->kern_tls_fin_short = 0;
12017 				txq->kern_tls_cbc = 0;
12018 				txq->kern_tls_gcm = 0;
12019 				mp_ring_reset_stats(txq->r);
12020 			}
12021 
12022 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
12023 			for_each_ofld_txq(vi, i, ofld_txq) {
12024 				ofld_txq->wrq.tx_wrs_direct = 0;
12025 				ofld_txq->wrq.tx_wrs_copied = 0;
12026 				counter_u64_zero(ofld_txq->tx_iscsi_pdus);
12027 				counter_u64_zero(ofld_txq->tx_iscsi_octets);
12028 				counter_u64_zero(ofld_txq->tx_iscsi_iso_wrs);
12029 				counter_u64_zero(ofld_txq->tx_toe_tls_records);
12030 				counter_u64_zero(ofld_txq->tx_toe_tls_octets);
12031 			}
12032 #endif
12033 #ifdef TCP_OFFLOAD
12034 			for_each_ofld_rxq(vi, i, ofld_rxq) {
12035 				ofld_rxq->fl.cl_allocated = 0;
12036 				ofld_rxq->fl.cl_recycled = 0;
12037 				ofld_rxq->fl.cl_fast_recycled = 0;
12038 				counter_u64_zero(
12039 				    ofld_rxq->rx_iscsi_ddp_setup_ok);
12040 				counter_u64_zero(
12041 				    ofld_rxq->rx_iscsi_ddp_setup_error);
12042 				ofld_rxq->rx_iscsi_ddp_pdus = 0;
12043 				ofld_rxq->rx_iscsi_ddp_octets = 0;
12044 				ofld_rxq->rx_iscsi_fl_pdus = 0;
12045 				ofld_rxq->rx_iscsi_fl_octets = 0;
12046 				ofld_rxq->rx_toe_tls_records = 0;
12047 				ofld_rxq->rx_toe_tls_octets = 0;
12048 			}
12049 #endif
12050 
12051 			if (IS_MAIN_VI(vi)) {
12052 				wrq = &sc->sge.ctrlq[pi->port_id];
12053 				wrq->tx_wrs_direct = 0;
12054 				wrq->tx_wrs_copied = 0;
12055 			}
12056 		}
12057 	}
12058 
12059 	return (0);
12060 }
12061 
12062 static int
12063 hold_clip_addr(struct adapter *sc, struct t4_clip_addr *ca)
12064 {
12065 #ifdef INET6
12066 	struct in6_addr in6;
12067 
12068 	bcopy(&ca->addr[0], &in6.s6_addr[0], sizeof(in6.s6_addr));
12069 	if (t4_get_clip_entry(sc, &in6, true) != NULL)
12070 		return (0);
12071 	else
12072 		return (EIO);
12073 #else
12074 	return (ENOTSUP);
12075 #endif
12076 }
12077 
12078 static int
12079 release_clip_addr(struct adapter *sc, struct t4_clip_addr *ca)
12080 {
12081 #ifdef INET6
12082 	struct in6_addr in6;
12083 
12084 	bcopy(&ca->addr[0], &in6.s6_addr[0], sizeof(in6.s6_addr));
12085 	return (t4_release_clip_addr(sc, &in6));
12086 #else
12087 	return (ENOTSUP);
12088 #endif
12089 }
12090 
12091 int
12092 t4_os_find_pci_capability(struct adapter *sc, int cap)
12093 {
12094 	int i;
12095 
12096 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
12097 }
12098 
12099 int
12100 t4_os_pci_save_state(struct adapter *sc)
12101 {
12102 	device_t dev;
12103 	struct pci_devinfo *dinfo;
12104 
12105 	dev = sc->dev;
12106 	dinfo = device_get_ivars(dev);
12107 
12108 	pci_cfg_save(dev, dinfo, 0);
12109 	return (0);
12110 }
12111 
12112 int
12113 t4_os_pci_restore_state(struct adapter *sc)
12114 {
12115 	device_t dev;
12116 	struct pci_devinfo *dinfo;
12117 
12118 	dev = sc->dev;
12119 	dinfo = device_get_ivars(dev);
12120 
12121 	pci_cfg_restore(dev, dinfo);
12122 	return (0);
12123 }
12124 
12125 void
12126 t4_os_portmod_changed(struct port_info *pi)
12127 {
12128 	struct adapter *sc = pi->adapter;
12129 	struct vi_info *vi;
12130 	if_t ifp;
12131 	static const char *mod_str[] = {
12132 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
12133 	};
12134 
12135 	KASSERT((pi->flags & FIXED_IFMEDIA) == 0,
12136 	    ("%s: port_type %u", __func__, pi->port_type));
12137 
12138 	vi = &pi->vi[0];
12139 	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
12140 		PORT_LOCK(pi);
12141 		build_medialist(pi);
12142 		if (pi->mod_type != FW_PORT_MOD_TYPE_NONE) {
12143 			fixup_link_config(pi);
12144 			apply_link_config(pi);
12145 		}
12146 		PORT_UNLOCK(pi);
12147 		end_synchronized_op(sc, LOCK_HELD);
12148 	}
12149 
12150 	ifp = vi->ifp;
12151 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
12152 		if_printf(ifp, "transceiver unplugged.\n");
12153 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
12154 		if_printf(ifp, "unknown transceiver inserted.\n");
12155 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
12156 		if_printf(ifp, "unsupported transceiver inserted.\n");
12157 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
12158 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
12159 		    port_top_speed(pi), mod_str[pi->mod_type]);
12160 	} else {
12161 		if_printf(ifp, "transceiver (type %d) inserted.\n",
12162 		    pi->mod_type);
12163 	}
12164 }
12165 
12166 void
12167 t4_os_link_changed(struct port_info *pi)
12168 {
12169 	struct vi_info *vi;
12170 	if_t ifp;
12171 	struct link_config *lc = &pi->link_cfg;
12172 	struct adapter *sc = pi->adapter;
12173 	int v;
12174 
12175 	PORT_LOCK_ASSERT_OWNED(pi);
12176 
12177 	if (is_t6(sc)) {
