xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision e9e2022e744c124adbe0372e0822c32d23dac8db)
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 #ifdef TCP_OFFLOAD
698 static int toe_capability(struct vi_info *, int);
699 #endif
700 static int mod_event(module_t, int, void *);
701 static int notify_siblings(device_t, int);
702 
703 struct {
704 	uint16_t device;
705 	char *desc;
706 } t4_pciids[] = {
707 	{0xa000, "Chelsio Terminator 4 FPGA"},
708 	{0x4400, "Chelsio T440-dbg"},
709 	{0x4401, "Chelsio T420-CR"},
710 	{0x4402, "Chelsio T422-CR"},
711 	{0x4403, "Chelsio T440-CR"},
712 	{0x4404, "Chelsio T420-BCH"},
713 	{0x4405, "Chelsio T440-BCH"},
714 	{0x4406, "Chelsio T440-CH"},
715 	{0x4407, "Chelsio T420-SO"},
716 	{0x4408, "Chelsio T420-CX"},
717 	{0x4409, "Chelsio T420-BT"},
718 	{0x440a, "Chelsio T404-BT"},
719 	{0x440e, "Chelsio T440-LP-CR"},
720 }, t5_pciids[] = {
721 	{0xb000, "Chelsio Terminator 5 FPGA"},
722 	{0x5400, "Chelsio T580-dbg"},
723 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
724 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
725 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
726 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
727 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
728 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
729 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
730 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
731 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
732 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
733 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
734 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
735 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
736 	{0x5418,  "Chelsio T540-BT"},		/* 4 x 10GBaseT */
737 	{0x5419,  "Chelsio T540-LP-BT"},	/* 4 x 10GBaseT */
738 	{0x541a,  "Chelsio T540-SO-BT"},	/* 4 x 10GBaseT, nomem */
739 	{0x541b,  "Chelsio T540-SO-CR"},	/* 4 x 10G, nomem */
740 
741 	/* Custom */
742 	{0x5483, "Custom T540-CR"},
743 	{0x5484, "Custom T540-BT"},
744 }, t6_pciids[] = {
745 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
746 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
747 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
748 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
749 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
750 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
751 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
752 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
753 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
754 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
755 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
756 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
757 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
758 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
759 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
760 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
761 
762 	/* Custom */
763 	{0x6480, "Custom T6225-CR"},
764 	{0x6481, "Custom T62100-CR"},
765 	{0x6482, "Custom T6225-CR"},
766 	{0x6483, "Custom T62100-CR"},
767 	{0x6484, "Custom T64100-CR"},
768 	{0x6485, "Custom T6240-SO"},
769 	{0x6486, "Custom T6225-SO-CR"},
770 	{0x6487, "Custom T6225-CR"},
771 };
772 
773 #ifdef TCP_OFFLOAD
774 /*
775  * service_iq_fl() has an iq and needs the fl.  Offset of fl from the iq should
776  * be exactly the same for both rxq and ofld_rxq.
777  */
778 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
779 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
780 #endif
781 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
782 
783 static int
784 t4_probe(device_t dev)
785 {
786 	int i;
787 	uint16_t v = pci_get_vendor(dev);
788 	uint16_t d = pci_get_device(dev);
789 	uint8_t f = pci_get_function(dev);
790 
791 	if (v != PCI_VENDOR_ID_CHELSIO)
792 		return (ENXIO);
793 
794 	/* Attach only to PF0 of the FPGA */
795 	if (d == 0xa000 && f != 0)
796 		return (ENXIO);
797 
798 	for (i = 0; i < nitems(t4_pciids); i++) {
799 		if (d == t4_pciids[i].device) {
800 			device_set_desc(dev, t4_pciids[i].desc);
801 			return (BUS_PROBE_DEFAULT);
802 		}
803 	}
804 
805 	return (ENXIO);
806 }
807 
808 static int
809 t5_probe(device_t dev)
810 {
811 	int i;
812 	uint16_t v = pci_get_vendor(dev);
813 	uint16_t d = pci_get_device(dev);
814 	uint8_t f = pci_get_function(dev);
815 
816 	if (v != PCI_VENDOR_ID_CHELSIO)
817 		return (ENXIO);
818 
819 	/* Attach only to PF0 of the FPGA */
820 	if (d == 0xb000 && f != 0)
821 		return (ENXIO);
822 
823 	for (i = 0; i < nitems(t5_pciids); i++) {
824 		if (d == t5_pciids[i].device) {
825 			device_set_desc(dev, t5_pciids[i].desc);
826 			return (BUS_PROBE_DEFAULT);
827 		}
828 	}
829 
830 	return (ENXIO);
831 }
832 
833 static int
834 t6_probe(device_t dev)
835 {
836 	int i;
837 	uint16_t v = pci_get_vendor(dev);
838 	uint16_t d = pci_get_device(dev);
839 
840 	if (v != PCI_VENDOR_ID_CHELSIO)
841 		return (ENXIO);
842 
843 	for (i = 0; i < nitems(t6_pciids); i++) {
844 		if (d == t6_pciids[i].device) {
845 			device_set_desc(dev, t6_pciids[i].desc);
846 			return (BUS_PROBE_DEFAULT);
847 		}
848 	}
849 
850 	return (ENXIO);
851 }
852 
853 static void
854 t5_attribute_workaround(device_t dev)
855 {
856 	device_t root_port;
857 	uint32_t v;
858 
859 	/*
860 	 * The T5 chips do not properly echo the No Snoop and Relaxed
861 	 * Ordering attributes when replying to a TLP from a Root
862 	 * Port.  As a workaround, find the parent Root Port and
863 	 * disable No Snoop and Relaxed Ordering.  Note that this
864 	 * affects all devices under this root port.
865 	 */
866 	root_port = pci_find_pcie_root_port(dev);
867 	if (root_port == NULL) {
868 		device_printf(dev, "Unable to find parent root port\n");
869 		return;
870 	}
871 
872 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
873 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
874 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
875 	    0)
876 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
877 		    device_get_nameunit(root_port));
878 }
879 
880 static const struct devnames devnames[] = {
881 	{
882 		.nexus_name = "t4nex",
883 		.ifnet_name = "cxgbe",
884 		.vi_ifnet_name = "vcxgbe",
885 		.pf03_drv_name = "t4iov",
886 		.vf_nexus_name = "t4vf",
887 		.vf_ifnet_name = "cxgbev"
888 	}, {
889 		.nexus_name = "t5nex",
890 		.ifnet_name = "cxl",
891 		.vi_ifnet_name = "vcxl",
892 		.pf03_drv_name = "t5iov",
893 		.vf_nexus_name = "t5vf",
894 		.vf_ifnet_name = "cxlv"
895 	}, {
896 		.nexus_name = "t6nex",
897 		.ifnet_name = "cc",
898 		.vi_ifnet_name = "vcc",
899 		.pf03_drv_name = "t6iov",
900 		.vf_nexus_name = "t6vf",
901 		.vf_ifnet_name = "ccv"
902 	}
903 };
904 
905 void
906 t4_init_devnames(struct adapter *sc)
907 {
908 	int id;
909 
910 	id = chip_id(sc);
911 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
912 		sc->names = &devnames[id - CHELSIO_T4];
913 	else {
914 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
915 		sc->names = NULL;
916 	}
917 }
918 
919 static int
920 t4_ifnet_unit(struct adapter *sc, struct port_info *pi)
921 {
922 	const char *parent, *name;
923 	long value;
924 	int line, unit;
925 
926 	line = 0;
927 	parent = device_get_nameunit(sc->dev);
928 	name = sc->names->ifnet_name;
929 	while (resource_find_dev(&line, name, &unit, "at", parent) == 0) {
930 		if (resource_long_value(name, unit, "port", &value) == 0 &&
931 		    value == pi->port_id)
932 			return (unit);
933 	}
934 	return (-1);
935 }
936 
937 static int
938 t4_attach(device_t dev)
939 {
940 	struct adapter *sc;
941 	int rc = 0, i, j, rqidx, tqidx, nports;
942 	struct make_dev_args mda;
943 	struct intrs_and_queues iaq;
944 	struct sge *s;
945 	uint32_t *buf;
946 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
947 	int ofld_tqidx;
948 #endif
949 #ifdef TCP_OFFLOAD
950 	int ofld_rqidx;
951 #endif
952 #ifdef DEV_NETMAP
953 	int nm_rqidx, nm_tqidx;
954 #endif
955 	int num_vis;
956 
957 	sc = device_get_softc(dev);
958 	sc->dev = dev;
959 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
960 
961 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
962 		t5_attribute_workaround(dev);
963 	pci_enable_busmaster(dev);
964 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
965 		uint32_t v;
966 
967 		pci_set_max_read_req(dev, 4096);
968 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
969 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
970 		if (pcie_relaxed_ordering == 0 &&
971 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
972 			v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
973 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
974 		} else if (pcie_relaxed_ordering == 1 &&
975 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
976 			v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
977 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
978 		}
979 	}
980 
981 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
982 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
983 	sc->traceq = -1;
984 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
985 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
986 	    device_get_nameunit(dev));
987 
988 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
989 	    device_get_nameunit(dev));
990 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
991 	t4_add_adapter(sc);
992 
993 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
994 	TAILQ_INIT(&sc->sfl);
995 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
996 
997 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
998 
999 	sc->policy = NULL;
1000 	rw_init(&sc->policy_lock, "connection offload policy");
1001 
1002 	rc = t4_map_bars_0_and_4(sc);
1003 	if (rc != 0)
1004 		goto done; /* error message displayed already */
1005 
1006 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
1007 
1008 	/* Prepare the adapter for operation. */
1009 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
1010 	rc = -t4_prep_adapter(sc, buf);
1011 	free(buf, M_CXGBE);
1012 	if (rc != 0) {
1013 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
1014 		goto done;
1015 	}
1016 
1017 	/*
1018 	 * This is the real PF# to which we're attaching.  Works from within PCI
1019 	 * passthrough environments too, where pci_get_function() could return a
1020 	 * different PF# depending on the passthrough configuration.  We need to
1021 	 * use the real PF# in all our communication with the firmware.
1022 	 */
1023 	j = t4_read_reg(sc, A_PL_WHOAMI);
1024 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
1025 	sc->mbox = sc->pf;
1026 
1027 	t4_init_devnames(sc);
1028 	if (sc->names == NULL) {
1029 		rc = ENOTSUP;
1030 		goto done; /* error message displayed already */
1031 	}
1032 
1033 	/*
1034 	 * Do this really early, with the memory windows set up even before the
1035 	 * character device.  The userland tool's register i/o and mem read
1036 	 * will work even in "recovery mode".
1037 	 */
1038 	setup_memwin(sc);
1039 	if (t4_init_devlog_params(sc, 0) == 0)
1040 		fixup_devlog_params(sc);
1041 	make_dev_args_init(&mda);
1042 	mda.mda_devsw = &t4_cdevsw;
1043 	mda.mda_uid = UID_ROOT;
1044 	mda.mda_gid = GID_WHEEL;
1045 	mda.mda_mode = 0600;
1046 	mda.mda_si_drv1 = sc;
1047 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
1048 	if (rc != 0)
1049 		device_printf(dev, "failed to create nexus char device: %d.\n",
1050 		    rc);
1051 
1052 	/* Go no further if recovery mode has been requested. */
1053 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
1054 		device_printf(dev, "recovery mode.\n");
1055 		goto done;
1056 	}
1057 
1058 #if defined(__i386__)
1059 	if ((cpu_feature & CPUID_CX8) == 0) {
1060 		device_printf(dev, "64 bit atomics not available.\n");
1061 		rc = ENOTSUP;
1062 		goto done;
1063 	}
1064 #endif
1065 
1066 	/* Contact the firmware and try to become the master driver. */
1067 	rc = contact_firmware(sc);
1068 	if (rc != 0)
1069 		goto done; /* error message displayed already */
1070 	MPASS(sc->flags & FW_OK);
1071 
1072 	rc = get_params__pre_init(sc);
1073 	if (rc != 0)
1074 		goto done; /* error message displayed already */
1075 
1076 	if (sc->flags & MASTER_PF) {
1077 		rc = partition_resources(sc);
1078 		if (rc != 0)
1079 			goto done; /* error message displayed already */
1080 		t4_intr_clear(sc);
1081 	}
1082 
1083 	rc = get_params__post_init(sc);
1084 	if (rc != 0)
1085 		goto done; /* error message displayed already */
1086 
1087 	rc = set_params__post_init(sc);
1088 	if (rc != 0)
1089 		goto done; /* error message displayed already */
1090 
1091 	rc = t4_map_bar_2(sc);
1092 	if (rc != 0)
1093 		goto done; /* error message displayed already */
1094 
1095 	rc = t4_create_dma_tag(sc);
1096 	if (rc != 0)
1097 		goto done; /* error message displayed already */
1098 
1099 	/*
1100 	 * First pass over all the ports - allocate VIs and initialize some
1101 	 * basic parameters like mac address, port type, etc.
1102 	 */
1103 	for_each_port(sc, i) {
1104 		struct port_info *pi;
1105 
1106 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
1107 		sc->port[i] = pi;
1108 
1109 		/* These must be set before t4_port_init */
1110 		pi->adapter = sc;
1111 		pi->port_id = i;
1112 		/*
1113 		 * XXX: vi[0] is special so we can't delay this allocation until
1114 		 * pi->nvi's final value is known.
1115 		 */
1116 		pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
1117 		    M_ZERO | M_WAITOK);
1118 
1119 		/*
1120 		 * Allocate the "main" VI and initialize parameters
1121 		 * like mac addr.
1122 		 */
1123 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1124 		if (rc != 0) {
1125 			device_printf(dev, "unable to initialize port %d: %d\n",
1126 			    i, rc);
1127 			free(pi->vi, M_CXGBE);
1128 			free(pi, M_CXGBE);
1129 			sc->port[i] = NULL;
1130 			goto done;
1131 		}
1132 
1133 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1134 		    device_get_nameunit(dev), i);
1135 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1136 		sc->chan_map[pi->tx_chan] = i;
1137 
1138 		/* All VIs on this port share this media. */
1139 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1140 		    cxgbe_media_status);
1141 
1142 		PORT_LOCK(pi);
1143 		init_link_config(pi);
1144 		fixup_link_config(pi);
1145 		build_medialist(pi);
1146 		if (fixed_ifmedia(pi))
1147 			pi->flags |= FIXED_IFMEDIA;
1148 		PORT_UNLOCK(pi);
1149 
1150 		pi->dev = device_add_child(dev, sc->names->ifnet_name,
1151 		    t4_ifnet_unit(sc, pi));
1152 		if (pi->dev == NULL) {
1153 			device_printf(dev,
1154 			    "failed to add device for port %d.\n", i);
1155 			rc = ENXIO;
1156 			goto done;
1157 		}
1158 		pi->vi[0].dev = pi->dev;
1159 		device_set_softc(pi->dev, pi);
1160 	}
1161 
1162 	/*
1163 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1164 	 */
1165 	nports = sc->params.nports;
1166 	rc = cfg_itype_and_nqueues(sc, &iaq);
1167 	if (rc != 0)
1168 		goto done; /* error message displayed already */
1169 
1170 	num_vis = iaq.num_vis;
1171 	sc->intr_type = iaq.intr_type;
1172 	sc->intr_count = iaq.nirq;
1173 
1174 	s = &sc->sge;
1175 	s->nrxq = nports * iaq.nrxq;
1176 	s->ntxq = nports * iaq.ntxq;
1177 	if (num_vis > 1) {
1178 		s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1179 		s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1180 	}
1181 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1182 	s->neq += nports;		/* ctrl queues: 1 per port */
1183 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1184 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1185 	if (is_offload(sc) || is_ethoffload(sc)) {
1186 		s->nofldtxq = nports * iaq.nofldtxq;
1187 		if (num_vis > 1)
1188 			s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1189 		s->neq += s->nofldtxq;
1190 
1191 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1192 		    M_CXGBE, M_ZERO | M_WAITOK);
1193 	}
1194 #endif
1195 #ifdef TCP_OFFLOAD
1196 	if (is_offload(sc)) {
1197 		s->nofldrxq = nports * iaq.nofldrxq;
1198 		if (num_vis > 1)
1199 			s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1200 		s->neq += s->nofldrxq;	/* free list */
1201 		s->niq += s->nofldrxq;
1202 
1203 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1204 		    M_CXGBE, M_ZERO | M_WAITOK);
1205 	}
1206 #endif
1207 #ifdef DEV_NETMAP
1208 	if (num_vis > 1) {
1209 		s->nnmrxq = nports * (num_vis - 1) * iaq.nnmrxq_vi;
1210 		s->nnmtxq = nports * (num_vis - 1) * iaq.nnmtxq_vi;
1211 	}
1212 	s->neq += s->nnmtxq + s->nnmrxq;
1213 	s->niq += s->nnmrxq;
1214 
1215 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1216 	    M_CXGBE, M_ZERO | M_WAITOK);
1217 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1218 	    M_CXGBE, M_ZERO | M_WAITOK);
1219 #endif
1220 
1221 	s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1222 	    M_ZERO | M_WAITOK);
1223 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1224 	    M_ZERO | M_WAITOK);
1225 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1226 	    M_ZERO | M_WAITOK);
1227 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1228 	    M_ZERO | M_WAITOK);
1229 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1230 	    M_ZERO | M_WAITOK);
1231 
1232 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1233 	    M_ZERO | M_WAITOK);
1234 
1235 	t4_init_l2t(sc, M_WAITOK);
1236 	t4_init_smt(sc, M_WAITOK);
1237 	t4_init_tx_sched(sc);
1238 #ifdef RATELIMIT
1239 	t4_init_etid_table(sc);
1240 #endif
1241 #ifdef INET6
1242 	t4_init_clip_table(sc);
1243 #endif
1244 	if (sc->vres.key.size != 0)
1245 		sc->key_map = vmem_create("T4TLS key map", sc->vres.key.start,
1246 		    sc->vres.key.size, 32, 0, M_FIRSTFIT | M_WAITOK);
1247 
1248 	/*
1249 	 * Second pass over the ports.  This time we know the number of rx and
1250 	 * tx queues that each port should get.
1251 	 */
1252 	rqidx = tqidx = 0;
1253 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1254 	ofld_tqidx = 0;
1255 #endif
1256 #ifdef TCP_OFFLOAD
1257 	ofld_rqidx = 0;
1258 #endif
1259 #ifdef DEV_NETMAP
1260 	nm_rqidx = nm_tqidx = 0;
1261 #endif
1262 	for_each_port(sc, i) {
1263 		struct port_info *pi = sc->port[i];
1264 		struct vi_info *vi;
1265 
1266 		if (pi == NULL)
1267 			continue;
1268 
1269 		pi->nvi = num_vis;
1270 		for_each_vi(pi, j, vi) {
1271 			vi->pi = pi;
1272 			vi->qsize_rxq = t4_qsize_rxq;
1273 			vi->qsize_txq = t4_qsize_txq;
1274 
1275 			vi->first_rxq = rqidx;
1276 			vi->first_txq = tqidx;
1277 			vi->tmr_idx = t4_tmr_idx;
1278 			vi->pktc_idx = t4_pktc_idx;
1279 			vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1280 			vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1281 
1282 			rqidx += vi->nrxq;
1283 			tqidx += vi->ntxq;
1284 
1285 			if (j == 0 && vi->ntxq > 1)
1286 				vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1287 			else
1288 				vi->rsrv_noflowq = 0;
1289 
1290 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1291 			vi->first_ofld_txq = ofld_tqidx;
1292 			vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1293 			ofld_tqidx += vi->nofldtxq;
1294 #endif
1295 #ifdef TCP_OFFLOAD
1296 			vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1297 			vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1298 			vi->first_ofld_rxq = ofld_rqidx;
1299 			vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1300 
1301 			ofld_rqidx += vi->nofldrxq;
1302 #endif
1303 #ifdef DEV_NETMAP
1304 			if (j > 0) {
1305 				vi->first_nm_rxq = nm_rqidx;
1306 				vi->first_nm_txq = nm_tqidx;
1307 				vi->nnmrxq = iaq.nnmrxq_vi;
1308 				vi->nnmtxq = iaq.nnmtxq_vi;
1309 				nm_rqidx += vi->nnmrxq;
1310 				nm_tqidx += vi->nnmtxq;
1311 			}
1312 #endif
1313 		}
1314 	}
1315 
1316 	rc = t4_setup_intr_handlers(sc);
1317 	if (rc != 0) {
1318 		device_printf(dev,
1319 		    "failed to setup interrupt handlers: %d\n", rc);
1320 		goto done;
1321 	}
1322 
1323 	rc = bus_generic_probe(dev);
1324 	if (rc != 0) {
1325 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1326 		goto done;
1327 	}
1328 
1329 	/*
1330 	 * Ensure thread-safe mailbox access (in debug builds).
1331 	 *
1332 	 * So far this was the only thread accessing the mailbox but various
1333 	 * ifnets and sysctls are about to be created and their handlers/ioctls
1334 	 * will access the mailbox from different threads.
