xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision d65cd7a57bf0600b722afc770838a5d0c1c3a8e1)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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 __FBSDID("$FreeBSD$");
32 
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_kern_tls.h"
37 #include "opt_ratelimit.h"
38 #include "opt_rss.h"
39 
40 #include <sys/param.h>
41 #include <sys/conf.h>
42 #include <sys/priv.h>
43 #include <sys/kernel.h>
44 #include <sys/bus.h>
45 #include <sys/module.h>
46 #include <sys/malloc.h>
47 #include <sys/queue.h>
48 #include <sys/taskqueue.h>
49 #include <sys/pciio.h>
50 #include <dev/pci/pcireg.h>
51 #include <dev/pci/pcivar.h>
52 #include <dev/pci/pci_private.h>
53 #include <sys/firmware.h>
54 #include <sys/sbuf.h>
55 #include <sys/smp.h>
56 #include <sys/socket.h>
57 #include <sys/sockio.h>
58 #include <sys/sysctl.h>
59 #include <net/ethernet.h>
60 #include <net/if.h>
61 #include <net/if_types.h>
62 #include <net/if_dl.h>
63 #include <net/if_vlan_var.h>
64 #ifdef RSS
65 #include <net/rss_config.h>
66 #endif
67 #include <netinet/in.h>
68 #include <netinet/ip.h>
69 #ifdef KERN_TLS
70 #include <netinet/tcp_seq.h>
71 #endif
72 #if defined(__i386__) || defined(__amd64__)
73 #include <machine/md_var.h>
74 #include <machine/cputypes.h>
75 #include <vm/vm.h>
76 #include <vm/pmap.h>
77 #endif
78 #ifdef DDB
79 #include <ddb/ddb.h>
80 #include <ddb/db_lex.h>
81 #endif
82 
83 #include "common/common.h"
84 #include "common/t4_msg.h"
85 #include "common/t4_regs.h"
86 #include "common/t4_regs_values.h"
87 #include "cudbg/cudbg.h"
88 #include "t4_clip.h"
89 #include "t4_ioctl.h"
90 #include "t4_l2t.h"
91 #include "t4_mp_ring.h"
92 #include "t4_if.h"
93 #include "t4_smt.h"
94 
95 /* T4 bus driver interface */
96 static int t4_probe(device_t);
97 static int t4_attach(device_t);
98 static int t4_detach(device_t);
99 static int t4_child_location_str(device_t, device_t, char *, size_t);
100 static int t4_ready(device_t);
101 static int t4_read_port_device(device_t, int, device_t *);
102 static device_method_t t4_methods[] = {
103 	DEVMETHOD(device_probe,		t4_probe),
104 	DEVMETHOD(device_attach,	t4_attach),
105 	DEVMETHOD(device_detach,	t4_detach),
106 
107 	DEVMETHOD(bus_child_location_str, t4_child_location_str),
108 
109 	DEVMETHOD(t4_is_main_ready,	t4_ready),
110 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
111 
112 	DEVMETHOD_END
113 };
114 static driver_t t4_driver = {
115 	"t4nex",
116 	t4_methods,
117 	sizeof(struct adapter)
118 };
119 
120 
121 /* T4 port (cxgbe) interface */
122 static int cxgbe_probe(device_t);
123 static int cxgbe_attach(device_t);
124 static int cxgbe_detach(device_t);
125 device_method_t cxgbe_methods[] = {
126 	DEVMETHOD(device_probe,		cxgbe_probe),
127 	DEVMETHOD(device_attach,	cxgbe_attach),
128 	DEVMETHOD(device_detach,	cxgbe_detach),
129 	{ 0, 0 }
130 };
131 static driver_t cxgbe_driver = {
132 	"cxgbe",
133 	cxgbe_methods,
134 	sizeof(struct port_info)
135 };
136 
137 /* T4 VI (vcxgbe) interface */
138 static int vcxgbe_probe(device_t);
139 static int vcxgbe_attach(device_t);
140 static int vcxgbe_detach(device_t);
141 static device_method_t vcxgbe_methods[] = {
142 	DEVMETHOD(device_probe,		vcxgbe_probe),
143 	DEVMETHOD(device_attach,	vcxgbe_attach),
144 	DEVMETHOD(device_detach,	vcxgbe_detach),
145 	{ 0, 0 }
146 };
147 static driver_t vcxgbe_driver = {
148 	"vcxgbe",
149 	vcxgbe_methods,
150 	sizeof(struct vi_info)
151 };
152 
153 static d_ioctl_t t4_ioctl;
154 
155 static struct cdevsw t4_cdevsw = {
156        .d_version = D_VERSION,
157        .d_ioctl = t4_ioctl,
158        .d_name = "t4nex",
159 };
160 
161 /* T5 bus driver interface */
162 static int t5_probe(device_t);
163 static device_method_t t5_methods[] = {
164 	DEVMETHOD(device_probe,		t5_probe),
165 	DEVMETHOD(device_attach,	t4_attach),
166 	DEVMETHOD(device_detach,	t4_detach),
167 
168 	DEVMETHOD(bus_child_location_str, t4_child_location_str),
169 
170 	DEVMETHOD(t4_is_main_ready,	t4_ready),
171 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
172 
173 	DEVMETHOD_END
174 };
175 static driver_t t5_driver = {
176 	"t5nex",
177 	t5_methods,
178 	sizeof(struct adapter)
179 };
180 
181 
182 /* T5 port (cxl) interface */
183 static driver_t cxl_driver = {
184 	"cxl",
185 	cxgbe_methods,
186 	sizeof(struct port_info)
187 };
188 
189 /* T5 VI (vcxl) interface */
190 static driver_t vcxl_driver = {
191 	"vcxl",
192 	vcxgbe_methods,
193 	sizeof(struct vi_info)
194 };
195 
196 /* T6 bus driver interface */
197 static int t6_probe(device_t);
198 static device_method_t t6_methods[] = {
199 	DEVMETHOD(device_probe,		t6_probe),
200 	DEVMETHOD(device_attach,	t4_attach),
201 	DEVMETHOD(device_detach,	t4_detach),
202 
203 	DEVMETHOD(bus_child_location_str, t4_child_location_str),
204 
205 	DEVMETHOD(t4_is_main_ready,	t4_ready),
206 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
207 
208 	DEVMETHOD_END
209 };
210 static driver_t t6_driver = {
211 	"t6nex",
212 	t6_methods,
213 	sizeof(struct adapter)
214 };
215 
216 
217 /* T6 port (cc) interface */
218 static driver_t cc_driver = {
219 	"cc",
220 	cxgbe_methods,
221 	sizeof(struct port_info)
222 };
223 
224 /* T6 VI (vcc) interface */
225 static driver_t vcc_driver = {
226 	"vcc",
227 	vcxgbe_methods,
228 	sizeof(struct vi_info)
229 };
230 
231 /* ifnet interface */
232 static void cxgbe_init(void *);
233 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
234 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
235 static void cxgbe_qflush(struct ifnet *);
236 #if defined(KERN_TLS) || defined(RATELIMIT)
237 static int cxgbe_snd_tag_alloc(struct ifnet *, union if_snd_tag_alloc_params *,
238     struct m_snd_tag **);
239 static int cxgbe_snd_tag_modify(struct m_snd_tag *,
240     union if_snd_tag_modify_params *);
241 static int cxgbe_snd_tag_query(struct m_snd_tag *,
242     union if_snd_tag_query_params *);
243 static void cxgbe_snd_tag_free(struct m_snd_tag *);
244 #endif
245 
246 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
247 
248 /*
249  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
250  * then ADAPTER_LOCK, then t4_uld_list_lock.
251  */
252 static struct sx t4_list_lock;
253 SLIST_HEAD(, adapter) t4_list;
254 #ifdef TCP_OFFLOAD
255 static struct sx t4_uld_list_lock;
256 SLIST_HEAD(, uld_info) t4_uld_list;
257 #endif
258 
259 /*
260  * Tunables.  See tweak_tunables() too.
261  *
262  * Each tunable is set to a default value here if it's known at compile-time.
263  * Otherwise it is set to -n as an indication to tweak_tunables() that it should
264  * provide a reasonable default (upto n) when the driver is loaded.
265  *
266  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
267  * T5 are under hw.cxl.
268  */
269 SYSCTL_NODE(_hw, OID_AUTO, cxgbe, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
270     "cxgbe(4) parameters");
271 SYSCTL_NODE(_hw, OID_AUTO, cxl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
272     "cxgbe(4) T5+ parameters");
273 SYSCTL_NODE(_hw_cxgbe, OID_AUTO, toe, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
274     "cxgbe(4) TOE parameters");
275 
276 /*
277  * Number of queues for tx and rx, NIC and offload.
278  */
279 #define NTXQ 16
280 int t4_ntxq = -NTXQ;
281 SYSCTL_INT(_hw_cxgbe, OID_AUTO, ntxq, CTLFLAG_RDTUN, &t4_ntxq, 0,
282     "Number of TX queues per port");
283 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq);	/* Old name, undocumented */
284 
285 #define NRXQ 8
286 int t4_nrxq = -NRXQ;
287 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nrxq, CTLFLAG_RDTUN, &t4_nrxq, 0,
288     "Number of RX queues per port");
289 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq);	/* Old name, undocumented */
290 
291 #define NTXQ_VI 1
292 static int t4_ntxq_vi = -NTXQ_VI;
293 SYSCTL_INT(_hw_cxgbe, OID_AUTO, ntxq_vi, CTLFLAG_RDTUN, &t4_ntxq_vi, 0,
294     "Number of TX queues per VI");
295 
296 #define NRXQ_VI 1
297 static int t4_nrxq_vi = -NRXQ_VI;
298 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nrxq_vi, CTLFLAG_RDTUN, &t4_nrxq_vi, 0,
299     "Number of RX queues per VI");
300 
301 static int t4_rsrv_noflowq = 0;
302 SYSCTL_INT(_hw_cxgbe, OID_AUTO, rsrv_noflowq, CTLFLAG_RDTUN, &t4_rsrv_noflowq,
303     0, "Reserve TX queue 0 of each VI for non-flowid packets");
304 
305 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
306 #define NOFLDTXQ 8
307 static int t4_nofldtxq = -NOFLDTXQ;
308 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldtxq, CTLFLAG_RDTUN, &t4_nofldtxq, 0,
309     "Number of offload TX queues per port");
310 
311 #define NOFLDRXQ 2
312 static int t4_nofldrxq = -NOFLDRXQ;
313 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldrxq, CTLFLAG_RDTUN, &t4_nofldrxq, 0,
314     "Number of offload RX queues per port");
315 
316 #define NOFLDTXQ_VI 1
317 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
318 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldtxq_vi, CTLFLAG_RDTUN, &t4_nofldtxq_vi, 0,
319     "Number of offload TX queues per VI");
320 
321 #define NOFLDRXQ_VI 1
322 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
323 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldrxq_vi, CTLFLAG_RDTUN, &t4_nofldrxq_vi, 0,
324     "Number of offload RX queues per VI");
325 
326 #define TMR_IDX_OFLD 1
327 int t4_tmr_idx_ofld = TMR_IDX_OFLD;
328 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_timer_idx_ofld, CTLFLAG_RDTUN,
329     &t4_tmr_idx_ofld, 0, "Holdoff timer index for offload queues");
330 
331 #define PKTC_IDX_OFLD (-1)
332 int t4_pktc_idx_ofld = PKTC_IDX_OFLD;
333 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_pktc_idx_ofld, CTLFLAG_RDTUN,
334     &t4_pktc_idx_ofld, 0, "holdoff packet counter index for offload queues");
335 
336 /* 0 means chip/fw default, non-zero number is value in microseconds */
337 static u_long t4_toe_keepalive_idle = 0;
338 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, keepalive_idle, CTLFLAG_RDTUN,
339     &t4_toe_keepalive_idle, 0, "TOE keepalive idle timer (us)");
340 
341 /* 0 means chip/fw default, non-zero number is value in microseconds */
342 static u_long t4_toe_keepalive_interval = 0;
343 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, keepalive_interval, CTLFLAG_RDTUN,
344     &t4_toe_keepalive_interval, 0, "TOE keepalive interval timer (us)");
345 
346 /* 0 means chip/fw default, non-zero number is # of keepalives before abort */
347 static int t4_toe_keepalive_count = 0;
348 SYSCTL_INT(_hw_cxgbe_toe, OID_AUTO, keepalive_count, CTLFLAG_RDTUN,
349     &t4_toe_keepalive_count, 0, "Number of TOE keepalive probes before abort");
350 
351 /* 0 means chip/fw default, non-zero number is value in microseconds */
352 static u_long t4_toe_rexmt_min = 0;
353 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, rexmt_min, CTLFLAG_RDTUN,
354     &t4_toe_rexmt_min, 0, "Minimum TOE retransmit interval (us)");
355 
356 /* 0 means chip/fw default, non-zero number is value in microseconds */
357 static u_long t4_toe_rexmt_max = 0;
358 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, rexmt_max, CTLFLAG_RDTUN,
359     &t4_toe_rexmt_max, 0, "Maximum TOE retransmit interval (us)");
360 
361 /* 0 means chip/fw default, non-zero number is # of rexmt before abort */
362 static int t4_toe_rexmt_count = 0;
363 SYSCTL_INT(_hw_cxgbe_toe, OID_AUTO, rexmt_count, CTLFLAG_RDTUN,
364     &t4_toe_rexmt_count, 0, "Number of TOE retransmissions before abort");
365 
366 /* -1 means chip/fw default, other values are raw backoff values to use */
367 static int t4_toe_rexmt_backoff[16] = {
368 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
369 };
370 SYSCTL_NODE(_hw_cxgbe_toe, OID_AUTO, rexmt_backoff,
371     CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
372     "cxgbe(4) TOE retransmit backoff values");
373 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 0, CTLFLAG_RDTUN,
374     &t4_toe_rexmt_backoff[0], 0, "");
375 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 1, CTLFLAG_RDTUN,
376     &t4_toe_rexmt_backoff[1], 0, "");
377 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 2, CTLFLAG_RDTUN,
378     &t4_toe_rexmt_backoff[2], 0, "");
379 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 3, CTLFLAG_RDTUN,
380     &t4_toe_rexmt_backoff[3], 0, "");
381 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 4, CTLFLAG_RDTUN,
382     &t4_toe_rexmt_backoff[4], 0, "");
383 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 5, CTLFLAG_RDTUN,
384     &t4_toe_rexmt_backoff[5], 0, "");
385 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 6, CTLFLAG_RDTUN,
386     &t4_toe_rexmt_backoff[6], 0, "");
387 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 7, CTLFLAG_RDTUN,
388     &t4_toe_rexmt_backoff[7], 0, "");
389 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 8, CTLFLAG_RDTUN,
390     &t4_toe_rexmt_backoff[8], 0, "");
391 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 9, CTLFLAG_RDTUN,
392     &t4_toe_rexmt_backoff[9], 0, "");
393 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 10, CTLFLAG_RDTUN,
394     &t4_toe_rexmt_backoff[10], 0, "");
395 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 11, CTLFLAG_RDTUN,
396     &t4_toe_rexmt_backoff[11], 0, "");
397 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 12, CTLFLAG_RDTUN,
398     &t4_toe_rexmt_backoff[12], 0, "");
399 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 13, CTLFLAG_RDTUN,
400     &t4_toe_rexmt_backoff[13], 0, "");
401 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 14, CTLFLAG_RDTUN,
402     &t4_toe_rexmt_backoff[14], 0, "");
403 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 15, CTLFLAG_RDTUN,
404     &t4_toe_rexmt_backoff[15], 0, "");
405 #endif
406 
407 #ifdef DEV_NETMAP
408 #define NN_MAIN_VI	(1 << 0)	/* Native netmap on the main VI */
409 #define NN_EXTRA_VI	(1 << 1)	/* Native netmap on the extra VI(s) */
410 static int t4_native_netmap = NN_EXTRA_VI;
411 SYSCTL_INT(_hw_cxgbe, OID_AUTO, native_netmap, CTLFLAG_RDTUN, &t4_native_netmap,
412     0, "Native netmap support.  bit 0 = main VI, bit 1 = extra VIs");
413 
414 #define NNMTXQ 8
415 static int t4_nnmtxq = -NNMTXQ;
416 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmtxq, CTLFLAG_RDTUN, &t4_nnmtxq, 0,
417     "Number of netmap TX queues");
418 
419 #define NNMRXQ 8
420 static int t4_nnmrxq = -NNMRXQ;
421 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmrxq, CTLFLAG_RDTUN, &t4_nnmrxq, 0,
422     "Number of netmap RX queues");
423 
424 #define NNMTXQ_VI 2
425 static int t4_nnmtxq_vi = -NNMTXQ_VI;
426 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmtxq_vi, CTLFLAG_RDTUN, &t4_nnmtxq_vi, 0,
427     "Number of netmap TX queues per VI");
428 
429 #define NNMRXQ_VI 2
430 static int t4_nnmrxq_vi = -NNMRXQ_VI;
431 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmrxq_vi, CTLFLAG_RDTUN, &t4_nnmrxq_vi, 0,
432     "Number of netmap RX queues per VI");
433 #endif
434 
435 /*
436  * Holdoff parameters for ports.
437  */
438 #define TMR_IDX 1
439 int t4_tmr_idx = TMR_IDX;
440 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_timer_idx, CTLFLAG_RDTUN, &t4_tmr_idx,
441     0, "Holdoff timer index");
442 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx);	/* Old name */
443 
444 #define PKTC_IDX (-1)
445 int t4_pktc_idx = PKTC_IDX;
446 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_pktc_idx, CTLFLAG_RDTUN, &t4_pktc_idx,
447     0, "Holdoff packet counter index");
448 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx);	/* Old name */
449 
450 /*
451  * Size (# of entries) of each tx and rx queue.
452  */
453 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
454 SYSCTL_INT(_hw_cxgbe, OID_AUTO, qsize_txq, CTLFLAG_RDTUN, &t4_qsize_txq, 0,
455     "Number of descriptors in each TX queue");
456 
457 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
458 SYSCTL_INT(_hw_cxgbe, OID_AUTO, qsize_rxq, CTLFLAG_RDTUN, &t4_qsize_rxq, 0,
459     "Number of descriptors in each RX queue");
460 
461 /*
462  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
463  */
464 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
465 SYSCTL_INT(_hw_cxgbe, OID_AUTO, interrupt_types, CTLFLAG_RDTUN, &t4_intr_types,
466     0, "Interrupt types allowed (bit 0 = INTx, 1 = MSI, 2 = MSI-X)");
467 
468 /*
469  * Configuration file.  All the _CF names here are special.
470  */
471 #define DEFAULT_CF	"default"
472 #define BUILTIN_CF	"built-in"
473 #define FLASH_CF	"flash"
474 #define UWIRE_CF	"uwire"
475 #define FPGA_CF		"fpga"
476 static char t4_cfg_file[32] = DEFAULT_CF;
477 SYSCTL_STRING(_hw_cxgbe, OID_AUTO, config_file, CTLFLAG_RDTUN, t4_cfg_file,
478     sizeof(t4_cfg_file), "Firmware configuration file");
479 
480 /*
481  * PAUSE settings (bit 0, 1, 2 = rx_pause, tx_pause, pause_autoneg respectively).
482  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
483  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
484  *            mark or when signalled to do so, 0 to never emit PAUSE.
485  * pause_autoneg = 1 means PAUSE will be negotiated if possible and the
486  *                 negotiated settings will override rx_pause/tx_pause.
487  *                 Otherwise rx_pause/tx_pause are applied forcibly.
488  */
489 static int t4_pause_settings = PAUSE_RX | PAUSE_TX | PAUSE_AUTONEG;
490 SYSCTL_INT(_hw_cxgbe, OID_AUTO, pause_settings, CTLFLAG_RDTUN,
491     &t4_pause_settings, 0,
492     "PAUSE settings (bit 0 = rx_pause, 1 = tx_pause, 2 = pause_autoneg)");
493 
494 /*
495  * Forward Error Correction settings (bit 0, 1 = RS, BASER respectively).
496  * -1 to run with the firmware default.  Same as FEC_AUTO (bit 5)
497  *  0 to disable FEC.
498  */
499 static int t4_fec = -1;
500 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fec, CTLFLAG_RDTUN, &t4_fec, 0,
501     "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
502 
503 /*
504  * Link autonegotiation.
505  * -1 to run with the firmware default.
506  *  0 to disable.
507  *  1 to enable.
508  */
509 static int t4_autoneg = -1;
510 SYSCTL_INT(_hw_cxgbe, OID_AUTO, autoneg, CTLFLAG_RDTUN, &t4_autoneg, 0,
511     "Link autonegotiation");
512 
513 /*
514  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
515  * encouraged respectively).  '-n' is the same as 'n' except the firmware
516  * version used in the checks is read from the firmware bundled with the driver.
517  */
518 static int t4_fw_install = 1;
519 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fw_install, CTLFLAG_RDTUN, &t4_fw_install, 0,
520     "Firmware auto-install (0 = prohibited, 1 = allowed, 2 = encouraged)");
521 
522 /*
523  * ASIC features that will be used.  Disable the ones you don't want so that the
524  * chip resources aren't wasted on features that will not be used.
525  */
526 static int t4_nbmcaps_allowed = 0;
527 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nbmcaps_allowed, CTLFLAG_RDTUN,
528     &t4_nbmcaps_allowed, 0, "Default NBM capabilities");
529 
530 static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
531 SYSCTL_INT(_hw_cxgbe, OID_AUTO, linkcaps_allowed, CTLFLAG_RDTUN,
532     &t4_linkcaps_allowed, 0, "Default link capabilities");
533 
534 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
535     FW_CAPS_CONFIG_SWITCH_EGRESS;
536 SYSCTL_INT(_hw_cxgbe, OID_AUTO, switchcaps_allowed, CTLFLAG_RDTUN,
537     &t4_switchcaps_allowed, 0, "Default switch capabilities");
538 
539 #ifdef RATELIMIT
540 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
541 	FW_CAPS_CONFIG_NIC_HASHFILTER | FW_CAPS_CONFIG_NIC_ETHOFLD;
542 #else
543 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
544 	FW_CAPS_CONFIG_NIC_HASHFILTER;
545 #endif
546 SYSCTL_INT(_hw_cxgbe, OID_AUTO, niccaps_allowed, CTLFLAG_RDTUN,
547     &t4_niccaps_allowed, 0, "Default NIC capabilities");
548 
549 static int t4_toecaps_allowed = -1;
550 SYSCTL_INT(_hw_cxgbe, OID_AUTO, toecaps_allowed, CTLFLAG_RDTUN,
551     &t4_toecaps_allowed, 0, "Default TCP offload capabilities");
552 
553 static int t4_rdmacaps_allowed = -1;
554 SYSCTL_INT(_hw_cxgbe, OID_AUTO, rdmacaps_allowed, CTLFLAG_RDTUN,
555     &t4_rdmacaps_allowed, 0, "Default RDMA capabilities");
556 
557 static int t4_cryptocaps_allowed = -1;
558 SYSCTL_INT(_hw_cxgbe, OID_AUTO, cryptocaps_allowed, CTLFLAG_RDTUN,
559     &t4_cryptocaps_allowed, 0, "Default crypto capabilities");
560 
561 static int t4_iscsicaps_allowed = -1;
562 SYSCTL_INT(_hw_cxgbe, OID_AUTO, iscsicaps_allowed, CTLFLAG_RDTUN,
563     &t4_iscsicaps_allowed, 0, "Default iSCSI capabilities");
564 
565 static int t4_fcoecaps_allowed = 0;
566 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fcoecaps_allowed, CTLFLAG_RDTUN,
567     &t4_fcoecaps_allowed, 0, "Default FCoE capabilities");
568 
569 static int t5_write_combine = 0;
570 SYSCTL_INT(_hw_cxl, OID_AUTO, write_combine, CTLFLAG_RDTUN, &t5_write_combine,
571     0, "Use WC instead of UC for BAR2");
572 
573 static int t4_num_vis = 1;
574 SYSCTL_INT(_hw_cxgbe, OID_AUTO, num_vis, CTLFLAG_RDTUN, &t4_num_vis, 0,
575     "Number of VIs per port");
576 
577 /*
578  * PCIe Relaxed Ordering.
579  * -1: driver should figure out a good value.
580  * 0: disable RO.
581  * 1: enable RO.
582  * 2: leave RO alone.
583  */
584 static int pcie_relaxed_ordering = -1;
585 SYSCTL_INT(_hw_cxgbe, OID_AUTO, pcie_relaxed_ordering, CTLFLAG_RDTUN,
586     &pcie_relaxed_ordering, 0,
587     "PCIe Relaxed Ordering: 0 = disable, 1 = enable, 2 = leave alone");
588 
589 static int t4_panic_on_fatal_err = 0;
590 SYSCTL_INT(_hw_cxgbe, OID_AUTO, panic_on_fatal_err, CTLFLAG_RDTUN,
591     &t4_panic_on_fatal_err, 0, "panic on fatal errors");
592 
593 #ifdef TCP_OFFLOAD
594 /*
595  * TOE tunables.
596  */
597 static int t4_cop_managed_offloading = 0;
598 SYSCTL_INT(_hw_cxgbe, OID_AUTO, cop_managed_offloading, CTLFLAG_RDTUN,
599     &t4_cop_managed_offloading, 0,
600     "COP (Connection Offload Policy) controls all TOE offload");
601 #endif
602 
603 #ifdef KERN_TLS
604 /*
605  * This enables KERN_TLS for all adapters if set.
606  */
607 static int t4_kern_tls = 0;
608 SYSCTL_INT(_hw_cxgbe, OID_AUTO, kern_tls, CTLFLAG_RDTUN, &t4_kern_tls, 0,
609     "Enable KERN_TLS mode for all supported adapters");
610 
611 SYSCTL_NODE(_hw_cxgbe, OID_AUTO, tls, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
612     "cxgbe(4) KERN_TLS parameters");
613 
614 static int t4_tls_inline_keys = 0;
615 SYSCTL_INT(_hw_cxgbe_tls, OID_AUTO, inline_keys, CTLFLAG_RDTUN,
616     &t4_tls_inline_keys, 0,
617     "Always pass TLS keys in work requests (1) or attempt to store TLS keys "
618     "in card memory.");
619 
620 static int t4_tls_combo_wrs = 0;
621 SYSCTL_INT(_hw_cxgbe_tls, OID_AUTO, combo_wrs, CTLFLAG_RDTUN, &t4_tls_combo_wrs,
622     0, "Attempt to combine TCB field updates with TLS record work requests.");
623 #endif
624 
625 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
626 static int vi_mac_funcs[] = {
627 	FW_VI_FUNC_ETH,
628 	FW_VI_FUNC_OFLD,
629 	FW_VI_FUNC_IWARP,
630 	FW_VI_FUNC_OPENISCSI,
631 	FW_VI_FUNC_OPENFCOE,
632 	FW_VI_FUNC_FOISCSI,
633 	FW_VI_FUNC_FOFCOE,
634 };
635 
636 struct intrs_and_queues {
637 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
638 	uint16_t num_vis;	/* number of VIs for each port */
639 	uint16_t nirq;		/* Total # of vectors */
640 	uint16_t ntxq;		/* # of NIC txq's for each port */
641 	uint16_t nrxq;		/* # of NIC rxq's for each port */
642 	uint16_t nofldtxq;	/* # of TOE/ETHOFLD txq's for each port */
643 	uint16_t nofldrxq;	/* # of TOE rxq's for each port */
644 	uint16_t nnmtxq;	/* # of netmap txq's */
645 	uint16_t nnmrxq;	/* # of netmap rxq's */
646 
647 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
648 	uint16_t ntxq_vi;	/* # of NIC txq's */
649 	uint16_t nrxq_vi;	/* # of NIC rxq's */
650 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
651 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
652 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
653 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
654 };
655 
656 static void setup_memwin(struct adapter *);
657 static void position_memwin(struct adapter *, int, uint32_t);
658 static int validate_mem_range(struct adapter *, uint32_t, uint32_t);
659 static int fwmtype_to_hwmtype(int);
660 static int validate_mt_off_len(struct adapter *, int, uint32_t, uint32_t,
661     uint32_t *);
662 static int fixup_devlog_params(struct adapter *);
663 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
664 static int contact_firmware(struct adapter *);
665 static int partition_resources(struct adapter *);
666 static int get_params__pre_init(struct adapter *);
667 static int set_params__pre_init(struct adapter *);
668 static int get_params__post_init(struct adapter *);
669 static int set_params__post_init(struct adapter *);
670 static void t4_set_desc(struct adapter *);
671 static bool fixed_ifmedia(struct port_info *);
672 static void build_medialist(struct port_info *);
673 static void init_link_config(struct port_info *);
674 static int fixup_link_config(struct port_info *);
675 static int apply_link_config(struct port_info *);
676 static int cxgbe_init_synchronized(struct vi_info *);
677 static int cxgbe_uninit_synchronized(struct vi_info *);
678 static void quiesce_txq(struct adapter *, struct sge_txq *);
679 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
680 static void quiesce_iq(struct adapter *, struct sge_iq *);
681 static void quiesce_fl(struct adapter *, struct sge_fl *);
682 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
683     driver_intr_t *, void *, char *);
684 static int t4_free_irq(struct adapter *, struct irq *);
685 static void t4_init_atid_table(struct adapter *);
686 static void t4_free_atid_table(struct adapter *);
687 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
688 static void vi_refresh_stats(struct adapter *, struct vi_info *);
689 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
690 static void cxgbe_tick(void *);
691 static void cxgbe_sysctls(struct port_info *);
692 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
693 static int sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS);
694 static int sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS);
695 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
696 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
697 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
698 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
699 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
700 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
701 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
702 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
703 static int sysctl_module_fec(SYSCTL_HANDLER_ARGS);
704 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
705 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
706 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
707 static int sysctl_vdd(SYSCTL_HANDLER_ARGS);
708 static int sysctl_reset_sensor(SYSCTL_HANDLER_ARGS);
709 static int sysctl_loadavg(SYSCTL_HANDLER_ARGS);
710 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
711 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
712 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
713 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
714 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
715 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
716 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
717 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
718 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
719 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
720 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
721 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
722 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
723 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
724 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
725 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
726 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
727 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
728 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
729 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
730 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
731 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
732 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
733 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
734 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
735 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
736 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
737 static int sysctl_cpus(SYSCTL_HANDLER_ARGS);
738 #ifdef TCP_OFFLOAD
739 static int sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS);
740 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
741 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
742 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
743 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
744 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
745 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
746 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
747 #endif
748 static int get_sge_context(struct adapter *, struct t4_sge_context *);
749 static int load_fw(struct adapter *, struct t4_data *);
750 static int load_cfg(struct adapter *, struct t4_data *);
751 static int load_boot(struct adapter *, struct t4_bootrom *);
752 static int load_bootcfg(struct adapter *, struct t4_data *);
753 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
754 static void free_offload_policy(struct t4_offload_policy *);
755 static int set_offload_policy(struct adapter *, struct t4_offload_policy *);
756 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
757 static int read_i2c(struct adapter *, struct t4_i2c_data *);
758 static int clear_stats(struct adapter *, u_int);
759 #ifdef TCP_OFFLOAD
760 static int toe_capability(struct vi_info *, int);
761 static void t4_async_event(void *, int);
762 #endif
763 static int mod_event(module_t, int, void *);
764 static int notify_siblings(device_t, int);
765 
766 struct {
767 	uint16_t device;
768 	char *desc;
769 } t4_pciids[] = {
770 	{0xa000, "Chelsio Terminator 4 FPGA"},
771 	{0x4400, "Chelsio T440-dbg"},
772 	{0x4401, "Chelsio T420-CR"},
773 	{0x4402, "Chelsio T422-CR"},
774 	{0x4403, "Chelsio T440-CR"},
775 	{0x4404, "Chelsio T420-BCH"},
776 	{0x4405, "Chelsio T440-BCH"},
777 	{0x4406, "Chelsio T440-CH"},
778 	{0x4407, "Chelsio T420-SO"},
779 	{0x4408, "Chelsio T420-CX"},
780 	{0x4409, "Chelsio T420-BT"},
781 	{0x440a, "Chelsio T404-BT"},
782 	{0x440e, "Chelsio T440-LP-CR"},
783 }, t5_pciids[] = {
784 	{0xb000, "Chelsio Terminator 5 FPGA"},
785 	{0x5400, "Chelsio T580-dbg"},
786 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
787 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
788 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
789 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
790 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
791 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
792 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
793 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
794 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
795 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
796 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
797 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
798 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
799 	{0x5418,  "Chelsio T540-BT"},		/* 4 x 10GBaseT */
800 	{0x5419,  "Chelsio T540-LP-BT"},	/* 4 x 10GBaseT */
801 	{0x541a,  "Chelsio T540-SO-BT"},	/* 4 x 10GBaseT, nomem */
802 	{0x541b,  "Chelsio T540-SO-CR"},	/* 4 x 10G, nomem */
803 
804 	/* Custom */
805 	{0x5483, "Custom T540-CR"},
806 	{0x5484, "Custom T540-BT"},
807 }, t6_pciids[] = {
808 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
809 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
810 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
811 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
812 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
813 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
814 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
815 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
816 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
817 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
818 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
819 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
820 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
821 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
822 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
823 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
824 
825 	/* Custom */
826 	{0x6480, "Custom T6225-CR"},
827 	{0x6481, "Custom T62100-CR"},
828 	{0x6482, "Custom T6225-CR"},
829 	{0x6483, "Custom T62100-CR"},
830 	{0x6484, "Custom T64100-CR"},
831 	{0x6485, "Custom T6240-SO"},
832 	{0x6486, "Custom T6225-SO-CR"},
833 	{0x6487, "Custom T6225-CR"},
834 };
835 
836 #ifdef TCP_OFFLOAD
837 /*
838  * service_iq_fl() has an iq and needs the fl.  Offset of fl from the iq should
839  * be exactly the same for both rxq and ofld_rxq.
840  */
841 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
842 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
843 #endif
844 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
845 
846 static int
847 t4_probe(device_t dev)
848 {
849 	int i;
850 	uint16_t v = pci_get_vendor(dev);
851 	uint16_t d = pci_get_device(dev);
852 	uint8_t f = pci_get_function(dev);
853 
854 	if (v != PCI_VENDOR_ID_CHELSIO)
855 		return (ENXIO);
856 
857 	/* Attach only to PF0 of the FPGA */
858 	if (d == 0xa000 && f != 0)
859 		return (ENXIO);
860 
861 	for (i = 0; i < nitems(t4_pciids); i++) {
862 		if (d == t4_pciids[i].device) {
863 			device_set_desc(dev, t4_pciids[i].desc);
864 			return (BUS_PROBE_DEFAULT);
865 		}
866 	}
867 
868 	return (ENXIO);
869 }
870 
871 static int
872 t5_probe(device_t dev)
873 {
874 	int i;
875 	uint16_t v = pci_get_vendor(dev);
876 	uint16_t d = pci_get_device(dev);
877 	uint8_t f = pci_get_function(dev);
878 
879 	if (v != PCI_VENDOR_ID_CHELSIO)
880 		return (ENXIO);
881 
882 	/* Attach only to PF0 of the FPGA */
883 	if (d == 0xb000 && f != 0)
884 		return (ENXIO);
885 
886 	for (i = 0; i < nitems(t5_pciids); i++) {
887 		if (d == t5_pciids[i].device) {
888 			device_set_desc(dev, t5_pciids[i].desc);
889 			return (BUS_PROBE_DEFAULT);
890 		}
891 	}
892 
893 	return (ENXIO);
894 }
895 
896 static int
897 t6_probe(device_t dev)
898 {
899 	int i;
900 	uint16_t v = pci_get_vendor(dev);
901 	uint16_t d = pci_get_device(dev);
902 
903 	if (v != PCI_VENDOR_ID_CHELSIO)
904 		return (ENXIO);
905 
906 	for (i = 0; i < nitems(t6_pciids); i++) {
907 		if (d == t6_pciids[i].device) {
908 			device_set_desc(dev, t6_pciids[i].desc);
909 			return (BUS_PROBE_DEFAULT);
910 		}
911 	}
912 
913 	return (ENXIO);
914 }
915 
916 static void
917 t5_attribute_workaround(device_t dev)
918 {
919 	device_t root_port;
920 	uint32_t v;
921 
922 	/*
923 	 * The T5 chips do not properly echo the No Snoop and Relaxed
924 	 * Ordering attributes when replying to a TLP from a Root
925 	 * Port.  As a workaround, find the parent Root Port and
926 	 * disable No Snoop and Relaxed Ordering.  Note that this
927 	 * affects all devices under this root port.
928 	 */
929 	root_port = pci_find_pcie_root_port(dev);
930 	if (root_port == NULL) {
931 		device_printf(dev, "Unable to find parent root port\n");
932 		return;
933 	}
934 
935 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
936 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
937 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
938 	    0)
939 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
940 		    device_get_nameunit(root_port));
941 }
942 
943 static const struct devnames devnames[] = {
944 	{
945 		.nexus_name = "t4nex",
946 		.ifnet_name = "cxgbe",
947 		.vi_ifnet_name = "vcxgbe",
948 		.pf03_drv_name = "t4iov",
949 		.vf_nexus_name = "t4vf",
950 		.vf_ifnet_name = "cxgbev"
951 	}, {
952 		.nexus_name = "t5nex",
953 		.ifnet_name = "cxl",
954 		.vi_ifnet_name = "vcxl",
955 		.pf03_drv_name = "t5iov",
956 		.vf_nexus_name = "t5vf",
957 		.vf_ifnet_name = "cxlv"
958 	}, {
959 		.nexus_name = "t6nex",
960 		.ifnet_name = "cc",
961 		.vi_ifnet_name = "vcc",
962 		.pf03_drv_name = "t6iov",
963 		.vf_nexus_name = "t6vf",
964 		.vf_ifnet_name = "ccv"
965 	}
966 };
967 
968 void
969 t4_init_devnames(struct adapter *sc)
970 {
971 	int id;
972 
973 	id = chip_id(sc);
974 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
975 		sc->names = &devnames[id - CHELSIO_T4];
976 	else {
977 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
978 		sc->names = NULL;
979 	}
980 }
981 
982 static int
983 t4_ifnet_unit(struct adapter *sc, struct port_info *pi)
984 {
985 	const char *parent, *name;
986 	long value;
987 	int line, unit;
988 
989 	line = 0;
990 	parent = device_get_nameunit(sc->dev);
991 	name = sc->names->ifnet_name;
992 	while (resource_find_dev(&line, name, &unit, "at", parent) == 0) {
993 		if (resource_long_value(name, unit, "port", &value) == 0 &&
994 		    value == pi->port_id)
995 			return (unit);
996 	}
997 	return (-1);
998 }
999 
1000 static int
1001 t4_attach(device_t dev)
1002 {
1003 	struct adapter *sc;
1004 	int rc = 0, i, j, rqidx, tqidx, nports;
1005 	struct make_dev_args mda;
1006 	struct intrs_and_queues iaq;
1007 	struct sge *s;
1008 	uint32_t *buf;
1009 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1010 	int ofld_tqidx;
1011 #endif
1012 #ifdef TCP_OFFLOAD
1013 	int ofld_rqidx;
1014 #endif
1015 #ifdef DEV_NETMAP
1016 	int nm_rqidx, nm_tqidx;
1017 #endif
1018 	int num_vis;
1019 
1020 	sc = device_get_softc(dev);
1021 	sc->dev = dev;
1022 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
1023 
1024 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
1025 		t5_attribute_workaround(dev);
1026 	pci_enable_busmaster(dev);
1027 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
1028 		uint32_t v;
1029 
1030 		pci_set_max_read_req(dev, 4096);
1031 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
1032 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
1033 		if (pcie_relaxed_ordering == 0 &&
1034 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
1035 			v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
1036 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
1037 		} else if (pcie_relaxed_ordering == 1 &&
1038 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
1039 			v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
1040 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
1041 		}
1042 	}
1043 
1044 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
1045 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
1046 	sc->traceq = -1;
1047 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
1048 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
1049 	    device_get_nameunit(dev));
1050 
1051 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
1052 	    device_get_nameunit(dev));
1053 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
1054 	t4_add_adapter(sc);
1055 
1056 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
1057 	TAILQ_INIT(&sc->sfl);
1058 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
1059 
1060 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
1061 
1062 	sc->policy = NULL;
1063 	rw_init(&sc->policy_lock, "connection offload policy");
1064 
1065 	callout_init(&sc->ktls_tick, 1);
1066 
1067 #ifdef TCP_OFFLOAD
1068 	TASK_INIT(&sc->async_event_task, 0, t4_async_event, sc);
1069 #endif
1070 
1071 	rc = t4_map_bars_0_and_4(sc);
1072 	if (rc != 0)
1073 		goto done; /* error message displayed already */
1074 
1075 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
1076 
1077 	/* Prepare the adapter for operation. */
1078 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
1079 	rc = -t4_prep_adapter(sc, buf);
1080 	free(buf, M_CXGBE);
1081 	if (rc != 0) {
1082 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
1083 		goto done;
1084 	}
1085 
1086 	/*
1087 	 * This is the real PF# to which we're attaching.  Works from within PCI
1088 	 * passthrough environments too, where pci_get_function() could return a
1089 	 * different PF# depending on the passthrough configuration.  We need to
1090 	 * use the real PF# in all our communication with the firmware.
1091 	 */
1092 	j = t4_read_reg(sc, A_PL_WHOAMI);
1093 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
1094 	sc->mbox = sc->pf;
1095 
1096 	t4_init_devnames(sc);
1097 	if (sc->names == NULL) {
1098 		rc = ENOTSUP;
1099 		goto done; /* error message displayed already */
1100 	}
1101 
1102 	/*
1103 	 * Do this really early, with the memory windows set up even before the
1104 	 * character device.  The userland tool's register i/o and mem read
1105 	 * will work even in "recovery mode".
1106 	 */
1107 	setup_memwin(sc);
1108 	if (t4_init_devlog_params(sc, 0) == 0)
1109 		fixup_devlog_params(sc);
1110 	make_dev_args_init(&mda);
1111 	mda.mda_devsw = &t4_cdevsw;
1112 	mda.mda_uid = UID_ROOT;
1113 	mda.mda_gid = GID_WHEEL;
1114 	mda.mda_mode = 0600;
1115 	mda.mda_si_drv1 = sc;
1116 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
1117 	if (rc != 0)
1118 		device_printf(dev, "failed to create nexus char device: %d.\n",
1119 		    rc);
1120 
1121 	/* Go no further if recovery mode has been requested. */
1122 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
1123 		device_printf(dev, "recovery mode.\n");
1124 		goto done;
1125 	}
1126 
1127 #if defined(__i386__)
1128 	if ((cpu_feature & CPUID_CX8) == 0) {
1129 		device_printf(dev, "64 bit atomics not available.\n");
1130 		rc = ENOTSUP;
1131 		goto done;
1132 	}
1133 #endif
1134 
1135 	/* Contact the firmware and try to become the master driver. */
1136 	rc = contact_firmware(sc);
1137 	if (rc != 0)
1138 		goto done; /* error message displayed already */
1139 	MPASS(sc->flags & FW_OK);
1140 
1141 	rc = get_params__pre_init(sc);
1142 	if (rc != 0)
1143 		goto done; /* error message displayed already */
1144 
1145 	if (sc->flags & MASTER_PF) {
1146 		rc = partition_resources(sc);
1147 		if (rc != 0)
1148 			goto done; /* error message displayed already */
1149 		t4_intr_clear(sc);
1150 	}
1151 
1152 	rc = get_params__post_init(sc);
1153 	if (rc != 0)
1154 		goto done; /* error message displayed already */
1155 
1156 	rc = set_params__post_init(sc);
1157 	if (rc != 0)
1158 		goto done; /* error message displayed already */
1159 
1160 	rc = t4_map_bar_2(sc);
1161 	if (rc != 0)
1162 		goto done; /* error message displayed already */
1163 
1164 	rc = t4_create_dma_tag(sc);
1165 	if (rc != 0)
1166 		goto done; /* error message displayed already */
1167 
1168 	/*
1169 	 * First pass over all the ports - allocate VIs and initialize some
1170 	 * basic parameters like mac address, port type, etc.
1171 	 */
1172 	for_each_port(sc, i) {
1173 		struct port_info *pi;
1174 
1175 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
1176 		sc->port[i] = pi;
1177 
1178 		/* These must be set before t4_port_init */
1179 		pi->adapter = sc;
1180 		pi->port_id = i;
1181 		/*
1182 		 * XXX: vi[0] is special so we can't delay this allocation until
1183 		 * pi->nvi's final value is known.
1184 		 */
1185 		pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
1186 		    M_ZERO | M_WAITOK);
1187 
1188 		/*
1189 		 * Allocate the "main" VI and initialize parameters
1190 		 * like mac addr.
