xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision f0483545503a78e16e256d46d458a2faae2f07ea)
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 firmware 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 get_params__post_init(struct adapter *);
612 static int set_params__post_init(struct adapter *);
613 static void t4_set_desc(struct adapter *);
614 static bool fixed_ifmedia(struct port_info *);
615 static void build_medialist(struct port_info *);
616 static void init_link_config(struct port_info *);
617 static int fixup_link_config(struct port_info *);
618 static int apply_link_config(struct port_info *);
619 static int cxgbe_init_synchronized(struct vi_info *);
620 static int cxgbe_uninit_synchronized(struct vi_info *);
621 static void quiesce_txq(struct adapter *, struct sge_txq *);
622 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
623 static void quiesce_iq(struct adapter *, struct sge_iq *);
624 static void quiesce_fl(struct adapter *, struct sge_fl *);
625 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
626     driver_intr_t *, void *, char *);
627 static int t4_free_irq(struct adapter *, struct irq *);
628 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
629 static void vi_refresh_stats(struct adapter *, struct vi_info *);
630 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
631 static void cxgbe_tick(void *);
632 static void cxgbe_sysctls(struct port_info *);
633 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
634 static int sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS);
635 static int sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS);
636 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
637 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
638 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
639 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
640 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
641 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
642 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
643 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
644 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
645 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
646 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
647 static int sysctl_loadavg(SYSCTL_HANDLER_ARGS);
648 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
649 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
650 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
651 static int sysctl_cim_la_t6(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 port_info *pi;
1376 
1377 	pi = device_get_softc(dev);
1378 	snprintf(buf, buflen, "port=%d", pi->port_id);
1379 	return (0);
1380 }
1381 
1382 static int
1383 t4_ready(device_t dev)
1384 {
1385 	struct adapter *sc;
1386 
1387 	sc = device_get_softc(dev);
1388 	if (sc->flags & FW_OK)
1389 		return (0);
1390 	return (ENXIO);
1391 }
1392 
1393 static int
1394 t4_read_port_device(device_t dev, int port, device_t *child)
1395 {
1396 	struct adapter *sc;
1397 	struct port_info *pi;
1398 
1399 	sc = device_get_softc(dev);
1400 	if (port < 0 || port >= MAX_NPORTS)
1401 		return (EINVAL);
1402 	pi = sc->port[port];
1403 	if (pi == NULL || pi->dev == NULL)
1404 		return (ENXIO);
1405 	*child = pi->dev;
1406 	return (0);
1407 }
1408 
1409 static int
1410 notify_siblings(device_t dev, int detaching)
1411 {
1412 	device_t sibling;
1413 	int error, i;
1414 
1415 	error = 0;
1416 	for (i = 0; i < PCI_FUNCMAX; i++) {
1417 		if (i == pci_get_function(dev))
1418 			continue;
1419 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1420 		    pci_get_slot(dev), i);
1421 		if (sibling == NULL || !device_is_attached(sibling))
1422 			continue;
1423 		if (detaching)
1424 			error = T4_DETACH_CHILD(sibling);
1425 		else
1426 			(void)T4_ATTACH_CHILD(sibling);
1427 		if (error)
1428 			break;
1429 	}
1430 	return (error);
1431 }
1432 
1433 /*
1434  * Idempotent
1435  */
1436 static int
1437 t4_detach(device_t dev)
1438 {
1439 	struct adapter *sc;
1440 	int rc;
1441 
1442 	sc = device_get_softc(dev);
1443 
1444 	rc = notify_siblings(dev, 1);
1445 	if (rc) {
1446 		device_printf(dev,
1447 		    "failed to detach sibling devices: %d\n", rc);
1448 		return (rc);
1449 	}
1450 
1451 	return (t4_detach_common(dev));
1452 }
1453 
1454 int
1455 t4_detach_common(device_t dev)
1456 {
1457 	struct adapter *sc;
1458 	struct port_info *pi;
1459 	int i, rc;
1460 
1461 	sc = device_get_softc(dev);
1462 
1463 	if (sc->cdev) {
1464 		destroy_dev(sc->cdev);
1465 		sc->cdev = NULL;
1466 	}
1467 
1468 	sc->flags &= ~CHK_MBOX_ACCESS;
1469 	if (sc->flags & FULL_INIT_DONE) {
1470 		if (!(sc->flags & IS_VF))
1471 			t4_intr_disable(sc);
1472 	}
1473 
1474 	if (device_is_attached(dev)) {
1475 		rc = bus_generic_detach(dev);
1476 		if (rc) {
1477 			device_printf(dev,
1478 			    "failed to detach child devices: %d\n", rc);
1479 			return (rc);
1480 		}
1481 	}
1482 
1483 	for (i = 0; i < sc->intr_count; i++)
1484 		t4_free_irq(sc, &sc->irq[i]);
1485 
1486 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1487 		t4_free_tx_sched(sc);
1488 
1489 	for (i = 0; i < MAX_NPORTS; i++) {
1490 		pi = sc->port[i];
1491 		if (pi) {
1492 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1493 			if (pi->dev)
1494 				device_delete_child(dev, pi->dev);
1495 
1496 			mtx_destroy(&pi->pi_lock);
1497 			free(pi->vi, M_CXGBE);
1498 			free(pi, M_CXGBE);
1499 		}
1500 	}
1501 
1502 	device_delete_children(dev);
1503 
1504 	if (sc->flags & FULL_INIT_DONE)
1505 		adapter_full_uninit(sc);
1506 
1507 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1508 		t4_fw_bye(sc, sc->mbox);
1509 
1510 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1511 		pci_release_msi(dev);
1512 
1513 	if (sc->regs_res)
1514 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1515 		    sc->regs_res);
1516 
1517 	if (sc->udbs_res)
1518 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1519 		    sc->udbs_res);
1520 
1521 	if (sc->msix_res)
1522 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1523 		    sc->msix_res);
1524 
1525 	if (sc->l2t)
1526 		t4_free_l2t(sc->l2t);
1527 	if (sc->smt)
1528 		t4_free_smt(sc->smt);
1529 #ifdef RATELIMIT
1530 	t4_free_etid_table(sc);
1531 #endif
1532 	if (sc->key_map)
1533 		vmem_destroy(sc->key_map);
1534 #ifdef INET6
1535 	t4_destroy_clip_table(sc);
1536 #endif
1537 
1538 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1539 	free(sc->sge.ofld_txq, M_CXGBE);
1540 #endif
1541 #ifdef TCP_OFFLOAD
1542 	free(sc->sge.ofld_rxq, M_CXGBE);
1543 #endif
1544 #ifdef DEV_NETMAP
1545 	free(sc->sge.nm_rxq, M_CXGBE);
1546 	free(sc->sge.nm_txq, M_CXGBE);
1547 #endif
1548 	free(sc->irq, M_CXGBE);
1549 	free(sc->sge.rxq, M_CXGBE);
1550 	free(sc->sge.txq, M_CXGBE);
1551 	free(sc->sge.ctrlq, M_CXGBE);
1552 	free(sc->sge.iqmap, M_CXGBE);
1553 	free(sc->sge.eqmap, M_CXGBE);
1554 	free(sc->tids.ftid_tab, M_CXGBE);
1555 	free(sc->tids.hpftid_tab, M_CXGBE);
1556 	free_hftid_hash(&sc->tids);
1557 	free(sc->tids.atid_tab, M_CXGBE);
1558 	free(sc->tids.tid_tab, M_CXGBE);
1559 	free(sc->tt.tls_rx_ports, M_CXGBE);
1560 	t4_destroy_dma_tag(sc);
1561 	if (mtx_initialized(&sc->sc_lock)) {
1562 		sx_xlock(&t4_list_lock);
1563 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1564 		sx_xunlock(&t4_list_lock);
1565 		mtx_destroy(&sc->sc_lock);
1566 	}
1567 
1568 	callout_drain(&sc->sfl_callout);
1569 	if (mtx_initialized(&sc->tids.ftid_lock)) {
1570 		mtx_destroy(&sc->tids.ftid_lock);
1571 		cv_destroy(&sc->tids.ftid_cv);
1572 	}
1573 	if (mtx_initialized(&sc->tids.atid_lock))
1574 		mtx_destroy(&sc->tids.atid_lock);
1575 	if (mtx_initialized(&sc->sfl_lock))
1576 		mtx_destroy(&sc->sfl_lock);
1577 	if (mtx_initialized(&sc->ifp_lock))
1578 		mtx_destroy(&sc->ifp_lock);
1579 	if (mtx_initialized(&sc->reg_lock))
1580 		mtx_destroy(&sc->reg_lock);
1581 
1582 	if (rw_initialized(&sc->policy_lock)) {
1583 		rw_destroy(&sc->policy_lock);
1584 #ifdef TCP_OFFLOAD
1585 		if (sc->policy != NULL)
1586 			free_offload_policy(sc->policy);
1587 #endif
1588 	}
1589 
1590 	for (i = 0; i < NUM_MEMWIN; i++) {
1591 		struct memwin *mw = &sc->memwin[i];
1592 
1593 		if (rw_initialized(&mw->mw_lock))
1594 			rw_destroy(&mw->mw_lock);
1595 	}
1596 
1597 	bzero(sc, sizeof(*sc));
1598 
1599 	return (0);
1600 }
1601 
1602 static int
1603 cxgbe_probe(device_t dev)
1604 {
1605 	char buf[128];
1606 	struct port_info *pi = device_get_softc(dev);
1607 
1608 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1609 	device_set_desc_copy(dev, buf);
1610 
1611 	return (BUS_PROBE_DEFAULT);
1612 }
1613 
1614 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1615     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1616     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS | \
1617     IFCAP_HWRXTSTMP)
1618 #define T4_CAP_ENABLE (T4_CAP)
1619 
1620 static int
1621 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1622 {
1623 	struct ifnet *ifp;
1624 	struct sbuf *sb;
1625 
1626 	vi->xact_addr_filt = -1;
1627 	callout_init(&vi->tick, 1);
1628 
1629 	/* Allocate an ifnet and set it up */
1630 	ifp = if_alloc(IFT_ETHER);
1631 	if (ifp == NULL) {
1632 		device_printf(dev, "Cannot allocate ifnet\n");
1633 		return (ENOMEM);
1634 	}
1635 	vi->ifp = ifp;
1636 	ifp->if_softc = vi;
1637 
1638 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1639 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1640 
1641 	ifp->if_init = cxgbe_init;
1642 	ifp->if_ioctl = cxgbe_ioctl;
1643 	ifp->if_transmit = cxgbe_transmit;
1644 	ifp->if_qflush = cxgbe_qflush;
1645 	ifp->if_get_counter = cxgbe_get_counter;
1646 #ifdef RATELIMIT
1647 	ifp->if_snd_tag_alloc = cxgbe_snd_tag_alloc;
1648 	ifp->if_snd_tag_modify = cxgbe_snd_tag_modify;
1649 	ifp->if_snd_tag_query = cxgbe_snd_tag_query;
1650 	ifp->if_snd_tag_free = cxgbe_snd_tag_free;
1651 #endif
1652 
1653 	ifp->if_capabilities = T4_CAP;
1654 	ifp->if_capenable = T4_CAP_ENABLE;
1655 #ifdef TCP_OFFLOAD
1656 	if (vi->nofldrxq != 0)
1657 		ifp->if_capabilities |= IFCAP_TOE;
1658 #endif
1659 #ifdef RATELIMIT
1660 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0) {
1661 		ifp->if_capabilities |= IFCAP_TXRTLMT;
1662 		ifp->if_capenable |= IFCAP_TXRTLMT;
1663 	}
1664 #endif
1665 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1666 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1667 
1668 	ifp->if_hw_tsomax = IP_MAXPACKET;
1669 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_TSO;
1670 #ifdef RATELIMIT
1671 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0)
1672 		ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_EO_TSO;
1673 #endif
1674 	ifp->if_hw_tsomaxsegsize = 65536;
1675 
1676 	ether_ifattach(ifp, vi->hw_addr);
1677 #ifdef DEV_NETMAP
1678 	if (vi->nnmrxq != 0)
1679 		cxgbe_nm_attach(vi);
1680 #endif
1681 	sb = sbuf_new_auto();
1682 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1683 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1684 	switch (ifp->if_capabilities & (IFCAP_TOE | IFCAP_TXRTLMT)) {
1685 	case IFCAP_TOE:
1686 		sbuf_printf(sb, "; %d txq (TOE)", vi->nofldtxq);
1687 		break;
1688 	case IFCAP_TOE | IFCAP_TXRTLMT:
1689 		sbuf_printf(sb, "; %d txq (TOE/ETHOFLD)", vi->nofldtxq);
1690 		break;
1691 	case IFCAP_TXRTLMT:
1692 		sbuf_printf(sb, "; %d txq (ETHOFLD)", vi->nofldtxq);
1693 		break;
1694 	}
1695 #endif
1696 #ifdef TCP_OFFLOAD
1697 	if (ifp->if_capabilities & IFCAP_TOE)
1698 		sbuf_printf(sb, ", %d rxq (TOE)", vi->nofldrxq);
1699 #endif
1700 #ifdef DEV_NETMAP
1701 	if (ifp->if_capabilities & IFCAP_NETMAP)
1702 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1703 		    vi->nnmtxq, vi->nnmrxq);
1704 #endif
1705 	sbuf_finish(sb);
1706 	device_printf(dev, "%s\n", sbuf_data(sb));
1707 	sbuf_delete(sb);
1708 
1709 	vi_sysctls(vi);
1710 
1711 	return (0);
1712 }
1713 
1714 static int
1715 cxgbe_attach(device_t dev)
1716 {
1717 	struct port_info *pi = device_get_softc(dev);
1718 	struct adapter *sc = pi->adapter;
1719 	struct vi_info *vi;
1720 	int i, rc;
1721 
1722 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1723 
1724 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1725 	if (rc)
1726 		return (rc);
1727 
1728 	for_each_vi(pi, i, vi) {
1729 		if (i == 0)
1730 			continue;
1731 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1732 		if (vi->dev == NULL) {
1733 			device_printf(dev, "failed to add VI %d\n", i);
1734 			continue;
1735 		}
1736 		device_set_softc(vi->dev, vi);
1737 	}
1738 
1739 	cxgbe_sysctls(pi);
1740 
1741 	bus_generic_attach(dev);
1742 
1743 	return (0);
1744 }
1745 
1746 static void
1747 cxgbe_vi_detach(struct vi_info *vi)
1748 {
1749 	struct ifnet *ifp = vi->ifp;
1750 
1751 	ether_ifdetach(ifp);
1752 
1753 	/* Let detach proceed even if these fail. */
1754 #ifdef DEV_NETMAP
1755 	if (ifp->if_capabilities & IFCAP_NETMAP)
1756 		cxgbe_nm_detach(vi);
1757 #endif
1758 	cxgbe_uninit_synchronized(vi);
1759 	callout_drain(&vi->tick);
1760 	vi_full_uninit(vi);
1761 
1762 	if_free(vi->ifp);
1763 	vi->ifp = NULL;
1764 }
1765 
1766 static int
1767 cxgbe_detach(device_t dev)
1768 {
1769 	struct port_info *pi = device_get_softc(dev);
1770 	struct adapter *sc = pi->adapter;
1771 	int rc;
1772 
1773 	/* Detach the extra VIs first. */
1774 	rc = bus_generic_detach(dev);
1775 	if (rc)
1776 		return (rc);
1777 	device_delete_children(dev);
1778 
1779 	doom_vi(sc, &pi->vi[0]);
1780 
1781 	if (pi->flags & HAS_TRACEQ) {
1782 		sc->traceq = -1;	/* cloner should not create ifnet */
1783 		t4_tracer_port_detach(sc);
1784 	}
1785 
1786 	cxgbe_vi_detach(&pi->vi[0]);
1787 	callout_drain(&pi->tick);
1788 	ifmedia_removeall(&pi->media);
1789 
1790 	end_synchronized_op(sc, 0);
1791 
1792 	return (0);
1793 }
1794 
1795 static void
1796 cxgbe_init(void *arg)
1797 {
1798 	struct vi_info *vi = arg;
1799 	struct adapter *sc = vi->pi->adapter;
1800 
1801 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1802 		return;
1803 	cxgbe_init_synchronized(vi);
1804 	end_synchronized_op(sc, 0);
1805 }
1806 
1807 static int
1808 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1809 {
1810 	int rc = 0, mtu, flags;
1811 	struct vi_info *vi = ifp->if_softc;
1812 	struct port_info *pi = vi->pi;
1813 	struct adapter *sc = pi->adapter;
1814 	struct ifreq *ifr = (struct ifreq *)data;
1815 	uint32_t mask;
1816 
1817 	switch (cmd) {
1818 	case SIOCSIFMTU:
1819 		mtu = ifr->ifr_mtu;
1820 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1821 			return (EINVAL);
1822 
1823 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1824 		if (rc)
1825 			return (rc);
1826 		ifp->if_mtu = mtu;
1827 		if (vi->flags & VI_INIT_DONE) {
1828 			t4_update_fl_bufsize(ifp);
1829 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1830 				rc = update_mac_settings(ifp, XGMAC_MTU);
1831 		}
1832 		end_synchronized_op(sc, 0);
1833 		break;
1834 
1835 	case SIOCSIFFLAGS:
1836 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4flg");
1837 		if (rc)
1838 			return (rc);
1839 
1840 		if (ifp->if_flags & IFF_UP) {
1841 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1842 				flags = vi->if_flags;
1843 				if ((ifp->if_flags ^ flags) &
1844 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1845 					rc = update_mac_settings(ifp,
1846 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1847 				}
1848 			} else {
1849 				rc = cxgbe_init_synchronized(vi);
1850 			}
1851 			vi->if_flags = ifp->if_flags;
1852 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1853 			rc = cxgbe_uninit_synchronized(vi);
1854 		}
1855 		end_synchronized_op(sc, 0);
1856 		break;
1857 
1858 	case SIOCADDMULTI:
1859 	case SIOCDELMULTI:
1860 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4multi");
1861 		if (rc)
1862 			return (rc);
1863 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1864 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1865 		end_synchronized_op(sc, 0);
1866 		break;
1867 
1868 	case SIOCSIFCAP:
1869 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1870 		if (rc)
1871 			return (rc);
1872 
1873 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1874 		if (mask & IFCAP_TXCSUM) {
1875 			ifp->if_capenable ^= IFCAP_TXCSUM;
1876 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1877 
1878 			if (IFCAP_TSO4 & ifp->if_capenable &&
1879 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1880 				ifp->if_capenable &= ~IFCAP_TSO4;
1881 				if_printf(ifp,
1882 				    "tso4 disabled due to -txcsum.\n");
1883 			}
1884 		}
1885 		if (mask & IFCAP_TXCSUM_IPV6) {
1886 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1887 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1888 
1889 			if (IFCAP_TSO6 & ifp->if_capenable &&
1890 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1891 				ifp->if_capenable &= ~IFCAP_TSO6;
1892 				if_printf(ifp,
1893 				    "tso6 disabled due to -txcsum6.\n");
1894 			}
1895 		}
1896 		if (mask & IFCAP_RXCSUM)
1897 			ifp->if_capenable ^= IFCAP_RXCSUM;
1898 		if (mask & IFCAP_RXCSUM_IPV6)
1899 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1900 
1901 		/*
1902 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1903 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1904 		 * sending a TSO request our way, so it's sufficient to toggle
1905 		 * IFCAP_TSOx only.
1906 		 */
1907 		if (mask & IFCAP_TSO4) {
1908 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1909 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1910 				if_printf(ifp, "enable txcsum first.\n");
1911 				rc = EAGAIN;
1912 				goto fail;
1913 			}
1914 			ifp->if_capenable ^= IFCAP_TSO4;
1915 		}
1916 		if (mask & IFCAP_TSO6) {
1917 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1918 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1919 				if_printf(ifp, "enable txcsum6 first.\n");
1920 				rc = EAGAIN;
1921 				goto fail;
1922 			}
1923 			ifp->if_capenable ^= IFCAP_TSO6;
1924 		}
1925 		if (mask & IFCAP_LRO) {
1926 #if defined(INET) || defined(INET6)
1927 			int i;
1928 			struct sge_rxq *rxq;
1929 
1930 			ifp->if_capenable ^= IFCAP_LRO;
1931 			for_each_rxq(vi, i, rxq) {
1932 				if (ifp->if_capenable & IFCAP_LRO)
1933 					rxq->iq.flags |= IQ_LRO_ENABLED;
1934 				else
1935 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1936 			}
1937 #endif
1938 		}
1939 #ifdef TCP_OFFLOAD
1940 		if (mask & IFCAP_TOE) {
1941 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1942 
1943 			rc = toe_capability(vi, enable);
1944 			if (rc != 0)
1945 				goto fail;
1946 
1947 			ifp->if_capenable ^= mask;
1948 		}
1949 #endif
1950 		if (mask & IFCAP_VLAN_HWTAGGING) {
1951 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1952 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1953 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1954 		}
1955 		if (mask & IFCAP_VLAN_MTU) {
1956 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1957 
1958 			/* Need to find out how to disable auto-mtu-inflation */
1959 		}
1960 		if (mask & IFCAP_VLAN_HWTSO)
1961 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1962 		if (mask & IFCAP_VLAN_HWCSUM)
1963 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1964 #ifdef RATELIMIT
1965 		if (mask & IFCAP_TXRTLMT)
1966 			ifp->if_capenable ^= IFCAP_TXRTLMT;
1967 #endif
1968 		if (mask & IFCAP_HWRXTSTMP) {
1969 			int i;
1970 			struct sge_rxq *rxq;
1971 
1972 			ifp->if_capenable ^= IFCAP_HWRXTSTMP;
1973 			for_each_rxq(vi, i, rxq) {
1974 				if (ifp->if_capenable & IFCAP_HWRXTSTMP)
1975 					rxq->iq.flags |= IQ_RX_TIMESTAMP;
1976 				else
1977 					rxq->iq.flags &= ~IQ_RX_TIMESTAMP;
1978 			}
1979 		}
1980 
1981 #ifdef VLAN_CAPABILITIES
1982 		VLAN_CAPABILITIES(ifp);
1983 #endif
1984 fail:
1985 		end_synchronized_op(sc, 0);
1986 		break;
1987 
1988 	case SIOCSIFMEDIA:
1989 	case SIOCGIFMEDIA:
1990 	case SIOCGIFXMEDIA:
1991 		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1992 		break;
1993 
1994 	case SIOCGI2C: {
1995 		struct ifi2creq i2c;
1996 
1997 		rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
1998 		if (rc != 0)
1999 			break;
2000 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
2001 			rc = EPERM;
2002 			break;
2003 		}
2004 		if (i2c.len > sizeof(i2c.data)) {
2005 			rc = EINVAL;
2006 			break;
2007 		}
2008 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
2009 		if (rc)
2010 			return (rc);
2011 		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
2012 		    i2c.offset, i2c.len, &i2c.data[0]);
2013 		end_synchronized_op(sc, 0);
2014 		if (rc == 0)
2015 			rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
2016 		break;
2017 	}
2018 
2019 	default:
2020 		rc = ether_ioctl(ifp, cmd, data);
2021 	}
2022 
2023 	return (rc);
2024 }
2025 
2026 static int
2027 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
2028 {
2029 	struct vi_info *vi = ifp->if_softc;
2030 	struct port_info *pi = vi->pi;
2031 	struct adapter *sc = pi->adapter;
2032 	struct sge_txq *txq;
2033 	void *items[1];
2034 	int rc;
2035 
2036 	M_ASSERTPKTHDR(m);
2037 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
2038 
2039 	if (__predict_false(pi->link_cfg.link_ok == false)) {
2040 		m_freem(m);
2041 		return (ENETDOWN);
2042 	}
2043 
2044 	rc = parse_pkt(sc, &m);
2045 	if (__predict_false(rc != 0)) {
2046 		MPASS(m == NULL);			/* was freed already */
2047 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
2048 		return (rc);
2049 	}
2050 #ifdef RATELIMIT
2051 	if (m->m_pkthdr.snd_tag != NULL) {
2052 		/* EAGAIN tells the stack we are not the correct interface. */
2053 		if (__predict_false(ifp != m->m_pkthdr.snd_tag->ifp)) {
2054 			m_freem(m);
2055 			return (EAGAIN);
2056 		}
2057 
2058 		return (ethofld_transmit(ifp, m));
2059 	}
2060 #endif
2061 
2062 	/* Select a txq. */
2063 	txq = &sc->sge.txq[vi->first_txq];
2064 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2065 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
2066 		    vi->rsrv_noflowq);
2067 
2068 	items[0] = m;
2069 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
2070 	if (__predict_false(rc != 0))
2071 		m_freem(m);
2072 
2073 	return (rc);
2074 }
2075 
2076 static void
2077 cxgbe_qflush(struct ifnet *ifp)
2078 {
2079 	struct vi_info *vi = ifp->if_softc;
2080 	struct sge_txq *txq;
2081 	int i;
2082 
2083 	/* queues do not exist if !VI_INIT_DONE. */
2084 	if (vi->flags & VI_INIT_DONE) {
2085 		for_each_txq(vi, i, txq) {
2086 			TXQ_LOCK(txq);
2087 			txq->eq.flags |= EQ_QFLUSH;
2088 			TXQ_UNLOCK(txq);
2089 			while (!mp_ring_is_idle(txq->r)) {
2090 				mp_ring_check_drainage(txq->r, 0);
2091 				pause("qflush", 1);
2092 			}
2093 			TXQ_LOCK(txq);
2094 			txq->eq.flags &= ~EQ_QFLUSH;
2095 			TXQ_UNLOCK(txq);
2096 		}
2097 	}
2098 	if_qflush(ifp);
2099 }
2100 
2101 static uint64_t
2102 vi_get_counter(struct ifnet *ifp, ift_counter c)
2103 {
2104 	struct vi_info *vi = ifp->if_softc;
2105 	struct fw_vi_stats_vf *s = &vi->stats;
2106 
2107 	vi_refresh_stats(vi->pi->adapter, vi);
2108 
2109 	switch (c) {
2110 	case IFCOUNTER_IPACKETS:
2111 		return (s->rx_bcast_frames + s->rx_mcast_frames +
2112 		    s->rx_ucast_frames);
2113 	case IFCOUNTER_IERRORS:
2114 		return (s->rx_err_frames);
2115 	case IFCOUNTER_OPACKETS:
2116 		return (s->tx_bcast_frames + s->tx_mcast_frames +
2117 		    s->tx_ucast_frames + s->tx_offload_frames);
2118 	case IFCOUNTER_OERRORS:
2119 		return (s->tx_drop_frames);
2120 	case IFCOUNTER_IBYTES:
2121 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
2122 		    s->rx_ucast_bytes);
2123 	case IFCOUNTER_OBYTES:
2124 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
2125 		    s->tx_ucast_bytes + s->tx_offload_bytes);
2126 	case IFCOUNTER_IMCASTS:
2127 		return (s->rx_mcast_frames);
2128 	case IFCOUNTER_OMCASTS:
2129 		return (s->tx_mcast_frames);
2130 	case IFCOUNTER_OQDROPS: {
2131 		uint64_t drops;
2132 
2133 		drops = 0;
2134 		if (vi->flags & VI_INIT_DONE) {
2135 			int i;
2136 			struct sge_txq *txq;
2137 
2138 			for_each_txq(vi, i, txq)
2139 				drops += counter_u64_fetch(txq->r->drops);
2140 		}
2141 
2142 		return (drops);
2143 
2144 	}
2145 
2146 	default:
2147 		return (if_get_counter_default(ifp, c));
2148 	}
2149 }
2150 
2151 uint64_t
2152 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
2153 {
2154 	struct vi_info *vi = ifp->if_softc;
2155 	struct port_info *pi = vi->pi;
2156 	struct adapter *sc = pi->adapter;
2157 	struct port_stats *s = &pi->stats;
2158 
2159 	if (pi->nvi > 1 || sc->flags & IS_VF)
2160 		return (vi_get_counter(ifp, c));
2161 
2162 	cxgbe_refresh_stats(sc, pi);
2163 
2164 	switch (c) {
2165 	case IFCOUNTER_IPACKETS:
2166 		return (s->rx_frames);
2167 
2168 	case IFCOUNTER_IERRORS:
2169 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
2170 		    s->rx_fcs_err + s->rx_len_err);
2171 
2172 	case IFCOUNTER_OPACKETS:
2173 		return (s->tx_frames);
2174 
2175 	case IFCOUNTER_OERRORS:
2176 		return (s->tx_error_frames);
2177 
2178 	case IFCOUNTER_IBYTES:
2179 		return (s->rx_octets);
2180 
2181 	case IFCOUNTER_OBYTES:
2182 		return (s->tx_octets);
2183 
2184 	case IFCOUNTER_IMCASTS:
2185 		return (s->rx_mcast_frames);
2186 
2187 	case IFCOUNTER_OMCASTS:
2188 		return (s->tx_mcast_frames);
2189 
2190 	case IFCOUNTER_IQDROPS:
2191 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
2192 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
2193 		    s->rx_trunc3 + pi->tnl_cong_drops);
2194 
2195 	case IFCOUNTER_OQDROPS: {
2196 		uint64_t drops;
2197 
2198 		drops = s->tx_drop;
2199 		if (vi->flags & VI_INIT_DONE) {
2200 			int i;
2201 			struct sge_txq *txq;
2202 
2203 			for_each_txq(vi, i, txq)
2204 				drops += counter_u64_fetch(txq->r->drops);
2205 		}
2206 
2207 		return (drops);
2208 
2209 	}
2210 
2211 	default:
2212 		return (if_get_counter_default(ifp, c));
2213 	}
2214 }
2215 
2216 /*
2217  * The kernel picks a media from the list we had provided but we still validate
2218  * the requeste.
2219  */
2220 int
2221 cxgbe_media_change(struct ifnet *ifp)
2222 {
2223 	struct vi_info *vi = ifp->if_softc;
2224 	struct port_info *pi = vi->pi;
2225 	struct ifmedia *ifm = &pi->media;
2226 	struct link_config *lc = &pi->link_cfg;
2227 	struct adapter *sc = pi->adapter;
2228 	int rc;
2229 
2230 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mec");
2231 	if (rc != 0)
2232 		return (rc);
2233 	PORT_LOCK(pi);
2234 	if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
2235 		/* ifconfig .. media autoselect */
2236 		if (!(lc->supported & FW_PORT_CAP32_ANEG)) {
2237 			rc = ENOTSUP; /* AN not supported by transceiver */
2238 			goto done;
2239 		}
2240 		lc->requested_aneg = AUTONEG_ENABLE;
2241 		lc->requested_speed = 0;
2242 		lc->requested_fc |= PAUSE_AUTONEG;
2243 	} else {
2244 		lc->requested_aneg = AUTONEG_DISABLE;
2245 		lc->requested_speed =
2246 		    ifmedia_baudrate(ifm->ifm_media) / 1000000;
2247 		lc->requested_fc = 0;
2248 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
2249 			lc->requested_fc |= PAUSE_RX;
2250 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
2251 			lc->requested_fc |= PAUSE_TX;
2252 	}
2253 	if (pi->up_vis > 0) {
2254 		fixup_link_config(pi);
2255 		rc = apply_link_config(pi);
2256 	}
2257 done:
2258 	PORT_UNLOCK(pi);
2259 	end_synchronized_op(sc, 0);
2260 	return (rc);
2261 }
2262 
2263 /*
2264  * Base media word (without ETHER, pause, link active, etc.) for the port at the
2265  * given speed.
2266  */
2267 static int
2268 port_mword(struct port_info *pi, uint32_t speed)
2269 {
2270 
2271 	MPASS(speed & M_FW_PORT_CAP32_SPEED);
2272 	MPASS(powerof2(speed));
2273 
2274 	switch(pi->port_type) {
2275 	case FW_PORT_TYPE_BT_SGMII:
2276 	case FW_PORT_TYPE_BT_XFI:
2277 	case FW_PORT_TYPE_BT_XAUI:
2278 		/* BaseT */
2279 		switch (speed) {
2280 		case FW_PORT_CAP32_SPEED_100M:
2281 			return (IFM_100_T);
2282 		case FW_PORT_CAP32_SPEED_1G:
2283 			return (IFM_1000_T);
2284 		case FW_PORT_CAP32_SPEED_10G:
2285 			return (IFM_10G_T);
2286 		}
2287 		break;
2288 	case FW_PORT_TYPE_KX4:
2289 		if (speed == FW_PORT_CAP32_SPEED_10G)
2290 			return (IFM_10G_KX4);
2291 		break;
2292 	case FW_PORT_TYPE_CX4:
2293 		if (speed == FW_PORT_CAP32_SPEED_10G)
2294 			return (IFM_10G_CX4);
2295 		break;
2296 	case FW_PORT_TYPE_KX:
2297 		if (speed == FW_PORT_CAP32_SPEED_1G)
2298 			return (IFM_1000_KX);
2299 		break;
2300 	case FW_PORT_TYPE_KR:
2301 	case FW_PORT_TYPE_BP_AP:
2302 	case FW_PORT_TYPE_BP4_AP:
2303 	case FW_PORT_TYPE_BP40_BA:
2304 	case FW_PORT_TYPE_KR4_100G:
2305 	case FW_PORT_TYPE_KR_SFP28:
2306 	case FW_PORT_TYPE_KR_XLAUI:
2307 		switch (speed) {
2308 		case FW_PORT_CAP32_SPEED_1G:
2309 			return (IFM_1000_KX);
2310 		case FW_PORT_CAP32_SPEED_10G:
2311 			return (IFM_10G_KR);
2312 		case FW_PORT_CAP32_SPEED_25G:
2313 			return (IFM_25G_KR);
2314 		case FW_PORT_CAP32_SPEED_40G:
2315 			return (IFM_40G_KR4);
2316 		case FW_PORT_CAP32_SPEED_50G:
2317 			return (IFM_50G_KR2);
2318 		case FW_PORT_CAP32_SPEED_100G:
2319 			return (IFM_100G_KR4);
2320 		}
2321 		break;
2322 	case FW_PORT_TYPE_FIBER_XFI:
2323 	case FW_PORT_TYPE_FIBER_XAUI:
2324 	case FW_PORT_TYPE_SFP:
2325 	case FW_PORT_TYPE_QSFP_10G:
2326 	case FW_PORT_TYPE_QSA:
2327 	case FW_PORT_TYPE_QSFP:
2328 	case FW_PORT_TYPE_CR4_QSFP:
2329 	case FW_PORT_TYPE_CR_QSFP:
2330 	case FW_PORT_TYPE_CR2_QSFP:
2331 	case FW_PORT_TYPE_SFP28:
2332 		/* Pluggable transceiver */
2333 		switch (pi->mod_type) {
2334 		case FW_PORT_MOD_TYPE_LR:
2335 			switch (speed) {
2336 			case FW_PORT_CAP32_SPEED_1G:
2337 				return (IFM_1000_LX);
2338 			case FW_PORT_CAP32_SPEED_10G:
2339 				return (IFM_10G_LR);
2340 			case FW_PORT_CAP32_SPEED_25G:
2341 				return (IFM_25G_LR);
2342 			case FW_PORT_CAP32_SPEED_40G:
2343 				return (IFM_40G_LR4);
2344 			case FW_PORT_CAP32_SPEED_50G:
2345 				return (IFM_50G_LR2);
2346 			case FW_PORT_CAP32_SPEED_100G:
2347 				return (IFM_100G_LR4);
2348 			}
2349 			break;
2350 		case FW_PORT_MOD_TYPE_SR:
2351 			switch (speed) {
2352 			case FW_PORT_CAP32_SPEED_1G:
2353 				return (IFM_1000_SX);
2354 			case FW_PORT_CAP32_SPEED_10G:
2355 				return (IFM_10G_SR);
2356 			case FW_PORT_CAP32_SPEED_25G:
2357 				return (IFM_25G_SR);
2358 			case FW_PORT_CAP32_SPEED_40G:
2359 				return (IFM_40G_SR4);
2360 			case FW_PORT_CAP32_SPEED_50G:
2361 				return (IFM_50G_SR2);
2362 			case FW_PORT_CAP32_SPEED_100G:
2363 				return (IFM_100G_SR4);
2364 			}
2365 			break;
2366 		case FW_PORT_MOD_TYPE_ER:
2367 			if (speed == FW_PORT_CAP32_SPEED_10G)
2368 				return (IFM_10G_ER);
2369 			break;
2370 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2371 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2372 			switch (speed) {
2373 			case FW_PORT_CAP32_SPEED_1G:
2374 				return (IFM_1000_CX);
2375 			case FW_PORT_CAP32_SPEED_10G:
2376 				return (IFM_10G_TWINAX);
2377 			case FW_PORT_CAP32_SPEED_25G:
2378 				return (IFM_25G_CR);
2379 			case FW_PORT_CAP32_SPEED_40G:
2380 				return (IFM_40G_CR4);
2381 			case FW_PORT_CAP32_SPEED_50G:
2382 				return (IFM_50G_CR2);
2383 			case FW_PORT_CAP32_SPEED_100G:
2384 				return (IFM_100G_CR4);
2385 			}
2386 			break;
2387 		case FW_PORT_MOD_TYPE_LRM:
2388 			if (speed == FW_PORT_CAP32_SPEED_10G)
2389 				return (IFM_10G_LRM);
2390 			break;
2391 		case FW_PORT_MOD_TYPE_NA:
2392 			MPASS(0);	/* Not pluggable? */
2393 			/* fall throough */
2394 		case FW_PORT_MOD_TYPE_ERROR:
2395 		case FW_PORT_MOD_TYPE_UNKNOWN:
2396 		case FW_PORT_MOD_TYPE_NOTSUPPORTED:
2397 			break;
2398 		case FW_PORT_MOD_TYPE_NONE:
2399 			return (IFM_NONE);
2400 		}
2401 		break;
2402 	case FW_PORT_TYPE_NONE:
2403 		return (IFM_NONE);
2404 	}
2405 
2406 	return (IFM_UNKNOWN);
2407 }
2408 
2409 void
2410 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2411 {
2412 	struct vi_info *vi = ifp->if_softc;
2413 	struct port_info *pi = vi->pi;
2414 	struct adapter *sc = pi->adapter;
2415 	struct link_config *lc = &pi->link_cfg;
2416 
2417 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4med") != 0)
2418 		return;
2419 	PORT_LOCK(pi);
2420 
2421 	if (pi->up_vis == 0) {
2422 		/*
2423 		 * If all the interfaces are administratively down the firmware
2424 		 * does not report transceiver changes.  Refresh port info here
2425 		 * so that ifconfig displays accurate ifmedia at all times.