1335 	 */
1336 	sc->flags |= CHK_MBOX_ACCESS;
1337 
1338 	rc = bus_generic_attach(dev);
1339 	if (rc != 0) {
1340 		device_printf(dev,
1341 		    "failed to attach all child ports: %d\n", rc);
1342 		goto done;
1343 	}
1344 
1345 	device_printf(dev,
1346 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1347 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1348 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1349 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1350 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1351 
1352 	t4_set_desc(sc);
1353 
1354 	notify_siblings(dev, 0);
1355 
1356 done:
1357 	if (rc != 0 && sc->cdev) {
1358 		/* cdev was created and so cxgbetool works; recover that way. */
1359 		device_printf(dev,
1360 		    "error during attach, adapter is now in recovery mode.\n");
1361 		rc = 0;
1362 	}
1363 
1364 	if (rc != 0)
1365 		t4_detach_common(dev);
1366 	else
1367 		t4_sysctls(sc);
1368 
1369 	return (rc);
1370 }
1371 
1372 static int
1373 t4_child_location_str(device_t bus, device_t dev, char *buf, size_t buflen)
1374 {
1375 	struct adapter *sc;
1376 	struct port_info *pi;
1377 	int i;
1378 
1379 	sc = device_get_softc(bus);
1380 	buf[0] = '\0';
1381 	for_each_port(sc, i) {
1382 		pi = sc->port[i];
1383 		if (pi != NULL && pi->dev == dev) {
1384 			snprintf(buf, buflen, "port=%d", pi->port_id);
1385 			break;
1386 		}
1387 	}
1388 	return (0);
1389 }
1390 
1391 static int
1392 t4_ready(device_t dev)
1393 {
1394 	struct adapter *sc;
1395 
1396 	sc = device_get_softc(dev);
1397 	if (sc->flags & FW_OK)
1398 		return (0);
1399 	return (ENXIO);
1400 }
1401 
1402 static int
1403 t4_read_port_device(device_t dev, int port, device_t *child)
1404 {
1405 	struct adapter *sc;
1406 	struct port_info *pi;
1407 
1408 	sc = device_get_softc(dev);
1409 	if (port < 0 || port >= MAX_NPORTS)
1410 		return (EINVAL);
1411 	pi = sc->port[port];
1412 	if (pi == NULL || pi->dev == NULL)
1413 		return (ENXIO);
1414 	*child = pi->dev;
1415 	return (0);
1416 }
1417 
1418 static int
1419 notify_siblings(device_t dev, int detaching)
1420 {
1421 	device_t sibling;
1422 	int error, i;
1423 
1424 	error = 0;
1425 	for (i = 0; i < PCI_FUNCMAX; i++) {
1426 		if (i == pci_get_function(dev))
1427 			continue;
1428 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1429 		    pci_get_slot(dev), i);
1430 		if (sibling == NULL || !device_is_attached(sibling))
1431 			continue;
1432 		if (detaching)
1433 			error = T4_DETACH_CHILD(sibling);
1434 		else
1435 			(void)T4_ATTACH_CHILD(sibling);
1436 		if (error)
1437 			break;
1438 	}
1439 	return (error);
1440 }
1441 
1442 /*
1443  * Idempotent
1444  */
1445 static int
1446 t4_detach(device_t dev)
1447 {
1448 	struct adapter *sc;
1449 	int rc;
1450 
1451 	sc = device_get_softc(dev);
1452 
1453 	rc = notify_siblings(dev, 1);
1454 	if (rc) {
1455 		device_printf(dev,
1456 		    "failed to detach sibling devices: %d\n", rc);
1457 		return (rc);
1458 	}
1459 
1460 	return (t4_detach_common(dev));
1461 }
1462 
1463 int
1464 t4_detach_common(device_t dev)
1465 {
1466 	struct adapter *sc;
1467 	struct port_info *pi;
1468 	int i, rc;
1469 
1470 	sc = device_get_softc(dev);
1471 
1472 	if (sc->cdev) {
1473 		destroy_dev(sc->cdev);
1474 		sc->cdev = NULL;
1475 	}
1476 
1477 	sc->flags &= ~CHK_MBOX_ACCESS;
1478 	if (sc->flags & FULL_INIT_DONE) {
1479 		if (!(sc->flags & IS_VF))
1480 			t4_intr_disable(sc);
1481 	}
1482 
1483 	if (device_is_attached(dev)) {
1484 		rc = bus_generic_detach(dev);
1485 		if (rc) {
1486 			device_printf(dev,
1487 			    "failed to detach child devices: %d\n", rc);
1488 			return (rc);
1489 		}
1490 	}
1491 
1492 	for (i = 0; i < sc->intr_count; i++)
1493 		t4_free_irq(sc, &sc->irq[i]);
1494 
1495 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1496 		t4_free_tx_sched(sc);
1497 
1498 	for (i = 0; i < MAX_NPORTS; i++) {
1499 		pi = sc->port[i];
1500 		if (pi) {
1501 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1502 			if (pi->dev)
1503 				device_delete_child(dev, pi->dev);
1504 
1505 			mtx_destroy(&pi->pi_lock);
1506 			free(pi->vi, M_CXGBE);
1507 			free(pi, M_CXGBE);
1508 		}
1509 	}
1510 
1511 	device_delete_children(dev);
1512 
1513 	if (sc->flags & FULL_INIT_DONE)
1514 		adapter_full_uninit(sc);
1515 
1516 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1517 		t4_fw_bye(sc, sc->mbox);
1518 
1519 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1520 		pci_release_msi(dev);
1521 
1522 	if (sc->regs_res)
1523 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1524 		    sc->regs_res);
1525 
1526 	if (sc->udbs_res)
1527 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1528 		    sc->udbs_res);
1529 
1530 	if (sc->msix_res)
1531 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1532 		    sc->msix_res);
1533 
1534 	if (sc->l2t)
1535 		t4_free_l2t(sc->l2t);
1536 	if (sc->smt)
1537 		t4_free_smt(sc->smt);
1538 #ifdef RATELIMIT
1539 	t4_free_etid_table(sc);
1540 #endif
1541 	if (sc->key_map)
1542 		vmem_destroy(sc->key_map);
1543 #ifdef INET6
1544 	t4_destroy_clip_table(sc);
1545 #endif
1546 
1547 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1548 	free(sc->sge.ofld_txq, M_CXGBE);
1549 #endif
1550 #ifdef TCP_OFFLOAD
1551 	free(sc->sge.ofld_rxq, M_CXGBE);
1552 #endif
1553 #ifdef DEV_NETMAP
1554 	free(sc->sge.nm_rxq, M_CXGBE);
1555 	free(sc->sge.nm_txq, M_CXGBE);
1556 #endif
1557 	free(sc->irq, M_CXGBE);
1558 	free(sc->sge.rxq, M_CXGBE);
1559 	free(sc->sge.txq, M_CXGBE);
1560 	free(sc->sge.ctrlq, M_CXGBE);
1561 	free(sc->sge.iqmap, M_CXGBE);
1562 	free(sc->sge.eqmap, M_CXGBE);
1563 	free(sc->tids.ftid_tab, M_CXGBE);
1564 	free(sc->tids.hpftid_tab, M_CXGBE);
1565 	free_hftid_hash(&sc->tids);
1566 	free(sc->tids.atid_tab, M_CXGBE);
1567 	free(sc->tids.tid_tab, M_CXGBE);
1568 	free(sc->tt.tls_rx_ports, M_CXGBE);
1569 	t4_destroy_dma_tag(sc);
1570 	if (mtx_initialized(&sc->sc_lock)) {
1571 		sx_xlock(&t4_list_lock);
1572 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1573 		sx_xunlock(&t4_list_lock);
1574 		mtx_destroy(&sc->sc_lock);
1575 	}
1576 
1577 	callout_drain(&sc->sfl_callout);
1578 	if (mtx_initialized(&sc->tids.ftid_lock)) {
1579 		mtx_destroy(&sc->tids.ftid_lock);
1580 		cv_destroy(&sc->tids.ftid_cv);
1581 	}
1582 	if (mtx_initialized(&sc->tids.atid_lock))
1583 		mtx_destroy(&sc->tids.atid_lock);
1584 	if (mtx_initialized(&sc->sfl_lock))
1585 		mtx_destroy(&sc->sfl_lock);
1586 	if (mtx_initialized(&sc->ifp_lock))
1587 		mtx_destroy(&sc->ifp_lock);
1588 	if (mtx_initialized(&sc->reg_lock))
1589 		mtx_destroy(&sc->reg_lock);
1590 
1591 	if (rw_initialized(&sc->policy_lock)) {
1592 		rw_destroy(&sc->policy_lock);
1593 #ifdef TCP_OFFLOAD
1594 		if (sc->policy != NULL)
1595 			free_offload_policy(sc->policy);
1596 #endif
1597 	}
1598 
1599 	for (i = 0; i < NUM_MEMWIN; i++) {
1600 		struct memwin *mw = &sc->memwin[i];
1601 
1602 		if (rw_initialized(&mw->mw_lock))
1603 			rw_destroy(&mw->mw_lock);
1604 	}
1605 
1606 	bzero(sc, sizeof(*sc));
1607 
1608 	return (0);
1609 }
1610 
1611 static int
1612 cxgbe_probe(device_t dev)
1613 {
1614 	char buf[128];
1615 	struct port_info *pi = device_get_softc(dev);
1616 
1617 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1618 	device_set_desc_copy(dev, buf);
1619 
1620 	return (BUS_PROBE_DEFAULT);
1621 }
1622 
1623 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1624     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1625     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS | \
1626     IFCAP_HWRXTSTMP)
1627 #define T4_CAP_ENABLE (T4_CAP)
1628 
1629 static int
1630 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1631 {
1632 	struct ifnet *ifp;
1633 	struct sbuf *sb;
1634 
1635 	vi->xact_addr_filt = -1;
1636 	callout_init(&vi->tick, 1);
1637 
1638 	/* Allocate an ifnet and set it up */
1639 	ifp = if_alloc(IFT_ETHER);
1640 	if (ifp == NULL) {
1641 		device_printf(dev, "Cannot allocate ifnet\n");
1642 		return (ENOMEM);
1643 	}
1644 	vi->ifp = ifp;
1645 	ifp->if_softc = vi;
1646 
1647 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1648 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1649 
1650 	ifp->if_init = cxgbe_init;
1651 	ifp->if_ioctl = cxgbe_ioctl;
1652 	ifp->if_transmit = cxgbe_transmit;
1653 	ifp->if_qflush = cxgbe_qflush;
1654 	ifp->if_get_counter = cxgbe_get_counter;
1655 #ifdef RATELIMIT
1656 	ifp->if_snd_tag_alloc = cxgbe_snd_tag_alloc;
1657 	ifp->if_snd_tag_modify = cxgbe_snd_tag_modify;
1658 	ifp->if_snd_tag_query = cxgbe_snd_tag_query;
1659 	ifp->if_snd_tag_free = cxgbe_snd_tag_free;
1660 #endif
1661 
1662 	ifp->if_capabilities = T4_CAP;
1663 	ifp->if_capenable = T4_CAP_ENABLE;
1664 #ifdef TCP_OFFLOAD
1665 	if (vi->nofldrxq != 0)
1666 		ifp->if_capabilities |= IFCAP_TOE;
1667 #endif
1668 #ifdef RATELIMIT
1669 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0) {
1670 		ifp->if_capabilities |= IFCAP_TXRTLMT;
1671 		ifp->if_capenable |= IFCAP_TXRTLMT;
1672 	}
1673 #endif
1674 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1675 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1676 
1677 	ifp->if_hw_tsomax = IP_MAXPACKET;
1678 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_TSO;
1679 #ifdef RATELIMIT
1680 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0)
1681 		ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_EO_TSO;
1682 #endif
1683 	ifp->if_hw_tsomaxsegsize = 65536;
1684 
1685 	ether_ifattach(ifp, vi->hw_addr);
1686 #ifdef DEV_NETMAP
1687 	if (vi->nnmrxq != 0)
1688 		cxgbe_nm_attach(vi);
1689 #endif
1690 	sb = sbuf_new_auto();
1691 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1692 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1693 	switch (ifp->if_capabilities & (IFCAP_TOE | IFCAP_TXRTLMT)) {
1694 	case IFCAP_TOE:
1695 		sbuf_printf(sb, "; %d txq (TOE)", vi->nofldtxq);
1696 		break;
1697 	case IFCAP_TOE | IFCAP_TXRTLMT:
1698 		sbuf_printf(sb, "; %d txq (TOE/ETHOFLD)", vi->nofldtxq);
1699 		break;
1700 	case IFCAP_TXRTLMT:
1701 		sbuf_printf(sb, "; %d txq (ETHOFLD)", vi->nofldtxq);
1702 		break;
1703 	}
1704 #endif
1705 #ifdef TCP_OFFLOAD
1706 	if (ifp->if_capabilities & IFCAP_TOE)
1707 		sbuf_printf(sb, ", %d rxq (TOE)", vi->nofldrxq);
1708 #endif
1709 #ifdef DEV_NETMAP
1710 	if (ifp->if_capabilities & IFCAP_NETMAP)
1711 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1712 		    vi->nnmtxq, vi->nnmrxq);
1713 #endif
1714 	sbuf_finish(sb);
1715 	device_printf(dev, "%s\n", sbuf_data(sb));
1716 	sbuf_delete(sb);
1717 
1718 	vi_sysctls(vi);
1719 
1720 	return (0);
1721 }
1722 
1723 static int
1724 cxgbe_attach(device_t dev)
1725 {
1726 	struct port_info *pi = device_get_softc(dev);
1727 	struct adapter *sc = pi->adapter;
1728 	struct vi_info *vi;
1729 	int i, rc;
1730 
1731 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1732 
1733 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1734 	if (rc)
1735 		return (rc);
1736 
1737 	for_each_vi(pi, i, vi) {
1738 		if (i == 0)
1739 			continue;
1740 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1741 		if (vi->dev == NULL) {
1742 			device_printf(dev, "failed to add VI %d\n", i);
1743 			continue;
1744 		}
1745 		device_set_softc(vi->dev, vi);
1746 	}
1747 
1748 	cxgbe_sysctls(pi);
1749 
1750 	bus_generic_attach(dev);
1751 
1752 	return (0);
1753 }
1754 
1755 static void
1756 cxgbe_vi_detach(struct vi_info *vi)
1757 {
1758 	struct ifnet *ifp = vi->ifp;
1759 
1760 	ether_ifdetach(ifp);
1761 
1762 	/* Let detach proceed even if these fail. */
1763 #ifdef DEV_NETMAP
1764 	if (ifp->if_capabilities & IFCAP_NETMAP)
1765 		cxgbe_nm_detach(vi);
1766 #endif
1767 	cxgbe_uninit_synchronized(vi);
1768 	callout_drain(&vi->tick);
1769 	vi_full_uninit(vi);
1770 
1771 	if_free(vi->ifp);
1772 	vi->ifp = NULL;
1773 }
1774 
1775 static int
1776 cxgbe_detach(device_t dev)
1777 {
1778 	struct port_info *pi = device_get_softc(dev);
1779 	struct adapter *sc = pi->adapter;
1780 	int rc;
1781 
1782 	/* Detach the extra VIs first. */
1783 	rc = bus_generic_detach(dev);
1784 	if (rc)
1785 		return (rc);
1786 	device_delete_children(dev);
1787 
1788 	doom_vi(sc, &pi->vi[0]);
1789 
1790 	if (pi->flags & HAS_TRACEQ) {
1791 		sc->traceq = -1;	/* cloner should not create ifnet */
1792 		t4_tracer_port_detach(sc);
1793 	}
1794 
1795 	cxgbe_vi_detach(&pi->vi[0]);
1796 	callout_drain(&pi->tick);
1797 	ifmedia_removeall(&pi->media);
1798 
1799 	end_synchronized_op(sc, 0);
1800 
1801 	return (0);
1802 }
1803 
1804 static void
1805 cxgbe_init(void *arg)
1806 {
1807 	struct vi_info *vi = arg;
1808 	struct adapter *sc = vi->pi->adapter;
1809 
1810 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1811 		return;
1812 	cxgbe_init_synchronized(vi);
1813 	end_synchronized_op(sc, 0);
1814 }
1815 
1816 static int
1817 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1818 {
1819 	int rc = 0, mtu, flags;
1820 	struct vi_info *vi = ifp->if_softc;
1821 	struct port_info *pi = vi->pi;
1822 	struct adapter *sc = pi->adapter;
1823 	struct ifreq *ifr = (struct ifreq *)data;
1824 	uint32_t mask;
1825 
1826 	switch (cmd) {
1827 	case SIOCSIFMTU:
1828 		mtu = ifr->ifr_mtu;
1829 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1830 			return (EINVAL);
1831 
1832 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1833 		if (rc)
1834 			return (rc);
1835 		ifp->if_mtu = mtu;
1836 		if (vi->flags & VI_INIT_DONE) {
1837 			t4_update_fl_bufsize(ifp);
1838 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1839 				rc = update_mac_settings(ifp, XGMAC_MTU);
1840 		}
1841 		end_synchronized_op(sc, 0);
1842 		break;
1843 
1844 	case SIOCSIFFLAGS:
1845 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4flg");
1846 		if (rc)
1847 			return (rc);
1848 
1849 		if (ifp->if_flags & IFF_UP) {
1850 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1851 				flags = vi->if_flags;
1852 				if ((ifp->if_flags ^ flags) &
1853 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1854 					rc = update_mac_settings(ifp,
1855 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1856 				}
1857 			} else {
1858 				rc = cxgbe_init_synchronized(vi);
1859 			}
1860 			vi->if_flags = ifp->if_flags;
1861 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1862 			rc = cxgbe_uninit_synchronized(vi);
1863 		}
1864 		end_synchronized_op(sc, 0);
1865 		break;
1866 
1867 	case SIOCADDMULTI:
1868 	case SIOCDELMULTI:
1869 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4multi");
1870 		if (rc)
1871 			return (rc);
1872 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1873 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1874 		end_synchronized_op(sc, 0);
1875 		break;
1876 
1877 	case SIOCSIFCAP:
1878 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1879 		if (rc)
1880 			return (rc);
1881 
1882 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1883 		if (mask & IFCAP_TXCSUM) {
1884 			ifp->if_capenable ^= IFCAP_TXCSUM;
1885 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1886 
1887 			if (IFCAP_TSO4 & ifp->if_capenable &&
1888 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1889 				ifp->if_capenable &= ~IFCAP_TSO4;
1890 				if_printf(ifp,
1891 				    "tso4 disabled due to -txcsum.\n");
1892 			}
1893 		}
1894 		if (mask & IFCAP_TXCSUM_IPV6) {
1895 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1896 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1897 
1898 			if (IFCAP_TSO6 & ifp->if_capenable &&
1899 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1900 				ifp->if_capenable &= ~IFCAP_TSO6;
1901 				if_printf(ifp,
1902 				    "tso6 disabled due to -txcsum6.\n");
1903 			}
1904 		}
1905 		if (mask & IFCAP_RXCSUM)
1906 			ifp->if_capenable ^= IFCAP_RXCSUM;
1907 		if (mask & IFCAP_RXCSUM_IPV6)
1908 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1909 
1910 		/*
1911 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1912 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1913 		 * sending a TSO request our way, so it's sufficient to toggle
1914 		 * IFCAP_TSOx only.
1915 		 */
1916 		if (mask & IFCAP_TSO4) {
1917 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1918 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1919 				if_printf(ifp, "enable txcsum first.\n");
1920 				rc = EAGAIN;
1921 				goto fail;
1922 			}
1923 			ifp->if_capenable ^= IFCAP_TSO4;
1924 		}
1925 		if (mask & IFCAP_TSO6) {
1926 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1927 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1928 				if_printf(ifp, "enable txcsum6 first.\n");
1929 				rc = EAGAIN;
1930 				goto fail;
1931 			}
1932 			ifp->if_capenable ^= IFCAP_TSO6;
1933 		}
1934 		if (mask & IFCAP_LRO) {
1935 #if defined(INET) || defined(INET6)
1936 			int i;
1937 			struct sge_rxq *rxq;
1938 
1939 			ifp->if_capenable ^= IFCAP_LRO;
1940 			for_each_rxq(vi, i, rxq) {
1941 				if (ifp->if_capenable & IFCAP_LRO)
1942 					rxq->iq.flags |= IQ_LRO_ENABLED;
1943 				else
1944 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1945 			}
1946 #endif
1947 		}
1948 #ifdef TCP_OFFLOAD
1949 		if (mask & IFCAP_TOE) {
1950 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1951 
1952 			rc = toe_capability(vi, enable);
1953 			if (rc != 0)
1954 				goto fail;
1955 
1956 			ifp->if_capenable ^= mask;
1957 		}
1958 #endif
1959 		if (mask & IFCAP_VLAN_HWTAGGING) {
1960 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1961 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1962 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1963 		}
1964 		if (mask & IFCAP_VLAN_MTU) {
1965 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1966 
1967 			/* Need to find out how to disable auto-mtu-inflation */
1968 		}
1969 		if (mask & IFCAP_VLAN_HWTSO)
1970 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1971 		if (mask & IFCAP_VLAN_HWCSUM)
1972 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1973 #ifdef RATELIMIT
1974 		if (mask & IFCAP_TXRTLMT)
1975 			ifp->if_capenable ^= IFCAP_TXRTLMT;
1976 #endif
1977 		if (mask & IFCAP_HWRXTSTMP) {
1978 			int i;
1979 			struct sge_rxq *rxq;
1980 
1981 			ifp->if_capenable ^= IFCAP_HWRXTSTMP;
1982 			for_each_rxq(vi, i, rxq) {
1983 				if (ifp->if_capenable & IFCAP_HWRXTSTMP)
1984 					rxq->iq.flags |= IQ_RX_TIMESTAMP;
1985 				else
1986 					rxq->iq.flags &= ~IQ_RX_TIMESTAMP;
1987 			}
1988 		}
1989 
1990 #ifdef VLAN_CAPABILITIES
1991 		VLAN_CAPABILITIES(ifp);
1992 #endif
1993 fail:
1994 		end_synchronized_op(sc, 0);
1995 		break;
1996 
1997 	case SIOCSIFMEDIA:
1998 	case SIOCGIFMEDIA:
1999 	case SIOCGIFXMEDIA:
2000 		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
2001 		break;
2002 
2003 	case SIOCGI2C: {
2004 		struct ifi2creq i2c;
2005 
2006 		rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
2007 		if (rc != 0)
2008 			break;
2009 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
2010 			rc = EPERM;
2011 			break;
2012 		}
2013 		if (i2c.len > sizeof(i2c.data)) {
2014 			rc = EINVAL;
2015 			break;
2016 		}
2017 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
2018 		if (rc)
2019 			return (rc);
2020 		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
2021 		    i2c.offset, i2c.len, &i2c.data[0]);
2022 		end_synchronized_op(sc, 0);
2023 		if (rc == 0)
2024 			rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
2025 		break;
2026 	}
2027 
2028 	default:
2029 		rc = ether_ioctl(ifp, cmd, data);
2030 	}
2031 
2032 	return (rc);
2033 }
2034 
2035 static int
2036 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
2037 {
2038 	struct vi_info *vi = ifp->if_softc;
2039 	struct port_info *pi = vi->pi;
2040 	struct adapter *sc = pi->adapter;
2041 	struct sge_txq *txq;
2042 	void *items[1];
2043 	int rc;
2044 
2045 	M_ASSERTPKTHDR(m);
2046 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
2047 
2048 	if (__predict_false(pi->link_cfg.link_ok == false)) {
2049 		m_freem(m);
2050 		return (ENETDOWN);
2051 	}
2052 
2053 	rc = parse_pkt(sc, &m);
2054 	if (__predict_false(rc != 0)) {
2055 		MPASS(m == NULL);			/* was freed already */
2056 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
2057 		return (rc);
2058 	}
2059 #ifdef RATELIMIT
2060 	if (m->m_pkthdr.snd_tag != NULL) {
2061 		/* EAGAIN tells the stack we are not the correct interface. */
2062 		if (__predict_false(ifp != m->m_pkthdr.snd_tag->ifp)) {
2063 			m_freem(m);
2064 			return (EAGAIN);
2065 		}
2066 
2067 		return (ethofld_transmit(ifp, m));
2068 	}
2069 #endif
2070 
2071 	/* Select a txq. */
2072 	txq = &sc->sge.txq[vi->first_txq];
2073 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2074 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
2075 		    vi->rsrv_noflowq);
2076 
2077 	items[0] = m;
2078 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
2079 	if (__predict_false(rc != 0))
2080 		m_freem(m);
2081 
2082 	return (rc);
2083 }
2084 
2085 static void
2086 cxgbe_qflush(struct ifnet *ifp)
2087 {
2088 	struct vi_info *vi = ifp->if_softc;
2089 	struct sge_txq *txq;
2090 	int i;
2091 
2092 	/* queues do not exist if !VI_INIT_DONE. */
2093 	if (vi->flags & VI_INIT_DONE) {
2094 		for_each_txq(vi, i, txq) {
2095 			TXQ_LOCK(txq);
2096 			txq->eq.flags |= EQ_QFLUSH;
2097 			TXQ_UNLOCK(txq);
2098 			while (!mp_ring_is_idle(txq->r)) {
2099 				mp_ring_check_drainage(txq->r, 0);
2100 				pause("qflush", 1);
2101 			}
2102 			TXQ_LOCK(txq);
2103 			txq->eq.flags &= ~EQ_QFLUSH;
2104 			TXQ_UNLOCK(txq);
2105 		}
2106 	}
2107 	if_qflush(ifp);
2108 }
2109 
2110 static uint64_t
2111 vi_get_counter(struct ifnet *ifp, ift_counter c)
2112 {
2113 	struct vi_info *vi = ifp->if_softc;
2114 	struct fw_vi_stats_vf *s = &vi->stats;
2115 
2116 	vi_refresh_stats(vi->pi->adapter, vi);
2117 
2118 	switch (c) {
2119 	case IFCOUNTER_IPACKETS:
2120 		return (s->rx_bcast_frames + s->rx_mcast_frames +
2121 		    s->rx_ucast_frames);
2122 	case IFCOUNTER_IERRORS:
2123 		return (s->rx_err_frames);
2124 	case IFCOUNTER_OPACKETS:
2125 		return (s->tx_bcast_frames + s->tx_mcast_frames +
2126 		    s->tx_ucast_frames + s->tx_offload_frames);
2127 	case IFCOUNTER_OERRORS:
2128 		return (s->tx_drop_frames);
2129 	case IFCOUNTER_IBYTES:
2130 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
2131 		    s->rx_ucast_bytes);
2132 	case IFCOUNTER_OBYTES:
2133 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
2134 		    s->tx_ucast_bytes + s->tx_offload_bytes);
2135 	case IFCOUNTER_IMCASTS:
2136 		return (s->rx_mcast_frames);
2137 	case IFCOUNTER_OMCASTS:
2138 		return (s->tx_mcast_frames);
2139 	case IFCOUNTER_OQDROPS: {
2140 		uint64_t drops;
2141 
2142 		drops = 0;
2143 		if (vi->flags & VI_INIT_DONE) {
2144 			int i;
2145 			struct sge_txq *txq;
2146 
2147 			for_each_txq(vi, i, txq)
2148 				drops += counter_u64_fetch(txq->r->drops);
2149 		}
2150 
2151 		return (drops);
2152 
2153 	}
2154 
2155 	default:
2156 		return (if_get_counter_default(ifp, c));
2157 	}
2158 }
2159 
2160 uint64_t
2161 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
2162 {
2163 	struct vi_info *vi = ifp->if_softc;
2164 	struct port_info *pi = vi->pi;
2165 	struct adapter *sc = pi->adapter;
2166 	struct port_stats *s = &pi->stats;
2167 
2168 	if (pi->nvi > 1 || sc->flags & IS_VF)
2169 		return (vi_get_counter(ifp, c));
2170 
2171 	cxgbe_refresh_stats(sc, pi);
2172 
2173 	switch (c) {
2174 	case IFCOUNTER_IPACKETS:
2175 		return (s->rx_frames);
2176 
2177 	case IFCOUNTER_IERRORS:
2178 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
2179 		    s->rx_fcs_err + s->rx_len_err);
2180 
2181 	case IFCOUNTER_OPACKETS:
2182 		return (s->tx_frames);
2183 
2184 	case IFCOUNTER_OERRORS:
2185 		return (s->tx_error_frames);
2186 
2187 	case IFCOUNTER_IBYTES:
2188 		return (s->rx_octets);
2189 
2190 	case IFCOUNTER_OBYTES:
2191 		return (s->tx_octets);
2192 
2193 	case IFCOUNTER_IMCASTS:
2194 		return (s->rx_mcast_frames);
2195 
2196 	case IFCOUNTER_OMCASTS:
2197 		return (s->tx_mcast_frames);
2198 
2199 	case IFCOUNTER_IQDROPS:
2200 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
2201 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
2202 		    s->rx_trunc3 + pi->tnl_cong_drops);
2203 
2204 	case IFCOUNTER_OQDROPS: {
2205 		uint64_t drops;
2206 
2207 		drops = s->tx_drop;
2208 		if (vi->flags & VI_INIT_DONE) {
2209 			int i;
2210 			struct sge_txq *txq;
2211 
2212 			for_each_txq(vi, i, txq)
2213 				drops += counter_u64_fetch(txq->r->drops);
2214 		}
2215 
2216 		return (drops);
2217 
2218 	}
2219 
2220 	default:
2221 		return (if_get_counter_default(ifp, c));
2222 	}
2223 }
2224 
2225 /*
2226  * The kernel picks a media from the list we had provided but we still validate
2227  * the requeste.