1191 		 */
1192 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1193 		if (rc != 0) {
1194 			device_printf(dev, "unable to initialize port %d: %d\n",
1195 			    i, rc);
1196 			free(pi->vi, M_CXGBE);
1197 			free(pi, M_CXGBE);
1198 			sc->port[i] = NULL;
1199 			goto done;
1200 		}
1201 
1202 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1203 		    device_get_nameunit(dev), i);
1204 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1205 		sc->chan_map[pi->tx_chan] = i;
1206 
1207 		/* All VIs on this port share this media. */
1208 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1209 		    cxgbe_media_status);
1210 
1211 		PORT_LOCK(pi);
1212 		init_link_config(pi);
1213 		fixup_link_config(pi);
1214 		build_medialist(pi);
1215 		if (fixed_ifmedia(pi))
1216 			pi->flags |= FIXED_IFMEDIA;
1217 		PORT_UNLOCK(pi);
1218 
1219 		pi->dev = device_add_child(dev, sc->names->ifnet_name,
1220 		    t4_ifnet_unit(sc, pi));
1221 		if (pi->dev == NULL) {
1222 			device_printf(dev,
1223 			    "failed to add device for port %d.\n", i);
1224 			rc = ENXIO;
1225 			goto done;
1226 		}
1227 		pi->vi[0].dev = pi->dev;
1228 		device_set_softc(pi->dev, pi);
1229 	}
1230 
1231 	/*
1232 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1233 	 */
1234 	nports = sc->params.nports;
1235 	rc = cfg_itype_and_nqueues(sc, &iaq);
1236 	if (rc != 0)
1237 		goto done; /* error message displayed already */
1238 
1239 	num_vis = iaq.num_vis;
1240 	sc->intr_type = iaq.intr_type;
1241 	sc->intr_count = iaq.nirq;
1242 
1243 	s = &sc->sge;
1244 	s->nrxq = nports * iaq.nrxq;
1245 	s->ntxq = nports * iaq.ntxq;
1246 	if (num_vis > 1) {
1247 		s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1248 		s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1249 	}
1250 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1251 	s->neq += nports;		/* ctrl queues: 1 per port */
1252 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1253 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1254 	if (is_offload(sc) || is_ethoffload(sc)) {
1255 		s->nofldtxq = nports * iaq.nofldtxq;
1256 		if (num_vis > 1)
1257 			s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1258 		s->neq += s->nofldtxq;
1259 
1260 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1261 		    M_CXGBE, M_ZERO | M_WAITOK);
1262 	}
1263 #endif
1264 #ifdef TCP_OFFLOAD
1265 	if (is_offload(sc)) {
1266 		s->nofldrxq = nports * iaq.nofldrxq;
1267 		if (num_vis > 1)
1268 			s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1269 		s->neq += s->nofldrxq;	/* free list */
1270 		s->niq += s->nofldrxq;
1271 
1272 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1273 		    M_CXGBE, M_ZERO | M_WAITOK);
1274 	}
1275 #endif
1276 #ifdef DEV_NETMAP
1277 	s->nnmrxq = 0;
1278 	s->nnmtxq = 0;
1279 	if (t4_native_netmap & NN_MAIN_VI) {
1280 		s->nnmrxq += nports * iaq.nnmrxq;
1281 		s->nnmtxq += nports * iaq.nnmtxq;
1282 	}
1283 	if (num_vis > 1 && t4_native_netmap & NN_EXTRA_VI) {
1284 		s->nnmrxq += nports * (num_vis - 1) * iaq.nnmrxq_vi;
1285 		s->nnmtxq += nports * (num_vis - 1) * iaq.nnmtxq_vi;
1286 	}
1287 	s->neq += s->nnmtxq + s->nnmrxq;
1288 	s->niq += s->nnmrxq;
1289 
1290 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1291 	    M_CXGBE, M_ZERO | M_WAITOK);
1292 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1293 	    M_CXGBE, M_ZERO | M_WAITOK);
1294 #endif
1295 
1296 	s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1297 	    M_ZERO | M_WAITOK);
1298 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1299 	    M_ZERO | M_WAITOK);
1300 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1301 	    M_ZERO | M_WAITOK);
1302 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1303 	    M_ZERO | M_WAITOK);
1304 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1305 	    M_ZERO | M_WAITOK);
1306 
1307 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1308 	    M_ZERO | M_WAITOK);
1309 
1310 	t4_init_l2t(sc, M_WAITOK);
1311 	t4_init_smt(sc, M_WAITOK);
1312 	t4_init_tx_sched(sc);
1313 	t4_init_atid_table(sc);
1314 #ifdef RATELIMIT
1315 	t4_init_etid_table(sc);
1316 #endif
1317 #ifdef INET6
1318 	t4_init_clip_table(sc);
1319 #endif
1320 	if (sc->vres.key.size != 0)
1321 		sc->key_map = vmem_create("T4TLS key map", sc->vres.key.start,
1322 		    sc->vres.key.size, 32, 0, M_FIRSTFIT | M_WAITOK);
1323 
1324 	/*
1325 	 * Second pass over the ports.  This time we know the number of rx and
1326 	 * tx queues that each port should get.
1327 	 */
1328 	rqidx = tqidx = 0;
1329 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1330 	ofld_tqidx = 0;
1331 #endif
1332 #ifdef TCP_OFFLOAD
1333 	ofld_rqidx = 0;
1334 #endif
1335 #ifdef DEV_NETMAP
1336 	nm_rqidx = nm_tqidx = 0;
1337 #endif
1338 	for_each_port(sc, i) {
1339 		struct port_info *pi = sc->port[i];
1340 		struct vi_info *vi;
1341 
1342 		if (pi == NULL)
1343 			continue;
1344 
1345 		pi->nvi = num_vis;
1346 		for_each_vi(pi, j, vi) {
1347 			vi->pi = pi;
1348 			vi->qsize_rxq = t4_qsize_rxq;
1349 			vi->qsize_txq = t4_qsize_txq;
1350 
1351 			vi->first_rxq = rqidx;
1352 			vi->first_txq = tqidx;
1353 			vi->tmr_idx = t4_tmr_idx;
1354 			vi->pktc_idx = t4_pktc_idx;
1355 			vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1356 			vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1357 
1358 			rqidx += vi->nrxq;
1359 			tqidx += vi->ntxq;
1360 
1361 			if (j == 0 && vi->ntxq > 1)
1362 				vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1363 			else
1364 				vi->rsrv_noflowq = 0;
1365 
1366 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1367 			vi->first_ofld_txq = ofld_tqidx;
1368 			vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1369 			ofld_tqidx += vi->nofldtxq;
1370 #endif
1371 #ifdef TCP_OFFLOAD
1372 			vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1373 			vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1374 			vi->first_ofld_rxq = ofld_rqidx;
1375 			vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1376 
1377 			ofld_rqidx += vi->nofldrxq;
1378 #endif
1379 #ifdef DEV_NETMAP
1380 			vi->first_nm_rxq = nm_rqidx;
1381 			vi->first_nm_txq = nm_tqidx;
1382 			if (j == 0) {
1383 				vi->nnmrxq = iaq.nnmrxq;
1384 				vi->nnmtxq = iaq.nnmtxq;
1385 			} else {
1386 				vi->nnmrxq = iaq.nnmrxq_vi;
1387 				vi->nnmtxq = iaq.nnmtxq_vi;
1388 			}
1389 			nm_rqidx += vi->nnmrxq;
1390 			nm_tqidx += vi->nnmtxq;
1391 #endif
1392 		}
1393 	}
1394 
1395 	rc = t4_setup_intr_handlers(sc);
1396 	if (rc != 0) {
1397 		device_printf(dev,
1398 		    "failed to setup interrupt handlers: %d\n", rc);
1399 		goto done;
1400 	}
1401 
1402 	rc = bus_generic_probe(dev);
1403 	if (rc != 0) {
1404 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1405 		goto done;
1406 	}
1407 
1408 	/*
1409 	 * Ensure thread-safe mailbox access (in debug builds).
1410 	 *
1411 	 * So far this was the only thread accessing the mailbox but various
1412 	 * ifnets and sysctls are about to be created and their handlers/ioctls
1413 	 * will access the mailbox from different threads.
1414 	 */
1415 	sc->flags |= CHK_MBOX_ACCESS;
1416 
1417 	rc = bus_generic_attach(dev);
1418 	if (rc != 0) {
1419 		device_printf(dev,
1420 		    "failed to attach all child ports: %d\n", rc);
1421 		goto done;
1422 	}
1423 
1424 	device_printf(dev,
1425 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1426 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1427 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1428 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1429 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1430 
1431 	t4_set_desc(sc);
1432 
1433 	notify_siblings(dev, 0);
1434 
1435 done:
1436 	if (rc != 0 && sc->cdev) {
1437 		/* cdev was created and so cxgbetool works; recover that way. */
1438 		device_printf(dev,
1439 		    "error during attach, adapter is now in recovery mode.\n");
1440 		rc = 0;
1441 	}
1442 
1443 	if (rc != 0)
1444 		t4_detach_common(dev);
1445 	else
1446 		t4_sysctls(sc);
1447 
1448 	return (rc);
1449 }
1450 
1451 static int
1452 t4_child_location_str(device_t bus, device_t dev, char *buf, size_t buflen)
1453 {
1454 	struct adapter *sc;
1455 	struct port_info *pi;
1456 	int i;
1457 
1458 	sc = device_get_softc(bus);
1459 	buf[0] = '\0';
1460 	for_each_port(sc, i) {
1461 		pi = sc->port[i];
1462 		if (pi != NULL && pi->dev == dev) {
1463 			snprintf(buf, buflen, "port=%d", pi->port_id);
1464 			break;
1465 		}
1466 	}
1467 	return (0);
1468 }
1469 
1470 static int
1471 t4_ready(device_t dev)
1472 {
1473 	struct adapter *sc;
1474 
1475 	sc = device_get_softc(dev);
1476 	if (sc->flags & FW_OK)
1477 		return (0);
1478 	return (ENXIO);
1479 }
1480 
1481 static int
1482 t4_read_port_device(device_t dev, int port, device_t *child)
1483 {
1484 	struct adapter *sc;
1485 	struct port_info *pi;
1486 
1487 	sc = device_get_softc(dev);
1488 	if (port < 0 || port >= MAX_NPORTS)
1489 		return (EINVAL);
1490 	pi = sc->port[port];
1491 	if (pi == NULL || pi->dev == NULL)
1492 		return (ENXIO);
1493 	*child = pi->dev;
1494 	return (0);
1495 }
1496 
1497 static int
1498 notify_siblings(device_t dev, int detaching)
1499 {
1500 	device_t sibling;
1501 	int error, i;
1502 
1503 	error = 0;
1504 	for (i = 0; i < PCI_FUNCMAX; i++) {
1505 		if (i == pci_get_function(dev))
1506 			continue;
1507 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1508 		    pci_get_slot(dev), i);
1509 		if (sibling == NULL || !device_is_attached(sibling))
1510 			continue;
1511 		if (detaching)
1512 			error = T4_DETACH_CHILD(sibling);
1513 		else
1514 			(void)T4_ATTACH_CHILD(sibling);
1515 		if (error)
1516 			break;
1517 	}
1518 	return (error);
1519 }
1520 
1521 /*
1522  * Idempotent
1523  */
1524 static int
1525 t4_detach(device_t dev)
1526 {
1527 	struct adapter *sc;
1528 	int rc;
1529 
1530 	sc = device_get_softc(dev);
1531 
1532 	rc = notify_siblings(dev, 1);
1533 	if (rc) {
1534 		device_printf(dev,
1535 		    "failed to detach sibling devices: %d\n", rc);
1536 		return (rc);
1537 	}
1538 
1539 	return (t4_detach_common(dev));
1540 }
1541 
1542 int
1543 t4_detach_common(device_t dev)
1544 {
1545 	struct adapter *sc;
1546 	struct port_info *pi;
1547 	int i, rc;
1548 
1549 	sc = device_get_softc(dev);
1550 
1551 	if (sc->cdev) {
1552 		destroy_dev(sc->cdev);
1553 		sc->cdev = NULL;
1554 	}
1555 
1556 	sx_xlock(&t4_list_lock);
1557 	SLIST_REMOVE(&t4_list, sc, adapter, link);
1558 	sx_xunlock(&t4_list_lock);
1559 
1560 	sc->flags &= ~CHK_MBOX_ACCESS;
1561 	if (sc->flags & FULL_INIT_DONE) {
1562 		if (!(sc->flags & IS_VF))
1563 			t4_intr_disable(sc);
1564 	}
1565 
1566 	if (device_is_attached(dev)) {
1567 		rc = bus_generic_detach(dev);
1568 		if (rc) {
1569 			device_printf(dev,
1570 			    "failed to detach child devices: %d\n", rc);
1571 			return (rc);
1572 		}
1573 	}
1574 
1575 #ifdef TCP_OFFLOAD
1576 	taskqueue_drain(taskqueue_thread, &sc->async_event_task);
1577 #endif
1578 
1579 	for (i = 0; i < sc->intr_count; i++)
1580 		t4_free_irq(sc, &sc->irq[i]);
1581 
1582 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1583 		t4_free_tx_sched(sc);
1584 
1585 	for (i = 0; i < MAX_NPORTS; i++) {
1586 		pi = sc->port[i];
1587 		if (pi) {
1588 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1589 			if (pi->dev)
1590 				device_delete_child(dev, pi->dev);
1591 
1592 			mtx_destroy(&pi->pi_lock);
1593 			free(pi->vi, M_CXGBE);
1594 			free(pi, M_CXGBE);
1595 		}
1596 	}
1597 
1598 	device_delete_children(dev);
1599 
1600 	if (sc->flags & FULL_INIT_DONE)
1601 		adapter_full_uninit(sc);
1602 
1603 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1604 		t4_fw_bye(sc, sc->mbox);
1605 
1606 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1607 		pci_release_msi(dev);
1608 
1609 	if (sc->regs_res)
1610 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1611 		    sc->regs_res);
1612 
1613 	if (sc->udbs_res)
1614 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1615 		    sc->udbs_res);
1616 
1617 	if (sc->msix_res)
1618 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1619 		    sc->msix_res);
1620 
1621 	if (sc->l2t)
1622 		t4_free_l2t(sc->l2t);
1623 	if (sc->smt)
1624 		t4_free_smt(sc->smt);
1625 	t4_free_atid_table(sc);
1626 #ifdef RATELIMIT
1627 	t4_free_etid_table(sc);
1628 #endif
1629 	if (sc->key_map)
1630 		vmem_destroy(sc->key_map);
1631 #ifdef INET6
1632 	t4_destroy_clip_table(sc);
1633 #endif
1634 
1635 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1636 	free(sc->sge.ofld_txq, M_CXGBE);
1637 #endif
1638 #ifdef TCP_OFFLOAD
1639 	free(sc->sge.ofld_rxq, M_CXGBE);
1640 #endif
1641 #ifdef DEV_NETMAP
1642 	free(sc->sge.nm_rxq, M_CXGBE);
1643 	free(sc->sge.nm_txq, M_CXGBE);
1644 #endif
1645 	free(sc->irq, M_CXGBE);
1646 	free(sc->sge.rxq, M_CXGBE);
1647 	free(sc->sge.txq, M_CXGBE);
1648 	free(sc->sge.ctrlq, M_CXGBE);
1649 	free(sc->sge.iqmap, M_CXGBE);
1650 	free(sc->sge.eqmap, M_CXGBE);
1651 	free(sc->tids.ftid_tab, M_CXGBE);
1652 	free(sc->tids.hpftid_tab, M_CXGBE);
1653 	free_hftid_hash(&sc->tids);
1654 	free(sc->tids.tid_tab, M_CXGBE);
1655 	free(sc->tt.tls_rx_ports, M_CXGBE);
1656 	t4_destroy_dma_tag(sc);
1657 
1658 	callout_drain(&sc->ktls_tick);
1659 	callout_drain(&sc->sfl_callout);
1660 	if (mtx_initialized(&sc->tids.ftid_lock)) {
1661 		mtx_destroy(&sc->tids.ftid_lock);
1662 		cv_destroy(&sc->tids.ftid_cv);
1663 	}
1664 	if (mtx_initialized(&sc->tids.atid_lock))
1665 		mtx_destroy(&sc->tids.atid_lock);
1666 	if (mtx_initialized(&sc->ifp_lock))
1667 		mtx_destroy(&sc->ifp_lock);
1668 
1669 	if (rw_initialized(&sc->policy_lock)) {
1670 		rw_destroy(&sc->policy_lock);
1671 #ifdef TCP_OFFLOAD
1672 		if (sc->policy != NULL)
1673 			free_offload_policy(sc->policy);
1674 #endif
1675 	}
1676 
1677 	for (i = 0; i < NUM_MEMWIN; i++) {
1678 		struct memwin *mw = &sc->memwin[i];
1679 
1680 		if (rw_initialized(&mw->mw_lock))
1681 			rw_destroy(&mw->mw_lock);
1682 	}
1683 
1684 	mtx_destroy(&sc->sfl_lock);
1685 	mtx_destroy(&sc->reg_lock);
1686 	mtx_destroy(&sc->sc_lock);
1687 
1688 	bzero(sc, sizeof(*sc));
1689 
1690 	return (0);
1691 }
1692 
1693 static int
1694 cxgbe_probe(device_t dev)
1695 {
1696 	char buf[128];
1697 	struct port_info *pi = device_get_softc(dev);
1698 
1699 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1700 	device_set_desc_copy(dev, buf);
1701 
1702 	return (BUS_PROBE_DEFAULT);
1703 }
1704 
1705 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1706     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1707     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS | \
1708     IFCAP_HWRXTSTMP | IFCAP_NOMAP)
1709 #define T4_CAP_ENABLE (T4_CAP)
1710 
1711 static int
1712 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1713 {
1714 	struct ifnet *ifp;
1715 	struct sbuf *sb;
1716 	struct pfil_head_args pa;
1717 
1718 	vi->xact_addr_filt = -1;
1719 	callout_init(&vi->tick, 1);
1720 
1721 	/* Allocate an ifnet and set it up */
1722 	ifp = if_alloc_dev(IFT_ETHER, dev);
1723 	if (ifp == NULL) {
1724 		device_printf(dev, "Cannot allocate ifnet\n");
1725 		return (ENOMEM);
1726 	}
1727 	vi->ifp = ifp;
1728 	ifp->if_softc = vi;
1729 
1730 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1731 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1732 
1733 	ifp->if_init = cxgbe_init;
1734 	ifp->if_ioctl = cxgbe_ioctl;
1735 	ifp->if_transmit = cxgbe_transmit;
1736 	ifp->if_qflush = cxgbe_qflush;
1737 	ifp->if_get_counter = cxgbe_get_counter;
1738 #if defined(KERN_TLS) || defined(RATELIMIT)
1739 	ifp->if_snd_tag_alloc = cxgbe_snd_tag_alloc;
1740 	ifp->if_snd_tag_modify = cxgbe_snd_tag_modify;
1741 	ifp->if_snd_tag_query = cxgbe_snd_tag_query;
1742 	ifp->if_snd_tag_free = cxgbe_snd_tag_free;
1743 #endif
1744 #ifdef RATELIMIT
1745 	ifp->if_ratelimit_query = cxgbe_ratelimit_query;
1746 #endif
1747 
1748 	ifp->if_capabilities = T4_CAP;
1749 	ifp->if_capenable = T4_CAP_ENABLE;
1750 #ifdef TCP_OFFLOAD
1751 	if (vi->nofldrxq != 0 && (vi->pi->adapter->flags & KERN_TLS_OK) == 0)
1752 		ifp->if_capabilities |= IFCAP_TOE;
1753 #endif
1754 #ifdef RATELIMIT
1755 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0) {
1756 		ifp->if_capabilities |= IFCAP_TXRTLMT;
1757 		ifp->if_capenable |= IFCAP_TXRTLMT;
1758 	}
1759 #endif
1760 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1761 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1762 
1763 	ifp->if_hw_tsomax = IP_MAXPACKET;
1764 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_TSO;
1765 #ifdef RATELIMIT
1766 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0)
1767 		ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_EO_TSO;
1768 #endif
1769 	ifp->if_hw_tsomaxsegsize = 65536;
1770 #ifdef KERN_TLS
1771 	if (vi->pi->adapter->flags & KERN_TLS_OK) {
1772 		ifp->if_capabilities |= IFCAP_TXTLS;
1773 		ifp->if_capenable |= IFCAP_TXTLS;
1774 	}
1775 #endif
1776 
1777 	ether_ifattach(ifp, vi->hw_addr);
1778 #ifdef DEV_NETMAP
1779 	if (vi->nnmrxq != 0)
1780 		cxgbe_nm_attach(vi);
1781 #endif
1782 	sb = sbuf_new_auto();
1783 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1784 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1785 	switch (ifp->if_capabilities & (IFCAP_TOE | IFCAP_TXRTLMT)) {
1786 	case IFCAP_TOE:
1787 		sbuf_printf(sb, "; %d txq (TOE)", vi->nofldtxq);
1788 		break;
1789 	case IFCAP_TOE | IFCAP_TXRTLMT:
1790 		sbuf_printf(sb, "; %d txq (TOE/ETHOFLD)", vi->nofldtxq);
1791 		break;
1792 	case IFCAP_TXRTLMT:
1793 		sbuf_printf(sb, "; %d txq (ETHOFLD)", vi->nofldtxq);
1794 		break;
1795 	}
1796 #endif
1797 #ifdef TCP_OFFLOAD
1798 	if (ifp->if_capabilities & IFCAP_TOE)
1799 		sbuf_printf(sb, ", %d rxq (TOE)", vi->nofldrxq);
1800 #endif
1801 #ifdef DEV_NETMAP
1802 	if (ifp->if_capabilities & IFCAP_NETMAP)
1803 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1804 		    vi->nnmtxq, vi->nnmrxq);
1805 #endif
1806 	sbuf_finish(sb);
1807 	device_printf(dev, "%s\n", sbuf_data(sb));
1808 	sbuf_delete(sb);
1809 
1810 	vi_sysctls(vi);
1811 
1812 	pa.pa_version = PFIL_VERSION;
1813 	pa.pa_flags = PFIL_IN;
1814 	pa.pa_type = PFIL_TYPE_ETHERNET;
1815 	pa.pa_headname = ifp->if_xname;
1816 	vi->pfil = pfil_head_register(&pa);
1817 
1818 	return (0);
1819 }
1820 
1821 static int
1822 cxgbe_attach(device_t dev)
1823 {
1824 	struct port_info *pi = device_get_softc(dev);
1825 	struct adapter *sc = pi->adapter;
1826 	struct vi_info *vi;
1827 	int i, rc;
1828 
1829 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1830 
1831 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1832 	if (rc)
1833 		return (rc);
1834 
1835 	for_each_vi(pi, i, vi) {
1836 		if (i == 0)
1837 			continue;
1838 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1839 		if (vi->dev == NULL) {
1840 			device_printf(dev, "failed to add VI %d\n", i);
1841 			continue;
1842 		}
1843 		device_set_softc(vi->dev, vi);
1844 	}
1845 
1846 	cxgbe_sysctls(pi);
1847 
1848 	bus_generic_attach(dev);
1849 
1850 	return (0);
1851 }
1852 
1853 static void
1854 cxgbe_vi_detach(struct vi_info *vi)
1855 {
1856 	struct ifnet *ifp = vi->ifp;
1857 
1858 	if (vi->pfil != NULL) {
1859 		pfil_head_unregister(vi->pfil);
1860 		vi->pfil = NULL;
1861 	}
1862 
1863 	ether_ifdetach(ifp);
1864 
1865 	/* Let detach proceed even if these fail. */
1866 #ifdef DEV_NETMAP
1867 	if (ifp->if_capabilities & IFCAP_NETMAP)
1868 		cxgbe_nm_detach(vi);
1869 #endif
1870 	cxgbe_uninit_synchronized(vi);
1871 	callout_drain(&vi->tick);
1872 	vi_full_uninit(vi);
1873 
1874 	if_free(vi->ifp);
1875 	vi->ifp = NULL;
1876 }
1877 
1878 static int
1879 cxgbe_detach(device_t dev)
1880 {
1881 	struct port_info *pi = device_get_softc(dev);
1882 	struct adapter *sc = pi->adapter;
1883 	int rc;
1884 
1885 	/* Detach the extra VIs first. */
1886 	rc = bus_generic_detach(dev);
1887 	if (rc)
1888 		return (rc);
1889 	device_delete_children(dev);
1890 
1891 	doom_vi(sc, &pi->vi[0]);
1892 
1893 	if (pi->flags & HAS_TRACEQ) {
1894 		sc->traceq = -1;	/* cloner should not create ifnet */
1895 		t4_tracer_port_detach(sc);
1896 	}
1897 
1898 	cxgbe_vi_detach(&pi->vi[0]);
1899 	callout_drain(&pi->tick);
1900 	ifmedia_removeall(&pi->media);
1901 
1902 	end_synchronized_op(sc, 0);
1903 
1904 	return (0);
1905 }
1906 
1907 static void
1908 cxgbe_init(void *arg)
1909 {
1910 	struct vi_info *vi = arg;
1911 	struct adapter *sc = vi->pi->adapter;
1912 
1913 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1914 		return;
1915 	cxgbe_init_synchronized(vi);
1916 	end_synchronized_op(sc, 0);
1917 }
1918 
1919 static int
1920 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1921 {
1922 	int rc = 0, mtu, flags;
1923 	struct vi_info *vi = ifp->if_softc;
1924 	struct port_info *pi = vi->pi;
1925 	struct adapter *sc = pi->adapter;
1926 	struct ifreq *ifr = (struct ifreq *)data;
1927 	uint32_t mask;
1928 
1929 	switch (cmd) {
1930 	case SIOCSIFMTU:
1931 		mtu = ifr->ifr_mtu;
1932 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1933 			return (EINVAL);
1934 
1935 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1936 		if (rc)
1937 			return (rc);
1938 		ifp->if_mtu = mtu;
1939 		if (vi->flags & VI_INIT_DONE) {
1940 			t4_update_fl_bufsize(ifp);
1941 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1942 				rc = update_mac_settings(ifp, XGMAC_MTU);
1943 		}
1944 		end_synchronized_op(sc, 0);
1945 		break;
1946 
1947 	case SIOCSIFFLAGS:
1948 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4flg");
1949 		if (rc)
1950 			return (rc);
1951 
1952 		if (ifp->if_flags & IFF_UP) {
1953 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1954 				flags = vi->if_flags;
1955 				if ((ifp->if_flags ^ flags) &
1956 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1957 					rc = update_mac_settings(ifp,
1958 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1959 				}
1960 			} else {
1961 				rc = cxgbe_init_synchronized(vi);
1962 			}
1963 			vi->if_flags = ifp->if_flags;
1964 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1965 			rc = cxgbe_uninit_synchronized(vi);
1966 		}
1967 		end_synchronized_op(sc, 0);
1968 		break;
1969 
1970 	case SIOCADDMULTI:
1971 	case SIOCDELMULTI:
1972 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4multi");
1973 		if (rc)
1974 			return (rc);
1975 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1976 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1977 		end_synchronized_op(sc, 0);
1978 		break;
1979 
1980 	case SIOCSIFCAP:
1981 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1982 		if (rc)
1983 			return (rc);
1984 
1985 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1986 		if (mask & IFCAP_TXCSUM) {
1987 			ifp->if_capenable ^= IFCAP_TXCSUM;
1988 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1989 
1990 			if (IFCAP_TSO4 & ifp->if_capenable &&
1991 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1992 				ifp->if_capenable &= ~IFCAP_TSO4;
1993 				if_printf(ifp,
1994 				    "tso4 disabled due to -txcsum.\n");
1995 			}
1996 		}
1997 		if (mask & IFCAP_TXCSUM_IPV6) {
1998 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1999 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
2000 
2001 			if (IFCAP_TSO6 & ifp->if_capenable &&
2002 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2003 				ifp->if_capenable &= ~IFCAP_TSO6;
2004 				if_printf(ifp,
2005 				    "tso6 disabled due to -txcsum6.\n");
2006 			}
2007 		}
2008 		if (mask & IFCAP_RXCSUM)
2009 			ifp->if_capenable ^= IFCAP_RXCSUM;
2010 		if (mask & IFCAP_RXCSUM_IPV6)
2011 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
2012 
2013 		/*
2014 		 * Note that we leave CSUM_TSO alone (it is always set).  The
2015 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
2016 		 * sending a TSO request our way, so it's sufficient to toggle
2017 		 * IFCAP_TSOx only.
2018 		 */
2019 		if (mask & IFCAP_TSO4) {
2020 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
2021 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
2022 				if_printf(ifp, "enable txcsum first.\n");
2023 				rc = EAGAIN;
2024 				goto fail;
2025 			}
2026 			ifp->if_capenable ^= IFCAP_TSO4;
2027 		}
2028 		if (mask & IFCAP_TSO6) {
2029 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
2030 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
2031 				if_printf(ifp, "enable txcsum6 first.\n");
2032 				rc = EAGAIN;
2033 				goto fail;
2034 			}
2035 			ifp->if_capenable ^= IFCAP_TSO6;
2036 		}
2037 		if (mask & IFCAP_LRO) {
2038 #if defined(INET) || defined(INET6)
2039 			int i;
2040 			struct sge_rxq *rxq;
2041 
2042 			ifp->if_capenable ^= IFCAP_LRO;
2043 			for_each_rxq(vi, i, rxq) {
2044 				if (ifp->if_capenable & IFCAP_LRO)
2045 					rxq->iq.flags |= IQ_LRO_ENABLED;
2046 				else
2047 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
2048 			}
2049 #endif
2050 		}
2051 #ifdef TCP_OFFLOAD
2052 		if (mask & IFCAP_TOE) {
2053 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
2054 
2055 			rc = toe_capability(vi, enable);
2056 			if (rc != 0)
2057 				goto fail;
2058 
2059 			ifp->if_capenable ^= mask;
2060 		}
2061 #endif
2062 		if (mask & IFCAP_VLAN_HWTAGGING) {
2063 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2064 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
2065 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
2066 		}
2067 		if (mask & IFCAP_VLAN_MTU) {
2068 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
2069 
2070 			/* Need to find out how to disable auto-mtu-inflation */
2071 		}
2072 		if (mask & IFCAP_VLAN_HWTSO)
2073 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
2074 		if (mask & IFCAP_VLAN_HWCSUM)
2075 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
2076 #ifdef RATELIMIT
2077 		if (mask & IFCAP_TXRTLMT)
2078 			ifp->if_capenable ^= IFCAP_TXRTLMT;
2079 #endif
2080 		if (mask & IFCAP_HWRXTSTMP) {
2081 			int i;
2082 			struct sge_rxq *rxq;
2083 
2084 			ifp->if_capenable ^= IFCAP_HWRXTSTMP;
2085 			for_each_rxq(vi, i, rxq) {
2086 				if (ifp->if_capenable & IFCAP_HWRXTSTMP)
2087 					rxq->iq.flags |= IQ_RX_TIMESTAMP;
2088 				else
2089 					rxq->iq.flags &= ~IQ_RX_TIMESTAMP;
2090 			}
2091 		}
2092 		if (mask & IFCAP_NOMAP)
2093 			ifp->if_capenable ^= IFCAP_NOMAP;
2094 
2095 #ifdef KERN_TLS
2096 		if (mask & IFCAP_TXTLS)
2097 			ifp->if_capenable ^= (mask & IFCAP_TXTLS);
2098 #endif
2099 
2100 #ifdef VLAN_CAPABILITIES
2101 		VLAN_CAPABILITIES(ifp);
2102 #endif
2103 fail:
2104 		end_synchronized_op(sc, 0);
2105 		break;
2106 
2107 	case SIOCSIFMEDIA:
2108 	case SIOCGIFMEDIA:
2109 	case SIOCGIFXMEDIA:
2110 		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
2111 		break;
2112 
2113 	case SIOCGI2C: {
2114 		struct ifi2creq i2c;
2115 
2116 		rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
2117 		if (rc != 0)
2118 			break;
2119 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
2120 			rc = EPERM;
2121 			break;
2122 		}
2123 		if (i2c.len > sizeof(i2c.data)) {
2124 			rc = EINVAL;
2125 			break;
2126 		}
2127 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
2128 		if (rc)
2129 			return (rc);
2130 		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
2131 		    i2c.offset, i2c.len, &i2c.data[0]);
2132 		end_synchronized_op(sc, 0);
2133 		if (rc == 0)
2134 			rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
2135 		break;
2136 	}
2137 
2138 	default:
2139 		rc = ether_ioctl(ifp, cmd, data);
2140 	}
2141 
2142 	return (rc);
2143 }
2144 
2145 static int
2146 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
2147 {
2148 	struct vi_info *vi = ifp->if_softc;
2149 	struct port_info *pi = vi->pi;
2150 	struct adapter *sc = pi->adapter;
2151 	struct sge_txq *txq;
2152 #ifdef RATELIMIT
2153 	struct cxgbe_snd_tag *cst;
2154 #endif
2155 	void *items[1];
2156 	int rc;
2157 
2158 	M_ASSERTPKTHDR(m);
2159 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
2160 #if defined(KERN_TLS) || defined(RATELIMIT)
2161 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
2162 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2163 #endif
2164 
2165 	if (__predict_false(pi->link_cfg.link_ok == false)) {
2166 		m_freem(m);
2167 		return (ENETDOWN);
2168 	}
2169 
2170 	rc = parse_pkt(sc, &m);
2171 	if (__predict_false(rc != 0)) {
2172 		MPASS(m == NULL);			/* was freed already */
2173 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
2174 		return (rc);
2175 	}
2176 #ifdef RATELIMIT
2177 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2178 		cst = mst_to_cst(m->m_pkthdr.snd_tag);
2179 		if (cst->type == IF_SND_TAG_TYPE_RATE_LIMIT)
2180 			return (ethofld_transmit(ifp, m));
2181 	}
2182 #endif
2183 
2184 	/* Select a txq. */
2185 	txq = &sc->sge.txq[vi->first_txq];
2186 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2187 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
2188 		    vi->rsrv_noflowq);
2189 
2190 	items[0] = m;
2191 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
2192 	if (__predict_false(rc != 0))
2193 		m_freem(m);
2194 
2195 	return (rc);
2196 }
2197 
2198 static void
2199 cxgbe_qflush(struct ifnet *ifp)
2200 {
2201 	struct vi_info *vi = ifp->if_softc;
2202 	struct sge_txq *txq;
2203 	int i;
2204 
2205 	/* queues do not exist if !VI_INIT_DONE. */
2206 	if (vi->flags & VI_INIT_DONE) {
2207 		for_each_txq(vi, i, txq) {
2208 			TXQ_LOCK(txq);
2209 			txq->eq.flags |= EQ_QFLUSH;
2210 			TXQ_UNLOCK(txq);
2211 			while (!mp_ring_is_idle(txq->r)) {
2212 				mp_ring_check_drainage(txq->r, 0);
2213 				pause("qflush", 1);
2214 			}
2215 			TXQ_LOCK(txq);
2216 			txq->eq.flags &= ~EQ_QFLUSH;
2217 			TXQ_UNLOCK(txq);
2218 		}
2219 	}
2220 	if_qflush(ifp);
2221 }
2222 
2223 static uint64_t
2224 vi_get_counter(struct ifnet *ifp, ift_counter c)
2225 {
2226 	struct vi_info *vi = ifp->if_softc;
2227 	struct fw_vi_stats_vf *s = &vi->stats;
2228 
2229 	vi_refresh_stats(vi->pi->adapter, vi);
2230 
2231 	switch (c) {
2232 	case IFCOUNTER_IPACKETS:
2233 		return (s->rx_bcast_frames + s->rx_mcast_frames +
2234 		    s->rx_ucast_frames);
2235 	case IFCOUNTER_IERRORS:
2236 		return (s->rx_err_frames);
2237 	case IFCOUNTER_OPACKETS:
2238 		return (s->tx_bcast_frames + s->tx_mcast_frames +
2239 		    s->tx_ucast_frames + s->tx_offload_frames);
2240 	case IFCOUNTER_OERRORS:
2241 		return (s->tx_drop_frames);
2242 	case IFCOUNTER_IBYTES:
2243 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
2244 		    s->rx_ucast_bytes);
2245 	case IFCOUNTER_OBYTES:
2246 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
2247 		    s->tx_ucast_bytes + s->tx_offload_bytes);
2248 	case IFCOUNTER_IMCASTS:
2249 		return (s->rx_mcast_frames);
2250 	case IFCOUNTER_OMCASTS:
2251 		return (s->tx_mcast_frames);
2252 	case IFCOUNTER_OQDROPS: {
2253 		uint64_t drops;
2254 
2255 		drops = 0;
2256 		if (vi->flags & VI_INIT_DONE) {
2257 			int i;
2258 			struct sge_txq *txq;
2259 
2260 			for_each_txq(vi, i, txq)
2261 				drops += counter_u64_fetch(txq->r->drops);
2262 		}
2263 
2264 		return (drops);
2265 
2266 	}
2267 
2268 	default:
2269 		return (if_get_counter_default(ifp, c));
2270 	}
2271 }
2272 
2273 uint64_t
2274 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
2275 {
2276 	struct vi_info *vi = ifp->if_softc;
2277 	struct port_info *pi = vi->pi;
2278 	struct adapter *sc = pi->adapter;
2279 	struct port_stats *s = &pi->stats;
2280 
2281 	if (pi->nvi > 1 || sc->flags & IS_VF)
2282 		return (vi_get_counter(ifp, c));
2283 
2284 	cxgbe_refresh_stats(sc, pi);
2285 
2286 	switch (c) {
2287 	case IFCOUNTER_IPACKETS:
2288 		return (s->rx_frames);
2289 
2290 	case IFCOUNTER_IERRORS:
2291 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
2292 		    s->rx_fcs_err + s->rx_len_err);
2293 
2294 	case IFCOUNTER_OPACKETS:
2295 		return (s->tx_frames);
2296 
2297 	case IFCOUNTER_OERRORS:
2298 		return (s->tx_error_frames);
2299 
2300 	case IFCOUNTER_IBYTES:
2301 		return (s->rx_octets);
2302 
2303 	case IFCOUNTER_OBYTES:
2304 		return (s->tx_octets);
2305 
2306 	case IFCOUNTER_IMCASTS:
2307 		return (s->rx_mcast_frames);
2308 
2309 	case IFCOUNTER_OMCASTS:
2310 		return (s->tx_mcast_frames);
2311 
2312 	case IFCOUNTER_IQDROPS:
2313 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
2314 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
2315 		    s->rx_trunc3 + pi->tnl_cong_drops);
2316 
2317 	case IFCOUNTER_OQDROPS: {
2318 		uint64_t drops;
2319 
2320 		drops = s->tx_drop;
2321 		if (vi->flags & VI_INIT_DONE) {
2322 			int i;
2323 			struct sge_txq *txq;
2324 
2325 			for_each_txq(vi, i, txq)
2326 				drops += counter_u64_fetch(txq->r->drops);
2327 		}
2328 
2329 		return (drops);
2330 
2331 	}
2332 
2333 	default:
2334 		return (if_get_counter_default(ifp, c));
2335 	}
2336 }
2337 
2338 #if defined(KERN_TLS) || defined(RATELIMIT)
2339 void
2340 cxgbe_snd_tag_init(struct cxgbe_snd_tag *cst, struct ifnet *ifp, int type)
2341 {
2342 
2343 	m_snd_tag_init(&cst->com, ifp);
2344 	cst->type = type;
2345 }
2346 
2347 static int
2348 cxgbe_snd_tag_alloc(struct ifnet *ifp, union if_snd_tag_alloc_params *params,
2349     struct m_snd_tag **pt)
2350 {
2351 	int error;
2352 
2353 	switch (params->hdr.type) {
2354 #ifdef RATELIMIT
2355 	case IF_SND_TAG_TYPE_RATE_LIMIT:
2356 		error = cxgbe_rate_tag_alloc(ifp, params, pt);
2357 		break;
2358 #endif
2359 #ifdef KERN_TLS
2360 	case IF_SND_TAG_TYPE_TLS:
2361 		error = cxgbe_tls_tag_alloc(ifp, params, pt);
2362 		break;
2363 #endif
2364 	default:
2365 		error = EOPNOTSUPP;
2366 	}
2367 	if (error == 0)
2368 		MPASS(mst_to_cst(*pt)->type == params->hdr.type);
2369 	return (error);
2370 }
2371 
2372 static int
2373 cxgbe_snd_tag_modify(struct m_snd_tag *mst,
2374     union if_snd_tag_modify_params *params)
2375 {
2376 	struct cxgbe_snd_tag *cst;
2377 
2378 	cst = mst_to_cst(mst);
2379 	switch (cst->type) {
2380 #ifdef RATELIMIT
2381 	case IF_SND_TAG_TYPE_RATE_LIMIT:
2382 		return (cxgbe_rate_tag_modify(mst, params));
2383 #endif
2384 	default:
2385 		return (EOPNOTSUPP);
2386 	}
2387 }
2388 
2389 static int
2390 cxgbe_snd_tag_query(struct m_snd_tag *mst,
2391     union if_snd_tag_query_params *params)
2392 {
2393 	struct cxgbe_snd_tag *cst;
2394 
2395 	cst = mst_to_cst(mst);
2396 	switch (cst->type) {
2397 #ifdef RATELIMIT
2398 	case IF_SND_TAG_TYPE_RATE_LIMIT:
2399 		return (cxgbe_rate_tag_query(mst, params));
2400 #endif
2401 	default:
2402 		return (EOPNOTSUPP);
2403 	}
2404 }
2405 
2406 static void
2407 cxgbe_snd_tag_free(struct m_snd_tag *mst)
2408 {
2409 	struct cxgbe_snd_tag *cst;
2410 
2411 	cst = mst_to_cst(mst);
2412 	switch (cst->type) {
2413 #ifdef RATELIMIT
2414 	case IF_SND_TAG_TYPE_RATE_LIMIT:
2415 		cxgbe_rate_tag_free(mst);
2416 		return;
2417 #endif
2418 #ifdef KERN_TLS
2419 	case IF_SND_TAG_TYPE_TLS:
2420 		cxgbe_tls_tag_free(mst);
2421 		return;
2422 #endif
2423 	default:
2424 		panic("shouldn't get here");
2425 	}
2426 }
2427 #endif
2428 
2429 /*
2430  * The kernel picks a media from the list we had provided but we still validate
2431  * the requeste.
2432  */
2433 int
2434 cxgbe_media_change(struct ifnet *ifp)
2435 {
2436 	struct vi_info *vi = ifp->if_softc;
2437 	struct port_info *pi = vi->pi;
2438 	struct ifmedia *ifm = &pi->media;
2439 	struct link_config *lc = &pi->link_cfg;
2440 	struct adapter *sc = pi->adapter;
2441 	int rc;
2442 
2443 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mec");
2444 	if (rc != 0)
2445 		return (rc);
2446 	PORT_LOCK(pi);
2447 	if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
2448 		/* ifconfig .. media autoselect */
2449 		if (!(lc->pcaps & FW_PORT_CAP32_ANEG)) {
2450 			rc = ENOTSUP; /* AN not supported by transceiver */
2451 			goto done;
2452 		}
2453 		lc->requested_aneg = AUTONEG_ENABLE;
2454 		lc->requested_speed = 0;
2455 		lc->requested_fc |= PAUSE_AUTONEG;
2456 	} else {
2457 		lc->requested_aneg = AUTONEG_DISABLE;
2458 		lc->requested_speed =
2459 		    ifmedia_baudrate(ifm->ifm_media) / 1000000;
2460 		lc->requested_fc = 0;
2461 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
2462 			lc->requested_fc |= PAUSE_RX;
2463 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
2464 			lc->requested_fc |= PAUSE_TX;
2465 	}
2466 	if (pi->up_vis > 0) {
2467 		fixup_link_config(pi);
2468 		rc = apply_link_config(pi);
2469 	}
2470 done:
2471 	PORT_UNLOCK(pi);
2472 	end_synchronized_op(sc, 0);
2473 	return (rc);
2474 }
2475 
2476 /*
2477  * Base media word (without ETHER, pause, link active, etc.) for the port at the
2478  * given speed.
2479  */
2480 static int
2481 port_mword(struct port_info *pi, uint32_t speed)
2482 {
2483 
2484 	MPASS(speed & M_FW_PORT_CAP32_SPEED);
2485 	MPASS(powerof2(speed));
2486 
2487 	switch(pi->port_type) {
2488 	case FW_PORT_TYPE_BT_SGMII:
2489 	case FW_PORT_TYPE_BT_XFI:
2490 	case FW_PORT_TYPE_BT_XAUI:
2491 		/* BaseT */
2492 		switch (speed) {
2493 		case FW_PORT_CAP32_SPEED_100M:
2494 			return (IFM_100_T);
2495 		case FW_PORT_CAP32_SPEED_1G:
2496 			return (IFM_1000_T);
2497 		case FW_PORT_CAP32_SPEED_10G:
2498 			return (IFM_10G_T);
2499 		}
2500 		break;
2501 	case FW_PORT_TYPE_KX4:
2502 		if (speed == FW_PORT_CAP32_SPEED_10G)
2503 			return (IFM_10G_KX4);
2504 		break;
2505 	case FW_PORT_TYPE_CX4:
2506 		if (speed == FW_PORT_CAP32_SPEED_10G)
2507 			return (IFM_10G_CX4);
2508 		break;
2509 	case FW_PORT_TYPE_KX:
2510 		if (speed == FW_PORT_CAP32_SPEED_1G)
2511 			return (IFM_1000_KX);
2512 		break;
2513 	case FW_PORT_TYPE_KR:
2514 	case FW_PORT_TYPE_BP_AP:
2515 	case FW_PORT_TYPE_BP4_AP:
2516 	case FW_PORT_TYPE_BP40_BA:
2517 	case FW_PORT_TYPE_KR4_100G:
2518 	case FW_PORT_TYPE_KR_SFP28:
2519 	case FW_PORT_TYPE_KR_XLAUI:
2520 		switch (speed) {
2521 		case FW_PORT_CAP32_SPEED_1G:
2522 			return (IFM_1000_KX);
2523 		case FW_PORT_CAP32_SPEED_10G:
2524 			return (IFM_10G_KR);
2525 		case FW_PORT_CAP32_SPEED_25G:
2526 			return (IFM_25G_KR);
2527 		case FW_PORT_CAP32_SPEED_40G:
2528 			return (IFM_40G_KR4);
2529 		case FW_PORT_CAP32_SPEED_50G:
2530 			return (IFM_50G_KR2);
2531 		case FW_PORT_CAP32_SPEED_100G:
2532 			return (IFM_100G_KR4);
2533 		}
2534 		break;
2535 	case FW_PORT_TYPE_FIBER_XFI:
2536 	case FW_PORT_TYPE_FIBER_XAUI:
2537 	case FW_PORT_TYPE_SFP:
2538 	case FW_PORT_TYPE_QSFP_10G:
2539 	case FW_PORT_TYPE_QSA:
2540 	case FW_PORT_TYPE_QSFP:
2541 	case FW_PORT_TYPE_CR4_QSFP:
2542 	case FW_PORT_TYPE_CR_QSFP:
2543 	case FW_PORT_TYPE_CR2_QSFP:
2544 	case FW_PORT_TYPE_SFP28:
2545 		/* Pluggable transceiver */
2546 		switch (pi->mod_type) {
2547 		case FW_PORT_MOD_TYPE_LR:
2548 			switch (speed) {
2549 			case FW_PORT_CAP32_SPEED_1G:
2550 				return (IFM_1000_LX);
2551 			case FW_PORT_CAP32_SPEED_10G:
2552 				return (IFM_10G_LR);
2553 			case FW_PORT_CAP32_SPEED_25G:
2554 				return (IFM_25G_LR);
2555 			case FW_PORT_CAP32_SPEED_40G:
2556 				return (IFM_40G_LR4);
2557 			case FW_PORT_CAP32_SPEED_50G:
2558 				return (IFM_50G_LR2);
2559 			case FW_PORT_CAP32_SPEED_100G:
2560 				return (IFM_100G_LR4);
2561 			}
2562 			break;
2563 		case FW_PORT_MOD_TYPE_SR:
2564 			switch (speed) {
2565 			case FW_PORT_CAP32_SPEED_1G:
2566 				return (IFM_1000_SX);
2567 			case FW_PORT_CAP32_SPEED_10G:
2568 				return (IFM_10G_SR);
2569 			case FW_PORT_CAP32_SPEED_25G:
2570 				return (IFM_25G_SR);
2571 			case FW_PORT_CAP32_SPEED_40G:
2572 				return (IFM_40G_SR4);
2573 			case FW_PORT_CAP32_SPEED_50G:
2574 				return (IFM_50G_SR2);
2575 			case FW_PORT_CAP32_SPEED_100G:
2576 				return (IFM_100G_SR4);
2577 			}
2578 			break;
2579 		case FW_PORT_MOD_TYPE_ER:
2580 			if (speed == FW_PORT_CAP32_SPEED_10G)
2581 				return (IFM_10G_ER);
2582 			break;
2583 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2584 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2585 			switch (speed) {
2586 			case FW_PORT_CAP32_SPEED_1G:
2587 				return (IFM_1000_CX);
2588 			case FW_PORT_CAP32_SPEED_10G:
2589 				return (IFM_10G_TWINAX);
2590 			case FW_PORT_CAP32_SPEED_25G:
2591 				return (IFM_25G_CR);
2592 			case FW_PORT_CAP32_SPEED_40G:
2593 				return (IFM_40G_CR4);
2594 			case FW_PORT_CAP32_SPEED_50G:
2595 				return (IFM_50G_CR2);
2596 			case FW_PORT_CAP32_SPEED_100G:
2597 				return (IFM_100G_CR4);
2598 			}
2599 			break;
2600 		case FW_PORT_MOD_TYPE_LRM:
2601 			if (speed == FW_PORT_CAP32_SPEED_10G)
2602 				return (IFM_10G_LRM);
2603 			break;
2604 		case FW_PORT_MOD_TYPE_NA:
2605 			MPASS(0);	/* Not pluggable? */
2606 			/* fall throough */
2607 		case FW_PORT_MOD_TYPE_ERROR:
2608 		case FW_PORT_MOD_TYPE_UNKNOWN:
2609 		case FW_PORT_MOD_TYPE_NOTSUPPORTED:
2610 			break;
2611 		case FW_PORT_MOD_TYPE_NONE:
2612 			return (IFM_NONE);
2613 		}
2614 		break;
2615 	case FW_PORT_TYPE_NONE:
2616 		return (IFM_NONE);
2617 	}
2618 
2619 	return (IFM_UNKNOWN);
2620 }
2621 
2622 void
2623 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2624 {
2625 	struct vi_info *vi = ifp->if_softc;
2626 	struct port_info *pi = vi->pi;
2627 	struct adapter *sc = pi->adapter;
2628 	struct link_config *lc = &pi->link_cfg;
2629 
2630 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4med") != 0)
2631 		return;
2632 	PORT_LOCK(pi);
2633 
2634 	if (pi->up_vis == 0) {
2635 		/*
2636 		 * If all the interfaces are administratively down the firmware
2637 		 * does not report transceiver changes.  Refresh port info here
2638 		 * so that ifconfig displays accurate ifmedia at all times.