2426 		 * This is the only reason we have a synchronized op in this
2427 		 * function.  Just PORT_LOCK would have been enough otherwise.
2428 		 */
2429 		t4_update_port_info(pi);
2430 		build_medialist(pi);
2431 	}
2432 
2433 	/* ifm_status */
2434 	ifmr->ifm_status = IFM_AVALID;
2435 	if (lc->link_ok == false)
2436 		goto done;
2437 	ifmr->ifm_status |= IFM_ACTIVE;
2438 
2439 	/* ifm_active */
2440 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2441 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2442 	if (lc->fc & PAUSE_RX)
2443 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2444 	if (lc->fc & PAUSE_TX)
2445 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2446 	ifmr->ifm_active |= port_mword(pi, speed_to_fwcap(lc->speed));
2447 done:
2448 	PORT_UNLOCK(pi);
2449 	end_synchronized_op(sc, 0);
2450 }
2451 
2452 static int
2453 vcxgbe_probe(device_t dev)
2454 {
2455 	char buf[128];
2456 	struct vi_info *vi = device_get_softc(dev);
2457 
2458 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2459 	    vi - vi->pi->vi);
2460 	device_set_desc_copy(dev, buf);
2461 
2462 	return (BUS_PROBE_DEFAULT);
2463 }
2464 
2465 static int
2466 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2467 {
2468 	int func, index, rc;
2469 	uint32_t param, val;
2470 
2471 	ASSERT_SYNCHRONIZED_OP(sc);
2472 
2473 	index = vi - pi->vi;
2474 	MPASS(index > 0);	/* This function deals with _extra_ VIs only */
2475 	KASSERT(index < nitems(vi_mac_funcs),
2476 	    ("%s: VI %s doesn't have a MAC func", __func__,
2477 	    device_get_nameunit(vi->dev)));
2478 	func = vi_mac_funcs[index];
2479 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2480 	    vi->hw_addr, &vi->rss_size, func, 0);
2481 	if (rc < 0) {
2482 		device_printf(vi->dev, "failed to allocate virtual interface %d"
2483 		    "for port %d: %d\n", index, pi->port_id, -rc);
2484 		return (-rc);
2485 	}
2486 	vi->viid = rc;
2487 	if (chip_id(sc) <= CHELSIO_T5)
2488 		vi->smt_idx = (rc & 0x7f) << 1;
2489 	else
2490 		vi->smt_idx = (rc & 0x7f);
2491 
2492 	if (vi->rss_size == 1) {
2493 		/*
2494 		 * This VI didn't get a slice of the RSS table.  Reduce the
2495 		 * number of VIs being created (hw.cxgbe.num_vis) or modify the
2496 		 * configuration file (nvi, rssnvi for this PF) if this is a
2497 		 * problem.
2498 		 */
2499 		device_printf(vi->dev, "RSS table not available.\n");
2500 		vi->rss_base = 0xffff;
2501 
2502 		return (0);
2503 	}
2504 
2505 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2506 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2507 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2508 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2509 	if (rc)
2510 		vi->rss_base = 0xffff;
2511 	else {
2512 		MPASS((val >> 16) == vi->rss_size);
2513 		vi->rss_base = val & 0xffff;
2514 	}
2515 
2516 	return (0);
2517 }
2518 
2519 static int
2520 vcxgbe_attach(device_t dev)
2521 {
2522 	struct vi_info *vi;
2523 	struct port_info *pi;
2524 	struct adapter *sc;
2525 	int rc;
2526 
2527 	vi = device_get_softc(dev);
2528 	pi = vi->pi;
2529 	sc = pi->adapter;
2530 
2531 	rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2532 	if (rc)
2533 		return (rc);
2534 	rc = alloc_extra_vi(sc, pi, vi);
2535 	end_synchronized_op(sc, 0);
2536 	if (rc)
2537 		return (rc);
2538 
2539 	rc = cxgbe_vi_attach(dev, vi);
2540 	if (rc) {
2541 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2542 		return (rc);
2543 	}
2544 	return (0);
2545 }
2546 
2547 static int
2548 vcxgbe_detach(device_t dev)
2549 {
2550 	struct vi_info *vi;
2551 	struct adapter *sc;
2552 
2553 	vi = device_get_softc(dev);
2554 	sc = vi->pi->adapter;
2555 
2556 	doom_vi(sc, vi);
2557 
2558 	cxgbe_vi_detach(vi);
2559 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2560 
2561 	end_synchronized_op(sc, 0);
2562 
2563 	return (0);
2564 }
2565 
2566 void
2567 t4_fatal_err(struct adapter *sc, bool fw_error)
2568 {
2569 
2570 	t4_shutdown_adapter(sc);
2571 	log(LOG_ALERT, "%s: encountered fatal error, adapter stopped.\n",
2572 	    device_get_nameunit(sc->dev));
2573 	if (t4_panic_on_fatal_err)
2574 		panic("panic requested on fatal error");
2575 
2576 	if (fw_error) {
2577 		ASSERT_SYNCHRONIZED_OP(sc);
2578 		sc->flags |= ADAP_ERR;
2579 	} else {
2580 		ADAPTER_LOCK(sc);
2581 		sc->flags |= ADAP_ERR;
2582 		ADAPTER_UNLOCK(sc);
2583 	}
2584 }
2585 
2586 void
2587 t4_add_adapter(struct adapter *sc)
2588 {
2589 	sx_xlock(&t4_list_lock);
2590 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2591 	sx_xunlock(&t4_list_lock);
2592 }
2593 
2594 int
2595 t4_map_bars_0_and_4(struct adapter *sc)
2596 {
2597 	sc->regs_rid = PCIR_BAR(0);
2598 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2599 	    &sc->regs_rid, RF_ACTIVE);
2600 	if (sc->regs_res == NULL) {
2601 		device_printf(sc->dev, "cannot map registers.\n");
2602 		return (ENXIO);
2603 	}
2604 	sc->bt = rman_get_bustag(sc->regs_res);
2605 	sc->bh = rman_get_bushandle(sc->regs_res);
2606 	sc->mmio_len = rman_get_size(sc->regs_res);
2607 	setbit(&sc->doorbells, DOORBELL_KDB);
2608 
2609 	sc->msix_rid = PCIR_BAR(4);
2610 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2611 	    &sc->msix_rid, RF_ACTIVE);
2612 	if (sc->msix_res == NULL) {
2613 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2614 		return (ENXIO);
2615 	}
2616 
2617 	return (0);
2618 }
2619 
2620 int
2621 t4_map_bar_2(struct adapter *sc)
2622 {
2623 
2624 	/*
2625 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2626 	 * to map it if RDMA is disabled.
2627 	 */
2628 	if (is_t4(sc) && sc->rdmacaps == 0)
2629 		return (0);
2630 
2631 	sc->udbs_rid = PCIR_BAR(2);
2632 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2633 	    &sc->udbs_rid, RF_ACTIVE);
2634 	if (sc->udbs_res == NULL) {
2635 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2636 		return (ENXIO);
2637 	}
2638 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2639 
2640 	if (chip_id(sc) >= CHELSIO_T5) {
2641 		setbit(&sc->doorbells, DOORBELL_UDB);
2642 #if defined(__i386__) || defined(__amd64__)
2643 		if (t5_write_combine) {
2644 			int rc, mode;
2645 
2646 			/*
2647 			 * Enable write combining on BAR2.  This is the
2648 			 * userspace doorbell BAR and is split into 128B
2649 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2650 			 * with an egress queue.  The first 64B has the doorbell
2651 			 * and the second 64B can be used to submit a tx work
2652 			 * request with an implicit doorbell.
2653 			 */
2654 
2655 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2656 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2657 			if (rc == 0) {
2658 				clrbit(&sc->doorbells, DOORBELL_UDB);
2659 				setbit(&sc->doorbells, DOORBELL_WCWR);
2660 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2661 			} else {
2662 				device_printf(sc->dev,
2663 				    "couldn't enable write combining: %d\n",
2664 				    rc);
2665 			}
2666 
2667 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2668 			t4_write_reg(sc, A_SGE_STAT_CFG,
2669 			    V_STATSOURCE_T5(7) | mode);
2670 		}
2671 #endif
2672 	}
2673 	sc->iwt.wc_en = isset(&sc->doorbells, DOORBELL_UDBWC) ? 1 : 0;
2674 
2675 	return (0);
2676 }
2677 
2678 struct memwin_init {
2679 	uint32_t base;
2680 	uint32_t aperture;
2681 };
2682 
2683 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2684 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2685 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2686 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2687 };
2688 
2689 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2690 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2691 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2692 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2693 };
2694 
2695 static void
2696 setup_memwin(struct adapter *sc)
2697 {
2698 	const struct memwin_init *mw_init;
2699 	struct memwin *mw;
2700 	int i;
2701 	uint32_t bar0;
2702 
2703 	if (is_t4(sc)) {
2704 		/*
2705 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2706 		 * mechanism.  Works from within PCI passthrough environments
2707 		 * too, where rman_get_start() can return a different value.  We
2708 		 * need to program the T4 memory window decoders with the actual
2709 		 * addresses that will be coming across the PCIe link.
2710 		 */
2711 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2712 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2713 
2714 		mw_init = &t4_memwin[0];
2715 	} else {
2716 		/* T5+ use the relative offset inside the PCIe BAR */
2717 		bar0 = 0;
2718 
2719 		mw_init = &t5_memwin[0];
2720 	}
2721 
2722 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2723 		rw_init(&mw->mw_lock, "memory window access");
2724 		mw->mw_base = mw_init->base;
2725 		mw->mw_aperture = mw_init->aperture;
2726 		mw->mw_curpos = 0;
2727 		t4_write_reg(sc,
2728 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2729 		    (mw->mw_base + bar0) | V_BIR(0) |
2730 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2731 		rw_wlock(&mw->mw_lock);
2732 		position_memwin(sc, i, 0);
2733 		rw_wunlock(&mw->mw_lock);
2734 	}
2735 
2736 	/* flush */
2737 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2738 }
2739 
2740 /*
2741  * Positions the memory window at the given address in the card's address space.
2742  * There are some alignment requirements and the actual position may be at an
2743  * address prior to the requested address.  mw->mw_curpos always has the actual
2744  * position of the window.
2745  */
2746 static void
2747 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2748 {
2749 	struct memwin *mw;
2750 	uint32_t pf;
2751 	uint32_t reg;
2752 
2753 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2754 	mw = &sc->memwin[idx];
2755 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2756 
2757 	if (is_t4(sc)) {
2758 		pf = 0;
2759 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2760 	} else {
2761 		pf = V_PFNUM(sc->pf);
2762 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2763 	}
2764 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2765 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2766 	t4_read_reg(sc, reg);	/* flush */
2767 }
2768 
2769 int
2770 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2771     int len, int rw)
2772 {
2773 	struct memwin *mw;
2774 	uint32_t mw_end, v;
2775 
2776 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2777 
2778 	/* Memory can only be accessed in naturally aligned 4 byte units */
2779 	if (addr & 3 || len & 3 || len <= 0)
2780 		return (EINVAL);
2781 
2782 	mw = &sc->memwin[idx];
2783 	while (len > 0) {
2784 		rw_rlock(&mw->mw_lock);
2785 		mw_end = mw->mw_curpos + mw->mw_aperture;
2786 		if (addr >= mw_end || addr < mw->mw_curpos) {
2787 			/* Will need to reposition the window */
2788 			if (!rw_try_upgrade(&mw->mw_lock)) {
2789 				rw_runlock(&mw->mw_lock);
2790 				rw_wlock(&mw->mw_lock);
2791 			}
2792 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2793 			position_memwin(sc, idx, addr);
2794 			rw_downgrade(&mw->mw_lock);
2795 			mw_end = mw->mw_curpos + mw->mw_aperture;
2796 		}
2797 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2798 		while (addr < mw_end && len > 0) {
2799 			if (rw == 0) {
2800 				v = t4_read_reg(sc, mw->mw_base + addr -
2801 				    mw->mw_curpos);
2802 				*val++ = le32toh(v);
2803 			} else {
2804 				v = *val++;
2805 				t4_write_reg(sc, mw->mw_base + addr -
2806 				    mw->mw_curpos, htole32(v));
2807 			}
2808 			addr += 4;
2809 			len -= 4;
2810 		}
2811 		rw_runlock(&mw->mw_lock);
2812 	}
2813 
2814 	return (0);
2815 }
2816 
2817 int
2818 alloc_atid_tab(struct tid_info *t, int flags)
2819 {
2820 	int i;
2821 
2822 	MPASS(t->natids > 0);
2823 	MPASS(t->atid_tab == NULL);
2824 
2825 	t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
2826 	    M_ZERO | flags);
2827 	if (t->atid_tab == NULL)
2828 		return (ENOMEM);
2829 	mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
2830 	t->afree = t->atid_tab;
2831 	t->atids_in_use = 0;
2832 	for (i = 1; i < t->natids; i++)
2833 		t->atid_tab[i - 1].next = &t->atid_tab[i];
2834 	t->atid_tab[t->natids - 1].next = NULL;
2835 
2836 	return (0);
2837 }
2838 
2839 void
2840 free_atid_tab(struct tid_info *t)
2841 {
2842 
2843 	KASSERT(t->atids_in_use == 0,
2844 	    ("%s: %d atids still in use.", __func__, t->atids_in_use));
2845 
2846 	if (mtx_initialized(&t->atid_lock))
2847 		mtx_destroy(&t->atid_lock);
2848 	free(t->atid_tab, M_CXGBE);
2849 	t->atid_tab = NULL;
2850 }
2851 
2852 int
2853 alloc_atid(struct adapter *sc, void *ctx)
2854 {
2855 	struct tid_info *t = &sc->tids;
2856 	int atid = -1;
2857 
2858 	mtx_lock(&t->atid_lock);
2859 	if (t->afree) {
2860 		union aopen_entry *p = t->afree;
2861 
2862 		atid = p - t->atid_tab;
2863 		MPASS(atid <= M_TID_TID);
2864 		t->afree = p->next;
2865 		p->data = ctx;
2866 		t->atids_in_use++;
2867 	}
2868 	mtx_unlock(&t->atid_lock);
2869 	return (atid);
2870 }
2871 
2872 void *
2873 lookup_atid(struct adapter *sc, int atid)
2874 {
2875 	struct tid_info *t = &sc->tids;
2876 
2877 	return (t->atid_tab[atid].data);
2878 }
2879 
2880 void
2881 free_atid(struct adapter *sc, int atid)
2882 {
2883 	struct tid_info *t = &sc->tids;
2884 	union aopen_entry *p = &t->atid_tab[atid];
2885 
2886 	mtx_lock(&t->atid_lock);
2887 	p->next = t->afree;
2888 	t->afree = p;
2889 	t->atids_in_use--;
2890 	mtx_unlock(&t->atid_lock);
2891 }
2892 
2893 static void
2894 queue_tid_release(struct adapter *sc, int tid)
2895 {
2896 
2897 	CXGBE_UNIMPLEMENTED("deferred tid release");
2898 }
2899 
2900 void
2901 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
2902 {
2903 	struct wrqe *wr;
2904 	struct cpl_tid_release *req;
2905 
2906 	wr = alloc_wrqe(sizeof(*req), ctrlq);
2907 	if (wr == NULL) {
2908 		queue_tid_release(sc, tid);	/* defer */
2909 		return;
2910 	}
2911 	req = wrtod(wr);
2912 
2913 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
2914 
2915 	t4_wrq_tx(sc, wr);
2916 }
2917 
2918 static int
2919 t4_range_cmp(const void *a, const void *b)
2920 {
2921 	return ((const struct t4_range *)a)->start -
2922 	       ((const struct t4_range *)b)->start;
2923 }
2924 
2925 /*
2926  * Verify that the memory range specified by the addr/len pair is valid within
2927  * the card's address space.
2928  */
2929 static int
2930 validate_mem_range(struct adapter *sc, uint32_t addr, uint32_t len)
2931 {
2932 	struct t4_range mem_ranges[4], *r, *next;
2933 	uint32_t em, addr_len;
2934 	int i, n, remaining;
2935 
2936 	/* Memory can only be accessed in naturally aligned 4 byte units */
2937 	if (addr & 3 || len & 3 || len == 0)
2938 		return (EINVAL);
2939 
2940 	/* Enabled memories */
2941 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2942 
2943 	r = &mem_ranges[0];
2944 	n = 0;
2945 	bzero(r, sizeof(mem_ranges));
2946 	if (em & F_EDRAM0_ENABLE) {
2947 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2948 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2949 		if (r->size > 0) {
2950 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2951 			if (addr >= r->start &&
2952 			    addr + len <= r->start + r->size)
2953 				return (0);
2954 			r++;
2955 			n++;
2956 		}
2957 	}
2958 	if (em & F_EDRAM1_ENABLE) {
2959 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2960 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2961 		if (r->size > 0) {
2962 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2963 			if (addr >= r->start &&
2964 			    addr + len <= r->start + r->size)
2965 				return (0);
2966 			r++;
2967 			n++;
2968 		}
2969 	}
2970 	if (em & F_EXT_MEM_ENABLE) {
2971 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2972 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2973 		if (r->size > 0) {
2974 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2975 			if (addr >= r->start &&
2976 			    addr + len <= r->start + r->size)
2977 				return (0);
2978 			r++;
2979 			n++;
2980 		}
2981 	}
2982 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2983 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2984 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2985 		if (r->size > 0) {
2986 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2987 			if (addr >= r->start &&
2988 			    addr + len <= r->start + r->size)
2989 				return (0);
2990 			r++;
2991 			n++;
2992 		}
2993 	}
2994 	MPASS(n <= nitems(mem_ranges));
2995 
2996 	if (n > 1) {
2997 		/* Sort and merge the ranges. */
2998 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2999 
3000 		/* Start from index 0 and examine the next n - 1 entries. */
3001 		r = &mem_ranges[0];
3002 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
3003 
3004 			MPASS(r->size > 0);	/* r is a valid entry. */
3005 			next = r + 1;
3006 			MPASS(next->size > 0);	/* and so is the next one. */
3007 
3008 			while (r->start + r->size >= next->start) {
3009 				/* Merge the next one into the current entry. */
3010 				r->size = max(r->start + r->size,
3011 				    next->start + next->size) - r->start;
3012 				n--;	/* One fewer entry in total. */
3013 				if (--remaining == 0)
3014 					goto done;	/* short circuit */
3015 				next++;
3016 			}
3017 			if (next != r + 1) {
3018 				/*
3019 				 * Some entries were merged into r and next
3020 				 * points to the first valid entry that couldn't
3021 				 * be merged.
3022 				 */
3023 				MPASS(next->size > 0);	/* must be valid */
3024 				memcpy(r + 1, next, remaining * sizeof(*r));
3025 #ifdef INVARIANTS
3026 				/*
3027 				 * This so that the foo->size assertion in the
3028 				 * next iteration of the loop do the right
3029 				 * thing for entries that were pulled up and are
3030 				 * no longer valid.
3031 				 */
3032 				MPASS(n < nitems(mem_ranges));
3033 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
3034 				    sizeof(struct t4_range));
3035 #endif
3036 			}
3037 		}
3038 done:
3039 		/* Done merging the ranges. */
3040 		MPASS(n > 0);
3041 		r = &mem_ranges[0];
3042 		for (i = 0; i < n; i++, r++) {
3043 			if (addr >= r->start &&
3044 			    addr + len <= r->start + r->size)
3045 				return (0);
3046 		}
3047 	}
3048 
3049 	return (EFAULT);
3050 }
3051 
3052 static int
3053 fwmtype_to_hwmtype(int mtype)
3054 {
3055 
3056 	switch (mtype) {
3057 	case FW_MEMTYPE_EDC0:
3058 		return (MEM_EDC0);
3059 	case FW_MEMTYPE_EDC1:
3060 		return (MEM_EDC1);
3061 	case FW_MEMTYPE_EXTMEM:
3062 		return (MEM_MC0);
3063 	case FW_MEMTYPE_EXTMEM1:
3064 		return (MEM_MC1);
3065 	default:
3066 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
3067 	}
3068 }
3069 
3070 /*
3071  * Verify that the memory range specified by the memtype/offset/len pair is
3072  * valid and lies entirely within the memtype specified.  The global address of
3073  * the start of the range is returned in addr.
3074  */
3075 static int
3076 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, uint32_t len,
3077     uint32_t *addr)
3078 {
3079 	uint32_t em, addr_len, maddr;
3080 
3081 	/* Memory can only be accessed in naturally aligned 4 byte units */
3082 	if (off & 3 || len & 3 || len == 0)
3083 		return (EINVAL);
3084 
3085 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
3086 	switch (fwmtype_to_hwmtype(mtype)) {
3087 	case MEM_EDC0:
3088 		if (!(em & F_EDRAM0_ENABLE))
3089 			return (EINVAL);
3090 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
3091 		maddr = G_EDRAM0_BASE(addr_len) << 20;
3092 		break;
3093 	case MEM_EDC1:
3094 		if (!(em & F_EDRAM1_ENABLE))
3095 			return (EINVAL);
3096 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
3097 		maddr = G_EDRAM1_BASE(addr_len) << 20;
3098 		break;
3099 	case MEM_MC:
3100 		if (!(em & F_EXT_MEM_ENABLE))
3101 			return (EINVAL);
3102 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
3103 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
3104 		break;
3105 	case MEM_MC1:
3106 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
3107 			return (EINVAL);
3108 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
3109 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
3110 		break;
3111 	default:
3112 		return (EINVAL);
3113 	}
3114 
3115 	*addr = maddr + off;	/* global address */
3116 	return (validate_mem_range(sc, *addr, len));
3117 }
3118 
3119 static int
3120 fixup_devlog_params(struct adapter *sc)
3121 {
3122 	struct devlog_params *dparams = &sc->params.devlog;
3123 	int rc;
3124 
3125 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
3126 	    dparams->size, &dparams->addr);
3127 
3128 	return (rc);
3129 }
3130 
3131 static void
3132 update_nirq(struct intrs_and_queues *iaq, int nports)
3133 {
3134 	int extra = T4_EXTRA_INTR;
3135 
3136 	iaq->nirq = extra;
3137 	iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
3138 	iaq->nirq += nports * (iaq->num_vis - 1) *
3139 	    max(iaq->nrxq_vi, iaq->nnmrxq_vi);
3140 	iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
3141 }
3142 
3143 /*
3144  * Adjust requirements to fit the number of interrupts available.
3145  */
3146 static void
3147 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
3148     int navail)
3149 {
3150 	int old_nirq;
3151 	const int nports = sc->params.nports;
3152 
3153 	MPASS(nports > 0);
3154 	MPASS(navail > 0);
3155 
3156 	bzero(iaq, sizeof(*iaq));
3157 	iaq->intr_type = itype;
3158 	iaq->num_vis = t4_num_vis;
3159 	iaq->ntxq = t4_ntxq;
3160 	iaq->ntxq_vi = t4_ntxq_vi;
3161 	iaq->nrxq = t4_nrxq;
3162 	iaq->nrxq_vi = t4_nrxq_vi;
3163 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
3164 	if (is_offload(sc) || is_ethoffload(sc)) {
3165 		iaq->nofldtxq = t4_nofldtxq;
3166 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
3167 	}
3168 #endif
3169 #ifdef TCP_OFFLOAD
3170 	if (is_offload(sc)) {
3171 		iaq->nofldrxq = t4_nofldrxq;
3172 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
3173 	}
3174 #endif
3175 #ifdef DEV_NETMAP
3176 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
3177 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
3178 #endif
3179 
3180 	update_nirq(iaq, nports);
3181 	if (iaq->nirq <= navail &&
3182 	    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3183 		/*
3184 		 * This is the normal case -- there are enough interrupts for
3185 		 * everything.
3186 		 */
3187 		goto done;
3188 	}
3189 
3190 	/*
3191 	 * If extra VIs have been configured try reducing their count and see if
3192 	 * that works.
3193 	 */
3194 	while (iaq->num_vis > 1) {
3195 		iaq->num_vis--;
3196 		update_nirq(iaq, nports);
3197 		if (iaq->nirq <= navail &&
3198 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3199 			device_printf(sc->dev, "virtual interfaces per port "
3200 			    "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
3201 			    "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
3202 			    "itype %d, navail %u, nirq %d.\n",
3203 			    iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
3204 			    iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
3205 			    itype, navail, iaq->nirq);
3206 			goto done;
3207 		}
3208 	}
3209 
3210 	/*
3211 	 * Extra VIs will not be created.  Log a message if they were requested.
3212 	 */
3213 	MPASS(iaq->num_vis == 1);
3214 	iaq->ntxq_vi = iaq->nrxq_vi = 0;
3215 	iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
3216 	iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
3217 	if (iaq->num_vis != t4_num_vis) {
3218 		device_printf(sc->dev, "extra virtual interfaces disabled.  "
3219 		    "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
3220 		    "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
3221 		    iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
3222 		    iaq->nnmrxq_vi, itype, navail, iaq->nirq);
3223 	}
3224 
3225 	/*
3226 	 * Keep reducing the number of NIC rx queues to the next lower power of
3227 	 * 2 (for even RSS distribution) and halving the TOE rx queues and see
3228 	 * if that works.
3229 	 */
3230 	do {
3231 		if (iaq->nrxq > 1) {
3232 			do {
3233 				iaq->nrxq--;
3234 			} while (!powerof2(iaq->nrxq));
3235 		}
3236 		if (iaq->nofldrxq > 1)
3237 			iaq->nofldrxq >>= 1;
3238 
3239 		old_nirq = iaq->nirq;
3240 		update_nirq(iaq, nports);
3241 		if (iaq->nirq <= navail &&
3242 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3243 			device_printf(sc->dev, "running with reduced number of "
3244 			    "rx queues because of shortage of interrupts.  "
3245 			    "nrxq=%u, nofldrxq=%u.  "
3246 			    "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
3247 			    iaq->nofldrxq, itype, navail, iaq->nirq);
3248 			goto done;
3249 		}
3250 	} while (old_nirq != iaq->nirq);
3251 
3252 	/* One interrupt for everything.  Ugh. */
3253 	device_printf(sc->dev, "running with minimal number of queues.  "
3254 	    "itype %d, navail %u.\n", itype, navail);
3255 	iaq->nirq = 1;
3256 	MPASS(iaq->nrxq == 1);
3257 	iaq->ntxq = 1;
3258 	if (iaq->nofldrxq > 1)
3259 		iaq->nofldtxq = 1;
3260 done:
3261 	MPASS(iaq->num_vis > 0);
3262 	if (iaq->num_vis > 1) {
3263 		MPASS(iaq->nrxq_vi > 0);
3264 		MPASS(iaq->ntxq_vi > 0);
3265 	}
3266 	MPASS(iaq->nirq > 0);
3267 	MPASS(iaq->nrxq > 0);
3268 	MPASS(iaq->ntxq > 0);
3269 	if (itype == INTR_MSI) {
3270 		MPASS(powerof2(iaq->nirq));
3271 	}
3272 }
3273 
3274 static int
3275 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
3276 {
3277 	int rc, itype, navail, nalloc;
3278 
3279 	for (itype = INTR_MSIX; itype; itype >>= 1) {
3280 
3281 		if ((itype & t4_intr_types) == 0)
3282 			continue;	/* not allowed */
3283 
3284 		if (itype == INTR_MSIX)
3285 			navail = pci_msix_count(sc->dev);
3286 		else if (itype == INTR_MSI)
3287 			navail = pci_msi_count(sc->dev);
3288 		else
3289 			navail = 1;
3290 restart:
3291 		if (navail == 0)
3292 			continue;
3293 
3294 		calculate_iaq(sc, iaq, itype, navail);
3295 		nalloc = iaq->nirq;
3296 		rc = 0;
3297 		if (itype == INTR_MSIX)
3298 			rc = pci_alloc_msix(sc->dev, &nalloc);
3299 		else if (itype == INTR_MSI)
3300 			rc = pci_alloc_msi(sc->dev, &nalloc);
3301 
3302 		if (rc == 0 && nalloc > 0) {
3303 			if (nalloc == iaq->nirq)
3304 				return (0);
3305 
3306 			/*
3307 			 * Didn't get the number requested.  Use whatever number
3308 			 * the kernel is willing to allocate.
3309 			 */
3310 			device_printf(sc->dev, "fewer vectors than requested, "
3311 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
3312 			    itype, iaq->nirq, nalloc);
3313 			pci_release_msi(sc->dev);
3314 			navail = nalloc;
3315 			goto restart;
3316 		}
3317 
3318 		device_printf(sc->dev,
3319 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
3320 		    itype, rc, iaq->nirq, nalloc);
3321 	}
3322 
3323 	device_printf(sc->dev,
3324 	    "failed to find a usable interrupt type.  "
3325 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
3326 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
3327 
3328 	return (ENXIO);
3329 }
3330 
3331 #define FW_VERSION(chip) ( \
3332     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
3333     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
3334     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
3335     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
3336 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
3337 
3338 /* Just enough of fw_hdr to cover all version info. */
3339 struct fw_h {
3340 	__u8	ver;
3341 	__u8	chip;
3342 	__be16	len512;
3343 	__be32	fw_ver;
3344 	__be32	tp_microcode_ver;
3345 	__u8	intfver_nic;
3346 	__u8	intfver_vnic;
3347 	__u8	intfver_ofld;
3348 	__u8	intfver_ri;
3349 	__u8	intfver_iscsipdu;
3350 	__u8	intfver_iscsi;
3351 	__u8	intfver_fcoepdu;
3352 	__u8	intfver_fcoe;
3353 };
3354 /* Spot check a couple of fields. */
3355 CTASSERT(offsetof(struct fw_h, fw_ver) == offsetof(struct fw_hdr, fw_ver));
3356 CTASSERT(offsetof(struct fw_h, intfver_nic) == offsetof(struct fw_hdr, intfver_nic));
3357 CTASSERT(offsetof(struct fw_h, intfver_fcoe) == offsetof(struct fw_hdr, intfver_fcoe));
3358 
3359 struct fw_info {
3360 	uint8_t chip;
3361 	char *kld_name;
3362 	char *fw_mod_name;
3363 	struct fw_h fw_h;
3364 } fw_info[] = {
3365 	{
3366 		.chip = CHELSIO_T4,
3367 		.kld_name = "t4fw_cfg",
3368 		.fw_mod_name = "t4fw",
3369 		.fw_h = {
3370 			.chip = FW_HDR_CHIP_T4,
3371 			.fw_ver = htobe32(FW_VERSION(T4)),
3372 			.intfver_nic = FW_INTFVER(T4, NIC),
3373 			.intfver_vnic = FW_INTFVER(T4, VNIC),
3374 			.intfver_ofld = FW_INTFVER(T4, OFLD),
3375 			.intfver_ri = FW_INTFVER(T4, RI),
3376 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
3377 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
3378 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
3379 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
3380 		},
3381 	}, {
3382 		.chip = CHELSIO_T5,
3383 		.kld_name = "t5fw_cfg",
3384 		.fw_mod_name = "t5fw",
3385 		.fw_h = {
3386 			.chip = FW_HDR_CHIP_T5,
3387 			.fw_ver = htobe32(FW_VERSION(T5)),
3388 			.intfver_nic = FW_INTFVER(T5, NIC),
3389 			.intfver_vnic = FW_INTFVER(T5, VNIC),
3390 			.intfver_ofld = FW_INTFVER(T5, OFLD),
3391 			.intfver_ri = FW_INTFVER(T5, RI),
3392 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
3393 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
3394 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
3395 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
3396 		},
3397 	}, {
3398 		.chip = CHELSIO_T6,
3399 		.kld_name = "t6fw_cfg",
3400 		.fw_mod_name = "t6fw",
3401 		.fw_h = {
3402 			.chip = FW_HDR_CHIP_T6,
3403 			.fw_ver = htobe32(FW_VERSION(T6)),
3404 			.intfver_nic = FW_INTFVER(T6, NIC),
3405 			.intfver_vnic = FW_INTFVER(T6, VNIC),
3406 			.intfver_ofld = FW_INTFVER(T6, OFLD),
3407 			.intfver_ri = FW_INTFVER(T6, RI),
3408 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3409 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
3410 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3411 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
3412 		},
3413 	}
3414 };
3415 
3416 static struct fw_info *
3417 find_fw_info(int chip)
3418 {
3419 	int i;
3420 
3421 	for (i = 0; i < nitems(fw_info); i++) {
3422 		if (fw_info[i].chip == chip)
3423 			return (&fw_info[i]);
3424 	}
3425 	return (NULL);
3426 }
3427 
3428 /*
3429  * Is the given firmware API compatible with the one the driver was compiled
3430  * with?
3431  */
3432 static int
3433 fw_compatible(const struct fw_h *hdr1, const struct fw_h *hdr2)
3434 {
3435 
3436 	/* short circuit if it's the exact same firmware version */
3437 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
3438 		return (1);
3439 
3440 	/*
3441 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
3442 	 * features that are supported in the driver.
3443 	 */
3444 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
3445 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
3446 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
3447 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
3448 		return (1);
3449 #undef SAME_INTF
3450 
3451 	return (0);
3452 }
3453 
3454 static int
3455 load_fw_module(struct adapter *sc, const struct firmware **dcfg,
3456     const struct firmware **fw)
3457 {
3458 	struct fw_info *fw_info;
3459 
3460 	*dcfg = NULL;
3461 	if (fw != NULL)
3462 		*fw = NULL;
3463 
3464 	fw_info = find_fw_info(chip_id(sc));
3465 	if (fw_info == NULL) {
3466 		device_printf(sc->dev,
3467 		    "unable to look up firmware information for chip %d.\n",
3468 		    chip_id(sc));
3469 		return (EINVAL);
3470 	}
3471 
3472 	*dcfg = firmware_get(fw_info->kld_name);
3473 	if (*dcfg != NULL) {
3474 		if (fw != NULL)
3475 			*fw = firmware_get(fw_info->fw_mod_name);
3476 		return (0);
3477 	}
3478 
3479 	return (ENOENT);
3480 }
3481 
3482 static void
3483 unload_fw_module(struct adapter *sc, const struct firmware *dcfg,
3484     const struct firmware *fw)
3485 {
3486 
3487 	if (fw != NULL)
3488 		firmware_put(fw, FIRMWARE_UNLOAD);
3489 	if (dcfg != NULL)
3490 		firmware_put(dcfg, FIRMWARE_UNLOAD);
3491 }
3492 
3493 /*
3494  * Return values:
3495  * 0 means no firmware install attempted.
3496  * ERESTART means a firmware install was attempted and was successful.
3497  * +ve errno means a firmware install was attempted but failed.