2228  */
2229 int
2230 cxgbe_media_change(struct ifnet *ifp)
2231 {
2232 	struct vi_info *vi = ifp->if_softc;
2233 	struct port_info *pi = vi->pi;
2234 	struct ifmedia *ifm = &pi->media;
2235 	struct link_config *lc = &pi->link_cfg;
2236 	struct adapter *sc = pi->adapter;
2237 	int rc;
2238 
2239 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mec");
2240 	if (rc != 0)
2241 		return (rc);
2242 	PORT_LOCK(pi);
2243 	if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
2244 		/* ifconfig .. media autoselect */
2245 		if (!(lc->supported & FW_PORT_CAP32_ANEG)) {
2246 			rc = ENOTSUP; /* AN not supported by transceiver */
2247 			goto done;
2248 		}
2249 		lc->requested_aneg = AUTONEG_ENABLE;
2250 		lc->requested_speed = 0;
2251 		lc->requested_fc |= PAUSE_AUTONEG;
2252 	} else {
2253 		lc->requested_aneg = AUTONEG_DISABLE;
2254 		lc->requested_speed =
2255 		    ifmedia_baudrate(ifm->ifm_media) / 1000000;
2256 		lc->requested_fc = 0;
2257 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
2258 			lc->requested_fc |= PAUSE_RX;
2259 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
2260 			lc->requested_fc |= PAUSE_TX;
2261 	}
2262 	if (pi->up_vis > 0) {
2263 		fixup_link_config(pi);
2264 		rc = apply_link_config(pi);
2265 	}
2266 done:
2267 	PORT_UNLOCK(pi);
2268 	end_synchronized_op(sc, 0);
2269 	return (rc);
2270 }
2271 
2272 /*
2273  * Base media word (without ETHER, pause, link active, etc.) for the port at the
2274  * given speed.
2275  */
2276 static int
2277 port_mword(struct port_info *pi, uint32_t speed)
2278 {
2279 
2280 	MPASS(speed & M_FW_PORT_CAP32_SPEED);
2281 	MPASS(powerof2(speed));
2282 
2283 	switch(pi->port_type) {
2284 	case FW_PORT_TYPE_BT_SGMII:
2285 	case FW_PORT_TYPE_BT_XFI:
2286 	case FW_PORT_TYPE_BT_XAUI:
2287 		/* BaseT */
2288 		switch (speed) {
2289 		case FW_PORT_CAP32_SPEED_100M:
2290 			return (IFM_100_T);
2291 		case FW_PORT_CAP32_SPEED_1G:
2292 			return (IFM_1000_T);
2293 		case FW_PORT_CAP32_SPEED_10G:
2294 			return (IFM_10G_T);
2295 		}
2296 		break;
2297 	case FW_PORT_TYPE_KX4:
2298 		if (speed == FW_PORT_CAP32_SPEED_10G)
2299 			return (IFM_10G_KX4);
2300 		break;
2301 	case FW_PORT_TYPE_CX4:
2302 		if (speed == FW_PORT_CAP32_SPEED_10G)
2303 			return (IFM_10G_CX4);
2304 		break;
2305 	case FW_PORT_TYPE_KX:
2306 		if (speed == FW_PORT_CAP32_SPEED_1G)
2307 			return (IFM_1000_KX);
2308 		break;
2309 	case FW_PORT_TYPE_KR:
2310 	case FW_PORT_TYPE_BP_AP:
2311 	case FW_PORT_TYPE_BP4_AP:
2312 	case FW_PORT_TYPE_BP40_BA:
2313 	case FW_PORT_TYPE_KR4_100G:
2314 	case FW_PORT_TYPE_KR_SFP28:
2315 	case FW_PORT_TYPE_KR_XLAUI:
2316 		switch (speed) {
2317 		case FW_PORT_CAP32_SPEED_1G:
2318 			return (IFM_1000_KX);
2319 		case FW_PORT_CAP32_SPEED_10G:
2320 			return (IFM_10G_KR);
2321 		case FW_PORT_CAP32_SPEED_25G:
2322 			return (IFM_25G_KR);
2323 		case FW_PORT_CAP32_SPEED_40G:
2324 			return (IFM_40G_KR4);
2325 		case FW_PORT_CAP32_SPEED_50G:
2326 			return (IFM_50G_KR2);
2327 		case FW_PORT_CAP32_SPEED_100G:
2328 			return (IFM_100G_KR4);
2329 		}
2330 		break;
2331 	case FW_PORT_TYPE_FIBER_XFI:
2332 	case FW_PORT_TYPE_FIBER_XAUI:
2333 	case FW_PORT_TYPE_SFP:
2334 	case FW_PORT_TYPE_QSFP_10G:
2335 	case FW_PORT_TYPE_QSA:
2336 	case FW_PORT_TYPE_QSFP:
2337 	case FW_PORT_TYPE_CR4_QSFP:
2338 	case FW_PORT_TYPE_CR_QSFP:
2339 	case FW_PORT_TYPE_CR2_QSFP:
2340 	case FW_PORT_TYPE_SFP28:
2341 		/* Pluggable transceiver */
2342 		switch (pi->mod_type) {
2343 		case FW_PORT_MOD_TYPE_LR:
2344 			switch (speed) {
2345 			case FW_PORT_CAP32_SPEED_1G:
2346 				return (IFM_1000_LX);
2347 			case FW_PORT_CAP32_SPEED_10G:
2348 				return (IFM_10G_LR);
2349 			case FW_PORT_CAP32_SPEED_25G:
2350 				return (IFM_25G_LR);
2351 			case FW_PORT_CAP32_SPEED_40G:
2352 				return (IFM_40G_LR4);
2353 			case FW_PORT_CAP32_SPEED_50G:
2354 				return (IFM_50G_LR2);
2355 			case FW_PORT_CAP32_SPEED_100G:
2356 				return (IFM_100G_LR4);
2357 			}
2358 			break;
2359 		case FW_PORT_MOD_TYPE_SR:
2360 			switch (speed) {
2361 			case FW_PORT_CAP32_SPEED_1G:
2362 				return (IFM_1000_SX);
2363 			case FW_PORT_CAP32_SPEED_10G:
2364 				return (IFM_10G_SR);
2365 			case FW_PORT_CAP32_SPEED_25G:
2366 				return (IFM_25G_SR);
2367 			case FW_PORT_CAP32_SPEED_40G:
2368 				return (IFM_40G_SR4);
2369 			case FW_PORT_CAP32_SPEED_50G:
2370 				return (IFM_50G_SR2);
2371 			case FW_PORT_CAP32_SPEED_100G:
2372 				return (IFM_100G_SR4);
2373 			}
2374 			break;
2375 		case FW_PORT_MOD_TYPE_ER:
2376 			if (speed == FW_PORT_CAP32_SPEED_10G)
2377 				return (IFM_10G_ER);
2378 			break;
2379 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2380 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2381 			switch (speed) {
2382 			case FW_PORT_CAP32_SPEED_1G:
2383 				return (IFM_1000_CX);
2384 			case FW_PORT_CAP32_SPEED_10G:
2385 				return (IFM_10G_TWINAX);
2386 			case FW_PORT_CAP32_SPEED_25G:
2387 				return (IFM_25G_CR);
2388 			case FW_PORT_CAP32_SPEED_40G:
2389 				return (IFM_40G_CR4);
2390 			case FW_PORT_CAP32_SPEED_50G:
2391 				return (IFM_50G_CR2);
2392 			case FW_PORT_CAP32_SPEED_100G:
2393 				return (IFM_100G_CR4);
2394 			}
2395 			break;
2396 		case FW_PORT_MOD_TYPE_LRM:
2397 			if (speed == FW_PORT_CAP32_SPEED_10G)
2398 				return (IFM_10G_LRM);
2399 			break;
2400 		case FW_PORT_MOD_TYPE_NA:
2401 			MPASS(0);	/* Not pluggable? */
2402 			/* fall throough */
2403 		case FW_PORT_MOD_TYPE_ERROR:
2404 		case FW_PORT_MOD_TYPE_UNKNOWN:
2405 		case FW_PORT_MOD_TYPE_NOTSUPPORTED:
2406 			break;
2407 		case FW_PORT_MOD_TYPE_NONE:
2408 			return (IFM_NONE);
2409 		}
2410 		break;
2411 	case FW_PORT_TYPE_NONE:
2412 		return (IFM_NONE);
2413 	}
2414 
2415 	return (IFM_UNKNOWN);
2416 }
2417 
2418 void
2419 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2420 {
2421 	struct vi_info *vi = ifp->if_softc;
2422 	struct port_info *pi = vi->pi;
2423 	struct adapter *sc = pi->adapter;
2424 	struct link_config *lc = &pi->link_cfg;
2425 
2426 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4med") != 0)
2427 		return;
2428 	PORT_LOCK(pi);
2429 
2430 	if (pi->up_vis == 0) {
2431 		/*
2432 		 * If all the interfaces are administratively down the firmware
2433 		 * does not report transceiver changes.  Refresh port info here
2434 		 * so that ifconfig displays accurate ifmedia at all times.
2435 		 * This is the only reason we have a synchronized op in this
2436 		 * function.  Just PORT_LOCK would have been enough otherwise.
2437 		 */
2438 		t4_update_port_info(pi);
2439 		build_medialist(pi);
2440 	}
2441 
2442 	/* ifm_status */
2443 	ifmr->ifm_status = IFM_AVALID;
2444 	if (lc->link_ok == false)
2445 		goto done;
2446 	ifmr->ifm_status |= IFM_ACTIVE;
2447 
2448 	/* ifm_active */
2449 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2450 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2451 	if (lc->fc & PAUSE_RX)
2452 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2453 	if (lc->fc & PAUSE_TX)
2454 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2455 	ifmr->ifm_active |= port_mword(pi, speed_to_fwcap(lc->speed));
2456 done:
2457 	PORT_UNLOCK(pi);
2458 	end_synchronized_op(sc, 0);
2459 }
2460 
2461 static int
2462 vcxgbe_probe(device_t dev)
2463 {
2464 	char buf[128];
2465 	struct vi_info *vi = device_get_softc(dev);
2466 
2467 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2468 	    vi - vi->pi->vi);
2469 	device_set_desc_copy(dev, buf);
2470 
2471 	return (BUS_PROBE_DEFAULT);
2472 }
2473 
2474 static int
2475 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2476 {
2477 	int func, index, rc;
2478 	uint32_t param, val;
2479 
2480 	ASSERT_SYNCHRONIZED_OP(sc);
2481 
2482 	index = vi - pi->vi;
2483 	MPASS(index > 0);	/* This function deals with _extra_ VIs only */
2484 	KASSERT(index < nitems(vi_mac_funcs),
2485 	    ("%s: VI %s doesn't have a MAC func", __func__,
2486 	    device_get_nameunit(vi->dev)));
2487 	func = vi_mac_funcs[index];
2488 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2489 	    vi->hw_addr, &vi->rss_size, &vi->vfvld, &vi->vin, func, 0);
2490 	if (rc < 0) {
2491 		device_printf(vi->dev, "failed to allocate virtual interface %d"
2492 		    "for port %d: %d\n", index, pi->port_id, -rc);
2493 		return (-rc);
2494 	}
2495 	vi->viid = rc;
2496 
2497 	if (vi->rss_size == 1) {
2498 		/*
2499 		 * This VI didn't get a slice of the RSS table.  Reduce the
2500 		 * number of VIs being created (hw.cxgbe.num_vis) or modify the
2501 		 * configuration file (nvi, rssnvi for this PF) if this is a
2502 		 * problem.
2503 		 */
2504 		device_printf(vi->dev, "RSS table not available.\n");
2505 		vi->rss_base = 0xffff;
2506 
2507 		return (0);
2508 	}
2509 
2510 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2511 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2512 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2513 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2514 	if (rc)
2515 		vi->rss_base = 0xffff;
2516 	else {
2517 		MPASS((val >> 16) == vi->rss_size);
2518 		vi->rss_base = val & 0xffff;
2519 	}
2520 
2521 	return (0);
2522 }
2523 
2524 static int
2525 vcxgbe_attach(device_t dev)
2526 {
2527 	struct vi_info *vi;
2528 	struct port_info *pi;
2529 	struct adapter *sc;
2530 	int rc;
2531 
2532 	vi = device_get_softc(dev);
2533 	pi = vi->pi;
2534 	sc = pi->adapter;
2535 
2536 	rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2537 	if (rc)
2538 		return (rc);
2539 	rc = alloc_extra_vi(sc, pi, vi);
2540 	end_synchronized_op(sc, 0);
2541 	if (rc)
2542 		return (rc);
2543 
2544 	rc = cxgbe_vi_attach(dev, vi);
2545 	if (rc) {
2546 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2547 		return (rc);
2548 	}
2549 	return (0);
2550 }
2551 
2552 static int
2553 vcxgbe_detach(device_t dev)
2554 {
2555 	struct vi_info *vi;
2556 	struct adapter *sc;
2557 
2558 	vi = device_get_softc(dev);
2559 	sc = vi->pi->adapter;
2560 
2561 	doom_vi(sc, vi);
2562 
2563 	cxgbe_vi_detach(vi);
2564 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2565 
2566 	end_synchronized_op(sc, 0);
2567 
2568 	return (0);
2569 }
2570 
2571 static struct callout fatal_callout;
2572 
2573 static void
2574 delayed_panic(void *arg)
2575 {
2576 	struct adapter *sc = arg;
2577 
2578 	panic("%s: panic on fatal error", device_get_nameunit(sc->dev));
2579 }
2580 
2581 void
2582 t4_fatal_err(struct adapter *sc, bool fw_error)
2583 {
2584 
2585 	t4_shutdown_adapter(sc);
2586 	log(LOG_ALERT, "%s: encountered fatal error, adapter stopped.\n",
2587 	    device_get_nameunit(sc->dev));
2588 	if (fw_error) {
2589 		ASSERT_SYNCHRONIZED_OP(sc);
2590 		sc->flags |= ADAP_ERR;
2591 	} else {
2592 		ADAPTER_LOCK(sc);
2593 		sc->flags |= ADAP_ERR;
2594 		ADAPTER_UNLOCK(sc);
2595 	}
2596 
2597 	if (t4_panic_on_fatal_err) {
2598 		log(LOG_ALERT, "%s: panic on fatal error after 30s",
2599 		    device_get_nameunit(sc->dev));
2600 		callout_reset(&fatal_callout, hz * 30, delayed_panic, sc);
2601 	}
2602 }
2603 
2604 void
2605 t4_add_adapter(struct adapter *sc)
2606 {
2607 	sx_xlock(&t4_list_lock);
2608 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2609 	sx_xunlock(&t4_list_lock);
2610 }
2611 
2612 int
2613 t4_map_bars_0_and_4(struct adapter *sc)
2614 {
2615 	sc->regs_rid = PCIR_BAR(0);
2616 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2617 	    &sc->regs_rid, RF_ACTIVE);
2618 	if (sc->regs_res == NULL) {
2619 		device_printf(sc->dev, "cannot map registers.\n");
2620 		return (ENXIO);
2621 	}
2622 	sc->bt = rman_get_bustag(sc->regs_res);
2623 	sc->bh = rman_get_bushandle(sc->regs_res);
2624 	sc->mmio_len = rman_get_size(sc->regs_res);
2625 	setbit(&sc->doorbells, DOORBELL_KDB);
2626 
2627 	sc->msix_rid = PCIR_BAR(4);
2628 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2629 	    &sc->msix_rid, RF_ACTIVE);
2630 	if (sc->msix_res == NULL) {
2631 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2632 		return (ENXIO);
2633 	}
2634 
2635 	return (0);
2636 }
2637 
2638 int
2639 t4_map_bar_2(struct adapter *sc)
2640 {
2641 
2642 	/*
2643 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2644 	 * to map it if RDMA is disabled.
2645 	 */
2646 	if (is_t4(sc) && sc->rdmacaps == 0)
2647 		return (0);
2648 
2649 	sc->udbs_rid = PCIR_BAR(2);
2650 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2651 	    &sc->udbs_rid, RF_ACTIVE);
2652 	if (sc->udbs_res == NULL) {
2653 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2654 		return (ENXIO);
2655 	}
2656 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2657 
2658 	if (chip_id(sc) >= CHELSIO_T5) {
2659 		setbit(&sc->doorbells, DOORBELL_UDB);
2660 #if defined(__i386__) || defined(__amd64__)
2661 		if (t5_write_combine) {
2662 			int rc, mode;
2663 
2664 			/*
2665 			 * Enable write combining on BAR2.  This is the
2666 			 * userspace doorbell BAR and is split into 128B
2667 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2668 			 * with an egress queue.  The first 64B has the doorbell
2669 			 * and the second 64B can be used to submit a tx work
2670 			 * request with an implicit doorbell.
2671 			 */
2672 
2673 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2674 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2675 			if (rc == 0) {
2676 				clrbit(&sc->doorbells, DOORBELL_UDB);
2677 				setbit(&sc->doorbells, DOORBELL_WCWR);
2678 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2679 			} else {
2680 				device_printf(sc->dev,
2681 				    "couldn't enable write combining: %d\n",
2682 				    rc);
2683 			}
2684 
2685 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2686 			t4_write_reg(sc, A_SGE_STAT_CFG,
2687 			    V_STATSOURCE_T5(7) | mode);
2688 		}
2689 #endif
2690 	}
2691 	sc->iwt.wc_en = isset(&sc->doorbells, DOORBELL_UDBWC) ? 1 : 0;
2692 
2693 	return (0);
2694 }
2695 
2696 struct memwin_init {
2697 	uint32_t base;
2698 	uint32_t aperture;
2699 };
2700 
2701 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2702 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2703 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2704 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2705 };
2706 
2707 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2708 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2709 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2710 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2711 };
2712 
2713 static void
2714 setup_memwin(struct adapter *sc)
2715 {
2716 	const struct memwin_init *mw_init;
2717 	struct memwin *mw;
2718 	int i;
2719 	uint32_t bar0;
2720 
2721 	if (is_t4(sc)) {
2722 		/*
2723 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2724 		 * mechanism.  Works from within PCI passthrough environments
2725 		 * too, where rman_get_start() can return a different value.  We
2726 		 * need to program the T4 memory window decoders with the actual
2727 		 * addresses that will be coming across the PCIe link.
2728 		 */
2729 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2730 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2731 
2732 		mw_init = &t4_memwin[0];
2733 	} else {
2734 		/* T5+ use the relative offset inside the PCIe BAR */
2735 		bar0 = 0;
2736 
2737 		mw_init = &t5_memwin[0];
2738 	}
2739 
2740 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2741 		rw_init(&mw->mw_lock, "memory window access");
2742 		mw->mw_base = mw_init->base;
2743 		mw->mw_aperture = mw_init->aperture;
2744 		mw->mw_curpos = 0;
2745 		t4_write_reg(sc,
2746 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2747 		    (mw->mw_base + bar0) | V_BIR(0) |
2748 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2749 		rw_wlock(&mw->mw_lock);
2750 		position_memwin(sc, i, 0);
2751 		rw_wunlock(&mw->mw_lock);
2752 	}
2753 
2754 	/* flush */
2755 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2756 }
2757 
2758 /*
2759  * Positions the memory window at the given address in the card's address space.
2760  * There are some alignment requirements and the actual position may be at an
2761  * address prior to the requested address.  mw->mw_curpos always has the actual
2762  * position of the window.
2763  */
2764 static void
2765 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2766 {
2767 	struct memwin *mw;
2768 	uint32_t pf;
2769 	uint32_t reg;
2770 
2771 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2772 	mw = &sc->memwin[idx];
2773 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2774 
2775 	if (is_t4(sc)) {
2776 		pf = 0;
2777 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2778 	} else {
2779 		pf = V_PFNUM(sc->pf);
2780 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2781 	}
2782 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2783 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2784 	t4_read_reg(sc, reg);	/* flush */
2785 }
2786 
2787 int
2788 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2789     int len, int rw)
2790 {
2791 	struct memwin *mw;
2792 	uint32_t mw_end, v;
2793 
2794 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2795 
2796 	/* Memory can only be accessed in naturally aligned 4 byte units */
2797 	if (addr & 3 || len & 3 || len <= 0)
2798 		return (EINVAL);
2799 
2800 	mw = &sc->memwin[idx];
2801 	while (len > 0) {
2802 		rw_rlock(&mw->mw_lock);
2803 		mw_end = mw->mw_curpos + mw->mw_aperture;
2804 		if (addr >= mw_end || addr < mw->mw_curpos) {
2805 			/* Will need to reposition the window */
2806 			if (!rw_try_upgrade(&mw->mw_lock)) {
2807 				rw_runlock(&mw->mw_lock);
2808 				rw_wlock(&mw->mw_lock);
2809 			}
2810 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2811 			position_memwin(sc, idx, addr);
2812 			rw_downgrade(&mw->mw_lock);
2813 			mw_end = mw->mw_curpos + mw->mw_aperture;
2814 		}
2815 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2816 		while (addr < mw_end && len > 0) {
2817 			if (rw == 0) {
2818 				v = t4_read_reg(sc, mw->mw_base + addr -
2819 				    mw->mw_curpos);
2820 				*val++ = le32toh(v);
2821 			} else {
2822 				v = *val++;
2823 				t4_write_reg(sc, mw->mw_base + addr -
2824 				    mw->mw_curpos, htole32(v));
2825 			}
2826 			addr += 4;
2827 			len -= 4;
2828 		}
2829 		rw_runlock(&mw->mw_lock);
2830 	}
2831 
2832 	return (0);
2833 }
2834 
2835 int
2836 alloc_atid_tab(struct tid_info *t, int flags)
2837 {
2838 	int i;
2839 
2840 	MPASS(t->natids > 0);
2841 	MPASS(t->atid_tab == NULL);
2842 
2843 	t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
2844 	    M_ZERO | flags);
2845 	if (t->atid_tab == NULL)
2846 		return (ENOMEM);
2847 	mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
2848 	t->afree = t->atid_tab;
2849 	t->atids_in_use = 0;
2850 	for (i = 1; i < t->natids; i++)
2851 		t->atid_tab[i - 1].next = &t->atid_tab[i];
2852 	t->atid_tab[t->natids - 1].next = NULL;
2853 
2854 	return (0);
2855 }
2856 
2857 void
2858 free_atid_tab(struct tid_info *t)
2859 {
2860 
2861 	KASSERT(t->atids_in_use == 0,
2862 	    ("%s: %d atids still in use.", __func__, t->atids_in_use));
2863 
2864 	if (mtx_initialized(&t->atid_lock))
2865 		mtx_destroy(&t->atid_lock);
2866 	free(t->atid_tab, M_CXGBE);
2867 	t->atid_tab = NULL;
2868 }
2869 
2870 int
2871 alloc_atid(struct adapter *sc, void *ctx)
2872 {
2873 	struct tid_info *t = &sc->tids;
2874 	int atid = -1;
2875 
2876 	mtx_lock(&t->atid_lock);
2877 	if (t->afree) {
2878 		union aopen_entry *p = t->afree;
2879 
2880 		atid = p - t->atid_tab;
2881 		MPASS(atid <= M_TID_TID);
2882 		t->afree = p->next;
2883 		p->data = ctx;
2884 		t->atids_in_use++;
2885 	}
2886 	mtx_unlock(&t->atid_lock);
2887 	return (atid);
2888 }
2889 
2890 void *
2891 lookup_atid(struct adapter *sc, int atid)
2892 {
2893 	struct tid_info *t = &sc->tids;
2894 
2895 	return (t->atid_tab[atid].data);
2896 }
2897 
2898 void
2899 free_atid(struct adapter *sc, int atid)
2900 {
2901 	struct tid_info *t = &sc->tids;
2902 	union aopen_entry *p = &t->atid_tab[atid];
2903 
2904 	mtx_lock(&t->atid_lock);
2905 	p->next = t->afree;
2906 	t->afree = p;
2907 	t->atids_in_use--;
2908 	mtx_unlock(&t->atid_lock);
2909 }
2910 
2911 static void
2912 queue_tid_release(struct adapter *sc, int tid)
2913 {
2914 
2915 	CXGBE_UNIMPLEMENTED("deferred tid release");
2916 }
2917 
2918 void
2919 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
2920 {
2921 	struct wrqe *wr;
2922 	struct cpl_tid_release *req;
2923 
2924 	wr = alloc_wrqe(sizeof(*req), ctrlq);
2925 	if (wr == NULL) {
2926 		queue_tid_release(sc, tid);	/* defer */
2927 		return;
2928 	}
2929 	req = wrtod(wr);
2930 
2931 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
2932 
2933 	t4_wrq_tx(sc, wr);
2934 }
2935 
2936 static int
2937 t4_range_cmp(const void *a, const void *b)
2938 {
2939 	return ((const struct t4_range *)a)->start -
2940 	       ((const struct t4_range *)b)->start;
2941 }
2942 
2943 /*
2944  * Verify that the memory range specified by the addr/len pair is valid within
2945  * the card's address space.
