xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision 4e1ef62a367de01ccb7156bfe6ec2d613d6eb860)
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 #if defined(__i386__) || defined(__amd64__)
67 #include <machine/md_var.h>
68 #include <machine/cputypes.h>
69 #include <vm/vm.h>
70 #include <vm/pmap.h>
71 #endif
72 #include <crypto/rijndael/rijndael.h>
73 #ifdef DDB
74 #include <ddb/ddb.h>
75 #include <ddb/db_lex.h>
76 #endif
77 
78 #include "common/common.h"
79 #include "common/t4_msg.h"
80 #include "common/t4_regs.h"
81 #include "common/t4_regs_values.h"
82 #include "cudbg/cudbg.h"
83 #include "t4_ioctl.h"
84 #include "t4_l2t.h"
85 #include "t4_mp_ring.h"
86 #include "t4_if.h"
87 #include "t4_smt.h"
88 
89 /* T4 bus driver interface */
90 static int t4_probe(device_t);
91 static int t4_attach(device_t);
92 static int t4_detach(device_t);
93 static int t4_ready(device_t);
94 static int t4_read_port_device(device_t, int, device_t *);
95 static device_method_t t4_methods[] = {
96 	DEVMETHOD(device_probe,		t4_probe),
97 	DEVMETHOD(device_attach,	t4_attach),
98 	DEVMETHOD(device_detach,	t4_detach),
99 
100 	DEVMETHOD(t4_is_main_ready,	t4_ready),
101 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
102 
103 	DEVMETHOD_END
104 };
105 static driver_t t4_driver = {
106 	"t4nex",
107 	t4_methods,
108 	sizeof(struct adapter)
109 };
110 
111 
112 /* T4 port (cxgbe) interface */
113 static int cxgbe_probe(device_t);
114 static int cxgbe_attach(device_t);
115 static int cxgbe_detach(device_t);
116 device_method_t cxgbe_methods[] = {
117 	DEVMETHOD(device_probe,		cxgbe_probe),
118 	DEVMETHOD(device_attach,	cxgbe_attach),
119 	DEVMETHOD(device_detach,	cxgbe_detach),
120 	{ 0, 0 }
121 };
122 static driver_t cxgbe_driver = {
123 	"cxgbe",
124 	cxgbe_methods,
125 	sizeof(struct port_info)
126 };
127 
128 /* T4 VI (vcxgbe) interface */
129 static int vcxgbe_probe(device_t);
130 static int vcxgbe_attach(device_t);
131 static int vcxgbe_detach(device_t);
132 static device_method_t vcxgbe_methods[] = {
133 	DEVMETHOD(device_probe,		vcxgbe_probe),
134 	DEVMETHOD(device_attach,	vcxgbe_attach),
135 	DEVMETHOD(device_detach,	vcxgbe_detach),
136 	{ 0, 0 }
137 };
138 static driver_t vcxgbe_driver = {
139 	"vcxgbe",
140 	vcxgbe_methods,
141 	sizeof(struct vi_info)
142 };
143 
144 static d_ioctl_t t4_ioctl;
145 
146 static struct cdevsw t4_cdevsw = {
147        .d_version = D_VERSION,
148        .d_ioctl = t4_ioctl,
149        .d_name = "t4nex",
150 };
151 
152 /* T5 bus driver interface */
153 static int t5_probe(device_t);
154 static device_method_t t5_methods[] = {
155 	DEVMETHOD(device_probe,		t5_probe),
156 	DEVMETHOD(device_attach,	t4_attach),
157 	DEVMETHOD(device_detach,	t4_detach),
158 
159 	DEVMETHOD(t4_is_main_ready,	t4_ready),
160 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
161 
162 	DEVMETHOD_END
163 };
164 static driver_t t5_driver = {
165 	"t5nex",
166 	t5_methods,
167 	sizeof(struct adapter)
168 };
169 
170 
171 /* T5 port (cxl) interface */
172 static driver_t cxl_driver = {
173 	"cxl",
174 	cxgbe_methods,
175 	sizeof(struct port_info)
176 };
177 
178 /* T5 VI (vcxl) interface */
179 static driver_t vcxl_driver = {
180 	"vcxl",
181 	vcxgbe_methods,
182 	sizeof(struct vi_info)
183 };
184 
185 /* T6 bus driver interface */
186 static int t6_probe(device_t);
187 static device_method_t t6_methods[] = {
188 	DEVMETHOD(device_probe,		t6_probe),
189 	DEVMETHOD(device_attach,	t4_attach),
190 	DEVMETHOD(device_detach,	t4_detach),
191 
192 	DEVMETHOD(t4_is_main_ready,	t4_ready),
193 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
194 
195 	DEVMETHOD_END
196 };
197 static driver_t t6_driver = {
198 	"t6nex",
199 	t6_methods,
200 	sizeof(struct adapter)
201 };
202 
203 
204 /* T6 port (cc) interface */
205 static driver_t cc_driver = {
206 	"cc",
207 	cxgbe_methods,
208 	sizeof(struct port_info)
209 };
210 
211 /* T6 VI (vcc) interface */
212 static driver_t vcc_driver = {
213 	"vcc",
214 	vcxgbe_methods,
215 	sizeof(struct vi_info)
216 };
217 
218 /* ifnet + media interface */
219 static void cxgbe_init(void *);
220 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
221 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
222 static void cxgbe_qflush(struct ifnet *);
223 static int cxgbe_media_change(struct ifnet *);
224 static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
225 
226 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
227 
228 /*
229  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
230  * then ADAPTER_LOCK, then t4_uld_list_lock.
231  */
232 static struct sx t4_list_lock;
233 SLIST_HEAD(, adapter) t4_list;
234 #ifdef TCP_OFFLOAD
235 static struct sx t4_uld_list_lock;
236 SLIST_HEAD(, uld_info) t4_uld_list;
237 #endif
238 
239 /*
240  * Tunables.  See tweak_tunables() too.
241  *
242  * Each tunable is set to a default value here if it's known at compile-time.
243  * Otherwise it is set to -n as an indication to tweak_tunables() that it should
244  * provide a reasonable default (upto n) when the driver is loaded.
245  *
246  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
247  * T5 are under hw.cxl.
248  */
249 
250 /*
251  * Number of queues for tx and rx, NIC and offload.
252  */
253 #define NTXQ 16
254 int t4_ntxq = -NTXQ;
255 TUNABLE_INT("hw.cxgbe.ntxq", &t4_ntxq);
256 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq);	/* Old name, undocumented */
257 
258 #define NRXQ 8
259 int t4_nrxq = -NRXQ;
260 TUNABLE_INT("hw.cxgbe.nrxq", &t4_nrxq);
261 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq);	/* Old name, undocumented */
262 
263 #define NTXQ_VI 1
264 static int t4_ntxq_vi = -NTXQ_VI;
265 TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
266 
267 #define NRXQ_VI 1
268 static int t4_nrxq_vi = -NRXQ_VI;
269 TUNABLE_INT("hw.cxgbe.nrxq_vi", &t4_nrxq_vi);
270 
271 static int t4_rsrv_noflowq = 0;
272 TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
273 
274 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
275 #define NOFLDTXQ 8
276 static int t4_nofldtxq = -NOFLDTXQ;
277 TUNABLE_INT("hw.cxgbe.nofldtxq", &t4_nofldtxq);
278 
279 #define NOFLDRXQ 2
280 static int t4_nofldrxq = -NOFLDRXQ;
281 TUNABLE_INT("hw.cxgbe.nofldrxq", &t4_nofldrxq);
282 
283 #define NOFLDTXQ_VI 1
284 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
285 TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
286 
287 #define NOFLDRXQ_VI 1
288 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
289 TUNABLE_INT("hw.cxgbe.nofldrxq_vi", &t4_nofldrxq_vi);
290 
291 #define TMR_IDX_OFLD 1
292 int t4_tmr_idx_ofld = TMR_IDX_OFLD;
293 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_ofld", &t4_tmr_idx_ofld);
294 
295 #define PKTC_IDX_OFLD (-1)
296 int t4_pktc_idx_ofld = PKTC_IDX_OFLD;
297 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_ofld", &t4_pktc_idx_ofld);
298 
299 /* 0 means chip/fw default, non-zero number is value in microseconds */
300 static u_long t4_toe_keepalive_idle = 0;
301 TUNABLE_ULONG("hw.cxgbe.toe.keepalive_idle", &t4_toe_keepalive_idle);
302 
303 /* 0 means chip/fw default, non-zero number is value in microseconds */
304 static u_long t4_toe_keepalive_interval = 0;
305 TUNABLE_ULONG("hw.cxgbe.toe.keepalive_interval", &t4_toe_keepalive_interval);
306 
307 /* 0 means chip/fw default, non-zero number is # of keepalives before abort */
308 static int t4_toe_keepalive_count = 0;
309 TUNABLE_INT("hw.cxgbe.toe.keepalive_count", &t4_toe_keepalive_count);
310 
311 /* 0 means chip/fw default, non-zero number is value in microseconds */
312 static u_long t4_toe_rexmt_min = 0;
313 TUNABLE_ULONG("hw.cxgbe.toe.rexmt_min", &t4_toe_rexmt_min);
314 
315 /* 0 means chip/fw default, non-zero number is value in microseconds */
316 static u_long t4_toe_rexmt_max = 0;
317 TUNABLE_ULONG("hw.cxgbe.toe.rexmt_max", &t4_toe_rexmt_max);
318 
319 /* 0 means chip/fw default, non-zero number is # of rexmt before abort */
320 static int t4_toe_rexmt_count = 0;
321 TUNABLE_INT("hw.cxgbe.toe.rexmt_count", &t4_toe_rexmt_count);
322 
323 /* -1 means chip/fw default, other values are raw backoff values to use */
324 static int t4_toe_rexmt_backoff[16] = {
325 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
326 };
327 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.0", &t4_toe_rexmt_backoff[0]);
328 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.1", &t4_toe_rexmt_backoff[1]);
329 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.2", &t4_toe_rexmt_backoff[2]);
330 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.3", &t4_toe_rexmt_backoff[3]);
331 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.4", &t4_toe_rexmt_backoff[4]);
332 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.5", &t4_toe_rexmt_backoff[5]);
333 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.6", &t4_toe_rexmt_backoff[6]);
334 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.7", &t4_toe_rexmt_backoff[7]);
335 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.8", &t4_toe_rexmt_backoff[8]);
336 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.9", &t4_toe_rexmt_backoff[9]);
337 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.10", &t4_toe_rexmt_backoff[10]);
338 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.11", &t4_toe_rexmt_backoff[11]);
339 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.12", &t4_toe_rexmt_backoff[12]);
340 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.13", &t4_toe_rexmt_backoff[13]);
341 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.14", &t4_toe_rexmt_backoff[14]);
342 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.15", &t4_toe_rexmt_backoff[15]);
343 #endif
344 
345 #ifdef DEV_NETMAP
346 #define NNMTXQ_VI 2
347 static int t4_nnmtxq_vi = -NNMTXQ_VI;
348 TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
349 
350 #define NNMRXQ_VI 2
351 static int t4_nnmrxq_vi = -NNMRXQ_VI;
352 TUNABLE_INT("hw.cxgbe.nnmrxq_vi", &t4_nnmrxq_vi);
353 #endif
354 
355 /*
356  * Holdoff parameters for ports.
357  */
358 #define TMR_IDX 1
359 int t4_tmr_idx = TMR_IDX;
360 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx", &t4_tmr_idx);
361 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx);	/* Old name */
362 
363 #define PKTC_IDX (-1)
364 int t4_pktc_idx = PKTC_IDX;
365 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx", &t4_pktc_idx);
366 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx);	/* Old name */
367 
368 /*
369  * Size (# of entries) of each tx and rx queue.
370  */
371 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
372 TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
373 
374 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
375 TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
376 
377 /*
378  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
379  */
380 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
381 TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
382 
383 /*
384  * Configuration file.  All the _CF names here are special.
385  */
386 #define DEFAULT_CF	"default"
387 #define BUILTIN_CF	"built-in"
388 #define FLASH_CF	"flash"
389 #define UWIRE_CF	"uwire"
390 #define FPGA_CF		"fpga"
391 static char t4_cfg_file[32] = DEFAULT_CF;
392 TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
393 
394 /*
395  * PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
396  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
397  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
398  *            mark or when signalled to do so, 0 to never emit PAUSE.
399  */
400 static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
401 TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
402 
403 /*
404  * Forward Error Correction settings (bit 0, 1, 2 = FEC_RS, FEC_BASER_RS,
405  * FEC_RESERVED respectively).
406  * -1 to run with the firmware default.
407  *  0 to disable FEC.
408  */
409 static int t4_fec = -1;
410 TUNABLE_INT("hw.cxgbe.fec", &t4_fec);
411 
412 /*
413  * Link autonegotiation.
414  * -1 to run with the firmware default.
415  *  0 to disable.
416  *  1 to enable.
417  */
418 static int t4_autoneg = -1;
419 TUNABLE_INT("hw.cxgbe.autoneg", &t4_autoneg);
420 
421 /*
422  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
423  * encouraged respectively).
424  */
425 static unsigned int t4_fw_install = 1;
426 TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
427 
428 /*
429  * ASIC features that will be used.  Disable the ones you don't want so that the
430  * chip resources aren't wasted on features that will not be used.
431  */
432 static int t4_nbmcaps_allowed = 0;
433 TUNABLE_INT("hw.cxgbe.nbmcaps_allowed", &t4_nbmcaps_allowed);
434 
435 static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
436 TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
437 
438 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
439     FW_CAPS_CONFIG_SWITCH_EGRESS;
440 TUNABLE_INT("hw.cxgbe.switchcaps_allowed", &t4_switchcaps_allowed);
441 
442 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
443 	FW_CAPS_CONFIG_NIC_HASHFILTER | FW_CAPS_CONFIG_NIC_ETHOFLD;
444 TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
445 
446 static int t4_toecaps_allowed = -1;
447 TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
448 
449 static int t4_rdmacaps_allowed = -1;
450 TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
451 
452 static int t4_cryptocaps_allowed = -1;
453 TUNABLE_INT("hw.cxgbe.cryptocaps_allowed", &t4_cryptocaps_allowed);
454 
455 static int t4_iscsicaps_allowed = -1;
456 TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
457 
458 static int t4_fcoecaps_allowed = 0;
459 TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
460 
461 static int t5_write_combine = 0;
462 TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
463 
464 static int t4_num_vis = 1;
465 TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
466 /*
467  * PCIe Relaxed Ordering.
468  * -1: driver should figure out a good value.
469  * 0: disable RO.
470  * 1: enable RO.
471  * 2: leave RO alone.
472  */
473 static int pcie_relaxed_ordering = -1;
474 TUNABLE_INT("hw.cxgbe.pcie_relaxed_ordering", &pcie_relaxed_ordering);
475 
476 static int t4_panic_on_fatal_err = 0;
477 TUNABLE_INT("hw.cxgbe.panic_on_fatal_err", &t4_panic_on_fatal_err);
478 
479 #ifdef TCP_OFFLOAD
480 /*
481  * TOE tunables.
482  */
483 static int t4_cop_managed_offloading = 0;
484 TUNABLE_INT("hw.cxgbe.cop_managed_offloading", &t4_cop_managed_offloading);
485 #endif
486 
487 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
488 static int vi_mac_funcs[] = {
489 	FW_VI_FUNC_ETH,
490 	FW_VI_FUNC_OFLD,
491 	FW_VI_FUNC_IWARP,
492 	FW_VI_FUNC_OPENISCSI,
493 	FW_VI_FUNC_OPENFCOE,
494 	FW_VI_FUNC_FOISCSI,
495 	FW_VI_FUNC_FOFCOE,
496 };
497 
498 struct intrs_and_queues {
499 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
500 	uint16_t num_vis;	/* number of VIs for each port */
501 	uint16_t nirq;		/* Total # of vectors */
502 	uint16_t ntxq;		/* # of NIC txq's for each port */
503 	uint16_t nrxq;		/* # of NIC rxq's for each port */
504 	uint16_t nofldtxq;	/* # of TOE/ETHOFLD txq's for each port */
505 	uint16_t nofldrxq;	/* # of TOE rxq's for each port */
506 
507 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
508 	uint16_t ntxq_vi;	/* # of NIC txq's */
509 	uint16_t nrxq_vi;	/* # of NIC rxq's */
510 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
511 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
512 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
513 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
514 };
515 
516 static void setup_memwin(struct adapter *);
517 static void position_memwin(struct adapter *, int, uint32_t);
518 static int validate_mem_range(struct adapter *, uint32_t, int);
519 static int fwmtype_to_hwmtype(int);
520 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
521     uint32_t *);
522 static int fixup_devlog_params(struct adapter *);
523 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
524 static int prep_firmware(struct adapter *);
525 static int partition_resources(struct adapter *, const struct firmware *,
526     const char *);
527 static int get_params__pre_init(struct adapter *);
528 static int get_params__post_init(struct adapter *);
529 static int set_params__post_init(struct adapter *);
530 static void t4_set_desc(struct adapter *);
531 static void build_medialist(struct port_info *, struct ifmedia *);
532 static void init_l1cfg(struct port_info *);
533 static int apply_l1cfg(struct port_info *);
534 static int cxgbe_init_synchronized(struct vi_info *);
535 static int cxgbe_uninit_synchronized(struct vi_info *);
536 static void quiesce_txq(struct adapter *, struct sge_txq *);
537 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
538 static void quiesce_iq(struct adapter *, struct sge_iq *);
539 static void quiesce_fl(struct adapter *, struct sge_fl *);
540 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
541     driver_intr_t *, void *, char *);
542 static int t4_free_irq(struct adapter *, struct irq *);
543 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
544 static void vi_refresh_stats(struct adapter *, struct vi_info *);
545 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
546 static void cxgbe_tick(void *);
547 static void cxgbe_sysctls(struct port_info *);
548 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
549 static int sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS);
550 static int sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS);
551 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
552 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
553 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
554 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
555 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
556 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
557 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
558 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
559 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
560 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
561 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
562 static int sysctl_loadavg(SYSCTL_HANDLER_ARGS);
563 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
564 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
565 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
566 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
567 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
568 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
569 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
570 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
571 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
572 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
573 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
574 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
575 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
576 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
577 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
578 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
579 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
580 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
581 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
582 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
583 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
584 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
585 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
586 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
587 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
588 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
589 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
590 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
591 static int sysctl_cpus(SYSCTL_HANDLER_ARGS);
592 #ifdef TCP_OFFLOAD
593 static int sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS);
594 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
595 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
596 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
597 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
598 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
599 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
600 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
601 #endif
602 static int get_sge_context(struct adapter *, struct t4_sge_context *);
603 static int load_fw(struct adapter *, struct t4_data *);
604 static int load_cfg(struct adapter *, struct t4_data *);
605 static int load_boot(struct adapter *, struct t4_bootrom *);
606 static int load_bootcfg(struct adapter *, struct t4_data *);
607 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
608 static void free_offload_policy(struct t4_offload_policy *);
609 static int set_offload_policy(struct adapter *, struct t4_offload_policy *);
610 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
611 static int read_i2c(struct adapter *, struct t4_i2c_data *);
612 #ifdef TCP_OFFLOAD
613 static int toe_capability(struct vi_info *, int);
614 #endif
615 static int mod_event(module_t, int, void *);
616 static int notify_siblings(device_t, int);
617 
618 struct {
619 	uint16_t device;
620 	char *desc;
621 } t4_pciids[] = {
622 	{0xa000, "Chelsio Terminator 4 FPGA"},
623 	{0x4400, "Chelsio T440-dbg"},
624 	{0x4401, "Chelsio T420-CR"},
625 	{0x4402, "Chelsio T422-CR"},
626 	{0x4403, "Chelsio T440-CR"},
627 	{0x4404, "Chelsio T420-BCH"},
628 	{0x4405, "Chelsio T440-BCH"},
629 	{0x4406, "Chelsio T440-CH"},
630 	{0x4407, "Chelsio T420-SO"},
631 	{0x4408, "Chelsio T420-CX"},
632 	{0x4409, "Chelsio T420-BT"},
633 	{0x440a, "Chelsio T404-BT"},
634 	{0x440e, "Chelsio T440-LP-CR"},
635 }, t5_pciids[] = {
636 	{0xb000, "Chelsio Terminator 5 FPGA"},
637 	{0x5400, "Chelsio T580-dbg"},
638 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
639 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
640 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
641 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
642 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
643 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
644 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
645 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
646 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
647 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
648 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
649 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
650 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
651 	{0x5418,  "Chelsio T540-BT"},		/* 4 x 10GBaseT */
652 	{0x5419,  "Chelsio T540-LP-BT"},	/* 4 x 10GBaseT */
653 	{0x541a,  "Chelsio T540-SO-BT"},	/* 4 x 10GBaseT, nomem */
654 	{0x541b,  "Chelsio T540-SO-CR"},	/* 4 x 10G, nomem */
655 }, t6_pciids[] = {
656 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
657 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
658 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
659 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
660 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
661 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
662 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
663 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
664 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
665 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
666 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
667 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
668 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
669 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
670 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
671 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
672 
673 	/* Custom */
674 	{0x6480, "Custom T6225-CR"},
675 	{0x6481, "Custom T62100-CR"},
676 	{0x6482, "Custom T6225-CR"},
677 	{0x6483, "Custom T62100-CR"},
678 	{0x6484, "Custom T64100-CR"},
679 	{0x6485, "Custom T6240-SO"},
680 	{0x6486, "Custom T6225-SO-CR"},
681 	{0x6487, "Custom T6225-CR"},
682 };
683 
684 #ifdef TCP_OFFLOAD
685 /*
686  * service_iq_fl() has an iq and needs the fl.  Offset of fl from the iq should
687  * be exactly the same for both rxq and ofld_rxq.
688  */
689 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
690 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
691 #endif
692 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
693 
694 static int
695 t4_probe(device_t dev)
696 {
697 	int i;
698 	uint16_t v = pci_get_vendor(dev);
699 	uint16_t d = pci_get_device(dev);
700 	uint8_t f = pci_get_function(dev);
701 
702 	if (v != PCI_VENDOR_ID_CHELSIO)
703 		return (ENXIO);
704 
705 	/* Attach only to PF0 of the FPGA */
706 	if (d == 0xa000 && f != 0)
707 		return (ENXIO);
708 
709 	for (i = 0; i < nitems(t4_pciids); i++) {
710 		if (d == t4_pciids[i].device) {
711 			device_set_desc(dev, t4_pciids[i].desc);
712 			return (BUS_PROBE_DEFAULT);
713 		}
714 	}
715 
716 	return (ENXIO);
717 }
718 
719 static int
720 t5_probe(device_t dev)
721 {
722 	int i;
723 	uint16_t v = pci_get_vendor(dev);
724 	uint16_t d = pci_get_device(dev);
725 	uint8_t f = pci_get_function(dev);
726 
727 	if (v != PCI_VENDOR_ID_CHELSIO)
728 		return (ENXIO);
729 
730 	/* Attach only to PF0 of the FPGA */
731 	if (d == 0xb000 && f != 0)
732 		return (ENXIO);
733 
734 	for (i = 0; i < nitems(t5_pciids); i++) {
735 		if (d == t5_pciids[i].device) {
736 			device_set_desc(dev, t5_pciids[i].desc);
737 			return (BUS_PROBE_DEFAULT);
738 		}
739 	}
740 
741 	return (ENXIO);
742 }
743 
744 static int
745 t6_probe(device_t dev)
746 {
747 	int i;
748 	uint16_t v = pci_get_vendor(dev);
749 	uint16_t d = pci_get_device(dev);
750 
751 	if (v != PCI_VENDOR_ID_CHELSIO)
752 		return (ENXIO);
753 
754 	for (i = 0; i < nitems(t6_pciids); i++) {
755 		if (d == t6_pciids[i].device) {
756 			device_set_desc(dev, t6_pciids[i].desc);
757 			return (BUS_PROBE_DEFAULT);
758 		}
759 	}
760 
761 	return (ENXIO);
762 }
763 
764 static void
765 t5_attribute_workaround(device_t dev)
766 {
767 	device_t root_port;
768 	uint32_t v;
769 
770 	/*
771 	 * The T5 chips do not properly echo the No Snoop and Relaxed
772 	 * Ordering attributes when replying to a TLP from a Root
773 	 * Port.  As a workaround, find the parent Root Port and
774 	 * disable No Snoop and Relaxed Ordering.  Note that this
775 	 * affects all devices under this root port.
776 	 */
777 	root_port = pci_find_pcie_root_port(dev);
778 	if (root_port == NULL) {
779 		device_printf(dev, "Unable to find parent root port\n");
780 		return;
781 	}
782 
783 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
784 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
785 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
786 	    0)
787 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
788 		    device_get_nameunit(root_port));
789 }
790 
791 static const struct devnames devnames[] = {
792 	{
793 		.nexus_name = "t4nex",
794 		.ifnet_name = "cxgbe",
795 		.vi_ifnet_name = "vcxgbe",
796 		.pf03_drv_name = "t4iov",
797 		.vf_nexus_name = "t4vf",
798 		.vf_ifnet_name = "cxgbev"
799 	}, {
800 		.nexus_name = "t5nex",
801 		.ifnet_name = "cxl",
802 		.vi_ifnet_name = "vcxl",
803 		.pf03_drv_name = "t5iov",
804 		.vf_nexus_name = "t5vf",
805 		.vf_ifnet_name = "cxlv"
806 	}, {
807 		.nexus_name = "t6nex",
808 		.ifnet_name = "cc",
809 		.vi_ifnet_name = "vcc",
810 		.pf03_drv_name = "t6iov",
811 		.vf_nexus_name = "t6vf",
812 		.vf_ifnet_name = "ccv"
813 	}
814 };
815 
816 void
817 t4_init_devnames(struct adapter *sc)
818 {
819 	int id;
820 
821 	id = chip_id(sc);
822 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
823 		sc->names = &devnames[id - CHELSIO_T4];
824 	else {
825 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
826 		sc->names = NULL;
827 	}
828 }
829 
830 static int
831 t4_attach(device_t dev)
832 {
833 	struct adapter *sc;
834 	int rc = 0, i, j, rqidx, tqidx, nports;
835 	struct make_dev_args mda;
836 	struct intrs_and_queues iaq;
837 	struct sge *s;
838 	uint32_t *buf;
839 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
840 	int ofld_tqidx;
841 #endif
842 #ifdef TCP_OFFLOAD
843 	int ofld_rqidx;
844 #endif
845 #ifdef DEV_NETMAP
846 	int nm_rqidx, nm_tqidx;
847 #endif
848 	int num_vis;
849 
850 	sc = device_get_softc(dev);
851 	sc->dev = dev;
852 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
853 
854 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
855 		t5_attribute_workaround(dev);
856 	pci_enable_busmaster(dev);
857 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
858 		uint32_t v;
859 
860 		pci_set_max_read_req(dev, 4096);
861 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
862 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
863 		if (pcie_relaxed_ordering == 0 &&
864 		    (v | PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
865 			v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
866 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
867 		} else if (pcie_relaxed_ordering == 1 &&
868 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
869 			v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
870 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
871 		}
872 	}
873 
874 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
875 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
876 	sc->traceq = -1;
877 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
878 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
879 	    device_get_nameunit(dev));
880 
881 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
882 	    device_get_nameunit(dev));
883 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
884 	t4_add_adapter(sc);
885 
886 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
887 	TAILQ_INIT(&sc->sfl);
888 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
889 
890 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
891 
892 	sc->policy = NULL;
893 	rw_init(&sc->policy_lock, "connection offload policy");
894 
895 	rc = t4_map_bars_0_and_4(sc);
896 	if (rc != 0)
897 		goto done; /* error message displayed already */
898 
899 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
900 
901 	/* Prepare the adapter for operation. */
902 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
903 	rc = -t4_prep_adapter(sc, buf);
904 	free(buf, M_CXGBE);
905 	if (rc != 0) {
906 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
907 		goto done;
908 	}
909 
910 	/*
911 	 * This is the real PF# to which we're attaching.  Works from within PCI
912 	 * passthrough environments too, where pci_get_function() could return a
913 	 * different PF# depending on the passthrough configuration.  We need to
914 	 * use the real PF# in all our communication with the firmware.
915 	 */
916 	j = t4_read_reg(sc, A_PL_WHOAMI);
917 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
918 	sc->mbox = sc->pf;
919 
920 	t4_init_devnames(sc);
921 	if (sc->names == NULL) {
922 		rc = ENOTSUP;
923 		goto done; /* error message displayed already */
924 	}
925 
926 	/*
927 	 * Do this really early, with the memory windows set up even before the
928 	 * character device.  The userland tool's register i/o and mem read
929 	 * will work even in "recovery mode".
930 	 */
931 	setup_memwin(sc);
932 	if (t4_init_devlog_params(sc, 0) == 0)
933 		fixup_devlog_params(sc);
934 	make_dev_args_init(&mda);
935 	mda.mda_devsw = &t4_cdevsw;
936 	mda.mda_uid = UID_ROOT;
937 	mda.mda_gid = GID_WHEEL;
938 	mda.mda_mode = 0600;
939 	mda.mda_si_drv1 = sc;
940 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
941 	if (rc != 0)
942 		device_printf(dev, "failed to create nexus char device: %d.\n",
943 		    rc);
944 
945 	/* Go no further if recovery mode has been requested. */
946 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
947 		device_printf(dev, "recovery mode.\n");
948 		goto done;
949 	}
950 
951 #if defined(__i386__)
952 	if ((cpu_feature & CPUID_CX8) == 0) {
953 		device_printf(dev, "64 bit atomics not available.\n");
954 		rc = ENOTSUP;
955 		goto done;
956 	}
957 #endif
958 
959 	/* Prepare the firmware for operation */
960 	rc = prep_firmware(sc);
961 	if (rc != 0)
962 		goto done; /* error message displayed already */
963 
964 	rc = get_params__post_init(sc);
965 	if (rc != 0)
966 		goto done; /* error message displayed already */
967 
968 	rc = set_params__post_init(sc);
969 	if (rc != 0)
970 		goto done; /* error message displayed already */
971 
972 	rc = t4_map_bar_2(sc);
973 	if (rc != 0)
974 		goto done; /* error message displayed already */
975 
976 	rc = t4_create_dma_tag(sc);
977 	if (rc != 0)
978 		goto done; /* error message displayed already */
979 
980 	/*
981 	 * First pass over all the ports - allocate VIs and initialize some
982 	 * basic parameters like mac address, port type, etc.
983 	 */
984 	for_each_port(sc, i) {
985 		struct port_info *pi;
986 
987 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
988 		sc->port[i] = pi;
989 
990 		/* These must be set before t4_port_init */
991 		pi->adapter = sc;
992 		pi->port_id = i;
993 		/*
994 		 * XXX: vi[0] is special so we can't delay this allocation until
995 		 * pi->nvi's final value is known.
996 		 */
997 		pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
998 		    M_ZERO | M_WAITOK);
999 
1000 		/*
1001 		 * Allocate the "main" VI and initialize parameters
1002 		 * like mac addr.
1003 		 */
1004 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1005 		if (rc != 0) {
1006 			device_printf(dev, "unable to initialize port %d: %d\n",
1007 			    i, rc);
1008 			free(pi->vi, M_CXGBE);
1009 			free(pi, M_CXGBE);
1010 			sc->port[i] = NULL;
1011 			goto done;
1012 		}
1013 
1014 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1015 		    device_get_nameunit(dev), i);
1016 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1017 		sc->chan_map[pi->tx_chan] = i;
1018 
1019 		/* All VIs on this port share this media. */
1020 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1021 		    cxgbe_media_status);
1022 
1023 		pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
1024 		if (pi->dev == NULL) {
1025 			device_printf(dev,
1026 			    "failed to add device for port %d.\n", i);
1027 			rc = ENXIO;
1028 			goto done;
1029 		}
1030 		pi->vi[0].dev = pi->dev;
1031 		device_set_softc(pi->dev, pi);
1032 	}
1033 
1034 	/*
1035 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1036 	 */
1037 	nports = sc->params.nports;
1038 	rc = cfg_itype_and_nqueues(sc, &iaq);
1039 	if (rc != 0)
1040 		goto done; /* error message displayed already */
1041 
1042 	num_vis = iaq.num_vis;
1043 	sc->intr_type = iaq.intr_type;
1044 	sc->intr_count = iaq.nirq;
1045 
1046 	s = &sc->sge;
1047 	s->nrxq = nports * iaq.nrxq;
1048 	s->ntxq = nports * iaq.ntxq;
1049 	if (num_vis > 1) {
1050 		s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1051 		s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1052 	}
1053 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1054 	s->neq += nports;		/* ctrl queues: 1 per port */
1055 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1056 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1057 	if (is_offload(sc) || is_ethoffload(sc)) {
1058 		s->nofldtxq = nports * iaq.nofldtxq;
1059 		if (num_vis > 1)
1060 			s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1061 		s->neq += s->nofldtxq;
1062 
1063 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1064 		    M_CXGBE, M_ZERO | M_WAITOK);
1065 	}
1066 #endif
1067 #ifdef TCP_OFFLOAD
1068 	if (is_offload(sc)) {
1069 		s->nofldrxq = nports * iaq.nofldrxq;
1070 		if (num_vis > 1)
1071 			s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1072 		s->neq += s->nofldrxq;	/* free list */
1073 		s->niq += s->nofldrxq;
1074 
1075 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1076 		    M_CXGBE, M_ZERO | M_WAITOK);
1077 	}
1078 #endif
1079 #ifdef DEV_NETMAP
1080 	if (num_vis > 1) {
1081 		s->nnmrxq = nports * (num_vis - 1) * iaq.nnmrxq_vi;
1082 		s->nnmtxq = nports * (num_vis - 1) * iaq.nnmtxq_vi;
1083 	}
1084 	s->neq += s->nnmtxq + s->nnmrxq;
1085 	s->niq += s->nnmrxq;
1086 
1087 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1088 	    M_CXGBE, M_ZERO | M_WAITOK);
1089 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1090 	    M_CXGBE, M_ZERO | M_WAITOK);
1091 #endif
1092 
1093 	s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1094 	    M_ZERO | M_WAITOK);
1095 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1096 	    M_ZERO | M_WAITOK);
1097 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1098 	    M_ZERO | M_WAITOK);
1099 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1100 	    M_ZERO | M_WAITOK);
1101 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1102 	    M_ZERO | M_WAITOK);
1103 
1104 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1105 	    M_ZERO | M_WAITOK);
1106 
1107 	t4_init_l2t(sc, M_WAITOK);
1108 	t4_init_smt(sc, M_WAITOK);
1109 	t4_init_tx_sched(sc);
1110 #ifdef RATELIMIT
1111 	t4_init_etid_table(sc);
1112 #endif
1113 
1114 	/*
1115 	 * Second pass over the ports.  This time we know the number of rx and
1116 	 * tx queues that each port should get.
1117 	 */
1118 	rqidx = tqidx = 0;
1119 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1120 	ofld_tqidx = 0;
1121 #endif
1122 #ifdef TCP_OFFLOAD
1123 	ofld_rqidx = 0;
1124 #endif
1125 #ifdef DEV_NETMAP
1126 	nm_rqidx = nm_tqidx = 0;
1127 #endif
1128 	for_each_port(sc, i) {
1129 		struct port_info *pi = sc->port[i];
1130 		struct vi_info *vi;
1131 
1132 		if (pi == NULL)
1133 			continue;
1134 
1135 		pi->nvi = num_vis;
1136 		for_each_vi(pi, j, vi) {
1137 			vi->pi = pi;
1138 			vi->qsize_rxq = t4_qsize_rxq;
1139 			vi->qsize_txq = t4_qsize_txq;
1140 
1141 			vi->first_rxq = rqidx;
1142 			vi->first_txq = tqidx;
1143 			vi->tmr_idx = t4_tmr_idx;
1144 			vi->pktc_idx = t4_pktc_idx;
1145 			vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1146 			vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1147 
1148 			rqidx += vi->nrxq;
1149 			tqidx += vi->ntxq;
1150 
1151 			if (j == 0 && vi->ntxq > 1)
1152 				vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1153 			else
1154 				vi->rsrv_noflowq = 0;
1155 
1156 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1157 			vi->first_ofld_txq = ofld_tqidx;
1158 			vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1159 			ofld_tqidx += vi->nofldtxq;
1160 #endif
1161 #ifdef TCP_OFFLOAD
1162 			vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1163 			vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1164 			vi->first_ofld_rxq = ofld_rqidx;
1165 			vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1166 
1167 			ofld_rqidx += vi->nofldrxq;
1168 #endif
1169 #ifdef DEV_NETMAP
1170 			if (j > 0) {
1171 				vi->first_nm_rxq = nm_rqidx;
1172 				vi->first_nm_txq = nm_tqidx;
1173 				vi->nnmrxq = iaq.nnmrxq_vi;
1174 				vi->nnmtxq = iaq.nnmtxq_vi;
1175 				nm_rqidx += vi->nnmrxq;
1176 				nm_tqidx += vi->nnmtxq;
1177 			}
1178 #endif
1179 		}
1180 	}
1181 
1182 	rc = t4_setup_intr_handlers(sc);
1183 	if (rc != 0) {
1184 		device_printf(dev,
1185 		    "failed to setup interrupt handlers: %d\n", rc);
1186 		goto done;
1187 	}
1188 
1189 	rc = bus_generic_probe(dev);
1190 	if (rc != 0) {
1191 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1192 		goto done;
1193 	}
1194 
1195 	/*
1196 	 * Ensure thread-safe mailbox access (in debug builds).