3498  */
3499 static int
3500 install_kld_firmware(struct adapter *sc, struct fw_h *card_fw,
3501     const struct fw_h *drv_fw, const char *reason, int *already)
3502 {
3503 	const struct firmware *cfg, *fw;
3504 	const uint32_t c = be32toh(card_fw->fw_ver);
3505 	uint32_t d, k;
3506 	int rc, fw_install;
3507 	struct fw_h bundled_fw;
3508 	bool load_attempted;
3509 
3510 	cfg = fw = NULL;
3511 	load_attempted = false;
3512 	fw_install = t4_fw_install < 0 ? -t4_fw_install : t4_fw_install;
3513 
3514 	if (reason != NULL)
3515 		goto install;
3516 
3517 	if ((sc->flags & FW_OK) == 0) {
3518 
3519 		if (c == 0xffffffff) {
3520 			reason = "missing";
3521 			goto install;
3522 		}
3523 
3524 		return (0);
3525 	}
3526 
3527 	memcpy(&bundled_fw, drv_fw, sizeof(bundled_fw));
3528 	if (t4_fw_install < 0) {
3529 		rc = load_fw_module(sc, &cfg, &fw);
3530 		if (rc != 0 || fw == NULL) {
3531 			device_printf(sc->dev,
3532 			    "failed to load firmware module: %d. cfg %p, fw %p;"
3533 			    " will use compiled-in firmware version for"
3534 			    "hw.cxgbe.fw_install checks.\n",
3535 			    rc, cfg, fw);
3536 		} else {
3537 			memcpy(&bundled_fw, fw->data, sizeof(bundled_fw));
3538 		}
3539 		load_attempted = true;
3540 	}
3541 	d = be32toh(bundled_fw.fw_ver);
3542 
3543 	if (!fw_compatible(card_fw, &bundled_fw)) {
3544 		reason = "incompatible or unusable";
3545 		goto install;
3546 	}
3547 
3548 	if (d > c) {
3549 		reason = "older than the version bundled with this driver";
3550 		goto install;
3551 	}
3552 
3553 	if (fw_install == 2 && d != c) {
3554 		reason = "different than the version bundled with this driver";
3555 		goto install;
3556 	}
3557 
3558 	/* No reason to do anything to the firmware already on the card. */
3559 	rc = 0;
3560 	goto done;
3561 
3562 install:
3563 	rc = 0;
3564 	if ((*already)++)
3565 		goto done;
3566 
3567 	if (fw_install == 0) {
3568 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3569 		    "but the driver is prohibited from installing a firmware "
3570 		    "on the card.\n",
3571 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3572 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3573 
3574 		goto done;
3575 	}
3576 
3577 	/*
3578 	 * We'll attempt to install a firmware.  Load the module first (if it
3579 	 * hasn't been loaded already).
3580 	 */
3581 	if (!load_attempted) {
3582 		rc = load_fw_module(sc, &cfg, &fw);
3583 		if (rc != 0 || fw == NULL) {
3584 			device_printf(sc->dev,
3585 			    "failed to load firmware module: %d. cfg %p, fw %p\n",
3586 			    rc, cfg, fw);
3587 			/* carry on */
3588 		}
3589 	}
3590 	if (fw == NULL) {
3591 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3592 		    "but the driver cannot take corrective action because it "
3593 		    "is unable to load the firmware module.\n",
3594 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3595 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3596 		rc = sc->flags & FW_OK ? 0 : ENOENT;
3597 		goto done;
3598 	}
3599 	k = be32toh(((const struct fw_hdr *)fw->data)->fw_ver);
3600 	if (k != d) {
3601 		MPASS(t4_fw_install > 0);
3602 		device_printf(sc->dev,
3603 		    "firmware in KLD (%u.%u.%u.%u) is not what the driver was "
3604 		    "expecting (%u.%u.%u.%u) and will not be used.\n",
3605 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3606 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k),
3607 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3608 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3609 		rc = sc->flags & FW_OK ? 0 : EINVAL;
3610 		goto done;
3611 	}
3612 
3613 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3614 	    "installing firmware %u.%u.%u.%u on card.\n",
3615 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3616 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3617 	    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3618 	    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3619 
3620 	rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3621 	if (rc != 0) {
3622 		device_printf(sc->dev, "failed to install firmware: %d\n", rc);
3623 	} else {
3624 		/* Installed successfully, update the cached header too. */
3625 		rc = ERESTART;
3626 		memcpy(card_fw, fw->data, sizeof(*card_fw));
3627 	}
3628 done:
3629 	unload_fw_module(sc, cfg, fw);
3630 
3631 	return (rc);
3632 }
3633 
3634 /*
3635  * Establish contact with the firmware and attempt to become the master driver.
3636  *
3637  * A firmware will be installed to the card if needed (if the driver is allowed
3638  * to do so).
3639  */
3640 static int
3641 contact_firmware(struct adapter *sc)
3642 {
3643 	int rc, already = 0;
3644 	enum dev_state state;
3645 	struct fw_info *fw_info;
3646 	struct fw_hdr *card_fw;		/* fw on the card */
3647 	const struct fw_h *drv_fw;
3648 
3649 	fw_info = find_fw_info(chip_id(sc));
3650 	if (fw_info == NULL) {
3651 		device_printf(sc->dev,
3652 		    "unable to look up firmware information for chip %d.\n",
3653 		    chip_id(sc));
3654 		return (EINVAL);
3655 	}
3656 	drv_fw = &fw_info->fw_h;
3657 
3658 	/* Read the header of the firmware on the card */
3659 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3660 restart:
3661 	rc = -t4_get_fw_hdr(sc, card_fw);
3662 	if (rc != 0) {
3663 		device_printf(sc->dev,
3664 		    "unable to read firmware header from card's flash: %d\n",
3665 		    rc);
3666 		goto done;
3667 	}
3668 
3669 	rc = install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw, NULL,
3670 	    &already);
3671 	if (rc == ERESTART)
3672 		goto restart;
3673 	if (rc != 0)
3674 		goto done;
3675 
3676 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3677 	if (rc < 0 || state == DEV_STATE_ERR) {
3678 		rc = -rc;
3679 		device_printf(sc->dev,
3680 		    "failed to connect to the firmware: %d, %d.  "
3681 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3682 #if 0
3683 		if (install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw,
3684 		    "not responding properly to HELLO", &already) == ERESTART)
3685 			goto restart;
3686 #endif
3687 		goto done;
3688 	}
3689 	MPASS(be32toh(card_fw->flags) & FW_HDR_FLAGS_RESET_HALT);
3690 	sc->flags |= FW_OK;	/* The firmware responded to the FW_HELLO. */
3691 
3692 	if (rc == sc->pf) {
3693 		sc->flags |= MASTER_PF;
3694 		rc = install_kld_firmware(sc, (struct fw_h *)card_fw, drv_fw,
3695 		    NULL, &already);
3696 		if (rc == ERESTART)
3697 			rc = 0;
3698 		else if (rc != 0)
3699 			goto done;
3700 	} else if (state == DEV_STATE_UNINIT) {
3701 		/*
3702 		 * We didn't get to be the master so we definitely won't be
3703 		 * configuring the chip.  It's a bug if someone else hasn't
3704 		 * configured it already.
3705 		 */
3706 		device_printf(sc->dev, "couldn't be master(%d), "
3707 		    "device not already initialized either(%d).  "
3708 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3709 		rc = EPROTO;
3710 		goto done;
3711 	} else {
3712 		/*
3713 		 * Some other PF is the master and has configured the chip.
3714 		 * This is allowed but untested.
3715 		 */
3716 		device_printf(sc->dev, "PF%d is master, device state %d.  "
3717 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3718 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", rc);
3719 		sc->cfcsum = 0;
3720 		rc = 0;
3721 	}
3722 done:
3723 	if (rc != 0 && sc->flags & FW_OK) {
3724 		t4_fw_bye(sc, sc->mbox);
3725 		sc->flags &= ~FW_OK;
3726 	}
3727 	free(card_fw, M_CXGBE);
3728 	return (rc);
3729 }
3730 
3731 static int
3732 copy_cfg_file_to_card(struct adapter *sc, char *cfg_file,
3733     uint32_t mtype, uint32_t moff)
3734 {
3735 	struct fw_info *fw_info;
3736 	const struct firmware *dcfg, *rcfg = NULL;
3737 	const uint32_t *cfdata;
3738 	uint32_t cflen, addr;
3739 	int rc;
3740 
3741 	load_fw_module(sc, &dcfg, NULL);
3742 
3743 	/* Card specific interpretation of "default". */
3744 	if (strncmp(cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3745 		if (pci_get_device(sc->dev) == 0x440a)
3746 			snprintf(cfg_file, sizeof(t4_cfg_file), UWIRE_CF);
3747 		if (is_fpga(sc))
3748 			snprintf(cfg_file, sizeof(t4_cfg_file), FPGA_CF);
3749 	}
3750 
3751 	if (strncmp(cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3752 		if (dcfg == NULL) {
3753 			device_printf(sc->dev,
3754 			    "KLD with default config is not available.\n");
3755 			rc = ENOENT;
3756 			goto done;
3757 		}
3758 		cfdata = dcfg->data;
3759 		cflen = dcfg->datasize & ~3;
3760 	} else {
3761 		char s[32];
3762 
3763 		fw_info = find_fw_info(chip_id(sc));
3764 		if (fw_info == NULL) {
3765 			device_printf(sc->dev,
3766 			    "unable to look up firmware information for chip %d.\n",
3767 			    chip_id(sc));
3768 			rc = EINVAL;
3769 			goto done;
3770 		}
3771 		snprintf(s, sizeof(s), "%s_%s", fw_info->kld_name, cfg_file);
3772 
3773 		rcfg = firmware_get(s);
3774 		if (rcfg == NULL) {
3775 			device_printf(sc->dev,
3776 			    "unable to load module \"%s\" for configuration "
3777 			    "profile \"%s\".\n", s, cfg_file);
3778 			rc = ENOENT;
3779 			goto done;
3780 		}
3781 		cfdata = rcfg->data;
3782 		cflen = rcfg->datasize & ~3;
3783 	}
3784 
3785 	if (cflen > FLASH_CFG_MAX_SIZE) {
3786 		device_printf(sc->dev,
3787 		    "config file too long (%d, max allowed is %d).\n",
3788 		    cflen, FLASH_CFG_MAX_SIZE);
3789 		rc = EINVAL;
3790 		goto done;
3791 	}
3792 
3793 	rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3794 	if (rc != 0) {
3795 		device_printf(sc->dev,
3796 		    "%s: addr (%d/0x%x) or len %d is not valid: %d.\n",
3797 		    __func__, mtype, moff, cflen, rc);
3798 		rc = EINVAL;
3799 		goto done;
3800 	}
3801 	write_via_memwin(sc, 2, addr, cfdata, cflen);
3802 done:
3803 	if (rcfg != NULL)
3804 		firmware_put(rcfg, FIRMWARE_UNLOAD);
3805 	unload_fw_module(sc, dcfg, NULL);
3806 	return (rc);
3807 }
3808 
3809 struct caps_allowed {
3810 	uint16_t nbmcaps;
3811 	uint16_t linkcaps;
3812 	uint16_t switchcaps;
3813 	uint16_t niccaps;
3814 	uint16_t toecaps;
3815 	uint16_t rdmacaps;
3816 	uint16_t cryptocaps;
3817 	uint16_t iscsicaps;
3818 	uint16_t fcoecaps;
3819 };
3820 
3821 #define FW_PARAM_DEV(param) \
3822 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3823 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3824 #define FW_PARAM_PFVF(param) \
3825 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3826 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3827 
3828 /*
3829  * Provide a configuration profile to the firmware and have it initialize the
3830  * chip accordingly.  This may involve uploading a configuration file to the
3831  * card.
3832  */
3833 static int
3834 apply_cfg_and_initialize(struct adapter *sc, char *cfg_file,
3835     const struct caps_allowed *caps_allowed)
3836 {
3837 	int rc;
3838 	struct fw_caps_config_cmd caps;
3839 	uint32_t mtype, moff, finicsum, cfcsum, param, val;
3840 
3841 	rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
3842 	if (rc != 0) {
3843 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3844 		return (rc);
3845 	}
3846 
3847 	bzero(&caps, sizeof(caps));
3848 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3849 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3850 	if (strncmp(cfg_file, BUILTIN_CF, sizeof(t4_cfg_file)) == 0) {
3851 		mtype = 0;
3852 		moff = 0;
3853 		caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3854 	} else if (strncmp(cfg_file, FLASH_CF, sizeof(t4_cfg_file)) == 0) {
3855 		mtype = FW_MEMTYPE_FLASH;
3856 		moff = t4_flash_cfg_addr(sc);
3857 		caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3858 		    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3859 		    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) |
3860 		    FW_LEN16(caps));
3861 	} else {
3862 		/*
3863 		 * Ask the firmware where it wants us to upload the config file.
3864 		 */
3865 		param = FW_PARAM_DEV(CF);
3866 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3867 		if (rc != 0) {
3868 			/* No support for config file?  Shouldn't happen. */
3869 			device_printf(sc->dev,
3870 			    "failed to query config file location: %d.\n", rc);
3871 			goto done;
3872 		}
3873 		mtype = G_FW_PARAMS_PARAM_Y(val);
3874 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
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 
3880 		rc = copy_cfg_file_to_card(sc, cfg_file, mtype, moff);
3881 		if (rc != 0) {
3882 			device_printf(sc->dev,
3883 			    "failed to upload config file to card: %d.\n", rc);
3884 			goto done;
3885 		}
3886 	}
3887 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3888 	if (rc != 0) {
3889 		device_printf(sc->dev, "failed to pre-process config file: %d "
3890 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3891 		goto done;
3892 	}
3893 
3894 	finicsum = be32toh(caps.finicsum);
3895 	cfcsum = be32toh(caps.cfcsum);	/* actual */
3896 	if (finicsum != cfcsum) {
3897 		device_printf(sc->dev,
3898 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3899 		    finicsum, cfcsum);
3900 	}
3901 	sc->cfcsum = cfcsum;
3902 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", cfg_file);
3903 
3904 	/*
3905 	 * Let the firmware know what features will (not) be used so it can tune
3906 	 * things accordingly.
3907 	 */
3908 #define LIMIT_CAPS(x) do { \
3909 	caps.x##caps &= htobe16(caps_allowed->x##caps); \
3910 } while (0)
3911 	LIMIT_CAPS(nbm);
3912 	LIMIT_CAPS(link);
3913 	LIMIT_CAPS(switch);
3914 	LIMIT_CAPS(nic);
3915 	LIMIT_CAPS(toe);
3916 	LIMIT_CAPS(rdma);
3917 	LIMIT_CAPS(crypto);
3918 	LIMIT_CAPS(iscsi);
3919 	LIMIT_CAPS(fcoe);
3920 #undef LIMIT_CAPS
3921 	if (caps.niccaps & htobe16(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
3922 		/*
3923 		 * TOE and hashfilters are mutually exclusive.  It is a config
3924 		 * file or firmware bug if both are reported as available.  Try
3925 		 * to cope with the situation in non-debug builds by disabling
3926 		 * TOE.
3927 		 */
3928 		MPASS(caps.toecaps == 0);
3929 
3930 		caps.toecaps = 0;
3931 		caps.rdmacaps = 0;
3932 		caps.iscsicaps = 0;
3933 	}
3934 
3935 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3936 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3937 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3938 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3939 	if (rc != 0) {
3940 		device_printf(sc->dev,
3941 		    "failed to process config file: %d.\n", rc);
3942 		goto done;
3943 	}
3944 
3945 	t4_tweak_chip_settings(sc);
3946 
3947 	/* get basic stuff going */
3948 	rc = -t4_fw_initialize(sc, sc->mbox);
3949 	if (rc != 0) {
3950 		device_printf(sc->dev, "fw_initialize failed: %d.\n", rc);
3951 		goto done;
3952 	}
3953 done:
3954 	return (rc);
3955 }
3956 
3957 /*
3958  * Partition chip resources for use between various PFs, VFs, etc.
3959  */
3960 static int
3961 partition_resources(struct adapter *sc)
3962 {
3963 	char cfg_file[sizeof(t4_cfg_file)];
3964 	struct caps_allowed caps_allowed;
3965 	int rc;
3966 	bool fallback;
3967 
3968 	/* Only the master driver gets to configure the chip resources. */
3969 	MPASS(sc->flags & MASTER_PF);
3970 
3971 #define COPY_CAPS(x) do { \
3972 	caps_allowed.x##caps = t4_##x##caps_allowed; \
3973 } while (0)
3974 	bzero(&caps_allowed, sizeof(caps_allowed));
3975 	COPY_CAPS(nbm);
3976 	COPY_CAPS(link);
3977 	COPY_CAPS(switch);
3978 	COPY_CAPS(nic);
3979 	COPY_CAPS(toe);
3980 	COPY_CAPS(rdma);
3981 	COPY_CAPS(crypto);
3982 	COPY_CAPS(iscsi);
3983 	COPY_CAPS(fcoe);
3984 	fallback = sc->debug_flags & DF_DISABLE_CFG_RETRY ? false : true;
3985 	snprintf(cfg_file, sizeof(cfg_file), "%s", t4_cfg_file);
3986 retry:
3987 	rc = apply_cfg_and_initialize(sc, cfg_file, &caps_allowed);
3988 	if (rc != 0 && fallback) {
3989 		device_printf(sc->dev,
3990 		    "failed (%d) to configure card with \"%s\" profile, "
3991 		    "will fall back to a basic configuration and retry.\n",
3992 		    rc, cfg_file);
3993 		snprintf(cfg_file, sizeof(cfg_file), "%s", BUILTIN_CF);
3994 		bzero(&caps_allowed, sizeof(caps_allowed));
3995 		COPY_CAPS(nbm);
3996 		COPY_CAPS(link);
3997 		COPY_CAPS(switch);
3998 		COPY_CAPS(nic);
3999 		fallback = false;
4000 		goto retry;
4001 	}
4002 #undef COPY_CAPS
4003 	return (rc);
4004 }
4005 
4006 /*
4007  * Retrieve parameters that are needed (or nice to have) very early.
4008  */
4009 static int
4010 get_params__pre_init(struct adapter *sc)
4011 {
4012 	int rc;
4013 	uint32_t param[2], val[2];
4014 
4015 	t4_get_version_info(sc);
4016 
4017 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
4018 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
4019 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
4020 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
4021 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
4022 
4023 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
4024 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
4025 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
4026 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
4027 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
4028 
4029 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
4030 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
4031 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
4032 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
4033 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
4034 
4035 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
4036 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
4037 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
4038 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
4039 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
4040 
4041 	param[0] = FW_PARAM_DEV(PORTVEC);
4042 	param[1] = FW_PARAM_DEV(CCLK);
4043 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4044 	if (rc != 0) {
4045 		device_printf(sc->dev,
4046 		    "failed to query parameters (pre_init): %d.\n", rc);
4047 		return (rc);
4048 	}
4049 
4050 	sc->params.portvec = val[0];
4051 	sc->params.nports = bitcount32(val[0]);
4052 	sc->params.vpd.cclk = val[1];
4053 
4054 	/* Read device log parameters. */
4055 	rc = -t4_init_devlog_params(sc, 1);
4056 	if (rc == 0)
4057 		fixup_devlog_params(sc);
4058 	else {
4059 		device_printf(sc->dev,
4060 		    "failed to get devlog parameters: %d.\n", rc);
4061 		rc = 0;	/* devlog isn't critical for device operation */
4062 	}
4063 
4064 	return (rc);
4065 }
4066 
4067 /*
4068  * Retrieve various parameters that are of interest to the driver.  The device
4069  * has been initialized by the firmware at this point.
4070  */
4071 static int
4072 get_params__post_init(struct adapter *sc)
4073 {
4074 	int rc;
4075 	uint32_t param[7], val[7];
4076 	struct fw_caps_config_cmd caps;
4077 
4078 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
4079 	param[1] = FW_PARAM_PFVF(EQ_START);
4080 	param[2] = FW_PARAM_PFVF(FILTER_START);
4081 	param[3] = FW_PARAM_PFVF(FILTER_END);
4082 	param[4] = FW_PARAM_PFVF(L2T_START);
4083 	param[5] = FW_PARAM_PFVF(L2T_END);
4084 	param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4085 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
4086 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
4087 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
4088 	if (rc != 0) {
4089 		device_printf(sc->dev,
4090 		    "failed to query parameters (post_init): %d.\n", rc);
4091 		return (rc);
4092 	}
4093 
4094 	sc->sge.iq_start = val[0];
4095 	sc->sge.eq_start = val[1];
4096 	if ((int)val[3] > (int)val[2]) {
4097 		sc->tids.ftid_base = val[2];
4098 		sc->tids.ftid_end = val[3];
4099 		sc->tids.nftids = val[3] - val[2] + 1;
4100 	}
4101 	sc->vres.l2t.start = val[4];
4102 	sc->vres.l2t.size = val[5] - val[4] + 1;
4103 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
4104 	    ("%s: L2 table size (%u) larger than expected (%u)",
4105 	    __func__, sc->vres.l2t.size, L2T_SIZE));
4106 	sc->params.core_vdd = val[6];
4107 
4108 	if (chip_id(sc) >= CHELSIO_T6) {
4109 
4110 #ifdef INVARIANTS
4111 		if (sc->params.fw_vers >=
4112 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4113 		    V_FW_HDR_FW_VER_MICRO(1) | V_FW_HDR_FW_VER_BUILD(0))) {
4114 			/*
4115 			 * Note that the code to enable the region should run
4116 			 * before t4_fw_initialize and not here.  This is just a
4117 			 * reminder to add said code.
4118 			 */
4119 			device_printf(sc->dev,
4120 			    "hpfilter region not enabled.\n");
4121 		}
4122 #endif
4123 
4124 		sc->tids.tid_base = t4_read_reg(sc,
4125 		    A_LE_DB_ACTIVE_TABLE_START_INDEX);
4126 
4127 		param[0] = FW_PARAM_PFVF(HPFILTER_START);
4128 		param[1] = FW_PARAM_PFVF(HPFILTER_END);
4129 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4130 		if (rc != 0) {
4131 			device_printf(sc->dev,
4132 			   "failed to query hpfilter parameters: %d.\n", rc);
4133 			return (rc);
4134 		}
4135 		if ((int)val[1] > (int)val[0]) {
4136 			sc->tids.hpftid_base = val[0];
4137 			sc->tids.hpftid_end = val[1];
4138 			sc->tids.nhpftids = val[1] - val[0] + 1;
4139 
4140 			/*
4141 			 * These should go off if the layout changes and the
4142 			 * driver needs to catch up.
4143 			 */
4144 			MPASS(sc->tids.hpftid_base == 0);
4145 			MPASS(sc->tids.tid_base == sc->tids.nhpftids);
4146 		}
4147 	}
4148 
4149 	/*
4150 	 * MPSBGMAP is queried separately because only recent firmwares support
4151 	 * it as a parameter and we don't want the compound query above to fail
4152 	 * on older firmwares.
4153 	 */
4154 	param[0] = FW_PARAM_DEV(MPSBGMAP);
4155 	val[0] = 0;
4156 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4157 	if (rc == 0)
4158 		sc->params.mps_bg_map = val[0];
4159 	else
4160 		sc->params.mps_bg_map = 0;
4161 
4162 	/*
4163 	 * Determine whether the firmware supports the filter2 work request.
4164 	 * This is queried separately for the same reason as MPSBGMAP above.
4165 	 */
4166 	param[0] = FW_PARAM_DEV(FILTER2_WR);
4167 	val[0] = 0;
4168 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4169 	if (rc == 0)
4170 		sc->params.filter2_wr_support = val[0] != 0;
4171 	else
4172 		sc->params.filter2_wr_support = 0;
4173 
4174 	/*
4175 	 * Find out whether we're allowed to use the ULPTX MEMWRITE DSGL.
4176 	 * This is queried separately for the same reason as other params above.
4177 	 */
4178 	param[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
4179 	val[0] = 0;
4180 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4181 	if (rc == 0)
4182 		sc->params.ulptx_memwrite_dsgl = val[0] != 0;
4183 	else
4184 		sc->params.ulptx_memwrite_dsgl = false;
4185 
4186 	/* get capabilites */
4187 	bzero(&caps, sizeof(caps));
4188 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4189 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
4190 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
4191 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
4192 	if (rc != 0) {
4193 		device_printf(sc->dev,
4194 		    "failed to get card capabilities: %d.\n", rc);
4195 		return (rc);
4196 	}
4197 
4198 #define READ_CAPS(x) do { \
4199 	sc->x = htobe16(caps.x); \
4200 } while (0)
4201 	READ_CAPS(nbmcaps);
4202 	READ_CAPS(linkcaps);
4203 	READ_CAPS(switchcaps);
4204 	READ_CAPS(niccaps);
4205 	READ_CAPS(toecaps);
4206 	READ_CAPS(rdmacaps);
4207 	READ_CAPS(cryptocaps);
4208 	READ_CAPS(iscsicaps);
4209 	READ_CAPS(fcoecaps);
4210 
4211 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_HASHFILTER) {
4212 		MPASS(chip_id(sc) > CHELSIO_T4);
4213 		MPASS(sc->toecaps == 0);
4214 		sc->toecaps = 0;
4215 
4216 		param[0] = FW_PARAM_DEV(NTID);
4217 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4218 		if (rc != 0) {
4219 			device_printf(sc->dev,
4220 			    "failed to query HASHFILTER parameters: %d.\n", rc);
4221 			return (rc);
4222 		}
4223 		sc->tids.ntids = val[0];
4224 		if (sc->params.fw_vers <
4225 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4226 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4227 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4228 			sc->tids.ntids -= sc->tids.nhpftids;
4229 		}
4230 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4231 		sc->params.hash_filter = 1;
4232 	}
4233 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
4234 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
4235 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
4236 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4237 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
4238 		if (rc != 0) {
4239 			device_printf(sc->dev,
4240 			    "failed to query NIC parameters: %d.\n", rc);
4241 			return (rc);
4242 		}
4243 		if ((int)val[1] > (int)val[0]) {
4244 			sc->tids.etid_base = val[0];
4245 			sc->tids.etid_end = val[1];
4246 			sc->tids.netids = val[1] - val[0] + 1;
4247 			sc->params.eo_wr_cred = val[2];
4248 			sc->params.ethoffload = 1;
4249 		}
4250 	}
4251 	if (sc->toecaps) {
4252 		/* query offload-related parameters */
4253 		param[0] = FW_PARAM_DEV(NTID);
4254 		param[1] = FW_PARAM_PFVF(SERVER_START);
4255 		param[2] = FW_PARAM_PFVF(SERVER_END);
4256 		param[3] = FW_PARAM_PFVF(TDDP_START);
4257 		param[4] = FW_PARAM_PFVF(TDDP_END);
4258 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4259 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4260 		if (rc != 0) {
4261 			device_printf(sc->dev,
4262 			    "failed to query TOE parameters: %d.\n", rc);
4263 			return (rc);
4264 		}
4265 		sc->tids.ntids = val[0];
4266 		if (sc->params.fw_vers <
4267 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4268 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4269 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4270 			sc->tids.ntids -= sc->tids.nhpftids;
4271 		}
4272 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4273 		if ((int)val[2] > (int)val[1]) {
4274 			sc->tids.stid_base = val[1];
4275 			sc->tids.nstids = val[2] - val[1] + 1;
4276 		}
4277 		sc->vres.ddp.start = val[3];
4278 		sc->vres.ddp.size = val[4] - val[3] + 1;
4279 		sc->params.ofldq_wr_cred = val[5];
4280 		sc->params.offload = 1;
4281 	} else {
4282 		/*
4283 		 * The firmware attempts memfree TOE configuration for -SO cards
4284 		 * and will report toecaps=0 if it runs out of resources (this
4285 		 * depends on the config file).  It may not report 0 for other
4286 		 * capabilities dependent on the TOE in this case.  Set them to
4287 		 * 0 here so that the driver doesn't bother tracking resources
4288 		 * that will never be used.
4289 		 */
4290 		sc->iscsicaps = 0;
4291 		sc->rdmacaps = 0;
4292 	}
4293 	if (sc->rdmacaps) {
4294 		param[0] = FW_PARAM_PFVF(STAG_START);
4295 		param[1] = FW_PARAM_PFVF(STAG_END);
4296 		param[2] = FW_PARAM_PFVF(RQ_START);
4297 		param[3] = FW_PARAM_PFVF(RQ_END);
4298 		param[4] = FW_PARAM_PFVF(PBL_START);
4299 		param[5] = FW_PARAM_PFVF(PBL_END);
4300 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4301 		if (rc != 0) {
4302 			device_printf(sc->dev,
4303 			    "failed to query RDMA parameters(1): %d.\n", rc);
4304 			return (rc);
4305 		}
4306 		sc->vres.stag.start = val[0];
4307 		sc->vres.stag.size = val[1] - val[0] + 1;
4308 		sc->vres.rq.start = val[2];
4309 		sc->vres.rq.size = val[3] - val[2] + 1;
4310 		sc->vres.pbl.start = val[4];
4311 		sc->vres.pbl.size = val[5] - val[4] + 1;
4312 
4313 		param[0] = FW_PARAM_PFVF(SQRQ_START);
4314 		param[1] = FW_PARAM_PFVF(SQRQ_END);
4315 		param[2] = FW_PARAM_PFVF(CQ_START);
4316 		param[3] = FW_PARAM_PFVF(CQ_END);
4317 		param[4] = FW_PARAM_PFVF(OCQ_START);
4318 		param[5] = FW_PARAM_PFVF(OCQ_END);
4319 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4320 		if (rc != 0) {
4321 			device_printf(sc->dev,
4322 			    "failed to query RDMA parameters(2): %d.\n", rc);
4323 			return (rc);
4324 		}
4325 		sc->vres.qp.start = val[0];
4326 		sc->vres.qp.size = val[1] - val[0] + 1;
4327 		sc->vres.cq.start = val[2];
4328 		sc->vres.cq.size = val[3] - val[2] + 1;
4329 		sc->vres.ocq.start = val[4];
4330 		sc->vres.ocq.size = val[5] - val[4] + 1;
4331 
4332 		param[0] = FW_PARAM_PFVF(SRQ_START);
4333 		param[1] = FW_PARAM_PFVF(SRQ_END);
4334 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
4335 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4336 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
4337 		if (rc != 0) {
4338 			device_printf(sc->dev,
4339 			    "failed to query RDMA parameters(3): %d.\n", rc);
4340 			return (rc);
4341 		}
4342 		sc->vres.srq.start = val[0];
4343 		sc->vres.srq.size = val[1] - val[0] + 1;
4344 		sc->params.max_ordird_qp = val[2];
4345 		sc->params.max_ird_adapter = val[3];
4346 	}
4347 	if (sc->iscsicaps) {
4348 		param[0] = FW_PARAM_PFVF(ISCSI_START);
4349 		param[1] = FW_PARAM_PFVF(ISCSI_END);
4350 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4351 		if (rc != 0) {
4352 			device_printf(sc->dev,
4353 			    "failed to query iSCSI parameters: %d.\n", rc);
4354 			return (rc);
4355 		}
4356 		sc->vres.iscsi.start = val[0];
4357 		sc->vres.iscsi.size = val[1] - val[0] + 1;
4358 	}
4359 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
4360 		param[0] = FW_PARAM_PFVF(TLS_START);
4361 		param[1] = FW_PARAM_PFVF(TLS_END);
4362 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4363 		if (rc != 0) {
4364 			device_printf(sc->dev,
4365 			    "failed to query TLS parameters: %d.\n", rc);
4366 			return (rc);
4367 		}
4368 		sc->vres.key.start = val[0];
4369 		sc->vres.key.size = val[1] - val[0] + 1;
4370 	}
4371 
4372 	t4_init_sge_params(sc);
4373 
4374 	/*
4375 	 * We've got the params we wanted to query via the firmware.  Now grab
4376 	 * some others directly from the chip.
4377 	 */
4378 	rc = t4_read_chip_settings(sc);
4379 
4380 	return (rc);
4381 }
4382 
4383 static int
4384 set_params__post_init(struct adapter *sc)
4385 {
4386 	uint32_t param, val;
4387 #ifdef TCP_OFFLOAD
4388 	int i, v, shift;
4389 #endif
4390 
4391 	/* ask for encapsulated CPLs */
4392 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4393 	val = 1;
4394 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4395 
4396 	/* Enable 32b port caps if the firmware supports it. */
4397 	param = FW_PARAM_PFVF(PORT_CAPS32);
4398 	val = 1;
4399 	if (t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val) == 0)
4400 		sc->params.port_caps32 = 1;
4401 
4402 	/* Let filter + maskhash steer to a part of the VI's RSS region. */
4403 	val = 1 << (G_MASKSIZE(t4_read_reg(sc, A_TP_RSS_CONFIG_TNL)) - 1);
4404 	t4_set_reg_field(sc, A_TP_RSS_CONFIG_TNL, V_MASKFILTER(M_MASKFILTER),
4405 	    V_MASKFILTER(val - 1));
4406 
4407 #ifdef TCP_OFFLOAD
4408 	/*
4409 	 * Override the TOE timers with user provided tunables.  This is not the
4410 	 * recommended way to change the timers (the firmware config file is) so
4411 	 * these tunables are not documented.
4412 	 *
4413 	 * All the timer tunables are in microseconds.
4414 	 */
4415 	if (t4_toe_keepalive_idle != 0) {
4416 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
4417 		v &= M_KEEPALIVEIDLE;
4418 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
4419 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
4420 	}
4421 	if (t4_toe_keepalive_interval != 0) {
4422 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
4423 		v &= M_KEEPALIVEINTVL;
4424 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
4425 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
4426 	}
4427 	if (t4_toe_keepalive_count != 0) {
4428 		v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
4429 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4430 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
4431 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
4432 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
4433 	}
4434 	if (t4_toe_rexmt_min != 0) {
4435 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
4436 		v &= M_RXTMIN;
4437 		t4_set_reg_field(sc, A_TP_RXT_MIN,
4438 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
4439 	}
4440 	if (t4_toe_rexmt_max != 0) {
4441 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
4442 		v &= M_RXTMAX;
4443 		t4_set_reg_field(sc, A_TP_RXT_MAX,
4444 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
4445 	}
4446 	if (t4_toe_rexmt_count != 0) {
4447 		v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
4448 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4449 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
4450 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
4451 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
4452 	}
4453 	for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
4454 		if (t4_toe_rexmt_backoff[i] != -1) {
4455 			v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
4456 			shift = (i & 3) << 3;
4457 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
4458 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
4459 		}
4460 	}
4461 #endif
4462 	return (0);
4463 }
4464 
4465 #undef FW_PARAM_PFVF
4466 #undef FW_PARAM_DEV
4467 
4468 static void
4469 t4_set_desc(struct adapter *sc)
4470 {
4471 	char buf[128];
4472 	struct adapter_params *p = &sc->params;
4473 
4474 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
4475 
4476 	device_set_desc_copy(sc->dev, buf);
4477 }
4478 
4479 static inline void
4480 ifmedia_add4(struct ifmedia *ifm, int m)
4481 {
4482 
4483 	ifmedia_add(ifm, m, 0, NULL);
4484 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE, 0, NULL);
4485 	ifmedia_add(ifm, m | IFM_ETH_RXPAUSE, 0, NULL);
4486 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE, 0, NULL);
4487 }
4488 
4489 /*
4490  * This is the selected media, which is not quite the same as the active media.
4491  * The media line in ifconfig is "media: Ethernet selected (active)" if selected
4492  * and active are not the same, and "media: Ethernet selected" otherwise.
4493  */
4494 static void
4495 set_current_media(struct port_info *pi)
4496 {
4497 	struct link_config *lc;
4498 	struct ifmedia *ifm;
4499 	int mword;
4500 	u_int speed;
4501 
4502 	PORT_LOCK_ASSERT_OWNED(pi);
4503 
4504 	/* Leave current media alone if it's already set to IFM_NONE. */
4505 	ifm = &pi->media;
4506 	if (ifm->ifm_cur != NULL &&
4507 	    IFM_SUBTYPE(ifm->ifm_cur->ifm_media) == IFM_NONE)
4508 		return;
4509 
4510 	lc = &pi->link_cfg;
4511 	if (lc->requested_aneg != AUTONEG_DISABLE &&
4512 	    lc->supported & FW_PORT_CAP32_ANEG) {
4513 		ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
4514 		return;
4515 	}
4516 	mword = IFM_ETHER | IFM_FDX;
4517 	if (lc->requested_fc & PAUSE_TX)
4518 		mword |= IFM_ETH_TXPAUSE;
4519 	if (lc->requested_fc & PAUSE_RX)
4520 		mword |= IFM_ETH_RXPAUSE;
4521 	if (lc->requested_speed == 0)
4522 		speed = port_top_speed(pi) * 1000;	/* Gbps -> Mbps */
4523 	else
4524 		speed = lc->requested_speed;
4525 	mword |= port_mword(pi, speed_to_fwcap(speed));
4526 	ifmedia_set(ifm, mword);
4527 }
4528 
4529 /*
4530  * Returns true if the ifmedia list for the port cannot change.
4531  */
4532 static bool
4533 fixed_ifmedia(struct port_info *pi)
4534 {
4535 
4536 	return (pi->port_type == FW_PORT_TYPE_BT_SGMII ||
4537 	    pi->port_type == FW_PORT_TYPE_BT_XFI ||
4538 	    pi->port_type == FW_PORT_TYPE_BT_XAUI ||
4539 	    pi->port_type == FW_PORT_TYPE_KX4 ||
4540 	    pi->port_type == FW_PORT_TYPE_KX ||
4541 	    pi->port_type == FW_PORT_TYPE_KR ||
4542 	    pi->port_type == FW_PORT_TYPE_BP_AP ||
4543 	    pi->port_type == FW_PORT_TYPE_BP4_AP ||
4544 	    pi->port_type == FW_PORT_TYPE_BP40_BA ||
4545 	    pi->port_type == FW_PORT_TYPE_KR4_100G ||
4546 	    pi->port_type == FW_PORT_TYPE_KR_SFP28 ||
4547 	    pi->port_type == FW_PORT_TYPE_KR_XLAUI);
4548 }
4549 
4550 static void
4551 build_medialist(struct port_info *pi)
4552 {
4553 	uint32_t ss, speed;
4554 	int unknown, mword, bit;
4555 	struct link_config *lc;
4556 	struct ifmedia *ifm;
4557 
4558 	PORT_LOCK_ASSERT_OWNED(pi);
4559 
4560 	if (pi->flags & FIXED_IFMEDIA)
4561 		return;
4562 
4563 	/*
4564 	 * Rebuild the ifmedia list.