2946  */
2947 static int
2948 validate_mem_range(struct adapter *sc, uint32_t addr, uint32_t len)
2949 {
2950 	struct t4_range mem_ranges[4], *r, *next;
2951 	uint32_t em, addr_len;
2952 	int i, n, remaining;
2953 
2954 	/* Memory can only be accessed in naturally aligned 4 byte units */
2955 	if (addr & 3 || len & 3 || len == 0)
2956 		return (EINVAL);
2957 
2958 	/* Enabled memories */
2959 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2960 
2961 	r = &mem_ranges[0];
2962 	n = 0;
2963 	bzero(r, sizeof(mem_ranges));
2964 	if (em & F_EDRAM0_ENABLE) {
2965 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2966 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2967 		if (r->size > 0) {
2968 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2969 			if (addr >= r->start &&
2970 			    addr + len <= r->start + r->size)
2971 				return (0);
2972 			r++;
2973 			n++;
2974 		}
2975 	}
2976 	if (em & F_EDRAM1_ENABLE) {
2977 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2978 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2979 		if (r->size > 0) {
2980 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2981 			if (addr >= r->start &&
2982 			    addr + len <= r->start + r->size)
2983 				return (0);
2984 			r++;
2985 			n++;
2986 		}
2987 	}
2988 	if (em & F_EXT_MEM_ENABLE) {
2989 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2990 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2991 		if (r->size > 0) {
2992 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2993 			if (addr >= r->start &&
2994 			    addr + len <= r->start + r->size)
2995 				return (0);
2996 			r++;
2997 			n++;
2998 		}
2999 	}
3000 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
3001 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
3002 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
3003 		if (r->size > 0) {
3004 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
3005 			if (addr >= r->start &&
3006 			    addr + len <= r->start + r->size)
3007 				return (0);
3008 			r++;
3009 			n++;
3010 		}
3011 	}
3012 	MPASS(n <= nitems(mem_ranges));
3013 
3014 	if (n > 1) {
3015 		/* Sort and merge the ranges. */
3016 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
3017 
3018 		/* Start from index 0 and examine the next n - 1 entries. */
3019 		r = &mem_ranges[0];
3020 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
3021 
3022 			MPASS(r->size > 0);	/* r is a valid entry. */
3023 			next = r + 1;
3024 			MPASS(next->size > 0);	/* and so is the next one. */
3025 
3026 			while (r->start + r->size >= next->start) {
3027 				/* Merge the next one into the current entry. */
3028 				r->size = max(r->start + r->size,
3029 				    next->start + next->size) - r->start;
3030 				n--;	/* One fewer entry in total. */
3031 				if (--remaining == 0)
3032 					goto done;	/* short circuit */
3033 				next++;
3034 			}
3035 			if (next != r + 1) {
3036 				/*
3037 				 * Some entries were merged into r and next
3038 				 * points to the first valid entry that couldn't
3039 				 * be merged.
3040 				 */
3041 				MPASS(next->size > 0);	/* must be valid */
3042 				memcpy(r + 1, next, remaining * sizeof(*r));
3043 #ifdef INVARIANTS
3044 				/*
3045 				 * This so that the foo->size assertion in the
3046 				 * next iteration of the loop do the right
3047 				 * thing for entries that were pulled up and are
3048 				 * no longer valid.
3049 				 */
3050 				MPASS(n < nitems(mem_ranges));
3051 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
3052 				    sizeof(struct t4_range));
3053 #endif
3054 			}
3055 		}
3056 done:
3057 		/* Done merging the ranges. */
3058 		MPASS(n > 0);
3059 		r = &mem_ranges[0];
3060 		for (i = 0; i < n; i++, r++) {
3061 			if (addr >= r->start &&
3062 			    addr + len <= r->start + r->size)
3063 				return (0);
3064 		}
3065 	}
3066 
3067 	return (EFAULT);
3068 }
3069 
3070 static int
3071 fwmtype_to_hwmtype(int mtype)
3072 {
3073 
3074 	switch (mtype) {
3075 	case FW_MEMTYPE_EDC0:
3076 		return (MEM_EDC0);
3077 	case FW_MEMTYPE_EDC1:
3078 		return (MEM_EDC1);
3079 	case FW_MEMTYPE_EXTMEM:
3080 		return (MEM_MC0);
3081 	case FW_MEMTYPE_EXTMEM1:
3082 		return (MEM_MC1);
3083 	default:
3084 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
3085 	}
3086 }
3087 
3088 /*
3089  * Verify that the memory range specified by the memtype/offset/len pair is
3090  * valid and lies entirely within the memtype specified.  The global address of
3091  * the start of the range is returned in addr.
3092  */
3093 static int
3094 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, uint32_t len,
3095     uint32_t *addr)
3096 {
3097 	uint32_t em, addr_len, maddr;
3098 
3099 	/* Memory can only be accessed in naturally aligned 4 byte units */
3100 	if (off & 3 || len & 3 || len == 0)
3101 		return (EINVAL);
3102 
3103 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
3104 	switch (fwmtype_to_hwmtype(mtype)) {
3105 	case MEM_EDC0:
3106 		if (!(em & F_EDRAM0_ENABLE))
3107 			return (EINVAL);
3108 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
3109 		maddr = G_EDRAM0_BASE(addr_len) << 20;
3110 		break;
3111 	case MEM_EDC1:
3112 		if (!(em & F_EDRAM1_ENABLE))
3113 			return (EINVAL);
3114 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
3115 		maddr = G_EDRAM1_BASE(addr_len) << 20;
3116 		break;
3117 	case MEM_MC:
3118 		if (!(em & F_EXT_MEM_ENABLE))
3119 			return (EINVAL);
3120 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
3121 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
3122 		break;
3123 	case MEM_MC1:
3124 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
3125 			return (EINVAL);
3126 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
3127 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
3128 		break;
3129 	default:
3130 		return (EINVAL);
3131 	}
3132 
3133 	*addr = maddr + off;	/* global address */
3134 	return (validate_mem_range(sc, *addr, len));
3135 }
3136 
3137 static int
3138 fixup_devlog_params(struct adapter *sc)
3139 {
3140 	struct devlog_params *dparams = &sc->params.devlog;
3141 	int rc;
3142 
3143 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
3144 	    dparams->size, &dparams->addr);
3145 
3146 	return (rc);
3147 }
3148 
3149 static void
3150 update_nirq(struct intrs_and_queues *iaq, int nports)
3151 {
3152 	int extra = T4_EXTRA_INTR;
3153 
3154 	iaq->nirq = extra;
3155 	iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
3156 	iaq->nirq += nports * (iaq->num_vis - 1) *
3157 	    max(iaq->nrxq_vi, iaq->nnmrxq_vi);
3158 	iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
3159 }
3160 
3161 /*
3162  * Adjust requirements to fit the number of interrupts available.
3163  */
3164 static void
3165 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
3166     int navail)
3167 {
3168 	int old_nirq;
3169 	const int nports = sc->params.nports;
3170 
3171 	MPASS(nports > 0);
3172 	MPASS(navail > 0);
3173 
3174 	bzero(iaq, sizeof(*iaq));
3175 	iaq->intr_type = itype;
3176 	iaq->num_vis = t4_num_vis;
3177 	iaq->ntxq = t4_ntxq;
3178 	iaq->ntxq_vi = t4_ntxq_vi;
3179 	iaq->nrxq = t4_nrxq;
3180 	iaq->nrxq_vi = t4_nrxq_vi;
3181 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
3182 	if (is_offload(sc) || is_ethoffload(sc)) {
3183 		iaq->nofldtxq = t4_nofldtxq;
3184 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
3185 	}
3186 #endif
3187 #ifdef TCP_OFFLOAD
3188 	if (is_offload(sc)) {
3189 		iaq->nofldrxq = t4_nofldrxq;
3190 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
3191 	}
3192 #endif
3193 #ifdef DEV_NETMAP
3194 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
3195 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
3196 #endif
3197 
3198 	update_nirq(iaq, nports);
3199 	if (iaq->nirq <= navail &&
3200 	    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3201 		/*
3202 		 * This is the normal case -- there are enough interrupts for
3203 		 * everything.
3204 		 */
3205 		goto done;
3206 	}
3207 
3208 	/*
3209 	 * If extra VIs have been configured try reducing their count and see if
3210 	 * that works.
3211 	 */
3212 	while (iaq->num_vis > 1) {
3213 		iaq->num_vis--;
3214 		update_nirq(iaq, nports);
3215 		if (iaq->nirq <= navail &&
3216 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3217 			device_printf(sc->dev, "virtual interfaces per port "
3218 			    "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
3219 			    "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
3220 			    "itype %d, navail %u, nirq %d.\n",
3221 			    iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
3222 			    iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
3223 			    itype, navail, iaq->nirq);
3224 			goto done;
3225 		}
3226 	}
3227 
3228 	/*
3229 	 * Extra VIs will not be created.  Log a message if they were requested.
3230 	 */
3231 	MPASS(iaq->num_vis == 1);
3232 	iaq->ntxq_vi = iaq->nrxq_vi = 0;
3233 	iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
3234 	iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
3235 	if (iaq->num_vis != t4_num_vis) {
3236 		device_printf(sc->dev, "extra virtual interfaces disabled.  "
3237 		    "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
3238 		    "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
3239 		    iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
3240 		    iaq->nnmrxq_vi, itype, navail, iaq->nirq);
3241 	}
3242 
3243 	/*
3244 	 * Keep reducing the number of NIC rx queues to the next lower power of
3245 	 * 2 (for even RSS distribution) and halving the TOE rx queues and see
3246 	 * if that works.
3247 	 */
3248 	do {
3249 		if (iaq->nrxq > 1) {
3250 			do {
3251 				iaq->nrxq--;
3252 			} while (!powerof2(iaq->nrxq));
3253 		}
3254 		if (iaq->nofldrxq > 1)
3255 			iaq->nofldrxq >>= 1;
3256 
3257 		old_nirq = iaq->nirq;
3258 		update_nirq(iaq, nports);
3259 		if (iaq->nirq <= navail &&
3260 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3261 			device_printf(sc->dev, "running with reduced number of "
3262 			    "rx queues because of shortage of interrupts.  "
3263 			    "nrxq=%u, nofldrxq=%u.  "
3264 			    "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
3265 			    iaq->nofldrxq, itype, navail, iaq->nirq);
3266 			goto done;
3267 		}
3268 	} while (old_nirq != iaq->nirq);
3269 
3270 	/* One interrupt for everything.  Ugh. */
3271 	device_printf(sc->dev, "running with minimal number of queues.  "
3272 	    "itype %d, navail %u.\n", itype, navail);
3273 	iaq->nirq = 1;
3274 	MPASS(iaq->nrxq == 1);
3275 	iaq->ntxq = 1;
3276 	if (iaq->nofldrxq > 1)
3277 		iaq->nofldtxq = 1;
3278 done:
3279 	MPASS(iaq->num_vis > 0);
3280 	if (iaq->num_vis > 1) {
3281 		MPASS(iaq->nrxq_vi > 0);
3282 		MPASS(iaq->ntxq_vi > 0);
3283 	}
3284 	MPASS(iaq->nirq > 0);
3285 	MPASS(iaq->nrxq > 0);
3286 	MPASS(iaq->ntxq > 0);
3287 	if (itype == INTR_MSI) {
3288 		MPASS(powerof2(iaq->nirq));
3289 	}
3290 }
3291 
3292 static int
3293 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
3294 {
3295 	int rc, itype, navail, nalloc;
3296 
3297 	for (itype = INTR_MSIX; itype; itype >>= 1) {
3298 
3299 		if ((itype & t4_intr_types) == 0)
3300 			continue;	/* not allowed */
3301 
3302 		if (itype == INTR_MSIX)
3303 			navail = pci_msix_count(sc->dev);
3304 		else if (itype == INTR_MSI)
3305 			navail = pci_msi_count(sc->dev);
3306 		else
3307 			navail = 1;
3308 restart:
3309 		if (navail == 0)
3310 			continue;
3311 
3312 		calculate_iaq(sc, iaq, itype, navail);
3313 		nalloc = iaq->nirq;
3314 		rc = 0;
3315 		if (itype == INTR_MSIX)
3316 			rc = pci_alloc_msix(sc->dev, &nalloc);
3317 		else if (itype == INTR_MSI)
3318 			rc = pci_alloc_msi(sc->dev, &nalloc);
3319 
3320 		if (rc == 0 && nalloc > 0) {
3321 			if (nalloc == iaq->nirq)
3322 				return (0);
3323 
3324 			/*
3325 			 * Didn't get the number requested.  Use whatever number
3326 			 * the kernel is willing to allocate.
3327 			 */
3328 			device_printf(sc->dev, "fewer vectors than requested, "
3329 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
3330 			    itype, iaq->nirq, nalloc);
3331 			pci_release_msi(sc->dev);
3332 			navail = nalloc;
3333 			goto restart;
3334 		}
3335 
3336 		device_printf(sc->dev,
3337 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
3338 		    itype, rc, iaq->nirq, nalloc);
3339 	}
3340 
3341 	device_printf(sc->dev,
3342 	    "failed to find a usable interrupt type.  "
3343 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
3344 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
3345 
3346 	return (ENXIO);
3347 }
3348 
3349 #define FW_VERSION(chip) ( \
3350     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
3351     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
3352     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
3353     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
3354 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
3355 
3356 /* Just enough of fw_hdr to cover all version info. */
3357 struct fw_h {
3358 	__u8	ver;
3359 	__u8	chip;
3360 	__be16	len512;
3361 	__be32	fw_ver;
3362 	__be32	tp_microcode_ver;
3363 	__u8	intfver_nic;
3364 	__u8	intfver_vnic;
3365 	__u8	intfver_ofld;
3366 	__u8	intfver_ri;
3367 	__u8	intfver_iscsipdu;
3368 	__u8	intfver_iscsi;
3369 	__u8	intfver_fcoepdu;
3370 	__u8	intfver_fcoe;
3371 };
3372 /* Spot check a couple of fields. */
3373 CTASSERT(offsetof(struct fw_h, fw_ver) == offsetof(struct fw_hdr, fw_ver));
3374 CTASSERT(offsetof(struct fw_h, intfver_nic) == offsetof(struct fw_hdr, intfver_nic));
3375 CTASSERT(offsetof(struct fw_h, intfver_fcoe) == offsetof(struct fw_hdr, intfver_fcoe));
3376 
3377 struct fw_info {
3378 	uint8_t chip;
3379 	char *kld_name;
3380 	char *fw_mod_name;
3381 	struct fw_h fw_h;
3382 } fw_info[] = {
3383 	{
3384 		.chip = CHELSIO_T4,
3385 		.kld_name = "t4fw_cfg",
3386 		.fw_mod_name = "t4fw",
3387 		.fw_h = {
3388 			.chip = FW_HDR_CHIP_T4,
3389 			.fw_ver = htobe32(FW_VERSION(T4)),
3390 			.intfver_nic = FW_INTFVER(T4, NIC),
3391 			.intfver_vnic = FW_INTFVER(T4, VNIC),
3392 			.intfver_ofld = FW_INTFVER(T4, OFLD),
3393 			.intfver_ri = FW_INTFVER(T4, RI),
3394 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
3395 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
3396 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
3397 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
3398 		},
3399 	}, {
3400 		.chip = CHELSIO_T5,
3401 		.kld_name = "t5fw_cfg",
3402 		.fw_mod_name = "t5fw",
3403 		.fw_h = {
3404 			.chip = FW_HDR_CHIP_T5,
3405 			.fw_ver = htobe32(FW_VERSION(T5)),
3406 			.intfver_nic = FW_INTFVER(T5, NIC),
3407 			.intfver_vnic = FW_INTFVER(T5, VNIC),
3408 			.intfver_ofld = FW_INTFVER(T5, OFLD),
3409 			.intfver_ri = FW_INTFVER(T5, RI),
3410 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
3411 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
3412 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
3413 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
3414 		},
3415 	}, {
3416 		.chip = CHELSIO_T6,
3417 		.kld_name = "t6fw_cfg",
3418 		.fw_mod_name = "t6fw",
3419 		.fw_h = {
3420 			.chip = FW_HDR_CHIP_T6,
3421 			.fw_ver = htobe32(FW_VERSION(T6)),
3422 			.intfver_nic = FW_INTFVER(T6, NIC),
3423 			.intfver_vnic = FW_INTFVER(T6, VNIC),
3424 			.intfver_ofld = FW_INTFVER(T6, OFLD),
3425 			.intfver_ri = FW_INTFVER(T6, RI),
3426 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3427 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
3428 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3429 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
3430 		},
3431 	}
3432 };
3433 
3434 static struct fw_info *
3435 find_fw_info(int chip)
3436 {
3437 	int i;
3438 
3439 	for (i = 0; i < nitems(fw_info); i++) {
3440 		if (fw_info[i].chip == chip)
3441 			return (&fw_info[i]);
3442 	}
3443 	return (NULL);
3444 }
3445 
3446 /*
3447  * Is the given firmware API compatible with the one the driver was compiled
3448  * with?
3449  */
3450 static int
3451 fw_compatible(const struct fw_h *hdr1, const struct fw_h *hdr2)
3452 {
3453 
3454 	/* short circuit if it's the exact same firmware version */
3455 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
3456 		return (1);
3457 
3458 	/*
3459 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
3460 	 * features that are supported in the driver.
3461 	 */
3462 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
3463 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
3464 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
3465 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
3466 		return (1);
3467 #undef SAME_INTF
3468 
3469 	return (0);
3470 }
3471 
3472 static int
3473 load_fw_module(struct adapter *sc, const struct firmware **dcfg,
3474     const struct firmware **fw)
3475 {
3476 	struct fw_info *fw_info;
3477 
3478 	*dcfg = NULL;
3479 	if (fw != NULL)
3480 		*fw = NULL;
3481 
3482 	fw_info = find_fw_info(chip_id(sc));
3483 	if (fw_info == NULL) {
3484 		device_printf(sc->dev,
3485 		    "unable to look up firmware information for chip %d.\n",
3486 		    chip_id(sc));
3487 		return (EINVAL);
3488 	}
3489 
3490 	*dcfg = firmware_get(fw_info->kld_name);
3491 	if (*dcfg != NULL) {
3492 		if (fw != NULL)
3493 			*fw = firmware_get(fw_info->fw_mod_name);
3494 		return (0);
3495 	}
3496 
3497 	return (ENOENT);
3498 }
3499 
3500 static void
3501 unload_fw_module(struct adapter *sc, const struct firmware *dcfg,
3502     const struct firmware *fw)
3503 {
3504 
3505 	if (fw != NULL)
3506 		firmware_put(fw, FIRMWARE_UNLOAD);
3507 	if (dcfg != NULL)
3508 		firmware_put(dcfg, FIRMWARE_UNLOAD);
3509 }
3510 
3511 /*
3512  * Return values:
3513  * 0 means no firmware install attempted.
3514  * ERESTART means a firmware install was attempted and was successful.
3515  * +ve errno means a firmware install was attempted but failed.
3516  */
3517 static int
3518 install_kld_firmware(struct adapter *sc, struct fw_h *card_fw,
3519     const struct fw_h *drv_fw, const char *reason, int *already)
3520 {
3521 	const struct firmware *cfg, *fw;
3522 	const uint32_t c = be32toh(card_fw->fw_ver);
3523 	uint32_t d, k;
3524 	int rc, fw_install;
3525 	struct fw_h bundled_fw;
3526 	bool load_attempted;
3527 
3528 	cfg = fw = NULL;
3529 	load_attempted = false;
3530 	fw_install = t4_fw_install < 0 ? -t4_fw_install : t4_fw_install;
3531 
3532 	if (reason != NULL)
3533 		goto install;
3534 
3535 	if ((sc->flags & FW_OK) == 0) {
3536 
3537 		if (c == 0xffffffff) {
3538 			reason = "missing";
3539 			goto install;
3540 		}
3541 
3542 		return (0);
3543 	}
3544 
3545 	memcpy(&bundled_fw, drv_fw, sizeof(bundled_fw));
3546 	if (t4_fw_install < 0) {
3547 		rc = load_fw_module(sc, &cfg, &fw);
3548 		if (rc != 0 || fw == NULL) {
3549 			device_printf(sc->dev,
3550 			    "failed to load firmware module: %d. cfg %p, fw %p;"
3551 			    " will use compiled-in firmware version for"
3552 			    "hw.cxgbe.fw_install checks.\n",
3553 			    rc, cfg, fw);
3554 		} else {
3555 			memcpy(&bundled_fw, fw->data, sizeof(bundled_fw));
3556 		}
3557 		load_attempted = true;
3558 	}
3559 	d = be32toh(bundled_fw.fw_ver);
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(nbm);
4015 		COPY_CAPS(link);
4016 		COPY_CAPS(switch);
4017 		COPY_CAPS(nic);
4018 		fallback = false;
4019 		goto retry;
4020 	}
4021 #undef COPY_CAPS
4022 	return (rc);
4023 }
4024 
4025 /*
4026  * Retrieve parameters that are needed (or nice to have) very early.
4027  */
4028 static int
4029 get_params__pre_init(struct adapter *sc)
4030 {
4031 	int rc;
4032 	uint32_t param[2], val[2];
4033 
4034 	t4_get_version_info(sc);
4035 
4036 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
4037 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
4038 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
4039 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
4040 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
4041 
4042 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
4043 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
4044 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
4045 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
4046 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
4047 
4048 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
4049 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
4050 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
4051 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
4052 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
4053 
4054 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
4055 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
4056 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
4057 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
4058 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
4059 
4060 	param[0] = FW_PARAM_DEV(PORTVEC);
4061 	param[1] = FW_PARAM_DEV(CCLK);
4062 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4063 	if (rc != 0) {
4064 		device_printf(sc->dev,
4065 		    "failed to query parameters (pre_init): %d.\n", rc);
4066 		return (rc);
4067 	}
4068 
4069 	sc->params.portvec = val[0];
4070 	sc->params.nports = bitcount32(val[0]);
4071 	sc->params.vpd.cclk = val[1];
4072 
4073 	/* Read device log parameters. */
4074 	rc = -t4_init_devlog_params(sc, 1);
4075 	if (rc == 0)
4076 		fixup_devlog_params(sc);
4077 	else {
4078 		device_printf(sc->dev,
4079 		    "failed to get devlog parameters: %d.\n", rc);
4080 		rc = 0;	/* devlog isn't critical for device operation */
4081 	}
4082 
4083 	return (rc);
4084 }
4085 
4086 /*
4087  * Any params that need to be set before FW_INITIALIZE.
4088  */
4089 static int
4090 set_params__pre_init(struct adapter *sc)
4091 {
4092 	int rc = 0;
4093 	uint32_t param, val;
4094 
4095 	if (chip_id(sc) >= CHELSIO_T6) {
4096 		param = FW_PARAM_DEV(HPFILTER_REGION_SUPPORT);
4097 		val = 1;
4098 		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4099 		/* firmwares < 1.20.1.0 do not have this param. */
4100 		if (rc == FW_EINVAL && sc->params.fw_vers <
4101 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4102 		    V_FW_HDR_FW_VER_MICRO(1) | V_FW_HDR_FW_VER_BUILD(0))) {
4103 			rc = 0;
4104 		}
4105 		if (rc != 0) {
4106 			device_printf(sc->dev,
4107 			    "failed to enable high priority filters :%d.\n",
4108 			    rc);
4109 		}
4110 	}
4111 
4112 	/* Enable opaque VIIDs with firmwares that support it. */
4113 	param = FW_PARAM_DEV(OPAQUE_VIID_SMT_EXTN);
4114 	val = 1;
4115 	rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4116 	if (rc == 0 && val == 1)
4117 		sc->params.viid_smt_extn_support = true;
4118 	else
4119 		sc->params.viid_smt_extn_support = false;
4120 
4121 	return (rc);
4122 }
4123 
4124 /*
4125  * Retrieve various parameters that are of interest to the driver.  The device
4126  * has been initialized by the firmware at this point.