12178 		if (lc->link_ok) {
12179 			if (lc->speed > 25000 ||
12180 			    (lc->speed == 25000 && lc->fec == FEC_RS)) {
12181 				pi->fcs_reg = T5_PORT_REG(pi->tx_chan,
12182 				    A_MAC_PORT_AFRAMECHECKSEQUENCEERRORS);
12183 			} else {
12184 				pi->fcs_reg = T5_PORT_REG(pi->tx_chan,
12185 				    A_MAC_PORT_MTIP_1G10G_RX_CRCERRORS);
12186 			}
12187 			pi->fcs_base = t4_read_reg64(sc, pi->fcs_reg);
12188 			pi->stats.rx_fcs_err = 0;
12189 		} else {
12190 			pi->fcs_reg = -1;
12191 		}
12192 	} else {
12193 		MPASS(pi->fcs_reg != -1);
12194 		MPASS(pi->fcs_base == 0);
12195 	}
12196 
12197 	for_each_vi(pi, v, vi) {
12198 		ifp = vi->ifp;
12199 		if (ifp == NULL)
12200 			continue;
12201 
12202 		if (lc->link_ok) {
12203 			if_setbaudrate(ifp, IF_Mbps(lc->speed));
12204 			if_link_state_change(ifp, LINK_STATE_UP);
12205 		} else {
12206 			if_link_state_change(ifp, LINK_STATE_DOWN);
12207 		}
12208 	}
12209 }
12210 
12211 void
12212 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
12213 {
12214 	struct adapter *sc;
12215 
12216 	sx_slock(&t4_list_lock);
12217 	SLIST_FOREACH(sc, &t4_list, link) {
12218 		/*
12219 		 * func should not make any assumptions about what state sc is
12220 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
12221 		 */
12222 		func(sc, arg);
12223 	}
12224 	sx_sunlock(&t4_list_lock);
12225 }
12226 
12227 static int
12228 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
12229     struct thread *td)
12230 {
12231 	int rc;
12232 	struct adapter *sc = dev->si_drv1;
12233 
12234 	rc = priv_check(td, PRIV_DRIVER);
12235 	if (rc != 0)
12236 		return (rc);
12237 
12238 	switch (cmd) {
12239 	case CHELSIO_T4_GETREG: {
12240 		struct t4_reg *edata = (struct t4_reg *)data;
12241 
12242 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
12243 			return (EFAULT);
12244 
12245 		mtx_lock(&sc->reg_lock);
12246 		if (hw_off_limits(sc))
12247 			rc = ENXIO;
12248 		else if (edata->size == 4)
12249 			edata->val = t4_read_reg(sc, edata->addr);
12250 		else if (edata->size == 8)
12251 			edata->val = t4_read_reg64(sc, edata->addr);
12252 		else
12253 			rc = EINVAL;
12254 		mtx_unlock(&sc->reg_lock);
12255 
12256 		break;
12257 	}
12258 	case CHELSIO_T4_SETREG: {
12259 		struct t4_reg *edata = (struct t4_reg *)data;
12260 
12261 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
12262 			return (EFAULT);
12263 
12264 		mtx_lock(&sc->reg_lock);
12265 		if (hw_off_limits(sc))
12266 			rc = ENXIO;
12267 		else if (edata->size == 4) {
12268 			if (edata->val & 0xffffffff00000000)
12269 				rc = EINVAL;
12270 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
12271 		} else if (edata->size == 8)
12272 			t4_write_reg64(sc, edata->addr, edata->val);
12273 		else
12274 			rc = EINVAL;
12275 		mtx_unlock(&sc->reg_lock);
12276 
12277 		break;
12278 	}
12279 	case CHELSIO_T4_REGDUMP: {
12280 		struct t4_regdump *regs = (struct t4_regdump *)data;
12281 		int reglen = t4_get_regs_len(sc);
12282 		uint8_t *buf;
12283 
12284 		if (regs->len < reglen) {
12285 			regs->len = reglen; /* hint to the caller */
12286 			return (ENOBUFS);
12287 		}
12288 
12289 		regs->len = reglen;
12290 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
12291 		mtx_lock(&sc->reg_lock);
12292 		if (hw_off_limits(sc))
12293 			rc = ENXIO;
12294 		else
12295 			get_regs(sc, regs, buf);
12296 		mtx_unlock(&sc->reg_lock);
12297 		if (rc == 0)
12298 			rc = copyout(buf, regs->data, reglen);
12299 		free(buf, M_CXGBE);
12300 		break;
12301 	}
12302 	case CHELSIO_T4_GET_FILTER_MODE:
12303 		rc = get_filter_mode(sc, (uint32_t *)data);
12304 		break;
12305 	case CHELSIO_T4_SET_FILTER_MODE:
12306 		rc = set_filter_mode(sc, *(uint32_t *)data);
12307 		break;
12308 	case CHELSIO_T4_SET_FILTER_MASK:
12309 		rc = set_filter_mask(sc, *(uint32_t *)data);
12310 		break;
12311 	case CHELSIO_T4_GET_FILTER:
12312 		rc = get_filter(sc, (struct t4_filter *)data);
12313 		break;
12314 	case CHELSIO_T4_SET_FILTER:
12315 		rc = set_filter(sc, (struct t4_filter *)data);
12316 		break;
12317 	case CHELSIO_T4_DEL_FILTER:
12318 		rc = del_filter(sc, (struct t4_filter *)data);
12319 		break;
12320 	case CHELSIO_T4_GET_SGE_CONTEXT:
12321 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
12322 		break;
12323 	case CHELSIO_T4_LOAD_FW:
12324 		rc = load_fw(sc, (struct t4_data *)data);
12325 		break;
12326 	case CHELSIO_T4_GET_MEM:
12327 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
12328 		break;
12329 	case CHELSIO_T4_GET_I2C:
12330 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
12331 		break;
12332 	case CHELSIO_T4_CLEAR_STATS:
12333 		rc = clear_stats(sc, *(uint32_t *)data);