4565 	 */
4566 	ifm = &pi->media;
4567 	ifmedia_removeall(ifm);
4568 	lc = &pi->link_cfg;
4569 	ss = G_FW_PORT_CAP32_SPEED(lc->supported); /* Supported Speeds */
4570 	if (__predict_false(ss == 0)) {	/* not supposed to happen. */
4571 		MPASS(ss != 0);
4572 no_media:
4573 		MPASS(LIST_EMPTY(&ifm->ifm_list));
4574 		ifmedia_add(ifm, IFM_ETHER | IFM_NONE, 0, NULL);
4575 		ifmedia_set(ifm, IFM_ETHER | IFM_NONE);
4576 		return;
4577 	}
4578 
4579 	unknown = 0;
4580 	for (bit = S_FW_PORT_CAP32_SPEED; bit < fls(ss); bit++) {
4581 		speed = 1 << bit;
4582 		MPASS(speed & M_FW_PORT_CAP32_SPEED);
4583 		if (ss & speed) {
4584 			mword = port_mword(pi, speed);
4585 			if (mword == IFM_NONE) {
4586 				goto no_media;
4587 			} else if (mword == IFM_UNKNOWN)
4588 				unknown++;
4589 			else
4590 				ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | mword);
4591 		}
4592 	}
4593 	if (unknown > 0) /* Add one unknown for all unknown media types. */
4594 		ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | IFM_UNKNOWN);
4595 	if (lc->supported & FW_PORT_CAP32_ANEG)
4596 		ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
4597 
4598 	set_current_media(pi);
4599 }
4600 
4601 /*
4602  * Initialize the requested fields in the link config based on driver tunables.
4603  */
4604 static void
4605 init_link_config(struct port_info *pi)
4606 {
4607 	struct link_config *lc = &pi->link_cfg;
4608 
4609 	PORT_LOCK_ASSERT_OWNED(pi);
4610 
4611 	lc->requested_speed = 0;
4612 
4613 	if (t4_autoneg == 0)
4614 		lc->requested_aneg = AUTONEG_DISABLE;
4615 	else if (t4_autoneg == 1)
4616 		lc->requested_aneg = AUTONEG_ENABLE;
4617 	else
4618 		lc->requested_aneg = AUTONEG_AUTO;
4619 
4620 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX |
4621 	    PAUSE_AUTONEG);
4622 
4623 	if (t4_fec == -1 || t4_fec & FEC_AUTO)
4624 		lc->requested_fec = FEC_AUTO;
4625 	else {
4626 		lc->requested_fec = FEC_NONE;
4627 		if (t4_fec & FEC_RS)
4628 			lc->requested_fec |= FEC_RS;
4629 		if (t4_fec & FEC_BASER_RS)
4630 			lc->requested_fec |= FEC_BASER_RS;
4631 	}
4632 }
4633 
4634 /*
4635  * Makes sure that all requested settings comply with what's supported by the
4636  * port.  Returns the number of settings that were invalid and had to be fixed.
4637  */
4638 static int
4639 fixup_link_config(struct port_info *pi)
4640 {
4641 	int n = 0;
4642 	struct link_config *lc = &pi->link_cfg;
4643 	uint32_t fwspeed;
4644 
4645 	PORT_LOCK_ASSERT_OWNED(pi);
4646 
4647 	/* Speed (when not autonegotiating) */
4648 	if (lc->requested_speed != 0) {
4649 		fwspeed = speed_to_fwcap(lc->requested_speed);
4650 		if ((fwspeed & lc->supported) == 0) {
4651 			n++;
4652 			lc->requested_speed = 0;
4653 		}
4654 	}
4655 
4656 	/* Link autonegotiation */
4657 	MPASS(lc->requested_aneg == AUTONEG_ENABLE ||
4658 	    lc->requested_aneg == AUTONEG_DISABLE ||
4659 	    lc->requested_aneg == AUTONEG_AUTO);
4660 	if (lc->requested_aneg == AUTONEG_ENABLE &&
4661 	    !(lc->supported & FW_PORT_CAP32_ANEG)) {
4662 		n++;
4663 		lc->requested_aneg = AUTONEG_AUTO;
4664 	}
4665 
4666 	/* Flow control */
4667 	MPASS((lc->requested_fc & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG)) == 0);
4668 	if (lc->requested_fc & PAUSE_TX &&
4669 	    !(lc->supported & FW_PORT_CAP32_FC_TX)) {
4670 		n++;
4671 		lc->requested_fc &= ~PAUSE_TX;
4672 	}
4673 	if (lc->requested_fc & PAUSE_RX &&
4674 	    !(lc->supported & FW_PORT_CAP32_FC_RX)) {
4675 		n++;
4676 		lc->requested_fc &= ~PAUSE_RX;
4677 	}
4678 	if (!(lc->requested_fc & PAUSE_AUTONEG) &&
4679 	    !(lc->supported & FW_PORT_CAP32_FORCE_PAUSE)) {
4680 		n++;
4681 		lc->requested_fc |= PAUSE_AUTONEG;
4682 	}
4683 
4684 	/* FEC */
4685 	if ((lc->requested_fec & FEC_RS &&
4686 	    !(lc->supported & FW_PORT_CAP32_FEC_RS)) ||
4687 	    (lc->requested_fec & FEC_BASER_RS &&
4688 	    !(lc->supported & FW_PORT_CAP32_FEC_BASER_RS))) {
4689 		n++;
4690 		lc->requested_fec = FEC_AUTO;
4691 	}
4692 
4693 	return (n);
4694 }
4695 
4696 /*
4697  * Apply the requested L1 settings, which are expected to be valid, to the
4698  * hardware.
4699  */
4700 static int
4701 apply_link_config(struct port_info *pi)
4702 {
4703 	struct adapter *sc = pi->adapter;
4704 	struct link_config *lc = &pi->link_cfg;
4705 	int rc;
4706 
4707 #ifdef INVARIANTS
4708 	ASSERT_SYNCHRONIZED_OP(sc);
4709 	PORT_LOCK_ASSERT_OWNED(pi);
4710 
4711 	if (lc->requested_aneg == AUTONEG_ENABLE)
4712 		MPASS(lc->supported & FW_PORT_CAP32_ANEG);
4713 	if (!(lc->requested_fc & PAUSE_AUTONEG))
4714 		MPASS(lc->supported & FW_PORT_CAP32_FORCE_PAUSE);
4715 	if (lc->requested_fc & PAUSE_TX)
4716 		MPASS(lc->supported & FW_PORT_CAP32_FC_TX);
4717 	if (lc->requested_fc & PAUSE_RX)
4718 		MPASS(lc->supported & FW_PORT_CAP32_FC_RX);
4719 	if (lc->requested_fec & FEC_RS)
4720 		MPASS(lc->supported & FW_PORT_CAP32_FEC_RS);
4721 	if (lc->requested_fec & FEC_BASER_RS)
4722 		MPASS(lc->supported & FW_PORT_CAP32_FEC_BASER_RS);
4723 #endif
4724 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
4725 	if (rc != 0) {
4726 		/* Don't complain if the VF driver gets back an EPERM. */
4727 		if (!(sc->flags & IS_VF) || rc != FW_EPERM)
4728 			device_printf(pi->dev, "l1cfg failed: %d\n", rc);
4729 	} else {
4730 		/*
4731 		 * An L1_CFG will almost always result in a link-change event if
4732 		 * the link is up, and the driver will refresh the actual
4733 		 * fec/fc/etc. when the notification is processed.  If the link
4734 		 * is down then the actual settings are meaningless.
4735 		 *
4736 		 * This takes care of the case where a change in the L1 settings
4737 		 * may not result in a notification.
4738 		 */
4739 		if (lc->link_ok && !(lc->requested_fc & PAUSE_AUTONEG))
4740 			lc->fc = lc->requested_fc & (PAUSE_TX | PAUSE_RX);
4741 	}
4742 	return (rc);
4743 }
4744 
4745 #define FW_MAC_EXACT_CHUNK	7
4746 
4747 /*
4748  * Program the port's XGMAC based on parameters in ifnet.  The caller also
4749  * indicates which parameters should be programmed (the rest are left alone).
4750  */
4751 int
4752 update_mac_settings(struct ifnet *ifp, int flags)
4753 {
4754 	int rc = 0;
4755 	struct vi_info *vi = ifp->if_softc;
4756 	struct port_info *pi = vi->pi;
4757 	struct adapter *sc = pi->adapter;
4758 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4759 
4760 	ASSERT_SYNCHRONIZED_OP(sc);
4761 	KASSERT(flags, ("%s: not told what to update.", __func__));
4762 
4763 	if (flags & XGMAC_MTU)
4764 		mtu = ifp->if_mtu;
4765 
4766 	if (flags & XGMAC_PROMISC)
4767 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4768 
4769 	if (flags & XGMAC_ALLMULTI)
4770 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4771 
4772 	if (flags & XGMAC_VLANEX)
4773 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4774 
4775 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4776 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4777 		    allmulti, 1, vlanex, false);
4778 		if (rc) {
4779 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4780 			    rc);
4781 			return (rc);
4782 		}
4783 	}
4784 
4785 	if (flags & XGMAC_UCADDR) {
4786 		uint8_t ucaddr[ETHER_ADDR_LEN];
4787 
4788 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4789 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4790 		    ucaddr, true, true);
4791 		if (rc < 0) {
4792 			rc = -rc;
4793 			if_printf(ifp, "change_mac failed: %d\n", rc);
4794 			return (rc);
4795 		} else {
4796 			vi->xact_addr_filt = rc;
4797 			rc = 0;
4798 		}
4799 	}
4800 
4801 	if (flags & XGMAC_MCADDRS) {
4802 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4803 		int del = 1;
4804 		uint64_t hash = 0;
4805 		struct ifmultiaddr *ifma;
4806 		int i = 0, j;
4807 
4808 		if_maddr_rlock(ifp);
4809 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4810 			if (ifma->ifma_addr->sa_family != AF_LINK)
4811 				continue;
4812 			mcaddr[i] =
4813 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4814 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4815 			i++;
4816 
4817 			if (i == FW_MAC_EXACT_CHUNK) {
4818 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4819 				    del, i, mcaddr, NULL, &hash, 0);
4820 				if (rc < 0) {
4821 					rc = -rc;
4822 					for (j = 0; j < i; j++) {
4823 						if_printf(ifp,
4824 						    "failed to add mc address"
4825 						    " %02x:%02x:%02x:"
4826 						    "%02x:%02x:%02x rc=%d\n",
4827 						    mcaddr[j][0], mcaddr[j][1],
4828 						    mcaddr[j][2], mcaddr[j][3],
4829 						    mcaddr[j][4], mcaddr[j][5],
4830 						    rc);
4831 					}
4832 					goto mcfail;
4833 				}
4834 				del = 0;
4835 				i = 0;
4836 			}
4837 		}
4838 		if (i > 0) {
4839 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4840 			    mcaddr, NULL, &hash, 0);
4841 			if (rc < 0) {
4842 				rc = -rc;
4843 				for (j = 0; j < i; j++) {
4844 					if_printf(ifp,
4845 					    "failed to add mc address"
4846 					    " %02x:%02x:%02x:"
4847 					    "%02x:%02x:%02x rc=%d\n",
4848 					    mcaddr[j][0], mcaddr[j][1],
4849 					    mcaddr[j][2], mcaddr[j][3],
4850 					    mcaddr[j][4], mcaddr[j][5],
4851 					    rc);
4852 				}
4853 				goto mcfail;
4854 			}
4855 		}
4856 
4857 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4858 		if (rc != 0)
4859 			if_printf(ifp, "failed to set mc address hash: %d", rc);
4860 mcfail:
4861 		if_maddr_runlock(ifp);
4862 	}
4863 
4864 	return (rc);
4865 }
4866 
4867 /*
4868  * {begin|end}_synchronized_op must be called from the same thread.
4869  */
4870 int
4871 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4872     char *wmesg)
4873 {
4874 	int rc, pri;
4875 
4876 #ifdef WITNESS
4877 	/* the caller thinks it's ok to sleep, but is it really? */
4878 	if (flags & SLEEP_OK)
4879 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4880 		    "begin_synchronized_op");
4881 #endif
4882 
4883 	if (INTR_OK)
4884 		pri = PCATCH;
4885 	else
4886 		pri = 0;
4887 
4888 	ADAPTER_LOCK(sc);
4889 	for (;;) {
4890 
4891 		if (vi && IS_DOOMED(vi)) {
4892 			rc = ENXIO;
4893 			goto done;
4894 		}
4895 
4896 		if (!IS_BUSY(sc)) {
4897 			rc = 0;
4898 			break;
4899 		}
4900 
4901 		if (!(flags & SLEEP_OK)) {
4902 			rc = EBUSY;
4903 			goto done;
4904 		}
4905 
4906 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4907 			rc = EINTR;
4908 			goto done;
4909 		}
4910 	}
4911 
4912 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4913 	SET_BUSY(sc);
4914 #ifdef INVARIANTS
4915 	sc->last_op = wmesg;
4916 	sc->last_op_thr = curthread;
4917 	sc->last_op_flags = flags;
4918 #endif
4919 
4920 done:
4921 	if (!(flags & HOLD_LOCK) || rc)
4922 		ADAPTER_UNLOCK(sc);
4923 
4924 	return (rc);
4925 }
4926 
4927 /*
4928  * Tell if_ioctl and if_init that the VI is going away.  This is
4929  * special variant of begin_synchronized_op and must be paired with a
4930  * call to end_synchronized_op.
4931  */
4932 void
4933 doom_vi(struct adapter *sc, struct vi_info *vi)
4934 {
4935 
4936 	ADAPTER_LOCK(sc);
4937 	SET_DOOMED(vi);
4938 	wakeup(&sc->flags);
4939 	while (IS_BUSY(sc))
4940 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4941 	SET_BUSY(sc);
4942 #ifdef INVARIANTS
4943 	sc->last_op = "t4detach";
4944 	sc->last_op_thr = curthread;
4945 	sc->last_op_flags = 0;
4946 #endif
4947 	ADAPTER_UNLOCK(sc);
4948 }
4949 
4950 /*
4951  * {begin|end}_synchronized_op must be called from the same thread.
4952  */
4953 void
4954 end_synchronized_op(struct adapter *sc, int flags)
4955 {
4956 
4957 	if (flags & LOCK_HELD)
4958 		ADAPTER_LOCK_ASSERT_OWNED(sc);
4959 	else
4960 		ADAPTER_LOCK(sc);
4961 
4962 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
4963 	CLR_BUSY(sc);
4964 	wakeup(&sc->flags);
4965 	ADAPTER_UNLOCK(sc);
4966 }
4967 
4968 static int
4969 cxgbe_init_synchronized(struct vi_info *vi)
4970 {
4971 	struct port_info *pi = vi->pi;
4972 	struct adapter *sc = pi->adapter;
4973 	struct ifnet *ifp = vi->ifp;
4974 	int rc = 0, i;
4975 	struct sge_txq *txq;
4976 
4977 	ASSERT_SYNCHRONIZED_OP(sc);
4978 
4979 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
4980 		return (0);	/* already running */
4981 
4982 	if (!(sc->flags & FULL_INIT_DONE) &&
4983 	    ((rc = adapter_full_init(sc)) != 0))
4984 		return (rc);	/* error message displayed already */
4985 
4986 	if (!(vi->flags & VI_INIT_DONE) &&
4987 	    ((rc = vi_full_init(vi)) != 0))
4988 		return (rc); /* error message displayed already */
4989 
4990 	rc = update_mac_settings(ifp, XGMAC_ALL);
4991 	if (rc)
4992 		goto done;	/* error message displayed already */
4993 
4994 	PORT_LOCK(pi);
4995 	if (pi->up_vis == 0) {
4996 		t4_update_port_info(pi);
4997 		fixup_link_config(pi);
4998 		build_medialist(pi);
4999 		apply_link_config(pi);
5000 	}
5001 
5002 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
5003 	if (rc != 0) {
5004 		if_printf(ifp, "enable_vi failed: %d\n", rc);
5005 		PORT_UNLOCK(pi);
5006 		goto done;
5007 	}
5008 
5009 	/*
5010 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
5011 	 * if this changes.
5012 	 */
5013 
5014 	for_each_txq(vi, i, txq) {
5015 		TXQ_LOCK(txq);
5016 		txq->eq.flags |= EQ_ENABLED;
5017 		TXQ_UNLOCK(txq);
5018 	}
5019 
5020 	/*
5021 	 * The first iq of the first port to come up is used for tracing.
5022 	 */
5023 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
5024 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
5025 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
5026 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
5027 		    V_QUEUENUMBER(sc->traceq));
5028 		pi->flags |= HAS_TRACEQ;
5029 	}
5030 
5031 	/* all ok */
5032 	pi->up_vis++;
5033 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
5034 
5035 	if (pi->nvi > 1 || sc->flags & IS_VF)
5036 		callout_reset(&vi->tick, hz, vi_tick, vi);
5037 	else
5038 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
5039 	PORT_UNLOCK(pi);
5040 done:
5041 	if (rc != 0)
5042 		cxgbe_uninit_synchronized(vi);
5043 
5044 	return (rc);
5045 }
5046 
5047 /*
5048  * Idempotent.
5049  */
5050 static int
5051 cxgbe_uninit_synchronized(struct vi_info *vi)
5052 {
5053 	struct port_info *pi = vi->pi;
5054 	struct adapter *sc = pi->adapter;
5055 	struct ifnet *ifp = vi->ifp;
5056 	int rc, i;
5057 	struct sge_txq *txq;
5058 
5059 	ASSERT_SYNCHRONIZED_OP(sc);
5060 
5061 	if (!(vi->flags & VI_INIT_DONE)) {
5062 		if (__predict_false(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
5063 			KASSERT(0, ("uninited VI is running"));
5064 			if_printf(ifp, "uninited VI with running ifnet.  "
5065 			    "vi->flags 0x%016lx, if_flags 0x%08x, "
5066 			    "if_drv_flags 0x%08x\n", vi->flags, ifp->if_flags,
5067 			    ifp->if_drv_flags);
5068 		}
5069 		return (0);
5070 	}
5071 
5072 	/*
5073 	 * Disable the VI so that all its data in either direction is discarded
5074 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
5075 	 * tick) intact as the TP can deliver negative advice or data that it's
5076 	 * holding in its RAM (for an offloaded connection) even after the VI is
5077 	 * disabled.
5078 	 */
5079 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
5080 	if (rc) {
5081 		if_printf(ifp, "disable_vi failed: %d\n", rc);
5082 		return (rc);
5083 	}
5084 
5085 	for_each_txq(vi, i, txq) {
5086 		TXQ_LOCK(txq);
5087 		txq->eq.flags &= ~EQ_ENABLED;
5088 		TXQ_UNLOCK(txq);
5089 	}
5090 
5091 	PORT_LOCK(pi);
5092 	if (pi->nvi > 1 || sc->flags & IS_VF)
5093 		callout_stop(&vi->tick);
5094 	else
5095 		callout_stop(&pi->tick);
5096 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
5097 		PORT_UNLOCK(pi);
5098 		return (0);
5099 	}
5100 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
5101 	pi->up_vis--;
5102 	if (pi->up_vis > 0) {
5103 		PORT_UNLOCK(pi);
5104 		return (0);
5105 	}
5106 
5107 	pi->link_cfg.link_ok = false;
5108 	pi->link_cfg.speed = 0;
5109 	pi->link_cfg.link_down_rc = 255;
5110 	t4_os_link_changed(pi);
5111 	PORT_UNLOCK(pi);
5112 
5113 	return (0);
5114 }
5115 
5116 /*
5117  * It is ok for this function to fail midway and return right away.  t4_detach
5118  * will walk the entire sc->irq list and clean up whatever is valid.
5119  */
5120 int
5121 t4_setup_intr_handlers(struct adapter *sc)
5122 {
5123 	int rc, rid, p, q, v;
5124 	char s[8];
5125 	struct irq *irq;
5126 	struct port_info *pi;
5127 	struct vi_info *vi;
5128 	struct sge *sge = &sc->sge;
5129 	struct sge_rxq *rxq;
5130 #ifdef TCP_OFFLOAD
5131 	struct sge_ofld_rxq *ofld_rxq;
5132 #endif
5133 #ifdef DEV_NETMAP
5134 	struct sge_nm_rxq *nm_rxq;
5135 #endif
5136 #ifdef RSS
5137 	int nbuckets = rss_getnumbuckets();
5138 #endif
5139 
5140 	/*
5141 	 * Setup interrupts.
5142 	 */
5143 	irq = &sc->irq[0];
5144 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
5145 	if (forwarding_intr_to_fwq(sc))
5146 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
5147 
5148 	/* Multiple interrupts. */
5149 	if (sc->flags & IS_VF)
5150 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
5151 		    ("%s: too few intr.", __func__));
5152 	else
5153 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
5154 		    ("%s: too few intr.", __func__));
5155 
5156 	/* The first one is always error intr on PFs */
5157 	if (!(sc->flags & IS_VF)) {
5158 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
5159 		if (rc != 0)
5160 			return (rc);
5161 		irq++;
5162 		rid++;
5163 	}
5164 
5165 	/* The second one is always the firmware event queue (first on VFs) */
5166 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
5167 	if (rc != 0)
5168 		return (rc);
5169 	irq++;
5170 	rid++;
5171 
5172 	for_each_port(sc, p) {
5173 		pi = sc->port[p];
5174 		for_each_vi(pi, v, vi) {
5175 			vi->first_intr = rid - 1;
5176 
5177 			if (vi->nnmrxq > 0) {
5178 				int n = max(vi->nrxq, vi->nnmrxq);
5179 
5180 				rxq = &sge->rxq[vi->first_rxq];
5181 #ifdef DEV_NETMAP
5182 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
5183 #endif
5184 				for (q = 0; q < n; q++) {
5185 					snprintf(s, sizeof(s), "%x%c%x", p,
5186 					    'a' + v, q);
5187 					if (q < vi->nrxq)
5188 						irq->rxq = rxq++;
5189 #ifdef DEV_NETMAP
5190 					if (q < vi->nnmrxq)
5191 						irq->nm_rxq = nm_rxq++;
5192 
5193 					if (irq->nm_rxq != NULL &&
5194 					    irq->rxq == NULL) {
5195 						/* Netmap rx only */
5196 						rc = t4_alloc_irq(sc, irq, rid,
5197 						    t4_nm_intr, irq->nm_rxq, s);
5198 					}
5199 					if (irq->nm_rxq != NULL &&
5200 					    irq->rxq != NULL) {
5201 						/* NIC and Netmap rx */
5202 						rc = t4_alloc_irq(sc, irq, rid,
5203 						    t4_vi_intr, irq, s);
5204 					}
5205 #endif
5206 					if (irq->rxq != NULL &&
5207 					    irq->nm_rxq == NULL) {
5208 						/* NIC rx only */
5209 						rc = t4_alloc_irq(sc, irq, rid,
5210 						    t4_intr, irq->rxq, s);
5211 					}
5212 					if (rc != 0)
5213 						return (rc);
5214 #ifdef RSS
5215 					if (q < vi->nrxq) {
5216 						bus_bind_intr(sc->dev, irq->res,
5217 						    rss_getcpu(q % nbuckets));
5218 					}
5219 #endif
5220 					irq++;
5221 					rid++;
5222 					vi->nintr++;
5223 				}
5224 			} else {
5225 				for_each_rxq(vi, q, rxq) {
5226 					snprintf(s, sizeof(s), "%x%c%x", p,
5227 					    'a' + v, q);
5228 					rc = t4_alloc_irq(sc, irq, rid,
5229 					    t4_intr, rxq, s);
5230 					if (rc != 0)
5231 						return (rc);
5232 #ifdef RSS
5233 					bus_bind_intr(sc->dev, irq->res,
5234 					    rss_getcpu(q % nbuckets));
5235 #endif
5236 					irq++;
5237 					rid++;
5238 					vi->nintr++;
5239 				}
5240 			}
5241 #ifdef TCP_OFFLOAD
5242 			for_each_ofld_rxq(vi, q, ofld_rxq) {
5243 				snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
5244 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
5245 				    ofld_rxq, s);
5246 				if (rc != 0)
5247 					return (rc);
5248 				irq++;
5249 				rid++;
5250 				vi->nintr++;
5251 			}
5252 #endif
5253 		}
5254 	}
5255 	MPASS(irq == &sc->irq[sc->intr_count]);
5256 
5257 	return (0);
5258 }
5259 
5260 int
5261 adapter_full_init(struct adapter *sc)
5262 {
5263 	int rc, i;
5264 #ifdef RSS
5265 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5266 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5267 #endif
5268 
5269 	ASSERT_SYNCHRONIZED_OP(sc);
5270 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5271 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
5272 	    ("%s: FULL_INIT_DONE already", __func__));
5273 
5274 	/*
5275 	 * queues that belong to the adapter (not any particular port).
5276 	 */
5277 	rc = t4_setup_adapter_queues(sc);
5278 	if (rc != 0)
5279 		goto done;
5280 
5281 	for (i = 0; i < nitems(sc->tq); i++) {
5282 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
5283 		    taskqueue_thread_enqueue, &sc->tq[i]);
5284 		if (sc->tq[i] == NULL) {
5285 			device_printf(sc->dev,
5286 			    "failed to allocate task queue %d\n", i);
5287 			rc = ENOMEM;
5288 			goto done;
5289 		}
5290 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
5291 		    device_get_nameunit(sc->dev), i);
5292 	}
5293 #ifdef RSS
5294 	MPASS(RSS_KEYSIZE == 40);
5295 	rss_getkey((void *)&raw_rss_key[0]);
5296 	for (i = 0; i < nitems(rss_key); i++) {
5297 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
5298 	}
5299 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
5300 #endif
5301 
5302 	if (!(sc->flags & IS_VF))
5303 		t4_intr_enable(sc);
5304 	sc->flags |= FULL_INIT_DONE;
5305 done:
5306 	if (rc != 0)
5307 		adapter_full_uninit(sc);
5308 
5309 	return (rc);
5310 }
5311 
5312 int
5313 adapter_full_uninit(struct adapter *sc)
5314 {
5315 	int i;
5316 
5317 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5318 
5319 	t4_teardown_adapter_queues(sc);
5320 
5321 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
5322 		taskqueue_free(sc->tq[i]);
5323 		sc->tq[i] = NULL;
5324 	}
5325 
5326 	sc->flags &= ~FULL_INIT_DONE;
5327 
5328 	return (0);
5329 }
5330 
5331 #ifdef RSS
5332 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
5333     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
5334     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
5335     RSS_HASHTYPE_RSS_UDP_IPV6)
5336 
5337 /* Translates kernel hash types to hardware. */
5338 static int
5339 hashconfig_to_hashen(int hashconfig)
5340 {
5341 	int hashen = 0;
5342 
5343 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
5344 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
5345 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
5346 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
5347 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
5348 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5349 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5350 	}
5351 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
5352 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5353 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5354 	}
5355 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
5356 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5357 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
5358 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5359 
5360 	return (hashen);
5361 }
5362 
5363 /* Translates hardware hash types to kernel. */
5364 static int
5365 hashen_to_hashconfig(int hashen)
5366 {
5367 	int hashconfig = 0;
5368 
5369 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
5370 		/*
5371 		 * If UDP hashing was enabled it must have been enabled for
5372 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
5373 		 * enabling any 4-tuple hash is nonsense configuration.
5374 		 */
5375 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5376 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
5377 
5378 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5379 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
5380 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5381 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
5382 	}
5383 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5384 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
5385 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5386 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
5387 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
5388 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
5389 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
5390 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
5391 
5392 	return (hashconfig);
5393 }
5394 #endif
5395 
5396 int
5397 vi_full_init(struct vi_info *vi)
5398 {
5399 	struct adapter *sc = vi->pi->adapter;
5400 	struct ifnet *ifp = vi->ifp;
5401 	uint16_t *rss;
5402 	struct sge_rxq *rxq;
5403 	int rc, i, j;
5404 #ifdef RSS
5405 	int nbuckets = rss_getnumbuckets();
5406 	int hashconfig = rss_gethashconfig();
5407 	int extra;
5408 #endif
5409 
5410 	ASSERT_SYNCHRONIZED_OP(sc);
5411 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
5412 	    ("%s: VI_INIT_DONE already", __func__));
5413 
5414 	sysctl_ctx_init(&vi->ctx);
5415 	vi->flags |= VI_SYSCTL_CTX;
5416 
5417 	/*
5418 	 * Allocate tx/rx/fl queues for this VI.
5419 	 */
5420 	rc = t4_setup_vi_queues(vi);
5421 	if (rc != 0)
5422 		goto done;	/* error message displayed already */
5423 
5424 	/*
5425 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
5426 	 */
5427 	if (vi->nrxq > vi->rss_size) {
5428 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
5429 		    "some queues will never receive traffic.\n", vi->nrxq,
5430 		    vi->rss_size);
5431 	} else if (vi->rss_size % vi->nrxq) {
5432 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
5433 		    "expect uneven traffic distribution.\n", vi->nrxq,
5434 		    vi->rss_size);
5435 	}
5436 #ifdef RSS
5437 	if (vi->nrxq != nbuckets) {
5438 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
5439 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
5440 	}
5441 #endif
5442 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
5443 	for (i = 0; i < vi->rss_size;) {
5444 #ifdef RSS
5445 		j = rss_get_indirection_to_bucket(i);
5446 		j %= vi->nrxq;
5447 		rxq = &sc->sge.rxq[vi->first_rxq + j];
5448 		rss[i++] = rxq->iq.abs_id;
5449 #else
5450 		for_each_rxq(vi, j, rxq) {
5451 			rss[i++] = rxq->iq.abs_id;
5452 			if (i == vi->rss_size)
5453 				break;
5454 		}
5455 #endif
5456 	}
5457 
5458 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
5459 	    vi->rss_size);
5460 	if (rc != 0) {
5461 		free(rss, M_CXGBE);
5462 		if_printf(ifp, "rss_config failed: %d\n", rc);
5463 		goto done;
5464 	}
5465 
5466 #ifdef RSS
5467 	vi->hashen = hashconfig_to_hashen(hashconfig);
5468 
5469 	/*
5470 	 * We may have had to enable some hashes even though the global config
5471 	 * wants them disabled.  This is a potential problem that must be
5472 	 * reported to the user.
5473 	 */
5474 	extra = hashen_to_hashconfig(vi->hashen) ^ hashconfig;
5475 
5476 	/*
5477 	 * If we consider only the supported hash types, then the enabled hashes
5478 	 * are a superset of the requested hashes.  In other words, there cannot
5479 	 * be any supported hash that was requested but not enabled, but there
5480 	 * can be hashes that were not requested but had to be enabled.
5481 	 */
5482 	extra &= SUPPORTED_RSS_HASHTYPES;
5483 	MPASS((extra & hashconfig) == 0);
5484 
5485 	if (extra) {
5486 		if_printf(ifp,
5487 		    "global RSS config (0x%x) cannot be accommodated.\n",
5488 		    hashconfig);
5489 	}
5490 	if (extra & RSS_HASHTYPE_RSS_IPV4)
5491 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
5492 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
5493 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
5494 	if (extra & RSS_HASHTYPE_RSS_IPV6)
5495 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
5496 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
5497 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
5498 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
5499 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
5500 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
5501 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
5502 #else
5503 	vi->hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
5504 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
5505 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5506 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
5507 #endif
5508 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, vi->hashen, rss[0], 0, 0);
5509 	if (rc != 0) {
5510 		free(rss, M_CXGBE);
5511 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
5512 		goto done;
5513 	}
5514 
5515 	vi->rss = rss;
5516 	vi->flags |= VI_INIT_DONE;
5517 done:
5518 	if (rc != 0)
5519 		vi_full_uninit(vi);
5520 
5521 	return (rc);
5522 }
5523 
5524 /*
5525  * Idempotent.
