xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision 51f61fc0c7ece7a30c737341e65455841bc3f04e)
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_ratelimit.h"
37 #include "opt_rss.h"
38 
39 #include <sys/param.h>
40 #include <sys/conf.h>
41 #include <sys/priv.h>
42 #include <sys/kernel.h>
43 #include <sys/bus.h>
44 #include <sys/module.h>
45 #include <sys/malloc.h>
46 #include <sys/queue.h>
47 #include <sys/taskqueue.h>
48 #include <sys/pciio.h>
49 #include <dev/pci/pcireg.h>
50 #include <dev/pci/pcivar.h>
51 #include <dev/pci/pci_private.h>
52 #include <sys/firmware.h>
53 #include <sys/sbuf.h>
54 #include <sys/smp.h>
55 #include <sys/socket.h>
56 #include <sys/sockio.h>
57 #include <sys/sysctl.h>
58 #include <net/ethernet.h>
59 #include <net/if.h>
60 #include <net/if_types.h>
61 #include <net/if_dl.h>
62 #include <net/if_vlan_var.h>
63 #ifdef RSS
64 #include <net/rss_config.h>
65 #endif
66 #include <netinet/in.h>
67 #include <netinet/ip.h>
68 #if defined(__i386__) || defined(__amd64__)
69 #include <machine/md_var.h>
70 #include <machine/cputypes.h>
71 #include <vm/vm.h>
72 #include <vm/pmap.h>
73 #endif
74 #include <crypto/rijndael/rijndael.h>
75 #ifdef DDB
76 #include <ddb/ddb.h>
77 #include <ddb/db_lex.h>
78 #endif
79 
80 #include "common/common.h"
81 #include "common/t4_msg.h"
82 #include "common/t4_regs.h"
83 #include "common/t4_regs_values.h"
84 #include "cudbg/cudbg.h"
85 #include "t4_clip.h"
86 #include "t4_ioctl.h"
87 #include "t4_l2t.h"
88 #include "t4_mp_ring.h"
89 #include "t4_if.h"
90 #include "t4_smt.h"
91 
92 /* T4 bus driver interface */
93 static int t4_probe(device_t);
94 static int t4_attach(device_t);
95 static int t4_detach(device_t);
96 static int t4_child_location_str(device_t, device_t, char *, size_t);
97 static int t4_ready(device_t);
98 static int t4_read_port_device(device_t, int, device_t *);
99 static device_method_t t4_methods[] = {
100 	DEVMETHOD(device_probe,		t4_probe),
101 	DEVMETHOD(device_attach,	t4_attach),
102 	DEVMETHOD(device_detach,	t4_detach),
103 
104 	DEVMETHOD(bus_child_location_str, t4_child_location_str),
105 
106 	DEVMETHOD(t4_is_main_ready,	t4_ready),
107 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
108 
109 	DEVMETHOD_END
110 };
111 static driver_t t4_driver = {
112 	"t4nex",
113 	t4_methods,
114 	sizeof(struct adapter)
115 };
116 
117 
118 /* T4 port (cxgbe) interface */
119 static int cxgbe_probe(device_t);
120 static int cxgbe_attach(device_t);
121 static int cxgbe_detach(device_t);
122 device_method_t cxgbe_methods[] = {
123 	DEVMETHOD(device_probe,		cxgbe_probe),
124 	DEVMETHOD(device_attach,	cxgbe_attach),
125 	DEVMETHOD(device_detach,	cxgbe_detach),
126 	{ 0, 0 }
127 };
128 static driver_t cxgbe_driver = {
129 	"cxgbe",
130 	cxgbe_methods,
131 	sizeof(struct port_info)
132 };
133 
134 /* T4 VI (vcxgbe) interface */
135 static int vcxgbe_probe(device_t);
136 static int vcxgbe_attach(device_t);
137 static int vcxgbe_detach(device_t);
138 static device_method_t vcxgbe_methods[] = {
139 	DEVMETHOD(device_probe,		vcxgbe_probe),
140 	DEVMETHOD(device_attach,	vcxgbe_attach),
141 	DEVMETHOD(device_detach,	vcxgbe_detach),
142 	{ 0, 0 }
143 };
144 static driver_t vcxgbe_driver = {
145 	"vcxgbe",
146 	vcxgbe_methods,
147 	sizeof(struct vi_info)
148 };
149 
150 static d_ioctl_t t4_ioctl;
151 
152 static struct cdevsw t4_cdevsw = {
153        .d_version = D_VERSION,
154        .d_ioctl = t4_ioctl,
155        .d_name = "t4nex",
156 };
157 
158 /* T5 bus driver interface */
159 static int t5_probe(device_t);
160 static device_method_t t5_methods[] = {
161 	DEVMETHOD(device_probe,		t5_probe),
162 	DEVMETHOD(device_attach,	t4_attach),
163 	DEVMETHOD(device_detach,	t4_detach),
164 
165 	DEVMETHOD(bus_child_location_str, t4_child_location_str),
166 
167 	DEVMETHOD(t4_is_main_ready,	t4_ready),
168 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
169 
170 	DEVMETHOD_END
171 };
172 static driver_t t5_driver = {
173 	"t5nex",
174 	t5_methods,
175 	sizeof(struct adapter)
176 };
177 
178 
179 /* T5 port (cxl) interface */
180 static driver_t cxl_driver = {
181 	"cxl",
182 	cxgbe_methods,
183 	sizeof(struct port_info)
184 };
185 
186 /* T5 VI (vcxl) interface */
187 static driver_t vcxl_driver = {
188 	"vcxl",
189 	vcxgbe_methods,
190 	sizeof(struct vi_info)
191 };
192 
193 /* T6 bus driver interface */
194 static int t6_probe(device_t);
195 static device_method_t t6_methods[] = {
196 	DEVMETHOD(device_probe,		t6_probe),
197 	DEVMETHOD(device_attach,	t4_attach),
198 	DEVMETHOD(device_detach,	t4_detach),
199 
200 	DEVMETHOD(bus_child_location_str, t4_child_location_str),
201 
202 	DEVMETHOD(t4_is_main_ready,	t4_ready),
203 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
204 
205 	DEVMETHOD_END
206 };
207 static driver_t t6_driver = {
208 	"t6nex",
209 	t6_methods,
210 	sizeof(struct adapter)
211 };
212 
213 
214 /* T6 port (cc) interface */
215 static driver_t cc_driver = {
216 	"cc",
217 	cxgbe_methods,
218 	sizeof(struct port_info)
219 };
220 
221 /* T6 VI (vcc) interface */
222 static driver_t vcc_driver = {
223 	"vcc",
224 	vcxgbe_methods,
225 	sizeof(struct vi_info)
226 };
227 
228 /* ifnet interface */
229 static void cxgbe_init(void *);
230 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
231 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
232 static void cxgbe_qflush(struct ifnet *);
233 
234 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
235 
236 /*
237  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
238  * then ADAPTER_LOCK, then t4_uld_list_lock.
239  */
240 static struct sx t4_list_lock;
241 SLIST_HEAD(, adapter) t4_list;
242 #ifdef TCP_OFFLOAD
243 static struct sx t4_uld_list_lock;
244 SLIST_HEAD(, uld_info) t4_uld_list;
245 #endif
246 
247 /*
248  * Tunables.  See tweak_tunables() too.
249  *
250  * Each tunable is set to a default value here if it's known at compile-time.
251  * Otherwise it is set to -n as an indication to tweak_tunables() that it should
252  * provide a reasonable default (upto n) when the driver is loaded.
253  *
254  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
255  * T5 are under hw.cxl.
256  */
257 SYSCTL_NODE(_hw, OID_AUTO, cxgbe, CTLFLAG_RD, 0, "cxgbe(4) parameters");
258 SYSCTL_NODE(_hw, OID_AUTO, cxl, CTLFLAG_RD, 0, "cxgbe(4) T5+ parameters");
259 SYSCTL_NODE(_hw_cxgbe, OID_AUTO, toe, CTLFLAG_RD, 0, "cxgbe(4) TOE parameters");
260 
261 /*
262  * Number of queues for tx and rx, NIC and offload.
263  */
264 #define NTXQ 16
265 int t4_ntxq = -NTXQ;
266 SYSCTL_INT(_hw_cxgbe, OID_AUTO, ntxq, CTLFLAG_RDTUN, &t4_ntxq, 0,
267     "Number of TX queues per port");
268 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq);	/* Old name, undocumented */
269 
270 #define NRXQ 8
271 int t4_nrxq = -NRXQ;
272 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nrxq, CTLFLAG_RDTUN, &t4_nrxq, 0,
273     "Number of RX queues per port");
274 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq);	/* Old name, undocumented */
275 
276 #define NTXQ_VI 1
277 static int t4_ntxq_vi = -NTXQ_VI;
278 SYSCTL_INT(_hw_cxgbe, OID_AUTO, ntxq_vi, CTLFLAG_RDTUN, &t4_ntxq_vi, 0,
279     "Number of TX queues per VI");
280 
281 #define NRXQ_VI 1
282 static int t4_nrxq_vi = -NRXQ_VI;
283 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nrxq_vi, CTLFLAG_RDTUN, &t4_nrxq_vi, 0,
284     "Number of RX queues per VI");
285 
286 static int t4_rsrv_noflowq = 0;
287 SYSCTL_INT(_hw_cxgbe, OID_AUTO, rsrv_noflowq, CTLFLAG_RDTUN, &t4_rsrv_noflowq,
288     0, "Reserve TX queue 0 of each VI for non-flowid packets");
289 
290 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
291 #define NOFLDTXQ 8
292 static int t4_nofldtxq = -NOFLDTXQ;
293 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldtxq, CTLFLAG_RDTUN, &t4_nofldtxq, 0,
294     "Number of offload TX queues per port");
295 
296 #define NOFLDRXQ 2
297 static int t4_nofldrxq = -NOFLDRXQ;
298 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldrxq, CTLFLAG_RDTUN, &t4_nofldrxq, 0,
299     "Number of offload RX queues per port");
300 
301 #define NOFLDTXQ_VI 1
302 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
303 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldtxq_vi, CTLFLAG_RDTUN, &t4_nofldtxq_vi, 0,
304     "Number of offload TX queues per VI");
305 
306 #define NOFLDRXQ_VI 1
307 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
308 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldrxq_vi, CTLFLAG_RDTUN, &t4_nofldrxq_vi, 0,
309     "Number of offload RX queues per VI");
310 
311 #define TMR_IDX_OFLD 1
312 int t4_tmr_idx_ofld = TMR_IDX_OFLD;
313 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_timer_idx_ofld, CTLFLAG_RDTUN,
314     &t4_tmr_idx_ofld, 0, "Holdoff timer index for offload queues");
315 
316 #define PKTC_IDX_OFLD (-1)
317 int t4_pktc_idx_ofld = PKTC_IDX_OFLD;
318 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_pktc_idx_ofld, CTLFLAG_RDTUN,
319     &t4_pktc_idx_ofld, 0, "holdoff packet counter index for offload queues");
320 
321 /* 0 means chip/fw default, non-zero number is value in microseconds */
322 static u_long t4_toe_keepalive_idle = 0;
323 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, keepalive_idle, CTLFLAG_RDTUN,
324     &t4_toe_keepalive_idle, 0, "TOE keepalive idle timer (us)");
325 
326 /* 0 means chip/fw default, non-zero number is value in microseconds */
327 static u_long t4_toe_keepalive_interval = 0;
328 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, keepalive_interval, CTLFLAG_RDTUN,
329     &t4_toe_keepalive_interval, 0, "TOE keepalive interval timer (us)");
330 
331 /* 0 means chip/fw default, non-zero number is # of keepalives before abort */
332 static int t4_toe_keepalive_count = 0;
333 SYSCTL_INT(_hw_cxgbe_toe, OID_AUTO, keepalive_count, CTLFLAG_RDTUN,
334     &t4_toe_keepalive_count, 0, "Number of TOE keepalive probes before abort");
335 
336 /* 0 means chip/fw default, non-zero number is value in microseconds */
337 static u_long t4_toe_rexmt_min = 0;
338 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, rexmt_min, CTLFLAG_RDTUN,
339     &t4_toe_rexmt_min, 0, "Minimum TOE retransmit interval (us)");
340 
341 /* 0 means chip/fw default, non-zero number is value in microseconds */
342 static u_long t4_toe_rexmt_max = 0;
343 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, rexmt_max, CTLFLAG_RDTUN,
344     &t4_toe_rexmt_max, 0, "Maximum TOE retransmit interval (us)");
345 
346 /* 0 means chip/fw default, non-zero number is # of rexmt before abort */
347 static int t4_toe_rexmt_count = 0;
348 SYSCTL_INT(_hw_cxgbe_toe, OID_AUTO, rexmt_count, CTLFLAG_RDTUN,
349     &t4_toe_rexmt_count, 0, "Number of TOE retransmissions before abort");
350 
351 /* -1 means chip/fw default, other values are raw backoff values to use */
352 static int t4_toe_rexmt_backoff[16] = {
353 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
354 };
355 SYSCTL_NODE(_hw_cxgbe_toe, OID_AUTO, rexmt_backoff, CTLFLAG_RD, 0,
356     "cxgbe(4) TOE retransmit backoff values");
357 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 0, CTLFLAG_RDTUN,
358     &t4_toe_rexmt_backoff[0], 0, "");
359 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 1, CTLFLAG_RDTUN,
360     &t4_toe_rexmt_backoff[1], 0, "");
361 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 2, CTLFLAG_RDTUN,
362     &t4_toe_rexmt_backoff[2], 0, "");
363 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 3, CTLFLAG_RDTUN,
364     &t4_toe_rexmt_backoff[3], 0, "");
365 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 4, CTLFLAG_RDTUN,
366     &t4_toe_rexmt_backoff[4], 0, "");
367 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 5, CTLFLAG_RDTUN,
368     &t4_toe_rexmt_backoff[5], 0, "");
369 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 6, CTLFLAG_RDTUN,
370     &t4_toe_rexmt_backoff[6], 0, "");
371 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 7, CTLFLAG_RDTUN,
372     &t4_toe_rexmt_backoff[7], 0, "");
373 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 8, CTLFLAG_RDTUN,
374     &t4_toe_rexmt_backoff[8], 0, "");
375 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 9, CTLFLAG_RDTUN,
376     &t4_toe_rexmt_backoff[9], 0, "");
377 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 10, CTLFLAG_RDTUN,
378     &t4_toe_rexmt_backoff[10], 0, "");
379 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 11, CTLFLAG_RDTUN,
380     &t4_toe_rexmt_backoff[11], 0, "");
381 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 12, CTLFLAG_RDTUN,
382     &t4_toe_rexmt_backoff[12], 0, "");
383 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 13, CTLFLAG_RDTUN,
384     &t4_toe_rexmt_backoff[13], 0, "");
385 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 14, CTLFLAG_RDTUN,
386     &t4_toe_rexmt_backoff[14], 0, "");
387 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 15, CTLFLAG_RDTUN,
388     &t4_toe_rexmt_backoff[15], 0, "");
389 #endif
390 
391 #ifdef DEV_NETMAP
392 #define NNMTXQ_VI 2
393 static int t4_nnmtxq_vi = -NNMTXQ_VI;
394 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmtxq_vi, CTLFLAG_RDTUN, &t4_nnmtxq_vi, 0,
395     "Number of netmap TX queues per VI");
396 
397 #define NNMRXQ_VI 2
398 static int t4_nnmrxq_vi = -NNMRXQ_VI;
399 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmrxq_vi, CTLFLAG_RDTUN, &t4_nnmrxq_vi, 0,
400     "Number of netmap RX queues per VI");
401 #endif
402 
403 /*
404  * Holdoff parameters for ports.
405  */
406 #define TMR_IDX 1
407 int t4_tmr_idx = TMR_IDX;
408 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_timer_idx, CTLFLAG_RDTUN, &t4_tmr_idx,
409     0, "Holdoff timer index");
410 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx);	/* Old name */
411 
412 #define PKTC_IDX (-1)
413 int t4_pktc_idx = PKTC_IDX;
414 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_pktc_idx, CTLFLAG_RDTUN, &t4_pktc_idx,
415     0, "Holdoff packet counter index");
416 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx);	/* Old name */
417 
418 /*
419  * Size (# of entries) of each tx and rx queue.
420  */
421 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
422 SYSCTL_INT(_hw_cxgbe, OID_AUTO, qsize_txq, CTLFLAG_RDTUN, &t4_qsize_txq, 0,
423     "Number of descriptors in each TX queue");
424 
425 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
426 SYSCTL_INT(_hw_cxgbe, OID_AUTO, qsize_rxq, CTLFLAG_RDTUN, &t4_qsize_rxq, 0,
427     "Number of descriptors in each RX queue");
428 
429 /*
430  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
431  */
432 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
433 SYSCTL_INT(_hw_cxgbe, OID_AUTO, interrupt_types, CTLFLAG_RDTUN, &t4_intr_types,
434     0, "Interrupt types allowed (bit 0 = INTx, 1 = MSI, 2 = MSI-X)");
435 
436 /*
437  * Configuration file.  All the _CF names here are special.
438  */
439 #define DEFAULT_CF	"default"
440 #define BUILTIN_CF	"built-in"
441 #define FLASH_CF	"flash"
442 #define UWIRE_CF	"uwire"
443 #define FPGA_CF		"fpga"
444 static char t4_cfg_file[32] = DEFAULT_CF;
445 SYSCTL_STRING(_hw_cxgbe, OID_AUTO, config_file, CTLFLAG_RDTUN, t4_cfg_file,
446     sizeof(t4_cfg_file), "Firmware configuration file");
447 
448 /*
449  * PAUSE settings (bit 0, 1, 2 = rx_pause, tx_pause, pause_autoneg respectively).
450  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
451  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
452  *            mark or when signalled to do so, 0 to never emit PAUSE.
453  * pause_autoneg = 1 means PAUSE will be negotiated if possible and the
454  *                 negotiated settings will override rx_pause/tx_pause.
455  *                 Otherwise rx_pause/tx_pause are applied forcibly.
456  */
457 static int t4_pause_settings = PAUSE_RX | PAUSE_TX | PAUSE_AUTONEG;
458 SYSCTL_INT(_hw_cxgbe, OID_AUTO, pause_settings, CTLFLAG_RDTUN,
459     &t4_pause_settings, 0,
460     "PAUSE settings (bit 0 = rx_pause, 1 = tx_pause, 2 = pause_autoneg)");
461 
462 /*
463  * Forward Error Correction settings (bit 0, 1 = RS, BASER respectively).
464  * -1 to run with the firmware default.  Same as FEC_AUTO (bit 5)
465  *  0 to disable FEC.
466  */
467 static int t4_fec = -1;
468 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fec, CTLFLAG_RDTUN, &t4_fec, 0,
469     "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
470 
471 /*
472  * Link autonegotiation.
473  * -1 to run with the firmware default.
474  *  0 to disable.
475  *  1 to enable.
476  */
477 static int t4_autoneg = -1;
478 SYSCTL_INT(_hw_cxgbe, OID_AUTO, autoneg, CTLFLAG_RDTUN, &t4_autoneg, 0,
479     "Link autonegotiation");
480 
481 /*
482  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
483  * encouraged respectively).  '-n' is the same as 'n' except the firmware
484  * version used in the checks is read from the firmware bundled with the driver.
485  */
486 static int t4_fw_install = 1;
487 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fw_install, CTLFLAG_RDTUN, &t4_fw_install, 0,
488     "Firmware auto-install (0 = prohibited, 1 = allowed, 2 = encouraged)");
489 
490 /*
491  * ASIC features that will be used.  Disable the ones you don't want so that the
492  * chip resources aren't wasted on features that will not be used.
493  */
494 static int t4_nbmcaps_allowed = 0;
495 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nbmcaps_allowed, CTLFLAG_RDTUN,
496     &t4_nbmcaps_allowed, 0, "Default NBM capabilities");
497 
498 static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
499 SYSCTL_INT(_hw_cxgbe, OID_AUTO, linkcaps_allowed, CTLFLAG_RDTUN,
500     &t4_linkcaps_allowed, 0, "Default link capabilities");
501 
502 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
503     FW_CAPS_CONFIG_SWITCH_EGRESS;
504 SYSCTL_INT(_hw_cxgbe, OID_AUTO, switchcaps_allowed, CTLFLAG_RDTUN,
505     &t4_switchcaps_allowed, 0, "Default switch capabilities");
506 
507 #ifdef RATELIMIT
508 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
509 	FW_CAPS_CONFIG_NIC_HASHFILTER | FW_CAPS_CONFIG_NIC_ETHOFLD;
510 #else
511 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
512 	FW_CAPS_CONFIG_NIC_HASHFILTER;
513 #endif
514 SYSCTL_INT(_hw_cxgbe, OID_AUTO, niccaps_allowed, CTLFLAG_RDTUN,
515     &t4_niccaps_allowed, 0, "Default NIC capabilities");
516 
517 static int t4_toecaps_allowed = -1;
518 SYSCTL_INT(_hw_cxgbe, OID_AUTO, toecaps_allowed, CTLFLAG_RDTUN,
519     &t4_toecaps_allowed, 0, "Default TCP offload capabilities");
520 
521 static int t4_rdmacaps_allowed = -1;
522 SYSCTL_INT(_hw_cxgbe, OID_AUTO, rdmacaps_allowed, CTLFLAG_RDTUN,
523     &t4_rdmacaps_allowed, 0, "Default RDMA capabilities");
524 
525 static int t4_cryptocaps_allowed = -1;
526 SYSCTL_INT(_hw_cxgbe, OID_AUTO, cryptocaps_allowed, CTLFLAG_RDTUN,
527     &t4_cryptocaps_allowed, 0, "Default crypto capabilities");
528 
529 static int t4_iscsicaps_allowed = -1;
530 SYSCTL_INT(_hw_cxgbe, OID_AUTO, iscsicaps_allowed, CTLFLAG_RDTUN,
531     &t4_iscsicaps_allowed, 0, "Default iSCSI capabilities");
532 
533 static int t4_fcoecaps_allowed = 0;
534 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fcoecaps_allowed, CTLFLAG_RDTUN,
535     &t4_fcoecaps_allowed, 0, "Default FCoE capabilities");
536 
537 static int t5_write_combine = 0;
538 SYSCTL_INT(_hw_cxl, OID_AUTO, write_combine, CTLFLAG_RDTUN, &t5_write_combine,
539     0, "Use WC instead of UC for BAR2");
540 
541 static int t4_num_vis = 1;
542 SYSCTL_INT(_hw_cxgbe, OID_AUTO, num_vis, CTLFLAG_RDTUN, &t4_num_vis, 0,
543     "Number of VIs per port");
544 
545 /*
546  * PCIe Relaxed Ordering.
547  * -1: driver should figure out a good value.
548  * 0: disable RO.
549  * 1: enable RO.
550  * 2: leave RO alone.
551  */
552 static int pcie_relaxed_ordering = -1;
553 SYSCTL_INT(_hw_cxgbe, OID_AUTO, pcie_relaxed_ordering, CTLFLAG_RDTUN,
554     &pcie_relaxed_ordering, 0,
555     "PCIe Relaxed Ordering: 0 = disable, 1 = enable, 2 = leave alone");
556 
557 static int t4_panic_on_fatal_err = 0;
558 SYSCTL_INT(_hw_cxgbe, OID_AUTO, panic_on_fatal_err, CTLFLAG_RDTUN,
559     &t4_panic_on_fatal_err, 0, "panic on fatal errors");
560 
561 #ifdef TCP_OFFLOAD
562 /*
563  * TOE tunables.
564  */
565 static int t4_cop_managed_offloading = 0;
566 SYSCTL_INT(_hw_cxgbe, OID_AUTO, cop_managed_offloading, CTLFLAG_RDTUN,
567     &t4_cop_managed_offloading, 0,
568     "COP (Connection Offload Policy) controls all TOE offload");
569 #endif
570 
571 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
572 static int vi_mac_funcs[] = {
573 	FW_VI_FUNC_ETH,
574 	FW_VI_FUNC_OFLD,
575 	FW_VI_FUNC_IWARP,
576 	FW_VI_FUNC_OPENISCSI,
577 	FW_VI_FUNC_OPENFCOE,
578 	FW_VI_FUNC_FOISCSI,
579 	FW_VI_FUNC_FOFCOE,
580 };
581 
582 struct intrs_and_queues {
583 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
584 	uint16_t num_vis;	/* number of VIs for each port */
585 	uint16_t nirq;		/* Total # of vectors */
586 	uint16_t ntxq;		/* # of NIC txq's for each port */
587 	uint16_t nrxq;		/* # of NIC rxq's for each port */
588 	uint16_t nofldtxq;	/* # of TOE/ETHOFLD txq's for each port */
589 	uint16_t nofldrxq;	/* # of TOE rxq's for each port */
590 
591 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
592 	uint16_t ntxq_vi;	/* # of NIC txq's */
593 	uint16_t nrxq_vi;	/* # of NIC rxq's */
594 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
595 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
596 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
597 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
598 };
599 
600 static void setup_memwin(struct adapter *);
601 static void position_memwin(struct adapter *, int, uint32_t);
602 static int validate_mem_range(struct adapter *, uint32_t, uint32_t);
603 static int fwmtype_to_hwmtype(int);
604 static int validate_mt_off_len(struct adapter *, int, uint32_t, uint32_t,
605     uint32_t *);
606 static int fixup_devlog_params(struct adapter *);
607 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
608 static int contact_firmware(struct adapter *);
609 static int partition_resources(struct adapter *);
610 static int get_params__pre_init(struct adapter *);
611 static int set_params__pre_init(struct adapter *);
612 static int get_params__post_init(struct adapter *);
613 static int set_params__post_init(struct adapter *);
614 static void t4_set_desc(struct adapter *);
615 static bool fixed_ifmedia(struct port_info *);
616 static void build_medialist(struct port_info *);
617 static void init_link_config(struct port_info *);
618 static int fixup_link_config(struct port_info *);
619 static int apply_link_config(struct port_info *);
620 static int cxgbe_init_synchronized(struct vi_info *);
621 static int cxgbe_uninit_synchronized(struct vi_info *);
622 static void quiesce_txq(struct adapter *, struct sge_txq *);
623 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
624 static void quiesce_iq(struct adapter *, struct sge_iq *);
625 static void quiesce_fl(struct adapter *, struct sge_fl *);
626 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
627     driver_intr_t *, void *, char *);
628 static int t4_free_irq(struct adapter *, struct irq *);
629 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
630 static void vi_refresh_stats(struct adapter *, struct vi_info *);
631 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
632 static void cxgbe_tick(void *);
633 static void cxgbe_sysctls(struct port_info *);
634 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
635 static int sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS);
636 static int sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS);
637 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
638 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
639 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
640 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
641 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
642 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
643 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
644 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
645 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
646 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
647 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
648 static int sysctl_loadavg(SYSCTL_HANDLER_ARGS);
649 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
650 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
651 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
652 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
653 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
654 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
655 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
656 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
657 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
658 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
659 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
660 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
661 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
662 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
663 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
664 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
665 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
666 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
667 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
668 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
669 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
670 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
671 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
672 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
673 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
674 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
675 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
676 static int sysctl_cpus(SYSCTL_HANDLER_ARGS);
677 #ifdef TCP_OFFLOAD
678 static int sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS);
679 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
680 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
681 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
682 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
683 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
684 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
685 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
686 #endif
687 static int get_sge_context(struct adapter *, struct t4_sge_context *);
688 static int load_fw(struct adapter *, struct t4_data *);
689 static int load_cfg(struct adapter *, struct t4_data *);
690 static int load_boot(struct adapter *, struct t4_bootrom *);
691 static int load_bootcfg(struct adapter *, struct t4_data *);
692 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
693 static void free_offload_policy(struct t4_offload_policy *);
694 static int set_offload_policy(struct adapter *, struct t4_offload_policy *);
695 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
696 static int read_i2c(struct adapter *, struct t4_i2c_data *);
697 static int clear_stats(struct adapter *, u_int);
698 #ifdef TCP_OFFLOAD
699 static int toe_capability(struct vi_info *, int);
700 #endif
701 static int mod_event(module_t, int, void *);
702 static int notify_siblings(device_t, int);
703 
704 struct {
705 	uint16_t device;
706 	char *desc;
707 } t4_pciids[] = {
708 	{0xa000, "Chelsio Terminator 4 FPGA"},
709 	{0x4400, "Chelsio T440-dbg"},
710 	{0x4401, "Chelsio T420-CR"},
711 	{0x4402, "Chelsio T422-CR"},
712 	{0x4403, "Chelsio T440-CR"},
713 	{0x4404, "Chelsio T420-BCH"},
714 	{0x4405, "Chelsio T440-BCH"},
715 	{0x4406, "Chelsio T440-CH"},
716 	{0x4407, "Chelsio T420-SO"},
717 	{0x4408, "Chelsio T420-CX"},
718 	{0x4409, "Chelsio T420-BT"},
719 	{0x440a, "Chelsio T404-BT"},
720 	{0x440e, "Chelsio T440-LP-CR"},
721 }, t5_pciids[] = {
722 	{0xb000, "Chelsio Terminator 5 FPGA"},
723 	{0x5400, "Chelsio T580-dbg"},
724 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
725 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
726 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
727 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
728 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
729 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
730 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
731 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
732 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
733 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
734 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
735 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
736 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
737 	{0x5418,  "Chelsio T540-BT"},		/* 4 x 10GBaseT */
738 	{0x5419,  "Chelsio T540-LP-BT"},	/* 4 x 10GBaseT */
739 	{0x541a,  "Chelsio T540-SO-BT"},	/* 4 x 10GBaseT, nomem */
740 	{0x541b,  "Chelsio T540-SO-CR"},	/* 4 x 10G, nomem */
741 
742 	/* Custom */
743 	{0x5483, "Custom T540-CR"},
744 	{0x5484, "Custom T540-BT"},
745 }, t6_pciids[] = {
746 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
747 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
748 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
749 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
750 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
751 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
752 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
753 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
754 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
755 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
756 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
757 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
758 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
759 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
760 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
761 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
762 
763 	/* Custom */
764 	{0x6480, "Custom T6225-CR"},
765 	{0x6481, "Custom T62100-CR"},
766 	{0x6482, "Custom T6225-CR"},
767 	{0x6483, "Custom T62100-CR"},
768 	{0x6484, "Custom T64100-CR"},
769 	{0x6485, "Custom T6240-SO"},
770 	{0x6486, "Custom T6225-SO-CR"},
771 	{0x6487, "Custom T6225-CR"},
772 };
773 
774 #ifdef TCP_OFFLOAD
775 /*
776  * service_iq_fl() has an iq and needs the fl.  Offset of fl from the iq should
777  * be exactly the same for both rxq and ofld_rxq.
778  */
779 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
780 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
781 #endif
782 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
783 
784 static int
785 t4_probe(device_t dev)
786 {
787 	int i;
788 	uint16_t v = pci_get_vendor(dev);
789 	uint16_t d = pci_get_device(dev);
790 	uint8_t f = pci_get_function(dev);
791 
792 	if (v != PCI_VENDOR_ID_CHELSIO)
793 		return (ENXIO);
794 
795 	/* Attach only to PF0 of the FPGA */
796 	if (d == 0xa000 && f != 0)
797 		return (ENXIO);
798 
799 	for (i = 0; i < nitems(t4_pciids); i++) {
800 		if (d == t4_pciids[i].device) {
801 			device_set_desc(dev, t4_pciids[i].desc);
802 			return (BUS_PROBE_DEFAULT);
803 		}
804 	}
805 
806 	return (ENXIO);
807 }
808 
809 static int
810 t5_probe(device_t dev)
811 {
812 	int i;
813 	uint16_t v = pci_get_vendor(dev);
814 	uint16_t d = pci_get_device(dev);
815 	uint8_t f = pci_get_function(dev);
816 
817 	if (v != PCI_VENDOR_ID_CHELSIO)
818 		return (ENXIO);
819 
820 	/* Attach only to PF0 of the FPGA */
821 	if (d == 0xb000 && f != 0)
822 		return (ENXIO);
823 
824 	for (i = 0; i < nitems(t5_pciids); i++) {
825 		if (d == t5_pciids[i].device) {
826 			device_set_desc(dev, t5_pciids[i].desc);
827 			return (BUS_PROBE_DEFAULT);
828 		}
829 	}
830 
831 	return (ENXIO);
832 }
833 
834 static int
835 t6_probe(device_t dev)
836 {
837 	int i;
838 	uint16_t v = pci_get_vendor(dev);
839 	uint16_t d = pci_get_device(dev);
840 
841 	if (v != PCI_VENDOR_ID_CHELSIO)
842 		return (ENXIO);
843 
844 	for (i = 0; i < nitems(t6_pciids); i++) {
845 		if (d == t6_pciids[i].device) {
846 			device_set_desc(dev, t6_pciids[i].desc);
847 			return (BUS_PROBE_DEFAULT);
848 		}
849 	}
850 
851 	return (ENXIO);
852 }
853 
854 static void
855 t5_attribute_workaround(device_t dev)
856 {
857 	device_t root_port;
858 	uint32_t v;
859 
860 	/*
861 	 * The T5 chips do not properly echo the No Snoop and Relaxed
862 	 * Ordering attributes when replying to a TLP from a Root
863 	 * Port.  As a workaround, find the parent Root Port and
864 	 * disable No Snoop and Relaxed Ordering.  Note that this
865 	 * affects all devices under this root port.
866 	 */
867 	root_port = pci_find_pcie_root_port(dev);
868 	if (root_port == NULL) {
869 		device_printf(dev, "Unable to find parent root port\n");
870 		return;
871 	}
872 
873 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
874 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
875 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
876 	    0)
877 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
878 		    device_get_nameunit(root_port));
879 }
880 
881 static const struct devnames devnames[] = {
882 	{
883 		.nexus_name = "t4nex",
884 		.ifnet_name = "cxgbe",
885 		.vi_ifnet_name = "vcxgbe",
886 		.pf03_drv_name = "t4iov",
887 		.vf_nexus_name = "t4vf",
888 		.vf_ifnet_name = "cxgbev"
889 	}, {
890 		.nexus_name = "t5nex",
891 		.ifnet_name = "cxl",
892 		.vi_ifnet_name = "vcxl",
893 		.pf03_drv_name = "t5iov",
894 		.vf_nexus_name = "t5vf",
895 		.vf_ifnet_name = "cxlv"
896 	}, {
897 		.nexus_name = "t6nex",
898 		.ifnet_name = "cc",
899 		.vi_ifnet_name = "vcc",
900 		.pf03_drv_name = "t6iov",
901 		.vf_nexus_name = "t6vf",
902 		.vf_ifnet_name = "ccv"
903 	}
904 };
905 
906 void
907 t4_init_devnames(struct adapter *sc)
908 {
909 	int id;
910 
911 	id = chip_id(sc);
912 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
913 		sc->names = &devnames[id - CHELSIO_T4];
914 	else {
915 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
916 		sc->names = NULL;
917 	}
918 }
919 
920 static int
921 t4_ifnet_unit(struct adapter *sc, struct port_info *pi)
922 {
923 	const char *parent, *name;
924 	long value;
925 	int line, unit;
926 
927 	line = 0;
928 	parent = device_get_nameunit(sc->dev);
929 	name = sc->names->ifnet_name;
930 	while (resource_find_dev(&line, name, &unit, "at", parent) == 0) {
931 		if (resource_long_value(name, unit, "port", &value) == 0 &&
932 		    value == pi->port_id)
933 			return (unit);
934 	}
935 	return (-1);
936 }
937 
938 static int
939 t4_attach(device_t dev)
940 {
941 	struct adapter *sc;
942 	int rc = 0, i, j, rqidx, tqidx, nports;
943 	struct make_dev_args mda;
944 	struct intrs_and_queues iaq;
945 	struct sge *s;
946 	uint32_t *buf;
947 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
948 	int ofld_tqidx;
949 #endif
950 #ifdef TCP_OFFLOAD
951 	int ofld_rqidx;
952 #endif
953 #ifdef DEV_NETMAP
954 	int nm_rqidx, nm_tqidx;
955 #endif
956 	int num_vis;
957 
958 	sc = device_get_softc(dev);
959 	sc->dev = dev;
960 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
961 
962 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
963 		t5_attribute_workaround(dev);
964 	pci_enable_busmaster(dev);
965 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
966 		uint32_t v;
967 
968 		pci_set_max_read_req(dev, 4096);
969 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
970 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
971 		if (pcie_relaxed_ordering == 0 &&
972 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
973 			v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
974 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
975 		} else if (pcie_relaxed_ordering == 1 &&
976 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
977 			v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
978 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
979 		}
980 	}
981 
982 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
983 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
984 	sc->traceq = -1;
985 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
986 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
987 	    device_get_nameunit(dev));
988 
989 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
990 	    device_get_nameunit(dev));
991 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
992 	t4_add_adapter(sc);
993 
994 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
995 	TAILQ_INIT(&sc->sfl);
996 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
997 
998 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
999 
1000 	sc->policy = NULL;
1001 	rw_init(&sc->policy_lock, "connection offload policy");
1002 
1003 	rc = t4_map_bars_0_and_4(sc);
1004 	if (rc != 0)
1005 		goto done; /* error message displayed already */
1006 
1007 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
1008 
1009 	/* Prepare the adapter for operation. */
1010 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
1011 	rc = -t4_prep_adapter(sc, buf);
1012 	free(buf, M_CXGBE);
1013 	if (rc != 0) {
1014 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
1015 		goto done;
1016 	}
1017 
1018 	/*
1019 	 * This is the real PF# to which we're attaching.  Works from within PCI
1020 	 * passthrough environments too, where pci_get_function() could return a
1021 	 * different PF# depending on the passthrough configuration.  We need to
1022 	 * use the real PF# in all our communication with the firmware.
1023 	 */
1024 	j = t4_read_reg(sc, A_PL_WHOAMI);
1025 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
1026 	sc->mbox = sc->pf;
1027 
1028 	t4_init_devnames(sc);
1029 	if (sc->names == NULL) {
1030 		rc = ENOTSUP;
1031 		goto done; /* error message displayed already */
1032 	}
1033 
1034 	/*
1035 	 * Do this really early, with the memory windows set up even before the
1036 	 * character device.  The userland tool's register i/o and mem read
1037 	 * will work even in "recovery mode".
1038 	 */
1039 	setup_memwin(sc);
1040 	if (t4_init_devlog_params(sc, 0) == 0)
1041 		fixup_devlog_params(sc);
1042 	make_dev_args_init(&mda);
1043 	mda.mda_devsw = &t4_cdevsw;
1044 	mda.mda_uid = UID_ROOT;
1045 	mda.mda_gid = GID_WHEEL;
1046 	mda.mda_mode = 0600;
1047 	mda.mda_si_drv1 = sc;
1048 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
1049 	if (rc != 0)
1050 		device_printf(dev, "failed to create nexus char device: %d.\n",
1051 		    rc);
1052 
1053 	/* Go no further if recovery mode has been requested. */
1054 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
1055 		device_printf(dev, "recovery mode.\n");
1056 		goto done;
1057 	}
1058 
1059 #if defined(__i386__)
1060 	if ((cpu_feature & CPUID_CX8) == 0) {
1061 		device_printf(dev, "64 bit atomics not available.\n");
1062 		rc = ENOTSUP;
1063 		goto done;
1064 	}
1065 #endif
1066 
1067 	/* Contact the firmware and try to become the master driver. */
1068 	rc = contact_firmware(sc);
1069 	if (rc != 0)
1070 		goto done; /* error message displayed already */
1071 	MPASS(sc->flags & FW_OK);
1072 
1073 	rc = get_params__pre_init(sc);
1074 	if (rc != 0)
1075 		goto done; /* error message displayed already */
1076 
1077 	if (sc->flags & MASTER_PF) {
1078 		rc = partition_resources(sc);
1079 		if (rc != 0)
1080 			goto done; /* error message displayed already */
1081 		t4_intr_clear(sc);
1082 	}
1083 
1084 	rc = get_params__post_init(sc);
1085 	if (rc != 0)
1086 		goto done; /* error message displayed already */
1087 
1088 	rc = set_params__post_init(sc);
1089 	if (rc != 0)
1090 		goto done; /* error message displayed already */
1091 
1092 	rc = t4_map_bar_2(sc);
1093 	if (rc != 0)
1094 		goto done; /* error message displayed already */
1095 
1096 	rc = t4_create_dma_tag(sc);
1097 	if (rc != 0)
1098 		goto done; /* error message displayed already */
1099 
1100 	/*
1101 	 * First pass over all the ports - allocate VIs and initialize some
1102 	 * basic parameters like mac address, port type, etc.
1103 	 */
1104 	for_each_port(sc, i) {
1105 		struct port_info *pi;
1106 
1107 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
1108 		sc->port[i] = pi;
1109 
1110 		/* These must be set before t4_port_init */
1111 		pi->adapter = sc;
1112 		pi->port_id = i;
1113 		/*
1114 		 * XXX: vi[0] is special so we can't delay this allocation until
1115 		 * pi->nvi's final value is known.
1116 		 */
1117 		pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
1118 		    M_ZERO | M_WAITOK);
1119 
1120 		/*
1121 		 * Allocate the "main" VI and initialize parameters
1122 		 * like mac addr.
1123 		 */
1124 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1125 		if (rc != 0) {
1126 			device_printf(dev, "unable to initialize port %d: %d\n",
1127 			    i, rc);
1128 			free(pi->vi, M_CXGBE);
1129 			free(pi, M_CXGBE);
1130 			sc->port[i] = NULL;
1131 			goto done;
1132 		}
1133 
1134 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1135 		    device_get_nameunit(dev), i);
1136 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1137 		sc->chan_map[pi->tx_chan] = i;
1138 
1139 		/* All VIs on this port share this media. */
1140 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1141 		    cxgbe_media_status);
1142 
1143 		PORT_LOCK(pi);
1144 		init_link_config(pi);
1145 		fixup_link_config(pi);
1146 		build_medialist(pi);
1147 		if (fixed_ifmedia(pi))
1148 			pi->flags |= FIXED_IFMEDIA;
1149 		PORT_UNLOCK(pi);
1150 
1151 		pi->dev = device_add_child(dev, sc->names->ifnet_name,
1152 		    t4_ifnet_unit(sc, pi));
1153 		if (pi->dev == NULL) {
1154 			device_printf(dev,
1155 			    "failed to add device for port %d.\n", i);
1156 			rc = ENXIO;
1157 			goto done;
1158 		}
1159 		pi->vi[0].dev = pi->dev;
1160 		device_set_softc(pi->dev, pi);
1161 	}
1162 
1163 	/*
1164 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1165 	 */
1166 	nports = sc->params.nports;
1167 	rc = cfg_itype_and_nqueues(sc, &iaq);
1168 	if (rc != 0)
1169 		goto done; /* error message displayed already */
1170 
1171 	num_vis = iaq.num_vis;
1172 	sc->intr_type = iaq.intr_type;
1173 	sc->intr_count = iaq.nirq;
1174 
1175 	s = &sc->sge;
1176 	s->nrxq = nports * iaq.nrxq;
1177 	s->ntxq = nports * iaq.ntxq;
1178 	if (num_vis > 1) {
1179 		s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1180 		s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1181 	}
1182 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1183 	s->neq += nports;		/* ctrl queues: 1 per port */
1184 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1185 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1186 	if (is_offload(sc) || is_ethoffload(sc)) {
1187 		s->nofldtxq = nports * iaq.nofldtxq;
1188 		if (num_vis > 1)
1189 			s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1190 		s->neq += s->nofldtxq;
1191 
1192 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1193 		    M_CXGBE, M_ZERO | M_WAITOK);
1194 	}
1195 #endif
1196 #ifdef TCP_OFFLOAD
1197 	if (is_offload(sc)) {
1198 		s->nofldrxq = nports * iaq.nofldrxq;
1199 		if (num_vis > 1)
1200 			s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1201 		s->neq += s->nofldrxq;	/* free list */
1202 		s->niq += s->nofldrxq;
1203 
1204 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1205 		    M_CXGBE, M_ZERO | M_WAITOK);
1206 	}
1207 #endif
1208 #ifdef DEV_NETMAP
1209 	if (num_vis > 1) {
1210 		s->nnmrxq = nports * (num_vis - 1) * iaq.nnmrxq_vi;
1211 		s->nnmtxq = nports * (num_vis - 1) * iaq.nnmtxq_vi;
1212 	}
1213 	s->neq += s->nnmtxq + s->nnmrxq;
1214 	s->niq += s->nnmrxq;
1215 
1216 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1217 	    M_CXGBE, M_ZERO | M_WAITOK);
1218 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1219 	    M_CXGBE, M_ZERO | M_WAITOK);
1220 #endif
1221 
1222 	s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1223 	    M_ZERO | M_WAITOK);
1224 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1225 	    M_ZERO | M_WAITOK);
1226 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1227 	    M_ZERO | M_WAITOK);
1228 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1229 	    M_ZERO | M_WAITOK);
1230 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1231 	    M_ZERO | M_WAITOK);
1232 
1233 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1234 	    M_ZERO | M_WAITOK);
1235 
1236 	t4_init_l2t(sc, M_WAITOK);
1237 	t4_init_smt(sc, M_WAITOK);
1238 	t4_init_tx_sched(sc);
1239 #ifdef RATELIMIT
1240 	t4_init_etid_table(sc);
1241 #endif
1242 #ifdef INET6
1243 	t4_init_clip_table(sc);
1244 #endif
1245 	if (sc->vres.key.size != 0)
1246 		sc->key_map = vmem_create("T4TLS key map", sc->vres.key.start,
1247 		    sc->vres.key.size, 32, 0, M_FIRSTFIT | M_WAITOK);
1248 
1249 	/*
1250 	 * Second pass over the ports.  This time we know the number of rx and
1251 	 * tx queues that each port should get.