1197 	 *
1198 	 * So far this was the only thread accessing the mailbox but various
1199 	 * ifnets and sysctls are about to be created and their handlers/ioctls
1200 	 * will access the mailbox from different threads.
1201 	 */
1202 	sc->flags |= CHK_MBOX_ACCESS;
1203 
1204 	rc = bus_generic_attach(dev);
1205 	if (rc != 0) {
1206 		device_printf(dev,
1207 		    "failed to attach all child ports: %d\n", rc);
1208 		goto done;
1209 	}
1210 
1211 	device_printf(dev,
1212 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1213 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1214 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1215 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1216 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1217 
1218 	t4_set_desc(sc);
1219 
1220 	notify_siblings(dev, 0);
1221 
1222 done:
1223 	if (rc != 0 && sc->cdev) {
1224 		/* cdev was created and so cxgbetool works; recover that way. */
1225 		device_printf(dev,
1226 		    "error during attach, adapter is now in recovery mode.\n");
1227 		rc = 0;
1228 	}
1229 
1230 	if (rc != 0)
1231 		t4_detach_common(dev);
1232 	else
1233 		t4_sysctls(sc);
1234 
1235 	return (rc);
1236 }
1237 
1238 static int
1239 t4_ready(device_t dev)
1240 {
1241 	struct adapter *sc;
1242 
1243 	sc = device_get_softc(dev);
1244 	if (sc->flags & FW_OK)
1245 		return (0);
1246 	return (ENXIO);
1247 }
1248 
1249 static int
1250 t4_read_port_device(device_t dev, int port, device_t *child)
1251 {
1252 	struct adapter *sc;
1253 	struct port_info *pi;
1254 
1255 	sc = device_get_softc(dev);
1256 	if (port < 0 || port >= MAX_NPORTS)
1257 		return (EINVAL);
1258 	pi = sc->port[port];
1259 	if (pi == NULL || pi->dev == NULL)
1260 		return (ENXIO);
1261 	*child = pi->dev;
1262 	return (0);
1263 }
1264 
1265 static int
1266 notify_siblings(device_t dev, int detaching)
1267 {
1268 	device_t sibling;
1269 	int error, i;
1270 
1271 	error = 0;
1272 	for (i = 0; i < PCI_FUNCMAX; i++) {
1273 		if (i == pci_get_function(dev))
1274 			continue;
1275 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1276 		    pci_get_slot(dev), i);
1277 		if (sibling == NULL || !device_is_attached(sibling))
1278 			continue;
1279 		if (detaching)
1280 			error = T4_DETACH_CHILD(sibling);
1281 		else
1282 			(void)T4_ATTACH_CHILD(sibling);
1283 		if (error)
1284 			break;
1285 	}
1286 	return (error);
1287 }
1288 
1289 /*
1290  * Idempotent
1291  */
1292 static int
1293 t4_detach(device_t dev)
1294 {
1295 	struct adapter *sc;
1296 	int rc;
1297 
1298 	sc = device_get_softc(dev);
1299 
1300 	rc = notify_siblings(dev, 1);
1301 	if (rc) {
1302 		device_printf(dev,
1303 		    "failed to detach sibling devices: %d\n", rc);
1304 		return (rc);
1305 	}
1306 
1307 	return (t4_detach_common(dev));
1308 }
1309 
1310 int
1311 t4_detach_common(device_t dev)
1312 {
1313 	struct adapter *sc;
1314 	struct port_info *pi;
1315 	int i, rc;
1316 
1317 	sc = device_get_softc(dev);
1318 
1319 	if (sc->cdev) {
1320 		destroy_dev(sc->cdev);
1321 		sc->cdev = NULL;
1322 	}
1323 
1324 	sc->flags &= ~CHK_MBOX_ACCESS;
1325 	if (sc->flags & FULL_INIT_DONE) {
1326 		if (!(sc->flags & IS_VF))
1327 			t4_intr_disable(sc);
1328 	}
1329 
1330 	if (device_is_attached(dev)) {
1331 		rc = bus_generic_detach(dev);
1332 		if (rc) {
1333 			device_printf(dev,
1334 			    "failed to detach child devices: %d\n", rc);
1335 			return (rc);
1336 		}
1337 	}
1338 
1339 	for (i = 0; i < sc->intr_count; i++)
1340 		t4_free_irq(sc, &sc->irq[i]);
1341 
1342 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1343 		t4_free_tx_sched(sc);
1344 
1345 	for (i = 0; i < MAX_NPORTS; i++) {
1346 		pi = sc->port[i];
1347 		if (pi) {
1348 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1349 			if (pi->dev)
1350 				device_delete_child(dev, pi->dev);
1351 
1352 			mtx_destroy(&pi->pi_lock);
1353 			free(pi->vi, M_CXGBE);
1354 			free(pi, M_CXGBE);
1355 		}
1356 	}
1357 
1358 	device_delete_children(dev);
1359 
1360 	if (sc->flags & FULL_INIT_DONE)
1361 		adapter_full_uninit(sc);
1362 
1363 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1364 		t4_fw_bye(sc, sc->mbox);
1365 
1366 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1367 		pci_release_msi(dev);
1368 
1369 	if (sc->regs_res)
1370 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1371 		    sc->regs_res);
1372 
1373 	if (sc->udbs_res)
1374 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1375 		    sc->udbs_res);
1376 
1377 	if (sc->msix_res)
1378 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1379 		    sc->msix_res);
1380 
1381 	if (sc->l2t)
1382 		t4_free_l2t(sc->l2t);
1383 	if (sc->smt)
1384 		t4_free_smt(sc->smt);
1385 #ifdef RATELIMIT
1386 	t4_free_etid_table(sc);
1387 #endif
1388 
1389 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1390 	free(sc->sge.ofld_txq, M_CXGBE);
1391 #endif
1392 #ifdef TCP_OFFLOAD
1393 	free(sc->sge.ofld_rxq, M_CXGBE);
1394 #endif
1395 #ifdef DEV_NETMAP
1396 	free(sc->sge.nm_rxq, M_CXGBE);
1397 	free(sc->sge.nm_txq, M_CXGBE);
1398 #endif
1399 	free(sc->irq, M_CXGBE);
1400 	free(sc->sge.rxq, M_CXGBE);
1401 	free(sc->sge.txq, M_CXGBE);
1402 	free(sc->sge.ctrlq, M_CXGBE);
1403 	free(sc->sge.iqmap, M_CXGBE);
1404 	free(sc->sge.eqmap, M_CXGBE);
1405 	free(sc->tids.ftid_tab, M_CXGBE);
1406 	free(sc->tids.hpftid_tab, M_CXGBE);
1407 	free_hftid_hash(&sc->tids);
1408 	free(sc->tids.atid_tab, M_CXGBE);
1409 	free(sc->tids.tid_tab, M_CXGBE);
1410 	free(sc->tt.tls_rx_ports, M_CXGBE);
1411 	t4_destroy_dma_tag(sc);
1412 	if (mtx_initialized(&sc->sc_lock)) {
1413 		sx_xlock(&t4_list_lock);
1414 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1415 		sx_xunlock(&t4_list_lock);
1416 		mtx_destroy(&sc->sc_lock);
1417 	}
1418 
1419 	callout_drain(&sc->sfl_callout);
1420 	if (mtx_initialized(&sc->tids.ftid_lock)) {
1421 		mtx_destroy(&sc->tids.ftid_lock);
1422 		cv_destroy(&sc->tids.ftid_cv);
1423 	}
1424 	if (mtx_initialized(&sc->tids.atid_lock))
1425 		mtx_destroy(&sc->tids.atid_lock);
1426 	if (mtx_initialized(&sc->sfl_lock))
1427 		mtx_destroy(&sc->sfl_lock);
1428 	if (mtx_initialized(&sc->ifp_lock))
1429 		mtx_destroy(&sc->ifp_lock);
1430 	if (mtx_initialized(&sc->reg_lock))
1431 		mtx_destroy(&sc->reg_lock);
1432 
1433 	if (rw_initialized(&sc->policy_lock)) {
1434 		rw_destroy(&sc->policy_lock);
1435 #ifdef TCP_OFFLOAD
1436 		if (sc->policy != NULL)
1437 			free_offload_policy(sc->policy);
1438 #endif
1439 	}
1440 
1441 	for (i = 0; i < NUM_MEMWIN; i++) {
1442 		struct memwin *mw = &sc->memwin[i];
1443 
1444 		if (rw_initialized(&mw->mw_lock))
1445 			rw_destroy(&mw->mw_lock);
1446 	}
1447 
1448 	bzero(sc, sizeof(*sc));
1449 
1450 	return (0);
1451 }
1452 
1453 static int
1454 cxgbe_probe(device_t dev)
1455 {
1456 	char buf[128];
1457 	struct port_info *pi = device_get_softc(dev);
1458 
1459 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1460 	device_set_desc_copy(dev, buf);
1461 
1462 	return (BUS_PROBE_DEFAULT);
1463 }
1464 
1465 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1466     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1467     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS | \
1468     IFCAP_HWRXTSTMP)
1469 #define T4_CAP_ENABLE (T4_CAP)
1470 
1471 static int
1472 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1473 {
1474 	struct ifnet *ifp;
1475 	struct sbuf *sb;
1476 
1477 	vi->xact_addr_filt = -1;
1478 	callout_init(&vi->tick, 1);
1479 
1480 	/* Allocate an ifnet and set it up */
1481 	ifp = if_alloc(IFT_ETHER);
1482 	if (ifp == NULL) {
1483 		device_printf(dev, "Cannot allocate ifnet\n");
1484 		return (ENOMEM);
1485 	}
1486 	vi->ifp = ifp;
1487 	ifp->if_softc = vi;
1488 
1489 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1490 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1491 
1492 	ifp->if_init = cxgbe_init;
1493 	ifp->if_ioctl = cxgbe_ioctl;
1494 	ifp->if_transmit = cxgbe_transmit;
1495 	ifp->if_qflush = cxgbe_qflush;
1496 	ifp->if_get_counter = cxgbe_get_counter;
1497 #ifdef RATELIMIT
1498 	ifp->if_snd_tag_alloc = cxgbe_snd_tag_alloc;
1499 	ifp->if_snd_tag_modify = cxgbe_snd_tag_modify;
1500 	ifp->if_snd_tag_query = cxgbe_snd_tag_query;
1501 	ifp->if_snd_tag_free = cxgbe_snd_tag_free;
1502 #endif
1503 
1504 	ifp->if_capabilities = T4_CAP;
1505 #ifdef TCP_OFFLOAD
1506 	if (vi->nofldrxq != 0)
1507 		ifp->if_capabilities |= IFCAP_TOE;
1508 #endif
1509 #ifdef DEV_NETMAP
1510 	if (vi->nnmrxq != 0)
1511 		ifp->if_capabilities |= IFCAP_NETMAP;
1512 #endif
1513 #ifdef RATELIMIT
1514 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0)
1515 		ifp->if_capabilities |= IFCAP_TXRTLMT;
1516 #endif
1517 	ifp->if_capenable = T4_CAP_ENABLE;
1518 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1519 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1520 
1521 	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1522 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1523 	ifp->if_hw_tsomaxsegsize = 65536;
1524 
1525 	ether_ifattach(ifp, vi->hw_addr);
1526 #ifdef DEV_NETMAP
1527 	if (ifp->if_capabilities & IFCAP_NETMAP)
1528 		cxgbe_nm_attach(vi);
1529 #endif
1530 	sb = sbuf_new_auto();
1531 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1532 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1533 	switch (ifp->if_capabilities & (IFCAP_TOE | IFCAP_TXRTLMT)) {
1534 	case IFCAP_TOE:
1535 		sbuf_printf(sb, "; %d txq (TOE)", vi->nofldtxq);
1536 		break;
1537 	case IFCAP_TOE | IFCAP_TXRTLMT:
1538 		sbuf_printf(sb, "; %d txq (TOE/ETHOFLD)", vi->nofldtxq);
1539 		break;
1540 	case IFCAP_TXRTLMT:
1541 		sbuf_printf(sb, "; %d txq (ETHOFLD)", vi->nofldtxq);
1542 		break;
1543 	}
1544 #endif
1545 #ifdef TCP_OFFLOAD
1546 	if (ifp->if_capabilities & IFCAP_TOE)
1547 		sbuf_printf(sb, ", %d rxq (TOE)", vi->nofldrxq);
1548 #endif
1549 #ifdef DEV_NETMAP
1550 	if (ifp->if_capabilities & IFCAP_NETMAP)
1551 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1552 		    vi->nnmtxq, vi->nnmrxq);
1553 #endif
1554 	sbuf_finish(sb);
1555 	device_printf(dev, "%s\n", sbuf_data(sb));
1556 	sbuf_delete(sb);
1557 
1558 	vi_sysctls(vi);
1559 
1560 	return (0);
1561 }
1562 
1563 static int
1564 cxgbe_attach(device_t dev)
1565 {
1566 	struct port_info *pi = device_get_softc(dev);
1567 	struct adapter *sc = pi->adapter;
1568 	struct vi_info *vi;
1569 	int i, rc;
1570 
1571 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1572 
1573 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1574 	if (rc)
1575 		return (rc);
1576 
1577 	for_each_vi(pi, i, vi) {
1578 		if (i == 0)
1579 			continue;
1580 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1581 		if (vi->dev == NULL) {
1582 			device_printf(dev, "failed to add VI %d\n", i);
1583 			continue;
1584 		}
1585 		device_set_softc(vi->dev, vi);
1586 	}
1587 
1588 	cxgbe_sysctls(pi);
1589 
1590 	bus_generic_attach(dev);
1591 
1592 	return (0);
1593 }
1594 
1595 static void
1596 cxgbe_vi_detach(struct vi_info *vi)
1597 {
1598 	struct ifnet *ifp = vi->ifp;
1599 
1600 	ether_ifdetach(ifp);
1601 
1602 	/* Let detach proceed even if these fail. */
1603 #ifdef DEV_NETMAP
1604 	if (ifp->if_capabilities & IFCAP_NETMAP)
1605 		cxgbe_nm_detach(vi);
1606 #endif
1607 	cxgbe_uninit_synchronized(vi);
1608 	callout_drain(&vi->tick);
1609 	vi_full_uninit(vi);
1610 
1611 	if_free(vi->ifp);
1612 	vi->ifp = NULL;
1613 }
1614 
1615 static int
1616 cxgbe_detach(device_t dev)
1617 {
1618 	struct port_info *pi = device_get_softc(dev);
1619 	struct adapter *sc = pi->adapter;
1620 	int rc;
1621 
1622 	/* Detach the extra VIs first. */
1623 	rc = bus_generic_detach(dev);
1624 	if (rc)
1625 		return (rc);
1626 	device_delete_children(dev);
1627 
1628 	doom_vi(sc, &pi->vi[0]);
1629 
1630 	if (pi->flags & HAS_TRACEQ) {
1631 		sc->traceq = -1;	/* cloner should not create ifnet */
1632 		t4_tracer_port_detach(sc);
1633 	}
1634 
1635 	cxgbe_vi_detach(&pi->vi[0]);
1636 	callout_drain(&pi->tick);
1637 	ifmedia_removeall(&pi->media);
1638 
1639 	end_synchronized_op(sc, 0);
1640 
1641 	return (0);
1642 }
1643 
1644 static void
1645 cxgbe_init(void *arg)
1646 {
1647 	struct vi_info *vi = arg;
1648 	struct adapter *sc = vi->pi->adapter;
1649 
1650 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1651 		return;
1652 	cxgbe_init_synchronized(vi);
1653 	end_synchronized_op(sc, 0);
1654 }
1655 
1656 static int
1657 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1658 {
1659 	int rc = 0, mtu, flags;
1660 	struct vi_info *vi = ifp->if_softc;
1661 	struct port_info *pi = vi->pi;
1662 	struct adapter *sc = pi->adapter;
1663 	struct ifreq *ifr = (struct ifreq *)data;
1664 	uint32_t mask;
1665 
1666 	switch (cmd) {
1667 	case SIOCSIFMTU:
1668 		mtu = ifr->ifr_mtu;
1669 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1670 			return (EINVAL);
1671 
1672 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1673 		if (rc)
1674 			return (rc);
1675 		ifp->if_mtu = mtu;
1676 		if (vi->flags & VI_INIT_DONE) {
1677 			t4_update_fl_bufsize(ifp);
1678 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1679 				rc = update_mac_settings(ifp, XGMAC_MTU);
1680 		}
1681 		end_synchronized_op(sc, 0);
1682 		break;
1683 
1684 	case SIOCSIFFLAGS:
1685 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4flg");
1686 		if (rc)
1687 			return (rc);
1688 
1689 		if (ifp->if_flags & IFF_UP) {
1690 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1691 				flags = vi->if_flags;
1692 				if ((ifp->if_flags ^ flags) &
1693 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1694 					rc = update_mac_settings(ifp,
1695 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1696 				}
1697 			} else {
1698 				rc = cxgbe_init_synchronized(vi);
1699 			}
1700 			vi->if_flags = ifp->if_flags;
1701 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1702 			rc = cxgbe_uninit_synchronized(vi);
1703 		}
1704 		end_synchronized_op(sc, 0);
1705 		break;
1706 
1707 	case SIOCADDMULTI:
1708 	case SIOCDELMULTI:
1709 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4multi");
1710 		if (rc)
1711 			return (rc);
1712 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1713 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1714 		end_synchronized_op(sc, 0);
1715 		break;
1716 
1717 	case SIOCSIFCAP:
1718 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1719 		if (rc)
1720 			return (rc);
1721 
1722 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1723 		if (mask & IFCAP_TXCSUM) {
1724 			ifp->if_capenable ^= IFCAP_TXCSUM;
1725 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1726 
1727 			if (IFCAP_TSO4 & ifp->if_capenable &&
1728 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1729 				ifp->if_capenable &= ~IFCAP_TSO4;
1730 				if_printf(ifp,
1731 				    "tso4 disabled due to -txcsum.\n");
1732 			}
1733 		}
1734 		if (mask & IFCAP_TXCSUM_IPV6) {
1735 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1736 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1737 
1738 			if (IFCAP_TSO6 & ifp->if_capenable &&
1739 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1740 				ifp->if_capenable &= ~IFCAP_TSO6;
1741 				if_printf(ifp,
1742 				    "tso6 disabled due to -txcsum6.\n");
1743 			}
1744 		}
1745 		if (mask & IFCAP_RXCSUM)
1746 			ifp->if_capenable ^= IFCAP_RXCSUM;
1747 		if (mask & IFCAP_RXCSUM_IPV6)
1748 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1749 
1750 		/*
1751 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1752 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1753 		 * sending a TSO request our way, so it's sufficient to toggle
1754 		 * IFCAP_TSOx only.
1755 		 */
1756 		if (mask & IFCAP_TSO4) {
1757 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1758 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1759 				if_printf(ifp, "enable txcsum first.\n");
1760 				rc = EAGAIN;
1761 				goto fail;
1762 			}
1763 			ifp->if_capenable ^= IFCAP_TSO4;
1764 		}
1765 		if (mask & IFCAP_TSO6) {
1766 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1767 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1768 				if_printf(ifp, "enable txcsum6 first.\n");
1769 				rc = EAGAIN;
1770 				goto fail;
1771 			}
1772 			ifp->if_capenable ^= IFCAP_TSO6;
1773 		}
1774 		if (mask & IFCAP_LRO) {
1775 #if defined(INET) || defined(INET6)
1776 			int i;
1777 			struct sge_rxq *rxq;
1778 
1779 			ifp->if_capenable ^= IFCAP_LRO;
1780 			for_each_rxq(vi, i, rxq) {
1781 				if (ifp->if_capenable & IFCAP_LRO)
1782 					rxq->iq.flags |= IQ_LRO_ENABLED;
1783 				else
1784 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1785 			}
1786 #endif
1787 		}
1788 #ifdef TCP_OFFLOAD
1789 		if (mask & IFCAP_TOE) {
1790 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1791 
1792 			rc = toe_capability(vi, enable);
1793 			if (rc != 0)
1794 				goto fail;
1795 
1796 			ifp->if_capenable ^= mask;
1797 		}
1798 #endif
1799 		if (mask & IFCAP_VLAN_HWTAGGING) {
1800 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1801 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1802 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1803 		}
1804 		if (mask & IFCAP_VLAN_MTU) {
1805 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1806 
1807 			/* Need to find out how to disable auto-mtu-inflation */
1808 		}
1809 		if (mask & IFCAP_VLAN_HWTSO)
1810 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1811 		if (mask & IFCAP_VLAN_HWCSUM)
1812 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1813 #ifdef RATELIMIT
1814 		if (mask & IFCAP_TXRTLMT)
1815 			ifp->if_capenable ^= IFCAP_TXRTLMT;
1816 #endif
1817 		if (mask & IFCAP_HWRXTSTMP) {
1818 			int i;
1819 			struct sge_rxq *rxq;
1820 
1821 			ifp->if_capenable ^= IFCAP_HWRXTSTMP;
1822 			for_each_rxq(vi, i, rxq) {
1823 				if (ifp->if_capenable & IFCAP_HWRXTSTMP)
1824 					rxq->iq.flags |= IQ_RX_TIMESTAMP;
1825 				else
1826 					rxq->iq.flags &= ~IQ_RX_TIMESTAMP;
1827 			}
1828 		}
1829 
1830 #ifdef VLAN_CAPABILITIES
1831 		VLAN_CAPABILITIES(ifp);
1832 #endif
1833 fail:
1834 		end_synchronized_op(sc, 0);
1835 		break;
1836 
1837 	case SIOCSIFMEDIA:
1838 	case SIOCGIFMEDIA:
1839 	case SIOCGIFXMEDIA:
1840 		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1841 		break;
1842 
1843 	case SIOCGI2C: {
1844 		struct ifi2creq i2c;
1845 
1846 		rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
1847 		if (rc != 0)
1848 			break;
1849 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1850 			rc = EPERM;
1851 			break;
1852 		}
1853 		if (i2c.len > sizeof(i2c.data)) {
1854 			rc = EINVAL;
1855 			break;
1856 		}
1857 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1858 		if (rc)
1859 			return (rc);
1860 		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
1861 		    i2c.offset, i2c.len, &i2c.data[0]);
1862 		end_synchronized_op(sc, 0);
1863 		if (rc == 0)
1864 			rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
1865 		break;
1866 	}
1867 
1868 	default:
1869 		rc = ether_ioctl(ifp, cmd, data);
1870 	}
1871 
1872 	return (rc);
1873 }
1874 
1875 static int
1876 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1877 {
1878 	struct vi_info *vi = ifp->if_softc;
1879 	struct port_info *pi = vi->pi;
1880 	struct adapter *sc = pi->adapter;
1881 	struct sge_txq *txq;
1882 	void *items[1];
1883 	int rc;
1884 
1885 	M_ASSERTPKTHDR(m);
1886 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1887 
1888 	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1889 		m_freem(m);
1890 		return (ENETDOWN);
1891 	}
1892 
1893 	rc = parse_pkt(sc, &m);
1894 	if (__predict_false(rc != 0)) {
1895 		MPASS(m == NULL);			/* was freed already */
1896 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1897 		return (rc);
1898 	}
1899 #ifdef RATELIMIT
1900 	if (m->m_pkthdr.snd_tag != NULL) {
1901 		/* EAGAIN tells the stack we are not the correct interface. */
1902 		if (__predict_false(ifp != m->m_pkthdr.snd_tag->ifp)) {
1903 			m_freem(m);
1904 			return (EAGAIN);
1905 		}
1906 
1907 		return (ethofld_transmit(ifp, m));
1908 	}
1909 #endif
1910 
1911 	/* Select a txq. */
1912 	txq = &sc->sge.txq[vi->first_txq];
1913 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1914 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1915 		    vi->rsrv_noflowq);
1916 
1917 	items[0] = m;
1918 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1919 	if (__predict_false(rc != 0))
1920 		m_freem(m);
1921 
1922 	return (rc);
1923 }
1924 
1925 static void
1926 cxgbe_qflush(struct ifnet *ifp)
1927 {
1928 	struct vi_info *vi = ifp->if_softc;
1929 	struct sge_txq *txq;
1930 	int i;
1931 
1932 	/* queues do not exist if !VI_INIT_DONE. */
1933 	if (vi->flags & VI_INIT_DONE) {
1934 		for_each_txq(vi, i, txq) {
1935 			TXQ_LOCK(txq);
1936 			txq->eq.flags |= EQ_QFLUSH;
1937 			TXQ_UNLOCK(txq);
1938 			while (!mp_ring_is_idle(txq->r)) {
1939 				mp_ring_check_drainage(txq->r, 0);
1940 				pause("qflush", 1);
1941 			}
1942 			TXQ_LOCK(txq);
1943 			txq->eq.flags &= ~EQ_QFLUSH;
1944 			TXQ_UNLOCK(txq);
1945 		}
1946 	}
1947 	if_qflush(ifp);
1948 }
1949 
1950 static uint64_t
1951 vi_get_counter(struct ifnet *ifp, ift_counter c)
1952 {
1953 	struct vi_info *vi = ifp->if_softc;
1954 	struct fw_vi_stats_vf *s = &vi->stats;
1955 
1956 	vi_refresh_stats(vi->pi->adapter, vi);
1957 
1958 	switch (c) {
1959 	case IFCOUNTER_IPACKETS:
1960 		return (s->rx_bcast_frames + s->rx_mcast_frames +
1961 		    s->rx_ucast_frames);
1962 	case IFCOUNTER_IERRORS:
1963 		return (s->rx_err_frames);
1964 	case IFCOUNTER_OPACKETS:
1965 		return (s->tx_bcast_frames + s->tx_mcast_frames +
1966 		    s->tx_ucast_frames + s->tx_offload_frames);
1967 	case IFCOUNTER_OERRORS:
1968 		return (s->tx_drop_frames);
1969 	case IFCOUNTER_IBYTES:
1970 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
1971 		    s->rx_ucast_bytes);
1972 	case IFCOUNTER_OBYTES:
1973 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
1974 		    s->tx_ucast_bytes + s->tx_offload_bytes);
1975 	case IFCOUNTER_IMCASTS:
1976 		return (s->rx_mcast_frames);
1977 	case IFCOUNTER_OMCASTS:
1978 		return (s->tx_mcast_frames);
1979 	case IFCOUNTER_OQDROPS: {
1980 		uint64_t drops;
1981 
1982 		drops = 0;
1983 		if (vi->flags & VI_INIT_DONE) {
1984 			int i;
1985 			struct sge_txq *txq;
1986 
1987 			for_each_txq(vi, i, txq)
1988 				drops += counter_u64_fetch(txq->r->drops);
1989 		}
1990 
1991 		return (drops);
1992 
1993 	}
1994 
1995 	default:
1996 		return (if_get_counter_default(ifp, c));
1997 	}
1998 }
1999 
2000 uint64_t
2001 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
2002 {
2003 	struct vi_info *vi = ifp->if_softc;
2004 	struct port_info *pi = vi->pi;
2005 	struct adapter *sc = pi->adapter;
2006 	struct port_stats *s = &pi->stats;
2007 
2008 	if (pi->nvi > 1 || sc->flags & IS_VF)
2009 		return (vi_get_counter(ifp, c));
2010 
2011 	cxgbe_refresh_stats(sc, pi);
2012 
2013 	switch (c) {
2014 	case IFCOUNTER_IPACKETS:
2015 		return (s->rx_frames);
2016 
2017 	case IFCOUNTER_IERRORS:
2018 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
2019 		    s->rx_fcs_err + s->rx_len_err);
2020 
2021 	case IFCOUNTER_OPACKETS:
2022 		return (s->tx_frames);
2023 
2024 	case IFCOUNTER_OERRORS:
2025 		return (s->tx_error_frames);
2026 
2027 	case IFCOUNTER_IBYTES:
2028 		return (s->rx_octets);
2029 
2030 	case IFCOUNTER_OBYTES:
2031 		return (s->tx_octets);
2032 
2033 	case IFCOUNTER_IMCASTS:
2034 		return (s->rx_mcast_frames);
2035 
2036 	case IFCOUNTER_OMCASTS:
2037 		return (s->tx_mcast_frames);
2038 
2039 	case IFCOUNTER_IQDROPS:
2040 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
2041 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
2042 		    s->rx_trunc3 + pi->tnl_cong_drops);
2043 
2044 	case IFCOUNTER_OQDROPS: {
2045 		uint64_t drops;
2046 
2047 		drops = s->tx_drop;
2048 		if (vi->flags & VI_INIT_DONE) {
2049 			int i;
2050 			struct sge_txq *txq;
2051 
2052 			for_each_txq(vi, i, txq)
2053 				drops += counter_u64_fetch(txq->r->drops);
2054 		}
2055 
2056 		return (drops);
2057 
2058 	}
2059 
2060 	default:
2061 		return (if_get_counter_default(ifp, c));
2062 	}
2063 }
2064 
2065 /*
2066  * The kernel picks a media from the list we had provided so we do not have to
2067  * validate the request.
2068  */
2069 static int
2070 cxgbe_media_change(struct ifnet *ifp)
2071 {
2072 	struct vi_info *vi = ifp->if_softc;
2073 	struct port_info *pi = vi->pi;
2074 	struct ifmedia *ifm = &pi->media;
2075 	struct link_config *lc = &pi->link_cfg;
2076 	struct adapter *sc = pi->adapter;
2077 	int rc;
2078 
2079 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mec");
2080 	if (rc != 0)
2081 		return (rc);
2082 	PORT_LOCK(pi);
2083 	if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
2084 		MPASS(lc->supported & FW_PORT_CAP_ANEG);
2085 		lc->requested_aneg = AUTONEG_ENABLE;
2086 	} else {
2087 		lc->requested_aneg = AUTONEG_DISABLE;
2088 		lc->requested_speed =
2089 		    ifmedia_baudrate(ifm->ifm_media) / 1000000;
2090 		lc->requested_fc = 0;
2091 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
2092 			lc->requested_fc |= PAUSE_RX;
2093 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
2094 			lc->requested_fc |= PAUSE_TX;
2095 	}
2096 	if (pi->up_vis > 0)
2097 		rc = apply_l1cfg(pi);
2098 	PORT_UNLOCK(pi);
2099 	end_synchronized_op(sc, 0);
2100 	return (rc);
2101 }
2102 
2103 /*
2104  * Mbps to FW_PORT_CAP_SPEED_* bit.
2105  */
2106 static uint16_t
2107 speed_to_fwspeed(int speed)
2108 {
2109 
2110 	switch (speed) {
2111 	case 100000:
2112 		return (FW_PORT_CAP_SPEED_100G);
2113 	case 40000:
2114 		return (FW_PORT_CAP_SPEED_40G);
2115 	case 25000:
2116 		return (FW_PORT_CAP_SPEED_25G);
2117 	case 10000:
2118 		return (FW_PORT_CAP_SPEED_10G);
2119 	case 1000:
2120 		return (FW_PORT_CAP_SPEED_1G);
2121 	case 100:
2122 		return (FW_PORT_CAP_SPEED_100M);
2123 	}
2124 
2125 	return (0);
2126 }
2127 
2128 /*
2129  * Base media word (without ETHER, pause, link active, etc.) for the port at the
2130  * given speed.
2131  */
2132 static int
2133 port_mword(struct port_info *pi, uint16_t speed)
2134 {
2135 
2136 	MPASS(speed & M_FW_PORT_CAP_SPEED);
2137 	MPASS(powerof2(speed));
2138 
2139 	switch(pi->port_type) {
2140 	case FW_PORT_TYPE_BT_SGMII:
2141 	case FW_PORT_TYPE_BT_XFI:
2142 	case FW_PORT_TYPE_BT_XAUI:
2143 		/* BaseT */
2144 		switch (speed) {
2145 		case FW_PORT_CAP_SPEED_100M:
2146 			return (IFM_100_T);
2147 		case FW_PORT_CAP_SPEED_1G:
2148 			return (IFM_1000_T);
2149 		case FW_PORT_CAP_SPEED_10G:
2150 			return (IFM_10G_T);
2151 		}
2152 		break;
2153 	case FW_PORT_TYPE_KX4:
2154 		if (speed == FW_PORT_CAP_SPEED_10G)
2155 			return (IFM_10G_KX4);
2156 		break;
2157 	case FW_PORT_TYPE_CX4:
2158 		if (speed == FW_PORT_CAP_SPEED_10G)
2159 			return (IFM_10G_CX4);
2160 		break;
2161 	case FW_PORT_TYPE_KX:
2162 		if (speed == FW_PORT_CAP_SPEED_1G)
2163 			return (IFM_1000_KX);
2164 		break;
2165 	case FW_PORT_TYPE_KR:
2166 	case FW_PORT_TYPE_BP_AP:
2167 	case FW_PORT_TYPE_BP4_AP:
2168 	case FW_PORT_TYPE_BP40_BA:
2169 	case FW_PORT_TYPE_KR4_100G:
2170 	case FW_PORT_TYPE_KR_SFP28:
2171 	case FW_PORT_TYPE_KR_XLAUI:
2172 		switch (speed) {
2173 		case FW_PORT_CAP_SPEED_1G:
2174 			return (IFM_1000_KX);
2175 		case FW_PORT_CAP_SPEED_10G:
2176 			return (IFM_10G_KR);
2177 		case FW_PORT_CAP_SPEED_25G:
2178 			return (IFM_25G_KR);
2179 		case FW_PORT_CAP_SPEED_40G:
2180 			return (IFM_40G_KR4);
2181 		case FW_PORT_CAP_SPEED_100G:
2182 			return (IFM_100G_KR4);
2183 		}
2184 		break;
2185 	case FW_PORT_TYPE_FIBER_XFI:
2186 	case FW_PORT_TYPE_FIBER_XAUI:
2187 	case FW_PORT_TYPE_SFP:
2188 	case FW_PORT_TYPE_QSFP_10G:
2189 	case FW_PORT_TYPE_QSA:
2190 	case FW_PORT_TYPE_QSFP:
2191 	case FW_PORT_TYPE_CR4_QSFP:
2192 	case FW_PORT_TYPE_CR_QSFP:
2193 	case FW_PORT_TYPE_CR2_QSFP:
2194 	case FW_PORT_TYPE_SFP28:
2195 		/* Pluggable transceiver */
2196 		switch (pi->mod_type) {
2197 		case FW_PORT_MOD_TYPE_LR:
2198 			switch (speed) {
2199 			case FW_PORT_CAP_SPEED_1G:
2200 				return (IFM_1000_LX);
2201 			case FW_PORT_CAP_SPEED_10G:
2202 				return (IFM_10G_LR);
2203 			case FW_PORT_CAP_SPEED_25G:
2204 				return (IFM_25G_LR);
2205 			case FW_PORT_CAP_SPEED_40G:
2206 				return (IFM_40G_LR4);
2207 			case FW_PORT_CAP_SPEED_100G:
2208 				return (IFM_100G_LR4);
2209 			}
2210 			break;
2211 		case FW_PORT_MOD_TYPE_SR:
2212 			switch (speed) {
2213 			case FW_PORT_CAP_SPEED_1G:
2214 				return (IFM_1000_SX);
2215 			case FW_PORT_CAP_SPEED_10G:
2216 				return (IFM_10G_SR);
2217 			case FW_PORT_CAP_SPEED_25G:
2218 				return (IFM_25G_SR);
2219 			case FW_PORT_CAP_SPEED_40G:
2220 				return (IFM_40G_SR4);
2221 			case FW_PORT_CAP_SPEED_100G:
2222 				return (IFM_100G_SR4);
2223 			}
2224 			break;
2225 		case FW_PORT_MOD_TYPE_ER:
2226 			if (speed == FW_PORT_CAP_SPEED_10G)
2227 				return (IFM_10G_ER);
2228 			break;
2229 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2230 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2231 			switch (speed) {
2232 			case FW_PORT_CAP_SPEED_1G:
2233 				return (IFM_1000_CX);
2234 			case FW_PORT_CAP_SPEED_10G:
2235 				return (IFM_10G_TWINAX);
2236 			case FW_PORT_CAP_SPEED_25G:
2237 				return (IFM_25G_CR);
2238 			case FW_PORT_CAP_SPEED_40G:
2239 				return (IFM_40G_CR4);
2240 			case FW_PORT_CAP_SPEED_100G:
2241 				return (IFM_100G_CR4);
2242 			}
2243 			break;
2244 		case FW_PORT_MOD_TYPE_LRM:
2245 			if (speed == FW_PORT_CAP_SPEED_10G)
2246 				return (IFM_10G_LRM);
2247 			break;
2248 		case FW_PORT_MOD_TYPE_NA:
2249 			MPASS(0);	/* Not pluggable? */
2250 			/* fall throough */
2251 		case FW_PORT_MOD_TYPE_ERROR:
2252 		case FW_PORT_MOD_TYPE_UNKNOWN:
2253 		case FW_PORT_MOD_TYPE_NOTSUPPORTED:
2254 			break;
2255 		case FW_PORT_MOD_TYPE_NONE:
2256 			return (IFM_NONE);
2257 		}
2258 		break;
2259 	case FW_PORT_TYPE_NONE:
2260 		return (IFM_NONE);
2261 	}
2262 
2263 	return (IFM_UNKNOWN);
2264 }
2265 
2266 static void
2267 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2268 {
2269 	struct vi_info *vi = ifp->if_softc;
2270 	struct port_info *pi = vi->pi;
2271 	struct adapter *sc = pi->adapter;
2272 	struct link_config *lc = &pi->link_cfg;
2273 
2274 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4med") != 0)
2275 		return;
2276 	PORT_LOCK(pi);
2277 
2278 	if (pi->up_vis == 0) {
2279 		/*
2280 		 * If all the interfaces are administratively down the firmware
2281 		 * does not report transceiver changes.  Refresh port info here
2282 		 * so that ifconfig displays accurate ifmedia at all times.