5526  */
5527 int
5528 vi_full_uninit(struct vi_info *vi)
5529 {
5530 	struct port_info *pi = vi->pi;
5531 	struct adapter *sc = pi->adapter;
5532 	int i;
5533 	struct sge_rxq *rxq;
5534 	struct sge_txq *txq;
5535 #ifdef TCP_OFFLOAD
5536 	struct sge_ofld_rxq *ofld_rxq;
5537 #endif
5538 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
5539 	struct sge_wrq *ofld_txq;
5540 #endif
5541 
5542 	if (vi->flags & VI_INIT_DONE) {
5543 
5544 		/* Need to quiesce queues.  */
5545 
5546 		/* XXX: Only for the first VI? */
5547 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
5548 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
5549 
5550 		for_each_txq(vi, i, txq) {
5551 			quiesce_txq(sc, txq);
5552 		}
5553 
5554 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
5555 		for_each_ofld_txq(vi, i, ofld_txq) {
5556 			quiesce_wrq(sc, ofld_txq);
5557 		}
5558 #endif
5559 
5560 		for_each_rxq(vi, i, rxq) {
5561 			quiesce_iq(sc, &rxq->iq);
5562 			quiesce_fl(sc, &rxq->fl);
5563 		}
5564 
5565 #ifdef TCP_OFFLOAD
5566 		for_each_ofld_rxq(vi, i, ofld_rxq) {
5567 			quiesce_iq(sc, &ofld_rxq->iq);
5568 			quiesce_fl(sc, &ofld_rxq->fl);
5569 		}
5570 #endif
5571 		free(vi->rss, M_CXGBE);
5572 		free(vi->nm_rss, M_CXGBE);
5573 	}
5574 
5575 	t4_teardown_vi_queues(vi);
5576 	vi->flags &= ~VI_INIT_DONE;
5577 
5578 	return (0);
5579 }
5580 
5581 static void
5582 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
5583 {
5584 	struct sge_eq *eq = &txq->eq;
5585 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
5586 
5587 	(void) sc;	/* unused */
5588 
5589 #ifdef INVARIANTS
5590 	TXQ_LOCK(txq);
5591 	MPASS((eq->flags & EQ_ENABLED) == 0);
5592 	TXQ_UNLOCK(txq);
5593 #endif
5594 
5595 	/* Wait for the mp_ring to empty. */
5596 	while (!mp_ring_is_idle(txq->r)) {
5597 		mp_ring_check_drainage(txq->r, 0);
5598 		pause("rquiesce", 1);
5599 	}
5600 
5601 	/* Then wait for the hardware to finish. */
5602 	while (spg->cidx != htobe16(eq->pidx))
5603 		pause("equiesce", 1);
5604 
5605 	/* Finally, wait for the driver to reclaim all descriptors. */
5606 	while (eq->cidx != eq->pidx)
5607 		pause("dquiesce", 1);
5608 }
5609 
5610 static void
5611 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
5612 {
5613 
5614 	/* XXXTX */
5615 }
5616 
5617 static void
5618 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
5619 {
5620 	(void) sc;	/* unused */
5621 
5622 	/* Synchronize with the interrupt handler */
5623 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
5624 		pause("iqfree", 1);
5625 }
5626 
5627 static void
5628 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
5629 {
5630 	mtx_lock(&sc->sfl_lock);
5631 	FL_LOCK(fl);
5632 	fl->flags |= FL_DOOMED;
5633 	FL_UNLOCK(fl);
5634 	callout_stop(&sc->sfl_callout);
5635 	mtx_unlock(&sc->sfl_lock);
5636 
5637 	KASSERT((fl->flags & FL_STARVING) == 0,
5638 	    ("%s: still starving", __func__));
5639 }
5640 
5641 static int
5642 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
5643     driver_intr_t *handler, void *arg, char *name)
5644 {
5645 	int rc;
5646 
5647 	irq->rid = rid;
5648 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
5649 	    RF_SHAREABLE | RF_ACTIVE);
5650 	if (irq->res == NULL) {
5651 		device_printf(sc->dev,
5652 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
5653 		return (ENOMEM);
5654 	}
5655 
5656 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
5657 	    NULL, handler, arg, &irq->tag);
5658 	if (rc != 0) {
5659 		device_printf(sc->dev,
5660 		    "failed to setup interrupt for rid %d, name %s: %d\n",
5661 		    rid, name, rc);
5662 	} else if (name)
5663 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
5664 
5665 	return (rc);
5666 }
5667 
5668 static int
5669 t4_free_irq(struct adapter *sc, struct irq *irq)
5670 {
5671 	if (irq->tag)
5672 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
5673 	if (irq->res)
5674 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
5675 
5676 	bzero(irq, sizeof(*irq));
5677 
5678 	return (0);
5679 }
5680 
5681 static void
5682 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
5683 {
5684 
5685 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
5686 	t4_get_regs(sc, buf, regs->len);
5687 }
5688 
5689 #define	A_PL_INDIR_CMD	0x1f8
5690 
5691 #define	S_PL_AUTOINC	31
5692 #define	M_PL_AUTOINC	0x1U
5693 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
5694 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
5695 
5696 #define	S_PL_VFID	20
5697 #define	M_PL_VFID	0xffU
5698 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
5699 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
5700 
5701 #define	S_PL_ADDR	0
5702 #define	M_PL_ADDR	0xfffffU
5703 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
5704 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
5705 
5706 #define	A_PL_INDIR_DATA	0x1fc
5707 
5708 static uint64_t
5709 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
5710 {
5711 	u32 stats[2];
5712 
5713 	mtx_assert(&sc->reg_lock, MA_OWNED);
5714 	if (sc->flags & IS_VF) {
5715 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
5716 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
5717 	} else {
5718 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5719 		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5720 		    V_PL_ADDR(VF_MPS_REG(reg)));
5721 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
5722 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
5723 	}
5724 	return (((uint64_t)stats[1]) << 32 | stats[0]);
5725 }
5726 
5727 static void
5728 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
5729     struct fw_vi_stats_vf *stats)
5730 {
5731 
5732 #define GET_STAT(name) \
5733 	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
5734 
5735 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
5736 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
5737 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
5738 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
5739 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
5740 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
5741 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
5742 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
5743 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
5744 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
5745 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
5746 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
5747 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
5748 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
5749 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
5750 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
5751 
5752 #undef GET_STAT
5753 }
5754 
5755 static void
5756 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
5757 {
5758 	int reg;
5759 
5760 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5761 	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5762 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
5763 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
5764 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
5765 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
5766 }
5767 
5768 static void
5769 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
5770 {
5771 	struct timeval tv;
5772 	const struct timeval interval = {0, 250000};	/* 250ms */
5773 
5774 	if (!(vi->flags & VI_INIT_DONE))
5775 		return;
5776 
5777 	getmicrotime(&tv);
5778 	timevalsub(&tv, &interval);
5779 	if (timevalcmp(&tv, &vi->last_refreshed, <))
5780 		return;
5781 
5782 	mtx_lock(&sc->reg_lock);
5783 	t4_get_vi_stats(sc, vi->viid, &vi->stats);
5784 	getmicrotime(&vi->last_refreshed);
5785 	mtx_unlock(&sc->reg_lock);
5786 }
5787 
5788 static void
5789 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5790 {
5791 	u_int i, v, tnl_cong_drops, bg_map;
5792 	struct timeval tv;
5793 	const struct timeval interval = {0, 250000};	/* 250ms */
5794 
5795 	getmicrotime(&tv);
5796 	timevalsub(&tv, &interval);
5797 	if (timevalcmp(&tv, &pi->last_refreshed, <))
5798 		return;
5799 
5800 	tnl_cong_drops = 0;
5801 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5802 	bg_map = pi->mps_bg_map;
5803 	while (bg_map) {
5804 		i = ffs(bg_map) - 1;
5805 		mtx_lock(&sc->reg_lock);
5806 		t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5807 		    A_TP_MIB_TNL_CNG_DROP_0 + i);
5808 		mtx_unlock(&sc->reg_lock);
5809 		tnl_cong_drops += v;
5810 		bg_map &= ~(1 << i);
5811 	}
5812 	pi->tnl_cong_drops = tnl_cong_drops;
5813 	getmicrotime(&pi->last_refreshed);
5814 }
5815 
5816 static void
5817 cxgbe_tick(void *arg)
5818 {
5819 	struct port_info *pi = arg;
5820 	struct adapter *sc = pi->adapter;
5821 
5822 	PORT_LOCK_ASSERT_OWNED(pi);
5823 	cxgbe_refresh_stats(sc, pi);
5824 
5825 	callout_schedule(&pi->tick, hz);
5826 }
5827 
5828 void
5829 vi_tick(void *arg)
5830 {
5831 	struct vi_info *vi = arg;
5832 	struct adapter *sc = vi->pi->adapter;
5833 
5834 	vi_refresh_stats(sc, vi);
5835 
5836 	callout_schedule(&vi->tick, hz);
5837 }
5838 
5839 /*
5840  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5841  */
5842 static char *caps_decoder[] = {
5843 	"\20\001IPMI\002NCSI",				/* 0: NBM */
5844 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
5845 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
5846 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
5847 	    "\006HASHFILTER\007ETHOFLD",
5848 	"\20\001TOE",					/* 4: TOE */
5849 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
5850 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
5851 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5852 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5853 	    "\007T10DIF"
5854 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5855 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
5856 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
5857 		    "\004PO_INITIATOR\005PO_TARGET",
5858 };
5859 
5860 void
5861 t4_sysctls(struct adapter *sc)
5862 {
5863 	struct sysctl_ctx_list *ctx;
5864 	struct sysctl_oid *oid;
5865 	struct sysctl_oid_list *children, *c0;
5866 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5867 
5868 	ctx = device_get_sysctl_ctx(sc->dev);
5869 
5870 	/*
5871 	 * dev.t4nex.X.
5872 	 */
5873 	oid = device_get_sysctl_tree(sc->dev);
5874 	c0 = children = SYSCTL_CHILDREN(oid);
5875 
5876 	sc->sc_do_rxcopy = 1;
5877 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5878 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5879 
5880 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5881 	    sc->params.nports, "# of ports");
5882 
5883 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5884 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, (uintptr_t)&sc->doorbells,
5885 	    sysctl_bitfield_8b, "A", "available doorbells");
5886 
5887 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5888 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
5889 
5890 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5891 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5892 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5893 	    "interrupt holdoff timer values (us)");
5894 
5895 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5896 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5897 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5898 	    "interrupt holdoff packet counter values");
5899 
5900 	t4_sge_sysctls(sc, ctx, children);
5901 
5902 	sc->lro_timeout = 100;
5903 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5904 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5905 
5906 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5907 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
5908 
5909 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5910 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5911 
5912 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5913 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5914 
5915 	if (sc->flags & IS_VF)
5916 		return;
5917 
5918 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5919 	    NULL, chip_rev(sc), "chip hardware revision");
5920 
5921 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5922 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5923 
5924 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5925 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5926 
5927 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5928 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5929 
5930 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5931 	    CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5932 
5933 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5934 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5935 
5936 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5937 	    sc->er_version, 0, "expansion ROM version");
5938 
5939 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5940 	    sc->bs_version, 0, "bootstrap firmware version");
5941 
5942 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5943 	    NULL, sc->params.scfg_vers, "serial config version");
5944 
5945 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5946 	    NULL, sc->params.vpd_vers, "VPD version");
5947 
5948 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5949 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5950 
5951 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5952 	    sc->cfcsum, "config file checksum");
5953 
5954 #define SYSCTL_CAP(name, n, text) \
5955 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5956 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], (uintptr_t)&sc->name, \
5957 	    sysctl_bitfield_16b, "A", "available " text " capabilities")
5958 
5959 	SYSCTL_CAP(nbmcaps, 0, "NBM");
5960 	SYSCTL_CAP(linkcaps, 1, "link");
5961 	SYSCTL_CAP(switchcaps, 2, "switch");
5962 	SYSCTL_CAP(niccaps, 3, "NIC");
5963 	SYSCTL_CAP(toecaps, 4, "TCP offload");
5964 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
5965 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5966 	SYSCTL_CAP(cryptocaps, 7, "crypto");
5967 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
5968 #undef SYSCTL_CAP
5969 
5970 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5971 	    NULL, sc->tids.nftids, "number of filters");
5972 
5973 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5974 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5975 	    "chip temperature (in Celsius)");
5976 
5977 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "loadavg", CTLTYPE_STRING |
5978 	    CTLFLAG_RD, sc, 0, sysctl_loadavg, "A",
5979 	    "microprocessor load averages (debug firmwares only)");
5980 
5981 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
5982 	    &sc->params.core_vdd, 0, "core Vdd (in mV)");
5983 
5984 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "local_cpus",
5985 	    CTLTYPE_STRING | CTLFLAG_RD, sc, LOCAL_CPUS,
5986 	    sysctl_cpus, "A", "local CPUs");
5987 
5988 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "intr_cpus",
5989 	    CTLTYPE_STRING | CTLFLAG_RD, sc, INTR_CPUS,
5990 	    sysctl_cpus, "A", "preferred CPUs for interrupts");
5991 
5992 	/*
5993 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
5994 	 */
5995 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5996 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5997 	    "logs and miscellaneous information");
5998 	children = SYSCTL_CHILDREN(oid);
5999 
6000 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
6001 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6002 	    sysctl_cctrl, "A", "congestion control");
6003 
6004 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
6005 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6006 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
6007 
6008 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
6009 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
6010 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
6011 
6012 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
6013 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
6014 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
6015 
6016 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
6017 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
6018 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
6019 
6020 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
6021 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
6022 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
6023 
6024 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
6025 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
6026 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
6027 
6028 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
6029 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6030 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
6031 	    "A", "CIM logic analyzer");
6032 
6033 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
6034 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6035 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
6036 
6037 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
6038 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
6039 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
6040 
6041 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
6042 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
6043 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
6044 
6045 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
6046 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
6047 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
6048 
6049 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
6050 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
6051 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
6052 
6053 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
6054 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
6055 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
6056 
6057 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
6058 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
6059 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
6060 
6061 	if (chip_id(sc) > CHELSIO_T4) {
6062 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
6063 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
6064 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
6065 
6066 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
6067 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
6068 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
6069 	}
6070 
6071 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
6072 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6073 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
6074 
6075 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
6076 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6077 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
6078 
6079 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
6080 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6081 	    sysctl_cpl_stats, "A", "CPL statistics");
6082 
6083 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
6084 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6085 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
6086 
6087 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
6088 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6089 	    sysctl_devlog, "A", "firmware's device log");
6090 
6091 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
6092 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6093 	    sysctl_fcoe_stats, "A", "FCoE statistics");
6094 
6095 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
6096 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6097 	    sysctl_hw_sched, "A", "hardware scheduler ");
6098 
6099 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
6100 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6101 	    sysctl_l2t, "A", "hardware L2 table");
6102 
6103 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "smt",
6104 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6105 	    sysctl_smt, "A", "hardware source MAC table");
6106 
6107 #ifdef INET6
6108 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "clip",
6109 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6110 	    sysctl_clip, "A", "active CLIP table entries");
6111 #endif
6112 
6113 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
6114 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6115 	    sysctl_lb_stats, "A", "loopback statistics");
6116 
6117 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
6118 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6119 	    sysctl_meminfo, "A", "memory regions");
6120 
6121 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
6122 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6123 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
6124 	    "A", "MPS TCAM entries");
6125 
6126 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
6127 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6128 	    sysctl_path_mtus, "A", "path MTUs");
6129 
6130 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
6131 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6132 	    sysctl_pm_stats, "A", "PM statistics");
6133 
6134 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
6135 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6136 	    sysctl_rdma_stats, "A", "RDMA statistics");
6137 
6138 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
6139 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6140 	    sysctl_tcp_stats, "A", "TCP statistics");
6141 
6142 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
6143 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6144 	    sysctl_tids, "A", "TID information");
6145 
6146 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
6147 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6148 	    sysctl_tp_err_stats, "A", "TP error statistics");
6149 
6150 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
6151 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
6152 	    "TP logic analyzer event capture mask");
6153 
6154 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
6155 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6156 	    sysctl_tp_la, "A", "TP logic analyzer");
6157 
6158 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
6159 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6160 	    sysctl_tx_rate, "A", "Tx rate");
6161 
6162 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
6163 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6164 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
6165 
6166 	if (chip_id(sc) >= CHELSIO_T5) {
6167 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
6168 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6169 		    sysctl_wcwr_stats, "A", "write combined work requests");
6170 	}
6171 
6172 #ifdef TCP_OFFLOAD
6173 	if (is_offload(sc)) {
6174 		int i;
6175 		char s[4];
6176 
6177 		/*
6178 		 * dev.t4nex.X.toe.
6179 		 */
6180 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
6181 		    NULL, "TOE parameters");
6182 		children = SYSCTL_CHILDREN(oid);
6183 
6184 		sc->tt.cong_algorithm = -1;
6185 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
6186 		    CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
6187 		    "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
6188 		    "3 = highspeed)");
6189 
6190 		sc->tt.sndbuf = 256 * 1024;
6191 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
6192 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
6193 
6194 		sc->tt.ddp = 0;
6195 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
6196 		    &sc->tt.ddp, 0, "DDP allowed");
6197 
6198 		sc->tt.rx_coalesce = 1;
6199 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
6200 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
6201 
6202 		sc->tt.tls = 0;
6203 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tls", CTLFLAG_RW,
6204 		    &sc->tt.tls, 0, "Inline TLS allowed");
6205 
6206 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls_rx_ports",
6207 		    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tls_rx_ports,
6208 		    "I", "TCP ports that use inline TLS+TOE RX");
6209 
6210 		sc->tt.tx_align = 1;
6211 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
6212 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
6213 
6214 		sc->tt.tx_zcopy = 0;
6215 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
6216 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
6217 		    "Enable zero-copy aio_write(2)");
6218 
6219 		sc->tt.cop_managed_offloading = !!t4_cop_managed_offloading;
6220 		SYSCTL_ADD_INT(ctx, children, OID_AUTO,
6221 		    "cop_managed_offloading", CTLFLAG_RW,
6222 		    &sc->tt.cop_managed_offloading, 0,
6223 		    "COP (Connection Offload Policy) controls all TOE offload");
6224 
6225 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
6226 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
6227 		    "TP timer tick (us)");
6228 
6229 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
6230 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
6231 		    "TCP timestamp tick (us)");
6232 
6233 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
6234 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
6235 		    "DACK tick (us)");
6236 
6237 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
6238 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
6239 		    "IU", "DACK timer (us)");
6240 
6241 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
6242 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
6243 		    sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
6244 
6245 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
6246 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
6247 		    sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
6248 
6249 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
6250 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
6251 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
6252 
6253 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
6254 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
6255 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
6256 
6257 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
6258 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
6259 		    sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
6260 
6261 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
6262 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
6263 		    sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
6264 
6265 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
6266 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
6267 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
6268 
6269 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
6270 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
6271 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
6272 
6273 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
6274 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
6275 		    sysctl_tp_shift_cnt, "IU",
6276 		    "Number of SYN retransmissions before abort");
6277 
6278 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
6279 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
6280 		    sysctl_tp_shift_cnt, "IU",
6281 		    "Number of retransmissions before abort");
6282 
6283 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
6284 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
6285 		    sysctl_tp_shift_cnt, "IU",
6286 		    "Number of keepalive probes before abort");
6287 
6288 		oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
6289 		    CTLFLAG_RD, NULL, "TOE retransmit backoffs");
6290 		children = SYSCTL_CHILDREN(oid);
6291 		for (i = 0; i < 16; i++) {
6292 			snprintf(s, sizeof(s), "%u", i);
6293 			SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
6294 			    CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
6295 			    "IU", "TOE retransmit backoff");
6296 		}
6297 	}
6298 #endif
6299 }
6300 
6301 void
6302 vi_sysctls(struct vi_info *vi)
6303 {
6304 	struct sysctl_ctx_list *ctx;
6305 	struct sysctl_oid *oid;
6306 	struct sysctl_oid_list *children;
6307 
6308 	ctx = device_get_sysctl_ctx(vi->dev);
6309 
6310 	/*
6311 	 * dev.v?(cxgbe|cxl).X.
6312 	 */
6313 	oid = device_get_sysctl_tree(vi->dev);
6314 	children = SYSCTL_CHILDREN(oid);
6315 
6316 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
6317 	    vi->viid, "VI identifer");
6318 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
6319 	    &vi->nrxq, 0, "# of rx queues");
6320 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
6321 	    &vi->ntxq, 0, "# of tx queues");
6322 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
6323 	    &vi->first_rxq, 0, "index of first rx queue");
6324 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
6325 	    &vi->first_txq, 0, "index of first tx queue");
6326 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_base", CTLFLAG_RD, NULL,
6327 	    vi->rss_base, "start of RSS indirection table");
6328 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
6329 	    vi->rss_size, "size of RSS indirection table");
6330 
6331 	if (IS_MAIN_VI(vi)) {
6332 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
6333 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
6334 		    "Reserve queue 0 for non-flowid packets");
6335 	}
6336 
6337 #ifdef TCP_OFFLOAD
6338 	if (vi->nofldrxq != 0) {
6339 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
6340 		    &vi->nofldrxq, 0,
6341 		    "# of rx queues for offloaded TCP connections");
6342 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
6343 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
6344 		    "index of first TOE rx queue");
6345 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
6346 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
6347 		    sysctl_holdoff_tmr_idx_ofld, "I",
6348 		    "holdoff timer index for TOE queues");
6349 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
6350 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
6351 		    sysctl_holdoff_pktc_idx_ofld, "I",
6352 		    "holdoff packet counter index for TOE queues");
6353 	}
6354 #endif
6355 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
6356 	if (vi->nofldtxq != 0) {
6357 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
6358 		    &vi->nofldtxq, 0,
6359 		    "# of tx queues for TOE/ETHOFLD");
6360 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
6361 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
6362 		    "index of first TOE/ETHOFLD tx queue");
6363 	}
6364 #endif
6365 #ifdef DEV_NETMAP
6366 	if (vi->nnmrxq != 0) {
6367 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
6368 		    &vi->nnmrxq, 0, "# of netmap rx queues");
6369 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
6370 		    &vi->nnmtxq, 0, "# of netmap tx queues");
6371 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
6372 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
6373 		    "index of first netmap rx queue");
6374 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
6375 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
6376 		    "index of first netmap tx queue");
6377 	}
6378 #endif
6379 
6380 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
6381 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
6382 	    "holdoff timer index");
6383 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
6384 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
6385 	    "holdoff packet counter index");
6386 
6387 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
6388 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
6389 	    "rx queue size");
6390 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
6391 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
6392 	    "tx queue size");
6393 }
6394 
6395 static void
6396 cxgbe_sysctls(struct port_info *pi)
6397 {
6398 	struct sysctl_ctx_list *ctx;
6399 	struct sysctl_oid *oid;
6400 	struct sysctl_oid_list *children, *children2;
6401 	struct adapter *sc = pi->adapter;
6402 	int i;
6403 	char name[16];
6404 	static char *tc_flags = {"\20\1USER\2SYNC\3ASYNC\4ERR"};
6405 
6406 	ctx = device_get_sysctl_ctx(pi->dev);
6407 
6408 	/*
6409 	 * dev.cxgbe.X.
6410 	 */
6411 	oid = device_get_sysctl_tree(pi->dev);
6412 	children = SYSCTL_CHILDREN(oid);
6413 
6414 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
6415 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
6416 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
6417 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
6418 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
6419 		    "PHY temperature (in Celsius)");
6420 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
6421 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
6422 		    "PHY firmware version");
6423 	}
6424 
6425 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
6426 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
6427     "PAUSE settings (bit 0 = rx_pause, 1 = tx_pause, 2 = pause_autoneg)");
6428 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
6429 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
6430 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
6431 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
6432 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
6433 	    "autonegotiation (-1 = not supported)");
6434 
6435 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
6436 	    port_top_speed(pi), "max speed (in Gbps)");
6437 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
6438 	    pi->mps_bg_map, "MPS buffer group map");
6439 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
6440 	    NULL, pi->rx_e_chan_map, "TP rx e-channel map");
6441 
6442 	if (sc->flags & IS_VF)
6443 		return;
6444 
6445 	/*
6446 	 * dev.(cxgbe|cxl).X.tc.
6447 	 */
6448 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
6449 	    "Tx scheduler traffic classes (cl_rl)");
6450 	children2 = SYSCTL_CHILDREN(oid);
6451 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "pktsize",
6452 	    CTLFLAG_RW, &pi->sched_params->pktsize, 0,
6453 	    "pktsize for per-flow cl-rl (0 means up to the driver )");
6454 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "burstsize",
6455 	    CTLFLAG_RW, &pi->sched_params->burstsize, 0,
6456 	    "burstsize for per-flow cl-rl (0 means up to the driver)");
6457 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
6458 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
6459 
6460 		snprintf(name, sizeof(name), "%d", i);
6461 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
6462 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
6463 		    "traffic class"));
6464 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "flags",
6465 		    CTLTYPE_STRING | CTLFLAG_RD, tc_flags, (uintptr_t)&tc->flags,
6466 		    sysctl_bitfield_8b, "A", "flags");
6467 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
6468 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
6469 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
6470 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
6471 		    sysctl_tc_params, "A", "traffic class parameters");
6472 	}
6473 
6474 	/*
6475 	 * dev.cxgbe.X.stats.
6476 	 */
6477 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
6478 	    NULL, "port statistics");
6479 	children = SYSCTL_CHILDREN(oid);
6480 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
6481 	    &pi->tx_parse_error, 0,
6482 	    "# of tx packets with invalid length or # of segments");
6483 
6484 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
6485 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
6486 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
6487 	    sysctl_handle_t4_reg64, "QU", desc)
6488 
6489 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
6490 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
6491 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
6492 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
6493 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
6494 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
6495 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
6496 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
6497 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
6498 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
6499 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
6500 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
6501 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
6502 	    "# of tx frames in this range",
6503 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
6504 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
6505 	    "# of tx frames in this range",
6506 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
6507 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
6508 	    "# of tx frames in this range",
6509 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
6510 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
6511 	    "# of tx frames in this range",
6512 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
6513 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
6514 	    "# of tx frames in this range",
6515 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
6516 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
6517 	    "# of tx frames in this range",
6518 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
6519 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
6520 	    "# of tx frames in this range",
6521 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
6522 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
6523 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
6524 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
6525 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
6526 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
6527 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
6528 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
6529 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
6530 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
6531 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
6532 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
6533 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
6534 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
6535 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
6536 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
6537 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
6538 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
6539 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
6540 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
6541 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
6542 
6543 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
6544 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
6545 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
6546 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
6547 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
6548 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
6549 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
6550 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
6551 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
6552 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
6553 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
6554 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
6555 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
6556 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
6557 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
6558 	    "# of frames received with bad FCS",
6559 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
6560 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
6561 	    "# of frames received with length error",
6562 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
6563 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
6564 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
6565 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
6566 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
6567 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
6568 	    "# of rx frames in this range",
6569 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
6570 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
6571 	    "# of rx frames in this range",
6572 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
6573 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
6574 	    "# of rx frames in this range",
6575 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
6576 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
6577 	    "# of rx frames in this range",
6578 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
6579 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
6580 	    "# of rx frames in this range",
6581 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
6582 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
6583 	    "# of rx frames in this range",
6584 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
6585 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
6586 	    "# of rx frames in this range",
6587 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
6588 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
6589 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
6590 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
6591 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
6592 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
6593 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
6594 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
6595 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
6596 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
6597 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
6598 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
6599 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
6600 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
6601 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
6602 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
6603 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
6604 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
6605 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
6606 
6607 #undef SYSCTL_ADD_T4_REG64
6608 
6609 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
6610 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
6611 	    &pi->stats.name, desc)
6612 
6613 	/* We get these from port_stats and they may be stale by up to 1s */
6614 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
6615 	    "# drops due to buffer-group 0 overflows");
6616 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
6617 	    "# drops due to buffer-group 1 overflows");
6618 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
6619 	    "# drops due to buffer-group 2 overflows");
6620 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
6621 	    "# drops due to buffer-group 3 overflows");
6622 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
6623 	    "# of buffer-group 0 truncated packets");
6624 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
6625 	    "# of buffer-group 1 truncated packets");
6626 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
6627 	    "# of buffer-group 2 truncated packets");
6628 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
6629 	    "# of buffer-group 3 truncated packets");
6630 
6631 #undef SYSCTL_ADD_T4_PORTSTAT
6632 
6633 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_records",
6634 	    CTLFLAG_RD, &pi->tx_tls_records,
6635 	    "# of TLS records transmitted");
6636 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_octets",
6637 	    CTLFLAG_RD, &pi->tx_tls_octets,
6638 	    "# of payload octets in transmitted TLS records");
6639 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_records",
6640 	    CTLFLAG_RD, &pi->rx_tls_records,
6641 	    "# of TLS records received");
6642 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_octets",
6643 	    CTLFLAG_RD, &pi->rx_tls_octets,
6644 	    "# of payload octets in received TLS records");
6645 }
6646 
6647 static int
6648 sysctl_int_array(SYSCTL_HANDLER_ARGS)
6649 {
6650 	int rc, *i, space = 0;
6651 	struct sbuf sb;
6652 
6653 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
6654 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
6655 		if (space)
6656 			sbuf_printf(&sb, " ");
6657 		sbuf_printf(&sb, "%d", *i);
6658 		space = 1;
6659 	}
6660 	rc = sbuf_finish(&sb);
6661 	sbuf_delete(&sb);
6662 	return (rc);
6663 }
6664 
6665 static int
6666 sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS)
6667 {
6668 	int rc;
6669 	struct sbuf *sb;
6670 
6671 	rc = sysctl_wire_old_buffer(req, 0);
6672 	if (rc != 0)
6673 		return(rc);
6674 
6675 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6676 	if (sb == NULL)
6677 		return (ENOMEM);
6678 
6679 	sbuf_printf(sb, "%b", *(uint8_t *)(uintptr_t)arg2, (char *)arg1);
6680 	rc = sbuf_finish(sb);
6681 	sbuf_delete(sb);
6682 
6683 	return (rc);
6684 }
6685 
6686 static int
6687 sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS)
6688 {
6689 	int rc;
6690 	struct sbuf *sb;
6691 
6692 	rc = sysctl_wire_old_buffer(req, 0);
6693 	if (rc != 0)
6694 		return(rc);
6695 
6696 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6697 	if (sb == NULL)
6698 		return (ENOMEM);
6699 
6700 	sbuf_printf(sb, "%b", *(uint16_t *)(uintptr_t)arg2, (char *)arg1);
6701 	rc = sbuf_finish(sb);
6702 	sbuf_delete(sb);
6703 
6704 	return (rc);
6705 }
6706 
6707 static int
6708 sysctl_btphy(SYSCTL_HANDLER_ARGS)
6709 {
6710 	struct port_info *pi = arg1;
6711 	int op = arg2;
6712 	struct adapter *sc = pi->adapter;
6713 	u_int v;
6714 	int rc;
6715 
6716 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
6717 	if (rc)
6718 		return (rc);
6719 	/* XXX: magic numbers */
6720 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
6721 	    &v);
6722 	end_synchronized_op(sc, 0);
6723 	if (rc)
6724 		return (rc);
6725 	if (op == 0)
6726 		v /= 256;
6727 
6728 	rc = sysctl_handle_int(oidp, &v, 0, req);
6729 	return (rc);
6730 }
6731 
6732 static int
6733 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
6734 {
6735 	struct vi_info *vi = arg1;
6736 	int rc, val;
6737 
6738 	val = vi->rsrv_noflowq;
6739 	rc = sysctl_handle_int(oidp, &val, 0, req);
6740 	if (rc != 0 || req->newptr == NULL)
6741 		return (rc);
6742 
6743 	if ((val >= 1) && (vi->ntxq > 1))
6744 		vi->rsrv_noflowq = 1;
6745 	else
6746 		vi->rsrv_noflowq = 0;
6747 
6748 	return (rc);
6749 }
6750 
6751 static int
6752 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
6753 {
6754 	struct vi_info *vi = arg1;
6755 	struct adapter *sc = vi->pi->adapter;
6756 	int idx, rc, i;
6757 	struct sge_rxq *rxq;
6758 	uint8_t v;
6759 
6760 	idx = vi->tmr_idx;
6761 
6762 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6763 	if (rc != 0 || req->newptr == NULL)
6764 		return (rc);
6765 
6766 	if (idx < 0 || idx >= SGE_NTIMERS)
6767 		return (EINVAL);
6768 
6769 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6770 	    "t4tmr");
6771 	if (rc)
6772 		return (rc);
6773 
6774 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
6775 	for_each_rxq(vi, i, rxq) {
6776 #ifdef atomic_store_rel_8
6777 		atomic_store_rel_8(&rxq->iq.intr_params, v);
6778 #else
6779 		rxq->iq.intr_params = v;
6780 #endif
6781 	}
6782 	vi->tmr_idx = idx;
6783 
6784 	end_synchronized_op(sc, LOCK_HELD);
6785 	return (0);
6786 }
6787 
6788 static int
6789 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
6790 {
6791 	struct vi_info *vi = arg1;
6792 	struct adapter *sc = vi->pi->adapter;
6793 	int idx, rc;
6794 
6795 	idx = vi->pktc_idx;
6796 
6797 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6798 	if (rc != 0 || req->newptr == NULL)
6799 		return (rc);
6800 
6801 	if (idx < -1 || idx >= SGE_NCOUNTERS)
6802 		return (EINVAL);
6803 
6804 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6805 	    "t4pktc");
6806 	if (rc)
6807 		return (rc);
6808 
6809 	if (vi->flags & VI_INIT_DONE)
6810 		rc = EBUSY; /* cannot be changed once the queues are created */
6811 	else
6812 		vi->pktc_idx = idx;
6813 
6814 	end_synchronized_op(sc, LOCK_HELD);
6815 	return (rc);
6816 }
6817 
6818 static int
6819 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
6820 {
6821 	struct vi_info *vi = arg1;
6822 	struct adapter *sc = vi->pi->adapter;
6823 	int qsize, rc;
6824 
6825 	qsize = vi->qsize_rxq;
6826 
6827 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6828 	if (rc != 0 || req->newptr == NULL)
6829 		return (rc);
6830 
6831 	if (qsize < 128 || (qsize & 7))
6832 		return (EINVAL);
6833 
6834 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6835 	    "t4rxqs");
6836 	if (rc)
6837 		return (rc);
6838 
6839 	if (vi->flags & VI_INIT_DONE)
6840 		rc = EBUSY; /* cannot be changed once the queues are created */
6841 	else
6842 		vi->qsize_rxq = qsize;
6843 
6844 	end_synchronized_op(sc, LOCK_HELD);
6845 	return (rc);
6846 }
6847 
6848 static int
6849 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
6850 {
6851 	struct vi_info *vi = arg1;
6852 	struct adapter *sc = vi->pi->adapter;
6853 	int qsize, rc;
6854 
6855 	qsize = vi->qsize_txq;
6856 
6857 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6858 	if (rc != 0 || req->newptr == NULL)
6859 		return (rc);
6860 
6861 	if (qsize < 128 || qsize > 65536)
6862 		return (EINVAL);
6863 
6864 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6865 	    "t4txqs");
6866 	if (rc)
6867 		return (rc);
6868 
6869 	if (vi->flags & VI_INIT_DONE)
6870 		rc = EBUSY; /* cannot be changed once the queues are created */
6871 	else
6872 		vi->qsize_txq = qsize;
6873 
6874 	end_synchronized_op(sc, LOCK_HELD);
6875 	return (rc);
6876 }
6877 
6878 static int
6879 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
6880 {
6881 	struct port_info *pi = arg1;
6882 	struct adapter *sc = pi->adapter;
6883 	struct link_config *lc = &pi->link_cfg;
6884 	int rc;
6885 
6886 	if (req->newptr == NULL) {
6887 		struct sbuf *sb;
6888 		static char *bits = "\20\1RX\2TX\3AUTO";
6889 
6890 		rc = sysctl_wire_old_buffer(req, 0);
6891 		if (rc != 0)
6892 			return(rc);
6893 
6894 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6895 		if (sb == NULL)
6896 			return (ENOMEM);
6897 
6898 		if (lc->link_ok) {
6899 			sbuf_printf(sb, "%b", (lc->fc & (PAUSE_TX | PAUSE_RX)) |
6900 			    (lc->requested_fc & PAUSE_AUTONEG), bits);
6901 		} else {
6902 			sbuf_printf(sb, "%b", lc->requested_fc & (PAUSE_TX |
6903 			    PAUSE_RX | PAUSE_AUTONEG), bits);
6904 		}
6905 		rc = sbuf_finish(sb);
6906 		sbuf_delete(sb);
6907 	} else {
6908 		char s[2];
6909 		int n;
6910 
6911 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX |
6912 		    PAUSE_AUTONEG));
6913 		s[1] = 0;
6914 
6915 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6916 		if (rc != 0)
6917 			return(rc);
6918 
6919 		if (s[1] != 0)
6920 			return (EINVAL);
6921 		if (s[0] < '0' || s[0] > '9')
6922 			return (EINVAL);	/* not a number */
6923 		n = s[0] - '0';
6924 		if (n & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG))
6925 			return (EINVAL);	/* some other bit is set too */
6926 
6927 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6928 		    "t4PAUSE");
6929 		if (rc)
6930 			return (rc);
6931 		PORT_LOCK(pi);
6932 		lc->requested_fc = n;
6933 		fixup_link_config(pi);
6934 		if (pi->up_vis > 0)
6935 			rc = apply_link_config(pi);
6936 		set_current_media(pi);
6937 		PORT_UNLOCK(pi);
6938 		end_synchronized_op(sc, 0);
6939 	}
6940 
6941 	return (rc);
6942 }
6943 
6944 static int
6945 sysctl_fec(SYSCTL_HANDLER_ARGS)
6946 {
6947 	struct port_info *pi = arg1;
6948 	struct adapter *sc = pi->adapter;
6949 	struct link_config *lc = &pi->link_cfg;
6950 	int rc;
6951 	int8_t old;
6952 
6953 	if (req->newptr == NULL) {
6954 		struct sbuf *sb;
6955 		static char *bits = "\20\1RS\2BASE-R\3RSVD1\4RSVD2\5RSVD3\6AUTO";
6956 
6957 		rc = sysctl_wire_old_buffer(req, 0);
6958 		if (rc != 0)
6959 			return(rc);
6960 
6961 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6962 		if (sb == NULL)
6963 			return (ENOMEM);
6964 
6965 		/*
6966 		 * Display the requested_fec when the link is down -- the actual
6967 		 * FEC makes sense only when the link is up.