1252 	 */
1253 	rqidx = tqidx = 0;
1254 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1255 	ofld_tqidx = 0;
1256 #endif
1257 #ifdef TCP_OFFLOAD
1258 	ofld_rqidx = 0;
1259 #endif
1260 #ifdef DEV_NETMAP
1261 	nm_rqidx = nm_tqidx = 0;
1262 #endif
1263 	for_each_port(sc, i) {
1264 		struct port_info *pi = sc->port[i];
1265 		struct vi_info *vi;
1266 
1267 		if (pi == NULL)
1268 			continue;
1269 
1270 		pi->nvi = num_vis;
1271 		for_each_vi(pi, j, vi) {
1272 			vi->pi = pi;
1273 			vi->qsize_rxq = t4_qsize_rxq;
1274 			vi->qsize_txq = t4_qsize_txq;
1275 
1276 			vi->first_rxq = rqidx;
1277 			vi->first_txq = tqidx;
1278 			vi->tmr_idx = t4_tmr_idx;
1279 			vi->pktc_idx = t4_pktc_idx;
1280 			vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1281 			vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1282 
1283 			rqidx += vi->nrxq;
1284 			tqidx += vi->ntxq;
1285 
1286 			if (j == 0 && vi->ntxq > 1)
1287 				vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1288 			else
1289 				vi->rsrv_noflowq = 0;
1290 
1291 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1292 			vi->first_ofld_txq = ofld_tqidx;
1293 			vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1294 			ofld_tqidx += vi->nofldtxq;
1295 #endif
1296 #ifdef TCP_OFFLOAD
1297 			vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1298 			vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1299 			vi->first_ofld_rxq = ofld_rqidx;
1300 			vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1301 
1302 			ofld_rqidx += vi->nofldrxq;
1303 #endif
1304 #ifdef DEV_NETMAP
1305 			if (j > 0) {
1306 				vi->first_nm_rxq = nm_rqidx;
1307 				vi->first_nm_txq = nm_tqidx;
1308 				vi->nnmrxq = iaq.nnmrxq_vi;
1309 				vi->nnmtxq = iaq.nnmtxq_vi;
1310 				nm_rqidx += vi->nnmrxq;
1311 				nm_tqidx += vi->nnmtxq;
1312 			}
1313 #endif
1314 		}
1315 	}
1316 
1317 	rc = t4_setup_intr_handlers(sc);
1318 	if (rc != 0) {
1319 		device_printf(dev,
1320 		    "failed to setup interrupt handlers: %d\n", rc);
1321 		goto done;
1322 	}
1323 
1324 	rc = bus_generic_probe(dev);
1325 	if (rc != 0) {
1326 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1327 		goto done;
1328 	}
1329 
1330 	/*
1331 	 * Ensure thread-safe mailbox access (in debug builds).
1332 	 *
1333 	 * So far this was the only thread accessing the mailbox but various
1334 	 * ifnets and sysctls are about to be created and their handlers/ioctls
1335 	 * will access the mailbox from different threads.
1336 	 */
1337 	sc->flags |= CHK_MBOX_ACCESS;
1338 
1339 	rc = bus_generic_attach(dev);
1340 	if (rc != 0) {
1341 		device_printf(dev,
1342 		    "failed to attach all child ports: %d\n", rc);
1343 		goto done;
1344 	}
1345 
1346 	device_printf(dev,
1347 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1348 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1349 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1350 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1351 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1352 
1353 	t4_set_desc(sc);
1354 
1355 	notify_siblings(dev, 0);
1356 
1357 done:
1358 	if (rc != 0 && sc->cdev) {
1359 		/* cdev was created and so cxgbetool works; recover that way. */
1360 		device_printf(dev,
1361 		    "error during attach, adapter is now in recovery mode.\n");
1362 		rc = 0;
1363 	}
1364 
1365 	if (rc != 0)
1366 		t4_detach_common(dev);
1367 	else
1368 		t4_sysctls(sc);
1369 
1370 	return (rc);
1371 }
1372 
1373 static int
1374 t4_child_location_str(device_t bus, device_t dev, char *buf, size_t buflen)
1375 {
1376 	struct adapter *sc;
1377 	struct port_info *pi;
1378 	int i;
1379 
1380 	sc = device_get_softc(bus);
1381 	buf[0] = '\0';
1382 	for_each_port(sc, i) {
1383 		pi = sc->port[i];
1384 		if (pi != NULL && pi->dev == dev) {
1385 			snprintf(buf, buflen, "port=%d", pi->port_id);
1386 			break;
1387 		}
1388 	}
1389 	return (0);
1390 }
1391 
1392 static int
1393 t4_ready(device_t dev)
1394 {
1395 	struct adapter *sc;
1396 
1397 	sc = device_get_softc(dev);
1398 	if (sc->flags & FW_OK)
1399 		return (0);
1400 	return (ENXIO);
1401 }
1402 
1403 static int
1404 t4_read_port_device(device_t dev, int port, device_t *child)
1405 {
1406 	struct adapter *sc;
1407 	struct port_info *pi;
1408 
1409 	sc = device_get_softc(dev);
1410 	if (port < 0 || port >= MAX_NPORTS)
1411 		return (EINVAL);
1412 	pi = sc->port[port];
1413 	if (pi == NULL || pi->dev == NULL)
1414 		return (ENXIO);
1415 	*child = pi->dev;
1416 	return (0);
1417 }
1418 
1419 static int
1420 notify_siblings(device_t dev, int detaching)
1421 {
1422 	device_t sibling;
1423 	int error, i;
1424 
1425 	error = 0;
1426 	for (i = 0; i < PCI_FUNCMAX; i++) {
1427 		if (i == pci_get_function(dev))
1428 			continue;
1429 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1430 		    pci_get_slot(dev), i);
1431 		if (sibling == NULL || !device_is_attached(sibling))
1432 			continue;
1433 		if (detaching)
1434 			error = T4_DETACH_CHILD(sibling);
1435 		else
1436 			(void)T4_ATTACH_CHILD(sibling);
1437 		if (error)
1438 			break;
1439 	}
1440 	return (error);
1441 }
1442 
1443 /*
1444  * Idempotent
1445  */
1446 static int
1447 t4_detach(device_t dev)
1448 {
1449 	struct adapter *sc;
1450 	int rc;
1451 
1452 	sc = device_get_softc(dev);
1453 
1454 	rc = notify_siblings(dev, 1);
1455 	if (rc) {
1456 		device_printf(dev,
1457 		    "failed to detach sibling devices: %d\n", rc);
1458 		return (rc);
1459 	}
1460 
1461 	return (t4_detach_common(dev));
1462 }
1463 
1464 int
1465 t4_detach_common(device_t dev)
1466 {
1467 	struct adapter *sc;
1468 	struct port_info *pi;
1469 	int i, rc;
1470 
1471 	sc = device_get_softc(dev);
1472 
1473 	if (sc->cdev) {
1474 		destroy_dev(sc->cdev);
1475 		sc->cdev = NULL;
1476 	}
1477 
1478 	sc->flags &= ~CHK_MBOX_ACCESS;
1479 	if (sc->flags & FULL_INIT_DONE) {
1480 		if (!(sc->flags & IS_VF))
1481 			t4_intr_disable(sc);
1482 	}
1483 
1484 	if (device_is_attached(dev)) {
1485 		rc = bus_generic_detach(dev);
1486 		if (rc) {
1487 			device_printf(dev,
1488 			    "failed to detach child devices: %d\n", rc);
1489 			return (rc);
1490 		}
1491 	}
1492 
1493 	for (i = 0; i < sc->intr_count; i++)
1494 		t4_free_irq(sc, &sc->irq[i]);
1495 
1496 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1497 		t4_free_tx_sched(sc);
1498 
1499 	for (i = 0; i < MAX_NPORTS; i++) {
1500 		pi = sc->port[i];
1501 		if (pi) {
1502 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1503 			if (pi->dev)
1504 				device_delete_child(dev, pi->dev);
1505 
1506 			mtx_destroy(&pi->pi_lock);
1507 			free(pi->vi, M_CXGBE);
1508 			free(pi, M_CXGBE);
1509 		}
1510 	}
1511 
1512 	device_delete_children(dev);
1513 
1514 	if (sc->flags & FULL_INIT_DONE)
1515 		adapter_full_uninit(sc);
1516 
1517 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1518 		t4_fw_bye(sc, sc->mbox);
1519 
1520 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1521 		pci_release_msi(dev);
1522 
1523 	if (sc->regs_res)
1524 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1525 		    sc->regs_res);
1526 
1527 	if (sc->udbs_res)
1528 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1529 		    sc->udbs_res);
1530 
1531 	if (sc->msix_res)
1532 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1533 		    sc->msix_res);
1534 
1535 	if (sc->l2t)
1536 		t4_free_l2t(sc->l2t);
1537 	if (sc->smt)
1538 		t4_free_smt(sc->smt);
1539 #ifdef RATELIMIT
1540 	t4_free_etid_table(sc);
1541 #endif
1542 	if (sc->key_map)
1543 		vmem_destroy(sc->key_map);
1544 #ifdef INET6
1545 	t4_destroy_clip_table(sc);
1546 #endif
1547 
1548 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1549 	free(sc->sge.ofld_txq, M_CXGBE);
1550 #endif
1551 #ifdef TCP_OFFLOAD
1552 	free(sc->sge.ofld_rxq, M_CXGBE);
1553 #endif
1554 #ifdef DEV_NETMAP
1555 	free(sc->sge.nm_rxq, M_CXGBE);
1556 	free(sc->sge.nm_txq, M_CXGBE);
1557 #endif
1558 	free(sc->irq, M_CXGBE);
1559 	free(sc->sge.rxq, M_CXGBE);
1560 	free(sc->sge.txq, M_CXGBE);
1561 	free(sc->sge.ctrlq, M_CXGBE);
1562 	free(sc->sge.iqmap, M_CXGBE);
1563 	free(sc->sge.eqmap, M_CXGBE);
1564 	free(sc->tids.ftid_tab, M_CXGBE);
1565 	free(sc->tids.hpftid_tab, M_CXGBE);
1566 	free_hftid_hash(&sc->tids);
1567 	free(sc->tids.atid_tab, M_CXGBE);
1568 	free(sc->tids.tid_tab, M_CXGBE);
1569 	free(sc->tt.tls_rx_ports, M_CXGBE);
1570 	t4_destroy_dma_tag(sc);
1571 	if (mtx_initialized(&sc->sc_lock)) {
1572 		sx_xlock(&t4_list_lock);
1573 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1574 		sx_xunlock(&t4_list_lock);
1575 		mtx_destroy(&sc->sc_lock);
1576 	}
1577 
1578 	callout_drain(&sc->sfl_callout);
1579 	if (mtx_initialized(&sc->tids.ftid_lock)) {
1580 		mtx_destroy(&sc->tids.ftid_lock);
1581 		cv_destroy(&sc->tids.ftid_cv);
1582 	}
1583 	if (mtx_initialized(&sc->tids.atid_lock))
1584 		mtx_destroy(&sc->tids.atid_lock);
1585 	if (mtx_initialized(&sc->sfl_lock))
1586 		mtx_destroy(&sc->sfl_lock);
1587 	if (mtx_initialized(&sc->ifp_lock))
1588 		mtx_destroy(&sc->ifp_lock);
1589 	if (mtx_initialized(&sc->reg_lock))
1590 		mtx_destroy(&sc->reg_lock);
1591 
1592 	if (rw_initialized(&sc->policy_lock)) {
1593 		rw_destroy(&sc->policy_lock);
1594 #ifdef TCP_OFFLOAD
1595 		if (sc->policy != NULL)
1596 			free_offload_policy(sc->policy);
1597 #endif
1598 	}
1599 
1600 	for (i = 0; i < NUM_MEMWIN; i++) {
1601 		struct memwin *mw = &sc->memwin[i];
1602 
1603 		if (rw_initialized(&mw->mw_lock))
1604 			rw_destroy(&mw->mw_lock);
1605 	}
1606 
1607 	bzero(sc, sizeof(*sc));
1608 
1609 	return (0);
1610 }
1611 
1612 static int
1613 cxgbe_probe(device_t dev)
1614 {
1615 	char buf[128];
1616 	struct port_info *pi = device_get_softc(dev);
1617 
1618 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1619 	device_set_desc_copy(dev, buf);
1620 
1621 	return (BUS_PROBE_DEFAULT);
1622 }
1623 
1624 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1625     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1626     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS | \
1627     IFCAP_HWRXTSTMP | IFCAP_NOMAP)
1628 #define T4_CAP_ENABLE (T4_CAP)
1629 
1630 static int
1631 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1632 {
1633 	struct ifnet *ifp;
1634 	struct sbuf *sb;
1635 
1636 	vi->xact_addr_filt = -1;
1637 	callout_init(&vi->tick, 1);
1638 
1639 	/* Allocate an ifnet and set it up */
1640 	ifp = if_alloc_dev(IFT_ETHER, dev);
1641 	if (ifp == NULL) {
1642 		device_printf(dev, "Cannot allocate ifnet\n");
1643 		return (ENOMEM);
1644 	}
1645 	vi->ifp = ifp;
1646 	ifp->if_softc = vi;
1647 
1648 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1649 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1650 
1651 	ifp->if_init = cxgbe_init;
1652 	ifp->if_ioctl = cxgbe_ioctl;
1653 	ifp->if_transmit = cxgbe_transmit;
1654 	ifp->if_qflush = cxgbe_qflush;
1655 	ifp->if_get_counter = cxgbe_get_counter;
1656 #ifdef RATELIMIT
1657 	ifp->if_snd_tag_alloc = cxgbe_snd_tag_alloc;
1658 	ifp->if_snd_tag_modify = cxgbe_snd_tag_modify;
1659 	ifp->if_snd_tag_query = cxgbe_snd_tag_query;
1660 	ifp->if_snd_tag_free = cxgbe_snd_tag_free;
1661 	ifp->if_ratelimit_query = cxgbe_ratelimit_query;
1662 #endif
1663 
1664 	ifp->if_capabilities = T4_CAP;
1665 	ifp->if_capenable = T4_CAP_ENABLE;
1666 #ifdef TCP_OFFLOAD
1667 	if (vi->nofldrxq != 0)
1668 		ifp->if_capabilities |= IFCAP_TOE;
1669 #endif
1670 #ifdef RATELIMIT
1671 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0) {
1672 		ifp->if_capabilities |= IFCAP_TXRTLMT;
1673 		ifp->if_capenable |= IFCAP_TXRTLMT;
1674 	}
1675 #endif
1676 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1677 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1678 
1679 	ifp->if_hw_tsomax = IP_MAXPACKET;
1680 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_TSO;
1681 #ifdef RATELIMIT
1682 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0)
1683 		ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_EO_TSO;
1684 #endif
1685 	ifp->if_hw_tsomaxsegsize = 65536;
1686 
1687 	ether_ifattach(ifp, vi->hw_addr);
1688 #ifdef DEV_NETMAP
1689 	if (vi->nnmrxq != 0)
1690 		cxgbe_nm_attach(vi);
1691 #endif
1692 	sb = sbuf_new_auto();
1693 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1694 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1695 	switch (ifp->if_capabilities & (IFCAP_TOE | IFCAP_TXRTLMT)) {
1696 	case IFCAP_TOE:
1697 		sbuf_printf(sb, "; %d txq (TOE)", vi->nofldtxq);
1698 		break;
1699 	case IFCAP_TOE | IFCAP_TXRTLMT:
1700 		sbuf_printf(sb, "; %d txq (TOE/ETHOFLD)", vi->nofldtxq);
1701 		break;
1702 	case IFCAP_TXRTLMT:
1703 		sbuf_printf(sb, "; %d txq (ETHOFLD)", vi->nofldtxq);
1704 		break;
1705 	}
1706 #endif
1707 #ifdef TCP_OFFLOAD
1708 	if (ifp->if_capabilities & IFCAP_TOE)
1709 		sbuf_printf(sb, ", %d rxq (TOE)", vi->nofldrxq);
1710 #endif
1711 #ifdef DEV_NETMAP
1712 	if (ifp->if_capabilities & IFCAP_NETMAP)
1713 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1714 		    vi->nnmtxq, vi->nnmrxq);
1715 #endif
1716 	sbuf_finish(sb);
1717 	device_printf(dev, "%s\n", sbuf_data(sb));
1718 	sbuf_delete(sb);
1719 
1720 	vi_sysctls(vi);
1721 
1722 	return (0);
1723 }
1724 
1725 static int
1726 cxgbe_attach(device_t dev)
1727 {
1728 	struct port_info *pi = device_get_softc(dev);
1729 	struct adapter *sc = pi->adapter;
1730 	struct vi_info *vi;
1731 	int i, rc;
1732 
1733 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1734 
1735 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1736 	if (rc)
1737 		return (rc);
1738 
1739 	for_each_vi(pi, i, vi) {
1740 		if (i == 0)
1741 			continue;
1742 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1743 		if (vi->dev == NULL) {
1744 			device_printf(dev, "failed to add VI %d\n", i);
1745 			continue;
1746 		}
1747 		device_set_softc(vi->dev, vi);
1748 	}
1749 
1750 	cxgbe_sysctls(pi);
1751 
1752 	bus_generic_attach(dev);
1753 
1754 	return (0);
1755 }
1756 
1757 static void
1758 cxgbe_vi_detach(struct vi_info *vi)
1759 {
1760 	struct ifnet *ifp = vi->ifp;
1761 
1762 	ether_ifdetach(ifp);
1763 
1764 	/* Let detach proceed even if these fail. */
1765 #ifdef DEV_NETMAP
1766 	if (ifp->if_capabilities & IFCAP_NETMAP)
1767 		cxgbe_nm_detach(vi);
1768 #endif
1769 	cxgbe_uninit_synchronized(vi);
1770 	callout_drain(&vi->tick);
1771 	vi_full_uninit(vi);
1772 
1773 	if_free(vi->ifp);
1774 	vi->ifp = NULL;
1775 }
1776 
1777 static int
1778 cxgbe_detach(device_t dev)
1779 {
1780 	struct port_info *pi = device_get_softc(dev);
1781 	struct adapter *sc = pi->adapter;
1782 	int rc;
1783 
1784 	/* Detach the extra VIs first. */
1785 	rc = bus_generic_detach(dev);
1786 	if (rc)
1787 		return (rc);
1788 	device_delete_children(dev);
1789 
1790 	doom_vi(sc, &pi->vi[0]);
1791 
1792 	if (pi->flags & HAS_TRACEQ) {
1793 		sc->traceq = -1;	/* cloner should not create ifnet */
1794 		t4_tracer_port_detach(sc);
1795 	}
1796 
1797 	cxgbe_vi_detach(&pi->vi[0]);
1798 	callout_drain(&pi->tick);
1799 	ifmedia_removeall(&pi->media);
1800 
1801 	end_synchronized_op(sc, 0);
1802 
1803 	return (0);
1804 }
1805 
1806 static void
1807 cxgbe_init(void *arg)
1808 {
1809 	struct vi_info *vi = arg;
1810 	struct adapter *sc = vi->pi->adapter;
1811 
1812 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1813 		return;
1814 	cxgbe_init_synchronized(vi);
1815 	end_synchronized_op(sc, 0);
1816 }
1817 
1818 static int
1819 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1820 {
1821 	int rc = 0, mtu, flags;
1822 	struct vi_info *vi = ifp->if_softc;
1823 	struct port_info *pi = vi->pi;
1824 	struct adapter *sc = pi->adapter;
1825 	struct ifreq *ifr = (struct ifreq *)data;
1826 	uint32_t mask;
1827 
1828 	switch (cmd) {
1829 	case SIOCSIFMTU:
1830 		mtu = ifr->ifr_mtu;
1831 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1832 			return (EINVAL);
1833 
1834 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1835 		if (rc)
1836 			return (rc);
1837 		ifp->if_mtu = mtu;
1838 		if (vi->flags & VI_INIT_DONE) {
1839 			t4_update_fl_bufsize(ifp);
1840 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1841 				rc = update_mac_settings(ifp, XGMAC_MTU);
1842 		}
1843 		end_synchronized_op(sc, 0);
1844 		break;
1845 
1846 	case SIOCSIFFLAGS:
1847 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4flg");
1848 		if (rc)
1849 			return (rc);
1850 
1851 		if (ifp->if_flags & IFF_UP) {
1852 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1853 				flags = vi->if_flags;
1854 				if ((ifp->if_flags ^ flags) &
1855 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1856 					rc = update_mac_settings(ifp,
1857 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1858 				}
1859 			} else {
1860 				rc = cxgbe_init_synchronized(vi);
1861 			}
1862 			vi->if_flags = ifp->if_flags;
1863 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1864 			rc = cxgbe_uninit_synchronized(vi);
1865 		}
1866 		end_synchronized_op(sc, 0);
1867 		break;
1868 
1869 	case SIOCADDMULTI:
1870 	case SIOCDELMULTI:
1871 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4multi");
1872 		if (rc)
1873 			return (rc);
1874 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1875 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1876 		end_synchronized_op(sc, 0);
1877 		break;
1878 
1879 	case SIOCSIFCAP:
1880 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1881 		if (rc)
1882 			return (rc);
1883 
1884 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1885 		if (mask & IFCAP_TXCSUM) {
1886 			ifp->if_capenable ^= IFCAP_TXCSUM;
1887 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1888 
1889 			if (IFCAP_TSO4 & ifp->if_capenable &&
1890 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1891 				ifp->if_capenable &= ~IFCAP_TSO4;
1892 				if_printf(ifp,
1893 				    "tso4 disabled due to -txcsum.\n");
1894 			}
1895 		}
1896 		if (mask & IFCAP_TXCSUM_IPV6) {
1897 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1898 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1899 
1900 			if (IFCAP_TSO6 & ifp->if_capenable &&
1901 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1902 				ifp->if_capenable &= ~IFCAP_TSO6;
1903 				if_printf(ifp,
1904 				    "tso6 disabled due to -txcsum6.\n");
1905 			}
1906 		}
1907 		if (mask & IFCAP_RXCSUM)
1908 			ifp->if_capenable ^= IFCAP_RXCSUM;
1909 		if (mask & IFCAP_RXCSUM_IPV6)
1910 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1911 
1912 		/*
1913 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1914 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1915 		 * sending a TSO request our way, so it's sufficient to toggle
1916 		 * IFCAP_TSOx only.
1917 		 */
1918 		if (mask & IFCAP_TSO4) {
1919 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1920 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1921 				if_printf(ifp, "enable txcsum first.\n");
1922 				rc = EAGAIN;
1923 				goto fail;
1924 			}
1925 			ifp->if_capenable ^= IFCAP_TSO4;
1926 		}
1927 		if (mask & IFCAP_TSO6) {
1928 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1929 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1930 				if_printf(ifp, "enable txcsum6 first.\n");
1931 				rc = EAGAIN;
1932 				goto fail;
1933 			}
1934 			ifp->if_capenable ^= IFCAP_TSO6;
1935 		}
1936 		if (mask & IFCAP_LRO) {
1937 #if defined(INET) || defined(INET6)
1938 			int i;
1939 			struct sge_rxq *rxq;
1940 
1941 			ifp->if_capenable ^= IFCAP_LRO;
1942 			for_each_rxq(vi, i, rxq) {
1943 				if (ifp->if_capenable & IFCAP_LRO)
1944 					rxq->iq.flags |= IQ_LRO_ENABLED;
1945 				else
1946 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1947 			}
1948 #endif
1949 		}
1950 #ifdef TCP_OFFLOAD
1951 		if (mask & IFCAP_TOE) {
1952 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1953 
1954 			rc = toe_capability(vi, enable);
1955 			if (rc != 0)
1956 				goto fail;
1957 
1958 			ifp->if_capenable ^= mask;
1959 		}
1960 #endif
1961 		if (mask & IFCAP_VLAN_HWTAGGING) {
1962 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1963 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1964 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1965 		}
1966 		if (mask & IFCAP_VLAN_MTU) {
1967 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1968 
1969 			/* Need to find out how to disable auto-mtu-inflation */
1970 		}
1971 		if (mask & IFCAP_VLAN_HWTSO)
1972 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1973 		if (mask & IFCAP_VLAN_HWCSUM)
1974 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1975 #ifdef RATELIMIT
1976 		if (mask & IFCAP_TXRTLMT)
1977 			ifp->if_capenable ^= IFCAP_TXRTLMT;
1978 #endif
1979 		if (mask & IFCAP_HWRXTSTMP) {
1980 			int i;
1981 			struct sge_rxq *rxq;
1982 
1983 			ifp->if_capenable ^= IFCAP_HWRXTSTMP;
1984 			for_each_rxq(vi, i, rxq) {
1985 				if (ifp->if_capenable & IFCAP_HWRXTSTMP)
1986 					rxq->iq.flags |= IQ_RX_TIMESTAMP;
1987 				else
1988 					rxq->iq.flags &= ~IQ_RX_TIMESTAMP;
1989 			}
1990 		}
1991 		if (mask & IFCAP_NOMAP)
1992 			ifp->if_capenable ^= IFCAP_NOMAP;
1993 
1994 #ifdef VLAN_CAPABILITIES
1995 		VLAN_CAPABILITIES(ifp);
1996 #endif
1997 fail:
1998 		end_synchronized_op(sc, 0);
1999 		break;
2000 
2001 	case SIOCSIFMEDIA:
2002 	case SIOCGIFMEDIA:
2003 	case SIOCGIFXMEDIA:
2004 		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
2005 		break;
2006 
2007 	case SIOCGI2C: {
2008 		struct ifi2creq i2c;
2009 
2010 		rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
2011 		if (rc != 0)
2012 			break;
2013 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
2014 			rc = EPERM;
2015 			break;
2016 		}
2017 		if (i2c.len > sizeof(i2c.data)) {
2018 			rc = EINVAL;
2019 			break;
2020 		}
2021 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
2022 		if (rc)
2023 			return (rc);
2024 		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
2025 		    i2c.offset, i2c.len, &i2c.data[0]);
2026 		end_synchronized_op(sc, 0);
2027 		if (rc == 0)
2028 			rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
2029 		break;
2030 	}
2031 
2032 	default:
2033 		rc = ether_ioctl(ifp, cmd, data);
2034 	}
2035 
2036 	return (rc);
2037 }
2038 
2039 static int
2040 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
2041 {
2042 	struct vi_info *vi = ifp->if_softc;
2043 	struct port_info *pi = vi->pi;
2044 	struct adapter *sc = pi->adapter;
2045 	struct sge_txq *txq;
2046 	void *items[1];
2047 	int rc;
2048 
2049 	M_ASSERTPKTHDR(m);
2050 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
2051 
2052 	if (__predict_false(pi->link_cfg.link_ok == false)) {
2053 		m_freem(m);
2054 		return (ENETDOWN);
2055 	}
2056 
2057 	rc = parse_pkt(sc, &m);
2058 	if (__predict_false(rc != 0)) {
2059 		MPASS(m == NULL);			/* was freed already */
2060 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
2061 		return (rc);
2062 	}
2063 #ifdef RATELIMIT
2064 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) {
2065 		MPASS(m->m_pkthdr.snd_tag->ifp == ifp);
2066 		return (ethofld_transmit(ifp, m));
2067 	}
2068 #endif
2069 
2070 	/* Select a txq. */
2071 	txq = &sc->sge.txq[vi->first_txq];
2072 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2073 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
2074 		    vi->rsrv_noflowq);
2075 
2076 	items[0] = m;
2077 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
2078 	if (__predict_false(rc != 0))
2079 		m_freem(m);
2080 
2081 	return (rc);
2082 }
2083 
2084 static void
2085 cxgbe_qflush(struct ifnet *ifp)
2086 {
2087 	struct vi_info *vi = ifp->if_softc;
2088 	struct sge_txq *txq;
2089 	int i;
2090 
2091 	/* queues do not exist if !VI_INIT_DONE. */
2092 	if (vi->flags & VI_INIT_DONE) {
2093 		for_each_txq(vi, i, txq) {
2094 			TXQ_LOCK(txq);
2095 			txq->eq.flags |= EQ_QFLUSH;
2096 			TXQ_UNLOCK(txq);
2097 			while (!mp_ring_is_idle(txq->r)) {
2098 				mp_ring_check_drainage(txq->r, 0);
2099 				pause("qflush", 1);
2100 			}
2101 			TXQ_LOCK(txq);
2102 			txq->eq.flags &= ~EQ_QFLUSH;
2103 			TXQ_UNLOCK(txq);
2104 		}
2105 	}
2106 	if_qflush(ifp);
2107 }
2108 
2109 static uint64_t
2110 vi_get_counter(struct ifnet *ifp, ift_counter c)
2111 {
2112 	struct vi_info *vi = ifp->if_softc;
2113 	struct fw_vi_stats_vf *s = &vi->stats;
2114 
2115 	vi_refresh_stats(vi->pi->adapter, vi);
2116 
2117 	switch (c) {
2118 	case IFCOUNTER_IPACKETS:
2119 		return (s->rx_bcast_frames + s->rx_mcast_frames +
2120 		    s->rx_ucast_frames);
2121 	case IFCOUNTER_IERRORS:
2122 		return (s->rx_err_frames);
2123 	case IFCOUNTER_OPACKETS:
2124 		return (s->tx_bcast_frames + s->tx_mcast_frames +
2125 		    s->tx_ucast_frames + s->tx_offload_frames);
2126 	case IFCOUNTER_OERRORS:
2127 		return (s->tx_drop_frames);
2128 	case IFCOUNTER_IBYTES:
2129 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
2130 		    s->rx_ucast_bytes);
2131 	case IFCOUNTER_OBYTES:
2132 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
2133 		    s->tx_ucast_bytes + s->tx_offload_bytes);
2134 	case IFCOUNTER_IMCASTS:
2135 		return (s->rx_mcast_frames);
2136 	case IFCOUNTER_OMCASTS:
2137 		return (s->tx_mcast_frames);
2138 	case IFCOUNTER_OQDROPS: {
2139 		uint64_t drops;
2140 
2141 		drops = 0;
2142 		if (vi->flags & VI_INIT_DONE) {
2143 			int i;
2144 			struct sge_txq *txq;
2145 
2146 			for_each_txq(vi, i, txq)
2147 				drops += counter_u64_fetch(txq->r->drops);
2148 		}
2149 
2150 		return (drops);
2151 
2152 	}
2153 
2154 	default:
2155 		return (if_get_counter_default(ifp, c));
2156 	}
2157 }
2158 
2159 uint64_t
2160 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
2161 {
2162 	struct vi_info *vi = ifp->if_softc;
2163 	struct port_info *pi = vi->pi;
2164 	struct adapter *sc = pi->adapter;
2165 	struct port_stats *s = &pi->stats;
2166 
2167 	if (pi->nvi > 1 || sc->flags & IS_VF)
2168 		return (vi_get_counter(ifp, c));
2169 
2170 	cxgbe_refresh_stats(sc, pi);
2171 
2172 	switch (c) {
2173 	case IFCOUNTER_IPACKETS:
2174 		return (s->rx_frames);
2175 
2176 	case IFCOUNTER_IERRORS:
2177 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
2178 		    s->rx_fcs_err + s->rx_len_err);
2179 
2180 	case IFCOUNTER_OPACKETS:
2181 		return (s->tx_frames);
2182 
2183 	case IFCOUNTER_OERRORS:
2184 		return (s->tx_error_frames);
2185 
2186 	case IFCOUNTER_IBYTES:
2187 		return (s->rx_octets);
2188 
2189 	case IFCOUNTER_OBYTES:
2190 		return (s->tx_octets);
2191 
2192 	case IFCOUNTER_IMCASTS:
2193 		return (s->rx_mcast_frames);
2194 
2195 	case IFCOUNTER_OMCASTS:
2196 		return (s->tx_mcast_frames);
2197 
2198 	case IFCOUNTER_IQDROPS:
2199 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
2200 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
2201 		    s->rx_trunc3 + pi->tnl_cong_drops);
2202 
2203 	case IFCOUNTER_OQDROPS: {
2204 		uint64_t drops;
2205 
2206 		drops = s->tx_drop;
2207 		if (vi->flags & VI_INIT_DONE) {
2208 			int i;
2209 			struct sge_txq *txq;
2210 
2211 			for_each_txq(vi, i, txq)
2212 				drops += counter_u64_fetch(txq->r->drops);
2213 		}
2214 
2215 		return (drops);
2216 
2217 	}
2218 
2219 	default:
2220 		return (if_get_counter_default(ifp, c));
2221 	}
2222 }
2223 
2224 /*
2225  * The kernel picks a media from the list we had provided but we still validate
2226  * the requeste.
2227  */
2228 int
2229 cxgbe_media_change(struct ifnet *ifp)
2230 {
2231 	struct vi_info *vi = ifp->if_softc;
2232 	struct port_info *pi = vi->pi;
2233 	struct ifmedia *ifm = &pi->media;
2234 	struct link_config *lc = &pi->link_cfg;
2235 	struct adapter *sc = pi->adapter;
2236 	int rc;
2237 
2238 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mec");
2239 	if (rc != 0)
2240 		return (rc);
2241 	PORT_LOCK(pi);
2242 	if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
2243 		/* ifconfig .. media autoselect */
2244 		if (!(lc->supported & FW_PORT_CAP32_ANEG)) {
2245 			rc = ENOTSUP; /* AN not supported by transceiver */
2246 			goto done;
2247 		}
2248 		lc->requested_aneg = AUTONEG_ENABLE;
2249 		lc->requested_speed = 0;
2250 		lc->requested_fc |= PAUSE_AUTONEG;
2251 	} else {
2252 		lc->requested_aneg = AUTONEG_DISABLE;
2253 		lc->requested_speed =
2254 		    ifmedia_baudrate(ifm->ifm_media) / 1000000;
2255 		lc->requested_fc = 0;
2256 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
2257 			lc->requested_fc |= PAUSE_RX;
2258 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
2259 			lc->requested_fc |= PAUSE_TX;
2260 	}
2261 	if (pi->up_vis > 0) {
2262 		fixup_link_config(pi);
2263 		rc = apply_link_config(pi);
2264 	}
2265 done:
2266 	PORT_UNLOCK(pi);
2267 	end_synchronized_op(sc, 0);
2268 	return (rc);
2269 }
2270 
2271 /*
2272  * Base media word (without ETHER, pause, link active, etc.) for the port at the
2273  * given speed.
2274  */
2275 static int
2276 port_mword(struct port_info *pi, uint32_t speed)
2277 {
2278 
2279 	MPASS(speed & M_FW_PORT_CAP32_SPEED);
2280 	MPASS(powerof2(speed));
2281 
2282 	switch(pi->port_type) {
2283 	case FW_PORT_TYPE_BT_SGMII:
2284 	case FW_PORT_TYPE_BT_XFI:
2285 	case FW_PORT_TYPE_BT_XAUI:
2286 		/* BaseT */
2287 		switch (speed) {
2288 		case FW_PORT_CAP32_SPEED_100M:
2289 			return (IFM_100_T);
2290 		case FW_PORT_CAP32_SPEED_1G:
2291 			return (IFM_1000_T);
2292 		case FW_PORT_CAP32_SPEED_10G:
2293 			return (IFM_10G_T);
2294 		}
2295 		break;
2296 	case FW_PORT_TYPE_KX4:
2297 		if (speed == FW_PORT_CAP32_SPEED_10G)
2298 			return (IFM_10G_KX4);
2299 		break;
2300 	case FW_PORT_TYPE_CX4:
2301 		if (speed == FW_PORT_CAP32_SPEED_10G)
2302 			return (IFM_10G_CX4);
2303 		break;
2304 	case FW_PORT_TYPE_KX:
2305 		if (speed == FW_PORT_CAP32_SPEED_1G)
2306 			return (IFM_1000_KX);
2307 		break;
2308 	case FW_PORT_TYPE_KR:
2309 	case FW_PORT_TYPE_BP_AP:
2310 	case FW_PORT_TYPE_BP4_AP:
2311 	case FW_PORT_TYPE_BP40_BA:
2312 	case FW_PORT_TYPE_KR4_100G:
2313 	case FW_PORT_TYPE_KR_SFP28:
2314 	case FW_PORT_TYPE_KR_XLAUI:
2315 		switch (speed) {
2316 		case FW_PORT_CAP32_SPEED_1G:
2317 			return (IFM_1000_KX);
2318 		case FW_PORT_CAP32_SPEED_10G:
2319 			return (IFM_10G_KR);
2320 		case FW_PORT_CAP32_SPEED_25G:
2321 			return (IFM_25G_KR);
2322 		case FW_PORT_CAP32_SPEED_40G:
2323 			return (IFM_40G_KR4);
2324 		case FW_PORT_CAP32_SPEED_50G:
2325 			return (IFM_50G_KR2);
2326 		case FW_PORT_CAP32_SPEED_100G:
2327 			return (IFM_100G_KR4);
2328 		}
2329 		break;
2330 	case FW_PORT_TYPE_FIBER_XFI:
2331 	case FW_PORT_TYPE_FIBER_XAUI:
2332 	case FW_PORT_TYPE_SFP:
2333 	case FW_PORT_TYPE_QSFP_10G:
2334 	case FW_PORT_TYPE_QSA:
2335 	case FW_PORT_TYPE_QSFP:
2336 	case FW_PORT_TYPE_CR4_QSFP:
2337 	case FW_PORT_TYPE_CR_QSFP:
2338 	case FW_PORT_TYPE_CR2_QSFP:
2339 	case FW_PORT_TYPE_SFP28:
2340 		/* Pluggable transceiver */
2341 		switch (pi->mod_type) {
2342 		case FW_PORT_MOD_TYPE_LR:
2343 			switch (speed) {
2344 			case FW_PORT_CAP32_SPEED_1G:
2345 				return (IFM_1000_LX);
2346 			case FW_PORT_CAP32_SPEED_10G:
2347 				return (IFM_10G_LR);
2348 			case FW_PORT_CAP32_SPEED_25G:
2349 				return (IFM_25G_LR);
2350 			case FW_PORT_CAP32_SPEED_40G:
2351 				return (IFM_40G_LR4);
2352 			case FW_PORT_CAP32_SPEED_50G:
2353 				return (IFM_50G_LR2);
2354 			case FW_PORT_CAP32_SPEED_100G:
2355 				return (IFM_100G_LR4);
2356 			}
2357 			break;
2358 		case FW_PORT_MOD_TYPE_SR:
2359 			switch (speed) {
2360 			case FW_PORT_CAP32_SPEED_1G:
2361 				return (IFM_1000_SX);
2362 			case FW_PORT_CAP32_SPEED_10G:
2363 				return (IFM_10G_SR);
2364 			case FW_PORT_CAP32_SPEED_25G:
2365 				return (IFM_25G_SR);
2366 			case FW_PORT_CAP32_SPEED_40G:
2367 				return (IFM_40G_SR4);
2368 			case FW_PORT_CAP32_SPEED_50G:
2369 				return (IFM_50G_SR2);
2370 			case FW_PORT_CAP32_SPEED_100G:
2371 				return (IFM_100G_SR4);
2372 			}
2373 			break;
2374 		case FW_PORT_MOD_TYPE_ER:
2375 			if (speed == FW_PORT_CAP32_SPEED_10G)
2376 				return (IFM_10G_ER);
2377 			break;
2378 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2379 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2380 			switch (speed) {
2381 			case FW_PORT_CAP32_SPEED_1G:
2382 				return (IFM_1000_CX);
2383 			case FW_PORT_CAP32_SPEED_10G:
2384 				return (IFM_10G_TWINAX);
2385 			case FW_PORT_CAP32_SPEED_25G:
2386 				return (IFM_25G_CR);
2387 			case FW_PORT_CAP32_SPEED_40G:
2388 				return (IFM_40G_CR4);
2389 			case FW_PORT_CAP32_SPEED_50G:
2390 				return (IFM_50G_CR2);
2391 			case FW_PORT_CAP32_SPEED_100G:
2392 				return (IFM_100G_CR4);
2393 			}
2394 			break;
2395 		case FW_PORT_MOD_TYPE_LRM:
2396 			if (speed == FW_PORT_CAP32_SPEED_10G)
2397 				return (IFM_10G_LRM);
2398 			break;
2399 		case FW_PORT_MOD_TYPE_NA:
2400 			MPASS(0);	/* Not pluggable? */
2401 			/* fall throough */
2402 		case FW_PORT_MOD_TYPE_ERROR:
2403 		case FW_PORT_MOD_TYPE_UNKNOWN:
2404 		case FW_PORT_MOD_TYPE_NOTSUPPORTED:
2405 			break;
2406 		case FW_PORT_MOD_TYPE_NONE:
2407 			return (IFM_NONE);
2408 		}
2409 		break;
2410 	case FW_PORT_TYPE_NONE:
2411 		return (IFM_NONE);
2412 	}
2413 
2414 	return (IFM_UNKNOWN);
2415 }
2416 
2417 void
2418 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2419 {
2420 	struct vi_info *vi = ifp->if_softc;
2421 	struct port_info *pi = vi->pi;
2422 	struct adapter *sc = pi->adapter;
2423 	struct link_config *lc = &pi->link_cfg;
2424 
2425 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4med") != 0)
2426 		return;
2427 	PORT_LOCK(pi);
2428 
2429 	if (pi->up_vis == 0) {
2430 		/*
2431 		 * If all the interfaces are administratively down the firmware
2432 		 * does not report transceiver changes.  Refresh port info here
2433 		 * so that ifconfig displays accurate ifmedia at all times.
2434 		 * This is the only reason we have a synchronized op in this
2435 		 * function.  Just PORT_LOCK would have been enough otherwise.