2283 		 * This is the only reason we have a synchronized op in this
2284 		 * function.  Just PORT_LOCK would have been enough otherwise.
2285 		 */
2286 		t4_update_port_info(pi);
2287 		build_medialist(pi, &pi->media);
2288 	}
2289 
2290 	/* ifm_status */
2291 	ifmr->ifm_status = IFM_AVALID;
2292 	if (lc->link_ok == 0)
2293 		goto done;
2294 	ifmr->ifm_status |= IFM_ACTIVE;
2295 
2296 	/* ifm_active */
2297 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2298 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2299 	if (lc->fc & PAUSE_RX)
2300 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2301 	if (lc->fc & PAUSE_TX)
2302 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2303 	ifmr->ifm_active |= port_mword(pi, speed_to_fwspeed(lc->speed));
2304 done:
2305 	PORT_UNLOCK(pi);
2306 	end_synchronized_op(sc, 0);
2307 }
2308 
2309 static int
2310 vcxgbe_probe(device_t dev)
2311 {
2312 	char buf[128];
2313 	struct vi_info *vi = device_get_softc(dev);
2314 
2315 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2316 	    vi - vi->pi->vi);
2317 	device_set_desc_copy(dev, buf);
2318 
2319 	return (BUS_PROBE_DEFAULT);
2320 }
2321 
2322 static int
2323 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2324 {
2325 	int func, index, rc;
2326 	uint32_t param, val;
2327 
2328 	ASSERT_SYNCHRONIZED_OP(sc);
2329 
2330 	index = vi - pi->vi;
2331 	MPASS(index > 0);	/* This function deals with _extra_ VIs only */
2332 	KASSERT(index < nitems(vi_mac_funcs),
2333 	    ("%s: VI %s doesn't have a MAC func", __func__,
2334 	    device_get_nameunit(vi->dev)));
2335 	func = vi_mac_funcs[index];
2336 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2337 	    vi->hw_addr, &vi->rss_size, func, 0);
2338 	if (rc < 0) {
2339 		device_printf(vi->dev, "failed to allocate virtual interface %d"
2340 		    "for port %d: %d\n", index, pi->port_id, -rc);
2341 		return (-rc);
2342 	}
2343 	vi->viid = rc;
2344 	if (chip_id(sc) <= CHELSIO_T5)
2345 		vi->smt_idx = (rc & 0x7f) << 1;
2346 	else
2347 		vi->smt_idx = (rc & 0x7f);
2348 
2349 	if (vi->rss_size == 1) {
2350 		/*
2351 		 * This VI didn't get a slice of the RSS table.  Reduce the
2352 		 * number of VIs being created (hw.cxgbe.num_vis) or modify the
2353 		 * configuration file (nvi, rssnvi for this PF) if this is a
2354 		 * problem.
2355 		 */
2356 		device_printf(vi->dev, "RSS table not available.\n");
2357 		vi->rss_base = 0xffff;
2358 
2359 		return (0);
2360 	}
2361 
2362 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2363 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2364 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2365 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2366 	if (rc)
2367 		vi->rss_base = 0xffff;
2368 	else {
2369 		MPASS((val >> 16) == vi->rss_size);
2370 		vi->rss_base = val & 0xffff;
2371 	}
2372 
2373 	return (0);
2374 }
2375 
2376 static int
2377 vcxgbe_attach(device_t dev)
2378 {
2379 	struct vi_info *vi;
2380 	struct port_info *pi;
2381 	struct adapter *sc;
2382 	int rc;
2383 
2384 	vi = device_get_softc(dev);
2385 	pi = vi->pi;
2386 	sc = pi->adapter;
2387 
2388 	rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2389 	if (rc)
2390 		return (rc);
2391 	rc = alloc_extra_vi(sc, pi, vi);
2392 	end_synchronized_op(sc, 0);
2393 	if (rc)
2394 		return (rc);
2395 
2396 	rc = cxgbe_vi_attach(dev, vi);
2397 	if (rc) {
2398 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2399 		return (rc);
2400 	}
2401 	return (0);
2402 }
2403 
2404 static int
2405 vcxgbe_detach(device_t dev)
2406 {
2407 	struct vi_info *vi;
2408 	struct adapter *sc;
2409 
2410 	vi = device_get_softc(dev);
2411 	sc = vi->pi->adapter;
2412 
2413 	doom_vi(sc, vi);
2414 
2415 	cxgbe_vi_detach(vi);
2416 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2417 
2418 	end_synchronized_op(sc, 0);
2419 
2420 	return (0);
2421 }
2422 
2423 void
2424 t4_fatal_err(struct adapter *sc)
2425 {
2426 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2427 	t4_intr_disable(sc);
2428 	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2429 	    device_get_nameunit(sc->dev));
2430 	if (t4_panic_on_fatal_err)
2431 		panic("panic requested on fatal error");
2432 }
2433 
2434 void
2435 t4_add_adapter(struct adapter *sc)
2436 {
2437 	sx_xlock(&t4_list_lock);
2438 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2439 	sx_xunlock(&t4_list_lock);
2440 }
2441 
2442 int
2443 t4_map_bars_0_and_4(struct adapter *sc)
2444 {
2445 	sc->regs_rid = PCIR_BAR(0);
2446 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2447 	    &sc->regs_rid, RF_ACTIVE);
2448 	if (sc->regs_res == NULL) {
2449 		device_printf(sc->dev, "cannot map registers.\n");
2450 		return (ENXIO);
2451 	}
2452 	sc->bt = rman_get_bustag(sc->regs_res);
2453 	sc->bh = rman_get_bushandle(sc->regs_res);
2454 	sc->mmio_len = rman_get_size(sc->regs_res);
2455 	setbit(&sc->doorbells, DOORBELL_KDB);
2456 
2457 	sc->msix_rid = PCIR_BAR(4);
2458 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2459 	    &sc->msix_rid, RF_ACTIVE);
2460 	if (sc->msix_res == NULL) {
2461 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2462 		return (ENXIO);
2463 	}
2464 
2465 	return (0);
2466 }
2467 
2468 int
2469 t4_map_bar_2(struct adapter *sc)
2470 {
2471 
2472 	/*
2473 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2474 	 * to map it if RDMA is disabled.
2475 	 */
2476 	if (is_t4(sc) && sc->rdmacaps == 0)
2477 		return (0);
2478 
2479 	sc->udbs_rid = PCIR_BAR(2);
2480 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2481 	    &sc->udbs_rid, RF_ACTIVE);
2482 	if (sc->udbs_res == NULL) {
2483 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2484 		return (ENXIO);
2485 	}
2486 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2487 
2488 	if (chip_id(sc) >= CHELSIO_T5) {
2489 		setbit(&sc->doorbells, DOORBELL_UDB);
2490 #if defined(__i386__) || defined(__amd64__)
2491 		if (t5_write_combine) {
2492 			int rc, mode;
2493 
2494 			/*
2495 			 * Enable write combining on BAR2.  This is the
2496 			 * userspace doorbell BAR and is split into 128B
2497 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2498 			 * with an egress queue.  The first 64B has the doorbell
2499 			 * and the second 64B can be used to submit a tx work
2500 			 * request with an implicit doorbell.
2501 			 */
2502 
2503 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2504 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2505 			if (rc == 0) {
2506 				clrbit(&sc->doorbells, DOORBELL_UDB);
2507 				setbit(&sc->doorbells, DOORBELL_WCWR);
2508 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2509 			} else {
2510 				device_printf(sc->dev,
2511 				    "couldn't enable write combining: %d\n",
2512 				    rc);
2513 			}
2514 
2515 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2516 			t4_write_reg(sc, A_SGE_STAT_CFG,
2517 			    V_STATSOURCE_T5(7) | mode);
2518 		}
2519 #endif
2520 	}
2521 	sc->iwt.wc_en = isset(&sc->doorbells, DOORBELL_UDBWC) ? 1 : 0;
2522 
2523 	return (0);
2524 }
2525 
2526 struct memwin_init {
2527 	uint32_t base;
2528 	uint32_t aperture;
2529 };
2530 
2531 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2532 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2533 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2534 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2535 };
2536 
2537 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2538 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2539 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2540 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2541 };
2542 
2543 static void
2544 setup_memwin(struct adapter *sc)
2545 {
2546 	const struct memwin_init *mw_init;
2547 	struct memwin *mw;
2548 	int i;
2549 	uint32_t bar0;
2550 
2551 	if (is_t4(sc)) {
2552 		/*
2553 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2554 		 * mechanism.  Works from within PCI passthrough environments
2555 		 * too, where rman_get_start() can return a different value.  We
2556 		 * need to program the T4 memory window decoders with the actual
2557 		 * addresses that will be coming across the PCIe link.
2558 		 */
2559 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2560 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2561 
2562 		mw_init = &t4_memwin[0];
2563 	} else {
2564 		/* T5+ use the relative offset inside the PCIe BAR */
2565 		bar0 = 0;
2566 
2567 		mw_init = &t5_memwin[0];
2568 	}
2569 
2570 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2571 		rw_init(&mw->mw_lock, "memory window access");
2572 		mw->mw_base = mw_init->base;
2573 		mw->mw_aperture = mw_init->aperture;
2574 		mw->mw_curpos = 0;
2575 		t4_write_reg(sc,
2576 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2577 		    (mw->mw_base + bar0) | V_BIR(0) |
2578 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2579 		rw_wlock(&mw->mw_lock);
2580 		position_memwin(sc, i, 0);
2581 		rw_wunlock(&mw->mw_lock);
2582 	}
2583 
2584 	/* flush */
2585 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2586 }
2587 
2588 /*
2589  * Positions the memory window at the given address in the card's address space.
2590  * There are some alignment requirements and the actual position may be at an
2591  * address prior to the requested address.  mw->mw_curpos always has the actual
2592  * position of the window.
2593  */
2594 static void
2595 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2596 {
2597 	struct memwin *mw;
2598 	uint32_t pf;
2599 	uint32_t reg;
2600 
2601 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2602 	mw = &sc->memwin[idx];
2603 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2604 
2605 	if (is_t4(sc)) {
2606 		pf = 0;
2607 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2608 	} else {
2609 		pf = V_PFNUM(sc->pf);
2610 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2611 	}
2612 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2613 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2614 	t4_read_reg(sc, reg);	/* flush */
2615 }
2616 
2617 int
2618 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2619     int len, int rw)
2620 {
2621 	struct memwin *mw;
2622 	uint32_t mw_end, v;
2623 
2624 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2625 
2626 	/* Memory can only be accessed in naturally aligned 4 byte units */
2627 	if (addr & 3 || len & 3 || len <= 0)
2628 		return (EINVAL);
2629 
2630 	mw = &sc->memwin[idx];
2631 	while (len > 0) {
2632 		rw_rlock(&mw->mw_lock);
2633 		mw_end = mw->mw_curpos + mw->mw_aperture;
2634 		if (addr >= mw_end || addr < mw->mw_curpos) {
2635 			/* Will need to reposition the window */
2636 			if (!rw_try_upgrade(&mw->mw_lock)) {
2637 				rw_runlock(&mw->mw_lock);
2638 				rw_wlock(&mw->mw_lock);
2639 			}
2640 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2641 			position_memwin(sc, idx, addr);
2642 			rw_downgrade(&mw->mw_lock);
2643 			mw_end = mw->mw_curpos + mw->mw_aperture;
2644 		}
2645 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2646 		while (addr < mw_end && len > 0) {
2647 			if (rw == 0) {
2648 				v = t4_read_reg(sc, mw->mw_base + addr -
2649 				    mw->mw_curpos);
2650 				*val++ = le32toh(v);
2651 			} else {
2652 				v = *val++;
2653 				t4_write_reg(sc, mw->mw_base + addr -
2654 				    mw->mw_curpos, htole32(v));
2655 			}
2656 			addr += 4;
2657 			len -= 4;
2658 		}
2659 		rw_runlock(&mw->mw_lock);
2660 	}
2661 
2662 	return (0);
2663 }
2664 
2665 int
2666 alloc_atid_tab(struct tid_info *t, int flags)
2667 {
2668 	int i;
2669 
2670 	MPASS(t->natids > 0);
2671 	MPASS(t->atid_tab == NULL);
2672 
2673 	t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
2674 	    M_ZERO | flags);
2675 	if (t->atid_tab == NULL)
2676 		return (ENOMEM);
2677 	mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
2678 	t->afree = t->atid_tab;
2679 	t->atids_in_use = 0;
2680 	for (i = 1; i < t->natids; i++)
2681 		t->atid_tab[i - 1].next = &t->atid_tab[i];
2682 	t->atid_tab[t->natids - 1].next = NULL;
2683 
2684 	return (0);
2685 }
2686 
2687 void
2688 free_atid_tab(struct tid_info *t)
2689 {
2690 
2691 	KASSERT(t->atids_in_use == 0,
2692 	    ("%s: %d atids still in use.", __func__, t->atids_in_use));
2693 
2694 	if (mtx_initialized(&t->atid_lock))
2695 		mtx_destroy(&t->atid_lock);
2696 	free(t->atid_tab, M_CXGBE);
2697 	t->atid_tab = NULL;
2698 }
2699 
2700 int
2701 alloc_atid(struct adapter *sc, void *ctx)
2702 {
2703 	struct tid_info *t = &sc->tids;
2704 	int atid = -1;
2705 
2706 	mtx_lock(&t->atid_lock);
2707 	if (t->afree) {
2708 		union aopen_entry *p = t->afree;
2709 
2710 		atid = p - t->atid_tab;
2711 		MPASS(atid <= M_TID_TID);
2712 		t->afree = p->next;
2713 		p->data = ctx;
2714 		t->atids_in_use++;
2715 	}
2716 	mtx_unlock(&t->atid_lock);
2717 	return (atid);
2718 }
2719 
2720 void *
2721 lookup_atid(struct adapter *sc, int atid)
2722 {
2723 	struct tid_info *t = &sc->tids;
2724 
2725 	return (t->atid_tab[atid].data);
2726 }
2727 
2728 void
2729 free_atid(struct adapter *sc, int atid)
2730 {
2731 	struct tid_info *t = &sc->tids;
2732 	union aopen_entry *p = &t->atid_tab[atid];
2733 
2734 	mtx_lock(&t->atid_lock);
2735 	p->next = t->afree;
2736 	t->afree = p;
2737 	t->atids_in_use--;
2738 	mtx_unlock(&t->atid_lock);
2739 }
2740 
2741 static void
2742 queue_tid_release(struct adapter *sc, int tid)
2743 {
2744 
2745 	CXGBE_UNIMPLEMENTED("deferred tid release");
2746 }
2747 
2748 void
2749 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
2750 {
2751 	struct wrqe *wr;
2752 	struct cpl_tid_release *req;
2753 
2754 	wr = alloc_wrqe(sizeof(*req), ctrlq);
2755 	if (wr == NULL) {
2756 		queue_tid_release(sc, tid);	/* defer */
2757 		return;
2758 	}
2759 	req = wrtod(wr);
2760 
2761 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
2762 
2763 	t4_wrq_tx(sc, wr);
2764 }
2765 
2766 static int
2767 t4_range_cmp(const void *a, const void *b)
2768 {
2769 	return ((const struct t4_range *)a)->start -
2770 	       ((const struct t4_range *)b)->start;
2771 }
2772 
2773 /*
2774  * Verify that the memory range specified by the addr/len pair is valid within
2775  * the card's address space.
2776  */
2777 static int
2778 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2779 {
2780 	struct t4_range mem_ranges[4], *r, *next;
2781 	uint32_t em, addr_len;
2782 	int i, n, remaining;
2783 
2784 	/* Memory can only be accessed in naturally aligned 4 byte units */
2785 	if (addr & 3 || len & 3 || len <= 0)
2786 		return (EINVAL);
2787 
2788 	/* Enabled memories */
2789 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2790 
2791 	r = &mem_ranges[0];
2792 	n = 0;
2793 	bzero(r, sizeof(mem_ranges));
2794 	if (em & F_EDRAM0_ENABLE) {
2795 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2796 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2797 		if (r->size > 0) {
2798 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2799 			if (addr >= r->start &&
2800 			    addr + len <= r->start + r->size)
2801 				return (0);
2802 			r++;
2803 			n++;
2804 		}
2805 	}
2806 	if (em & F_EDRAM1_ENABLE) {
2807 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2808 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2809 		if (r->size > 0) {
2810 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2811 			if (addr >= r->start &&
2812 			    addr + len <= r->start + r->size)
2813 				return (0);
2814 			r++;
2815 			n++;
2816 		}
2817 	}
2818 	if (em & F_EXT_MEM_ENABLE) {
2819 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2820 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2821 		if (r->size > 0) {
2822 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2823 			if (addr >= r->start &&
2824 			    addr + len <= r->start + r->size)
2825 				return (0);
2826 			r++;
2827 			n++;
2828 		}
2829 	}
2830 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2831 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2832 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2833 		if (r->size > 0) {
2834 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2835 			if (addr >= r->start &&
2836 			    addr + len <= r->start + r->size)
2837 				return (0);
2838 			r++;
2839 			n++;
2840 		}
2841 	}
2842 	MPASS(n <= nitems(mem_ranges));
2843 
2844 	if (n > 1) {
2845 		/* Sort and merge the ranges. */
2846 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2847 
2848 		/* Start from index 0 and examine the next n - 1 entries. */
2849 		r = &mem_ranges[0];
2850 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2851 
2852 			MPASS(r->size > 0);	/* r is a valid entry. */
2853 			next = r + 1;
2854 			MPASS(next->size > 0);	/* and so is the next one. */
2855 
2856 			while (r->start + r->size >= next->start) {
2857 				/* Merge the next one into the current entry. */
2858 				r->size = max(r->start + r->size,
2859 				    next->start + next->size) - r->start;
2860 				n--;	/* One fewer entry in total. */
2861 				if (--remaining == 0)
2862 					goto done;	/* short circuit */
2863 				next++;
2864 			}
2865 			if (next != r + 1) {
2866 				/*
2867 				 * Some entries were merged into r and next
2868 				 * points to the first valid entry that couldn't
2869 				 * be merged.
2870 				 */
2871 				MPASS(next->size > 0);	/* must be valid */
2872 				memcpy(r + 1, next, remaining * sizeof(*r));
2873 #ifdef INVARIANTS
2874 				/*
2875 				 * This so that the foo->size assertion in the
2876 				 * next iteration of the loop do the right
2877 				 * thing for entries that were pulled up and are
2878 				 * no longer valid.
2879 				 */
2880 				MPASS(n < nitems(mem_ranges));
2881 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2882 				    sizeof(struct t4_range));
2883 #endif
2884 			}
2885 		}
2886 done:
2887 		/* Done merging the ranges. */
2888 		MPASS(n > 0);
2889 		r = &mem_ranges[0];
2890 		for (i = 0; i < n; i++, r++) {
2891 			if (addr >= r->start &&
2892 			    addr + len <= r->start + r->size)
2893 				return (0);
2894 		}
2895 	}
2896 
2897 	return (EFAULT);
2898 }
2899 
2900 static int
2901 fwmtype_to_hwmtype(int mtype)
2902 {
2903 
2904 	switch (mtype) {
2905 	case FW_MEMTYPE_EDC0:
2906 		return (MEM_EDC0);
2907 	case FW_MEMTYPE_EDC1:
2908 		return (MEM_EDC1);
2909 	case FW_MEMTYPE_EXTMEM:
2910 		return (MEM_MC0);
2911 	case FW_MEMTYPE_EXTMEM1:
2912 		return (MEM_MC1);
2913 	default:
2914 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2915 	}
2916 }
2917 
2918 /*
2919  * Verify that the memory range specified by the memtype/offset/len pair is
2920  * valid and lies entirely within the memtype specified.  The global address of
2921  * the start of the range is returned in addr.
2922  */
2923 static int
2924 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2925     uint32_t *addr)
2926 {
2927 	uint32_t em, addr_len, maddr;
2928 
2929 	/* Memory can only be accessed in naturally aligned 4 byte units */
2930 	if (off & 3 || len & 3 || len == 0)
2931 		return (EINVAL);
2932 
2933 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2934 	switch (fwmtype_to_hwmtype(mtype)) {
2935 	case MEM_EDC0:
2936 		if (!(em & F_EDRAM0_ENABLE))
2937 			return (EINVAL);
2938 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2939 		maddr = G_EDRAM0_BASE(addr_len) << 20;
2940 		break;
2941 	case MEM_EDC1:
2942 		if (!(em & F_EDRAM1_ENABLE))
2943 			return (EINVAL);
2944 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2945 		maddr = G_EDRAM1_BASE(addr_len) << 20;
2946 		break;
2947 	case MEM_MC:
2948 		if (!(em & F_EXT_MEM_ENABLE))
2949 			return (EINVAL);
2950 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2951 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2952 		break;
2953 	case MEM_MC1:
2954 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2955 			return (EINVAL);
2956 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2957 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2958 		break;
2959 	default:
2960 		return (EINVAL);
2961 	}
2962 
2963 	*addr = maddr + off;	/* global address */
2964 	return (validate_mem_range(sc, *addr, len));
2965 }
2966 
2967 static int
2968 fixup_devlog_params(struct adapter *sc)
2969 {
2970 	struct devlog_params *dparams = &sc->params.devlog;
2971 	int rc;
2972 
2973 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2974 	    dparams->size, &dparams->addr);
2975 
2976 	return (rc);
2977 }
2978 
2979 static void
2980 update_nirq(struct intrs_and_queues *iaq, int nports)
2981 {
2982 	int extra = T4_EXTRA_INTR;
2983 
2984 	iaq->nirq = extra;
2985 	iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
2986 	iaq->nirq += nports * (iaq->num_vis - 1) *
2987 	    max(iaq->nrxq_vi, iaq->nnmrxq_vi);
2988 	iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
2989 }
2990 
2991 /*
2992  * Adjust requirements to fit the number of interrupts available.
2993  */
2994 static void
2995 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
2996     int navail)
2997 {
2998 	int old_nirq;
2999 	const int nports = sc->params.nports;
3000 
3001 	MPASS(nports > 0);
3002 	MPASS(navail > 0);
3003 
3004 	bzero(iaq, sizeof(*iaq));
3005 	iaq->intr_type = itype;
3006 	iaq->num_vis = t4_num_vis;
3007 	iaq->ntxq = t4_ntxq;
3008 	iaq->ntxq_vi = t4_ntxq_vi;
3009 	iaq->nrxq = t4_nrxq;
3010 	iaq->nrxq_vi = t4_nrxq_vi;
3011 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
3012 	if (is_offload(sc) || is_ethoffload(sc)) {
3013 		iaq->nofldtxq = t4_nofldtxq;
3014 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
3015 	}
3016 #endif
3017 #ifdef TCP_OFFLOAD
3018 	if (is_offload(sc)) {
3019 		iaq->nofldrxq = t4_nofldrxq;
3020 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
3021 	}
3022 #endif
3023 #ifdef DEV_NETMAP
3024 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
3025 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
3026 #endif
3027 
3028 	update_nirq(iaq, nports);
3029 	if (iaq->nirq <= navail &&
3030 	    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3031 		/*
3032 		 * This is the normal case -- there are enough interrupts for
3033 		 * everything.
3034 		 */
3035 		goto done;
3036 	}
3037 
3038 	/*
3039 	 * If extra VIs have been configured try reducing their count and see if
3040 	 * that works.
3041 	 */
3042 	while (iaq->num_vis > 1) {
3043 		iaq->num_vis--;
3044 		update_nirq(iaq, nports);
3045 		if (iaq->nirq <= navail &&
3046 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3047 			device_printf(sc->dev, "virtual interfaces per port "
3048 			    "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
3049 			    "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
3050 			    "itype %d, navail %u, nirq %d.\n",
3051 			    iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
3052 			    iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
3053 			    itype, navail, iaq->nirq);
3054 			goto done;
3055 		}
3056 	}
3057 
3058 	/*
3059 	 * Extra VIs will not be created.  Log a message if they were requested.
3060 	 */
3061 	MPASS(iaq->num_vis == 1);
3062 	iaq->ntxq_vi = iaq->nrxq_vi = 0;
3063 	iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
3064 	iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
3065 	if (iaq->num_vis != t4_num_vis) {
3066 		device_printf(sc->dev, "extra virtual interfaces disabled.  "
3067 		    "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
3068 		    "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
3069 		    iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
3070 		    iaq->nnmrxq_vi, itype, navail, iaq->nirq);
3071 	}
3072 
3073 	/*
3074 	 * Keep reducing the number of NIC rx queues to the next lower power of
3075 	 * 2 (for even RSS distribution) and halving the TOE rx queues and see
3076 	 * if that works.
3077 	 */
3078 	do {
3079 		if (iaq->nrxq > 1) {
3080 			do {
3081 				iaq->nrxq--;
3082 			} while (!powerof2(iaq->nrxq));
3083 		}
3084 		if (iaq->nofldrxq > 1)
3085 			iaq->nofldrxq >>= 1;
3086 
3087 		old_nirq = iaq->nirq;
3088 		update_nirq(iaq, nports);
3089 		if (iaq->nirq <= navail &&
3090 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3091 			device_printf(sc->dev, "running with reduced number of "
3092 			    "rx queues because of shortage of interrupts.  "
3093 			    "nrxq=%u, nofldrxq=%u.  "
3094 			    "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
3095 			    iaq->nofldrxq, itype, navail, iaq->nirq);
3096 			goto done;
3097 		}
3098 	} while (old_nirq != iaq->nirq);
3099 
3100 	/* One interrupt for everything.  Ugh. */
3101 	device_printf(sc->dev, "running with minimal number of queues.  "
3102 	    "itype %d, navail %u.\n", itype, navail);
3103 	iaq->nirq = 1;
3104 	MPASS(iaq->nrxq == 1);
3105 	iaq->ntxq = 1;
3106 	if (iaq->nofldrxq > 1)
3107 		iaq->nofldtxq = 1;
3108 done:
3109 	MPASS(iaq->num_vis > 0);
3110 	if (iaq->num_vis > 1) {
3111 		MPASS(iaq->nrxq_vi > 0);
3112 		MPASS(iaq->ntxq_vi > 0);
3113 	}
3114 	MPASS(iaq->nirq > 0);
3115 	MPASS(iaq->nrxq > 0);
3116 	MPASS(iaq->ntxq > 0);
3117 	if (itype == INTR_MSI) {
3118 		MPASS(powerof2(iaq->nirq));
3119 	}
3120 }
3121 
3122 static int
3123 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
3124 {
3125 	int rc, itype, navail, nalloc;
3126 
3127 	for (itype = INTR_MSIX; itype; itype >>= 1) {
3128 
3129 		if ((itype & t4_intr_types) == 0)
3130 			continue;	/* not allowed */
3131 
3132 		if (itype == INTR_MSIX)
3133 			navail = pci_msix_count(sc->dev);
3134 		else if (itype == INTR_MSI)
3135 			navail = pci_msi_count(sc->dev);
3136 		else
3137 			navail = 1;
3138 restart:
3139 		if (navail == 0)
3140 			continue;
3141 
3142 		calculate_iaq(sc, iaq, itype, navail);
3143 		nalloc = iaq->nirq;
3144 		rc = 0;
3145 		if (itype == INTR_MSIX)
3146 			rc = pci_alloc_msix(sc->dev, &nalloc);
3147 		else if (itype == INTR_MSI)
3148 			rc = pci_alloc_msi(sc->dev, &nalloc);
3149 
3150 		if (rc == 0 && nalloc > 0) {
3151 			if (nalloc == iaq->nirq)
3152 				return (0);
3153 
3154 			/*
3155 			 * Didn't get the number requested.  Use whatever number
3156 			 * the kernel is willing to allocate.
3157 			 */
3158 			device_printf(sc->dev, "fewer vectors than requested, "
3159 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
3160 			    itype, iaq->nirq, nalloc);
3161 			pci_release_msi(sc->dev);
3162 			navail = nalloc;
3163 			goto restart;
3164 		}
3165 
3166 		device_printf(sc->dev,
3167 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
3168 		    itype, rc, iaq->nirq, nalloc);
3169 	}
3170 
3171 	device_printf(sc->dev,
3172 	    "failed to find a usable interrupt type.  "
3173 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
3174 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
3175 
3176 	return (ENXIO);
3177 }
3178 
3179 #define FW_VERSION(chip) ( \
3180     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
3181     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
3182     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
3183     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
3184 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
3185 
3186 struct fw_info {
3187 	uint8_t chip;
3188 	char *kld_name;
3189 	char *fw_mod_name;
3190 	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
3191 } fw_info[] = {
3192 	{
3193 		.chip = CHELSIO_T4,
3194 		.kld_name = "t4fw_cfg",
3195 		.fw_mod_name = "t4fw",
3196 		.fw_hdr = {
3197 			.chip = FW_HDR_CHIP_T4,
3198 			.fw_ver = htobe32(FW_VERSION(T4)),
3199 			.intfver_nic = FW_INTFVER(T4, NIC),
3200 			.intfver_vnic = FW_INTFVER(T4, VNIC),
3201 			.intfver_ofld = FW_INTFVER(T4, OFLD),
3202 			.intfver_ri = FW_INTFVER(T4, RI),
3203 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
3204 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
3205 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
3206 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
3207 		},
3208 	}, {
3209 		.chip = CHELSIO_T5,
3210 		.kld_name = "t5fw_cfg",
3211 		.fw_mod_name = "t5fw",
3212 		.fw_hdr = {
3213 			.chip = FW_HDR_CHIP_T5,
3214 			.fw_ver = htobe32(FW_VERSION(T5)),
3215 			.intfver_nic = FW_INTFVER(T5, NIC),
3216 			.intfver_vnic = FW_INTFVER(T5, VNIC),
3217 			.intfver_ofld = FW_INTFVER(T5, OFLD),
3218 			.intfver_ri = FW_INTFVER(T5, RI),
3219 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
3220 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
3221 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
3222 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
3223 		},
3224 	}, {
3225 		.chip = CHELSIO_T6,
3226 		.kld_name = "t6fw_cfg",
3227 		.fw_mod_name = "t6fw",
3228 		.fw_hdr = {
3229 			.chip = FW_HDR_CHIP_T6,
3230 			.fw_ver = htobe32(FW_VERSION(T6)),
3231 			.intfver_nic = FW_INTFVER(T6, NIC),
3232 			.intfver_vnic = FW_INTFVER(T6, VNIC),
3233 			.intfver_ofld = FW_INTFVER(T6, OFLD),
3234 			.intfver_ri = FW_INTFVER(T6, RI),
3235 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3236 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
3237 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3238 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
3239 		},
3240 	}
3241 };
3242 
3243 static struct fw_info *
3244 find_fw_info(int chip)
3245 {
3246 	int i;
3247 
3248 	for (i = 0; i < nitems(fw_info); i++) {
3249 		if (fw_info[i].chip == chip)
3250 			return (&fw_info[i]);
3251 	}
3252 	return (NULL);
3253 }
3254 
3255 /*
3256  * Is the given firmware API compatible with the one the driver was compiled
3257  * with?
3258  */
3259 static int
3260 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
3261 {
3262 
3263 	/* short circuit if it's the exact same firmware version */
3264 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
3265 		return (1);
3266 
3267 	/*
3268 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
3269 	 * features that are supported in the driver.
3270 	 */
3271 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
3272 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
3273 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
3274 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
3275 		return (1);
3276 #undef SAME_INTF
3277 
3278 	return (0);
3279 }
3280 
3281 /*
3282  * The firmware in the KLD is usable, but should it be installed?  This routine
3283  * explains itself in detail if it indicates the KLD firmware should be
3284  * installed.
3285  */
3286 static int
3287 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
3288 {
3289 	const char *reason;
3290 
3291 	if (!card_fw_usable) {
3292 		reason = "incompatible or unusable";
3293 		goto install;
3294 	}
3295 
3296 	if (k > c) {
3297 		reason = "older than the version bundled with this driver";
3298 		goto install;
3299 	}
3300 
3301 	if (t4_fw_install == 2 && k != c) {
3302 		reason = "different than the version bundled with this driver";
3303 		goto install;
3304 	}
3305 
3306 	return (0);
3307 
3308 install:
3309 	if (t4_fw_install == 0) {
3310 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3311 		    "but the driver is prohibited from installing a different "
3312 		    "firmware on the card.\n",
3313 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3314 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3315 
3316 		return (0);
3317 	}
3318 
3319 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3320 	    "installing firmware %u.%u.%u.%u on card.\n",
3321 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3322 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3323 	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3324 	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3325 
3326 	return (1);
3327 }
3328 
3329 /*
3330  * Establish contact with the firmware and determine if we are the master driver
3331  * or not, and whether we are responsible for chip initialization.
3332  */
3333 static int
3334 prep_firmware(struct adapter *sc)
3335 {
3336 	const struct firmware *fw = NULL, *default_cfg;
3337 	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
3338 	enum dev_state state;
3339 	struct fw_info *fw_info;
3340 	struct fw_hdr *card_fw;		/* fw on the card */
3341 	const struct fw_hdr *kld_fw;	/* fw in the KLD */
3342 	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
3343 					   against */
3344 
3345 	/* This is the firmware whose headers the driver was compiled against */
3346 	fw_info = find_fw_info(chip_id(sc));
3347 	if (fw_info == NULL) {
3348 		device_printf(sc->dev,
3349 		    "unable to look up firmware information for chip %d.\n",
3350 		    chip_id(sc));
3351 		return (EINVAL);
3352 	}
3353 	drv_fw = &fw_info->fw_hdr;
3354 
3355 	/*
3356 	 * The firmware KLD contains many modules.  The KLD name is also the
3357 	 * name of the module that contains the default config file.