4127  */
4128 static int
4129 get_params__post_init(struct adapter *sc)
4130 {
4131 	int rc;
4132 	uint32_t param[7], val[7];
4133 	struct fw_caps_config_cmd caps;
4134 
4135 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
4136 	param[1] = FW_PARAM_PFVF(EQ_START);
4137 	param[2] = FW_PARAM_PFVF(FILTER_START);
4138 	param[3] = FW_PARAM_PFVF(FILTER_END);
4139 	param[4] = FW_PARAM_PFVF(L2T_START);
4140 	param[5] = FW_PARAM_PFVF(L2T_END);
4141 	param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4142 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
4143 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
4144 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
4145 	if (rc != 0) {
4146 		device_printf(sc->dev,
4147 		    "failed to query parameters (post_init): %d.\n", rc);
4148 		return (rc);
4149 	}
4150 
4151 	sc->sge.iq_start = val[0];
4152 	sc->sge.eq_start = val[1];
4153 	if ((int)val[3] > (int)val[2]) {
4154 		sc->tids.ftid_base = val[2];
4155 		sc->tids.ftid_end = val[3];
4156 		sc->tids.nftids = val[3] - val[2] + 1;
4157 	}
4158 	sc->vres.l2t.start = val[4];
4159 	sc->vres.l2t.size = val[5] - val[4] + 1;
4160 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
4161 	    ("%s: L2 table size (%u) larger than expected (%u)",
4162 	    __func__, sc->vres.l2t.size, L2T_SIZE));
4163 	sc->params.core_vdd = val[6];
4164 
4165 	if (chip_id(sc) >= CHELSIO_T6) {
4166 
4167 		sc->tids.tid_base = t4_read_reg(sc,
4168 		    A_LE_DB_ACTIVE_TABLE_START_INDEX);
4169 
4170 		param[0] = FW_PARAM_PFVF(HPFILTER_START);
4171 		param[1] = FW_PARAM_PFVF(HPFILTER_END);
4172 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4173 		if (rc != 0) {
4174 			device_printf(sc->dev,
4175 			   "failed to query hpfilter parameters: %d.\n", rc);
4176 			return (rc);
4177 		}
4178 		if ((int)val[1] > (int)val[0]) {
4179 			sc->tids.hpftid_base = val[0];
4180 			sc->tids.hpftid_end = val[1];
4181 			sc->tids.nhpftids = val[1] - val[0] + 1;
4182 
4183 			/*
4184 			 * These should go off if the layout changes and the
4185 			 * driver needs to catch up.
4186 			 */
4187 			MPASS(sc->tids.hpftid_base == 0);
4188 			MPASS(sc->tids.tid_base == sc->tids.nhpftids);
4189 		}
4190 	}
4191 
4192 	/*
4193 	 * MPSBGMAP is queried separately because only recent firmwares support
4194 	 * it as a parameter and we don't want the compound query above to fail
4195 	 * on older firmwares.
4196 	 */
4197 	param[0] = FW_PARAM_DEV(MPSBGMAP);
4198 	val[0] = 0;
4199 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4200 	if (rc == 0)
4201 		sc->params.mps_bg_map = val[0];
4202 	else
4203 		sc->params.mps_bg_map = 0;
4204 
4205 	/*
4206 	 * Determine whether the firmware supports the filter2 work request.
4207 	 * This is queried separately for the same reason as MPSBGMAP above.
4208 	 */
4209 	param[0] = FW_PARAM_DEV(FILTER2_WR);
4210 	val[0] = 0;
4211 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4212 	if (rc == 0)
4213 		sc->params.filter2_wr_support = val[0] != 0;
4214 	else
4215 		sc->params.filter2_wr_support = 0;
4216 
4217 	/*
4218 	 * Find out whether we're allowed to use the ULPTX MEMWRITE DSGL.
4219 	 * This is queried separately for the same reason as other params above.
4220 	 */
4221 	param[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
4222 	val[0] = 0;
4223 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4224 	if (rc == 0)
4225 		sc->params.ulptx_memwrite_dsgl = val[0] != 0;
4226 	else
4227 		sc->params.ulptx_memwrite_dsgl = false;
4228 
4229 	/* get capabilites */
4230 	bzero(&caps, sizeof(caps));
4231 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4232 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
4233 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
4234 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
4235 	if (rc != 0) {
4236 		device_printf(sc->dev,
4237 		    "failed to get card capabilities: %d.\n", rc);
4238 		return (rc);
4239 	}
4240 
4241 #define READ_CAPS(x) do { \
4242 	sc->x = htobe16(caps.x); \
4243 } while (0)
4244 	READ_CAPS(nbmcaps);
4245 	READ_CAPS(linkcaps);
4246 	READ_CAPS(switchcaps);
4247 	READ_CAPS(niccaps);
4248 	READ_CAPS(toecaps);
4249 	READ_CAPS(rdmacaps);
4250 	READ_CAPS(cryptocaps);
4251 	READ_CAPS(iscsicaps);
4252 	READ_CAPS(fcoecaps);
4253 
4254 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_HASHFILTER) {
4255 		MPASS(chip_id(sc) > CHELSIO_T4);
4256 		MPASS(sc->toecaps == 0);
4257 		sc->toecaps = 0;
4258 
4259 		param[0] = FW_PARAM_DEV(NTID);
4260 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4261 		if (rc != 0) {
4262 			device_printf(sc->dev,
4263 			    "failed to query HASHFILTER parameters: %d.\n", rc);
4264 			return (rc);
4265 		}
4266 		sc->tids.ntids = val[0];
4267 		if (sc->params.fw_vers <
4268 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4269 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4270 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4271 			sc->tids.ntids -= sc->tids.nhpftids;
4272 		}
4273 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4274 		sc->params.hash_filter = 1;
4275 	}
4276 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
4277 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
4278 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
4279 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4280 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
4281 		if (rc != 0) {
4282 			device_printf(sc->dev,
4283 			    "failed to query NIC parameters: %d.\n", rc);
4284 			return (rc);
4285 		}
4286 		if ((int)val[1] > (int)val[0]) {
4287 			sc->tids.etid_base = val[0];
4288 			sc->tids.etid_end = val[1];
4289 			sc->tids.netids = val[1] - val[0] + 1;
4290 			sc->params.eo_wr_cred = val[2];
4291 			sc->params.ethoffload = 1;
4292 		}
4293 	}
4294 	if (sc->toecaps) {
4295 		/* query offload-related parameters */
4296 		param[0] = FW_PARAM_DEV(NTID);
4297 		param[1] = FW_PARAM_PFVF(SERVER_START);
4298 		param[2] = FW_PARAM_PFVF(SERVER_END);
4299 		param[3] = FW_PARAM_PFVF(TDDP_START);
4300 		param[4] = FW_PARAM_PFVF(TDDP_END);
4301 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4302 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4303 		if (rc != 0) {
4304 			device_printf(sc->dev,
4305 			    "failed to query TOE parameters: %d.\n", rc);
4306 			return (rc);
4307 		}
4308 		sc->tids.ntids = val[0];
4309 		if (sc->params.fw_vers <
4310 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4311 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4312 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4313 			sc->tids.ntids -= sc->tids.nhpftids;
4314 		}
4315 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4316 		if ((int)val[2] > (int)val[1]) {
4317 			sc->tids.stid_base = val[1];
4318 			sc->tids.nstids = val[2] - val[1] + 1;
4319 		}
4320 		sc->vres.ddp.start = val[3];
4321 		sc->vres.ddp.size = val[4] - val[3] + 1;
4322 		sc->params.ofldq_wr_cred = val[5];
4323 		sc->params.offload = 1;
4324 	} else {
4325 		/*
4326 		 * The firmware attempts memfree TOE configuration for -SO cards
4327 		 * and will report toecaps=0 if it runs out of resources (this
4328 		 * depends on the config file).  It may not report 0 for other
4329 		 * capabilities dependent on the TOE in this case.  Set them to
4330 		 * 0 here so that the driver doesn't bother tracking resources
4331 		 * that will never be used.
4332 		 */
4333 		sc->iscsicaps = 0;
4334 		sc->rdmacaps = 0;
4335 	}
4336 	if (sc->rdmacaps) {
4337 		param[0] = FW_PARAM_PFVF(STAG_START);
4338 		param[1] = FW_PARAM_PFVF(STAG_END);
4339 		param[2] = FW_PARAM_PFVF(RQ_START);
4340 		param[3] = FW_PARAM_PFVF(RQ_END);
4341 		param[4] = FW_PARAM_PFVF(PBL_START);
4342 		param[5] = FW_PARAM_PFVF(PBL_END);
4343 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4344 		if (rc != 0) {
4345 			device_printf(sc->dev,
4346 			    "failed to query RDMA parameters(1): %d.\n", rc);
4347 			return (rc);
4348 		}
4349 		sc->vres.stag.start = val[0];
4350 		sc->vres.stag.size = val[1] - val[0] + 1;
4351 		sc->vres.rq.start = val[2];
4352 		sc->vres.rq.size = val[3] - val[2] + 1;
4353 		sc->vres.pbl.start = val[4];
4354 		sc->vres.pbl.size = val[5] - val[4] + 1;
4355 
4356 		param[0] = FW_PARAM_PFVF(SQRQ_START);
4357 		param[1] = FW_PARAM_PFVF(SQRQ_END);
4358 		param[2] = FW_PARAM_PFVF(CQ_START);
4359 		param[3] = FW_PARAM_PFVF(CQ_END);
4360 		param[4] = FW_PARAM_PFVF(OCQ_START);
4361 		param[5] = FW_PARAM_PFVF(OCQ_END);
4362 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4363 		if (rc != 0) {
4364 			device_printf(sc->dev,
4365 			    "failed to query RDMA parameters(2): %d.\n", rc);
4366 			return (rc);
4367 		}
4368 		sc->vres.qp.start = val[0];
4369 		sc->vres.qp.size = val[1] - val[0] + 1;
4370 		sc->vres.cq.start = val[2];
4371 		sc->vres.cq.size = val[3] - val[2] + 1;
4372 		sc->vres.ocq.start = val[4];
4373 		sc->vres.ocq.size = val[5] - val[4] + 1;
4374 
4375 		param[0] = FW_PARAM_PFVF(SRQ_START);
4376 		param[1] = FW_PARAM_PFVF(SRQ_END);
4377 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
4378 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4379 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
4380 		if (rc != 0) {
4381 			device_printf(sc->dev,
4382 			    "failed to query RDMA parameters(3): %d.\n", rc);
4383 			return (rc);
4384 		}
4385 		sc->vres.srq.start = val[0];
4386 		sc->vres.srq.size = val[1] - val[0] + 1;
4387 		sc->params.max_ordird_qp = val[2];
4388 		sc->params.max_ird_adapter = val[3];
4389 	}
4390 	if (sc->iscsicaps) {
4391 		param[0] = FW_PARAM_PFVF(ISCSI_START);
4392 		param[1] = FW_PARAM_PFVF(ISCSI_END);
4393 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4394 		if (rc != 0) {
4395 			device_printf(sc->dev,
4396 			    "failed to query iSCSI parameters: %d.\n", rc);
4397 			return (rc);
4398 		}
4399 		sc->vres.iscsi.start = val[0];
4400 		sc->vres.iscsi.size = val[1] - val[0] + 1;
4401 	}
4402 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
4403 		param[0] = FW_PARAM_PFVF(TLS_START);
4404 		param[1] = FW_PARAM_PFVF(TLS_END);
4405 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4406 		if (rc != 0) {
4407 			device_printf(sc->dev,
4408 			    "failed to query TLS parameters: %d.\n", rc);
4409 			return (rc);
4410 		}
4411 		sc->vres.key.start = val[0];
4412 		sc->vres.key.size = val[1] - val[0] + 1;
4413 	}
4414 
4415 	t4_init_sge_params(sc);
4416 
4417 	/*
4418 	 * We've got the params we wanted to query via the firmware.  Now grab
4419 	 * some others directly from the chip.
4420 	 */
4421 	rc = t4_read_chip_settings(sc);
4422 
4423 	return (rc);
4424 }
4425 
4426 static int
4427 set_params__post_init(struct adapter *sc)
4428 {
4429 	uint32_t param, val;
4430 #ifdef TCP_OFFLOAD
4431 	int i, v, shift;
4432 #endif
4433 
4434 	/* ask for encapsulated CPLs */
4435 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4436 	val = 1;
4437 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4438 
4439 	/* Enable 32b port caps if the firmware supports it. */
4440 	param = FW_PARAM_PFVF(PORT_CAPS32);
4441 	val = 1;
4442 	if (t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val) == 0)
4443 		sc->params.port_caps32 = 1;
4444 
4445 	/* Let filter + maskhash steer to a part of the VI's RSS region. */
4446 	val = 1 << (G_MASKSIZE(t4_read_reg(sc, A_TP_RSS_CONFIG_TNL)) - 1);
4447 	t4_set_reg_field(sc, A_TP_RSS_CONFIG_TNL, V_MASKFILTER(M_MASKFILTER),
4448 	    V_MASKFILTER(val - 1));
4449 
4450 #ifdef TCP_OFFLOAD
4451 	/*
4452 	 * Override the TOE timers with user provided tunables.  This is not the
4453 	 * recommended way to change the timers (the firmware config file is) so
4454 	 * these tunables are not documented.
4455 	 *
4456 	 * All the timer tunables are in microseconds.
4457 	 */
4458 	if (t4_toe_keepalive_idle != 0) {
4459 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
4460 		v &= M_KEEPALIVEIDLE;
4461 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
4462 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
4463 	}
4464 	if (t4_toe_keepalive_interval != 0) {
4465 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
4466 		v &= M_KEEPALIVEINTVL;
4467 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
4468 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
4469 	}
4470 	if (t4_toe_keepalive_count != 0) {
4471 		v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
4472 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4473 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
4474 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
4475 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
4476 	}
4477 	if (t4_toe_rexmt_min != 0) {
4478 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
4479 		v &= M_RXTMIN;
4480 		t4_set_reg_field(sc, A_TP_RXT_MIN,
4481 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
4482 	}
4483 	if (t4_toe_rexmt_max != 0) {
4484 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
4485 		v &= M_RXTMAX;
4486 		t4_set_reg_field(sc, A_TP_RXT_MAX,
4487 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
4488 	}
4489 	if (t4_toe_rexmt_count != 0) {
4490 		v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
4491 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4492 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
4493 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
4494 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
4495 	}
4496 	for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
4497 		if (t4_toe_rexmt_backoff[i] != -1) {
4498 			v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
4499 			shift = (i & 3) << 3;
4500 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
4501 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
4502 		}
4503 	}
4504 #endif
4505 	return (0);
4506 }
4507 
4508 #undef FW_PARAM_PFVF
4509 #undef FW_PARAM_DEV
4510 
4511 static void
4512 t4_set_desc(struct adapter *sc)
4513 {
4514 	char buf[128];
4515 	struct adapter_params *p = &sc->params;
4516 
4517 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
4518 
4519 	device_set_desc_copy(sc->dev, buf);
4520 }
4521 
4522 static inline void
4523 ifmedia_add4(struct ifmedia *ifm, int m)
4524 {
4525 
4526 	ifmedia_add(ifm, m, 0, NULL);
4527 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE, 0, NULL);
4528 	ifmedia_add(ifm, m | IFM_ETH_RXPAUSE, 0, NULL);
4529 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE, 0, NULL);
4530 }
4531 
4532 /*
4533  * This is the selected media, which is not quite the same as the active media.
4534  * The media line in ifconfig is "media: Ethernet selected (active)" if selected
4535  * and active are not the same, and "media: Ethernet selected" otherwise.
4536  */
4537 static void
4538 set_current_media(struct port_info *pi)
4539 {
4540 	struct link_config *lc;
4541 	struct ifmedia *ifm;
4542 	int mword;
4543 	u_int speed;
4544 
4545 	PORT_LOCK_ASSERT_OWNED(pi);
4546 
4547 	/* Leave current media alone if it's already set to IFM_NONE. */
4548 	ifm = &pi->media;
4549 	if (ifm->ifm_cur != NULL &&
4550 	    IFM_SUBTYPE(ifm->ifm_cur->ifm_media) == IFM_NONE)
4551 		return;
4552 
4553 	lc = &pi->link_cfg;
4554 	if (lc->requested_aneg != AUTONEG_DISABLE &&
4555 	    lc->supported & FW_PORT_CAP32_ANEG) {
4556 		ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
4557 		return;
4558 	}
4559 	mword = IFM_ETHER | IFM_FDX;
4560 	if (lc->requested_fc & PAUSE_TX)
4561 		mword |= IFM_ETH_TXPAUSE;
4562 	if (lc->requested_fc & PAUSE_RX)
4563 		mword |= IFM_ETH_RXPAUSE;
4564 	if (lc->requested_speed == 0)
4565 		speed = port_top_speed(pi) * 1000;	/* Gbps -> Mbps */
4566 	else
4567 		speed = lc->requested_speed;
4568 	mword |= port_mword(pi, speed_to_fwcap(speed));
4569 	ifmedia_set(ifm, mword);
4570 }
4571 
4572 /*
4573  * Returns true if the ifmedia list for the port cannot change.
4574  */
4575 static bool
4576 fixed_ifmedia(struct port_info *pi)
4577 {
4578 
4579 	return (pi->port_type == FW_PORT_TYPE_BT_SGMII ||
4580 	    pi->port_type == FW_PORT_TYPE_BT_XFI ||
4581 	    pi->port_type == FW_PORT_TYPE_BT_XAUI ||
4582 	    pi->port_type == FW_PORT_TYPE_KX4 ||
4583 	    pi->port_type == FW_PORT_TYPE_KX ||
4584 	    pi->port_type == FW_PORT_TYPE_KR ||
4585 	    pi->port_type == FW_PORT_TYPE_BP_AP ||
4586 	    pi->port_type == FW_PORT_TYPE_BP4_AP ||
4587 	    pi->port_type == FW_PORT_TYPE_BP40_BA ||
4588 	    pi->port_type == FW_PORT_TYPE_KR4_100G ||
4589 	    pi->port_type == FW_PORT_TYPE_KR_SFP28 ||
4590 	    pi->port_type == FW_PORT_TYPE_KR_XLAUI);
4591 }
4592 
4593 static void
4594 build_medialist(struct port_info *pi)
4595 {
4596 	uint32_t ss, speed;
4597 	int unknown, mword, bit;
4598 	struct link_config *lc;
4599 	struct ifmedia *ifm;
4600 
4601 	PORT_LOCK_ASSERT_OWNED(pi);
4602 
4603 	if (pi->flags & FIXED_IFMEDIA)
4604 		return;
4605 
4606 	/*
4607 	 * Rebuild the ifmedia list.
4608 	 */
4609 	ifm = &pi->media;
4610 	ifmedia_removeall(ifm);
4611 	lc = &pi->link_cfg;
4612 	ss = G_FW_PORT_CAP32_SPEED(lc->supported); /* Supported Speeds */
4613 	if (__predict_false(ss == 0)) {	/* not supposed to happen. */
4614 		MPASS(ss != 0);
4615 no_media:
4616 		MPASS(LIST_EMPTY(&ifm->ifm_list));
4617 		ifmedia_add(ifm, IFM_ETHER | IFM_NONE, 0, NULL);
4618 		ifmedia_set(ifm, IFM_ETHER | IFM_NONE);
4619 		return;
4620 	}
4621 
4622 	unknown = 0;
4623 	for (bit = S_FW_PORT_CAP32_SPEED; bit < fls(ss); bit++) {
4624 		speed = 1 << bit;
4625 		MPASS(speed & M_FW_PORT_CAP32_SPEED);
4626 		if (ss & speed) {
4627 			mword = port_mword(pi, speed);
4628 			if (mword == IFM_NONE) {
4629 				goto no_media;
4630 			} else if (mword == IFM_UNKNOWN)
4631 				unknown++;
4632 			else
4633 				ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | mword);
4634 		}
4635 	}
4636 	if (unknown > 0) /* Add one unknown for all unknown media types. */
4637 		ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | IFM_UNKNOWN);
4638 	if (lc->supported & FW_PORT_CAP32_ANEG)
4639 		ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
4640 
4641 	set_current_media(pi);
4642 }
4643 
4644 /*
4645  * Initialize the requested fields in the link config based on driver tunables.
4646  */
4647 static void
4648 init_link_config(struct port_info *pi)
4649 {
4650 	struct link_config *lc = &pi->link_cfg;
4651 
4652 	PORT_LOCK_ASSERT_OWNED(pi);
4653 
4654 	lc->requested_speed = 0;
4655 
4656 	if (t4_autoneg == 0)
4657 		lc->requested_aneg = AUTONEG_DISABLE;
4658 	else if (t4_autoneg == 1)
4659 		lc->requested_aneg = AUTONEG_ENABLE;
4660 	else
4661 		lc->requested_aneg = AUTONEG_AUTO;
4662 
4663 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX |
4664 	    PAUSE_AUTONEG);
4665 
4666 	if (t4_fec == -1 || t4_fec & FEC_AUTO)
4667 		lc->requested_fec = FEC_AUTO;
4668 	else {
4669 		lc->requested_fec = FEC_NONE;
4670 		if (t4_fec & FEC_RS)
4671 			lc->requested_fec |= FEC_RS;
4672 		if (t4_fec & FEC_BASER_RS)
4673 			lc->requested_fec |= FEC_BASER_RS;
4674 	}
4675 }
4676 
4677 /*
4678  * Makes sure that all requested settings comply with what's supported by the
4679  * port.  Returns the number of settings that were invalid and had to be fixed.
4680  */
4681 static int
4682 fixup_link_config(struct port_info *pi)
4683 {
4684 	int n = 0;
4685 	struct link_config *lc = &pi->link_cfg;
4686 	uint32_t fwspeed;
4687 
4688 	PORT_LOCK_ASSERT_OWNED(pi);
4689 
4690 	/* Speed (when not autonegotiating) */
4691 	if (lc->requested_speed != 0) {
4692 		fwspeed = speed_to_fwcap(lc->requested_speed);
4693 		if ((fwspeed & lc->supported) == 0) {
4694 			n++;
4695 			lc->requested_speed = 0;
4696 		}
4697 	}
4698 
4699 	/* Link autonegotiation */
4700 	MPASS(lc->requested_aneg == AUTONEG_ENABLE ||
4701 	    lc->requested_aneg == AUTONEG_DISABLE ||
4702 	    lc->requested_aneg == AUTONEG_AUTO);
4703 	if (lc->requested_aneg == AUTONEG_ENABLE &&
4704 	    !(lc->supported & FW_PORT_CAP32_ANEG)) {
4705 		n++;
4706 		lc->requested_aneg = AUTONEG_AUTO;
4707 	}
4708 
4709 	/* Flow control */
4710 	MPASS((lc->requested_fc & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG)) == 0);
4711 	if (lc->requested_fc & PAUSE_TX &&
4712 	    !(lc->supported & FW_PORT_CAP32_FC_TX)) {
4713 		n++;
4714 		lc->requested_fc &= ~PAUSE_TX;
4715 	}
4716 	if (lc->requested_fc & PAUSE_RX &&
4717 	    !(lc->supported & FW_PORT_CAP32_FC_RX)) {
4718 		n++;
4719 		lc->requested_fc &= ~PAUSE_RX;
4720 	}
4721 	if (!(lc->requested_fc & PAUSE_AUTONEG) &&
4722 	    !(lc->supported & FW_PORT_CAP32_FORCE_PAUSE)) {
4723 		n++;
4724 		lc->requested_fc |= PAUSE_AUTONEG;
4725 	}
4726 
4727 	/* FEC */
4728 	if ((lc->requested_fec & FEC_RS &&
4729 	    !(lc->supported & FW_PORT_CAP32_FEC_RS)) ||
4730 	    (lc->requested_fec & FEC_BASER_RS &&
4731 	    !(lc->supported & FW_PORT_CAP32_FEC_BASER_RS))) {
4732 		n++;
4733 		lc->requested_fec = FEC_AUTO;
4734 	}
4735 
4736 	return (n);
4737 }
4738 
4739 /*
4740  * Apply the requested L1 settings, which are expected to be valid, to the
4741  * hardware.
4742  */
4743 static int
4744 apply_link_config(struct port_info *pi)
4745 {
4746 	struct adapter *sc = pi->adapter;
4747 	struct link_config *lc = &pi->link_cfg;
4748 	int rc;
4749 
4750 #ifdef INVARIANTS
4751 	ASSERT_SYNCHRONIZED_OP(sc);
4752 	PORT_LOCK_ASSERT_OWNED(pi);
4753 
4754 	if (lc->requested_aneg == AUTONEG_ENABLE)
4755 		MPASS(lc->supported & FW_PORT_CAP32_ANEG);
4756 	if (!(lc->requested_fc & PAUSE_AUTONEG))
4757 		MPASS(lc->supported & FW_PORT_CAP32_FORCE_PAUSE);
4758 	if (lc->requested_fc & PAUSE_TX)
4759 		MPASS(lc->supported & FW_PORT_CAP32_FC_TX);
4760 	if (lc->requested_fc & PAUSE_RX)
4761 		MPASS(lc->supported & FW_PORT_CAP32_FC_RX);
4762 	if (lc->requested_fec & FEC_RS)
4763 		MPASS(lc->supported & FW_PORT_CAP32_FEC_RS);
4764 	if (lc->requested_fec & FEC_BASER_RS)
4765 		MPASS(lc->supported & FW_PORT_CAP32_FEC_BASER_RS);
4766 #endif
4767 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
4768 	if (rc != 0) {
4769 		/* Don't complain if the VF driver gets back an EPERM. */
4770 		if (!(sc->flags & IS_VF) || rc != FW_EPERM)
4771 			device_printf(pi->dev, "l1cfg failed: %d\n", rc);
4772 	} else {
4773 		/*
4774 		 * An L1_CFG will almost always result in a link-change event if
4775 		 * the link is up, and the driver will refresh the actual
4776 		 * fec/fc/etc. when the notification is processed.  If the link
4777 		 * is down then the actual settings are meaningless.