2639 		 * This is the only reason we have a synchronized op in this
2640 		 * function.  Just PORT_LOCK would have been enough otherwise.
2641 		 */
2642 		t4_update_port_info(pi);
2643 		build_medialist(pi);
2644 	}
2645 
2646 	/* ifm_status */
2647 	ifmr->ifm_status = IFM_AVALID;
2648 	if (lc->link_ok == false)
2649 		goto done;
2650 	ifmr->ifm_status |= IFM_ACTIVE;
2651 
2652 	/* ifm_active */
2653 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2654 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2655 	if (lc->fc & PAUSE_RX)
2656 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2657 	if (lc->fc & PAUSE_TX)
2658 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2659 	ifmr->ifm_active |= port_mword(pi, speed_to_fwcap(lc->speed));
2660 done:
2661 	PORT_UNLOCK(pi);
2662 	end_synchronized_op(sc, 0);
2663 }
2664 
2665 static int
2666 vcxgbe_probe(device_t dev)
2667 {
2668 	char buf[128];
2669 	struct vi_info *vi = device_get_softc(dev);
2670 
2671 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2672 	    vi - vi->pi->vi);
2673 	device_set_desc_copy(dev, buf);
2674 
2675 	return (BUS_PROBE_DEFAULT);
2676 }
2677 
2678 static int
2679 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2680 {
2681 	int func, index, rc;
2682 	uint32_t param, val;
2683 
2684 	ASSERT_SYNCHRONIZED_OP(sc);
2685 
2686 	index = vi - pi->vi;
2687 	MPASS(index > 0);	/* This function deals with _extra_ VIs only */
2688 	KASSERT(index < nitems(vi_mac_funcs),
2689 	    ("%s: VI %s doesn't have a MAC func", __func__,
2690 	    device_get_nameunit(vi->dev)));
2691 	func = vi_mac_funcs[index];
2692 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2693 	    vi->hw_addr, &vi->rss_size, &vi->vfvld, &vi->vin, func, 0);
2694 	if (rc < 0) {
2695 		device_printf(vi->dev, "failed to allocate virtual interface %d"
2696 		    "for port %d: %d\n", index, pi->port_id, -rc);
2697 		return (-rc);
2698 	}
2699 	vi->viid = rc;
2700 
2701 	if (vi->rss_size == 1) {
2702 		/*
2703 		 * This VI didn't get a slice of the RSS table.  Reduce the
2704 		 * number of VIs being created (hw.cxgbe.num_vis) or modify the
2705 		 * configuration file (nvi, rssnvi for this PF) if this is a
2706 		 * problem.
2707 		 */
2708 		device_printf(vi->dev, "RSS table not available.\n");
2709 		vi->rss_base = 0xffff;
2710 
2711 		return (0);
2712 	}
2713 
2714 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2715 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2716 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2717 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2718 	if (rc)
2719 		vi->rss_base = 0xffff;
2720 	else {
2721 		MPASS((val >> 16) == vi->rss_size);
2722 		vi->rss_base = val & 0xffff;
2723 	}
2724 
2725 	return (0);
2726 }
2727 
2728 static int
2729 vcxgbe_attach(device_t dev)
2730 {
2731 	struct vi_info *vi;
2732 	struct port_info *pi;
2733 	struct adapter *sc;
2734 	int rc;
2735 
2736 	vi = device_get_softc(dev);
2737 	pi = vi->pi;
2738 	sc = pi->adapter;
2739 
2740 	rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2741 	if (rc)
2742 		return (rc);
2743 	rc = alloc_extra_vi(sc, pi, vi);
2744 	end_synchronized_op(sc, 0);
2745 	if (rc)
2746 		return (rc);
2747 
2748 	rc = cxgbe_vi_attach(dev, vi);
2749 	if (rc) {
2750 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2751 		return (rc);
2752 	}
2753 	return (0);
2754 }
2755 
2756 static int
2757 vcxgbe_detach(device_t dev)
2758 {
2759 	struct vi_info *vi;
2760 	struct adapter *sc;
2761 
2762 	vi = device_get_softc(dev);
2763 	sc = vi->pi->adapter;
2764 
2765 	doom_vi(sc, vi);
2766 
2767 	cxgbe_vi_detach(vi);
2768 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2769 
2770 	end_synchronized_op(sc, 0);
2771 
2772 	return (0);
2773 }
2774 
2775 static struct callout fatal_callout;
2776 
2777 static void
2778 delayed_panic(void *arg)
2779 {
2780 	struct adapter *sc = arg;
2781 
2782 	panic("%s: panic on fatal error", device_get_nameunit(sc->dev));
2783 }
2784 
2785 void
2786 t4_fatal_err(struct adapter *sc, bool fw_error)
2787 {
2788 
2789 	t4_shutdown_adapter(sc);
2790 	log(LOG_ALERT, "%s: encountered fatal error, adapter stopped.\n",
2791 	    device_get_nameunit(sc->dev));
2792 	if (fw_error) {
2793 		ASSERT_SYNCHRONIZED_OP(sc);
2794 		sc->flags |= ADAP_ERR;
2795 	} else {
2796 		ADAPTER_LOCK(sc);
2797 		sc->flags |= ADAP_ERR;
2798 		ADAPTER_UNLOCK(sc);
2799 	}
2800 #ifdef TCP_OFFLOAD
2801 	taskqueue_enqueue(taskqueue_thread, &sc->async_event_task);
2802 #endif
2803 
2804 	if (t4_panic_on_fatal_err) {
2805 		log(LOG_ALERT, "%s: panic on fatal error after 30s",
2806 		    device_get_nameunit(sc->dev));
2807 		callout_reset(&fatal_callout, hz * 30, delayed_panic, sc);
2808 	}
2809 }
2810 
2811 void
2812 t4_add_adapter(struct adapter *sc)
2813 {
2814 	sx_xlock(&t4_list_lock);
2815 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2816 	sx_xunlock(&t4_list_lock);
2817 }
2818 
2819 int
2820 t4_map_bars_0_and_4(struct adapter *sc)
2821 {
2822 	sc->regs_rid = PCIR_BAR(0);
2823 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2824 	    &sc->regs_rid, RF_ACTIVE);
2825 	if (sc->regs_res == NULL) {
2826 		device_printf(sc->dev, "cannot map registers.\n");
2827 		return (ENXIO);
2828 	}
2829 	sc->bt = rman_get_bustag(sc->regs_res);
2830 	sc->bh = rman_get_bushandle(sc->regs_res);
2831 	sc->mmio_len = rman_get_size(sc->regs_res);
2832 	setbit(&sc->doorbells, DOORBELL_KDB);
2833 
2834 	sc->msix_rid = PCIR_BAR(4);
2835 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2836 	    &sc->msix_rid, RF_ACTIVE);
2837 	if (sc->msix_res == NULL) {
2838 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2839 		return (ENXIO);
2840 	}
2841 
2842 	return (0);
2843 }
2844 
2845 int
2846 t4_map_bar_2(struct adapter *sc)
2847 {
2848 
2849 	/*
2850 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2851 	 * to map it if RDMA is disabled.
2852 	 */
2853 	if (is_t4(sc) && sc->rdmacaps == 0)
2854 		return (0);
2855 
2856 	sc->udbs_rid = PCIR_BAR(2);
2857 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2858 	    &sc->udbs_rid, RF_ACTIVE);
2859 	if (sc->udbs_res == NULL) {
2860 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2861 		return (ENXIO);
2862 	}
2863 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2864 
2865 	if (chip_id(sc) >= CHELSIO_T5) {
2866 		setbit(&sc->doorbells, DOORBELL_UDB);
2867 #if defined(__i386__) || defined(__amd64__)
2868 		if (t5_write_combine) {
2869 			int rc, mode;
2870 
2871 			/*
2872 			 * Enable write combining on BAR2.  This is the
2873 			 * userspace doorbell BAR and is split into 128B
2874 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2875 			 * with an egress queue.  The first 64B has the doorbell
2876 			 * and the second 64B can be used to submit a tx work
2877 			 * request with an implicit doorbell.
2878 			 */
2879 
2880 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2881 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2882 			if (rc == 0) {
2883 				clrbit(&sc->doorbells, DOORBELL_UDB);
2884 				setbit(&sc->doorbells, DOORBELL_WCWR);
2885 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2886 			} else {
2887 				device_printf(sc->dev,
2888 				    "couldn't enable write combining: %d\n",
2889 				    rc);
2890 			}
2891 
2892 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2893 			t4_write_reg(sc, A_SGE_STAT_CFG,
2894 			    V_STATSOURCE_T5(7) | mode);
2895 		}
2896 #endif
2897 	}
2898 	sc->iwt.wc_en = isset(&sc->doorbells, DOORBELL_UDBWC) ? 1 : 0;
2899 
2900 	return (0);
2901 }
2902 
2903 struct memwin_init {
2904 	uint32_t base;
2905 	uint32_t aperture;
2906 };
2907 
2908 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2909 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2910 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2911 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2912 };
2913 
2914 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2915 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2916 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2917 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2918 };
2919 
2920 static void
2921 setup_memwin(struct adapter *sc)
2922 {
2923 	const struct memwin_init *mw_init;
2924 	struct memwin *mw;
2925 	int i;
2926 	uint32_t bar0;
2927 
2928 	if (is_t4(sc)) {
2929 		/*
2930 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2931 		 * mechanism.  Works from within PCI passthrough environments
2932 		 * too, where rman_get_start() can return a different value.  We
2933 		 * need to program the T4 memory window decoders with the actual
2934 		 * addresses that will be coming across the PCIe link.
2935 		 */
2936 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2937 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2938 
2939 		mw_init = &t4_memwin[0];
2940 	} else {
2941 		/* T5+ use the relative offset inside the PCIe BAR */
2942 		bar0 = 0;
2943 
2944 		mw_init = &t5_memwin[0];
2945 	}
2946 
2947 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2948 		rw_init(&mw->mw_lock, "memory window access");
2949 		mw->mw_base = mw_init->base;
2950 		mw->mw_aperture = mw_init->aperture;
2951 		mw->mw_curpos = 0;
2952 		t4_write_reg(sc,
2953 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2954 		    (mw->mw_base + bar0) | V_BIR(0) |
2955 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2956 		rw_wlock(&mw->mw_lock);
2957 		position_memwin(sc, i, 0);
2958 		rw_wunlock(&mw->mw_lock);
2959 	}
2960 
2961 	/* flush */
2962 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2963 }
2964 
2965 /*
2966  * Positions the memory window at the given address in the card's address space.
2967  * There are some alignment requirements and the actual position may be at an
2968  * address prior to the requested address.  mw->mw_curpos always has the actual
2969  * position of the window.
2970  */
2971 static void
2972 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2973 {
2974 	struct memwin *mw;
2975 	uint32_t pf;
2976 	uint32_t reg;
2977 
2978 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2979 	mw = &sc->memwin[idx];
2980 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2981 
2982 	if (is_t4(sc)) {
2983 		pf = 0;
2984 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2985 	} else {
2986 		pf = V_PFNUM(sc->pf);
2987 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2988 	}
2989 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2990 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2991 	t4_read_reg(sc, reg);	/* flush */
2992 }
2993 
2994 int
2995 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2996     int len, int rw)
2997 {
2998 	struct memwin *mw;
2999 	uint32_t mw_end, v;
3000 
3001 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
3002 
3003 	/* Memory can only be accessed in naturally aligned 4 byte units */
3004 	if (addr & 3 || len & 3 || len <= 0)
3005 		return (EINVAL);
3006 
3007 	mw = &sc->memwin[idx];
3008 	while (len > 0) {
3009 		rw_rlock(&mw->mw_lock);
3010 		mw_end = mw->mw_curpos + mw->mw_aperture;
3011 		if (addr >= mw_end || addr < mw->mw_curpos) {
3012 			/* Will need to reposition the window */
3013 			if (!rw_try_upgrade(&mw->mw_lock)) {
3014 				rw_runlock(&mw->mw_lock);
3015 				rw_wlock(&mw->mw_lock);
3016 			}
3017 			rw_assert(&mw->mw_lock, RA_WLOCKED);
3018 			position_memwin(sc, idx, addr);
3019 			rw_downgrade(&mw->mw_lock);
3020 			mw_end = mw->mw_curpos + mw->mw_aperture;
3021 		}
3022 		rw_assert(&mw->mw_lock, RA_RLOCKED);
3023 		while (addr < mw_end && len > 0) {
3024 			if (rw == 0) {
3025 				v = t4_read_reg(sc, mw->mw_base + addr -
3026 				    mw->mw_curpos);
3027 				*val++ = le32toh(v);
3028 			} else {
3029 				v = *val++;
3030 				t4_write_reg(sc, mw->mw_base + addr -
3031 				    mw->mw_curpos, htole32(v));
3032 			}
3033 			addr += 4;
3034 			len -= 4;
3035 		}
3036 		rw_runlock(&mw->mw_lock);
3037 	}
3038 
3039 	return (0);
3040 }
3041 
3042 static void
3043 t4_init_atid_table(struct adapter *sc)
3044 {
3045 	struct tid_info *t;
3046 	int i;
3047 
3048 	t = &sc->tids;
3049 	if (t->natids == 0)
3050 		return;
3051 
3052 	MPASS(t->atid_tab == NULL);
3053 
3054 	t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
3055 	    M_ZERO | M_WAITOK);
3056 	mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
3057 	t->afree = t->atid_tab;
3058 	t->atids_in_use = 0;
3059 	for (i = 1; i < t->natids; i++)
3060 		t->atid_tab[i - 1].next = &t->atid_tab[i];
3061 	t->atid_tab[t->natids - 1].next = NULL;
3062 }
3063 
3064 static void
3065 t4_free_atid_table(struct adapter *sc)
3066 {
3067 	struct tid_info *t;
3068 
3069 	t = &sc->tids;
3070 
3071 	KASSERT(t->atids_in_use == 0,
3072 	    ("%s: %d atids still in use.", __func__, t->atids_in_use));
3073 
3074 	if (mtx_initialized(&t->atid_lock))
3075 		mtx_destroy(&t->atid_lock);
3076 	free(t->atid_tab, M_CXGBE);
3077 	t->atid_tab = NULL;
3078 }
3079 
3080 int
3081 alloc_atid(struct adapter *sc, void *ctx)
3082 {
3083 	struct tid_info *t = &sc->tids;
3084 	int atid = -1;
3085 
3086 	mtx_lock(&t->atid_lock);
3087 	if (t->afree) {
3088 		union aopen_entry *p = t->afree;
3089 
3090 		atid = p - t->atid_tab;
3091 		MPASS(atid <= M_TID_TID);
3092 		t->afree = p->next;
3093 		p->data = ctx;
3094 		t->atids_in_use++;
3095 	}
3096 	mtx_unlock(&t->atid_lock);
3097 	return (atid);
3098 }
3099 
3100 void *
3101 lookup_atid(struct adapter *sc, int atid)
3102 {
3103 	struct tid_info *t = &sc->tids;
3104 
3105 	return (t->atid_tab[atid].data);
3106 }
3107 
3108 void
3109 free_atid(struct adapter *sc, int atid)
3110 {
3111 	struct tid_info *t = &sc->tids;
3112 	union aopen_entry *p = &t->atid_tab[atid];
3113 
3114 	mtx_lock(&t->atid_lock);
3115 	p->next = t->afree;
3116 	t->afree = p;
3117 	t->atids_in_use--;
3118 	mtx_unlock(&t->atid_lock);
3119 }
3120 
3121 static void
3122 queue_tid_release(struct adapter *sc, int tid)
3123 {
3124 
3125 	CXGBE_UNIMPLEMENTED("deferred tid release");
3126 }
3127 
3128 void
3129 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
3130 {
3131 	struct wrqe *wr;
3132 	struct cpl_tid_release *req;
3133 
3134 	wr = alloc_wrqe(sizeof(*req), ctrlq);
3135 	if (wr == NULL) {
3136 		queue_tid_release(sc, tid);	/* defer */
3137 		return;
3138 	}
3139 	req = wrtod(wr);
3140 
3141 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
3142 
3143 	t4_wrq_tx(sc, wr);
3144 }
3145 
3146 static int
3147 t4_range_cmp(const void *a, const void *b)
3148 {
3149 	return ((const struct t4_range *)a)->start -
3150 	       ((const struct t4_range *)b)->start;
3151 }
3152 
3153 /*
3154  * Verify that the memory range specified by the addr/len pair is valid within
3155  * the card's address space.
3156  */
3157 static int
3158 validate_mem_range(struct adapter *sc, uint32_t addr, uint32_t len)
3159 {
3160 	struct t4_range mem_ranges[4], *r, *next;
3161 	uint32_t em, addr_len;
3162 	int i, n, remaining;
3163 
3164 	/* Memory can only be accessed in naturally aligned 4 byte units */
3165 	if (addr & 3 || len & 3 || len == 0)
3166 		return (EINVAL);
3167 
3168 	/* Enabled memories */
3169 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
3170 
3171 	r = &mem_ranges[0];
3172 	n = 0;
3173 	bzero(r, sizeof(mem_ranges));
3174 	if (em & F_EDRAM0_ENABLE) {
3175 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
3176 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
3177 		if (r->size > 0) {
3178 			r->start = G_EDRAM0_BASE(addr_len) << 20;
3179 			if (addr >= r->start &&
3180 			    addr + len <= r->start + r->size)
3181 				return (0);
3182 			r++;
3183 			n++;
3184 		}
3185 	}
3186 	if (em & F_EDRAM1_ENABLE) {
3187 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
3188 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
3189 		if (r->size > 0) {
3190 			r->start = G_EDRAM1_BASE(addr_len) << 20;
3191 			if (addr >= r->start &&
3192 			    addr + len <= r->start + r->size)
3193 				return (0);
3194 			r++;
3195 			n++;
3196 		}
3197 	}
3198 	if (em & F_EXT_MEM_ENABLE) {
3199 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
3200 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
3201 		if (r->size > 0) {
3202 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
3203 			if (addr >= r->start &&
3204 			    addr + len <= r->start + r->size)
3205 				return (0);
3206 			r++;
3207 			n++;
3208 		}
3209 	}
3210 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
3211 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
3212 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
3213 		if (r->size > 0) {
3214 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
3215 			if (addr >= r->start &&
3216 			    addr + len <= r->start + r->size)
3217 				return (0);
3218 			r++;
3219 			n++;
3220 		}
3221 	}
3222 	MPASS(n <= nitems(mem_ranges));
3223 
3224 	if (n > 1) {
3225 		/* Sort and merge the ranges. */
3226 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
3227 
3228 		/* Start from index 0 and examine the next n - 1 entries. */
3229 		r = &mem_ranges[0];
3230 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
3231 
3232 			MPASS(r->size > 0);	/* r is a valid entry. */
3233 			next = r + 1;
3234 			MPASS(next->size > 0);	/* and so is the next one. */
3235 
3236 			while (r->start + r->size >= next->start) {
3237 				/* Merge the next one into the current entry. */
3238 				r->size = max(r->start + r->size,
3239 				    next->start + next->size) - r->start;
3240 				n--;	/* One fewer entry in total. */
3241 				if (--remaining == 0)
3242 					goto done;	/* short circuit */
3243 				next++;
3244 			}
3245 			if (next != r + 1) {
3246 				/*
3247 				 * Some entries were merged into r and next
3248 				 * points to the first valid entry that couldn't
3249 				 * be merged.
3250 				 */
3251 				MPASS(next->size > 0);	/* must be valid */
3252 				memcpy(r + 1, next, remaining * sizeof(*r));
3253 #ifdef INVARIANTS
3254 				/*
3255 				 * This so that the foo->size assertion in the
3256 				 * next iteration of the loop do the right
3257 				 * thing for entries that were pulled up and are
3258 				 * no longer valid.
3259 				 */
3260 				MPASS(n < nitems(mem_ranges));
3261 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
3262 				    sizeof(struct t4_range));
3263 #endif
3264 			}
3265 		}
3266 done:
3267 		/* Done merging the ranges. */
3268 		MPASS(n > 0);
3269 		r = &mem_ranges[0];
3270 		for (i = 0; i < n; i++, r++) {
3271 			if (addr >= r->start &&
3272 			    addr + len <= r->start + r->size)
3273 				return (0);
3274 		}
3275 	}
3276 
3277 	return (EFAULT);
3278 }
3279 
3280 static int
3281 fwmtype_to_hwmtype(int mtype)
3282 {
3283 
3284 	switch (mtype) {
3285 	case FW_MEMTYPE_EDC0:
3286 		return (MEM_EDC0);
3287 	case FW_MEMTYPE_EDC1:
3288 		return (MEM_EDC1);
3289 	case FW_MEMTYPE_EXTMEM:
3290 		return (MEM_MC0);
3291 	case FW_MEMTYPE_EXTMEM1:
3292 		return (MEM_MC1);
3293 	default:
3294 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
3295 	}
3296 }
3297 
3298 /*
3299  * Verify that the memory range specified by the memtype/offset/len pair is
3300  * valid and lies entirely within the memtype specified.  The global address of
3301  * the start of the range is returned in addr.
3302  */
3303 static int
3304 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, uint32_t len,
3305     uint32_t *addr)
3306 {
3307 	uint32_t em, addr_len, maddr;
3308 
3309 	/* Memory can only be accessed in naturally aligned 4 byte units */
3310 	if (off & 3 || len & 3 || len == 0)
3311 		return (EINVAL);
3312 
3313 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
3314 	switch (fwmtype_to_hwmtype(mtype)) {
3315 	case MEM_EDC0:
3316 		if (!(em & F_EDRAM0_ENABLE))
3317 			return (EINVAL);
3318 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
3319 		maddr = G_EDRAM0_BASE(addr_len) << 20;
3320 		break;
3321 	case MEM_EDC1:
3322 		if (!(em & F_EDRAM1_ENABLE))
3323 			return (EINVAL);
3324 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
3325 		maddr = G_EDRAM1_BASE(addr_len) << 20;
3326 		break;
3327 	case MEM_MC:
3328 		if (!(em & F_EXT_MEM_ENABLE))
3329 			return (EINVAL);
3330 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
3331 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
3332 		break;
3333 	case MEM_MC1:
3334 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
3335 			return (EINVAL);
3336 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
3337 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
3338 		break;
3339 	default:
3340 		return (EINVAL);
3341 	}
3342 
3343 	*addr = maddr + off;	/* global address */
3344 	return (validate_mem_range(sc, *addr, len));
3345 }
3346 
3347 static int
3348 fixup_devlog_params(struct adapter *sc)
3349 {
3350 	struct devlog_params *dparams = &sc->params.devlog;
3351 	int rc;
3352 
3353 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
3354 	    dparams->size, &dparams->addr);
3355 
3356 	return (rc);
3357 }
3358 
3359 static void
3360 update_nirq(struct intrs_and_queues *iaq, int nports)
3361 {
3362 
3363 	iaq->nirq = T4_EXTRA_INTR;
3364 	iaq->nirq += nports * max(iaq->nrxq, iaq->nnmrxq);
3365 	iaq->nirq += nports * iaq->nofldrxq;
3366 	iaq->nirq += nports * (iaq->num_vis - 1) *
3367 	    max(iaq->nrxq_vi, iaq->nnmrxq_vi);
3368 	iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
3369 }
3370 
3371 /*
3372  * Adjust requirements to fit the number of interrupts available.
3373  */
3374 static void
3375 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
3376     int navail)
3377 {
3378 	int old_nirq;
3379 	const int nports = sc->params.nports;
3380 
3381 	MPASS(nports > 0);
3382 	MPASS(navail > 0);
3383 
3384 	bzero(iaq, sizeof(*iaq));
3385 	iaq->intr_type = itype;
3386 	iaq->num_vis = t4_num_vis;
3387 	iaq->ntxq = t4_ntxq;
3388 	iaq->ntxq_vi = t4_ntxq_vi;
3389 	iaq->nrxq = t4_nrxq;
3390 	iaq->nrxq_vi = t4_nrxq_vi;
3391 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
3392 	if (is_offload(sc) || is_ethoffload(sc)) {
3393 		iaq->nofldtxq = t4_nofldtxq;
3394 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
3395 	}
3396 #endif
3397 #ifdef TCP_OFFLOAD
3398 	if (is_offload(sc)) {
3399 		iaq->nofldrxq = t4_nofldrxq;
3400 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
3401 	}
3402 #endif
3403 #ifdef DEV_NETMAP
3404 	if (t4_native_netmap & NN_MAIN_VI) {
3405 		iaq->nnmtxq = t4_nnmtxq;
3406 		iaq->nnmrxq = t4_nnmrxq;
3407 	}
3408 	if (t4_native_netmap & NN_EXTRA_VI) {
3409 		iaq->nnmtxq_vi = t4_nnmtxq_vi;
3410 		iaq->nnmrxq_vi = t4_nnmrxq_vi;
3411 	}
3412 #endif
3413 
3414 	update_nirq(iaq, nports);
3415 	if (iaq->nirq <= navail &&
3416 	    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3417 		/*
3418 		 * This is the normal case -- there are enough interrupts for
3419 		 * everything.
3420 		 */
3421 		goto done;
3422 	}
3423 
3424 	/*
3425 	 * If extra VIs have been configured try reducing their count and see if
3426 	 * that works.
3427 	 */
3428 	while (iaq->num_vis > 1) {
3429 		iaq->num_vis--;
3430 		update_nirq(iaq, nports);
3431 		if (iaq->nirq <= navail &&
3432 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3433 			device_printf(sc->dev, "virtual interfaces per port "
3434 			    "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
3435 			    "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
3436 			    "itype %d, navail %u, nirq %d.\n",
3437 			    iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
3438 			    iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
3439 			    itype, navail, iaq->nirq);
3440 			goto done;
3441 		}
3442 	}
3443 
3444 	/*
3445 	 * Extra VIs will not be created.  Log a message if they were requested.
3446 	 */
3447 	MPASS(iaq->num_vis == 1);
3448 	iaq->ntxq_vi = iaq->nrxq_vi = 0;
3449 	iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
3450 	iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
3451 	if (iaq->num_vis != t4_num_vis) {
3452 		device_printf(sc->dev, "extra virtual interfaces disabled.  "
3453 		    "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
3454 		    "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
3455 		    iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
3456 		    iaq->nnmrxq_vi, itype, navail, iaq->nirq);
3457 	}
3458 
3459 	/*
3460 	 * Keep reducing the number of NIC rx queues to the next lower power of
3461 	 * 2 (for even RSS distribution) and halving the TOE rx queues and see
3462 	 * if that works.
3463 	 */
3464 	do {
3465 		if (iaq->nrxq > 1) {
3466 			do {
3467 				iaq->nrxq--;
3468 			} while (!powerof2(iaq->nrxq));
3469 			if (iaq->nnmrxq > iaq->nrxq)
3470 				iaq->nnmrxq = iaq->nrxq;
3471 		}
3472 		if (iaq->nofldrxq > 1)
3473 			iaq->nofldrxq >>= 1;
3474 
3475 		old_nirq = iaq->nirq;
3476 		update_nirq(iaq, nports);
3477 		if (iaq->nirq <= navail &&
3478 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3479 			device_printf(sc->dev, "running with reduced number of "
3480 			    "rx queues because of shortage of interrupts.  "
3481 			    "nrxq=%u, nofldrxq=%u.  "
3482 			    "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
3483 			    iaq->nofldrxq, itype, navail, iaq->nirq);
3484 			goto done;
3485 		}
3486 	} while (old_nirq != iaq->nirq);
3487 
3488 	/* One interrupt for everything.  Ugh. */
3489 	device_printf(sc->dev, "running with minimal number of queues.  "
3490 	    "itype %d, navail %u.\n", itype, navail);
3491 	iaq->nirq = 1;
3492 	iaq->nrxq = 1;
3493 	iaq->ntxq = 1;
3494 	if (iaq->nofldrxq > 0) {
3495 		iaq->nofldrxq = 1;
3496 		iaq->nofldtxq = 1;
3497 	}
3498 	iaq->nnmtxq = 0;
3499 	iaq->nnmrxq = 0;
3500 done:
3501 	MPASS(iaq->num_vis > 0);
3502 	if (iaq->num_vis > 1) {
3503 		MPASS(iaq->nrxq_vi > 0);
3504 		MPASS(iaq->ntxq_vi > 0);
3505 	}
3506 	MPASS(iaq->nirq > 0);
3507 	MPASS(iaq->nrxq > 0);
3508 	MPASS(iaq->ntxq > 0);
3509 	if (itype == INTR_MSI) {
3510 		MPASS(powerof2(iaq->nirq));
3511 	}
3512 }
3513 
3514 static int
3515 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
3516 {
3517 	int rc, itype, navail, nalloc;
3518 
3519 	for (itype = INTR_MSIX; itype; itype >>= 1) {
3520 
3521 		if ((itype & t4_intr_types) == 0)
3522 			continue;	/* not allowed */
3523 
3524 		if (itype == INTR_MSIX)
3525 			navail = pci_msix_count(sc->dev);
3526 		else if (itype == INTR_MSI)
3527 			navail = pci_msi_count(sc->dev);
3528 		else
3529 			navail = 1;
3530 restart:
3531 		if (navail == 0)
3532 			continue;
3533 
3534 		calculate_iaq(sc, iaq, itype, navail);
3535 		nalloc = iaq->nirq;
3536 		rc = 0;
3537 		if (itype == INTR_MSIX)
3538 			rc = pci_alloc_msix(sc->dev, &nalloc);
3539 		else if (itype == INTR_MSI)
3540 			rc = pci_alloc_msi(sc->dev, &nalloc);
3541 
3542 		if (rc == 0 && nalloc > 0) {
3543 			if (nalloc == iaq->nirq)
3544 				return (0);
3545 
3546 			/*
3547 			 * Didn't get the number requested.  Use whatever number
3548 			 * the kernel is willing to allocate.
3549 			 */
3550 			device_printf(sc->dev, "fewer vectors than requested, "
3551 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
3552 			    itype, iaq->nirq, nalloc);
3553 			pci_release_msi(sc->dev);
3554 			navail = nalloc;
3555 			goto restart;
3556 		}
3557 
3558 		device_printf(sc->dev,
3559 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
3560 		    itype, rc, iaq->nirq, nalloc);
3561 	}
3562 
3563 	device_printf(sc->dev,
3564 	    "failed to find a usable interrupt type.  "
3565 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
3566 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
3567 
3568 	return (ENXIO);
3569 }
3570 
3571 #define FW_VERSION(chip) ( \
3572     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
3573     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
3574     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
3575     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
3576 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
3577 
3578 /* Just enough of fw_hdr to cover all version info. */
3579 struct fw_h {
3580 	__u8	ver;
3581 	__u8	chip;
3582 	__be16	len512;
3583 	__be32	fw_ver;
3584 	__be32	tp_microcode_ver;
3585 	__u8	intfver_nic;
3586 	__u8	intfver_vnic;
3587 	__u8	intfver_ofld;
3588 	__u8	intfver_ri;
3589 	__u8	intfver_iscsipdu;
3590 	__u8	intfver_iscsi;
3591 	__u8	intfver_fcoepdu;
3592 	__u8	intfver_fcoe;
3593 };
3594 /* Spot check a couple of fields. */
3595 CTASSERT(offsetof(struct fw_h, fw_ver) == offsetof(struct fw_hdr, fw_ver));
3596 CTASSERT(offsetof(struct fw_h, intfver_nic) == offsetof(struct fw_hdr, intfver_nic));
3597 CTASSERT(offsetof(struct fw_h, intfver_fcoe) == offsetof(struct fw_hdr, intfver_fcoe));
3598 
3599 struct fw_info {
3600 	uint8_t chip;
3601 	char *kld_name;
3602 	char *fw_mod_name;
3603 	struct fw_h fw_h;
3604 } fw_info[] = {
3605 	{
3606 		.chip = CHELSIO_T4,
3607 		.kld_name = "t4fw_cfg",
3608 		.fw_mod_name = "t4fw",
3609 		.fw_h = {
3610 			.chip = FW_HDR_CHIP_T4,
3611 			.fw_ver = htobe32(FW_VERSION(T4)),
3612 			.intfver_nic = FW_INTFVER(T4, NIC),
3613 			.intfver_vnic = FW_INTFVER(T4, VNIC),
3614 			.intfver_ofld = FW_INTFVER(T4, OFLD),
3615 			.intfver_ri = FW_INTFVER(T4, RI),
3616 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
3617 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
3618 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
3619 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
3620 		},
3621 	}, {
3622 		.chip = CHELSIO_T5,
3623 		.kld_name = "t5fw_cfg",
3624 		.fw_mod_name = "t5fw",
3625 		.fw_h = {
3626 			.chip = FW_HDR_CHIP_T5,
3627 			.fw_ver = htobe32(FW_VERSION(T5)),
3628 			.intfver_nic = FW_INTFVER(T5, NIC),
3629 			.intfver_vnic = FW_INTFVER(T5, VNIC),
3630 			.intfver_ofld = FW_INTFVER(T5, OFLD),
3631 			.intfver_ri = FW_INTFVER(T5, RI),
3632 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
3633 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
3634 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
3635 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
3636 		},
3637 	}, {
3638 		.chip = CHELSIO_T6,
3639 		.kld_name = "t6fw_cfg",
3640 		.fw_mod_name = "t6fw",
3641 		.fw_h = {
3642 			.chip = FW_HDR_CHIP_T6,
3643 			.fw_ver = htobe32(FW_VERSION(T6)),
3644 			.intfver_nic = FW_INTFVER(T6, NIC),
3645 			.intfver_vnic = FW_INTFVER(T6, VNIC),
3646 			.intfver_ofld = FW_INTFVER(T6, OFLD),
3647 			.intfver_ri = FW_INTFVER(T6, RI),
3648 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3649 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
3650 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3651 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
3652 		},
3653 	}
3654 };
3655 
3656 static struct fw_info *
3657 find_fw_info(int chip)
3658 {
3659 	int i;
3660 
3661 	for (i = 0; i < nitems(fw_info); i++) {
3662 		if (fw_info[i].chip == chip)
3663 			return (&fw_info[i]);
3664 	}
3665 	return (NULL);
3666 }
3667 
3668 /*
3669  * Is the given firmware API compatible with the one the driver was compiled
3670  * with?
3671  */
3672 static int
3673 fw_compatible(const struct fw_h *hdr1, const struct fw_h *hdr2)
3674 {
3675 
3676 	/* short circuit if it's the exact same firmware version */
3677 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
3678 		return (1);
3679 
3680 	/*
3681 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
3682 	 * features that are supported in the driver.
3683 	 */
3684 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
3685 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
3686 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
3687 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
3688 		return (1);
3689 #undef SAME_INTF
3690 
3691 	return (0);
3692 }
3693 
3694 static int
3695 load_fw_module(struct adapter *sc, const struct firmware **dcfg,
3696     const struct firmware **fw)
3697 {
3698 	struct fw_info *fw_info;
3699 
3700 	*dcfg = NULL;
3701 	if (fw != NULL)
3702 		*fw = NULL;
3703 
3704 	fw_info = find_fw_info(chip_id(sc));
3705 	if (fw_info == NULL) {
3706 		device_printf(sc->dev,
3707 		    "unable to look up firmware information for chip %d.\n",
3708 		    chip_id(sc));
3709 		return (EINVAL);
3710 	}
3711 
3712 	*dcfg = firmware_get(fw_info->kld_name);
3713 	if (*dcfg != NULL) {
3714 		if (fw != NULL)
3715 			*fw = firmware_get(fw_info->fw_mod_name);
3716 		return (0);
3717 	}
3718 
3719 	return (ENOENT);
3720 }
3721 
3722 static void
3723 unload_fw_module(struct adapter *sc, const struct firmware *dcfg,
3724     const struct firmware *fw)
3725 {
3726 
3727 	if (fw != NULL)
3728 		firmware_put(fw, FIRMWARE_UNLOAD);
3729 	if (dcfg != NULL)
3730 		firmware_put(dcfg, FIRMWARE_UNLOAD);
3731 }
3732 
3733 /*
3734  * Return values:
3735  * 0 means no firmware install attempted.
3736  * ERESTART means a firmware install was attempted and was successful.
3737  * +ve errno means a firmware install was attempted but failed.
3738  */
3739 static int
3740 install_kld_firmware(struct adapter *sc, struct fw_h *card_fw,
3741     const struct fw_h *drv_fw, const char *reason, int *already)
3742 {
3743 	const struct firmware *cfg, *fw;
3744 	const uint32_t c = be32toh(card_fw->fw_ver);
3745 	uint32_t d, k;
3746 	int rc, fw_install;
3747 	struct fw_h bundled_fw;
3748 	bool load_attempted;
3749 
3750 	cfg = fw = NULL;
3751 	load_attempted = false;
3752 	fw_install = t4_fw_install < 0 ? -t4_fw_install : t4_fw_install;
3753 
3754 	memcpy(&bundled_fw, drv_fw, sizeof(bundled_fw));
3755 	if (t4_fw_install < 0) {
3756 		rc = load_fw_module(sc, &cfg, &fw);
3757 		if (rc != 0 || fw == NULL) {
3758 			device_printf(sc->dev,
3759 			    "failed to load firmware module: %d. cfg %p, fw %p;"
3760 			    " will use compiled-in firmware version for"
3761 			    "hw.cxgbe.fw_install checks.\n",
3762 			    rc, cfg, fw);
3763 		} else {
3764 			memcpy(&bundled_fw, fw->data, sizeof(bundled_fw));
3765 		}
3766 		load_attempted = true;
3767 	}
3768 	d = be32toh(bundled_fw.fw_ver);
3769 
3770 	if (reason != NULL)
3771 		goto install;
3772 
3773 	if ((sc->flags & FW_OK) == 0) {
3774 
3775 		if (c == 0xffffffff) {
3776 			reason = "missing";
3777 			goto install;
3778 		}
3779 
3780 		rc = 0;
3781 		goto done;
3782 	}
3783 
3784 	if (!fw_compatible(card_fw, &bundled_fw)) {
3785 		reason = "incompatible or unusable";
3786 		goto install;
3787 	}
3788 
3789 	if (d > c) {
3790 		reason = "older than the version bundled with this driver";
3791 		goto install;
3792 	}
3793 
3794 	if (fw_install == 2 && d != c) {
3795 		reason = "different than the version bundled with this driver";
3796 		goto install;
3797 	}
3798 
3799 	/* No reason to do anything to the firmware already on the card. */
3800 	rc = 0;
3801 	goto done;
3802 
3803 install:
3804 	rc = 0;
3805 	if ((*already)++)
3806 		goto done;
3807 
3808 	if (fw_install == 0) {
3809 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3810 		    "but the driver is prohibited from installing a firmware "
3811 		    "on the card.\n",
3812 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3813 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3814 
3815 		goto done;
3816 	}
3817 
3818 	/*
3819 	 * We'll attempt to install a firmware.  Load the module first (if it
3820 	 * hasn't been loaded already).
3821 	 */
3822 	if (!load_attempted) {
3823 		rc = load_fw_module(sc, &cfg, &fw);
3824 		if (rc != 0 || fw == NULL) {
3825 			device_printf(sc->dev,
3826 			    "failed to load firmware module: %d. cfg %p, fw %p\n",
3827 			    rc, cfg, fw);
3828 			/* carry on */
3829 		}
3830 	}
3831 	if (fw == NULL) {
3832 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3833 		    "but the driver cannot take corrective action because it "
3834 		    "is unable to load the firmware module.\n",
3835 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3836 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3837 		rc = sc->flags & FW_OK ? 0 : ENOENT;
3838 		goto done;
3839 	}
3840 	k = be32toh(((const struct fw_hdr *)fw->data)->fw_ver);
3841 	if (k != d) {
3842 		MPASS(t4_fw_install > 0);
3843 		device_printf(sc->dev,
3844 		    "firmware in KLD (%u.%u.%u.%u) is not what the driver was "
3845 		    "expecting (%u.%u.%u.%u) and will not be used.\n",
3846 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3847 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k),
3848 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3849 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3850 		rc = sc->flags & FW_OK ? 0 : EINVAL;
3851 		goto done;
3852 	}
3853 
3854 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3855 	    "installing firmware %u.%u.%u.%u on card.\n",
3856 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3857 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3858 	    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3859 	    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3860 
3861 	rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3862 	if (rc != 0) {
3863 		device_printf(sc->dev, "failed to install firmware: %d\n", rc);
3864 	} else {
3865 		/* Installed successfully, update the cached header too. */
3866 		rc = ERESTART;
3867 		memcpy(card_fw, fw->data, sizeof(*card_fw));
3868 	}
3869 done:
3870 	unload_fw_module(sc, cfg, fw);
3871 
3872 	return (rc);
3873 }
3874 
3875 /*
3876  * Establish contact with the firmware and attempt to become the master driver.
3877  *
3878  * A firmware will be installed to the card if needed (if the driver is allowed
3879  * to do so).
3880  */
3881 static int
3882 contact_firmware(struct adapter *sc)
3883 {
3884 	int rc, already = 0;
3885 	enum dev_state state;
3886 	struct fw_info *fw_info;
3887 	struct fw_hdr *card_fw;		/* fw on the card */
3888 	const struct fw_h *drv_fw;
3889 
3890 	fw_info = find_fw_info(chip_id(sc));
3891 	if (fw_info == NULL) {
3892 		device_printf(sc->dev,
3893 		    "unable to look up firmware information for chip %d.\n",
3894 		    chip_id(sc));
3895 		return (EINVAL);
3896 	}
3897 	drv_fw = &fw_info->fw_h;
3898 
3899 	/* Read the header of the firmware on the card */
3900 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3901 restart:
3902 	rc = -t4_get_fw_hdr(sc, card_fw);
3903 	if (rc != 0) {
3904 		device_printf(sc->dev,
3905 		    "unable to read firmware header from card's flash: %d\n",
3906 		    rc);
3907 		goto done;
3908 	}
3909 
3910 	rc = install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw, NULL,
3911 	    &already);
3912 	if (rc == ERESTART)
3913 		goto restart;
3914 	if (rc != 0)
3915 		goto done;
3916 
3917 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3918 	if (rc < 0 || state == DEV_STATE_ERR) {
3919 		rc = -rc;
3920 		device_printf(sc->dev,
3921 		    "failed to connect to the firmware: %d, %d.  "
3922 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3923 #if 0
3924 		if (install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw,
3925 		    "not responding properly to HELLO", &already) == ERESTART)
3926 			goto restart;
3927 #endif
3928 		goto done;
3929 	}
3930 	MPASS(be32toh(card_fw->flags) & FW_HDR_FLAGS_RESET_HALT);
3931 	sc->flags |= FW_OK;	/* The firmware responded to the FW_HELLO. */
3932 
3933 	if (rc == sc->pf) {
3934 		sc->flags |= MASTER_PF;
3935 		rc = install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw,
3936 		    NULL, &already);
3937 		if (rc == ERESTART)
3938 			rc = 0;
3939 		else if (rc != 0)
3940 			goto done;
3941 	} else if (state == DEV_STATE_UNINIT) {
3942 		/*
3943 		 * We didn't get to be the master so we definitely won't be
3944 		 * configuring the chip.  It's a bug if someone else hasn't
3945 		 * configured it already.
3946 		 */
3947 		device_printf(sc->dev, "couldn't be master(%d), "
3948 		    "device not already initialized either(%d).  "
3949 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3950 		rc = EPROTO;
3951 		goto done;
3952 	} else {
3953 		/*
3954 		 * Some other PF is the master and has configured the chip.
3955 		 * This is allowed but untested.