2436 		 */
2437 		t4_update_port_info(pi);
2438 		build_medialist(pi);
2439 	}
2440 
2441 	/* ifm_status */
2442 	ifmr->ifm_status = IFM_AVALID;
2443 	if (lc->link_ok == false)
2444 		goto done;
2445 	ifmr->ifm_status |= IFM_ACTIVE;
2446 
2447 	/* ifm_active */
2448 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2449 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2450 	if (lc->fc & PAUSE_RX)
2451 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2452 	if (lc->fc & PAUSE_TX)
2453 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2454 	ifmr->ifm_active |= port_mword(pi, speed_to_fwcap(lc->speed));
2455 done:
2456 	PORT_UNLOCK(pi);
2457 	end_synchronized_op(sc, 0);
2458 }
2459 
2460 static int
2461 vcxgbe_probe(device_t dev)
2462 {
2463 	char buf[128];
2464 	struct vi_info *vi = device_get_softc(dev);
2465 
2466 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2467 	    vi - vi->pi->vi);
2468 	device_set_desc_copy(dev, buf);
2469 
2470 	return (BUS_PROBE_DEFAULT);
2471 }
2472 
2473 static int
2474 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2475 {
2476 	int func, index, rc;
2477 	uint32_t param, val;
2478 
2479 	ASSERT_SYNCHRONIZED_OP(sc);
2480 
2481 	index = vi - pi->vi;
2482 	MPASS(index > 0);	/* This function deals with _extra_ VIs only */
2483 	KASSERT(index < nitems(vi_mac_funcs),
2484 	    ("%s: VI %s doesn't have a MAC func", __func__,
2485 	    device_get_nameunit(vi->dev)));
2486 	func = vi_mac_funcs[index];
2487 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2488 	    vi->hw_addr, &vi->rss_size, &vi->vfvld, &vi->vin, func, 0);
2489 	if (rc < 0) {
2490 		device_printf(vi->dev, "failed to allocate virtual interface %d"
2491 		    "for port %d: %d\n", index, pi->port_id, -rc);
2492 		return (-rc);
2493 	}
2494 	vi->viid = rc;
2495 
2496 	if (vi->rss_size == 1) {
2497 		/*
2498 		 * This VI didn't get a slice of the RSS table.  Reduce the
2499 		 * number of VIs being created (hw.cxgbe.num_vis) or modify the
2500 		 * configuration file (nvi, rssnvi for this PF) if this is a
2501 		 * problem.
2502 		 */
2503 		device_printf(vi->dev, "RSS table not available.\n");
2504 		vi->rss_base = 0xffff;
2505 
2506 		return (0);
2507 	}
2508 
2509 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2510 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2511 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2512 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2513 	if (rc)
2514 		vi->rss_base = 0xffff;
2515 	else {
2516 		MPASS((val >> 16) == vi->rss_size);
2517 		vi->rss_base = val & 0xffff;
2518 	}
2519 
2520 	return (0);
2521 }
2522 
2523 static int
2524 vcxgbe_attach(device_t dev)
2525 {
2526 	struct vi_info *vi;
2527 	struct port_info *pi;
2528 	struct adapter *sc;
2529 	int rc;
2530 
2531 	vi = device_get_softc(dev);
2532 	pi = vi->pi;
2533 	sc = pi->adapter;
2534 
2535 	rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2536 	if (rc)
2537 		return (rc);
2538 	rc = alloc_extra_vi(sc, pi, vi);
2539 	end_synchronized_op(sc, 0);
2540 	if (rc)
2541 		return (rc);
2542 
2543 	rc = cxgbe_vi_attach(dev, vi);
2544 	if (rc) {
2545 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2546 		return (rc);
2547 	}
2548 	return (0);
2549 }
2550 
2551 static int
2552 vcxgbe_detach(device_t dev)
2553 {
2554 	struct vi_info *vi;
2555 	struct adapter *sc;
2556 
2557 	vi = device_get_softc(dev);
2558 	sc = vi->pi->adapter;
2559 
2560 	doom_vi(sc, vi);
2561 
2562 	cxgbe_vi_detach(vi);
2563 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2564 
2565 	end_synchronized_op(sc, 0);
2566 
2567 	return (0);
2568 }
2569 
2570 static struct callout fatal_callout;
2571 
2572 static void
2573 delayed_panic(void *arg)
2574 {
2575 	struct adapter *sc = arg;
2576 
2577 	panic("%s: panic on fatal error", device_get_nameunit(sc->dev));
2578 }
2579 
2580 void
2581 t4_fatal_err(struct adapter *sc, bool fw_error)
2582 {
2583 
2584 	t4_shutdown_adapter(sc);
2585 	log(LOG_ALERT, "%s: encountered fatal error, adapter stopped.\n",
2586 	    device_get_nameunit(sc->dev));
2587 	if (fw_error) {
2588 		ASSERT_SYNCHRONIZED_OP(sc);
2589 		sc->flags |= ADAP_ERR;
2590 	} else {
2591 		ADAPTER_LOCK(sc);
2592 		sc->flags |= ADAP_ERR;
2593 		ADAPTER_UNLOCK(sc);
2594 	}
2595 
2596 	if (t4_panic_on_fatal_err) {
2597 		log(LOG_ALERT, "%s: panic on fatal error after 30s",
2598 		    device_get_nameunit(sc->dev));
2599 		callout_reset(&fatal_callout, hz * 30, delayed_panic, sc);
2600 	}
2601 }
2602 
2603 void
2604 t4_add_adapter(struct adapter *sc)
2605 {
2606 	sx_xlock(&t4_list_lock);
2607 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2608 	sx_xunlock(&t4_list_lock);
2609 }
2610 
2611 int
2612 t4_map_bars_0_and_4(struct adapter *sc)
2613 {
2614 	sc->regs_rid = PCIR_BAR(0);
2615 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2616 	    &sc->regs_rid, RF_ACTIVE);
2617 	if (sc->regs_res == NULL) {
2618 		device_printf(sc->dev, "cannot map registers.\n");
2619 		return (ENXIO);
2620 	}
2621 	sc->bt = rman_get_bustag(sc->regs_res);
2622 	sc->bh = rman_get_bushandle(sc->regs_res);
2623 	sc->mmio_len = rman_get_size(sc->regs_res);
2624 	setbit(&sc->doorbells, DOORBELL_KDB);
2625 
2626 	sc->msix_rid = PCIR_BAR(4);
2627 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2628 	    &sc->msix_rid, RF_ACTIVE);
2629 	if (sc->msix_res == NULL) {
2630 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2631 		return (ENXIO);
2632 	}
2633 
2634 	return (0);
2635 }
2636 
2637 int
2638 t4_map_bar_2(struct adapter *sc)
2639 {
2640 
2641 	/*
2642 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2643 	 * to map it if RDMA is disabled.
2644 	 */
2645 	if (is_t4(sc) && sc->rdmacaps == 0)
2646 		return (0);
2647 
2648 	sc->udbs_rid = PCIR_BAR(2);
2649 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2650 	    &sc->udbs_rid, RF_ACTIVE);
2651 	if (sc->udbs_res == NULL) {
2652 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2653 		return (ENXIO);
2654 	}
2655 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2656 
2657 	if (chip_id(sc) >= CHELSIO_T5) {
2658 		setbit(&sc->doorbells, DOORBELL_UDB);
2659 #if defined(__i386__) || defined(__amd64__)
2660 		if (t5_write_combine) {
2661 			int rc, mode;
2662 
2663 			/*
2664 			 * Enable write combining on BAR2.  This is the
2665 			 * userspace doorbell BAR and is split into 128B
2666 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2667 			 * with an egress queue.  The first 64B has the doorbell
2668 			 * and the second 64B can be used to submit a tx work
2669 			 * request with an implicit doorbell.
2670 			 */
2671 
2672 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2673 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2674 			if (rc == 0) {
2675 				clrbit(&sc->doorbells, DOORBELL_UDB);
2676 				setbit(&sc->doorbells, DOORBELL_WCWR);
2677 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2678 			} else {
2679 				device_printf(sc->dev,
2680 				    "couldn't enable write combining: %d\n",
2681 				    rc);
2682 			}
2683 
2684 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2685 			t4_write_reg(sc, A_SGE_STAT_CFG,
2686 			    V_STATSOURCE_T5(7) | mode);
2687 		}
2688 #endif
2689 	}
2690 	sc->iwt.wc_en = isset(&sc->doorbells, DOORBELL_UDBWC) ? 1 : 0;
2691 
2692 	return (0);
2693 }
2694 
2695 struct memwin_init {
2696 	uint32_t base;
2697 	uint32_t aperture;
2698 };
2699 
2700 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2701 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2702 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2703 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2704 };
2705 
2706 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2707 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2708 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2709 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2710 };
2711 
2712 static void
2713 setup_memwin(struct adapter *sc)
2714 {
2715 	const struct memwin_init *mw_init;
2716 	struct memwin *mw;
2717 	int i;
2718 	uint32_t bar0;
2719 
2720 	if (is_t4(sc)) {
2721 		/*
2722 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2723 		 * mechanism.  Works from within PCI passthrough environments
2724 		 * too, where rman_get_start() can return a different value.  We
2725 		 * need to program the T4 memory window decoders with the actual
2726 		 * addresses that will be coming across the PCIe link.
2727 		 */
2728 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2729 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2730 
2731 		mw_init = &t4_memwin[0];
2732 	} else {
2733 		/* T5+ use the relative offset inside the PCIe BAR */
2734 		bar0 = 0;
2735 
2736 		mw_init = &t5_memwin[0];
2737 	}
2738 
2739 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2740 		rw_init(&mw->mw_lock, "memory window access");
2741 		mw->mw_base = mw_init->base;
2742 		mw->mw_aperture = mw_init->aperture;
2743 		mw->mw_curpos = 0;
2744 		t4_write_reg(sc,
2745 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2746 		    (mw->mw_base + bar0) | V_BIR(0) |
2747 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2748 		rw_wlock(&mw->mw_lock);
2749 		position_memwin(sc, i, 0);
2750 		rw_wunlock(&mw->mw_lock);
2751 	}
2752 
2753 	/* flush */
2754 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2755 }
2756 
2757 /*
2758  * Positions the memory window at the given address in the card's address space.
2759  * There are some alignment requirements and the actual position may be at an
2760  * address prior to the requested address.  mw->mw_curpos always has the actual
2761  * position of the window.
2762  */
2763 static void
2764 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2765 {
2766 	struct memwin *mw;
2767 	uint32_t pf;
2768 	uint32_t reg;
2769 
2770 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2771 	mw = &sc->memwin[idx];
2772 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2773 
2774 	if (is_t4(sc)) {
2775 		pf = 0;
2776 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2777 	} else {
2778 		pf = V_PFNUM(sc->pf);
2779 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2780 	}
2781 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2782 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2783 	t4_read_reg(sc, reg);	/* flush */
2784 }
2785 
2786 int
2787 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2788     int len, int rw)
2789 {
2790 	struct memwin *mw;
2791 	uint32_t mw_end, v;
2792 
2793 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2794 
2795 	/* Memory can only be accessed in naturally aligned 4 byte units */
2796 	if (addr & 3 || len & 3 || len <= 0)
2797 		return (EINVAL);
2798 
2799 	mw = &sc->memwin[idx];
2800 	while (len > 0) {
2801 		rw_rlock(&mw->mw_lock);
2802 		mw_end = mw->mw_curpos + mw->mw_aperture;
2803 		if (addr >= mw_end || addr < mw->mw_curpos) {
2804 			/* Will need to reposition the window */
2805 			if (!rw_try_upgrade(&mw->mw_lock)) {
2806 				rw_runlock(&mw->mw_lock);
2807 				rw_wlock(&mw->mw_lock);
2808 			}
2809 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2810 			position_memwin(sc, idx, addr);
2811 			rw_downgrade(&mw->mw_lock);
2812 			mw_end = mw->mw_curpos + mw->mw_aperture;
2813 		}
2814 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2815 		while (addr < mw_end && len > 0) {
2816 			if (rw == 0) {
2817 				v = t4_read_reg(sc, mw->mw_base + addr -
2818 				    mw->mw_curpos);
2819 				*val++ = le32toh(v);
2820 			} else {
2821 				v = *val++;
2822 				t4_write_reg(sc, mw->mw_base + addr -
2823 				    mw->mw_curpos, htole32(v));
2824 			}
2825 			addr += 4;
2826 			len -= 4;
2827 		}
2828 		rw_runlock(&mw->mw_lock);
2829 	}
2830 
2831 	return (0);
2832 }
2833 
2834 int
2835 alloc_atid_tab(struct tid_info *t, int flags)
2836 {
2837 	int i;
2838 
2839 	MPASS(t->natids > 0);
2840 	MPASS(t->atid_tab == NULL);
2841 
2842 	t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
2843 	    M_ZERO | flags);
2844 	if (t->atid_tab == NULL)
2845 		return (ENOMEM);
2846 	mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
2847 	t->afree = t->atid_tab;
2848 	t->atids_in_use = 0;
2849 	for (i = 1; i < t->natids; i++)
2850 		t->atid_tab[i - 1].next = &t->atid_tab[i];
2851 	t->atid_tab[t->natids - 1].next = NULL;
2852 
2853 	return (0);
2854 }
2855 
2856 void
2857 free_atid_tab(struct tid_info *t)
2858 {
2859 
2860 	KASSERT(t->atids_in_use == 0,
2861 	    ("%s: %d atids still in use.", __func__, t->atids_in_use));
2862 
2863 	if (mtx_initialized(&t->atid_lock))
2864 		mtx_destroy(&t->atid_lock);
2865 	free(t->atid_tab, M_CXGBE);
2866 	t->atid_tab = NULL;
2867 }
2868 
2869 int
2870 alloc_atid(struct adapter *sc, void *ctx)
2871 {
2872 	struct tid_info *t = &sc->tids;
2873 	int atid = -1;
2874 
2875 	mtx_lock(&t->atid_lock);
2876 	if (t->afree) {
2877 		union aopen_entry *p = t->afree;
2878 
2879 		atid = p - t->atid_tab;
2880 		MPASS(atid <= M_TID_TID);
2881 		t->afree = p->next;
2882 		p->data = ctx;
2883 		t->atids_in_use++;
2884 	}
2885 	mtx_unlock(&t->atid_lock);
2886 	return (atid);
2887 }
2888 
2889 void *
2890 lookup_atid(struct adapter *sc, int atid)
2891 {
2892 	struct tid_info *t = &sc->tids;
2893 
2894 	return (t->atid_tab[atid].data);
2895 }
2896 
2897 void
2898 free_atid(struct adapter *sc, int atid)
2899 {
2900 	struct tid_info *t = &sc->tids;
2901 	union aopen_entry *p = &t->atid_tab[atid];
2902 
2903 	mtx_lock(&t->atid_lock);
2904 	p->next = t->afree;
2905 	t->afree = p;
2906 	t->atids_in_use--;
2907 	mtx_unlock(&t->atid_lock);
2908 }
2909 
2910 static void
2911 queue_tid_release(struct adapter *sc, int tid)
2912 {
2913 
2914 	CXGBE_UNIMPLEMENTED("deferred tid release");
2915 }
2916 
2917 void
2918 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
2919 {
2920 	struct wrqe *wr;
2921 	struct cpl_tid_release *req;
2922 
2923 	wr = alloc_wrqe(sizeof(*req), ctrlq);
2924 	if (wr == NULL) {
2925 		queue_tid_release(sc, tid);	/* defer */
2926 		return;
2927 	}
2928 	req = wrtod(wr);
2929 
2930 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
2931 
2932 	t4_wrq_tx(sc, wr);
2933 }
2934 
2935 static int
2936 t4_range_cmp(const void *a, const void *b)
2937 {
2938 	return ((const struct t4_range *)a)->start -
2939 	       ((const struct t4_range *)b)->start;
2940 }
2941 
2942 /*
2943  * Verify that the memory range specified by the addr/len pair is valid within
2944  * the card's address space.
2945  */
2946 static int
2947 validate_mem_range(struct adapter *sc, uint32_t addr, uint32_t len)
2948 {
2949 	struct t4_range mem_ranges[4], *r, *next;
2950 	uint32_t em, addr_len;
2951 	int i, n, remaining;
2952 
2953 	/* Memory can only be accessed in naturally aligned 4 byte units */
2954 	if (addr & 3 || len & 3 || len == 0)
2955 		return (EINVAL);
2956 
2957 	/* Enabled memories */
2958 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2959 
2960 	r = &mem_ranges[0];
2961 	n = 0;
2962 	bzero(r, sizeof(mem_ranges));
2963 	if (em & F_EDRAM0_ENABLE) {
2964 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2965 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2966 		if (r->size > 0) {
2967 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2968 			if (addr >= r->start &&
2969 			    addr + len <= r->start + r->size)
2970 				return (0);
2971 			r++;
2972 			n++;
2973 		}
2974 	}
2975 	if (em & F_EDRAM1_ENABLE) {
2976 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2977 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2978 		if (r->size > 0) {
2979 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2980 			if (addr >= r->start &&
2981 			    addr + len <= r->start + r->size)
2982 				return (0);
2983 			r++;
2984 			n++;
2985 		}
2986 	}
2987 	if (em & F_EXT_MEM_ENABLE) {
2988 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2989 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2990 		if (r->size > 0) {
2991 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2992 			if (addr >= r->start &&
2993 			    addr + len <= r->start + r->size)
2994 				return (0);
2995 			r++;
2996 			n++;
2997 		}
2998 	}
2999 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
3000 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
3001 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
3002 		if (r->size > 0) {
3003 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
3004 			if (addr >= r->start &&
3005 			    addr + len <= r->start + r->size)
3006 				return (0);
3007 			r++;
3008 			n++;
3009 		}
3010 	}
3011 	MPASS(n <= nitems(mem_ranges));
3012 
3013 	if (n > 1) {
3014 		/* Sort and merge the ranges. */
3015 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
3016 
3017 		/* Start from index 0 and examine the next n - 1 entries. */
3018 		r = &mem_ranges[0];
3019 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
3020 
3021 			MPASS(r->size > 0);	/* r is a valid entry. */
3022 			next = r + 1;
3023 			MPASS(next->size > 0);	/* and so is the next one. */
3024 
3025 			while (r->start + r->size >= next->start) {
3026 				/* Merge the next one into the current entry. */
3027 				r->size = max(r->start + r->size,
3028 				    next->start + next->size) - r->start;
3029 				n--;	/* One fewer entry in total. */
3030 				if (--remaining == 0)
3031 					goto done;	/* short circuit */
3032 				next++;
3033 			}
3034 			if (next != r + 1) {
3035 				/*
3036 				 * Some entries were merged into r and next
3037 				 * points to the first valid entry that couldn't
3038 				 * be merged.
3039 				 */
3040 				MPASS(next->size > 0);	/* must be valid */
3041 				memcpy(r + 1, next, remaining * sizeof(*r));
3042 #ifdef INVARIANTS
3043 				/*
3044 				 * This so that the foo->size assertion in the
3045 				 * next iteration of the loop do the right
3046 				 * thing for entries that were pulled up and are
3047 				 * no longer valid.
3048 				 */
3049 				MPASS(n < nitems(mem_ranges));
3050 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
3051 				    sizeof(struct t4_range));
3052 #endif
3053 			}
3054 		}
3055 done:
3056 		/* Done merging the ranges. */
3057 		MPASS(n > 0);
3058 		r = &mem_ranges[0];
3059 		for (i = 0; i < n; i++, r++) {
3060 			if (addr >= r->start &&
3061 			    addr + len <= r->start + r->size)
3062 				return (0);
3063 		}
3064 	}
3065 
3066 	return (EFAULT);
3067 }
3068 
3069 static int
3070 fwmtype_to_hwmtype(int mtype)
3071 {
3072 
3073 	switch (mtype) {
3074 	case FW_MEMTYPE_EDC0:
3075 		return (MEM_EDC0);
3076 	case FW_MEMTYPE_EDC1:
3077 		return (MEM_EDC1);
3078 	case FW_MEMTYPE_EXTMEM:
3079 		return (MEM_MC0);
3080 	case FW_MEMTYPE_EXTMEM1:
3081 		return (MEM_MC1);
3082 	default:
3083 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
3084 	}
3085 }
3086 
3087 /*
3088  * Verify that the memory range specified by the memtype/offset/len pair is
3089  * valid and lies entirely within the memtype specified.  The global address of
3090  * the start of the range is returned in addr.
3091  */
3092 static int
3093 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, uint32_t len,
3094     uint32_t *addr)
3095 {
3096 	uint32_t em, addr_len, maddr;
3097 
3098 	/* Memory can only be accessed in naturally aligned 4 byte units */
3099 	if (off & 3 || len & 3 || len == 0)
3100 		return (EINVAL);
3101 
3102 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
3103 	switch (fwmtype_to_hwmtype(mtype)) {
3104 	case MEM_EDC0:
3105 		if (!(em & F_EDRAM0_ENABLE))
3106 			return (EINVAL);
3107 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
3108 		maddr = G_EDRAM0_BASE(addr_len) << 20;
3109 		break;
3110 	case MEM_EDC1:
3111 		if (!(em & F_EDRAM1_ENABLE))
3112 			return (EINVAL);
3113 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
3114 		maddr = G_EDRAM1_BASE(addr_len) << 20;
3115 		break;
3116 	case MEM_MC:
3117 		if (!(em & F_EXT_MEM_ENABLE))
3118 			return (EINVAL);
3119 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
3120 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
3121 		break;
3122 	case MEM_MC1:
3123 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
3124 			return (EINVAL);
3125 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
3126 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
3127 		break;
3128 	default:
3129 		return (EINVAL);
3130 	}
3131 
3132 	*addr = maddr + off;	/* global address */
3133 	return (validate_mem_range(sc, *addr, len));
3134 }
3135 
3136 static int
3137 fixup_devlog_params(struct adapter *sc)
3138 {
3139 	struct devlog_params *dparams = &sc->params.devlog;
3140 	int rc;
3141 
3142 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
3143 	    dparams->size, &dparams->addr);
3144 
3145 	return (rc);
3146 }
3147 
3148 static void
3149 update_nirq(struct intrs_and_queues *iaq, int nports)
3150 {
3151 	int extra = T4_EXTRA_INTR;
3152 
3153 	iaq->nirq = extra;
3154 	iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
3155 	iaq->nirq += nports * (iaq->num_vis - 1) *
3156 	    max(iaq->nrxq_vi, iaq->nnmrxq_vi);
3157 	iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
3158 }
3159 
3160 /*
3161  * Adjust requirements to fit the number of interrupts available.
3162  */
3163 static void
3164 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
3165     int navail)
3166 {
3167 	int old_nirq;
3168 	const int nports = sc->params.nports;
3169 
3170 	MPASS(nports > 0);
3171 	MPASS(navail > 0);
3172 
3173 	bzero(iaq, sizeof(*iaq));
3174 	iaq->intr_type = itype;
3175 	iaq->num_vis = t4_num_vis;
3176 	iaq->ntxq = t4_ntxq;
3177 	iaq->ntxq_vi = t4_ntxq_vi;
3178 	iaq->nrxq = t4_nrxq;
3179 	iaq->nrxq_vi = t4_nrxq_vi;
3180 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
3181 	if (is_offload(sc) || is_ethoffload(sc)) {
3182 		iaq->nofldtxq = t4_nofldtxq;
3183 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
3184 	}
3185 #endif
3186 #ifdef TCP_OFFLOAD
3187 	if (is_offload(sc)) {
3188 		iaq->nofldrxq = t4_nofldrxq;
3189 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
3190 	}
3191 #endif
3192 #ifdef DEV_NETMAP
3193 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
3194 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
3195 #endif
3196 
3197 	update_nirq(iaq, nports);
3198 	if (iaq->nirq <= navail &&
3199 	    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3200 		/*
3201 		 * This is the normal case -- there are enough interrupts for
3202 		 * everything.
3203 		 */
3204 		goto done;
3205 	}
3206 
3207 	/*
3208 	 * If extra VIs have been configured try reducing their count and see if
3209 	 * that works.
3210 	 */
3211 	while (iaq->num_vis > 1) {
3212 		iaq->num_vis--;
3213 		update_nirq(iaq, nports);
3214 		if (iaq->nirq <= navail &&
3215 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3216 			device_printf(sc->dev, "virtual interfaces per port "
3217 			    "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
3218 			    "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
3219 			    "itype %d, navail %u, nirq %d.\n",
3220 			    iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
3221 			    iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
3222 			    itype, navail, iaq->nirq);
3223 			goto done;
3224 		}
3225 	}
3226 
3227 	/*
3228 	 * Extra VIs will not be created.  Log a message if they were requested.
3229 	 */
3230 	MPASS(iaq->num_vis == 1);
3231 	iaq->ntxq_vi = iaq->nrxq_vi = 0;
3232 	iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
3233 	iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
3234 	if (iaq->num_vis != t4_num_vis) {
3235 		device_printf(sc->dev, "extra virtual interfaces disabled.  "
3236 		    "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
3237 		    "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
3238 		    iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
3239 		    iaq->nnmrxq_vi, itype, navail, iaq->nirq);
3240 	}
3241 
3242 	/*
3243 	 * Keep reducing the number of NIC rx queues to the next lower power of
3244 	 * 2 (for even RSS distribution) and halving the TOE rx queues and see
3245 	 * if that works.
3246 	 */
3247 	do {
3248 		if (iaq->nrxq > 1) {
3249 			do {
3250 				iaq->nrxq--;
3251 			} while (!powerof2(iaq->nrxq));
3252 		}
3253 		if (iaq->nofldrxq > 1)
3254 			iaq->nofldrxq >>= 1;
3255 
3256 		old_nirq = iaq->nirq;
3257 		update_nirq(iaq, nports);
3258 		if (iaq->nirq <= navail &&
3259 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3260 			device_printf(sc->dev, "running with reduced number of "
3261 			    "rx queues because of shortage of interrupts.  "
3262 			    "nrxq=%u, nofldrxq=%u.  "
3263 			    "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
3264 			    iaq->nofldrxq, itype, navail, iaq->nirq);
3265 			goto done;
3266 		}
3267 	} while (old_nirq != iaq->nirq);
3268 
3269 	/* One interrupt for everything.  Ugh. */
3270 	device_printf(sc->dev, "running with minimal number of queues.  "
3271 	    "itype %d, navail %u.\n", itype, navail);
3272 	iaq->nirq = 1;
3273 	MPASS(iaq->nrxq == 1);
3274 	iaq->ntxq = 1;
3275 	if (iaq->nofldrxq > 1)
3276 		iaq->nofldtxq = 1;
3277 done:
3278 	MPASS(iaq->num_vis > 0);
3279 	if (iaq->num_vis > 1) {
3280 		MPASS(iaq->nrxq_vi > 0);
3281 		MPASS(iaq->ntxq_vi > 0);
3282 	}
3283 	MPASS(iaq->nirq > 0);
3284 	MPASS(iaq->nrxq > 0);
3285 	MPASS(iaq->ntxq > 0);
3286 	if (itype == INTR_MSI) {
3287 		MPASS(powerof2(iaq->nirq));
3288 	}
3289 }
3290 
3291 static int
3292 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
3293 {
3294 	int rc, itype, navail, nalloc;
3295 
3296 	for (itype = INTR_MSIX; itype; itype >>= 1) {
3297 
3298 		if ((itype & t4_intr_types) == 0)
3299 			continue;	/* not allowed */
3300 
3301 		if (itype == INTR_MSIX)
3302 			navail = pci_msix_count(sc->dev);
3303 		else if (itype == INTR_MSI)
3304 			navail = pci_msi_count(sc->dev);
3305 		else
3306 			navail = 1;
3307 restart:
3308 		if (navail == 0)
3309 			continue;
3310 
3311 		calculate_iaq(sc, iaq, itype, navail);
3312 		nalloc = iaq->nirq;
3313 		rc = 0;
3314 		if (itype == INTR_MSIX)
3315 			rc = pci_alloc_msix(sc->dev, &nalloc);
3316 		else if (itype == INTR_MSI)
3317 			rc = pci_alloc_msi(sc->dev, &nalloc);
3318 
3319 		if (rc == 0 && nalloc > 0) {
3320 			if (nalloc == iaq->nirq)
3321 				return (0);
3322 
3323 			/*
3324 			 * Didn't get the number requested.  Use whatever number
3325 			 * the kernel is willing to allocate.
3326 			 */
3327 			device_printf(sc->dev, "fewer vectors than requested, "
3328 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
3329 			    itype, iaq->nirq, nalloc);
3330 			pci_release_msi(sc->dev);
3331 			navail = nalloc;
3332 			goto restart;
3333 		}
3334 
3335 		device_printf(sc->dev,
3336 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
3337 		    itype, rc, iaq->nirq, nalloc);
3338 	}
3339 
3340 	device_printf(sc->dev,
3341 	    "failed to find a usable interrupt type.  "
3342 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
3343 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
3344 
3345 	return (ENXIO);
3346 }
3347 
3348 #define FW_VERSION(chip) ( \
3349     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
3350     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
3351     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
3352     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
3353 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
3354 
3355 /* Just enough of fw_hdr to cover all version info. */
3356 struct fw_h {
3357 	__u8	ver;
3358 	__u8	chip;
3359 	__be16	len512;
3360 	__be32	fw_ver;
3361 	__be32	tp_microcode_ver;
3362 	__u8	intfver_nic;
3363 	__u8	intfver_vnic;
3364 	__u8	intfver_ofld;
3365 	__u8	intfver_ri;
3366 	__u8	intfver_iscsipdu;
3367 	__u8	intfver_iscsi;
3368 	__u8	intfver_fcoepdu;
3369 	__u8	intfver_fcoe;
3370 };
3371 /* Spot check a couple of fields. */
3372 CTASSERT(offsetof(struct fw_h, fw_ver) == offsetof(struct fw_hdr, fw_ver));
3373 CTASSERT(offsetof(struct fw_h, intfver_nic) == offsetof(struct fw_hdr, intfver_nic));
3374 CTASSERT(offsetof(struct fw_h, intfver_fcoe) == offsetof(struct fw_hdr, intfver_fcoe));
3375 
3376 struct fw_info {
3377 	uint8_t chip;
3378 	char *kld_name;
3379 	char *fw_mod_name;
3380 	struct fw_h fw_h;
3381 } fw_info[] = {
3382 	{
3383 		.chip = CHELSIO_T4,
3384 		.kld_name = "t4fw_cfg",
3385 		.fw_mod_name = "t4fw",
3386 		.fw_h = {
3387 			.chip = FW_HDR_CHIP_T4,
3388 			.fw_ver = htobe32(FW_VERSION(T4)),
3389 			.intfver_nic = FW_INTFVER(T4, NIC),
3390 			.intfver_vnic = FW_INTFVER(T4, VNIC),
3391 			.intfver_ofld = FW_INTFVER(T4, OFLD),
3392 			.intfver_ri = FW_INTFVER(T4, RI),
3393 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
3394 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
3395 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
3396 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
3397 		},
3398 	}, {
3399 		.chip = CHELSIO_T5,
3400 		.kld_name = "t5fw_cfg",
3401 		.fw_mod_name = "t5fw",
3402 		.fw_h = {
3403 			.chip = FW_HDR_CHIP_T5,
3404 			.fw_ver = htobe32(FW_VERSION(T5)),
3405 			.intfver_nic = FW_INTFVER(T5, NIC),
3406 			.intfver_vnic = FW_INTFVER(T5, VNIC),
3407 			.intfver_ofld = FW_INTFVER(T5, OFLD),
3408 			.intfver_ri = FW_INTFVER(T5, RI),
3409 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
3410 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
3411 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
3412 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
3413 		},
3414 	}, {
3415 		.chip = CHELSIO_T6,
3416 		.kld_name = "t6fw_cfg",
3417 		.fw_mod_name = "t6fw",
3418 		.fw_h = {
3419 			.chip = FW_HDR_CHIP_T6,
3420 			.fw_ver = htobe32(FW_VERSION(T6)),
3421 			.intfver_nic = FW_INTFVER(T6, NIC),
3422 			.intfver_vnic = FW_INTFVER(T6, VNIC),
3423 			.intfver_ofld = FW_INTFVER(T6, OFLD),
3424 			.intfver_ri = FW_INTFVER(T6, RI),
3425 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3426 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
3427 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3428 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
3429 		},
3430 	}
3431 };
3432 
3433 static struct fw_info *
3434 find_fw_info(int chip)
3435 {
3436 	int i;
3437 
3438 	for (i = 0; i < nitems(fw_info); i++) {
3439 		if (fw_info[i].chip == chip)
3440 			return (&fw_info[i]);
3441 	}
3442 	return (NULL);
3443 }
3444 
3445 /*
3446  * Is the given firmware API compatible with the one the driver was compiled
3447  * with?
3448  */
3449 static int
3450 fw_compatible(const struct fw_h *hdr1, const struct fw_h *hdr2)
3451 {
3452 
3453 	/* short circuit if it's the exact same firmware version */
3454 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
3455 		return (1);
3456 
3457 	/*
3458 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
3459 	 * features that are supported in the driver.
3460 	 */
3461 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
3462 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
3463 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
3464 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
3465 		return (1);
3466 #undef SAME_INTF
3467 
3468 	return (0);
3469 }
3470 
3471 static int
3472 load_fw_module(struct adapter *sc, const struct firmware **dcfg,
3473     const struct firmware **fw)
3474 {
3475 	struct fw_info *fw_info;
3476 
3477 	*dcfg = NULL;
3478 	if (fw != NULL)
3479 		*fw = NULL;
3480 
3481 	fw_info = find_fw_info(chip_id(sc));
3482 	if (fw_info == NULL) {
3483 		device_printf(sc->dev,
3484 		    "unable to look up firmware information for chip %d.\n",
3485 		    chip_id(sc));
3486 		return (EINVAL);
3487 	}
3488 
3489 	*dcfg = firmware_get(fw_info->kld_name);
3490 	if (*dcfg != NULL) {
3491 		if (fw != NULL)
3492 			*fw = firmware_get(fw_info->fw_mod_name);
3493 		return (0);
3494 	}
3495 
3496 	return (ENOENT);
3497 }
3498 
3499 static void
3500 unload_fw_module(struct adapter *sc, const struct firmware *dcfg,
3501     const struct firmware *fw)
3502 {
3503 
3504 	if (fw != NULL)
3505 		firmware_put(fw, FIRMWARE_UNLOAD);
3506 	if (dcfg != NULL)
3507 		firmware_put(dcfg, FIRMWARE_UNLOAD);
3508 }
3509 
3510 /*
3511  * Return values:
3512  * 0 means no firmware install attempted.
3513  * ERESTART means a firmware install was attempted and was successful.
3514  * +ve errno means a firmware install was attempted but failed.
3515  */
3516 static int
3517 install_kld_firmware(struct adapter *sc, struct fw_h *card_fw,
3518     const struct fw_h *drv_fw, const char *reason, int *already)
3519 {
3520 	const struct firmware *cfg, *fw;
3521 	const uint32_t c = be32toh(card_fw->fw_ver);
3522 	uint32_t d, k;
3523 	int rc, fw_install;
3524 	struct fw_h bundled_fw;
3525 	bool load_attempted;
3526 
3527 	cfg = fw = NULL;
3528 	load_attempted = false;
3529 	fw_install = t4_fw_install < 0 ? -t4_fw_install : t4_fw_install;
3530 
3531 	memcpy(&bundled_fw, drv_fw, sizeof(bundled_fw));
3532 	if (t4_fw_install < 0) {
3533 		rc = load_fw_module(sc, &cfg, &fw);
3534 		if (rc != 0 || fw == NULL) {
3535 			device_printf(sc->dev,
3536 			    "failed to load firmware module: %d. cfg %p, fw %p;"
3537 			    " will use compiled-in firmware version for"
3538 			    "hw.cxgbe.fw_install checks.\n",
3539 			    rc, cfg, fw);
3540 		} else {
3541 			memcpy(&bundled_fw, fw->data, sizeof(bundled_fw));
3542 		}
3543 		load_attempted = true;
3544 	}
3545 	d = be32toh(bundled_fw.fw_ver);
3546 
3547 	if (reason != NULL)
3548 		goto install;
3549 
3550 	if ((sc->flags & FW_OK) == 0) {
3551 
3552 		if (c == 0xffffffff) {
3553 			reason = "missing";
3554 			goto install;
3555 		}
3556 
3557 		rc = 0;
3558 		goto done;
3559 	}
3560 
3561 	if (!fw_compatible(card_fw, &bundled_fw)) {
3562 		reason = "incompatible or unusable";
3563 		goto install;
3564 	}
3565 
3566 	if (d > c) {
3567 		reason = "older than the version bundled with this driver";
3568 		goto install;
3569 	}
3570 
3571 	if (fw_install == 2 && d != c) {
3572 		reason = "different than the version bundled with this driver";
3573 		goto install;
3574 	}
3575 
3576 	/* No reason to do anything to the firmware already on the card. */
3577 	rc = 0;
3578 	goto done;
3579 
3580 install:
3581 	rc = 0;
3582 	if ((*already)++)
3583 		goto done;
3584 
3585 	if (fw_install == 0) {
3586 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3587 		    "but the driver is prohibited from installing a firmware "
3588 		    "on the card.\n",
3589 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3590 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3591 
3592 		goto done;
3593 	}
3594 
3595 	/*
3596 	 * We'll attempt to install a firmware.  Load the module first (if it
3597 	 * hasn't been loaded already).
3598 	 */
3599 	if (!load_attempted) {
3600 		rc = load_fw_module(sc, &cfg, &fw);
3601 		if (rc != 0 || fw == NULL) {
3602 			device_printf(sc->dev,
3603 			    "failed to load firmware module: %d. cfg %p, fw %p\n",
3604 			    rc, cfg, fw);
3605 			/* carry on */
3606 		}
3607 	}
3608 	if (fw == NULL) {
3609 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3610 		    "but the driver cannot take corrective action because it "
3611 		    "is unable to load the firmware module.\n",
3612 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3613 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3614 		rc = sc->flags & FW_OK ? 0 : ENOENT;
3615 		goto done;
3616 	}
3617 	k = be32toh(((const struct fw_hdr *)fw->data)->fw_ver);
3618 	if (k != d) {
3619 		MPASS(t4_fw_install > 0);
3620 		device_printf(sc->dev,
3621 		    "firmware in KLD (%u.%u.%u.%u) is not what the driver was "
3622 		    "expecting (%u.%u.%u.%u) and will not be used.\n",
3623 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3624 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k),
3625 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3626 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3627 		rc = sc->flags & FW_OK ? 0 : EINVAL;
3628 		goto done;
3629 	}
3630 
3631 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3632 	    "installing firmware %u.%u.%u.%u on card.\n",
3633 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3634 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3635 	    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3636 	    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3637 
3638 	rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3639 	if (rc != 0) {
3640 		device_printf(sc->dev, "failed to install firmware: %d\n", rc);
3641 	} else {
3642 		/* Installed successfully, update the cached header too. */
3643 		rc = ERESTART;
3644 		memcpy(card_fw, fw->data, sizeof(*card_fw));
3645 	}
3646 done:
3647 	unload_fw_module(sc, cfg, fw);
3648 
3649 	return (rc);
3650 }
3651 
3652 /*
3653  * Establish contact with the firmware and attempt to become the master driver.
3654  *
3655  * A firmware will be installed to the card if needed (if the driver is allowed
3656  * to do so).
3657  */
3658 static int
3659 contact_firmware(struct adapter *sc)
3660 {
3661 	int rc, already = 0;
3662 	enum dev_state state;
3663 	struct fw_info *fw_info;
3664 	struct fw_hdr *card_fw;		/* fw on the card */
3665 	const struct fw_h *drv_fw;
3666 
3667 	fw_info = find_fw_info(chip_id(sc));
3668 	if (fw_info == NULL) {
3669 		device_printf(sc->dev,
3670 		    "unable to look up firmware information for chip %d.\n",
3671 		    chip_id(sc));
3672 		return (EINVAL);
3673 	}
3674 	drv_fw = &fw_info->fw_h;
3675 
3676 	/* Read the header of the firmware on the card */
3677 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3678 restart:
3679 	rc = -t4_get_fw_hdr(sc, card_fw);
3680 	if (rc != 0) {
3681 		device_printf(sc->dev,
3682 		    "unable to read firmware header from card's flash: %d\n",
3683 		    rc);
3684 		goto done;
3685 	}
3686 
3687 	rc = install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw, NULL,
3688 	    &already);
3689 	if (rc == ERESTART)
3690 		goto restart;
3691 	if (rc != 0)
3692 		goto done;
3693 
3694 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3695 	if (rc < 0 || state == DEV_STATE_ERR) {
3696 		rc = -rc;
3697 		device_printf(sc->dev,
3698 		    "failed to connect to the firmware: %d, %d.  "
3699 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3700 #if 0
3701 		if (install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw,
3702 		    "not responding properly to HELLO", &already) == ERESTART)
3703 			goto restart;
3704 #endif
3705 		goto done;
3706 	}
3707 	MPASS(be32toh(card_fw->flags) & FW_HDR_FLAGS_RESET_HALT);
3708 	sc->flags |= FW_OK;	/* The firmware responded to the FW_HELLO. */
3709 
3710 	if (rc == sc->pf) {
3711 		sc->flags |= MASTER_PF;
3712 		rc = install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw,
3713 		    NULL, &already);
3714 		if (rc == ERESTART)
3715 			rc = 0;
3716 		else if (rc != 0)
3717 			goto done;
3718 	} else if (state == DEV_STATE_UNINIT) {
3719 		/*
3720 		 * We didn't get to be the master so we definitely won't be
3721 		 * configuring the chip.  It's a bug if someone else hasn't
3722 		 * configured it already.
3723 		 */
3724 		device_printf(sc->dev, "couldn't be master(%d), "
3725 		    "device not already initialized either(%d).  "
3726 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3727 		rc = EPROTO;
3728 		goto done;
3729 	} else {
3730 		/*
3731 		 * Some other PF is the master and has configured the chip.
3732 		 * This is allowed but untested.