4778 		 *
4779 		 * This takes care of the case where a change in the L1 settings
4780 		 * may not result in a notification.
4781 		 */
4782 		if (lc->link_ok && !(lc->requested_fc & PAUSE_AUTONEG))
4783 			lc->fc = lc->requested_fc & (PAUSE_TX | PAUSE_RX);
4784 	}
4785 	return (rc);
4786 }
4787 
4788 #define FW_MAC_EXACT_CHUNK	7
4789 
4790 /*
4791  * Program the port's XGMAC based on parameters in ifnet.  The caller also
4792  * indicates which parameters should be programmed (the rest are left alone).
4793  */
4794 int
4795 update_mac_settings(struct ifnet *ifp, int flags)
4796 {
4797 	int rc = 0;
4798 	struct vi_info *vi = ifp->if_softc;
4799 	struct port_info *pi = vi->pi;
4800 	struct adapter *sc = pi->adapter;
4801 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4802 
4803 	ASSERT_SYNCHRONIZED_OP(sc);
4804 	KASSERT(flags, ("%s: not told what to update.", __func__));
4805 
4806 	if (flags & XGMAC_MTU)
4807 		mtu = ifp->if_mtu;
4808 
4809 	if (flags & XGMAC_PROMISC)
4810 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4811 
4812 	if (flags & XGMAC_ALLMULTI)
4813 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4814 
4815 	if (flags & XGMAC_VLANEX)
4816 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4817 
4818 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4819 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4820 		    allmulti, 1, vlanex, false);
4821 		if (rc) {
4822 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4823 			    rc);
4824 			return (rc);
4825 		}
4826 	}
4827 
4828 	if (flags & XGMAC_UCADDR) {
4829 		uint8_t ucaddr[ETHER_ADDR_LEN];
4830 
4831 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4832 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4833 		    ucaddr, true, &vi->smt_idx);
4834 		if (rc < 0) {
4835 			rc = -rc;
4836 			if_printf(ifp, "change_mac failed: %d\n", rc);
4837 			return (rc);
4838 		} else {
4839 			vi->xact_addr_filt = rc;
4840 			rc = 0;
4841 		}
4842 	}
4843 
4844 	if (flags & XGMAC_MCADDRS) {
4845 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4846 		int del = 1;
4847 		uint64_t hash = 0;
4848 		struct ifmultiaddr *ifma;
4849 		int i = 0, j;
4850 
4851 		if_maddr_rlock(ifp);
4852 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4853 			if (ifma->ifma_addr->sa_family != AF_LINK)
4854 				continue;
4855 			mcaddr[i] =
4856 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4857 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4858 			i++;
4859 
4860 			if (i == FW_MAC_EXACT_CHUNK) {
4861 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4862 				    del, i, mcaddr, NULL, &hash, 0);
4863 				if (rc < 0) {
4864 					rc = -rc;
4865 					for (j = 0; j < i; j++) {
4866 						if_printf(ifp,
4867 						    "failed to add mc address"
4868 						    " %02x:%02x:%02x:"
4869 						    "%02x:%02x:%02x rc=%d\n",
4870 						    mcaddr[j][0], mcaddr[j][1],
4871 						    mcaddr[j][2], mcaddr[j][3],
4872 						    mcaddr[j][4], mcaddr[j][5],
4873 						    rc);
4874 					}
4875 					goto mcfail;
4876 				}
4877 				del = 0;
4878 				i = 0;
4879 			}
4880 		}
4881 		if (i > 0) {
4882 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4883 			    mcaddr, NULL, &hash, 0);
4884 			if (rc < 0) {
4885 				rc = -rc;
4886 				for (j = 0; j < i; j++) {
4887 					if_printf(ifp,
4888 					    "failed to add mc address"
4889 					    " %02x:%02x:%02x:"
4890 					    "%02x:%02x:%02x rc=%d\n",
4891 					    mcaddr[j][0], mcaddr[j][1],
4892 					    mcaddr[j][2], mcaddr[j][3],
4893 					    mcaddr[j][4], mcaddr[j][5],
4894 					    rc);
4895 				}
4896 				goto mcfail;
4897 			}
4898 		}
4899 
4900 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4901 		if (rc != 0)
4902 			if_printf(ifp, "failed to set mc address hash: %d", rc);
4903 mcfail:
4904 		if_maddr_runlock(ifp);
4905 	}
4906 
4907 	return (rc);
4908 }
4909 
4910 /*
4911  * {begin|end}_synchronized_op must be called from the same thread.
4912  */
4913 int
4914 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4915     char *wmesg)
4916 {
4917 	int rc, pri;
4918 
4919 #ifdef WITNESS
4920 	/* the caller thinks it's ok to sleep, but is it really? */
4921 	if (flags & SLEEP_OK)
4922 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4923 		    "begin_synchronized_op");
4924 #endif
4925 
4926 	if (INTR_OK)
4927 		pri = PCATCH;
4928 	else
4929 		pri = 0;
4930 
4931 	ADAPTER_LOCK(sc);
4932 	for (;;) {
4933 
4934 		if (vi && IS_DOOMED(vi)) {
4935 			rc = ENXIO;
4936 			goto done;
4937 		}
4938 
4939 		if (!IS_BUSY(sc)) {
4940 			rc = 0;
4941 			break;
4942 		}
4943 
4944 		if (!(flags & SLEEP_OK)) {
4945 			rc = EBUSY;
4946 			goto done;
4947 		}
4948 
4949 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4950 			rc = EINTR;
4951 			goto done;
4952 		}
4953 	}
4954 
4955 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4956 	SET_BUSY(sc);
4957 #ifdef INVARIANTS
4958 	sc->last_op = wmesg;
4959 	sc->last_op_thr = curthread;
4960 	sc->last_op_flags = flags;
4961 #endif
4962 
4963 done:
4964 	if (!(flags & HOLD_LOCK) || rc)
4965 		ADAPTER_UNLOCK(sc);
4966 
4967 	return (rc);
4968 }
4969 
4970 /*
4971  * Tell if_ioctl and if_init that the VI is going away.  This is
4972  * special variant of begin_synchronized_op and must be paired with a
4973  * call to end_synchronized_op.
4974  */
4975 void
4976 doom_vi(struct adapter *sc, struct vi_info *vi)
4977 {
4978 
4979 	ADAPTER_LOCK(sc);
4980 	SET_DOOMED(vi);
4981 	wakeup(&sc->flags);
4982 	while (IS_BUSY(sc))
4983 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4984 	SET_BUSY(sc);
4985 #ifdef INVARIANTS
4986 	sc->last_op = "t4detach";
4987 	sc->last_op_thr = curthread;
4988 	sc->last_op_flags = 0;
4989 #endif
4990 	ADAPTER_UNLOCK(sc);
4991 }
4992 
4993 /*
4994  * {begin|end}_synchronized_op must be called from the same thread.
4995  */
4996 void
4997 end_synchronized_op(struct adapter *sc, int flags)
4998 {
4999 
5000 	if (flags & LOCK_HELD)
5001 		ADAPTER_LOCK_ASSERT_OWNED(sc);
5002 	else
5003 		ADAPTER_LOCK(sc);
5004 
5005 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
5006 	CLR_BUSY(sc);
5007 	wakeup(&sc->flags);
5008 	ADAPTER_UNLOCK(sc);
5009 }
5010 
5011 static int
5012 cxgbe_init_synchronized(struct vi_info *vi)
5013 {
5014 	struct port_info *pi = vi->pi;
5015 	struct adapter *sc = pi->adapter;
5016 	struct ifnet *ifp = vi->ifp;
5017 	int rc = 0, i;
5018 	struct sge_txq *txq;
5019 
5020 	ASSERT_SYNCHRONIZED_OP(sc);
5021 
5022 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
5023 		return (0);	/* already running */
5024 
5025 	if (!(sc->flags & FULL_INIT_DONE) &&
5026 	    ((rc = adapter_full_init(sc)) != 0))
5027 		return (rc);	/* error message displayed already */
5028 
5029 	if (!(vi->flags & VI_INIT_DONE) &&
5030 	    ((rc = vi_full_init(vi)) != 0))
5031 		return (rc); /* error message displayed already */
5032 
5033 	rc = update_mac_settings(ifp, XGMAC_ALL);
5034 	if (rc)
5035 		goto done;	/* error message displayed already */
5036 
5037 	PORT_LOCK(pi);
5038 	if (pi->up_vis == 0) {
5039 		t4_update_port_info(pi);
5040 		fixup_link_config(pi);
5041 		build_medialist(pi);
5042 		apply_link_config(pi);
5043 	}
5044 
5045 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
5046 	if (rc != 0) {
5047 		if_printf(ifp, "enable_vi failed: %d\n", rc);
5048 		PORT_UNLOCK(pi);
5049 		goto done;
5050 	}
5051 
5052 	/*
5053 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
5054 	 * if this changes.
5055 	 */
5056 
5057 	for_each_txq(vi, i, txq) {
5058 		TXQ_LOCK(txq);
5059 		txq->eq.flags |= EQ_ENABLED;
5060 		TXQ_UNLOCK(txq);
5061 	}
5062 
5063 	/*
5064 	 * The first iq of the first port to come up is used for tracing.
5065 	 */
5066 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
5067 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
5068 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
5069 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
5070 		    V_QUEUENUMBER(sc->traceq));
5071 		pi->flags |= HAS_TRACEQ;
5072 	}
5073 
5074 	/* all ok */
5075 	pi->up_vis++;
5076 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
5077 
5078 	if (pi->nvi > 1 || sc->flags & IS_VF)
5079 		callout_reset(&vi->tick, hz, vi_tick, vi);
5080 	else
5081 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
5082 	if (pi->link_cfg.link_ok)
5083 		t4_os_link_changed(pi);
5084 	PORT_UNLOCK(pi);
5085 done:
5086 	if (rc != 0)
5087 		cxgbe_uninit_synchronized(vi);
5088 
5089 	return (rc);
5090 }
5091 
5092 /*
5093  * Idempotent.
5094  */
5095 static int
5096 cxgbe_uninit_synchronized(struct vi_info *vi)
5097 {
5098 	struct port_info *pi = vi->pi;
5099 	struct adapter *sc = pi->adapter;
5100 	struct ifnet *ifp = vi->ifp;
5101 	int rc, i;
5102 	struct sge_txq *txq;
5103 
5104 	ASSERT_SYNCHRONIZED_OP(sc);
5105 
5106 	if (!(vi->flags & VI_INIT_DONE)) {
5107 		if (__predict_false(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
5108 			KASSERT(0, ("uninited VI is running"));
5109 			if_printf(ifp, "uninited VI with running ifnet.  "
5110 			    "vi->flags 0x%016lx, if_flags 0x%08x, "
5111 			    "if_drv_flags 0x%08x\n", vi->flags, ifp->if_flags,
5112 			    ifp->if_drv_flags);
5113 		}
5114 		return (0);
5115 	}
5116 
5117 	/*
5118 	 * Disable the VI so that all its data in either direction is discarded
5119 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
5120 	 * tick) intact as the TP can deliver negative advice or data that it's
5121 	 * holding in its RAM (for an offloaded connection) even after the VI is
5122 	 * disabled.
5123 	 */
5124 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
5125 	if (rc) {
5126 		if_printf(ifp, "disable_vi failed: %d\n", rc);
5127 		return (rc);
5128 	}
5129 
5130 	for_each_txq(vi, i, txq) {
5131 		TXQ_LOCK(txq);
5132 		txq->eq.flags &= ~EQ_ENABLED;
5133 		TXQ_UNLOCK(txq);
5134 	}
5135 
5136 	PORT_LOCK(pi);
5137 	if (pi->nvi > 1 || sc->flags & IS_VF)
5138 		callout_stop(&vi->tick);
5139 	else
5140 		callout_stop(&pi->tick);
5141 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
5142 		PORT_UNLOCK(pi);
5143 		return (0);
5144 	}
5145 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
5146 	pi->up_vis--;
5147 	if (pi->up_vis > 0) {
5148 		PORT_UNLOCK(pi);
5149 		return (0);
5150 	}
5151 
5152 	pi->link_cfg.link_ok = false;
5153 	pi->link_cfg.speed = 0;
5154 	pi->link_cfg.link_down_rc = 255;
5155 	t4_os_link_changed(pi);
5156 	PORT_UNLOCK(pi);
5157 
5158 	return (0);
5159 }
5160 
5161 /*
5162  * It is ok for this function to fail midway and return right away.  t4_detach
5163  * will walk the entire sc->irq list and clean up whatever is valid.
5164  */
5165 int
5166 t4_setup_intr_handlers(struct adapter *sc)
5167 {
5168 	int rc, rid, p, q, v;
5169 	char s[8];
5170 	struct irq *irq;
5171 	struct port_info *pi;
5172 	struct vi_info *vi;
5173 	struct sge *sge = &sc->sge;
5174 	struct sge_rxq *rxq;
5175 #ifdef TCP_OFFLOAD
5176 	struct sge_ofld_rxq *ofld_rxq;
5177 #endif
5178 #ifdef DEV_NETMAP
5179 	struct sge_nm_rxq *nm_rxq;
5180 #endif
5181 #ifdef RSS
5182 	int nbuckets = rss_getnumbuckets();
5183 #endif
5184 
5185 	/*
5186 	 * Setup interrupts.
5187 	 */
5188 	irq = &sc->irq[0];
5189 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
5190 	if (forwarding_intr_to_fwq(sc))
5191 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
5192 
5193 	/* Multiple interrupts. */
5194 	if (sc->flags & IS_VF)
5195 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
5196 		    ("%s: too few intr.", __func__));
5197 	else
5198 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
5199 		    ("%s: too few intr.", __func__));
5200 
5201 	/* The first one is always error intr on PFs */
5202 	if (!(sc->flags & IS_VF)) {
5203 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
5204 		if (rc != 0)
5205 			return (rc);
5206 		irq++;
5207 		rid++;
5208 	}
5209 
5210 	/* The second one is always the firmware event queue (first on VFs) */
5211 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
5212 	if (rc != 0)
5213 		return (rc);
5214 	irq++;
5215 	rid++;
5216 
5217 	for_each_port(sc, p) {
5218 		pi = sc->port[p];
5219 		for_each_vi(pi, v, vi) {
5220 			vi->first_intr = rid - 1;
5221 
5222 			if (vi->nnmrxq > 0) {
5223 				int n = max(vi->nrxq, vi->nnmrxq);
5224 
5225 				rxq = &sge->rxq[vi->first_rxq];
5226 #ifdef DEV_NETMAP
5227 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
5228 #endif
5229 				for (q = 0; q < n; q++) {
5230 					snprintf(s, sizeof(s), "%x%c%x", p,
5231 					    'a' + v, q);
5232 					if (q < vi->nrxq)
5233 						irq->rxq = rxq++;
5234 #ifdef DEV_NETMAP
5235 					if (q < vi->nnmrxq)
5236 						irq->nm_rxq = nm_rxq++;
5237 
5238 					if (irq->nm_rxq != NULL &&
5239 					    irq->rxq == NULL) {
5240 						/* Netmap rx only */
5241 						rc = t4_alloc_irq(sc, irq, rid,
5242 						    t4_nm_intr, irq->nm_rxq, s);
5243 					}
5244 					if (irq->nm_rxq != NULL &&
5245 					    irq->rxq != NULL) {
5246 						/* NIC and Netmap rx */
5247 						rc = t4_alloc_irq(sc, irq, rid,
5248 						    t4_vi_intr, irq, s);
5249 					}
5250 #endif
5251 					if (irq->rxq != NULL &&
5252 					    irq->nm_rxq == NULL) {
5253 						/* NIC rx only */
5254 						rc = t4_alloc_irq(sc, irq, rid,
5255 						    t4_intr, irq->rxq, s);
5256 					}
5257 					if (rc != 0)
5258 						return (rc);
5259 #ifdef RSS
5260 					if (q < vi->nrxq) {
5261 						bus_bind_intr(sc->dev, irq->res,
5262 						    rss_getcpu(q % nbuckets));
5263 					}
5264 #endif
5265 					irq++;
5266 					rid++;
5267 					vi->nintr++;
5268 				}
5269 			} else {
5270 				for_each_rxq(vi, q, rxq) {
5271 					snprintf(s, sizeof(s), "%x%c%x", p,
5272 					    'a' + v, q);
5273 					rc = t4_alloc_irq(sc, irq, rid,
5274 					    t4_intr, rxq, s);
5275 					if (rc != 0)
5276 						return (rc);
5277 #ifdef RSS
5278 					bus_bind_intr(sc->dev, irq->res,
5279 					    rss_getcpu(q % nbuckets));
5280 #endif
5281 					irq++;
5282 					rid++;
5283 					vi->nintr++;
5284 				}
5285 			}
5286 #ifdef TCP_OFFLOAD
5287 			for_each_ofld_rxq(vi, q, ofld_rxq) {
5288 				snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
5289 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
5290 				    ofld_rxq, s);
5291 				if (rc != 0)
5292 					return (rc);
5293 				irq++;
5294 				rid++;
5295 				vi->nintr++;
5296 			}
5297 #endif
5298 		}
5299 	}
5300 	MPASS(irq == &sc->irq[sc->intr_count]);
5301 
5302 	return (0);
5303 }
5304 
5305 int
5306 adapter_full_init(struct adapter *sc)
5307 {
5308 	int rc, i;
5309 #ifdef RSS
5310 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5311 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5312 #endif
5313 
5314 	ASSERT_SYNCHRONIZED_OP(sc);
5315 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5316 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
5317 	    ("%s: FULL_INIT_DONE already", __func__));
5318 
5319 	/*
5320 	 * queues that belong to the adapter (not any particular port).
5321 	 */
5322 	rc = t4_setup_adapter_queues(sc);
5323 	if (rc != 0)
5324 		goto done;
5325 
5326 	for (i = 0; i < nitems(sc->tq); i++) {
5327 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
5328 		    taskqueue_thread_enqueue, &sc->tq[i]);
5329 		if (sc->tq[i] == NULL) {
5330 			device_printf(sc->dev,
5331 			    "failed to allocate task queue %d\n", i);
5332 			rc = ENOMEM;
5333 			goto done;
5334 		}
5335 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
5336 		    device_get_nameunit(sc->dev), i);
5337 	}
5338 #ifdef RSS
5339 	MPASS(RSS_KEYSIZE == 40);
5340 	rss_getkey((void *)&raw_rss_key[0]);
5341 	for (i = 0; i < nitems(rss_key); i++) {
5342 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
5343 	}
5344 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
5345 #endif
5346 
5347 	if (!(sc->flags & IS_VF))
5348 		t4_intr_enable(sc);
5349 	sc->flags |= FULL_INIT_DONE;
5350 done:
5351 	if (rc != 0)
5352 		adapter_full_uninit(sc);
5353 
5354 	return (rc);
5355 }
5356 
5357 int
5358 adapter_full_uninit(struct adapter *sc)
5359 {
5360 	int i;
5361 
5362 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5363 
5364 	t4_teardown_adapter_queues(sc);
5365 
5366 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
5367 		taskqueue_free(sc->tq[i]);
5368 		sc->tq[i] = NULL;
5369 	}
5370 
5371 	sc->flags &= ~FULL_INIT_DONE;
5372 
5373 	return (0);
5374 }
5375 
5376 #ifdef RSS
5377 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
5378     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
5379     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
5380     RSS_HASHTYPE_RSS_UDP_IPV6)
5381 
5382 /* Translates kernel hash types to hardware. */
5383 static int
5384 hashconfig_to_hashen(int hashconfig)
5385 {
5386 	int hashen = 0;
5387 
5388 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
5389 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
5390 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
5391 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
5392 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
5393 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5394 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5395 	}
5396 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
5397 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5398 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5399 	}
5400 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
5401 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5402 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
5403 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5404 
5405 	return (hashen);
5406 }
5407 
5408 /* Translates hardware hash types to kernel. */
5409 static int
5410 hashen_to_hashconfig(int hashen)
5411 {
5412 	int hashconfig = 0;
5413 
5414 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
5415 		/*
5416 		 * If UDP hashing was enabled it must have been enabled for
5417 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
5418 		 * enabling any 4-tuple hash is nonsense configuration.
5419 		 */
5420 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5421 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
5422 
5423 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5424 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
5425 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5426 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
5427 	}
5428 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5429 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
5430 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5431 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
5432 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
5433 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
5434 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
5435 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
5436 
5437 	return (hashconfig);
5438 }
5439 #endif
5440 
5441 int
5442 vi_full_init(struct vi_info *vi)
5443 {
5444 	struct adapter *sc = vi->pi->adapter;
5445 	struct ifnet *ifp = vi->ifp;
5446 	uint16_t *rss;
5447 	struct sge_rxq *rxq;
5448 	int rc, i, j;
5449 #ifdef RSS
5450 	int nbuckets = rss_getnumbuckets();
5451 	int hashconfig = rss_gethashconfig();
5452 	int extra;
5453 #endif
5454 
5455 	ASSERT_SYNCHRONIZED_OP(sc);
5456 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
5457 	    ("%s: VI_INIT_DONE already", __func__));
5458 
5459 	sysctl_ctx_init(&vi->ctx);
5460 	vi->flags |= VI_SYSCTL_CTX;
5461 
5462 	/*
5463 	 * Allocate tx/rx/fl queues for this VI.
5464 	 */
5465 	rc = t4_setup_vi_queues(vi);
5466 	if (rc != 0)
5467 		goto done;	/* error message displayed already */
5468 
5469 	/*
5470 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
5471 	 */
5472 	if (vi->nrxq > vi->rss_size) {
5473 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
5474 		    "some queues will never receive traffic.\n", vi->nrxq,
5475 		    vi->rss_size);
5476 	} else if (vi->rss_size % vi->nrxq) {
5477 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
5478 		    "expect uneven traffic distribution.\n", vi->nrxq,
5479 		    vi->rss_size);
5480 	}
5481 #ifdef RSS
5482 	if (vi->nrxq != nbuckets) {
5483 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
5484 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
5485 	}
5486 #endif
5487 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
5488 	for (i = 0; i < vi->rss_size;) {
5489 #ifdef RSS
5490 		j = rss_get_indirection_to_bucket(i);
5491 		j %= vi->nrxq;
5492 		rxq = &sc->sge.rxq[vi->first_rxq + j];
5493 		rss[i++] = rxq->iq.abs_id;
5494 #else
5495 		for_each_rxq(vi, j, rxq) {
5496 			rss[i++] = rxq->iq.abs_id;
5497 			if (i == vi->rss_size)
5498 				break;
5499 		}
5500 #endif
5501 	}
5502 
5503 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
5504 	    vi->rss_size);
5505 	if (rc != 0) {
5506 		free(rss, M_CXGBE);
5507 		if_printf(ifp, "rss_config failed: %d\n", rc);
5508 		goto done;
5509 	}
5510 
5511 #ifdef RSS
5512 	vi->hashen = hashconfig_to_hashen(hashconfig);
5513 
5514 	/*
5515 	 * We may have had to enable some hashes even though the global config
5516 	 * wants them disabled.  This is a potential problem that must be
5517 	 * reported to the user.
5518 	 */
5519 	extra = hashen_to_hashconfig(vi->hashen) ^ hashconfig;
5520 
5521 	/*
5522 	 * If we consider only the supported hash types, then the enabled hashes
5523 	 * are a superset of the requested hashes.  In other words, there cannot
5524 	 * be any supported hash that was requested but not enabled, but there
5525 	 * can be hashes that were not requested but had to be enabled.
5526 	 */
5527 	extra &= SUPPORTED_RSS_HASHTYPES;
5528 	MPASS((extra & hashconfig) == 0);
5529 
5530 	if (extra) {
5531 		if_printf(ifp,
5532 		    "global RSS config (0x%x) cannot be accommodated.\n",
5533 		    hashconfig);
5534 	}
5535 	if (extra & RSS_HASHTYPE_RSS_IPV4)
5536 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
5537 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
5538 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
5539 	if (extra & RSS_HASHTYPE_RSS_IPV6)
5540 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
5541 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
5542 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
5543 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
5544 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
5545 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
5546 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
5547 #else
5548 	vi->hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
5549 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
5550 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5551 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
5552 #endif
5553 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, vi->hashen, rss[0], 0, 0);
5554 	if (rc != 0) {
5555 		free(rss, M_CXGBE);
5556 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
5557 		goto done;
5558 	}
5559 
5560 	vi->rss = rss;
5561 	vi->flags |= VI_INIT_DONE;
5562 done:
5563 	if (rc != 0)
5564 		vi_full_uninit(vi);
5565 
5566 	return (rc);
5567 }
5568 
5569 /*
5570  * Idempotent.