12334 		break;
12335 	case CHELSIO_T4_SCHED_CLASS:
12336 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
12337 		break;
12338 	case CHELSIO_T4_SCHED_QUEUE:
12339 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
12340 		break;
12341 	case CHELSIO_T4_GET_TRACER:
12342 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
12343 		break;
12344 	case CHELSIO_T4_SET_TRACER:
12345 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
12346 		break;
12347 	case CHELSIO_T4_LOAD_CFG:
12348 		rc = load_cfg(sc, (struct t4_data *)data);
12349 		break;
12350 	case CHELSIO_T4_LOAD_BOOT:
12351 		rc = load_boot(sc, (struct t4_bootrom *)data);
12352 		break;
12353 	case CHELSIO_T4_LOAD_BOOTCFG:
12354 		rc = load_bootcfg(sc, (struct t4_data *)data);
12355 		break;
12356 	case CHELSIO_T4_CUDBG_DUMP:
12357 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
12358 		break;
12359 	case CHELSIO_T4_SET_OFLD_POLICY:
12360 		rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
12361 		break;
12362 	case CHELSIO_T4_HOLD_CLIP_ADDR:
12363 		rc = hold_clip_addr(sc, (struct t4_clip_addr *)data);
12364 		break;
12365 	case CHELSIO_T4_RELEASE_CLIP_ADDR:
12366 		rc = release_clip_addr(sc, (struct t4_clip_addr *)data);
12367 		break;
12368 	default:
12369 		rc = ENOTTY;
12370 	}
12371 
12372 	return (rc);
12373 }
12374 
12375 #ifdef TCP_OFFLOAD
12376 static int
12377 toe_capability(struct vi_info *vi, bool enable)
12378 {
12379 	int rc;
12380 	struct port_info *pi = vi->pi;
12381 	struct adapter *sc = pi->adapter;
12382 
12383 	ASSERT_SYNCHRONIZED_OP(sc);
12384 
12385 	if (!is_offload(sc))
12386 		return (ENODEV);
12387 	if (hw_off_limits(sc))
12388 		return (ENXIO);
12389 
12390 	if (enable) {
12391 #ifdef KERN_TLS
12392 		if (sc->flags & KERN_TLS_ON && is_t6(sc)) {
12393 			int i, j, n;
12394 			struct port_info *p;
12395 			struct vi_info *v;
12396 
12397 			/*
12398 			 * Reconfigure hardware for TOE if TXTLS is not enabled
12399 			 * on any ifnet.
12400 			 */
12401 			n = 0;
12402 			for_each_port(sc, i) {
12403 				p = sc->port[i];
12404 				for_each_vi(p, j, v) {
12405 					if (if_getcapenable(v->ifp) & IFCAP_TXTLS) {
12406 						CH_WARN(sc,
12407 						    "%s has NIC TLS enabled.\n",
12408 						    device_get_nameunit(v->dev));
12409 						n++;
12410 					}
12411 				}
12412 			}
12413 			if (n > 0) {
12414 				CH_WARN(sc, "Disable NIC TLS on all interfaces "
12415 				    "associated with this adapter before "
12416 				    "trying to enable TOE.\n");
12417 				return (EAGAIN);
12418 			}
12419 			rc = t6_config_kern_tls(sc, false);
12420 			if (rc)
12421 				return (rc);
12422 		}
12423 #endif
12424 		if ((if_getcapenable(vi->ifp) & IFCAP_TOE) != 0) {
12425 			/* TOE is already enabled. */
12426 			return (0);
12427 		}
12428 
12429 		/*
12430 		 * We need the port's queues around so that we're able to send
12431 		 * and receive CPLs to/from the TOE even if the ifnet for this
12432 		 * port has never been UP'd administratively.
12433 		 */
12434 		if (!(vi->flags & VI_INIT_DONE) && ((rc = vi_init(vi)) != 0))
12435 			return (rc);
12436 		if (!(pi->vi[0].flags & VI_INIT_DONE) &&
12437 		    ((rc = vi_init(&pi->vi[0])) != 0))
12438 			return (rc);
12439 
12440 		if (isset(&sc->offload_map, pi->port_id)) {
12441 			/* TOE is enabled on another VI of this port. */
12442 			pi->uld_vis++;
12443 			return (0);
12444 		}
12445 
12446 		if (!uld_active(sc, ULD_TOM)) {
12447 			rc = t4_activate_uld(sc, ULD_TOM);
12448 			if (rc == EAGAIN) {
12449 				log(LOG_WARNING,
12450 				    "You must kldload t4_tom.ko before trying "
12451 				    "to enable TOE on a cxgbe interface.\n");
12452 			}
12453 			if (rc != 0)
12454 				return (rc);
12455 			KASSERT(sc->tom_softc != NULL,
12456 			    ("%s: TOM activated but softc NULL", __func__));
12457 			KASSERT(uld_active(sc, ULD_TOM),
12458 			    ("%s: TOM activated but flag not set", __func__));
12459 		}
12460 
12461 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
12462 		if (!uld_active(sc, ULD_IWARP))
12463 			(void) t4_activate_uld(sc, ULD_IWARP);
12464 		if (!uld_active(sc, ULD_ISCSI))
12465 			(void) t4_activate_uld(sc, ULD_ISCSI);
12466 
12467 		pi->uld_vis++;
12468 		setbit(&sc->offload_map, pi->port_id);
12469 	} else {
12470 		pi->uld_vis--;
12471 
12472 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
12473 			return (0);
12474 
12475 		KASSERT(uld_active(sc, ULD_TOM),
12476 		    ("%s: TOM never initialized?", __func__));
12477 		clrbit(&sc->offload_map, pi->port_id);
12478 	}
12479 
12480 	return (0);
12481 }
12482 
12483 /*
12484  * Add an upper layer driver to the global list.