3358 	 */
3359 	default_cfg = firmware_get(fw_info->kld_name);
3360 
3361 	/* This is the firmware in the KLD */
3362 	fw = firmware_get(fw_info->fw_mod_name);
3363 	if (fw != NULL) {
3364 		kld_fw = (const void *)fw->data;
3365 		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3366 	} else {
3367 		kld_fw = NULL;
3368 		kld_fw_usable = 0;
3369 	}
3370 
3371 	/* Read the header of the firmware on the card */
3372 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3373 	rc = -t4_read_flash(sc, FLASH_FW_START,
3374 	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
3375 	if (rc == 0) {
3376 		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
3377 		if (card_fw->fw_ver == be32toh(0xffffffff)) {
3378 			uint32_t d = be32toh(kld_fw->fw_ver);
3379 
3380 			if (!kld_fw_usable) {
3381 				device_printf(sc->dev,
3382 				    "no firmware on the card and no usable "
3383 				    "firmware bundled with the driver.\n");
3384 				rc = EIO;
3385 				goto done;
3386 			} else if (t4_fw_install == 0) {
3387 				device_printf(sc->dev,
3388 				    "no firmware on the card and the driver "
3389 				    "is prohibited from installing new "
3390 				    "firmware.\n");
3391 				rc = EIO;
3392 				goto done;
3393 			}
3394 
3395 			device_printf(sc->dev, "no firmware on the card, "
3396 			    "installing firmware %d.%d.%d.%d\n",
3397 			    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3398 			    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3399 			rc = t4_fw_forceinstall(sc, fw->data, fw->datasize);
3400 			if (rc < 0) {
3401 				rc = -rc;
3402 				device_printf(sc->dev,
3403 				    "firmware install failed: %d.\n", rc);
3404 				goto done;
3405 			}
3406 			memcpy(card_fw, kld_fw, sizeof(*card_fw));
3407 			card_fw_usable = 1;
3408 			need_fw_reset = 0;
3409 		}
3410 	} else {
3411 		device_printf(sc->dev,
3412 		    "Unable to read card's firmware header: %d\n", rc);
3413 		card_fw_usable = 0;
3414 	}
3415 
3416 	/* Contact firmware. */
3417 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3418 	if (rc < 0 || state == DEV_STATE_ERR) {
3419 		rc = -rc;
3420 		device_printf(sc->dev,
3421 		    "failed to connect to the firmware: %d, %d.\n", rc, state);
3422 		goto done;
3423 	}
3424 	pf = rc;
3425 	if (pf == sc->mbox)
3426 		sc->flags |= MASTER_PF;
3427 	else if (state == DEV_STATE_UNINIT) {
3428 		/*
3429 		 * We didn't get to be the master so we definitely won't be
3430 		 * configuring the chip.  It's a bug if someone else hasn't
3431 		 * configured it already.
3432 		 */
3433 		device_printf(sc->dev, "couldn't be master(%d), "
3434 		    "device not already initialized either(%d).\n", rc, state);
3435 		rc = EPROTO;
3436 		goto done;
3437 	}
3438 
3439 	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3440 	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3441 		/*
3442 		 * Common case: the firmware on the card is an exact match and
3443 		 * the KLD is an exact match too, or the KLD is
3444 		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
3445 		 * here -- use cxgbetool loadfw if you want to reinstall the
3446 		 * same firmware as the one on the card.
3447 		 */
3448 	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3449 	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3450 	    be32toh(card_fw->fw_ver))) {
3451 
3452 		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3453 		if (rc != 0) {
3454 			device_printf(sc->dev,
3455 			    "failed to install firmware: %d\n", rc);
3456 			goto done;
3457 		}
3458 
3459 		/* Installed successfully, update the cached header too. */
3460 		memcpy(card_fw, kld_fw, sizeof(*card_fw));
3461 		card_fw_usable = 1;
3462 		need_fw_reset = 0;	/* already reset as part of load_fw */
3463 	}
3464 
3465 	if (!card_fw_usable) {
3466 		uint32_t d, c, k;
3467 
3468 		d = ntohl(drv_fw->fw_ver);
3469 		c = ntohl(card_fw->fw_ver);
3470 		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3471 
3472 		device_printf(sc->dev, "Cannot find a usable firmware: "
3473 		    "fw_install %d, chip state %d, "
3474 		    "driver compiled with %d.%d.%d.%d, "
3475 		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3476 		    t4_fw_install, state,
3477 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3478 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3479 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3480 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3481 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3482 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3483 		rc = EINVAL;
3484 		goto done;
3485 	}
3486 
3487 	/* Reset device */
3488 	if (need_fw_reset &&
3489 	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3490 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3491 		if (rc != ETIMEDOUT && rc != EIO)
3492 			t4_fw_bye(sc, sc->mbox);
3493 		goto done;
3494 	}
3495 	sc->flags |= FW_OK;
3496 
3497 	rc = get_params__pre_init(sc);
3498 	if (rc != 0)
3499 		goto done; /* error message displayed already */
3500 
3501 	/* Partition adapter resources as specified in the config file. */
3502 	if (state == DEV_STATE_UNINIT) {
3503 
3504 		KASSERT(sc->flags & MASTER_PF,
3505 		    ("%s: trying to change chip settings when not master.",
3506 		    __func__));
3507 
3508 		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3509 		if (rc != 0)
3510 			goto done;	/* error message displayed already */
3511 
3512 		t4_tweak_chip_settings(sc);
3513 
3514 		/* get basic stuff going */
3515 		rc = -t4_fw_initialize(sc, sc->mbox);
3516 		if (rc != 0) {
3517 			device_printf(sc->dev, "fw init failed: %d.\n", rc);
3518 			goto done;
3519 		}
3520 	} else {
3521 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3522 		sc->cfcsum = 0;
3523 	}
3524 
3525 done:
3526 	free(card_fw, M_CXGBE);
3527 	if (fw != NULL)
3528 		firmware_put(fw, FIRMWARE_UNLOAD);
3529 	if (default_cfg != NULL)
3530 		firmware_put(default_cfg, FIRMWARE_UNLOAD);
3531 
3532 	return (rc);
3533 }
3534 
3535 #define FW_PARAM_DEV(param) \
3536 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3537 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3538 #define FW_PARAM_PFVF(param) \
3539 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3540 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3541 
3542 /*
3543  * Partition chip resources for use between various PFs, VFs, etc.
3544  */
3545 static int
3546 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3547     const char *name_prefix)
3548 {
3549 	const struct firmware *cfg = NULL;
3550 	int rc = 0;
3551 	struct fw_caps_config_cmd caps;
3552 	uint32_t mtype, moff, finicsum, cfcsum;
3553 
3554 	/*
3555 	 * Figure out what configuration file to use.  Pick the default config
3556 	 * file for the card if the user hasn't specified one explicitly.
3557 	 */
3558 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3559 	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3560 		/* Card specific overrides go here. */
3561 		if (pci_get_device(sc->dev) == 0x440a)
3562 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3563 		if (is_fpga(sc))
3564 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3565 	} else if (strncmp(t4_cfg_file, BUILTIN_CF, sizeof(t4_cfg_file)) == 0)
3566 		goto use_built_in_config;	/* go straight to config. */
3567 
3568 	/*
3569 	 * We need to load another module if the profile is anything except
3570 	 * "default" or "flash".
3571 	 */
3572 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3573 	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3574 		char s[32];
3575 
3576 		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3577 		cfg = firmware_get(s);
3578 		if (cfg == NULL) {
3579 			if (default_cfg != NULL) {
3580 				device_printf(sc->dev,
3581 				    "unable to load module \"%s\" for "
3582 				    "configuration profile \"%s\", will use "
3583 				    "the default config file instead.\n",
3584 				    s, sc->cfg_file);
3585 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3586 				    "%s", DEFAULT_CF);
3587 			} else {
3588 				device_printf(sc->dev,
3589 				    "unable to load module \"%s\" for "
3590 				    "configuration profile \"%s\", will use "
3591 				    "the config file on the card's flash "
3592 				    "instead.\n", s, sc->cfg_file);
3593 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3594 				    "%s", FLASH_CF);
3595 			}
3596 		}
3597 	}
3598 
3599 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3600 	    default_cfg == NULL) {
3601 		device_printf(sc->dev,
3602 		    "default config file not available, will use the config "
3603 		    "file on the card's flash instead.\n");
3604 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3605 	}
3606 
3607 	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3608 		u_int cflen;
3609 		const uint32_t *cfdata;
3610 		uint32_t param, val, addr;
3611 
3612 		KASSERT(cfg != NULL || default_cfg != NULL,
3613 		    ("%s: no config to upload", __func__));
3614 
3615 		/*
3616 		 * Ask the firmware where it wants us to upload the config file.
3617 		 */
3618 		param = FW_PARAM_DEV(CF);
3619 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3620 		if (rc != 0) {
3621 			/* No support for config file?  Shouldn't happen. */
3622 			device_printf(sc->dev,
3623 			    "failed to query config file location: %d.\n", rc);
3624 			goto done;
3625 		}
3626 		mtype = G_FW_PARAMS_PARAM_Y(val);
3627 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3628 
3629 		/*
3630 		 * XXX: sheer laziness.  We deliberately added 4 bytes of
3631 		 * useless stuffing/comments at the end of the config file so
3632 		 * it's ok to simply throw away the last remaining bytes when
3633 		 * the config file is not an exact multiple of 4.  This also
3634 		 * helps with the validate_mt_off_len check.
3635 		 */
3636 		if (cfg != NULL) {
3637 			cflen = cfg->datasize & ~3;
3638 			cfdata = cfg->data;
3639 		} else {
3640 			cflen = default_cfg->datasize & ~3;
3641 			cfdata = default_cfg->data;
3642 		}
3643 
3644 		if (cflen > FLASH_CFG_MAX_SIZE) {
3645 			device_printf(sc->dev,
3646 			    "config file too long (%d, max allowed is %d).  "
3647 			    "Will try to use the config on the card, if any.\n",
3648 			    cflen, FLASH_CFG_MAX_SIZE);
3649 			goto use_config_on_flash;
3650 		}
3651 
3652 		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3653 		if (rc != 0) {
3654 			device_printf(sc->dev,
3655 			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3656 			    "Will try to use the config on the card, if any.\n",
3657 			    __func__, mtype, moff, cflen, rc);
3658 			goto use_config_on_flash;
3659 		}
3660 		write_via_memwin(sc, 2, addr, cfdata, cflen);
3661 	} else {
3662 use_config_on_flash:
3663 		mtype = FW_MEMTYPE_FLASH;
3664 		moff = t4_flash_cfg_addr(sc);
3665 	}
3666 
3667 	bzero(&caps, sizeof(caps));
3668 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3669 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3670 	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3671 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3672 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3673 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3674 	if (rc != 0) {
3675 		device_printf(sc->dev,
3676 		    "failed to pre-process config file: %d "
3677 		    "(mtype %d, moff 0x%x).  Will reset the firmware and retry "
3678 		    "with the built-in configuration.\n", rc, mtype, moff);
3679 
3680 	    	rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
3681 		if (rc != 0) {
3682 			device_printf(sc->dev,
3683 			    "firmware reset failed: %d.\n", rc);
3684 			if (rc != ETIMEDOUT && rc != EIO) {
3685 				t4_fw_bye(sc, sc->mbox);
3686 				sc->flags &= ~FW_OK;
3687 			}
3688 			goto done;
3689 		}
3690 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", "built-in");
3691 use_built_in_config:
3692 		bzero(&caps, sizeof(caps));
3693 		caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3694 		    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3695 		caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3696 		rc = t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3697 		if (rc != 0) {
3698 			device_printf(sc->dev,
3699 			    "built-in configuration failed: %d.\n", rc);
3700 			goto done;
3701 		}
3702 	}
3703 
3704 	finicsum = be32toh(caps.finicsum);
3705 	cfcsum = be32toh(caps.cfcsum);
3706 	if (finicsum != cfcsum) {
3707 		device_printf(sc->dev,
3708 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3709 		    finicsum, cfcsum);
3710 	}
3711 	sc->cfcsum = cfcsum;
3712 
3713 #define LIMIT_CAPS(x) do { \
3714 	caps.x &= htobe16(t4_##x##_allowed); \
3715 } while (0)
3716 
3717 	/*
3718 	 * Let the firmware know what features will (not) be used so it can tune
3719 	 * things accordingly.
3720 	 */
3721 	LIMIT_CAPS(nbmcaps);
3722 	LIMIT_CAPS(linkcaps);
3723 	LIMIT_CAPS(switchcaps);
3724 	LIMIT_CAPS(niccaps);
3725 	LIMIT_CAPS(toecaps);
3726 	LIMIT_CAPS(rdmacaps);
3727 	LIMIT_CAPS(cryptocaps);
3728 	LIMIT_CAPS(iscsicaps);
3729 	LIMIT_CAPS(fcoecaps);
3730 #undef LIMIT_CAPS
3731 
3732 	if (caps.niccaps & htobe16(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
3733 		/*
3734 		 * TOE and hashfilters are mutually exclusive.  It is a config
3735 		 * file or firmware bug if both are reported as available.  Try
3736 		 * to cope with the situation in non-debug builds by disabling
3737 		 * TOE.
3738 		 */
3739 		MPASS(caps.toecaps == 0);
3740 
3741 		caps.toecaps = 0;
3742 		caps.rdmacaps = 0;
3743 		caps.iscsicaps = 0;
3744 	}
3745 
3746 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3747 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3748 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3749 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3750 	if (rc != 0) {
3751 		device_printf(sc->dev,
3752 		    "failed to process config file: %d.\n", rc);
3753 	}
3754 done:
3755 	if (cfg != NULL)
3756 		firmware_put(cfg, FIRMWARE_UNLOAD);
3757 	return (rc);
3758 }
3759 
3760 /*
3761  * Retrieve parameters that are needed (or nice to have) very early.
3762  */
3763 static int
3764 get_params__pre_init(struct adapter *sc)
3765 {
3766 	int rc;
3767 	uint32_t param[2], val[2];
3768 
3769 	t4_get_version_info(sc);
3770 
3771 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3772 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3773 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3774 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3775 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3776 
3777 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3778 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3779 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3780 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3781 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3782 
3783 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3784 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3785 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3786 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3787 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3788 
3789 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3790 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3791 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3792 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3793 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3794 
3795 	param[0] = FW_PARAM_DEV(PORTVEC);
3796 	param[1] = FW_PARAM_DEV(CCLK);
3797 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3798 	if (rc != 0) {
3799 		device_printf(sc->dev,
3800 		    "failed to query parameters (pre_init): %d.\n", rc);
3801 		return (rc);
3802 	}
3803 
3804 	sc->params.portvec = val[0];
3805 	sc->params.nports = bitcount32(val[0]);
3806 	sc->params.vpd.cclk = val[1];
3807 
3808 	/* Read device log parameters. */
3809 	rc = -t4_init_devlog_params(sc, 1);
3810 	if (rc == 0)
3811 		fixup_devlog_params(sc);
3812 	else {
3813 		device_printf(sc->dev,
3814 		    "failed to get devlog parameters: %d.\n", rc);
3815 		rc = 0;	/* devlog isn't critical for device operation */
3816 	}
3817 
3818 	return (rc);
3819 }
3820 
3821 /*
3822  * Retrieve various parameters that are of interest to the driver.  The device
3823  * has been initialized by the firmware at this point.
3824  */
3825 static int
3826 get_params__post_init(struct adapter *sc)
3827 {
3828 	int rc;
3829 	uint32_t param[7], val[7];
3830 	struct fw_caps_config_cmd caps;
3831 
3832 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3833 	param[1] = FW_PARAM_PFVF(EQ_START);
3834 	param[2] = FW_PARAM_PFVF(FILTER_START);
3835 	param[3] = FW_PARAM_PFVF(FILTER_END);
3836 	param[4] = FW_PARAM_PFVF(L2T_START);
3837 	param[5] = FW_PARAM_PFVF(L2T_END);
3838 	param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
3839 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
3840 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
3841 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
3842 	if (rc != 0) {
3843 		device_printf(sc->dev,
3844 		    "failed to query parameters (post_init): %d.\n", rc);
3845 		return (rc);
3846 	}
3847 
3848 	sc->sge.iq_start = val[0];
3849 	sc->sge.eq_start = val[1];
3850 	if (val[3] > val[2]) {
3851 		sc->tids.ftid_base = val[2];
3852 		sc->tids.ftid_end = val[3];
3853 		sc->tids.nftids = val[3] - val[2] + 1;
3854 	}
3855 	sc->vres.l2t.start = val[4];
3856 	sc->vres.l2t.size = val[5] - val[4] + 1;
3857 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3858 	    ("%s: L2 table size (%u) larger than expected (%u)",
3859 	    __func__, sc->vres.l2t.size, L2T_SIZE));
3860 	sc->params.core_vdd = val[6];
3861 
3862 	if (chip_id(sc) >= CHELSIO_T6) {
3863 
3864 #ifdef INVARIANTS
3865 		if (sc->params.fw_vers >=
3866 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
3867 		    V_FW_HDR_FW_VER_MICRO(1) | V_FW_HDR_FW_VER_BUILD(0))) {
3868 			/*
3869 			 * Note that the code to enable the region should run
3870 			 * before t4_fw_initialize and not here.  This is just a
3871 			 * reminder to add said code.
3872 			 */
3873 			device_printf(sc->dev,
3874 			    "hpfilter region not enabled.\n");
3875 		}
3876 #endif
3877 
3878 		sc->tids.tid_base = t4_read_reg(sc,
3879 		    A_LE_DB_ACTIVE_TABLE_START_INDEX);
3880 
3881 		param[0] = FW_PARAM_PFVF(HPFILTER_START);
3882 		param[1] = FW_PARAM_PFVF(HPFILTER_END);
3883 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3884 		if (rc != 0) {
3885 			device_printf(sc->dev,
3886 			   "failed to query hpfilter parameters: %d.\n", rc);
3887 			return (rc);
3888 		}
3889 		if ((int)val[1] > (int)val[0]) {
3890 			sc->tids.hpftid_base = val[0];
3891 			sc->tids.hpftid_end = val[1];
3892 			sc->tids.nhpftids = val[1] - val[0] + 1;
3893 
3894 			/*
3895 			 * These should go off if the layout changes and the
3896 			 * driver needs to catch up.
3897 			 */
3898 			MPASS(sc->tids.hpftid_base == 0);
3899 			MPASS(sc->tids.tid_base == sc->tids.nhpftids);
3900 		}
3901 	}
3902 
3903 	/*
3904 	 * MPSBGMAP is queried separately because only recent firmwares support
3905 	 * it as a parameter and we don't want the compound query above to fail
3906 	 * on older firmwares.
3907 	 */
3908 	param[0] = FW_PARAM_DEV(MPSBGMAP);
3909 	val[0] = 0;
3910 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3911 	if (rc == 0)
3912 		sc->params.mps_bg_map = val[0];
3913 	else
3914 		sc->params.mps_bg_map = 0;
3915 
3916 	/*
3917 	 * Determine whether the firmware supports the filter2 work request.
3918 	 * This is queried separately for the same reason as MPSBGMAP above.
3919 	 */
3920 	param[0] = FW_PARAM_DEV(FILTER2_WR);
3921 	val[0] = 0;
3922 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3923 	if (rc == 0)
3924 		sc->params.filter2_wr_support = val[0] != 0;
3925 	else
3926 		sc->params.filter2_wr_support = 0;
3927 
3928 	/* get capabilites */
3929 	bzero(&caps, sizeof(caps));
3930 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3931 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3932 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3933 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3934 	if (rc != 0) {
3935 		device_printf(sc->dev,
3936 		    "failed to get card capabilities: %d.\n", rc);
3937 		return (rc);
3938 	}
3939 
3940 #define READ_CAPS(x) do { \
3941 	sc->x = htobe16(caps.x); \
3942 } while (0)
3943 	READ_CAPS(nbmcaps);
3944 	READ_CAPS(linkcaps);
3945 	READ_CAPS(switchcaps);
3946 	READ_CAPS(niccaps);
3947 	READ_CAPS(toecaps);
3948 	READ_CAPS(rdmacaps);
3949 	READ_CAPS(cryptocaps);
3950 	READ_CAPS(iscsicaps);
3951 	READ_CAPS(fcoecaps);
3952 
3953 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_HASHFILTER) {
3954 		MPASS(chip_id(sc) > CHELSIO_T4);
3955 		MPASS(sc->toecaps == 0);
3956 		sc->toecaps = 0;
3957 
3958 		param[0] = FW_PARAM_DEV(NTID);
3959 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3960 		if (rc != 0) {
3961 			device_printf(sc->dev,
3962 			    "failed to query HASHFILTER parameters: %d.\n", rc);
3963 			return (rc);
3964 		}
3965 		sc->tids.ntids = val[0];
3966 		if (sc->params.fw_vers <
3967 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
3968 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
3969 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
3970 			sc->tids.ntids -= sc->tids.nhpftids;
3971 		}
3972 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3973 		sc->params.hash_filter = 1;
3974 	}
3975 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3976 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3977 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3978 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3979 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3980 		if (rc != 0) {
3981 			device_printf(sc->dev,
3982 			    "failed to query NIC parameters: %d.\n", rc);
3983 			return (rc);
3984 		}
3985 		if (val[1] > val[0]) {
3986 			sc->tids.etid_base = val[0];
3987 			sc->tids.etid_end = val[1];
3988 			sc->tids.netids = val[1] - val[0] + 1;
3989 			sc->params.eo_wr_cred = val[2];
3990 			sc->params.ethoffload = 1;
3991 		}
3992 	}
3993 	if (sc->toecaps) {
3994 		/* query offload-related parameters */
3995 		param[0] = FW_PARAM_DEV(NTID);
3996 		param[1] = FW_PARAM_PFVF(SERVER_START);
3997 		param[2] = FW_PARAM_PFVF(SERVER_END);
3998 		param[3] = FW_PARAM_PFVF(TDDP_START);
3999 		param[4] = FW_PARAM_PFVF(TDDP_END);
4000 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4001 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4002 		if (rc != 0) {
4003 			device_printf(sc->dev,
4004 			    "failed to query TOE parameters: %d.\n", rc);
4005 			return (rc);
4006 		}
4007 		sc->tids.ntids = val[0];
4008 		if (sc->params.fw_vers <
4009 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4010 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4011 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4012 			sc->tids.ntids -= sc->tids.nhpftids;
4013 		}
4014 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4015 		if (val[2] > val[1]) {
4016 			sc->tids.stid_base = val[1];
4017 			sc->tids.nstids = val[2] - val[1] + 1;
4018 		}
4019 		sc->vres.ddp.start = val[3];
4020 		sc->vres.ddp.size = val[4] - val[3] + 1;
4021 		sc->params.ofldq_wr_cred = val[5];
4022 		sc->params.offload = 1;
4023 	} else {
4024 		/*
4025 		 * The firmware attempts memfree TOE configuration for -SO cards
4026 		 * and will report toecaps=0 if it runs out of resources (this
4027 		 * depends on the config file).  It may not report 0 for other
4028 		 * capabilities dependent on the TOE in this case.  Set them to
4029 		 * 0 here so that the driver doesn't bother tracking resources
4030 		 * that will never be used.
4031 		 */
4032 		sc->iscsicaps = 0;
4033 		sc->rdmacaps = 0;
4034 	}
4035 	if (sc->rdmacaps) {
4036 		param[0] = FW_PARAM_PFVF(STAG_START);
4037 		param[1] = FW_PARAM_PFVF(STAG_END);
4038 		param[2] = FW_PARAM_PFVF(RQ_START);
4039 		param[3] = FW_PARAM_PFVF(RQ_END);
4040 		param[4] = FW_PARAM_PFVF(PBL_START);
4041 		param[5] = FW_PARAM_PFVF(PBL_END);
4042 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4043 		if (rc != 0) {
4044 			device_printf(sc->dev,
4045 			    "failed to query RDMA parameters(1): %d.\n", rc);
4046 			return (rc);
4047 		}
4048 		sc->vres.stag.start = val[0];
4049 		sc->vres.stag.size = val[1] - val[0] + 1;
4050 		sc->vres.rq.start = val[2];
4051 		sc->vres.rq.size = val[3] - val[2] + 1;
4052 		sc->vres.pbl.start = val[4];
4053 		sc->vres.pbl.size = val[5] - val[4] + 1;
4054 
4055 		param[0] = FW_PARAM_PFVF(SQRQ_START);
4056 		param[1] = FW_PARAM_PFVF(SQRQ_END);
4057 		param[2] = FW_PARAM_PFVF(CQ_START);
4058 		param[3] = FW_PARAM_PFVF(CQ_END);
4059 		param[4] = FW_PARAM_PFVF(OCQ_START);
4060 		param[5] = FW_PARAM_PFVF(OCQ_END);
4061 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4062 		if (rc != 0) {
4063 			device_printf(sc->dev,
4064 			    "failed to query RDMA parameters(2): %d.\n", rc);
4065 			return (rc);
4066 		}
4067 		sc->vres.qp.start = val[0];
4068 		sc->vres.qp.size = val[1] - val[0] + 1;
4069 		sc->vres.cq.start = val[2];
4070 		sc->vres.cq.size = val[3] - val[2] + 1;
4071 		sc->vres.ocq.start = val[4];
4072 		sc->vres.ocq.size = val[5] - val[4] + 1;
4073 
4074 		param[0] = FW_PARAM_PFVF(SRQ_START);
4075 		param[1] = FW_PARAM_PFVF(SRQ_END);
4076 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
4077 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4078 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
4079 		if (rc != 0) {
4080 			device_printf(sc->dev,
4081 			    "failed to query RDMA parameters(3): %d.\n", rc);
4082 			return (rc);
4083 		}
4084 		sc->vres.srq.start = val[0];
4085 		sc->vres.srq.size = val[1] - val[0] + 1;
4086 		sc->params.max_ordird_qp = val[2];
4087 		sc->params.max_ird_adapter = val[3];
4088 	}
4089 	if (sc->iscsicaps) {
4090 		param[0] = FW_PARAM_PFVF(ISCSI_START);
4091 		param[1] = FW_PARAM_PFVF(ISCSI_END);
4092 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4093 		if (rc != 0) {
4094 			device_printf(sc->dev,
4095 			    "failed to query iSCSI parameters: %d.\n", rc);
4096 			return (rc);
4097 		}
4098 		sc->vres.iscsi.start = val[0];
4099 		sc->vres.iscsi.size = val[1] - val[0] + 1;
4100 	}
4101 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
4102 		param[0] = FW_PARAM_PFVF(TLS_START);
4103 		param[1] = FW_PARAM_PFVF(TLS_END);
4104 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4105 		if (rc != 0) {
4106 			device_printf(sc->dev,
4107 			    "failed to query TLS parameters: %d.\n", rc);
4108 			return (rc);
4109 		}
4110 		sc->vres.key.start = val[0];
4111 		sc->vres.key.size = val[1] - val[0] + 1;
4112 	}
4113 
4114 	t4_init_sge_params(sc);
4115 
4116 	/*
4117 	 * We've got the params we wanted to query via the firmware.  Now grab
4118 	 * some others directly from the chip.
4119 	 */
4120 	rc = t4_read_chip_settings(sc);
4121 
4122 	return (rc);
4123 }
4124 
4125 static int
4126 set_params__post_init(struct adapter *sc)
4127 {
4128 	uint32_t param, val;
4129 #ifdef TCP_OFFLOAD
4130 	int i, v, shift;
4131 #endif
4132 
4133 	/* ask for encapsulated CPLs */
4134 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4135 	val = 1;
4136 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4137 
4138 #ifdef TCP_OFFLOAD
4139 	/*
4140 	 * Override the TOE timers with user provided tunables.  This is not the
4141 	 * recommended way to change the timers (the firmware config file is) so
4142 	 * these tunables are not documented.
4143 	 *
4144 	 * All the timer tunables are in microseconds.
4145 	 */
4146 	if (t4_toe_keepalive_idle != 0) {
4147 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
4148 		v &= M_KEEPALIVEIDLE;
4149 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
4150 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
4151 	}
4152 	if (t4_toe_keepalive_interval != 0) {
4153 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
4154 		v &= M_KEEPALIVEINTVL;
4155 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
4156 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
4157 	}
4158 	if (t4_toe_keepalive_count != 0) {
4159 		v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
4160 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4161 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
4162 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
4163 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
4164 	}
4165 	if (t4_toe_rexmt_min != 0) {
4166 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
4167 		v &= M_RXTMIN;
4168 		t4_set_reg_field(sc, A_TP_RXT_MIN,
4169 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
4170 	}
4171 	if (t4_toe_rexmt_max != 0) {
4172 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
4173 		v &= M_RXTMAX;
4174 		t4_set_reg_field(sc, A_TP_RXT_MAX,
4175 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
4176 	}
4177 	if (t4_toe_rexmt_count != 0) {
4178 		v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
4179 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4180 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
4181 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
4182 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
4183 	}
4184 	for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
4185 		if (t4_toe_rexmt_backoff[i] != -1) {
4186 			v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
4187 			shift = (i & 3) << 3;
4188 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
4189 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
4190 		}
4191 	}
4192 #endif
4193 	return (0);
4194 }
4195 
4196 #undef FW_PARAM_PFVF
4197 #undef FW_PARAM_DEV
4198 
4199 static void
4200 t4_set_desc(struct adapter *sc)
4201 {
4202 	char buf[128];
4203 	struct adapter_params *p = &sc->params;
4204 
4205 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
4206 
4207 	device_set_desc_copy(sc->dev, buf);
4208 }
4209 
4210 static inline void
4211 ifmedia_add4(struct ifmedia *ifm, int m)
4212 {
4213 
4214 	ifmedia_add(ifm, m, 0, NULL);
4215 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE, 0, NULL);
4216 	ifmedia_add(ifm, m | IFM_ETH_RXPAUSE, 0, NULL);
4217 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE, 0, NULL);
4218 }
4219 
4220 static void
4221 set_current_media(struct port_info *pi, struct ifmedia *ifm)
4222 {
4223 	struct link_config *lc;
4224 	int mword;
4225 
4226 	PORT_LOCK_ASSERT_OWNED(pi);
4227 
4228 	/* Leave current media alone if it's already set to IFM_NONE. */
4229 	if (ifm->ifm_cur != NULL &&
4230 	    IFM_SUBTYPE(ifm->ifm_cur->ifm_media) == IFM_NONE)
4231 		return;
4232 
4233 	lc = &pi->link_cfg;
4234 	if (lc->requested_aneg == AUTONEG_ENABLE &&
4235 	    lc->supported & FW_PORT_CAP_ANEG) {
4236 		ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
4237 		return;
4238 	}
4239 	mword = IFM_ETHER | IFM_FDX;
4240 	if (lc->requested_fc & PAUSE_TX)
4241 		mword |= IFM_ETH_TXPAUSE;
4242 	if (lc->requested_fc & PAUSE_RX)
4243 		mword |= IFM_ETH_RXPAUSE;
4244 	mword |= port_mword(pi, speed_to_fwspeed(lc->requested_speed));
4245 	ifmedia_set(ifm, mword);
4246 }
4247 
4248 static void
4249 build_medialist(struct port_info *pi, struct ifmedia *ifm)
4250 {
4251 	uint16_t ss, speed;
4252 	int unknown, mword, bit;
4253 	struct link_config *lc;
4254 
4255 	PORT_LOCK_ASSERT_OWNED(pi);
4256 
4257 	if (pi->flags & FIXED_IFMEDIA)
4258 		return;
4259 
4260 	/*
4261 	 * First setup all the requested_ fields so that they comply with what's
4262 	 * supported by the port + transceiver.  Note that this clobbers any
4263 	 * user preferences set via sysctl_pause_settings or sysctl_autoneg.
4264 	 */
4265 	init_l1cfg(pi);
4266 
4267 	/*
4268 	 * Now (re)build the ifmedia list.
4269 	 */
4270 	ifmedia_removeall(ifm);
4271 	lc = &pi->link_cfg;
4272 	ss = G_FW_PORT_CAP_SPEED(lc->supported); /* Supported Speeds */
4273 	if (__predict_false(ss == 0)) {	/* not supposed to happen. */
4274 		MPASS(ss != 0);
4275 no_media:
4276 		MPASS(LIST_EMPTY(&ifm->ifm_list));
4277 		ifmedia_add(ifm, IFM_ETHER | IFM_NONE, 0, NULL);
4278 		ifmedia_set(ifm, IFM_ETHER | IFM_NONE);
4279 		return;
4280 	}
4281 
4282 	unknown = 0;
4283 	for (bit = 0; bit < fls(ss); bit++) {
4284 		speed = 1 << bit;
4285 		MPASS(speed & M_FW_PORT_CAP_SPEED);
4286 		if (ss & speed) {
4287 			mword = port_mword(pi, speed);
4288 			if (mword == IFM_NONE) {
4289 				goto no_media;
4290 			} else if (mword == IFM_UNKNOWN)
4291 				unknown++;
4292 			else
4293 				ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | mword);
4294 		}
4295 	}
4296 	if (unknown > 0) /* Add one unknown for all unknown media types. */
4297 		ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | IFM_UNKNOWN);
4298 	if (lc->supported & FW_PORT_CAP_ANEG)
4299 		ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
4300 
4301 	set_current_media(pi, ifm);
4302 }
4303 
4304 /*
4305  * Update all the requested_* fields in the link config to something valid (and
4306  * reasonable).
4307  */
4308 static void
4309 init_l1cfg(struct port_info *pi)
4310 {
4311 	struct link_config *lc = &pi->link_cfg;
4312 
4313 	PORT_LOCK_ASSERT_OWNED(pi);
4314 
4315 	/* Gbps -> Mbps */
4316 	lc->requested_speed = port_top_speed(pi) * 1000;
4317 
4318 	if (t4_autoneg != 0 && lc->supported & FW_PORT_CAP_ANEG) {
4319 		lc->requested_aneg = AUTONEG_ENABLE;
4320 	} else {
4321 		lc->requested_aneg = AUTONEG_DISABLE;
4322 	}
4323 
4324 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX);
4325 
4326 	if (t4_fec != -1) {
4327 		if (t4_fec & FEC_RS && lc->supported & FW_PORT_CAP_FEC_RS) {
4328 			lc->requested_fec = FEC_RS;
4329 		} else if (t4_fec & FEC_BASER_RS &&
4330 		    lc->supported & FW_PORT_CAP_FEC_BASER_RS) {
4331 			lc->requested_fec = FEC_BASER_RS;
4332 		} else {
4333 			lc->requested_fec = 0;
4334 		}
4335 	} else {
4336 		/* Use the suggested value provided by the firmware in acaps */
4337 		if (lc->advertising & FW_PORT_CAP_FEC_RS &&
4338 		    lc->supported & FW_PORT_CAP_FEC_RS) {
4339 			lc->requested_fec = FEC_RS;
4340 		} else if (lc->advertising & FW_PORT_CAP_FEC_BASER_RS &&
4341 		    lc->supported & FW_PORT_CAP_FEC_BASER_RS) {
4342 			lc->requested_fec = FEC_BASER_RS;
4343 		} else {
4344 			lc->requested_fec = 0;
4345 		}
4346 	}
4347 }
4348 
4349 /*
4350  * Apply the settings in requested_* to the hardware.  The parameters are
4351  * expected to be sane.
4352  */
4353 static int
4354 apply_l1cfg(struct port_info *pi)
4355 {
4356 	struct adapter *sc = pi->adapter;
4357 	struct link_config *lc = &pi->link_cfg;
4358 	int rc;
4359 #ifdef INVARIANTS
4360 	uint16_t fwspeed;
4361 
4362 	ASSERT_SYNCHRONIZED_OP(sc);
4363 	PORT_LOCK_ASSERT_OWNED(pi);
4364 
4365 	if (lc->requested_aneg == AUTONEG_ENABLE)
4366 		MPASS(lc->supported & FW_PORT_CAP_ANEG);
4367 	if (lc->requested_fc & PAUSE_TX)
4368 		MPASS(lc->supported & FW_PORT_CAP_FC_TX);
4369 	if (lc->requested_fc & PAUSE_RX)
4370 		MPASS(lc->supported & FW_PORT_CAP_FC_RX);
4371 	if (lc->requested_fec == FEC_RS)
4372 		MPASS(lc->supported & FW_PORT_CAP_FEC_RS);
4373 	if (lc->requested_fec == FEC_BASER_RS)
4374 		MPASS(lc->supported & FW_PORT_CAP_FEC_BASER_RS);
4375 	fwspeed = speed_to_fwspeed(lc->requested_speed);
4376 	MPASS(fwspeed != 0);
4377 	MPASS(lc->supported & fwspeed);
4378 #endif
4379 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
4380 	if (rc != 0) {
4381 		device_printf(pi->dev, "l1cfg failed: %d\n", rc);
4382 	} else {
4383 		lc->fc = lc->requested_fc;
4384 		lc->fec = lc->requested_fec;
4385 	}
4386 	return (rc);
4387 }
4388 
4389 #define FW_MAC_EXACT_CHUNK	7
4390 
4391 /*
4392  * Program the port's XGMAC based on parameters in ifnet.  The caller also
4393  * indicates which parameters should be programmed (the rest are left alone).