5571  */
5572 int
5573 vi_full_uninit(struct vi_info *vi)
5574 {
5575 	struct port_info *pi = vi->pi;
5576 	struct adapter *sc = pi->adapter;
5577 	int i;
5578 	struct sge_rxq *rxq;
5579 	struct sge_txq *txq;
5580 #ifdef TCP_OFFLOAD
5581 	struct sge_ofld_rxq *ofld_rxq;
5582 #endif
5583 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
5584 	struct sge_wrq *ofld_txq;
5585 #endif
5586 
5587 	if (vi->flags & VI_INIT_DONE) {
5588 
5589 		/* Need to quiesce queues.  */
5590 
5591 		/* XXX: Only for the first VI? */
5592 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
5593 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
5594 
5595 		for_each_txq(vi, i, txq) {
5596 			quiesce_txq(sc, txq);
5597 		}
5598 
5599 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
5600 		for_each_ofld_txq(vi, i, ofld_txq) {
5601 			quiesce_wrq(sc, ofld_txq);
5602 		}
5603 #endif
5604 
5605 		for_each_rxq(vi, i, rxq) {
5606 			quiesce_iq(sc, &rxq->iq);
5607 			quiesce_fl(sc, &rxq->fl);
5608 		}
5609 
5610 #ifdef TCP_OFFLOAD
5611 		for_each_ofld_rxq(vi, i, ofld_rxq) {
5612 			quiesce_iq(sc, &ofld_rxq->iq);
5613 			quiesce_fl(sc, &ofld_rxq->fl);
5614 		}
5615 #endif
5616 		free(vi->rss, M_CXGBE);
5617 		free(vi->nm_rss, M_CXGBE);
5618 	}
5619 
5620 	t4_teardown_vi_queues(vi);
5621 	vi->flags &= ~VI_INIT_DONE;
5622 
5623 	return (0);
5624 }
5625 
5626 static void
5627 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
5628 {
5629 	struct sge_eq *eq = &txq->eq;
5630 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
5631 
5632 	(void) sc;	/* unused */
5633 
5634 #ifdef INVARIANTS
5635 	TXQ_LOCK(txq);
5636 	MPASS((eq->flags & EQ_ENABLED) == 0);
5637 	TXQ_UNLOCK(txq);
5638 #endif
5639 
5640 	/* Wait for the mp_ring to empty. */
5641 	while (!mp_ring_is_idle(txq->r)) {
5642 		mp_ring_check_drainage(txq->r, 0);
5643 		pause("rquiesce", 1);
5644 	}
5645 
5646 	/* Then wait for the hardware to finish. */
5647 	while (spg->cidx != htobe16(eq->pidx))
5648 		pause("equiesce", 1);
5649 
5650 	/* Finally, wait for the driver to reclaim all descriptors. */
5651 	while (eq->cidx != eq->pidx)
5652 		pause("dquiesce", 1);
5653 }
5654 
5655 static void
5656 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
5657 {
5658 
5659 	/* XXXTX */
5660 }
5661 
5662 static void
5663 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
5664 {
5665 	(void) sc;	/* unused */
5666 
5667 	/* Synchronize with the interrupt handler */
5668 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
5669 		pause("iqfree", 1);
5670 }
5671 
5672 static void
5673 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
5674 {
5675 	mtx_lock(&sc->sfl_lock);
5676 	FL_LOCK(fl);
5677 	fl->flags |= FL_DOOMED;
5678 	FL_UNLOCK(fl);
5679 	callout_stop(&sc->sfl_callout);
5680 	mtx_unlock(&sc->sfl_lock);
5681 
5682 	KASSERT((fl->flags & FL_STARVING) == 0,
5683 	    ("%s: still starving", __func__));
5684 }
5685 
5686 static int
5687 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
5688     driver_intr_t *handler, void *arg, char *name)
5689 {
5690 	int rc;
5691 
5692 	irq->rid = rid;
5693 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
5694 	    RF_SHAREABLE | RF_ACTIVE);
5695 	if (irq->res == NULL) {
5696 		device_printf(sc->dev,
5697 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
5698 		return (ENOMEM);
5699 	}
5700 
5701 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
5702 	    NULL, handler, arg, &irq->tag);
5703 	if (rc != 0) {
5704 		device_printf(sc->dev,
5705 		    "failed to setup interrupt for rid %d, name %s: %d\n",
5706 		    rid, name, rc);
5707 	} else if (name)
5708 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
5709 
5710 	return (rc);
5711 }
5712 
5713 static int
5714 t4_free_irq(struct adapter *sc, struct irq *irq)
5715 {
5716 	if (irq->tag)
5717 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
5718 	if (irq->res)
5719 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
5720 
5721 	bzero(irq, sizeof(*irq));
5722 
5723 	return (0);
5724 }
5725 
5726 static void
5727 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
5728 {
5729 
5730 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
5731 	t4_get_regs(sc, buf, regs->len);
5732 }
5733 
5734 #define	A_PL_INDIR_CMD	0x1f8
5735 
5736 #define	S_PL_AUTOINC	31
5737 #define	M_PL_AUTOINC	0x1U
5738 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
5739 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
5740 
5741 #define	S_PL_VFID	20
5742 #define	M_PL_VFID	0xffU
5743 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
5744 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
5745 
5746 #define	S_PL_ADDR	0
5747 #define	M_PL_ADDR	0xfffffU
5748 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
5749 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
5750 
5751 #define	A_PL_INDIR_DATA	0x1fc
5752 
5753 static uint64_t
5754 read_vf_stat(struct adapter *sc, u_int vin, int reg)
5755 {
5756 	u32 stats[2];
5757 
5758 	mtx_assert(&sc->reg_lock, MA_OWNED);
5759 	if (sc->flags & IS_VF) {
5760 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
5761 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
5762 	} else {
5763 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5764 		    V_PL_VFID(vin) | V_PL_ADDR(VF_MPS_REG(reg)));
5765 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
5766 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
5767 	}
5768 	return (((uint64_t)stats[1]) << 32 | stats[0]);
5769 }
5770 
5771 static void
5772 t4_get_vi_stats(struct adapter *sc, u_int vin, struct fw_vi_stats_vf *stats)
5773 {
5774 
5775 #define GET_STAT(name) \
5776 	read_vf_stat(sc, vin, A_MPS_VF_STAT_##name##_L)
5777 
5778 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
5779 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
5780 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
5781 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
5782 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
5783 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
5784 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
5785 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
5786 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
5787 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
5788 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
5789 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
5790 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
5791 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
5792 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
5793 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
5794 
5795 #undef GET_STAT
5796 }
5797 
5798 static void
5799 t4_clr_vi_stats(struct adapter *sc, u_int vin)
5800 {
5801 	int reg;
5802 
5803 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) | V_PL_VFID(vin) |
5804 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
5805 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
5806 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
5807 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
5808 }
5809 
5810 static void
5811 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
5812 {
5813 	struct timeval tv;
5814 	const struct timeval interval = {0, 250000};	/* 250ms */
5815 
5816 	if (!(vi->flags & VI_INIT_DONE))
5817 		return;
5818 
5819 	getmicrotime(&tv);
5820 	timevalsub(&tv, &interval);
5821 	if (timevalcmp(&tv, &vi->last_refreshed, <))
5822 		return;
5823 
5824 	mtx_lock(&sc->reg_lock);
5825 	t4_get_vi_stats(sc, vi->vin, &vi->stats);
5826 	getmicrotime(&vi->last_refreshed);
5827 	mtx_unlock(&sc->reg_lock);
5828 }
5829 
5830 static void
5831 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5832 {
5833 	u_int i, v, tnl_cong_drops, bg_map;
5834 	struct timeval tv;
5835 	const struct timeval interval = {0, 250000};	/* 250ms */
5836 
5837 	getmicrotime(&tv);
5838 	timevalsub(&tv, &interval);
5839 	if (timevalcmp(&tv, &pi->last_refreshed, <))
5840 		return;
5841 
5842 	tnl_cong_drops = 0;
5843 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5844 	bg_map = pi->mps_bg_map;
5845 	while (bg_map) {
5846 		i = ffs(bg_map) - 1;
5847 		mtx_lock(&sc->reg_lock);
5848 		t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5849 		    A_TP_MIB_TNL_CNG_DROP_0 + i);
5850 		mtx_unlock(&sc->reg_lock);
5851 		tnl_cong_drops += v;
5852 		bg_map &= ~(1 << i);
5853 	}
5854 	pi->tnl_cong_drops = tnl_cong_drops;
5855 	getmicrotime(&pi->last_refreshed);
5856 }
5857 
5858 static void
5859 cxgbe_tick(void *arg)
5860 {
5861 	struct port_info *pi = arg;
5862 	struct adapter *sc = pi->adapter;
5863 
5864 	PORT_LOCK_ASSERT_OWNED(pi);
5865 	cxgbe_refresh_stats(sc, pi);
5866 
5867 	callout_schedule(&pi->tick, hz);
5868 }
5869 
5870 void
5871 vi_tick(void *arg)
5872 {
5873 	struct vi_info *vi = arg;
5874 	struct adapter *sc = vi->pi->adapter;
5875 
5876 	vi_refresh_stats(sc, vi);
5877 
5878 	callout_schedule(&vi->tick, hz);
5879 }
5880 
5881 /*
5882  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5883  */
5884 static char *caps_decoder[] = {
5885 	"\20\001IPMI\002NCSI",				/* 0: NBM */
5886 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
5887 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
5888 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
5889 	    "\006HASHFILTER\007ETHOFLD",
5890 	"\20\001TOE",					/* 4: TOE */
5891 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
5892 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
5893 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5894 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5895 	    "\007T10DIF"
5896 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5897 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
5898 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
5899 		    "\004PO_INITIATOR\005PO_TARGET",
5900 };
5901 
5902 void
5903 t4_sysctls(struct adapter *sc)
5904 {
5905 	struct sysctl_ctx_list *ctx;
5906 	struct sysctl_oid *oid;
5907 	struct sysctl_oid_list *children, *c0;
5908 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5909 
5910 	ctx = device_get_sysctl_ctx(sc->dev);
5911 
5912 	/*
5913 	 * dev.t4nex.X.
5914 	 */
5915 	oid = device_get_sysctl_tree(sc->dev);
5916 	c0 = children = SYSCTL_CHILDREN(oid);
5917 
5918 	sc->sc_do_rxcopy = 1;
5919 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5920 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5921 
5922 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5923 	    sc->params.nports, "# of ports");
5924 
5925 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5926 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, (uintptr_t)&sc->doorbells,
5927 	    sysctl_bitfield_8b, "A", "available doorbells");
5928 
5929 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5930 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
5931 
5932 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5933 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5934 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5935 	    "interrupt holdoff timer values (us)");
5936 
5937 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5938 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5939 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5940 	    "interrupt holdoff packet counter values");
5941 
5942 	t4_sge_sysctls(sc, ctx, children);
5943 
5944 	sc->lro_timeout = 100;
5945 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5946 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5947 
5948 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5949 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
5950 
5951 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5952 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5953 
5954 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5955 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5956 
5957 	if (sc->flags & IS_VF)
5958 		return;
5959 
5960 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5961 	    NULL, chip_rev(sc), "chip hardware revision");
5962 
5963 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5964 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5965 
5966 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5967 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5968 
5969 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5970 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5971 
5972 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5973 	    CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5974 
5975 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5976 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5977 
5978 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5979 	    sc->er_version, 0, "expansion ROM version");
5980 
5981 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5982 	    sc->bs_version, 0, "bootstrap firmware version");
5983 
5984 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5985 	    NULL, sc->params.scfg_vers, "serial config version");
5986 
5987 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5988 	    NULL, sc->params.vpd_vers, "VPD version");
5989 
5990 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5991 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5992 
5993 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5994 	    sc->cfcsum, "config file checksum");
5995 
5996 #define SYSCTL_CAP(name, n, text) \
5997 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5998 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], (uintptr_t)&sc->name, \
5999 	    sysctl_bitfield_16b, "A", "available " text " capabilities")
6000 
6001 	SYSCTL_CAP(nbmcaps, 0, "NBM");
6002 	SYSCTL_CAP(linkcaps, 1, "link");
6003 	SYSCTL_CAP(switchcaps, 2, "switch");
6004 	SYSCTL_CAP(niccaps, 3, "NIC");
6005 	SYSCTL_CAP(toecaps, 4, "TCP offload");
6006 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
6007 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
6008 	SYSCTL_CAP(cryptocaps, 7, "crypto");
6009 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
6010 #undef SYSCTL_CAP
6011 
6012 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
6013 	    NULL, sc->tids.nftids, "number of filters");
6014 
6015 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
6016 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
6017 	    "chip temperature (in Celsius)");
6018 
6019 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "loadavg", CTLTYPE_STRING |
6020 	    CTLFLAG_RD, sc, 0, sysctl_loadavg, "A",
6021 	    "microprocessor load averages (debug firmwares only)");
6022 
6023 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
6024 	    &sc->params.core_vdd, 0, "core Vdd (in mV)");
6025 
6026 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "local_cpus",
6027 	    CTLTYPE_STRING | CTLFLAG_RD, sc, LOCAL_CPUS,
6028 	    sysctl_cpus, "A", "local CPUs");
6029 
6030 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "intr_cpus",
6031 	    CTLTYPE_STRING | CTLFLAG_RD, sc, INTR_CPUS,
6032 	    sysctl_cpus, "A", "preferred CPUs for interrupts");
6033 
6034 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "swintr", CTLFLAG_RW,
6035 	    &sc->swintr, 0, "software triggered interrupts");
6036 
6037 	/*
6038 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
6039 	 */
6040 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
6041 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
6042 	    "logs and miscellaneous information");
6043 	children = SYSCTL_CHILDREN(oid);
6044 
6045 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
6046 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6047 	    sysctl_cctrl, "A", "congestion control");
6048 
6049 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
6050 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6051 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
6052 
6053 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
6054 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
6055 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
6056 
6057 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
6058 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
6059 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
6060 
6061 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
6062 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
6063 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
6064 
6065 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
6066 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
6067 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
6068 
6069 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
6070 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
6071 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
6072 
6073 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
6074 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_cim_la,
6075 	    "A", "CIM logic analyzer");
6076 
6077 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
6078 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6079 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
6080 
6081 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
6082 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
6083 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
6084 
6085 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
6086 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
6087 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
6088 
6089 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
6090 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
6091 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
6092 
6093 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
6094 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
6095 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
6096 
6097 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
6098 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
6099 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
6100 
6101 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
6102 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
6103 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
6104 
6105 	if (chip_id(sc) > CHELSIO_T4) {
6106 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
6107 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
6108 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
6109 
6110 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
6111 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
6112 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
6113 	}
6114 
6115 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
6116 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6117 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
6118 
6119 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
6120 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6121 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
6122 
6123 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
6124 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6125 	    sysctl_cpl_stats, "A", "CPL statistics");
6126 
6127 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
6128 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6129 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
6130 
6131 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
6132 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6133 	    sysctl_devlog, "A", "firmware's device log");
6134 
6135 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
6136 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6137 	    sysctl_fcoe_stats, "A", "FCoE statistics");
6138 
6139 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
6140 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6141 	    sysctl_hw_sched, "A", "hardware scheduler ");
6142 
6143 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
6144 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6145 	    sysctl_l2t, "A", "hardware L2 table");
6146 
6147 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "smt",
6148 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6149 	    sysctl_smt, "A", "hardware source MAC table");
6150 
6151 #ifdef INET6
6152 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "clip",
6153 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6154 	    sysctl_clip, "A", "active CLIP table entries");
6155 #endif
6156 
6157 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
6158 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6159 	    sysctl_lb_stats, "A", "loopback statistics");
6160 
6161 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
6162 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6163 	    sysctl_meminfo, "A", "memory regions");
6164 
6165 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
6166 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6167 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
6168 	    "A", "MPS TCAM entries");
6169 
6170 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
6171 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6172 	    sysctl_path_mtus, "A", "path MTUs");
6173 
6174 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
6175 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6176 	    sysctl_pm_stats, "A", "PM statistics");
6177 
6178 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
6179 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6180 	    sysctl_rdma_stats, "A", "RDMA statistics");
6181 
6182 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
6183 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6184 	    sysctl_tcp_stats, "A", "TCP statistics");
6185 
6186 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
6187 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6188 	    sysctl_tids, "A", "TID information");
6189 
6190 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
6191 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6192 	    sysctl_tp_err_stats, "A", "TP error statistics");
6193 
6194 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
6195 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
6196 	    "TP logic analyzer event capture mask");
6197 
6198 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
6199 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6200 	    sysctl_tp_la, "A", "TP logic analyzer");
6201 
6202 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
6203 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6204 	    sysctl_tx_rate, "A", "Tx rate");
6205 
6206 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
6207 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6208 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
6209 
6210 	if (chip_id(sc) >= CHELSIO_T5) {
6211 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
6212 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6213 		    sysctl_wcwr_stats, "A", "write combined work requests");
6214 	}
6215 
6216 #ifdef TCP_OFFLOAD
6217 	if (is_offload(sc)) {
6218 		int i;
6219 		char s[4];
6220 
6221 		/*
6222 		 * dev.t4nex.X.toe.
6223 		 */
6224 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
6225 		    NULL, "TOE parameters");
6226 		children = SYSCTL_CHILDREN(oid);
6227 
6228 		sc->tt.cong_algorithm = -1;
6229 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
6230 		    CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
6231 		    "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
6232 		    "3 = highspeed)");
6233 
6234 		sc->tt.sndbuf = 256 * 1024;
6235 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
6236 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
6237 
6238 		sc->tt.ddp = 0;
6239 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
6240 		    &sc->tt.ddp, 0, "DDP allowed");
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 	} 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 int
9847 t4_os_find_pci_capability(struct adapter *sc, int cap)
9848 {
9849 	int i;
9850 
9851 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9852 }
9853 
9854 int
9855 t4_os_pci_save_state(struct adapter *sc)
9856 {
9857 	device_t dev;
9858 	struct pci_devinfo *dinfo;
9859 
9860 	dev = sc->dev;
9861 	dinfo = device_get_ivars(dev);
9862 
9863 	pci_cfg_save(dev, dinfo, 0);
9864 	return (0);
9865 }
9866 
9867 int
9868 t4_os_pci_restore_state(struct adapter *sc)
9869 {
9870 	device_t dev;
9871 	struct pci_devinfo *dinfo;
9872 
9873 	dev = sc->dev;
9874 	dinfo = device_get_ivars(dev);
9875 
9876 	pci_cfg_restore(dev, dinfo);
9877 	return (0);
9878 }
9879 
9880 void
9881 t4_os_portmod_changed(struct port_info *pi)
9882 {
9883 	struct adapter *sc = pi->adapter;
9884 	struct vi_info *vi;
9885 	struct ifnet *ifp;
9886 	static const char *mod_str[] = {
9887 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9888 	};
9889 
9890 	KASSERT((pi->flags & FIXED_IFMEDIA) == 0,
9891 	    ("%s: port_type %u", __func__, pi->port_type));
9892 
9893 	vi = &pi->vi[0];
9894 	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9895 		PORT_LOCK(pi);
9896 		build_medialist(pi);
9897 		if (pi->mod_type != FW_PORT_MOD_TYPE_NONE) {
9898 			fixup_link_config(pi);
9899 			apply_link_config(pi);
9900 		}
9901 		PORT_UNLOCK(pi);
9902 		end_synchronized_op(sc, LOCK_HELD);
9903 	}
9904 
9905 	ifp = vi->ifp;
9906 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9907 		if_printf(ifp, "transceiver unplugged.\n");
9908 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9909 		if_printf(ifp, "unknown transceiver inserted.\n");
9910 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9911 		if_printf(ifp, "unsupported transceiver inserted.\n");
9912 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9913 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9914 		    port_top_speed(pi), mod_str[pi->mod_type]);
9915 	} else {
9916 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9917 		    pi->mod_type);
9918 	}
9919 }
9920 
9921 void
9922 t4_os_link_changed(struct port_info *pi)
9923 {
9924 	struct vi_info *vi;
9925 	struct ifnet *ifp;
9926 	struct link_config *lc;
9927 	int v;
9928 
9929 	PORT_LOCK_ASSERT_OWNED(pi);
9930 
9931 	for_each_vi(pi, v, vi) {
9932 		ifp = vi->ifp;
9933 		if (ifp == NULL)
9934 			continue;
9935 
9936 		lc = &pi->link_cfg;
9937 		if (lc->link_ok) {
9938 			ifp->if_baudrate = IF_Mbps(lc->speed);
9939 			if_link_state_change(ifp, LINK_STATE_UP);
9940 		} else {
9941 			if_link_state_change(ifp, LINK_STATE_DOWN);
9942 		}
9943 	}
9944 }
9945 
9946 void
9947 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9948 {
9949 	struct adapter *sc;
9950 
9951 	sx_slock(&t4_list_lock);
9952 	SLIST_FOREACH(sc, &t4_list, link) {
9953 		/*
9954 		 * func should not make any assumptions about what state sc is
9955 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9956 		 */
9957 		func(sc, arg);
9958 	}
9959 	sx_sunlock(&t4_list_lock);
9960 }
9961 
9962 static int
9963 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9964     struct thread *td)
9965 {
9966 	int rc;
9967 	struct adapter *sc = dev->si_drv1;
9968 
9969 	rc = priv_check(td, PRIV_DRIVER);
9970 	if (rc != 0)
9971 		return (rc);
9972 
9973 	switch (cmd) {
9974 	case CHELSIO_T4_GETREG: {
9975 		struct t4_reg *edata = (struct t4_reg *)data;
9976 
9977 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9978 			return (EFAULT);
9979 
9980 		if (edata->size == 4)
9981 			edata->val = t4_read_reg(sc, edata->addr);
9982 		else if (edata->size == 8)
9983 			edata->val = t4_read_reg64(sc, edata->addr);
9984 		else
9985 			return (EINVAL);
9986 
9987 		break;
9988 	}
9989 	case CHELSIO_T4_SETREG: {
9990 		struct t4_reg *edata = (struct t4_reg *)data;
9991 
9992 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9993 			return (EFAULT);
9994 
9995 		if (edata->size == 4) {
9996 			if (edata->val & 0xffffffff00000000)
9997 				return (EINVAL);
9998 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9999 		} else if (edata->size == 8)
10000 			t4_write_reg64(sc, edata->addr, edata->val);
10001 		else
10002 			return (EINVAL);
10003 		break;
10004 	}
10005 	case CHELSIO_T4_REGDUMP: {
10006 		struct t4_regdump *regs = (struct t4_regdump *)data;
10007 		int reglen = t4_get_regs_len(sc);
10008 		uint8_t *buf;
10009 
10010 		if (regs->len < reglen) {
10011 			regs->len = reglen; /* hint to the caller */
10012 			return (ENOBUFS);
10013 		}
10014 
10015 		regs->len = reglen;
10016 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
10017 		get_regs(sc, regs, buf);
10018 		rc = copyout(buf, regs->data, reglen);
10019 		free(buf, M_CXGBE);
10020 		break;
10021 	}
10022 	case CHELSIO_T4_GET_FILTER_MODE:
10023 		rc = get_filter_mode(sc, (uint32_t *)data);
10024 		break;
10025 	case CHELSIO_T4_SET_FILTER_MODE:
10026 		rc = set_filter_mode(sc, *(uint32_t *)data);
10027 		break;
10028 	case CHELSIO_T4_GET_FILTER:
10029 		rc = get_filter(sc, (struct t4_filter *)data);
10030 		break;
10031 	case CHELSIO_T4_SET_FILTER:
10032 		rc = set_filter(sc, (struct t4_filter *)data);
10033 		break;
10034 	case CHELSIO_T4_DEL_FILTER:
10035 		rc = del_filter(sc, (struct t4_filter *)data);
10036 		break;
10037 	case CHELSIO_T4_GET_SGE_CONTEXT:
10038 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
10039 		break;
10040 	case CHELSIO_T4_LOAD_FW:
10041 		rc = load_fw(sc, (struct t4_data *)data);
10042 		break;
10043 	case CHELSIO_T4_GET_MEM:
10044 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
10045 		break;
10046 	case CHELSIO_T4_GET_I2C:
10047 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
10048 		break;
10049 	case CHELSIO_T4_CLEAR_STATS: {
10050 		int i, v, bg_map;
10051 		u_int port_id = *(uint32_t *)data;
10052 		struct port_info *pi;
10053 		struct vi_info *vi;
10054 
10055 		if (port_id >= sc->params.nports)
10056 			return (EINVAL);
10057 		pi = sc->port[port_id];
10058 		if (pi == NULL)
10059 			return (EIO);
10060 
10061 		/* MAC stats */
10062 		t4_clr_port_stats(sc, pi->tx_chan);
10063 		pi->tx_parse_error = 0;
10064 		pi->tnl_cong_drops = 0;
10065 		mtx_lock(&sc->reg_lock);
10066 		for_each_vi(pi, v, vi) {
10067 			if (vi->flags & VI_INIT_DONE)
10068 				t4_clr_vi_stats(sc, vi->vin);
10069 		}
10070 		bg_map = pi->mps_bg_map;
10071 		v = 0;	/* reuse */
10072 		while (bg_map) {
10073 			i = ffs(bg_map) - 1;
10074 			t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
10075 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
10076 			bg_map &= ~(1 << i);
10077 		}
10078 		mtx_unlock(&sc->reg_lock);
10079 
10080 		/*
10081 		 * Since this command accepts a port, clear stats for
10082 		 * all VIs on this port.