12485  */
12486 int
12487 t4_register_uld(struct uld_info *ui)
12488 {
12489 	int rc = 0;
12490 	struct uld_info *u;
12491 
12492 	sx_xlock(&t4_uld_list_lock);
12493 	SLIST_FOREACH(u, &t4_uld_list, link) {
12494 	    if (u->uld_id == ui->uld_id) {
12495 		    rc = EEXIST;
12496 		    goto done;
12497 	    }
12498 	}
12499 
12500 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
12501 	ui->refcount = 0;
12502 done:
12503 	sx_xunlock(&t4_uld_list_lock);
12504 	return (rc);
12505 }
12506 
12507 int
12508 t4_unregister_uld(struct uld_info *ui)
12509 {
12510 	int rc = EINVAL;
12511 	struct uld_info *u;
12512 
12513 	sx_xlock(&t4_uld_list_lock);
12514 
12515 	SLIST_FOREACH(u, &t4_uld_list, link) {
12516 	    if (u == ui) {
12517 		    if (ui->refcount > 0) {
12518 			    rc = EBUSY;
12519 			    goto done;
12520 		    }
12521 
12522 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
12523 		    rc = 0;
12524 		    goto done;
12525 	    }
12526 	}
12527 done:
12528 	sx_xunlock(&t4_uld_list_lock);
12529 	return (rc);
12530 }
12531 
12532 int
12533 t4_activate_uld(struct adapter *sc, int id)
12534 {
12535 	int rc;
12536 	struct uld_info *ui;
12537 
12538 	ASSERT_SYNCHRONIZED_OP(sc);
12539 
12540 	if (id < 0 || id > ULD_MAX)
12541 		return (EINVAL);
12542 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
12543 
12544 	sx_slock(&t4_uld_list_lock);
12545 
12546 	SLIST_FOREACH(ui, &t4_uld_list, link) {
12547 		if (ui->uld_id == id) {
12548 			if (!(sc->flags & FULL_INIT_DONE)) {
12549 				rc = adapter_init(sc);
12550 				if (rc != 0)
12551 					break;
12552 			}
12553 
12554 			rc = ui->activate(sc);
12555 			if (rc == 0) {
12556 				setbit(&sc->active_ulds, id);
12557 				ui->refcount++;
12558 			}
12559 			break;
12560 		}
12561 	}
12562 
12563 	sx_sunlock(&t4_uld_list_lock);
12564 
12565 	return (rc);
12566 }
12567 
12568 int
12569 t4_deactivate_uld(struct adapter *sc, int id)
12570 {
12571 	int rc;
12572 	struct uld_info *ui;
12573 
12574 	ASSERT_SYNCHRONIZED_OP(sc);
12575 
12576 	if (id < 0 || id > ULD_MAX)
12577 		return (EINVAL);
12578 	rc = ENXIO;
12579 
12580 	sx_slock(&t4_uld_list_lock);
12581 
12582 	SLIST_FOREACH(ui, &t4_uld_list, link) {
12583 		if (ui->uld_id == id) {
12584 			rc = ui->deactivate(sc);
12585 			if (rc == 0) {
12586 				clrbit(&sc->active_ulds, id);
12587 				ui->refcount--;
12588 			}
12589 			break;
12590 		}
12591 	}
12592 
12593 	sx_sunlock(&t4_uld_list_lock);
12594 
12595 	return (rc);
12596 }
12597 
12598 static int
12599 t4_deactivate_all_uld(struct adapter *sc)
12600 {
12601 	int rc;
12602 	struct uld_info *ui;
12603 
12604 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK, "t4detuld");
12605 	if (rc != 0)
12606 		return (ENXIO);
12607 
12608 	sx_slock(&t4_uld_list_lock);
12609 
12610 	SLIST_FOREACH(ui, &t4_uld_list, link) {
12611 		if (isset(&sc->active_ulds, ui->uld_id)) {
12612 			rc = ui->deactivate(sc);
12613 			if (rc != 0)
12614 				break;
12615 			clrbit(&sc->active_ulds, ui->uld_id);
12616 			ui->refcount--;
12617 		}
12618 	}
12619 
12620 	sx_sunlock(&t4_uld_list_lock);
12621 	end_synchronized_op(sc, 0);
12622 
12623 	return (rc);
12624 }
12625 
12626 static void
12627 t4_async_event(struct adapter *sc)
12628 {
12629 	struct uld_info *ui;
12630 
12631 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4async") != 0)
12632 		return;
12633 	sx_slock(&t4_uld_list_lock);
12634 	SLIST_FOREACH(ui, &t4_uld_list, link) {
12635 		if (ui->uld_id == ULD_IWARP) {
12636 			ui->async_event(sc);
12637 			break;
12638 		}
12639 	}
12640 	sx_sunlock(&t4_uld_list_lock);
12641 	end_synchronized_op(sc, 0);
12642 }
12643 
12644 int
12645 uld_active(struct adapter *sc, int uld_id)
12646 {
12647 
12648 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
12649 
12650 	return (isset(&sc->active_ulds, uld_id));
12651 }
12652 #endif
12653 
12654 #ifdef KERN_TLS
12655 static int
12656 ktls_capability(struct adapter *sc, bool enable)
12657 {
12658 	ASSERT_SYNCHRONIZED_OP(sc);
12659 
12660 	if (!is_ktls(sc))
12661 		return (ENODEV);
12662 	if (!is_t6(sc))
12663 		return (0);
12664 	if (hw_off_limits(sc))
12665 		return (ENXIO);
12666 
12667 	if (enable) {
12668 		if (sc->flags & KERN_TLS_ON)
12669 			return (0);	/* already on */
12670 		if (sc->offload_map != 0) {
12671 			CH_WARN(sc,
12672 			    "Disable TOE on all interfaces associated with "
12673 			    "this adapter before trying to enable NIC TLS.\n");
12674 			return (EAGAIN);
12675 		}
12676 		return (t6_config_kern_tls(sc, true));
12677 	} else {
12678 		/*
12679 		 * Nothing to do for disable.  If TOE is enabled sometime later
12680 		 * then toe_capability will reconfigure the hardware.