3733 		 */
3734 		device_printf(sc->dev, "PF%d is master, device state %d.  "
3735 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3736 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", rc);
3737 		sc->cfcsum = 0;
3738 		rc = 0;
3739 	}
3740 done:
3741 	if (rc != 0 && sc->flags & FW_OK) {
3742 		t4_fw_bye(sc, sc->mbox);
3743 		sc->flags &= ~FW_OK;
3744 	}
3745 	free(card_fw, M_CXGBE);
3746 	return (rc);
3747 }
3748 
3749 static int
3750 copy_cfg_file_to_card(struct adapter *sc, char *cfg_file,
3751     uint32_t mtype, uint32_t moff)
3752 {
3753 	struct fw_info *fw_info;
3754 	const struct firmware *dcfg, *rcfg = NULL;
3755 	const uint32_t *cfdata;
3756 	uint32_t cflen, addr;
3757 	int rc;
3758 
3759 	load_fw_module(sc, &dcfg, NULL);
3760 
3761 	/* Card specific interpretation of "default". */
3762 	if (strncmp(cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3763 		if (pci_get_device(sc->dev) == 0x440a)
3764 			snprintf(cfg_file, sizeof(t4_cfg_file), UWIRE_CF);
3765 		if (is_fpga(sc))
3766 			snprintf(cfg_file, sizeof(t4_cfg_file), FPGA_CF);
3767 	}
3768 
3769 	if (strncmp(cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3770 		if (dcfg == NULL) {
3771 			device_printf(sc->dev,
3772 			    "KLD with default config is not available.\n");
3773 			rc = ENOENT;
3774 			goto done;
3775 		}
3776 		cfdata = dcfg->data;
3777 		cflen = dcfg->datasize & ~3;
3778 	} else {
3779 		char s[32];
3780 
3781 		fw_info = find_fw_info(chip_id(sc));
3782 		if (fw_info == NULL) {
3783 			device_printf(sc->dev,
3784 			    "unable to look up firmware information for chip %d.\n",
3785 			    chip_id(sc));
3786 			rc = EINVAL;
3787 			goto done;
3788 		}
3789 		snprintf(s, sizeof(s), "%s_%s", fw_info->kld_name, cfg_file);
3790 
3791 		rcfg = firmware_get(s);
3792 		if (rcfg == NULL) {
3793 			device_printf(sc->dev,
3794 			    "unable to load module \"%s\" for configuration "
3795 			    "profile \"%s\".\n", s, cfg_file);
3796 			rc = ENOENT;
3797 			goto done;
3798 		}
3799 		cfdata = rcfg->data;
3800 		cflen = rcfg->datasize & ~3;
3801 	}
3802 
3803 	if (cflen > FLASH_CFG_MAX_SIZE) {
3804 		device_printf(sc->dev,
3805 		    "config file too long (%d, max allowed is %d).\n",
3806 		    cflen, FLASH_CFG_MAX_SIZE);
3807 		rc = EINVAL;
3808 		goto done;
3809 	}
3810 
3811 	rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3812 	if (rc != 0) {
3813 		device_printf(sc->dev,
3814 		    "%s: addr (%d/0x%x) or len %d is not valid: %d.\n",
3815 		    __func__, mtype, moff, cflen, rc);
3816 		rc = EINVAL;
3817 		goto done;
3818 	}
3819 	write_via_memwin(sc, 2, addr, cfdata, cflen);
3820 done:
3821 	if (rcfg != NULL)
3822 		firmware_put(rcfg, FIRMWARE_UNLOAD);
3823 	unload_fw_module(sc, dcfg, NULL);
3824 	return (rc);
3825 }
3826 
3827 struct caps_allowed {
3828 	uint16_t nbmcaps;
3829 	uint16_t linkcaps;
3830 	uint16_t switchcaps;
3831 	uint16_t niccaps;
3832 	uint16_t toecaps;
3833 	uint16_t rdmacaps;
3834 	uint16_t cryptocaps;
3835 	uint16_t iscsicaps;
3836 	uint16_t fcoecaps;
3837 };
3838 
3839 #define FW_PARAM_DEV(param) \
3840 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3841 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3842 #define FW_PARAM_PFVF(param) \
3843 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3844 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3845 
3846 /*
3847  * Provide a configuration profile to the firmware and have it initialize the
3848  * chip accordingly.  This may involve uploading a configuration file to the
3849  * card.
3850  */
3851 static int
3852 apply_cfg_and_initialize(struct adapter *sc, char *cfg_file,
3853     const struct caps_allowed *caps_allowed)
3854 {
3855 	int rc;
3856 	struct fw_caps_config_cmd caps;
3857 	uint32_t mtype, moff, finicsum, cfcsum, param, val;
3858 
3859 	rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
3860 	if (rc != 0) {
3861 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3862 		return (rc);
3863 	}
3864 
3865 	bzero(&caps, sizeof(caps));
3866 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3867 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3868 	if (strncmp(cfg_file, BUILTIN_CF, sizeof(t4_cfg_file)) == 0) {
3869 		mtype = 0;
3870 		moff = 0;
3871 		caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3872 	} else if (strncmp(cfg_file, FLASH_CF, sizeof(t4_cfg_file)) == 0) {
3873 		mtype = FW_MEMTYPE_FLASH;
3874 		moff = t4_flash_cfg_addr(sc);
3875 		caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3876 		    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3877 		    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) |
3878 		    FW_LEN16(caps));
3879 	} else {
3880 		/*
3881 		 * Ask the firmware where it wants us to upload the config file.
3882 		 */
3883 		param = FW_PARAM_DEV(CF);
3884 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3885 		if (rc != 0) {
3886 			/* No support for config file?  Shouldn't happen. */
3887 			device_printf(sc->dev,
3888 			    "failed to query config file location: %d.\n", rc);
3889 			goto done;
3890 		}
3891 		mtype = G_FW_PARAMS_PARAM_Y(val);
3892 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3893 		caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3894 		    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3895 		    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) |
3896 		    FW_LEN16(caps));
3897 
3898 		rc = copy_cfg_file_to_card(sc, cfg_file, mtype, moff);
3899 		if (rc != 0) {
3900 			device_printf(sc->dev,
3901 			    "failed to upload config file to card: %d.\n", rc);
3902 			goto done;
3903 		}
3904 	}
3905 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3906 	if (rc != 0) {
3907 		device_printf(sc->dev, "failed to pre-process config file: %d "
3908 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3909 		goto done;
3910 	}
3911 
3912 	finicsum = be32toh(caps.finicsum);
3913 	cfcsum = be32toh(caps.cfcsum);	/* actual */
3914 	if (finicsum != cfcsum) {
3915 		device_printf(sc->dev,
3916 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3917 		    finicsum, cfcsum);
3918 	}
3919 	sc->cfcsum = cfcsum;
3920 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", cfg_file);
3921 
3922 	/*
3923 	 * Let the firmware know what features will (not) be used so it can tune
3924 	 * things accordingly.
3925 	 */
3926 #define LIMIT_CAPS(x) do { \
3927 	caps.x##caps &= htobe16(caps_allowed->x##caps); \
3928 } while (0)
3929 	LIMIT_CAPS(nbm);
3930 	LIMIT_CAPS(link);
3931 	LIMIT_CAPS(switch);
3932 	LIMIT_CAPS(nic);
3933 	LIMIT_CAPS(toe);
3934 	LIMIT_CAPS(rdma);
3935 	LIMIT_CAPS(crypto);
3936 	LIMIT_CAPS(iscsi);
3937 	LIMIT_CAPS(fcoe);
3938 #undef LIMIT_CAPS
3939 	if (caps.niccaps & htobe16(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
3940 		/*
3941 		 * TOE and hashfilters are mutually exclusive.  It is a config
3942 		 * file or firmware bug if both are reported as available.  Try
3943 		 * to cope with the situation in non-debug builds by disabling
3944 		 * TOE.
3945 		 */
3946 		MPASS(caps.toecaps == 0);
3947 
3948 		caps.toecaps = 0;
3949 		caps.rdmacaps = 0;
3950 		caps.iscsicaps = 0;
3951 	}
3952 
3953 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3954 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3955 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3956 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3957 	if (rc != 0) {
3958 		device_printf(sc->dev,
3959 		    "failed to process config file: %d.\n", rc);
3960 		goto done;
3961 	}
3962 
3963 	t4_tweak_chip_settings(sc);
3964 	set_params__pre_init(sc);
3965 
3966 	/* get basic stuff going */
3967 	rc = -t4_fw_initialize(sc, sc->mbox);
3968 	if (rc != 0) {
3969 		device_printf(sc->dev, "fw_initialize failed: %d.\n", rc);
3970 		goto done;
3971 	}
3972 done:
3973 	return (rc);
3974 }
3975 
3976 /*
3977  * Partition chip resources for use between various PFs, VFs, etc.
3978  */
3979 static int
3980 partition_resources(struct adapter *sc)
3981 {
3982 	char cfg_file[sizeof(t4_cfg_file)];
3983 	struct caps_allowed caps_allowed;
3984 	int rc;
3985 	bool fallback;
3986 
3987 	/* Only the master driver gets to configure the chip resources. */
3988 	MPASS(sc->flags & MASTER_PF);
3989 
3990 #define COPY_CAPS(x) do { \
3991 	caps_allowed.x##caps = t4_##x##caps_allowed; \
3992 } while (0)
3993 	bzero(&caps_allowed, sizeof(caps_allowed));
3994 	COPY_CAPS(nbm);
3995 	COPY_CAPS(link);
3996 	COPY_CAPS(switch);
3997 	COPY_CAPS(nic);
3998 	COPY_CAPS(toe);
3999 	COPY_CAPS(rdma);
4000 	COPY_CAPS(crypto);
4001 	COPY_CAPS(iscsi);
4002 	COPY_CAPS(fcoe);
4003 	fallback = sc->debug_flags & DF_DISABLE_CFG_RETRY ? false : true;
4004 	snprintf(cfg_file, sizeof(cfg_file), "%s", t4_cfg_file);
4005 retry:
4006 	rc = apply_cfg_and_initialize(sc, cfg_file, &caps_allowed);
4007 	if (rc != 0 && fallback) {
4008 		device_printf(sc->dev,
4009 		    "failed (%d) to configure card with \"%s\" profile, "
4010 		    "will fall back to a basic configuration and retry.\n",
4011 		    rc, cfg_file);
4012 		snprintf(cfg_file, sizeof(cfg_file), "%s", BUILTIN_CF);
4013 		bzero(&caps_allowed, sizeof(caps_allowed));
4014 		COPY_CAPS(switch);
4015 		caps_allowed.niccaps = FW_CAPS_CONFIG_NIC;
4016 		fallback = false;
4017 		goto retry;
4018 	}
4019 #undef COPY_CAPS
4020 	return (rc);
4021 }
4022 
4023 /*
4024  * Retrieve parameters that are needed (or nice to have) very early.
4025  */
4026 static int
4027 get_params__pre_init(struct adapter *sc)
4028 {
4029 	int rc;
4030 	uint32_t param[2], val[2];
4031 
4032 	t4_get_version_info(sc);
4033 
4034 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
4035 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
4036 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
4037 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
4038 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
4039 
4040 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
4041 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
4042 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
4043 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
4044 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
4045 
4046 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
4047 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
4048 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
4049 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
4050 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
4051 
4052 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
4053 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
4054 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
4055 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
4056 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
4057 
4058 	param[0] = FW_PARAM_DEV(PORTVEC);
4059 	param[1] = FW_PARAM_DEV(CCLK);
4060 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4061 	if (rc != 0) {
4062 		device_printf(sc->dev,
4063 		    "failed to query parameters (pre_init): %d.\n", rc);
4064 		return (rc);
4065 	}
4066 
4067 	sc->params.portvec = val[0];
4068 	sc->params.nports = bitcount32(val[0]);
4069 	sc->params.vpd.cclk = val[1];
4070 
4071 	/* Read device log parameters. */
4072 	rc = -t4_init_devlog_params(sc, 1);
4073 	if (rc == 0)
4074 		fixup_devlog_params(sc);
4075 	else {
4076 		device_printf(sc->dev,
4077 		    "failed to get devlog parameters: %d.\n", rc);
4078 		rc = 0;	/* devlog isn't critical for device operation */
4079 	}
4080 
4081 	return (rc);
4082 }
4083 
4084 /*
4085  * Any params that need to be set before FW_INITIALIZE.
4086  */
4087 static int
4088 set_params__pre_init(struct adapter *sc)
4089 {
4090 	int rc = 0;
4091 	uint32_t param, val;
4092 
4093 	if (chip_id(sc) >= CHELSIO_T6) {
4094 		param = FW_PARAM_DEV(HPFILTER_REGION_SUPPORT);
4095 		val = 1;
4096 		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4097 		/* firmwares < 1.20.1.0 do not have this param. */
4098 		if (rc == FW_EINVAL && sc->params.fw_vers <
4099 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4100 		    V_FW_HDR_FW_VER_MICRO(1) | V_FW_HDR_FW_VER_BUILD(0))) {
4101 			rc = 0;
4102 		}
4103 		if (rc != 0) {
4104 			device_printf(sc->dev,
4105 			    "failed to enable high priority filters :%d.\n",
4106 			    rc);
4107 		}
4108 	}
4109 
4110 	/* Enable opaque VIIDs with firmwares that support it. */
4111 	param = FW_PARAM_DEV(OPAQUE_VIID_SMT_EXTN);
4112 	val = 1;
4113 	rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4114 	if (rc == 0 && val == 1)
4115 		sc->params.viid_smt_extn_support = true;
4116 	else
4117 		sc->params.viid_smt_extn_support = false;
4118 
4119 	return (rc);
4120 }
4121 
4122 /*
4123  * Retrieve various parameters that are of interest to the driver.  The device
4124  * has been initialized by the firmware at this point.
4125  */
4126 static int
4127 get_params__post_init(struct adapter *sc)
4128 {
4129 	int rc;
4130 	uint32_t param[7], val[7];
4131 	struct fw_caps_config_cmd caps;
4132 
4133 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
4134 	param[1] = FW_PARAM_PFVF(EQ_START);
4135 	param[2] = FW_PARAM_PFVF(FILTER_START);
4136 	param[3] = FW_PARAM_PFVF(FILTER_END);
4137 	param[4] = FW_PARAM_PFVF(L2T_START);
4138 	param[5] = FW_PARAM_PFVF(L2T_END);
4139 	param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4140 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
4141 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
4142 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
4143 	if (rc != 0) {
4144 		device_printf(sc->dev,
4145 		    "failed to query parameters (post_init): %d.\n", rc);
4146 		return (rc);
4147 	}
4148 
4149 	sc->sge.iq_start = val[0];
4150 	sc->sge.eq_start = val[1];
4151 	if ((int)val[3] > (int)val[2]) {
4152 		sc->tids.ftid_base = val[2];
4153 		sc->tids.ftid_end = val[3];
4154 		sc->tids.nftids = val[3] - val[2] + 1;
4155 	}
4156 	sc->vres.l2t.start = val[4];
4157 	sc->vres.l2t.size = val[5] - val[4] + 1;
4158 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
4159 	    ("%s: L2 table size (%u) larger than expected (%u)",
4160 	    __func__, sc->vres.l2t.size, L2T_SIZE));
4161 	sc->params.core_vdd = val[6];
4162 
4163 	if (chip_id(sc) >= CHELSIO_T6) {
4164 
4165 		sc->tids.tid_base = t4_read_reg(sc,
4166 		    A_LE_DB_ACTIVE_TABLE_START_INDEX);
4167 
4168 		param[0] = FW_PARAM_PFVF(HPFILTER_START);
4169 		param[1] = FW_PARAM_PFVF(HPFILTER_END);
4170 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4171 		if (rc != 0) {
4172 			device_printf(sc->dev,
4173 			   "failed to query hpfilter parameters: %d.\n", rc);
4174 			return (rc);
4175 		}
4176 		if ((int)val[1] > (int)val[0]) {
4177 			sc->tids.hpftid_base = val[0];
4178 			sc->tids.hpftid_end = val[1];
4179 			sc->tids.nhpftids = val[1] - val[0] + 1;
4180 
4181 			/*
4182 			 * These should go off if the layout changes and the
4183 			 * driver needs to catch up.
4184 			 */
4185 			MPASS(sc->tids.hpftid_base == 0);
4186 			MPASS(sc->tids.tid_base == sc->tids.nhpftids);
4187 		}
4188 	}
4189 
4190 	/*
4191 	 * MPSBGMAP is queried separately because only recent firmwares support
4192 	 * it as a parameter and we don't want the compound query above to fail
4193 	 * on older firmwares.
4194 	 */
4195 	param[0] = FW_PARAM_DEV(MPSBGMAP);
4196 	val[0] = 0;
4197 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4198 	if (rc == 0)
4199 		sc->params.mps_bg_map = val[0];
4200 	else
4201 		sc->params.mps_bg_map = 0;
4202 
4203 	/*
4204 	 * Determine whether the firmware supports the filter2 work request.
4205 	 * This is queried separately for the same reason as MPSBGMAP above.
4206 	 */
4207 	param[0] = FW_PARAM_DEV(FILTER2_WR);
4208 	val[0] = 0;
4209 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4210 	if (rc == 0)
4211 		sc->params.filter2_wr_support = val[0] != 0;
4212 	else
4213 		sc->params.filter2_wr_support = 0;
4214 
4215 	/*
4216 	 * Find out whether we're allowed to use the ULPTX MEMWRITE DSGL.
4217 	 * This is queried separately for the same reason as other params above.
4218 	 */
4219 	param[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
4220 	val[0] = 0;
4221 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4222 	if (rc == 0)
4223 		sc->params.ulptx_memwrite_dsgl = val[0] != 0;
4224 	else
4225 		sc->params.ulptx_memwrite_dsgl = false;
4226 
4227 	/* get capabilites */
4228 	bzero(&caps, sizeof(caps));
4229 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4230 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
4231 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
4232 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
4233 	if (rc != 0) {
4234 		device_printf(sc->dev,
4235 		    "failed to get card capabilities: %d.\n", rc);
4236 		return (rc);
4237 	}
4238 
4239 #define READ_CAPS(x) do { \
4240 	sc->x = htobe16(caps.x); \
4241 } while (0)
4242 	READ_CAPS(nbmcaps);
4243 	READ_CAPS(linkcaps);
4244 	READ_CAPS(switchcaps);
4245 	READ_CAPS(niccaps);
4246 	READ_CAPS(toecaps);
4247 	READ_CAPS(rdmacaps);
4248 	READ_CAPS(cryptocaps);
4249 	READ_CAPS(iscsicaps);
4250 	READ_CAPS(fcoecaps);
4251 
4252 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_HASHFILTER) {
4253 		MPASS(chip_id(sc) > CHELSIO_T4);
4254 		MPASS(sc->toecaps == 0);
4255 		sc->toecaps = 0;
4256 
4257 		param[0] = FW_PARAM_DEV(NTID);
4258 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4259 		if (rc != 0) {
4260 			device_printf(sc->dev,
4261 			    "failed to query HASHFILTER parameters: %d.\n", rc);
4262 			return (rc);
4263 		}
4264 		sc->tids.ntids = val[0];
4265 		if (sc->params.fw_vers <
4266 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4267 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4268 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4269 			sc->tids.ntids -= sc->tids.nhpftids;
4270 		}
4271 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4272 		sc->params.hash_filter = 1;
4273 	}
4274 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
4275 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
4276 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
4277 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4278 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
4279 		if (rc != 0) {
4280 			device_printf(sc->dev,
4281 			    "failed to query NIC parameters: %d.\n", rc);
4282 			return (rc);
4283 		}
4284 		if ((int)val[1] > (int)val[0]) {
4285 			sc->tids.etid_base = val[0];
4286 			sc->tids.etid_end = val[1];
4287 			sc->tids.netids = val[1] - val[0] + 1;
4288 			sc->params.eo_wr_cred = val[2];
4289 			sc->params.ethoffload = 1;
4290 		}
4291 	}
4292 	if (sc->toecaps) {
4293 		/* query offload-related parameters */
4294 		param[0] = FW_PARAM_DEV(NTID);
4295 		param[1] = FW_PARAM_PFVF(SERVER_START);
4296 		param[2] = FW_PARAM_PFVF(SERVER_END);
4297 		param[3] = FW_PARAM_PFVF(TDDP_START);
4298 		param[4] = FW_PARAM_PFVF(TDDP_END);
4299 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4300 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4301 		if (rc != 0) {
4302 			device_printf(sc->dev,
4303 			    "failed to query TOE parameters: %d.\n", rc);
4304 			return (rc);
4305 		}
4306 		sc->tids.ntids = val[0];
4307 		if (sc->params.fw_vers <
4308 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4309 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4310 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4311 			sc->tids.ntids -= sc->tids.nhpftids;
4312 		}
4313 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4314 		if ((int)val[2] > (int)val[1]) {
4315 			sc->tids.stid_base = val[1];
4316 			sc->tids.nstids = val[2] - val[1] + 1;
4317 		}
4318 		sc->vres.ddp.start = val[3];
4319 		sc->vres.ddp.size = val[4] - val[3] + 1;
4320 		sc->params.ofldq_wr_cred = val[5];
4321 		sc->params.offload = 1;
4322 	} else {
4323 		/*
4324 		 * The firmware attempts memfree TOE configuration for -SO cards
4325 		 * and will report toecaps=0 if it runs out of resources (this
4326 		 * depends on the config file).  It may not report 0 for other
4327 		 * capabilities dependent on the TOE in this case.  Set them to
4328 		 * 0 here so that the driver doesn't bother tracking resources
4329 		 * that will never be used.
4330 		 */
4331 		sc->iscsicaps = 0;
4332 		sc->rdmacaps = 0;
4333 	}
4334 	if (sc->rdmacaps) {
4335 		param[0] = FW_PARAM_PFVF(STAG_START);
4336 		param[1] = FW_PARAM_PFVF(STAG_END);
4337 		param[2] = FW_PARAM_PFVF(RQ_START);
4338 		param[3] = FW_PARAM_PFVF(RQ_END);
4339 		param[4] = FW_PARAM_PFVF(PBL_START);
4340 		param[5] = FW_PARAM_PFVF(PBL_END);
4341 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4342 		if (rc != 0) {
4343 			device_printf(sc->dev,
4344 			    "failed to query RDMA parameters(1): %d.\n", rc);
4345 			return (rc);
4346 		}
4347 		sc->vres.stag.start = val[0];
4348 		sc->vres.stag.size = val[1] - val[0] + 1;
4349 		sc->vres.rq.start = val[2];
4350 		sc->vres.rq.size = val[3] - val[2] + 1;
4351 		sc->vres.pbl.start = val[4];
4352 		sc->vres.pbl.size = val[5] - val[4] + 1;
4353 
4354 		param[0] = FW_PARAM_PFVF(SQRQ_START);
4355 		param[1] = FW_PARAM_PFVF(SQRQ_END);
4356 		param[2] = FW_PARAM_PFVF(CQ_START);
4357 		param[3] = FW_PARAM_PFVF(CQ_END);
4358 		param[4] = FW_PARAM_PFVF(OCQ_START);
4359 		param[5] = FW_PARAM_PFVF(OCQ_END);
4360 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4361 		if (rc != 0) {
4362 			device_printf(sc->dev,
4363 			    "failed to query RDMA parameters(2): %d.\n", rc);
4364 			return (rc);
4365 		}
4366 		sc->vres.qp.start = val[0];
4367 		sc->vres.qp.size = val[1] - val[0] + 1;
4368 		sc->vres.cq.start = val[2];
4369 		sc->vres.cq.size = val[3] - val[2] + 1;
4370 		sc->vres.ocq.start = val[4];
4371 		sc->vres.ocq.size = val[5] - val[4] + 1;
4372 
4373 		param[0] = FW_PARAM_PFVF(SRQ_START);
4374 		param[1] = FW_PARAM_PFVF(SRQ_END);
4375 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
4376 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4377 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
4378 		if (rc != 0) {
4379 			device_printf(sc->dev,
4380 			    "failed to query RDMA parameters(3): %d.\n", rc);
4381 			return (rc);
4382 		}
4383 		sc->vres.srq.start = val[0];
4384 		sc->vres.srq.size = val[1] - val[0] + 1;
4385 		sc->params.max_ordird_qp = val[2];
4386 		sc->params.max_ird_adapter = val[3];
4387 	}
4388 	if (sc->iscsicaps) {
4389 		param[0] = FW_PARAM_PFVF(ISCSI_START);
4390 		param[1] = FW_PARAM_PFVF(ISCSI_END);
4391 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4392 		if (rc != 0) {
4393 			device_printf(sc->dev,
4394 			    "failed to query iSCSI parameters: %d.\n", rc);
4395 			return (rc);
4396 		}
4397 		sc->vres.iscsi.start = val[0];
4398 		sc->vres.iscsi.size = val[1] - val[0] + 1;
4399 	}
4400 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
4401 		param[0] = FW_PARAM_PFVF(TLS_START);
4402 		param[1] = FW_PARAM_PFVF(TLS_END);
4403 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4404 		if (rc != 0) {
4405 			device_printf(sc->dev,
4406 			    "failed to query TLS parameters: %d.\n", rc);
4407 			return (rc);
4408 		}
4409 		sc->vres.key.start = val[0];
4410 		sc->vres.key.size = val[1] - val[0] + 1;
4411 	}
4412 
4413 	t4_init_sge_params(sc);
4414 
4415 	/*
4416 	 * We've got the params we wanted to query via the firmware.  Now grab
4417 	 * some others directly from the chip.
4418 	 */
4419 	rc = t4_read_chip_settings(sc);
4420 
4421 	return (rc);
4422 }
4423 
4424 static int
4425 set_params__post_init(struct adapter *sc)
4426 {
4427 	uint32_t param, val;
4428 #ifdef TCP_OFFLOAD
4429 	int i, v, shift;
4430 #endif
4431 
4432 	/* ask for encapsulated CPLs */
4433 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4434 	val = 1;
4435 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4436 
4437 	/* Enable 32b port caps if the firmware supports it. */
4438 	param = FW_PARAM_PFVF(PORT_CAPS32);
4439 	val = 1;
4440 	if (t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val) == 0)
4441 		sc->params.port_caps32 = 1;
4442 
4443 	/* Let filter + maskhash steer to a part of the VI's RSS region. */
4444 	val = 1 << (G_MASKSIZE(t4_read_reg(sc, A_TP_RSS_CONFIG_TNL)) - 1);
4445 	t4_set_reg_field(sc, A_TP_RSS_CONFIG_TNL, V_MASKFILTER(M_MASKFILTER),
4446 	    V_MASKFILTER(val - 1));
4447 
4448 #ifdef TCP_OFFLOAD
4449 	/*
4450 	 * Override the TOE timers with user provided tunables.  This is not the
4451 	 * recommended way to change the timers (the firmware config file is) so
4452 	 * these tunables are not documented.
4453 	 *
4454 	 * All the timer tunables are in microseconds.
4455 	 */
4456 	if (t4_toe_keepalive_idle != 0) {
4457 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
4458 		v &= M_KEEPALIVEIDLE;
4459 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
4460 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
4461 	}
4462 	if (t4_toe_keepalive_interval != 0) {
4463 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
4464 		v &= M_KEEPALIVEINTVL;
4465 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
4466 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
4467 	}
4468 	if (t4_toe_keepalive_count != 0) {
4469 		v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
4470 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4471 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
4472 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
4473 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
4474 	}
4475 	if (t4_toe_rexmt_min != 0) {
4476 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
4477 		v &= M_RXTMIN;
4478 		t4_set_reg_field(sc, A_TP_RXT_MIN,
4479 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
4480 	}
4481 	if (t4_toe_rexmt_max != 0) {
4482 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
4483 		v &= M_RXTMAX;
4484 		t4_set_reg_field(sc, A_TP_RXT_MAX,
4485 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
4486 	}
4487 	if (t4_toe_rexmt_count != 0) {
4488 		v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
4489 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4490 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
4491 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
4492 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
4493 	}
4494 	for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
4495 		if (t4_toe_rexmt_backoff[i] != -1) {
4496 			v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
4497 			shift = (i & 3) << 3;
4498 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
4499 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
4500 		}
4501 	}
4502 #endif
4503 	return (0);
4504 }
4505 
4506 #undef FW_PARAM_PFVF
4507 #undef FW_PARAM_DEV
4508 
4509 static void
4510 t4_set_desc(struct adapter *sc)
4511 {
4512 	char buf[128];
4513 	struct adapter_params *p = &sc->params;
4514 
4515 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
4516 
4517 	device_set_desc_copy(sc->dev, buf);
4518 }
4519 
4520 static inline void
4521 ifmedia_add4(struct ifmedia *ifm, int m)
4522 {
4523 
4524 	ifmedia_add(ifm, m, 0, NULL);
4525 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE, 0, NULL);
4526 	ifmedia_add(ifm, m | IFM_ETH_RXPAUSE, 0, NULL);
4527 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE, 0, NULL);
4528 }
4529 
4530 /*
4531  * This is the selected media, which is not quite the same as the active media.
4532  * The media line in ifconfig is "media: Ethernet selected (active)" if selected
4533  * and active are not the same, and "media: Ethernet selected" otherwise.
4534  */
4535 static void
4536 set_current_media(struct port_info *pi)
4537 {
4538 	struct link_config *lc;
4539 	struct ifmedia *ifm;
4540 	int mword;
4541 	u_int speed;
4542 
4543 	PORT_LOCK_ASSERT_OWNED(pi);
4544 
4545 	/* Leave current media alone if it's already set to IFM_NONE. */
4546 	ifm = &pi->media;
4547 	if (ifm->ifm_cur != NULL &&
4548 	    IFM_SUBTYPE(ifm->ifm_cur->ifm_media) == IFM_NONE)
4549 		return;
4550 
4551 	lc = &pi->link_cfg;
4552 	if (lc->requested_aneg != AUTONEG_DISABLE &&
4553 	    lc->supported & FW_PORT_CAP32_ANEG) {
4554 		ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
4555 		return;
4556 	}
4557 	mword = IFM_ETHER | IFM_FDX;
4558 	if (lc->requested_fc & PAUSE_TX)
4559 		mword |= IFM_ETH_TXPAUSE;
4560 	if (lc->requested_fc & PAUSE_RX)
4561 		mword |= IFM_ETH_RXPAUSE;
4562 	if (lc->requested_speed == 0)
4563 		speed = port_top_speed(pi) * 1000;	/* Gbps -> Mbps */
4564 	else
4565 		speed = lc->requested_speed;
4566 	mword |= port_mword(pi, speed_to_fwcap(speed));
4567 	ifmedia_set(ifm, mword);
4568 }
4569 
4570 /*
4571  * Returns true if the ifmedia list for the port cannot change.
4572  */
4573 static bool
4574 fixed_ifmedia(struct port_info *pi)
4575 {
4576 
4577 	return (pi->port_type == FW_PORT_TYPE_BT_SGMII ||
4578 	    pi->port_type == FW_PORT_TYPE_BT_XFI ||
4579 	    pi->port_type == FW_PORT_TYPE_BT_XAUI ||
4580 	    pi->port_type == FW_PORT_TYPE_KX4 ||
4581 	    pi->port_type == FW_PORT_TYPE_KX ||
4582 	    pi->port_type == FW_PORT_TYPE_KR ||
4583 	    pi->port_type == FW_PORT_TYPE_BP_AP ||
4584 	    pi->port_type == FW_PORT_TYPE_BP4_AP ||
4585 	    pi->port_type == FW_PORT_TYPE_BP40_BA ||
4586 	    pi->port_type == FW_PORT_TYPE_KR4_100G ||
4587 	    pi->port_type == FW_PORT_TYPE_KR_SFP28 ||
4588 	    pi->port_type == FW_PORT_TYPE_KR_XLAUI);
4589 }
4590 
4591 static void
4592 build_medialist(struct port_info *pi)
4593 {
4594 	uint32_t ss, speed;
4595 	int unknown, mword, bit;
4596 	struct link_config *lc;
4597 	struct ifmedia *ifm;
4598 
4599 	PORT_LOCK_ASSERT_OWNED(pi);
4600 
4601 	if (pi->flags & FIXED_IFMEDIA)
4602 		return;
4603 
4604 	/*
4605 	 * Rebuild the ifmedia list.
4606 	 */
4607 	ifm = &pi->media;
4608 	ifmedia_removeall(ifm);
4609 	lc = &pi->link_cfg;
4610 	ss = G_FW_PORT_CAP32_SPEED(lc->supported); /* Supported Speeds */
4611 	if (__predict_false(ss == 0)) {	/* not supposed to happen. */
4612 		MPASS(ss != 0);
4613 no_media:
4614 		MPASS(LIST_EMPTY(&ifm->ifm_list));
4615 		ifmedia_add(ifm, IFM_ETHER | IFM_NONE, 0, NULL);
4616 		ifmedia_set(ifm, IFM_ETHER | IFM_NONE);
4617 		return;
4618 	}
4619 
4620 	unknown = 0;
4621 	for (bit = S_FW_PORT_CAP32_SPEED; bit < fls(ss); bit++) {
4622 		speed = 1 << bit;
4623 		MPASS(speed & M_FW_PORT_CAP32_SPEED);
4624 		if (ss & speed) {
4625 			mword = port_mword(pi, speed);
4626 			if (mword == IFM_NONE) {
4627 				goto no_media;
4628 			} else if (mword == IFM_UNKNOWN)
4629 				unknown++;
4630 			else
4631 				ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | mword);
4632 		}
4633 	}
4634 	if (unknown > 0) /* Add one unknown for all unknown media types. */
4635 		ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | IFM_UNKNOWN);
4636 	if (lc->supported & FW_PORT_CAP32_ANEG)
4637 		ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
4638 
4639 	set_current_media(pi);
4640 }
4641 
4642 /*
4643  * Initialize the requested fields in the link config based on driver tunables.
4644  */
4645 static void
4646 init_link_config(struct port_info *pi)
4647 {
4648 	struct link_config *lc = &pi->link_cfg;
4649 
4650 	PORT_LOCK_ASSERT_OWNED(pi);
4651 
4652 	lc->requested_speed = 0;
4653 
4654 	if (t4_autoneg == 0)
4655 		lc->requested_aneg = AUTONEG_DISABLE;
4656 	else if (t4_autoneg == 1)
4657 		lc->requested_aneg = AUTONEG_ENABLE;
4658 	else
4659 		lc->requested_aneg = AUTONEG_AUTO;
4660 
4661 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX |
4662 	    PAUSE_AUTONEG);
4663 
4664 	if (t4_fec == -1 || t4_fec & FEC_AUTO)
4665 		lc->requested_fec = FEC_AUTO;
4666 	else {
4667 		lc->requested_fec = FEC_NONE;
4668 		if (t4_fec & FEC_RS)
4669 			lc->requested_fec |= FEC_RS;
4670 		if (t4_fec & FEC_BASER_RS)
4671 			lc->requested_fec |= FEC_BASER_RS;
4672 	}
4673 }
4674 
4675 /*
4676  * Makes sure that all requested settings comply with what's supported by the
4677  * port.  Returns the number of settings that were invalid and had to be fixed.
4678  */
4679 static int
4680 fixup_link_config(struct port_info *pi)
4681 {
4682 	int n = 0;
4683 	struct link_config *lc = &pi->link_cfg;
4684 	uint32_t fwspeed;
4685 
4686 	PORT_LOCK_ASSERT_OWNED(pi);
4687 
4688 	/* Speed (when not autonegotiating) */
4689 	if (lc->requested_speed != 0) {
4690 		fwspeed = speed_to_fwcap(lc->requested_speed);
4691 		if ((fwspeed & lc->supported) == 0) {
4692 			n++;
4693 			lc->requested_speed = 0;
4694 		}
4695 	}
4696 
4697 	/* Link autonegotiation */
4698 	MPASS(lc->requested_aneg == AUTONEG_ENABLE ||
4699 	    lc->requested_aneg == AUTONEG_DISABLE ||
4700 	    lc->requested_aneg == AUTONEG_AUTO);
4701 	if (lc->requested_aneg == AUTONEG_ENABLE &&
4702 	    !(lc->supported & FW_PORT_CAP32_ANEG)) {
4703 		n++;
4704 		lc->requested_aneg = AUTONEG_AUTO;
4705 	}
4706 
4707 	/* Flow control */
4708 	MPASS((lc->requested_fc & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG)) == 0);
4709 	if (lc->requested_fc & PAUSE_TX &&
4710 	    !(lc->supported & FW_PORT_CAP32_FC_TX)) {
4711 		n++;
4712 		lc->requested_fc &= ~PAUSE_TX;
4713 	}
4714 	if (lc->requested_fc & PAUSE_RX &&
4715 	    !(lc->supported & FW_PORT_CAP32_FC_RX)) {
4716 		n++;
4717 		lc->requested_fc &= ~PAUSE_RX;
4718 	}
4719 	if (!(lc->requested_fc & PAUSE_AUTONEG) &&
4720 	    !(lc->supported & FW_PORT_CAP32_FORCE_PAUSE)) {
4721 		n++;
4722 		lc->requested_fc |= PAUSE_AUTONEG;
4723 	}
4724 
4725 	/* FEC */
4726 	if ((lc->requested_fec & FEC_RS &&
4727 	    !(lc->supported & FW_PORT_CAP32_FEC_RS)) ||
4728 	    (lc->requested_fec & FEC_BASER_RS &&
4729 	    !(lc->supported & FW_PORT_CAP32_FEC_BASER_RS))) {
4730 		n++;
4731 		lc->requested_fec = FEC_AUTO;
4732 	}
4733 
4734 	return (n);
4735 }
4736 
4737 /*
4738  * Apply the requested L1 settings, which are expected to be valid, to the
4739  * hardware.
4740  */
4741 static int
4742 apply_link_config(struct port_info *pi)
4743 {
4744 	struct adapter *sc = pi->adapter;
4745 	struct link_config *lc = &pi->link_cfg;
4746 	int rc;
4747 
4748 #ifdef INVARIANTS
4749 	ASSERT_SYNCHRONIZED_OP(sc);
4750 	PORT_LOCK_ASSERT_OWNED(pi);
4751 
4752 	if (lc->requested_aneg == AUTONEG_ENABLE)
4753 		MPASS(lc->supported & FW_PORT_CAP32_ANEG);
4754 	if (!(lc->requested_fc & PAUSE_AUTONEG))
4755 		MPASS(lc->supported & FW_PORT_CAP32_FORCE_PAUSE);
4756 	if (lc->requested_fc & PAUSE_TX)
4757 		MPASS(lc->supported & FW_PORT_CAP32_FC_TX);
4758 	if (lc->requested_fc & PAUSE_RX)
4759 		MPASS(lc->supported & FW_PORT_CAP32_FC_RX);
4760 	if (lc->requested_fec & FEC_RS)
4761 		MPASS(lc->supported & FW_PORT_CAP32_FEC_RS);
4762 	if (lc->requested_fec & FEC_BASER_RS)
4763 		MPASS(lc->supported & FW_PORT_CAP32_FEC_BASER_RS);
4764 #endif
4765 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
4766 	if (rc != 0) {
4767 		/* Don't complain if the VF driver gets back an EPERM. */
4768 		if (!(sc->flags & IS_VF) || rc != FW_EPERM)
4769 			device_printf(pi->dev, "l1cfg failed: %d\n", rc);
4770 	} else {
4771 		/*
4772 		 * An L1_CFG will almost always result in a link-change event if
4773 		 * the link is up, and the driver will refresh the actual
4774 		 * fec/fc/etc. when the notification is processed.  If the link
4775 		 * is down then the actual settings are meaningless.
4776 		 *
4777 		 * This takes care of the case where a change in the L1 settings
4778 		 * may not result in a notification.
4779 		 */
4780 		if (lc->link_ok && !(lc->requested_fc & PAUSE_AUTONEG))
4781 			lc->fc = lc->requested_fc & (PAUSE_TX | PAUSE_RX);
4782 	}
4783 	return (rc);
4784 }
4785 
4786 #define FW_MAC_EXACT_CHUNK	7
4787 
4788 /*
4789  * Program the port's XGMAC based on parameters in ifnet.  The caller also
4790  * indicates which parameters should be programmed (the rest are left alone).
4791  */
4792 int
4793 update_mac_settings(struct ifnet *ifp, int flags)
4794 {
4795 	int rc = 0;
4796 	struct vi_info *vi = ifp->if_softc;
4797 	struct port_info *pi = vi->pi;
4798 	struct adapter *sc = pi->adapter;
4799 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4800 
4801 	ASSERT_SYNCHRONIZED_OP(sc);
4802 	KASSERT(flags, ("%s: not told what to update.", __func__));
4803 
4804 	if (flags & XGMAC_MTU)
4805 		mtu = ifp->if_mtu;
4806 
4807 	if (flags & XGMAC_PROMISC)
4808 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4809 
4810 	if (flags & XGMAC_ALLMULTI)
4811 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4812 
4813 	if (flags & XGMAC_VLANEX)
4814 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4815 
4816 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4817 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4818 		    allmulti, 1, vlanex, false);
4819 		if (rc) {
4820 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4821 			    rc);
4822 			return (rc);
4823 		}
4824 	}
4825 
4826 	if (flags & XGMAC_UCADDR) {
4827 		uint8_t ucaddr[ETHER_ADDR_LEN];
4828 
4829 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4830 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4831 		    ucaddr, true, &vi->smt_idx);
4832 		if (rc < 0) {
4833 			rc = -rc;
4834 			if_printf(ifp, "change_mac failed: %d\n", rc);
4835 			return (rc);
4836 		} else {
4837 			vi->xact_addr_filt = rc;
4838 			rc = 0;
4839 		}
4840 	}
4841 
4842 	if (flags & XGMAC_MCADDRS) {
4843 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4844 		int del = 1;
4845 		uint64_t hash = 0;
4846 		struct ifmultiaddr *ifma;
4847 		int i = 0, j;
4848 
4849 		if_maddr_rlock(ifp);
4850 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4851 			if (ifma->ifma_addr->sa_family != AF_LINK)
4852 				continue;
4853 			mcaddr[i] =
4854 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4855 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4856 			i++;
4857 
4858 			if (i == FW_MAC_EXACT_CHUNK) {
4859 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4860 				    del, i, mcaddr, NULL, &hash, 0);
4861 				if (rc < 0) {
4862 					rc = -rc;
4863 					for (j = 0; j < i; j++) {
4864 						if_printf(ifp,
4865 						    "failed to add mc address"
4866 						    " %02x:%02x:%02x:"
4867 						    "%02x:%02x:%02x rc=%d\n",
4868 						    mcaddr[j][0], mcaddr[j][1],
4869 						    mcaddr[j][2], mcaddr[j][3],
4870 						    mcaddr[j][4], mcaddr[j][5],
4871 						    rc);
4872 					}
4873 					goto mcfail;
4874 				}
4875 				del = 0;
4876 				i = 0;
4877 			}
4878 		}
4879 		if (i > 0) {
4880 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4881 			    mcaddr, NULL, &hash, 0);
4882 			if (rc < 0) {
4883 				rc = -rc;
4884 				for (j = 0; j < i; j++) {
4885 					if_printf(ifp,
4886 					    "failed to add mc address"
4887 					    " %02x:%02x:%02x:"
4888 					    "%02x:%02x:%02x rc=%d\n",
4889 					    mcaddr[j][0], mcaddr[j][1],
4890 					    mcaddr[j][2], mcaddr[j][3],
4891 					    mcaddr[j][4], mcaddr[j][5],
4892 					    rc);
4893 				}
4894 				goto mcfail;
4895 			}
4896 		}
4897 
4898 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4899 		if (rc != 0)
4900 			if_printf(ifp, "failed to set mc address hash: %d", rc);
4901 mcfail:
4902 		if_maddr_runlock(ifp);
4903 	}
4904 
4905 	return (rc);
4906 }
4907 
4908 /*
4909  * {begin|end}_synchronized_op must be called from the same thread.
4910  */
4911 int
4912 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4913     char *wmesg)
4914 {
4915 	int rc, pri;
4916 
4917 #ifdef WITNESS
4918 	/* the caller thinks it's ok to sleep, but is it really? */
4919 	if (flags & SLEEP_OK)
4920 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4921 		    "begin_synchronized_op");
4922 #endif
4923 
4924 	if (INTR_OK)
4925 		pri = PCATCH;
4926 	else
4927 		pri = 0;
4928 
4929 	ADAPTER_LOCK(sc);
4930 	for (;;) {
4931 
4932 		if (vi && IS_DOOMED(vi)) {
4933 			rc = ENXIO;
4934 			goto done;
4935 		}
4936 
4937 		if (!IS_BUSY(sc)) {
4938 			rc = 0;
4939 			break;
4940 		}
4941 
4942 		if (!(flags & SLEEP_OK)) {
4943 			rc = EBUSY;
4944 			goto done;
4945 		}
4946 
4947 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4948 			rc = EINTR;
4949 			goto done;
4950 		}
4951 	}
4952 
4953 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4954 	SET_BUSY(sc);
4955 #ifdef INVARIANTS
4956 	sc->last_op = wmesg;
4957 	sc->last_op_thr = curthread;
4958 	sc->last_op_flags = flags;
4959 #endif
4960 
4961 done:
4962 	if (!(flags & HOLD_LOCK) || rc)
4963 		ADAPTER_UNLOCK(sc);
4964 
4965 	return (rc);
4966 }
4967 
4968 /*
4969  * Tell if_ioctl and if_init that the VI is going away.  This is
4970  * special variant of begin_synchronized_op and must be paired with a
4971  * call to end_synchronized_op.