3956 		 */
3957 		device_printf(sc->dev, "PF%d is master, device state %d.  "
3958 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3959 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", rc);
3960 		sc->cfcsum = 0;
3961 		rc = 0;
3962 	}
3963 done:
3964 	if (rc != 0 && sc->flags & FW_OK) {
3965 		t4_fw_bye(sc, sc->mbox);
3966 		sc->flags &= ~FW_OK;
3967 	}
3968 	free(card_fw, M_CXGBE);
3969 	return (rc);
3970 }
3971 
3972 static int
3973 copy_cfg_file_to_card(struct adapter *sc, char *cfg_file,
3974     uint32_t mtype, uint32_t moff)
3975 {
3976 	struct fw_info *fw_info;
3977 	const struct firmware *dcfg, *rcfg = NULL;
3978 	const uint32_t *cfdata;
3979 	uint32_t cflen, addr;
3980 	int rc;
3981 
3982 	load_fw_module(sc, &dcfg, NULL);
3983 
3984 	/* Card specific interpretation of "default". */
3985 	if (strncmp(cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3986 		if (pci_get_device(sc->dev) == 0x440a)
3987 			snprintf(cfg_file, sizeof(t4_cfg_file), UWIRE_CF);
3988 		if (is_fpga(sc))
3989 			snprintf(cfg_file, sizeof(t4_cfg_file), FPGA_CF);
3990 	}
3991 
3992 	if (strncmp(cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3993 		if (dcfg == NULL) {
3994 			device_printf(sc->dev,
3995 			    "KLD with default config is not available.\n");
3996 			rc = ENOENT;
3997 			goto done;
3998 		}
3999 		cfdata = dcfg->data;
4000 		cflen = dcfg->datasize & ~3;
4001 	} else {
4002 		char s[32];
4003 
4004 		fw_info = find_fw_info(chip_id(sc));
4005 		if (fw_info == NULL) {
4006 			device_printf(sc->dev,
4007 			    "unable to look up firmware information for chip %d.\n",
4008 			    chip_id(sc));
4009 			rc = EINVAL;
4010 			goto done;
4011 		}
4012 		snprintf(s, sizeof(s), "%s_%s", fw_info->kld_name, cfg_file);
4013 
4014 		rcfg = firmware_get(s);
4015 		if (rcfg == NULL) {
4016 			device_printf(sc->dev,
4017 			    "unable to load module \"%s\" for configuration "
4018 			    "profile \"%s\".\n", s, cfg_file);
4019 			rc = ENOENT;
4020 			goto done;
4021 		}
4022 		cfdata = rcfg->data;
4023 		cflen = rcfg->datasize & ~3;
4024 	}
4025 
4026 	if (cflen > FLASH_CFG_MAX_SIZE) {
4027 		device_printf(sc->dev,
4028 		    "config file too long (%d, max allowed is %d).\n",
4029 		    cflen, FLASH_CFG_MAX_SIZE);
4030 		rc = EINVAL;
4031 		goto done;
4032 	}
4033 
4034 	rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
4035 	if (rc != 0) {
4036 		device_printf(sc->dev,
4037 		    "%s: addr (%d/0x%x) or len %d is not valid: %d.\n",
4038 		    __func__, mtype, moff, cflen, rc);
4039 		rc = EINVAL;
4040 		goto done;
4041 	}
4042 	write_via_memwin(sc, 2, addr, cfdata, cflen);
4043 done:
4044 	if (rcfg != NULL)
4045 		firmware_put(rcfg, FIRMWARE_UNLOAD);
4046 	unload_fw_module(sc, dcfg, NULL);
4047 	return (rc);
4048 }
4049 
4050 struct caps_allowed {
4051 	uint16_t nbmcaps;
4052 	uint16_t linkcaps;
4053 	uint16_t switchcaps;
4054 	uint16_t niccaps;
4055 	uint16_t toecaps;
4056 	uint16_t rdmacaps;
4057 	uint16_t cryptocaps;
4058 	uint16_t iscsicaps;
4059 	uint16_t fcoecaps;
4060 };
4061 
4062 #define FW_PARAM_DEV(param) \
4063 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
4064 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
4065 #define FW_PARAM_PFVF(param) \
4066 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
4067 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
4068 
4069 /*
4070  * Provide a configuration profile to the firmware and have it initialize the
4071  * chip accordingly.  This may involve uploading a configuration file to the
4072  * card.
4073  */
4074 static int
4075 apply_cfg_and_initialize(struct adapter *sc, char *cfg_file,
4076     const struct caps_allowed *caps_allowed)
4077 {
4078 	int rc;
4079 	struct fw_caps_config_cmd caps;
4080 	uint32_t mtype, moff, finicsum, cfcsum, param, val;
4081 
4082 	rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
4083 	if (rc != 0) {
4084 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
4085 		return (rc);
4086 	}
4087 
4088 	bzero(&caps, sizeof(caps));
4089 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4090 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
4091 	if (strncmp(cfg_file, BUILTIN_CF, sizeof(t4_cfg_file)) == 0) {
4092 		mtype = 0;
4093 		moff = 0;
4094 		caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
4095 	} else if (strncmp(cfg_file, FLASH_CF, sizeof(t4_cfg_file)) == 0) {
4096 		mtype = FW_MEMTYPE_FLASH;
4097 		moff = t4_flash_cfg_addr(sc);
4098 		caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
4099 		    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
4100 		    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) |
4101 		    FW_LEN16(caps));
4102 	} else {
4103 		/*
4104 		 * Ask the firmware where it wants us to upload the config file.
4105 		 */
4106 		param = FW_PARAM_DEV(CF);
4107 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4108 		if (rc != 0) {
4109 			/* No support for config file?  Shouldn't happen. */
4110 			device_printf(sc->dev,
4111 			    "failed to query config file location: %d.\n", rc);
4112 			goto done;
4113 		}
4114 		mtype = G_FW_PARAMS_PARAM_Y(val);
4115 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
4116 		caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
4117 		    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
4118 		    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) |
4119 		    FW_LEN16(caps));
4120 
4121 		rc = copy_cfg_file_to_card(sc, cfg_file, mtype, moff);
4122 		if (rc != 0) {
4123 			device_printf(sc->dev,
4124 			    "failed to upload config file to card: %d.\n", rc);
4125 			goto done;
4126 		}
4127 	}
4128 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
4129 	if (rc != 0) {
4130 		device_printf(sc->dev, "failed to pre-process config file: %d "
4131 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
4132 		goto done;
4133 	}
4134 
4135 	finicsum = be32toh(caps.finicsum);
4136 	cfcsum = be32toh(caps.cfcsum);	/* actual */
4137 	if (finicsum != cfcsum) {
4138 		device_printf(sc->dev,
4139 		    "WARNING: config file checksum mismatch: %08x %08x\n",
4140 		    finicsum, cfcsum);
4141 	}
4142 	sc->cfcsum = cfcsum;
4143 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", cfg_file);
4144 
4145 	/*
4146 	 * Let the firmware know what features will (not) be used so it can tune
4147 	 * things accordingly.
4148 	 */
4149 #define LIMIT_CAPS(x) do { \
4150 	caps.x##caps &= htobe16(caps_allowed->x##caps); \
4151 } while (0)
4152 	LIMIT_CAPS(nbm);
4153 	LIMIT_CAPS(link);
4154 	LIMIT_CAPS(switch);
4155 	LIMIT_CAPS(nic);
4156 	LIMIT_CAPS(toe);
4157 	LIMIT_CAPS(rdma);
4158 	LIMIT_CAPS(crypto);
4159 	LIMIT_CAPS(iscsi);
4160 	LIMIT_CAPS(fcoe);
4161 #undef LIMIT_CAPS
4162 	if (caps.niccaps & htobe16(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
4163 		/*
4164 		 * TOE and hashfilters are mutually exclusive.  It is a config
4165 		 * file or firmware bug if both are reported as available.  Try
4166 		 * to cope with the situation in non-debug builds by disabling
4167 		 * TOE.
4168 		 */
4169 		MPASS(caps.toecaps == 0);
4170 
4171 		caps.toecaps = 0;
4172 		caps.rdmacaps = 0;
4173 		caps.iscsicaps = 0;
4174 	}
4175 
4176 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4177 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
4178 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
4179 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
4180 	if (rc != 0) {
4181 		device_printf(sc->dev,
4182 		    "failed to process config file: %d.\n", rc);
4183 		goto done;
4184 	}
4185 
4186 	t4_tweak_chip_settings(sc);
4187 	set_params__pre_init(sc);
4188 
4189 	/* get basic stuff going */
4190 	rc = -t4_fw_initialize(sc, sc->mbox);
4191 	if (rc != 0) {
4192 		device_printf(sc->dev, "fw_initialize failed: %d.\n", rc);
4193 		goto done;
4194 	}
4195 done:
4196 	return (rc);
4197 }
4198 
4199 /*
4200  * Partition chip resources for use between various PFs, VFs, etc.
4201  */
4202 static int
4203 partition_resources(struct adapter *sc)
4204 {
4205 	char cfg_file[sizeof(t4_cfg_file)];
4206 	struct caps_allowed caps_allowed;
4207 	int rc;
4208 	bool fallback;
4209 
4210 	/* Only the master driver gets to configure the chip resources. */
4211 	MPASS(sc->flags & MASTER_PF);
4212 
4213 #define COPY_CAPS(x) do { \
4214 	caps_allowed.x##caps = t4_##x##caps_allowed; \
4215 } while (0)
4216 	bzero(&caps_allowed, sizeof(caps_allowed));
4217 	COPY_CAPS(nbm);
4218 	COPY_CAPS(link);
4219 	COPY_CAPS(switch);
4220 	COPY_CAPS(nic);
4221 	COPY_CAPS(toe);
4222 	COPY_CAPS(rdma);
4223 	COPY_CAPS(crypto);
4224 	COPY_CAPS(iscsi);
4225 	COPY_CAPS(fcoe);
4226 	fallback = sc->debug_flags & DF_DISABLE_CFG_RETRY ? false : true;
4227 	snprintf(cfg_file, sizeof(cfg_file), "%s", t4_cfg_file);
4228 retry:
4229 	rc = apply_cfg_and_initialize(sc, cfg_file, &caps_allowed);
4230 	if (rc != 0 && fallback) {
4231 		device_printf(sc->dev,
4232 		    "failed (%d) to configure card with \"%s\" profile, "
4233 		    "will fall back to a basic configuration and retry.\n",
4234 		    rc, cfg_file);
4235 		snprintf(cfg_file, sizeof(cfg_file), "%s", BUILTIN_CF);
4236 		bzero(&caps_allowed, sizeof(caps_allowed));
4237 		COPY_CAPS(switch);
4238 		caps_allowed.niccaps = FW_CAPS_CONFIG_NIC;
4239 		fallback = false;
4240 		goto retry;
4241 	}
4242 #undef COPY_CAPS
4243 	return (rc);
4244 }
4245 
4246 /*
4247  * Retrieve parameters that are needed (or nice to have) very early.
4248  */
4249 static int
4250 get_params__pre_init(struct adapter *sc)
4251 {
4252 	int rc;
4253 	uint32_t param[2], val[2];
4254 
4255 	t4_get_version_info(sc);
4256 
4257 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
4258 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
4259 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
4260 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
4261 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
4262 
4263 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
4264 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
4265 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
4266 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
4267 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
4268 
4269 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
4270 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
4271 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
4272 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
4273 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
4274 
4275 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
4276 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
4277 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
4278 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
4279 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
4280 
4281 	param[0] = FW_PARAM_DEV(PORTVEC);
4282 	param[1] = FW_PARAM_DEV(CCLK);
4283 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4284 	if (rc != 0) {
4285 		device_printf(sc->dev,
4286 		    "failed to query parameters (pre_init): %d.\n", rc);
4287 		return (rc);
4288 	}
4289 
4290 	sc->params.portvec = val[0];
4291 	sc->params.nports = bitcount32(val[0]);
4292 	sc->params.vpd.cclk = val[1];
4293 
4294 	/* Read device log parameters. */
4295 	rc = -t4_init_devlog_params(sc, 1);
4296 	if (rc == 0)
4297 		fixup_devlog_params(sc);
4298 	else {
4299 		device_printf(sc->dev,
4300 		    "failed to get devlog parameters: %d.\n", rc);
4301 		rc = 0;	/* devlog isn't critical for device operation */
4302 	}
4303 
4304 	return (rc);
4305 }
4306 
4307 /*
4308  * Any params that need to be set before FW_INITIALIZE.
4309  */
4310 static int
4311 set_params__pre_init(struct adapter *sc)
4312 {
4313 	int rc = 0;
4314 	uint32_t param, val;
4315 
4316 	if (chip_id(sc) >= CHELSIO_T6) {
4317 		param = FW_PARAM_DEV(HPFILTER_REGION_SUPPORT);
4318 		val = 1;
4319 		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4320 		/* firmwares < 1.20.1.0 do not have this param. */
4321 		if (rc == FW_EINVAL &&
4322 		    sc->params.fw_vers < FW_VERSION32(1, 20, 1, 0)) {
4323 			rc = 0;
4324 		}
4325 		if (rc != 0) {
4326 			device_printf(sc->dev,
4327 			    "failed to enable high priority filters :%d.\n",
4328 			    rc);
4329 		}
4330 	}
4331 
4332 	/* Enable opaque VIIDs with firmwares that support it. */
4333 	param = FW_PARAM_DEV(OPAQUE_VIID_SMT_EXTN);
4334 	val = 1;
4335 	rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4336 	if (rc == 0 && val == 1)
4337 		sc->params.viid_smt_extn_support = true;
4338 	else
4339 		sc->params.viid_smt_extn_support = false;
4340 
4341 	return (rc);
4342 }
4343 
4344 /*
4345  * Retrieve various parameters that are of interest to the driver.  The device
4346  * has been initialized by the firmware at this point.
4347  */
4348 static int
4349 get_params__post_init(struct adapter *sc)
4350 {
4351 	int rc;
4352 	uint32_t param[7], val[7];
4353 	struct fw_caps_config_cmd caps;
4354 
4355 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
4356 	param[1] = FW_PARAM_PFVF(EQ_START);
4357 	param[2] = FW_PARAM_PFVF(FILTER_START);
4358 	param[3] = FW_PARAM_PFVF(FILTER_END);
4359 	param[4] = FW_PARAM_PFVF(L2T_START);
4360 	param[5] = FW_PARAM_PFVF(L2T_END);
4361 	param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4362 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
4363 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
4364 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
4365 	if (rc != 0) {
4366 		device_printf(sc->dev,
4367 		    "failed to query parameters (post_init): %d.\n", rc);
4368 		return (rc);
4369 	}
4370 
4371 	sc->sge.iq_start = val[0];
4372 	sc->sge.eq_start = val[1];
4373 	if ((int)val[3] > (int)val[2]) {
4374 		sc->tids.ftid_base = val[2];
4375 		sc->tids.ftid_end = val[3];
4376 		sc->tids.nftids = val[3] - val[2] + 1;
4377 	}
4378 	sc->vres.l2t.start = val[4];
4379 	sc->vres.l2t.size = val[5] - val[4] + 1;
4380 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
4381 	    ("%s: L2 table size (%u) larger than expected (%u)",
4382 	    __func__, sc->vres.l2t.size, L2T_SIZE));
4383 	sc->params.core_vdd = val[6];
4384 
4385 	if (chip_id(sc) >= CHELSIO_T6) {
4386 
4387 		sc->tids.tid_base = t4_read_reg(sc,
4388 		    A_LE_DB_ACTIVE_TABLE_START_INDEX);
4389 
4390 		param[0] = FW_PARAM_PFVF(HPFILTER_START);
4391 		param[1] = FW_PARAM_PFVF(HPFILTER_END);
4392 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4393 		if (rc != 0) {
4394 			device_printf(sc->dev,
4395 			   "failed to query hpfilter parameters: %d.\n", rc);
4396 			return (rc);
4397 		}
4398 		if ((int)val[1] > (int)val[0]) {
4399 			sc->tids.hpftid_base = val[0];
4400 			sc->tids.hpftid_end = val[1];
4401 			sc->tids.nhpftids = val[1] - val[0] + 1;
4402 
4403 			/*
4404 			 * These should go off if the layout changes and the
4405 			 * driver needs to catch up.
4406 			 */
4407 			MPASS(sc->tids.hpftid_base == 0);
4408 			MPASS(sc->tids.tid_base == sc->tids.nhpftids);
4409 		}
4410 	}
4411 
4412 	/*
4413 	 * MPSBGMAP is queried separately because only recent firmwares support
4414 	 * it as a parameter and we don't want the compound query above to fail
4415 	 * on older firmwares.
4416 	 */
4417 	param[0] = FW_PARAM_DEV(MPSBGMAP);
4418 	val[0] = 0;
4419 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4420 	if (rc == 0)
4421 		sc->params.mps_bg_map = val[0];
4422 	else
4423 		sc->params.mps_bg_map = 0;
4424 
4425 	/*
4426 	 * Determine whether the firmware supports the filter2 work request.
4427 	 * This is queried separately for the same reason as MPSBGMAP above.
4428 	 */
4429 	param[0] = FW_PARAM_DEV(FILTER2_WR);
4430 	val[0] = 0;
4431 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4432 	if (rc == 0)
4433 		sc->params.filter2_wr_support = val[0] != 0;
4434 	else
4435 		sc->params.filter2_wr_support = 0;
4436 
4437 	/*
4438 	 * Find out whether we're allowed to use the ULPTX MEMWRITE DSGL.
4439 	 * This is queried separately for the same reason as other params above.
4440 	 */
4441 	param[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
4442 	val[0] = 0;
4443 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4444 	if (rc == 0)
4445 		sc->params.ulptx_memwrite_dsgl = val[0] != 0;
4446 	else
4447 		sc->params.ulptx_memwrite_dsgl = false;
4448 
4449 	/* FW_RI_FR_NSMR_TPTE_WR support */
4450 	param[0] = FW_PARAM_DEV(RI_FR_NSMR_TPTE_WR);
4451 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4452 	if (rc == 0)
4453 		sc->params.fr_nsmr_tpte_wr_support = val[0] != 0;
4454 	else
4455 		sc->params.fr_nsmr_tpte_wr_support = false;
4456 
4457 	/* get capabilites */
4458 	bzero(&caps, sizeof(caps));
4459 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4460 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
4461 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
4462 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
4463 	if (rc != 0) {
4464 		device_printf(sc->dev,
4465 		    "failed to get card capabilities: %d.\n", rc);
4466 		return (rc);
4467 	}
4468 
4469 #define READ_CAPS(x) do { \
4470 	sc->x = htobe16(caps.x); \
4471 } while (0)
4472 	READ_CAPS(nbmcaps);
4473 	READ_CAPS(linkcaps);
4474 	READ_CAPS(switchcaps);
4475 	READ_CAPS(niccaps);
4476 	READ_CAPS(toecaps);
4477 	READ_CAPS(rdmacaps);
4478 	READ_CAPS(cryptocaps);
4479 	READ_CAPS(iscsicaps);
4480 	READ_CAPS(fcoecaps);
4481 
4482 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_HASHFILTER) {
4483 		MPASS(chip_id(sc) > CHELSIO_T4);
4484 		MPASS(sc->toecaps == 0);
4485 		sc->toecaps = 0;
4486 
4487 		param[0] = FW_PARAM_DEV(NTID);
4488 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4489 		if (rc != 0) {
4490 			device_printf(sc->dev,
4491 			    "failed to query HASHFILTER parameters: %d.\n", rc);
4492 			return (rc);
4493 		}
4494 		sc->tids.ntids = val[0];
4495 		if (sc->params.fw_vers < FW_VERSION32(1, 20, 5, 0)) {
4496 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4497 			sc->tids.ntids -= sc->tids.nhpftids;
4498 		}
4499 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4500 		sc->params.hash_filter = 1;
4501 	}
4502 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
4503 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
4504 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
4505 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4506 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
4507 		if (rc != 0) {
4508 			device_printf(sc->dev,
4509 			    "failed to query NIC parameters: %d.\n", rc);
4510 			return (rc);
4511 		}
4512 		if ((int)val[1] > (int)val[0]) {
4513 			sc->tids.etid_base = val[0];
4514 			sc->tids.etid_end = val[1];
4515 			sc->tids.netids = val[1] - val[0] + 1;
4516 			sc->params.eo_wr_cred = val[2];
4517 			sc->params.ethoffload = 1;
4518 		}
4519 	}
4520 	if (sc->toecaps) {
4521 		/* query offload-related parameters */
4522 		param[0] = FW_PARAM_DEV(NTID);
4523 		param[1] = FW_PARAM_PFVF(SERVER_START);
4524 		param[2] = FW_PARAM_PFVF(SERVER_END);
4525 		param[3] = FW_PARAM_PFVF(TDDP_START);
4526 		param[4] = FW_PARAM_PFVF(TDDP_END);
4527 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4528 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4529 		if (rc != 0) {
4530 			device_printf(sc->dev,
4531 			    "failed to query TOE parameters: %d.\n", rc);
4532 			return (rc);
4533 		}
4534 		sc->tids.ntids = val[0];
4535 		if (sc->params.fw_vers < FW_VERSION32(1, 20, 5, 0)) {
4536 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4537 			sc->tids.ntids -= sc->tids.nhpftids;
4538 		}
4539 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4540 		if ((int)val[2] > (int)val[1]) {
4541 			sc->tids.stid_base = val[1];
4542 			sc->tids.nstids = val[2] - val[1] + 1;
4543 		}
4544 		sc->vres.ddp.start = val[3];
4545 		sc->vres.ddp.size = val[4] - val[3] + 1;
4546 		sc->params.ofldq_wr_cred = val[5];
4547 		sc->params.offload = 1;
4548 	} else {
4549 		/*
4550 		 * The firmware attempts memfree TOE configuration for -SO cards
4551 		 * and will report toecaps=0 if it runs out of resources (this
4552 		 * depends on the config file).  It may not report 0 for other
4553 		 * capabilities dependent on the TOE in this case.  Set them to
4554 		 * 0 here so that the driver doesn't bother tracking resources
4555 		 * that will never be used.
4556 		 */
4557 		sc->iscsicaps = 0;
4558 		sc->rdmacaps = 0;
4559 	}
4560 	if (sc->rdmacaps) {
4561 		param[0] = FW_PARAM_PFVF(STAG_START);
4562 		param[1] = FW_PARAM_PFVF(STAG_END);
4563 		param[2] = FW_PARAM_PFVF(RQ_START);
4564 		param[3] = FW_PARAM_PFVF(RQ_END);
4565 		param[4] = FW_PARAM_PFVF(PBL_START);
4566 		param[5] = FW_PARAM_PFVF(PBL_END);
4567 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4568 		if (rc != 0) {
4569 			device_printf(sc->dev,
4570 			    "failed to query RDMA parameters(1): %d.\n", rc);
4571 			return (rc);
4572 		}
4573 		sc->vres.stag.start = val[0];
4574 		sc->vres.stag.size = val[1] - val[0] + 1;
4575 		sc->vres.rq.start = val[2];
4576 		sc->vres.rq.size = val[3] - val[2] + 1;
4577 		sc->vres.pbl.start = val[4];
4578 		sc->vres.pbl.size = val[5] - val[4] + 1;
4579 
4580 		param[0] = FW_PARAM_PFVF(SQRQ_START);
4581 		param[1] = FW_PARAM_PFVF(SQRQ_END);
4582 		param[2] = FW_PARAM_PFVF(CQ_START);
4583 		param[3] = FW_PARAM_PFVF(CQ_END);
4584 		param[4] = FW_PARAM_PFVF(OCQ_START);
4585 		param[5] = FW_PARAM_PFVF(OCQ_END);
4586 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4587 		if (rc != 0) {
4588 			device_printf(sc->dev,
4589 			    "failed to query RDMA parameters(2): %d.\n", rc);
4590 			return (rc);
4591 		}
4592 		sc->vres.qp.start = val[0];
4593 		sc->vres.qp.size = val[1] - val[0] + 1;
4594 		sc->vres.cq.start = val[2];
4595 		sc->vres.cq.size = val[3] - val[2] + 1;
4596 		sc->vres.ocq.start = val[4];
4597 		sc->vres.ocq.size = val[5] - val[4] + 1;
4598 
4599 		param[0] = FW_PARAM_PFVF(SRQ_START);
4600 		param[1] = FW_PARAM_PFVF(SRQ_END);
4601 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
4602 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4603 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
4604 		if (rc != 0) {
4605 			device_printf(sc->dev,
4606 			    "failed to query RDMA parameters(3): %d.\n", rc);
4607 			return (rc);
4608 		}
4609 		sc->vres.srq.start = val[0];
4610 		sc->vres.srq.size = val[1] - val[0] + 1;
4611 		sc->params.max_ordird_qp = val[2];
4612 		sc->params.max_ird_adapter = val[3];
4613 	}
4614 	if (sc->iscsicaps) {
4615 		param[0] = FW_PARAM_PFVF(ISCSI_START);
4616 		param[1] = FW_PARAM_PFVF(ISCSI_END);
4617 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4618 		if (rc != 0) {
4619 			device_printf(sc->dev,
4620 			    "failed to query iSCSI parameters: %d.\n", rc);
4621 			return (rc);
4622 		}
4623 		sc->vres.iscsi.start = val[0];
4624 		sc->vres.iscsi.size = val[1] - val[0] + 1;
4625 	}
4626 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
4627 		param[0] = FW_PARAM_PFVF(TLS_START);
4628 		param[1] = FW_PARAM_PFVF(TLS_END);
4629 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4630 		if (rc != 0) {
4631 			device_printf(sc->dev,
4632 			    "failed to query TLS parameters: %d.\n", rc);
4633 			return (rc);
4634 		}
4635 		sc->vres.key.start = val[0];
4636 		sc->vres.key.size = val[1] - val[0] + 1;
4637 	}
4638 
4639 	t4_init_sge_params(sc);
4640 
4641 	/*
4642 	 * We've got the params we wanted to query via the firmware.  Now grab
4643 	 * some others directly from the chip.
4644 	 */
4645 	rc = t4_read_chip_settings(sc);
4646 
4647 	return (rc);
4648 }
4649 
4650 #ifdef KERN_TLS
4651 static void
4652 ktls_tick(void *arg)
4653 {
4654 	struct adapter *sc;
4655 	uint32_t tstamp;
4656 
4657 	sc = arg;
4658 
4659 	tstamp = tcp_ts_getticks();
4660 	t4_write_reg(sc, A_TP_SYNC_TIME_HI, tstamp >> 1);
4661 	t4_write_reg(sc, A_TP_SYNC_TIME_LO, tstamp << 31);
4662 
4663 	callout_schedule_sbt(&sc->ktls_tick, SBT_1MS, 0, C_HARDCLOCK);
4664 }
4665 
4666 static void
4667 t4_enable_kern_tls(struct adapter *sc)
4668 {
4669 	uint32_t m, v;
4670 
4671 	m = F_ENABLECBYP;
4672 	v = F_ENABLECBYP;
4673 	t4_set_reg_field(sc, A_TP_PARA_REG6, m, v);
4674 
4675 	m = F_CPL_FLAGS_UPDATE_EN | F_SEQ_UPDATE_EN;
4676 	v = F_CPL_FLAGS_UPDATE_EN | F_SEQ_UPDATE_EN;
4677 	t4_set_reg_field(sc, A_ULP_TX_CONFIG, m, v);
4678 
4679 	m = F_NICMODE;
4680 	v = F_NICMODE;
4681 	t4_set_reg_field(sc, A_TP_IN_CONFIG, m, v);
4682 
4683 	m = F_LOOKUPEVERYPKT;
4684 	v = 0;
4685 	t4_set_reg_field(sc, A_TP_INGRESS_CONFIG, m, v);
4686 
4687 	m = F_TXDEFERENABLE | F_DISABLEWINDOWPSH | F_DISABLESEPPSHFLAG;
4688 	v = F_DISABLEWINDOWPSH;
4689 	t4_set_reg_field(sc, A_TP_PC_CONFIG, m, v);
4690 
4691 	m = V_TIMESTAMPRESOLUTION(M_TIMESTAMPRESOLUTION);
4692 	v = V_TIMESTAMPRESOLUTION(0x1f);
4693 	t4_set_reg_field(sc, A_TP_TIMER_RESOLUTION, m, v);
4694 
4695 	sc->flags |= KERN_TLS_OK;
4696 
4697 	sc->tlst.inline_keys = t4_tls_inline_keys;
4698 	sc->tlst.combo_wrs = t4_tls_combo_wrs;
4699 }
4700 #endif
4701 
4702 static int
4703 set_params__post_init(struct adapter *sc)
4704 {
4705 	uint32_t param, val;
4706 #ifdef TCP_OFFLOAD
4707 	int i, v, shift;
4708 #endif
4709 
4710 	/* ask for encapsulated CPLs */
4711 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4712 	val = 1;
4713 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4714 
4715 	/* Enable 32b port caps if the firmware supports it. */
4716 	param = FW_PARAM_PFVF(PORT_CAPS32);
4717 	val = 1;
4718 	if (t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val) == 0)
4719 		sc->params.port_caps32 = 1;
4720 
4721 	/* Let filter + maskhash steer to a part of the VI's RSS region. */
4722 	val = 1 << (G_MASKSIZE(t4_read_reg(sc, A_TP_RSS_CONFIG_TNL)) - 1);
4723 	t4_set_reg_field(sc, A_TP_RSS_CONFIG_TNL, V_MASKFILTER(M_MASKFILTER),
4724 	    V_MASKFILTER(val - 1));
4725 
4726 #ifdef TCP_OFFLOAD
4727 	/*
4728 	 * Override the TOE timers with user provided tunables.  This is not the
4729 	 * recommended way to change the timers (the firmware config file is) so
4730 	 * these tunables are not documented.
4731 	 *
4732 	 * All the timer tunables are in microseconds.
4733 	 */
4734 	if (t4_toe_keepalive_idle != 0) {
4735 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
4736 		v &= M_KEEPALIVEIDLE;
4737 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
4738 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
4739 	}
4740 	if (t4_toe_keepalive_interval != 0) {
4741 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
4742 		v &= M_KEEPALIVEINTVL;
4743 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
4744 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
4745 	}
4746 	if (t4_toe_keepalive_count != 0) {
4747 		v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
4748 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4749 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
4750 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
4751 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
4752 	}
4753 	if (t4_toe_rexmt_min != 0) {
4754 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
4755 		v &= M_RXTMIN;
4756 		t4_set_reg_field(sc, A_TP_RXT_MIN,
4757 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
4758 	}
4759 	if (t4_toe_rexmt_max != 0) {
4760 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
4761 		v &= M_RXTMAX;
4762 		t4_set_reg_field(sc, A_TP_RXT_MAX,
4763 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
4764 	}
4765 	if (t4_toe_rexmt_count != 0) {
4766 		v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
4767 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4768 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
4769 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
4770 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
4771 	}
4772 	for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
4773 		if (t4_toe_rexmt_backoff[i] != -1) {
4774 			v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
4775 			shift = (i & 3) << 3;
4776 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
4777 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
4778 		}
4779 	}
4780 #endif
4781 
4782 #ifdef KERN_TLS
4783 	if (t4_kern_tls != 0 && sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS &&
4784 	    sc->toecaps & FW_CAPS_CONFIG_TOE)
4785 		t4_enable_kern_tls(sc);
4786 #endif
4787 	return (0);
4788 }
4789 
4790 #undef FW_PARAM_PFVF
4791 #undef FW_PARAM_DEV
4792 
4793 static void
4794 t4_set_desc(struct adapter *sc)
4795 {
4796 	char buf[128];
4797 	struct adapter_params *p = &sc->params;
4798 
4799 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
4800 
4801 	device_set_desc_copy(sc->dev, buf);
4802 }
4803 
4804 static inline void
4805 ifmedia_add4(struct ifmedia *ifm, int m)
4806 {
4807 
4808 	ifmedia_add(ifm, m, 0, NULL);
4809 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE, 0, NULL);
4810 	ifmedia_add(ifm, m | IFM_ETH_RXPAUSE, 0, NULL);
4811 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE, 0, NULL);
4812 }
4813 
4814 /*
4815  * This is the selected media, which is not quite the same as the active media.
4816  * The media line in ifconfig is "media: Ethernet selected (active)" if selected
4817  * and active are not the same, and "media: Ethernet selected" otherwise.
4818  */
4819 static void
4820 set_current_media(struct port_info *pi)
4821 {
4822 	struct link_config *lc;
4823 	struct ifmedia *ifm;
4824 	int mword;
4825 	u_int speed;
4826 
4827 	PORT_LOCK_ASSERT_OWNED(pi);
4828 
4829 	/* Leave current media alone if it's already set to IFM_NONE. */
4830 	ifm = &pi->media;
4831 	if (ifm->ifm_cur != NULL &&
4832 	    IFM_SUBTYPE(ifm->ifm_cur->ifm_media) == IFM_NONE)
4833 		return;
4834 
4835 	lc = &pi->link_cfg;
4836 	if (lc->requested_aneg != AUTONEG_DISABLE &&
4837 	    lc->pcaps & FW_PORT_CAP32_ANEG) {
4838 		ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
4839 		return;
4840 	}
4841 	mword = IFM_ETHER | IFM_FDX;
4842 	if (lc->requested_fc & PAUSE_TX)
4843 		mword |= IFM_ETH_TXPAUSE;
4844 	if (lc->requested_fc & PAUSE_RX)
4845 		mword |= IFM_ETH_RXPAUSE;
4846 	if (lc->requested_speed == 0)
4847 		speed = port_top_speed(pi) * 1000;	/* Gbps -> Mbps */
4848 	else
4849 		speed = lc->requested_speed;
4850 	mword |= port_mword(pi, speed_to_fwcap(speed));
4851 	ifmedia_set(ifm, mword);
4852 }
4853 
4854 /*
4855  * Returns true if the ifmedia list for the port cannot change.
4856  */
4857 static bool
4858 fixed_ifmedia(struct port_info *pi)
4859 {
4860 
4861 	return (pi->port_type == FW_PORT_TYPE_BT_SGMII ||
4862 	    pi->port_type == FW_PORT_TYPE_BT_XFI ||
4863 	    pi->port_type == FW_PORT_TYPE_BT_XAUI ||
4864 	    pi->port_type == FW_PORT_TYPE_KX4 ||
4865 	    pi->port_type == FW_PORT_TYPE_KX ||
4866 	    pi->port_type == FW_PORT_TYPE_KR ||
4867 	    pi->port_type == FW_PORT_TYPE_BP_AP ||
4868 	    pi->port_type == FW_PORT_TYPE_BP4_AP ||
4869 	    pi->port_type == FW_PORT_TYPE_BP40_BA ||
4870 	    pi->port_type == FW_PORT_TYPE_KR4_100G ||
4871 	    pi->port_type == FW_PORT_TYPE_KR_SFP28 ||
4872 	    pi->port_type == FW_PORT_TYPE_KR_XLAUI);
4873 }
4874 
4875 static void
4876 build_medialist(struct port_info *pi)
4877 {
4878 	uint32_t ss, speed;
4879 	int unknown, mword, bit;
4880 	struct link_config *lc;
4881 	struct ifmedia *ifm;
4882 
4883 	PORT_LOCK_ASSERT_OWNED(pi);
4884 
4885 	if (pi->flags & FIXED_IFMEDIA)
4886 		return;
4887 
4888 	/*
4889 	 * Rebuild the ifmedia list.
4890 	 */
4891 	ifm = &pi->media;
4892 	ifmedia_removeall(ifm);
4893 	lc = &pi->link_cfg;
4894 	ss = G_FW_PORT_CAP32_SPEED(lc->pcaps); /* Supported Speeds */
4895 	if (__predict_false(ss == 0)) {	/* not supposed to happen. */
4896 		MPASS(ss != 0);
4897 no_media:
4898 		MPASS(LIST_EMPTY(&ifm->ifm_list));
4899 		ifmedia_add(ifm, IFM_ETHER | IFM_NONE, 0, NULL);
4900 		ifmedia_set(ifm, IFM_ETHER | IFM_NONE);
4901 		return;
4902 	}
4903 
4904 	unknown = 0;
4905 	for (bit = S_FW_PORT_CAP32_SPEED; bit < fls(ss); bit++) {
4906 		speed = 1 << bit;
4907 		MPASS(speed & M_FW_PORT_CAP32_SPEED);
4908 		if (ss & speed) {
4909 			mword = port_mword(pi, speed);
4910 			if (mword == IFM_NONE) {
4911 				goto no_media;
4912 			} else if (mword == IFM_UNKNOWN)
4913 				unknown++;
4914 			else
4915 				ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | mword);
4916 		}
4917 	}
4918 	if (unknown > 0) /* Add one unknown for all unknown media types. */
4919 		ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | IFM_UNKNOWN);
4920 	if (lc->pcaps & FW_PORT_CAP32_ANEG)
4921 		ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
4922 
4923 	set_current_media(pi);
4924 }
4925 
4926 /*
4927  * Initialize the requested fields in the link config based on driver tunables.
4928  */
4929 static void
4930 init_link_config(struct port_info *pi)
4931 {
4932 	struct link_config *lc = &pi->link_cfg;
4933 
4934 	PORT_LOCK_ASSERT_OWNED(pi);
4935 
4936 	lc->requested_speed = 0;
4937 
4938 	if (t4_autoneg == 0)
4939 		lc->requested_aneg = AUTONEG_DISABLE;
4940 	else if (t4_autoneg == 1)
4941 		lc->requested_aneg = AUTONEG_ENABLE;
4942 	else
4943 		lc->requested_aneg = AUTONEG_AUTO;
4944 
4945 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX |
4946 	    PAUSE_AUTONEG);
4947 
4948 	if (t4_fec & FEC_AUTO)
4949 		lc->requested_fec = FEC_AUTO;
4950 	else if (t4_fec == 0)
4951 		lc->requested_fec = FEC_NONE;
4952 	else {
4953 		/* -1 is handled by the FEC_AUTO block above and not here. */
4954 		lc->requested_fec = t4_fec &
4955 		    (FEC_RS | FEC_BASER_RS | FEC_NONE | FEC_MODULE);
4956 		if (lc->requested_fec == 0)
4957 			lc->requested_fec = FEC_AUTO;
4958 	}
4959 }
4960 
4961 /*
4962  * Makes sure that all requested settings comply with what's supported by the
4963  * port.  Returns the number of settings that were invalid and had to be fixed.
4964  */
4965 static int
4966 fixup_link_config(struct port_info *pi)
4967 {
4968 	int n = 0;
4969 	struct link_config *lc = &pi->link_cfg;
4970 	uint32_t fwspeed;
4971 
4972 	PORT_LOCK_ASSERT_OWNED(pi);
4973 
4974 	/* Speed (when not autonegotiating) */
4975 	if (lc->requested_speed != 0) {
4976 		fwspeed = speed_to_fwcap(lc->requested_speed);
4977 		if ((fwspeed & lc->pcaps) == 0) {
4978 			n++;
4979 			lc->requested_speed = 0;
4980 		}
4981 	}
4982 
4983 	/* Link autonegotiation */
4984 	MPASS(lc->requested_aneg == AUTONEG_ENABLE ||
4985 	    lc->requested_aneg == AUTONEG_DISABLE ||
4986 	    lc->requested_aneg == AUTONEG_AUTO);
4987 	if (lc->requested_aneg == AUTONEG_ENABLE &&
4988 	    !(lc->pcaps & FW_PORT_CAP32_ANEG)) {
4989 		n++;
4990 		lc->requested_aneg = AUTONEG_AUTO;
4991 	}
4992 
4993 	/* Flow control */
4994 	MPASS((lc->requested_fc & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG)) == 0);
4995 	if (lc->requested_fc & PAUSE_TX &&
4996 	    !(lc->pcaps & FW_PORT_CAP32_FC_TX)) {
4997 		n++;
4998 		lc->requested_fc &= ~PAUSE_TX;
4999 	}
5000 	if (lc->requested_fc & PAUSE_RX &&
5001 	    !(lc->pcaps & FW_PORT_CAP32_FC_RX)) {
5002 		n++;
5003 		lc->requested_fc &= ~PAUSE_RX;
5004 	}
5005 	if (!(lc->requested_fc & PAUSE_AUTONEG) &&
5006 	    !(lc->pcaps & FW_PORT_CAP32_FORCE_PAUSE)) {
5007 		n++;
5008 		lc->requested_fc |= PAUSE_AUTONEG;
5009 	}
5010 
5011 	/* FEC */
5012 	if ((lc->requested_fec & FEC_RS &&
5013 	    !(lc->pcaps & FW_PORT_CAP32_FEC_RS)) ||
5014 	    (lc->requested_fec & FEC_BASER_RS &&
5015 	    !(lc->pcaps & FW_PORT_CAP32_FEC_BASER_RS))) {
5016 		n++;
5017 		lc->requested_fec = FEC_AUTO;
5018 	}
5019 
5020 	return (n);
5021 }
5022 
5023 /*
5024  * Apply the requested L1 settings, which are expected to be valid, to the
5025  * hardware.
5026  */
5027 static int
5028 apply_link_config(struct port_info *pi)
5029 {
5030 	struct adapter *sc = pi->adapter;
5031 	struct link_config *lc = &pi->link_cfg;
5032 	int rc;
5033 
5034 #ifdef INVARIANTS
5035 	ASSERT_SYNCHRONIZED_OP(sc);
5036 	PORT_LOCK_ASSERT_OWNED(pi);
5037 
5038 	if (lc->requested_aneg == AUTONEG_ENABLE)
5039 		MPASS(lc->pcaps & FW_PORT_CAP32_ANEG);
5040 	if (!(lc->requested_fc & PAUSE_AUTONEG))
5041 		MPASS(lc->pcaps & FW_PORT_CAP32_FORCE_PAUSE);
5042 	if (lc->requested_fc & PAUSE_TX)
5043 		MPASS(lc->pcaps & FW_PORT_CAP32_FC_TX);
5044 	if (lc->requested_fc & PAUSE_RX)
5045 		MPASS(lc->pcaps & FW_PORT_CAP32_FC_RX);
5046 	if (lc->requested_fec & FEC_RS)
5047 		MPASS(lc->pcaps & FW_PORT_CAP32_FEC_RS);
5048 	if (lc->requested_fec & FEC_BASER_RS)
5049 		MPASS(lc->pcaps & FW_PORT_CAP32_FEC_BASER_RS);
5050 #endif
5051 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5052 	if (rc != 0) {
5053 		/* Don't complain if the VF driver gets back an EPERM. */
5054 		if (!(sc->flags & IS_VF) || rc != FW_EPERM)
5055 			device_printf(pi->dev, "l1cfg failed: %d\n", rc);
5056 	} else {
5057 		/*
5058 		 * An L1_CFG will almost always result in a link-change event if
5059 		 * the link is up, and the driver will refresh the actual
5060 		 * fec/fc/etc. when the notification is processed.  If the link
5061 		 * is down then the actual settings are meaningless.
5062 		 *
5063 		 * This takes care of the case where a change in the L1 settings
5064 		 * may not result in a notification.
5065 		 */
5066 		if (lc->link_ok && !(lc->requested_fc & PAUSE_AUTONEG))
5067 			lc->fc = lc->requested_fc & (PAUSE_TX | PAUSE_RX);
5068 	}
5069 	return (rc);
5070 }
5071 
5072 #define FW_MAC_EXACT_CHUNK	7
5073 struct mcaddr_ctx {
5074 	struct ifnet *ifp;
5075 	const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
5076 	uint64_t hash;
5077 	int i;
5078 	int del;
5079 	int rc;
5080 };
5081 
5082 static u_int
5083 add_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
5084 {
5085 	struct mcaddr_ctx *ctx = arg;
5086 	struct vi_info *vi = ctx->ifp->if_softc;
5087 	struct port_info *pi = vi->pi;
5088 	struct adapter *sc = pi->adapter;
5089 
5090 	if (ctx->rc < 0)
5091 		return (0);
5092 
5093 	ctx->mcaddr[ctx->i] = LLADDR(sdl);
5094 	MPASS(ETHER_IS_MULTICAST(ctx->mcaddr[ctx->i]));
5095 	ctx->i++;
5096 
5097 	if (ctx->i == FW_MAC_EXACT_CHUNK) {
5098 		ctx->rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, ctx->del,
5099 		    ctx->i, ctx->mcaddr, NULL, &ctx->hash, 0);
5100 		if (ctx->rc < 0) {
5101 			int j;
5102 
5103 			for (j = 0; j < ctx->i; j++) {
5104 				if_printf(ctx->ifp,
5105 				    "failed to add mc address"
5106 				    " %02x:%02x:%02x:"
5107 				    "%02x:%02x:%02x rc=%d\n",
5108 				    ctx->mcaddr[j][0], ctx->mcaddr[j][1],
5109 				    ctx->mcaddr[j][2], ctx->mcaddr[j][3],
5110 				    ctx->mcaddr[j][4], ctx->mcaddr[j][5],
5111 				    -ctx->rc);
5112 			}
5113 			return (0);
5114 		}
5115 		ctx->del = 0;
5116 		ctx->i = 0;
5117 	}
5118 
5119 	return (1);
5120 }
5121 
5122 /*
5123  * Program the port's XGMAC based on parameters in ifnet.  The caller also
5124  * indicates which parameters should be programmed (the rest are left alone).
5125  */
5126 int
5127 update_mac_settings(struct ifnet *ifp, int flags)
5128 {
5129 	int rc = 0;
5130 	struct vi_info *vi = ifp->if_softc;
5131 	struct port_info *pi = vi->pi;
5132 	struct adapter *sc = pi->adapter;
5133 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
5134 
5135 	ASSERT_SYNCHRONIZED_OP(sc);
5136 	KASSERT(flags, ("%s: not told what to update.", __func__));
5137 
5138 	if (flags & XGMAC_MTU)
5139 		mtu = ifp->if_mtu;
5140 
5141 	if (flags & XGMAC_PROMISC)
5142 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
5143 
5144 	if (flags & XGMAC_ALLMULTI)
5145 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
5146 
5147 	if (flags & XGMAC_VLANEX)
5148 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
5149 
5150 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
5151 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
5152 		    allmulti, 1, vlanex, false);
5153 		if (rc) {
5154 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
5155 			    rc);
5156 			return (rc);
5157 		}
5158 	}
5159 
5160 	if (flags & XGMAC_UCADDR) {
5161 		uint8_t ucaddr[ETHER_ADDR_LEN];
5162 
5163 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
5164 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
5165 		    ucaddr, true, &vi->smt_idx);
5166 		if (rc < 0) {
5167 			rc = -rc;
5168 			if_printf(ifp, "change_mac failed: %d\n", rc);
5169 			return (rc);
5170 		} else {
5171 			vi->xact_addr_filt = rc;
5172 			rc = 0;
5173 		}
5174 	}
5175 
5176 	if (flags & XGMAC_MCADDRS) {
5177 		struct epoch_tracker et;
5178 		struct mcaddr_ctx ctx;
5179 		int j;
5180 
5181 		ctx.ifp = ifp;
5182 		ctx.hash = 0;
5183 		ctx.i = 0;
5184 		ctx.del = 1;
5185 		ctx.rc = 0;
5186 		/*
5187 		 * Unlike other drivers, we accumulate list of pointers into
5188 		 * interface address lists and we need to keep it safe even
5189 		 * after if_foreach_llmaddr() returns, thus we must enter the
5190 		 * network epoch.