12681 		 */
12682 		return (0);
12683 	}
12684 }
12685 #endif
12686 
12687 /*
12688  * t  = ptr to tunable.
12689  * nc = number of CPUs.
12690  * c  = compiled in default for that tunable.
12691  */
12692 static void
12693 calculate_nqueues(int *t, int nc, const int c)
12694 {
12695 	int nq;
12696 
12697 	if (*t > 0)
12698 		return;
12699 	nq = *t < 0 ? -*t : c;
12700 	*t = min(nc, nq);
12701 }
12702 
12703 /*
12704  * Come up with reasonable defaults for some of the tunables, provided they're
12705  * not set by the user (in which case we'll use the values as is).
12706  */
12707 static void
12708 tweak_tunables(void)
12709 {
12710 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
12711 
12712 	if (t4_ntxq < 1) {
12713 #ifdef RSS
12714 		t4_ntxq = rss_getnumbuckets();
12715 #else
12716 		calculate_nqueues(&t4_ntxq, nc, NTXQ);
12717 #endif
12718 	}
12719 
12720 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
12721 
12722 	if (t4_nrxq < 1) {
12723 #ifdef RSS
12724 		t4_nrxq = rss_getnumbuckets();
12725 #else
12726 		calculate_nqueues(&t4_nrxq, nc, NRXQ);
12727 #endif
12728 	}
12729 
12730 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
12731 
12732 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
12733 	calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
12734 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
12735 #endif
12736 #ifdef TCP_OFFLOAD
12737 	calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
12738 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
12739 #endif
12740 
12741 #if defined(TCP_OFFLOAD) || defined(KERN_TLS)
12742 	if (t4_toecaps_allowed == -1)
12743 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
12744 #else
12745 	if (t4_toecaps_allowed == -1)
12746 		t4_toecaps_allowed = 0;
12747 #endif
12748 
12749 #ifdef TCP_OFFLOAD
12750 	if (t4_rdmacaps_allowed == -1) {
12751 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
12752 		    FW_CAPS_CONFIG_RDMA_RDMAC;
12753 	}
12754 
12755 	if (t4_iscsicaps_allowed == -1) {
12756 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
12757 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
12758 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
12759 	}
12760 
12761 	if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
12762 		t4_tmr_idx_ofld = TMR_IDX_OFLD;
12763 
12764 	if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
12765 		t4_pktc_idx_ofld = PKTC_IDX_OFLD;
12766 #else
12767 	if (t4_rdmacaps_allowed == -1)
12768 		t4_rdmacaps_allowed = 0;
12769 
12770 	if (t4_iscsicaps_allowed == -1)
12771 		t4_iscsicaps_allowed = 0;
12772 #endif
12773 
12774 #ifdef DEV_NETMAP
12775 	calculate_nqueues(&t4_nnmtxq, nc, NNMTXQ);
12776 	calculate_nqueues(&t4_nnmrxq, nc, NNMRXQ);
12777 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
12778 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
12779 #endif
12780 
12781 	if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
12782 		t4_tmr_idx = TMR_IDX;
12783 
12784 	if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
12785 		t4_pktc_idx = PKTC_IDX;
12786 
12787 	if (t4_qsize_txq < 128)
12788 		t4_qsize_txq = 128;
12789 
12790 	if (t4_qsize_rxq < 128)
12791 		t4_qsize_rxq = 128;
12792 	while (t4_qsize_rxq & 7)
12793 		t4_qsize_rxq++;
12794 
12795 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
12796 
12797 	/*
12798 	 * Number of VIs to create per-port.  The first VI is the "main" regular
12799 	 * VI for the port.  The rest are additional virtual interfaces on the
12800 	 * same physical port.  Note that the main VI does not have native
12801 	 * netmap support but the extra VIs do.
12802 	 *
12803 	 * Limit the number of VIs per port to the number of available
12804 	 * MAC addresses per port.