6968 		 */
6969 		if (lc->link_ok) {
6970 			sbuf_printf(sb, "%b", (lc->fec & M_FW_PORT_CAP32_FEC) |
6971 			    (lc->requested_fec & FEC_AUTO), bits);
6972 		} else {
6973 			sbuf_printf(sb, "%b", lc->requested_fec, bits);
6974 		}
6975 		rc = sbuf_finish(sb);
6976 		sbuf_delete(sb);
6977 	} else {
6978 		char s[3];
6979 		int n;
6980 
6981 		snprintf(s, sizeof(s), "%d",
6982 		    lc->requested_fec == FEC_AUTO ? -1 :
6983 		    lc->requested_fec & M_FW_PORT_CAP32_FEC);
6984 
6985 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6986 		if (rc != 0)
6987 			return(rc);
6988 
6989 		n = strtol(&s[0], NULL, 0);
6990 		if (n < 0 || n & FEC_AUTO)
6991 			n = FEC_AUTO;
6992 		else {
6993 			if (n & ~M_FW_PORT_CAP32_FEC)
6994 				return (EINVAL);/* some other bit is set too */
6995 			if (!powerof2(n))
6996 				return (EINVAL);/* one bit can be set at most */
6997 		}
6998 
6999 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
7000 		    "t4fec");
7001 		if (rc)
7002 			return (rc);
7003 		PORT_LOCK(pi);
7004 		old = lc->requested_fec;
7005 		if (n == FEC_AUTO)
7006 			lc->requested_fec = FEC_AUTO;
7007 		else if (n == 0)
7008 			lc->requested_fec = FEC_NONE;
7009 		else {
7010 			if ((lc->supported | V_FW_PORT_CAP32_FEC(n)) !=
7011 			    lc->supported) {
7012 				rc = ENOTSUP;
7013 				goto done;
7014 			}
7015 			lc->requested_fec = n;
7016 		}
7017 		fixup_link_config(pi);
7018 		if (pi->up_vis > 0) {
7019 			rc = apply_link_config(pi);
7020 			if (rc != 0) {
7021 				lc->requested_fec = old;
7022 				if (rc == FW_EPROTO)
7023 					rc = ENOTSUP;
7024 			}
7025 		}
7026 done:
7027 		PORT_UNLOCK(pi);
7028 		end_synchronized_op(sc, 0);
7029 	}
7030 
7031 	return (rc);
7032 }
7033 
7034 static int
7035 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
7036 {
7037 	struct port_info *pi = arg1;
7038 	struct adapter *sc = pi->adapter;
7039 	struct link_config *lc = &pi->link_cfg;
7040 	int rc, val;
7041 
7042 	if (lc->supported & FW_PORT_CAP32_ANEG)
7043 		val = lc->requested_aneg == AUTONEG_DISABLE ? 0 : 1;
7044 	else
7045 		val = -1;
7046 	rc = sysctl_handle_int(oidp, &val, 0, req);
7047 	if (rc != 0 || req->newptr == NULL)
7048 		return (rc);
7049 	if (val == 0)
7050 		val = AUTONEG_DISABLE;
7051 	else if (val == 1)
7052 		val = AUTONEG_ENABLE;
7053 	else
7054 		val = AUTONEG_AUTO;
7055 
7056 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
7057 	    "t4aneg");
7058 	if (rc)
7059 		return (rc);
7060 	PORT_LOCK(pi);
7061 	if (val == AUTONEG_ENABLE && !(lc->supported & FW_PORT_CAP32_ANEG)) {
7062 		rc = ENOTSUP;
7063 		goto done;
7064 	}
7065 	lc->requested_aneg = val;
7066 	fixup_link_config(pi);
7067 	if (pi->up_vis > 0)
7068 		rc = apply_link_config(pi);
7069 	set_current_media(pi);
7070 done:
7071 	PORT_UNLOCK(pi);
7072 	end_synchronized_op(sc, 0);
7073 	return (rc);
7074 }
7075 
7076 static int
7077 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
7078 {
7079 	struct adapter *sc = arg1;
7080 	int reg = arg2;
7081 	uint64_t val;
7082 
7083 	val = t4_read_reg64(sc, reg);
7084 
7085 	return (sysctl_handle_64(oidp, &val, 0, req));
7086 }
7087 
7088 static int
7089 sysctl_temperature(SYSCTL_HANDLER_ARGS)
7090 {
7091 	struct adapter *sc = arg1;
7092 	int rc, t;
7093 	uint32_t param, val;
7094 
7095 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
7096 	if (rc)
7097 		return (rc);
7098 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
7099 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
7100 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
7101 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
7102 	end_synchronized_op(sc, 0);
7103 	if (rc)
7104 		return (rc);
7105 
7106 	/* unknown is returned as 0 but we display -1 in that case */
7107 	t = val == 0 ? -1 : val;
7108 
7109 	rc = sysctl_handle_int(oidp, &t, 0, req);
7110 	return (rc);
7111 }
7112 
7113 static int
7114 sysctl_loadavg(SYSCTL_HANDLER_ARGS)
7115 {
7116 	struct adapter *sc = arg1;
7117 	struct sbuf *sb;
7118 	int rc;
7119 	uint32_t param, val;
7120 
7121 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4lavg");
7122 	if (rc)
7123 		return (rc);
7124 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
7125 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_LOAD);
7126 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
7127 	end_synchronized_op(sc, 0);
7128 	if (rc)
7129 		return (rc);
7130 
7131 	rc = sysctl_wire_old_buffer(req, 0);
7132 	if (rc != 0)
7133 		return (rc);
7134 
7135 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7136 	if (sb == NULL)
7137 		return (ENOMEM);
7138 
7139 	if (val == 0xffffffff) {
7140 		/* Only debug and custom firmwares report load averages. */
7141 		sbuf_printf(sb, "not available");
7142 	} else {
7143 		sbuf_printf(sb, "%d %d %d", val & 0xff, (val >> 8) & 0xff,
7144 		    (val >> 16) & 0xff);
7145 	}
7146 	rc = sbuf_finish(sb);
7147 	sbuf_delete(sb);
7148 
7149 	return (rc);
7150 }
7151 
7152 static int
7153 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
7154 {
7155 	struct adapter *sc = arg1;
7156 	struct sbuf *sb;
7157 	int rc, i;
7158 	uint16_t incr[NMTUS][NCCTRL_WIN];
7159 	static const char *dec_fac[] = {
7160 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
7161 		"0.9375"
7162 	};
7163 
7164 	rc = sysctl_wire_old_buffer(req, 0);
7165 	if (rc != 0)
7166 		return (rc);
7167 
7168 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7169 	if (sb == NULL)
7170 		return (ENOMEM);
7171 
7172 	t4_read_cong_tbl(sc, incr);
7173 
7174 	for (i = 0; i < NCCTRL_WIN; ++i) {
7175 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
7176 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
7177 		    incr[5][i], incr[6][i], incr[7][i]);
7178 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
7179 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
7180 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
7181 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
7182 	}
7183 
7184 	rc = sbuf_finish(sb);
7185 	sbuf_delete(sb);
7186 
7187 	return (rc);
7188 }
7189 
7190 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
7191 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
7192 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
7193 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
7194 };
7195 
7196 static int
7197 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
7198 {
7199 	struct adapter *sc = arg1;
7200 	struct sbuf *sb;
7201 	int rc, i, n, qid = arg2;
7202 	uint32_t *buf, *p;
7203 	char *qtype;
7204 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
7205 
7206 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
7207 	    ("%s: bad qid %d\n", __func__, qid));
7208 
7209 	if (qid < CIM_NUM_IBQ) {
7210 		/* inbound queue */
7211 		qtype = "IBQ";
7212 		n = 4 * CIM_IBQ_SIZE;
7213 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
7214 		rc = t4_read_cim_ibq(sc, qid, buf, n);
7215 	} else {
7216 		/* outbound queue */
7217 		qtype = "OBQ";
7218 		qid -= CIM_NUM_IBQ;
7219 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
7220 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
7221 		rc = t4_read_cim_obq(sc, qid, buf, n);
7222 	}
7223 
7224 	if (rc < 0) {
7225 		rc = -rc;
7226 		goto done;
7227 	}
7228 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
7229 
7230 	rc = sysctl_wire_old_buffer(req, 0);
7231 	if (rc != 0)
7232 		goto done;
7233 
7234 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7235 	if (sb == NULL) {
7236 		rc = ENOMEM;
7237 		goto done;
7238 	}
7239 
7240 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
7241 	for (i = 0, p = buf; i < n; i += 16, p += 4)
7242 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
7243 		    p[2], p[3]);
7244 
7245 	rc = sbuf_finish(sb);
7246 	sbuf_delete(sb);
7247 done:
7248 	free(buf, M_CXGBE);
7249 	return (rc);
7250 }
7251 
7252 static int
7253 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
7254 {
7255 	struct adapter *sc = arg1;
7256 	u_int cfg;
7257 	struct sbuf *sb;
7258 	uint32_t *buf, *p;
7259 	int rc;
7260 
7261 	MPASS(chip_id(sc) <= CHELSIO_T5);
7262 
7263 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
7264 	if (rc != 0)
7265 		return (rc);
7266 
7267 	rc = sysctl_wire_old_buffer(req, 0);
7268 	if (rc != 0)
7269 		return (rc);
7270 
7271 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7272 	if (sb == NULL)
7273 		return (ENOMEM);
7274 
7275 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
7276 	    M_ZERO | M_WAITOK);
7277 
7278 	rc = -t4_cim_read_la(sc, buf, NULL);
7279 	if (rc != 0)
7280 		goto done;
7281 
7282 	sbuf_printf(sb, "Status   Data      PC%s",
7283 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
7284 	    "     LS0Stat  LS0Addr             LS0Data");
7285 
7286 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
7287 		if (cfg & F_UPDBGLACAPTPCONLY) {
7288 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
7289 			    p[6], p[7]);
7290 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
7291 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
7292 			    p[4] & 0xff, p[5] >> 8);
7293 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
7294 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7295 			    p[1] & 0xf, p[2] >> 4);
7296 		} else {
7297 			sbuf_printf(sb,
7298 			    "\n  %02x   %x%07x %x%07x %08x %08x "
7299 			    "%08x%08x%08x%08x",
7300 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7301 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
7302 			    p[6], p[7]);
7303 		}
7304 	}
7305 
7306 	rc = sbuf_finish(sb);
7307 	sbuf_delete(sb);
7308 done:
7309 	free(buf, M_CXGBE);
7310 	return (rc);
7311 }
7312 
7313 static int
7314 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
7315 {
7316 	struct adapter *sc = arg1;
7317 	u_int cfg;
7318 	struct sbuf *sb;
7319 	uint32_t *buf, *p;
7320 	int rc;
7321 
7322 	MPASS(chip_id(sc) > CHELSIO_T5);
7323 
7324 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
7325 	if (rc != 0)
7326 		return (rc);
7327 
7328 	rc = sysctl_wire_old_buffer(req, 0);
7329 	if (rc != 0)
7330 		return (rc);
7331 
7332 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7333 	if (sb == NULL)
7334 		return (ENOMEM);
7335 
7336 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
7337 	    M_ZERO | M_WAITOK);
7338 
7339 	rc = -t4_cim_read_la(sc, buf, NULL);
7340 	if (rc != 0)
7341 		goto done;
7342 
7343 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
7344 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
7345 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
7346 
7347 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
7348 		if (cfg & F_UPDBGLACAPTPCONLY) {
7349 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
7350 			    p[3] & 0xff, p[2], p[1], p[0]);
7351 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
7352 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
7353 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
7354 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
7355 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
7356 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
7357 			    p[6] >> 16);
7358 		} else {
7359 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
7360 			    "%08x %08x %08x %08x %08x %08x",
7361 			    (p[9] >> 16) & 0xff,
7362 			    p[9] & 0xffff, p[8] >> 16,
7363 			    p[8] & 0xffff, p[7] >> 16,
7364 			    p[7] & 0xffff, p[6] >> 16,
7365 			    p[2], p[1], p[0], p[5], p[4], p[3]);
7366 		}
7367 	}
7368 
7369 	rc = sbuf_finish(sb);
7370 	sbuf_delete(sb);
7371 done:
7372 	free(buf, M_CXGBE);
7373 	return (rc);
7374 }
7375 
7376 static int
7377 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
7378 {
7379 	struct adapter *sc = arg1;
7380 	u_int i;
7381 	struct sbuf *sb;
7382 	uint32_t *buf, *p;
7383 	int rc;
7384 
7385 	rc = sysctl_wire_old_buffer(req, 0);
7386 	if (rc != 0)
7387 		return (rc);
7388 
7389 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7390 	if (sb == NULL)
7391 		return (ENOMEM);
7392 
7393 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
7394 	    M_ZERO | M_WAITOK);
7395 
7396 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
7397 	p = buf;
7398 
7399 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7400 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
7401 		    p[1], p[0]);
7402 	}
7403 
7404 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
7405 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7406 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
7407 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
7408 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
7409 		    (p[1] >> 2) | ((p[2] & 3) << 30),
7410 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
7411 		    p[0] & 1);
7412 	}
7413 
7414 	rc = sbuf_finish(sb);
7415 	sbuf_delete(sb);
7416 	free(buf, M_CXGBE);
7417 	return (rc);
7418 }
7419 
7420 static int
7421 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
7422 {
7423 	struct adapter *sc = arg1;
7424 	u_int i;
7425 	struct sbuf *sb;
7426 	uint32_t *buf, *p;
7427 	int rc;
7428 
7429 	rc = sysctl_wire_old_buffer(req, 0);
7430 	if (rc != 0)
7431 		return (rc);
7432 
7433 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7434 	if (sb == NULL)
7435 		return (ENOMEM);
7436 
7437 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
7438 	    M_ZERO | M_WAITOK);
7439 
7440 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
7441 	p = buf;
7442 
7443 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
7444 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7445 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
7446 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
7447 		    p[4], p[3], p[2], p[1], p[0]);
7448 	}
7449 
7450 	sbuf_printf(sb, "\n\nCntl ID               Data");
7451 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7452 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
7453 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
7454 	}
7455 
7456 	rc = sbuf_finish(sb);
7457 	sbuf_delete(sb);
7458 	free(buf, M_CXGBE);
7459 	return (rc);
7460 }
7461 
7462 static int
7463 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
7464 {
7465 	struct adapter *sc = arg1;
7466 	struct sbuf *sb;
7467 	int rc, i;
7468 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7469 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7470 	uint16_t thres[CIM_NUM_IBQ];
7471 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
7472 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
7473 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
7474 
7475 	cim_num_obq = sc->chip_params->cim_num_obq;
7476 	if (is_t4(sc)) {
7477 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
7478 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
7479 	} else {
7480 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
7481 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
7482 	}
7483 	nq = CIM_NUM_IBQ + cim_num_obq;
7484 
7485 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
7486 	if (rc == 0)
7487 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
7488 	if (rc != 0)
7489 		return (rc);
7490 
7491 	t4_read_cimq_cfg(sc, base, size, thres);
7492 
7493 	rc = sysctl_wire_old_buffer(req, 0);
7494 	if (rc != 0)
7495 		return (rc);
7496 
7497 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7498 	if (sb == NULL)
7499 		return (ENOMEM);
7500 
7501 	sbuf_printf(sb,
7502 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
7503 
7504 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
7505 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
7506 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
7507 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7508 		    G_QUEREMFLITS(p[2]) * 16);
7509 	for ( ; i < nq; i++, p += 4, wr += 2)
7510 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
7511 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
7512 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7513 		    G_QUEREMFLITS(p[2]) * 16);
7514 
7515 	rc = sbuf_finish(sb);
7516 	sbuf_delete(sb);
7517 
7518 	return (rc);
7519 }
7520 
7521 static int
7522 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
7523 {
7524 	struct adapter *sc = arg1;
7525 	struct sbuf *sb;
7526 	int rc;
7527 	struct tp_cpl_stats stats;
7528 
7529 	rc = sysctl_wire_old_buffer(req, 0);
7530 	if (rc != 0)
7531 		return (rc);
7532 
7533 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7534 	if (sb == NULL)
7535 		return (ENOMEM);
7536 
7537 	mtx_lock(&sc->reg_lock);
7538 	t4_tp_get_cpl_stats(sc, &stats, 0);
7539 	mtx_unlock(&sc->reg_lock);
7540 
7541 	if (sc->chip_params->nchan > 2) {
7542 		sbuf_printf(sb, "                 channel 0  channel 1"
7543 		    "  channel 2  channel 3");
7544 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
7545 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
7546 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
7547 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
7548 	} else {
7549 		sbuf_printf(sb, "                 channel 0  channel 1");
7550 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
7551 		    stats.req[0], stats.req[1]);
7552 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
7553 		    stats.rsp[0], stats.rsp[1]);
7554 	}
7555 
7556 	rc = sbuf_finish(sb);
7557 	sbuf_delete(sb);
7558 
7559 	return (rc);
7560 }
7561 
7562 static int
7563 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
7564 {
7565 	struct adapter *sc = arg1;
7566 	struct sbuf *sb;
7567 	int rc;
7568 	struct tp_usm_stats stats;
7569 
7570 	rc = sysctl_wire_old_buffer(req, 0);
7571 	if (rc != 0)
7572 		return(rc);
7573 
7574 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7575 	if (sb == NULL)
7576 		return (ENOMEM);
7577 
7578 	t4_get_usm_stats(sc, &stats, 1);
7579 
7580 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
7581 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
7582 	sbuf_printf(sb, "Drops:  %u", stats.drops);
7583 
7584 	rc = sbuf_finish(sb);
7585 	sbuf_delete(sb);
7586 
7587 	return (rc);
7588 }
7589 
7590 static const char * const devlog_level_strings[] = {
7591 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
7592 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
7593 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
7594 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
7595 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
7596 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
7597 };
7598 
7599 static const char * const devlog_facility_strings[] = {
7600 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
7601 	[FW_DEVLOG_FACILITY_CF]		= "CF",
7602 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
7603 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
7604 	[FW_DEVLOG_FACILITY_RES]	= "RES",
7605 	[FW_DEVLOG_FACILITY_HW]		= "HW",
7606 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
7607 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
7608 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
7609 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
7610 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
7611 	[FW_DEVLOG_FACILITY_VI]		= "VI",
7612 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
7613 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
7614 	[FW_DEVLOG_FACILITY_TM]		= "TM",
7615 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
7616 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
7617 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
7618 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
7619 	[FW_DEVLOG_FACILITY_RI]		= "RI",
7620 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
7621 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
7622 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
7623 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
7624 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
7625 };
7626 
7627 static int
7628 sysctl_devlog(SYSCTL_HANDLER_ARGS)
7629 {
7630 	struct adapter *sc = arg1;
7631 	struct devlog_params *dparams = &sc->params.devlog;
7632 	struct fw_devlog_e *buf, *e;
7633 	int i, j, rc, nentries, first = 0;
7634 	struct sbuf *sb;
7635 	uint64_t ftstamp = UINT64_MAX;
7636 
7637 	if (dparams->addr == 0)
7638 		return (ENXIO);
7639 
7640 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
7641 	if (buf == NULL)
7642 		return (ENOMEM);
7643 
7644 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
7645 	if (rc != 0)
7646 		goto done;
7647 
7648 	nentries = dparams->size / sizeof(struct fw_devlog_e);
7649 	for (i = 0; i < nentries; i++) {
7650 		e = &buf[i];
7651 
7652 		if (e->timestamp == 0)
7653 			break;	/* end */
7654 
7655 		e->timestamp = be64toh(e->timestamp);
7656 		e->seqno = be32toh(e->seqno);
7657 		for (j = 0; j < 8; j++)
7658 			e->params[j] = be32toh(e->params[j]);
7659 
7660 		if (e->timestamp < ftstamp) {
7661 			ftstamp = e->timestamp;
7662 			first = i;
7663 		}
7664 	}
7665 
7666 	if (buf[first].timestamp == 0)
7667 		goto done;	/* nothing in the log */
7668 
7669 	rc = sysctl_wire_old_buffer(req, 0);
7670 	if (rc != 0)
7671 		goto done;
7672 
7673 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7674 	if (sb == NULL) {
7675 		rc = ENOMEM;
7676 		goto done;
7677 	}
7678 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
7679 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
7680 
7681 	i = first;
7682 	do {
7683 		e = &buf[i];
7684 		if (e->timestamp == 0)
7685 			break;	/* end */
7686 
7687 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
7688 		    e->seqno, e->timestamp,
7689 		    (e->level < nitems(devlog_level_strings) ?
7690 			devlog_level_strings[e->level] : "UNKNOWN"),
7691 		    (e->facility < nitems(devlog_facility_strings) ?
7692 			devlog_facility_strings[e->facility] : "UNKNOWN"));
7693 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
7694 		    e->params[2], e->params[3], e->params[4],
7695 		    e->params[5], e->params[6], e->params[7]);
7696 
7697 		if (++i == nentries)
7698 			i = 0;
7699 	} while (i != first);
7700 
7701 	rc = sbuf_finish(sb);
7702 	sbuf_delete(sb);
7703 done:
7704 	free(buf, M_CXGBE);
7705 	return (rc);
7706 }
7707 
7708 static int
7709 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
7710 {
7711 	struct adapter *sc = arg1;
7712 	struct sbuf *sb;
7713 	int rc;
7714 	struct tp_fcoe_stats stats[MAX_NCHAN];
7715 	int i, nchan = sc->chip_params->nchan;
7716 
7717 	rc = sysctl_wire_old_buffer(req, 0);
7718 	if (rc != 0)
7719 		return (rc);
7720 
7721 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7722 	if (sb == NULL)
7723 		return (ENOMEM);
7724 
7725 	for (i = 0; i < nchan; i++)
7726 		t4_get_fcoe_stats(sc, i, &stats[i], 1);
7727 
7728 	if (nchan > 2) {
7729 		sbuf_printf(sb, "                   channel 0        channel 1"
7730 		    "        channel 2        channel 3");
7731 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
7732 		    stats[0].octets_ddp, stats[1].octets_ddp,
7733 		    stats[2].octets_ddp, stats[3].octets_ddp);
7734 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
7735 		    stats[0].frames_ddp, stats[1].frames_ddp,
7736 		    stats[2].frames_ddp, stats[3].frames_ddp);
7737 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
7738 		    stats[0].frames_drop, stats[1].frames_drop,
7739 		    stats[2].frames_drop, stats[3].frames_drop);
7740 	} else {
7741 		sbuf_printf(sb, "                   channel 0        channel 1");
7742 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
7743 		    stats[0].octets_ddp, stats[1].octets_ddp);
7744 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
7745 		    stats[0].frames_ddp, stats[1].frames_ddp);
7746 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
7747 		    stats[0].frames_drop, stats[1].frames_drop);
7748 	}
7749 
7750 	rc = sbuf_finish(sb);
7751 	sbuf_delete(sb);
7752 
7753 	return (rc);
7754 }
7755 
7756 static int
7757 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
7758 {
7759 	struct adapter *sc = arg1;
7760 	struct sbuf *sb;
7761 	int rc, i;
7762 	unsigned int map, kbps, ipg, mode;
7763 	unsigned int pace_tab[NTX_SCHED];
7764 
7765 	rc = sysctl_wire_old_buffer(req, 0);
7766 	if (rc != 0)
7767 		return (rc);
7768 
7769 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7770 	if (sb == NULL)
7771 		return (ENOMEM);
7772 
7773 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
7774 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
7775 	t4_read_pace_tbl(sc, pace_tab);
7776 
7777 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
7778 	    "Class IPG (0.1 ns)   Flow IPG (us)");
7779 
7780 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
7781 		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
7782 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
7783 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
7784 		if (kbps)
7785 			sbuf_printf(sb, "%9u     ", kbps);
7786 		else
7787 			sbuf_printf(sb, " disabled     ");
7788 
7789 		if (ipg)
7790 			sbuf_printf(sb, "%13u        ", ipg);
7791 		else
7792 			sbuf_printf(sb, "     disabled        ");
7793 
7794 		if (pace_tab[i])
7795 			sbuf_printf(sb, "%10u", pace_tab[i]);
7796 		else
7797 			sbuf_printf(sb, "  disabled");
7798 	}
7799 
7800 	rc = sbuf_finish(sb);
7801 	sbuf_delete(sb);
7802 
7803 	return (rc);
7804 }
7805 
7806 static int
7807 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
7808 {
7809 	struct adapter *sc = arg1;
7810 	struct sbuf *sb;
7811 	int rc, i, j;
7812 	uint64_t *p0, *p1;
7813 	struct lb_port_stats s[2];
7814 	static const char *stat_name[] = {
7815 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
7816 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
7817 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
7818 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
7819 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
7820 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
7821 		"BG2FramesTrunc:", "BG3FramesTrunc:"
7822 	};
7823 
7824 	rc = sysctl_wire_old_buffer(req, 0);
7825 	if (rc != 0)
7826 		return (rc);
7827 
7828 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7829 	if (sb == NULL)
7830 		return (ENOMEM);
7831 
7832 	memset(s, 0, sizeof(s));
7833 
7834 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
7835 		t4_get_lb_stats(sc, i, &s[0]);
7836 		t4_get_lb_stats(sc, i + 1, &s[1]);
7837 
7838 		p0 = &s[0].octets;
7839 		p1 = &s[1].octets;
7840 		sbuf_printf(sb, "%s                       Loopback %u"
7841 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
7842 
7843 		for (j = 0; j < nitems(stat_name); j++)
7844 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
7845 				   *p0++, *p1++);
7846 	}
7847 
7848 	rc = sbuf_finish(sb);
7849 	sbuf_delete(sb);
7850 
7851 	return (rc);
7852 }
7853 
7854 static int
7855 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
7856 {
7857 	int rc = 0;
7858 	struct port_info *pi = arg1;
7859 	struct link_config *lc = &pi->link_cfg;
7860 	struct sbuf *sb;
7861 
7862 	rc = sysctl_wire_old_buffer(req, 0);
7863 	if (rc != 0)
7864 		return(rc);
7865 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
7866 	if (sb == NULL)
7867 		return (ENOMEM);
7868 
7869 	if (lc->link_ok || lc->link_down_rc == 255)
7870 		sbuf_printf(sb, "n/a");
7871 	else
7872 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
7873 
7874 	rc = sbuf_finish(sb);
7875 	sbuf_delete(sb);
7876 
7877 	return (rc);
7878 }
7879 
7880 struct mem_desc {
7881 	unsigned int base;
7882 	unsigned int limit;
7883 	unsigned int idx;
7884 };
7885 
7886 static int
7887 mem_desc_cmp(const void *a, const void *b)
7888 {
7889 	return ((const struct mem_desc *)a)->base -
7890 	       ((const struct mem_desc *)b)->base;
7891 }
7892 
7893 static void
7894 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
7895     unsigned int to)
7896 {
7897 	unsigned int size;
7898 
7899 	if (from == to)
7900 		return;
7901 
7902 	size = to - from + 1;
7903 	if (size == 0)
7904 		return;
7905 
7906 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
7907 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
7908 }
7909 
7910 static int
7911 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
7912 {
7913 	struct adapter *sc = arg1;
7914 	struct sbuf *sb;
7915 	int rc, i, n;
7916 	uint32_t lo, hi, used, alloc;
7917 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
7918 	static const char *region[] = {
7919 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
7920 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
7921 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
7922 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
7923 		"RQUDP region:", "PBL region:", "TXPBL region:",
7924 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
7925 		"On-chip queues:", "TLS keys:",
7926 	};
7927 	struct mem_desc avail[4];
7928 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
7929 	struct mem_desc *md = mem;
7930 
7931 	rc = sysctl_wire_old_buffer(req, 0);
7932 	if (rc != 0)
7933 		return (rc);
7934 
7935 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7936 	if (sb == NULL)
7937 		return (ENOMEM);
7938 
7939 	for (i = 0; i < nitems(mem); i++) {
7940 		mem[i].limit = 0;
7941 		mem[i].idx = i;
7942 	}
7943 
7944 	/* Find and sort the populated memory ranges */
7945 	i = 0;
7946 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
7947 	if (lo & F_EDRAM0_ENABLE) {
7948 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
7949 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
7950 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
7951 		avail[i].idx = 0;
7952 		i++;
7953 	}
7954 	if (lo & F_EDRAM1_ENABLE) {
7955 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
7956 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
7957 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
7958 		avail[i].idx = 1;
7959 		i++;
7960 	}
7961 	if (lo & F_EXT_MEM_ENABLE) {
7962 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
7963 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
7964 		avail[i].limit = avail[i].base +
7965 		    (G_EXT_MEM_SIZE(hi) << 20);
7966 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
7967 		i++;
7968 	}
7969 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
7970 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
7971 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
7972 		avail[i].limit = avail[i].base +
7973 		    (G_EXT_MEM1_SIZE(hi) << 20);
7974 		avail[i].idx = 4;
7975 		i++;
7976 	}
7977 	if (!i)                                    /* no memory available */
7978 		return 0;
7979 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
7980 
7981 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
7982 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
7983 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
7984 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
7985 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
7986 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
7987 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
7988 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
7989 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
7990 
7991 	/* the next few have explicit upper bounds */
7992 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
7993 	md->limit = md->base - 1 +
7994 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
7995 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
7996 	md++;
7997 
7998 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
7999 	md->limit = md->base - 1 +
8000 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
8001 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
8002 	md++;
8003 
8004 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
8005 		if (chip_id(sc) <= CHELSIO_T5)
8006 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
8007 		else
8008 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
8009 		md->limit = 0;
8010 	} else {
8011 		md->base = 0;
8012 		md->idx = nitems(region);  /* hide it */
8013 	}
8014 	md++;
8015 
8016 #define ulp_region(reg) \
8017 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
8018 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
8019 
8020 	ulp_region(RX_ISCSI);
8021 	ulp_region(RX_TDDP);
8022 	ulp_region(TX_TPT);
8023 	ulp_region(RX_STAG);
8024 	ulp_region(RX_RQ);
8025 	ulp_region(RX_RQUDP);
8026 	ulp_region(RX_PBL);
8027 	ulp_region(TX_PBL);
8028 #undef ulp_region
8029 
8030 	md->base = 0;
8031 	md->idx = nitems(region);
8032 	if (!is_t4(sc)) {
8033 		uint32_t size = 0;
8034 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
8035 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
8036 
8037 		if (is_t5(sc)) {
8038 			if (sge_ctrl & F_VFIFO_ENABLE)
8039 				size = G_DBVFIFO_SIZE(fifo_size);
8040 		} else
8041 			size = G_T6_DBVFIFO_SIZE(fifo_size);
8042 
8043 		if (size) {
8044 			md->base = G_BASEADDR(t4_read_reg(sc,
8045 			    A_SGE_DBVFIFO_BADDR));
8046 			md->limit = md->base + (size << 2) - 1;
8047 		}
8048 	}
8049 	md++;
8050 
8051 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
8052 	md->limit = 0;
8053 	md++;
8054 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
8055 	md->limit = 0;
8056 	md++;
8057 
8058 	md->base = sc->vres.ocq.start;
8059 	if (sc->vres.ocq.size)
8060 		md->limit = md->base + sc->vres.ocq.size - 1;
8061 	else
8062 		md->idx = nitems(region);  /* hide it */
8063 	md++;
8064 
8065 	md->base = sc->vres.key.start;
8066 	if (sc->vres.key.size)
8067 		md->limit = md->base + sc->vres.key.size - 1;
8068 	else
8069 		md->idx = nitems(region);  /* hide it */
8070 	md++;
8071 
8072 	/* add any address-space holes, there can be up to 3 */
8073 	for (n = 0; n < i - 1; n++)
8074 		if (avail[n].limit < avail[n + 1].base)
8075 			(md++)->base = avail[n].limit;
8076 	if (avail[n].limit)
8077 		(md++)->base = avail[n].limit;
8078 
8079 	n = md - mem;
8080 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
8081 
8082 	for (lo = 0; lo < i; lo++)
8083 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
8084 				avail[lo].limit - 1);
8085 
8086 	sbuf_printf(sb, "\n");
8087 	for (i = 0; i < n; i++) {
8088 		if (mem[i].idx >= nitems(region))
8089 			continue;                        /* skip holes */
8090 		if (!mem[i].limit)
8091 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
8092 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
8093 				mem[i].limit);
8094 	}
8095 
8096 	sbuf_printf(sb, "\n");
8097 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
8098 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
8099 	mem_region_show(sb, "uP RAM:", lo, hi);
8100 
8101 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
8102 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
8103 	mem_region_show(sb, "uP Extmem2:", lo, hi);
8104 
8105 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
8106 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
8107 		   G_PMRXMAXPAGE(lo),
8108 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
8109 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
8110 
8111 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
8112 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
8113 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
8114 		   G_PMTXMAXPAGE(lo),
8115 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
8116 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
8117 	sbuf_printf(sb, "%u p-structs\n",
8118 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
8119 
8120 	for (i = 0; i < 4; i++) {
8121 		if (chip_id(sc) > CHELSIO_T5)
8122 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
8123 		else
8124 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
8125 		if (is_t5(sc)) {
8126 			used = G_T5_USED(lo);
8127 			alloc = G_T5_ALLOC(lo);
8128 		} else {
8129 			used = G_USED(lo);
8130 			alloc = G_ALLOC(lo);
8131 		}
8132 		/* For T6 these are MAC buffer groups */
8133 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
8134 		    i, used, alloc);
8135 	}
8136 	for (i = 0; i < sc->chip_params->nchan; i++) {
8137 		if (chip_id(sc) > CHELSIO_T5)
8138 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
8139 		else
8140 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
8141 		if (is_t5(sc)) {
8142 			used = G_T5_USED(lo);
8143 			alloc = G_T5_ALLOC(lo);
8144 		} else {
8145 			used = G_USED(lo);
8146 			alloc = G_ALLOC(lo);
8147 		}
8148 		/* For T6 these are MAC buffer groups */
8149 		sbuf_printf(sb,
8150 		    "\nLoopback %d using %u pages out of %u allocated",
8151 		    i, used, alloc);
8152 	}
8153 
8154 	rc = sbuf_finish(sb);
8155 	sbuf_delete(sb);
8156 
8157 	return (rc);
8158 }
8159 
8160 static inline void
8161 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
8162 {
8163 	*mask = x | y;
8164 	y = htobe64(y);
8165 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
8166 }
8167 
8168 static int
8169 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
8170 {
8171 	struct adapter *sc = arg1;
8172 	struct sbuf *sb;
8173 	int rc, i;
8174 
8175 	MPASS(chip_id(sc) <= CHELSIO_T5);
8176 
8177 	rc = sysctl_wire_old_buffer(req, 0);
8178 	if (rc != 0)
8179 		return (rc);
8180 
8181 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8182 	if (sb == NULL)
8183 		return (ENOMEM);
8184 
8185 	sbuf_printf(sb,
8186 	    "Idx  Ethernet address     Mask     Vld Ports PF"
8187 	    "  VF              Replication             P0 P1 P2 P3  ML");
8188 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
8189 		uint64_t tcamx, tcamy, mask;
8190 		uint32_t cls_lo, cls_hi;
8191 		uint8_t addr[ETHER_ADDR_LEN];
8192 
8193 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
8194 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
8195 		if (tcamx & tcamy)
8196 			continue;
8197 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
8198 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
8199 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
8200 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
8201 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
8202 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
8203 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
8204 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
8205 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
8206 
8207 		if (cls_lo & F_REPLICATE) {
8208 			struct fw_ldst_cmd ldst_cmd;
8209 
8210 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
8211 			ldst_cmd.op_to_addrspace =
8212 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
8213 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
8214 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
8215 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
8216 			ldst_cmd.u.mps.rplc.fid_idx =
8217 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
8218 				V_FW_LDST_CMD_IDX(i));
8219 
8220 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8221 			    "t4mps");
8222 			if (rc)
8223 				break;
8224 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
8225 			    sizeof(ldst_cmd), &ldst_cmd);
8226 			end_synchronized_op(sc, 0);
8227 
8228 			if (rc != 0) {
8229 				sbuf_printf(sb, "%36d", rc);
8230 				rc = 0;
8231 			} else {
8232 				sbuf_printf(sb, " %08x %08x %08x %08x",
8233 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
8234 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
8235 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
8236 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
8237 			}
8238 		} else
8239 			sbuf_printf(sb, "%36s", "");
8240 
8241 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
8242 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
8243 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
8244 	}
8245 
8246 	if (rc)
8247 		(void) sbuf_finish(sb);
8248 	else
8249 		rc = sbuf_finish(sb);
8250 	sbuf_delete(sb);
8251 
8252 	return (rc);
8253 }
8254 
8255 static int
8256 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
8257 {
8258 	struct adapter *sc = arg1;
8259 	struct sbuf *sb;
8260 	int rc, i;
8261 
8262 	MPASS(chip_id(sc) > CHELSIO_T5);
8263 
8264 	rc = sysctl_wire_old_buffer(req, 0);
8265 	if (rc != 0)
8266 		return (rc);
8267 
8268 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8269 	if (sb == NULL)
8270 		return (ENOMEM);
8271 
8272 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
8273 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
8274 	    "                           Replication"