5191 		 */
5192 		NET_EPOCH_ENTER(et);
5193 		if_foreach_llmaddr(ifp, add_maddr, &ctx);
5194 		if (ctx.rc < 0) {
5195 			NET_EPOCH_EXIT(et);
5196 			rc = -ctx.rc;
5197 			return (rc);
5198 		}
5199 		if (ctx.i > 0) {
5200 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
5201 			    ctx.del, ctx.i, ctx.mcaddr, NULL, &ctx.hash, 0);
5202 			NET_EPOCH_EXIT(et);
5203 			if (rc < 0) {
5204 				rc = -rc;
5205 				for (j = 0; j < ctx.i; j++) {
5206 					if_printf(ifp,
5207 					    "failed to add mc address"
5208 					    " %02x:%02x:%02x:"
5209 					    "%02x:%02x:%02x rc=%d\n",
5210 					    ctx.mcaddr[j][0], ctx.mcaddr[j][1],
5211 					    ctx.mcaddr[j][2], ctx.mcaddr[j][3],
5212 					    ctx.mcaddr[j][4], ctx.mcaddr[j][5],
5213 					    rc);
5214 				}
5215 				return (rc);
5216 			}
5217 		} else
5218 			NET_EPOCH_EXIT(et);
5219 
5220 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, ctx.hash, 0);
5221 		if (rc != 0)
5222 			if_printf(ifp, "failed to set mc address hash: %d", rc);
5223 	}
5224 
5225 	return (rc);
5226 }
5227 
5228 /*
5229  * {begin|end}_synchronized_op must be called from the same thread.
5230  */
5231 int
5232 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
5233     char *wmesg)
5234 {
5235 	int rc, pri;
5236 
5237 #ifdef WITNESS
5238 	/* the caller thinks it's ok to sleep, but is it really? */
5239 	if (flags & SLEEP_OK)
5240 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
5241 		    "begin_synchronized_op");
5242 #endif
5243 
5244 	if (INTR_OK)
5245 		pri = PCATCH;
5246 	else
5247 		pri = 0;
5248 
5249 	ADAPTER_LOCK(sc);
5250 	for (;;) {
5251 
5252 		if (vi && IS_DOOMED(vi)) {
5253 			rc = ENXIO;
5254 			goto done;
5255 		}
5256 
5257 		if (!IS_BUSY(sc)) {
5258 			rc = 0;
5259 			break;
5260 		}
5261 
5262 		if (!(flags & SLEEP_OK)) {
5263 			rc = EBUSY;
5264 			goto done;
5265 		}
5266 
5267 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
5268 			rc = EINTR;
5269 			goto done;
5270 		}
5271 	}
5272 
5273 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
5274 	SET_BUSY(sc);
5275 #ifdef INVARIANTS
5276 	sc->last_op = wmesg;
5277 	sc->last_op_thr = curthread;
5278 	sc->last_op_flags = flags;
5279 #endif
5280 
5281 done:
5282 	if (!(flags & HOLD_LOCK) || rc)
5283 		ADAPTER_UNLOCK(sc);
5284 
5285 	return (rc);
5286 }
5287 
5288 /*
5289  * Tell if_ioctl and if_init that the VI is going away.  This is
5290  * special variant of begin_synchronized_op and must be paired with a
5291  * call to end_synchronized_op.
5292  */
5293 void
5294 doom_vi(struct adapter *sc, struct vi_info *vi)
5295 {
5296 
5297 	ADAPTER_LOCK(sc);
5298 	SET_DOOMED(vi);
5299 	wakeup(&sc->flags);
5300 	while (IS_BUSY(sc))
5301 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
5302 	SET_BUSY(sc);
5303 #ifdef INVARIANTS
5304 	sc->last_op = "t4detach";
5305 	sc->last_op_thr = curthread;
5306 	sc->last_op_flags = 0;
5307 #endif
5308 	ADAPTER_UNLOCK(sc);
5309 }
5310 
5311 /*
5312  * {begin|end}_synchronized_op must be called from the same thread.
5313  */
5314 void
5315 end_synchronized_op(struct adapter *sc, int flags)
5316 {
5317 
5318 	if (flags & LOCK_HELD)
5319 		ADAPTER_LOCK_ASSERT_OWNED(sc);
5320 	else
5321 		ADAPTER_LOCK(sc);
5322 
5323 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
5324 	CLR_BUSY(sc);
5325 	wakeup(&sc->flags);
5326 	ADAPTER_UNLOCK(sc);
5327 }
5328 
5329 static int
5330 cxgbe_init_synchronized(struct vi_info *vi)
5331 {
5332 	struct port_info *pi = vi->pi;
5333 	struct adapter *sc = pi->adapter;
5334 	struct ifnet *ifp = vi->ifp;
5335 	int rc = 0, i;
5336 	struct sge_txq *txq;
5337 
5338 	ASSERT_SYNCHRONIZED_OP(sc);
5339 
5340 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
5341 		return (0);	/* already running */
5342 
5343 	if (!(sc->flags & FULL_INIT_DONE) &&
5344 	    ((rc = adapter_full_init(sc)) != 0))
5345 		return (rc);	/* error message displayed already */
5346 
5347 	if (!(vi->flags & VI_INIT_DONE) &&
5348 	    ((rc = vi_full_init(vi)) != 0))
5349 		return (rc); /* error message displayed already */
5350 
5351 	rc = update_mac_settings(ifp, XGMAC_ALL);
5352 	if (rc)
5353 		goto done;	/* error message displayed already */
5354 
5355 	PORT_LOCK(pi);
5356 	if (pi->up_vis == 0) {
5357 		t4_update_port_info(pi);
5358 		fixup_link_config(pi);
5359 		build_medialist(pi);
5360 		apply_link_config(pi);
5361 	}
5362 
5363 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
5364 	if (rc != 0) {
5365 		if_printf(ifp, "enable_vi failed: %d\n", rc);
5366 		PORT_UNLOCK(pi);
5367 		goto done;
5368 	}
5369 
5370 	/*
5371 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
5372 	 * if this changes.
5373 	 */
5374 
5375 	for_each_txq(vi, i, txq) {
5376 		TXQ_LOCK(txq);
5377 		txq->eq.flags |= EQ_ENABLED;
5378 		TXQ_UNLOCK(txq);
5379 	}
5380 
5381 	/*
5382 	 * The first iq of the first port to come up is used for tracing.
5383 	 */
5384 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
5385 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
5386 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
5387 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
5388 		    V_QUEUENUMBER(sc->traceq));
5389 		pi->flags |= HAS_TRACEQ;
5390 	}
5391 
5392 	/* all ok */
5393 	pi->up_vis++;
5394 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
5395 
5396 	if (pi->nvi > 1 || sc->flags & IS_VF)
5397 		callout_reset(&vi->tick, hz, vi_tick, vi);
5398 	else
5399 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
5400 	if (pi->link_cfg.link_ok)
5401 		t4_os_link_changed(pi);
5402 	PORT_UNLOCK(pi);
5403 done:
5404 	if (rc != 0)
5405 		cxgbe_uninit_synchronized(vi);
5406 
5407 	return (rc);
5408 }
5409 
5410 /*
5411  * Idempotent.
5412  */
5413 static int
5414 cxgbe_uninit_synchronized(struct vi_info *vi)
5415 {
5416 	struct port_info *pi = vi->pi;
5417 	struct adapter *sc = pi->adapter;
5418 	struct ifnet *ifp = vi->ifp;
5419 	int rc, i;
5420 	struct sge_txq *txq;
5421 
5422 	ASSERT_SYNCHRONIZED_OP(sc);
5423 
5424 	if (!(vi->flags & VI_INIT_DONE)) {
5425 		if (__predict_false(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
5426 			KASSERT(0, ("uninited VI is running"));
5427 			if_printf(ifp, "uninited VI with running ifnet.  "
5428 			    "vi->flags 0x%016lx, if_flags 0x%08x, "
5429 			    "if_drv_flags 0x%08x\n", vi->flags, ifp->if_flags,
5430 			    ifp->if_drv_flags);
5431 		}
5432 		return (0);
5433 	}
5434 
5435 	/*
5436 	 * Disable the VI so that all its data in either direction is discarded
5437 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
5438 	 * tick) intact as the TP can deliver negative advice or data that it's
5439 	 * holding in its RAM (for an offloaded connection) even after the VI is
5440 	 * disabled.
5441 	 */
5442 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
5443 	if (rc) {
5444 		if_printf(ifp, "disable_vi failed: %d\n", rc);
5445 		return (rc);
5446 	}
5447 
5448 	for_each_txq(vi, i, txq) {
5449 		TXQ_LOCK(txq);
5450 		txq->eq.flags &= ~EQ_ENABLED;
5451 		TXQ_UNLOCK(txq);
5452 	}
5453 
5454 	PORT_LOCK(pi);
5455 	if (pi->nvi > 1 || sc->flags & IS_VF)
5456 		callout_stop(&vi->tick);
5457 	else
5458 		callout_stop(&pi->tick);
5459 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
5460 		PORT_UNLOCK(pi);
5461 		return (0);
5462 	}
5463 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
5464 	pi->up_vis--;
5465 	if (pi->up_vis > 0) {
5466 		PORT_UNLOCK(pi);
5467 		return (0);
5468 	}
5469 
5470 	pi->link_cfg.link_ok = false;
5471 	pi->link_cfg.speed = 0;
5472 	pi->link_cfg.link_down_rc = 255;
5473 	t4_os_link_changed(pi);
5474 	PORT_UNLOCK(pi);
5475 
5476 	return (0);
5477 }
5478 
5479 /*
5480  * It is ok for this function to fail midway and return right away.  t4_detach
5481  * will walk the entire sc->irq list and clean up whatever is valid.
5482  */
5483 int
5484 t4_setup_intr_handlers(struct adapter *sc)
5485 {
5486 	int rc, rid, p, q, v;
5487 	char s[8];
5488 	struct irq *irq;
5489 	struct port_info *pi;
5490 	struct vi_info *vi;
5491 	struct sge *sge = &sc->sge;
5492 	struct sge_rxq *rxq;
5493 #ifdef TCP_OFFLOAD
5494 	struct sge_ofld_rxq *ofld_rxq;
5495 #endif
5496 #ifdef DEV_NETMAP
5497 	struct sge_nm_rxq *nm_rxq;
5498 #endif
5499 #ifdef RSS
5500 	int nbuckets = rss_getnumbuckets();
5501 #endif
5502 
5503 	/*
5504 	 * Setup interrupts.
5505 	 */
5506 	irq = &sc->irq[0];
5507 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
5508 	if (forwarding_intr_to_fwq(sc))
5509 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
5510 
5511 	/* Multiple interrupts. */
5512 	if (sc->flags & IS_VF)
5513 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
5514 		    ("%s: too few intr.", __func__));
5515 	else
5516 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
5517 		    ("%s: too few intr.", __func__));
5518 
5519 	/* The first one is always error intr on PFs */
5520 	if (!(sc->flags & IS_VF)) {
5521 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
5522 		if (rc != 0)
5523 			return (rc);
5524 		irq++;
5525 		rid++;
5526 	}
5527 
5528 	/* The second one is always the firmware event queue (first on VFs) */
5529 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
5530 	if (rc != 0)
5531 		return (rc);
5532 	irq++;
5533 	rid++;
5534 
5535 	for_each_port(sc, p) {
5536 		pi = sc->port[p];
5537 		for_each_vi(pi, v, vi) {
5538 			vi->first_intr = rid - 1;
5539 
5540 			if (vi->nnmrxq > 0) {
5541 				int n = max(vi->nrxq, vi->nnmrxq);
5542 
5543 				rxq = &sge->rxq[vi->first_rxq];
5544 #ifdef DEV_NETMAP
5545 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
5546 #endif
5547 				for (q = 0; q < n; q++) {
5548 					snprintf(s, sizeof(s), "%x%c%x", p,
5549 					    'a' + v, q);
5550 					if (q < vi->nrxq)
5551 						irq->rxq = rxq++;
5552 #ifdef DEV_NETMAP
5553 					if (q < vi->nnmrxq)
5554 						irq->nm_rxq = nm_rxq++;
5555 
5556 					if (irq->nm_rxq != NULL &&
5557 					    irq->rxq == NULL) {
5558 						/* Netmap rx only */
5559 						rc = t4_alloc_irq(sc, irq, rid,
5560 						    t4_nm_intr, irq->nm_rxq, s);
5561 					}
5562 					if (irq->nm_rxq != NULL &&
5563 					    irq->rxq != NULL) {
5564 						/* NIC and Netmap rx */
5565 						rc = t4_alloc_irq(sc, irq, rid,
5566 						    t4_vi_intr, irq, s);
5567 					}
5568 #endif
5569 					if (irq->rxq != NULL &&
5570 					    irq->nm_rxq == NULL) {
5571 						/* NIC rx only */
5572 						rc = t4_alloc_irq(sc, irq, rid,
5573 						    t4_intr, irq->rxq, s);
5574 					}
5575 					if (rc != 0)
5576 						return (rc);
5577 #ifdef RSS
5578 					if (q < vi->nrxq) {
5579 						bus_bind_intr(sc->dev, irq->res,
5580 						    rss_getcpu(q % nbuckets));
5581 					}
5582 #endif
5583 					irq++;
5584 					rid++;
5585 					vi->nintr++;
5586 				}
5587 			} else {
5588 				for_each_rxq(vi, q, rxq) {
5589 					snprintf(s, sizeof(s), "%x%c%x", p,
5590 					    'a' + v, q);
5591 					rc = t4_alloc_irq(sc, irq, rid,
5592 					    t4_intr, rxq, s);
5593 					if (rc != 0)
5594 						return (rc);
5595 #ifdef RSS
5596 					bus_bind_intr(sc->dev, irq->res,
5597 					    rss_getcpu(q % nbuckets));
5598 #endif
5599 					irq++;
5600 					rid++;
5601 					vi->nintr++;
5602 				}
5603 			}
5604 #ifdef TCP_OFFLOAD
5605 			for_each_ofld_rxq(vi, q, ofld_rxq) {
5606 				snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
5607 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
5608 				    ofld_rxq, s);
5609 				if (rc != 0)
5610 					return (rc);
5611 				irq++;
5612 				rid++;
5613 				vi->nintr++;
5614 			}
5615 #endif
5616 		}
5617 	}
5618 	MPASS(irq == &sc->irq[sc->intr_count]);
5619 
5620 	return (0);
5621 }
5622 
5623 int
5624 adapter_full_init(struct adapter *sc)
5625 {
5626 	int rc, i;
5627 #ifdef RSS
5628 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5629 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5630 #endif
5631 
5632 	ASSERT_SYNCHRONIZED_OP(sc);
5633 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5634 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
5635 	    ("%s: FULL_INIT_DONE already", __func__));
5636 
5637 	/*
5638 	 * queues that belong to the adapter (not any particular port).
5639 	 */
5640 	rc = t4_setup_adapter_queues(sc);
5641 	if (rc != 0)
5642 		goto done;
5643 
5644 	for (i = 0; i < nitems(sc->tq); i++) {
5645 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
5646 		    taskqueue_thread_enqueue, &sc->tq[i]);
5647 		if (sc->tq[i] == NULL) {
5648 			device_printf(sc->dev,
5649 			    "failed to allocate task queue %d\n", i);
5650 			rc = ENOMEM;
5651 			goto done;
5652 		}
5653 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
5654 		    device_get_nameunit(sc->dev), i);
5655 	}
5656 #ifdef RSS
5657 	MPASS(RSS_KEYSIZE == 40);
5658 	rss_getkey((void *)&raw_rss_key[0]);
5659 	for (i = 0; i < nitems(rss_key); i++) {
5660 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
5661 	}
5662 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
5663 #endif
5664 
5665 	if (!(sc->flags & IS_VF))
5666 		t4_intr_enable(sc);
5667 #ifdef KERN_TLS
5668 	if (sc->flags & KERN_TLS_OK)
5669 		callout_reset_sbt(&sc->ktls_tick, SBT_1MS, 0, ktls_tick, sc,
5670 		    C_HARDCLOCK);
5671 #endif
5672 	sc->flags |= FULL_INIT_DONE;
5673 done:
5674 	if (rc != 0)
5675 		adapter_full_uninit(sc);
5676 
5677 	return (rc);
5678 }
5679 
5680 int
5681 adapter_full_uninit(struct adapter *sc)
5682 {
5683 	int i;
5684 
5685 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5686 
5687 	t4_teardown_adapter_queues(sc);
5688 
5689 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
5690 		taskqueue_free(sc->tq[i]);
5691 		sc->tq[i] = NULL;
5692 	}
5693 
5694 	sc->flags &= ~FULL_INIT_DONE;
5695 
5696 	return (0);
5697 }
5698 
5699 #ifdef RSS
5700 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
5701     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
5702     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
5703     RSS_HASHTYPE_RSS_UDP_IPV6)
5704 
5705 /* Translates kernel hash types to hardware. */
5706 static int
5707 hashconfig_to_hashen(int hashconfig)
5708 {
5709 	int hashen = 0;
5710 
5711 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
5712 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
5713 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
5714 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
5715 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
5716 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5717 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5718 	}
5719 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
5720 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5721 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5722 	}
5723 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
5724 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5725 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
5726 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5727 
5728 	return (hashen);
5729 }
5730 
5731 /* Translates hardware hash types to kernel. */
5732 static int
5733 hashen_to_hashconfig(int hashen)
5734 {
5735 	int hashconfig = 0;
5736 
5737 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
5738 		/*
5739 		 * If UDP hashing was enabled it must have been enabled for
5740 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
5741 		 * enabling any 4-tuple hash is nonsense configuration.
5742 		 */
5743 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5744 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
5745 
5746 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5747 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
5748 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5749 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
5750 	}
5751 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5752 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
5753 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5754 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
5755 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
5756 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
5757 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
5758 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
5759 
5760 	return (hashconfig);
5761 }
5762 #endif
5763 
5764 int
5765 vi_full_init(struct vi_info *vi)
5766 {
5767 	struct adapter *sc = vi->pi->adapter;
5768 	struct ifnet *ifp = vi->ifp;
5769 	uint16_t *rss;
5770 	struct sge_rxq *rxq;
5771 	int rc, i, j;
5772 #ifdef RSS
5773 	int nbuckets = rss_getnumbuckets();
5774 	int hashconfig = rss_gethashconfig();
5775 	int extra;
5776 #endif
5777 
5778 	ASSERT_SYNCHRONIZED_OP(sc);
5779 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
5780 	    ("%s: VI_INIT_DONE already", __func__));
5781 
5782 	sysctl_ctx_init(&vi->ctx);
5783 	vi->flags |= VI_SYSCTL_CTX;
5784 
5785 	/*
5786 	 * Allocate tx/rx/fl queues for this VI.
5787 	 */
5788 	rc = t4_setup_vi_queues(vi);
5789 	if (rc != 0)
5790 		goto done;	/* error message displayed already */
5791 
5792 	/*
5793 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
5794 	 */
5795 	if (vi->nrxq > vi->rss_size) {
5796 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
5797 		    "some queues will never receive traffic.\n", vi->nrxq,
5798 		    vi->rss_size);
5799 	} else if (vi->rss_size % vi->nrxq) {
5800 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
5801 		    "expect uneven traffic distribution.\n", vi->nrxq,
5802 		    vi->rss_size);
5803 	}
5804 #ifdef RSS
5805 	if (vi->nrxq != nbuckets) {
5806 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
5807 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
5808 	}
5809 #endif
5810 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
5811 	for (i = 0; i < vi->rss_size;) {
5812 #ifdef RSS
5813 		j = rss_get_indirection_to_bucket(i);
5814 		j %= vi->nrxq;
5815 		rxq = &sc->sge.rxq[vi->first_rxq + j];
5816 		rss[i++] = rxq->iq.abs_id;
5817 #else
5818 		for_each_rxq(vi, j, rxq) {
5819 			rss[i++] = rxq->iq.abs_id;
5820 			if (i == vi->rss_size)
5821 				break;
5822 		}
5823 #endif
5824 	}
5825 
5826 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
5827 	    vi->rss_size);
5828 	if (rc != 0) {
5829 		free(rss, M_CXGBE);
5830 		if_printf(ifp, "rss_config failed: %d\n", rc);
5831 		goto done;
5832 	}
5833 
5834 #ifdef RSS
5835 	vi->hashen = hashconfig_to_hashen(hashconfig);
5836 
5837 	/*
5838 	 * We may have had to enable some hashes even though the global config
5839 	 * wants them disabled.  This is a potential problem that must be
5840 	 * reported to the user.
5841 	 */
5842 	extra = hashen_to_hashconfig(vi->hashen) ^ hashconfig;
5843 
5844 	/*
5845 	 * If we consider only the supported hash types, then the enabled hashes
5846 	 * are a superset of the requested hashes.  In other words, there cannot
5847 	 * be any supported hash that was requested but not enabled, but there
5848 	 * can be hashes that were not requested but had to be enabled.
5849 	 */
5850 	extra &= SUPPORTED_RSS_HASHTYPES;
5851 	MPASS((extra & hashconfig) == 0);
5852 
5853 	if (extra) {
5854 		if_printf(ifp,
5855 		    "global RSS config (0x%x) cannot be accommodated.\n",
5856 		    hashconfig);
5857 	}
5858 	if (extra & RSS_HASHTYPE_RSS_IPV4)
5859 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
5860 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
5861 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
5862 	if (extra & RSS_HASHTYPE_RSS_IPV6)
5863 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
5864 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
5865 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
5866 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
5867 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
5868 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
5869 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
5870 #else
5871 	vi->hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
5872 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
5873 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5874 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
5875 #endif
5876 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, vi->hashen, rss[0], 0, 0);
5877 	if (rc != 0) {
5878 		free(rss, M_CXGBE);
5879 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
5880 		goto done;
5881 	}
5882 
5883 	vi->rss = rss;
5884 	vi->flags |= VI_INIT_DONE;
5885 done:
5886 	if (rc != 0)
5887 		vi_full_uninit(vi);
5888 
5889 	return (rc);
5890 }
5891 
5892 /*
5893  * Idempotent.
5894  */
5895 int
5896 vi_full_uninit(struct vi_info *vi)
5897 {
5898 	struct port_info *pi = vi->pi;
5899 	struct adapter *sc = pi->adapter;
5900 	int i;
5901 	struct sge_rxq *rxq;
5902 	struct sge_txq *txq;
5903 #ifdef TCP_OFFLOAD
5904 	struct sge_ofld_rxq *ofld_rxq;
5905 #endif
5906 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
5907 	struct sge_wrq *ofld_txq;
5908 #endif
5909 
5910 	if (vi->flags & VI_INIT_DONE) {
5911 
5912 		/* Need to quiesce queues.  */
5913 
5914 		/* XXX: Only for the first VI? */
5915 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
5916 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
5917 
5918 		for_each_txq(vi, i, txq) {
5919 			quiesce_txq(sc, txq);
5920 		}
5921 
5922 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
5923 		for_each_ofld_txq(vi, i, ofld_txq) {
5924 			quiesce_wrq(sc, ofld_txq);
5925 		}
5926 #endif
5927 
5928 		for_each_rxq(vi, i, rxq) {
5929 			quiesce_iq(sc, &rxq->iq);
5930 			quiesce_fl(sc, &rxq->fl);
5931 		}
5932 
5933 #ifdef TCP_OFFLOAD
5934 		for_each_ofld_rxq(vi, i, ofld_rxq) {
5935 			quiesce_iq(sc, &ofld_rxq->iq);
5936 			quiesce_fl(sc, &ofld_rxq->fl);
5937 		}
5938 #endif
5939 		free(vi->rss, M_CXGBE);
5940 		free(vi->nm_rss, M_CXGBE);
5941 	}
5942 
5943 	t4_teardown_vi_queues(vi);
5944 	vi->flags &= ~VI_INIT_DONE;
5945 
5946 	return (0);
5947 }
5948 
5949 static void
5950 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
5951 {
5952 	struct sge_eq *eq = &txq->eq;
5953 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
5954 
5955 	(void) sc;	/* unused */
5956 
5957 #ifdef INVARIANTS
5958 	TXQ_LOCK(txq);
5959 	MPASS((eq->flags & EQ_ENABLED) == 0);
5960 	TXQ_UNLOCK(txq);
5961 #endif
5962 
5963 	/* Wait for the mp_ring to empty. */
5964 	while (!mp_ring_is_idle(txq->r)) {
5965 		mp_ring_check_drainage(txq->r, 0);
5966 		pause("rquiesce", 1);
5967 	}
5968 
5969 	/* Then wait for the hardware to finish. */
5970 	while (spg->cidx != htobe16(eq->pidx))
5971 		pause("equiesce", 1);
5972 
5973 	/* Finally, wait for the driver to reclaim all descriptors. */
5974 	while (eq->cidx != eq->pidx)
5975 		pause("dquiesce", 1);
5976 }
5977 
5978 static void
5979 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
5980 {
5981 
5982 	/* XXXTX */
5983 }
5984 
5985 static void
5986 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
5987 {
5988 	(void) sc;	/* unused */
5989 
5990 	/* Synchronize with the interrupt handler */
5991 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
5992 		pause("iqfree", 1);
5993 }
5994 
5995 static void
5996 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
5997 {
5998 	mtx_lock(&sc->sfl_lock);
5999 	FL_LOCK(fl);
6000 	fl->flags |= FL_DOOMED;
6001 	FL_UNLOCK(fl);
6002 	callout_stop(&sc->sfl_callout);
6003 	mtx_unlock(&sc->sfl_lock);
6004 
6005 	KASSERT((fl->flags & FL_STARVING) == 0,
6006 	    ("%s: still starving", __func__));
6007 }
6008 
6009 static int
6010 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
6011     driver_intr_t *handler, void *arg, char *name)
6012 {
6013 	int rc;
6014 
6015 	irq->rid = rid;
6016 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
6017 	    RF_SHAREABLE | RF_ACTIVE);
6018 	if (irq->res == NULL) {
6019 		device_printf(sc->dev,
6020 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
6021 		return (ENOMEM);
6022 	}
6023 
6024 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
6025 	    NULL, handler, arg, &irq->tag);
6026 	if (rc != 0) {
6027 		device_printf(sc->dev,
6028 		    "failed to setup interrupt for rid %d, name %s: %d\n",
6029 		    rid, name, rc);
6030 	} else if (name)
6031 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
6032 
6033 	return (rc);
6034 }
6035 
6036 static int
6037 t4_free_irq(struct adapter *sc, struct irq *irq)
6038 {
6039 	if (irq->tag)
6040 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
6041 	if (irq->res)
6042 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
6043 
6044 	bzero(irq, sizeof(*irq));
6045 
6046 	return (0);
6047 }
6048 
6049 static void
6050 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
6051 {
6052 
6053 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
6054 	t4_get_regs(sc, buf, regs->len);
6055 }
6056 
6057 #define	A_PL_INDIR_CMD	0x1f8
6058 
6059 #define	S_PL_AUTOINC	31
6060 #define	M_PL_AUTOINC	0x1U
6061 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
6062 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
6063 
6064 #define	S_PL_VFID	20
6065 #define	M_PL_VFID	0xffU
6066 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
6067 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
6068 
6069 #define	S_PL_ADDR	0
6070 #define	M_PL_ADDR	0xfffffU
6071 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
6072 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
6073 
6074 #define	A_PL_INDIR_DATA	0x1fc
6075 
6076 static uint64_t
6077 read_vf_stat(struct adapter *sc, u_int vin, int reg)
6078 {
6079 	u32 stats[2];
6080 
6081 	mtx_assert(&sc->reg_lock, MA_OWNED);
6082 	if (sc->flags & IS_VF) {
6083 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
6084 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
6085 	} else {
6086 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
6087 		    V_PL_VFID(vin) | V_PL_ADDR(VF_MPS_REG(reg)));
6088 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
6089 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
6090 	}
6091 	return (((uint64_t)stats[1]) << 32 | stats[0]);
6092 }
6093 
6094 static void
6095 t4_get_vi_stats(struct adapter *sc, u_int vin, struct fw_vi_stats_vf *stats)
6096 {
6097 
6098 #define GET_STAT(name) \
6099 	read_vf_stat(sc, vin, A_MPS_VF_STAT_##name##_L)
6100 
6101 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
6102 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
6103 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
6104 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
6105 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
6106 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
6107 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
6108 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
6109 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
6110 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
6111 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
6112 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
6113 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
6114 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
6115 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
6116 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
6117 
6118 #undef GET_STAT
6119 }
6120 
6121 static void
6122 t4_clr_vi_stats(struct adapter *sc, u_int vin)
6123 {
6124 	int reg;
6125 
6126 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) | V_PL_VFID(vin) |
6127 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
6128 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
6129 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
6130 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
6131 }
6132 
6133 static void
6134 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
6135 {
6136 	struct timeval tv;
6137 	const struct timeval interval = {0, 250000};	/* 250ms */
6138 
6139 	if (!(vi->flags & VI_INIT_DONE))
6140 		return;
6141 
6142 	getmicrotime(&tv);
6143 	timevalsub(&tv, &interval);
6144 	if (timevalcmp(&tv, &vi->last_refreshed, <))
6145 		return;
6146 
6147 	mtx_lock(&sc->reg_lock);
6148 	t4_get_vi_stats(sc, vi->vin, &vi->stats);
6149 	getmicrotime(&vi->last_refreshed);
6150 	mtx_unlock(&sc->reg_lock);
6151 }
6152 
6153 static void
6154 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
6155 {
6156 	u_int i, v, tnl_cong_drops, chan_map;
6157 	struct timeval tv;
6158 	const struct timeval interval = {0, 250000};	/* 250ms */
6159 
6160 	getmicrotime(&tv);
6161 	timevalsub(&tv, &interval);
6162 	if (timevalcmp(&tv, &pi->last_refreshed, <))
6163 		return;
6164 
6165 	tnl_cong_drops = 0;
6166 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
6167 	chan_map = pi->rx_e_chan_map;
6168 	while (chan_map) {
6169 		i = ffs(chan_map) - 1;
6170 		mtx_lock(&sc->reg_lock);
6171 		t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
6172 		    A_TP_MIB_TNL_CNG_DROP_0 + i);
6173 		mtx_unlock(&sc->reg_lock);
6174 		tnl_cong_drops += v;
6175 		chan_map &= ~(1 << i);
6176 	}
6177 	pi->tnl_cong_drops = tnl_cong_drops;
6178 	getmicrotime(&pi->last_refreshed);
6179 }
6180 
6181 static void
6182 cxgbe_tick(void *arg)
6183 {
6184 	struct port_info *pi = arg;
6185 	struct adapter *sc = pi->adapter;
6186 
6187 	PORT_LOCK_ASSERT_OWNED(pi);
6188 	cxgbe_refresh_stats(sc, pi);
6189 
6190 	callout_schedule(&pi->tick, hz);
6191 }
6192 
6193 void
6194 vi_tick(void *arg)
6195 {
6196 	struct vi_info *vi = arg;
6197 	struct adapter *sc = vi->pi->adapter;
6198 
6199 	vi_refresh_stats(sc, vi);
6200 
6201 	callout_schedule(&vi->tick, hz);
6202 }
6203 
6204 /*
6205  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
6206  */
6207 static char *caps_decoder[] = {
6208 	"\20\001IPMI\002NCSI",				/* 0: NBM */
6209 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
6210 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
6211 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
6212 	    "\006HASHFILTER\007ETHOFLD",
6213 	"\20\001TOE",					/* 4: TOE */
6214 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
6215 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
6216 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
6217 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
6218 	    "\007T10DIF"
6219 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
6220 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
6221 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
6222 		    "\004PO_INITIATOR\005PO_TARGET",
6223 };
6224 
6225 void
6226 t4_sysctls(struct adapter *sc)
6227 {
6228 	struct sysctl_ctx_list *ctx;
6229 	struct sysctl_oid *oid;
6230 	struct sysctl_oid_list *children, *c0;
6231 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
6232 
6233 	ctx = device_get_sysctl_ctx(sc->dev);
6234 
6235 	/*
6236 	 * dev.t4nex.X.
6237 	 */
6238 	oid = device_get_sysctl_tree(sc->dev);
6239 	c0 = children = SYSCTL_CHILDREN(oid);
6240 
6241 	sc->sc_do_rxcopy = 1;
6242 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
6243 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
6244 
6245 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
6246 	    sc->params.nports, "# of ports");
6247 
6248 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
6249 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, doorbells,
6250 	    (uintptr_t)&sc->doorbells, sysctl_bitfield_8b, "A",
6251 	    "available doorbells");
6252 
6253 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
6254 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
6255 
6256 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
6257 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
6258 	    sc->params.sge.timer_val, sizeof(sc->params.sge.timer_val),
6259 	    sysctl_int_array, "A", "interrupt holdoff timer values (us)");
6260 
6261 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
6262 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
6263 	    sc->params.sge.counter_val, sizeof(sc->params.sge.counter_val),
6264 	    sysctl_int_array, "A", "interrupt holdoff packet counter values");
6265 
6266 	t4_sge_sysctls(sc, ctx, children);
6267 
6268 	sc->lro_timeout = 100;
6269 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
6270 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
6271 
6272 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
6273 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
6274 
6275 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
6276 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
6277 
6278 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
6279 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
6280 
6281 	if (sc->flags & IS_VF)
6282 		return;
6283 
6284 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
6285 	    NULL, chip_rev(sc), "chip hardware revision");
6286 
6287 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
6288 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
6289 
6290 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
6291 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
6292 
6293 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
6294 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
6295 
6296 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
6297 	    CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
6298 
6299 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
6300 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
6301 
6302 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
6303 	    sc->er_version, 0, "expansion ROM version");
6304 
6305 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
6306 	    sc->bs_version, 0, "bootstrap firmware version");
6307 
6308 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
6309 	    NULL, sc->params.scfg_vers, "serial config version");
6310 
6311 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
6312 	    NULL, sc->params.vpd_vers, "VPD version");
6313 
6314 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
6315 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
6316 
6317 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
6318 	    sc->cfcsum, "config file checksum");
6319 
6320 #define SYSCTL_CAP(name, n, text) \
6321 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
6322 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, caps_decoder[n], \
6323 	    (uintptr_t)&sc->name, sysctl_bitfield_16b, "A", \
6324 	    "available " text " capabilities")
6325 
6326 	SYSCTL_CAP(nbmcaps, 0, "NBM");
6327 	SYSCTL_CAP(linkcaps, 1, "link");
6328 	SYSCTL_CAP(switchcaps, 2, "switch");
6329 	SYSCTL_CAP(niccaps, 3, "NIC");
6330 	SYSCTL_CAP(toecaps, 4, "TCP offload");
6331 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
6332 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
6333 	SYSCTL_CAP(cryptocaps, 7, "crypto");
6334 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
6335 #undef SYSCTL_CAP
6336 
6337 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
6338 	    NULL, sc->tids.nftids, "number of filters");
6339 
6340 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
6341 	    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6342 	    sysctl_temperature, "I", "chip temperature (in Celsius)");
6343 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "reset_sensor",
6344 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, sc, 0,
6345 	    sysctl_reset_sensor, "I", "reset the chip's temperature sensor.");
6346 
6347 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "loadavg",
6348 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6349 	    sysctl_loadavg, "A",
6350 	    "microprocessor load averages (debug firmwares only)");
6351 
6352 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "core_vdd",
6353 	    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0, sysctl_vdd,
6354 	    "I", "core Vdd (in mV)");
6355 
6356 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "local_cpus",
6357 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, LOCAL_CPUS,
6358 	    sysctl_cpus, "A", "local CPUs");
6359 
6360 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "intr_cpus",
6361 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, INTR_CPUS,
6362 	    sysctl_cpus, "A", "preferred CPUs for interrupts");
6363 
6364 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "swintr", CTLFLAG_RW,
6365 	    &sc->swintr, 0, "software triggered interrupts");
6366 
6367 	/*
6368 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
6369 	 */
6370 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
6371 	    CTLFLAG_RD | CTLFLAG_SKIP | CTLFLAG_MPSAFE, NULL,
6372 	    "logs and miscellaneous information");
6373 	children = SYSCTL_CHILDREN(oid);
6374 
6375 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
6376 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6377 	    sysctl_cctrl, "A", "congestion control");
6378 
6379 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
6380 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6381 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
6382 
6383 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
6384 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 1,
6385 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
6386 
6387 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
6388 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 2,
6389 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
6390 
6391 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
6392 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 3,
6393 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
6394 
6395 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
6396 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 4,
6397 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
6398 
6399 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
6400 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 5,
6401 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
6402 
6403 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
6404 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6405 	    sysctl_cim_la, "A", "CIM logic analyzer");
6406 
6407 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
6408 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6409 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
6410 
6411 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
6412 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6413 	    0 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
6414 
6415 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
6416 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6417 	    1 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
6418 
6419 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
6420 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6421 	    2 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
6422 
6423 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
6424 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6425 	    3 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
6426 
6427 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
6428 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6429 	    4 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
6430 
6431 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
6432 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6433 	    5 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
6434 
6435 	if (chip_id(sc) > CHELSIO_T4) {
6436 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
6437 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6438 		    6 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A",
6439 		    "CIM OBQ 6 (SGE0-RX)");
6440 
6441 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
6442 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6443 		    7 + CIM_NUM_IBQ, sysctl_cim_ibq_obq, "A",
6444 		    "CIM OBQ 7 (SGE1-RX)");
6445 	}
6446 
6447 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
6448 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6449 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
6450 
6451 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
6452 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6453 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
6454 
6455 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
6456 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6457 	    sysctl_cpl_stats, "A", "CPL statistics");
6458 
6459 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
6460 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6461 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
6462 
6463 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
6464 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6465 	    sysctl_devlog, "A", "firmware's device log");
6466 
6467 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
6468 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6469 	    sysctl_fcoe_stats, "A", "FCoE statistics");
6470 
6471 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
6472 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6473 	    sysctl_hw_sched, "A", "hardware scheduler ");
6474 
6475 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
6476 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6477 	    sysctl_l2t, "A", "hardware L2 table");
6478 
6479 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "smt",
6480 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6481 	    sysctl_smt, "A", "hardware source MAC table");
6482 
6483 #ifdef INET6
6484 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "clip",
6485 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6486 	    sysctl_clip, "A", "active CLIP table entries");
6487 #endif
6488 
6489 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
6490 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6491 	    sysctl_lb_stats, "A", "loopback statistics");
6492 
6493 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
6494 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6495 	    sysctl_meminfo, "A", "memory regions");
6496 
6497 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
6498 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6499 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
6500 	    "A", "MPS TCAM entries");
6501 
6502 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
6503 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6504 	    sysctl_path_mtus, "A", "path MTUs");
6505 
6506 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
6507 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6508 	    sysctl_pm_stats, "A", "PM statistics");
6509 
6510 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
6511 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6512 	    sysctl_rdma_stats, "A", "RDMA statistics");
6513 
6514 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
6515 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6516 	    sysctl_tcp_stats, "A", "TCP statistics");
6517 
6518 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
6519 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6520 	    sysctl_tids, "A", "TID information");
6521 
6522 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
6523 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6524 	    sysctl_tp_err_stats, "A", "TP error statistics");
6525 
6526 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
6527 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, sc, 0,
6528 	    sysctl_tp_la_mask, "I", "TP logic analyzer event capture mask");
6529 
6530 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
6531 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6532 	    sysctl_tp_la, "A", "TP logic analyzer");
6533 
6534 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
6535 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6536 	    sysctl_tx_rate, "A", "Tx rate");
6537 
6538 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
6539 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6540 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
6541 
6542 	if (chip_id(sc) >= CHELSIO_T5) {
6543 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
6544 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6545 		    sysctl_wcwr_stats, "A", "write combined work requests");
6546 	}
6547 
6548 #ifdef KERN_TLS
6549 	if (sc->flags & KERN_TLS_OK) {
6550 		/*
6551 		 * dev.t4nex.0.tls.
6552 		 */
6553 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "tls",
6554 		    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "KERN_TLS parameters");
6555 		children = SYSCTL_CHILDREN(oid);
6556 
6557 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "inline_keys",
6558 		    CTLFLAG_RW, &sc->tlst.inline_keys, 0, "Always pass TLS "
6559 		    "keys in work requests (1) or attempt to store TLS keys "
6560 		    "in card memory.");
6561 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "combo_wrs",
6562 		    CTLFLAG_RW, &sc->tlst.combo_wrs, 0, "Attempt to combine "
6563 		    "TCB field updates with TLS record work requests.");
6564 	}
6565 #endif
6566 
6567 #ifdef TCP_OFFLOAD
6568 	if (is_offload(sc)) {
6569 		int i;
6570 		char s[4];
6571 
6572 		/*
6573 		 * dev.t4nex.X.toe.
6574 		 */
6575 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe",
6576 		    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "TOE parameters");
6577 		children = SYSCTL_CHILDREN(oid);
6578 
6579 		sc->tt.cong_algorithm = -1;
6580 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
6581 		    CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
6582 		    "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
6583 		    "3 = highspeed)");
6584 
6585 		sc->tt.sndbuf = -1;
6586 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
6587 		    &sc->tt.sndbuf, 0, "hardware send buffer");
6588 
6589 		sc->tt.ddp = 0;
6590 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp",
6591 		    CTLFLAG_RW | CTLFLAG_SKIP, &sc->tt.ddp, 0, "");
6592 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_zcopy", CTLFLAG_RW,
6593 		    &sc->tt.ddp, 0, "Enable zero-copy aio_read(2)");
6594 
6595 		sc->tt.rx_coalesce = -1;
6596 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
6597 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
6598 
6599 		sc->tt.tls = 0;
6600 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tls", CTLFLAG_RW,
6601 		    &sc->tt.tls, 0, "Inline TLS allowed");
6602 
6603 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls_rx_ports",
6604 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, sc, 0,
6605 		    sysctl_tls_rx_ports, "I",
6606 		    "TCP ports that use inline TLS+TOE RX");
6607 
6608 		sc->tt.tx_align = -1;
6609 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
6610 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
6611 
6612 		sc->tt.tx_zcopy = 0;
6613 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
6614 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
6615 		    "Enable zero-copy aio_write(2)");
6616 
6617 		sc->tt.cop_managed_offloading = !!t4_cop_managed_offloading;
6618 		SYSCTL_ADD_INT(ctx, children, OID_AUTO,
6619 		    "cop_managed_offloading", CTLFLAG_RW,
6620 		    &sc->tt.cop_managed_offloading, 0,
6621 		    "COP (Connection Offload Policy) controls all TOE offload");
6622 
6623 		sc->tt.autorcvbuf_inc = 16 * 1024;
6624 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "autorcvbuf_inc",
6625 		    CTLFLAG_RW, &sc->tt.autorcvbuf_inc, 0,
6626 		    "autorcvbuf increment");
6627 
6628 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
6629 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6630 		    sysctl_tp_tick, "A", "TP timer tick (us)");
6631 
6632 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
6633 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 1,
6634 		    sysctl_tp_tick, "A", "TCP timestamp tick (us)");
6635 
6636 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
6637 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 2,
6638 		    sysctl_tp_tick, "A", "DACK tick (us)");
6639 
6640 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
6641 		    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, 0,
6642 		    sysctl_tp_dack_timer, "IU", "DACK timer (us)");
6643 
6644 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
6645 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6646 		    A_TP_RXT_MIN, sysctl_tp_timer, "LU",
6647 		    "Minimum retransmit interval (us)");
6648 
6649 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
6650 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6651 		    A_TP_RXT_MAX, sysctl_tp_timer, "LU",
6652 		    "Maximum retransmit interval (us)");
6653 
6654 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
6655 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6656 		    A_TP_PERS_MIN, sysctl_tp_timer, "LU",
6657 		    "Persist timer min (us)");
6658 
6659 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
6660 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6661 		    A_TP_PERS_MAX, sysctl_tp_timer, "LU",
6662 		    "Persist timer max (us)");
6663 
6664 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
6665 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6666 		    A_TP_KEEP_IDLE, sysctl_tp_timer, "LU",
6667 		    "Keepalive idle timer (us)");
6668 
6669 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
6670 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6671 		    A_TP_KEEP_INTVL, sysctl_tp_timer, "LU",
6672 		    "Keepalive interval timer (us)");
6673 
6674 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
6675 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6676 		    A_TP_INIT_SRTT, sysctl_tp_timer, "LU", "Initial SRTT (us)");
6677 
6678 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
6679 		    CTLTYPE_ULONG | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6680 		    A_TP_FINWAIT2_TIMER, sysctl_tp_timer, "LU",
6681 		    "FINWAIT2 timer (us)");
6682 
6683 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
6684 		    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6685 		    S_SYNSHIFTMAX, sysctl_tp_shift_cnt, "IU",
6686 		    "Number of SYN retransmissions before abort");
6687 
6688 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
6689 		    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6690 		    S_RXTSHIFTMAXR2, sysctl_tp_shift_cnt, "IU",
6691 		    "Number of retransmissions before abort");
6692 
6693 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
6694 		    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6695 		    S_KEEPALIVEMAXR2, sysctl_tp_shift_cnt, "IU",
6696 		    "Number of keepalive probes before abort");
6697 
6698 		oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
6699 		    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
6700 		    "TOE retransmit backoffs");
6701 		children = SYSCTL_CHILDREN(oid);
6702 		for (i = 0; i < 16; i++) {
6703 			snprintf(s, sizeof(s), "%u", i);
6704 			SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
6705 			    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6706 			    i, sysctl_tp_backoff, "IU",
6707 			    "TOE retransmit backoff");
6708 		}
6709 	}
6710 #endif
6711 }
6712 
6713 void
6714 vi_sysctls(struct vi_info *vi)
6715 {
6716 	struct sysctl_ctx_list *ctx;
6717 	struct sysctl_oid *oid;
6718 	struct sysctl_oid_list *children;
6719 
6720 	ctx = device_get_sysctl_ctx(vi->dev);
6721 
6722 	/*
6723 	 * dev.v?(cxgbe|cxl).X.