4394  */
4395 int
4396 update_mac_settings(struct ifnet *ifp, int flags)
4397 {
4398 	int rc = 0;
4399 	struct vi_info *vi = ifp->if_softc;
4400 	struct port_info *pi = vi->pi;
4401 	struct adapter *sc = pi->adapter;
4402 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4403 
4404 	ASSERT_SYNCHRONIZED_OP(sc);
4405 	KASSERT(flags, ("%s: not told what to update.", __func__));
4406 
4407 	if (flags & XGMAC_MTU)
4408 		mtu = ifp->if_mtu;
4409 
4410 	if (flags & XGMAC_PROMISC)
4411 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4412 
4413 	if (flags & XGMAC_ALLMULTI)
4414 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4415 
4416 	if (flags & XGMAC_VLANEX)
4417 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4418 
4419 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4420 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4421 		    allmulti, 1, vlanex, false);
4422 		if (rc) {
4423 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4424 			    rc);
4425 			return (rc);
4426 		}
4427 	}
4428 
4429 	if (flags & XGMAC_UCADDR) {
4430 		uint8_t ucaddr[ETHER_ADDR_LEN];
4431 
4432 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4433 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4434 		    ucaddr, true, true);
4435 		if (rc < 0) {
4436 			rc = -rc;
4437 			if_printf(ifp, "change_mac failed: %d\n", rc);
4438 			return (rc);
4439 		} else {
4440 			vi->xact_addr_filt = rc;
4441 			rc = 0;
4442 		}
4443 	}
4444 
4445 	if (flags & XGMAC_MCADDRS) {
4446 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4447 		int del = 1;
4448 		uint64_t hash = 0;
4449 		struct ifmultiaddr *ifma;
4450 		int i = 0, j;
4451 
4452 		if_maddr_rlock(ifp);
4453 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4454 			if (ifma->ifma_addr->sa_family != AF_LINK)
4455 				continue;
4456 			mcaddr[i] =
4457 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4458 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4459 			i++;
4460 
4461 			if (i == FW_MAC_EXACT_CHUNK) {
4462 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4463 				    del, i, mcaddr, NULL, &hash, 0);
4464 				if (rc < 0) {
4465 					rc = -rc;
4466 					for (j = 0; j < i; j++) {
4467 						if_printf(ifp,
4468 						    "failed to add mc address"
4469 						    " %02x:%02x:%02x:"
4470 						    "%02x:%02x:%02x rc=%d\n",
4471 						    mcaddr[j][0], mcaddr[j][1],
4472 						    mcaddr[j][2], mcaddr[j][3],
4473 						    mcaddr[j][4], mcaddr[j][5],
4474 						    rc);
4475 					}
4476 					goto mcfail;
4477 				}
4478 				del = 0;
4479 				i = 0;
4480 			}
4481 		}
4482 		if (i > 0) {
4483 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4484 			    mcaddr, NULL, &hash, 0);
4485 			if (rc < 0) {
4486 				rc = -rc;
4487 				for (j = 0; j < i; j++) {
4488 					if_printf(ifp,
4489 					    "failed to add mc address"
4490 					    " %02x:%02x:%02x:"
4491 					    "%02x:%02x:%02x rc=%d\n",
4492 					    mcaddr[j][0], mcaddr[j][1],
4493 					    mcaddr[j][2], mcaddr[j][3],
4494 					    mcaddr[j][4], mcaddr[j][5],
4495 					    rc);
4496 				}
4497 				goto mcfail;
4498 			}
4499 		}
4500 
4501 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4502 		if (rc != 0)
4503 			if_printf(ifp, "failed to set mc address hash: %d", rc);
4504 mcfail:
4505 		if_maddr_runlock(ifp);
4506 	}
4507 
4508 	return (rc);
4509 }
4510 
4511 /*
4512  * {begin|end}_synchronized_op must be called from the same thread.
4513  */
4514 int
4515 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4516     char *wmesg)
4517 {
4518 	int rc, pri;
4519 
4520 #ifdef WITNESS
4521 	/* the caller thinks it's ok to sleep, but is it really? */
4522 	if (flags & SLEEP_OK)
4523 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4524 		    "begin_synchronized_op");
4525 #endif
4526 
4527 	if (INTR_OK)
4528 		pri = PCATCH;
4529 	else
4530 		pri = 0;
4531 
4532 	ADAPTER_LOCK(sc);
4533 	for (;;) {
4534 
4535 		if (vi && IS_DOOMED(vi)) {
4536 			rc = ENXIO;
4537 			goto done;
4538 		}
4539 
4540 		if (!IS_BUSY(sc)) {
4541 			rc = 0;
4542 			break;
4543 		}
4544 
4545 		if (!(flags & SLEEP_OK)) {
4546 			rc = EBUSY;
4547 			goto done;
4548 		}
4549 
4550 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4551 			rc = EINTR;
4552 			goto done;
4553 		}
4554 	}
4555 
4556 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4557 	SET_BUSY(sc);
4558 #ifdef INVARIANTS
4559 	sc->last_op = wmesg;
4560 	sc->last_op_thr = curthread;
4561 	sc->last_op_flags = flags;
4562 #endif
4563 
4564 done:
4565 	if (!(flags & HOLD_LOCK) || rc)
4566 		ADAPTER_UNLOCK(sc);
4567 
4568 	return (rc);
4569 }
4570 
4571 /*
4572  * Tell if_ioctl and if_init that the VI is going away.  This is
4573  * special variant of begin_synchronized_op and must be paired with a
4574  * call to end_synchronized_op.
4575  */
4576 void
4577 doom_vi(struct adapter *sc, struct vi_info *vi)
4578 {
4579 
4580 	ADAPTER_LOCK(sc);
4581 	SET_DOOMED(vi);
4582 	wakeup(&sc->flags);
4583 	while (IS_BUSY(sc))
4584 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4585 	SET_BUSY(sc);
4586 #ifdef INVARIANTS
4587 	sc->last_op = "t4detach";
4588 	sc->last_op_thr = curthread;
4589 	sc->last_op_flags = 0;
4590 #endif
4591 	ADAPTER_UNLOCK(sc);
4592 }
4593 
4594 /*
4595  * {begin|end}_synchronized_op must be called from the same thread.
4596  */
4597 void
4598 end_synchronized_op(struct adapter *sc, int flags)
4599 {
4600 
4601 	if (flags & LOCK_HELD)
4602 		ADAPTER_LOCK_ASSERT_OWNED(sc);
4603 	else
4604 		ADAPTER_LOCK(sc);
4605 
4606 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
4607 	CLR_BUSY(sc);
4608 	wakeup(&sc->flags);
4609 	ADAPTER_UNLOCK(sc);
4610 }
4611 
4612 static int
4613 cxgbe_init_synchronized(struct vi_info *vi)
4614 {
4615 	struct port_info *pi = vi->pi;
4616 	struct adapter *sc = pi->adapter;
4617 	struct ifnet *ifp = vi->ifp;
4618 	int rc = 0, i;
4619 	struct sge_txq *txq;
4620 
4621 	ASSERT_SYNCHRONIZED_OP(sc);
4622 
4623 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
4624 		return (0);	/* already running */
4625 
4626 	if (!(sc->flags & FULL_INIT_DONE) &&
4627 	    ((rc = adapter_full_init(sc)) != 0))
4628 		return (rc);	/* error message displayed already */
4629 
4630 	if (!(vi->flags & VI_INIT_DONE) &&
4631 	    ((rc = vi_full_init(vi)) != 0))
4632 		return (rc); /* error message displayed already */
4633 
4634 	rc = update_mac_settings(ifp, XGMAC_ALL);
4635 	if (rc)
4636 		goto done;	/* error message displayed already */
4637 
4638 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
4639 	if (rc != 0) {
4640 		if_printf(ifp, "enable_vi failed: %d\n", rc);
4641 		goto done;
4642 	}
4643 
4644 	/*
4645 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
4646 	 * if this changes.
4647 	 */
4648 
4649 	for_each_txq(vi, i, txq) {
4650 		TXQ_LOCK(txq);
4651 		txq->eq.flags |= EQ_ENABLED;
4652 		TXQ_UNLOCK(txq);
4653 	}
4654 
4655 	/*
4656 	 * The first iq of the first port to come up is used for tracing.
4657 	 */
4658 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4659 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4660 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4661 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4662 		    V_QUEUENUMBER(sc->traceq));
4663 		pi->flags |= HAS_TRACEQ;
4664 	}
4665 
4666 	/* all ok */
4667 	PORT_LOCK(pi);
4668 	if (pi->up_vis++ == 0) {
4669 		t4_update_port_info(pi);
4670 		build_medialist(pi, &pi->media);
4671 		apply_l1cfg(pi);
4672 	}
4673 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
4674 
4675 	if (pi->nvi > 1 || sc->flags & IS_VF)
4676 		callout_reset(&vi->tick, hz, vi_tick, vi);
4677 	else
4678 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4679 	PORT_UNLOCK(pi);
4680 done:
4681 	if (rc != 0)
4682 		cxgbe_uninit_synchronized(vi);
4683 
4684 	return (rc);
4685 }
4686 
4687 /*
4688  * Idempotent.
4689  */
4690 static int
4691 cxgbe_uninit_synchronized(struct vi_info *vi)
4692 {
4693 	struct port_info *pi = vi->pi;
4694 	struct adapter *sc = pi->adapter;
4695 	struct ifnet *ifp = vi->ifp;
4696 	int rc, i;
4697 	struct sge_txq *txq;
4698 
4699 	ASSERT_SYNCHRONIZED_OP(sc);
4700 
4701 	if (!(vi->flags & VI_INIT_DONE)) {
4702 		if (__predict_false(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4703 			KASSERT(0, ("uninited VI is running"));
4704 			if_printf(ifp, "uninited VI with running ifnet.  "
4705 			    "vi->flags 0x%016lx, if_flags 0x%08x, "
4706 			    "if_drv_flags 0x%08x\n", vi->flags, ifp->if_flags,
4707 			    ifp->if_drv_flags);
4708 		}
4709 		return (0);
4710 	}
4711 
4712 	/*
4713 	 * Disable the VI so that all its data in either direction is discarded
4714 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4715 	 * tick) intact as the TP can deliver negative advice or data that it's
4716 	 * holding in its RAM (for an offloaded connection) even after the VI is
4717 	 * disabled.
4718 	 */
4719 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4720 	if (rc) {
4721 		if_printf(ifp, "disable_vi failed: %d\n", rc);
4722 		return (rc);
4723 	}
4724 
4725 	for_each_txq(vi, i, txq) {
4726 		TXQ_LOCK(txq);
4727 		txq->eq.flags &= ~EQ_ENABLED;
4728 		TXQ_UNLOCK(txq);
4729 	}
4730 
4731 	PORT_LOCK(pi);
4732 	if (pi->nvi > 1 || sc->flags & IS_VF)
4733 		callout_stop(&vi->tick);
4734 	else
4735 		callout_stop(&pi->tick);
4736 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4737 		PORT_UNLOCK(pi);
4738 		return (0);
4739 	}
4740 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4741 	pi->up_vis--;
4742 	if (pi->up_vis > 0) {
4743 		PORT_UNLOCK(pi);
4744 		return (0);
4745 	}
4746 
4747 	pi->link_cfg.link_ok = 0;
4748 	pi->link_cfg.speed = 0;
4749 	pi->link_cfg.link_down_rc = 255;
4750 	t4_os_link_changed(pi);
4751 	pi->old_link_cfg = pi->link_cfg;
4752 	PORT_UNLOCK(pi);
4753 
4754 	return (0);
4755 }
4756 
4757 /*
4758  * It is ok for this function to fail midway and return right away.  t4_detach
4759  * will walk the entire sc->irq list and clean up whatever is valid.
4760  */
4761 int
4762 t4_setup_intr_handlers(struct adapter *sc)
4763 {
4764 	int rc, rid, p, q, v;
4765 	char s[8];
4766 	struct irq *irq;
4767 	struct port_info *pi;
4768 	struct vi_info *vi;
4769 	struct sge *sge = &sc->sge;
4770 	struct sge_rxq *rxq;
4771 #ifdef TCP_OFFLOAD
4772 	struct sge_ofld_rxq *ofld_rxq;
4773 #endif
4774 #ifdef DEV_NETMAP
4775 	struct sge_nm_rxq *nm_rxq;
4776 #endif
4777 #ifdef RSS
4778 	int nbuckets = rss_getnumbuckets();
4779 #endif
4780 
4781 	/*
4782 	 * Setup interrupts.
4783 	 */
4784 	irq = &sc->irq[0];
4785 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
4786 	if (forwarding_intr_to_fwq(sc))
4787 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4788 
4789 	/* Multiple interrupts. */
4790 	if (sc->flags & IS_VF)
4791 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4792 		    ("%s: too few intr.", __func__));
4793 	else
4794 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4795 		    ("%s: too few intr.", __func__));
4796 
4797 	/* The first one is always error intr on PFs */
4798 	if (!(sc->flags & IS_VF)) {
4799 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4800 		if (rc != 0)
4801 			return (rc);
4802 		irq++;
4803 		rid++;
4804 	}
4805 
4806 	/* The second one is always the firmware event queue (first on VFs) */
4807 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4808 	if (rc != 0)
4809 		return (rc);
4810 	irq++;
4811 	rid++;
4812 
4813 	for_each_port(sc, p) {
4814 		pi = sc->port[p];
4815 		for_each_vi(pi, v, vi) {
4816 			vi->first_intr = rid - 1;
4817 
4818 			if (vi->nnmrxq > 0) {
4819 				int n = max(vi->nrxq, vi->nnmrxq);
4820 
4821 				rxq = &sge->rxq[vi->first_rxq];
4822 #ifdef DEV_NETMAP
4823 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4824 #endif
4825 				for (q = 0; q < n; q++) {
4826 					snprintf(s, sizeof(s), "%x%c%x", p,
4827 					    'a' + v, q);
4828 					if (q < vi->nrxq)
4829 						irq->rxq = rxq++;
4830 #ifdef DEV_NETMAP
4831 					if (q < vi->nnmrxq)
4832 						irq->nm_rxq = nm_rxq++;
4833 
4834 					if (irq->nm_rxq != NULL &&
4835 					    irq->rxq == NULL) {
4836 						/* Netmap rx only */
4837 						rc = t4_alloc_irq(sc, irq, rid,
4838 						    t4_nm_intr, irq->nm_rxq, s);
4839 					}
4840 					if (irq->nm_rxq != NULL &&
4841 					    irq->rxq != NULL) {
4842 						/* NIC and Netmap rx */
4843 						rc = t4_alloc_irq(sc, irq, rid,
4844 						    t4_vi_intr, irq, s);
4845 					}
4846 #endif
4847 					if (irq->rxq != NULL &&
4848 					    irq->nm_rxq == NULL) {
4849 						/* NIC rx only */
4850 						rc = t4_alloc_irq(sc, irq, rid,
4851 						    t4_intr, irq->rxq, s);
4852 					}
4853 					if (rc != 0)
4854 						return (rc);
4855 #ifdef RSS
4856 					if (q < vi->nrxq) {
4857 						bus_bind_intr(sc->dev, irq->res,
4858 						    rss_getcpu(q % nbuckets));
4859 					}
4860 #endif
4861 					irq++;
4862 					rid++;
4863 					vi->nintr++;
4864 				}
4865 			} else {
4866 				for_each_rxq(vi, q, rxq) {
4867 					snprintf(s, sizeof(s), "%x%c%x", p,
4868 					    'a' + v, q);
4869 					rc = t4_alloc_irq(sc, irq, rid,
4870 					    t4_intr, rxq, s);
4871 					if (rc != 0)
4872 						return (rc);
4873 #ifdef RSS
4874 					bus_bind_intr(sc->dev, irq->res,
4875 					    rss_getcpu(q % nbuckets));
4876 #endif
4877 					irq++;
4878 					rid++;
4879 					vi->nintr++;
4880 				}
4881 			}
4882 #ifdef TCP_OFFLOAD
4883 			for_each_ofld_rxq(vi, q, ofld_rxq) {
4884 				snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
4885 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
4886 				    ofld_rxq, s);
4887 				if (rc != 0)
4888 					return (rc);
4889 				irq++;
4890 				rid++;
4891 				vi->nintr++;
4892 			}
4893 #endif
4894 		}
4895 	}
4896 	MPASS(irq == &sc->irq[sc->intr_count]);
4897 
4898 	return (0);
4899 }
4900 
4901 int
4902 adapter_full_init(struct adapter *sc)
4903 {
4904 	int rc, i;
4905 #ifdef RSS
4906 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4907 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4908 #endif
4909 
4910 	ASSERT_SYNCHRONIZED_OP(sc);
4911 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4912 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
4913 	    ("%s: FULL_INIT_DONE already", __func__));
4914 
4915 	/*
4916 	 * queues that belong to the adapter (not any particular port).
4917 	 */
4918 	rc = t4_setup_adapter_queues(sc);
4919 	if (rc != 0)
4920 		goto done;
4921 
4922 	for (i = 0; i < nitems(sc->tq); i++) {
4923 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
4924 		    taskqueue_thread_enqueue, &sc->tq[i]);
4925 		if (sc->tq[i] == NULL) {
4926 			device_printf(sc->dev,
4927 			    "failed to allocate task queue %d\n", i);
4928 			rc = ENOMEM;
4929 			goto done;
4930 		}
4931 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
4932 		    device_get_nameunit(sc->dev), i);
4933 	}
4934 #ifdef RSS
4935 	MPASS(RSS_KEYSIZE == 40);
4936 	rss_getkey((void *)&raw_rss_key[0]);
4937 	for (i = 0; i < nitems(rss_key); i++) {
4938 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
4939 	}
4940 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
4941 #endif
4942 
4943 	if (!(sc->flags & IS_VF))
4944 		t4_intr_enable(sc);
4945 	sc->flags |= FULL_INIT_DONE;
4946 done:
4947 	if (rc != 0)
4948 		adapter_full_uninit(sc);
4949 
4950 	return (rc);
4951 }
4952 
4953 int
4954 adapter_full_uninit(struct adapter *sc)
4955 {
4956 	int i;
4957 
4958 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4959 
4960 	t4_teardown_adapter_queues(sc);
4961 
4962 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
4963 		taskqueue_free(sc->tq[i]);
4964 		sc->tq[i] = NULL;
4965 	}
4966 
4967 	sc->flags &= ~FULL_INIT_DONE;
4968 
4969 	return (0);
4970 }
4971 
4972 #ifdef RSS
4973 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
4974     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
4975     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
4976     RSS_HASHTYPE_RSS_UDP_IPV6)
4977 
4978 /* Translates kernel hash types to hardware. */
4979 static int
4980 hashconfig_to_hashen(int hashconfig)
4981 {
4982 	int hashen = 0;
4983 
4984 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
4985 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
4986 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
4987 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
4988 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
4989 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4990 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4991 	}
4992 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
4993 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4994 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4995 	}
4996 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
4997 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4998 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
4999 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5000 
5001 	return (hashen);
5002 }
5003 
5004 /* Translates hardware hash types to kernel. */
5005 static int
5006 hashen_to_hashconfig(int hashen)
5007 {
5008 	int hashconfig = 0;
5009 
5010 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
5011 		/*
5012 		 * If UDP hashing was enabled it must have been enabled for
5013 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
5014 		 * enabling any 4-tuple hash is nonsense configuration.
5015 		 */
5016 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5017 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
5018 
5019 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5020 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
5021 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5022 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
5023 	}
5024 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5025 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
5026 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5027 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
5028 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
5029 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
5030 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
5031 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
5032 
5033 	return (hashconfig);
5034 }
5035 #endif
5036 
5037 int
5038 vi_full_init(struct vi_info *vi)
5039 {
5040 	struct adapter *sc = vi->pi->adapter;
5041 	struct ifnet *ifp = vi->ifp;
5042 	uint16_t *rss;
5043 	struct sge_rxq *rxq;
5044 	int rc, i, j, hashen;
5045 #ifdef RSS
5046 	int nbuckets = rss_getnumbuckets();
5047 	int hashconfig = rss_gethashconfig();
5048 	int extra;
5049 #endif
5050 
5051 	ASSERT_SYNCHRONIZED_OP(sc);
5052 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
5053 	    ("%s: VI_INIT_DONE already", __func__));
5054 
5055 	sysctl_ctx_init(&vi->ctx);
5056 	vi->flags |= VI_SYSCTL_CTX;
5057 
5058 	/*
5059 	 * Allocate tx/rx/fl queues for this VI.
5060 	 */
5061 	rc = t4_setup_vi_queues(vi);
5062 	if (rc != 0)
5063 		goto done;	/* error message displayed already */
5064 
5065 	/*
5066 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
5067 	 */
5068 	if (vi->nrxq > vi->rss_size) {
5069 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
5070 		    "some queues will never receive traffic.\n", vi->nrxq,
5071 		    vi->rss_size);
5072 	} else if (vi->rss_size % vi->nrxq) {
5073 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
5074 		    "expect uneven traffic distribution.\n", vi->nrxq,
5075 		    vi->rss_size);
5076 	}
5077 #ifdef RSS
5078 	if (vi->nrxq != nbuckets) {
5079 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
5080 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
5081 	}
5082 #endif
5083 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
5084 	for (i = 0; i < vi->rss_size;) {
5085 #ifdef RSS
5086 		j = rss_get_indirection_to_bucket(i);
5087 		j %= vi->nrxq;
5088 		rxq = &sc->sge.rxq[vi->first_rxq + j];
5089 		rss[i++] = rxq->iq.abs_id;
5090 #else
5091 		for_each_rxq(vi, j, rxq) {
5092 			rss[i++] = rxq->iq.abs_id;
5093 			if (i == vi->rss_size)
5094 				break;
5095 		}
5096 #endif
5097 	}
5098 
5099 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
5100 	    vi->rss_size);
5101 	if (rc != 0) {
5102 		if_printf(ifp, "rss_config failed: %d\n", rc);
5103 		goto done;
5104 	}
5105 
5106 #ifdef RSS
5107 	hashen = hashconfig_to_hashen(hashconfig);
5108 
5109 	/*
5110 	 * We may have had to enable some hashes even though the global config
5111 	 * wants them disabled.  This is a potential problem that must be
5112 	 * reported to the user.
5113 	 */
5114 	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
5115 
5116 	/*
5117 	 * If we consider only the supported hash types, then the enabled hashes
5118 	 * are a superset of the requested hashes.  In other words, there cannot
5119 	 * be any supported hash that was requested but not enabled, but there
5120 	 * can be hashes that were not requested but had to be enabled.
5121 	 */
5122 	extra &= SUPPORTED_RSS_HASHTYPES;
5123 	MPASS((extra & hashconfig) == 0);
5124 
5125 	if (extra) {
5126 		if_printf(ifp,
5127 		    "global RSS config (0x%x) cannot be accommodated.\n",
5128 		    hashconfig);
5129 	}
5130 	if (extra & RSS_HASHTYPE_RSS_IPV4)
5131 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
5132 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
5133 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
5134 	if (extra & RSS_HASHTYPE_RSS_IPV6)
5135 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
5136 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
5137 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
5138 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
5139 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
5140 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
5141 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
5142 #else
5143 	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
5144 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
5145 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5146 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
5147 #endif
5148 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
5149 	if (rc != 0) {
5150 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
5151 		goto done;
5152 	}
5153 
5154 	vi->rss = rss;
5155 	vi->flags |= VI_INIT_DONE;
5156 done:
5157 	if (rc != 0)
5158 		vi_full_uninit(vi);
5159 
5160 	return (rc);
5161 }
5162 
5163 /*
5164  * Idempotent.
5165  */
5166 int
5167 vi_full_uninit(struct vi_info *vi)
5168 {
5169 	struct port_info *pi = vi->pi;
5170 	struct adapter *sc = pi->adapter;
5171 	int i;
5172 	struct sge_rxq *rxq;
5173 	struct sge_txq *txq;
5174 #ifdef TCP_OFFLOAD
5175 	struct sge_ofld_rxq *ofld_rxq;
5176 	struct sge_wrq *ofld_txq;
5177 #endif
5178 
5179 	if (vi->flags & VI_INIT_DONE) {
5180 
5181 		/* Need to quiesce queues.  */
5182 
5183 		/* XXX: Only for the first VI? */
5184 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
5185 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
5186 
5187 		for_each_txq(vi, i, txq) {
5188 			quiesce_txq(sc, txq);
5189 		}
5190 
5191 #ifdef TCP_OFFLOAD
5192 		for_each_ofld_txq(vi, i, ofld_txq) {
5193 			quiesce_wrq(sc, ofld_txq);
5194 		}
5195 #endif
5196 
5197 		for_each_rxq(vi, i, rxq) {
5198 			quiesce_iq(sc, &rxq->iq);
5199 			quiesce_fl(sc, &rxq->fl);
5200 		}
5201 
5202 #ifdef TCP_OFFLOAD
5203 		for_each_ofld_rxq(vi, i, ofld_rxq) {
5204 			quiesce_iq(sc, &ofld_rxq->iq);
5205 			quiesce_fl(sc, &ofld_rxq->fl);
5206 		}
5207 #endif
5208 		free(vi->rss, M_CXGBE);
5209 		free(vi->nm_rss, M_CXGBE);
5210 	}
5211 
5212 	t4_teardown_vi_queues(vi);
5213 	vi->flags &= ~VI_INIT_DONE;
5214 
5215 	return (0);
5216 }
5217 
5218 static void
5219 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
5220 {
5221 	struct sge_eq *eq = &txq->eq;
5222 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
5223 
5224 	(void) sc;	/* unused */
5225 
5226 #ifdef INVARIANTS
5227 	TXQ_LOCK(txq);
5228 	MPASS((eq->flags & EQ_ENABLED) == 0);
5229 	TXQ_UNLOCK(txq);
5230 #endif
5231 
5232 	/* Wait for the mp_ring to empty. */
5233 	while (!mp_ring_is_idle(txq->r)) {
5234 		mp_ring_check_drainage(txq->r, 0);
5235 		pause("rquiesce", 1);
5236 	}
5237 
5238 	/* Then wait for the hardware to finish. */
5239 	while (spg->cidx != htobe16(eq->pidx))
5240 		pause("equiesce", 1);
5241 
5242 	/* Finally, wait for the driver to reclaim all descriptors. */
5243 	while (eq->cidx != eq->pidx)
5244 		pause("dquiesce", 1);
5245 }
5246 
5247 static void
5248 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
5249 {
5250 
5251 	/* XXXTX */
5252 }
5253 
5254 static void
5255 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
5256 {
5257 	(void) sc;	/* unused */
5258 
5259 	/* Synchronize with the interrupt handler */
5260 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
5261 		pause("iqfree", 1);
5262 }
5263 
5264 static void
5265 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
5266 {
5267 	mtx_lock(&sc->sfl_lock);
5268 	FL_LOCK(fl);
5269 	fl->flags |= FL_DOOMED;
5270 	FL_UNLOCK(fl);
5271 	callout_stop(&sc->sfl_callout);
5272 	mtx_unlock(&sc->sfl_lock);
5273 
5274 	KASSERT((fl->flags & FL_STARVING) == 0,
5275 	    ("%s: still starving", __func__));
5276 }
5277 
5278 static int
5279 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
5280     driver_intr_t *handler, void *arg, char *name)
5281 {
5282 	int rc;
5283 
5284 	irq->rid = rid;
5285 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
5286 	    RF_SHAREABLE | RF_ACTIVE);
5287 	if (irq->res == NULL) {
5288 		device_printf(sc->dev,
5289 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
5290 		return (ENOMEM);
5291 	}
5292 
5293 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
5294 	    NULL, handler, arg, &irq->tag);
5295 	if (rc != 0) {
5296 		device_printf(sc->dev,
5297 		    "failed to setup interrupt for rid %d, name %s: %d\n",
5298 		    rid, name, rc);
5299 	} else if (name)
5300 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
5301 
5302 	return (rc);
5303 }
5304 
5305 static int
5306 t4_free_irq(struct adapter *sc, struct irq *irq)
5307 {
5308 	if (irq->tag)
5309 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
5310 	if (irq->res)
5311 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
5312 
5313 	bzero(irq, sizeof(*irq));
5314 
5315 	return (0);
5316 }
5317 
5318 static void
5319 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
5320 {
5321 
5322 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
5323 	t4_get_regs(sc, buf, regs->len);
5324 }
5325 
5326 #define	A_PL_INDIR_CMD	0x1f8
5327 
5328 #define	S_PL_AUTOINC	31
5329 #define	M_PL_AUTOINC	0x1U
5330 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
5331 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
5332 
5333 #define	S_PL_VFID	20
5334 #define	M_PL_VFID	0xffU
5335 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
5336 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
5337 
5338 #define	S_PL_ADDR	0
5339 #define	M_PL_ADDR	0xfffffU
5340 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
5341 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
5342 
5343 #define	A_PL_INDIR_DATA	0x1fc
5344 
5345 static uint64_t
5346 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
5347 {
5348 	u32 stats[2];
5349 
5350 	mtx_assert(&sc->reg_lock, MA_OWNED);
5351 	if (sc->flags & IS_VF) {
5352 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
5353 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
5354 	} else {
5355 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5356 		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5357 		    V_PL_ADDR(VF_MPS_REG(reg)));
5358 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
5359 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
5360 	}
5361 	return (((uint64_t)stats[1]) << 32 | stats[0]);
5362 }
5363 
5364 static void
5365 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
5366     struct fw_vi_stats_vf *stats)
5367 {
5368 
5369 #define GET_STAT(name) \
5370 	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
5371 
5372 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
5373 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
5374 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
5375 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
5376 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
5377 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
5378 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
5379 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
5380 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
5381 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
5382 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
5383 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
5384 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
5385 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
5386 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
5387 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
5388 
5389 #undef GET_STAT
5390 }
5391 
5392 static void
5393 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
5394 {
5395 	int reg;
5396 
5397 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5398 	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5399 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
5400 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
5401 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
5402 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
5403 }
5404 
5405 static void
5406 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
5407 {
5408 	struct timeval tv;
5409 	const struct timeval interval = {0, 250000};	/* 250ms */
5410 
5411 	if (!(vi->flags & VI_INIT_DONE))
5412 		return;
5413 
5414 	getmicrotime(&tv);
5415 	timevalsub(&tv, &interval);
5416 	if (timevalcmp(&tv, &vi->last_refreshed, <))
5417 		return;
5418 
5419 	mtx_lock(&sc->reg_lock);
5420 	t4_get_vi_stats(sc, vi->viid, &vi->stats);
5421 	getmicrotime(&vi->last_refreshed);
5422 	mtx_unlock(&sc->reg_lock);
5423 }
5424 
5425 static void
5426 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5427 {
5428 	u_int i, v, tnl_cong_drops, bg_map;
5429 	struct timeval tv;
5430 	const struct timeval interval = {0, 250000};	/* 250ms */
5431 
5432 	getmicrotime(&tv);
5433 	timevalsub(&tv, &interval);
5434 	if (timevalcmp(&tv, &pi->last_refreshed, <))
5435 		return;
5436 
5437 	tnl_cong_drops = 0;
5438 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5439 	bg_map = pi->mps_bg_map;
5440 	while (bg_map) {
5441 		i = ffs(bg_map) - 1;
5442 		mtx_lock(&sc->reg_lock);
5443 		t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5444 		    A_TP_MIB_TNL_CNG_DROP_0 + i);
5445 		mtx_unlock(&sc->reg_lock);
5446 		tnl_cong_drops += v;
5447 		bg_map &= ~(1 << i);
5448 	}
5449 	pi->tnl_cong_drops = tnl_cong_drops;
5450 	getmicrotime(&pi->last_refreshed);
5451 }
5452 
5453 static void
5454 cxgbe_tick(void *arg)
5455 {
5456 	struct port_info *pi = arg;
5457 	struct adapter *sc = pi->adapter;
5458 
5459 	PORT_LOCK_ASSERT_OWNED(pi);
5460 	cxgbe_refresh_stats(sc, pi);
5461 
5462 	callout_schedule(&pi->tick, hz);
5463 }
5464 
5465 void
5466 vi_tick(void *arg)
5467 {
5468 	struct vi_info *vi = arg;
5469 	struct adapter *sc = vi->pi->adapter;
5470 
5471 	vi_refresh_stats(sc, vi);
5472 
5473 	callout_schedule(&vi->tick, hz);
5474 }
5475 
5476 /*
5477  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5478  */
5479 static char *caps_decoder[] = {
5480 	"\20\001IPMI\002NCSI",				/* 0: NBM */
5481 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
5482 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
5483 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
5484 	    "\006HASHFILTER\007ETHOFLD",
5485 	"\20\001TOE",					/* 4: TOE */
5486 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
5487 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
5488 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5489 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5490 	    "\007T10DIF"
5491 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5492 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
5493 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
5494 		    "\004PO_INITIATOR\005PO_TARGET",
5495 };
5496 
5497 void
5498 t4_sysctls(struct adapter *sc)
5499 {
5500 	struct sysctl_ctx_list *ctx;
5501 	struct sysctl_oid *oid;
5502 	struct sysctl_oid_list *children, *c0;
5503 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5504 
5505 	ctx = device_get_sysctl_ctx(sc->dev);
5506 
5507 	/*
5508 	 * dev.t4nex.X.