8275 	    "                                    P0 P1 P2 P3  ML\n");
8276 
8277 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
8278 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
8279 		uint16_t ivlan;
8280 		uint64_t tcamx, tcamy, val, mask;
8281 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
8282 		uint8_t addr[ETHER_ADDR_LEN];
8283 
8284 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
8285 		if (i < 256)
8286 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
8287 		else
8288 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
8289 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8290 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8291 		tcamy = G_DMACH(val) << 32;
8292 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8293 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8294 		lookup_type = G_DATALKPTYPE(data2);
8295 		port_num = G_DATAPORTNUM(data2);
8296 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8297 			/* Inner header VNI */
8298 			vniy = ((data2 & F_DATAVIDH2) << 23) |
8299 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8300 			dip_hit = data2 & F_DATADIPHIT;
8301 			vlan_vld = 0;
8302 		} else {
8303 			vniy = 0;
8304 			dip_hit = 0;
8305 			vlan_vld = data2 & F_DATAVIDH2;
8306 			ivlan = G_VIDL(val);
8307 		}
8308 
8309 		ctl |= V_CTLXYBITSEL(1);
8310 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8311 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8312 		tcamx = G_DMACH(val) << 32;
8313 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8314 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8315 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8316 			/* Inner header VNI mask */
8317 			vnix = ((data2 & F_DATAVIDH2) << 23) |
8318 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8319 		} else
8320 			vnix = 0;
8321 
8322 		if (tcamx & tcamy)
8323 			continue;
8324 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
8325 
8326 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
8327 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
8328 
8329 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8330 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8331 			    "%012jx %06x %06x    -    -   %3c"
8332 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
8333 			    addr[1], addr[2], addr[3], addr[4], addr[5],
8334 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
8335 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8336 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8337 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8338 		} else {
8339 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8340 			    "%012jx    -       -   ", i, addr[0], addr[1],
8341 			    addr[2], addr[3], addr[4], addr[5],
8342 			    (uintmax_t)mask);
8343 
8344 			if (vlan_vld)
8345 				sbuf_printf(sb, "%4u   Y     ", ivlan);
8346 			else
8347 				sbuf_printf(sb, "  -    N     ");
8348 
8349 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
8350 			    lookup_type ? 'I' : 'O', port_num,
8351 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8352 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8353 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8354 		}
8355 
8356 
8357 		if (cls_lo & F_T6_REPLICATE) {
8358 			struct fw_ldst_cmd ldst_cmd;
8359 
8360 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
8361 			ldst_cmd.op_to_addrspace =
8362 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
8363 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
8364 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
8365 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
8366 			ldst_cmd.u.mps.rplc.fid_idx =
8367 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
8368 				V_FW_LDST_CMD_IDX(i));
8369 
8370 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8371 			    "t6mps");
8372 			if (rc)
8373 				break;
8374 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
8375 			    sizeof(ldst_cmd), &ldst_cmd);
8376 			end_synchronized_op(sc, 0);
8377 
8378 			if (rc != 0) {
8379 				sbuf_printf(sb, "%72d", rc);
8380 				rc = 0;
8381 			} else {
8382 				sbuf_printf(sb, " %08x %08x %08x %08x"
8383 				    " %08x %08x %08x %08x",
8384 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
8385 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
8386 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
8387 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
8388 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
8389 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
8390 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
8391 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
8392 			}
8393 		} else
8394 			sbuf_printf(sb, "%72s", "");
8395 
8396 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
8397 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
8398 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
8399 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
8400 	}
8401 
8402 	if (rc)
8403 		(void) sbuf_finish(sb);
8404 	else
8405 		rc = sbuf_finish(sb);
8406 	sbuf_delete(sb);
8407 
8408 	return (rc);
8409 }
8410 
8411 static int
8412 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
8413 {
8414 	struct adapter *sc = arg1;
8415 	struct sbuf *sb;
8416 	int rc;
8417 	uint16_t mtus[NMTUS];
8418 
8419 	rc = sysctl_wire_old_buffer(req, 0);
8420 	if (rc != 0)
8421 		return (rc);
8422 
8423 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8424 	if (sb == NULL)
8425 		return (ENOMEM);
8426 
8427 	t4_read_mtu_tbl(sc, mtus, NULL);
8428 
8429 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
8430 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
8431 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
8432 	    mtus[14], mtus[15]);
8433 
8434 	rc = sbuf_finish(sb);
8435 	sbuf_delete(sb);
8436 
8437 	return (rc);
8438 }
8439 
8440 static int
8441 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
8442 {
8443 	struct adapter *sc = arg1;
8444 	struct sbuf *sb;
8445 	int rc, i;
8446 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
8447 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
8448 	static const char *tx_stats[MAX_PM_NSTATS] = {
8449 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
8450 		"Tx FIFO wait", NULL, "Tx latency"
8451 	};
8452 	static const char *rx_stats[MAX_PM_NSTATS] = {
8453 		"Read:", "Write bypass:", "Write mem:", "Flush:",
8454 		"Rx FIFO wait", NULL, "Rx latency"
8455 	};
8456 
8457 	rc = sysctl_wire_old_buffer(req, 0);
8458 	if (rc != 0)
8459 		return (rc);
8460 
8461 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8462 	if (sb == NULL)
8463 		return (ENOMEM);
8464 
8465 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
8466 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
8467 
8468 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
8469 	for (i = 0; i < 4; i++) {
8470 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8471 		    tx_cyc[i]);
8472 	}
8473 
8474 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
8475 	for (i = 0; i < 4; i++) {
8476 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8477 		    rx_cyc[i]);
8478 	}
8479 
8480 	if (chip_id(sc) > CHELSIO_T5) {
8481 		sbuf_printf(sb,
8482 		    "\n              Total wait      Total occupancy");
8483 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8484 		    tx_cyc[i]);
8485 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8486 		    rx_cyc[i]);
8487 
8488 		i += 2;
8489 		MPASS(i < nitems(tx_stats));
8490 
8491 		sbuf_printf(sb,
8492 		    "\n                   Reads           Total wait");
8493 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8494 		    tx_cyc[i]);
8495 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8496 		    rx_cyc[i]);
8497 	}
8498 
8499 	rc = sbuf_finish(sb);
8500 	sbuf_delete(sb);
8501 
8502 	return (rc);
8503 }
8504 
8505 static int
8506 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
8507 {
8508 	struct adapter *sc = arg1;
8509 	struct sbuf *sb;
8510 	int rc;
8511 	struct tp_rdma_stats stats;
8512 
8513 	rc = sysctl_wire_old_buffer(req, 0);
8514 	if (rc != 0)
8515 		return (rc);
8516 
8517 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8518 	if (sb == NULL)
8519 		return (ENOMEM);
8520 
8521 	mtx_lock(&sc->reg_lock);
8522 	t4_tp_get_rdma_stats(sc, &stats, 0);
8523 	mtx_unlock(&sc->reg_lock);
8524 
8525 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
8526 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
8527 
8528 	rc = sbuf_finish(sb);
8529 	sbuf_delete(sb);
8530 
8531 	return (rc);
8532 }
8533 
8534 static int
8535 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
8536 {
8537 	struct adapter *sc = arg1;
8538 	struct sbuf *sb;
8539 	int rc;
8540 	struct tp_tcp_stats v4, v6;
8541 
8542 	rc = sysctl_wire_old_buffer(req, 0);
8543 	if (rc != 0)
8544 		return (rc);
8545 
8546 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8547 	if (sb == NULL)
8548 		return (ENOMEM);
8549 
8550 	mtx_lock(&sc->reg_lock);
8551 	t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
8552 	mtx_unlock(&sc->reg_lock);
8553 
8554 	sbuf_printf(sb,
8555 	    "                                IP                 IPv6\n");
8556 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
8557 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
8558 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
8559 	    v4.tcp_in_segs, v6.tcp_in_segs);
8560 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
8561 	    v4.tcp_out_segs, v6.tcp_out_segs);
8562 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
8563 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
8564 
8565 	rc = sbuf_finish(sb);
8566 	sbuf_delete(sb);
8567 
8568 	return (rc);
8569 }
8570 
8571 static int
8572 sysctl_tids(SYSCTL_HANDLER_ARGS)
8573 {
8574 	struct adapter *sc = arg1;
8575 	struct sbuf *sb;
8576 	int rc;
8577 	struct tid_info *t = &sc->tids;
8578 
8579 	rc = sysctl_wire_old_buffer(req, 0);
8580 	if (rc != 0)
8581 		return (rc);
8582 
8583 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8584 	if (sb == NULL)
8585 		return (ENOMEM);
8586 
8587 	if (t->natids) {
8588 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
8589 		    t->atids_in_use);
8590 	}
8591 
8592 	if (t->nhpftids) {
8593 		sbuf_printf(sb, "HPFTID range: %u-%u, in use: %u\n",
8594 		    t->hpftid_base, t->hpftid_end, t->hpftids_in_use);
8595 	}
8596 
8597 	if (t->ntids) {
8598 		sbuf_printf(sb, "TID range: ");
8599 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
8600 			uint32_t b, hb;
8601 
8602 			if (chip_id(sc) <= CHELSIO_T5) {
8603 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
8604 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
8605 			} else {
8606 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
8607 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
8608 			}
8609 
8610 			if (b)
8611 				sbuf_printf(sb, "%u-%u, ", t->tid_base, b - 1);
8612 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
8613 		} else
8614 			sbuf_printf(sb, "%u-%u", t->tid_base, t->ntids - 1);
8615 		sbuf_printf(sb, ", in use: %u\n",
8616 		    atomic_load_acq_int(&t->tids_in_use));
8617 	}
8618 
8619 	if (t->nstids) {
8620 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
8621 		    t->stid_base + t->nstids - 1, t->stids_in_use);
8622 	}
8623 
8624 	if (t->nftids) {
8625 		sbuf_printf(sb, "FTID range: %u-%u, in use: %u\n", t->ftid_base,
8626 		    t->ftid_end, t->ftids_in_use);
8627 	}
8628 
8629 	if (t->netids) {
8630 		sbuf_printf(sb, "ETID range: %u-%u, in use: %u\n", t->etid_base,
8631 		    t->etid_base + t->netids - 1, t->etids_in_use);
8632 	}
8633 
8634 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
8635 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
8636 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
8637 
8638 	rc = sbuf_finish(sb);
8639 	sbuf_delete(sb);
8640 
8641 	return (rc);
8642 }
8643 
8644 static int
8645 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
8646 {
8647 	struct adapter *sc = arg1;
8648 	struct sbuf *sb;
8649 	int rc;
8650 	struct tp_err_stats stats;
8651 
8652 	rc = sysctl_wire_old_buffer(req, 0);
8653 	if (rc != 0)
8654 		return (rc);
8655 
8656 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8657 	if (sb == NULL)
8658 		return (ENOMEM);
8659 
8660 	mtx_lock(&sc->reg_lock);
8661 	t4_tp_get_err_stats(sc, &stats, 0);
8662 	mtx_unlock(&sc->reg_lock);
8663 
8664 	if (sc->chip_params->nchan > 2) {
8665 		sbuf_printf(sb, "                 channel 0  channel 1"
8666 		    "  channel 2  channel 3\n");
8667 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
8668 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
8669 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
8670 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
8671 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
8672 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
8673 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
8674 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
8675 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
8676 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
8677 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
8678 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
8679 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
8680 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
8681 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
8682 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
8683 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
8684 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
8685 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
8686 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
8687 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
8688 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
8689 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
8690 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
8691 	} else {
8692 		sbuf_printf(sb, "                 channel 0  channel 1\n");
8693 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
8694 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
8695 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
8696 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
8697 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
8698 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
8699 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
8700 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
8701 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
8702 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
8703 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
8704 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
8705 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
8706 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
8707 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
8708 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
8709 	}
8710 
8711 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
8712 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
8713 
8714 	rc = sbuf_finish(sb);
8715 	sbuf_delete(sb);
8716 
8717 	return (rc);
8718 }
8719 
8720 static int
8721 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
8722 {
8723 	struct adapter *sc = arg1;
8724 	struct tp_params *tpp = &sc->params.tp;
8725 	u_int mask;
8726 	int rc;
8727 
8728 	mask = tpp->la_mask >> 16;
8729 	rc = sysctl_handle_int(oidp, &mask, 0, req);
8730 	if (rc != 0 || req->newptr == NULL)
8731 		return (rc);
8732 	if (mask > 0xffff)
8733 		return (EINVAL);
8734 	tpp->la_mask = mask << 16;
8735 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
8736 
8737 	return (0);
8738 }
8739 
8740 struct field_desc {
8741 	const char *name;
8742 	u_int start;
8743 	u_int width;
8744 };
8745 
8746 static void
8747 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
8748 {
8749 	char buf[32];
8750 	int line_size = 0;
8751 
8752 	while (f->name) {
8753 		uint64_t mask = (1ULL << f->width) - 1;
8754 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
8755 		    ((uintmax_t)v >> f->start) & mask);
8756 
8757 		if (line_size + len >= 79) {
8758 			line_size = 8;
8759 			sbuf_printf(sb, "\n        ");
8760 		}
8761 		sbuf_printf(sb, "%s ", buf);
8762 		line_size += len + 1;
8763 		f++;
8764 	}
8765 	sbuf_printf(sb, "\n");
8766 }
8767 
8768 static const struct field_desc tp_la0[] = {
8769 	{ "RcfOpCodeOut", 60, 4 },
8770 	{ "State", 56, 4 },
8771 	{ "WcfState", 52, 4 },
8772 	{ "RcfOpcSrcOut", 50, 2 },
8773 	{ "CRxError", 49, 1 },
8774 	{ "ERxError", 48, 1 },
8775 	{ "SanityFailed", 47, 1 },
8776 	{ "SpuriousMsg", 46, 1 },
8777 	{ "FlushInputMsg", 45, 1 },
8778 	{ "FlushInputCpl", 44, 1 },
8779 	{ "RssUpBit", 43, 1 },
8780 	{ "RssFilterHit", 42, 1 },
8781 	{ "Tid", 32, 10 },
8782 	{ "InitTcb", 31, 1 },
8783 	{ "LineNumber", 24, 7 },
8784 	{ "Emsg", 23, 1 },
8785 	{ "EdataOut", 22, 1 },
8786 	{ "Cmsg", 21, 1 },
8787 	{ "CdataOut", 20, 1 },
8788 	{ "EreadPdu", 19, 1 },
8789 	{ "CreadPdu", 18, 1 },
8790 	{ "TunnelPkt", 17, 1 },
8791 	{ "RcfPeerFin", 16, 1 },
8792 	{ "RcfReasonOut", 12, 4 },
8793 	{ "TxCchannel", 10, 2 },
8794 	{ "RcfTxChannel", 8, 2 },
8795 	{ "RxEchannel", 6, 2 },
8796 	{ "RcfRxChannel", 5, 1 },
8797 	{ "RcfDataOutSrdy", 4, 1 },
8798 	{ "RxDvld", 3, 1 },
8799 	{ "RxOoDvld", 2, 1 },
8800 	{ "RxCongestion", 1, 1 },
8801 	{ "TxCongestion", 0, 1 },
8802 	{ NULL }
8803 };
8804 
8805 static const struct field_desc tp_la1[] = {
8806 	{ "CplCmdIn", 56, 8 },
8807 	{ "CplCmdOut", 48, 8 },
8808 	{ "ESynOut", 47, 1 },
8809 	{ "EAckOut", 46, 1 },
8810 	{ "EFinOut", 45, 1 },
8811 	{ "ERstOut", 44, 1 },
8812 	{ "SynIn", 43, 1 },
8813 	{ "AckIn", 42, 1 },
8814 	{ "FinIn", 41, 1 },
8815 	{ "RstIn", 40, 1 },
8816 	{ "DataIn", 39, 1 },
8817 	{ "DataInVld", 38, 1 },
8818 	{ "PadIn", 37, 1 },
8819 	{ "RxBufEmpty", 36, 1 },
8820 	{ "RxDdp", 35, 1 },
8821 	{ "RxFbCongestion", 34, 1 },
8822 	{ "TxFbCongestion", 33, 1 },
8823 	{ "TxPktSumSrdy", 32, 1 },
8824 	{ "RcfUlpType", 28, 4 },
8825 	{ "Eread", 27, 1 },
8826 	{ "Ebypass", 26, 1 },
8827 	{ "Esave", 25, 1 },
8828 	{ "Static0", 24, 1 },
8829 	{ "Cread", 23, 1 },
8830 	{ "Cbypass", 22, 1 },
8831 	{ "Csave", 21, 1 },
8832 	{ "CPktOut", 20, 1 },
8833 	{ "RxPagePoolFull", 18, 2 },
8834 	{ "RxLpbkPkt", 17, 1 },
8835 	{ "TxLpbkPkt", 16, 1 },
8836 	{ "RxVfValid", 15, 1 },
8837 	{ "SynLearned", 14, 1 },
8838 	{ "SetDelEntry", 13, 1 },
8839 	{ "SetInvEntry", 12, 1 },
8840 	{ "CpcmdDvld", 11, 1 },
8841 	{ "CpcmdSave", 10, 1 },
8842 	{ "RxPstructsFull", 8, 2 },
8843 	{ "EpcmdDvld", 7, 1 },
8844 	{ "EpcmdFlush", 6, 1 },
8845 	{ "EpcmdTrimPrefix", 5, 1 },
8846 	{ "EpcmdTrimPostfix", 4, 1 },
8847 	{ "ERssIp4Pkt", 3, 1 },
8848 	{ "ERssIp6Pkt", 2, 1 },
8849 	{ "ERssTcpUdpPkt", 1, 1 },
8850 	{ "ERssFceFipPkt", 0, 1 },
8851 	{ NULL }
8852 };
8853 
8854 static const struct field_desc tp_la2[] = {
8855 	{ "CplCmdIn", 56, 8 },
8856 	{ "MpsVfVld", 55, 1 },
8857 	{ "MpsPf", 52, 3 },
8858 	{ "MpsVf", 44, 8 },
8859 	{ "SynIn", 43, 1 },
8860 	{ "AckIn", 42, 1 },
8861 	{ "FinIn", 41, 1 },
8862 	{ "RstIn", 40, 1 },
8863 	{ "DataIn", 39, 1 },
8864 	{ "DataInVld", 38, 1 },
8865 	{ "PadIn", 37, 1 },
8866 	{ "RxBufEmpty", 36, 1 },
8867 	{ "RxDdp", 35, 1 },
8868 	{ "RxFbCongestion", 34, 1 },
8869 	{ "TxFbCongestion", 33, 1 },
8870 	{ "TxPktSumSrdy", 32, 1 },
8871 	{ "RcfUlpType", 28, 4 },
8872 	{ "Eread", 27, 1 },
8873 	{ "Ebypass", 26, 1 },
8874 	{ "Esave", 25, 1 },
8875 	{ "Static0", 24, 1 },
8876 	{ "Cread", 23, 1 },
8877 	{ "Cbypass", 22, 1 },
8878 	{ "Csave", 21, 1 },
8879 	{ "CPktOut", 20, 1 },
8880 	{ "RxPagePoolFull", 18, 2 },
8881 	{ "RxLpbkPkt", 17, 1 },
8882 	{ "TxLpbkPkt", 16, 1 },
8883 	{ "RxVfValid", 15, 1 },
8884 	{ "SynLearned", 14, 1 },
8885 	{ "SetDelEntry", 13, 1 },
8886 	{ "SetInvEntry", 12, 1 },
8887 	{ "CpcmdDvld", 11, 1 },
8888 	{ "CpcmdSave", 10, 1 },
8889 	{ "RxPstructsFull", 8, 2 },
8890 	{ "EpcmdDvld", 7, 1 },
8891 	{ "EpcmdFlush", 6, 1 },
8892 	{ "EpcmdTrimPrefix", 5, 1 },
8893 	{ "EpcmdTrimPostfix", 4, 1 },
8894 	{ "ERssIp4Pkt", 3, 1 },
8895 	{ "ERssIp6Pkt", 2, 1 },
8896 	{ "ERssTcpUdpPkt", 1, 1 },
8897 	{ "ERssFceFipPkt", 0, 1 },
8898 	{ NULL }
8899 };
8900 
8901 static void
8902 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
8903 {
8904 
8905 	field_desc_show(sb, *p, tp_la0);
8906 }
8907 
8908 static void
8909 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
8910 {
8911 
8912 	if (idx)
8913 		sbuf_printf(sb, "\n");
8914 	field_desc_show(sb, p[0], tp_la0);
8915 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8916 		field_desc_show(sb, p[1], tp_la0);
8917 }
8918 
8919 static void
8920 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
8921 {
8922 
8923 	if (idx)
8924 		sbuf_printf(sb, "\n");
8925 	field_desc_show(sb, p[0], tp_la0);
8926 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8927 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
8928 }
8929 
8930 static int
8931 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
8932 {
8933 	struct adapter *sc = arg1;
8934 	struct sbuf *sb;
8935 	uint64_t *buf, *p;
8936 	int rc;
8937 	u_int i, inc;
8938 	void (*show_func)(struct sbuf *, uint64_t *, int);
8939 
8940 	rc = sysctl_wire_old_buffer(req, 0);
8941 	if (rc != 0)
8942 		return (rc);
8943 
8944 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8945 	if (sb == NULL)
8946 		return (ENOMEM);
8947 
8948 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
8949 
8950 	t4_tp_read_la(sc, buf, NULL);
8951 	p = buf;
8952 
8953 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
8954 	case 2:
8955 		inc = 2;
8956 		show_func = tp_la_show2;
8957 		break;
8958 	case 3:
8959 		inc = 2;
8960 		show_func = tp_la_show3;
8961 		break;
8962 	default:
8963 		inc = 1;
8964 		show_func = tp_la_show;
8965 	}
8966 
8967 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
8968 		(*show_func)(sb, p, i);
8969 
8970 	rc = sbuf_finish(sb);
8971 	sbuf_delete(sb);
8972 	free(buf, M_CXGBE);
8973 	return (rc);
8974 }
8975 
8976 static int
8977 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
8978 {
8979 	struct adapter *sc = arg1;
8980 	struct sbuf *sb;
8981 	int rc;
8982 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
8983 
8984 	rc = sysctl_wire_old_buffer(req, 0);
8985 	if (rc != 0)
8986 		return (rc);
8987 
8988 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8989 	if (sb == NULL)
8990 		return (ENOMEM);
8991 
8992 	t4_get_chan_txrate(sc, nrate, orate);
8993 
8994 	if (sc->chip_params->nchan > 2) {
8995 		sbuf_printf(sb, "              channel 0   channel 1"
8996 		    "   channel 2   channel 3\n");
8997 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
8998 		    nrate[0], nrate[1], nrate[2], nrate[3]);
8999 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
9000 		    orate[0], orate[1], orate[2], orate[3]);
9001 	} else {
9002 		sbuf_printf(sb, "              channel 0   channel 1\n");
9003 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
9004 		    nrate[0], nrate[1]);
9005 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
9006 		    orate[0], orate[1]);
9007 	}
9008 
9009 	rc = sbuf_finish(sb);
9010 	sbuf_delete(sb);
9011 
9012 	return (rc);
9013 }
9014 
9015 static int
9016 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
9017 {
9018 	struct adapter *sc = arg1;
9019 	struct sbuf *sb;
9020 	uint32_t *buf, *p;
9021 	int rc, i;
9022 
9023 	rc = sysctl_wire_old_buffer(req, 0);
9024 	if (rc != 0)
9025 		return (rc);
9026 
9027 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9028 	if (sb == NULL)
9029 		return (ENOMEM);
9030 
9031 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
9032 	    M_ZERO | M_WAITOK);
9033 
9034 	t4_ulprx_read_la(sc, buf);
9035 	p = buf;
9036 
9037 	sbuf_printf(sb, "      Pcmd        Type   Message"
9038 	    "                Data");
9039 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
9040 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
9041 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
9042 	}
9043 
9044 	rc = sbuf_finish(sb);
9045 	sbuf_delete(sb);
9046 	free(buf, M_CXGBE);
9047 	return (rc);
9048 }
9049 
9050 static int
9051 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
9052 {
9053 	struct adapter *sc = arg1;
9054 	struct sbuf *sb;
9055 	int rc, v;
9056 
9057 	MPASS(chip_id(sc) >= CHELSIO_T5);
9058 
9059 	rc = sysctl_wire_old_buffer(req, 0);
9060 	if (rc != 0)
9061 		return (rc);
9062 
9063 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9064 	if (sb == NULL)
9065 		return (ENOMEM);
9066 
9067 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
9068 	if (G_STATSOURCE_T5(v) == 7) {
9069 		int mode;
9070 
9071 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
9072 		if (mode == 0) {
9073 			sbuf_printf(sb, "total %d, incomplete %d",
9074 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
9075 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
9076 		} else if (mode == 1) {
9077 			sbuf_printf(sb, "total %d, data overflow %d",
9078 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
9079 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
9080 		} else {
9081 			sbuf_printf(sb, "unknown mode %d", mode);
9082 		}
9083 	}
9084 	rc = sbuf_finish(sb);
9085 	sbuf_delete(sb);
9086 
9087 	return (rc);
9088 }
9089 
9090 static int
9091 sysctl_cpus(SYSCTL_HANDLER_ARGS)
9092 {
9093 	struct adapter *sc = arg1;
9094 	enum cpu_sets op = arg2;
9095 	cpuset_t cpuset;
9096 	struct sbuf *sb;
9097 	int i, rc;
9098 
9099 	MPASS(op == LOCAL_CPUS || op == INTR_CPUS);
9100 
9101 	CPU_ZERO(&cpuset);
9102 	rc = bus_get_cpus(sc->dev, op, sizeof(cpuset), &cpuset);
9103 	if (rc != 0)
9104 		return (rc);
9105 
9106 	rc = sysctl_wire_old_buffer(req, 0);
9107 	if (rc != 0)
9108 		return (rc);
9109 
9110 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9111 	if (sb == NULL)
9112 		return (ENOMEM);
9113 
9114 	CPU_FOREACH(i)
9115 		sbuf_printf(sb, "%d ", i);
9116 	rc = sbuf_finish(sb);
9117 	sbuf_delete(sb);
9118 
9119 	return (rc);
9120 }
9121 
9122 #ifdef TCP_OFFLOAD
9123 static int
9124 sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS)
9125 {
9126 	struct adapter *sc = arg1;
9127 	int *old_ports, *new_ports;
9128 	int i, new_count, rc;
9129 
9130 	if (req->newptr == NULL && req->oldptr == NULL)
9131 		return (SYSCTL_OUT(req, NULL, imax(sc->tt.num_tls_rx_ports, 1) *
9132 		    sizeof(sc->tt.tls_rx_ports[0])));
9133 
9134 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tlsrx");
9135 	if (rc)
9136 		return (rc);
9137 
9138 	if (sc->tt.num_tls_rx_ports == 0) {
9139 		i = -1;
9140 		rc = SYSCTL_OUT(req, &i, sizeof(i));
9141 	} else
9142 		rc = SYSCTL_OUT(req, sc->tt.tls_rx_ports,
9143 		    sc->tt.num_tls_rx_ports * sizeof(sc->tt.tls_rx_ports[0]));
9144 	if (rc == 0 && req->newptr != NULL) {
9145 		new_count = req->newlen / sizeof(new_ports[0]);
9146 		new_ports = malloc(new_count * sizeof(new_ports[0]), M_CXGBE,
9147 		    M_WAITOK);
9148 		rc = SYSCTL_IN(req, new_ports, new_count *
9149 		    sizeof(new_ports[0]));
9150 		if (rc)
9151 			goto err;
9152 
9153 		/* Allow setting to a single '-1' to clear the list. */
9154 		if (new_count == 1 && new_ports[0] == -1) {
9155 			ADAPTER_LOCK(sc);
9156 			old_ports = sc->tt.tls_rx_ports;
9157 			sc->tt.tls_rx_ports = NULL;
9158 			sc->tt.num_tls_rx_ports = 0;
9159 			ADAPTER_UNLOCK(sc);
9160 			free(old_ports, M_CXGBE);
9161 		} else {
9162 			for (i = 0; i < new_count; i++) {
9163 				if (new_ports[i] < 1 ||
9164 				    new_ports[i] > IPPORT_MAX) {
9165 					rc = EINVAL;
9166 					goto err;
9167 				}
9168 			}
9169 
9170 			ADAPTER_LOCK(sc);
9171 			old_ports = sc->tt.tls_rx_ports;
9172 			sc->tt.tls_rx_ports = new_ports;
9173 			sc->tt.num_tls_rx_ports = new_count;
9174 			ADAPTER_UNLOCK(sc);
9175 			free(old_ports, M_CXGBE);
9176 			new_ports = NULL;
9177 		}
9178 	err:
9179 		free(new_ports, M_CXGBE);
9180 	}
9181 	end_synchronized_op(sc, 0);
9182 	return (rc);
9183 }
9184 
9185 static void
9186 unit_conv(char *buf, size_t len, u_int val, u_int factor)
9187 {
9188 	u_int rem = val % factor;
9189 
9190 	if (rem == 0)
9191 		snprintf(buf, len, "%u", val / factor);
9192 	else {
9193 		while (rem % 10 == 0)
9194 			rem /= 10;
9195 		snprintf(buf, len, "%u.%u", val / factor, rem);
9196 	}
9197 }
9198 
9199 static int
9200 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
9201 {
9202 	struct adapter *sc = arg1;
9203 	char buf[16];
9204 	u_int res, re;
9205 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9206 
9207 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
9208 	switch (arg2) {
9209 	case 0:
9210 		/* timer_tick */
9211 		re = G_TIMERRESOLUTION(res);
9212 		break;
9213 	case 1:
9214 		/* TCP timestamp tick */
9215 		re = G_TIMESTAMPRESOLUTION(res);
9216 		break;
9217 	case 2:
9218 		/* DACK tick */
9219 		re = G_DELAYEDACKRESOLUTION(res);
9220 		break;
9221 	default:
9222 		return (EDOOFUS);
9223 	}
9224 
9225 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
9226 
9227 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
9228 }
9229 
9230 static int
9231 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
9232 {
9233 	struct adapter *sc = arg1;
9234 	u_int res, dack_re, v;
9235 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9236 
9237 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
9238 	dack_re = G_DELAYEDACKRESOLUTION(res);
9239 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
9240 
9241 	return (sysctl_handle_int(oidp, &v, 0, req));
9242 }
9243 
9244 static int
9245 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
9246 {
9247 	struct adapter *sc = arg1;
9248 	int reg = arg2;
9249 	u_int tre;
9250 	u_long tp_tick_us, v;
9251 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9252 
9253 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
9254 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
9255 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
9256 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
9257 
9258 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
9259 	tp_tick_us = (cclk_ps << tre) / 1000000;
9260 
9261 	if (reg == A_TP_INIT_SRTT)
9262 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
9263 	else
9264 		v = tp_tick_us * t4_read_reg(sc, reg);
9265 
9266 	return (sysctl_handle_long(oidp, &v, 0, req));
9267 }
9268 
9269 /*
9270  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
9271  * passed to this function.
9272  */
9273 static int
9274 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
9275 {
9276 	struct adapter *sc = arg1;
9277 	int idx = arg2;
9278 	u_int v;
9279 
9280 	MPASS(idx >= 0 && idx <= 24);
9281 
9282 	v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
9283 
9284 	return (sysctl_handle_int(oidp, &v, 0, req));
9285 }
9286 
9287 static int
9288 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
9289 {
9290 	struct adapter *sc = arg1;
9291 	int idx = arg2;
9292 	u_int shift, v, r;
9293 
9294 	MPASS(idx >= 0 && idx < 16);
9295 
9296 	r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
9297 	shift = (idx & 3) << 3;
9298 	v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
9299 
9300 	return (sysctl_handle_int(oidp, &v, 0, req));
9301 }
9302 
9303 static int
9304 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
9305 {
9306 	struct vi_info *vi = arg1;
9307 	struct adapter *sc = vi->pi->adapter;
9308 	int idx, rc, i;
9309 	struct sge_ofld_rxq *ofld_rxq;
9310 	uint8_t v;
9311 
9312 	idx = vi->ofld_tmr_idx;
9313 
9314 	rc = sysctl_handle_int(oidp, &idx, 0, req);
9315 	if (rc != 0 || req->newptr == NULL)
9316 		return (rc);
9317 
9318 	if (idx < 0 || idx >= SGE_NTIMERS)
9319 		return (EINVAL);
9320 
9321 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9322 	    "t4otmr");
9323 	if (rc)
9324 		return (rc);
9325 
9326 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
9327 	for_each_ofld_rxq(vi, i, ofld_rxq) {
9328 #ifdef atomic_store_rel_8
9329 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
9330 #else
9331 		ofld_rxq->iq.intr_params = v;
9332 #endif
9333 	}
9334 	vi->ofld_tmr_idx = idx;
9335 
9336 	end_synchronized_op(sc, LOCK_HELD);
9337 	return (0);
9338 }
9339 
9340 static int
9341 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
9342 {
9343 	struct vi_info *vi = arg1;
9344 	struct adapter *sc = vi->pi->adapter;
9345 	int idx, rc;
9346 
9347 	idx = vi->ofld_pktc_idx;
9348 
9349 	rc = sysctl_handle_int(oidp, &idx, 0, req);
9350 	if (rc != 0 || req->newptr == NULL)
9351 		return (rc);
9352 
9353 	if (idx < -1 || idx >= SGE_NCOUNTERS)
9354 		return (EINVAL);
9355 
9356 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9357 	    "t4opktc");
9358 	if (rc)
9359 		return (rc);
9360 
9361 	if (vi->flags & VI_INIT_DONE)
9362 		rc = EBUSY; /* cannot be changed once the queues are created */
9363 	else
9364 		vi->ofld_pktc_idx = idx;
9365 
9366 	end_synchronized_op(sc, LOCK_HELD);
9367 	return (rc);
9368 }
9369 #endif
9370 
9371 static int
9372 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
9373 {
9374 	int rc;
9375 
9376 	if (cntxt->cid > M_CTXTQID)
9377 		return (EINVAL);
9378 
9379 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
9380 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
9381 		return (EINVAL);
9382 
9383 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
9384 	if (rc)
9385 		return (rc);
9386 
9387 	if (sc->flags & FW_OK) {
9388 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
9389 		    &cntxt->data[0]);
9390 		if (rc == 0)
9391 			goto done;
9392 	}
9393 
9394 	/*
9395 	 * Read via firmware failed or wasn't even attempted.  Read directly via
9396 	 * the backdoor.
9397 	 */
9398 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
9399 done:
9400 	end_synchronized_op(sc, 0);
9401 	return (rc);
9402 }
9403 
9404 static int
9405 load_fw(struct adapter *sc, struct t4_data *fw)
9406 {
9407 	int rc;
9408 	uint8_t *fw_data;
9409 
9410 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
9411 	if (rc)
9412 		return (rc);
9413 
9414 	/*
9415 	 * The firmware, with the sole exception of the memory parity error
9416 	 * handler, runs from memory and not flash.  It is almost always safe to
9417 	 * install a new firmware on a running system.  Just set bit 1 in
9418 	 * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
9419 	 */
9420 	if (sc->flags & FULL_INIT_DONE &&
9421 	    (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
9422 		rc = EBUSY;
9423 		goto done;
9424 	}
9425 
9426 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
9427 	if (fw_data == NULL) {
9428 		rc = ENOMEM;
9429 		goto done;
9430 	}
9431 
9432 	rc = copyin(fw->data, fw_data, fw->len);
9433 	if (rc == 0)
9434 		rc = -t4_load_fw(sc, fw_data, fw->len);
9435 
9436 	free(fw_data, M_CXGBE);
9437 done:
9438 	end_synchronized_op(sc, 0);
9439 	return (rc);
9440 }
9441 
9442 static int
9443 load_cfg(struct adapter *sc, struct t4_data *cfg)
9444 {
9445 	int rc;
9446 	uint8_t *cfg_data = NULL;
9447 
9448 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9449 	if (rc)
9450 		return (rc);
9451 
9452 	if (cfg->len == 0) {
9453 		/* clear */
9454 		rc = -t4_load_cfg(sc, NULL, 0);
9455 		goto done;
9456 	}
9457 
9458 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
9459 	if (cfg_data == NULL) {
9460 		rc = ENOMEM;
9461 		goto done;
9462 	}
9463 
9464 	rc = copyin(cfg->data, cfg_data, cfg->len);
9465 	if (rc == 0)
9466 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
9467 
9468 	free(cfg_data, M_CXGBE);
9469 done:
9470 	end_synchronized_op(sc, 0);
9471 	return (rc);
9472 }
9473 
9474 static int
9475 load_boot(struct adapter *sc, struct t4_bootrom *br)
9476 {
9477 	int rc;
9478 	uint8_t *br_data = NULL;
9479 	u_int offset;
9480 
9481 	if (br->len > 1024 * 1024)
9482 		return (EFBIG);
9483 
9484 	if (br->pf_offset == 0) {
9485 		/* pfidx */
9486 		if (br->pfidx_addr > 7)
9487 			return (EINVAL);
9488 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
9489 		    A_PCIE_PF_EXPROM_OFST)));
9490 	} else if (br->pf_offset == 1) {
9491 		/* offset */
9492 		offset = G_OFFSET(br->pfidx_addr);
9493 	} else {
9494 		return (EINVAL);
9495 	}
9496 
9497 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
9498 	if (rc)
9499 		return (rc);
9500 
9501 	if (br->len == 0) {
9502 		/* clear */
9503 		rc = -t4_load_boot(sc, NULL, offset, 0);
9504 		goto done;
9505 	}
9506 
9507 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
9508 	if (br_data == NULL) {
9509 		rc = ENOMEM;
9510 		goto done;
9511 	}
9512 
9513 	rc = copyin(br->data, br_data, br->len);
9514 	if (rc == 0)
9515 		rc = -t4_load_boot(sc, br_data, offset, br->len);
9516 
9517 	free(br_data, M_CXGBE);
9518 done:
9519 	end_synchronized_op(sc, 0);
9520 	return (rc);
9521 }
9522 
9523 static int
9524 load_bootcfg(struct adapter *sc, struct t4_data *bc)
9525 {
9526 	int rc;
9527 	uint8_t *bc_data = NULL;
9528 
9529 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9530 	if (rc)
9531 		return (rc);
9532 
9533 	if (bc->len == 0) {
9534 		/* clear */
9535 		rc = -t4_load_bootcfg(sc, NULL, 0);
9536 		goto done;
9537 	}
9538 
9539 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
9540 	if (bc_data == NULL) {
9541 		rc = ENOMEM;
9542 		goto done;
9543 	}
9544 
9545 	rc = copyin(bc->data, bc_data, bc->len);
9546 	if (rc == 0)
9547 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
9548 
9549 	free(bc_data, M_CXGBE);
9550 done:
9551 	end_synchronized_op(sc, 0);
9552 	return (rc);
9553 }
9554 
9555 static int
9556 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
9557 {
9558 	int rc;
9559 	struct cudbg_init *cudbg;
9560 	void *handle, *buf;
9561 
9562 	/* buf is large, don't block if no memory is available */
9563 	buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
9564 	if (buf == NULL)
9565 		return (ENOMEM);
9566 
9567 	handle = cudbg_alloc_handle();
9568 	if (handle == NULL) {
9569 		rc = ENOMEM;
9570 		goto done;
9571 	}
9572 
9573 	cudbg = cudbg_get_init(handle);
9574 	cudbg->adap = sc;
9575 	cudbg->print = (cudbg_print_cb)printf;
9576 
9577 #ifndef notyet
9578 	device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
9579 	    __func__, dump->wr_flash, dump->len, dump->data);
9580 #endif
9581 
9582 	if (dump->wr_flash)
9583 		cudbg->use_flash = 1;
9584 	MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
9585 	memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
9586 
9587 	rc = cudbg_collect(handle, buf, &dump->len);
9588 	if (rc != 0)
9589 		goto done;
9590 
9591 	rc = copyout(buf, dump->data, dump->len);
9592 done:
9593 	cudbg_free_handle(handle);
9594 	free(buf, M_CXGBE);
9595 	return (rc);
9596 }
9597 
9598 static void
9599 free_offload_policy(struct t4_offload_policy *op)
9600 {
9601 	struct offload_rule *r;
9602 	int i;
9603 
9604 	if (op == NULL)
9605 		return;
9606 
9607 	r = &op->rule[0];
9608 	for (i = 0; i < op->nrules; i++, r++) {
9609 		free(r->bpf_prog.bf_insns, M_CXGBE);
9610 	}
9611 	free(op->rule, M_CXGBE);
9612 	free(op, M_CXGBE);
9613 }
9614 
9615 static int
9616 set_offload_policy(struct adapter *sc, struct t4_offload_policy *uop)
9617 {
9618 	int i, rc, len;
9619 	struct t4_offload_policy *op, *old;
9620 	struct bpf_program *bf;
9621 	const struct offload_settings *s;
9622 	struct offload_rule *r;
9623 	void *u;
9624 
9625 	if (!is_offload(sc))
9626 		return (ENODEV);
9627 
9628 	if (uop->nrules == 0) {
9629 		/* Delete installed policies. */
9630 		op = NULL;
9631 		goto set_policy;
9632 	} if (uop->nrules > 256) { /* arbitrary */
9633 		return (E2BIG);
9634 	}
9635 
9636 	/* Copy userspace offload policy to kernel */
9637 	op = malloc(sizeof(*op), M_CXGBE, M_ZERO | M_WAITOK);
9638 	op->nrules = uop->nrules;
9639 	len = op->nrules * sizeof(struct offload_rule);
9640 	op->rule = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9641 	rc = copyin(uop->rule, op->rule, len);
9642 	if (rc) {
9643 		free(op->rule, M_CXGBE);
9644 		free(op, M_CXGBE);
9645 		return (rc);
9646 	}
9647 
9648 	r = &op->rule[0];
9649 	for (i = 0; i < op->nrules; i++, r++) {
9650 
9651 		/* Validate open_type */
9652 		if (r->open_type != OPEN_TYPE_LISTEN &&
9653 		    r->open_type != OPEN_TYPE_ACTIVE &&
9654 		    r->open_type != OPEN_TYPE_PASSIVE &&
9655 		    r->open_type != OPEN_TYPE_DONTCARE) {
9656 error:
9657 			/*
9658 			 * Rules 0 to i have malloc'd filters that need to be
9659 			 * freed.  Rules i+1 to nrules have userspace pointers
9660 			 * and should be left alone.