6724 	 */
6725 	oid = device_get_sysctl_tree(vi->dev);
6726 	children = SYSCTL_CHILDREN(oid);
6727 
6728 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
6729 	    vi->viid, "VI identifer");
6730 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
6731 	    &vi->nrxq, 0, "# of rx queues");
6732 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
6733 	    &vi->ntxq, 0, "# of tx queues");
6734 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
6735 	    &vi->first_rxq, 0, "index of first rx queue");
6736 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
6737 	    &vi->first_txq, 0, "index of first tx queue");
6738 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_base", CTLFLAG_RD, NULL,
6739 	    vi->rss_base, "start of RSS indirection table");
6740 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
6741 	    vi->rss_size, "size of RSS indirection table");
6742 
6743 	if (IS_MAIN_VI(vi)) {
6744 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
6745 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, vi, 0,
6746 		    sysctl_noflowq, "IU",
6747 		    "Reserve queue 0 for non-flowid packets");
6748 	}
6749 
6750 #ifdef TCP_OFFLOAD
6751 	if (vi->nofldrxq != 0) {
6752 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
6753 		    &vi->nofldrxq, 0,
6754 		    "# of rx queues for offloaded TCP connections");
6755 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
6756 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
6757 		    "index of first TOE rx queue");
6758 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
6759 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, vi, 0,
6760 		    sysctl_holdoff_tmr_idx_ofld, "I",
6761 		    "holdoff timer index for TOE queues");
6762 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
6763 		    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, vi, 0,
6764 		    sysctl_holdoff_pktc_idx_ofld, "I",
6765 		    "holdoff packet counter index for TOE queues");
6766 	}
6767 #endif
6768 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
6769 	if (vi->nofldtxq != 0) {
6770 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
6771 		    &vi->nofldtxq, 0,
6772 		    "# of tx queues for TOE/ETHOFLD");
6773 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
6774 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
6775 		    "index of first TOE/ETHOFLD tx queue");
6776 	}
6777 #endif
6778 #ifdef DEV_NETMAP
6779 	if (vi->nnmrxq != 0) {
6780 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
6781 		    &vi->nnmrxq, 0, "# of netmap rx queues");
6782 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
6783 		    &vi->nnmtxq, 0, "# of netmap tx queues");
6784 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
6785 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
6786 		    "index of first netmap rx queue");
6787 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
6788 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
6789 		    "index of first netmap tx queue");
6790 	}
6791 #endif
6792 
6793 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
6794 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, vi, 0,
6795 	    sysctl_holdoff_tmr_idx, "I", "holdoff timer index");
6796 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
6797 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, vi, 0,
6798 	    sysctl_holdoff_pktc_idx, "I", "holdoff packet counter index");
6799 
6800 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
6801 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, vi, 0,
6802 	    sysctl_qsize_rxq, "I", "rx queue size");
6803 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
6804 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, vi, 0,
6805 	    sysctl_qsize_txq, "I", "tx queue size");
6806 }
6807 
6808 static void
6809 cxgbe_sysctls(struct port_info *pi)
6810 {
6811 	struct sysctl_ctx_list *ctx;
6812 	struct sysctl_oid *oid;
6813 	struct sysctl_oid_list *children, *children2;
6814 	struct adapter *sc = pi->adapter;
6815 	int i;
6816 	char name[16];
6817 	static char *tc_flags = {"\20\1USER\2SYNC\3ASYNC\4ERR"};
6818 
6819 	ctx = device_get_sysctl_ctx(pi->dev);
6820 
6821 	/*
6822 	 * dev.cxgbe.X.
6823 	 */
6824 	oid = device_get_sysctl_tree(pi->dev);
6825 	children = SYSCTL_CHILDREN(oid);
6826 
6827 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc",
6828 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, pi, 0,
6829 	    sysctl_linkdnrc, "A", "reason why link is down");
6830 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
6831 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
6832 		    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_NEEDGIANT, pi, 0,
6833 		    sysctl_btphy, "I", "PHY temperature (in Celsius)");
6834 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
6835 		    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_NEEDGIANT, pi, 1,
6836 		    sysctl_btphy, "I", "PHY firmware version");
6837 	}
6838 
6839 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
6840 	    CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, pi, 0,
6841 	    sysctl_pause_settings, "A",
6842 	    "PAUSE settings (bit 0 = rx_pause, 1 = tx_pause, 2 = pause_autoneg)");
6843 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
6844 	    CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, pi, 0,
6845 	    sysctl_fec, "A",
6846 	    "FECs to use (bit 0 = RS, 1 = FC, 2 = none, 5 = auto, 6 = module)");
6847 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "module_fec",
6848 	    CTLTYPE_STRING | CTLFLAG_NEEDGIANT, pi, 0, sysctl_module_fec, "A",
6849 	    "FEC recommended by the cable/transceiver");
6850 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
6851 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, pi, 0,
6852 	    sysctl_autoneg, "I",
6853 	    "autonegotiation (-1 = not supported)");
6854 
6855 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "pcaps", CTLFLAG_RD,
6856 	    &pi->link_cfg.pcaps, 0, "port capabilities");
6857 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "acaps", CTLFLAG_RD,
6858 	    &pi->link_cfg.acaps, 0, "advertised capabilities");
6859 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lpacaps", CTLFLAG_RD,
6860 	    &pi->link_cfg.lpacaps, 0, "link partner advertised capabilities");
6861 
6862 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
6863 	    port_top_speed(pi), "max speed (in Gbps)");
6864 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
6865 	    pi->mps_bg_map, "MPS buffer group map");
6866 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
6867 	    NULL, pi->rx_e_chan_map, "TP rx e-channel map");
6868 
6869 	if (sc->flags & IS_VF)
6870 		return;
6871 
6872 	/*
6873 	 * dev.(cxgbe|cxl).X.tc.
6874 	 */
6875 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc",
6876 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
6877 	    "Tx scheduler traffic classes (cl_rl)");
6878 	children2 = SYSCTL_CHILDREN(oid);
6879 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "pktsize",
6880 	    CTLFLAG_RW, &pi->sched_params->pktsize, 0,
6881 	    "pktsize for per-flow cl-rl (0 means up to the driver )");
6882 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "burstsize",
6883 	    CTLFLAG_RW, &pi->sched_params->burstsize, 0,
6884 	    "burstsize for per-flow cl-rl (0 means up to the driver)");
6885 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
6886 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
6887 
6888 		snprintf(name, sizeof(name), "%d", i);
6889 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
6890 		    SYSCTL_CHILDREN(oid), OID_AUTO, name,
6891 		    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "traffic class"));
6892 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "flags",
6893 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, tc_flags,
6894 		    (uintptr_t)&tc->flags, sysctl_bitfield_8b, "A", "flags");
6895 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
6896 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
6897 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
6898 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc,
6899 		    (pi->port_id << 16) | i, sysctl_tc_params, "A",
6900 		    "traffic class parameters");
6901 	}
6902 
6903 	/*
6904 	 * dev.cxgbe.X.stats.
6905 	 */
6906 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats",
6907 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "port statistics");
6908 	children = SYSCTL_CHILDREN(oid);
6909 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
6910 	    &pi->tx_parse_error, 0,
6911 	    "# of tx packets with invalid length or # of segments");
6912 
6913 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
6914     SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
6915         CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_NEEDGIANT, sc, reg, \
6916         sysctl_handle_t4_reg64, "QU", desc)
6917 
6918 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
6919 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
6920 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
6921 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
6922 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
6923 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
6924 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
6925 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
6926 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
6927 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
6928 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
6929 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
6930 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
6931 	    "# of tx frames in this range",
6932 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
6933 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
6934 	    "# of tx frames in this range",
6935 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
6936 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
6937 	    "# of tx frames in this range",
6938 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
6939 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
6940 	    "# of tx frames in this range",
6941 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
6942 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
6943 	    "# of tx frames in this range",
6944 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
6945 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
6946 	    "# of tx frames in this range",
6947 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
6948 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
6949 	    "# of tx frames in this range",
6950 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
6951 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
6952 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
6953 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
6954 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
6955 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
6956 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
6957 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
6958 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
6959 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
6960 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
6961 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
6962 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
6963 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
6964 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
6965 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
6966 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
6967 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
6968 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
6969 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
6970 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
6971 
6972 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
6973 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
6974 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
6975 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
6976 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
6977 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
6978 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
6979 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
6980 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
6981 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
6982 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
6983 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
6984 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
6985 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
6986 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
6987 	    "# of frames received with bad FCS",
6988 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
6989 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
6990 	    "# of frames received with length error",
6991 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
6992 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
6993 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
6994 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
6995 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
6996 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
6997 	    "# of rx frames in this range",
6998 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
6999 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
7000 	    "# of rx frames in this range",
7001 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
7002 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
7003 	    "# of rx frames in this range",
7004 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
7005 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
7006 	    "# of rx frames in this range",
7007 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
7008 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
7009 	    "# of rx frames in this range",
7010 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
7011 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
7012 	    "# of rx frames in this range",
7013 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
7014 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
7015 	    "# of rx frames in this range",
7016 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
7017 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
7018 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
7019 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
7020 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
7021 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
7022 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
7023 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
7024 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
7025 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
7026 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
7027 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
7028 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
7029 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
7030 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
7031 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
7032 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
7033 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
7034 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
7035 
7036 #undef SYSCTL_ADD_T4_REG64
7037 
7038 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
7039 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
7040 	    &pi->stats.name, desc)
7041 
7042 	/* We get these from port_stats and they may be stale by up to 1s */
7043 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
7044 	    "# drops due to buffer-group 0 overflows");
7045 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
7046 	    "# drops due to buffer-group 1 overflows");
7047 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
7048 	    "# drops due to buffer-group 2 overflows");
7049 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
7050 	    "# drops due to buffer-group 3 overflows");
7051 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
7052 	    "# of buffer-group 0 truncated packets");
7053 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
7054 	    "# of buffer-group 1 truncated packets");
7055 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
7056 	    "# of buffer-group 2 truncated packets");
7057 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
7058 	    "# of buffer-group 3 truncated packets");
7059 
7060 #undef SYSCTL_ADD_T4_PORTSTAT
7061 
7062 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_toe_tls_records",
7063 	    CTLFLAG_RD, &pi->tx_toe_tls_records,
7064 	    "# of TOE TLS records transmitted");
7065 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_toe_tls_octets",
7066 	    CTLFLAG_RD, &pi->tx_toe_tls_octets,
7067 	    "# of payload octets in transmitted TOE TLS records");
7068 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_toe_tls_records",
7069 	    CTLFLAG_RD, &pi->rx_toe_tls_records,
7070 	    "# of TOE TLS records received");
7071 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_toe_tls_octets",
7072 	    CTLFLAG_RD, &pi->rx_toe_tls_octets,
7073 	    "# of payload octets in received TOE TLS records");
7074 }
7075 
7076 static int
7077 sysctl_int_array(SYSCTL_HANDLER_ARGS)
7078 {
7079 	int rc, *i, space = 0;
7080 	struct sbuf sb;
7081 
7082 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
7083 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
7084 		if (space)
7085 			sbuf_printf(&sb, " ");
7086 		sbuf_printf(&sb, "%d", *i);
7087 		space = 1;
7088 	}
7089 	rc = sbuf_finish(&sb);
7090 	sbuf_delete(&sb);
7091 	return (rc);
7092 }
7093 
7094 static int
7095 sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS)
7096 {
7097 	int rc;
7098 	struct sbuf *sb;
7099 
7100 	rc = sysctl_wire_old_buffer(req, 0);
7101 	if (rc != 0)
7102 		return(rc);
7103 
7104 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
7105 	if (sb == NULL)
7106 		return (ENOMEM);
7107 
7108 	sbuf_printf(sb, "%b", *(uint8_t *)(uintptr_t)arg2, (char *)arg1);
7109 	rc = sbuf_finish(sb);
7110 	sbuf_delete(sb);
7111 
7112 	return (rc);
7113 }
7114 
7115 static int
7116 sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS)
7117 {
7118 	int rc;
7119 	struct sbuf *sb;
7120 
7121 	rc = sysctl_wire_old_buffer(req, 0);
7122 	if (rc != 0)
7123 		return(rc);
7124 
7125 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
7126 	if (sb == NULL)
7127 		return (ENOMEM);
7128 
7129 	sbuf_printf(sb, "%b", *(uint16_t *)(uintptr_t)arg2, (char *)arg1);
7130 	rc = sbuf_finish(sb);
7131 	sbuf_delete(sb);
7132 
7133 	return (rc);
7134 }
7135 
7136 static int
7137 sysctl_btphy(SYSCTL_HANDLER_ARGS)
7138 {
7139 	struct port_info *pi = arg1;
7140 	int op = arg2;
7141 	struct adapter *sc = pi->adapter;
7142 	u_int v;
7143 	int rc;
7144 
7145 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
7146 	if (rc)
7147 		return (rc);
7148 	/* XXX: magic numbers */
7149 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
7150 	    &v);
7151 	end_synchronized_op(sc, 0);
7152 	if (rc)
7153 		return (rc);
7154 	if (op == 0)
7155 		v /= 256;
7156 
7157 	rc = sysctl_handle_int(oidp, &v, 0, req);
7158 	return (rc);
7159 }
7160 
7161 static int
7162 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
7163 {
7164 	struct vi_info *vi = arg1;
7165 	int rc, val;
7166 
7167 	val = vi->rsrv_noflowq;
7168 	rc = sysctl_handle_int(oidp, &val, 0, req);
7169 	if (rc != 0 || req->newptr == NULL)
7170 		return (rc);
7171 
7172 	if ((val >= 1) && (vi->ntxq > 1))
7173 		vi->rsrv_noflowq = 1;
7174 	else
7175 		vi->rsrv_noflowq = 0;
7176 
7177 	return (rc);
7178 }
7179 
7180 static int
7181 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
7182 {
7183 	struct vi_info *vi = arg1;
7184 	struct adapter *sc = vi->pi->adapter;
7185 	int idx, rc, i;
7186 	struct sge_rxq *rxq;
7187 	uint8_t v;
7188 
7189 	idx = vi->tmr_idx;
7190 
7191 	rc = sysctl_handle_int(oidp, &idx, 0, req);
7192 	if (rc != 0 || req->newptr == NULL)
7193 		return (rc);
7194 
7195 	if (idx < 0 || idx >= SGE_NTIMERS)
7196 		return (EINVAL);
7197 
7198 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
7199 	    "t4tmr");
7200 	if (rc)
7201 		return (rc);
7202 
7203 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
7204 	for_each_rxq(vi, i, rxq) {
7205 #ifdef atomic_store_rel_8
7206 		atomic_store_rel_8(&rxq->iq.intr_params, v);
7207 #else
7208 		rxq->iq.intr_params = v;
7209 #endif
7210 	}
7211 	vi->tmr_idx = idx;
7212 
7213 	end_synchronized_op(sc, LOCK_HELD);
7214 	return (0);
7215 }
7216 
7217 static int
7218 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
7219 {
7220 	struct vi_info *vi = arg1;
7221 	struct adapter *sc = vi->pi->adapter;
7222 	int idx, rc;
7223 
7224 	idx = vi->pktc_idx;
7225 
7226 	rc = sysctl_handle_int(oidp, &idx, 0, req);
7227 	if (rc != 0 || req->newptr == NULL)
7228 		return (rc);
7229 
7230 	if (idx < -1 || idx >= SGE_NCOUNTERS)
7231 		return (EINVAL);
7232 
7233 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
7234 	    "t4pktc");
7235 	if (rc)
7236 		return (rc);
7237 
7238 	if (vi->flags & VI_INIT_DONE)
7239 		rc = EBUSY; /* cannot be changed once the queues are created */
7240 	else
7241 		vi->pktc_idx = idx;
7242 
7243 	end_synchronized_op(sc, LOCK_HELD);
7244 	return (rc);
7245 }
7246 
7247 static int
7248 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
7249 {
7250 	struct vi_info *vi = arg1;
7251 	struct adapter *sc = vi->pi->adapter;
7252 	int qsize, rc;
7253 
7254 	qsize = vi->qsize_rxq;
7255 
7256 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
7257 	if (rc != 0 || req->newptr == NULL)
7258 		return (rc);
7259 
7260 	if (qsize < 128 || (qsize & 7))
7261 		return (EINVAL);
7262 
7263 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
7264 	    "t4rxqs");
7265 	if (rc)
7266 		return (rc);
7267 
7268 	if (vi->flags & VI_INIT_DONE)
7269 		rc = EBUSY; /* cannot be changed once the queues are created */
7270 	else
7271 		vi->qsize_rxq = qsize;
7272 
7273 	end_synchronized_op(sc, LOCK_HELD);
7274 	return (rc);
7275 }
7276 
7277 static int
7278 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
7279 {
7280 	struct vi_info *vi = arg1;
7281 	struct adapter *sc = vi->pi->adapter;
7282 	int qsize, rc;
7283 
7284 	qsize = vi->qsize_txq;
7285 
7286 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
7287 	if (rc != 0 || req->newptr == NULL)
7288 		return (rc);
7289 
7290 	if (qsize < 128 || qsize > 65536)
7291 		return (EINVAL);
7292 
7293 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
7294 	    "t4txqs");
7295 	if (rc)
7296 		return (rc);
7297 
7298 	if (vi->flags & VI_INIT_DONE)
7299 		rc = EBUSY; /* cannot be changed once the queues are created */
7300 	else
7301 		vi->qsize_txq = qsize;
7302 
7303 	end_synchronized_op(sc, LOCK_HELD);
7304 	return (rc);
7305 }
7306 
7307 static int
7308 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
7309 {
7310 	struct port_info *pi = arg1;
7311 	struct adapter *sc = pi->adapter;
7312 	struct link_config *lc = &pi->link_cfg;
7313 	int rc;
7314 
7315 	if (req->newptr == NULL) {
7316 		struct sbuf *sb;
7317 		static char *bits = "\20\1RX\2TX\3AUTO";
7318 
7319 		rc = sysctl_wire_old_buffer(req, 0);
7320 		if (rc != 0)
7321 			return(rc);
7322 
7323 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
7324 		if (sb == NULL)
7325 			return (ENOMEM);
7326 
7327 		if (lc->link_ok) {
7328 			sbuf_printf(sb, "%b", (lc->fc & (PAUSE_TX | PAUSE_RX)) |
7329 			    (lc->requested_fc & PAUSE_AUTONEG), bits);
7330 		} else {
7331 			sbuf_printf(sb, "%b", lc->requested_fc & (PAUSE_TX |
7332 			    PAUSE_RX | PAUSE_AUTONEG), bits);
7333 		}
7334 		rc = sbuf_finish(sb);
7335 		sbuf_delete(sb);
7336 	} else {
7337 		char s[2];
7338 		int n;
7339 
7340 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX |
7341 		    PAUSE_AUTONEG));
7342 		s[1] = 0;
7343 
7344 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
7345 		if (rc != 0)
7346 			return(rc);
7347 
7348 		if (s[1] != 0)
7349 			return (EINVAL);
7350 		if (s[0] < '0' || s[0] > '9')
7351 			return (EINVAL);	/* not a number */
7352 		n = s[0] - '0';
7353 		if (n & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG))
7354 			return (EINVAL);	/* some other bit is set too */
7355 
7356 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
7357 		    "t4PAUSE");
7358 		if (rc)
7359 			return (rc);
7360 		PORT_LOCK(pi);
7361 		lc->requested_fc = n;
7362 		fixup_link_config(pi);
7363 		if (pi->up_vis > 0)
7364 			rc = apply_link_config(pi);
7365 		set_current_media(pi);
7366 		PORT_UNLOCK(pi);
7367 		end_synchronized_op(sc, 0);
7368 	}
7369 
7370 	return (rc);
7371 }
7372 
7373 static int
7374 sysctl_fec(SYSCTL_HANDLER_ARGS)
7375 {
7376 	struct port_info *pi = arg1;
7377 	struct adapter *sc = pi->adapter;
7378 	struct link_config *lc = &pi->link_cfg;
7379 	int rc;
7380 	int8_t old;
7381 
7382 	if (req->newptr == NULL) {
7383 		struct sbuf *sb;
7384 		static char *bits = "\20\1RS-FEC\2FC-FEC\3NO-FEC\4RSVD2"
7385 		    "\5RSVD3\6auto\7module";
7386 
7387 		rc = sysctl_wire_old_buffer(req, 0);
7388 		if (rc != 0)
7389 			return(rc);
7390 
7391 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
7392 		if (sb == NULL)
7393 			return (ENOMEM);
7394 
7395 		/*
7396 		 * Display the requested_fec when the link is down -- the actual
7397 		 * FEC makes sense only when the link is up.
7398 		 */
7399 		if (lc->link_ok) {
7400 			sbuf_printf(sb, "%b", (lc->fec & M_FW_PORT_CAP32_FEC) |
7401 			    (lc->requested_fec & (FEC_AUTO | FEC_MODULE)),
7402 			    bits);
7403 		} else {
7404 			sbuf_printf(sb, "%b", lc->requested_fec, bits);
7405 		}
7406 		rc = sbuf_finish(sb);
7407 		sbuf_delete(sb);
7408 	} else {
7409 		char s[8];
7410 		int n;
7411 
7412 		snprintf(s, sizeof(s), "%d",
7413 		    lc->requested_fec == FEC_AUTO ? -1 :
7414 		    lc->requested_fec & (M_FW_PORT_CAP32_FEC | FEC_MODULE));
7415 
7416 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
7417 		if (rc != 0)
7418 			return(rc);
7419 
7420 		n = strtol(&s[0], NULL, 0);
7421 		if (n < 0 || n & FEC_AUTO)
7422 			n = FEC_AUTO;
7423 		else if (n & ~(M_FW_PORT_CAP32_FEC | FEC_MODULE))
7424 			return (EINVAL);/* some other bit is set too */
7425 
7426 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
7427 		    "t4fec");
7428 		if (rc)
7429 			return (rc);
7430 		PORT_LOCK(pi);
7431 		old = lc->requested_fec;
7432 		if (n == FEC_AUTO)
7433 			lc->requested_fec = FEC_AUTO;
7434 		else if (n == 0 || n == FEC_NONE)
7435 			lc->requested_fec = FEC_NONE;
7436 		else {
7437 			if ((lc->pcaps |
7438 			    V_FW_PORT_CAP32_FEC(n & M_FW_PORT_CAP32_FEC)) !=
7439 			    lc->pcaps) {
7440 				rc = ENOTSUP;
7441 				goto done;
7442 			}
7443 			lc->requested_fec = n & (M_FW_PORT_CAP32_FEC |
7444 			    FEC_MODULE);
7445 		}
7446 		fixup_link_config(pi);
7447 		if (pi->up_vis > 0) {
7448 			rc = apply_link_config(pi);
7449 			if (rc != 0) {
7450 				lc->requested_fec = old;
7451 				if (rc == FW_EPROTO)
7452 					rc = ENOTSUP;
7453 			}
7454 		}
7455 done:
7456 		PORT_UNLOCK(pi);
7457 		end_synchronized_op(sc, 0);
7458 	}
7459 
7460 	return (rc);
7461 }
7462 
7463 static int
7464 sysctl_module_fec(SYSCTL_HANDLER_ARGS)
7465 {
7466 	struct port_info *pi = arg1;
7467 	struct adapter *sc = pi->adapter;
7468 	struct link_config *lc = &pi->link_cfg;
7469 	int rc;
7470 	int8_t fec;
7471 	struct sbuf *sb;
7472 	static char *bits = "\20\1RS-FEC\2FC-FEC\3NO-FEC\4RSVD2\5RSVD3";
7473 
7474 	rc = sysctl_wire_old_buffer(req, 0);
7475 	if (rc != 0)
7476 		return (rc);
7477 
7478 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
7479 	if (sb == NULL)
7480 		return (ENOMEM);
7481 
7482 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mfec") != 0)
7483 		return (EBUSY);
7484 	PORT_LOCK(pi);
7485 	if (pi->up_vis == 0) {
7486 		/*
7487 		 * If all the interfaces are administratively down the firmware
7488 		 * does not report transceiver changes.  Refresh port info here.
7489 		 * This is the only reason we have a synchronized op in this
7490 		 * function.  Just PORT_LOCK would have been enough otherwise.
7491 		 */
7492 		t4_update_port_info(pi);
7493 	}
7494 
7495 	fec = lc->fec_hint;
7496 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE ||
7497 	    !fec_supported(lc->pcaps)) {
7498 		sbuf_printf(sb, "n/a");
7499 	} else {
7500 		if (fec == 0)
7501 			fec = FEC_NONE;
7502 		sbuf_printf(sb, "%b", fec & M_FW_PORT_CAP32_FEC, bits);
7503 	}
7504 	rc = sbuf_finish(sb);
7505 	sbuf_delete(sb);
7506 
7507 	PORT_UNLOCK(pi);
7508 	end_synchronized_op(sc, 0);
7509 
7510 	return (rc);
7511 }
7512 
7513 static int
7514 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
7515 {
7516 	struct port_info *pi = arg1;
7517 	struct adapter *sc = pi->adapter;
7518 	struct link_config *lc = &pi->link_cfg;
7519 	int rc, val;
7520 
7521 	if (lc->pcaps & FW_PORT_CAP32_ANEG)
7522 		val = lc->requested_aneg == AUTONEG_DISABLE ? 0 : 1;
7523 	else
7524 		val = -1;
7525 	rc = sysctl_handle_int(oidp, &val, 0, req);
7526 	if (rc != 0 || req->newptr == NULL)
7527 		return (rc);
7528 	if (val == 0)
7529 		val = AUTONEG_DISABLE;
7530 	else if (val == 1)
7531 		val = AUTONEG_ENABLE;
7532 	else
7533 		val = AUTONEG_AUTO;
7534 
7535 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
7536 	    "t4aneg");
7537 	if (rc)
7538 		return (rc);
7539 	PORT_LOCK(pi);
7540 	if (val == AUTONEG_ENABLE && !(lc->pcaps & FW_PORT_CAP32_ANEG)) {
7541 		rc = ENOTSUP;
7542 		goto done;
7543 	}
7544 	lc->requested_aneg = val;
7545 	fixup_link_config(pi);
7546 	if (pi->up_vis > 0)
7547 		rc = apply_link_config(pi);
7548 	set_current_media(pi);
7549 done:
7550 	PORT_UNLOCK(pi);
7551 	end_synchronized_op(sc, 0);
7552 	return (rc);
7553 }
7554 
7555 static int
7556 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
7557 {
7558 	struct adapter *sc = arg1;
7559 	int reg = arg2;
7560 	uint64_t val;
7561 
7562 	val = t4_read_reg64(sc, reg);
7563 
7564 	return (sysctl_handle_64(oidp, &val, 0, req));
7565 }
7566 
7567 static int
7568 sysctl_temperature(SYSCTL_HANDLER_ARGS)
7569 {
7570 	struct adapter *sc = arg1;
7571 	int rc, t;
7572 	uint32_t param, val;
7573 
7574 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
7575 	if (rc)
7576 		return (rc);
7577 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
7578 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
7579 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
7580 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
7581 	end_synchronized_op(sc, 0);
7582 	if (rc)
7583 		return (rc);
7584 
7585 	/* unknown is returned as 0 but we display -1 in that case */
7586 	t = val == 0 ? -1 : val;
7587 
7588 	rc = sysctl_handle_int(oidp, &t, 0, req);
7589 	return (rc);
7590 }
7591 
7592 static int
7593 sysctl_vdd(SYSCTL_HANDLER_ARGS)
7594 {
7595 	struct adapter *sc = arg1;
7596 	int rc;
7597 	uint32_t param, val;
7598 
7599 	if (sc->params.core_vdd == 0) {
7600 		rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7601 		    "t4vdd");
7602 		if (rc)
7603 			return (rc);
7604 		param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
7605 		    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
7606 		    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
7607 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
7608 		end_synchronized_op(sc, 0);
7609 		if (rc)
7610 			return (rc);
7611 		sc->params.core_vdd = val;
7612 	}
7613 
7614 	return (sysctl_handle_int(oidp, &sc->params.core_vdd, 0, req));
7615 }
7616 
7617 static int
7618 sysctl_reset_sensor(SYSCTL_HANDLER_ARGS)
7619 {
7620 	struct adapter *sc = arg1;
7621 	int rc, v;
7622 	uint32_t param, val;
7623 
7624 	v = sc->sensor_resets;
7625 	rc = sysctl_handle_int(oidp, &v, 0, req);
7626 	if (rc != 0 || req->newptr == NULL || v <= 0)
7627 		return (rc);
7628 
7629 	if (sc->params.fw_vers < FW_VERSION32(1, 24, 7, 0) ||
7630 	    chip_id(sc) < CHELSIO_T5)
7631 		return (ENOTSUP);
7632 
7633 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4srst");
7634 	if (rc)
7635 		return (rc);
7636 	param = (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
7637 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
7638 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_RESET_TMP_SENSOR));
7639 	val = 1;
7640 	rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
7641 	end_synchronized_op(sc, 0);
7642 	if (rc == 0)
7643 		sc->sensor_resets++;
7644 	return (rc);
7645 }
7646 
7647 static int
7648 sysctl_loadavg(SYSCTL_HANDLER_ARGS)
7649 {
7650 	struct adapter *sc = arg1;
7651 	struct sbuf *sb;
7652 	int rc;
7653 	uint32_t param, val;
7654 
7655 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4lavg");
7656 	if (rc)
7657 		return (rc);
7658 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
7659 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_LOAD);
7660 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
7661 	end_synchronized_op(sc, 0);
7662 	if (rc)
7663 		return (rc);
7664 
7665 	rc = sysctl_wire_old_buffer(req, 0);
7666 	if (rc != 0)
7667 		return (rc);
7668 
7669 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7670 	if (sb == NULL)
7671 		return (ENOMEM);
7672 
7673 	if (val == 0xffffffff) {
7674 		/* Only debug and custom firmwares report load averages. */
7675 		sbuf_printf(sb, "not available");
7676 	} else {
7677 		sbuf_printf(sb, "%d %d %d", val & 0xff, (val >> 8) & 0xff,
7678 		    (val >> 16) & 0xff);
7679 	}
7680 	rc = sbuf_finish(sb);
7681 	sbuf_delete(sb);
7682 
7683 	return (rc);
7684 }
7685 
7686 static int
7687 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
7688 {
7689 	struct adapter *sc = arg1;
7690 	struct sbuf *sb;
7691 	int rc, i;
7692 	uint16_t incr[NMTUS][NCCTRL_WIN];
7693 	static const char *dec_fac[] = {
7694 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
7695 		"0.9375"
7696 	};
7697 
7698 	rc = sysctl_wire_old_buffer(req, 0);
7699 	if (rc != 0)
7700 		return (rc);
7701 
7702 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7703 	if (sb == NULL)
7704 		return (ENOMEM);
7705 
7706 	t4_read_cong_tbl(sc, incr);
7707 
7708 	for (i = 0; i < NCCTRL_WIN; ++i) {
7709 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
7710 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
7711 		    incr[5][i], incr[6][i], incr[7][i]);
7712 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
7713 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
7714 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
7715 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
7716 	}
7717 
7718 	rc = sbuf_finish(sb);
7719 	sbuf_delete(sb);
7720 
7721 	return (rc);
7722 }
7723 
7724 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
7725 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
7726 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
7727 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
7728 };
7729 
7730 static int
7731 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
7732 {
7733 	struct adapter *sc = arg1;
7734 	struct sbuf *sb;
7735 	int rc, i, n, qid = arg2;
7736 	uint32_t *buf, *p;
7737 	char *qtype;
7738 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
7739 
7740 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
7741 	    ("%s: bad qid %d\n", __func__, qid));
7742 
7743 	if (qid < CIM_NUM_IBQ) {
7744 		/* inbound queue */
7745 		qtype = "IBQ";
7746 		n = 4 * CIM_IBQ_SIZE;
7747 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
7748 		rc = t4_read_cim_ibq(sc, qid, buf, n);
7749 	} else {
7750 		/* outbound queue */
7751 		qtype = "OBQ";
7752 		qid -= CIM_NUM_IBQ;
7753 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
7754 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
7755 		rc = t4_read_cim_obq(sc, qid, buf, n);
7756 	}
7757 
7758 	if (rc < 0) {
7759 		rc = -rc;
7760 		goto done;
7761 	}
7762 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
7763 
7764 	rc = sysctl_wire_old_buffer(req, 0);
7765 	if (rc != 0)
7766 		goto done;
7767 
7768 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7769 	if (sb == NULL) {
7770 		rc = ENOMEM;
7771 		goto done;
7772 	}
7773 
7774 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
7775 	for (i = 0, p = buf; i < n; i += 16, p += 4)
7776 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
7777 		    p[2], p[3]);
7778 
7779 	rc = sbuf_finish(sb);
7780 	sbuf_delete(sb);
7781 done:
7782 	free(buf, M_CXGBE);
7783 	return (rc);
7784 }
7785 
7786 static void
7787 sbuf_cim_la4(struct adapter *sc, struct sbuf *sb, uint32_t *buf, uint32_t cfg)
7788 {
7789 	uint32_t *p;
7790 
7791 	sbuf_printf(sb, "Status   Data      PC%s",
7792 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
7793 	    "     LS0Stat  LS0Addr             LS0Data");
7794 
7795 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
7796 		if (cfg & F_UPDBGLACAPTPCONLY) {
7797 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
7798 			    p[6], p[7]);
7799 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
7800 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
7801 			    p[4] & 0xff, p[5] >> 8);
7802 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
7803 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7804 			    p[1] & 0xf, p[2] >> 4);
7805 		} else {
7806 			sbuf_printf(sb,
7807 			    "\n  %02x   %x%07x %x%07x %08x %08x "
7808 			    "%08x%08x%08x%08x",
7809 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7810 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
7811 			    p[6], p[7]);
7812 		}
7813 	}
7814 }
7815 
7816 static void
7817 sbuf_cim_la6(struct adapter *sc, struct sbuf *sb, uint32_t *buf, uint32_t cfg)
7818 {
7819 	uint32_t *p;
7820 
7821 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
7822 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
7823 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
7824 
7825 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
7826 		if (cfg & F_UPDBGLACAPTPCONLY) {
7827 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
7828 			    p[3] & 0xff, p[2], p[1], p[0]);
7829 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
7830 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
7831 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
7832 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
7833 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
7834 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
7835 			    p[6] >> 16);
7836 		} else {
7837 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
7838 			    "%08x %08x %08x %08x %08x %08x",
7839 			    (p[9] >> 16) & 0xff,
7840 			    p[9] & 0xffff, p[8] >> 16,
7841 			    p[8] & 0xffff, p[7] >> 16,
7842 			    p[7] & 0xffff, p[6] >> 16,
7843 			    p[2], p[1], p[0], p[5], p[4], p[3]);
7844 		}
7845 	}
7846 }
7847 
7848 static int
7849 sbuf_cim_la(struct adapter *sc, struct sbuf *sb, int flags)
7850 {
7851 	uint32_t cfg, *buf;
7852 	int rc;
7853 
7854 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
7855 	if (rc != 0)
7856 		return (rc);
7857 
7858 	MPASS(flags == M_WAITOK || flags == M_NOWAIT);
7859 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
7860 	    M_ZERO | flags);
7861 	if (buf == NULL)
7862 		return (ENOMEM);
7863 
7864 	rc = -t4_cim_read_la(sc, buf, NULL);
7865 	if (rc != 0)
7866 		goto done;
7867 	if (chip_id(sc) < CHELSIO_T6)
7868 		sbuf_cim_la4(sc, sb, buf, cfg);
7869 	else
7870 		sbuf_cim_la6(sc, sb, buf, cfg);
7871 
7872 done:
7873 	free(buf, M_CXGBE);
7874 	return (rc);
7875 }
7876 
7877 static int
7878 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
7879 {
7880 	struct adapter *sc = arg1;
7881 	struct sbuf *sb;
7882 	int rc;
7883 
7884 	rc = sysctl_wire_old_buffer(req, 0);
7885 	if (rc != 0)
7886 		return (rc);
7887 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7888 	if (sb == NULL)
7889 		return (ENOMEM);
7890 
7891 	rc = sbuf_cim_la(sc, sb, M_WAITOK);
7892 	if (rc == 0)
7893 		rc = sbuf_finish(sb);
7894 	sbuf_delete(sb);
7895 	return (rc);
7896 }
7897 
7898 bool
7899 t4_os_dump_cimla(struct adapter *sc, int arg, bool verbose)
7900 {
7901 	struct sbuf sb;
7902 	int rc;
7903 
7904 	if (sbuf_new(&sb, NULL, 4096, SBUF_AUTOEXTEND) != &sb)
7905 		return (false);
7906 	rc = sbuf_cim_la(sc, &sb, M_NOWAIT);
7907 	if (rc == 0) {
7908 		rc = sbuf_finish(&sb);
7909 		if (rc == 0) {
7910 			log(LOG_DEBUG, "%s: CIM LA dump follows.\n%s",
7911 		    		device_get_nameunit(sc->dev), sbuf_data(&sb));
7912 		}
7913 	}
7914 	sbuf_delete(&sb);
7915 	return (false);
7916 }
7917 
7918 static int
7919 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
7920 {
7921 	struct adapter *sc = arg1;
7922 	u_int i;
7923 	struct sbuf *sb;
7924 	uint32_t *buf, *p;
7925 	int rc;
7926 
7927 	rc = sysctl_wire_old_buffer(req, 0);
7928 	if (rc != 0)
7929 		return (rc);
7930 
7931 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7932 	if (sb == NULL)
7933 		return (ENOMEM);
7934 
7935 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
7936 	    M_ZERO | M_WAITOK);
7937 
7938 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
7939 	p = buf;
7940 
7941 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7942 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
7943 		    p[1], p[0]);
7944 	}
7945 
7946 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
7947 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7948 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
7949 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
7950 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
7951 		    (p[1] >> 2) | ((p[2] & 3) << 30),
7952 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
7953 		    p[0] & 1);
7954 	}
7955 
7956 	rc = sbuf_finish(sb);
7957 	sbuf_delete(sb);
7958 	free(buf, M_CXGBE);
7959 	return (rc);
7960 }
7961 
7962 static int
7963 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
7964 {
7965 	struct adapter *sc = arg1;
7966 	u_int i;
7967 	struct sbuf *sb;
7968 	uint32_t *buf, *p;
7969 	int rc;
7970 
7971 	rc = sysctl_wire_old_buffer(req, 0);
7972 	if (rc != 0)
7973 		return (rc);
7974 
7975 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7976 	if (sb == NULL)
7977 		return (ENOMEM);
7978 
7979 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
7980 	    M_ZERO | M_WAITOK);
7981 
7982 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
7983 	p = buf;
7984 
7985 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
7986 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7987 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
7988 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
7989 		    p[4], p[3], p[2], p[1], p[0]);
7990 	}
7991 
7992 	sbuf_printf(sb, "\n\nCntl ID               Data");
7993 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7994 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
7995 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
7996 	}
7997 
7998 	rc = sbuf_finish(sb);
7999 	sbuf_delete(sb);
8000 	free(buf, M_CXGBE);
8001 	return (rc);
8002 }
8003 
8004 static int
8005 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
8006 {
8007 	struct adapter *sc = arg1;
8008 	struct sbuf *sb;
8009 	int rc, i;
8010 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
8011 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
8012 	uint16_t thres[CIM_NUM_IBQ];
8013 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
8014 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
8015 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
8016 
8017 	cim_num_obq = sc->chip_params->cim_num_obq;
8018 	if (is_t4(sc)) {
8019 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
8020 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
8021 	} else {
8022 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
8023 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
8024 	}
8025 	nq = CIM_NUM_IBQ + cim_num_obq;
8026 
8027 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
8028 	if (rc == 0)
8029 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
8030 	if (rc != 0)
8031 		return (rc);
8032 
8033 	t4_read_cimq_cfg(sc, base, size, thres);
8034 
8035 	rc = sysctl_wire_old_buffer(req, 0);
8036 	if (rc != 0)
8037 		return (rc);
8038 
8039 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
8040 	if (sb == NULL)
8041 		return (ENOMEM);
8042 
8043 	sbuf_printf(sb,
8044 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
8045 
8046 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
8047 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
8048 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
8049 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
8050 		    G_QUEREMFLITS(p[2]) * 16);
8051 	for ( ; i < nq; i++, p += 4, wr += 2)
8052 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
8053 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
8054 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
8055 		    G_QUEREMFLITS(p[2]) * 16);
8056 
8057 	rc = sbuf_finish(sb);
8058 	sbuf_delete(sb);
8059 
8060 	return (rc);
8061 }
8062 
8063 static int
8064 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
8065 {
8066 	struct adapter *sc = arg1;
8067 	struct sbuf *sb;
8068 	int rc;
8069 	struct tp_cpl_stats stats;
8070 
8071 	rc = sysctl_wire_old_buffer(req, 0);
8072 	if (rc != 0)
8073 		return (rc);
8074 
8075 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8076 	if (sb == NULL)
8077 		return (ENOMEM);
8078 
8079 	mtx_lock(&sc->reg_lock);
8080 	t4_tp_get_cpl_stats(sc, &stats, 0);
8081 	mtx_unlock(&sc->reg_lock);
8082 
8083 	if (sc->chip_params->nchan > 2) {
8084 		sbuf_printf(sb, "                 channel 0  channel 1"
8085 		    "  channel 2  channel 3");
8086 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
8087 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
8088 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
8089 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
8090 	} else {
8091 		sbuf_printf(sb, "                 channel 0  channel 1");
8092 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
8093 		    stats.req[0], stats.req[1]);
8094 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
8095 		    stats.rsp[0], stats.rsp[1]);
8096 	}
8097 
8098 	rc = sbuf_finish(sb);
8099 	sbuf_delete(sb);
8100 
8101 	return (rc);
8102 }
8103 
8104 static int
8105 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
8106 {
8107 	struct adapter *sc = arg1;
8108 	struct sbuf *sb;
8109 	int rc;
8110 	struct tp_usm_stats stats;
8111 
8112 	rc = sysctl_wire_old_buffer(req, 0);
8113 	if (rc != 0)
8114 		return(rc);
8115 
8116 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8117 	if (sb == NULL)
8118 		return (ENOMEM);
8119 
8120 	t4_get_usm_stats(sc, &stats, 1);
8121 
8122 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
8123 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
8124 	sbuf_printf(sb, "Drops:  %u", stats.drops);
8125 
8126 	rc = sbuf_finish(sb);
8127 	sbuf_delete(sb);
8128 
8129 	return (rc);
8130 }
8131 
8132 static const char * const devlog_level_strings[] = {
8133 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
8134 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
8135 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
8136 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
8137 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
8138 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
8139 };
8140 
8141 static const char * const devlog_facility_strings[] = {
8142 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
8143 	[FW_DEVLOG_FACILITY_CF]		= "CF",
8144 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
8145 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
8146 	[FW_DEVLOG_FACILITY_RES]	= "RES",
8147 	[FW_DEVLOG_FACILITY_HW]		= "HW",
8148 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
8149 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
8150 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
8151 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
8152 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
8153 	[FW_DEVLOG_FACILITY_VI]		= "VI",
8154 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
8155 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
8156 	[FW_DEVLOG_FACILITY_TM]		= "TM",
8157 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
8158 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
8159 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
8160 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
8161 	[FW_DEVLOG_FACILITY_RI]		= "RI",
8162 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
8163 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
8164 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
8165 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
8166 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
8167 };
8168 
8169 static int
8170 sbuf_devlog(struct adapter *sc, struct sbuf *sb, int flags)
8171 {
8172 	int i, j, rc, nentries, first = 0;
8173 	struct devlog_params *dparams = &sc->params.devlog;
8174 	struct fw_devlog_e *buf, *e;
8175 	uint64_t ftstamp = UINT64_MAX;
8176 
8177 	if (dparams->addr == 0)
8178 		return (ENXIO);
8179 
8180 	MPASS(flags == M_WAITOK || flags == M_NOWAIT);
8181 	buf = malloc(dparams->size, M_CXGBE, M_ZERO | flags);
8182 	if (buf == NULL)
8183 		return (ENOMEM);
8184 
8185 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
8186 	if (rc != 0)
8187 		goto done;
8188 
8189 	nentries = dparams->size / sizeof(struct fw_devlog_e);
8190 	for (i = 0; i < nentries; i++) {
8191 		e = &buf[i];
8192 
8193 		if (e->timestamp == 0)
8194 			break;	/* end */
8195 
8196 		e->timestamp = be64toh(e->timestamp);
8197 		e->seqno = be32toh(e->seqno);
8198 		for (j = 0; j < 8; j++)
8199 			e->params[j] = be32toh(e->params[j]);
8200 
8201 		if (e->timestamp < ftstamp) {
8202 			ftstamp = e->timestamp;
8203 			first = i;
8204 		}
8205 	}
8206 
8207 	if (buf[first].timestamp == 0)
8208 		goto done;	/* nothing in the log */
8209 
8210 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
8211 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
8212 
8213 	i = first;
8214 	do {
8215 		e = &buf[i];
8216 		if (e->timestamp == 0)
8217 			break;	/* end */
8218 
8219 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
8220 		    e->seqno, e->timestamp,
8221 		    (e->level < nitems(devlog_level_strings) ?
8222 			devlog_level_strings[e->level] : "UNKNOWN"),
8223 		    (e->facility < nitems(devlog_facility_strings) ?