4972  */
4973 void
4974 doom_vi(struct adapter *sc, struct vi_info *vi)
4975 {
4976 
4977 	ADAPTER_LOCK(sc);
4978 	SET_DOOMED(vi);
4979 	wakeup(&sc->flags);
4980 	while (IS_BUSY(sc))
4981 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4982 	SET_BUSY(sc);
4983 #ifdef INVARIANTS
4984 	sc->last_op = "t4detach";
4985 	sc->last_op_thr = curthread;
4986 	sc->last_op_flags = 0;
4987 #endif
4988 	ADAPTER_UNLOCK(sc);
4989 }
4990 
4991 /*
4992  * {begin|end}_synchronized_op must be called from the same thread.
4993  */
4994 void
4995 end_synchronized_op(struct adapter *sc, int flags)
4996 {
4997 
4998 	if (flags & LOCK_HELD)
4999 		ADAPTER_LOCK_ASSERT_OWNED(sc);
5000 	else
5001 		ADAPTER_LOCK(sc);
5002 
5003 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
5004 	CLR_BUSY(sc);
5005 	wakeup(&sc->flags);
5006 	ADAPTER_UNLOCK(sc);
5007 }
5008 
5009 static int
5010 cxgbe_init_synchronized(struct vi_info *vi)
5011 {
5012 	struct port_info *pi = vi->pi;
5013 	struct adapter *sc = pi->adapter;
5014 	struct ifnet *ifp = vi->ifp;
5015 	int rc = 0, i;
5016 	struct sge_txq *txq;
5017 
5018 	ASSERT_SYNCHRONIZED_OP(sc);
5019 
5020 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
5021 		return (0);	/* already running */
5022 
5023 	if (!(sc->flags & FULL_INIT_DONE) &&
5024 	    ((rc = adapter_full_init(sc)) != 0))
5025 		return (rc);	/* error message displayed already */
5026 
5027 	if (!(vi->flags & VI_INIT_DONE) &&
5028 	    ((rc = vi_full_init(vi)) != 0))
5029 		return (rc); /* error message displayed already */
5030 
5031 	rc = update_mac_settings(ifp, XGMAC_ALL);
5032 	if (rc)
5033 		goto done;	/* error message displayed already */
5034 
5035 	PORT_LOCK(pi);
5036 	if (pi->up_vis == 0) {
5037 		t4_update_port_info(pi);
5038 		fixup_link_config(pi);
5039 		build_medialist(pi);
5040 		apply_link_config(pi);
5041 	}
5042 
5043 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
5044 	if (rc != 0) {
5045 		if_printf(ifp, "enable_vi failed: %d\n", rc);
5046 		PORT_UNLOCK(pi);
5047 		goto done;
5048 	}
5049 
5050 	/*
5051 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
5052 	 * if this changes.
5053 	 */
5054 
5055 	for_each_txq(vi, i, txq) {
5056 		TXQ_LOCK(txq);
5057 		txq->eq.flags |= EQ_ENABLED;
5058 		TXQ_UNLOCK(txq);
5059 	}
5060 
5061 	/*
5062 	 * The first iq of the first port to come up is used for tracing.
5063 	 */
5064 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
5065 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
5066 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
5067 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
5068 		    V_QUEUENUMBER(sc->traceq));
5069 		pi->flags |= HAS_TRACEQ;
5070 	}
5071 
5072 	/* all ok */
5073 	pi->up_vis++;
5074 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
5075 
5076 	if (pi->nvi > 1 || sc->flags & IS_VF)
5077 		callout_reset(&vi->tick, hz, vi_tick, vi);
5078 	else
5079 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
5080 	if (pi->link_cfg.link_ok)
5081 		t4_os_link_changed(pi);
5082 	PORT_UNLOCK(pi);
5083 done:
5084 	if (rc != 0)
5085 		cxgbe_uninit_synchronized(vi);
5086 
5087 	return (rc);
5088 }
5089 
5090 /*
5091  * Idempotent.
5092  */
5093 static int
5094 cxgbe_uninit_synchronized(struct vi_info *vi)
5095 {
5096 	struct port_info *pi = vi->pi;
5097 	struct adapter *sc = pi->adapter;
5098 	struct ifnet *ifp = vi->ifp;
5099 	int rc, i;
5100 	struct sge_txq *txq;
5101 
5102 	ASSERT_SYNCHRONIZED_OP(sc);
5103 
5104 	if (!(vi->flags & VI_INIT_DONE)) {
5105 		if (__predict_false(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
5106 			KASSERT(0, ("uninited VI is running"));
5107 			if_printf(ifp, "uninited VI with running ifnet.  "
5108 			    "vi->flags 0x%016lx, if_flags 0x%08x, "
5109 			    "if_drv_flags 0x%08x\n", vi->flags, ifp->if_flags,
5110 			    ifp->if_drv_flags);
5111 		}
5112 		return (0);
5113 	}
5114 
5115 	/*
5116 	 * Disable the VI so that all its data in either direction is discarded
5117 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
5118 	 * tick) intact as the TP can deliver negative advice or data that it's
5119 	 * holding in its RAM (for an offloaded connection) even after the VI is
5120 	 * disabled.
5121 	 */
5122 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
5123 	if (rc) {
5124 		if_printf(ifp, "disable_vi failed: %d\n", rc);
5125 		return (rc);
5126 	}
5127 
5128 	for_each_txq(vi, i, txq) {
5129 		TXQ_LOCK(txq);
5130 		txq->eq.flags &= ~EQ_ENABLED;
5131 		TXQ_UNLOCK(txq);
5132 	}
5133 
5134 	PORT_LOCK(pi);
5135 	if (pi->nvi > 1 || sc->flags & IS_VF)
5136 		callout_stop(&vi->tick);
5137 	else
5138 		callout_stop(&pi->tick);
5139 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
5140 		PORT_UNLOCK(pi);
5141 		return (0);
5142 	}
5143 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
5144 	pi->up_vis--;
5145 	if (pi->up_vis > 0) {
5146 		PORT_UNLOCK(pi);
5147 		return (0);
5148 	}
5149 
5150 	pi->link_cfg.link_ok = false;
5151 	pi->link_cfg.speed = 0;
5152 	pi->link_cfg.link_down_rc = 255;
5153 	t4_os_link_changed(pi);
5154 	PORT_UNLOCK(pi);
5155 
5156 	return (0);
5157 }
5158 
5159 /*
5160  * It is ok for this function to fail midway and return right away.  t4_detach
5161  * will walk the entire sc->irq list and clean up whatever is valid.
5162  */
5163 int
5164 t4_setup_intr_handlers(struct adapter *sc)
5165 {
5166 	int rc, rid, p, q, v;
5167 	char s[8];
5168 	struct irq *irq;
5169 	struct port_info *pi;
5170 	struct vi_info *vi;
5171 	struct sge *sge = &sc->sge;
5172 	struct sge_rxq *rxq;
5173 #ifdef TCP_OFFLOAD
5174 	struct sge_ofld_rxq *ofld_rxq;
5175 #endif
5176 #ifdef DEV_NETMAP
5177 	struct sge_nm_rxq *nm_rxq;
5178 #endif
5179 #ifdef RSS
5180 	int nbuckets = rss_getnumbuckets();
5181 #endif
5182 
5183 	/*
5184 	 * Setup interrupts.
5185 	 */
5186 	irq = &sc->irq[0];
5187 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
5188 	if (forwarding_intr_to_fwq(sc))
5189 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
5190 
5191 	/* Multiple interrupts. */
5192 	if (sc->flags & IS_VF)
5193 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
5194 		    ("%s: too few intr.", __func__));
5195 	else
5196 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
5197 		    ("%s: too few intr.", __func__));
5198 
5199 	/* The first one is always error intr on PFs */
5200 	if (!(sc->flags & IS_VF)) {
5201 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
5202 		if (rc != 0)
5203 			return (rc);
5204 		irq++;
5205 		rid++;
5206 	}
5207 
5208 	/* The second one is always the firmware event queue (first on VFs) */
5209 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
5210 	if (rc != 0)
5211 		return (rc);
5212 	irq++;
5213 	rid++;
5214 
5215 	for_each_port(sc, p) {
5216 		pi = sc->port[p];
5217 		for_each_vi(pi, v, vi) {
5218 			vi->first_intr = rid - 1;
5219 
5220 			if (vi->nnmrxq > 0) {
5221 				int n = max(vi->nrxq, vi->nnmrxq);
5222 
5223 				rxq = &sge->rxq[vi->first_rxq];
5224 #ifdef DEV_NETMAP
5225 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
5226 #endif
5227 				for (q = 0; q < n; q++) {
5228 					snprintf(s, sizeof(s), "%x%c%x", p,
5229 					    'a' + v, q);
5230 					if (q < vi->nrxq)
5231 						irq->rxq = rxq++;
5232 #ifdef DEV_NETMAP
5233 					if (q < vi->nnmrxq)
5234 						irq->nm_rxq = nm_rxq++;
5235 
5236 					if (irq->nm_rxq != NULL &&
5237 					    irq->rxq == NULL) {
5238 						/* Netmap rx only */
5239 						rc = t4_alloc_irq(sc, irq, rid,
5240 						    t4_nm_intr, irq->nm_rxq, s);
5241 					}
5242 					if (irq->nm_rxq != NULL &&
5243 					    irq->rxq != NULL) {
5244 						/* NIC and Netmap rx */
5245 						rc = t4_alloc_irq(sc, irq, rid,
5246 						    t4_vi_intr, irq, s);
5247 					}
5248 #endif
5249 					if (irq->rxq != NULL &&
5250 					    irq->nm_rxq == NULL) {
5251 						/* NIC rx only */
5252 						rc = t4_alloc_irq(sc, irq, rid,
5253 						    t4_intr, irq->rxq, s);
5254 					}
5255 					if (rc != 0)
5256 						return (rc);
5257 #ifdef RSS
5258 					if (q < vi->nrxq) {
5259 						bus_bind_intr(sc->dev, irq->res,
5260 						    rss_getcpu(q % nbuckets));
5261 					}
5262 #endif
5263 					irq++;
5264 					rid++;
5265 					vi->nintr++;
5266 				}
5267 			} else {
5268 				for_each_rxq(vi, q, rxq) {
5269 					snprintf(s, sizeof(s), "%x%c%x", p,
5270 					    'a' + v, q);
5271 					rc = t4_alloc_irq(sc, irq, rid,
5272 					    t4_intr, rxq, s);
5273 					if (rc != 0)
5274 						return (rc);
5275 #ifdef RSS
5276 					bus_bind_intr(sc->dev, irq->res,
5277 					    rss_getcpu(q % nbuckets));
5278 #endif
5279 					irq++;
5280 					rid++;
5281 					vi->nintr++;
5282 				}
5283 			}
5284 #ifdef TCP_OFFLOAD
5285 			for_each_ofld_rxq(vi, q, ofld_rxq) {
5286 				snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
5287 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
5288 				    ofld_rxq, s);
5289 				if (rc != 0)
5290 					return (rc);
5291 				irq++;
5292 				rid++;
5293 				vi->nintr++;
5294 			}
5295 #endif
5296 		}
5297 	}
5298 	MPASS(irq == &sc->irq[sc->intr_count]);
5299 
5300 	return (0);
5301 }
5302 
5303 int
5304 adapter_full_init(struct adapter *sc)
5305 {
5306 	int rc, i;
5307 #ifdef RSS
5308 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5309 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5310 #endif
5311 
5312 	ASSERT_SYNCHRONIZED_OP(sc);
5313 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5314 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
5315 	    ("%s: FULL_INIT_DONE already", __func__));
5316 
5317 	/*
5318 	 * queues that belong to the adapter (not any particular port).
5319 	 */
5320 	rc = t4_setup_adapter_queues(sc);
5321 	if (rc != 0)
5322 		goto done;
5323 
5324 	for (i = 0; i < nitems(sc->tq); i++) {
5325 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
5326 		    taskqueue_thread_enqueue, &sc->tq[i]);
5327 		if (sc->tq[i] == NULL) {
5328 			device_printf(sc->dev,
5329 			    "failed to allocate task queue %d\n", i);
5330 			rc = ENOMEM;
5331 			goto done;
5332 		}
5333 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
5334 		    device_get_nameunit(sc->dev), i);
5335 	}
5336 #ifdef RSS
5337 	MPASS(RSS_KEYSIZE == 40);
5338 	rss_getkey((void *)&raw_rss_key[0]);
5339 	for (i = 0; i < nitems(rss_key); i++) {
5340 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
5341 	}
5342 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
5343 #endif
5344 
5345 	if (!(sc->flags & IS_VF))
5346 		t4_intr_enable(sc);
5347 	sc->flags |= FULL_INIT_DONE;
5348 done:
5349 	if (rc != 0)
5350 		adapter_full_uninit(sc);
5351 
5352 	return (rc);
5353 }
5354 
5355 int
5356 adapter_full_uninit(struct adapter *sc)
5357 {
5358 	int i;
5359 
5360 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5361 
5362 	t4_teardown_adapter_queues(sc);
5363 
5364 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
5365 		taskqueue_free(sc->tq[i]);
5366 		sc->tq[i] = NULL;
5367 	}
5368 
5369 	sc->flags &= ~FULL_INIT_DONE;
5370 
5371 	return (0);
5372 }
5373 
5374 #ifdef RSS
5375 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
5376     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
5377     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
5378     RSS_HASHTYPE_RSS_UDP_IPV6)
5379 
5380 /* Translates kernel hash types to hardware. */
5381 static int
5382 hashconfig_to_hashen(int hashconfig)
5383 {
5384 	int hashen = 0;
5385 
5386 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
5387 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
5388 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
5389 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
5390 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
5391 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5392 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5393 	}
5394 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
5395 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5396 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5397 	}
5398 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
5399 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5400 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
5401 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5402 
5403 	return (hashen);
5404 }
5405 
5406 /* Translates hardware hash types to kernel. */
5407 static int
5408 hashen_to_hashconfig(int hashen)
5409 {
5410 	int hashconfig = 0;
5411 
5412 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
5413 		/*
5414 		 * If UDP hashing was enabled it must have been enabled for
5415 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
5416 		 * enabling any 4-tuple hash is nonsense configuration.
5417 		 */
5418 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5419 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
5420 
5421 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5422 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
5423 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5424 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
5425 	}
5426 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5427 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
5428 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5429 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
5430 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
5431 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
5432 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
5433 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
5434 
5435 	return (hashconfig);
5436 }
5437 #endif
5438 
5439 int
5440 vi_full_init(struct vi_info *vi)
5441 {
5442 	struct adapter *sc = vi->pi->adapter;
5443 	struct ifnet *ifp = vi->ifp;
5444 	uint16_t *rss;
5445 	struct sge_rxq *rxq;
5446 	int rc, i, j;
5447 #ifdef RSS
5448 	int nbuckets = rss_getnumbuckets();
5449 	int hashconfig = rss_gethashconfig();
5450 	int extra;
5451 #endif
5452 
5453 	ASSERT_SYNCHRONIZED_OP(sc);
5454 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
5455 	    ("%s: VI_INIT_DONE already", __func__));
5456 
5457 	sysctl_ctx_init(&vi->ctx);
5458 	vi->flags |= VI_SYSCTL_CTX;
5459 
5460 	/*
5461 	 * Allocate tx/rx/fl queues for this VI.
5462 	 */
5463 	rc = t4_setup_vi_queues(vi);
5464 	if (rc != 0)
5465 		goto done;	/* error message displayed already */
5466 
5467 	/*
5468 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
5469 	 */
5470 	if (vi->nrxq > vi->rss_size) {
5471 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
5472 		    "some queues will never receive traffic.\n", vi->nrxq,
5473 		    vi->rss_size);
5474 	} else if (vi->rss_size % vi->nrxq) {
5475 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
5476 		    "expect uneven traffic distribution.\n", vi->nrxq,
5477 		    vi->rss_size);
5478 	}
5479 #ifdef RSS
5480 	if (vi->nrxq != nbuckets) {
5481 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
5482 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
5483 	}
5484 #endif
5485 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
5486 	for (i = 0; i < vi->rss_size;) {
5487 #ifdef RSS
5488 		j = rss_get_indirection_to_bucket(i);
5489 		j %= vi->nrxq;
5490 		rxq = &sc->sge.rxq[vi->first_rxq + j];
5491 		rss[i++] = rxq->iq.abs_id;
5492 #else
5493 		for_each_rxq(vi, j, rxq) {
5494 			rss[i++] = rxq->iq.abs_id;
5495 			if (i == vi->rss_size)
5496 				break;
5497 		}
5498 #endif
5499 	}
5500 
5501 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
5502 	    vi->rss_size);
5503 	if (rc != 0) {
5504 		free(rss, M_CXGBE);
5505 		if_printf(ifp, "rss_config failed: %d\n", rc);
5506 		goto done;
5507 	}
5508 
5509 #ifdef RSS
5510 	vi->hashen = hashconfig_to_hashen(hashconfig);
5511 
5512 	/*
5513 	 * We may have had to enable some hashes even though the global config
5514 	 * wants them disabled.  This is a potential problem that must be
5515 	 * reported to the user.
5516 	 */
5517 	extra = hashen_to_hashconfig(vi->hashen) ^ hashconfig;
5518 
5519 	/*
5520 	 * If we consider only the supported hash types, then the enabled hashes
5521 	 * are a superset of the requested hashes.  In other words, there cannot
5522 	 * be any supported hash that was requested but not enabled, but there
5523 	 * can be hashes that were not requested but had to be enabled.
5524 	 */
5525 	extra &= SUPPORTED_RSS_HASHTYPES;
5526 	MPASS((extra & hashconfig) == 0);
5527 
5528 	if (extra) {
5529 		if_printf(ifp,
5530 		    "global RSS config (0x%x) cannot be accommodated.\n",
5531 		    hashconfig);
5532 	}
5533 	if (extra & RSS_HASHTYPE_RSS_IPV4)
5534 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
5535 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
5536 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
5537 	if (extra & RSS_HASHTYPE_RSS_IPV6)
5538 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
5539 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
5540 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
5541 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
5542 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
5543 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
5544 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
5545 #else
5546 	vi->hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
5547 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
5548 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5549 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
5550 #endif
5551 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, vi->hashen, rss[0], 0, 0);
5552 	if (rc != 0) {
5553 		free(rss, M_CXGBE);
5554 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
5555 		goto done;
5556 	}
5557 
5558 	vi->rss = rss;
5559 	vi->flags |= VI_INIT_DONE;
5560 done:
5561 	if (rc != 0)
5562 		vi_full_uninit(vi);
5563 
5564 	return (rc);
5565 }
5566 
5567 /*
5568  * Idempotent.
5569  */
5570 int
5571 vi_full_uninit(struct vi_info *vi)
5572 {
5573 	struct port_info *pi = vi->pi;
5574 	struct adapter *sc = pi->adapter;
5575 	int i;
5576 	struct sge_rxq *rxq;
5577 	struct sge_txq *txq;
5578 #ifdef TCP_OFFLOAD
5579 	struct sge_ofld_rxq *ofld_rxq;
5580 #endif
5581 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
5582 	struct sge_wrq *ofld_txq;
5583 #endif
5584 
5585 	if (vi->flags & VI_INIT_DONE) {
5586 
5587 		/* Need to quiesce queues.  */
5588 
5589 		/* XXX: Only for the first VI? */
5590 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
5591 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
5592 
5593 		for_each_txq(vi, i, txq) {
5594 			quiesce_txq(sc, txq);
5595 		}
5596 
5597 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
5598 		for_each_ofld_txq(vi, i, ofld_txq) {
5599 			quiesce_wrq(sc, ofld_txq);
5600 		}
5601 #endif
5602 
5603 		for_each_rxq(vi, i, rxq) {
5604 			quiesce_iq(sc, &rxq->iq);
5605 			quiesce_fl(sc, &rxq->fl);
5606 		}
5607 
5608 #ifdef TCP_OFFLOAD
5609 		for_each_ofld_rxq(vi, i, ofld_rxq) {
5610 			quiesce_iq(sc, &ofld_rxq->iq);
5611 			quiesce_fl(sc, &ofld_rxq->fl);
5612 		}
5613 #endif
5614 		free(vi->rss, M_CXGBE);
5615 		free(vi->nm_rss, M_CXGBE);
5616 	}
5617 
5618 	t4_teardown_vi_queues(vi);
5619 	vi->flags &= ~VI_INIT_DONE;
5620 
5621 	return (0);
5622 }
5623 
5624 static void
5625 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
5626 {
5627 	struct sge_eq *eq = &txq->eq;
5628 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
5629 
5630 	(void) sc;	/* unused */
5631 
5632 #ifdef INVARIANTS
5633 	TXQ_LOCK(txq);
5634 	MPASS((eq->flags & EQ_ENABLED) == 0);
5635 	TXQ_UNLOCK(txq);
5636 #endif
5637 
5638 	/* Wait for the mp_ring to empty. */
5639 	while (!mp_ring_is_idle(txq->r)) {
5640 		mp_ring_check_drainage(txq->r, 0);
5641 		pause("rquiesce", 1);
5642 	}
5643 
5644 	/* Then wait for the hardware to finish. */
5645 	while (spg->cidx != htobe16(eq->pidx))
5646 		pause("equiesce", 1);
5647 
5648 	/* Finally, wait for the driver to reclaim all descriptors. */
5649 	while (eq->cidx != eq->pidx)
5650 		pause("dquiesce", 1);
5651 }
5652 
5653 static void
5654 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
5655 {
5656 
5657 	/* XXXTX */
5658 }
5659 
5660 static void
5661 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
5662 {
5663 	(void) sc;	/* unused */
5664 
5665 	/* Synchronize with the interrupt handler */
5666 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
5667 		pause("iqfree", 1);
5668 }
5669 
5670 static void
5671 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
5672 {
5673 	mtx_lock(&sc->sfl_lock);
5674 	FL_LOCK(fl);
5675 	fl->flags |= FL_DOOMED;
5676 	FL_UNLOCK(fl);
5677 	callout_stop(&sc->sfl_callout);
5678 	mtx_unlock(&sc->sfl_lock);
5679 
5680 	KASSERT((fl->flags & FL_STARVING) == 0,
5681 	    ("%s: still starving", __func__));
5682 }
5683 
5684 static int
5685 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
5686     driver_intr_t *handler, void *arg, char *name)
5687 {
5688 	int rc;
5689 
5690 	irq->rid = rid;
5691 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
5692 	    RF_SHAREABLE | RF_ACTIVE);
5693 	if (irq->res == NULL) {
5694 		device_printf(sc->dev,
5695 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
5696 		return (ENOMEM);
5697 	}
5698 
5699 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
5700 	    NULL, handler, arg, &irq->tag);
5701 	if (rc != 0) {
5702 		device_printf(sc->dev,
5703 		    "failed to setup interrupt for rid %d, name %s: %d\n",
5704 		    rid, name, rc);
5705 	} else if (name)
5706 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
5707 
5708 	return (rc);
5709 }
5710 
5711 static int
5712 t4_free_irq(struct adapter *sc, struct irq *irq)
5713 {
5714 	if (irq->tag)
5715 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
5716 	if (irq->res)
5717 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
5718 
5719 	bzero(irq, sizeof(*irq));
5720 
5721 	return (0);
5722 }
5723 
5724 static void
5725 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
5726 {
5727 
5728 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
5729 	t4_get_regs(sc, buf, regs->len);
5730 }
5731 
5732 #define	A_PL_INDIR_CMD	0x1f8
5733 
5734 #define	S_PL_AUTOINC	31
5735 #define	M_PL_AUTOINC	0x1U
5736 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
5737 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
5738 
5739 #define	S_PL_VFID	20
5740 #define	M_PL_VFID	0xffU
5741 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
5742 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
5743 
5744 #define	S_PL_ADDR	0
5745 #define	M_PL_ADDR	0xfffffU
5746 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
5747 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
5748 
5749 #define	A_PL_INDIR_DATA	0x1fc
5750 
5751 static uint64_t
5752 read_vf_stat(struct adapter *sc, u_int vin, int reg)
5753 {
5754 	u32 stats[2];
5755 
5756 	mtx_assert(&sc->reg_lock, MA_OWNED);
5757 	if (sc->flags & IS_VF) {
5758 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
5759 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
5760 	} else {
5761 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5762 		    V_PL_VFID(vin) | V_PL_ADDR(VF_MPS_REG(reg)));
5763 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
5764 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
5765 	}
5766 	return (((uint64_t)stats[1]) << 32 | stats[0]);
5767 }
5768 
5769 static void
5770 t4_get_vi_stats(struct adapter *sc, u_int vin, struct fw_vi_stats_vf *stats)
5771 {
5772 
5773 #define GET_STAT(name) \
5774 	read_vf_stat(sc, vin, A_MPS_VF_STAT_##name##_L)
5775 
5776 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
5777 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
5778 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
5779 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
5780 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
5781 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
5782 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
5783 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
5784 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
5785 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
5786 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
5787 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
5788 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
5789 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
5790 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
5791 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
5792 
5793 #undef GET_STAT
5794 }
5795 
5796 static void
5797 t4_clr_vi_stats(struct adapter *sc, u_int vin)
5798 {
5799 	int reg;
5800 
5801 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) | V_PL_VFID(vin) |
5802 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
5803 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
5804 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
5805 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
5806 }
5807 
5808 static void
5809 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
5810 {
5811 	struct timeval tv;
5812 	const struct timeval interval = {0, 250000};	/* 250ms */
5813 
5814 	if (!(vi->flags & VI_INIT_DONE))
5815 		return;
5816 
5817 	getmicrotime(&tv);
5818 	timevalsub(&tv, &interval);
5819 	if (timevalcmp(&tv, &vi->last_refreshed, <))
5820 		return;
5821 
5822 	mtx_lock(&sc->reg_lock);
5823 	t4_get_vi_stats(sc, vi->vin, &vi->stats);
5824 	getmicrotime(&vi->last_refreshed);
5825 	mtx_unlock(&sc->reg_lock);
5826 }
5827 
5828 static void
5829 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5830 {
5831 	u_int i, v, tnl_cong_drops, bg_map;
5832 	struct timeval tv;
5833 	const struct timeval interval = {0, 250000};	/* 250ms */
5834 
5835 	getmicrotime(&tv);
5836 	timevalsub(&tv, &interval);
5837 	if (timevalcmp(&tv, &pi->last_refreshed, <))
5838 		return;
5839 
5840 	tnl_cong_drops = 0;
5841 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5842 	bg_map = pi->mps_bg_map;
5843 	while (bg_map) {
5844 		i = ffs(bg_map) - 1;
5845 		mtx_lock(&sc->reg_lock);
5846 		t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5847 		    A_TP_MIB_TNL_CNG_DROP_0 + i);
5848 		mtx_unlock(&sc->reg_lock);
5849 		tnl_cong_drops += v;
5850 		bg_map &= ~(1 << i);
5851 	}
5852 	pi->tnl_cong_drops = tnl_cong_drops;
5853 	getmicrotime(&pi->last_refreshed);
5854 }
5855 
5856 static void
5857 cxgbe_tick(void *arg)
5858 {
5859 	struct port_info *pi = arg;
5860 	struct adapter *sc = pi->adapter;
5861 
5862 	PORT_LOCK_ASSERT_OWNED(pi);
5863 	cxgbe_refresh_stats(sc, pi);
5864 
5865 	callout_schedule(&pi->tick, hz);
5866 }
5867 
5868 void
5869 vi_tick(void *arg)
5870 {
5871 	struct vi_info *vi = arg;
5872 	struct adapter *sc = vi->pi->adapter;
5873 
5874 	vi_refresh_stats(sc, vi);
5875 
5876 	callout_schedule(&vi->tick, hz);
5877 }
5878 
5879 /*
5880  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5881  */
5882 static char *caps_decoder[] = {
5883 	"\20\001IPMI\002NCSI",				/* 0: NBM */
5884 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
5885 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
5886 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
5887 	    "\006HASHFILTER\007ETHOFLD",
5888 	"\20\001TOE",					/* 4: TOE */
5889 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
5890 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
5891 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5892 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5893 	    "\007T10DIF"
5894 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5895 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
5896 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
5897 		    "\004PO_INITIATOR\005PO_TARGET",
5898 };
5899 
5900 void
5901 t4_sysctls(struct adapter *sc)
5902 {
5903 	struct sysctl_ctx_list *ctx;
5904 	struct sysctl_oid *oid;
5905 	struct sysctl_oid_list *children, *c0;
5906 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5907 
5908 	ctx = device_get_sysctl_ctx(sc->dev);
5909 
5910 	/*
5911 	 * dev.t4nex.X.
5912 	 */
5913 	oid = device_get_sysctl_tree(sc->dev);
5914 	c0 = children = SYSCTL_CHILDREN(oid);
5915 
5916 	sc->sc_do_rxcopy = 1;
5917 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5918 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5919 
5920 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5921 	    sc->params.nports, "# of ports");
5922 
5923 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5924 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, (uintptr_t)&sc->doorbells,
5925 	    sysctl_bitfield_8b, "A", "available doorbells");
5926 
5927 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5928 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
5929 
5930 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5931 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5932 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5933 	    "interrupt holdoff timer values (us)");
5934 
5935 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5936 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5937 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5938 	    "interrupt holdoff packet counter values");
5939 
5940 	t4_sge_sysctls(sc, ctx, children);
5941 
5942 	sc->lro_timeout = 100;
5943 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5944 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5945 
5946 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5947 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
5948 
5949 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5950 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5951 
5952 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5953 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5954 
5955 	if (sc->flags & IS_VF)
5956 		return;
5957 
5958 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5959 	    NULL, chip_rev(sc), "chip hardware revision");
5960 
5961 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5962 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5963 
5964 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5965 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5966 
5967 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5968 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5969 
5970 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5971 	    CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5972 
5973 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5974 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5975 
5976 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5977 	    sc->er_version, 0, "expansion ROM version");
5978 
5979 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5980 	    sc->bs_version, 0, "bootstrap firmware version");
5981 
5982 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5983 	    NULL, sc->params.scfg_vers, "serial config version");
5984 
5985 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5986 	    NULL, sc->params.vpd_vers, "VPD version");
5987 
5988 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5989 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5990 
5991 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5992 	    sc->cfcsum, "config file checksum");
5993 
5994 #define SYSCTL_CAP(name, n, text) \
5995 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5996 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], (uintptr_t)&sc->name, \
5997 	    sysctl_bitfield_16b, "A", "available " text " capabilities")
5998 
5999 	SYSCTL_CAP(nbmcaps, 0, "NBM");
6000 	SYSCTL_CAP(linkcaps, 1, "link");
6001 	SYSCTL_CAP(switchcaps, 2, "switch");
6002 	SYSCTL_CAP(niccaps, 3, "NIC");
6003 	SYSCTL_CAP(toecaps, 4, "TCP offload");
6004 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
6005 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
6006 	SYSCTL_CAP(cryptocaps, 7, "crypto");
6007 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
6008 #undef SYSCTL_CAP
6009 
6010 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
6011 	    NULL, sc->tids.nftids, "number of filters");
6012 
6013 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
6014 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
6015 	    "chip temperature (in Celsius)");
6016 
6017 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "loadavg", CTLTYPE_STRING |
6018 	    CTLFLAG_RD, sc, 0, sysctl_loadavg, "A",
6019 	    "microprocessor load averages (debug firmwares only)");
6020 
6021 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
6022 	    &sc->params.core_vdd, 0, "core Vdd (in mV)");
6023 
6024 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "local_cpus",
6025 	    CTLTYPE_STRING | CTLFLAG_RD, sc, LOCAL_CPUS,
6026 	    sysctl_cpus, "A", "local CPUs");
6027 
6028 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "intr_cpus",
6029 	    CTLTYPE_STRING | CTLFLAG_RD, sc, INTR_CPUS,
6030 	    sysctl_cpus, "A", "preferred CPUs for interrupts");
6031 
6032 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "swintr", CTLFLAG_RW,
6033 	    &sc->swintr, 0, "software triggered interrupts");
6034 
6035 	/*
6036 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
6037 	 */
6038 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
6039 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
6040 	    "logs and miscellaneous information");
6041 	children = SYSCTL_CHILDREN(oid);
6042 
6043 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
6044 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6045 	    sysctl_cctrl, "A", "congestion control");
6046 
6047 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
6048 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6049 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
6050 
6051 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
6052 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
6053 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
6054 
6055 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
6056 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
6057 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
6058 
6059 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
6060 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
6061 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
6062 
6063 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
6064 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
6065 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
6066 
6067 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
6068 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
6069 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
6070 
6071 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
6072 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_cim_la,
6073 	    "A", "CIM logic analyzer");
6074 
6075 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
6076 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6077 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
6078 
6079 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
6080 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
6081 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
6082 
6083 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
6084 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
6085 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
6086 
6087 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
6088 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
6089 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
6090 
6091 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
6092 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
6093 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
6094 
6095 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
6096 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
6097 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
6098 
6099 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
6100 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
6101 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
6102 
6103 	if (chip_id(sc) > CHELSIO_T4) {
6104 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
6105 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
6106 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
6107 
6108 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
6109 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
6110 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
6111 	}
6112 
6113 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
6114 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6115 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
6116 
6117 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
6118 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6119 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
6120 
6121 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
6122 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6123 	    sysctl_cpl_stats, "A", "CPL statistics");
6124 
6125 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
6126 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6127 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
6128 
6129 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
6130 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6131 	    sysctl_devlog, "A", "firmware's device log");
6132 
6133 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
6134 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6135 	    sysctl_fcoe_stats, "A", "FCoE statistics");
6136 
6137 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
6138 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6139 	    sysctl_hw_sched, "A", "hardware scheduler ");
6140 
6141 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
6142 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6143 	    sysctl_l2t, "A", "hardware L2 table");
6144 
6145 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "smt",
6146 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6147 	    sysctl_smt, "A", "hardware source MAC table");
6148 
6149 #ifdef INET6
6150 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "clip",
6151 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6152 	    sysctl_clip, "A", "active CLIP table entries");
6153 #endif
6154 
6155 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
6156 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6157 	    sysctl_lb_stats, "A", "loopback statistics");
6158 
6159 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
6160 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6161 	    sysctl_meminfo, "A", "memory regions");
6162 
6163 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
6164 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6165 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
6166 	    "A", "MPS TCAM entries");
6167 
6168 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
6169 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6170 	    sysctl_path_mtus, "A", "path MTUs");
6171 
6172 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
6173 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6174 	    sysctl_pm_stats, "A", "PM statistics");
6175 
6176 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
6177 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6178 	    sysctl_rdma_stats, "A", "RDMA statistics");
6179 
6180 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
6181 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6182 	    sysctl_tcp_stats, "A", "TCP statistics");
6183 
6184 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
6185 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6186 	    sysctl_tids, "A", "TID information");
6187 
6188 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
6189 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6190 	    sysctl_tp_err_stats, "A", "TP error statistics");
6191 
6192 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
6193 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
6194 	    "TP logic analyzer event capture mask");
6195 
6196 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
6197 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6198 	    sysctl_tp_la, "A", "TP logic analyzer");
6199 
6200 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
6201 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6202 	    sysctl_tx_rate, "A", "Tx rate");
6203 
6204 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
6205 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6206 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
6207 
6208 	if (chip_id(sc) >= CHELSIO_T5) {
6209 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
6210 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6211 		    sysctl_wcwr_stats, "A", "write combined work requests");
6212 	}
6213 
6214 #ifdef TCP_OFFLOAD
6215 	if (is_offload(sc)) {
6216 		int i;
6217 		char s[4];
6218 
6219 		/*
6220 		 * dev.t4nex.X.toe.
6221 		 */
6222 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
6223 		    NULL, "TOE parameters");
6224 		children = SYSCTL_CHILDREN(oid);
6225 
6226 		sc->tt.cong_algorithm = -1;
6227 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
6228 		    CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
6229 		    "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
6230 		    "3 = highspeed)");
6231 
6232 		sc->tt.sndbuf = 256 * 1024;
6233 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
6234 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
6235 
6236 		sc->tt.ddp = 0;
6237 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp",
6238 		    CTLFLAG_RW | CTLFLAG_SKIP, &sc->tt.ddp, 0, "");
6239 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_zcopy", CTLFLAG_RW,
6240 		    &sc->tt.ddp, 0, "Enable zero-copy aio_read(2)");
6241 
6242 		sc->tt.rx_coalesce = 1;
6243 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
6244 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
6245 
6246 		sc->tt.tls = 0;
6247 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tls", CTLFLAG_RW,
6248 		    &sc->tt.tls, 0, "Inline TLS allowed");
6249 
6250 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls_rx_ports",
6251 		    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tls_rx_ports,
6252 		    "I", "TCP ports that use inline TLS+TOE RX");
6253 
6254 		sc->tt.tx_align = 1;
6255 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
6256 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
6257 
6258 		sc->tt.tx_zcopy = 0;
6259 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
6260 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
6261 		    "Enable zero-copy aio_write(2)");
6262 
6263 		sc->tt.cop_managed_offloading = !!t4_cop_managed_offloading;
6264 		SYSCTL_ADD_INT(ctx, children, OID_AUTO,
6265 		    "cop_managed_offloading", CTLFLAG_RW,
6266 		    &sc->tt.cop_managed_offloading, 0,
6267 		    "COP (Connection Offload Policy) controls all TOE offload");
6268 
6269 		sc->tt.autorcvbuf_inc = 16 * 1024;
6270 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "autorcvbuf_inc",
6271 		    CTLFLAG_RW, &sc->tt.autorcvbuf_inc, 0,
6272 		    "autorcvbuf increment");
6273 
6274 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
6275 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
6276 		    "TP timer tick (us)");
6277 
6278 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
6279 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
6280 		    "TCP timestamp tick (us)");
6281 
6282 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
6283 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
6284 		    "DACK tick (us)");
6285 
6286 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
6287 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
6288 		    "IU", "DACK timer (us)");
6289 
6290 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
6291 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
6292 		    sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
6293 
6294 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
6295 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
6296 		    sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
6297 
6298 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
6299 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
6300 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
6301 
6302 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
6303 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
6304 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
6305 
6306 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
6307 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
6308 		    sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
6309 
6310 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
6311 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
6312 		    sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
6313 
6314 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
6315 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
6316 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
6317 
6318 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
6319 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
6320 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
6321 
6322 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
6323 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
6324 		    sysctl_tp_shift_cnt, "IU",
6325 		    "Number of SYN retransmissions before abort");
6326 
6327 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
6328 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
6329 		    sysctl_tp_shift_cnt, "IU",
6330 		    "Number of retransmissions before abort");
6331 
6332 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
6333 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
6334 		    sysctl_tp_shift_cnt, "IU",
6335 		    "Number of keepalive probes before abort");
6336 
6337 		oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
6338 		    CTLFLAG_RD, NULL, "TOE retransmit backoffs");
6339 		children = SYSCTL_CHILDREN(oid);
6340 		for (i = 0; i < 16; i++) {
6341 			snprintf(s, sizeof(s), "%u", i);
6342 			SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
6343 			    CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
6344 			    "IU", "TOE retransmit backoff");
6345 		}
6346 	}
6347 #endif
6348 }
6349 
6350 void
6351 vi_sysctls(struct vi_info *vi)
6352 {
6353 	struct sysctl_ctx_list *ctx;
6354 	struct sysctl_oid *oid;
6355 	struct sysctl_oid_list *children;
6356 
6357 	ctx = device_get_sysctl_ctx(vi->dev);
6358 
6359 	/*
6360 	 * dev.v?(cxgbe|cxl).X.
6361 	 */
6362 	oid = device_get_sysctl_tree(vi->dev);
6363 	children = SYSCTL_CHILDREN(oid);
6364 
6365 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
6366 	    vi->viid, "VI identifer");
6367 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
6368 	    &vi->nrxq, 0, "# of rx queues");
6369 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
6370 	    &vi->ntxq, 0, "# of tx queues");
6371 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
6372 	    &vi->first_rxq, 0, "index of first rx queue");
6373 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
6374 	    &vi->first_txq, 0, "index of first tx queue");
6375 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_base", CTLFLAG_RD, NULL,
6376 	    vi->rss_base, "start of RSS indirection table");
6377 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
6378 	    vi->rss_size, "size of RSS indirection table");
6379 
6380 	if (IS_MAIN_VI(vi)) {
6381 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
6382 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
6383 		    "Reserve queue 0 for non-flowid packets");
6384 	}
6385 
6386 #ifdef TCP_OFFLOAD
6387 	if (vi->nofldrxq != 0) {
6388 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
6389 		    &vi->nofldrxq, 0,
6390 		    "# of rx queues for offloaded TCP connections");
6391 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
6392 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
6393 		    "index of first TOE rx queue");
6394 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
6395 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
6396 		    sysctl_holdoff_tmr_idx_ofld, "I",
6397 		    "holdoff timer index for TOE queues");
6398 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
6399 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
6400 		    sysctl_holdoff_pktc_idx_ofld, "I",
6401 		    "holdoff packet counter index for TOE queues");
6402 	}
6403 #endif
6404 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
6405 	if (vi->nofldtxq != 0) {
6406 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
6407 		    &vi->nofldtxq, 0,
6408 		    "# of tx queues for TOE/ETHOFLD");
6409 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
6410 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
6411 		    "index of first TOE/ETHOFLD tx queue");
6412 	}
6413 #endif
6414 #ifdef DEV_NETMAP
6415 	if (vi->nnmrxq != 0) {
6416 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
6417 		    &vi->nnmrxq, 0, "# of netmap rx queues");
6418 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
6419 		    &vi->nnmtxq, 0, "# of netmap tx queues");
6420 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
6421 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
6422 		    "index of first netmap rx queue");
6423 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
6424 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
6425 		    "index of first netmap tx queue");
6426 	}
6427 #endif
6428 
6429 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
6430 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
6431 	    "holdoff timer index");
6432 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
6433 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
6434 	    "holdoff packet counter index");
6435 
6436 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
6437 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
6438 	    "rx queue size");
6439 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
6440 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
6441 	    "tx queue size");
6442 }
6443 
6444 static void
6445 cxgbe_sysctls(struct port_info *pi)
6446 {
6447 	struct sysctl_ctx_list *ctx;
6448 	struct sysctl_oid *oid;
6449 	struct sysctl_oid_list *children, *children2;
6450 	struct adapter *sc = pi->adapter;
6451 	int i;
6452 	char name[16];
6453 	static char *tc_flags = {"\20\1USER\2SYNC\3ASYNC\4ERR"};
6454 
6455 	ctx = device_get_sysctl_ctx(pi->dev);
6456 
6457 	/*
6458 	 * dev.cxgbe.X.
6459 	 */
6460 	oid = device_get_sysctl_tree(pi->dev);
6461 	children = SYSCTL_CHILDREN(oid);
6462 
6463 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
6464 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
6465 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
6466 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
6467 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
6468 		    "PHY temperature (in Celsius)");
6469 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
6470 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
6471 		    "PHY firmware version");
6472 	}
6473 
6474 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
6475 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
6476     "PAUSE settings (bit 0 = rx_pause, 1 = tx_pause, 2 = pause_autoneg)");
6477 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
6478 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
6479 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
6480 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
6481 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
6482 	    "autonegotiation (-1 = not supported)");
6483 
6484 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
6485 	    port_top_speed(pi), "max speed (in Gbps)");
6486 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
6487 	    pi->mps_bg_map, "MPS buffer group map");
6488 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
6489 	    NULL, pi->rx_e_chan_map, "TP rx e-channel map");
6490 
6491 	if (sc->flags & IS_VF)
6492 		return;
6493 
6494 	/*
6495 	 * dev.(cxgbe|cxl).X.tc.