12805 	 */
12806 	if (t4_num_vis < 1)
12807 		t4_num_vis = 1;
12808 	if (t4_num_vis > nitems(vi_mac_funcs)) {
12809 		t4_num_vis = nitems(vi_mac_funcs);
12810 		printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
12811 	}
12812 
12813 	if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
12814 		pcie_relaxed_ordering = 1;
12815 #if defined(__i386__) || defined(__amd64__)
12816 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
12817 			pcie_relaxed_ordering = 0;
12818 #endif
12819 	}
12820 }
12821 
12822 #ifdef DDB
12823 static void
12824 t4_dump_tcb(struct adapter *sc, int tid)
12825 {
12826 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
12827 
12828 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
12829 	save = t4_read_reg(sc, reg);
12830 	base = sc->memwin[2].mw_base;
12831 
12832 	/* Dump TCB for the tid */
12833 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
12834 	tcb_addr += tid * TCB_SIZE;
12835 
12836 	if (is_t4(sc)) {
12837 		pf = 0;
12838 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
12839 	} else {
12840 		pf = V_PFNUM(sc->pf);
12841 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
12842 	}
12843 	t4_write_reg(sc, reg, win_pos | pf);
12844 	t4_read_reg(sc, reg);
12845 
12846 	off = tcb_addr - win_pos;
12847 	for (i = 0; i < 4; i++) {
12848 		uint32_t buf[8];
12849 		for (j = 0; j < 8; j++, off += 4)
12850 			buf[j] = htonl(t4_read_reg(sc, base + off));
12851 
12852 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
12853 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
12854 		    buf[7]);
12855 	}
12856 
12857 	t4_write_reg(sc, reg, save);
12858 	t4_read_reg(sc, reg);
12859 }
12860 
12861 static void
12862 t4_dump_devlog(struct adapter *sc)
12863 {
12864 	struct devlog_params *dparams = &sc->params.devlog;
12865 	struct fw_devlog_e e;
12866 	int i, first, j, m, nentries, rc;
12867 	uint64_t ftstamp = UINT64_MAX;
12868 
12869 	if (dparams->start == 0) {
12870 		db_printf("devlog params not valid\n");
12871 		return;
12872 	}
12873 
12874 	nentries = dparams->size / sizeof(struct fw_devlog_e);
12875 	m = fwmtype_to_hwmtype(dparams->memtype);
12876 
12877 	/* Find the first entry. */
12878 	first = -1;
12879 	for (i = 0; i < nentries && !db_pager_quit; i++) {
12880 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
12881 		    sizeof(e), (void *)&e);
12882 		if (rc != 0)
12883 			break;
12884 
12885 		if (e.timestamp == 0)
12886 			break;
12887 
12888 		e.timestamp = be64toh(e.timestamp);
12889 		if (e.timestamp < ftstamp) {
12890 			ftstamp = e.timestamp;
12891 			first = i;
12892 		}
12893 	}
12894 
12895 	if (first == -1)
12896 		return;
12897 
12898 	i = first;
12899 	do {
12900 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
12901 		    sizeof(e), (void *)&e);
12902 		if (rc != 0)
12903 			return;
12904 
12905 		if (e.timestamp == 0)
12906 			return;
12907 
12908 		e.timestamp = be64toh(e.timestamp);
12909 		e.seqno = be32toh(e.seqno);
12910 		for (j = 0; j < 8; j++)
12911 			e.params[j] = be32toh(e.params[j]);
12912 
12913 		db_printf("%10d  %15ju  %8s  %8s  ",
12914 		    e.seqno, e.timestamp,
12915 		    (e.level < nitems(devlog_level_strings) ?
12916 			devlog_level_strings[e.level] : "UNKNOWN"),
12917 		    (e.facility < nitems(devlog_facility_strings) ?
12918 			devlog_facility_strings[e.facility] : "UNKNOWN"));
12919 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
12920 		    e.params[3], e.params[4], e.params[5], e.params[6],
12921 		    e.params[7]);
12922 
12923 		if (++i == nentries)
12924 			i = 0;
12925 	} while (i != first && !db_pager_quit);
12926 }
12927 
12928 static DB_DEFINE_TABLE(show, t4, show_t4);
12929 
12930 DB_TABLE_COMMAND_FLAGS(show_t4, devlog, db_show_devlog, CS_OWN)
12931 {
12932 	device_t dev;
12933 	int t;
12934 	bool valid;
12935 
12936 	valid = false;
12937 	t = db_read_token();
12938 	if (t == tIDENT) {
12939 		dev = device_lookup_by_name(db_tok_string);
12940 		valid = true;
12941 	}
12942 	db_skip_to_eol();
12943 	if (!valid) {
12944 		db_printf("usage: show t4 devlog <nexus>\n");
12945 		return;
12946 	}
12947 
12948 	if (dev == NULL) {
12949 		db_printf("device not found\n");
12950 		return;
12951 	}
12952 
12953 	t4_dump_devlog(device_get_softc(dev));
12954 }
12955 
12956 DB_TABLE_COMMAND_FLAGS(show_t4, tcb, db_show_t4tcb, CS_OWN)
12957 {
12958 	device_t dev;
12959 	int radix, tid, t;
12960 	bool valid;
12961 
12962 	valid = false;
12963 	radix = db_radix;
12964 	db_radix = 10;
12965 	t = db_read_token();
12966 	if (t == tIDENT) {
12967 		dev = device_lookup_by_name(db_tok_string);