8224 			devlog_facility_strings[e->facility] : "UNKNOWN"));
8225 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
8226 		    e->params[2], e->params[3], e->params[4],
8227 		    e->params[5], e->params[6], e->params[7]);
8228 
8229 		if (++i == nentries)
8230 			i = 0;
8231 	} while (i != first);
8232 done:
8233 	free(buf, M_CXGBE);
8234 	return (rc);
8235 }
8236 
8237 static int
8238 sysctl_devlog(SYSCTL_HANDLER_ARGS)
8239 {
8240 	struct adapter *sc = arg1;
8241 	int rc;
8242 	struct sbuf *sb;
8243 
8244 	rc = sysctl_wire_old_buffer(req, 0);
8245 	if (rc != 0)
8246 		return (rc);
8247 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8248 	if (sb == NULL)
8249 		return (ENOMEM);
8250 
8251 	rc = sbuf_devlog(sc, sb, M_WAITOK);
8252 	if (rc == 0)
8253 		rc = sbuf_finish(sb);
8254 	sbuf_delete(sb);
8255 	return (rc);
8256 }
8257 
8258 void
8259 t4_os_dump_devlog(struct adapter *sc)
8260 {
8261 	int rc;
8262 	struct sbuf sb;
8263 
8264 	if (sbuf_new(&sb, NULL, 4096, SBUF_AUTOEXTEND) != &sb)
8265 		return;
8266 	rc = sbuf_devlog(sc, &sb, M_NOWAIT);
8267 	if (rc == 0) {
8268 		rc = sbuf_finish(&sb);
8269 		if (rc == 0) {
8270 			log(LOG_DEBUG, "%s: device log follows.\n%s",
8271 		    		device_get_nameunit(sc->dev), sbuf_data(&sb));
8272 		}
8273 	}
8274 	sbuf_delete(&sb);
8275 }
8276 
8277 static int
8278 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
8279 {
8280 	struct adapter *sc = arg1;
8281 	struct sbuf *sb;
8282 	int rc;
8283 	struct tp_fcoe_stats stats[MAX_NCHAN];
8284 	int i, nchan = sc->chip_params->nchan;
8285 
8286 	rc = sysctl_wire_old_buffer(req, 0);
8287 	if (rc != 0)
8288 		return (rc);
8289 
8290 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8291 	if (sb == NULL)
8292 		return (ENOMEM);
8293 
8294 	for (i = 0; i < nchan; i++)
8295 		t4_get_fcoe_stats(sc, i, &stats[i], 1);
8296 
8297 	if (nchan > 2) {
8298 		sbuf_printf(sb, "                   channel 0        channel 1"
8299 		    "        channel 2        channel 3");
8300 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
8301 		    stats[0].octets_ddp, stats[1].octets_ddp,
8302 		    stats[2].octets_ddp, stats[3].octets_ddp);
8303 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
8304 		    stats[0].frames_ddp, stats[1].frames_ddp,
8305 		    stats[2].frames_ddp, stats[3].frames_ddp);
8306 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
8307 		    stats[0].frames_drop, stats[1].frames_drop,
8308 		    stats[2].frames_drop, stats[3].frames_drop);
8309 	} else {
8310 		sbuf_printf(sb, "                   channel 0        channel 1");
8311 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
8312 		    stats[0].octets_ddp, stats[1].octets_ddp);
8313 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
8314 		    stats[0].frames_ddp, stats[1].frames_ddp);
8315 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
8316 		    stats[0].frames_drop, stats[1].frames_drop);
8317 	}
8318 
8319 	rc = sbuf_finish(sb);
8320 	sbuf_delete(sb);
8321 
8322 	return (rc);
8323 }
8324 
8325 static int
8326 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
8327 {
8328 	struct adapter *sc = arg1;
8329 	struct sbuf *sb;
8330 	int rc, i;
8331 	unsigned int map, kbps, ipg, mode;
8332 	unsigned int pace_tab[NTX_SCHED];
8333 
8334 	rc = sysctl_wire_old_buffer(req, 0);
8335 	if (rc != 0)
8336 		return (rc);
8337 
8338 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8339 	if (sb == NULL)
8340 		return (ENOMEM);
8341 
8342 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
8343 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
8344 	t4_read_pace_tbl(sc, pace_tab);
8345 
8346 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
8347 	    "Class IPG (0.1 ns)   Flow IPG (us)");
8348 
8349 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
8350 		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
8351 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
8352 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
8353 		if (kbps)
8354 			sbuf_printf(sb, "%9u     ", kbps);
8355 		else
8356 			sbuf_printf(sb, " disabled     ");
8357 
8358 		if (ipg)
8359 			sbuf_printf(sb, "%13u        ", ipg);
8360 		else
8361 			sbuf_printf(sb, "     disabled        ");
8362 
8363 		if (pace_tab[i])
8364 			sbuf_printf(sb, "%10u", pace_tab[i]);
8365 		else
8366 			sbuf_printf(sb, "  disabled");
8367 	}
8368 
8369 	rc = sbuf_finish(sb);
8370 	sbuf_delete(sb);
8371 
8372 	return (rc);
8373 }
8374 
8375 static int
8376 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
8377 {
8378 	struct adapter *sc = arg1;
8379 	struct sbuf *sb;
8380 	int rc, i, j;
8381 	uint64_t *p0, *p1;
8382 	struct lb_port_stats s[2];
8383 	static const char *stat_name[] = {
8384 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
8385 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
8386 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
8387 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
8388 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
8389 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
8390 		"BG2FramesTrunc:", "BG3FramesTrunc:"
8391 	};
8392 
8393 	rc = sysctl_wire_old_buffer(req, 0);
8394 	if (rc != 0)
8395 		return (rc);
8396 
8397 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8398 	if (sb == NULL)
8399 		return (ENOMEM);
8400 
8401 	memset(s, 0, sizeof(s));
8402 
8403 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
8404 		t4_get_lb_stats(sc, i, &s[0]);
8405 		t4_get_lb_stats(sc, i + 1, &s[1]);
8406 
8407 		p0 = &s[0].octets;
8408 		p1 = &s[1].octets;
8409 		sbuf_printf(sb, "%s                       Loopback %u"
8410 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
8411 
8412 		for (j = 0; j < nitems(stat_name); j++)
8413 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
8414 				   *p0++, *p1++);
8415 	}
8416 
8417 	rc = sbuf_finish(sb);
8418 	sbuf_delete(sb);
8419 
8420 	return (rc);
8421 }
8422 
8423 static int
8424 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
8425 {
8426 	int rc = 0;
8427 	struct port_info *pi = arg1;
8428 	struct link_config *lc = &pi->link_cfg;
8429 	struct sbuf *sb;
8430 
8431 	rc = sysctl_wire_old_buffer(req, 0);
8432 	if (rc != 0)
8433 		return(rc);
8434 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
8435 	if (sb == NULL)
8436 		return (ENOMEM);
8437 
8438 	if (lc->link_ok || lc->link_down_rc == 255)
8439 		sbuf_printf(sb, "n/a");
8440 	else
8441 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
8442 
8443 	rc = sbuf_finish(sb);
8444 	sbuf_delete(sb);
8445 
8446 	return (rc);
8447 }
8448 
8449 struct mem_desc {
8450 	unsigned int base;
8451 	unsigned int limit;
8452 	unsigned int idx;
8453 };
8454 
8455 static int
8456 mem_desc_cmp(const void *a, const void *b)
8457 {
8458 	return ((const struct mem_desc *)a)->base -
8459 	       ((const struct mem_desc *)b)->base;
8460 }
8461 
8462 static void
8463 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
8464     unsigned int to)
8465 {
8466 	unsigned int size;
8467 
8468 	if (from == to)
8469 		return;
8470 
8471 	size = to - from + 1;
8472 	if (size == 0)
8473 		return;
8474 
8475 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
8476 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
8477 }
8478 
8479 static int
8480 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
8481 {
8482 	struct adapter *sc = arg1;
8483 	struct sbuf *sb;
8484 	int rc, i, n;
8485 	uint32_t lo, hi, used, alloc;
8486 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
8487 	static const char *region[] = {
8488 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
8489 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
8490 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
8491 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
8492 		"RQUDP region:", "PBL region:", "TXPBL region:",
8493 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
8494 		"On-chip queues:", "TLS keys:",
8495 	};
8496 	struct mem_desc avail[4];
8497 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
8498 	struct mem_desc *md = mem;
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, 4096, req);
8505 	if (sb == NULL)
8506 		return (ENOMEM);
8507 
8508 	for (i = 0; i < nitems(mem); i++) {
8509 		mem[i].limit = 0;
8510 		mem[i].idx = i;
8511 	}
8512 
8513 	/* Find and sort the populated memory ranges */
8514 	i = 0;
8515 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
8516 	if (lo & F_EDRAM0_ENABLE) {
8517 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
8518 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
8519 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
8520 		avail[i].idx = 0;
8521 		i++;
8522 	}
8523 	if (lo & F_EDRAM1_ENABLE) {
8524 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
8525 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
8526 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
8527 		avail[i].idx = 1;
8528 		i++;
8529 	}
8530 	if (lo & F_EXT_MEM_ENABLE) {
8531 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
8532 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
8533 		avail[i].limit = avail[i].base +
8534 		    (G_EXT_MEM_SIZE(hi) << 20);
8535 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
8536 		i++;
8537 	}
8538 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
8539 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
8540 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
8541 		avail[i].limit = avail[i].base +
8542 		    (G_EXT_MEM1_SIZE(hi) << 20);
8543 		avail[i].idx = 4;
8544 		i++;
8545 	}
8546 	if (!i)                                    /* no memory available */
8547 		return 0;
8548 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
8549 
8550 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
8551 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
8552 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
8553 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
8554 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
8555 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
8556 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
8557 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
8558 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
8559 
8560 	/* the next few have explicit upper bounds */
8561 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
8562 	md->limit = md->base - 1 +
8563 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
8564 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
8565 	md++;
8566 
8567 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
8568 	md->limit = md->base - 1 +
8569 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
8570 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
8571 	md++;
8572 
8573 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
8574 		if (chip_id(sc) <= CHELSIO_T5)
8575 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
8576 		else
8577 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
8578 		md->limit = 0;
8579 	} else {
8580 		md->base = 0;
8581 		md->idx = nitems(region);  /* hide it */
8582 	}
8583 	md++;
8584 
8585 #define ulp_region(reg) \
8586 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
8587 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
8588 
8589 	ulp_region(RX_ISCSI);
8590 	ulp_region(RX_TDDP);
8591 	ulp_region(TX_TPT);
8592 	ulp_region(RX_STAG);
8593 	ulp_region(RX_RQ);
8594 	ulp_region(RX_RQUDP);
8595 	ulp_region(RX_PBL);
8596 	ulp_region(TX_PBL);
8597 #undef ulp_region
8598 
8599 	md->base = 0;
8600 	md->idx = nitems(region);
8601 	if (!is_t4(sc)) {
8602 		uint32_t size = 0;
8603 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
8604 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
8605 
8606 		if (is_t5(sc)) {
8607 			if (sge_ctrl & F_VFIFO_ENABLE)
8608 				size = G_DBVFIFO_SIZE(fifo_size);
8609 		} else
8610 			size = G_T6_DBVFIFO_SIZE(fifo_size);
8611 
8612 		if (size) {
8613 			md->base = G_BASEADDR(t4_read_reg(sc,
8614 			    A_SGE_DBVFIFO_BADDR));
8615 			md->limit = md->base + (size << 2) - 1;
8616 		}
8617 	}
8618 	md++;
8619 
8620 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
8621 	md->limit = 0;
8622 	md++;
8623 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
8624 	md->limit = 0;
8625 	md++;
8626 
8627 	md->base = sc->vres.ocq.start;
8628 	if (sc->vres.ocq.size)
8629 		md->limit = md->base + sc->vres.ocq.size - 1;
8630 	else
8631 		md->idx = nitems(region);  /* hide it */
8632 	md++;
8633 
8634 	md->base = sc->vres.key.start;
8635 	if (sc->vres.key.size)
8636 		md->limit = md->base + sc->vres.key.size - 1;
8637 	else
8638 		md->idx = nitems(region);  /* hide it */
8639 	md++;
8640 
8641 	/* add any address-space holes, there can be up to 3 */
8642 	for (n = 0; n < i - 1; n++)
8643 		if (avail[n].limit < avail[n + 1].base)
8644 			(md++)->base = avail[n].limit;
8645 	if (avail[n].limit)
8646 		(md++)->base = avail[n].limit;
8647 
8648 	n = md - mem;
8649 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
8650 
8651 	for (lo = 0; lo < i; lo++)
8652 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
8653 				avail[lo].limit - 1);
8654 
8655 	sbuf_printf(sb, "\n");
8656 	for (i = 0; i < n; i++) {
8657 		if (mem[i].idx >= nitems(region))
8658 			continue;                        /* skip holes */
8659 		if (!mem[i].limit)
8660 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
8661 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
8662 				mem[i].limit);
8663 	}
8664 
8665 	sbuf_printf(sb, "\n");
8666 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
8667 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
8668 	mem_region_show(sb, "uP RAM:", lo, hi);
8669 
8670 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
8671 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
8672 	mem_region_show(sb, "uP Extmem2:", lo, hi);
8673 
8674 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
8675 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
8676 		   G_PMRXMAXPAGE(lo),
8677 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
8678 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
8679 
8680 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
8681 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
8682 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
8683 		   G_PMTXMAXPAGE(lo),
8684 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
8685 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
8686 	sbuf_printf(sb, "%u p-structs\n",
8687 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
8688 
8689 	for (i = 0; i < 4; i++) {
8690 		if (chip_id(sc) > CHELSIO_T5)
8691 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
8692 		else
8693 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
8694 		if (is_t5(sc)) {
8695 			used = G_T5_USED(lo);
8696 			alloc = G_T5_ALLOC(lo);
8697 		} else {
8698 			used = G_USED(lo);
8699 			alloc = G_ALLOC(lo);
8700 		}
8701 		/* For T6 these are MAC buffer groups */
8702 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
8703 		    i, used, alloc);
8704 	}
8705 	for (i = 0; i < sc->chip_params->nchan; i++) {
8706 		if (chip_id(sc) > CHELSIO_T5)
8707 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
8708 		else
8709 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
8710 		if (is_t5(sc)) {
8711 			used = G_T5_USED(lo);
8712 			alloc = G_T5_ALLOC(lo);
8713 		} else {
8714 			used = G_USED(lo);
8715 			alloc = G_ALLOC(lo);
8716 		}
8717 		/* For T6 these are MAC buffer groups */
8718 		sbuf_printf(sb,
8719 		    "\nLoopback %d using %u pages out of %u allocated",
8720 		    i, used, alloc);
8721 	}
8722 
8723 	rc = sbuf_finish(sb);
8724 	sbuf_delete(sb);
8725 
8726 	return (rc);
8727 }
8728 
8729 static inline void
8730 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
8731 {
8732 	*mask = x | y;
8733 	y = htobe64(y);
8734 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
8735 }
8736 
8737 static int
8738 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
8739 {
8740 	struct adapter *sc = arg1;
8741 	struct sbuf *sb;
8742 	int rc, i;
8743 
8744 	MPASS(chip_id(sc) <= CHELSIO_T5);
8745 
8746 	rc = sysctl_wire_old_buffer(req, 0);
8747 	if (rc != 0)
8748 		return (rc);
8749 
8750 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8751 	if (sb == NULL)
8752 		return (ENOMEM);
8753 
8754 	sbuf_printf(sb,
8755 	    "Idx  Ethernet address     Mask     Vld Ports PF"
8756 	    "  VF              Replication             P0 P1 P2 P3  ML");
8757 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
8758 		uint64_t tcamx, tcamy, mask;
8759 		uint32_t cls_lo, cls_hi;
8760 		uint8_t addr[ETHER_ADDR_LEN];
8761 
8762 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
8763 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
8764 		if (tcamx & tcamy)
8765 			continue;
8766 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
8767 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
8768 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
8769 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
8770 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
8771 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
8772 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
8773 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
8774 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
8775 
8776 		if (cls_lo & F_REPLICATE) {
8777 			struct fw_ldst_cmd ldst_cmd;
8778 
8779 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
8780 			ldst_cmd.op_to_addrspace =
8781 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
8782 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
8783 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
8784 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
8785 			ldst_cmd.u.mps.rplc.fid_idx =
8786 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
8787 				V_FW_LDST_CMD_IDX(i));
8788 
8789 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8790 			    "t4mps");
8791 			if (rc)
8792 				break;
8793 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
8794 			    sizeof(ldst_cmd), &ldst_cmd);
8795 			end_synchronized_op(sc, 0);
8796 
8797 			if (rc != 0) {
8798 				sbuf_printf(sb, "%36d", rc);
8799 				rc = 0;
8800 			} else {
8801 				sbuf_printf(sb, " %08x %08x %08x %08x",
8802 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
8803 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
8804 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
8805 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
8806 			}
8807 		} else
8808 			sbuf_printf(sb, "%36s", "");
8809 
8810 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
8811 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
8812 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
8813 	}
8814 
8815 	if (rc)
8816 		(void) sbuf_finish(sb);
8817 	else
8818 		rc = sbuf_finish(sb);
8819 	sbuf_delete(sb);
8820 
8821 	return (rc);
8822 }
8823 
8824 static int
8825 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
8826 {
8827 	struct adapter *sc = arg1;
8828 	struct sbuf *sb;
8829 	int rc, i;
8830 
8831 	MPASS(chip_id(sc) > CHELSIO_T5);
8832 
8833 	rc = sysctl_wire_old_buffer(req, 0);
8834 	if (rc != 0)
8835 		return (rc);
8836 
8837 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8838 	if (sb == NULL)
8839 		return (ENOMEM);
8840 
8841 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
8842 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
8843 	    "                           Replication"
8844 	    "                                    P0 P1 P2 P3  ML\n");
8845 
8846 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
8847 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
8848 		uint16_t ivlan;
8849 		uint64_t tcamx, tcamy, val, mask;
8850 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
8851 		uint8_t addr[ETHER_ADDR_LEN];
8852 
8853 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
8854 		if (i < 256)
8855 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
8856 		else
8857 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
8858 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8859 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8860 		tcamy = G_DMACH(val) << 32;
8861 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8862 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8863 		lookup_type = G_DATALKPTYPE(data2);
8864 		port_num = G_DATAPORTNUM(data2);
8865 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8866 			/* Inner header VNI */
8867 			vniy = ((data2 & F_DATAVIDH2) << 23) |
8868 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8869 			dip_hit = data2 & F_DATADIPHIT;
8870 			vlan_vld = 0;
8871 		} else {
8872 			vniy = 0;
8873 			dip_hit = 0;
8874 			vlan_vld = data2 & F_DATAVIDH2;
8875 			ivlan = G_VIDL(val);
8876 		}
8877 
8878 		ctl |= V_CTLXYBITSEL(1);
8879 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8880 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8881 		tcamx = G_DMACH(val) << 32;
8882 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8883 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8884 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8885 			/* Inner header VNI mask */
8886 			vnix = ((data2 & F_DATAVIDH2) << 23) |
8887 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8888 		} else
8889 			vnix = 0;
8890 
8891 		if (tcamx & tcamy)
8892 			continue;
8893 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
8894 
8895 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
8896 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
8897 
8898 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8899 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8900 			    "%012jx %06x %06x    -    -   %3c"
8901 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
8902 			    addr[1], addr[2], addr[3], addr[4], addr[5],
8903 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
8904 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8905 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8906 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8907 		} else {
8908 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8909 			    "%012jx    -       -   ", i, addr[0], addr[1],
8910 			    addr[2], addr[3], addr[4], addr[5],
8911 			    (uintmax_t)mask);
8912 
8913 			if (vlan_vld)
8914 				sbuf_printf(sb, "%4u   Y     ", ivlan);
8915 			else
8916 				sbuf_printf(sb, "  -    N     ");
8917 
8918 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
8919 			    lookup_type ? 'I' : 'O', port_num,
8920 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8921 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8922 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8923 		}
8924 
8925 
8926 		if (cls_lo & F_T6_REPLICATE) {
8927 			struct fw_ldst_cmd ldst_cmd;
8928 
8929 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
8930 			ldst_cmd.op_to_addrspace =
8931 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
8932 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
8933 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
8934 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
8935 			ldst_cmd.u.mps.rplc.fid_idx =
8936 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
8937 				V_FW_LDST_CMD_IDX(i));
8938 
8939 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8940 			    "t6mps");
8941 			if (rc)
8942 				break;
8943 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
8944 			    sizeof(ldst_cmd), &ldst_cmd);
8945 			end_synchronized_op(sc, 0);
8946 
8947 			if (rc != 0) {
8948 				sbuf_printf(sb, "%72d", rc);
8949 				rc = 0;
8950 			} else {
8951 				sbuf_printf(sb, " %08x %08x %08x %08x"
8952 				    " %08x %08x %08x %08x",
8953 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
8954 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
8955 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
8956 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
8957 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
8958 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
8959 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
8960 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
8961 			}
8962 		} else
8963 			sbuf_printf(sb, "%72s", "");
8964 
8965 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
8966 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
8967 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
8968 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
8969 	}
8970 
8971 	if (rc)
8972 		(void) sbuf_finish(sb);
8973 	else
8974 		rc = sbuf_finish(sb);
8975 	sbuf_delete(sb);
8976 
8977 	return (rc);
8978 }
8979 
8980 static int
8981 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
8982 {
8983 	struct adapter *sc = arg1;
8984 	struct sbuf *sb;
8985 	int rc;
8986 	uint16_t mtus[NMTUS];
8987 
8988 	rc = sysctl_wire_old_buffer(req, 0);
8989 	if (rc != 0)
8990 		return (rc);
8991 
8992 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8993 	if (sb == NULL)
8994 		return (ENOMEM);
8995 
8996 	t4_read_mtu_tbl(sc, mtus, NULL);
8997 
8998 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
8999 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
9000 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
9001 	    mtus[14], mtus[15]);
9002 
9003 	rc = sbuf_finish(sb);
9004 	sbuf_delete(sb);
9005 
9006 	return (rc);
9007 }
9008 
9009 static int
9010 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
9011 {
9012 	struct adapter *sc = arg1;
9013 	struct sbuf *sb;
9014 	int rc, i;
9015 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
9016 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
9017 	static const char *tx_stats[MAX_PM_NSTATS] = {
9018 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
9019 		"Tx FIFO wait", NULL, "Tx latency"
9020 	};
9021 	static const char *rx_stats[MAX_PM_NSTATS] = {
9022 		"Read:", "Write bypass:", "Write mem:", "Flush:",
9023 		"Rx FIFO wait", NULL, "Rx latency"
9024 	};
9025 
9026 	rc = sysctl_wire_old_buffer(req, 0);
9027 	if (rc != 0)
9028 		return (rc);
9029 
9030 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
9031 	if (sb == NULL)
9032 		return (ENOMEM);
9033 
9034 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
9035 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
9036 
9037 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
9038 	for (i = 0; i < 4; i++) {
9039 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
9040 		    tx_cyc[i]);
9041 	}
9042 
9043 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
9044 	for (i = 0; i < 4; i++) {
9045 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
9046 		    rx_cyc[i]);
9047 	}
9048 
9049 	if (chip_id(sc) > CHELSIO_T5) {
9050 		sbuf_printf(sb,
9051 		    "\n              Total wait      Total occupancy");
9052 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
9053 		    tx_cyc[i]);
9054 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
9055 		    rx_cyc[i]);
9056 
9057 		i += 2;
9058 		MPASS(i < nitems(tx_stats));
9059 
9060 		sbuf_printf(sb,
9061 		    "\n                   Reads           Total wait");
9062 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
9063 		    tx_cyc[i]);
9064 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
9065 		    rx_cyc[i]);
9066 	}
9067 
9068 	rc = sbuf_finish(sb);
9069 	sbuf_delete(sb);
9070 
9071 	return (rc);
9072 }
9073 
9074 static int
9075 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
9076 {
9077 	struct adapter *sc = arg1;
9078 	struct sbuf *sb;
9079 	int rc;
9080 	struct tp_rdma_stats stats;
9081 
9082 	rc = sysctl_wire_old_buffer(req, 0);
9083 	if (rc != 0)
9084 		return (rc);
9085 
9086 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
9087 	if (sb == NULL)
9088 		return (ENOMEM);
9089 
9090 	mtx_lock(&sc->reg_lock);
9091 	t4_tp_get_rdma_stats(sc, &stats, 0);
9092 	mtx_unlock(&sc->reg_lock);
9093 
9094 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
9095 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
9096 
9097 	rc = sbuf_finish(sb);
9098 	sbuf_delete(sb);
9099 
9100 	return (rc);
9101 }
9102 
9103 static int
9104 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
9105 {
9106 	struct adapter *sc = arg1;
9107 	struct sbuf *sb;
9108 	int rc;
9109 	struct tp_tcp_stats v4, v6;
9110 
9111 	rc = sysctl_wire_old_buffer(req, 0);
9112 	if (rc != 0)
9113 		return (rc);
9114 
9115 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
9116 	if (sb == NULL)
9117 		return (ENOMEM);
9118 
9119 	mtx_lock(&sc->reg_lock);
9120 	t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
9121 	mtx_unlock(&sc->reg_lock);
9122 
9123 	sbuf_printf(sb,
9124 	    "                                IP                 IPv6\n");
9125 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
9126 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
9127 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
9128 	    v4.tcp_in_segs, v6.tcp_in_segs);
9129 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
9130 	    v4.tcp_out_segs, v6.tcp_out_segs);
9131 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
9132 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
9133 
9134 	rc = sbuf_finish(sb);
9135 	sbuf_delete(sb);
9136 
9137 	return (rc);
9138 }
9139 
9140 static int
9141 sysctl_tids(SYSCTL_HANDLER_ARGS)
9142 {
9143 	struct adapter *sc = arg1;
9144 	struct sbuf *sb;
9145 	int rc;
9146 	struct tid_info *t = &sc->tids;
9147 
9148 	rc = sysctl_wire_old_buffer(req, 0);
9149 	if (rc != 0)
9150 		return (rc);
9151 
9152 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
9153 	if (sb == NULL)
9154 		return (ENOMEM);
9155 
9156 	if (t->natids) {
9157 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
9158 		    t->atids_in_use);
9159 	}
9160 
9161 	if (t->nhpftids) {
9162 		sbuf_printf(sb, "HPFTID range: %u-%u, in use: %u\n",
9163 		    t->hpftid_base, t->hpftid_end, t->hpftids_in_use);
9164 	}
9165 
9166 	if (t->ntids) {
9167 		sbuf_printf(sb, "TID range: ");
9168 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
9169 			uint32_t b, hb;
9170 
9171 			if (chip_id(sc) <= CHELSIO_T5) {
9172 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
9173 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
9174 			} else {
9175 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
9176 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
9177 			}
9178 
9179 			if (b)
9180 				sbuf_printf(sb, "%u-%u, ", t->tid_base, b - 1);
9181 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
9182 		} else
9183 			sbuf_printf(sb, "%u-%u", t->tid_base, t->ntids - 1);
9184 		sbuf_printf(sb, ", in use: %u\n",
9185 		    atomic_load_acq_int(&t->tids_in_use));
9186 	}
9187 
9188 	if (t->nstids) {
9189 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
9190 		    t->stid_base + t->nstids - 1, t->stids_in_use);
9191 	}
9192 
9193 	if (t->nftids) {
9194 		sbuf_printf(sb, "FTID range: %u-%u, in use: %u\n", t->ftid_base,
9195 		    t->ftid_end, t->ftids_in_use);
9196 	}
9197 
9198 	if (t->netids) {
9199 		sbuf_printf(sb, "ETID range: %u-%u, in use: %u\n", t->etid_base,
9200 		    t->etid_base + t->netids - 1, t->etids_in_use);
9201 	}
9202 
9203 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
9204 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
9205 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
9206 
9207 	rc = sbuf_finish(sb);
9208 	sbuf_delete(sb);
9209 
9210 	return (rc);
9211 }
9212 
9213 static int
9214 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
9215 {
9216 	struct adapter *sc = arg1;
9217 	struct sbuf *sb;
9218 	int rc;
9219 	struct tp_err_stats stats;
9220 
9221 	rc = sysctl_wire_old_buffer(req, 0);
9222 	if (rc != 0)
9223 		return (rc);
9224 
9225 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
9226 	if (sb == NULL)
9227 		return (ENOMEM);
9228 
9229 	mtx_lock(&sc->reg_lock);
9230 	t4_tp_get_err_stats(sc, &stats, 0);
9231 	mtx_unlock(&sc->reg_lock);
9232 
9233 	if (sc->chip_params->nchan > 2) {
9234 		sbuf_printf(sb, "                 channel 0  channel 1"
9235 		    "  channel 2  channel 3\n");
9236 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
9237 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
9238 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
9239 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
9240 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
9241 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
9242 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
9243 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
9244 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
9245 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
9246 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
9247 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
9248 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
9249 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
9250 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
9251 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
9252 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
9253 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
9254 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
9255 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
9256 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
9257 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
9258 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
9259 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
9260 	} else {
9261 		sbuf_printf(sb, "                 channel 0  channel 1\n");
9262 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
9263 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
9264 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
9265 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
9266 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
9267 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
9268 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
9269 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
9270 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
9271 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
9272 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
9273 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
9274 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
9275 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
9276 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
9277 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
9278 	}
9279 
9280 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
9281 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
9282 
9283 	rc = sbuf_finish(sb);
9284 	sbuf_delete(sb);
9285 
9286 	return (rc);
9287 }
9288 
9289 static int
9290 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
9291 {
9292 	struct adapter *sc = arg1;
9293 	struct tp_params *tpp = &sc->params.tp;
9294 	u_int mask;
9295 	int rc;
9296 
9297 	mask = tpp->la_mask >> 16;
9298 	rc = sysctl_handle_int(oidp, &mask, 0, req);
9299 	if (rc != 0 || req->newptr == NULL)
9300 		return (rc);
9301 	if (mask > 0xffff)
9302 		return (EINVAL);
9303 	tpp->la_mask = mask << 16;
9304 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
9305 
9306 	return (0);
9307 }
9308 
9309 struct field_desc {
9310 	const char *name;
9311 	u_int start;
9312 	u_int width;
9313 };
9314 
9315 static void
9316 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
9317 {
9318 	char buf[32];
9319 	int line_size = 0;
9320 
9321 	while (f->name) {
9322 		uint64_t mask = (1ULL << f->width) - 1;
9323 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
9324 		    ((uintmax_t)v >> f->start) & mask);
9325 
9326 		if (line_size + len >= 79) {
9327 			line_size = 8;
9328 			sbuf_printf(sb, "\n        ");
9329 		}
9330 		sbuf_printf(sb, "%s ", buf);
9331 		line_size += len + 1;
9332 		f++;
9333 	}
9334 	sbuf_printf(sb, "\n");
9335 }
9336 
9337 static const struct field_desc tp_la0[] = {
9338 	{ "RcfOpCodeOut", 60, 4 },
9339 	{ "State", 56, 4 },
9340 	{ "WcfState", 52, 4 },
9341 	{ "RcfOpcSrcOut", 50, 2 },
9342 	{ "CRxError", 49, 1 },
9343 	{ "ERxError", 48, 1 },
9344 	{ "SanityFailed", 47, 1 },
9345 	{ "SpuriousMsg", 46, 1 },
9346 	{ "FlushInputMsg", 45, 1 },
9347 	{ "FlushInputCpl", 44, 1 },
9348 	{ "RssUpBit", 43, 1 },
9349 	{ "RssFilterHit", 42, 1 },
9350 	{ "Tid", 32, 10 },
9351 	{ "InitTcb", 31, 1 },
9352 	{ "LineNumber", 24, 7 },
9353 	{ "Emsg", 23, 1 },
9354 	{ "EdataOut", 22, 1 },
9355 	{ "Cmsg", 21, 1 },
9356 	{ "CdataOut", 20, 1 },
9357 	{ "EreadPdu", 19, 1 },
9358 	{ "CreadPdu", 18, 1 },
9359 	{ "TunnelPkt", 17, 1 },
9360 	{ "RcfPeerFin", 16, 1 },
9361 	{ "RcfReasonOut", 12, 4 },
9362 	{ "TxCchannel", 10, 2 },
9363 	{ "RcfTxChannel", 8, 2 },
9364 	{ "RxEchannel", 6, 2 },
9365 	{ "RcfRxChannel", 5, 1 },
9366 	{ "RcfDataOutSrdy", 4, 1 },
9367 	{ "RxDvld", 3, 1 },
9368 	{ "RxOoDvld", 2, 1 },
9369 	{ "RxCongestion", 1, 1 },
9370 	{ "TxCongestion", 0, 1 },
9371 	{ NULL }
9372 };
9373 
9374 static const struct field_desc tp_la1[] = {
9375 	{ "CplCmdIn", 56, 8 },
9376 	{ "CplCmdOut", 48, 8 },
9377 	{ "ESynOut", 47, 1 },
9378 	{ "EAckOut", 46, 1 },
9379 	{ "EFinOut", 45, 1 },
9380 	{ "ERstOut", 44, 1 },
9381 	{ "SynIn", 43, 1 },
9382 	{ "AckIn", 42, 1 },
9383 	{ "FinIn", 41, 1 },
9384 	{ "RstIn", 40, 1 },
9385 	{ "DataIn", 39, 1 },
9386 	{ "DataInVld", 38, 1 },
9387 	{ "PadIn", 37, 1 },
9388 	{ "RxBufEmpty", 36, 1 },
9389 	{ "RxDdp", 35, 1 },
9390 	{ "RxFbCongestion", 34, 1 },
9391 	{ "TxFbCongestion", 33, 1 },
9392 	{ "TxPktSumSrdy", 32, 1 },
9393 	{ "RcfUlpType", 28, 4 },
9394 	{ "Eread", 27, 1 },
9395 	{ "Ebypass", 26, 1 },
9396 	{ "Esave", 25, 1 },
9397 	{ "Static0", 24, 1 },
9398 	{ "Cread", 23, 1 },
9399 	{ "Cbypass", 22, 1 },
9400 	{ "Csave", 21, 1 },
9401 	{ "CPktOut", 20, 1 },
9402 	{ "RxPagePoolFull", 18, 2 },
9403 	{ "RxLpbkPkt", 17, 1 },
9404 	{ "TxLpbkPkt", 16, 1 },
9405 	{ "RxVfValid", 15, 1 },
9406 	{ "SynLearned", 14, 1 },
9407 	{ "SetDelEntry", 13, 1 },
9408 	{ "SetInvEntry", 12, 1 },
9409 	{ "CpcmdDvld", 11, 1 },
9410 	{ "CpcmdSave", 10, 1 },
9411 	{ "RxPstructsFull", 8, 2 },
9412 	{ "EpcmdDvld", 7, 1 },
9413 	{ "EpcmdFlush", 6, 1 },
9414 	{ "EpcmdTrimPrefix", 5, 1 },
9415 	{ "EpcmdTrimPostfix", 4, 1 },
9416 	{ "ERssIp4Pkt", 3, 1 },
9417 	{ "ERssIp6Pkt", 2, 1 },
9418 	{ "ERssTcpUdpPkt", 1, 1 },
9419 	{ "ERssFceFipPkt", 0, 1 },
9420 	{ NULL }
9421 };
9422 
9423 static const struct field_desc tp_la2[] = {
9424 	{ "CplCmdIn", 56, 8 },
9425 	{ "MpsVfVld", 55, 1 },
9426 	{ "MpsPf", 52, 3 },
9427 	{ "MpsVf", 44, 8 },
9428 	{ "SynIn", 43, 1 },
9429 	{ "AckIn", 42, 1 },
9430 	{ "FinIn", 41, 1 },
9431 	{ "RstIn", 40, 1 },
9432 	{ "DataIn", 39, 1 },
9433 	{ "DataInVld", 38, 1 },
9434 	{ "PadIn", 37, 1 },
9435 	{ "RxBufEmpty", 36, 1 },
9436 	{ "RxDdp", 35, 1 },
9437 	{ "RxFbCongestion", 34, 1 },
9438 	{ "TxFbCongestion", 33, 1 },
9439 	{ "TxPktSumSrdy", 32, 1 },
9440 	{ "RcfUlpType", 28, 4 },
9441 	{ "Eread", 27, 1 },
9442 	{ "Ebypass", 26, 1 },
9443 	{ "Esave", 25, 1 },
9444 	{ "Static0", 24, 1 },
9445 	{ "Cread", 23, 1 },
9446 	{ "Cbypass", 22, 1 },
9447 	{ "Csave", 21, 1 },
9448 	{ "CPktOut", 20, 1 },
9449 	{ "RxPagePoolFull", 18, 2 },
9450 	{ "RxLpbkPkt", 17, 1 },
9451 	{ "TxLpbkPkt", 16, 1 },
9452 	{ "RxVfValid", 15, 1 },
9453 	{ "SynLearned", 14, 1 },
9454 	{ "SetDelEntry", 13, 1 },
9455 	{ "SetInvEntry", 12, 1 },
9456 	{ "CpcmdDvld", 11, 1 },
9457 	{ "CpcmdSave", 10, 1 },
9458 	{ "RxPstructsFull", 8, 2 },
9459 	{ "EpcmdDvld", 7, 1 },
9460 	{ "EpcmdFlush", 6, 1 },
9461 	{ "EpcmdTrimPrefix", 5, 1 },
9462 	{ "EpcmdTrimPostfix", 4, 1 },
9463 	{ "ERssIp4Pkt", 3, 1 },
9464 	{ "ERssIp6Pkt", 2, 1 },
9465 	{ "ERssTcpUdpPkt", 1, 1 },
9466 	{ "ERssFceFipPkt", 0, 1 },
9467 	{ NULL }
9468 };
9469 
9470 static void
9471 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
9472 {
9473 
9474 	field_desc_show(sb, *p, tp_la0);
9475 }
9476 
9477 static void
9478 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
9479 {
9480 
9481 	if (idx)
9482 		sbuf_printf(sb, "\n");
9483 	field_desc_show(sb, p[0], tp_la0);
9484 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
9485 		field_desc_show(sb, p[1], tp_la0);
9486 }
9487 
9488 static void
9489 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
9490 {
9491 
9492 	if (idx)
9493 		sbuf_printf(sb, "\n");
9494 	field_desc_show(sb, p[0], tp_la0);
9495 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
9496 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
9497 }
9498 
9499 static int
9500 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
9501 {
9502 	struct adapter *sc = arg1;
9503 	struct sbuf *sb;
9504 	uint64_t *buf, *p;
9505 	int rc;
9506 	u_int i, inc;
9507 	void (*show_func)(struct sbuf *, uint64_t *, int);
9508 
9509 	rc = sysctl_wire_old_buffer(req, 0);
9510 	if (rc != 0)
9511 		return (rc);
9512 
9513 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9514 	if (sb == NULL)
9515 		return (ENOMEM);
9516 
9517 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
9518 
9519 	t4_tp_read_la(sc, buf, NULL);
9520 	p = buf;
9521 
9522 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
9523 	case 2:
9524 		inc = 2;
9525 		show_func = tp_la_show2;
9526 		break;
9527 	case 3:
9528 		inc = 2;
9529 		show_func = tp_la_show3;
9530 		break;
9531 	default:
9532 		inc = 1;
9533 		show_func = tp_la_show;
9534 	}
9535 
9536 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
9537 		(*show_func)(sb, p, i);
9538 
9539 	rc = sbuf_finish(sb);
9540 	sbuf_delete(sb);
9541 	free(buf, M_CXGBE);
9542 	return (rc);
9543 }
9544 
9545 static int
9546 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
9547 {
9548 	struct adapter *sc = arg1;
9549 	struct sbuf *sb;
9550 	int rc;
9551 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
9552 
9553 	rc = sysctl_wire_old_buffer(req, 0);
9554 	if (rc != 0)
9555 		return (rc);
9556 
9557 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
9558 	if (sb == NULL)
9559 		return (ENOMEM);
9560 
9561 	t4_get_chan_txrate(sc, nrate, orate);
9562 
9563 	if (sc->chip_params->nchan > 2) {
9564 		sbuf_printf(sb, "              channel 0   channel 1"
9565 		    "   channel 2   channel 3\n");
9566 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
9567 		    nrate[0], nrate[1], nrate[2], nrate[3]);
9568 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
9569 		    orate[0], orate[1], orate[2], orate[3]);
9570 	} else {
9571 		sbuf_printf(sb, "              channel 0   channel 1\n");
9572 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
9573 		    nrate[0], nrate[1]);
9574 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
9575 		    orate[0], orate[1]);
9576 	}
9577 
9578 	rc = sbuf_finish(sb);
9579 	sbuf_delete(sb);
9580 
9581 	return (rc);
9582 }
9583 
9584 static int
9585 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
9586 {
9587 	struct adapter *sc = arg1;
9588 	struct sbuf *sb;
9589 	uint32_t *buf, *p;
9590 	int rc, i;
9591 
9592 	rc = sysctl_wire_old_buffer(req, 0);
9593 	if (rc != 0)
9594 		return (rc);
9595 
9596 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9597 	if (sb == NULL)
9598 		return (ENOMEM);
9599 
9600 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
9601 	    M_ZERO | M_WAITOK);
9602 
9603 	t4_ulprx_read_la(sc, buf);
9604 	p = buf;
9605 
9606 	sbuf_printf(sb, "      Pcmd        Type   Message"
9607 	    "                Data");
9608 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
9609 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
9610 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
9611 	}
9612 
9613 	rc = sbuf_finish(sb);
9614 	sbuf_delete(sb);
9615 	free(buf, M_CXGBE);
9616 	return (rc);
9617 }
9618 
9619 static int
9620 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
9621 {
9622 	struct adapter *sc = arg1;
9623 	struct sbuf *sb;
9624 	int rc, v;
9625 
9626 	MPASS(chip_id(sc) >= CHELSIO_T5);
9627 
9628 	rc = sysctl_wire_old_buffer(req, 0);
9629 	if (rc != 0)
9630 		return (rc);
9631 
9632 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9633 	if (sb == NULL)
9634 		return (ENOMEM);
9635 
9636 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
9637 	if (G_STATSOURCE_T5(v) == 7) {
9638 		int mode;
9639 
9640 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
9641 		if (mode == 0) {
9642 			sbuf_printf(sb, "total %d, incomplete %d",
9643 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
9644 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
9645 		} else if (mode == 1) {
9646 			sbuf_printf(sb, "total %d, data overflow %d",
9647 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
9648 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
9649 		} else {
9650 			sbuf_printf(sb, "unknown mode %d", mode);
9651 		}
9652 	}
9653 	rc = sbuf_finish(sb);
9654 	sbuf_delete(sb);
9655 
9656 	return (rc);
9657 }
9658 
9659 static int
9660 sysctl_cpus(SYSCTL_HANDLER_ARGS)
9661 {
9662 	struct adapter *sc = arg1;
9663 	enum cpu_sets op = arg2;
9664 	cpuset_t cpuset;
9665 	struct sbuf *sb;
9666 	int i, rc;
9667 
9668 	MPASS(op == LOCAL_CPUS || op == INTR_CPUS);
9669 
9670 	CPU_ZERO(&cpuset);
9671 	rc = bus_get_cpus(sc->dev, op, sizeof(cpuset), &cpuset);
9672 	if (rc != 0)
9673 		return (rc);
9674 
9675 	rc = sysctl_wire_old_buffer(req, 0);
9676 	if (rc != 0)
9677 		return (rc);
9678 
9679 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9680 	if (sb == NULL)
9681 		return (ENOMEM);
9682 
9683 	CPU_FOREACH(i)
9684 		sbuf_printf(sb, "%d ", i);
9685 	rc = sbuf_finish(sb);
9686 	sbuf_delete(sb);
9687 
9688 	return (rc);
9689 }
9690 
9691 #ifdef TCP_OFFLOAD
9692 static int
9693 sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS)
9694 {
9695 	struct adapter *sc = arg1;
9696 	int *old_ports, *new_ports;
9697 	int i, new_count, rc;
9698 
9699 	if (req->newptr == NULL && req->oldptr == NULL)
9700 		return (SYSCTL_OUT(req, NULL, imax(sc->tt.num_tls_rx_ports, 1) *
9701 		    sizeof(sc->tt.tls_rx_ports[0])));
9702 
9703 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tlsrx");
9704 	if (rc)
9705 		return (rc);
9706 
9707 	if (sc->tt.num_tls_rx_ports == 0) {
9708 		i = -1;
9709 		rc = SYSCTL_OUT(req, &i, sizeof(i));
9710 	} else
9711 		rc = SYSCTL_OUT(req, sc->tt.tls_rx_ports,
9712 		    sc->tt.num_tls_rx_ports * sizeof(sc->tt.tls_rx_ports[0]));
9713 	if (rc == 0 && req->newptr != NULL) {
9714 		new_count = req->newlen / sizeof(new_ports[0]);
9715 		new_ports = malloc(new_count * sizeof(new_ports[0]), M_CXGBE,
9716 		    M_WAITOK);
9717 		rc = SYSCTL_IN(req, new_ports, new_count *
9718 		    sizeof(new_ports[0]));
9719 		if (rc)
9720 			goto err;
9721 
9722 		/* Allow setting to a single '-1' to clear the list. */
9723 		if (new_count == 1 && new_ports[0] == -1) {
9724 			ADAPTER_LOCK(sc);
9725 			old_ports = sc->tt.tls_rx_ports;
9726 			sc->tt.tls_rx_ports = NULL;
9727 			sc->tt.num_tls_rx_ports = 0;
9728 			ADAPTER_UNLOCK(sc);
9729 			free(old_ports, M_CXGBE);
9730 		} else {
9731 			for (i = 0; i < new_count; i++) {
9732 				if (new_ports[i] < 1 ||
9733 				    new_ports[i] > IPPORT_MAX) {
9734 					rc = EINVAL;
9735 					goto err;
9736 				}
9737 			}
9738 
9739 			ADAPTER_LOCK(sc);
9740 			old_ports = sc->tt.tls_rx_ports;
9741 			sc->tt.tls_rx_ports = new_ports;
9742 			sc->tt.num_tls_rx_ports = new_count;
9743 			ADAPTER_UNLOCK(sc);
9744 			free(old_ports, M_CXGBE);
9745 			new_ports = NULL;
9746 		}
9747 	err:
9748 		free(new_ports, M_CXGBE);
9749 	}
9750 	end_synchronized_op(sc, 0);
9751 	return (rc);
9752 }
9753 
9754 static void
9755 unit_conv(char *buf, size_t len, u_int val, u_int factor)
9756 {
9757 	u_int rem = val % factor;
9758 
9759 	if (rem == 0)
9760 		snprintf(buf, len, "%u", val / factor);
9761 	else {
9762 		while (rem % 10 == 0)
9763 			rem /= 10;
9764 		snprintf(buf, len, "%u.%u", val / factor, rem);
9765 	}
9766 }
9767 
9768 static int
9769 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
9770 {
9771 	struct adapter *sc = arg1;
9772 	char buf[16];
9773 	u_int res, re;
9774 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9775 
9776 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
9777 	switch (arg2) {
9778 	case 0:
9779 		/* timer_tick */
9780 		re = G_TIMERRESOLUTION(res);
9781 		break;
9782 	case 1:
9783 		/* TCP timestamp tick */
9784 		re = G_TIMESTAMPRESOLUTION(res);
9785 		break;
9786 	case 2:
9787 		/* DACK tick */
9788 		re = G_DELAYEDACKRESOLUTION(res);
9789 		break;
9790 	default:
9791 		return (EDOOFUS);
9792 	}
9793 
9794 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
9795 
9796 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
9797 }
9798 
9799 static int
9800 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
9801 {
9802 	struct adapter *sc = arg1;
9803 	u_int res, dack_re, v;
9804 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9805 
9806 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
9807 	dack_re = G_DELAYEDACKRESOLUTION(res);
9808 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
9809 
9810 	return (sysctl_handle_int(oidp, &v, 0, req));
9811 }
9812 
9813 static int
9814 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
9815 {
9816 	struct adapter *sc = arg1;
9817 	int reg = arg2;
9818 	u_int tre;
9819 	u_long tp_tick_us, v;
9820 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9821 
9822 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
9823 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
9824 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
9825 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
9826 
9827 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
9828 	tp_tick_us = (cclk_ps << tre) / 1000000;
9829 
9830 	if (reg == A_TP_INIT_SRTT)
9831 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
9832 	else
9833 		v = tp_tick_us * t4_read_reg(sc, reg);
9834 
9835 	return (sysctl_handle_long(oidp, &v, 0, req));
9836 }
9837 
9838 /*
9839  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
9840  * passed to this function.