5509 	 */
5510 	oid = device_get_sysctl_tree(sc->dev);
5511 	c0 = children = SYSCTL_CHILDREN(oid);
5512 
5513 	sc->sc_do_rxcopy = 1;
5514 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5515 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5516 
5517 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5518 	    sc->params.nports, "# of ports");
5519 
5520 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5521 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, (uintptr_t)&sc->doorbells,
5522 	    sysctl_bitfield_8b, "A", "available doorbells");
5523 
5524 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5525 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
5526 
5527 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5528 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5529 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5530 	    "interrupt holdoff timer values (us)");
5531 
5532 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5533 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5534 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5535 	    "interrupt holdoff packet counter values");
5536 
5537 	t4_sge_sysctls(sc, ctx, children);
5538 
5539 	sc->lro_timeout = 100;
5540 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5541 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5542 
5543 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5544 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
5545 
5546 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5547 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5548 
5549 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5550 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5551 
5552 	if (sc->flags & IS_VF)
5553 		return;
5554 
5555 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5556 	    NULL, chip_rev(sc), "chip hardware revision");
5557 
5558 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5559 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5560 
5561 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5562 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5563 
5564 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5565 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5566 
5567 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5568 	    CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5569 
5570 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5571 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5572 
5573 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5574 	    sc->er_version, 0, "expansion ROM version");
5575 
5576 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5577 	    sc->bs_version, 0, "bootstrap firmware version");
5578 
5579 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5580 	    NULL, sc->params.scfg_vers, "serial config version");
5581 
5582 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5583 	    NULL, sc->params.vpd_vers, "VPD version");
5584 
5585 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5586 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5587 
5588 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5589 	    sc->cfcsum, "config file checksum");
5590 
5591 #define SYSCTL_CAP(name, n, text) \
5592 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5593 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], (uintptr_t)&sc->name, \
5594 	    sysctl_bitfield_16b, "A", "available " text " capabilities")
5595 
5596 	SYSCTL_CAP(nbmcaps, 0, "NBM");
5597 	SYSCTL_CAP(linkcaps, 1, "link");
5598 	SYSCTL_CAP(switchcaps, 2, "switch");
5599 	SYSCTL_CAP(niccaps, 3, "NIC");
5600 	SYSCTL_CAP(toecaps, 4, "TCP offload");
5601 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
5602 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5603 	SYSCTL_CAP(cryptocaps, 7, "crypto");
5604 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
5605 #undef SYSCTL_CAP
5606 
5607 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5608 	    NULL, sc->tids.nftids, "number of filters");
5609 
5610 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5611 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5612 	    "chip temperature (in Celsius)");
5613 
5614 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "loadavg", CTLTYPE_STRING |
5615 	    CTLFLAG_RD, sc, 0, sysctl_loadavg, "A",
5616 	    "microprocessor load averages (debug firmwares only)");
5617 
5618 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
5619 	    &sc->params.core_vdd, 0, "core Vdd (in mV)");
5620 
5621 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "local_cpus",
5622 	    CTLTYPE_STRING | CTLFLAG_RD, sc, LOCAL_CPUS,
5623 	    sysctl_cpus, "A", "local CPUs");
5624 
5625 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "intr_cpus",
5626 	    CTLTYPE_STRING | CTLFLAG_RD, sc, INTR_CPUS,
5627 	    sysctl_cpus, "A", "preferred CPUs for interrupts");
5628 
5629 	/*
5630 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
5631 	 */
5632 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5633 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5634 	    "logs and miscellaneous information");
5635 	children = SYSCTL_CHILDREN(oid);
5636 
5637 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
5638 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5639 	    sysctl_cctrl, "A", "congestion control");
5640 
5641 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
5642 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5643 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
5644 
5645 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
5646 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
5647 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
5648 
5649 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
5650 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
5651 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
5652 
5653 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
5654 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
5655 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
5656 
5657 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
5658 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
5659 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
5660 
5661 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
5662 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
5663 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
5664 
5665 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
5666 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5667 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
5668 	    "A", "CIM logic analyzer");
5669 
5670 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
5671 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5672 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
5673 
5674 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
5675 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
5676 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
5677 
5678 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
5679 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
5680 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
5681 
5682 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
5683 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
5684 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
5685 
5686 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5687 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5688 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5689 
5690 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5691 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5692 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5693 
5694 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5695 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5696 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5697 
5698 	if (chip_id(sc) > CHELSIO_T4) {
5699 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5700 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5701 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5702 
5703 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5704 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5705 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5706 	}
5707 
5708 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5709 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5710 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5711 
5712 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5713 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5714 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
5715 
5716 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5717 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5718 	    sysctl_cpl_stats, "A", "CPL statistics");
5719 
5720 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5721 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5722 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5723 
5724 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5725 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5726 	    sysctl_devlog, "A", "firmware's device log");
5727 
5728 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5729 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5730 	    sysctl_fcoe_stats, "A", "FCoE statistics");
5731 
5732 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5733 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5734 	    sysctl_hw_sched, "A", "hardware scheduler ");
5735 
5736 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5737 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5738 	    sysctl_l2t, "A", "hardware L2 table");
5739 
5740 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "smt",
5741 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5742 	    sysctl_smt, "A", "hardware source MAC table");
5743 
5744 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5745 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5746 	    sysctl_lb_stats, "A", "loopback statistics");
5747 
5748 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5749 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5750 	    sysctl_meminfo, "A", "memory regions");
5751 
5752 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5753 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5754 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5755 	    "A", "MPS TCAM entries");
5756 
5757 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5758 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5759 	    sysctl_path_mtus, "A", "path MTUs");
5760 
5761 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5762 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5763 	    sysctl_pm_stats, "A", "PM statistics");
5764 
5765 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5766 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5767 	    sysctl_rdma_stats, "A", "RDMA statistics");
5768 
5769 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5770 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5771 	    sysctl_tcp_stats, "A", "TCP statistics");
5772 
5773 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5774 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5775 	    sysctl_tids, "A", "TID information");
5776 
5777 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5778 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5779 	    sysctl_tp_err_stats, "A", "TP error statistics");
5780 
5781 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5782 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5783 	    "TP logic analyzer event capture mask");
5784 
5785 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5786 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5787 	    sysctl_tp_la, "A", "TP logic analyzer");
5788 
5789 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5790 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5791 	    sysctl_tx_rate, "A", "Tx rate");
5792 
5793 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5794 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5795 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5796 
5797 	if (chip_id(sc) >= CHELSIO_T5) {
5798 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5799 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5800 		    sysctl_wcwr_stats, "A", "write combined work requests");
5801 	}
5802 
5803 #ifdef TCP_OFFLOAD
5804 	if (is_offload(sc)) {
5805 		int i;
5806 		char s[4];
5807 
5808 		/*
5809 		 * dev.t4nex.X.toe.
5810 		 */
5811 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5812 		    NULL, "TOE parameters");
5813 		children = SYSCTL_CHILDREN(oid);
5814 
5815 		sc->tt.cong_algorithm = -1;
5816 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
5817 		    CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
5818 		    "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
5819 		    "3 = highspeed)");
5820 
5821 		sc->tt.sndbuf = 256 * 1024;
5822 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5823 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
5824 
5825 		sc->tt.ddp = 0;
5826 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5827 		    &sc->tt.ddp, 0, "DDP allowed");
5828 
5829 		sc->tt.rx_coalesce = 1;
5830 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5831 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5832 
5833 		sc->tt.tls = 0;
5834 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tls", CTLFLAG_RW,
5835 		    &sc->tt.tls, 0, "Inline TLS allowed");
5836 
5837 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls_rx_ports",
5838 		    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tls_rx_ports,
5839 		    "I", "TCP ports that use inline TLS+TOE RX");
5840 
5841 		sc->tt.tx_align = 1;
5842 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5843 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5844 
5845 		sc->tt.tx_zcopy = 0;
5846 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
5847 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
5848 		    "Enable zero-copy aio_write(2)");
5849 
5850 		sc->tt.cop_managed_offloading = !!t4_cop_managed_offloading;
5851 		SYSCTL_ADD_INT(ctx, children, OID_AUTO,
5852 		    "cop_managed_offloading", CTLFLAG_RW,
5853 		    &sc->tt.cop_managed_offloading, 0,
5854 		    "COP (Connection Offload Policy) controls all TOE offload");
5855 
5856 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
5857 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
5858 		    "TP timer tick (us)");
5859 
5860 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
5861 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
5862 		    "TCP timestamp tick (us)");
5863 
5864 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5865 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5866 		    "DACK tick (us)");
5867 
5868 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5869 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5870 		    "IU", "DACK timer (us)");
5871 
5872 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5873 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5874 		    sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
5875 
5876 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5877 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5878 		    sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
5879 
5880 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5881 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5882 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
5883 
5884 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5885 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5886 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
5887 
5888 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5889 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5890 		    sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
5891 
5892 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
5893 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
5894 		    sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
5895 
5896 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
5897 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
5898 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
5899 
5900 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
5901 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
5902 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
5903 
5904 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
5905 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
5906 		    sysctl_tp_shift_cnt, "IU",
5907 		    "Number of SYN retransmissions before abort");
5908 
5909 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
5910 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
5911 		    sysctl_tp_shift_cnt, "IU",
5912 		    "Number of retransmissions before abort");
5913 
5914 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
5915 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
5916 		    sysctl_tp_shift_cnt, "IU",
5917 		    "Number of keepalive probes before abort");
5918 
5919 		oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
5920 		    CTLFLAG_RD, NULL, "TOE retransmit backoffs");
5921 		children = SYSCTL_CHILDREN(oid);
5922 		for (i = 0; i < 16; i++) {
5923 			snprintf(s, sizeof(s), "%u", i);
5924 			SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
5925 			    CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
5926 			    "IU", "TOE retransmit backoff");
5927 		}
5928 	}
5929 #endif
5930 }
5931 
5932 void
5933 vi_sysctls(struct vi_info *vi)
5934 {
5935 	struct sysctl_ctx_list *ctx;
5936 	struct sysctl_oid *oid;
5937 	struct sysctl_oid_list *children;
5938 
5939 	ctx = device_get_sysctl_ctx(vi->dev);
5940 
5941 	/*
5942 	 * dev.v?(cxgbe|cxl).X.
5943 	 */
5944 	oid = device_get_sysctl_tree(vi->dev);
5945 	children = SYSCTL_CHILDREN(oid);
5946 
5947 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
5948 	    vi->viid, "VI identifer");
5949 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
5950 	    &vi->nrxq, 0, "# of rx queues");
5951 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5952 	    &vi->ntxq, 0, "# of tx queues");
5953 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5954 	    &vi->first_rxq, 0, "index of first rx queue");
5955 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5956 	    &vi->first_txq, 0, "index of first tx queue");
5957 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
5958 	    vi->rss_size, "size of RSS indirection table");
5959 
5960 	if (IS_MAIN_VI(vi)) {
5961 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
5962 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
5963 		    "Reserve queue 0 for non-flowid packets");
5964 	}
5965 
5966 #ifdef TCP_OFFLOAD
5967 	if (vi->nofldrxq != 0) {
5968 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5969 		    &vi->nofldrxq, 0,
5970 		    "# of rx queues for offloaded TCP connections");
5971 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5972 		    &vi->nofldtxq, 0,
5973 		    "# of tx queues for offloaded TCP connections");
5974 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5975 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
5976 		    "index of first TOE rx queue");
5977 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5978 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
5979 		    "index of first TOE tx queue");
5980 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
5981 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5982 		    sysctl_holdoff_tmr_idx_ofld, "I",
5983 		    "holdoff timer index for TOE queues");
5984 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
5985 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5986 		    sysctl_holdoff_pktc_idx_ofld, "I",
5987 		    "holdoff packet counter index for TOE queues");
5988 	}
5989 #endif
5990 #ifdef DEV_NETMAP
5991 	if (vi->nnmrxq != 0) {
5992 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5993 		    &vi->nnmrxq, 0, "# of netmap rx queues");
5994 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5995 		    &vi->nnmtxq, 0, "# of netmap tx queues");
5996 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5997 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
5998 		    "index of first netmap rx queue");
5999 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
6000 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
6001 		    "index of first netmap tx queue");
6002 	}
6003 #endif
6004 
6005 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
6006 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
6007 	    "holdoff timer index");
6008 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
6009 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
6010 	    "holdoff packet counter index");
6011 
6012 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
6013 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
6014 	    "rx queue size");
6015 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
6016 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
6017 	    "tx queue size");
6018 }
6019 
6020 static void
6021 cxgbe_sysctls(struct port_info *pi)
6022 {
6023 	struct sysctl_ctx_list *ctx;
6024 	struct sysctl_oid *oid;
6025 	struct sysctl_oid_list *children, *children2;
6026 	struct adapter *sc = pi->adapter;
6027 	int i;
6028 	char name[16];
6029 	static char *tc_flags = {"\20\1USER\2SYNC\3ASYNC\4ERR"};
6030 
6031 	ctx = device_get_sysctl_ctx(pi->dev);
6032 
6033 	/*
6034 	 * dev.cxgbe.X.
6035 	 */
6036 	oid = device_get_sysctl_tree(pi->dev);
6037 	children = SYSCTL_CHILDREN(oid);
6038 
6039 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
6040 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
6041 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
6042 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
6043 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
6044 		    "PHY temperature (in Celsius)");
6045 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
6046 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
6047 		    "PHY firmware version");
6048 	}
6049 
6050 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
6051 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
6052 	    "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
6053 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
6054 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
6055 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
6056 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
6057 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
6058 	    "autonegotiation (-1 = not supported)");
6059 
6060 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
6061 	    port_top_speed(pi), "max speed (in Gbps)");
6062 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
6063 	    pi->mps_bg_map, "MPS buffer group map");
6064 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
6065 	    NULL, pi->rx_e_chan_map, "TP rx e-channel map");
6066 
6067 	if (sc->flags & IS_VF)
6068 		return;
6069 
6070 	/*
6071 	 * dev.(cxgbe|cxl).X.tc.
6072 	 */
6073 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
6074 	    "Tx scheduler traffic classes (cl_rl)");
6075 	children2 = SYSCTL_CHILDREN(oid);
6076 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "pktsize",
6077 	    CTLFLAG_RW, &pi->sched_params->pktsize, 0,
6078 	    "pktsize for per-flow cl-rl (0 means up to the driver )");
6079 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "burstsize",
6080 	    CTLFLAG_RW, &pi->sched_params->burstsize, 0,
6081 	    "burstsize for per-flow cl-rl (0 means up to the driver)");
6082 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
6083 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
6084 
6085 		snprintf(name, sizeof(name), "%d", i);
6086 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
6087 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
6088 		    "traffic class"));
6089 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "flags",
6090 		    CTLTYPE_STRING | CTLFLAG_RD, tc_flags, (uintptr_t)&tc->flags,
6091 		    sysctl_bitfield_8b, "A", "flags");
6092 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
6093 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
6094 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
6095 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
6096 		    sysctl_tc_params, "A", "traffic class parameters");
6097 	}
6098 
6099 	/*
6100 	 * dev.cxgbe.X.stats.
6101 	 */
6102 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
6103 	    NULL, "port statistics");
6104 	children = SYSCTL_CHILDREN(oid);
6105 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
6106 	    &pi->tx_parse_error, 0,
6107 	    "# of tx packets with invalid length or # of segments");
6108 
6109 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
6110 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
6111 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
6112 	    sysctl_handle_t4_reg64, "QU", desc)
6113 
6114 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
6115 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
6116 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
6117 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
6118 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
6119 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
6120 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
6121 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
6122 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
6123 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
6124 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
6125 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
6126 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
6127 	    "# of tx frames in this range",
6128 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
6129 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
6130 	    "# of tx frames in this range",
6131 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
6132 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
6133 	    "# of tx frames in this range",
6134 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
6135 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
6136 	    "# of tx frames in this range",
6137 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
6138 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
6139 	    "# of tx frames in this range",
6140 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
6141 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
6142 	    "# of tx frames in this range",
6143 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
6144 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
6145 	    "# of tx frames in this range",
6146 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
6147 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
6148 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
6149 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
6150 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
6151 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
6152 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
6153 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
6154 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
6155 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
6156 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
6157 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
6158 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
6159 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
6160 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
6161 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
6162 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
6163 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
6164 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
6165 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
6166 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
6167 
6168 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
6169 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
6170 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
6171 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
6172 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
6173 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
6174 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
6175 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
6176 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
6177 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
6178 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
6179 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
6180 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
6181 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
6182 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
6183 	    "# of frames received with bad FCS",
6184 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
6185 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
6186 	    "# of frames received with length error",
6187 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
6188 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
6189 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
6190 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
6191 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
6192 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
6193 	    "# of rx frames in this range",
6194 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
6195 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
6196 	    "# of rx frames in this range",
6197 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
6198 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
6199 	    "# of rx frames in this range",
6200 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
6201 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
6202 	    "# of rx frames in this range",
6203 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
6204 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
6205 	    "# of rx frames in this range",
6206 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
6207 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
6208 	    "# of rx frames in this range",
6209 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
6210 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
6211 	    "# of rx frames in this range",
6212 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
6213 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
6214 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
6215 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
6216 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
6217 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
6218 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
6219 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
6220 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
6221 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
6222 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
6223 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
6224 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
6225 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
6226 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
6227 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
6228 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
6229 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
6230 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
6231 
6232 #undef SYSCTL_ADD_T4_REG64
6233 
6234 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
6235 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
6236 	    &pi->stats.name, desc)
6237 
6238 	/* We get these from port_stats and they may be stale by up to 1s */
6239 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
6240 	    "# drops due to buffer-group 0 overflows");
6241 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
6242 	    "# drops due to buffer-group 1 overflows");
6243 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
6244 	    "# drops due to buffer-group 2 overflows");
6245 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
6246 	    "# drops due to buffer-group 3 overflows");
6247 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
6248 	    "# of buffer-group 0 truncated packets");
6249 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
6250 	    "# of buffer-group 1 truncated packets");
6251 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
6252 	    "# of buffer-group 2 truncated packets");
6253 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
6254 	    "# of buffer-group 3 truncated packets");
6255 
6256 #undef SYSCTL_ADD_T4_PORTSTAT
6257 
6258 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_records",
6259 	    CTLFLAG_RD, &pi->tx_tls_records,
6260 	    "# of TLS records transmitted");
6261 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_octets",
6262 	    CTLFLAG_RD, &pi->tx_tls_octets,
6263 	    "# of payload octets in transmitted TLS records");
6264 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_records",
6265 	    CTLFLAG_RD, &pi->rx_tls_records,
6266 	    "# of TLS records received");
6267 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_octets",
6268 	    CTLFLAG_RD, &pi->rx_tls_octets,
6269 	    "# of payload octets in received TLS records");
6270 }
6271 
6272 static int
6273 sysctl_int_array(SYSCTL_HANDLER_ARGS)
6274 {
6275 	int rc, *i, space = 0;
6276 	struct sbuf sb;
6277 
6278 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
6279 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
6280 		if (space)
6281 			sbuf_printf(&sb, " ");
6282 		sbuf_printf(&sb, "%d", *i);
6283 		space = 1;
6284 	}
6285 	rc = sbuf_finish(&sb);
6286 	sbuf_delete(&sb);
6287 	return (rc);
6288 }
6289 
6290 static int
6291 sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS)
6292 {
6293 	int rc;
6294 	struct sbuf *sb;
6295 
6296 	rc = sysctl_wire_old_buffer(req, 0);
6297 	if (rc != 0)
6298 		return(rc);
6299 
6300 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6301 	if (sb == NULL)
6302 		return (ENOMEM);
6303 
6304 	sbuf_printf(sb, "%b", *(uint8_t *)(uintptr_t)arg2, (char *)arg1);
6305 	rc = sbuf_finish(sb);
6306 	sbuf_delete(sb);
6307 
6308 	return (rc);
6309 }
6310 
6311 static int
6312 sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS)
6313 {
6314 	int rc;
6315 	struct sbuf *sb;
6316 
6317 	rc = sysctl_wire_old_buffer(req, 0);
6318 	if (rc != 0)
6319 		return(rc);
6320 
6321 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6322 	if (sb == NULL)
6323 		return (ENOMEM);
6324 
6325 	sbuf_printf(sb, "%b", *(uint16_t *)(uintptr_t)arg2, (char *)arg1);
6326 	rc = sbuf_finish(sb);
6327 	sbuf_delete(sb);
6328 
6329 	return (rc);
6330 }
6331 
6332 static int
6333 sysctl_btphy(SYSCTL_HANDLER_ARGS)
6334 {
6335 	struct port_info *pi = arg1;
6336 	int op = arg2;
6337 	struct adapter *sc = pi->adapter;
6338 	u_int v;
6339 	int rc;
6340 
6341 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
6342 	if (rc)
6343 		return (rc);
6344 	/* XXX: magic numbers */
6345 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
6346 	    &v);
6347 	end_synchronized_op(sc, 0);
6348 	if (rc)
6349 		return (rc);
6350 	if (op == 0)
6351 		v /= 256;
6352 
6353 	rc = sysctl_handle_int(oidp, &v, 0, req);
6354 	return (rc);
6355 }
6356 
6357 static int
6358 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
6359 {
6360 	struct vi_info *vi = arg1;
6361 	int rc, val;
6362 
6363 	val = vi->rsrv_noflowq;
6364 	rc = sysctl_handle_int(oidp, &val, 0, req);
6365 	if (rc != 0 || req->newptr == NULL)
6366 		return (rc);
6367 
6368 	if ((val >= 1) && (vi->ntxq > 1))
6369 		vi->rsrv_noflowq = 1;
6370 	else
6371 		vi->rsrv_noflowq = 0;
6372 
6373 	return (rc);
6374 }
6375 
6376 static int
6377 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
6378 {
6379 	struct vi_info *vi = arg1;
6380 	struct adapter *sc = vi->pi->adapter;
6381 	int idx, rc, i;
6382 	struct sge_rxq *rxq;
6383 	uint8_t v;
6384 
6385 	idx = vi->tmr_idx;
6386 
6387 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6388 	if (rc != 0 || req->newptr == NULL)
6389 		return (rc);
6390 
6391 	if (idx < 0 || idx >= SGE_NTIMERS)
6392 		return (EINVAL);
6393 
6394 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6395 	    "t4tmr");
6396 	if (rc)
6397 		return (rc);
6398 
6399 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
6400 	for_each_rxq(vi, i, rxq) {
6401 #ifdef atomic_store_rel_8
6402 		atomic_store_rel_8(&rxq->iq.intr_params, v);
6403 #else
6404 		rxq->iq.intr_params = v;
6405 #endif
6406 	}
6407 	vi->tmr_idx = idx;
6408 
6409 	end_synchronized_op(sc, LOCK_HELD);
6410 	return (0);
6411 }
6412 
6413 static int
6414 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
6415 {
6416 	struct vi_info *vi = arg1;
6417 	struct adapter *sc = vi->pi->adapter;
6418 	int idx, rc;
6419 
6420 	idx = vi->pktc_idx;
6421 
6422 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6423 	if (rc != 0 || req->newptr == NULL)
6424 		return (rc);
6425 
6426 	if (idx < -1 || idx >= SGE_NCOUNTERS)
6427 		return (EINVAL);
6428 
6429 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6430 	    "t4pktc");
6431 	if (rc)
6432 		return (rc);
6433 
6434 	if (vi->flags & VI_INIT_DONE)
6435 		rc = EBUSY; /* cannot be changed once the queues are created */
6436 	else
6437 		vi->pktc_idx = idx;
6438 
6439 	end_synchronized_op(sc, LOCK_HELD);
6440 	return (rc);
6441 }
6442 
6443 static int
6444 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
6445 {
6446 	struct vi_info *vi = arg1;
6447 	struct adapter *sc = vi->pi->adapter;
6448 	int qsize, rc;
6449 
6450 	qsize = vi->qsize_rxq;
6451 
6452 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6453 	if (rc != 0 || req->newptr == NULL)
6454 		return (rc);
6455 
6456 	if (qsize < 128 || (qsize & 7))
6457 		return (EINVAL);
6458 
6459 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6460 	    "t4rxqs");
6461 	if (rc)
6462 		return (rc);
6463 
6464 	if (vi->flags & VI_INIT_DONE)
6465 		rc = EBUSY; /* cannot be changed once the queues are created */
6466 	else
6467 		vi->qsize_rxq = qsize;
6468 
6469 	end_synchronized_op(sc, LOCK_HELD);
6470 	return (rc);
6471 }
6472 
6473 static int
6474 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
6475 {
6476 	struct vi_info *vi = arg1;
6477 	struct adapter *sc = vi->pi->adapter;
6478 	int qsize, rc;
6479 
6480 	qsize = vi->qsize_txq;
6481 
6482 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6483 	if (rc != 0 || req->newptr == NULL)
6484 		return (rc);
6485 
6486 	if (qsize < 128 || qsize > 65536)
6487 		return (EINVAL);
6488 
6489 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6490 	    "t4txqs");
6491 	if (rc)
6492 		return (rc);
6493 
6494 	if (vi->flags & VI_INIT_DONE)
6495 		rc = EBUSY; /* cannot be changed once the queues are created */
6496 	else
6497 		vi->qsize_txq = qsize;
6498 
6499 	end_synchronized_op(sc, LOCK_HELD);
6500 	return (rc);
6501 }
6502 
6503 static int
6504 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
6505 {
6506 	struct port_info *pi = arg1;
6507 	struct adapter *sc = pi->adapter;
6508 	struct link_config *lc = &pi->link_cfg;
6509 	int rc;
6510 
6511 	if (req->newptr == NULL) {
6512 		struct sbuf *sb;
6513 		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
6514 
6515 		rc = sysctl_wire_old_buffer(req, 0);
6516 		if (rc != 0)
6517 			return(rc);
6518 
6519 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6520 		if (sb == NULL)
6521 			return (ENOMEM);
6522 
6523 		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
6524 		rc = sbuf_finish(sb);
6525 		sbuf_delete(sb);
6526 	} else {
6527 		char s[2];
6528 		int n;
6529 
6530 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
6531 		s[1] = 0;
6532 
6533 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6534 		if (rc != 0)
6535 			return(rc);
6536 
6537 		if (s[1] != 0)
6538 			return (EINVAL);
6539 		if (s[0] < '0' || s[0] > '9')
6540 			return (EINVAL);	/* not a number */
6541 		n = s[0] - '0';
6542 		if (n & ~(PAUSE_TX | PAUSE_RX))
6543 			return (EINVAL);	/* some other bit is set too */
6544 
6545 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6546 		    "t4PAUSE");
6547 		if (rc)
6548 			return (rc);
6549 		PORT_LOCK(pi);
6550 		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
6551 			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
6552 			lc->requested_fc |= n;
6553 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6554 			if (rc == 0) {
6555 				lc->fc = lc->requested_fc;
6556 				set_current_media(pi, &pi->media);
6557 			}
6558 		}
6559 		PORT_UNLOCK(pi);
6560 		end_synchronized_op(sc, 0);
6561 	}
6562 
6563 	return (rc);
6564 }
6565 
6566 static int
6567 sysctl_fec(SYSCTL_HANDLER_ARGS)
6568 {
6569 	struct port_info *pi = arg1;
6570 	struct adapter *sc = pi->adapter;
6571 	struct link_config *lc = &pi->link_cfg;
6572 	int rc;
6573 
6574 	if (req->newptr == NULL) {
6575 		struct sbuf *sb;
6576 		static char *bits = "\20\1RS\2BASER_RS\3RESERVED";
6577 
6578 		rc = sysctl_wire_old_buffer(req, 0);
6579 		if (rc != 0)
6580 			return(rc);
6581 
6582 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6583 		if (sb == NULL)
6584 			return (ENOMEM);
6585 
6586 		sbuf_printf(sb, "%b", lc->fec & M_FW_PORT_CAP_FEC, bits);
6587 		rc = sbuf_finish(sb);
6588 		sbuf_delete(sb);
6589 	} else {
6590 		char s[2];
6591 		int n;
6592 
6593 		s[0] = '0' + (lc->requested_fec & M_FW_PORT_CAP_FEC);
6594 		s[1] = 0;
6595 
6596 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6597 		if (rc != 0)
6598 			return(rc);
6599 
6600 		if (s[1] != 0)
6601 			return (EINVAL);
6602 		if (s[0] < '0' || s[0] > '9')
6603 			return (EINVAL);	/* not a number */
6604 		n = s[0] - '0';
6605 		if (n & ~M_FW_PORT_CAP_FEC)
6606 			return (EINVAL);	/* some other bit is set too */
6607 		if (!powerof2(n))
6608 			return (EINVAL);	/* one bit can be set at most */
6609 
6610 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6611 		    "t4fec");
6612 		if (rc)
6613 			return (rc);
6614 		PORT_LOCK(pi);
6615 		if ((lc->requested_fec & M_FW_PORT_CAP_FEC) != n) {
6616 			lc->requested_fec = n &
6617 			    G_FW_PORT_CAP_FEC(lc->supported);
6618 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6619 			if (rc == 0) {
6620 				lc->fec = lc->requested_fec;
6621 			}
6622 		}
6623 		PORT_UNLOCK(pi);
6624 		end_synchronized_op(sc, 0);
6625 	}
6626 
6627 	return (rc);
6628 }
6629 
6630 static int
6631 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
6632 {
6633 	struct port_info *pi = arg1;
6634 	struct adapter *sc = pi->adapter;
6635 	struct link_config *lc = &pi->link_cfg;
6636 	int rc, val, old;
6637 
6638 	if (lc->supported & FW_PORT_CAP_ANEG)
6639 		val = lc->requested_aneg == AUTONEG_ENABLE ? 1 : 0;
6640 	else
6641 		val = -1;
6642 	rc = sysctl_handle_int(oidp, &val, 0, req);
6643 	if (rc != 0 || req->newptr == NULL)
6644 		return (rc);
6645 	if (val == 0)
6646 		val = AUTONEG_DISABLE;
6647 	else if (val == 1)
6648 		val = AUTONEG_ENABLE;
6649 	else
6650 		return (EINVAL);
6651 
6652 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6653 	    "t4aneg");
6654 	if (rc)
6655 		return (rc);
6656 	PORT_LOCK(pi);
6657 	if ((lc->supported & FW_PORT_CAP_ANEG) == 0) {
6658 		rc = ENOTSUP;
6659 		goto done;
6660 	}
6661 	if (lc->requested_aneg == val) {
6662 		rc = 0;	/* no change, do nothing. */
6663 		goto done;
6664 	}
6665 	old = lc->requested_aneg;
6666 	lc->requested_aneg = val;
6667 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6668 	if (rc != 0)
6669 		lc->requested_aneg = old;
6670 	else
6671 		set_current_media(pi, &pi->media);
6672 done:
6673 	PORT_UNLOCK(pi);
6674 	end_synchronized_op(sc, 0);
6675 	return (rc);
6676 }
6677 
6678 static int
6679 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
6680 {
6681 	struct adapter *sc = arg1;
6682 	int reg = arg2;
6683 	uint64_t val;
6684 
6685 	val = t4_read_reg64(sc, reg);
6686 
6687 	return (sysctl_handle_64(oidp, &val, 0, req));
6688 }
6689 
6690 static int
6691 sysctl_temperature(SYSCTL_HANDLER_ARGS)
6692 {
6693 	struct adapter *sc = arg1;
6694 	int rc, t;
6695 	uint32_t param, val;
6696 
6697 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
6698 	if (rc)
6699 		return (rc);
6700 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6701 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
6702 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
6703 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6704 	end_synchronized_op(sc, 0);
6705 	if (rc)
6706 		return (rc);
6707 
6708 	/* unknown is returned as 0 but we display -1 in that case */
6709 	t = val == 0 ? -1 : val;
6710 
6711 	rc = sysctl_handle_int(oidp, &t, 0, req);
6712 	return (rc);
6713 }
6714 
6715 static int
6716 sysctl_loadavg(SYSCTL_HANDLER_ARGS)
6717 {
6718 	struct adapter *sc = arg1;
6719 	struct sbuf *sb;
6720 	int rc;
6721 	uint32_t param, val;
6722 
6723 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4lavg");
6724 	if (rc)
6725 		return (rc);
6726 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6727 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_LOAD);
6728 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6729 	end_synchronized_op(sc, 0);
6730 	if (rc)
6731 		return (rc);
6732 
6733 	rc = sysctl_wire_old_buffer(req, 0);
6734 	if (rc != 0)
6735 		return (rc);
6736 
6737 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6738 	if (sb == NULL)
6739 		return (ENOMEM);
6740 
6741 	if (val == 0xffffffff) {
6742 		/* Only debug and custom firmwares report load averages. */
6743 		sbuf_printf(sb, "not available");
6744 	} else {
6745 		sbuf_printf(sb, "%d %d %d", val & 0xff, (val >> 8) & 0xff,
6746 		    (val >> 16) & 0xff);
6747 	}
6748 	rc = sbuf_finish(sb);
6749 	sbuf_delete(sb);
6750 
6751 	return (rc);
6752 }
6753 
6754 static int
6755 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
6756 {
6757 	struct adapter *sc = arg1;
6758 	struct sbuf *sb;
6759 	int rc, i;
6760 	uint16_t incr[NMTUS][NCCTRL_WIN];
6761 	static const char *dec_fac[] = {
6762 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
6763 		"0.9375"
6764 	};
6765 
6766 	rc = sysctl_wire_old_buffer(req, 0);
6767 	if (rc != 0)
6768 		return (rc);
6769 
6770 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6771 	if (sb == NULL)
6772 		return (ENOMEM);
6773 
6774 	t4_read_cong_tbl(sc, incr);
6775 
6776 	for (i = 0; i < NCCTRL_WIN; ++i) {
6777 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
6778 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
6779 		    incr[5][i], incr[6][i], incr[7][i]);
6780 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
6781 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
6782 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
6783 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
6784 	}
6785 
6786 	rc = sbuf_finish(sb);
6787 	sbuf_delete(sb);
6788 
6789 	return (rc);
6790 }
6791 
6792 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
6793 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
6794 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
6795 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
6796 };
6797 
6798 static int
6799 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
6800 {
6801 	struct adapter *sc = arg1;
6802 	struct sbuf *sb;
6803 	int rc, i, n, qid = arg2;
6804 	uint32_t *buf, *p;
6805 	char *qtype;
6806 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
6807 
6808 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
6809 	    ("%s: bad qid %d\n", __func__, qid));
6810 
6811 	if (qid < CIM_NUM_IBQ) {
6812 		/* inbound queue */
6813 		qtype = "IBQ";
6814 		n = 4 * CIM_IBQ_SIZE;
6815 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6816 		rc = t4_read_cim_ibq(sc, qid, buf, n);
6817 	} else {
6818 		/* outbound queue */
6819 		qtype = "OBQ";
6820 		qid -= CIM_NUM_IBQ;
6821 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
6822 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6823 		rc = t4_read_cim_obq(sc, qid, buf, n);
6824 	}
6825 
6826 	if (rc < 0) {
6827 		rc = -rc;
6828 		goto done;
6829 	}
6830 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
6831 
6832 	rc = sysctl_wire_old_buffer(req, 0);
6833 	if (rc != 0)
6834 		goto done;
6835 
6836 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6837 	if (sb == NULL) {
6838 		rc = ENOMEM;
6839 		goto done;
6840 	}
6841 
6842 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
6843 	for (i = 0, p = buf; i < n; i += 16, p += 4)
6844 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
6845 		    p[2], p[3]);
6846 
6847 	rc = sbuf_finish(sb);
6848 	sbuf_delete(sb);
6849 done:
6850 	free(buf, M_CXGBE);
6851 	return (rc);
6852 }
6853 
6854 static int
6855 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
6856 {
6857 	struct adapter *sc = arg1;
6858 	u_int cfg;
6859 	struct sbuf *sb;
6860 	uint32_t *buf, *p;
6861 	int rc;
6862 
6863 	MPASS(chip_id(sc) <= CHELSIO_T5);
6864 
6865 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6866 	if (rc != 0)
6867 		return (rc);
6868 
6869 	rc = sysctl_wire_old_buffer(req, 0);
6870 	if (rc != 0)
6871 		return (rc);
6872 
6873 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6874 	if (sb == NULL)
6875 		return (ENOMEM);
6876 
6877 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6878 	    M_ZERO | M_WAITOK);
6879 
6880 	rc = -t4_cim_read_la(sc, buf, NULL);
6881 	if (rc != 0)
6882 		goto done;
6883 
6884 	sbuf_printf(sb, "Status   Data      PC%s",
6885 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6886 	    "     LS0Stat  LS0Addr             LS0Data");
6887 
6888 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
6889 		if (cfg & F_UPDBGLACAPTPCONLY) {
6890 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
6891 			    p[6], p[7]);
6892 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
6893 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
6894 			    p[4] & 0xff, p[5] >> 8);
6895 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
6896 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6897 			    p[1] & 0xf, p[2] >> 4);
6898 		} else {
6899 			sbuf_printf(sb,
6900 			    "\n  %02x   %x%07x %x%07x %08x %08x "
6901 			    "%08x%08x%08x%08x",
6902 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6903 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
6904 			    p[6], p[7]);
6905 		}
6906 	}
6907 
6908 	rc = sbuf_finish(sb);
6909 	sbuf_delete(sb);
6910 done:
6911 	free(buf, M_CXGBE);
6912 	return (rc);
6913 }
6914 
6915 static int
6916 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
6917 {
6918 	struct adapter *sc = arg1;
6919 	u_int cfg;
6920 	struct sbuf *sb;
6921 	uint32_t *buf, *p;
6922 	int rc;
6923 
6924 	MPASS(chip_id(sc) > CHELSIO_T5);
6925 
6926 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6927 	if (rc != 0)
6928 		return (rc);
6929 
6930 	rc = sysctl_wire_old_buffer(req, 0);
6931 	if (rc != 0)
6932 		return (rc);
6933 
6934 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6935 	if (sb == NULL)
6936 		return (ENOMEM);
6937 
6938 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6939 	    M_ZERO | M_WAITOK);
6940 
6941 	rc = -t4_cim_read_la(sc, buf, NULL);
6942 	if (rc != 0)
6943 		goto done;
6944 
6945 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
6946 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6947 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
6948 
6949 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
6950 		if (cfg & F_UPDBGLACAPTPCONLY) {
6951 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
6952 			    p[3] & 0xff, p[2], p[1], p[0]);
6953 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
6954 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
6955 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
6956 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
6957 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
6958 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
6959 			    p[6] >> 16);
6960 		} else {
6961 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
6962 			    "%08x %08x %08x %08x %08x %08x",
6963 			    (p[9] >> 16) & 0xff,
6964 			    p[9] & 0xffff, p[8] >> 16,
6965 			    p[8] & 0xffff, p[7] >> 16,
6966 			    p[7] & 0xffff, p[6] >> 16,
6967 			    p[2], p[1], p[0], p[5], p[4], p[3]);
6968 		}
6969 	}
6970 
6971 	rc = sbuf_finish(sb);
6972 	sbuf_delete(sb);
6973 done:
6974 	free(buf, M_CXGBE);
6975 	return (rc);
6976 }
6977 
6978 static int
6979 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
6980 {
6981 	struct adapter *sc = arg1;
6982 	u_int i;
6983 	struct sbuf *sb;
6984 	uint32_t *buf, *p;
6985 	int rc;
6986 
6987 	rc = sysctl_wire_old_buffer(req, 0);
6988 	if (rc != 0)
6989 		return (rc);
6990 
6991 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6992 	if (sb == NULL)
6993 		return (ENOMEM);
6994 
6995 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
6996 	    M_ZERO | M_WAITOK);
6997 
6998 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
6999 	p = buf;
7000 
7001 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7002 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
7003 		    p[1], p[0]);
7004 	}
7005 
7006 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
7007 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7008 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
7009 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
7010 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
7011 		    (p[1] >> 2) | ((p[2] & 3) << 30),
7012 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
7013 		    p[0] & 1);
7014 	}
7015 
7016 	rc = sbuf_finish(sb);
7017 	sbuf_delete(sb);
7018 	free(buf, M_CXGBE);
7019 	return (rc);
7020 }
7021 
7022 static int
7023 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
7024 {
7025 	struct adapter *sc = arg1;
7026 	u_int i;
7027 	struct sbuf *sb;
7028 	uint32_t *buf, *p;
7029 	int rc;
7030 
7031 	rc = sysctl_wire_old_buffer(req, 0);
7032 	if (rc != 0)
7033 		return (rc);
7034 
7035 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7036 	if (sb == NULL)
7037 		return (ENOMEM);
7038 
7039 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
7040 	    M_ZERO | M_WAITOK);
7041 
7042 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
7043 	p = buf;
7044 
7045 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
7046 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7047 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
7048 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
7049 		    p[4], p[3], p[2], p[1], p[0]);
7050 	}
7051 
7052 	sbuf_printf(sb, "\n\nCntl ID               Data");
7053 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7054 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
7055 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
7056 	}
7057 
7058 	rc = sbuf_finish(sb);
7059 	sbuf_delete(sb);
7060 	free(buf, M_CXGBE);
7061 	return (rc);
7062 }
7063 
7064 static int
7065 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
7066 {
7067 	struct adapter *sc = arg1;
7068 	struct sbuf *sb;
7069 	int rc, i;
7070 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7071 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7072 	uint16_t thres[CIM_NUM_IBQ];
7073 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
7074 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
7075 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
7076 
7077 	cim_num_obq = sc->chip_params->cim_num_obq;
7078 	if (is_t4(sc)) {
7079 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
7080 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
7081 	} else {
7082 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
7083 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
7084 	}
7085 	nq = CIM_NUM_IBQ + cim_num_obq;
7086 
7087 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
7088 	if (rc == 0)
7089 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
7090 	if (rc != 0)
7091 		return (rc);
7092 
7093 	t4_read_cimq_cfg(sc, base, size, thres);
7094 
7095 	rc = sysctl_wire_old_buffer(req, 0);
7096 	if (rc != 0)
7097 		return (rc);
7098 
7099 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7100 	if (sb == NULL)
7101 		return (ENOMEM);
7102 
7103 	sbuf_printf(sb,
7104 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
7105 
7106 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
7107 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
7108 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
7109 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7110 		    G_QUEREMFLITS(p[2]) * 16);
7111 	for ( ; i < nq; i++, p += 4, wr += 2)
7112 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
7113 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
7114 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7115 		    G_QUEREMFLITS(p[2]) * 16);
7116 
7117 	rc = sbuf_finish(sb);
7118 	sbuf_delete(sb);
7119 
7120 	return (rc);
7121 }
7122 
7123 static int
7124 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
7125 {
7126 	struct adapter *sc = arg1;
7127 	struct sbuf *sb;
7128 	int rc;
7129 	struct tp_cpl_stats stats;
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, 256, req);
7136 	if (sb == NULL)
7137 		return (ENOMEM);
7138 
7139 	mtx_lock(&sc->reg_lock);
7140 	t4_tp_get_cpl_stats(sc, &stats, 0);
7141 	mtx_unlock(&sc->reg_lock);
7142 
7143 	if (sc->chip_params->nchan > 2) {
7144 		sbuf_printf(sb, "                 channel 0  channel 1"
7145 		    "  channel 2  channel 3");
7146 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
7147 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
7148 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
7149 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
7150 	} else {
7151 		sbuf_printf(sb, "                 channel 0  channel 1");
7152 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
7153 		    stats.req[0], stats.req[1]);
7154 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
7155 		    stats.rsp[0], stats.rsp[1]);
7156 	}
7157 
7158 	rc = sbuf_finish(sb);
7159 	sbuf_delete(sb);
7160 
7161 	return (rc);
7162 }
7163 
7164 static int
7165 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
7166 {
7167 	struct adapter *sc = arg1;
7168 	struct sbuf *sb;
7169 	int rc;
7170 	struct tp_usm_stats stats;
7171 
7172 	rc = sysctl_wire_old_buffer(req, 0);
7173 	if (rc != 0)
7174 		return(rc);
7175 
7176 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7177 	if (sb == NULL)
7178 		return (ENOMEM);
7179 
7180 	t4_get_usm_stats(sc, &stats, 1);
7181 
7182 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
7183 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
7184 	sbuf_printf(sb, "Drops:  %u", stats.drops);
7185 
7186 	rc = sbuf_finish(sb);
7187 	sbuf_delete(sb);
7188 
7189 	return (rc);
7190 }
7191 
7192 static const char * const devlog_level_strings[] = {
7193 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
7194 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
7195 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
7196 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
7197 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
7198 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
7199 };
7200 
7201 static const char * const devlog_facility_strings[] = {
7202 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
7203 	[FW_DEVLOG_FACILITY_CF]		= "CF",
7204 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
7205 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
7206 	[FW_DEVLOG_FACILITY_RES]	= "RES",
7207 	[FW_DEVLOG_FACILITY_HW]		= "HW",
7208 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
7209 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
7210 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
7211 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
7212 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
7213 	[FW_DEVLOG_FACILITY_VI]		= "VI",
7214 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
7215 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
7216 	[FW_DEVLOG_FACILITY_TM]		= "TM",
7217 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
7218 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
7219 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
7220 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
7221 	[FW_DEVLOG_FACILITY_RI]		= "RI",
7222 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
7223 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
7224 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
7225 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
7226 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
7227 };
7228 
7229 static int
7230 sysctl_devlog(SYSCTL_HANDLER_ARGS)
7231 {
7232 	struct adapter *sc = arg1;
7233 	struct devlog_params *dparams = &sc->params.devlog;
7234 	struct fw_devlog_e *buf, *e;
7235 	int i, j, rc, nentries, first = 0;
7236 	struct sbuf *sb;
7237 	uint64_t ftstamp = UINT64_MAX;
7238 
7239 	if (dparams->addr == 0)
7240 		return (ENXIO);
7241 
7242 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
7243 	if (buf == NULL)
7244 		return (ENOMEM);
7245 
7246 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
7247 	if (rc != 0)
7248 		goto done;
7249 
7250 	nentries = dparams->size / sizeof(struct fw_devlog_e);
7251 	for (i = 0; i < nentries; i++) {
7252 		e = &buf[i];
7253 
7254 		if (e->timestamp == 0)
7255 			break;	/* end */
7256 
7257 		e->timestamp = be64toh(e->timestamp);
7258 		e->seqno = be32toh(e->seqno);
7259 		for (j = 0; j < 8; j++)
7260 			e->params[j] = be32toh(e->params[j]);
7261 
7262 		if (e->timestamp < ftstamp) {
7263 			ftstamp = e->timestamp;
7264 			first = i;
7265 		}
7266 	}
7267 
7268 	if (buf[first].timestamp == 0)
7269 		goto done;	/* nothing in the log */
7270 
7271 	rc = sysctl_wire_old_buffer(req, 0);
7272 	if (rc != 0)
7273 		goto done;
7274 
7275 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7276 	if (sb == NULL) {
7277 		rc = ENOMEM;
7278 		goto done;
7279 	}
7280 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
7281 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
7282 
7283 	i = first;
7284 	do {
7285 		e = &buf[i];
7286 		if (e->timestamp == 0)
7287 			break;	/* end */
7288 
7289 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
7290 		    e->seqno, e->timestamp,
7291 		    (e->level < nitems(devlog_level_strings) ?