6496 	 */
6497 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
6498 	    "Tx scheduler traffic classes (cl_rl)");
6499 	children2 = SYSCTL_CHILDREN(oid);
6500 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "pktsize",
6501 	    CTLFLAG_RW, &pi->sched_params->pktsize, 0,
6502 	    "pktsize for per-flow cl-rl (0 means up to the driver )");
6503 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "burstsize",
6504 	    CTLFLAG_RW, &pi->sched_params->burstsize, 0,
6505 	    "burstsize for per-flow cl-rl (0 means up to the driver)");
6506 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
6507 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
6508 
6509 		snprintf(name, sizeof(name), "%d", i);
6510 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
6511 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
6512 		    "traffic class"));
6513 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "flags",
6514 		    CTLTYPE_STRING | CTLFLAG_RD, tc_flags, (uintptr_t)&tc->flags,
6515 		    sysctl_bitfield_8b, "A", "flags");
6516 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
6517 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
6518 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
6519 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
6520 		    sysctl_tc_params, "A", "traffic class parameters");
6521 	}
6522 
6523 	/*
6524 	 * dev.cxgbe.X.stats.
6525 	 */
6526 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
6527 	    NULL, "port statistics");
6528 	children = SYSCTL_CHILDREN(oid);
6529 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
6530 	    &pi->tx_parse_error, 0,
6531 	    "# of tx packets with invalid length or # of segments");
6532 
6533 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
6534 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
6535 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
6536 	    sysctl_handle_t4_reg64, "QU", desc)
6537 
6538 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
6539 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
6540 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
6541 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
6542 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
6543 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
6544 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
6545 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
6546 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
6547 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
6548 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
6549 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
6550 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
6551 	    "# of tx frames in this range",
6552 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
6553 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
6554 	    "# of tx frames in this range",
6555 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
6556 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
6557 	    "# of tx frames in this range",
6558 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
6559 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
6560 	    "# of tx frames in this range",
6561 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
6562 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
6563 	    "# of tx frames in this range",
6564 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
6565 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
6566 	    "# of tx frames in this range",
6567 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
6568 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
6569 	    "# of tx frames in this range",
6570 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
6571 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
6572 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
6573 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
6574 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
6575 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
6576 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
6577 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
6578 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
6579 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
6580 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
6581 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
6582 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
6583 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
6584 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
6585 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
6586 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
6587 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
6588 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
6589 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
6590 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
6591 
6592 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
6593 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
6594 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
6595 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
6596 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
6597 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
6598 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
6599 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
6600 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
6601 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
6602 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
6603 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
6604 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
6605 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
6606 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
6607 	    "# of frames received with bad FCS",
6608 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
6609 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
6610 	    "# of frames received with length error",
6611 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
6612 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
6613 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
6614 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
6615 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
6616 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
6617 	    "# of rx frames in this range",
6618 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
6619 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
6620 	    "# of rx frames in this range",
6621 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
6622 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
6623 	    "# of rx frames in this range",
6624 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
6625 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
6626 	    "# of rx frames in this range",
6627 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
6628 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
6629 	    "# of rx frames in this range",
6630 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
6631 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
6632 	    "# of rx frames in this range",
6633 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
6634 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
6635 	    "# of rx frames in this range",
6636 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
6637 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
6638 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
6639 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
6640 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
6641 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
6642 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
6643 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
6644 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
6645 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
6646 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
6647 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
6648 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
6649 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
6650 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
6651 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
6652 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
6653 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
6654 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
6655 
6656 #undef SYSCTL_ADD_T4_REG64
6657 
6658 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
6659 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
6660 	    &pi->stats.name, desc)
6661 
6662 	/* We get these from port_stats and they may be stale by up to 1s */
6663 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
6664 	    "# drops due to buffer-group 0 overflows");
6665 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
6666 	    "# drops due to buffer-group 1 overflows");
6667 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
6668 	    "# drops due to buffer-group 2 overflows");
6669 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
6670 	    "# drops due to buffer-group 3 overflows");
6671 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
6672 	    "# of buffer-group 0 truncated packets");
6673 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
6674 	    "# of buffer-group 1 truncated packets");
6675 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
6676 	    "# of buffer-group 2 truncated packets");
6677 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
6678 	    "# of buffer-group 3 truncated packets");
6679 
6680 #undef SYSCTL_ADD_T4_PORTSTAT
6681 
6682 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_records",
6683 	    CTLFLAG_RD, &pi->tx_tls_records,
6684 	    "# of TLS records transmitted");
6685 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_octets",
6686 	    CTLFLAG_RD, &pi->tx_tls_octets,
6687 	    "# of payload octets in transmitted TLS records");
6688 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_records",
6689 	    CTLFLAG_RD, &pi->rx_tls_records,
6690 	    "# of TLS records received");
6691 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_octets",
6692 	    CTLFLAG_RD, &pi->rx_tls_octets,
6693 	    "# of payload octets in received TLS records");
6694 }
6695 
6696 static int
6697 sysctl_int_array(SYSCTL_HANDLER_ARGS)
6698 {
6699 	int rc, *i, space = 0;
6700 	struct sbuf sb;
6701 
6702 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
6703 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
6704 		if (space)
6705 			sbuf_printf(&sb, " ");
6706 		sbuf_printf(&sb, "%d", *i);
6707 		space = 1;
6708 	}
6709 	rc = sbuf_finish(&sb);
6710 	sbuf_delete(&sb);
6711 	return (rc);
6712 }
6713 
6714 static int
6715 sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS)
6716 {
6717 	int rc;
6718 	struct sbuf *sb;
6719 
6720 	rc = sysctl_wire_old_buffer(req, 0);
6721 	if (rc != 0)
6722 		return(rc);
6723 
6724 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6725 	if (sb == NULL)
6726 		return (ENOMEM);
6727 
6728 	sbuf_printf(sb, "%b", *(uint8_t *)(uintptr_t)arg2, (char *)arg1);
6729 	rc = sbuf_finish(sb);
6730 	sbuf_delete(sb);
6731 
6732 	return (rc);
6733 }
6734 
6735 static int
6736 sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS)
6737 {
6738 	int rc;
6739 	struct sbuf *sb;
6740 
6741 	rc = sysctl_wire_old_buffer(req, 0);
6742 	if (rc != 0)
6743 		return(rc);
6744 
6745 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6746 	if (sb == NULL)
6747 		return (ENOMEM);
6748 
6749 	sbuf_printf(sb, "%b", *(uint16_t *)(uintptr_t)arg2, (char *)arg1);
6750 	rc = sbuf_finish(sb);
6751 	sbuf_delete(sb);
6752 
6753 	return (rc);
6754 }
6755 
6756 static int
6757 sysctl_btphy(SYSCTL_HANDLER_ARGS)
6758 {
6759 	struct port_info *pi = arg1;
6760 	int op = arg2;
6761 	struct adapter *sc = pi->adapter;
6762 	u_int v;
6763 	int rc;
6764 
6765 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
6766 	if (rc)
6767 		return (rc);
6768 	/* XXX: magic numbers */
6769 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
6770 	    &v);
6771 	end_synchronized_op(sc, 0);
6772 	if (rc)
6773 		return (rc);
6774 	if (op == 0)
6775 		v /= 256;
6776 
6777 	rc = sysctl_handle_int(oidp, &v, 0, req);
6778 	return (rc);
6779 }
6780 
6781 static int
6782 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
6783 {
6784 	struct vi_info *vi = arg1;
6785 	int rc, val;
6786 
6787 	val = vi->rsrv_noflowq;
6788 	rc = sysctl_handle_int(oidp, &val, 0, req);
6789 	if (rc != 0 || req->newptr == NULL)
6790 		return (rc);
6791 
6792 	if ((val >= 1) && (vi->ntxq > 1))
6793 		vi->rsrv_noflowq = 1;
6794 	else
6795 		vi->rsrv_noflowq = 0;
6796 
6797 	return (rc);
6798 }
6799 
6800 static int
6801 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
6802 {
6803 	struct vi_info *vi = arg1;
6804 	struct adapter *sc = vi->pi->adapter;
6805 	int idx, rc, i;
6806 	struct sge_rxq *rxq;
6807 	uint8_t v;
6808 
6809 	idx = vi->tmr_idx;
6810 
6811 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6812 	if (rc != 0 || req->newptr == NULL)
6813 		return (rc);
6814 
6815 	if (idx < 0 || idx >= SGE_NTIMERS)
6816 		return (EINVAL);
6817 
6818 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6819 	    "t4tmr");
6820 	if (rc)
6821 		return (rc);
6822 
6823 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
6824 	for_each_rxq(vi, i, rxq) {
6825 #ifdef atomic_store_rel_8
6826 		atomic_store_rel_8(&rxq->iq.intr_params, v);
6827 #else
6828 		rxq->iq.intr_params = v;
6829 #endif
6830 	}
6831 	vi->tmr_idx = idx;
6832 
6833 	end_synchronized_op(sc, LOCK_HELD);
6834 	return (0);
6835 }
6836 
6837 static int
6838 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
6839 {
6840 	struct vi_info *vi = arg1;
6841 	struct adapter *sc = vi->pi->adapter;
6842 	int idx, rc;
6843 
6844 	idx = vi->pktc_idx;
6845 
6846 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6847 	if (rc != 0 || req->newptr == NULL)
6848 		return (rc);
6849 
6850 	if (idx < -1 || idx >= SGE_NCOUNTERS)
6851 		return (EINVAL);
6852 
6853 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6854 	    "t4pktc");
6855 	if (rc)
6856 		return (rc);
6857 
6858 	if (vi->flags & VI_INIT_DONE)
6859 		rc = EBUSY; /* cannot be changed once the queues are created */
6860 	else
6861 		vi->pktc_idx = idx;
6862 
6863 	end_synchronized_op(sc, LOCK_HELD);
6864 	return (rc);
6865 }
6866 
6867 static int
6868 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
6869 {
6870 	struct vi_info *vi = arg1;
6871 	struct adapter *sc = vi->pi->adapter;
6872 	int qsize, rc;
6873 
6874 	qsize = vi->qsize_rxq;
6875 
6876 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6877 	if (rc != 0 || req->newptr == NULL)
6878 		return (rc);
6879 
6880 	if (qsize < 128 || (qsize & 7))
6881 		return (EINVAL);
6882 
6883 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6884 	    "t4rxqs");
6885 	if (rc)
6886 		return (rc);
6887 
6888 	if (vi->flags & VI_INIT_DONE)
6889 		rc = EBUSY; /* cannot be changed once the queues are created */
6890 	else
6891 		vi->qsize_rxq = qsize;
6892 
6893 	end_synchronized_op(sc, LOCK_HELD);
6894 	return (rc);
6895 }
6896 
6897 static int
6898 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
6899 {
6900 	struct vi_info *vi = arg1;
6901 	struct adapter *sc = vi->pi->adapter;
6902 	int qsize, rc;
6903 
6904 	qsize = vi->qsize_txq;
6905 
6906 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6907 	if (rc != 0 || req->newptr == NULL)
6908 		return (rc);
6909 
6910 	if (qsize < 128 || qsize > 65536)
6911 		return (EINVAL);
6912 
6913 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6914 	    "t4txqs");
6915 	if (rc)
6916 		return (rc);
6917 
6918 	if (vi->flags & VI_INIT_DONE)
6919 		rc = EBUSY; /* cannot be changed once the queues are created */
6920 	else
6921 		vi->qsize_txq = qsize;
6922 
6923 	end_synchronized_op(sc, LOCK_HELD);
6924 	return (rc);
6925 }
6926 
6927 static int
6928 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
6929 {
6930 	struct port_info *pi = arg1;
6931 	struct adapter *sc = pi->adapter;
6932 	struct link_config *lc = &pi->link_cfg;
6933 	int rc;
6934 
6935 	if (req->newptr == NULL) {
6936 		struct sbuf *sb;
6937 		static char *bits = "\20\1RX\2TX\3AUTO";
6938 
6939 		rc = sysctl_wire_old_buffer(req, 0);
6940 		if (rc != 0)
6941 			return(rc);
6942 
6943 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6944 		if (sb == NULL)
6945 			return (ENOMEM);
6946 
6947 		if (lc->link_ok) {
6948 			sbuf_printf(sb, "%b", (lc->fc & (PAUSE_TX | PAUSE_RX)) |
6949 			    (lc->requested_fc & PAUSE_AUTONEG), bits);
6950 		} else {
6951 			sbuf_printf(sb, "%b", lc->requested_fc & (PAUSE_TX |
6952 			    PAUSE_RX | PAUSE_AUTONEG), bits);
6953 		}
6954 		rc = sbuf_finish(sb);
6955 		sbuf_delete(sb);
6956 	} else {
6957 		char s[2];
6958 		int n;
6959 
6960 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX |
6961 		    PAUSE_AUTONEG));
6962 		s[1] = 0;
6963 
6964 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6965 		if (rc != 0)
6966 			return(rc);
6967 
6968 		if (s[1] != 0)
6969 			return (EINVAL);
6970 		if (s[0] < '0' || s[0] > '9')
6971 			return (EINVAL);	/* not a number */
6972 		n = s[0] - '0';
6973 		if (n & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG))
6974 			return (EINVAL);	/* some other bit is set too */
6975 
6976 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6977 		    "t4PAUSE");
6978 		if (rc)
6979 			return (rc);
6980 		PORT_LOCK(pi);
6981 		lc->requested_fc = n;
6982 		fixup_link_config(pi);
6983 		if (pi->up_vis > 0)
6984 			rc = apply_link_config(pi);
6985 		set_current_media(pi);
6986 		PORT_UNLOCK(pi);
6987 		end_synchronized_op(sc, 0);
6988 	}
6989 
6990 	return (rc);
6991 }
6992 
6993 static int
6994 sysctl_fec(SYSCTL_HANDLER_ARGS)
6995 {
6996 	struct port_info *pi = arg1;
6997 	struct adapter *sc = pi->adapter;
6998 	struct link_config *lc = &pi->link_cfg;
6999 	int rc;
7000 	int8_t old;
7001 
7002 	if (req->newptr == NULL) {
7003 		struct sbuf *sb;
7004 		static char *bits = "\20\1RS\2BASE-R\3RSVD1\4RSVD2\5RSVD3\6AUTO";
7005 
7006 		rc = sysctl_wire_old_buffer(req, 0);
7007 		if (rc != 0)
7008 			return(rc);
7009 
7010 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
7011 		if (sb == NULL)
7012 			return (ENOMEM);
7013 
7014 		/*
7015 		 * Display the requested_fec when the link is down -- the actual
7016 		 * FEC makes sense only when the link is up.
7017 		 */
7018 		if (lc->link_ok) {
7019 			sbuf_printf(sb, "%b", (lc->fec & M_FW_PORT_CAP32_FEC) |
7020 			    (lc->requested_fec & FEC_AUTO), bits);
7021 		} else {
7022 			sbuf_printf(sb, "%b", lc->requested_fec, bits);
7023 		}
7024 		rc = sbuf_finish(sb);
7025 		sbuf_delete(sb);
7026 	} else {
7027 		char s[3];
7028 		int n;
7029 
7030 		snprintf(s, sizeof(s), "%d",
7031 		    lc->requested_fec == FEC_AUTO ? -1 :
7032 		    lc->requested_fec & M_FW_PORT_CAP32_FEC);
7033 
7034 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
7035 		if (rc != 0)
7036 			return(rc);
7037 
7038 		n = strtol(&s[0], NULL, 0);
7039 		if (n < 0 || n & FEC_AUTO)
7040 			n = FEC_AUTO;
7041 		else {
7042 			if (n & ~M_FW_PORT_CAP32_FEC)
7043 				return (EINVAL);/* some other bit is set too */
7044 			if (!powerof2(n))
7045 				return (EINVAL);/* one bit can be set at most */
7046 		}
7047 
7048 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
7049 		    "t4fec");
7050 		if (rc)
7051 			return (rc);
7052 		PORT_LOCK(pi);
7053 		old = lc->requested_fec;
7054 		if (n == FEC_AUTO)
7055 			lc->requested_fec = FEC_AUTO;
7056 		else if (n == 0)
7057 			lc->requested_fec = FEC_NONE;
7058 		else {
7059 			if ((lc->supported | V_FW_PORT_CAP32_FEC(n)) !=
7060 			    lc->supported) {
7061 				rc = ENOTSUP;
7062 				goto done;
7063 			}
7064 			lc->requested_fec = n;
7065 		}
7066 		fixup_link_config(pi);
7067 		if (pi->up_vis > 0) {
7068 			rc = apply_link_config(pi);
7069 			if (rc != 0) {
7070 				lc->requested_fec = old;
7071 				if (rc == FW_EPROTO)
7072 					rc = ENOTSUP;
7073 			}
7074 		}
7075 done:
7076 		PORT_UNLOCK(pi);
7077 		end_synchronized_op(sc, 0);
7078 	}
7079 
7080 	return (rc);
7081 }
7082 
7083 static int
7084 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
7085 {
7086 	struct port_info *pi = arg1;
7087 	struct adapter *sc = pi->adapter;
7088 	struct link_config *lc = &pi->link_cfg;
7089 	int rc, val;
7090 
7091 	if (lc->supported & FW_PORT_CAP32_ANEG)
7092 		val = lc->requested_aneg == AUTONEG_DISABLE ? 0 : 1;
7093 	else
7094 		val = -1;
7095 	rc = sysctl_handle_int(oidp, &val, 0, req);
7096 	if (rc != 0 || req->newptr == NULL)
7097 		return (rc);
7098 	if (val == 0)
7099 		val = AUTONEG_DISABLE;
7100 	else if (val == 1)
7101 		val = AUTONEG_ENABLE;
7102 	else
7103 		val = AUTONEG_AUTO;
7104 
7105 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
7106 	    "t4aneg");
7107 	if (rc)
7108 		return (rc);
7109 	PORT_LOCK(pi);
7110 	if (val == AUTONEG_ENABLE && !(lc->supported & FW_PORT_CAP32_ANEG)) {
7111 		rc = ENOTSUP;
7112 		goto done;
7113 	}
7114 	lc->requested_aneg = val;
7115 	fixup_link_config(pi);
7116 	if (pi->up_vis > 0)
7117 		rc = apply_link_config(pi);
7118 	set_current_media(pi);
7119 done:
7120 	PORT_UNLOCK(pi);
7121 	end_synchronized_op(sc, 0);
7122 	return (rc);
7123 }
7124 
7125 static int
7126 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
7127 {
7128 	struct adapter *sc = arg1;
7129 	int reg = arg2;
7130 	uint64_t val;
7131 
7132 	val = t4_read_reg64(sc, reg);
7133 
7134 	return (sysctl_handle_64(oidp, &val, 0, req));
7135 }
7136 
7137 static int
7138 sysctl_temperature(SYSCTL_HANDLER_ARGS)
7139 {
7140 	struct adapter *sc = arg1;
7141 	int rc, t;
7142 	uint32_t param, val;
7143 
7144 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
7145 	if (rc)
7146 		return (rc);
7147 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
7148 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
7149 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
7150 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
7151 	end_synchronized_op(sc, 0);
7152 	if (rc)
7153 		return (rc);
7154 
7155 	/* unknown is returned as 0 but we display -1 in that case */
7156 	t = val == 0 ? -1 : val;
7157 
7158 	rc = sysctl_handle_int(oidp, &t, 0, req);
7159 	return (rc);
7160 }
7161 
7162 static int
7163 sysctl_loadavg(SYSCTL_HANDLER_ARGS)
7164 {
7165 	struct adapter *sc = arg1;
7166 	struct sbuf *sb;
7167 	int rc;
7168 	uint32_t param, val;
7169 
7170 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4lavg");
7171 	if (rc)
7172 		return (rc);
7173 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
7174 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_LOAD);
7175 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
7176 	end_synchronized_op(sc, 0);
7177 	if (rc)
7178 		return (rc);
7179 
7180 	rc = sysctl_wire_old_buffer(req, 0);
7181 	if (rc != 0)
7182 		return (rc);
7183 
7184 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7185 	if (sb == NULL)
7186 		return (ENOMEM);
7187 
7188 	if (val == 0xffffffff) {
7189 		/* Only debug and custom firmwares report load averages. */
7190 		sbuf_printf(sb, "not available");
7191 	} else {
7192 		sbuf_printf(sb, "%d %d %d", val & 0xff, (val >> 8) & 0xff,
7193 		    (val >> 16) & 0xff);
7194 	}
7195 	rc = sbuf_finish(sb);
7196 	sbuf_delete(sb);
7197 
7198 	return (rc);
7199 }
7200 
7201 static int
7202 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
7203 {
7204 	struct adapter *sc = arg1;
7205 	struct sbuf *sb;
7206 	int rc, i;
7207 	uint16_t incr[NMTUS][NCCTRL_WIN];
7208 	static const char *dec_fac[] = {
7209 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
7210 		"0.9375"
7211 	};
7212 
7213 	rc = sysctl_wire_old_buffer(req, 0);
7214 	if (rc != 0)
7215 		return (rc);
7216 
7217 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7218 	if (sb == NULL)
7219 		return (ENOMEM);
7220 
7221 	t4_read_cong_tbl(sc, incr);
7222 
7223 	for (i = 0; i < NCCTRL_WIN; ++i) {
7224 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
7225 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
7226 		    incr[5][i], incr[6][i], incr[7][i]);
7227 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
7228 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
7229 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
7230 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
7231 	}
7232 
7233 	rc = sbuf_finish(sb);
7234 	sbuf_delete(sb);
7235 
7236 	return (rc);
7237 }
7238 
7239 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
7240 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
7241 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
7242 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
7243 };
7244 
7245 static int
7246 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
7247 {
7248 	struct adapter *sc = arg1;
7249 	struct sbuf *sb;
7250 	int rc, i, n, qid = arg2;
7251 	uint32_t *buf, *p;
7252 	char *qtype;
7253 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
7254 
7255 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
7256 	    ("%s: bad qid %d\n", __func__, qid));
7257 
7258 	if (qid < CIM_NUM_IBQ) {
7259 		/* inbound queue */
7260 		qtype = "IBQ";
7261 		n = 4 * CIM_IBQ_SIZE;
7262 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
7263 		rc = t4_read_cim_ibq(sc, qid, buf, n);
7264 	} else {
7265 		/* outbound queue */
7266 		qtype = "OBQ";
7267 		qid -= CIM_NUM_IBQ;
7268 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
7269 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
7270 		rc = t4_read_cim_obq(sc, qid, buf, n);
7271 	}
7272 
7273 	if (rc < 0) {
7274 		rc = -rc;
7275 		goto done;
7276 	}
7277 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
7278 
7279 	rc = sysctl_wire_old_buffer(req, 0);
7280 	if (rc != 0)
7281 		goto done;
7282 
7283 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7284 	if (sb == NULL) {
7285 		rc = ENOMEM;
7286 		goto done;
7287 	}
7288 
7289 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
7290 	for (i = 0, p = buf; i < n; i += 16, p += 4)
7291 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
7292 		    p[2], p[3]);
7293 
7294 	rc = sbuf_finish(sb);
7295 	sbuf_delete(sb);
7296 done:
7297 	free(buf, M_CXGBE);
7298 	return (rc);
7299 }
7300 
7301 static void
7302 sbuf_cim_la4(struct adapter *sc, struct sbuf *sb, uint32_t *buf, uint32_t cfg)
7303 {
7304 	uint32_t *p;
7305 
7306 	sbuf_printf(sb, "Status   Data      PC%s",
7307 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
7308 	    "     LS0Stat  LS0Addr             LS0Data");
7309 
7310 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
7311 		if (cfg & F_UPDBGLACAPTPCONLY) {
7312 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
7313 			    p[6], p[7]);
7314 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
7315 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
7316 			    p[4] & 0xff, p[5] >> 8);
7317 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
7318 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7319 			    p[1] & 0xf, p[2] >> 4);
7320 		} else {
7321 			sbuf_printf(sb,
7322 			    "\n  %02x   %x%07x %x%07x %08x %08x "
7323 			    "%08x%08x%08x%08x",
7324 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7325 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
7326 			    p[6], p[7]);
7327 		}
7328 	}
7329 }
7330 
7331 static void
7332 sbuf_cim_la6(struct adapter *sc, struct sbuf *sb, uint32_t *buf, uint32_t cfg)
7333 {
7334 	uint32_t *p;
7335 
7336 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
7337 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
7338 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
7339 
7340 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
7341 		if (cfg & F_UPDBGLACAPTPCONLY) {
7342 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
7343 			    p[3] & 0xff, p[2], p[1], p[0]);
7344 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
7345 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
7346 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
7347 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
7348 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
7349 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
7350 			    p[6] >> 16);
7351 		} else {
7352 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
7353 			    "%08x %08x %08x %08x %08x %08x",
7354 			    (p[9] >> 16) & 0xff,
7355 			    p[9] & 0xffff, p[8] >> 16,
7356 			    p[8] & 0xffff, p[7] >> 16,
7357 			    p[7] & 0xffff, p[6] >> 16,
7358 			    p[2], p[1], p[0], p[5], p[4], p[3]);
7359 		}
7360 	}
7361 }
7362 
7363 static int
7364 sbuf_cim_la(struct adapter *sc, struct sbuf *sb, int flags)
7365 {
7366 	uint32_t cfg, *buf;
7367 	int rc;
7368 
7369 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
7370 	if (rc != 0)
7371 		return (rc);
7372 
7373 	MPASS(flags == M_WAITOK || flags == M_NOWAIT);
7374 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
7375 	    M_ZERO | flags);
7376 	if (buf == NULL)
7377 		return (ENOMEM);
7378 
7379 	rc = -t4_cim_read_la(sc, buf, NULL);
7380 	if (rc != 0)
7381 		goto done;
7382 	if (chip_id(sc) < CHELSIO_T6)
7383 		sbuf_cim_la4(sc, sb, buf, cfg);
7384 	else
7385 		sbuf_cim_la6(sc, sb, buf, cfg);
7386 
7387 done:
7388 	free(buf, M_CXGBE);
7389 	return (rc);
7390 }
7391 
7392 static int
7393 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
7394 {
7395 	struct adapter *sc = arg1;
7396 	struct sbuf *sb;
7397 	int rc;
7398 
7399 	rc = sysctl_wire_old_buffer(req, 0);
7400 	if (rc != 0)
7401 		return (rc);
7402 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7403 	if (sb == NULL)
7404 		return (ENOMEM);
7405 
7406 	rc = sbuf_cim_la(sc, sb, M_WAITOK);
7407 	if (rc == 0)
7408 		rc = sbuf_finish(sb);
7409 	sbuf_delete(sb);
7410 	return (rc);
7411 }
7412 
7413 bool
7414 t4_os_dump_cimla(struct adapter *sc, int arg, bool verbose)
7415 {
7416 	struct sbuf sb;
7417 	int rc;
7418 
7419 	if (sbuf_new(&sb, NULL, 4096, SBUF_AUTOEXTEND) != &sb)
7420 		return (false);
7421 	rc = sbuf_cim_la(sc, &sb, M_NOWAIT);
7422 	if (rc == 0) {
7423 		rc = sbuf_finish(&sb);
7424 		if (rc == 0) {
7425 			log(LOG_DEBUG, "%s: CIM LA dump follows.\n%s",
7426 		    		device_get_nameunit(sc->dev), sbuf_data(&sb));
7427 		}
7428 	}
7429 	sbuf_delete(&sb);
7430 	return (false);
7431 }
7432 
7433 static int
7434 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
7435 {
7436 	struct adapter *sc = arg1;
7437 	u_int i;
7438 	struct sbuf *sb;
7439 	uint32_t *buf, *p;
7440 	int rc;
7441 
7442 	rc = sysctl_wire_old_buffer(req, 0);
7443 	if (rc != 0)
7444 		return (rc);
7445 
7446 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7447 	if (sb == NULL)
7448 		return (ENOMEM);
7449 
7450 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
7451 	    M_ZERO | M_WAITOK);
7452 
7453 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
7454 	p = buf;
7455 
7456 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7457 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
7458 		    p[1], p[0]);
7459 	}
7460 
7461 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
7462 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7463 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
7464 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
7465 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
7466 		    (p[1] >> 2) | ((p[2] & 3) << 30),
7467 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
7468 		    p[0] & 1);
7469 	}
7470 
7471 	rc = sbuf_finish(sb);
7472 	sbuf_delete(sb);
7473 	free(buf, M_CXGBE);
7474 	return (rc);
7475 }
7476 
7477 static int
7478 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
7479 {
7480 	struct adapter *sc = arg1;
7481 	u_int i;
7482 	struct sbuf *sb;
7483 	uint32_t *buf, *p;
7484 	int rc;
7485 
7486 	rc = sysctl_wire_old_buffer(req, 0);
7487 	if (rc != 0)
7488 		return (rc);
7489 
7490 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7491 	if (sb == NULL)
7492 		return (ENOMEM);
7493 
7494 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
7495 	    M_ZERO | M_WAITOK);
7496 
7497 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
7498 	p = buf;
7499 
7500 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
7501 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7502 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
7503 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
7504 		    p[4], p[3], p[2], p[1], p[0]);
7505 	}
7506 
7507 	sbuf_printf(sb, "\n\nCntl ID               Data");
7508 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7509 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
7510 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
7511 	}
7512 
7513 	rc = sbuf_finish(sb);
7514 	sbuf_delete(sb);
7515 	free(buf, M_CXGBE);
7516 	return (rc);
7517 }
7518 
7519 static int
7520 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
7521 {
7522 	struct adapter *sc = arg1;
7523 	struct sbuf *sb;
7524 	int rc, i;
7525 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7526 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7527 	uint16_t thres[CIM_NUM_IBQ];
7528 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
7529 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
7530 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
7531 
7532 	cim_num_obq = sc->chip_params->cim_num_obq;
7533 	if (is_t4(sc)) {
7534 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
7535 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
7536 	} else {
7537 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
7538 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
7539 	}
7540 	nq = CIM_NUM_IBQ + cim_num_obq;
7541 
7542 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
7543 	if (rc == 0)
7544 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
7545 	if (rc != 0)
7546 		return (rc);
7547 
7548 	t4_read_cimq_cfg(sc, base, size, thres);
7549 
7550 	rc = sysctl_wire_old_buffer(req, 0);
7551 	if (rc != 0)
7552 		return (rc);
7553 
7554 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7555 	if (sb == NULL)
7556 		return (ENOMEM);
7557 
7558 	sbuf_printf(sb,
7559 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
7560 
7561 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
7562 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
7563 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
7564 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7565 		    G_QUEREMFLITS(p[2]) * 16);
7566 	for ( ; i < nq; i++, p += 4, wr += 2)
7567 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
7568 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
7569 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7570 		    G_QUEREMFLITS(p[2]) * 16);
7571 
7572 	rc = sbuf_finish(sb);
7573 	sbuf_delete(sb);
7574 
7575 	return (rc);
7576 }
7577 
7578 static int
7579 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
7580 {
7581 	struct adapter *sc = arg1;
7582 	struct sbuf *sb;
7583 	int rc;
7584 	struct tp_cpl_stats stats;
7585 
7586 	rc = sysctl_wire_old_buffer(req, 0);
7587 	if (rc != 0)
7588 		return (rc);
7589 
7590 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7591 	if (sb == NULL)
7592 		return (ENOMEM);
7593 
7594 	mtx_lock(&sc->reg_lock);
7595 	t4_tp_get_cpl_stats(sc, &stats, 0);
7596 	mtx_unlock(&sc->reg_lock);
7597 
7598 	if (sc->chip_params->nchan > 2) {
7599 		sbuf_printf(sb, "                 channel 0  channel 1"
7600 		    "  channel 2  channel 3");
7601 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
7602 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
7603 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
7604 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
7605 	} else {
7606 		sbuf_printf(sb, "                 channel 0  channel 1");
7607 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
7608 		    stats.req[0], stats.req[1]);
7609 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
7610 		    stats.rsp[0], stats.rsp[1]);
7611 	}
7612 
7613 	rc = sbuf_finish(sb);
7614 	sbuf_delete(sb);
7615 
7616 	return (rc);
7617 }
7618 
7619 static int
7620 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
7621 {
7622 	struct adapter *sc = arg1;
7623 	struct sbuf *sb;
7624 	int rc;
7625 	struct tp_usm_stats stats;
7626 
7627 	rc = sysctl_wire_old_buffer(req, 0);
7628 	if (rc != 0)
7629 		return(rc);
7630 
7631 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7632 	if (sb == NULL)
7633 		return (ENOMEM);
7634 
7635 	t4_get_usm_stats(sc, &stats, 1);
7636 
7637 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
7638 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
7639 	sbuf_printf(sb, "Drops:  %u", stats.drops);
7640 
7641 	rc = sbuf_finish(sb);
7642 	sbuf_delete(sb);
7643 
7644 	return (rc);
7645 }
7646 
7647 static const char * const devlog_level_strings[] = {
7648 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
7649 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
7650 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
7651 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
7652 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
7653 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
7654 };
7655 
7656 static const char * const devlog_facility_strings[] = {
7657 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
7658 	[FW_DEVLOG_FACILITY_CF]		= "CF",
7659 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
7660 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
7661 	[FW_DEVLOG_FACILITY_RES]	= "RES",
7662 	[FW_DEVLOG_FACILITY_HW]		= "HW",
7663 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
7664 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
7665 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
7666 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
7667 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
7668 	[FW_DEVLOG_FACILITY_VI]		= "VI",
7669 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
7670 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
7671 	[FW_DEVLOG_FACILITY_TM]		= "TM",
7672 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
7673 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
7674 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
7675 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
7676 	[FW_DEVLOG_FACILITY_RI]		= "RI",
7677 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
7678 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
7679 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
7680 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
7681 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
7682 };
7683 
7684 static int
7685 sbuf_devlog(struct adapter *sc, struct sbuf *sb, int flags)
7686 {
7687 	int i, j, rc, nentries, first = 0;
7688 	struct devlog_params *dparams = &sc->params.devlog;
7689 	struct fw_devlog_e *buf, *e;
7690 	uint64_t ftstamp = UINT64_MAX;
7691 
7692 	if (dparams->addr == 0)
7693 		return (ENXIO);
7694 
7695 	MPASS(flags == M_WAITOK || flags == M_NOWAIT);
7696 	buf = malloc(dparams->size, M_CXGBE, M_ZERO | flags);
7697 	if (buf == NULL)
7698 		return (ENOMEM);
7699 
7700 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
7701 	if (rc != 0)
7702 		goto done;
7703 
7704 	nentries = dparams->size / sizeof(struct fw_devlog_e);
7705 	for (i = 0; i < nentries; i++) {
7706 		e = &buf[i];
7707 
7708 		if (e->timestamp == 0)
7709 			break;	/* end */
7710 
7711 		e->timestamp = be64toh(e->timestamp);
7712 		e->seqno = be32toh(e->seqno);
7713 		for (j = 0; j < 8; j++)
7714 			e->params[j] = be32toh(e->params[j]);
7715 
7716 		if (e->timestamp < ftstamp) {
7717 			ftstamp = e->timestamp;
7718 			first = i;
7719 		}
7720 	}
7721 
7722 	if (buf[first].timestamp == 0)
7723 		goto done;	/* nothing in the log */
7724 
7725 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
7726 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
7727 
7728 	i = first;
7729 	do {
7730 		e = &buf[i];
7731 		if (e->timestamp == 0)
7732 			break;	/* end */
7733 
7734 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
7735 		    e->seqno, e->timestamp,
7736 		    (e->level < nitems(devlog_level_strings) ?
7737 			devlog_level_strings[e->level] : "UNKNOWN"),
7738 		    (e->facility < nitems(devlog_facility_strings) ?