9841  */
9842 static int
9843 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
9844 {
9845 	struct adapter *sc = arg1;
9846 	int idx = arg2;
9847 	u_int v;
9848 
9849 	MPASS(idx >= 0 && idx <= 24);
9850 
9851 	v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
9852 
9853 	return (sysctl_handle_int(oidp, &v, 0, req));
9854 }
9855 
9856 static int
9857 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
9858 {
9859 	struct adapter *sc = arg1;
9860 	int idx = arg2;
9861 	u_int shift, v, r;
9862 
9863 	MPASS(idx >= 0 && idx < 16);
9864 
9865 	r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
9866 	shift = (idx & 3) << 3;
9867 	v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
9868 
9869 	return (sysctl_handle_int(oidp, &v, 0, req));
9870 }
9871 
9872 static int
9873 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
9874 {
9875 	struct vi_info *vi = arg1;
9876 	struct adapter *sc = vi->pi->adapter;
9877 	int idx, rc, i;
9878 	struct sge_ofld_rxq *ofld_rxq;
9879 	uint8_t v;
9880 
9881 	idx = vi->ofld_tmr_idx;
9882 
9883 	rc = sysctl_handle_int(oidp, &idx, 0, req);
9884 	if (rc != 0 || req->newptr == NULL)
9885 		return (rc);
9886 
9887 	if (idx < 0 || idx >= SGE_NTIMERS)
9888 		return (EINVAL);
9889 
9890 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9891 	    "t4otmr");
9892 	if (rc)
9893 		return (rc);
9894 
9895 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
9896 	for_each_ofld_rxq(vi, i, ofld_rxq) {
9897 #ifdef atomic_store_rel_8
9898 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
9899 #else
9900 		ofld_rxq->iq.intr_params = v;
9901 #endif
9902 	}
9903 	vi->ofld_tmr_idx = idx;
9904 
9905 	end_synchronized_op(sc, LOCK_HELD);
9906 	return (0);
9907 }
9908 
9909 static int
9910 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
9911 {
9912 	struct vi_info *vi = arg1;
9913 	struct adapter *sc = vi->pi->adapter;
9914 	int idx, rc;
9915 
9916 	idx = vi->ofld_pktc_idx;
9917 
9918 	rc = sysctl_handle_int(oidp, &idx, 0, req);
9919 	if (rc != 0 || req->newptr == NULL)
9920 		return (rc);
9921 
9922 	if (idx < -1 || idx >= SGE_NCOUNTERS)
9923 		return (EINVAL);
9924 
9925 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9926 	    "t4opktc");
9927 	if (rc)
9928 		return (rc);
9929 
9930 	if (vi->flags & VI_INIT_DONE)
9931 		rc = EBUSY; /* cannot be changed once the queues are created */
9932 	else
9933 		vi->ofld_pktc_idx = idx;
9934 
9935 	end_synchronized_op(sc, LOCK_HELD);
9936 	return (rc);
9937 }
9938 #endif
9939 
9940 static int
9941 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
9942 {
9943 	int rc;
9944 
9945 	if (cntxt->cid > M_CTXTQID)
9946 		return (EINVAL);
9947 
9948 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
9949 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
9950 		return (EINVAL);
9951 
9952 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
9953 	if (rc)
9954 		return (rc);
9955 
9956 	if (sc->flags & FW_OK) {
9957 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
9958 		    &cntxt->data[0]);
9959 		if (rc == 0)
9960 			goto done;
9961 	}
9962 
9963 	/*
9964 	 * Read via firmware failed or wasn't even attempted.  Read directly via
9965 	 * the backdoor.
9966 	 */
9967 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
9968 done:
9969 	end_synchronized_op(sc, 0);
9970 	return (rc);
9971 }
9972 
9973 static int
9974 load_fw(struct adapter *sc, struct t4_data *fw)
9975 {
9976 	int rc;
9977 	uint8_t *fw_data;
9978 
9979 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
9980 	if (rc)
9981 		return (rc);
9982 
9983 	/*
9984 	 * The firmware, with the sole exception of the memory parity error
9985 	 * handler, runs from memory and not flash.  It is almost always safe to
9986 	 * install a new firmware on a running system.  Just set bit 1 in
9987 	 * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
9988 	 */
9989 	if (sc->flags & FULL_INIT_DONE &&
9990 	    (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
9991 		rc = EBUSY;
9992 		goto done;
9993 	}
9994 
9995 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
9996 	if (fw_data == NULL) {
9997 		rc = ENOMEM;
9998 		goto done;
9999 	}
10000 
10001 	rc = copyin(fw->data, fw_data, fw->len);
10002 	if (rc == 0)
10003 		rc = -t4_load_fw(sc, fw_data, fw->len);
10004 
10005 	free(fw_data, M_CXGBE);
10006 done:
10007 	end_synchronized_op(sc, 0);
10008 	return (rc);
10009 }
10010 
10011 static int
10012 load_cfg(struct adapter *sc, struct t4_data *cfg)
10013 {
10014 	int rc;
10015 	uint8_t *cfg_data = NULL;
10016 
10017 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
10018 	if (rc)
10019 		return (rc);
10020 
10021 	if (cfg->len == 0) {
10022 		/* clear */
10023 		rc = -t4_load_cfg(sc, NULL, 0);
10024 		goto done;
10025 	}
10026 
10027 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
10028 	if (cfg_data == NULL) {
10029 		rc = ENOMEM;
10030 		goto done;
10031 	}
10032 
10033 	rc = copyin(cfg->data, cfg_data, cfg->len);
10034 	if (rc == 0)
10035 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
10036 
10037 	free(cfg_data, M_CXGBE);
10038 done:
10039 	end_synchronized_op(sc, 0);
10040 	return (rc);
10041 }
10042 
10043 static int
10044 load_boot(struct adapter *sc, struct t4_bootrom *br)
10045 {
10046 	int rc;
10047 	uint8_t *br_data = NULL;
10048 	u_int offset;
10049 
10050 	if (br->len > 1024 * 1024)
10051 		return (EFBIG);
10052 
10053 	if (br->pf_offset == 0) {
10054 		/* pfidx */
10055 		if (br->pfidx_addr > 7)
10056 			return (EINVAL);
10057 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
10058 		    A_PCIE_PF_EXPROM_OFST)));
10059 	} else if (br->pf_offset == 1) {
10060 		/* offset */
10061 		offset = G_OFFSET(br->pfidx_addr);
10062 	} else {
10063 		return (EINVAL);
10064 	}
10065 
10066 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
10067 	if (rc)
10068 		return (rc);
10069 
10070 	if (br->len == 0) {
10071 		/* clear */
10072 		rc = -t4_load_boot(sc, NULL, offset, 0);
10073 		goto done;
10074 	}
10075 
10076 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
10077 	if (br_data == NULL) {
10078 		rc = ENOMEM;
10079 		goto done;
10080 	}
10081 
10082 	rc = copyin(br->data, br_data, br->len);
10083 	if (rc == 0)
10084 		rc = -t4_load_boot(sc, br_data, offset, br->len);
10085 
10086 	free(br_data, M_CXGBE);
10087 done:
10088 	end_synchronized_op(sc, 0);
10089 	return (rc);
10090 }
10091 
10092 static int
10093 load_bootcfg(struct adapter *sc, struct t4_data *bc)
10094 {
10095 	int rc;
10096 	uint8_t *bc_data = NULL;
10097 
10098 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
10099 	if (rc)
10100 		return (rc);
10101 
10102 	if (bc->len == 0) {
10103 		/* clear */
10104 		rc = -t4_load_bootcfg(sc, NULL, 0);
10105 		goto done;
10106 	}
10107 
10108 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
10109 	if (bc_data == NULL) {
10110 		rc = ENOMEM;
10111 		goto done;
10112 	}
10113 
10114 	rc = copyin(bc->data, bc_data, bc->len);
10115 	if (rc == 0)
10116 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
10117 
10118 	free(bc_data, M_CXGBE);
10119 done:
10120 	end_synchronized_op(sc, 0);
10121 	return (rc);
10122 }
10123 
10124 static int
10125 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
10126 {
10127 	int rc;
10128 	struct cudbg_init *cudbg;
10129 	void *handle, *buf;
10130 
10131 	/* buf is large, don't block if no memory is available */
10132 	buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
10133 	if (buf == NULL)
10134 		return (ENOMEM);
10135 
10136 	handle = cudbg_alloc_handle();
10137 	if (handle == NULL) {
10138 		rc = ENOMEM;
10139 		goto done;
10140 	}
10141 
10142 	cudbg = cudbg_get_init(handle);
10143 	cudbg->adap = sc;
10144 	cudbg->print = (cudbg_print_cb)printf;
10145 
10146 #ifndef notyet
10147 	device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
10148 	    __func__, dump->wr_flash, dump->len, dump->data);
10149 #endif
10150 
10151 	if (dump->wr_flash)
10152 		cudbg->use_flash = 1;
10153 	MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
10154 	memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
10155 
10156 	rc = cudbg_collect(handle, buf, &dump->len);
10157 	if (rc != 0)
10158 		goto done;
10159 
10160 	rc = copyout(buf, dump->data, dump->len);
10161 done:
10162 	cudbg_free_handle(handle);
10163 	free(buf, M_CXGBE);
10164 	return (rc);
10165 }
10166 
10167 static void
10168 free_offload_policy(struct t4_offload_policy *op)
10169 {
10170 	struct offload_rule *r;
10171 	int i;
10172 
10173 	if (op == NULL)
10174 		return;
10175 
10176 	r = &op->rule[0];
10177 	for (i = 0; i < op->nrules; i++, r++) {
10178 		free(r->bpf_prog.bf_insns, M_CXGBE);
10179 	}
10180 	free(op->rule, M_CXGBE);
10181 	free(op, M_CXGBE);
10182 }
10183 
10184 static int
10185 set_offload_policy(struct adapter *sc, struct t4_offload_policy *uop)
10186 {
10187 	int i, rc, len;
10188 	struct t4_offload_policy *op, *old;
10189 	struct bpf_program *bf;
10190 	const struct offload_settings *s;
10191 	struct offload_rule *r;
10192 	void *u;
10193 
10194 	if (!is_offload(sc))
10195 		return (ENODEV);
10196 
10197 	if (uop->nrules == 0) {
10198 		/* Delete installed policies. */
10199 		op = NULL;
10200 		goto set_policy;
10201 	} else if (uop->nrules > 256) { /* arbitrary */
10202 		return (E2BIG);
10203 	}
10204 
10205 	/* Copy userspace offload policy to kernel */
10206 	op = malloc(sizeof(*op), M_CXGBE, M_ZERO | M_WAITOK);
10207 	op->nrules = uop->nrules;
10208 	len = op->nrules * sizeof(struct offload_rule);
10209 	op->rule = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
10210 	rc = copyin(uop->rule, op->rule, len);
10211 	if (rc) {
10212 		free(op->rule, M_CXGBE);
10213 		free(op, M_CXGBE);
10214 		return (rc);
10215 	}
10216 
10217 	r = &op->rule[0];
10218 	for (i = 0; i < op->nrules; i++, r++) {
10219 
10220 		/* Validate open_type */
10221 		if (r->open_type != OPEN_TYPE_LISTEN &&
10222 		    r->open_type != OPEN_TYPE_ACTIVE &&
10223 		    r->open_type != OPEN_TYPE_PASSIVE &&
10224 		    r->open_type != OPEN_TYPE_DONTCARE) {
10225 error:
10226 			/*
10227 			 * Rules 0 to i have malloc'd filters that need to be
10228 			 * freed.  Rules i+1 to nrules have userspace pointers
10229 			 * and should be left alone.
10230 			 */
10231 			op->nrules = i;
10232 			free_offload_policy(op);
10233 			return (rc);
10234 		}
10235 
10236 		/* Validate settings */
10237 		s = &r->settings;
10238 		if ((s->offload != 0 && s->offload != 1) ||
10239 		    s->cong_algo < -1 || s->cong_algo > CONG_ALG_HIGHSPEED ||
10240 		    s->sched_class < -1 ||
10241 		    s->sched_class >= sc->chip_params->nsched_cls) {
10242 			rc = EINVAL;
10243 			goto error;
10244 		}
10245 
10246 		bf = &r->bpf_prog;
10247 		u = bf->bf_insns;	/* userspace ptr */
10248 		bf->bf_insns = NULL;
10249 		if (bf->bf_len == 0) {
10250 			/* legal, matches everything */
10251 			continue;
10252 		}
10253 		len = bf->bf_len * sizeof(*bf->bf_insns);
10254 		bf->bf_insns = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
10255 		rc = copyin(u, bf->bf_insns, len);
10256 		if (rc != 0)
10257 			goto error;
10258 
10259 		if (!bpf_validate(bf->bf_insns, bf->bf_len)) {
10260 			rc = EINVAL;
10261 			goto error;
10262 		}
10263 	}
10264 set_policy:
10265 	rw_wlock(&sc->policy_lock);
10266 	old = sc->policy;
10267 	sc->policy = op;
10268 	rw_wunlock(&sc->policy_lock);
10269 	free_offload_policy(old);
10270 
10271 	return (0);
10272 }
10273 
10274 #define MAX_READ_BUF_SIZE (128 * 1024)
10275 static int
10276 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
10277 {
10278 	uint32_t addr, remaining, n;
10279 	uint32_t *buf;
10280 	int rc;
10281 	uint8_t *dst;
10282 
10283 	rc = validate_mem_range(sc, mr->addr, mr->len);
10284 	if (rc != 0)
10285 		return (rc);
10286 
10287 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
10288 	addr = mr->addr;
10289 	remaining = mr->len;
10290 	dst = (void *)mr->data;
10291 
10292 	while (remaining) {
10293 		n = min(remaining, MAX_READ_BUF_SIZE);
10294 		read_via_memwin(sc, 2, addr, buf, n);
10295 
10296 		rc = copyout(buf, dst, n);
10297 		if (rc != 0)
10298 			break;
10299 
10300 		dst += n;
10301 		remaining -= n;
10302 		addr += n;
10303 	}
10304 
10305 	free(buf, M_CXGBE);
10306 	return (rc);
10307 }
10308 #undef MAX_READ_BUF_SIZE
10309 
10310 static int
10311 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
10312 {
10313 	int rc;
10314 
10315 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
10316 		return (EINVAL);
10317 
10318 	if (i2cd->len > sizeof(i2cd->data))
10319 		return (EFBIG);
10320 
10321 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
10322 	if (rc)
10323 		return (rc);
10324 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
10325 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
10326 	end_synchronized_op(sc, 0);
10327 
10328 	return (rc);
10329 }
10330 
10331 static int
10332 clear_stats(struct adapter *sc, u_int port_id)
10333 {
10334 	int i, v, chan_map;
10335 	struct port_info *pi;
10336 	struct vi_info *vi;
10337 	struct sge_rxq *rxq;
10338 	struct sge_txq *txq;
10339 	struct sge_wrq *wrq;
10340 #ifdef TCP_OFFLOAD
10341 	struct sge_ofld_rxq *ofld_rxq;
10342 #endif
10343 
10344 	if (port_id >= sc->params.nports)
10345 		return (EINVAL);
10346 	pi = sc->port[port_id];
10347 	if (pi == NULL)
10348 		return (EIO);
10349 
10350 	/* MAC stats */
10351 	t4_clr_port_stats(sc, pi->tx_chan);
10352 	pi->tx_parse_error = 0;
10353 	pi->tnl_cong_drops = 0;
10354 	mtx_lock(&sc->reg_lock);
10355 	for_each_vi(pi, v, vi) {
10356 		if (vi->flags & VI_INIT_DONE)
10357 			t4_clr_vi_stats(sc, vi->vin);
10358 	}
10359 	chan_map = pi->rx_e_chan_map;
10360 	v = 0;	/* reuse */
10361 	while (chan_map) {
10362 		i = ffs(chan_map) - 1;
10363 		t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
10364 		    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
10365 		chan_map &= ~(1 << i);
10366 	}
10367 	mtx_unlock(&sc->reg_lock);
10368 
10369 	/*
10370 	 * Since this command accepts a port, clear stats for
10371 	 * all VIs on this port.
10372 	 */
10373 	for_each_vi(pi, v, vi) {
10374 		if (vi->flags & VI_INIT_DONE) {
10375 
10376 			for_each_rxq(vi, i, rxq) {
10377 #if defined(INET) || defined(INET6)
10378 				rxq->lro.lro_queued = 0;
10379 				rxq->lro.lro_flushed = 0;
10380 #endif
10381 				rxq->rxcsum = 0;
10382 				rxq->vlan_extraction = 0;
10383 
10384 				rxq->fl.cl_allocated = 0;
10385 				rxq->fl.cl_recycled = 0;
10386 				rxq->fl.cl_fast_recycled = 0;
10387 			}
10388 
10389 			for_each_txq(vi, i, txq) {
10390 				txq->txcsum = 0;
10391 				txq->tso_wrs = 0;
10392 				txq->vlan_insertion = 0;
10393 				txq->imm_wrs = 0;
10394 				txq->sgl_wrs = 0;
10395 				txq->txpkt_wrs = 0;
10396 				txq->txpkts0_wrs = 0;
10397 				txq->txpkts1_wrs = 0;
10398 				txq->txpkts0_pkts = 0;
10399 				txq->txpkts1_pkts = 0;
10400 				txq->raw_wrs = 0;
10401 				txq->kern_tls_records = 0;
10402 				txq->kern_tls_short = 0;
10403 				txq->kern_tls_partial = 0;
10404 				txq->kern_tls_full = 0;
10405 				txq->kern_tls_octets = 0;
10406 				txq->kern_tls_waste = 0;
10407 				txq->kern_tls_options = 0;
10408 				txq->kern_tls_header = 0;
10409 				txq->kern_tls_fin = 0;
10410 				txq->kern_tls_fin_short = 0;
10411 				txq->kern_tls_cbc = 0;
10412 				txq->kern_tls_gcm = 0;
10413 				mp_ring_reset_stats(txq->r);
10414 			}
10415 
10416 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
10417 			for_each_ofld_txq(vi, i, wrq) {
10418 				wrq->tx_wrs_direct = 0;
10419 				wrq->tx_wrs_copied = 0;
10420 			}
10421 #endif
10422 #ifdef TCP_OFFLOAD
10423 			for_each_ofld_rxq(vi, i, ofld_rxq) {
10424 				ofld_rxq->fl.cl_allocated = 0;
10425 				ofld_rxq->fl.cl_recycled = 0;
10426 				ofld_rxq->fl.cl_fast_recycled = 0;
10427 			}
10428 #endif
10429 
10430 			if (IS_MAIN_VI(vi)) {
10431 				wrq = &sc->sge.ctrlq[pi->port_id];
10432 				wrq->tx_wrs_direct = 0;
10433 				wrq->tx_wrs_copied = 0;
10434 			}
10435 		}
10436 	}
10437 
10438 	return (0);
10439 }
10440 
10441 int
10442 t4_os_find_pci_capability(struct adapter *sc, int cap)
10443 {
10444 	int i;
10445 
10446 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
10447 }
10448 
10449 int
10450 t4_os_pci_save_state(struct adapter *sc)
10451 {
10452 	device_t dev;
10453 	struct pci_devinfo *dinfo;
10454 
10455 	dev = sc->dev;
10456 	dinfo = device_get_ivars(dev);
10457 
10458 	pci_cfg_save(dev, dinfo, 0);
10459 	return (0);
10460 }
10461 
10462 int
10463 t4_os_pci_restore_state(struct adapter *sc)
10464 {
10465 	device_t dev;
10466 	struct pci_devinfo *dinfo;
10467 
10468 	dev = sc->dev;
10469 	dinfo = device_get_ivars(dev);
10470 
10471 	pci_cfg_restore(dev, dinfo);
10472 	return (0);
10473 }
10474 
10475 void
10476 t4_os_portmod_changed(struct port_info *pi)
10477 {
10478 	struct adapter *sc = pi->adapter;
10479 	struct vi_info *vi;
10480 	struct ifnet *ifp;
10481 	static const char *mod_str[] = {
10482 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
10483 	};
10484 
10485 	KASSERT((pi->flags & FIXED_IFMEDIA) == 0,
10486 	    ("%s: port_type %u", __func__, pi->port_type));
10487 
10488 	vi = &pi->vi[0];
10489 	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
10490 		PORT_LOCK(pi);
10491 		build_medialist(pi);
10492 		if (pi->mod_type != FW_PORT_MOD_TYPE_NONE) {
10493 			fixup_link_config(pi);
10494 			apply_link_config(pi);
10495 		}
10496 		PORT_UNLOCK(pi);
10497 		end_synchronized_op(sc, LOCK_HELD);
10498 	}
10499 
10500 	ifp = vi->ifp;
10501 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
10502 		if_printf(ifp, "transceiver unplugged.\n");
10503 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
10504 		if_printf(ifp, "unknown transceiver inserted.\n");
10505 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
10506 		if_printf(ifp, "unsupported transceiver inserted.\n");
10507 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
10508 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
10509 		    port_top_speed(pi), mod_str[pi->mod_type]);
10510 	} else {
10511 		if_printf(ifp, "transceiver (type %d) inserted.\n",
10512 		    pi->mod_type);
10513 	}
10514 }
10515 
10516 void
10517 t4_os_link_changed(struct port_info *pi)
10518 {
10519 	struct vi_info *vi;
10520 	struct ifnet *ifp;
10521 	struct link_config *lc;
10522 	int v;
10523 
10524 	PORT_LOCK_ASSERT_OWNED(pi);
10525 
10526 	for_each_vi(pi, v, vi) {
10527 		ifp = vi->ifp;
10528 		if (ifp == NULL)
10529 			continue;
10530 
10531 		lc = &pi->link_cfg;
10532 		if (lc->link_ok) {
10533 			ifp->if_baudrate = IF_Mbps(lc->speed);
10534 			if_link_state_change(ifp, LINK_STATE_UP);
10535 		} else {
10536 			if_link_state_change(ifp, LINK_STATE_DOWN);
10537 		}
10538 	}
10539 }
10540 
10541 void
10542 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
10543 {
10544 	struct adapter *sc;
10545 
10546 	sx_slock(&t4_list_lock);
10547 	SLIST_FOREACH(sc, &t4_list, link) {
10548 		/*
10549 		 * func should not make any assumptions about what state sc is
10550 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
10551 		 */
10552 		func(sc, arg);
10553 	}
10554 	sx_sunlock(&t4_list_lock);
10555 }
10556 
10557 static int
10558 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
10559     struct thread *td)
10560 {
10561 	int rc;
10562 	struct adapter *sc = dev->si_drv1;
10563 
10564 	rc = priv_check(td, PRIV_DRIVER);
10565 	if (rc != 0)
10566 		return (rc);
10567 
10568 	switch (cmd) {
10569 	case CHELSIO_T4_GETREG: {
10570 		struct t4_reg *edata = (struct t4_reg *)data;
10571 
10572 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
10573 			return (EFAULT);
10574 
10575 		if (edata->size == 4)
10576 			edata->val = t4_read_reg(sc, edata->addr);
10577 		else if (edata->size == 8)
10578 			edata->val = t4_read_reg64(sc, edata->addr);
10579 		else
10580 			return (EINVAL);
10581 
10582 		break;
10583 	}
10584 	case CHELSIO_T4_SETREG: {
10585 		struct t4_reg *edata = (struct t4_reg *)data;
10586 
10587 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
10588 			return (EFAULT);
10589 
10590 		if (edata->size == 4) {
10591 			if (edata->val & 0xffffffff00000000)
10592 				return (EINVAL);
10593 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
10594 		} else if (edata->size == 8)
10595 			t4_write_reg64(sc, edata->addr, edata->val);
10596 		else
10597 			return (EINVAL);
10598 		break;
10599 	}
10600 	case CHELSIO_T4_REGDUMP: {
10601 		struct t4_regdump *regs = (struct t4_regdump *)data;
10602 		int reglen = t4_get_regs_len(sc);
10603 		uint8_t *buf;
10604 
10605 		if (regs->len < reglen) {
10606 			regs->len = reglen; /* hint to the caller */
10607 			return (ENOBUFS);
10608 		}
10609 
10610 		regs->len = reglen;
10611 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
10612 		get_regs(sc, regs, buf);
10613 		rc = copyout(buf, regs->data, reglen);
10614 		free(buf, M_CXGBE);
10615 		break;
10616 	}
10617 	case CHELSIO_T4_GET_FILTER_MODE:
10618 		rc = get_filter_mode(sc, (uint32_t *)data);
10619 		break;
10620 	case CHELSIO_T4_SET_FILTER_MODE:
10621 		rc = set_filter_mode(sc, *(uint32_t *)data);
10622 		break;
10623 	case CHELSIO_T4_GET_FILTER:
10624 		rc = get_filter(sc, (struct t4_filter *)data);
10625 		break;
10626 	case CHELSIO_T4_SET_FILTER:
10627 		rc = set_filter(sc, (struct t4_filter *)data);
10628 		break;
10629 	case CHELSIO_T4_DEL_FILTER:
10630 		rc = del_filter(sc, (struct t4_filter *)data);
10631 		break;
10632 	case CHELSIO_T4_GET_SGE_CONTEXT:
10633 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
10634 		break;
10635 	case CHELSIO_T4_LOAD_FW:
10636 		rc = load_fw(sc, (struct t4_data *)data);
10637 		break;
10638 	case CHELSIO_T4_GET_MEM:
10639 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
10640 		break;
10641 	case CHELSIO_T4_GET_I2C:
10642 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
10643 		break;
10644 	case CHELSIO_T4_CLEAR_STATS:
10645 		rc = clear_stats(sc, *(uint32_t *)data);
10646 		break;
10647 	case CHELSIO_T4_SCHED_CLASS:
10648 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
10649 		break;
10650 	case CHELSIO_T4_SCHED_QUEUE:
10651 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
10652 		break;
10653 	case CHELSIO_T4_GET_TRACER:
10654 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
10655 		break;
10656 	case CHELSIO_T4_SET_TRACER:
10657 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
10658 		break;
10659 	case CHELSIO_T4_LOAD_CFG:
10660 		rc = load_cfg(sc, (struct t4_data *)data);
10661 		break;
10662 	case CHELSIO_T4_LOAD_BOOT:
10663 		rc = load_boot(sc, (struct t4_bootrom *)data);
10664 		break;
10665 	case CHELSIO_T4_LOAD_BOOTCFG:
10666 		rc = load_bootcfg(sc, (struct t4_data *)data);
10667 		break;
10668 	case CHELSIO_T4_CUDBG_DUMP:
10669 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
10670 		break;
10671 	case CHELSIO_T4_SET_OFLD_POLICY:
10672 		rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
10673 		break;
10674 	default:
10675 		rc = ENOTTY;
10676 	}
10677 
10678 	return (rc);
10679 }
10680 
10681 #ifdef TCP_OFFLOAD
10682 static int
10683 toe_capability(struct vi_info *vi, int enable)
10684 {
10685 	int rc;
10686 	struct port_info *pi = vi->pi;
10687 	struct adapter *sc = pi->adapter;
10688 
10689 	ASSERT_SYNCHRONIZED_OP(sc);
10690 
10691 	if (!is_offload(sc))
10692 		return (ENODEV);
10693 
10694 	if (enable) {
10695 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
10696 			/* TOE is already enabled. */
10697 			return (0);
10698 		}
10699 
10700 		/*
10701 		 * We need the port's queues around so that we're able to send
10702 		 * and receive CPLs to/from the TOE even if the ifnet for this
10703 		 * port has never been UP'd administratively.
10704 		 */
10705 		if (!(vi->flags & VI_INIT_DONE)) {
10706 			rc = vi_full_init(vi);
10707 			if (rc)
10708 				return (rc);
10709 		}
10710 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
10711 			rc = vi_full_init(&pi->vi[0]);
10712 			if (rc)
10713 				return (rc);
10714 		}
10715 
10716 		if (isset(&sc->offload_map, pi->port_id)) {
10717 			/* TOE is enabled on another VI of this port. */
10718 			pi->uld_vis++;
10719 			return (0);
10720 		}
10721 
10722 		if (!uld_active(sc, ULD_TOM)) {
10723 			rc = t4_activate_uld(sc, ULD_TOM);
10724 			if (rc == EAGAIN) {
10725 				log(LOG_WARNING,
10726 				    "You must kldload t4_tom.ko before trying "
10727 				    "to enable TOE on a cxgbe interface.\n");
10728 			}
10729 			if (rc != 0)
10730 				return (rc);
10731 			KASSERT(sc->tom_softc != NULL,
10732 			    ("%s: TOM activated but softc NULL", __func__));
10733 			KASSERT(uld_active(sc, ULD_TOM),
10734 			    ("%s: TOM activated but flag not set", __func__));
10735 		}
10736 
10737 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
10738 		if (!uld_active(sc, ULD_IWARP))
10739 			(void) t4_activate_uld(sc, ULD_IWARP);
10740 		if (!uld_active(sc, ULD_ISCSI))
10741 			(void) t4_activate_uld(sc, ULD_ISCSI);
10742 
10743 		pi->uld_vis++;
10744 		setbit(&sc->offload_map, pi->port_id);
10745 	} else {
10746 		pi->uld_vis--;
10747 
10748 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
10749 			return (0);
10750 
10751 		KASSERT(uld_active(sc, ULD_TOM),
10752 		    ("%s: TOM never initialized?", __func__));
10753 		clrbit(&sc->offload_map, pi->port_id);
10754 	}
10755 
10756 	return (0);
10757 }
10758 
10759 /*
10760  * Add an upper layer driver to the global list.
10761  */
10762 int
10763 t4_register_uld(struct uld_info *ui)
10764 {
10765 	int rc = 0;
10766 	struct uld_info *u;
10767 
10768 	sx_xlock(&t4_uld_list_lock);
10769 	SLIST_FOREACH(u, &t4_uld_list, link) {
10770 	    if (u->uld_id == ui->uld_id) {
10771 		    rc = EEXIST;
10772 		    goto done;
10773 	    }
10774 	}
10775 
10776 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
10777 	ui->refcount = 0;
10778 done:
10779 	sx_xunlock(&t4_uld_list_lock);
10780 	return (rc);
10781 }
10782 
10783 int
10784 t4_unregister_uld(struct uld_info *ui)
10785 {
10786 	int rc = EINVAL;
10787 	struct uld_info *u;
10788 
10789 	sx_xlock(&t4_uld_list_lock);
10790 
10791 	SLIST_FOREACH(u, &t4_uld_list, link) {
10792 	    if (u == ui) {
10793 		    if (ui->refcount > 0) {
10794 			    rc = EBUSY;
10795 			    goto done;
10796 		    }
10797 
10798 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
10799 		    rc = 0;
10800 		    goto done;
10801 	    }
10802 	}
10803 done:
10804 	sx_xunlock(&t4_uld_list_lock);
10805 	return (rc);
10806 }
10807 
10808 int
10809 t4_activate_uld(struct adapter *sc, int id)
10810 {
10811 	int rc;
10812 	struct uld_info *ui;
10813 
10814 	ASSERT_SYNCHRONIZED_OP(sc);
10815 
10816 	if (id < 0 || id > ULD_MAX)
10817 		return (EINVAL);
10818 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
10819 
10820 	sx_slock(&t4_uld_list_lock);
10821 
10822 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10823 		if (ui->uld_id == id) {
10824 			if (!(sc->flags & FULL_INIT_DONE)) {
10825 				rc = adapter_full_init(sc);
10826 				if (rc != 0)
10827 					break;
10828 			}
10829 
10830 			rc = ui->activate(sc);
10831 			if (rc == 0) {
10832 				setbit(&sc->active_ulds, id);
10833 				ui->refcount++;
10834 			}
10835 			break;
10836 		}
10837 	}
10838 
10839 	sx_sunlock(&t4_uld_list_lock);
10840 
10841 	return (rc);
10842 }
10843 
10844 int
10845 t4_deactivate_uld(struct adapter *sc, int id)
10846 {
10847 	int rc;
10848 	struct uld_info *ui;
10849 
10850 	ASSERT_SYNCHRONIZED_OP(sc);
10851 
10852 	if (id < 0 || id > ULD_MAX)
10853 		return (EINVAL);
10854 	rc = ENXIO;
10855 
10856 	sx_slock(&t4_uld_list_lock);
10857 
10858 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10859 		if (ui->uld_id == id) {
10860 			rc = ui->deactivate(sc);
10861 			if (rc == 0) {
10862 				clrbit(&sc->active_ulds, id);
10863 				ui->refcount--;
10864 			}
10865 			break;
10866 		}
10867 	}
10868 
10869 	sx_sunlock(&t4_uld_list_lock);
10870 
10871 	return (rc);
10872 }
10873 
10874 static void
10875 t4_async_event(void *arg, int n)
10876 {
10877 	struct uld_info *ui;
10878 	struct adapter *sc = (struct adapter *)arg;
10879 
10880 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4async") != 0)
10881 		return;
10882 	sx_slock(&t4_uld_list_lock);
10883 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10884 		if (ui->uld_id == ULD_IWARP) {
10885 			ui->async_event(sc);
10886 			break;
10887 		}
10888 	}
10889 	sx_sunlock(&t4_uld_list_lock);
10890 	end_synchronized_op(sc, 0);
10891 }
10892 
10893 int
10894 uld_active(struct adapter *sc, int uld_id)
10895 {
10896 
10897 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
10898 
10899 	return (isset(&sc->active_ulds, uld_id));
10900 }
10901 #endif
10902 
10903 /*
10904  * t  = ptr to tunable.
10905  * nc = number of CPUs.
10906  * c  = compiled in default for that tunable.
10907  */
10908 static void
10909 calculate_nqueues(int *t, int nc, const int c)
10910 {
10911 	int nq;
10912 
10913 	if (*t > 0)
10914 		return;
10915 	nq = *t < 0 ? -*t : c;
10916 	*t = min(nc, nq);
10917 }
10918 
10919 /*
10920  * Come up with reasonable defaults for some of the tunables, provided they're
10921  * not set by the user (in which case we'll use the values as is).
10922  */
10923 static void
10924 tweak_tunables(void)
10925 {
10926 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
10927 
10928 	if (t4_ntxq < 1) {
10929 #ifdef RSS
10930 		t4_ntxq = rss_getnumbuckets();
10931 #else
10932 		calculate_nqueues(&t4_ntxq, nc, NTXQ);
10933 #endif
10934 	}
10935 
10936 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
10937 
10938 	if (t4_nrxq < 1) {
10939 #ifdef RSS
10940 		t4_nrxq = rss_getnumbuckets();
10941 #else
10942 		calculate_nqueues(&t4_nrxq, nc, NRXQ);
10943 #endif
10944 	}
10945 
10946 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
10947 
10948 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
10949 	calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
10950 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
10951 #endif
10952 #ifdef TCP_OFFLOAD
10953 	calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
10954 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
10955 #endif
10956 
10957 #if defined(TCP_OFFLOAD) || defined(KERN_TLS)
10958 	if (t4_toecaps_allowed == -1)
10959 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
10960 #else
10961 	if (t4_toecaps_allowed == -1)
10962 		t4_toecaps_allowed = 0;
10963 #endif
10964 
10965 #ifdef TCP_OFFLOAD
10966 	if (t4_rdmacaps_allowed == -1) {
10967 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
10968 		    FW_CAPS_CONFIG_RDMA_RDMAC;
10969 	}
10970 
10971 	if (t4_iscsicaps_allowed == -1) {
10972 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
10973 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
10974 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
10975 	}
10976 
10977 	if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
10978 		t4_tmr_idx_ofld = TMR_IDX_OFLD;
10979 
10980 	if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
10981 		t4_pktc_idx_ofld = PKTC_IDX_OFLD;
10982 #else
10983 	if (t4_rdmacaps_allowed == -1)
10984 		t4_rdmacaps_allowed = 0;
10985 
10986 	if (t4_iscsicaps_allowed == -1)
10987 		t4_iscsicaps_allowed = 0;
10988 #endif
10989 
10990 #ifdef DEV_NETMAP
10991 	calculate_nqueues(&t4_nnmtxq, nc, NNMTXQ);
10992 	calculate_nqueues(&t4_nnmrxq, nc, NNMRXQ);
10993 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
10994 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
10995 #endif
10996 
10997 	if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
10998 		t4_tmr_idx = TMR_IDX;
10999 
11000 	if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
11001 		t4_pktc_idx = PKTC_IDX;
11002 
11003 	if (t4_qsize_txq < 128)
11004 		t4_qsize_txq = 128;
11005 
11006 	if (t4_qsize_rxq < 128)
11007 		t4_qsize_rxq = 128;
11008 	while (t4_qsize_rxq & 7)
11009 		t4_qsize_rxq++;
11010 
11011 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
11012 
11013 	/*
11014 	 * Number of VIs to create per-port.  The first VI is the "main" regular
11015 	 * VI for the port.  The rest are additional virtual interfaces on the
11016 	 * same physical port.  Note that the main VI does not have native
11017 	 * netmap support but the extra VIs do.
11018 	 *
11019 	 * Limit the number of VIs per port to the number of available
11020 	 * MAC addresses per port.
11021 	 */
11022 	if (t4_num_vis < 1)
11023 		t4_num_vis = 1;
11024 	if (t4_num_vis > nitems(vi_mac_funcs)) {
11025 		t4_num_vis = nitems(vi_mac_funcs);
11026 		printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
11027 	}
11028 
11029 	if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
11030 		pcie_relaxed_ordering = 1;
11031 #if defined(__i386__) || defined(__amd64__)
11032 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
11033 			pcie_relaxed_ordering = 0;
11034 #endif
11035 	}
11036 }
11037 
11038 #ifdef DDB
11039 static void
11040 t4_dump_tcb(struct adapter *sc, int tid)
11041 {
11042 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
11043 
11044 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
11045 	save = t4_read_reg(sc, reg);
11046 	base = sc->memwin[2].mw_base;
11047 
11048 	/* Dump TCB for the tid */
11049 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
11050 	tcb_addr += tid * TCB_SIZE;
11051 
11052 	if (is_t4(sc)) {
11053 		pf = 0;
11054 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
11055 	} else {
11056 		pf = V_PFNUM(sc->pf);
11057 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
11058 	}
11059 	t4_write_reg(sc, reg, win_pos | pf);
11060 	t4_read_reg(sc, reg);
11061 
11062 	off = tcb_addr - win_pos;
11063 	for (i = 0; i < 4; i++) {
11064 		uint32_t buf[8];
11065 		for (j = 0; j < 8; j++, off += 4)
11066 			buf[j] = htonl(t4_read_reg(sc, base + off));
11067 
11068 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
11069 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
11070 		    buf[7]);
11071 	}
11072 
11073 	t4_write_reg(sc, reg, save);
11074 	t4_read_reg(sc, reg);
11075 }
11076 
11077 static void
11078 t4_dump_devlog(struct adapter *sc)
11079 {
11080 	struct devlog_params *dparams = &sc->params.devlog;
11081 	struct fw_devlog_e e;
11082 	int i, first, j, m, nentries, rc;
11083 	uint64_t ftstamp = UINT64_MAX;
11084 
11085 	if (dparams->start == 0) {
11086 		db_printf("devlog params not valid\n");
11087 		return;
11088 	}
11089 
11090 	nentries = dparams->size / sizeof(struct fw_devlog_e);
11091 	m = fwmtype_to_hwmtype(dparams->memtype);
11092 
11093 	/* Find the first entry. */
11094 	first = -1;
11095 	for (i = 0; i < nentries && !db_pager_quit; i++) {
11096 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
11097 		    sizeof(e), (void *)&e);
11098 		if (rc != 0)
11099 			break;
11100 
11101 		if (e.timestamp == 0)
11102 			break;
11103 
11104 		e.timestamp = be64toh(e.timestamp);
11105 		if (e.timestamp < ftstamp) {
11106 			ftstamp = e.timestamp;
11107 			first = i;
11108 		}
11109 	}
11110 
11111 	if (first == -1)
11112 		return;
11113 
11114 	i = first;
11115 	do {
11116 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
11117 		    sizeof(e), (void *)&e);
11118 		if (rc != 0)
11119 			return;
11120 
11121 		if (e.timestamp == 0)
11122 			return;
11123 
11124 		e.timestamp = be64toh(e.timestamp);
11125 		e.seqno = be32toh(e.seqno);
11126 		for (j = 0; j < 8; j++)
11127 			e.params[j] = be32toh(e.params[j]);
11128 
11129 		db_printf("%10d  %15ju  %8s  %8s  ",
11130 		    e.seqno, e.timestamp,
11131 		    (e.level < nitems(devlog_level_strings) ?
11132 			devlog_level_strings[e.level] : "UNKNOWN"),
11133 		    (e.facility < nitems(devlog_facility_strings) ?
11134 			devlog_facility_strings[e.facility] : "UNKNOWN"));
11135 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
11136 		    e.params[3], e.params[4], e.params[5], e.params[6],
11137 		    e.params[7]);
11138 
11139 		if (++i == nentries)
11140 			i = 0;
11141 	} while (i != first && !db_pager_quit);
11142 }
11143 
11144 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
11145 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
11146 
11147 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
11148 {
11149 	device_t dev;
11150 	int t;
11151 	bool valid;
11152 
11153 	valid = false;
11154 	t = db_read_token();
11155 	if (t == tIDENT) {
11156 		dev = device_lookup_by_name(db_tok_string);
11157 		valid = true;
11158 	}
11159 	db_skip_to_eol();
11160 	if (!valid) {
11161 		db_printf("usage: show t4 devlog <nexus>\n");
11162 		return;
11163 	}
11164 
11165 	if (dev == NULL) {
11166 		db_printf("device not found\n");
11167 		return;
11168 	}
11169 
11170 	t4_dump_devlog(device_get_softc(dev));
11171 }
11172 
11173 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
11174 {
11175 	device_t dev;
11176 	int radix, tid, t;
11177 	bool valid;
11178 
11179 	valid = false;
11180 	radix = db_radix;
11181 	db_radix = 10;
11182 	t = db_read_token();
11183 	if (t == tIDENT) {
11184 		dev = device_lookup_by_name(db_tok_string);
11185 		t = db_read_token();
11186 		if (t == tNUMBER) {
11187 			tid = db_tok_number;
11188 			valid = true;
11189 		}
11190 	}
11191 	db_radix = radix;
11192 	db_skip_to_eol();
11193 	if (!valid) {
11194 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
11195 		return;
11196 	}
11197 
11198 	if (dev == NULL) {
11199 		db_printf("device not found\n");
11200 		return;
11201 	}
11202 	if (tid < 0) {
11203 		db_printf("invalid tid\n");
11204 		return;
11205 	}
11206 
11207 	t4_dump_tcb(device_get_softc(dev), tid);
11208 }
11209 #endif
11210 
11211 static struct sx mlu;	/* mod load unload */
11212 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
11213 
11214 static int
11215 mod_event(module_t mod, int cmd, void *arg)
11216 {
11217 	int rc = 0;
11218 	static int loaded = 0;
11219 
11220 	switch (cmd) {
11221 	case MOD_LOAD:
11222 		sx_xlock(&mlu);
11223 		if (loaded++ == 0) {
11224 			t4_sge_modload();
11225 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
11226 			    t4_filter_rpl, CPL_COOKIE_FILTER);
11227 			t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
11228 			    do_l2t_write_rpl, CPL_COOKIE_FILTER);
11229 			t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL,
11230 			    t4_hashfilter_ao_rpl, CPL_COOKIE_HASHFILTER);
11231 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
11232 			    t4_hashfilter_tcb_rpl, CPL_COOKIE_HASHFILTER);
11233 			t4_register_shared_cpl_handler(CPL_ABORT_RPL_RSS,
11234 			    t4_del_hashfilter_rpl, CPL_COOKIE_HASHFILTER);
11235 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
11236 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
11237 			t4_register_cpl_handler(CPL_SMT_WRITE_RPL,
11238 			    do_smt_write_rpl);
11239 			sx_init(&t4_list_lock, "T4/T5 adapters");
11240 			SLIST_INIT(&t4_list);
11241 			callout_init(&fatal_callout, 1);
11242 #ifdef TCP_OFFLOAD
11243 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
11244 			SLIST_INIT(&t4_uld_list);
11245 #endif
11246 #ifdef INET6
11247 			t4_clip_modload();
11248 #endif
11249 #ifdef KERN_TLS
11250 			t6_ktls_modload();
11251 #endif
11252 			t4_tracer_modload();
11253 			tweak_tunables();
11254 		}
11255 		sx_xunlock(&mlu);
11256 		break;
11257 
11258 	case MOD_UNLOAD:
11259 		sx_xlock(&mlu);
11260 		if (--loaded == 0) {
11261 			int tries;
11262 
11263 			sx_slock(&t4_list_lock);
11264 			if (!SLIST_EMPTY(&t4_list)) {
11265 				rc = EBUSY;
11266 				sx_sunlock(&t4_list_lock);
11267 				goto done_unload;
11268 			}
11269 #ifdef TCP_OFFLOAD
11270 			sx_slock(&t4_uld_list_lock);
11271 			if (!SLIST_EMPTY(&t4_uld_list)) {
11272 				rc = EBUSY;
11273 				sx_sunlock(&t4_uld_list_lock);
11274 				sx_sunlock(&t4_list_lock);
11275 				goto done_unload;
11276 			}
11277 #endif
11278 			tries = 0;
11279 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
11280 				uprintf("%ju clusters with custom free routine "
11281 				    "still is use.\n", t4_sge_extfree_refs());
11282 				pause("t4unload", 2 * hz);
11283 			}
11284 #ifdef TCP_OFFLOAD
11285 			sx_sunlock(&t4_uld_list_lock);
11286 #endif
11287 			sx_sunlock(&t4_list_lock);
11288 
11289 			if (t4_sge_extfree_refs() == 0) {
11290 				t4_tracer_modunload();
11291 #ifdef KERN_TLS
11292 				t6_ktls_modunload();
11293 #endif
11294 #ifdef INET6
11295 				t4_clip_modunload();
11296 #endif
11297 #ifdef TCP_OFFLOAD
11298 				sx_destroy(&t4_uld_list_lock);
11299 #endif
11300 				sx_destroy(&t4_list_lock);
11301 				t4_sge_modunload();
11302 				loaded = 0;
11303 			} else {
11304 				rc = EBUSY;
11305 				loaded++;	/* undo earlier decrement */
11306 			}
11307 		}
11308 done_unload:
11309 		sx_xunlock(&mlu);
11310 		break;
11311 	}
11312 
11313 	return (rc);
11314 }
11315 
11316 static devclass_t t4_devclass, t5_devclass, t6_devclass;
11317 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
11318 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
11319 
11320 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
11321 MODULE_VERSION(t4nex, 1);
11322 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
11323 #ifdef DEV_NETMAP
11324 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
11325 #endif /* DEV_NETMAP */
11326 
11327 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
11328 MODULE_VERSION(t5nex, 1);
11329 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
11330 #ifdef DEV_NETMAP
11331 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
11332 #endif /* DEV_NETMAP */
11333 
11334 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
11335 MODULE_VERSION(t6nex, 1);
11336 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
11337 #ifdef DEV_NETMAP
11338 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
11339 #endif /* DEV_NETMAP */
11340 
11341 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
11342 MODULE_VERSION(cxgbe, 1);
11343 
11344 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
11345 MODULE_VERSION(cxl, 1);
11346 
11347 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
11348 MODULE_VERSION(cc, 1);
11349 
11350 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
11351 MODULE_VERSION(vcxgbe, 1);
11352 
11353 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
11354 MODULE_VERSION(vcxl, 1);
11355 
11356 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
11357 MODULE_VERSION(vcc, 1);
11358