7292 			devlog_level_strings[e->level] : "UNKNOWN"),
7293 		    (e->facility < nitems(devlog_facility_strings) ?
7294 			devlog_facility_strings[e->facility] : "UNKNOWN"));
7295 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
7296 		    e->params[2], e->params[3], e->params[4],
7297 		    e->params[5], e->params[6], e->params[7]);
7298 
7299 		if (++i == nentries)
7300 			i = 0;
7301 	} while (i != first);
7302 
7303 	rc = sbuf_finish(sb);
7304 	sbuf_delete(sb);
7305 done:
7306 	free(buf, M_CXGBE);
7307 	return (rc);
7308 }
7309 
7310 static int
7311 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
7312 {
7313 	struct adapter *sc = arg1;
7314 	struct sbuf *sb;
7315 	int rc;
7316 	struct tp_fcoe_stats stats[MAX_NCHAN];
7317 	int i, nchan = sc->chip_params->nchan;
7318 
7319 	rc = sysctl_wire_old_buffer(req, 0);
7320 	if (rc != 0)
7321 		return (rc);
7322 
7323 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7324 	if (sb == NULL)
7325 		return (ENOMEM);
7326 
7327 	for (i = 0; i < nchan; i++)
7328 		t4_get_fcoe_stats(sc, i, &stats[i], 1);
7329 
7330 	if (nchan > 2) {
7331 		sbuf_printf(sb, "                   channel 0        channel 1"
7332 		    "        channel 2        channel 3");
7333 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
7334 		    stats[0].octets_ddp, stats[1].octets_ddp,
7335 		    stats[2].octets_ddp, stats[3].octets_ddp);
7336 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
7337 		    stats[0].frames_ddp, stats[1].frames_ddp,
7338 		    stats[2].frames_ddp, stats[3].frames_ddp);
7339 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
7340 		    stats[0].frames_drop, stats[1].frames_drop,
7341 		    stats[2].frames_drop, stats[3].frames_drop);
7342 	} else {
7343 		sbuf_printf(sb, "                   channel 0        channel 1");
7344 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
7345 		    stats[0].octets_ddp, stats[1].octets_ddp);
7346 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
7347 		    stats[0].frames_ddp, stats[1].frames_ddp);
7348 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
7349 		    stats[0].frames_drop, stats[1].frames_drop);
7350 	}
7351 
7352 	rc = sbuf_finish(sb);
7353 	sbuf_delete(sb);
7354 
7355 	return (rc);
7356 }
7357 
7358 static int
7359 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
7360 {
7361 	struct adapter *sc = arg1;
7362 	struct sbuf *sb;
7363 	int rc, i;
7364 	unsigned int map, kbps, ipg, mode;
7365 	unsigned int pace_tab[NTX_SCHED];
7366 
7367 	rc = sysctl_wire_old_buffer(req, 0);
7368 	if (rc != 0)
7369 		return (rc);
7370 
7371 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7372 	if (sb == NULL)
7373 		return (ENOMEM);
7374 
7375 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
7376 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
7377 	t4_read_pace_tbl(sc, pace_tab);
7378 
7379 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
7380 	    "Class IPG (0.1 ns)   Flow IPG (us)");
7381 
7382 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
7383 		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
7384 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
7385 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
7386 		if (kbps)
7387 			sbuf_printf(sb, "%9u     ", kbps);
7388 		else
7389 			sbuf_printf(sb, " disabled     ");
7390 
7391 		if (ipg)
7392 			sbuf_printf(sb, "%13u        ", ipg);
7393 		else
7394 			sbuf_printf(sb, "     disabled        ");
7395 
7396 		if (pace_tab[i])
7397 			sbuf_printf(sb, "%10u", pace_tab[i]);
7398 		else
7399 			sbuf_printf(sb, "  disabled");
7400 	}
7401 
7402 	rc = sbuf_finish(sb);
7403 	sbuf_delete(sb);
7404 
7405 	return (rc);
7406 }
7407 
7408 static int
7409 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
7410 {
7411 	struct adapter *sc = arg1;
7412 	struct sbuf *sb;
7413 	int rc, i, j;
7414 	uint64_t *p0, *p1;
7415 	struct lb_port_stats s[2];
7416 	static const char *stat_name[] = {
7417 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
7418 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
7419 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
7420 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
7421 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
7422 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
7423 		"BG2FramesTrunc:", "BG3FramesTrunc:"
7424 	};
7425 
7426 	rc = sysctl_wire_old_buffer(req, 0);
7427 	if (rc != 0)
7428 		return (rc);
7429 
7430 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7431 	if (sb == NULL)
7432 		return (ENOMEM);
7433 
7434 	memset(s, 0, sizeof(s));
7435 
7436 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
7437 		t4_get_lb_stats(sc, i, &s[0]);
7438 		t4_get_lb_stats(sc, i + 1, &s[1]);
7439 
7440 		p0 = &s[0].octets;
7441 		p1 = &s[1].octets;
7442 		sbuf_printf(sb, "%s                       Loopback %u"
7443 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
7444 
7445 		for (j = 0; j < nitems(stat_name); j++)
7446 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
7447 				   *p0++, *p1++);
7448 	}
7449 
7450 	rc = sbuf_finish(sb);
7451 	sbuf_delete(sb);
7452 
7453 	return (rc);
7454 }
7455 
7456 static int
7457 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
7458 {
7459 	int rc = 0;
7460 	struct port_info *pi = arg1;
7461 	struct link_config *lc = &pi->link_cfg;
7462 	struct sbuf *sb;
7463 
7464 	rc = sysctl_wire_old_buffer(req, 0);
7465 	if (rc != 0)
7466 		return(rc);
7467 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
7468 	if (sb == NULL)
7469 		return (ENOMEM);
7470 
7471 	if (lc->link_ok || lc->link_down_rc == 255)
7472 		sbuf_printf(sb, "n/a");
7473 	else
7474 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
7475 
7476 	rc = sbuf_finish(sb);
7477 	sbuf_delete(sb);
7478 
7479 	return (rc);
7480 }
7481 
7482 struct mem_desc {
7483 	unsigned int base;
7484 	unsigned int limit;
7485 	unsigned int idx;
7486 };
7487 
7488 static int
7489 mem_desc_cmp(const void *a, const void *b)
7490 {
7491 	return ((const struct mem_desc *)a)->base -
7492 	       ((const struct mem_desc *)b)->base;
7493 }
7494 
7495 static void
7496 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
7497     unsigned int to)
7498 {
7499 	unsigned int size;
7500 
7501 	if (from == to)
7502 		return;
7503 
7504 	size = to - from + 1;
7505 	if (size == 0)
7506 		return;
7507 
7508 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
7509 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
7510 }
7511 
7512 static int
7513 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
7514 {
7515 	struct adapter *sc = arg1;
7516 	struct sbuf *sb;
7517 	int rc, i, n;
7518 	uint32_t lo, hi, used, alloc;
7519 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
7520 	static const char *region[] = {
7521 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
7522 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
7523 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
7524 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
7525 		"RQUDP region:", "PBL region:", "TXPBL region:",
7526 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
7527 		"On-chip queues:", "TLS keys:",
7528 	};
7529 	struct mem_desc avail[4];
7530 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
7531 	struct mem_desc *md = mem;
7532 
7533 	rc = sysctl_wire_old_buffer(req, 0);
7534 	if (rc != 0)
7535 		return (rc);
7536 
7537 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7538 	if (sb == NULL)
7539 		return (ENOMEM);
7540 
7541 	for (i = 0; i < nitems(mem); i++) {
7542 		mem[i].limit = 0;
7543 		mem[i].idx = i;
7544 	}
7545 
7546 	/* Find and sort the populated memory ranges */
7547 	i = 0;
7548 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
7549 	if (lo & F_EDRAM0_ENABLE) {
7550 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
7551 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
7552 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
7553 		avail[i].idx = 0;
7554 		i++;
7555 	}
7556 	if (lo & F_EDRAM1_ENABLE) {
7557 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
7558 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
7559 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
7560 		avail[i].idx = 1;
7561 		i++;
7562 	}
7563 	if (lo & F_EXT_MEM_ENABLE) {
7564 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
7565 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
7566 		avail[i].limit = avail[i].base +
7567 		    (G_EXT_MEM_SIZE(hi) << 20);
7568 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
7569 		i++;
7570 	}
7571 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
7572 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
7573 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
7574 		avail[i].limit = avail[i].base +
7575 		    (G_EXT_MEM1_SIZE(hi) << 20);
7576 		avail[i].idx = 4;
7577 		i++;
7578 	}
7579 	if (!i)                                    /* no memory available */
7580 		return 0;
7581 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
7582 
7583 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
7584 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
7585 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
7586 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
7587 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
7588 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
7589 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
7590 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
7591 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
7592 
7593 	/* the next few have explicit upper bounds */
7594 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
7595 	md->limit = md->base - 1 +
7596 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
7597 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
7598 	md++;
7599 
7600 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
7601 	md->limit = md->base - 1 +
7602 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
7603 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
7604 	md++;
7605 
7606 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7607 		if (chip_id(sc) <= CHELSIO_T5)
7608 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
7609 		else
7610 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
7611 		md->limit = 0;
7612 	} else {
7613 		md->base = 0;
7614 		md->idx = nitems(region);  /* hide it */
7615 	}
7616 	md++;
7617 
7618 #define ulp_region(reg) \
7619 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
7620 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
7621 
7622 	ulp_region(RX_ISCSI);
7623 	ulp_region(RX_TDDP);
7624 	ulp_region(TX_TPT);
7625 	ulp_region(RX_STAG);
7626 	ulp_region(RX_RQ);
7627 	ulp_region(RX_RQUDP);
7628 	ulp_region(RX_PBL);
7629 	ulp_region(TX_PBL);
7630 #undef ulp_region
7631 
7632 	md->base = 0;
7633 	md->idx = nitems(region);
7634 	if (!is_t4(sc)) {
7635 		uint32_t size = 0;
7636 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
7637 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
7638 
7639 		if (is_t5(sc)) {
7640 			if (sge_ctrl & F_VFIFO_ENABLE)
7641 				size = G_DBVFIFO_SIZE(fifo_size);
7642 		} else
7643 			size = G_T6_DBVFIFO_SIZE(fifo_size);
7644 
7645 		if (size) {
7646 			md->base = G_BASEADDR(t4_read_reg(sc,
7647 			    A_SGE_DBVFIFO_BADDR));
7648 			md->limit = md->base + (size << 2) - 1;
7649 		}
7650 	}
7651 	md++;
7652 
7653 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
7654 	md->limit = 0;
7655 	md++;
7656 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
7657 	md->limit = 0;
7658 	md++;
7659 
7660 	md->base = sc->vres.ocq.start;
7661 	if (sc->vres.ocq.size)
7662 		md->limit = md->base + sc->vres.ocq.size - 1;
7663 	else
7664 		md->idx = nitems(region);  /* hide it */
7665 	md++;
7666 
7667 	md->base = sc->vres.key.start;
7668 	if (sc->vres.key.size)
7669 		md->limit = md->base + sc->vres.key.size - 1;
7670 	else
7671 		md->idx = nitems(region);  /* hide it */
7672 	md++;
7673 
7674 	/* add any address-space holes, there can be up to 3 */
7675 	for (n = 0; n < i - 1; n++)
7676 		if (avail[n].limit < avail[n + 1].base)
7677 			(md++)->base = avail[n].limit;
7678 	if (avail[n].limit)
7679 		(md++)->base = avail[n].limit;
7680 
7681 	n = md - mem;
7682 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
7683 
7684 	for (lo = 0; lo < i; lo++)
7685 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
7686 				avail[lo].limit - 1);
7687 
7688 	sbuf_printf(sb, "\n");
7689 	for (i = 0; i < n; i++) {
7690 		if (mem[i].idx >= nitems(region))
7691 			continue;                        /* skip holes */
7692 		if (!mem[i].limit)
7693 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
7694 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
7695 				mem[i].limit);
7696 	}
7697 
7698 	sbuf_printf(sb, "\n");
7699 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
7700 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
7701 	mem_region_show(sb, "uP RAM:", lo, hi);
7702 
7703 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
7704 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
7705 	mem_region_show(sb, "uP Extmem2:", lo, hi);
7706 
7707 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
7708 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
7709 		   G_PMRXMAXPAGE(lo),
7710 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
7711 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
7712 
7713 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
7714 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
7715 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
7716 		   G_PMTXMAXPAGE(lo),
7717 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
7718 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
7719 	sbuf_printf(sb, "%u p-structs\n",
7720 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
7721 
7722 	for (i = 0; i < 4; i++) {
7723 		if (chip_id(sc) > CHELSIO_T5)
7724 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
7725 		else
7726 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
7727 		if (is_t5(sc)) {
7728 			used = G_T5_USED(lo);
7729 			alloc = G_T5_ALLOC(lo);
7730 		} else {
7731 			used = G_USED(lo);
7732 			alloc = G_ALLOC(lo);
7733 		}
7734 		/* For T6 these are MAC buffer groups */
7735 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
7736 		    i, used, alloc);
7737 	}
7738 	for (i = 0; i < sc->chip_params->nchan; i++) {
7739 		if (chip_id(sc) > CHELSIO_T5)
7740 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
7741 		else
7742 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
7743 		if (is_t5(sc)) {
7744 			used = G_T5_USED(lo);
7745 			alloc = G_T5_ALLOC(lo);
7746 		} else {
7747 			used = G_USED(lo);
7748 			alloc = G_ALLOC(lo);
7749 		}
7750 		/* For T6 these are MAC buffer groups */
7751 		sbuf_printf(sb,
7752 		    "\nLoopback %d using %u pages out of %u allocated",
7753 		    i, used, alloc);
7754 	}
7755 
7756 	rc = sbuf_finish(sb);
7757 	sbuf_delete(sb);
7758 
7759 	return (rc);
7760 }
7761 
7762 static inline void
7763 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
7764 {
7765 	*mask = x | y;
7766 	y = htobe64(y);
7767 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
7768 }
7769 
7770 static int
7771 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
7772 {
7773 	struct adapter *sc = arg1;
7774 	struct sbuf *sb;
7775 	int rc, i;
7776 
7777 	MPASS(chip_id(sc) <= CHELSIO_T5);
7778 
7779 	rc = sysctl_wire_old_buffer(req, 0);
7780 	if (rc != 0)
7781 		return (rc);
7782 
7783 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7784 	if (sb == NULL)
7785 		return (ENOMEM);
7786 
7787 	sbuf_printf(sb,
7788 	    "Idx  Ethernet address     Mask     Vld Ports PF"
7789 	    "  VF              Replication             P0 P1 P2 P3  ML");
7790 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7791 		uint64_t tcamx, tcamy, mask;
7792 		uint32_t cls_lo, cls_hi;
7793 		uint8_t addr[ETHER_ADDR_LEN];
7794 
7795 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
7796 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
7797 		if (tcamx & tcamy)
7798 			continue;
7799 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7800 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7801 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7802 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
7803 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
7804 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
7805 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
7806 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
7807 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
7808 
7809 		if (cls_lo & F_REPLICATE) {
7810 			struct fw_ldst_cmd ldst_cmd;
7811 
7812 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7813 			ldst_cmd.op_to_addrspace =
7814 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7815 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7816 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7817 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7818 			ldst_cmd.u.mps.rplc.fid_idx =
7819 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7820 				V_FW_LDST_CMD_IDX(i));
7821 
7822 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7823 			    "t4mps");
7824 			if (rc)
7825 				break;
7826 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7827 			    sizeof(ldst_cmd), &ldst_cmd);
7828 			end_synchronized_op(sc, 0);
7829 
7830 			if (rc != 0) {
7831 				sbuf_printf(sb, "%36d", rc);
7832 				rc = 0;
7833 			} else {
7834 				sbuf_printf(sb, " %08x %08x %08x %08x",
7835 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7836 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7837 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7838 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7839 			}
7840 		} else
7841 			sbuf_printf(sb, "%36s", "");
7842 
7843 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
7844 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
7845 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
7846 	}
7847 
7848 	if (rc)
7849 		(void) sbuf_finish(sb);
7850 	else
7851 		rc = sbuf_finish(sb);
7852 	sbuf_delete(sb);
7853 
7854 	return (rc);
7855 }
7856 
7857 static int
7858 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
7859 {
7860 	struct adapter *sc = arg1;
7861 	struct sbuf *sb;
7862 	int rc, i;
7863 
7864 	MPASS(chip_id(sc) > CHELSIO_T5);
7865 
7866 	rc = sysctl_wire_old_buffer(req, 0);
7867 	if (rc != 0)
7868 		return (rc);
7869 
7870 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7871 	if (sb == NULL)
7872 		return (ENOMEM);
7873 
7874 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
7875 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
7876 	    "                           Replication"
7877 	    "                                    P0 P1 P2 P3  ML\n");
7878 
7879 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7880 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
7881 		uint16_t ivlan;
7882 		uint64_t tcamx, tcamy, val, mask;
7883 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
7884 		uint8_t addr[ETHER_ADDR_LEN];
7885 
7886 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
7887 		if (i < 256)
7888 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
7889 		else
7890 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
7891 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7892 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7893 		tcamy = G_DMACH(val) << 32;
7894 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7895 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7896 		lookup_type = G_DATALKPTYPE(data2);
7897 		port_num = G_DATAPORTNUM(data2);
7898 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7899 			/* Inner header VNI */
7900 			vniy = ((data2 & F_DATAVIDH2) << 23) |
7901 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7902 			dip_hit = data2 & F_DATADIPHIT;
7903 			vlan_vld = 0;
7904 		} else {
7905 			vniy = 0;
7906 			dip_hit = 0;
7907 			vlan_vld = data2 & F_DATAVIDH2;
7908 			ivlan = G_VIDL(val);
7909 		}
7910 
7911 		ctl |= V_CTLXYBITSEL(1);
7912 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7913 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7914 		tcamx = G_DMACH(val) << 32;
7915 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7916 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7917 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7918 			/* Inner header VNI mask */
7919 			vnix = ((data2 & F_DATAVIDH2) << 23) |
7920 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7921 		} else
7922 			vnix = 0;
7923 
7924 		if (tcamx & tcamy)
7925 			continue;
7926 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7927 
7928 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7929 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7930 
7931 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7932 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7933 			    "%012jx %06x %06x    -    -   %3c"
7934 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
7935 			    addr[1], addr[2], addr[3], addr[4], addr[5],
7936 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
7937 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7938 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7939 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7940 		} else {
7941 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7942 			    "%012jx    -       -   ", i, addr[0], addr[1],
7943 			    addr[2], addr[3], addr[4], addr[5],
7944 			    (uintmax_t)mask);
7945 
7946 			if (vlan_vld)
7947 				sbuf_printf(sb, "%4u   Y     ", ivlan);
7948 			else
7949 				sbuf_printf(sb, "  -    N     ");
7950 
7951 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
7952 			    lookup_type ? 'I' : 'O', port_num,
7953 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7954 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7955 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7956 		}
7957 
7958 
7959 		if (cls_lo & F_T6_REPLICATE) {
7960 			struct fw_ldst_cmd ldst_cmd;
7961 
7962 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7963 			ldst_cmd.op_to_addrspace =
7964 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7965 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7966 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7967 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7968 			ldst_cmd.u.mps.rplc.fid_idx =
7969 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7970 				V_FW_LDST_CMD_IDX(i));
7971 
7972 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7973 			    "t6mps");
7974 			if (rc)
7975 				break;
7976 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7977 			    sizeof(ldst_cmd), &ldst_cmd);
7978 			end_synchronized_op(sc, 0);
7979 
7980 			if (rc != 0) {
7981 				sbuf_printf(sb, "%72d", rc);
7982 				rc = 0;
7983 			} else {
7984 				sbuf_printf(sb, " %08x %08x %08x %08x"
7985 				    " %08x %08x %08x %08x",
7986 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
7987 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
7988 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
7989 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
7990 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7991 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7992 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7993 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7994 			}
7995 		} else
7996 			sbuf_printf(sb, "%72s", "");
7997 
7998 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
7999 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
8000 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
8001 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
8002 	}
8003 
8004 	if (rc)
8005 		(void) sbuf_finish(sb);
8006 	else
8007 		rc = sbuf_finish(sb);
8008 	sbuf_delete(sb);
8009 
8010 	return (rc);
8011 }
8012 
8013 static int
8014 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
8015 {
8016 	struct adapter *sc = arg1;
8017 	struct sbuf *sb;
8018 	int rc;
8019 	uint16_t mtus[NMTUS];
8020 
8021 	rc = sysctl_wire_old_buffer(req, 0);
8022 	if (rc != 0)
8023 		return (rc);
8024 
8025 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8026 	if (sb == NULL)
8027 		return (ENOMEM);
8028 
8029 	t4_read_mtu_tbl(sc, mtus, NULL);
8030 
8031 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
8032 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
8033 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
8034 	    mtus[14], mtus[15]);
8035 
8036 	rc = sbuf_finish(sb);
8037 	sbuf_delete(sb);
8038 
8039 	return (rc);
8040 }
8041 
8042 static int
8043 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
8044 {
8045 	struct adapter *sc = arg1;
8046 	struct sbuf *sb;
8047 	int rc, i;
8048 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
8049 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
8050 	static const char *tx_stats[MAX_PM_NSTATS] = {
8051 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
8052 		"Tx FIFO wait", NULL, "Tx latency"
8053 	};
8054 	static const char *rx_stats[MAX_PM_NSTATS] = {
8055 		"Read:", "Write bypass:", "Write mem:", "Flush:",
8056 		"Rx FIFO wait", NULL, "Rx latency"
8057 	};
8058 
8059 	rc = sysctl_wire_old_buffer(req, 0);
8060 	if (rc != 0)
8061 		return (rc);
8062 
8063 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8064 	if (sb == NULL)
8065 		return (ENOMEM);
8066 
8067 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
8068 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
8069 
8070 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
8071 	for (i = 0; i < 4; i++) {
8072 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8073 		    tx_cyc[i]);
8074 	}
8075 
8076 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
8077 	for (i = 0; i < 4; i++) {
8078 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8079 		    rx_cyc[i]);
8080 	}
8081 
8082 	if (chip_id(sc) > CHELSIO_T5) {
8083 		sbuf_printf(sb,
8084 		    "\n              Total wait      Total occupancy");
8085 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8086 		    tx_cyc[i]);
8087 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8088 		    rx_cyc[i]);
8089 
8090 		i += 2;
8091 		MPASS(i < nitems(tx_stats));
8092 
8093 		sbuf_printf(sb,
8094 		    "\n                   Reads           Total wait");
8095 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8096 		    tx_cyc[i]);
8097 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8098 		    rx_cyc[i]);
8099 	}
8100 
8101 	rc = sbuf_finish(sb);
8102 	sbuf_delete(sb);
8103 
8104 	return (rc);
8105 }
8106 
8107 static int
8108 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
8109 {
8110 	struct adapter *sc = arg1;
8111 	struct sbuf *sb;
8112 	int rc;
8113 	struct tp_rdma_stats stats;
8114 
8115 	rc = sysctl_wire_old_buffer(req, 0);
8116 	if (rc != 0)
8117 		return (rc);
8118 
8119 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8120 	if (sb == NULL)
8121 		return (ENOMEM);
8122 
8123 	mtx_lock(&sc->reg_lock);
8124 	t4_tp_get_rdma_stats(sc, &stats, 0);
8125 	mtx_unlock(&sc->reg_lock);
8126 
8127 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
8128 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
8129 
8130 	rc = sbuf_finish(sb);
8131 	sbuf_delete(sb);
8132 
8133 	return (rc);
8134 }
8135 
8136 static int
8137 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
8138 {
8139 	struct adapter *sc = arg1;
8140 	struct sbuf *sb;
8141 	int rc;
8142 	struct tp_tcp_stats v4, v6;
8143 
8144 	rc = sysctl_wire_old_buffer(req, 0);
8145 	if (rc != 0)
8146 		return (rc);
8147 
8148 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8149 	if (sb == NULL)
8150 		return (ENOMEM);
8151 
8152 	mtx_lock(&sc->reg_lock);
8153 	t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
8154 	mtx_unlock(&sc->reg_lock);
8155 
8156 	sbuf_printf(sb,
8157 	    "                                IP                 IPv6\n");
8158 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
8159 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
8160 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
8161 	    v4.tcp_in_segs, v6.tcp_in_segs);
8162 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
8163 	    v4.tcp_out_segs, v6.tcp_out_segs);
8164 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
8165 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
8166 
8167 	rc = sbuf_finish(sb);
8168 	sbuf_delete(sb);
8169 
8170 	return (rc);
8171 }
8172 
8173 static int
8174 sysctl_tids(SYSCTL_HANDLER_ARGS)
8175 {
8176 	struct adapter *sc = arg1;
8177 	struct sbuf *sb;
8178 	int rc;
8179 	struct tid_info *t = &sc->tids;
8180 
8181 	rc = sysctl_wire_old_buffer(req, 0);
8182 	if (rc != 0)
8183 		return (rc);
8184 
8185 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8186 	if (sb == NULL)
8187 		return (ENOMEM);
8188 
8189 	if (t->natids) {
8190 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
8191 		    t->atids_in_use);
8192 	}
8193 
8194 	if (t->nhpftids) {
8195 		sbuf_printf(sb, "HPFTID range: %u-%u, in use: %u\n",
8196 		    t->hpftid_base, t->hpftid_end, t->hpftids_in_use);
8197 	}
8198 
8199 	if (t->ntids) {
8200 		sbuf_printf(sb, "TID range: ");
8201 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
8202 			uint32_t b, hb;
8203 
8204 			if (chip_id(sc) <= CHELSIO_T5) {
8205 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
8206 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
8207 			} else {
8208 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
8209 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
8210 			}
8211 
8212 			if (b)
8213 				sbuf_printf(sb, "%u-%u, ", t->tid_base, b - 1);
8214 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
8215 		} else
8216 			sbuf_printf(sb, "%u-%u", t->tid_base, t->ntids - 1);
8217 		sbuf_printf(sb, ", in use: %u\n",
8218 		    atomic_load_acq_int(&t->tids_in_use));
8219 	}
8220 
8221 	if (t->nstids) {
8222 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
8223 		    t->stid_base + t->nstids - 1, t->stids_in_use);
8224 	}
8225 
8226 	if (t->nftids) {
8227 		sbuf_printf(sb, "FTID range: %u-%u, in use: %u\n", t->ftid_base,
8228 		    t->ftid_end, t->ftids_in_use);
8229 	}
8230 
8231 	if (t->netids) {
8232 		sbuf_printf(sb, "ETID range: %u-%u, in use: %u\n", t->etid_base,
8233 		    t->etid_base + t->netids - 1, t->etids_in_use);
8234 	}
8235 
8236 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
8237 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
8238 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
8239 
8240 	rc = sbuf_finish(sb);
8241 	sbuf_delete(sb);
8242 
8243 	return (rc);
8244 }
8245 
8246 static int
8247 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
8248 {
8249 	struct adapter *sc = arg1;
8250 	struct sbuf *sb;
8251 	int rc;
8252 	struct tp_err_stats stats;
8253 
8254 	rc = sysctl_wire_old_buffer(req, 0);
8255 	if (rc != 0)
8256 		return (rc);
8257 
8258 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8259 	if (sb == NULL)
8260 		return (ENOMEM);
8261 
8262 	mtx_lock(&sc->reg_lock);
8263 	t4_tp_get_err_stats(sc, &stats, 0);
8264 	mtx_unlock(&sc->reg_lock);
8265 
8266 	if (sc->chip_params->nchan > 2) {
8267 		sbuf_printf(sb, "                 channel 0  channel 1"
8268 		    "  channel 2  channel 3\n");
8269 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
8270 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
8271 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
8272 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
8273 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
8274 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
8275 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
8276 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
8277 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