7739 			devlog_facility_strings[e->facility] : "UNKNOWN"));
7740 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
7741 		    e->params[2], e->params[3], e->params[4],
7742 		    e->params[5], e->params[6], e->params[7]);
7743 
7744 		if (++i == nentries)
7745 			i = 0;
7746 	} while (i != first);
7747 done:
7748 	free(buf, M_CXGBE);
7749 	return (rc);
7750 }
7751 
7752 static int
7753 sysctl_devlog(SYSCTL_HANDLER_ARGS)
7754 {
7755 	struct adapter *sc = arg1;
7756 	int rc;
7757 	struct sbuf *sb;
7758 
7759 	rc = sysctl_wire_old_buffer(req, 0);
7760 	if (rc != 0)
7761 		return (rc);
7762 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7763 	if (sb == NULL)
7764 		return (ENOMEM);
7765 
7766 	rc = sbuf_devlog(sc, sb, M_WAITOK);
7767 	if (rc == 0)
7768 		rc = sbuf_finish(sb);
7769 	sbuf_delete(sb);
7770 	return (rc);
7771 }
7772 
7773 void
7774 t4_os_dump_devlog(struct adapter *sc)
7775 {
7776 	int rc;
7777 	struct sbuf sb;
7778 
7779 	if (sbuf_new(&sb, NULL, 4096, SBUF_AUTOEXTEND) != &sb)
7780 		return;
7781 	rc = sbuf_devlog(sc, &sb, M_NOWAIT);
7782 	if (rc == 0) {
7783 		rc = sbuf_finish(&sb);
7784 		if (rc == 0) {
7785 			log(LOG_DEBUG, "%s: device log follows.\n%s",
7786 		    		device_get_nameunit(sc->dev), sbuf_data(&sb));
7787 		}
7788 	}
7789 	sbuf_delete(&sb);
7790 }
7791 
7792 static int
7793 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
7794 {
7795 	struct adapter *sc = arg1;
7796 	struct sbuf *sb;
7797 	int rc;
7798 	struct tp_fcoe_stats stats[MAX_NCHAN];
7799 	int i, nchan = sc->chip_params->nchan;
7800 
7801 	rc = sysctl_wire_old_buffer(req, 0);
7802 	if (rc != 0)
7803 		return (rc);
7804 
7805 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7806 	if (sb == NULL)
7807 		return (ENOMEM);
7808 
7809 	for (i = 0; i < nchan; i++)
7810 		t4_get_fcoe_stats(sc, i, &stats[i], 1);
7811 
7812 	if (nchan > 2) {
7813 		sbuf_printf(sb, "                   channel 0        channel 1"
7814 		    "        channel 2        channel 3");
7815 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
7816 		    stats[0].octets_ddp, stats[1].octets_ddp,
7817 		    stats[2].octets_ddp, stats[3].octets_ddp);
7818 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
7819 		    stats[0].frames_ddp, stats[1].frames_ddp,
7820 		    stats[2].frames_ddp, stats[3].frames_ddp);
7821 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
7822 		    stats[0].frames_drop, stats[1].frames_drop,
7823 		    stats[2].frames_drop, stats[3].frames_drop);
7824 	} else {
7825 		sbuf_printf(sb, "                   channel 0        channel 1");
7826 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
7827 		    stats[0].octets_ddp, stats[1].octets_ddp);
7828 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
7829 		    stats[0].frames_ddp, stats[1].frames_ddp);
7830 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
7831 		    stats[0].frames_drop, stats[1].frames_drop);
7832 	}
7833 
7834 	rc = sbuf_finish(sb);
7835 	sbuf_delete(sb);
7836 
7837 	return (rc);
7838 }
7839 
7840 static int
7841 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
7842 {
7843 	struct adapter *sc = arg1;
7844 	struct sbuf *sb;
7845 	int rc, i;
7846 	unsigned int map, kbps, ipg, mode;
7847 	unsigned int pace_tab[NTX_SCHED];
7848 
7849 	rc = sysctl_wire_old_buffer(req, 0);
7850 	if (rc != 0)
7851 		return (rc);
7852 
7853 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7854 	if (sb == NULL)
7855 		return (ENOMEM);
7856 
7857 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
7858 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
7859 	t4_read_pace_tbl(sc, pace_tab);
7860 
7861 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
7862 	    "Class IPG (0.1 ns)   Flow IPG (us)");
7863 
7864 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
7865 		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
7866 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
7867 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
7868 		if (kbps)
7869 			sbuf_printf(sb, "%9u     ", kbps);
7870 		else
7871 			sbuf_printf(sb, " disabled     ");
7872 
7873 		if (ipg)
7874 			sbuf_printf(sb, "%13u        ", ipg);
7875 		else
7876 			sbuf_printf(sb, "     disabled        ");
7877 
7878 		if (pace_tab[i])
7879 			sbuf_printf(sb, "%10u", pace_tab[i]);
7880 		else
7881 			sbuf_printf(sb, "  disabled");
7882 	}
7883 
7884 	rc = sbuf_finish(sb);
7885 	sbuf_delete(sb);
7886 
7887 	return (rc);
7888 }
7889 
7890 static int
7891 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
7892 {
7893 	struct adapter *sc = arg1;
7894 	struct sbuf *sb;
7895 	int rc, i, j;
7896 	uint64_t *p0, *p1;
7897 	struct lb_port_stats s[2];
7898 	static const char *stat_name[] = {
7899 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
7900 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
7901 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
7902 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
7903 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
7904 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
7905 		"BG2FramesTrunc:", "BG3FramesTrunc:"
7906 	};
7907 
7908 	rc = sysctl_wire_old_buffer(req, 0);
7909 	if (rc != 0)
7910 		return (rc);
7911 
7912 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7913 	if (sb == NULL)
7914 		return (ENOMEM);
7915 
7916 	memset(s, 0, sizeof(s));
7917 
7918 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
7919 		t4_get_lb_stats(sc, i, &s[0]);
7920 		t4_get_lb_stats(sc, i + 1, &s[1]);
7921 
7922 		p0 = &s[0].octets;
7923 		p1 = &s[1].octets;
7924 		sbuf_printf(sb, "%s                       Loopback %u"
7925 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
7926 
7927 		for (j = 0; j < nitems(stat_name); j++)
7928 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
7929 				   *p0++, *p1++);
7930 	}
7931 
7932 	rc = sbuf_finish(sb);
7933 	sbuf_delete(sb);
7934 
7935 	return (rc);
7936 }
7937 
7938 static int
7939 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
7940 {
7941 	int rc = 0;
7942 	struct port_info *pi = arg1;
7943 	struct link_config *lc = &pi->link_cfg;
7944 	struct sbuf *sb;
7945 
7946 	rc = sysctl_wire_old_buffer(req, 0);
7947 	if (rc != 0)
7948 		return(rc);
7949 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
7950 	if (sb == NULL)
7951 		return (ENOMEM);
7952 
7953 	if (lc->link_ok || lc->link_down_rc == 255)
7954 		sbuf_printf(sb, "n/a");
7955 	else
7956 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
7957 
7958 	rc = sbuf_finish(sb);
7959 	sbuf_delete(sb);
7960 
7961 	return (rc);
7962 }
7963 
7964 struct mem_desc {
7965 	unsigned int base;
7966 	unsigned int limit;
7967 	unsigned int idx;
7968 };
7969 
7970 static int
7971 mem_desc_cmp(const void *a, const void *b)
7972 {
7973 	return ((const struct mem_desc *)a)->base -
7974 	       ((const struct mem_desc *)b)->base;
7975 }
7976 
7977 static void
7978 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
7979     unsigned int to)
7980 {
7981 	unsigned int size;
7982 
7983 	if (from == to)
7984 		return;
7985 
7986 	size = to - from + 1;
7987 	if (size == 0)
7988 		return;
7989 
7990 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
7991 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
7992 }
7993 
7994 static int
7995 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
7996 {
7997 	struct adapter *sc = arg1;
7998 	struct sbuf *sb;
7999 	int rc, i, n;
8000 	uint32_t lo, hi, used, alloc;
8001 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
8002 	static const char *region[] = {
8003 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
8004 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
8005 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
8006 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
8007 		"RQUDP region:", "PBL region:", "TXPBL region:",
8008 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
8009 		"On-chip queues:", "TLS keys:",
8010 	};
8011 	struct mem_desc avail[4];
8012 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
8013 	struct mem_desc *md = mem;
8014 
8015 	rc = sysctl_wire_old_buffer(req, 0);
8016 	if (rc != 0)
8017 		return (rc);
8018 
8019 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8020 	if (sb == NULL)
8021 		return (ENOMEM);
8022 
8023 	for (i = 0; i < nitems(mem); i++) {
8024 		mem[i].limit = 0;
8025 		mem[i].idx = i;
8026 	}
8027 
8028 	/* Find and sort the populated memory ranges */
8029 	i = 0;
8030 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
8031 	if (lo & F_EDRAM0_ENABLE) {
8032 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
8033 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
8034 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
8035 		avail[i].idx = 0;
8036 		i++;
8037 	}
8038 	if (lo & F_EDRAM1_ENABLE) {
8039 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
8040 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
8041 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
8042 		avail[i].idx = 1;
8043 		i++;
8044 	}
8045 	if (lo & F_EXT_MEM_ENABLE) {
8046 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
8047 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
8048 		avail[i].limit = avail[i].base +
8049 		    (G_EXT_MEM_SIZE(hi) << 20);
8050 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
8051 		i++;
8052 	}
8053 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
8054 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
8055 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
8056 		avail[i].limit = avail[i].base +
8057 		    (G_EXT_MEM1_SIZE(hi) << 20);
8058 		avail[i].idx = 4;
8059 		i++;
8060 	}
8061 	if (!i)                                    /* no memory available */
8062 		return 0;
8063 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
8064 
8065 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
8066 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
8067 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
8068 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
8069 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
8070 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
8071 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
8072 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
8073 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
8074 
8075 	/* the next few have explicit upper bounds */
8076 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
8077 	md->limit = md->base - 1 +
8078 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
8079 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
8080 	md++;
8081 
8082 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
8083 	md->limit = md->base - 1 +
8084 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
8085 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
8086 	md++;
8087 
8088 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
8089 		if (chip_id(sc) <= CHELSIO_T5)
8090 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
8091 		else
8092 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
8093 		md->limit = 0;
8094 	} else {
8095 		md->base = 0;
8096 		md->idx = nitems(region);  /* hide it */
8097 	}
8098 	md++;
8099 
8100 #define ulp_region(reg) \
8101 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
8102 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
8103 
8104 	ulp_region(RX_ISCSI);
8105 	ulp_region(RX_TDDP);
8106 	ulp_region(TX_TPT);
8107 	ulp_region(RX_STAG);
8108 	ulp_region(RX_RQ);
8109 	ulp_region(RX_RQUDP);
8110 	ulp_region(RX_PBL);
8111 	ulp_region(TX_PBL);
8112 #undef ulp_region
8113 
8114 	md->base = 0;
8115 	md->idx = nitems(region);
8116 	if (!is_t4(sc)) {
8117 		uint32_t size = 0;
8118 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
8119 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
8120 
8121 		if (is_t5(sc)) {
8122 			if (sge_ctrl & F_VFIFO_ENABLE)
8123 				size = G_DBVFIFO_SIZE(fifo_size);
8124 		} else
8125 			size = G_T6_DBVFIFO_SIZE(fifo_size);
8126 
8127 		if (size) {
8128 			md->base = G_BASEADDR(t4_read_reg(sc,
8129 			    A_SGE_DBVFIFO_BADDR));
8130 			md->limit = md->base + (size << 2) - 1;
8131 		}
8132 	}
8133 	md++;
8134 
8135 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
8136 	md->limit = 0;
8137 	md++;
8138 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
8139 	md->limit = 0;
8140 	md++;
8141 
8142 	md->base = sc->vres.ocq.start;
8143 	if (sc->vres.ocq.size)
8144 		md->limit = md->base + sc->vres.ocq.size - 1;
8145 	else
8146 		md->idx = nitems(region);  /* hide it */
8147 	md++;
8148 
8149 	md->base = sc->vres.key.start;
8150 	if (sc->vres.key.size)
8151 		md->limit = md->base + sc->vres.key.size - 1;
8152 	else
8153 		md->idx = nitems(region);  /* hide it */
8154 	md++;
8155 
8156 	/* add any address-space holes, there can be up to 3 */
8157 	for (n = 0; n < i - 1; n++)
8158 		if (avail[n].limit < avail[n + 1].base)
8159 			(md++)->base = avail[n].limit;
8160 	if (avail[n].limit)
8161 		(md++)->base = avail[n].limit;
8162 
8163 	n = md - mem;
8164 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
8165 
8166 	for (lo = 0; lo < i; lo++)
8167 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
8168 				avail[lo].limit - 1);
8169 
8170 	sbuf_printf(sb, "\n");
8171 	for (i = 0; i < n; i++) {
8172 		if (mem[i].idx >= nitems(region))
8173 			continue;                        /* skip holes */
8174 		if (!mem[i].limit)
8175 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
8176 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
8177 				mem[i].limit);
8178 	}
8179 
8180 	sbuf_printf(sb, "\n");
8181 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
8182 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
8183 	mem_region_show(sb, "uP RAM:", lo, hi);
8184 
8185 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
8186 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
8187 	mem_region_show(sb, "uP Extmem2:", lo, hi);
8188 
8189 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
8190 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
8191 		   G_PMRXMAXPAGE(lo),
8192 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
8193 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
8194 
8195 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
8196 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
8197 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
8198 		   G_PMTXMAXPAGE(lo),
8199 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
8200 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
8201 	sbuf_printf(sb, "%u p-structs\n",
8202 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
8203 
8204 	for (i = 0; i < 4; i++) {
8205 		if (chip_id(sc) > CHELSIO_T5)
8206 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
8207 		else
8208 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
8209 		if (is_t5(sc)) {
8210 			used = G_T5_USED(lo);
8211 			alloc = G_T5_ALLOC(lo);
8212 		} else {
8213 			used = G_USED(lo);
8214 			alloc = G_ALLOC(lo);
8215 		}
8216 		/* For T6 these are MAC buffer groups */
8217 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
8218 		    i, used, alloc);
8219 	}
8220 	for (i = 0; i < sc->chip_params->nchan; i++) {
8221 		if (chip_id(sc) > CHELSIO_T5)
8222 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
8223 		else
8224 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
8225 		if (is_t5(sc)) {
8226 			used = G_T5_USED(lo);
8227 			alloc = G_T5_ALLOC(lo);
8228 		} else {
8229 			used = G_USED(lo);
8230 			alloc = G_ALLOC(lo);
8231 		}
8232 		/* For T6 these are MAC buffer groups */
8233 		sbuf_printf(sb,
8234 		    "\nLoopback %d using %u pages out of %u allocated",
8235 		    i, used, alloc);
8236 	}
8237 
8238 	rc = sbuf_finish(sb);
8239 	sbuf_delete(sb);
8240 
8241 	return (rc);
8242 }
8243 
8244 static inline void
8245 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
8246 {
8247 	*mask = x | y;
8248 	y = htobe64(y);
8249 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
8250 }
8251 
8252 static int
8253 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
8254 {
8255 	struct adapter *sc = arg1;
8256 	struct sbuf *sb;
8257 	int rc, i;
8258 
8259 	MPASS(chip_id(sc) <= CHELSIO_T5);
8260 
8261 	rc = sysctl_wire_old_buffer(req, 0);
8262 	if (rc != 0)
8263 		return (rc);
8264 
8265 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8266 	if (sb == NULL)
8267 		return (ENOMEM);
8268 
8269 	sbuf_printf(sb,
8270 	    "Idx  Ethernet address     Mask     Vld Ports PF"
8271 	    "  VF              Replication             P0 P1 P2 P3  ML");
8272 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
8273 		uint64_t tcamx, tcamy, mask;
8274 		uint32_t cls_lo, cls_hi;
8275 		uint8_t addr[ETHER_ADDR_LEN];
8276 
8277 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
8278 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
8279 		if (tcamx & tcamy)
8280 			continue;
8281 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
8282 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
8283 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
8284 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
8285 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
8286 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
8287 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
8288 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
8289 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
8290 
8291 		if (cls_lo & F_REPLICATE) {
8292 			struct fw_ldst_cmd ldst_cmd;
8293 
8294 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
8295 			ldst_cmd.op_to_addrspace =
8296 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
8297 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
8298 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
8299 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
8300 			ldst_cmd.u.mps.rplc.fid_idx =
8301 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
8302 				V_FW_LDST_CMD_IDX(i));
8303 
8304 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8305 			    "t4mps");
8306 			if (rc)
8307 				break;
8308 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
8309 			    sizeof(ldst_cmd), &ldst_cmd);
8310 			end_synchronized_op(sc, 0);
8311 
8312 			if (rc != 0) {
8313 				sbuf_printf(sb, "%36d", rc);
8314 				rc = 0;
8315 			} else {
8316 				sbuf_printf(sb, " %08x %08x %08x %08x",
8317 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
8318 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
8319 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
8320 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
8321 			}
8322 		} else
8323 			sbuf_printf(sb, "%36s", "");
8324 
8325 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
8326 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
8327 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
8328 	}
8329 
8330 	if (rc)
8331 		(void) sbuf_finish(sb);
8332 	else
8333 		rc = sbuf_finish(sb);
8334 	sbuf_delete(sb);
8335 
8336 	return (rc);
8337 }
8338 
8339 static int
8340 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
8341 {
8342 	struct adapter *sc = arg1;
8343 	struct sbuf *sb;
8344 	int rc, i;
8345 
8346 	MPASS(chip_id(sc) > CHELSIO_T5);
8347 
8348 	rc = sysctl_wire_old_buffer(req, 0);
8349 	if (rc != 0)
8350 		return (rc);
8351 
8352 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8353 	if (sb == NULL)
8354 		return (ENOMEM);
8355 
8356 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
8357 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
8358 	    "                           Replication"
8359 	    "                                    P0 P1 P2 P3  ML\n");
8360 
8361 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
8362 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
8363 		uint16_t ivlan;
8364 		uint64_t tcamx, tcamy, val, mask;
8365 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
8366 		uint8_t addr[ETHER_ADDR_LEN];
8367 
8368 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
8369 		if (i < 256)
8370 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
8371 		else
8372 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
8373 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8374 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8375 		tcamy = G_DMACH(val) << 32;
8376 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8377 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8378 		lookup_type = G_DATALKPTYPE(data2);
8379 		port_num = G_DATAPORTNUM(data2);
8380 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8381 			/* Inner header VNI */
8382 			vniy = ((data2 & F_DATAVIDH2) << 23) |
8383 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8384 			dip_hit = data2 & F_DATADIPHIT;
8385 			vlan_vld = 0;
8386 		} else {
8387 			vniy = 0;
8388 			dip_hit = 0;
8389 			vlan_vld = data2 & F_DATAVIDH2;
8390 			ivlan = G_VIDL(val);
8391 		}
8392 
8393 		ctl |= V_CTLXYBITSEL(1);
8394 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8395 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8396 		tcamx = G_DMACH(val) << 32;
8397 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8398 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8399 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8400 			/* Inner header VNI mask */
8401 			vnix = ((data2 & F_DATAVIDH2) << 23) |
8402 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8403 		} else
8404 			vnix = 0;
8405 
8406 		if (tcamx & tcamy)
8407 			continue;
8408 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
8409 
8410 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
8411 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
8412 
8413 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8414 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8415 			    "%012jx %06x %06x    -    -   %3c"
8416 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
8417 			    addr[1], addr[2], addr[3], addr[4], addr[5],
8418 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
8419 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8420 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8421 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8422 		} else {
8423 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8424 			    "%012jx    -       -   ", i, addr[0], addr[1],
8425 			    addr[2], addr[3], addr[4], addr[5],
8426 			    (uintmax_t)mask);
8427 
8428 			if (vlan_vld)
8429 				sbuf_printf(sb, "%4u   Y     ", ivlan);
8430 			else
8431 				sbuf_printf(sb, "  -    N     ");
8432 
8433 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
8434 			    lookup_type ? 'I' : 'O', port_num,
8435 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8436 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8437 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8438 		}
8439 
8440 
8441 		if (cls_lo & F_T6_REPLICATE) {
8442 			struct fw_ldst_cmd ldst_cmd;
8443 
8444 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
8445 			ldst_cmd.op_to_addrspace =
8446 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
8447 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
8448 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
8449 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
8450 			ldst_cmd.u.mps.rplc.fid_idx =
8451 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
8452 				V_FW_LDST_CMD_IDX(i));
8453 
8454 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8455 			    "t6mps");
8456 			if (rc)
8457 				break;
8458 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
8459 			    sizeof(ldst_cmd), &ldst_cmd);
8460 			end_synchronized_op(sc, 0);
8461 
8462 			if (rc != 0) {
8463 				sbuf_printf(sb, "%72d", rc);
8464 				rc = 0;
8465 			} else {
8466 				sbuf_printf(sb, " %08x %08x %08x %08x"
8467 				    " %08x %08x %08x %08x",
8468 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
8469 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
8470 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
8471 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
8472 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
8473 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
8474 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
8475 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
8476 			}
8477 		} else
8478 			sbuf_printf(sb, "%72s", "");
8479 
8480 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
8481 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
8482 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
8483 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
8484 	}
8485 
8486 	if (rc)
8487 		(void) sbuf_finish(sb);
8488 	else
8489 		rc = sbuf_finish(sb);
8490 	sbuf_delete(sb);
8491 
8492 	return (rc);
8493 }
8494 
8495 static int
8496 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
8497 {
8498 	struct adapter *sc = arg1;
8499 	struct sbuf *sb;
8500 	int rc;
8501 	uint16_t mtus[NMTUS];
8502 
8503 	rc = sysctl_wire_old_buffer(req, 0);
8504 	if (rc != 0)
8505 		return (rc);
8506 
8507 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8508 	if (sb == NULL)
8509 		return (ENOMEM);
8510 
8511 	t4_read_mtu_tbl(sc, mtus, NULL);
8512 
8513 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
8514 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
8515 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
8516 	    mtus[14], mtus[15]);
8517 
8518 	rc = sbuf_finish(sb);
8519 	sbuf_delete(sb);
8520 
8521 	return (rc);
8522 }
8523 
8524 static int
8525 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
8526 {
8527 	struct adapter *sc = arg1;
8528 	struct sbuf *sb;
8529 	int rc, i;
8530 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
8531 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
8532 	static const char *tx_stats[MAX_PM_NSTATS] = {
8533 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
8534 		"Tx FIFO wait", NULL, "Tx latency"
8535 	};
8536 	static const char *rx_stats[MAX_PM_NSTATS] = {
8537 		"Read:", "Write bypass:", "Write mem:", "Flush:",
8538 		"Rx FIFO wait", NULL, "Rx latency"
8539 	};
8540 
8541 	rc = sysctl_wire_old_buffer(req, 0);
8542 	if (rc != 0)
8543 		return (rc);
8544 
8545 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8546 	if (sb == NULL)
8547 		return (ENOMEM);
8548 
8549 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
8550 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
8551 
8552 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
8553 	for (i = 0; i < 4; i++) {
8554 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8555 		    tx_cyc[i]);
8556 	}
8557 
8558 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
8559 	for (i = 0; i < 4; i++) {
8560 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8561 		    rx_cyc[i]);
8562 	}
8563 
8564 	if (chip_id(sc) > CHELSIO_T5) {
8565 		sbuf_printf(sb,
8566 		    "\n              Total wait      Total occupancy");
8567 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8568 		    tx_cyc[i]);
8569 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8570 		    rx_cyc[i]);
8571 
8572 		i += 2;
8573 		MPASS(i < nitems(tx_stats));
8574 
8575 		sbuf_printf(sb,
8576 		    "\n                   Reads           Total wait");
8577 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8578 		    tx_cyc[i]);
8579 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8580 		    rx_cyc[i]);
8581 	}
8582 
8583 	rc = sbuf_finish(sb);
8584 	sbuf_delete(sb);
8585 
8586 	return (rc);
8587 }
8588 
8589 static int
8590 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
8591 {
8592 	struct adapter *sc = arg1;
8593 	struct sbuf *sb;
8594 	int rc;
8595 	struct tp_rdma_stats stats;
8596 
8597 	rc = sysctl_wire_old_buffer(req, 0);
8598 	if (rc != 0)
8599 		return (rc);
8600 
8601 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8602 	if (sb == NULL)
8603 		return (ENOMEM);
8604 
8605 	mtx_lock(&sc->reg_lock);
8606 	t4_tp_get_rdma_stats(sc, &stats, 0);
8607 	mtx_unlock(&sc->reg_lock);
8608 
8609 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
8610 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
8611 
8612 	rc = sbuf_finish(sb);
8613 	sbuf_delete(sb);
8614 
8615 	return (rc);
8616 }
8617 
8618 static int
8619 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
8620 {
8621 	struct adapter *sc = arg1;
8622 	struct sbuf *sb;
8623 	int rc;
8624 	struct tp_tcp_stats v4, v6;
8625 
8626 	rc = sysctl_wire_old_buffer(req, 0);
8627 	if (rc != 0)
8628 		return (rc);
8629 
8630 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8631 	if (sb == NULL)
8632 		return (ENOMEM);
8633 
8634 	mtx_lock(&sc->reg_lock);
8635 	t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
8636 	mtx_unlock(&sc->reg_lock);
8637 
8638 	sbuf_printf(sb,
8639 	    "                                IP                 IPv6\n");
8640 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
8641 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
8642 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
8643 	    v4.tcp_in_segs, v6.tcp_in_segs);
8644 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
8645 	    v4.tcp_out_segs, v6.tcp_out_segs);
8646 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
8647 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
8648 
8649 	rc = sbuf_finish(sb);
8650 	sbuf_delete(sb);
8651 
8652 	return (rc);
8653 }
8654 
8655 static int
8656 sysctl_tids(SYSCTL_HANDLER_ARGS)
8657 {
8658 	struct adapter *sc = arg1;
8659 	struct sbuf *sb;
8660 	int rc;
8661 	struct tid_info *t = &sc->tids;
8662 
8663 	rc = sysctl_wire_old_buffer(req, 0);
8664 	if (rc != 0)
8665 		return (rc);
8666 
8667 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8668 	if (sb == NULL)
8669 		return (ENOMEM);
8670 
8671 	if (t->natids) {
8672 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
8673 		    t->atids_in_use);
8674 	}
8675 
8676 	if (t->nhpftids) {
8677 		sbuf_printf(sb, "HPFTID range: %u-%u, in use: %u\n",
8678 		    t->hpftid_base, t->hpftid_end, t->hpftids_in_use);
8679 	}
8680 
8681 	if (t->ntids) {
8682 		sbuf_printf(sb, "TID range: ");
8683 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
8684 			uint32_t b, hb;
8685 
8686 			if (chip_id(sc) <= CHELSIO_T5) {
8687 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
8688 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
8689 			} else {
8690 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
8691 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
8692 			}
8693 
8694 			if (b)
8695 				sbuf_printf(sb, "%u-%u, ", t->tid_base, b - 1);
8696 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
8697 		} else
8698 			sbuf_printf(sb, "%u-%u", t->tid_base, t->ntids - 1);
8699 		sbuf_printf(sb, ", in use: %u\n",
8700 		    atomic_load_acq_int(&t->tids_in_use));
8701 	}
8702 
8703 	if (t->nstids) {
8704 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
8705 		    t->stid_base + t->nstids - 1, t->stids_in_use);
8706 	}
8707 
8708 	if (t->nftids) {
8709 		sbuf_printf(sb, "FTID range: %u-%u, in use: %u\n", t->ftid_base,
8710 		    t->ftid_end, t->ftids_in_use);
8711 	}
8712 
8713 	if (t->netids) {
8714 		sbuf_printf(sb, "ETID range: %u-%u, in use: %u\n", t->etid_base,
8715 		    t->etid_base + t->netids - 1, t->etids_in_use);
8716 	}
8717 
8718 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
8719 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
8720 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
8721 
8722 	rc = sbuf_finish(sb);
8723 	sbuf_delete(sb);
8724 
8725 	return (rc);
8726 }
8727 
8728 static int
8729 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
8730 {
8731 	struct adapter *sc = arg1;
8732 	struct sbuf *sb;
8733 	int rc;
8734 	struct tp_err_stats stats;
8735 
8736 	rc = sysctl_wire_old_buffer(req, 0);
8737 	if (rc != 0)
8738 		return (rc);
8739 
8740 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8741 	if (sb == NULL)
8742 		return (ENOMEM);
8743 
8744 	mtx_lock(&sc->reg_lock);
8745 	t4_tp_get_err_stats(sc, &stats, 0);
8746 	mtx_unlock(&sc->reg_lock);
8747 
8748 	if (sc->chip_params->nchan > 2) {
8749 		sbuf_printf(sb, "                 channel 0  channel 1"
8750 		    "  channel 2  channel 3\n");
8751 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
8752 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
8753 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
8754 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
8755 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
8756 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
8757 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
8758 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
8759 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
8760 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
8761 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
8762 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
8763 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
8764 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
8765 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
8766 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
8767 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
8768 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
8769 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
8770 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
8771 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
8772 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
8773 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
8774 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
8775 	} else {
8776 		sbuf_printf(sb, "                 channel 0  channel 1\n");
8777 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
8778 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
8779 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
8780 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
8781 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
8782 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
8783 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
8784 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
8785 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
8786 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
8787 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
8788 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
8789 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
8790 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
8791 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
8792 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
8793 	}
8794 
8795 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
8796 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
8797 
8798 	rc = sbuf_finish(sb);
8799 	sbuf_delete(sb);
8800 
8801 	return (rc);
8802 }
8803 
8804 static int
8805 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
8806 {
8807 	struct adapter *sc = arg1;
8808 	struct tp_params *tpp = &sc->params.tp;
8809 	u_int mask;
8810 	int rc;
8811 
8812 	mask = tpp->la_mask >> 16;
8813 	rc = sysctl_handle_int(oidp, &mask, 0, req);
8814 	if (rc != 0 || req->newptr == NULL)
8815 		return (rc);
8816 	if (mask > 0xffff)
8817 		return (EINVAL);
8818 	tpp->la_mask = mask << 16;
8819 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
8820 
8821 	return (0);
8822 }
8823 
8824 struct field_desc {
8825 	const char *name;
8826 	u_int start;
8827 	u_int width;
8828 };
8829 
8830 static void
8831 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
8832 {
8833 	char buf[32];
8834 	int line_size = 0;
8835 
8836 	while (f->name) {
8837 		uint64_t mask = (1ULL << f->width) - 1;
8838 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
8839 		    ((uintmax_t)v >> f->start) & mask);
8840 
8841 		if (line_size + len >= 79) {
8842 			line_size = 8;
8843 			sbuf_printf(sb, "\n        ");
8844 		}
8845 		sbuf_printf(sb, "%s ", buf);
8846 		line_size += len + 1;
8847 		f++;
8848 	}
8849 	sbuf_printf(sb, "\n");
8850 }
8851 
8852 static const struct field_desc tp_la0[] = {
8853 	{ "RcfOpCodeOut", 60, 4 },
8854 	{ "State", 56, 4 },
8855 	{ "WcfState", 52, 4 },
8856 	{ "RcfOpcSrcOut", 50, 2 },
8857 	{ "CRxError", 49, 1 },
8858 	{ "ERxError", 48, 1 },
8859 	{ "SanityFailed", 47, 1 },
8860 	{ "SpuriousMsg", 46, 1 },
8861 	{ "FlushInputMsg", 45, 1 },
8862 	{ "FlushInputCpl", 44, 1 },
8863 	{ "RssUpBit", 43, 1 },
8864 	{ "RssFilterHit", 42, 1 },
8865 	{ "Tid", 32, 10 },
8866 	{ "InitTcb", 31, 1 },
8867 	{ "LineNumber", 24, 7 },
8868 	{ "Emsg", 23, 1 },
8869 	{ "EdataOut", 22, 1 },
8870 	{ "Cmsg", 21, 1 },
8871 	{ "CdataOut", 20, 1 },
8872 	{ "EreadPdu", 19, 1 },
8873 	{ "CreadPdu", 18, 1 },
8874 	{ "TunnelPkt", 17, 1 },
8875 	{ "RcfPeerFin", 16, 1 },
8876 	{ "RcfReasonOut", 12, 4 },
8877 	{ "TxCchannel", 10, 2 },
8878 	{ "RcfTxChannel", 8, 2 },
8879 	{ "RxEchannel", 6, 2 },
8880 	{ "RcfRxChannel", 5, 1 },
8881 	{ "RcfDataOutSrdy", 4, 1 },
8882 	{ "RxDvld", 3, 1 },
8883 	{ "RxOoDvld", 2, 1 },
8884 	{ "RxCongestion", 1, 1 },
8885 	{ "TxCongestion", 0, 1 },
8886 	{ NULL }
8887 };
8888 
8889 static const struct field_desc tp_la1[] = {
8890 	{ "CplCmdIn", 56, 8 },
8891 	{ "CplCmdOut", 48, 8 },
8892 	{ "ESynOut", 47, 1 },
8893 	{ "EAckOut", 46, 1 },
8894 	{ "EFinOut", 45, 1 },
8895 	{ "ERstOut", 44, 1 },
8896 	{ "SynIn", 43, 1 },
8897 	{ "AckIn", 42, 1 },
8898 	{ "FinIn", 41, 1 },
8899 	{ "RstIn", 40, 1 },
8900 	{ "DataIn", 39, 1 },
8901 	{ "DataInVld", 38, 1 },
8902 	{ "PadIn", 37, 1 },
8903 	{ "RxBufEmpty", 36, 1 },
8904 	{ "RxDdp", 35, 1 },
8905 	{ "RxFbCongestion", 34, 1 },
8906 	{ "TxFbCongestion", 33, 1 },
8907 	{ "TxPktSumSrdy", 32, 1 },
8908 	{ "RcfUlpType", 28, 4 },
8909 	{ "Eread", 27, 1 },
8910 	{ "Ebypass", 26, 1 },
8911 	{ "Esave", 25, 1 },
8912 	{ "Static0", 24, 1 },
8913 	{ "Cread", 23, 1 },
8914 	{ "Cbypass", 22, 1 },
8915 	{ "Csave", 21, 1 },
8916 	{ "CPktOut", 20, 1 },
8917 	{ "RxPagePoolFull", 18, 2 },
8918 	{ "RxLpbkPkt", 17, 1 },
8919 	{ "TxLpbkPkt", 16, 1 },
8920 	{ "RxVfValid", 15, 1 },
8921 	{ "SynLearned", 14, 1 },
8922 	{ "SetDelEntry", 13, 1 },
8923 	{ "SetInvEntry", 12, 1 },
8924 	{ "CpcmdDvld", 11, 1 },
8925 	{ "CpcmdSave", 10, 1 },
8926 	{ "RxPstructsFull", 8, 2 },
8927 	{ "EpcmdDvld", 7, 1 },
8928 	{ "EpcmdFlush", 6, 1 },
8929 	{ "EpcmdTrimPrefix", 5, 1 },
8930 	{ "EpcmdTrimPostfix", 4, 1 },
8931 	{ "ERssIp4Pkt", 3, 1 },
8932 	{ "ERssIp6Pkt", 2, 1 },
8933 	{ "ERssTcpUdpPkt", 1, 1 },
8934 	{ "ERssFceFipPkt", 0, 1 },
8935 	{ NULL }
8936 };
8937 
8938 static const struct field_desc tp_la2[] = {
8939 	{ "CplCmdIn", 56, 8 },
8940 	{ "MpsVfVld", 55, 1 },
8941 	{ "MpsPf", 52, 3 },
8942 	{ "MpsVf", 44, 8 },
8943 	{ "SynIn", 43, 1 },
8944 	{ "AckIn", 42, 1 },
8945 	{ "FinIn", 41, 1 },
8946 	{ "RstIn", 40, 1 },
8947 	{ "DataIn", 39, 1 },
8948 	{ "DataInVld", 38, 1 },
8949 	{ "PadIn", 37, 1 },
8950 	{ "RxBufEmpty", 36, 1 },
8951 	{ "RxDdp", 35, 1 },
8952 	{ "RxFbCongestion", 34, 1 },
8953 	{ "TxFbCongestion", 33, 1 },
8954 	{ "TxPktSumSrdy", 32, 1 },
8955 	{ "RcfUlpType", 28, 4 },
8956 	{ "Eread", 27, 1 },
8957 	{ "Ebypass", 26, 1 },
8958 	{ "Esave", 25, 1 },
8959 	{ "Static0", 24, 1 },
8960 	{ "Cread", 23, 1 },
8961 	{ "Cbypass", 22, 1 },
8962 	{ "Csave", 21, 1 },
8963 	{ "CPktOut", 20, 1 },
8964 	{ "RxPagePoolFull", 18, 2 },
8965 	{ "RxLpbkPkt", 17, 1 },
8966 	{ "TxLpbkPkt", 16, 1 },
8967 	{ "RxVfValid", 15, 1 },
8968 	{ "SynLearned", 14, 1 },
8969 	{ "SetDelEntry", 13, 1 },
8970 	{ "SetInvEntry", 12, 1 },
8971 	{ "CpcmdDvld", 11, 1 },
8972 	{ "CpcmdSave", 10, 1 },
8973 	{ "RxPstructsFull", 8, 2 },
8974 	{ "EpcmdDvld", 7, 1 },
8975 	{ "EpcmdFlush", 6, 1 },
8976 	{ "EpcmdTrimPrefix", 5, 1 },
8977 	{ "EpcmdTrimPostfix", 4, 1 },
8978 	{ "ERssIp4Pkt", 3, 1 },
8979 	{ "ERssIp6Pkt", 2, 1 },
8980 	{ "ERssTcpUdpPkt", 1, 1 },
8981 	{ "ERssFceFipPkt", 0, 1 },
8982 	{ NULL }
8983 };
8984 
8985 static void
8986 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
8987 {
8988 
8989 	field_desc_show(sb, *p, tp_la0);
8990 }
8991 
8992 static void
8993 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
8994 {
8995 
8996 	if (idx)
8997 		sbuf_printf(sb, "\n");
8998 	field_desc_show(sb, p[0], tp_la0);
8999 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
9000 		field_desc_show(sb, p[1], tp_la0);
9001 }
9002 
9003 static void
9004 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
9005 {
9006 
9007 	if (idx)
9008 		sbuf_printf(sb, "\n");
9009 	field_desc_show(sb, p[0], tp_la0);
9010 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
9011 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
9012 }
9013 
9014 static int
9015 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
9016 {
9017 	struct adapter *sc = arg1;
9018 	struct sbuf *sb;
9019 	uint64_t *buf, *p;
9020 	int rc;
9021 	u_int i, inc;
9022 	void (*show_func)(struct sbuf *, uint64_t *, int);
9023 
9024 	rc = sysctl_wire_old_buffer(req, 0);
9025 	if (rc != 0)
9026 		return (rc);
9027 
9028 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9029 	if (sb == NULL)
9030 		return (ENOMEM);
9031 
9032 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
9033 
9034 	t4_tp_read_la(sc, buf, NULL);
9035 	p = buf;
9036 
9037 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
9038 	case 2:
9039 		inc = 2;
9040 		show_func = tp_la_show2;
9041 		break;
9042 	case 3:
9043 		inc = 2;
9044 		show_func = tp_la_show3;
9045 		break;
9046 	default:
9047 		inc = 1;
9048 		show_func = tp_la_show;
9049 	}
9050 
9051 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
9052 		(*show_func)(sb, p, i);
9053 
9054 	rc = sbuf_finish(sb);
9055 	sbuf_delete(sb);
9056 	free(buf, M_CXGBE);
9057 	return (rc);
9058 }
9059 
9060 static int
9061 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
9062 {
9063 	struct adapter *sc = arg1;
9064 	struct sbuf *sb;
9065 	int rc;
9066 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
9067 
9068 	rc = sysctl_wire_old_buffer(req, 0);
9069 	if (rc != 0)
9070 		return (rc);
9071 
9072 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
9073 	if (sb == NULL)
9074 		return (ENOMEM);
9075 
9076 	t4_get_chan_txrate(sc, nrate, orate);
9077 
9078 	if (sc->chip_params->nchan > 2) {
9079 		sbuf_printf(sb, "              channel 0   channel 1"
9080 		    "   channel 2   channel 3\n");
9081 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
9082 		    nrate[0], nrate[1], nrate[2], nrate[3]);
9083 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
9084 		    orate[0], orate[1], orate[2], orate[3]);
9085 	} else {
9086 		sbuf_printf(sb, "              channel 0   channel 1\n");
9087 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
9088 		    nrate[0], nrate[1]);
9089 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
9090 		    orate[0], orate[1]);
9091 	}
9092 
9093 	rc = sbuf_finish(sb);
9094 	sbuf_delete(sb);
9095 
9096 	return (rc);
9097 }
9098 
9099 static int
9100 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
9101 {
9102 	struct adapter *sc = arg1;
9103 	struct sbuf *sb;
9104 	uint32_t *buf, *p;
9105 	int rc, i;
9106 
9107 	rc = sysctl_wire_old_buffer(req, 0);
9108 	if (rc != 0)
9109 		return (rc);
9110 
9111 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9112 	if (sb == NULL)
9113 		return (ENOMEM);
9114 
9115 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
9116 	    M_ZERO | M_WAITOK);
9117 
9118 	t4_ulprx_read_la(sc, buf);
9119 	p = buf;
9120 
9121 	sbuf_printf(sb, "      Pcmd        Type   Message"
9122 	    "                Data");
9123 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
9124 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
9125 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
9126 	}
9127 
9128 	rc = sbuf_finish(sb);
9129 	sbuf_delete(sb);
9130 	free(buf, M_CXGBE);
9131 	return (rc);
9132 }
9133 
9134 static int
9135 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
9136 {
9137 	struct adapter *sc = arg1;
9138 	struct sbuf *sb;
9139 	int rc, v;
9140 
9141 	MPASS(chip_id(sc) >= CHELSIO_T5);
9142 
9143 	rc = sysctl_wire_old_buffer(req, 0);
9144 	if (rc != 0)
9145 		return (rc);
9146 
9147 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9148 	if (sb == NULL)
9149 		return (ENOMEM);
9150 
9151 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
9152 	if (G_STATSOURCE_T5(v) == 7) {
9153 		int mode;
9154 
9155 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
9156 		if (mode == 0) {
9157 			sbuf_printf(sb, "total %d, incomplete %d",
9158 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
9159 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
9160 		} else if (mode == 1) {
9161 			sbuf_printf(sb, "total %d, data overflow %d",
9162 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
9163 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
9164 		} else {
9165 			sbuf_printf(sb, "unknown mode %d", mode);
9166 		}
9167 	}
9168 	rc = sbuf_finish(sb);
9169 	sbuf_delete(sb);
9170 
9171 	return (rc);
9172 }
9173 
9174 static int
9175 sysctl_cpus(SYSCTL_HANDLER_ARGS)
9176 {
9177 	struct adapter *sc = arg1;
9178 	enum cpu_sets op = arg2;
9179 	cpuset_t cpuset;
9180 	struct sbuf *sb;
9181 	int i, rc;
9182 
9183 	MPASS(op == LOCAL_CPUS || op == INTR_CPUS);
9184 
9185 	CPU_ZERO(&cpuset);
9186 	rc = bus_get_cpus(sc->dev, op, sizeof(cpuset), &cpuset);
9187 	if (rc != 0)
9188 		return (rc);
9189 
9190 	rc = sysctl_wire_old_buffer(req, 0);
9191 	if (rc != 0)
9192 		return (rc);
9193 
9194 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9195 	if (sb == NULL)
9196 		return (ENOMEM);
9197 
9198 	CPU_FOREACH(i)
9199 		sbuf_printf(sb, "%d ", i);
9200 	rc = sbuf_finish(sb);
9201 	sbuf_delete(sb);
9202 
9203 	return (rc);
9204 }
9205 
9206 #ifdef TCP_OFFLOAD
9207 static int
9208 sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS)
9209 {
9210 	struct adapter *sc = arg1;
9211 	int *old_ports, *new_ports;
9212 	int i, new_count, rc;
9213 
9214 	if (req->newptr == NULL && req->oldptr == NULL)
9215 		return (SYSCTL_OUT(req, NULL, imax(sc->tt.num_tls_rx_ports, 1) *
9216 		    sizeof(sc->tt.tls_rx_ports[0])));
9217 
9218 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tlsrx");
9219 	if (rc)
9220 		return (rc);
9221 
9222 	if (sc->tt.num_tls_rx_ports == 0) {
9223 		i = -1;
9224 		rc = SYSCTL_OUT(req, &i, sizeof(i));
9225 	} else
9226 		rc = SYSCTL_OUT(req, sc->tt.tls_rx_ports,
9227 		    sc->tt.num_tls_rx_ports * sizeof(sc->tt.tls_rx_ports[0]));
9228 	if (rc == 0 && req->newptr != NULL) {
9229 		new_count = req->newlen / sizeof(new_ports[0]);
9230 		new_ports = malloc(new_count * sizeof(new_ports[0]), M_CXGBE,
9231 		    M_WAITOK);
9232 		rc = SYSCTL_IN(req, new_ports, new_count *
9233 		    sizeof(new_ports[0]));
9234 		if (rc)
9235 			goto err;
9236 
9237 		/* Allow setting to a single '-1' to clear the list. */
9238 		if (new_count == 1 && new_ports[0] == -1) {
9239 			ADAPTER_LOCK(sc);
9240 			old_ports = sc->tt.tls_rx_ports;
9241 			sc->tt.tls_rx_ports = NULL;
9242 			sc->tt.num_tls_rx_ports = 0;
9243 			ADAPTER_UNLOCK(sc);
9244 			free(old_ports, M_CXGBE);
9245 		} else {
9246 			for (i = 0; i < new_count; i++) {
9247 				if (new_ports[i] < 1 ||
9248 				    new_ports[i] > IPPORT_MAX) {
9249 					rc = EINVAL;
9250 					goto err;
9251 				}
9252 			}
9253 
9254 			ADAPTER_LOCK(sc);
9255 			old_ports = sc->tt.tls_rx_ports;
9256 			sc->tt.tls_rx_ports = new_ports;
9257 			sc->tt.num_tls_rx_ports = new_count;
9258 			ADAPTER_UNLOCK(sc);
9259 			free(old_ports, M_CXGBE);
9260 			new_ports = NULL;
9261 		}
9262 	err:
9263 		free(new_ports, M_CXGBE);
9264 	}
9265 	end_synchronized_op(sc, 0);
9266 	return (rc);
9267 }
9268 
9269 static void
9270 unit_conv(char *buf, size_t len, u_int val, u_int factor)
9271 {
9272 	u_int rem = val % factor;
9273 
9274 	if (rem == 0)
9275 		snprintf(buf, len, "%u", val / factor);
9276 	else {
9277 		while (rem % 10 == 0)
9278 			rem /= 10;
9279 		snprintf(buf, len, "%u.%u", val / factor, rem);
9280 	}
9281 }
9282 
9283 static int
9284 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
9285 {
9286 	struct adapter *sc = arg1;
9287 	char buf[16];
9288 	u_int res, re;
9289 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9290 
9291 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
9292 	switch (arg2) {
9293 	case 0:
9294 		/* timer_tick */
9295 		re = G_TIMERRESOLUTION(res);
9296 		break;
9297 	case 1:
9298 		/* TCP timestamp tick */
9299 		re = G_TIMESTAMPRESOLUTION(res);
9300 		break;
9301 	case 2:
9302 		/* DACK tick */
9303 		re = G_DELAYEDACKRESOLUTION(res);
9304 		break;
9305 	default:
9306 		return (EDOOFUS);
9307 	}
9308 
9309 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
9310 
9311 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
9312 }
9313 
9314 static int
9315 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
9316 {
9317 	struct adapter *sc = arg1;
9318 	u_int res, dack_re, v;
9319 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9320 
9321 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
9322 	dack_re = G_DELAYEDACKRESOLUTION(res);
9323 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
9324 
9325 	return (sysctl_handle_int(oidp, &v, 0, req));
9326 }
9327 
9328 static int
9329 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
9330 {
9331 	struct adapter *sc = arg1;
9332 	int reg = arg2;
9333 	u_int tre;
9334 	u_long tp_tick_us, v;
9335 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9336 
9337 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
9338 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
9339 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
9340 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
9341 
9342 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
9343 	tp_tick_us = (cclk_ps << tre) / 1000000;
9344 
9345 	if (reg == A_TP_INIT_SRTT)
9346 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
9347 	else
9348 		v = tp_tick_us * t4_read_reg(sc, reg);
9349 
9350 	return (sysctl_handle_long(oidp, &v, 0, req));
9351 }
9352 
9353 /*
9354  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
9355  * passed to this function.
9356  */
9357 static int
9358 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
9359 {
9360 	struct adapter *sc = arg1;
9361 	int idx = arg2;
9362 	u_int v;
9363 
9364 	MPASS(idx >= 0 && idx <= 24);
9365 
9366 	v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
9367 
9368 	return (sysctl_handle_int(oidp, &v, 0, req));
9369 }
9370 
9371 static int
9372 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
9373 {
9374 	struct adapter *sc = arg1;
9375 	int idx = arg2;
9376 	u_int shift, v, r;
9377 
9378 	MPASS(idx >= 0 && idx < 16);
9379 
9380 	r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
9381 	shift = (idx & 3) << 3;
9382 	v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
9383 
9384 	return (sysctl_handle_int(oidp, &v, 0, req));
9385 }
9386 
9387 static int
9388 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
9389 {
9390 	struct vi_info *vi = arg1;
9391 	struct adapter *sc = vi->pi->adapter;
9392 	int idx, rc, i;
9393 	struct sge_ofld_rxq *ofld_rxq;
9394 	uint8_t v;
9395 
9396 	idx = vi->ofld_tmr_idx;
9397 
9398 	rc = sysctl_handle_int(oidp, &idx, 0, req);
9399 	if (rc != 0 || req->newptr == NULL)
9400 		return (rc);
9401 
9402 	if (idx < 0 || idx >= SGE_NTIMERS)
9403 		return (EINVAL);
9404 
9405 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9406 	    "t4otmr");
9407 	if (rc)
9408 		return (rc);
9409 
9410 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
9411 	for_each_ofld_rxq(vi, i, ofld_rxq) {
9412 #ifdef atomic_store_rel_8
9413 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
9414 #else
9415 		ofld_rxq->iq.intr_params = v;
9416 #endif
9417 	}
9418 	vi->ofld_tmr_idx = idx;
9419 
9420 	end_synchronized_op(sc, LOCK_HELD);
9421 	return (0);
9422 }
9423 
9424 static int
9425 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
9426 {
9427 	struct vi_info *vi = arg1;
9428 	struct adapter *sc = vi->pi->adapter;
9429 	int idx, rc;
9430 
9431 	idx = vi->ofld_pktc_idx;
9432 
9433 	rc = sysctl_handle_int(oidp, &idx, 0, req);
9434 	if (rc != 0 || req->newptr == NULL)
9435 		return (rc);
9436 
9437 	if (idx < -1 || idx >= SGE_NCOUNTERS)
9438 		return (EINVAL);
9439 
9440 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9441 	    "t4opktc");
9442 	if (rc)
9443 		return (rc);
9444 
9445 	if (vi->flags & VI_INIT_DONE)
9446 		rc = EBUSY; /* cannot be changed once the queues are created */
9447 	else
9448 		vi->ofld_pktc_idx = idx;
9449 
9450 	end_synchronized_op(sc, LOCK_HELD);
9451 	return (rc);
9452 }
9453 #endif
9454 
9455 static int
9456 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
9457 {
9458 	int rc;
9459 
9460 	if (cntxt->cid > M_CTXTQID)
9461 		return (EINVAL);
9462 
9463 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
9464 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
9465 		return (EINVAL);
9466 
9467 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
9468 	if (rc)
9469 		return (rc);
9470 
9471 	if (sc->flags & FW_OK) {
9472 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
9473 		    &cntxt->data[0]);
9474 		if (rc == 0)
9475 			goto done;
9476 	}
9477 
9478 	/*
9479 	 * Read via firmware failed or wasn't even attempted.  Read directly via
9480 	 * the backdoor.