9661 			 */
9662 			op->nrules = i;
9663 			free_offload_policy(op);
9664 			return (rc);
9665 		}
9666 
9667 		/* Validate settings */
9668 		s = &r->settings;
9669 		if ((s->offload != 0 && s->offload != 1) ||
9670 		    s->cong_algo < -1 || s->cong_algo > CONG_ALG_HIGHSPEED ||
9671 		    s->sched_class < -1 ||
9672 		    s->sched_class >= sc->chip_params->nsched_cls) {
9673 			rc = EINVAL;
9674 			goto error;
9675 		}
9676 
9677 		bf = &r->bpf_prog;
9678 		u = bf->bf_insns;	/* userspace ptr */
9679 		bf->bf_insns = NULL;
9680 		if (bf->bf_len == 0) {
9681 			/* legal, matches everything */
9682 			continue;
9683 		}
9684 		len = bf->bf_len * sizeof(*bf->bf_insns);
9685 		bf->bf_insns = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9686 		rc = copyin(u, bf->bf_insns, len);
9687 		if (rc != 0)
9688 			goto error;
9689 
9690 		if (!bpf_validate(bf->bf_insns, bf->bf_len)) {
9691 			rc = EINVAL;
9692 			goto error;
9693 		}
9694 	}
9695 set_policy:
9696 	rw_wlock(&sc->policy_lock);
9697 	old = sc->policy;
9698 	sc->policy = op;
9699 	rw_wunlock(&sc->policy_lock);
9700 	free_offload_policy(old);
9701 
9702 	return (0);
9703 }
9704 
9705 #define MAX_READ_BUF_SIZE (128 * 1024)
9706 static int
9707 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
9708 {
9709 	uint32_t addr, remaining, n;
9710 	uint32_t *buf;
9711 	int rc;
9712 	uint8_t *dst;
9713 
9714 	rc = validate_mem_range(sc, mr->addr, mr->len);
9715 	if (rc != 0)
9716 		return (rc);
9717 
9718 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
9719 	addr = mr->addr;
9720 	remaining = mr->len;
9721 	dst = (void *)mr->data;
9722 
9723 	while (remaining) {
9724 		n = min(remaining, MAX_READ_BUF_SIZE);
9725 		read_via_memwin(sc, 2, addr, buf, n);
9726 
9727 		rc = copyout(buf, dst, n);
9728 		if (rc != 0)
9729 			break;
9730 
9731 		dst += n;
9732 		remaining -= n;
9733 		addr += n;
9734 	}
9735 
9736 	free(buf, M_CXGBE);
9737 	return (rc);
9738 }
9739 #undef MAX_READ_BUF_SIZE
9740 
9741 static int
9742 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
9743 {
9744 	int rc;
9745 
9746 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
9747 		return (EINVAL);
9748 
9749 	if (i2cd->len > sizeof(i2cd->data))
9750 		return (EFBIG);
9751 
9752 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9753 	if (rc)
9754 		return (rc);
9755 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9756 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
9757 	end_synchronized_op(sc, 0);
9758 
9759 	return (rc);
9760 }
9761 
9762 int
9763 t4_os_find_pci_capability(struct adapter *sc, int cap)
9764 {
9765 	int i;
9766 
9767 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9768 }
9769 
9770 int
9771 t4_os_pci_save_state(struct adapter *sc)
9772 {
9773 	device_t dev;
9774 	struct pci_devinfo *dinfo;
9775 
9776 	dev = sc->dev;
9777 	dinfo = device_get_ivars(dev);
9778 
9779 	pci_cfg_save(dev, dinfo, 0);
9780 	return (0);
9781 }
9782 
9783 int
9784 t4_os_pci_restore_state(struct adapter *sc)
9785 {
9786 	device_t dev;
9787 	struct pci_devinfo *dinfo;
9788 
9789 	dev = sc->dev;
9790 	dinfo = device_get_ivars(dev);
9791 
9792 	pci_cfg_restore(dev, dinfo);
9793 	return (0);
9794 }
9795 
9796 void
9797 t4_os_portmod_changed(struct port_info *pi)
9798 {
9799 	struct adapter *sc = pi->adapter;
9800 	struct vi_info *vi;
9801 	struct ifnet *ifp;
9802 	static const char *mod_str[] = {
9803 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9804 	};
9805 
9806 	KASSERT((pi->flags & FIXED_IFMEDIA) == 0,
9807 	    ("%s: port_type %u", __func__, pi->port_type));
9808 
9809 	vi = &pi->vi[0];
9810 	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9811 		PORT_LOCK(pi);
9812 		build_medialist(pi);
9813 		if (pi->mod_type != FW_PORT_MOD_TYPE_NONE) {
9814 			fixup_link_config(pi);
9815 			apply_link_config(pi);
9816 		}
9817 		PORT_UNLOCK(pi);
9818 		end_synchronized_op(sc, LOCK_HELD);
9819 	}
9820 
9821 	ifp = vi->ifp;
9822 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9823 		if_printf(ifp, "transceiver unplugged.\n");
9824 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9825 		if_printf(ifp, "unknown transceiver inserted.\n");
9826 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9827 		if_printf(ifp, "unsupported transceiver inserted.\n");
9828 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9829 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9830 		    port_top_speed(pi), mod_str[pi->mod_type]);
9831 	} else {
9832 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9833 		    pi->mod_type);
9834 	}
9835 }
9836 
9837 void
9838 t4_os_link_changed(struct port_info *pi)
9839 {
9840 	struct vi_info *vi;
9841 	struct ifnet *ifp;
9842 	struct link_config *lc;
9843 	int v;
9844 
9845 	PORT_LOCK_ASSERT_OWNED(pi);
9846 
9847 	for_each_vi(pi, v, vi) {
9848 		ifp = vi->ifp;
9849 		if (ifp == NULL)
9850 			continue;
9851 
9852 		lc = &pi->link_cfg;
9853 		if (lc->link_ok) {
9854 			ifp->if_baudrate = IF_Mbps(lc->speed);
9855 			if_link_state_change(ifp, LINK_STATE_UP);
9856 		} else {
9857 			if_link_state_change(ifp, LINK_STATE_DOWN);
9858 		}
9859 	}
9860 }
9861 
9862 void
9863 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9864 {
9865 	struct adapter *sc;
9866 
9867 	sx_slock(&t4_list_lock);
9868 	SLIST_FOREACH(sc, &t4_list, link) {
9869 		/*
9870 		 * func should not make any assumptions about what state sc is
9871 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9872 		 */
9873 		func(sc, arg);
9874 	}
9875 	sx_sunlock(&t4_list_lock);
9876 }
9877 
9878 static int
9879 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9880     struct thread *td)
9881 {
9882 	int rc;
9883 	struct adapter *sc = dev->si_drv1;
9884 
9885 	rc = priv_check(td, PRIV_DRIVER);
9886 	if (rc != 0)
9887 		return (rc);
9888 
9889 	switch (cmd) {
9890 	case CHELSIO_T4_GETREG: {
9891 		struct t4_reg *edata = (struct t4_reg *)data;
9892 
9893 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9894 			return (EFAULT);
9895 
9896 		if (edata->size == 4)
9897 			edata->val = t4_read_reg(sc, edata->addr);
9898 		else if (edata->size == 8)
9899 			edata->val = t4_read_reg64(sc, edata->addr);
9900 		else
9901 			return (EINVAL);
9902 
9903 		break;
9904 	}
9905 	case CHELSIO_T4_SETREG: {
9906 		struct t4_reg *edata = (struct t4_reg *)data;
9907 
9908 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9909 			return (EFAULT);
9910 
9911 		if (edata->size == 4) {
9912 			if (edata->val & 0xffffffff00000000)
9913 				return (EINVAL);
9914 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9915 		} else if (edata->size == 8)
9916 			t4_write_reg64(sc, edata->addr, edata->val);
9917 		else
9918 			return (EINVAL);
9919 		break;
9920 	}
9921 	case CHELSIO_T4_REGDUMP: {
9922 		struct t4_regdump *regs = (struct t4_regdump *)data;
9923 		int reglen = t4_get_regs_len(sc);
9924 		uint8_t *buf;
9925 
9926 		if (regs->len < reglen) {
9927 			regs->len = reglen; /* hint to the caller */
9928 			return (ENOBUFS);
9929 		}
9930 
9931 		regs->len = reglen;
9932 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9933 		get_regs(sc, regs, buf);
9934 		rc = copyout(buf, regs->data, reglen);
9935 		free(buf, M_CXGBE);
9936 		break;
9937 	}
9938 	case CHELSIO_T4_GET_FILTER_MODE:
9939 		rc = get_filter_mode(sc, (uint32_t *)data);
9940 		break;
9941 	case CHELSIO_T4_SET_FILTER_MODE:
9942 		rc = set_filter_mode(sc, *(uint32_t *)data);
9943 		break;
9944 	case CHELSIO_T4_GET_FILTER:
9945 		rc = get_filter(sc, (struct t4_filter *)data);
9946 		break;
9947 	case CHELSIO_T4_SET_FILTER:
9948 		rc = set_filter(sc, (struct t4_filter *)data);
9949 		break;
9950 	case CHELSIO_T4_DEL_FILTER:
9951 		rc = del_filter(sc, (struct t4_filter *)data);
9952 		break;
9953 	case CHELSIO_T4_GET_SGE_CONTEXT:
9954 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9955 		break;
9956 	case CHELSIO_T4_LOAD_FW:
9957 		rc = load_fw(sc, (struct t4_data *)data);
9958 		break;
9959 	case CHELSIO_T4_GET_MEM:
9960 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9961 		break;
9962 	case CHELSIO_T4_GET_I2C:
9963 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9964 		break;
9965 	case CHELSIO_T4_CLEAR_STATS: {
9966 		int i, v, bg_map;
9967 		u_int port_id = *(uint32_t *)data;
9968 		struct port_info *pi;
9969 		struct vi_info *vi;
9970 
9971 		if (port_id >= sc->params.nports)
9972 			return (EINVAL);
9973 		pi = sc->port[port_id];
9974 		if (pi == NULL)
9975 			return (EIO);
9976 
9977 		/* MAC stats */
9978 		t4_clr_port_stats(sc, pi->tx_chan);
9979 		pi->tx_parse_error = 0;
9980 		pi->tnl_cong_drops = 0;
9981 		mtx_lock(&sc->reg_lock);
9982 		for_each_vi(pi, v, vi) {
9983 			if (vi->flags & VI_INIT_DONE)
9984 				t4_clr_vi_stats(sc, vi->viid);
9985 		}
9986 		bg_map = pi->mps_bg_map;
9987 		v = 0;	/* reuse */
9988 		while (bg_map) {
9989 			i = ffs(bg_map) - 1;
9990 			t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
9991 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
9992 			bg_map &= ~(1 << i);
9993 		}
9994 		mtx_unlock(&sc->reg_lock);
9995 
9996 		/*
9997 		 * Since this command accepts a port, clear stats for
9998 		 * all VIs on this port.
9999 		 */
10000 		for_each_vi(pi, v, vi) {
10001 			if (vi->flags & VI_INIT_DONE) {
10002 				struct sge_rxq *rxq;
10003 				struct sge_txq *txq;
10004 				struct sge_wrq *wrq;
10005 
10006 				for_each_rxq(vi, i, rxq) {
10007 #if defined(INET) || defined(INET6)
10008 					rxq->lro.lro_queued = 0;
10009 					rxq->lro.lro_flushed = 0;
10010 #endif
10011 					rxq->rxcsum = 0;
10012 					rxq->vlan_extraction = 0;
10013 				}
10014 
10015 				for_each_txq(vi, i, txq) {
10016 					txq->txcsum = 0;
10017 					txq->tso_wrs = 0;
10018 					txq->vlan_insertion = 0;
10019 					txq->imm_wrs = 0;
10020 					txq->sgl_wrs = 0;
10021 					txq->txpkt_wrs = 0;
10022 					txq->txpkts0_wrs = 0;
10023 					txq->txpkts1_wrs = 0;
10024 					txq->txpkts0_pkts = 0;
10025 					txq->txpkts1_pkts = 0;
10026 					txq->raw_wrs = 0;
10027 					mp_ring_reset_stats(txq->r);
10028 				}
10029 
10030 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
10031 				/* nothing to clear for each ofld_rxq */
10032 
10033 				for_each_ofld_txq(vi, i, wrq) {
10034 					wrq->tx_wrs_direct = 0;
10035 					wrq->tx_wrs_copied = 0;
10036 				}
10037 #endif
10038 
10039 				if (IS_MAIN_VI(vi)) {
10040 					wrq = &sc->sge.ctrlq[pi->port_id];
10041 					wrq->tx_wrs_direct = 0;
10042 					wrq->tx_wrs_copied = 0;
10043 				}
10044 			}
10045 		}
10046 		break;
10047 	}
10048 	case CHELSIO_T4_SCHED_CLASS:
10049 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
10050 		break;
10051 	case CHELSIO_T4_SCHED_QUEUE:
10052 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
10053 		break;
10054 	case CHELSIO_T4_GET_TRACER:
10055 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
10056 		break;
10057 	case CHELSIO_T4_SET_TRACER:
10058 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
10059 		break;
10060 	case CHELSIO_T4_LOAD_CFG:
10061 		rc = load_cfg(sc, (struct t4_data *)data);
10062 		break;
10063 	case CHELSIO_T4_LOAD_BOOT:
10064 		rc = load_boot(sc, (struct t4_bootrom *)data);
10065 		break;
10066 	case CHELSIO_T4_LOAD_BOOTCFG:
10067 		rc = load_bootcfg(sc, (struct t4_data *)data);
10068 		break;
10069 	case CHELSIO_T4_CUDBG_DUMP:
10070 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
10071 		break;
10072 	case CHELSIO_T4_SET_OFLD_POLICY:
10073 		rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
10074 		break;
10075 	default:
10076 		rc = ENOTTY;
10077 	}
10078 
10079 	return (rc);
10080 }
10081 
10082 #ifdef TCP_OFFLOAD
10083 static int
10084 toe_capability(struct vi_info *vi, int enable)
10085 {
10086 	int rc;
10087 	struct port_info *pi = vi->pi;
10088 	struct adapter *sc = pi->adapter;
10089 
10090 	ASSERT_SYNCHRONIZED_OP(sc);
10091 
10092 	if (!is_offload(sc))
10093 		return (ENODEV);
10094 
10095 	if (enable) {
10096 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
10097 			/* TOE is already enabled. */
10098 			return (0);
10099 		}
10100 
10101 		/*
10102 		 * We need the port's queues around so that we're able to send
10103 		 * and receive CPLs to/from the TOE even if the ifnet for this
10104 		 * port has never been UP'd administratively.
10105 		 */
10106 		if (!(vi->flags & VI_INIT_DONE)) {
10107 			rc = vi_full_init(vi);
10108 			if (rc)
10109 				return (rc);
10110 		}
10111 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
10112 			rc = vi_full_init(&pi->vi[0]);
10113 			if (rc)
10114 				return (rc);
10115 		}
10116 
10117 		if (isset(&sc->offload_map, pi->port_id)) {
10118 			/* TOE is enabled on another VI of this port. */
10119 			pi->uld_vis++;
10120 			return (0);
10121 		}
10122 
10123 		if (!uld_active(sc, ULD_TOM)) {
10124 			rc = t4_activate_uld(sc, ULD_TOM);
10125 			if (rc == EAGAIN) {
10126 				log(LOG_WARNING,
10127 				    "You must kldload t4_tom.ko before trying "
10128 				    "to enable TOE on a cxgbe interface.\n");
10129 			}
10130 			if (rc != 0)
10131 				return (rc);
10132 			KASSERT(sc->tom_softc != NULL,
10133 			    ("%s: TOM activated but softc NULL", __func__));
10134 			KASSERT(uld_active(sc, ULD_TOM),
10135 			    ("%s: TOM activated but flag not set", __func__));
10136 		}
10137 
10138 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
10139 		if (!uld_active(sc, ULD_IWARP))
10140 			(void) t4_activate_uld(sc, ULD_IWARP);
10141 		if (!uld_active(sc, ULD_ISCSI))
10142 			(void) t4_activate_uld(sc, ULD_ISCSI);
10143 
10144 		pi->uld_vis++;
10145 		setbit(&sc->offload_map, pi->port_id);
10146 	} else {
10147 		pi->uld_vis--;
10148 
10149 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
10150 			return (0);
10151 
10152 		KASSERT(uld_active(sc, ULD_TOM),
10153 		    ("%s: TOM never initialized?", __func__));
10154 		clrbit(&sc->offload_map, pi->port_id);
10155 	}
10156 
10157 	return (0);
10158 }
10159 
10160 /*
10161  * Add an upper layer driver to the global list.
10162  */
10163 int
10164 t4_register_uld(struct uld_info *ui)
10165 {
10166 	int rc = 0;
10167 	struct uld_info *u;
10168 
10169 	sx_xlock(&t4_uld_list_lock);
10170 	SLIST_FOREACH(u, &t4_uld_list, link) {
10171 	    if (u->uld_id == ui->uld_id) {
10172 		    rc = EEXIST;
10173 		    goto done;
10174 	    }
10175 	}
10176 
10177 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
10178 	ui->refcount = 0;
10179 done:
10180 	sx_xunlock(&t4_uld_list_lock);
10181 	return (rc);
10182 }
10183 
10184 int
10185 t4_unregister_uld(struct uld_info *ui)
10186 {
10187 	int rc = EINVAL;
10188 	struct uld_info *u;
10189 
10190 	sx_xlock(&t4_uld_list_lock);
10191 
10192 	SLIST_FOREACH(u, &t4_uld_list, link) {
10193 	    if (u == ui) {
10194 		    if (ui->refcount > 0) {
10195 			    rc = EBUSY;
10196 			    goto done;
10197 		    }
10198 
10199 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
10200 		    rc = 0;
10201 		    goto done;
10202 	    }
10203 	}
10204 done:
10205 	sx_xunlock(&t4_uld_list_lock);
10206 	return (rc);
10207 }
10208 
10209 int
10210 t4_activate_uld(struct adapter *sc, int id)
10211 {
10212 	int rc;
10213 	struct uld_info *ui;
10214 
10215 	ASSERT_SYNCHRONIZED_OP(sc);
10216 
10217 	if (id < 0 || id > ULD_MAX)
10218 		return (EINVAL);
10219 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
10220 
10221 	sx_slock(&t4_uld_list_lock);
10222 
10223 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10224 		if (ui->uld_id == id) {
10225 			if (!(sc->flags & FULL_INIT_DONE)) {
10226 				rc = adapter_full_init(sc);
10227 				if (rc != 0)
10228 					break;
10229 			}
10230 
10231 			rc = ui->activate(sc);
10232 			if (rc == 0) {
10233 				setbit(&sc->active_ulds, id);
10234 				ui->refcount++;
10235 			}
10236 			break;
10237 		}
10238 	}
10239 
10240 	sx_sunlock(&t4_uld_list_lock);
10241 
10242 	return (rc);
10243 }
10244 
10245 int
10246 t4_deactivate_uld(struct adapter *sc, int id)
10247 {
10248 	int rc;
10249 	struct uld_info *ui;
10250 
10251 	ASSERT_SYNCHRONIZED_OP(sc);
10252 
10253 	if (id < 0 || id > ULD_MAX)
10254 		return (EINVAL);
10255 	rc = ENXIO;
10256 
10257 	sx_slock(&t4_uld_list_lock);
10258 
10259 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10260 		if (ui->uld_id == id) {
10261 			rc = ui->deactivate(sc);
10262 			if (rc == 0) {
10263 				clrbit(&sc->active_ulds, id);
10264 				ui->refcount--;
10265 			}
10266 			break;
10267 		}
10268 	}
10269 
10270 	sx_sunlock(&t4_uld_list_lock);
10271 
10272 	return (rc);
10273 }
10274 
10275 int
10276 uld_active(struct adapter *sc, int uld_id)
10277 {
10278 
10279 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
10280 
10281 	return (isset(&sc->active_ulds, uld_id));
10282 }
10283 #endif
10284 
10285 /*
10286  * t  = ptr to tunable.
10287  * nc = number of CPUs.
10288  * c  = compiled in default for that tunable.
10289  */
10290 static void
10291 calculate_nqueues(int *t, int nc, const int c)
10292 {
10293 	int nq;
10294 
10295 	if (*t > 0)
10296 		return;
10297 	nq = *t < 0 ? -*t : c;
10298 	*t = min(nc, nq);
10299 }
10300 
10301 /*
10302  * Come up with reasonable defaults for some of the tunables, provided they're
10303  * not set by the user (in which case we'll use the values as is).
10304  */
10305 static void
10306 tweak_tunables(void)
10307 {
10308 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
10309 
10310 	if (t4_ntxq < 1) {
10311 #ifdef RSS
10312 		t4_ntxq = rss_getnumbuckets();
10313 #else
10314 		calculate_nqueues(&t4_ntxq, nc, NTXQ);
10315 #endif
10316 	}
10317 
10318 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
10319 
10320 	if (t4_nrxq < 1) {
10321 #ifdef RSS
10322 		t4_nrxq = rss_getnumbuckets();
10323 #else
10324 		calculate_nqueues(&t4_nrxq, nc, NRXQ);
10325 #endif
10326 	}
10327 
10328 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
10329 
10330 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
10331 	calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
10332 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
10333 #endif
10334 #ifdef TCP_OFFLOAD
10335 	calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
10336 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
10337 
10338 	if (t4_toecaps_allowed == -1)
10339 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
10340 
10341 	if (t4_rdmacaps_allowed == -1) {
10342 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
10343 		    FW_CAPS_CONFIG_RDMA_RDMAC;
10344 	}
10345 
10346 	if (t4_iscsicaps_allowed == -1) {
10347 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
10348 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
10349 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
10350 	}
10351 
10352 	if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
10353 		t4_tmr_idx_ofld = TMR_IDX_OFLD;
10354 
10355 	if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
10356 		t4_pktc_idx_ofld = PKTC_IDX_OFLD;
10357 #else
10358 	if (t4_toecaps_allowed == -1)
10359 		t4_toecaps_allowed = 0;
10360 
10361 	if (t4_rdmacaps_allowed == -1)
10362 		t4_rdmacaps_allowed = 0;
10363 
10364 	if (t4_iscsicaps_allowed == -1)
10365 		t4_iscsicaps_allowed = 0;
10366 #endif
10367 
10368 #ifdef DEV_NETMAP
10369 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
10370 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
10371 #endif
10372 
10373 	if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
10374 		t4_tmr_idx = TMR_IDX;
10375 
10376 	if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
10377 		t4_pktc_idx = PKTC_IDX;
10378 
10379 	if (t4_qsize_txq < 128)
10380 		t4_qsize_txq = 128;
10381 
10382 	if (t4_qsize_rxq < 128)
10383 		t4_qsize_rxq = 128;
10384 	while (t4_qsize_rxq & 7)
10385 		t4_qsize_rxq++;
10386 
10387 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
10388 
10389 	/*
10390 	 * Number of VIs to create per-port.  The first VI is the "main" regular
10391 	 * VI for the port.  The rest are additional virtual interfaces on the
10392 	 * same physical port.  Note that the main VI does not have native
10393 	 * netmap support but the extra VIs do.
10394 	 *
10395 	 * Limit the number of VIs per port to the number of available
10396 	 * MAC addresses per port.
10397 	 */
10398 	if (t4_num_vis < 1)
10399 		t4_num_vis = 1;
10400 	if (t4_num_vis > nitems(vi_mac_funcs)) {
10401 		t4_num_vis = nitems(vi_mac_funcs);
10402 		printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
10403 	}
10404 
10405 	if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
10406 		pcie_relaxed_ordering = 1;
10407 #if defined(__i386__) || defined(__amd64__)
10408 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
10409 			pcie_relaxed_ordering = 0;
10410 #endif
10411 	}
10412 }
10413 
10414 #ifdef DDB
10415 static void
10416 t4_dump_tcb(struct adapter *sc, int tid)
10417 {
10418 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
10419 
10420 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
10421 	save = t4_read_reg(sc, reg);
10422 	base = sc->memwin[2].mw_base;
10423 
10424 	/* Dump TCB for the tid */
10425 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
10426 	tcb_addr += tid * TCB_SIZE;
10427 
10428 	if (is_t4(sc)) {
10429 		pf = 0;
10430 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
10431 	} else {
10432 		pf = V_PFNUM(sc->pf);
10433 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
10434 	}
10435 	t4_write_reg(sc, reg, win_pos | pf);
10436 	t4_read_reg(sc, reg);
10437 
10438 	off = tcb_addr - win_pos;
10439 	for (i = 0; i < 4; i++) {
10440 		uint32_t buf[8];
10441 		for (j = 0; j < 8; j++, off += 4)
10442 			buf[j] = htonl(t4_read_reg(sc, base + off));
10443 
10444 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
10445 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
10446 		    buf[7]);
10447 	}
10448 
10449 	t4_write_reg(sc, reg, save);
10450 	t4_read_reg(sc, reg);
10451 }
10452 
10453 static void
10454 t4_dump_devlog(struct adapter *sc)
10455 {
10456 	struct devlog_params *dparams = &sc->params.devlog;
10457 	struct fw_devlog_e e;
10458 	int i, first, j, m, nentries, rc;
10459 	uint64_t ftstamp = UINT64_MAX;
10460 
10461 	if (dparams->start == 0) {
10462 		db_printf("devlog params not valid\n");
10463 		return;
10464 	}
10465 
10466 	nentries = dparams->size / sizeof(struct fw_devlog_e);
10467 	m = fwmtype_to_hwmtype(dparams->memtype);
10468 
10469 	/* Find the first entry. */
10470 	first = -1;
10471 	for (i = 0; i < nentries && !db_pager_quit; i++) {
10472 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10473 		    sizeof(e), (void *)&e);
10474 		if (rc != 0)
10475 			break;
10476 
10477 		if (e.timestamp == 0)
10478 			break;
10479 
10480 		e.timestamp = be64toh(e.timestamp);
10481 		if (e.timestamp < ftstamp) {
10482 			ftstamp = e.timestamp;
10483 			first = i;
10484 		}
10485 	}
10486 
10487 	if (first == -1)
10488 		return;
10489 
10490 	i = first;
10491 	do {
10492 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10493 		    sizeof(e), (void *)&e);
10494 		if (rc != 0)
10495 			return;
10496 
10497 		if (e.timestamp == 0)
10498 			return;
10499 
10500 		e.timestamp = be64toh(e.timestamp);
10501 		e.seqno = be32toh(e.seqno);
10502 		for (j = 0; j < 8; j++)
10503 			e.params[j] = be32toh(e.params[j]);
10504 
10505 		db_printf("%10d  %15ju  %8s  %8s  ",
10506 		    e.seqno, e.timestamp,
10507 		    (e.level < nitems(devlog_level_strings) ?
10508 			devlog_level_strings[e.level] : "UNKNOWN"),
10509 		    (e.facility < nitems(devlog_facility_strings) ?
10510 			devlog_facility_strings[e.facility] : "UNKNOWN"));
10511 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
10512 		    e.params[3], e.params[4], e.params[5], e.params[6],
10513 		    e.params[7]);
10514 
10515 		if (++i == nentries)
10516 			i = 0;
10517 	} while (i != first && !db_pager_quit);
10518 }
10519 
10520 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
10521 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
10522 
10523 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
10524 {
10525 	device_t dev;
10526 	int t;
10527 	bool valid;
10528 
10529 	valid = false;
10530 	t = db_read_token();
10531 	if (t == tIDENT) {
10532 		dev = device_lookup_by_name(db_tok_string);
10533 		valid = true;
10534 	}
10535 	db_skip_to_eol();
10536 	if (!valid) {
10537 		db_printf("usage: show t4 devlog <nexus>\n");
10538 		return;
10539 	}
10540 
10541 	if (dev == NULL) {
10542 		db_printf("device not found\n");
10543 		return;
10544 	}
10545 
10546 	t4_dump_devlog(device_get_softc(dev));
10547 }
10548 
10549 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
10550 {
10551 	device_t dev;
10552 	int radix, tid, t;
10553 	bool valid;
10554 
10555 	valid = false;
10556 	radix = db_radix;
10557 	db_radix = 10;
10558 	t = db_read_token();
10559 	if (t == tIDENT) {
10560 		dev = device_lookup_by_name(db_tok_string);
10561 		t = db_read_token();
10562 		if (t == tNUMBER) {
10563 			tid = db_tok_number;
10564 			valid = true;
10565 		}
10566 	}
10567 	db_radix = radix;
10568 	db_skip_to_eol();
10569 	if (!valid) {
10570 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
10571 		return;
10572 	}
10573 
10574 	if (dev == NULL) {
10575 		db_printf("device not found\n");
10576 		return;
10577 	}
10578 	if (tid < 0) {
10579 		db_printf("invalid tid\n");
10580 		return;
10581 	}
10582 
10583 	t4_dump_tcb(device_get_softc(dev), tid);
10584 }
10585 #endif
10586 
10587 /*
10588  * Borrowed from cesa_prep_aes_key().
10589  *
10590  * NB: The crypto engine wants the words in the decryption key in reverse
10591  * order.
10592  */
10593 void
10594 t4_aes_getdeckey(void *dec_key, const void *enc_key, unsigned int kbits)
10595 {
10596 	uint32_t ek[4 * (RIJNDAEL_MAXNR + 1)];
10597 	uint32_t *dkey;
10598 	int i;
10599 
10600 	rijndaelKeySetupEnc(ek, enc_key, kbits);
10601 	dkey = dec_key;
10602 	dkey += (kbits / 8) / 4;
10603 
10604 	switch (kbits) {
10605 	case 128:
10606 		for (i = 0; i < 4; i++)
10607 			*--dkey = htobe32(ek[4 * 10 + i]);
10608 		break;
10609 	case 192:
10610 		for (i = 0; i < 2; i++)
10611 			*--dkey = htobe32(ek[4 * 11 + 2 + i]);
10612 		for (i = 0; i < 4; i++)
10613 			*--dkey = htobe32(ek[4 * 12 + i]);
10614 		break;
10615 	case 256:
10616 		for (i = 0; i < 4; i++)
10617 			*--dkey = htobe32(ek[4 * 13 + i]);
10618 		for (i = 0; i < 4; i++)
10619 			*--dkey = htobe32(ek[4 * 14 + i]);
10620 		break;
10621 	}
10622 	MPASS(dkey == dec_key);
10623 }
10624 
10625 static struct sx mlu;	/* mod load unload */
10626 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
10627 
10628 static int
10629 mod_event(module_t mod, int cmd, void *arg)
10630 {
10631 	int rc = 0;
10632 	static int loaded = 0;
10633 
10634 	switch (cmd) {
10635 	case MOD_LOAD:
10636 		sx_xlock(&mlu);
10637 		if (loaded++ == 0) {
10638 			t4_sge_modload();
10639 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10640 			    t4_filter_rpl, CPL_COOKIE_FILTER);
10641 			t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
10642 			    do_l2t_write_rpl, CPL_COOKIE_FILTER);
10643 			t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL,
10644 			    t4_hashfilter_ao_rpl, CPL_COOKIE_HASHFILTER);
10645 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10646 			    t4_hashfilter_tcb_rpl, CPL_COOKIE_HASHFILTER);
10647 			t4_register_shared_cpl_handler(CPL_ABORT_RPL_RSS,
10648 			    t4_del_hashfilter_rpl, CPL_COOKIE_HASHFILTER);
10649 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
10650 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
10651 			t4_register_cpl_handler(CPL_SMT_WRITE_RPL,
10652 			    do_smt_write_rpl);
10653 			sx_init(&t4_list_lock, "T4/T5 adapters");
10654 			SLIST_INIT(&t4_list);
10655 #ifdef TCP_OFFLOAD
10656 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
10657 			SLIST_INIT(&t4_uld_list);
10658 #endif
10659 #ifdef INET6
10660 			t4_clip_modload();
10661 #endif
10662 			t4_tracer_modload();
10663 			tweak_tunables();
10664 		}
10665 		sx_xunlock(&mlu);
10666 		break;
10667 
10668 	case MOD_UNLOAD:
10669 		sx_xlock(&mlu);
10670 		if (--loaded == 0) {
10671 			int tries;
10672 
10673 			sx_slock(&t4_list_lock);
10674 			if (!SLIST_EMPTY(&t4_list)) {
10675 				rc = EBUSY;
10676 				sx_sunlock(&t4_list_lock);
10677 				goto done_unload;
10678 			}
10679 #ifdef TCP_OFFLOAD
10680 			sx_slock(&t4_uld_list_lock);
10681 			if (!SLIST_EMPTY(&t4_uld_list)) {
10682 				rc = EBUSY;
10683 				sx_sunlock(&t4_uld_list_lock);
10684 				sx_sunlock(&t4_list_lock);
10685 				goto done_unload;
10686 			}
10687 #endif
10688 			tries = 0;
10689 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
10690 				uprintf("%ju clusters with custom free routine "
10691 				    "still is use.\n", t4_sge_extfree_refs());
10692 				pause("t4unload", 2 * hz);
10693 			}
10694 #ifdef TCP_OFFLOAD
10695 			sx_sunlock(&t4_uld_list_lock);
10696 #endif
10697 			sx_sunlock(&t4_list_lock);
10698 
10699 			if (t4_sge_extfree_refs() == 0) {
10700 				t4_tracer_modunload();
10701 #ifdef INET6
10702 				t4_clip_modunload();
10703 #endif
10704 #ifdef TCP_OFFLOAD
10705 				sx_destroy(&t4_uld_list_lock);
10706 #endif
10707 				sx_destroy(&t4_list_lock);
10708 				t4_sge_modunload();
10709 				loaded = 0;
10710 			} else {
10711 				rc = EBUSY;
10712 				loaded++;	/* undo earlier decrement */
10713 			}
10714 		}
10715 done_unload:
10716 		sx_xunlock(&mlu);
10717 		break;
10718 	}
10719 
10720 	return (rc);
10721 }
10722 
10723 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10724 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10725 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10726 
10727 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10728 MODULE_VERSION(t4nex, 1);
10729 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10730 #ifdef DEV_NETMAP
10731 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10732 #endif /* DEV_NETMAP */
10733 
10734 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10735 MODULE_VERSION(t5nex, 1);
10736 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10737 #ifdef DEV_NETMAP
10738 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10739 #endif /* DEV_NETMAP */
10740 
10741 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10742 MODULE_VERSION(t6nex, 1);
10743 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10744 #ifdef DEV_NETMAP
10745 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10746 #endif /* DEV_NETMAP */
10747 
10748 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10749 MODULE_VERSION(cxgbe, 1);
10750 
10751 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10752 MODULE_VERSION(cxl, 1);
10753 
10754 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10755 MODULE_VERSION(cc, 1);
10756 
10757 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10758 MODULE_VERSION(vcxgbe, 1);
10759 
10760 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10761 MODULE_VERSION(vcxl, 1);
10762 
10763 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10764 MODULE_VERSION(vcc, 1);
10765