12968 		t = db_read_token();
12969 		if (t == tNUMBER) {
12970 			tid = db_tok_number;
12971 			valid = true;
12972 		}
12973 	}
12974 	db_radix = radix;
12975 	db_skip_to_eol();
12976 	if (!valid) {
12977 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
12978 		return;
12979 	}
12980 
12981 	if (dev == NULL) {
12982 		db_printf("device not found\n");
12983 		return;
12984 	}
12985 	if (tid < 0) {
12986 		db_printf("invalid tid\n");
12987 		return;
12988 	}
12989 
12990 	t4_dump_tcb(device_get_softc(dev), tid);
12991 }
12992 #endif
12993 
12994 static eventhandler_tag vxlan_start_evtag;
12995 static eventhandler_tag vxlan_stop_evtag;
12996 
12997 struct vxlan_evargs {
12998 	if_t ifp;
12999 	uint16_t port;
13000 };
13001 
13002 static void
13003 enable_vxlan_rx(struct adapter *sc)
13004 {
13005 	int i, rc;
13006 	struct port_info *pi;
13007 	uint8_t match_all_mac[ETHER_ADDR_LEN] = {0};
13008 
13009 	ASSERT_SYNCHRONIZED_OP(sc);
13010 
13011 	t4_write_reg(sc, A_MPS_RX_VXLAN_TYPE, V_VXLAN(sc->vxlan_port) |
13012 	    F_VXLAN_EN);
13013 	for_each_port(sc, i) {
13014 		pi = sc->port[i];
13015 		if (pi->vxlan_tcam_entry == true)
13016 			continue;
13017 		rc = t4_alloc_raw_mac_filt(sc, pi->vi[0].viid, match_all_mac,
13018 		    match_all_mac, sc->rawf_base + pi->port_id, 1, pi->port_id,
13019 		    true);
13020 		if (rc < 0) {
13021 			rc = -rc;
13022 			CH_ERR(&pi->vi[0],
13023 			    "failed to add VXLAN TCAM entry: %d.\n", rc);
13024 		} else {
13025 			MPASS(rc == sc->rawf_base + pi->port_id);
13026 			pi->vxlan_tcam_entry = true;
13027 		}
13028 	}
13029 }
13030 
13031 static void
13032 t4_vxlan_start(struct adapter *sc, void *arg)
13033 {
13034 	struct vxlan_evargs *v = arg;
13035 
13036 	if (sc->nrawf == 0 || chip_id(sc) <= CHELSIO_T5)
13037 		return;
13038 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4vxst") != 0)
13039 		return;
13040 
13041 	if (sc->vxlan_refcount == 0) {
13042 		sc->vxlan_port = v->port;
13043 		sc->vxlan_refcount = 1;
13044 		if (!hw_off_limits(sc))
13045 			enable_vxlan_rx(sc);
13046 	} else if (sc->vxlan_port == v->port) {
13047 		sc->vxlan_refcount++;
13048 	} else {
13049 		CH_ERR(sc, "VXLAN already configured on port  %d; "
13050 		    "ignoring attempt to configure it on port %d\n",
13051 		    sc->vxlan_port, v->port);
13052 	}
13053 	end_synchronized_op(sc, 0);
13054 }
13055 
13056 static void
13057 t4_vxlan_stop(struct adapter *sc, void *arg)
13058 {
13059 	struct vxlan_evargs *v = arg;
13060 
13061 	if (sc->nrawf == 0 || chip_id(sc) <= CHELSIO_T5)
13062 		return;
13063 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4vxsp") != 0)
13064 		return;
13065 
13066 	/*
13067 	 * VXLANs may have been configured before the driver was loaded so we
13068 	 * may see more stops than starts.  This is not handled cleanly but at
13069 	 * least we keep the refcount sane.
13070 	 */
13071 	if (sc->vxlan_port != v->port)
13072 		goto done;
13073 	if (sc->vxlan_refcount == 0) {
13074 		CH_ERR(sc, "VXLAN operation on port %d was stopped earlier; "
13075 		    "ignoring attempt to stop it again.\n", sc->vxlan_port);
13076 	} else if (--sc->vxlan_refcount == 0 && !hw_off_limits(sc))
13077 		t4_set_reg_field(sc, A_MPS_RX_VXLAN_TYPE, F_VXLAN_EN, 0);
13078 done:
13079 	end_synchronized_op(sc, 0);
13080 }
13081 
13082 static void
13083 t4_vxlan_start_handler(void *arg __unused, if_t ifp,
13084     sa_family_t family, u_int port)
13085 {
13086 	struct vxlan_evargs v;
13087 
13088 	MPASS(family == AF_INET || family == AF_INET6);
13089 	v.ifp = ifp;
13090 	v.port = port;
13091 
13092 	t4_iterate(t4_vxlan_start, &v);
13093 }
13094 
13095 static void
13096 t4_vxlan_stop_handler(void *arg __unused, if_t ifp, sa_family_t family,
13097     u_int port)
13098 {
13099 	struct vxlan_evargs v;
13100 
13101 	MPASS(family == AF_INET || family == AF_INET6);
13102 	v.ifp = ifp;
13103 	v.port = port;
13104 
13105 	t4_iterate(t4_vxlan_stop, &v);
13106 }
13107 
13108 
13109 static struct sx mlu;	/* mod load unload */
13110 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
13111 
13112 static int
13113 mod_event(module_t mod, int cmd, void *arg)
13114 {
13115 	int rc = 0;
13116 	static int loaded = 0;
13117 
13118 	switch (cmd) {
13119 	case MOD_LOAD:
13120 		sx_xlock(&mlu);
13121 		if (loaded++ == 0) {
13122 			t4_sge_modload();
13123 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
13124 			    t4_filter_rpl, CPL_COOKIE_FILTER);
13125 			t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
13126 			    do_l2t_write_rpl, CPL_COOKIE_FILTER);
13127 			t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL,