9481 	 */
9482 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
9483 done:
9484 	end_synchronized_op(sc, 0);
9485 	return (rc);
9486 }
9487 
9488 static int
9489 load_fw(struct adapter *sc, struct t4_data *fw)
9490 {
9491 	int rc;
9492 	uint8_t *fw_data;
9493 
9494 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
9495 	if (rc)
9496 		return (rc);
9497 
9498 	/*
9499 	 * The firmware, with the sole exception of the memory parity error
9500 	 * handler, runs from memory and not flash.  It is almost always safe to
9501 	 * install a new firmware on a running system.  Just set bit 1 in
9502 	 * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
9503 	 */
9504 	if (sc->flags & FULL_INIT_DONE &&
9505 	    (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
9506 		rc = EBUSY;
9507 		goto done;
9508 	}
9509 
9510 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
9511 	if (fw_data == NULL) {
9512 		rc = ENOMEM;
9513 		goto done;
9514 	}
9515 
9516 	rc = copyin(fw->data, fw_data, fw->len);
9517 	if (rc == 0)
9518 		rc = -t4_load_fw(sc, fw_data, fw->len);
9519 
9520 	free(fw_data, M_CXGBE);
9521 done:
9522 	end_synchronized_op(sc, 0);
9523 	return (rc);
9524 }
9525 
9526 static int
9527 load_cfg(struct adapter *sc, struct t4_data *cfg)
9528 {
9529 	int rc;
9530 	uint8_t *cfg_data = NULL;
9531 
9532 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9533 	if (rc)
9534 		return (rc);
9535 
9536 	if (cfg->len == 0) {
9537 		/* clear */
9538 		rc = -t4_load_cfg(sc, NULL, 0);
9539 		goto done;
9540 	}
9541 
9542 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
9543 	if (cfg_data == NULL) {
9544 		rc = ENOMEM;
9545 		goto done;
9546 	}
9547 
9548 	rc = copyin(cfg->data, cfg_data, cfg->len);
9549 	if (rc == 0)
9550 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
9551 
9552 	free(cfg_data, M_CXGBE);
9553 done:
9554 	end_synchronized_op(sc, 0);
9555 	return (rc);
9556 }
9557 
9558 static int
9559 load_boot(struct adapter *sc, struct t4_bootrom *br)
9560 {
9561 	int rc;
9562 	uint8_t *br_data = NULL;
9563 	u_int offset;
9564 
9565 	if (br->len > 1024 * 1024)
9566 		return (EFBIG);
9567 
9568 	if (br->pf_offset == 0) {
9569 		/* pfidx */
9570 		if (br->pfidx_addr > 7)
9571 			return (EINVAL);
9572 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
9573 		    A_PCIE_PF_EXPROM_OFST)));
9574 	} else if (br->pf_offset == 1) {
9575 		/* offset */
9576 		offset = G_OFFSET(br->pfidx_addr);
9577 	} else {
9578 		return (EINVAL);
9579 	}
9580 
9581 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
9582 	if (rc)
9583 		return (rc);
9584 
9585 	if (br->len == 0) {
9586 		/* clear */
9587 		rc = -t4_load_boot(sc, NULL, offset, 0);
9588 		goto done;
9589 	}
9590 
9591 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
9592 	if (br_data == NULL) {
9593 		rc = ENOMEM;
9594 		goto done;
9595 	}
9596 
9597 	rc = copyin(br->data, br_data, br->len);
9598 	if (rc == 0)
9599 		rc = -t4_load_boot(sc, br_data, offset, br->len);
9600 
9601 	free(br_data, M_CXGBE);
9602 done:
9603 	end_synchronized_op(sc, 0);
9604 	return (rc);
9605 }
9606 
9607 static int
9608 load_bootcfg(struct adapter *sc, struct t4_data *bc)
9609 {
9610 	int rc;
9611 	uint8_t *bc_data = NULL;
9612 
9613 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9614 	if (rc)
9615 		return (rc);
9616 
9617 	if (bc->len == 0) {
9618 		/* clear */
9619 		rc = -t4_load_bootcfg(sc, NULL, 0);
9620 		goto done;
9621 	}
9622 
9623 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
9624 	if (bc_data == NULL) {
9625 		rc = ENOMEM;
9626 		goto done;
9627 	}
9628 
9629 	rc = copyin(bc->data, bc_data, bc->len);
9630 	if (rc == 0)
9631 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
9632 
9633 	free(bc_data, M_CXGBE);
9634 done:
9635 	end_synchronized_op(sc, 0);
9636 	return (rc);
9637 }
9638 
9639 static int
9640 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
9641 {
9642 	int rc;
9643 	struct cudbg_init *cudbg;
9644 	void *handle, *buf;
9645 
9646 	/* buf is large, don't block if no memory is available */
9647 	buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
9648 	if (buf == NULL)
9649 		return (ENOMEM);
9650 
9651 	handle = cudbg_alloc_handle();
9652 	if (handle == NULL) {
9653 		rc = ENOMEM;
9654 		goto done;
9655 	}
9656 
9657 	cudbg = cudbg_get_init(handle);
9658 	cudbg->adap = sc;
9659 	cudbg->print = (cudbg_print_cb)printf;
9660 
9661 #ifndef notyet
9662 	device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
9663 	    __func__, dump->wr_flash, dump->len, dump->data);
9664 #endif
9665 
9666 	if (dump->wr_flash)
9667 		cudbg->use_flash = 1;
9668 	MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
9669 	memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
9670 
9671 	rc = cudbg_collect(handle, buf, &dump->len);
9672 	if (rc != 0)
9673 		goto done;
9674 
9675 	rc = copyout(buf, dump->data, dump->len);
9676 done:
9677 	cudbg_free_handle(handle);
9678 	free(buf, M_CXGBE);
9679 	return (rc);
9680 }
9681 
9682 static void
9683 free_offload_policy(struct t4_offload_policy *op)
9684 {
9685 	struct offload_rule *r;
9686 	int i;
9687 
9688 	if (op == NULL)
9689 		return;
9690 
9691 	r = &op->rule[0];
9692 	for (i = 0; i < op->nrules; i++, r++) {
9693 		free(r->bpf_prog.bf_insns, M_CXGBE);
9694 	}
9695 	free(op->rule, M_CXGBE);
9696 	free(op, M_CXGBE);
9697 }
9698 
9699 static int
9700 set_offload_policy(struct adapter *sc, struct t4_offload_policy *uop)
9701 {
9702 	int i, rc, len;
9703 	struct t4_offload_policy *op, *old;
9704 	struct bpf_program *bf;
9705 	const struct offload_settings *s;
9706 	struct offload_rule *r;
9707 	void *u;
9708 
9709 	if (!is_offload(sc))
9710 		return (ENODEV);
9711 
9712 	if (uop->nrules == 0) {
9713 		/* Delete installed policies. */
9714 		op = NULL;
9715 		goto set_policy;
9716 	} else if (uop->nrules > 256) { /* arbitrary */
9717 		return (E2BIG);
9718 	}
9719 
9720 	/* Copy userspace offload policy to kernel */
9721 	op = malloc(sizeof(*op), M_CXGBE, M_ZERO | M_WAITOK);
9722 	op->nrules = uop->nrules;
9723 	len = op->nrules * sizeof(struct offload_rule);
9724 	op->rule = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9725 	rc = copyin(uop->rule, op->rule, len);
9726 	if (rc) {
9727 		free(op->rule, M_CXGBE);
9728 		free(op, M_CXGBE);
9729 		return (rc);
9730 	}
9731 
9732 	r = &op->rule[0];
9733 	for (i = 0; i < op->nrules; i++, r++) {
9734 
9735 		/* Validate open_type */
9736 		if (r->open_type != OPEN_TYPE_LISTEN &&
9737 		    r->open_type != OPEN_TYPE_ACTIVE &&
9738 		    r->open_type != OPEN_TYPE_PASSIVE &&
9739 		    r->open_type != OPEN_TYPE_DONTCARE) {
9740 error:
9741 			/*
9742 			 * Rules 0 to i have malloc'd filters that need to be
9743 			 * freed.  Rules i+1 to nrules have userspace pointers
9744 			 * and should be left alone.
9745 			 */
9746 			op->nrules = i;
9747 			free_offload_policy(op);
9748 			return (rc);
9749 		}
9750 
9751 		/* Validate settings */
9752 		s = &r->settings;
9753 		if ((s->offload != 0 && s->offload != 1) ||
9754 		    s->cong_algo < -1 || s->cong_algo > CONG_ALG_HIGHSPEED ||
9755 		    s->sched_class < -1 ||
9756 		    s->sched_class >= sc->chip_params->nsched_cls) {
9757 			rc = EINVAL;
9758 			goto error;
9759 		}
9760 
9761 		bf = &r->bpf_prog;
9762 		u = bf->bf_insns;	/* userspace ptr */
9763 		bf->bf_insns = NULL;
9764 		if (bf->bf_len == 0) {
9765 			/* legal, matches everything */
9766 			continue;
9767 		}
9768 		len = bf->bf_len * sizeof(*bf->bf_insns);
9769 		bf->bf_insns = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9770 		rc = copyin(u, bf->bf_insns, len);
9771 		if (rc != 0)
9772 			goto error;
9773 
9774 		if (!bpf_validate(bf->bf_insns, bf->bf_len)) {
9775 			rc = EINVAL;
9776 			goto error;
9777 		}
9778 	}
9779 set_policy:
9780 	rw_wlock(&sc->policy_lock);
9781 	old = sc->policy;
9782 	sc->policy = op;
9783 	rw_wunlock(&sc->policy_lock);
9784 	free_offload_policy(old);
9785 
9786 	return (0);
9787 }
9788 
9789 #define MAX_READ_BUF_SIZE (128 * 1024)
9790 static int
9791 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
9792 {
9793 	uint32_t addr, remaining, n;
9794 	uint32_t *buf;
9795 	int rc;
9796 	uint8_t *dst;
9797 
9798 	rc = validate_mem_range(sc, mr->addr, mr->len);
9799 	if (rc != 0)
9800 		return (rc);
9801 
9802 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
9803 	addr = mr->addr;
9804 	remaining = mr->len;
9805 	dst = (void *)mr->data;
9806 
9807 	while (remaining) {
9808 		n = min(remaining, MAX_READ_BUF_SIZE);
9809 		read_via_memwin(sc, 2, addr, buf, n);
9810 
9811 		rc = copyout(buf, dst, n);
9812 		if (rc != 0)
9813 			break;
9814 
9815 		dst += n;
9816 		remaining -= n;
9817 		addr += n;
9818 	}
9819 
9820 	free(buf, M_CXGBE);
9821 	return (rc);
9822 }
9823 #undef MAX_READ_BUF_SIZE
9824 
9825 static int
9826 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
9827 {
9828 	int rc;
9829 
9830 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
9831 		return (EINVAL);
9832 
9833 	if (i2cd->len > sizeof(i2cd->data))
9834 		return (EFBIG);
9835 
9836 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9837 	if (rc)
9838 		return (rc);
9839 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9840 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
9841 	end_synchronized_op(sc, 0);
9842 
9843 	return (rc);
9844 }
9845 
9846 static int
9847 clear_stats(struct adapter *sc, u_int port_id)
9848 {
9849 	int i, v, bg_map;
9850 	struct port_info *pi;
9851 	struct vi_info *vi;
9852 	struct sge_rxq *rxq;
9853 	struct sge_txq *txq;
9854 	struct sge_wrq *wrq;
9855 #ifdef TCP_OFFLOAD
9856 	struct sge_ofld_rxq *ofld_rxq;
9857 #endif
9858 
9859 	if (port_id >= sc->params.nports)
9860 		return (EINVAL);
9861 	pi = sc->port[port_id];
9862 	if (pi == NULL)
9863 		return (EIO);
9864 
9865 	/* MAC stats */
9866 	t4_clr_port_stats(sc, pi->tx_chan);
9867 	pi->tx_parse_error = 0;
9868 	pi->tnl_cong_drops = 0;
9869 	mtx_lock(&sc->reg_lock);
9870 	for_each_vi(pi, v, vi) {
9871 		if (vi->flags & VI_INIT_DONE)
9872 			t4_clr_vi_stats(sc, vi->vin);
9873 	}
9874 	bg_map = pi->mps_bg_map;
9875 	v = 0;	/* reuse */
9876 	while (bg_map) {
9877 		i = ffs(bg_map) - 1;
9878 		t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
9879 		    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
9880 		bg_map &= ~(1 << i);
9881 	}
9882 	mtx_unlock(&sc->reg_lock);
9883 
9884 	/*
9885 	 * Since this command accepts a port, clear stats for
9886 	 * all VIs on this port.
9887 	 */
9888 	for_each_vi(pi, v, vi) {
9889 		if (vi->flags & VI_INIT_DONE) {
9890 
9891 			for_each_rxq(vi, i, rxq) {
9892 #if defined(INET) || defined(INET6)
9893 				rxq->lro.lro_queued = 0;
9894 				rxq->lro.lro_flushed = 0;
9895 #endif
9896 				rxq->rxcsum = 0;
9897 				rxq->vlan_extraction = 0;
9898 
9899 				rxq->fl.mbuf_allocated = 0;
9900 				rxq->fl.mbuf_inlined = 0;
9901 				rxq->fl.cl_allocated = 0;
9902 				rxq->fl.cl_recycled = 0;
9903 				rxq->fl.cl_fast_recycled = 0;
9904 			}
9905 
9906 			for_each_txq(vi, i, txq) {
9907 				txq->txcsum = 0;
9908 				txq->tso_wrs = 0;
9909 				txq->vlan_insertion = 0;
9910 				txq->imm_wrs = 0;
9911 				txq->sgl_wrs = 0;
9912 				txq->txpkt_wrs = 0;
9913 				txq->txpkts0_wrs = 0;
9914 				txq->txpkts1_wrs = 0;
9915 				txq->txpkts0_pkts = 0;
9916 				txq->txpkts1_pkts = 0;
9917 				txq->raw_wrs = 0;
9918 				mp_ring_reset_stats(txq->r);
9919 			}
9920 
9921 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
9922 			for_each_ofld_txq(vi, i, wrq) {
9923 				wrq->tx_wrs_direct = 0;
9924 				wrq->tx_wrs_copied = 0;
9925 			}
9926 #endif
9927 #ifdef TCP_OFFLOAD
9928 			for_each_ofld_rxq(vi, i, ofld_rxq) {
9929 				ofld_rxq->fl.mbuf_allocated = 0;
9930 				ofld_rxq->fl.mbuf_inlined = 0;
9931 				ofld_rxq->fl.cl_allocated = 0;
9932 				ofld_rxq->fl.cl_recycled = 0;
9933 				ofld_rxq->fl.cl_fast_recycled = 0;
9934 			}
9935 #endif
9936 
9937 			if (IS_MAIN_VI(vi)) {
9938 				wrq = &sc->sge.ctrlq[pi->port_id];
9939 				wrq->tx_wrs_direct = 0;
9940 				wrq->tx_wrs_copied = 0;
9941 			}
9942 		}
9943 	}
9944 
9945 	return (0);
9946 }
9947 
9948 int
9949 t4_os_find_pci_capability(struct adapter *sc, int cap)
9950 {
9951 	int i;
9952 
9953 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9954 }
9955 
9956 int
9957 t4_os_pci_save_state(struct adapter *sc)
9958 {
9959 	device_t dev;
9960 	struct pci_devinfo *dinfo;
9961 
9962 	dev = sc->dev;
9963 	dinfo = device_get_ivars(dev);
9964 
9965 	pci_cfg_save(dev, dinfo, 0);
9966 	return (0);
9967 }
9968 
9969 int
9970 t4_os_pci_restore_state(struct adapter *sc)
9971 {
9972 	device_t dev;
9973 	struct pci_devinfo *dinfo;
9974 
9975 	dev = sc->dev;
9976 	dinfo = device_get_ivars(dev);
9977 
9978 	pci_cfg_restore(dev, dinfo);
9979 	return (0);
9980 }
9981 
9982 void
9983 t4_os_portmod_changed(struct port_info *pi)
9984 {
9985 	struct adapter *sc = pi->adapter;
9986 	struct vi_info *vi;
9987 	struct ifnet *ifp;
9988 	static const char *mod_str[] = {
9989 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9990 	};
9991 
9992 	KASSERT((pi->flags & FIXED_IFMEDIA) == 0,
9993 	    ("%s: port_type %u", __func__, pi->port_type));
9994 
9995 	vi = &pi->vi[0];
9996 	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9997 		PORT_LOCK(pi);
9998 		build_medialist(pi);
9999 		if (pi->mod_type != FW_PORT_MOD_TYPE_NONE) {
10000 			fixup_link_config(pi);
10001 			apply_link_config(pi);
10002 		}
10003 		PORT_UNLOCK(pi);
10004 		end_synchronized_op(sc, LOCK_HELD);
10005 	}
10006 
10007 	ifp = vi->ifp;
10008 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
10009 		if_printf(ifp, "transceiver unplugged.\n");
10010 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
10011 		if_printf(ifp, "unknown transceiver inserted.\n");
10012 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
10013 		if_printf(ifp, "unsupported transceiver inserted.\n");
10014 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
10015 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
10016 		    port_top_speed(pi), mod_str[pi->mod_type]);
10017 	} else {
10018 		if_printf(ifp, "transceiver (type %d) inserted.\n",
10019 		    pi->mod_type);
10020 	}
10021 }
10022 
10023 void
10024 t4_os_link_changed(struct port_info *pi)
10025 {
10026 	struct vi_info *vi;
10027 	struct ifnet *ifp;
10028 	struct link_config *lc;
10029 	int v;
10030 
10031 	PORT_LOCK_ASSERT_OWNED(pi);
10032 
10033 	for_each_vi(pi, v, vi) {
10034 		ifp = vi->ifp;
10035 		if (ifp == NULL)
10036 			continue;
10037 
10038 		lc = &pi->link_cfg;
10039 		if (lc->link_ok) {
10040 			ifp->if_baudrate = IF_Mbps(lc->speed);
10041 			if_link_state_change(ifp, LINK_STATE_UP);
10042 		} else {
10043 			if_link_state_change(ifp, LINK_STATE_DOWN);
10044 		}
10045 	}
10046 }
10047 
10048 void
10049 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
10050 {
10051 	struct adapter *sc;
10052 
10053 	sx_slock(&t4_list_lock);
10054 	SLIST_FOREACH(sc, &t4_list, link) {
10055 		/*
10056 		 * func should not make any assumptions about what state sc is
10057 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
10058 		 */
10059 		func(sc, arg);
10060 	}
10061 	sx_sunlock(&t4_list_lock);
10062 }
10063 
10064 static int
10065 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
10066     struct thread *td)
10067 {
10068 	int rc;
10069 	struct adapter *sc = dev->si_drv1;
10070 
10071 	rc = priv_check(td, PRIV_DRIVER);
10072 	if (rc != 0)
10073 		return (rc);
10074 
10075 	switch (cmd) {
10076 	case CHELSIO_T4_GETREG: {
10077 		struct t4_reg *edata = (struct t4_reg *)data;
10078 
10079 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
10080 			return (EFAULT);
10081 
10082 		if (edata->size == 4)
10083 			edata->val = t4_read_reg(sc, edata->addr);
10084 		else if (edata->size == 8)
10085 			edata->val = t4_read_reg64(sc, edata->addr);
10086 		else
10087 			return (EINVAL);
10088 
10089 		break;
10090 	}
10091 	case CHELSIO_T4_SETREG: {
10092 		struct t4_reg *edata = (struct t4_reg *)data;
10093 
10094 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
10095 			return (EFAULT);
10096 
10097 		if (edata->size == 4) {
10098 			if (edata->val & 0xffffffff00000000)
10099 				return (EINVAL);
10100 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
10101 		} else if (edata->size == 8)
10102 			t4_write_reg64(sc, edata->addr, edata->val);
10103 		else
10104 			return (EINVAL);
10105 		break;
10106 	}
10107 	case CHELSIO_T4_REGDUMP: {
10108 		struct t4_regdump *regs = (struct t4_regdump *)data;
10109 		int reglen = t4_get_regs_len(sc);
10110 		uint8_t *buf;
10111 
10112 		if (regs->len < reglen) {
10113 			regs->len = reglen; /* hint to the caller */
10114 			return (ENOBUFS);
10115 		}
10116 
10117 		regs->len = reglen;
10118 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
10119 		get_regs(sc, regs, buf);
10120 		rc = copyout(buf, regs->data, reglen);
10121 		free(buf, M_CXGBE);
10122 		break;
10123 	}
10124 	case CHELSIO_T4_GET_FILTER_MODE:
10125 		rc = get_filter_mode(sc, (uint32_t *)data);
10126 		break;
10127 	case CHELSIO_T4_SET_FILTER_MODE:
10128 		rc = set_filter_mode(sc, *(uint32_t *)data);
10129 		break;
10130 	case CHELSIO_T4_GET_FILTER:
10131 		rc = get_filter(sc, (struct t4_filter *)data);
10132 		break;
10133 	case CHELSIO_T4_SET_FILTER:
10134 		rc = set_filter(sc, (struct t4_filter *)data);
10135 		break;
10136 	case CHELSIO_T4_DEL_FILTER:
10137 		rc = del_filter(sc, (struct t4_filter *)data);
10138 		break;
10139 	case CHELSIO_T4_GET_SGE_CONTEXT:
10140 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
10141 		break;
10142 	case CHELSIO_T4_LOAD_FW:
10143 		rc = load_fw(sc, (struct t4_data *)data);
10144 		break;
10145 	case CHELSIO_T4_GET_MEM:
10146 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
10147 		break;
10148 	case CHELSIO_T4_GET_I2C:
10149 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
10150 		break;
10151 	case CHELSIO_T4_CLEAR_STATS:
10152 		rc = clear_stats(sc, *(uint32_t *)data);
10153 		break;
10154 	case CHELSIO_T4_SCHED_CLASS:
10155 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
10156 		break;
10157 	case CHELSIO_T4_SCHED_QUEUE:
10158 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
10159 		break;
10160 	case CHELSIO_T4_GET_TRACER:
10161 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
10162 		break;
10163 	case CHELSIO_T4_SET_TRACER:
10164 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
10165 		break;
10166 	case CHELSIO_T4_LOAD_CFG:
10167 		rc = load_cfg(sc, (struct t4_data *)data);
10168 		break;
10169 	case CHELSIO_T4_LOAD_BOOT:
10170 		rc = load_boot(sc, (struct t4_bootrom *)data);
10171 		break;
10172 	case CHELSIO_T4_LOAD_BOOTCFG:
10173 		rc = load_bootcfg(sc, (struct t4_data *)data);
10174 		break;
10175 	case CHELSIO_T4_CUDBG_DUMP:
10176 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
10177 		break;
10178 	case CHELSIO_T4_SET_OFLD_POLICY:
10179 		rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
10180 		break;
10181 	default:
10182 		rc = ENOTTY;
10183 	}
10184 
10185 	return (rc);
10186 }
10187 
10188 #ifdef TCP_OFFLOAD
10189 static int
10190 toe_capability(struct vi_info *vi, int enable)
10191 {
10192 	int rc;
10193 	struct port_info *pi = vi->pi;
10194 	struct adapter *sc = pi->adapter;
10195 
10196 	ASSERT_SYNCHRONIZED_OP(sc);
10197 
10198 	if (!is_offload(sc))
10199 		return (ENODEV);
10200 
10201 	if (enable) {
10202 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
10203 			/* TOE is already enabled. */
10204 			return (0);
10205 		}
10206 
10207 		/*
10208 		 * We need the port's queues around so that we're able to send
10209 		 * and receive CPLs to/from the TOE even if the ifnet for this
10210 		 * port has never been UP'd administratively.
10211 		 */
10212 		if (!(vi->flags & VI_INIT_DONE)) {
10213 			rc = vi_full_init(vi);
10214 			if (rc)
10215 				return (rc);
10216 		}
10217 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
10218 			rc = vi_full_init(&pi->vi[0]);
10219 			if (rc)
10220 				return (rc);
10221 		}
10222 
10223 		if (isset(&sc->offload_map, pi->port_id)) {
10224 			/* TOE is enabled on another VI of this port. */
10225 			pi->uld_vis++;
10226 			return (0);
10227 		}
10228 
10229 		if (!uld_active(sc, ULD_TOM)) {
10230 			rc = t4_activate_uld(sc, ULD_TOM);
10231 			if (rc == EAGAIN) {
10232 				log(LOG_WARNING,
10233 				    "You must kldload t4_tom.ko before trying "
10234 				    "to enable TOE on a cxgbe interface.\n");
10235 			}
10236 			if (rc != 0)
10237 				return (rc);
10238 			KASSERT(sc->tom_softc != NULL,
10239 			    ("%s: TOM activated but softc NULL", __func__));
10240 			KASSERT(uld_active(sc, ULD_TOM),
10241 			    ("%s: TOM activated but flag not set", __func__));
10242 		}
10243 
10244 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
10245 		if (!uld_active(sc, ULD_IWARP))
10246 			(void) t4_activate_uld(sc, ULD_IWARP);
10247 		if (!uld_active(sc, ULD_ISCSI))
10248 			(void) t4_activate_uld(sc, ULD_ISCSI);
10249 
10250 		pi->uld_vis++;
10251 		setbit(&sc->offload_map, pi->port_id);
10252 	} else {
10253 		pi->uld_vis--;
10254 
10255 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
10256 			return (0);
10257 
10258 		KASSERT(uld_active(sc, ULD_TOM),
10259 		    ("%s: TOM never initialized?", __func__));
10260 		clrbit(&sc->offload_map, pi->port_id);
10261 	}
10262 
10263 	return (0);
10264 }
10265 
10266 /*
10267  * Add an upper layer driver to the global list.
10268  */
10269 int
10270 t4_register_uld(struct uld_info *ui)
10271 {
10272 	int rc = 0;
10273 	struct uld_info *u;
10274 
10275 	sx_xlock(&t4_uld_list_lock);
10276 	SLIST_FOREACH(u, &t4_uld_list, link) {
10277 	    if (u->uld_id == ui->uld_id) {
10278 		    rc = EEXIST;
10279 		    goto done;
10280 	    }
10281 	}
10282 
10283 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
10284 	ui->refcount = 0;
10285 done:
10286 	sx_xunlock(&t4_uld_list_lock);
10287 	return (rc);
10288 }
10289 
10290 int
10291 t4_unregister_uld(struct uld_info *ui)
10292 {
10293 	int rc = EINVAL;
10294 	struct uld_info *u;
10295 
10296 	sx_xlock(&t4_uld_list_lock);
10297 
10298 	SLIST_FOREACH(u, &t4_uld_list, link) {
10299 	    if (u == ui) {
10300 		    if (ui->refcount > 0) {
10301 			    rc = EBUSY;
10302 			    goto done;
10303 		    }
10304 
10305 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
10306 		    rc = 0;
10307 		    goto done;
10308 	    }
10309 	}
10310 done:
10311 	sx_xunlock(&t4_uld_list_lock);
10312 	return (rc);
10313 }
10314 
10315 int
10316 t4_activate_uld(struct adapter *sc, int id)
10317 {
10318 	int rc;
10319 	struct uld_info *ui;
10320 
10321 	ASSERT_SYNCHRONIZED_OP(sc);
10322 
10323 	if (id < 0 || id > ULD_MAX)
10324 		return (EINVAL);
10325 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
10326 
10327 	sx_slock(&t4_uld_list_lock);
10328 
10329 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10330 		if (ui->uld_id == id) {
10331 			if (!(sc->flags & FULL_INIT_DONE)) {
10332 				rc = adapter_full_init(sc);
10333 				if (rc != 0)
10334 					break;
10335 			}
10336 
10337 			rc = ui->activate(sc);
10338 			if (rc == 0) {
10339 				setbit(&sc->active_ulds, id);
10340 				ui->refcount++;
10341 			}
10342 			break;
10343 		}
10344 	}
10345 
10346 	sx_sunlock(&t4_uld_list_lock);
10347 
10348 	return (rc);
10349 }
10350 
10351 int
10352 t4_deactivate_uld(struct adapter *sc, int id)
10353 {
10354 	int rc;
10355 	struct uld_info *ui;
10356 
10357 	ASSERT_SYNCHRONIZED_OP(sc);
10358 
10359 	if (id < 0 || id > ULD_MAX)
10360 		return (EINVAL);
10361 	rc = ENXIO;
10362 
10363 	sx_slock(&t4_uld_list_lock);
10364 
10365 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10366 		if (ui->uld_id == id) {
10367 			rc = ui->deactivate(sc);
10368 			if (rc == 0) {
10369 				clrbit(&sc->active_ulds, id);
10370 				ui->refcount--;
10371 			}
10372 			break;
10373 		}
10374 	}
10375 
10376 	sx_sunlock(&t4_uld_list_lock);
10377 
10378 	return (rc);
10379 }
10380 
10381 int
10382 uld_active(struct adapter *sc, int uld_id)
10383 {
10384 
10385 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
10386 
10387 	return (isset(&sc->active_ulds, uld_id));
10388 }
10389 #endif
10390 
10391 /*
10392  * t  = ptr to tunable.
10393  * nc = number of CPUs.
10394  * c  = compiled in default for that tunable.
10395  */
10396 static void
10397 calculate_nqueues(int *t, int nc, const int c)
10398 {
10399 	int nq;
10400 
10401 	if (*t > 0)
10402 		return;
10403 	nq = *t < 0 ? -*t : c;
10404 	*t = min(nc, nq);
10405 }
10406 
10407 /*
10408  * Come up with reasonable defaults for some of the tunables, provided they're
10409  * not set by the user (in which case we'll use the values as is).
10410  */
10411 static void
10412 tweak_tunables(void)
10413 {
10414 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
10415 
10416 	if (t4_ntxq < 1) {
10417 #ifdef RSS
10418 		t4_ntxq = rss_getnumbuckets();
10419 #else
10420 		calculate_nqueues(&t4_ntxq, nc, NTXQ);
10421 #endif
10422 	}
10423 
10424 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
10425 
10426 	if (t4_nrxq < 1) {
10427 #ifdef RSS
10428 		t4_nrxq = rss_getnumbuckets();
10429 #else
10430 		calculate_nqueues(&t4_nrxq, nc, NRXQ);
10431 #endif
10432 	}
10433 
10434 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
10435 
10436 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
10437 	calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
10438 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
10439 #endif
10440 #ifdef TCP_OFFLOAD
10441 	calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
10442 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
10443 
10444 	if (t4_toecaps_allowed == -1)
10445 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
10446 
10447 	if (t4_rdmacaps_allowed == -1) {
10448 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
10449 		    FW_CAPS_CONFIG_RDMA_RDMAC;
10450 	}
10451 
10452 	if (t4_iscsicaps_allowed == -1) {
10453 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
10454 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
10455 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
10456 	}
10457 
10458 	if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
10459 		t4_tmr_idx_ofld = TMR_IDX_OFLD;
10460 
10461 	if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
10462 		t4_pktc_idx_ofld = PKTC_IDX_OFLD;
10463 #else
10464 	if (t4_toecaps_allowed == -1)
10465 		t4_toecaps_allowed = 0;
10466 
10467 	if (t4_rdmacaps_allowed == -1)
10468 		t4_rdmacaps_allowed = 0;
10469 
10470 	if (t4_iscsicaps_allowed == -1)
10471 		t4_iscsicaps_allowed = 0;
10472 #endif
10473 
10474 #ifdef DEV_NETMAP
10475 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
10476 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
10477 #endif
10478 
10479 	if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
10480 		t4_tmr_idx = TMR_IDX;
10481 
10482 	if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
10483 		t4_pktc_idx = PKTC_IDX;
10484 
10485 	if (t4_qsize_txq < 128)
10486 		t4_qsize_txq = 128;
10487 
10488 	if (t4_qsize_rxq < 128)
10489 		t4_qsize_rxq = 128;
10490 	while (t4_qsize_rxq & 7)
10491 		t4_qsize_rxq++;
10492 
10493 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
10494 
10495 	/*
10496 	 * Number of VIs to create per-port.  The first VI is the "main" regular
10497 	 * VI for the port.  The rest are additional virtual interfaces on the
10498 	 * same physical port.  Note that the main VI does not have native
10499 	 * netmap support but the extra VIs do.
10500 	 *
10501 	 * Limit the number of VIs per port to the number of available
10502 	 * MAC addresses per port.
10503 	 */
10504 	if (t4_num_vis < 1)
10505 		t4_num_vis = 1;
10506 	if (t4_num_vis > nitems(vi_mac_funcs)) {
10507 		t4_num_vis = nitems(vi_mac_funcs);
10508 		printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
10509 	}
10510 
10511 	if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
10512 		pcie_relaxed_ordering = 1;
10513 #if defined(__i386__) || defined(__amd64__)
10514 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
10515 			pcie_relaxed_ordering = 0;
10516 #endif
10517 	}
10518 }
10519 
10520 #ifdef DDB
10521 static void
10522 t4_dump_tcb(struct adapter *sc, int tid)
10523 {
10524 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
10525 
10526 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
10527 	save = t4_read_reg(sc, reg);
10528 	base = sc->memwin[2].mw_base;
10529 
10530 	/* Dump TCB for the tid */
10531 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
10532 	tcb_addr += tid * TCB_SIZE;
10533 
10534 	if (is_t4(sc)) {
10535 		pf = 0;
10536 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
10537 	} else {
10538 		pf = V_PFNUM(sc->pf);
10539 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
10540 	}
10541 	t4_write_reg(sc, reg, win_pos | pf);
10542 	t4_read_reg(sc, reg);
10543 
10544 	off = tcb_addr - win_pos;
10545 	for (i = 0; i < 4; i++) {
10546 		uint32_t buf[8];
10547 		for (j = 0; j < 8; j++, off += 4)
10548 			buf[j] = htonl(t4_read_reg(sc, base + off));
10549 
10550 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
10551 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
10552 		    buf[7]);
10553 	}
10554 
10555 	t4_write_reg(sc, reg, save);
10556 	t4_read_reg(sc, reg);
10557 }
10558 
10559 static void
10560 t4_dump_devlog(struct adapter *sc)
10561 {
10562 	struct devlog_params *dparams = &sc->params.devlog;
10563 	struct fw_devlog_e e;
10564 	int i, first, j, m, nentries, rc;
10565 	uint64_t ftstamp = UINT64_MAX;
10566 
10567 	if (dparams->start == 0) {
10568 		db_printf("devlog params not valid\n");
10569 		return;
10570 	}
10571 
10572 	nentries = dparams->size / sizeof(struct fw_devlog_e);
10573 	m = fwmtype_to_hwmtype(dparams->memtype);
10574 
10575 	/* Find the first entry. */
10576 	first = -1;
10577 	for (i = 0; i < nentries && !db_pager_quit; i++) {
10578 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10579 		    sizeof(e), (void *)&e);
10580 		if (rc != 0)
10581 			break;
10582 
10583 		if (e.timestamp == 0)
10584 			break;
10585 
10586 		e.timestamp = be64toh(e.timestamp);
10587 		if (e.timestamp < ftstamp) {
10588 			ftstamp = e.timestamp;
10589 			first = i;
10590 		}
10591 	}
10592 
10593 	if (first == -1)
10594 		return;
10595 
10596 	i = first;
10597 	do {
10598 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10599 		    sizeof(e), (void *)&e);
10600 		if (rc != 0)
10601 			return;
10602 
10603 		if (e.timestamp == 0)
10604 			return;
10605 
10606 		e.timestamp = be64toh(e.timestamp);
10607 		e.seqno = be32toh(e.seqno);
10608 		for (j = 0; j < 8; j++)
10609 			e.params[j] = be32toh(e.params[j]);
10610 
10611 		db_printf("%10d  %15ju  %8s  %8s  ",
10612 		    e.seqno, e.timestamp,
10613 		    (e.level < nitems(devlog_level_strings) ?
10614 			devlog_level_strings[e.level] : "UNKNOWN"),
10615 		    (e.facility < nitems(devlog_facility_strings) ?
10616 			devlog_facility_strings[e.facility] : "UNKNOWN"));
10617 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
10618 		    e.params[3], e.params[4], e.params[5], e.params[6],
10619 		    e.params[7]);
10620 
10621 		if (++i == nentries)
10622 			i = 0;
10623 	} while (i != first && !db_pager_quit);
10624 }
10625 
10626 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
10627 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
10628 
10629 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
10630 {
10631 	device_t dev;
10632 	int t;
10633 	bool valid;
10634 
10635 	valid = false;
10636 	t = db_read_token();
10637 	if (t == tIDENT) {
10638 		dev = device_lookup_by_name(db_tok_string);
10639 		valid = true;
10640 	}
10641 	db_skip_to_eol();
10642 	if (!valid) {
10643 		db_printf("usage: show t4 devlog <nexus>\n");
10644 		return;
10645 	}
10646 
10647 	if (dev == NULL) {
10648 		db_printf("device not found\n");
10649 		return;
10650 	}
10651 
10652 	t4_dump_devlog(device_get_softc(dev));
10653 }
10654 
10655 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
10656 {
10657 	device_t dev;
10658 	int radix, tid, t;
10659 	bool valid;
10660 
10661 	valid = false;
10662 	radix = db_radix;
10663 	db_radix = 10;
10664 	t = db_read_token();
10665 	if (t == tIDENT) {
10666 		dev = device_lookup_by_name(db_tok_string);
10667 		t = db_read_token();
10668 		if (t == tNUMBER) {
10669 			tid = db_tok_number;
10670 			valid = true;
10671 		}
10672 	}
10673 	db_radix = radix;
10674 	db_skip_to_eol();
10675 	if (!valid) {
10676 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
10677 		return;
10678 	}
10679 
10680 	if (dev == NULL) {
10681 		db_printf("device not found\n");
10682 		return;
10683 	}
10684 	if (tid < 0) {
10685 		db_printf("invalid tid\n");
10686 		return;
10687 	}
10688 
10689 	t4_dump_tcb(device_get_softc(dev), tid);
10690 }
10691 #endif
10692 
10693 /*
10694  * Borrowed from cesa_prep_aes_key().
10695  *
10696  * NB: The crypto engine wants the words in the decryption key in reverse
10697  * order.
10698  */
10699 void
10700 t4_aes_getdeckey(void *dec_key, const void *enc_key, unsigned int kbits)
10701 {
10702 	uint32_t ek[4 * (RIJNDAEL_MAXNR + 1)];
10703 	uint32_t *dkey;
10704 	int i;
10705 
10706 	rijndaelKeySetupEnc(ek, enc_key, kbits);
10707 	dkey = dec_key;
10708 	dkey += (kbits / 8) / 4;
10709 
10710 	switch (kbits) {
10711 	case 128:
10712 		for (i = 0; i < 4; i++)
10713 			*--dkey = htobe32(ek[4 * 10 + i]);
10714 		break;
10715 	case 192:
10716 		for (i = 0; i < 2; i++)
10717 			*--dkey = htobe32(ek[4 * 11 + 2 + i]);
10718 		for (i = 0; i < 4; i++)
10719 			*--dkey = htobe32(ek[4 * 12 + i]);
10720 		break;
10721 	case 256:
10722 		for (i = 0; i < 4; i++)
10723 			*--dkey = htobe32(ek[4 * 13 + i]);
10724 		for (i = 0; i < 4; i++)
10725 			*--dkey = htobe32(ek[4 * 14 + i]);
10726 		break;
10727 	}
10728 	MPASS(dkey == dec_key);
10729 }
10730 
10731 static struct sx mlu;	/* mod load unload */
10732 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
10733 
10734 static int
10735 mod_event(module_t mod, int cmd, void *arg)
10736 {
10737 	int rc = 0;
10738 	static int loaded = 0;
10739 
10740 	switch (cmd) {
10741 	case MOD_LOAD:
10742 		sx_xlock(&mlu);
10743 		if (loaded++ == 0) {
10744 			t4_sge_modload();
10745 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10746 			    t4_filter_rpl, CPL_COOKIE_FILTER);
10747 			t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
10748 			    do_l2t_write_rpl, CPL_COOKIE_FILTER);
10749 			t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL,
10750 			    t4_hashfilter_ao_rpl, CPL_COOKIE_HASHFILTER);
10751 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10752 			    t4_hashfilter_tcb_rpl, CPL_COOKIE_HASHFILTER);
10753 			t4_register_shared_cpl_handler(CPL_ABORT_RPL_RSS,
10754 			    t4_del_hashfilter_rpl, CPL_COOKIE_HASHFILTER);
10755 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
10756 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
10757 			t4_register_cpl_handler(CPL_SMT_WRITE_RPL,
10758 			    do_smt_write_rpl);
10759 			sx_init(&t4_list_lock, "T4/T5 adapters");
10760 			SLIST_INIT(&t4_list);
10761 			callout_init(&fatal_callout, 1);
10762 #ifdef TCP_OFFLOAD
10763 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
10764 			SLIST_INIT(&t4_uld_list);
10765 #endif
10766 #ifdef INET6
10767 			t4_clip_modload();
10768 #endif
10769 			t4_tracer_modload();
10770 			tweak_tunables();
10771 		}
10772 		sx_xunlock(&mlu);
10773 		break;
10774 
10775 	case MOD_UNLOAD:
10776 		sx_xlock(&mlu);
10777 		if (--loaded == 0) {
10778 			int tries;
10779 
10780 			sx_slock(&t4_list_lock);
10781 			if (!SLIST_EMPTY(&t4_list)) {
10782 				rc = EBUSY;
10783 				sx_sunlock(&t4_list_lock);
10784 				goto done_unload;
10785 			}
10786 #ifdef TCP_OFFLOAD
10787 			sx_slock(&t4_uld_list_lock);
10788 			if (!SLIST_EMPTY(&t4_uld_list)) {
10789 				rc = EBUSY;
10790 				sx_sunlock(&t4_uld_list_lock);
10791 				sx_sunlock(&t4_list_lock);
10792 				goto done_unload;
10793 			}
10794 #endif
10795 			tries = 0;
10796 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
10797 				uprintf("%ju clusters with custom free routine "
10798 				    "still is use.\n", t4_sge_extfree_refs());
10799 				pause("t4unload", 2 * hz);
10800 			}
10801 #ifdef TCP_OFFLOAD
10802 			sx_sunlock(&t4_uld_list_lock);
10803 #endif
10804 			sx_sunlock(&t4_list_lock);
10805 
10806 			if (t4_sge_extfree_refs() == 0) {
10807 				t4_tracer_modunload();
10808 #ifdef INET6
10809 				t4_clip_modunload();
10810 #endif
10811 #ifdef TCP_OFFLOAD
10812 				sx_destroy(&t4_uld_list_lock);
10813 #endif
10814 				sx_destroy(&t4_list_lock);
10815 				t4_sge_modunload();
10816 				loaded = 0;
10817 			} else {
10818 				rc = EBUSY;
10819 				loaded++;	/* undo earlier decrement */
10820 			}
10821 		}
10822 done_unload:
10823 		sx_xunlock(&mlu);
10824 		break;
10825 	}
10826 
10827 	return (rc);
10828 }
10829 
10830 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10831 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10832 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10833 
10834 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10835 MODULE_VERSION(t4nex, 1);
10836 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10837 #ifdef DEV_NETMAP
10838 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10839 #endif /* DEV_NETMAP */
10840 
10841 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10842 MODULE_VERSION(t5nex, 1);
10843 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10844 #ifdef DEV_NETMAP
10845 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10846 #endif /* DEV_NETMAP */
10847 
10848 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10849 MODULE_VERSION(t6nex, 1);
10850 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10851 #ifdef DEV_NETMAP
10852 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10853 #endif /* DEV_NETMAP */
10854 
10855 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10856 MODULE_VERSION(cxgbe, 1);
10857 
10858 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10859 MODULE_VERSION(cxl, 1);
10860 
10861 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10862 MODULE_VERSION(cc, 1);
10863 
10864 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10865 MODULE_VERSION(vcxgbe, 1);
10866 
10867 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10868 MODULE_VERSION(vcxl, 1);
10869 
10870 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10871 MODULE_VERSION(vcc, 1);
10872