10083 		 */
10084 		for_each_vi(pi, v, vi) {
10085 			if (vi->flags & VI_INIT_DONE) {
10086 				struct sge_rxq *rxq;
10087 				struct sge_txq *txq;
10088 				struct sge_wrq *wrq;
10089 
10090 				for_each_rxq(vi, i, rxq) {
10091 #if defined(INET) || defined(INET6)
10092 					rxq->lro.lro_queued = 0;
10093 					rxq->lro.lro_flushed = 0;
10094 #endif
10095 					rxq->rxcsum = 0;
10096 					rxq->vlan_extraction = 0;
10097 				}
10098 
10099 				for_each_txq(vi, i, txq) {
10100 					txq->txcsum = 0;
10101 					txq->tso_wrs = 0;
10102 					txq->vlan_insertion = 0;
10103 					txq->imm_wrs = 0;
10104 					txq->sgl_wrs = 0;
10105 					txq->txpkt_wrs = 0;
10106 					txq->txpkts0_wrs = 0;
10107 					txq->txpkts1_wrs = 0;
10108 					txq->txpkts0_pkts = 0;
10109 					txq->txpkts1_pkts = 0;
10110 					txq->raw_wrs = 0;
10111 					mp_ring_reset_stats(txq->r);
10112 				}
10113 
10114 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
10115 				/* nothing to clear for each ofld_rxq */
10116 
10117 				for_each_ofld_txq(vi, i, wrq) {
10118 					wrq->tx_wrs_direct = 0;
10119 					wrq->tx_wrs_copied = 0;
10120 				}
10121 #endif
10122 
10123 				if (IS_MAIN_VI(vi)) {
10124 					wrq = &sc->sge.ctrlq[pi->port_id];
10125 					wrq->tx_wrs_direct = 0;
10126 					wrq->tx_wrs_copied = 0;
10127 				}
10128 			}
10129 		}
10130 		break;
10131 	}
10132 	case CHELSIO_T4_SCHED_CLASS:
10133 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
10134 		break;
10135 	case CHELSIO_T4_SCHED_QUEUE:
10136 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
10137 		break;
10138 	case CHELSIO_T4_GET_TRACER:
10139 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
10140 		break;
10141 	case CHELSIO_T4_SET_TRACER:
10142 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
10143 		break;
10144 	case CHELSIO_T4_LOAD_CFG:
10145 		rc = load_cfg(sc, (struct t4_data *)data);
10146 		break;
10147 	case CHELSIO_T4_LOAD_BOOT:
10148 		rc = load_boot(sc, (struct t4_bootrom *)data);
10149 		break;
10150 	case CHELSIO_T4_LOAD_BOOTCFG:
10151 		rc = load_bootcfg(sc, (struct t4_data *)data);
10152 		break;
10153 	case CHELSIO_T4_CUDBG_DUMP:
10154 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
10155 		break;
10156 	case CHELSIO_T4_SET_OFLD_POLICY:
10157 		rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
10158 		break;
10159 	default:
10160 		rc = ENOTTY;
10161 	}
10162 
10163 	return (rc);
10164 }
10165 
10166 #ifdef TCP_OFFLOAD
10167 static int
10168 toe_capability(struct vi_info *vi, int enable)
10169 {
10170 	int rc;
10171 	struct port_info *pi = vi->pi;
10172 	struct adapter *sc = pi->adapter;
10173 
10174 	ASSERT_SYNCHRONIZED_OP(sc);
10175 
10176 	if (!is_offload(sc))
10177 		return (ENODEV);
10178 
10179 	if (enable) {
10180 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
10181 			/* TOE is already enabled. */
10182 			return (0);
10183 		}
10184 
10185 		/*
10186 		 * We need the port's queues around so that we're able to send
10187 		 * and receive CPLs to/from the TOE even if the ifnet for this
10188 		 * port has never been UP'd administratively.
10189 		 */
10190 		if (!(vi->flags & VI_INIT_DONE)) {
10191 			rc = vi_full_init(vi);
10192 			if (rc)
10193 				return (rc);
10194 		}
10195 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
10196 			rc = vi_full_init(&pi->vi[0]);
10197 			if (rc)
10198 				return (rc);
10199 		}
10200 
10201 		if (isset(&sc->offload_map, pi->port_id)) {
10202 			/* TOE is enabled on another VI of this port. */
10203 			pi->uld_vis++;
10204 			return (0);
10205 		}
10206 
10207 		if (!uld_active(sc, ULD_TOM)) {
10208 			rc = t4_activate_uld(sc, ULD_TOM);
10209 			if (rc == EAGAIN) {
10210 				log(LOG_WARNING,
10211 				    "You must kldload t4_tom.ko before trying "
10212 				    "to enable TOE on a cxgbe interface.\n");
10213 			}
10214 			if (rc != 0)
10215 				return (rc);
10216 			KASSERT(sc->tom_softc != NULL,
10217 			    ("%s: TOM activated but softc NULL", __func__));
10218 			KASSERT(uld_active(sc, ULD_TOM),
10219 			    ("%s: TOM activated but flag not set", __func__));
10220 		}
10221 
10222 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
10223 		if (!uld_active(sc, ULD_IWARP))
10224 			(void) t4_activate_uld(sc, ULD_IWARP);
10225 		if (!uld_active(sc, ULD_ISCSI))
10226 			(void) t4_activate_uld(sc, ULD_ISCSI);
10227 
10228 		pi->uld_vis++;
10229 		setbit(&sc->offload_map, pi->port_id);
10230 	} else {
10231 		pi->uld_vis--;
10232 
10233 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
10234 			return (0);
10235 
10236 		KASSERT(uld_active(sc, ULD_TOM),
10237 		    ("%s: TOM never initialized?", __func__));
10238 		clrbit(&sc->offload_map, pi->port_id);
10239 	}
10240 
10241 	return (0);
10242 }
10243 
10244 /*
10245  * Add an upper layer driver to the global list.
10246  */
10247 int
10248 t4_register_uld(struct uld_info *ui)
10249 {
10250 	int rc = 0;
10251 	struct uld_info *u;
10252 
10253 	sx_xlock(&t4_uld_list_lock);
10254 	SLIST_FOREACH(u, &t4_uld_list, link) {
10255 	    if (u->uld_id == ui->uld_id) {
10256 		    rc = EEXIST;
10257 		    goto done;
10258 	    }
10259 	}
10260 
10261 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
10262 	ui->refcount = 0;
10263 done:
10264 	sx_xunlock(&t4_uld_list_lock);
10265 	return (rc);
10266 }
10267 
10268 int
10269 t4_unregister_uld(struct uld_info *ui)
10270 {
10271 	int rc = EINVAL;
10272 	struct uld_info *u;
10273 
10274 	sx_xlock(&t4_uld_list_lock);
10275 
10276 	SLIST_FOREACH(u, &t4_uld_list, link) {
10277 	    if (u == ui) {
10278 		    if (ui->refcount > 0) {
10279 			    rc = EBUSY;
10280 			    goto done;
10281 		    }
10282 
10283 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
10284 		    rc = 0;
10285 		    goto done;
10286 	    }
10287 	}
10288 done:
10289 	sx_xunlock(&t4_uld_list_lock);
10290 	return (rc);
10291 }
10292 
10293 int
10294 t4_activate_uld(struct adapter *sc, int id)
10295 {
10296 	int rc;
10297 	struct uld_info *ui;
10298 
10299 	ASSERT_SYNCHRONIZED_OP(sc);
10300 
10301 	if (id < 0 || id > ULD_MAX)
10302 		return (EINVAL);
10303 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
10304 
10305 	sx_slock(&t4_uld_list_lock);
10306 
10307 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10308 		if (ui->uld_id == id) {
10309 			if (!(sc->flags & FULL_INIT_DONE)) {
10310 				rc = adapter_full_init(sc);
10311 				if (rc != 0)
10312 					break;
10313 			}
10314 
10315 			rc = ui->activate(sc);
10316 			if (rc == 0) {
10317 				setbit(&sc->active_ulds, id);
10318 				ui->refcount++;
10319 			}
10320 			break;
10321 		}
10322 	}
10323 
10324 	sx_sunlock(&t4_uld_list_lock);
10325 
10326 	return (rc);
10327 }
10328 
10329 int
10330 t4_deactivate_uld(struct adapter *sc, int id)
10331 {
10332 	int rc;
10333 	struct uld_info *ui;
10334 
10335 	ASSERT_SYNCHRONIZED_OP(sc);
10336 
10337 	if (id < 0 || id > ULD_MAX)
10338 		return (EINVAL);
10339 	rc = ENXIO;
10340 
10341 	sx_slock(&t4_uld_list_lock);
10342 
10343 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10344 		if (ui->uld_id == id) {
10345 			rc = ui->deactivate(sc);
10346 			if (rc == 0) {
10347 				clrbit(&sc->active_ulds, id);
10348 				ui->refcount--;
10349 			}
10350 			break;
10351 		}
10352 	}
10353 
10354 	sx_sunlock(&t4_uld_list_lock);
10355 
10356 	return (rc);
10357 }
10358 
10359 int
10360 uld_active(struct adapter *sc, int uld_id)
10361 {
10362 
10363 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
10364 
10365 	return (isset(&sc->active_ulds, uld_id));
10366 }
10367 #endif
10368 
10369 /*
10370  * t  = ptr to tunable.
10371  * nc = number of CPUs.
10372  * c  = compiled in default for that tunable.
10373  */
10374 static void
10375 calculate_nqueues(int *t, int nc, const int c)
10376 {
10377 	int nq;
10378 
10379 	if (*t > 0)
10380 		return;
10381 	nq = *t < 0 ? -*t : c;
10382 	*t = min(nc, nq);
10383 }
10384 
10385 /*
10386  * Come up with reasonable defaults for some of the tunables, provided they're
10387  * not set by the user (in which case we'll use the values as is).
10388  */
10389 static void
10390 tweak_tunables(void)
10391 {
10392 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
10393 
10394 	if (t4_ntxq < 1) {
10395 #ifdef RSS
10396 		t4_ntxq = rss_getnumbuckets();
10397 #else
10398 		calculate_nqueues(&t4_ntxq, nc, NTXQ);
10399 #endif
10400 	}
10401 
10402 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
10403 
10404 	if (t4_nrxq < 1) {
10405 #ifdef RSS
10406 		t4_nrxq = rss_getnumbuckets();
10407 #else
10408 		calculate_nqueues(&t4_nrxq, nc, NRXQ);
10409 #endif
10410 	}
10411 
10412 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
10413 
10414 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
10415 	calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
10416 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
10417 #endif
10418 #ifdef TCP_OFFLOAD
10419 	calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
10420 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
10421 
10422 	if (t4_toecaps_allowed == -1)
10423 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
10424 
10425 	if (t4_rdmacaps_allowed == -1) {
10426 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
10427 		    FW_CAPS_CONFIG_RDMA_RDMAC;
10428 	}
10429 
10430 	if (t4_iscsicaps_allowed == -1) {
10431 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
10432 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
10433 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
10434 	}
10435 
10436 	if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
10437 		t4_tmr_idx_ofld = TMR_IDX_OFLD;
10438 
10439 	if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
10440 		t4_pktc_idx_ofld = PKTC_IDX_OFLD;
10441 #else
10442 	if (t4_toecaps_allowed == -1)
10443 		t4_toecaps_allowed = 0;
10444 
10445 	if (t4_rdmacaps_allowed == -1)
10446 		t4_rdmacaps_allowed = 0;
10447 
10448 	if (t4_iscsicaps_allowed == -1)
10449 		t4_iscsicaps_allowed = 0;
10450 #endif
10451 
10452 #ifdef DEV_NETMAP
10453 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
10454 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
10455 #endif
10456 
10457 	if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
10458 		t4_tmr_idx = TMR_IDX;
10459 
10460 	if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
10461 		t4_pktc_idx = PKTC_IDX;
10462 
10463 	if (t4_qsize_txq < 128)
10464 		t4_qsize_txq = 128;
10465 
10466 	if (t4_qsize_rxq < 128)
10467 		t4_qsize_rxq = 128;
10468 	while (t4_qsize_rxq & 7)
10469 		t4_qsize_rxq++;
10470 
10471 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
10472 
10473 	/*
10474 	 * Number of VIs to create per-port.  The first VI is the "main" regular
10475 	 * VI for the port.  The rest are additional virtual interfaces on the
10476 	 * same physical port.  Note that the main VI does not have native
10477 	 * netmap support but the extra VIs do.
10478 	 *
10479 	 * Limit the number of VIs per port to the number of available
10480 	 * MAC addresses per port.
10481 	 */
10482 	if (t4_num_vis < 1)
10483 		t4_num_vis = 1;
10484 	if (t4_num_vis > nitems(vi_mac_funcs)) {
10485 		t4_num_vis = nitems(vi_mac_funcs);
10486 		printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
10487 	}
10488 
10489 	if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
10490 		pcie_relaxed_ordering = 1;
10491 #if defined(__i386__) || defined(__amd64__)
10492 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
10493 			pcie_relaxed_ordering = 0;
10494 #endif
10495 	}
10496 }
10497 
10498 #ifdef DDB
10499 static void
10500 t4_dump_tcb(struct adapter *sc, int tid)
10501 {
10502 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
10503 
10504 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
10505 	save = t4_read_reg(sc, reg);
10506 	base = sc->memwin[2].mw_base;
10507 
10508 	/* Dump TCB for the tid */
10509 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
10510 	tcb_addr += tid * TCB_SIZE;
10511 
10512 	if (is_t4(sc)) {
10513 		pf = 0;
10514 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
10515 	} else {
10516 		pf = V_PFNUM(sc->pf);
10517 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
10518 	}
10519 	t4_write_reg(sc, reg, win_pos | pf);
10520 	t4_read_reg(sc, reg);
10521 
10522 	off = tcb_addr - win_pos;
10523 	for (i = 0; i < 4; i++) {
10524 		uint32_t buf[8];
10525 		for (j = 0; j < 8; j++, off += 4)
10526 			buf[j] = htonl(t4_read_reg(sc, base + off));
10527 
10528 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
10529 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
10530 		    buf[7]);
10531 	}
10532 
10533 	t4_write_reg(sc, reg, save);
10534 	t4_read_reg(sc, reg);
10535 }
10536 
10537 static void
10538 t4_dump_devlog(struct adapter *sc)
10539 {
10540 	struct devlog_params *dparams = &sc->params.devlog;
10541 	struct fw_devlog_e e;
10542 	int i, first, j, m, nentries, rc;
10543 	uint64_t ftstamp = UINT64_MAX;
10544 
10545 	if (dparams->start == 0) {
10546 		db_printf("devlog params not valid\n");
10547 		return;
10548 	}
10549 
10550 	nentries = dparams->size / sizeof(struct fw_devlog_e);
10551 	m = fwmtype_to_hwmtype(dparams->memtype);
10552 
10553 	/* Find the first entry. */
10554 	first = -1;
10555 	for (i = 0; i < nentries && !db_pager_quit; i++) {
10556 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10557 		    sizeof(e), (void *)&e);
10558 		if (rc != 0)
10559 			break;
10560 
10561 		if (e.timestamp == 0)
10562 			break;
10563 
10564 		e.timestamp = be64toh(e.timestamp);
10565 		if (e.timestamp < ftstamp) {
10566 			ftstamp = e.timestamp;
10567 			first = i;
10568 		}
10569 	}
10570 
10571 	if (first == -1)
10572 		return;
10573 
10574 	i = first;
10575 	do {
10576 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10577 		    sizeof(e), (void *)&e);
10578 		if (rc != 0)
10579 			return;
10580 
10581 		if (e.timestamp == 0)
10582 			return;
10583 
10584 		e.timestamp = be64toh(e.timestamp);
10585 		e.seqno = be32toh(e.seqno);
10586 		for (j = 0; j < 8; j++)
10587 			e.params[j] = be32toh(e.params[j]);
10588 
10589 		db_printf("%10d  %15ju  %8s  %8s  ",
10590 		    e.seqno, e.timestamp,
10591 		    (e.level < nitems(devlog_level_strings) ?
10592 			devlog_level_strings[e.level] : "UNKNOWN"),
10593 		    (e.facility < nitems(devlog_facility_strings) ?
10594 			devlog_facility_strings[e.facility] : "UNKNOWN"));
10595 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
10596 		    e.params[3], e.params[4], e.params[5], e.params[6],
10597 		    e.params[7]);
10598 
10599 		if (++i == nentries)
10600 			i = 0;
10601 	} while (i != first && !db_pager_quit);
10602 }
10603 
10604 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
10605 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
10606 
10607 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
10608 {
10609 	device_t dev;
10610 	int t;
10611 	bool valid;
10612 
10613 	valid = false;
10614 	t = db_read_token();
10615 	if (t == tIDENT) {
10616 		dev = device_lookup_by_name(db_tok_string);
10617 		valid = true;
10618 	}
10619 	db_skip_to_eol();
10620 	if (!valid) {
10621 		db_printf("usage: show t4 devlog <nexus>\n");
10622 		return;
10623 	}
10624 
10625 	if (dev == NULL) {
10626 		db_printf("device not found\n");
10627 		return;
10628 	}
10629 
10630 	t4_dump_devlog(device_get_softc(dev));
10631 }
10632 
10633 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
10634 {
10635 	device_t dev;
10636 	int radix, tid, t;
10637 	bool valid;
10638 
10639 	valid = false;
10640 	radix = db_radix;
10641 	db_radix = 10;
10642 	t = db_read_token();
10643 	if (t == tIDENT) {
10644 		dev = device_lookup_by_name(db_tok_string);
10645 		t = db_read_token();
10646 		if (t == tNUMBER) {
10647 			tid = db_tok_number;
10648 			valid = true;
10649 		}
10650 	}
10651 	db_radix = radix;
10652 	db_skip_to_eol();
10653 	if (!valid) {
10654 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
10655 		return;
10656 	}
10657 
10658 	if (dev == NULL) {
10659 		db_printf("device not found\n");
10660 		return;
10661 	}
10662 	if (tid < 0) {
10663 		db_printf("invalid tid\n");
10664 		return;
10665 	}
10666 
10667 	t4_dump_tcb(device_get_softc(dev), tid);
10668 }
10669 #endif
10670 
10671 /*
10672  * Borrowed from cesa_prep_aes_key().
10673  *
10674  * NB: The crypto engine wants the words in the decryption key in reverse
10675  * order.
10676  */
10677 void
10678 t4_aes_getdeckey(void *dec_key, const void *enc_key, unsigned int kbits)
10679 {
10680 	uint32_t ek[4 * (RIJNDAEL_MAXNR + 1)];
10681 	uint32_t *dkey;
10682 	int i;
10683 
10684 	rijndaelKeySetupEnc(ek, enc_key, kbits);
10685 	dkey = dec_key;
10686 	dkey += (kbits / 8) / 4;
10687 
10688 	switch (kbits) {
10689 	case 128:
10690 		for (i = 0; i < 4; i++)
10691 			*--dkey = htobe32(ek[4 * 10 + i]);
10692 		break;
10693 	case 192:
10694 		for (i = 0; i < 2; i++)
10695 			*--dkey = htobe32(ek[4 * 11 + 2 + i]);
10696 		for (i = 0; i < 4; i++)
10697 			*--dkey = htobe32(ek[4 * 12 + i]);
10698 		break;
10699 	case 256:
10700 		for (i = 0; i < 4; i++)
10701 			*--dkey = htobe32(ek[4 * 13 + i]);
10702 		for (i = 0; i < 4; i++)
10703 			*--dkey = htobe32(ek[4 * 14 + i]);
10704 		break;
10705 	}
10706 	MPASS(dkey == dec_key);
10707 }
10708 
10709 static struct sx mlu;	/* mod load unload */
10710 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
10711 
10712 static int
10713 mod_event(module_t mod, int cmd, void *arg)
10714 {
10715 	int rc = 0;
10716 	static int loaded = 0;
10717 
10718 	switch (cmd) {
10719 	case MOD_LOAD:
10720 		sx_xlock(&mlu);
10721 		if (loaded++ == 0) {
10722 			t4_sge_modload();
10723 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10724 			    t4_filter_rpl, CPL_COOKIE_FILTER);
10725 			t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
10726 			    do_l2t_write_rpl, CPL_COOKIE_FILTER);
10727 			t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL,
10728 			    t4_hashfilter_ao_rpl, CPL_COOKIE_HASHFILTER);
10729 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10730 			    t4_hashfilter_tcb_rpl, CPL_COOKIE_HASHFILTER);
10731 			t4_register_shared_cpl_handler(CPL_ABORT_RPL_RSS,
10732 			    t4_del_hashfilter_rpl, CPL_COOKIE_HASHFILTER);
10733 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
10734 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
10735 			t4_register_cpl_handler(CPL_SMT_WRITE_RPL,
10736 			    do_smt_write_rpl);
10737 			sx_init(&t4_list_lock, "T4/T5 adapters");
10738 			SLIST_INIT(&t4_list);
10739 			callout_init(&fatal_callout, 1);
10740 #ifdef TCP_OFFLOAD
10741 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
10742 			SLIST_INIT(&t4_uld_list);
10743 #endif
10744 #ifdef INET6
10745 			t4_clip_modload();
10746 #endif
10747 			t4_tracer_modload();
10748 			tweak_tunables();
10749 		}
10750 		sx_xunlock(&mlu);
10751 		break;
10752 
10753 	case MOD_UNLOAD:
10754 		sx_xlock(&mlu);
10755 		if (--loaded == 0) {
10756 			int tries;
10757 
10758 			sx_slock(&t4_list_lock);
10759 			if (!SLIST_EMPTY(&t4_list)) {
10760 				rc = EBUSY;
10761 				sx_sunlock(&t4_list_lock);
10762 				goto done_unload;
10763 			}
10764 #ifdef TCP_OFFLOAD
10765 			sx_slock(&t4_uld_list_lock);
10766 			if (!SLIST_EMPTY(&t4_uld_list)) {
10767 				rc = EBUSY;
10768 				sx_sunlock(&t4_uld_list_lock);
10769 				sx_sunlock(&t4_list_lock);
10770 				goto done_unload;
10771 			}
10772 #endif
10773 			tries = 0;
10774 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
10775 				uprintf("%ju clusters with custom free routine "
10776 				    "still is use.\n", t4_sge_extfree_refs());
10777 				pause("t4unload", 2 * hz);
10778 			}
10779 #ifdef TCP_OFFLOAD
10780 			sx_sunlock(&t4_uld_list_lock);
10781 #endif
10782 			sx_sunlock(&t4_list_lock);
10783 
10784 			if (t4_sge_extfree_refs() == 0) {
10785 				t4_tracer_modunload();
10786 #ifdef INET6
10787 				t4_clip_modunload();
10788 #endif
10789 #ifdef TCP_OFFLOAD
10790 				sx_destroy(&t4_uld_list_lock);
10791 #endif
10792 				sx_destroy(&t4_list_lock);
10793 				t4_sge_modunload();
10794 				loaded = 0;
10795 			} else {
10796 				rc = EBUSY;
10797 				loaded++;	/* undo earlier decrement */
10798 			}
10799 		}
10800 done_unload:
10801 		sx_xunlock(&mlu);
10802 		break;
10803 	}
10804 
10805 	return (rc);
10806 }
10807 
10808 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10809 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10810 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10811 
10812 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10813 MODULE_VERSION(t4nex, 1);
10814 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10815 #ifdef DEV_NETMAP
10816 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10817 #endif /* DEV_NETMAP */
10818 
10819 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10820 MODULE_VERSION(t5nex, 1);
10821 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10822 #ifdef DEV_NETMAP
10823 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10824 #endif /* DEV_NETMAP */
10825 
10826 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10827 MODULE_VERSION(t6nex, 1);
10828 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10829 #ifdef DEV_NETMAP
10830 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10831 #endif /* DEV_NETMAP */
10832 
10833 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10834 MODULE_VERSION(cxgbe, 1);
10835 
10836 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10837 MODULE_VERSION(cxl, 1);
10838 
10839 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10840 MODULE_VERSION(cc, 1);
10841 
10842 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10843 MODULE_VERSION(vcxgbe, 1);
10844 
10845 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10846 MODULE_VERSION(vcxl, 1);
10847 
10848 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10849 MODULE_VERSION(vcc, 1);
10850