13128 			    t4_hashfilter_ao_rpl, CPL_COOKIE_HASHFILTER);
13129 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
13130 			    t4_hashfilter_tcb_rpl, CPL_COOKIE_HASHFILTER);
13131 			t4_register_shared_cpl_handler(CPL_ABORT_RPL_RSS,
13132 			    t4_del_hashfilter_rpl, CPL_COOKIE_HASHFILTER);
13133 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
13134 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
13135 			t4_register_cpl_handler(CPL_SMT_WRITE_RPL,
13136 			    do_smt_write_rpl);
13137 			sx_init(&t4_list_lock, "T4/T5 adapters");
13138 			SLIST_INIT(&t4_list);
13139 			callout_init(&fatal_callout, 1);
13140 #ifdef TCP_OFFLOAD
13141 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
13142 			SLIST_INIT(&t4_uld_list);
13143 #endif
13144 #ifdef INET6
13145 			t4_clip_modload();
13146 #endif
13147 #ifdef KERN_TLS
13148 			t6_ktls_modload();
13149 #endif
13150 			t4_tracer_modload();
13151 			tweak_tunables();
13152 			vxlan_start_evtag =
13153 			    EVENTHANDLER_REGISTER(vxlan_start,
13154 				t4_vxlan_start_handler, NULL,
13155 				EVENTHANDLER_PRI_ANY);
13156 			vxlan_stop_evtag =
13157 			    EVENTHANDLER_REGISTER(vxlan_stop,
13158 				t4_vxlan_stop_handler, NULL,
13159 				EVENTHANDLER_PRI_ANY);
13160 			reset_tq = taskqueue_create("t4_rst_tq", M_WAITOK,
13161 			    taskqueue_thread_enqueue, &reset_tq);
13162 			taskqueue_start_threads(&reset_tq, 1, PI_SOFT,
13163 			    "t4_rst_thr");
13164 		}
13165 		sx_xunlock(&mlu);
13166 		break;
13167 
13168 	case MOD_UNLOAD:
13169 		sx_xlock(&mlu);
13170 		if (--loaded == 0) {
13171 			int tries;
13172 
13173 			taskqueue_free(reset_tq);
13174 			sx_slock(&t4_list_lock);
13175 			if (!SLIST_EMPTY(&t4_list)) {
13176 				rc = EBUSY;
13177 				sx_sunlock(&t4_list_lock);
13178 				goto done_unload;
13179 			}
13180 #ifdef TCP_OFFLOAD
13181 			sx_slock(&t4_uld_list_lock);
13182 			if (!SLIST_EMPTY(&t4_uld_list)) {
13183 				rc = EBUSY;
13184 				sx_sunlock(&t4_uld_list_lock);
13185 				sx_sunlock(&t4_list_lock);
13186 				goto done_unload;
13187 			}
13188 #endif
13189 			tries = 0;
13190 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
13191 				uprintf("%ju clusters with custom free routine "
13192 				    "still is use.\n", t4_sge_extfree_refs());
13193 				pause("t4unload", 2 * hz);
13194 			}
13195 #ifdef TCP_OFFLOAD
13196 			sx_sunlock(&t4_uld_list_lock);
13197 #endif
13198 			sx_sunlock(&t4_list_lock);
13199 
13200 			if (t4_sge_extfree_refs() == 0) {
13201 				EVENTHANDLER_DEREGISTER(vxlan_start,
13202 				    vxlan_start_evtag);
13203 				EVENTHANDLER_DEREGISTER(vxlan_stop,
13204 				    vxlan_stop_evtag);
13205 				t4_tracer_modunload();
13206 #ifdef KERN_TLS
13207 				t6_ktls_modunload();
13208 #endif
13209 #ifdef INET6
13210 				t4_clip_modunload();
13211 #endif
13212 #ifdef TCP_OFFLOAD
13213 				sx_destroy(&t4_uld_list_lock);
13214 #endif
13215 				sx_destroy(&t4_list_lock);
13216 				t4_sge_modunload();
13217 				loaded = 0;
13218 			} else {
13219 				rc = EBUSY;
13220 				loaded++;	/* undo earlier decrement */
13221 			}
13222 		}
13223 done_unload:
13224 		sx_xunlock(&mlu);
13225 		break;
13226 	}
13227 
13228 	return (rc);
13229 }
13230 
13231 DRIVER_MODULE(t4nex, pci, t4_driver, mod_event, 0);
13232 MODULE_VERSION(t4nex, 1);
13233 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
13234 #ifdef DEV_NETMAP
13235 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
13236 #endif /* DEV_NETMAP */
13237 
13238 DRIVER_MODULE(t5nex, pci, t5_driver, mod_event, 0);
13239 MODULE_VERSION(t5nex, 1);
13240 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
13241 #ifdef DEV_NETMAP
13242 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
13243 #endif /* DEV_NETMAP */
13244 
13245 DRIVER_MODULE(t6nex, pci, t6_driver, mod_event, 0);
13246 MODULE_VERSION(t6nex, 1);
13247 MODULE_DEPEND(t6nex, crypto, 1, 1, 1);
13248 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
13249 #ifdef DEV_NETMAP
13250 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
13251 #endif /* DEV_NETMAP */
13252 
13253 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, 0, 0);
13254 MODULE_VERSION(cxgbe, 1);
13255 
13256 DRIVER_MODULE(cxl, t5nex, cxl_driver, 0, 0);
13257 MODULE_VERSION(cxl, 1);
13258 
13259 DRIVER_MODULE(cc, t6nex, cc_driver, 0, 0);
13260 MODULE_VERSION(cc, 1);
13261 
13262 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, 0, 0);
13263 MODULE_VERSION(vcxgbe, 1);
13264 
13265 DRIVER_MODULE(vcxl, cxl, vcxl_driver, 0, 0);
13266 MODULE_VERSION(vcxl, 1);
13267 
13268 DRIVER_MODULE(vcc, cc, vcc_driver, 0, 0);
13269 MODULE_VERSION(vcc, 1);
13270