8278 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
8279 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
8280 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
8281 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
8282 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
8283 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
8284 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
8285 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
8286 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
8287 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
8288 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
8289 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
8290 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
8291 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
8292 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
8293 	} else {
8294 		sbuf_printf(sb, "                 channel 0  channel 1\n");
8295 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
8296 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
8297 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
8298 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
8299 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
8300 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
8301 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
8302 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
8303 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
8304 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
8305 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
8306 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
8307 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
8308 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
8309 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
8310 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
8311 	}
8312 
8313 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
8314 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
8315 
8316 	rc = sbuf_finish(sb);
8317 	sbuf_delete(sb);
8318 
8319 	return (rc);
8320 }
8321 
8322 static int
8323 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
8324 {
8325 	struct adapter *sc = arg1;
8326 	struct tp_params *tpp = &sc->params.tp;
8327 	u_int mask;
8328 	int rc;
8329 
8330 	mask = tpp->la_mask >> 16;
8331 	rc = sysctl_handle_int(oidp, &mask, 0, req);
8332 	if (rc != 0 || req->newptr == NULL)
8333 		return (rc);
8334 	if (mask > 0xffff)
8335 		return (EINVAL);
8336 	tpp->la_mask = mask << 16;
8337 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
8338 
8339 	return (0);
8340 }
8341 
8342 struct field_desc {
8343 	const char *name;
8344 	u_int start;
8345 	u_int width;
8346 };
8347 
8348 static void
8349 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
8350 {
8351 	char buf[32];
8352 	int line_size = 0;
8353 
8354 	while (f->name) {
8355 		uint64_t mask = (1ULL << f->width) - 1;
8356 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
8357 		    ((uintmax_t)v >> f->start) & mask);
8358 
8359 		if (line_size + len >= 79) {
8360 			line_size = 8;
8361 			sbuf_printf(sb, "\n        ");
8362 		}
8363 		sbuf_printf(sb, "%s ", buf);
8364 		line_size += len + 1;
8365 		f++;
8366 	}
8367 	sbuf_printf(sb, "\n");
8368 }
8369 
8370 static const struct field_desc tp_la0[] = {
8371 	{ "RcfOpCodeOut", 60, 4 },
8372 	{ "State", 56, 4 },
8373 	{ "WcfState", 52, 4 },
8374 	{ "RcfOpcSrcOut", 50, 2 },
8375 	{ "CRxError", 49, 1 },
8376 	{ "ERxError", 48, 1 },
8377 	{ "SanityFailed", 47, 1 },
8378 	{ "SpuriousMsg", 46, 1 },
8379 	{ "FlushInputMsg", 45, 1 },
8380 	{ "FlushInputCpl", 44, 1 },
8381 	{ "RssUpBit", 43, 1 },
8382 	{ "RssFilterHit", 42, 1 },
8383 	{ "Tid", 32, 10 },
8384 	{ "InitTcb", 31, 1 },
8385 	{ "LineNumber", 24, 7 },
8386 	{ "Emsg", 23, 1 },
8387 	{ "EdataOut", 22, 1 },
8388 	{ "Cmsg", 21, 1 },
8389 	{ "CdataOut", 20, 1 },
8390 	{ "EreadPdu", 19, 1 },
8391 	{ "CreadPdu", 18, 1 },
8392 	{ "TunnelPkt", 17, 1 },
8393 	{ "RcfPeerFin", 16, 1 },
8394 	{ "RcfReasonOut", 12, 4 },
8395 	{ "TxCchannel", 10, 2 },
8396 	{ "RcfTxChannel", 8, 2 },
8397 	{ "RxEchannel", 6, 2 },
8398 	{ "RcfRxChannel", 5, 1 },
8399 	{ "RcfDataOutSrdy", 4, 1 },
8400 	{ "RxDvld", 3, 1 },
8401 	{ "RxOoDvld", 2, 1 },
8402 	{ "RxCongestion", 1, 1 },
8403 	{ "TxCongestion", 0, 1 },
8404 	{ NULL }
8405 };
8406 
8407 static const struct field_desc tp_la1[] = {
8408 	{ "CplCmdIn", 56, 8 },
8409 	{ "CplCmdOut", 48, 8 },
8410 	{ "ESynOut", 47, 1 },
8411 	{ "EAckOut", 46, 1 },
8412 	{ "EFinOut", 45, 1 },
8413 	{ "ERstOut", 44, 1 },
8414 	{ "SynIn", 43, 1 },
8415 	{ "AckIn", 42, 1 },
8416 	{ "FinIn", 41, 1 },
8417 	{ "RstIn", 40, 1 },
8418 	{ "DataIn", 39, 1 },
8419 	{ "DataInVld", 38, 1 },
8420 	{ "PadIn", 37, 1 },
8421 	{ "RxBufEmpty", 36, 1 },
8422 	{ "RxDdp", 35, 1 },
8423 	{ "RxFbCongestion", 34, 1 },
8424 	{ "TxFbCongestion", 33, 1 },
8425 	{ "TxPktSumSrdy", 32, 1 },
8426 	{ "RcfUlpType", 28, 4 },
8427 	{ "Eread", 27, 1 },
8428 	{ "Ebypass", 26, 1 },
8429 	{ "Esave", 25, 1 },
8430 	{ "Static0", 24, 1 },
8431 	{ "Cread", 23, 1 },
8432 	{ "Cbypass", 22, 1 },
8433 	{ "Csave", 21, 1 },
8434 	{ "CPktOut", 20, 1 },
8435 	{ "RxPagePoolFull", 18, 2 },
8436 	{ "RxLpbkPkt", 17, 1 },
8437 	{ "TxLpbkPkt", 16, 1 },
8438 	{ "RxVfValid", 15, 1 },
8439 	{ "SynLearned", 14, 1 },
8440 	{ "SetDelEntry", 13, 1 },
8441 	{ "SetInvEntry", 12, 1 },
8442 	{ "CpcmdDvld", 11, 1 },
8443 	{ "CpcmdSave", 10, 1 },
8444 	{ "RxPstructsFull", 8, 2 },
8445 	{ "EpcmdDvld", 7, 1 },
8446 	{ "EpcmdFlush", 6, 1 },
8447 	{ "EpcmdTrimPrefix", 5, 1 },
8448 	{ "EpcmdTrimPostfix", 4, 1 },
8449 	{ "ERssIp4Pkt", 3, 1 },
8450 	{ "ERssIp6Pkt", 2, 1 },
8451 	{ "ERssTcpUdpPkt", 1, 1 },
8452 	{ "ERssFceFipPkt", 0, 1 },
8453 	{ NULL }
8454 };
8455 
8456 static const struct field_desc tp_la2[] = {
8457 	{ "CplCmdIn", 56, 8 },
8458 	{ "MpsVfVld", 55, 1 },
8459 	{ "MpsPf", 52, 3 },
8460 	{ "MpsVf", 44, 8 },
8461 	{ "SynIn", 43, 1 },
8462 	{ "AckIn", 42, 1 },
8463 	{ "FinIn", 41, 1 },
8464 	{ "RstIn", 40, 1 },
8465 	{ "DataIn", 39, 1 },
8466 	{ "DataInVld", 38, 1 },
8467 	{ "PadIn", 37, 1 },
8468 	{ "RxBufEmpty", 36, 1 },
8469 	{ "RxDdp", 35, 1 },
8470 	{ "RxFbCongestion", 34, 1 },
8471 	{ "TxFbCongestion", 33, 1 },
8472 	{ "TxPktSumSrdy", 32, 1 },
8473 	{ "RcfUlpType", 28, 4 },
8474 	{ "Eread", 27, 1 },
8475 	{ "Ebypass", 26, 1 },
8476 	{ "Esave", 25, 1 },
8477 	{ "Static0", 24, 1 },
8478 	{ "Cread", 23, 1 },
8479 	{ "Cbypass", 22, 1 },
8480 	{ "Csave", 21, 1 },
8481 	{ "CPktOut", 20, 1 },
8482 	{ "RxPagePoolFull", 18, 2 },
8483 	{ "RxLpbkPkt", 17, 1 },
8484 	{ "TxLpbkPkt", 16, 1 },
8485 	{ "RxVfValid", 15, 1 },
8486 	{ "SynLearned", 14, 1 },
8487 	{ "SetDelEntry", 13, 1 },
8488 	{ "SetInvEntry", 12, 1 },
8489 	{ "CpcmdDvld", 11, 1 },
8490 	{ "CpcmdSave", 10, 1 },
8491 	{ "RxPstructsFull", 8, 2 },
8492 	{ "EpcmdDvld", 7, 1 },
8493 	{ "EpcmdFlush", 6, 1 },
8494 	{ "EpcmdTrimPrefix", 5, 1 },
8495 	{ "EpcmdTrimPostfix", 4, 1 },
8496 	{ "ERssIp4Pkt", 3, 1 },
8497 	{ "ERssIp6Pkt", 2, 1 },
8498 	{ "ERssTcpUdpPkt", 1, 1 },
8499 	{ "ERssFceFipPkt", 0, 1 },
8500 	{ NULL }
8501 };
8502 
8503 static void
8504 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
8505 {
8506 
8507 	field_desc_show(sb, *p, tp_la0);
8508 }
8509 
8510 static void
8511 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
8512 {
8513 
8514 	if (idx)
8515 		sbuf_printf(sb, "\n");
8516 	field_desc_show(sb, p[0], tp_la0);
8517 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8518 		field_desc_show(sb, p[1], tp_la0);
8519 }
8520 
8521 static void
8522 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
8523 {
8524 
8525 	if (idx)
8526 		sbuf_printf(sb, "\n");
8527 	field_desc_show(sb, p[0], tp_la0);
8528 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8529 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
8530 }
8531 
8532 static int
8533 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
8534 {
8535 	struct adapter *sc = arg1;
8536 	struct sbuf *sb;
8537 	uint64_t *buf, *p;
8538 	int rc;
8539 	u_int i, inc;
8540 	void (*show_func)(struct sbuf *, uint64_t *, int);
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, 4096, req);
8547 	if (sb == NULL)
8548 		return (ENOMEM);
8549 
8550 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
8551 
8552 	t4_tp_read_la(sc, buf, NULL);
8553 	p = buf;
8554 
8555 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
8556 	case 2:
8557 		inc = 2;
8558 		show_func = tp_la_show2;
8559 		break;
8560 	case 3:
8561 		inc = 2;
8562 		show_func = tp_la_show3;
8563 		break;
8564 	default:
8565 		inc = 1;
8566 		show_func = tp_la_show;
8567 	}
8568 
8569 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
8570 		(*show_func)(sb, p, i);
8571 
8572 	rc = sbuf_finish(sb);
8573 	sbuf_delete(sb);
8574 	free(buf, M_CXGBE);
8575 	return (rc);
8576 }
8577 
8578 static int
8579 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
8580 {
8581 	struct adapter *sc = arg1;
8582 	struct sbuf *sb;
8583 	int rc;
8584 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
8585 
8586 	rc = sysctl_wire_old_buffer(req, 0);
8587 	if (rc != 0)
8588 		return (rc);
8589 
8590 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8591 	if (sb == NULL)
8592 		return (ENOMEM);
8593 
8594 	t4_get_chan_txrate(sc, nrate, orate);
8595 
8596 	if (sc->chip_params->nchan > 2) {
8597 		sbuf_printf(sb, "              channel 0   channel 1"
8598 		    "   channel 2   channel 3\n");
8599 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
8600 		    nrate[0], nrate[1], nrate[2], nrate[3]);
8601 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
8602 		    orate[0], orate[1], orate[2], orate[3]);
8603 	} else {
8604 		sbuf_printf(sb, "              channel 0   channel 1\n");
8605 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
8606 		    nrate[0], nrate[1]);
8607 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
8608 		    orate[0], orate[1]);
8609 	}
8610 
8611 	rc = sbuf_finish(sb);
8612 	sbuf_delete(sb);
8613 
8614 	return (rc);
8615 }
8616 
8617 static int
8618 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
8619 {
8620 	struct adapter *sc = arg1;
8621 	struct sbuf *sb;
8622 	uint32_t *buf, *p;
8623 	int rc, i;
8624 
8625 	rc = sysctl_wire_old_buffer(req, 0);
8626 	if (rc != 0)
8627 		return (rc);
8628 
8629 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8630 	if (sb == NULL)
8631 		return (ENOMEM);
8632 
8633 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
8634 	    M_ZERO | M_WAITOK);
8635 
8636 	t4_ulprx_read_la(sc, buf);
8637 	p = buf;
8638 
8639 	sbuf_printf(sb, "      Pcmd        Type   Message"
8640 	    "                Data");
8641 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
8642 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
8643 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
8644 	}
8645 
8646 	rc = sbuf_finish(sb);
8647 	sbuf_delete(sb);
8648 	free(buf, M_CXGBE);
8649 	return (rc);
8650 }
8651 
8652 static int
8653 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
8654 {
8655 	struct adapter *sc = arg1;
8656 	struct sbuf *sb;
8657 	int rc, v;
8658 
8659 	MPASS(chip_id(sc) >= CHELSIO_T5);
8660 
8661 	rc = sysctl_wire_old_buffer(req, 0);
8662 	if (rc != 0)
8663 		return (rc);
8664 
8665 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8666 	if (sb == NULL)
8667 		return (ENOMEM);
8668 
8669 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
8670 	if (G_STATSOURCE_T5(v) == 7) {
8671 		int mode;
8672 
8673 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
8674 		if (mode == 0) {
8675 			sbuf_printf(sb, "total %d, incomplete %d",
8676 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8677 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8678 		} else if (mode == 1) {
8679 			sbuf_printf(sb, "total %d, data overflow %d",
8680 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8681 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8682 		} else {
8683 			sbuf_printf(sb, "unknown mode %d", mode);
8684 		}
8685 	}
8686 	rc = sbuf_finish(sb);
8687 	sbuf_delete(sb);
8688 
8689 	return (rc);
8690 }
8691 
8692 static int
8693 sysctl_cpus(SYSCTL_HANDLER_ARGS)
8694 {
8695 	struct adapter *sc = arg1;
8696 	enum cpu_sets op = arg2;
8697 	cpuset_t cpuset;
8698 	struct sbuf *sb;
8699 	int i, rc;
8700 
8701 	MPASS(op == LOCAL_CPUS || op == INTR_CPUS);
8702 
8703 	CPU_ZERO(&cpuset);
8704 	rc = bus_get_cpus(sc->dev, op, sizeof(cpuset), &cpuset);
8705 	if (rc != 0)
8706 		return (rc);
8707 
8708 	rc = sysctl_wire_old_buffer(req, 0);
8709 	if (rc != 0)
8710 		return (rc);
8711 
8712 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8713 	if (sb == NULL)
8714 		return (ENOMEM);
8715 
8716 	CPU_FOREACH(i)
8717 		sbuf_printf(sb, "%d ", i);
8718 	rc = sbuf_finish(sb);
8719 	sbuf_delete(sb);
8720 
8721 	return (rc);
8722 }
8723 
8724 #ifdef TCP_OFFLOAD
8725 static int
8726 sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS)
8727 {
8728 	struct adapter *sc = arg1;
8729 	int *old_ports, *new_ports;
8730 	int i, new_count, rc;
8731 
8732 	if (req->newptr == NULL && req->oldptr == NULL)
8733 		return (SYSCTL_OUT(req, NULL, imax(sc->tt.num_tls_rx_ports, 1) *
8734 		    sizeof(sc->tt.tls_rx_ports[0])));
8735 
8736 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tlsrx");
8737 	if (rc)
8738 		return (rc);
8739 
8740 	if (sc->tt.num_tls_rx_ports == 0) {
8741 		i = -1;
8742 		rc = SYSCTL_OUT(req, &i, sizeof(i));
8743 	} else
8744 		rc = SYSCTL_OUT(req, sc->tt.tls_rx_ports,
8745 		    sc->tt.num_tls_rx_ports * sizeof(sc->tt.tls_rx_ports[0]));
8746 	if (rc == 0 && req->newptr != NULL) {
8747 		new_count = req->newlen / sizeof(new_ports[0]);
8748 		new_ports = malloc(new_count * sizeof(new_ports[0]), M_CXGBE,
8749 		    M_WAITOK);
8750 		rc = SYSCTL_IN(req, new_ports, new_count *
8751 		    sizeof(new_ports[0]));
8752 		if (rc)
8753 			goto err;
8754 
8755 		/* Allow setting to a single '-1' to clear the list. */
8756 		if (new_count == 1 && new_ports[0] == -1) {
8757 			ADAPTER_LOCK(sc);
8758 			old_ports = sc->tt.tls_rx_ports;
8759 			sc->tt.tls_rx_ports = NULL;
8760 			sc->tt.num_tls_rx_ports = 0;
8761 			ADAPTER_UNLOCK(sc);
8762 			free(old_ports, M_CXGBE);
8763 		} else {
8764 			for (i = 0; i < new_count; i++) {
8765 				if (new_ports[i] < 1 ||
8766 				    new_ports[i] > IPPORT_MAX) {
8767 					rc = EINVAL;
8768 					goto err;
8769 				}
8770 			}
8771 
8772 			ADAPTER_LOCK(sc);
8773 			old_ports = sc->tt.tls_rx_ports;
8774 			sc->tt.tls_rx_ports = new_ports;
8775 			sc->tt.num_tls_rx_ports = new_count;
8776 			ADAPTER_UNLOCK(sc);
8777 			free(old_ports, M_CXGBE);
8778 			new_ports = NULL;
8779 		}
8780 	err:
8781 		free(new_ports, M_CXGBE);
8782 	}
8783 	end_synchronized_op(sc, 0);
8784 	return (rc);
8785 }
8786 
8787 static void
8788 unit_conv(char *buf, size_t len, u_int val, u_int factor)
8789 {
8790 	u_int rem = val % factor;
8791 
8792 	if (rem == 0)
8793 		snprintf(buf, len, "%u", val / factor);
8794 	else {
8795 		while (rem % 10 == 0)
8796 			rem /= 10;
8797 		snprintf(buf, len, "%u.%u", val / factor, rem);
8798 	}
8799 }
8800 
8801 static int
8802 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
8803 {
8804 	struct adapter *sc = arg1;
8805 	char buf[16];
8806 	u_int res, re;
8807 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8808 
8809 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8810 	switch (arg2) {
8811 	case 0:
8812 		/* timer_tick */
8813 		re = G_TIMERRESOLUTION(res);
8814 		break;
8815 	case 1:
8816 		/* TCP timestamp tick */
8817 		re = G_TIMESTAMPRESOLUTION(res);
8818 		break;
8819 	case 2:
8820 		/* DACK tick */
8821 		re = G_DELAYEDACKRESOLUTION(res);
8822 		break;
8823 	default:
8824 		return (EDOOFUS);
8825 	}
8826 
8827 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
8828 
8829 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
8830 }
8831 
8832 static int
8833 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
8834 {
8835 	struct adapter *sc = arg1;
8836 	u_int res, dack_re, v;
8837 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8838 
8839 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8840 	dack_re = G_DELAYEDACKRESOLUTION(res);
8841 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
8842 
8843 	return (sysctl_handle_int(oidp, &v, 0, req));
8844 }
8845 
8846 static int
8847 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
8848 {
8849 	struct adapter *sc = arg1;
8850 	int reg = arg2;
8851 	u_int tre;
8852 	u_long tp_tick_us, v;
8853 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8854 
8855 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
8856 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
8857 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
8858 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
8859 
8860 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
8861 	tp_tick_us = (cclk_ps << tre) / 1000000;
8862 
8863 	if (reg == A_TP_INIT_SRTT)
8864 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
8865 	else
8866 		v = tp_tick_us * t4_read_reg(sc, reg);
8867 
8868 	return (sysctl_handle_long(oidp, &v, 0, req));
8869 }
8870 
8871 /*
8872  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
8873  * passed to this function.
8874  */
8875 static int
8876 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
8877 {
8878 	struct adapter *sc = arg1;
8879 	int idx = arg2;
8880 	u_int v;
8881 
8882 	MPASS(idx >= 0 && idx <= 24);
8883 
8884 	v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
8885 
8886 	return (sysctl_handle_int(oidp, &v, 0, req));
8887 }
8888 
8889 static int
8890 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
8891 {
8892 	struct adapter *sc = arg1;
8893 	int idx = arg2;
8894 	u_int shift, v, r;
8895 
8896 	MPASS(idx >= 0 && idx < 16);
8897 
8898 	r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
8899 	shift = (idx & 3) << 3;
8900 	v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
8901 
8902 	return (sysctl_handle_int(oidp, &v, 0, req));
8903 }
8904 
8905 static int
8906 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
8907 {
8908 	struct vi_info *vi = arg1;
8909 	struct adapter *sc = vi->pi->adapter;
8910 	int idx, rc, i;
8911 	struct sge_ofld_rxq *ofld_rxq;
8912 	uint8_t v;
8913 
8914 	idx = vi->ofld_tmr_idx;
8915 
8916 	rc = sysctl_handle_int(oidp, &idx, 0, req);
8917 	if (rc != 0 || req->newptr == NULL)
8918 		return (rc);
8919 
8920 	if (idx < 0 || idx >= SGE_NTIMERS)
8921 		return (EINVAL);
8922 
8923 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8924 	    "t4otmr");
8925 	if (rc)
8926 		return (rc);
8927 
8928 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
8929 	for_each_ofld_rxq(vi, i, ofld_rxq) {
8930 #ifdef atomic_store_rel_8
8931 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
8932 #else
8933 		ofld_rxq->iq.intr_params = v;
8934 #endif
8935 	}
8936 	vi->ofld_tmr_idx = idx;
8937 
8938 	end_synchronized_op(sc, LOCK_HELD);
8939 	return (0);
8940 }
8941 
8942 static int
8943 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
8944 {
8945 	struct vi_info *vi = arg1;
8946 	struct adapter *sc = vi->pi->adapter;
8947 	int idx, rc;
8948 
8949 	idx = vi->ofld_pktc_idx;
8950 
8951 	rc = sysctl_handle_int(oidp, &idx, 0, req);
8952 	if (rc != 0 || req->newptr == NULL)
8953 		return (rc);
8954 
8955 	if (idx < -1 || idx >= SGE_NCOUNTERS)
8956 		return (EINVAL);
8957 
8958 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8959 	    "t4opktc");
8960 	if (rc)
8961 		return (rc);
8962 
8963 	if (vi->flags & VI_INIT_DONE)
8964 		rc = EBUSY; /* cannot be changed once the queues are created */
8965 	else
8966 		vi->ofld_pktc_idx = idx;
8967 
8968 	end_synchronized_op(sc, LOCK_HELD);
8969 	return (rc);
8970 }
8971 #endif
8972 
8973 static int
8974 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
8975 {
8976 	int rc;
8977 
8978 	if (cntxt->cid > M_CTXTQID)
8979 		return (EINVAL);
8980 
8981 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
8982 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
8983 		return (EINVAL);
8984 
8985 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
8986 	if (rc)
8987 		return (rc);
8988 
8989 	if (sc->flags & FW_OK) {
8990 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
8991 		    &cntxt->data[0]);
8992 		if (rc == 0)
8993 			goto done;
8994 	}
8995 
8996 	/*
8997 	 * Read via firmware failed or wasn't even attempted.  Read directly via
8998 	 * the backdoor.
8999 	 */
9000 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
9001 done:
9002 	end_synchronized_op(sc, 0);
9003 	return (rc);
9004 }
9005 
9006 static int
9007 load_fw(struct adapter *sc, struct t4_data *fw)
9008 {
9009 	int rc;
9010 	uint8_t *fw_data;
9011 
9012 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
9013 	if (rc)
9014 		return (rc);
9015 
9016 	/*
9017 	 * The firmware, with the sole exception of the memory parity error
9018 	 * handler, runs from memory and not flash.  It is almost always safe to
9019 	 * install a new firmware on a running system.  Just set bit 1 in
9020 	 * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
9021 	 */
9022 	if (sc->flags & FULL_INIT_DONE &&
9023 	    (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
9024 		rc = EBUSY;
9025 		goto done;
9026 	}
9027 
9028 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
9029 	if (fw_data == NULL) {
9030 		rc = ENOMEM;
9031 		goto done;
9032 	}
9033 
9034 	rc = copyin(fw->data, fw_data, fw->len);
9035 	if (rc == 0)
9036 		rc = -t4_load_fw(sc, fw_data, fw->len);
9037 
9038 	free(fw_data, M_CXGBE);
9039 done:
9040 	end_synchronized_op(sc, 0);
9041 	return (rc);
9042 }
9043 
9044 static int
9045 load_cfg(struct adapter *sc, struct t4_data *cfg)
9046 {
9047 	int rc;
9048 	uint8_t *cfg_data = NULL;
9049 
9050 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9051 	if (rc)
9052 		return (rc);
9053 
9054 	if (cfg->len == 0) {
9055 		/* clear */
9056 		rc = -t4_load_cfg(sc, NULL, 0);
9057 		goto done;
9058 	}
9059 
9060 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
9061 	if (cfg_data == NULL) {
9062 		rc = ENOMEM;
9063 		goto done;
9064 	}
9065 
9066 	rc = copyin(cfg->data, cfg_data, cfg->len);
9067 	if (rc == 0)
9068 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
9069 
9070 	free(cfg_data, M_CXGBE);
9071 done:
9072 	end_synchronized_op(sc, 0);
9073 	return (rc);
9074 }
9075 
9076 static int
9077 load_boot(struct adapter *sc, struct t4_bootrom *br)
9078 {
9079 	int rc;
9080 	uint8_t *br_data = NULL;
9081 	u_int offset;
9082 
9083 	if (br->len > 1024 * 1024)
9084 		return (EFBIG);
9085 
9086 	if (br->pf_offset == 0) {
9087 		/* pfidx */
9088 		if (br->pfidx_addr > 7)
9089 			return (EINVAL);
9090 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
9091 		    A_PCIE_PF_EXPROM_OFST)));
9092 	} else if (br->pf_offset == 1) {
9093 		/* offset */
9094 		offset = G_OFFSET(br->pfidx_addr);
9095 	} else {
9096 		return (EINVAL);
9097 	}
9098 
9099 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
9100 	if (rc)
9101 		return (rc);
9102 
9103 	if (br->len == 0) {
9104 		/* clear */
9105 		rc = -t4_load_boot(sc, NULL, offset, 0);
9106 		goto done;
9107 	}
9108 
9109 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
9110 	if (br_data == NULL) {
9111 		rc = ENOMEM;
9112 		goto done;
9113 	}
9114 
9115 	rc = copyin(br->data, br_data, br->len);
9116 	if (rc == 0)
9117 		rc = -t4_load_boot(sc, br_data, offset, br->len);
9118 
9119 	free(br_data, M_CXGBE);
9120 done:
9121 	end_synchronized_op(sc, 0);
9122 	return (rc);
9123 }
9124 
9125 static int
9126 load_bootcfg(struct adapter *sc, struct t4_data *bc)
9127 {
9128 	int rc;
9129 	uint8_t *bc_data = NULL;
9130 
9131 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9132 	if (rc)
9133 		return (rc);
9134 
9135 	if (bc->len == 0) {
9136 		/* clear */
9137 		rc = -t4_load_bootcfg(sc, NULL, 0);
9138 		goto done;
9139 	}
9140 
9141 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
9142 	if (bc_data == NULL) {
9143 		rc = ENOMEM;
9144 		goto done;
9145 	}
9146 
9147 	rc = copyin(bc->data, bc_data, bc->len);
9148 	if (rc == 0)
9149 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
9150 
9151 	free(bc_data, M_CXGBE);
9152 done:
9153 	end_synchronized_op(sc, 0);
9154 	return (rc);
9155 }
9156 
9157 static int
9158 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
9159 {
9160 	int rc;
9161 	struct cudbg_init *cudbg;
9162 	void *handle, *buf;
9163 
9164 	/* buf is large, don't block if no memory is available */
9165 	buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
9166 	if (buf == NULL)
9167 		return (ENOMEM);
9168 
9169 	handle = cudbg_alloc_handle();
9170 	if (handle == NULL) {
9171 		rc = ENOMEM;
9172 		goto done;
9173 	}
9174 
9175 	cudbg = cudbg_get_init(handle);
9176 	cudbg->adap = sc;
9177 	cudbg->print = (cudbg_print_cb)printf;
9178 
9179 #ifndef notyet
9180 	device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
9181 	    __func__, dump->wr_flash, dump->len, dump->data);
9182 #endif
9183 
9184 	if (dump->wr_flash)
9185 		cudbg->use_flash = 1;
9186 	MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
9187 	memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
9188 
9189 	rc = cudbg_collect(handle, buf, &dump->len);
9190 	if (rc != 0)
9191 		goto done;
9192 
9193 	rc = copyout(buf, dump->data, dump->len);
9194 done:
9195 	cudbg_free_handle(handle);
9196 	free(buf, M_CXGBE);
9197 	return (rc);
9198 }
9199 
9200 static void
9201 free_offload_policy(struct t4_offload_policy *op)
9202 {
9203 	struct offload_rule *r;
9204 	int i;
9205 
9206 	if (op == NULL)
9207 		return;
9208 
9209 	r = &op->rule[0];
9210 	for (i = 0; i < op->nrules; i++, r++) {
9211 		free(r->bpf_prog.bf_insns, M_CXGBE);
9212 	}
9213 	free(op->rule, M_CXGBE);
9214 	free(op, M_CXGBE);
9215 }
9216 
9217 static int
9218 set_offload_policy(struct adapter *sc, struct t4_offload_policy *uop)
9219 {
9220 	int i, rc, len;
9221 	struct t4_offload_policy *op, *old;
9222 	struct bpf_program *bf;
9223 	const struct offload_settings *s;
9224 	struct offload_rule *r;
9225 	void *u;
9226 
9227 	if (!is_offload(sc))
9228 		return (ENODEV);
9229 
9230 	if (uop->nrules == 0) {
9231 		/* Delete installed policies. */
9232 		op = NULL;
9233 		goto set_policy;
9234 	} if (uop->nrules > 256) { /* arbitrary */
9235 		return (E2BIG);
9236 	}
9237 
9238 	/* Copy userspace offload policy to kernel */
9239 	op = malloc(sizeof(*op), M_CXGBE, M_ZERO | M_WAITOK);
9240 	op->nrules = uop->nrules;
9241 	len = op->nrules * sizeof(struct offload_rule);
9242 	op->rule = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9243 	rc = copyin(uop->rule, op->rule, len);
9244 	if (rc) {
9245 		free(op->rule, M_CXGBE);
9246 		free(op, M_CXGBE);
9247 		return (rc);
9248 	}
9249 
9250 	r = &op->rule[0];
9251 	for (i = 0; i < op->nrules; i++, r++) {
9252 
9253 		/* Validate open_type */
9254 		if (r->open_type != OPEN_TYPE_LISTEN &&
9255 		    r->open_type != OPEN_TYPE_ACTIVE &&
9256 		    r->open_type != OPEN_TYPE_PASSIVE &&
9257 		    r->open_type != OPEN_TYPE_DONTCARE) {
9258 error:
9259 			/*
9260 			 * Rules 0 to i have malloc'd filters that need to be
9261 			 * freed.  Rules i+1 to nrules have userspace pointers
9262 			 * and should be left alone.
9263 			 */
9264 			op->nrules = i;
9265 			free_offload_policy(op);
9266 			return (rc);
9267 		}
9268 
9269 		/* Validate settings */
9270 		s = &r->settings;
9271 		if ((s->offload != 0 && s->offload != 1) ||
9272 		    s->cong_algo < -1 || s->cong_algo > CONG_ALG_HIGHSPEED ||
9273 		    s->sched_class < -1 ||
9274 		    s->sched_class >= sc->chip_params->nsched_cls) {
9275 			rc = EINVAL;
9276 			goto error;
9277 		}
9278 
9279 		bf = &r->bpf_prog;
9280 		u = bf->bf_insns;	/* userspace ptr */
9281 		bf->bf_insns = NULL;
9282 		if (bf->bf_len == 0) {
9283 			/* legal, matches everything */
9284 			continue;
9285 		}
9286 		len = bf->bf_len * sizeof(*bf->bf_insns);
9287 		bf->bf_insns = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9288 		rc = copyin(u, bf->bf_insns, len);
9289 		if (rc != 0)
9290 			goto error;
9291 
9292 		if (!bpf_validate(bf->bf_insns, bf->bf_len)) {
9293 			rc = EINVAL;
9294 			goto error;
9295 		}
9296 	}
9297 set_policy:
9298 	rw_wlock(&sc->policy_lock);
9299 	old = sc->policy;
9300 	sc->policy = op;
9301 	rw_wunlock(&sc->policy_lock);
9302 	free_offload_policy(old);
9303 
9304 	return (0);
9305 }
9306 
9307 #define MAX_READ_BUF_SIZE (128 * 1024)
9308 static int
9309 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
9310 {
9311 	uint32_t addr, remaining, n;
9312 	uint32_t *buf;
9313 	int rc;
9314 	uint8_t *dst;
9315 
9316 	rc = validate_mem_range(sc, mr->addr, mr->len);
9317 	if (rc != 0)
9318 		return (rc);
9319 
9320 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
9321 	addr = mr->addr;
9322 	remaining = mr->len;
9323 	dst = (void *)mr->data;
9324 
9325 	while (remaining) {
9326 		n = min(remaining, MAX_READ_BUF_SIZE);
9327 		read_via_memwin(sc, 2, addr, buf, n);
9328 
9329 		rc = copyout(buf, dst, n);
9330 		if (rc != 0)
9331 			break;
9332 
9333 		dst += n;
9334 		remaining -= n;
9335 		addr += n;
9336 	}
9337 
9338 	free(buf, M_CXGBE);
9339 	return (rc);
9340 }
9341 #undef MAX_READ_BUF_SIZE
9342 
9343 static int
9344 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
9345 {
9346 	int rc;
9347 
9348 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
9349 		return (EINVAL);
9350 
9351 	if (i2cd->len > sizeof(i2cd->data))
9352 		return (EFBIG);
9353 
9354 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9355 	if (rc)
9356 		return (rc);
9357 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9358 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
9359 	end_synchronized_op(sc, 0);
9360 
9361 	return (rc);
9362 }
9363 
9364 int
9365 t4_os_find_pci_capability(struct adapter *sc, int cap)
9366 {
9367 	int i;
9368 
9369 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9370 }
9371 
9372 int
9373 t4_os_pci_save_state(struct adapter *sc)
9374 {
9375 	device_t dev;
9376 	struct pci_devinfo *dinfo;
9377 
9378 	dev = sc->dev;
9379 	dinfo = device_get_ivars(dev);
9380 
9381 	pci_cfg_save(dev, dinfo, 0);
9382 	return (0);
9383 }
9384 
9385 int
9386 t4_os_pci_restore_state(struct adapter *sc)
9387 {
9388 	device_t dev;
9389 	struct pci_devinfo *dinfo;
9390 
9391 	dev = sc->dev;
9392 	dinfo = device_get_ivars(dev);
9393 
9394 	pci_cfg_restore(dev, dinfo);
9395 	return (0);
9396 }
9397 
9398 void
9399 t4_os_portmod_changed(struct port_info *pi)
9400 {
9401 	struct adapter *sc = pi->adapter;
9402 	struct vi_info *vi;
9403 	struct ifnet *ifp;
9404 	static const char *mod_str[] = {
9405 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9406 	};
9407 
9408 	MPASS((pi->flags & FIXED_IFMEDIA) == 0);
9409 
9410 	vi = &pi->vi[0];
9411 	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9412 		PORT_LOCK(pi);
9413 		build_medialist(pi, &pi->media);
9414 		apply_l1cfg(pi);
9415 		PORT_UNLOCK(pi);
9416 		end_synchronized_op(sc, LOCK_HELD);
9417 	}
9418 
9419 	ifp = vi->ifp;
9420 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9421 		if_printf(ifp, "transceiver unplugged.\n");
9422 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9423 		if_printf(ifp, "unknown transceiver inserted.\n");
9424 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9425 		if_printf(ifp, "unsupported transceiver inserted.\n");
9426 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9427 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9428 		    port_top_speed(pi), mod_str[pi->mod_type]);
9429 	} else {
9430 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9431 		    pi->mod_type);
9432 	}
9433 }
9434 
9435 void
9436 t4_os_link_changed(struct port_info *pi)
9437 {
9438 	struct vi_info *vi;
9439 	struct ifnet *ifp;
9440 	struct link_config *lc;
9441 	int v;
9442 
9443 	PORT_LOCK_ASSERT_OWNED(pi);
9444 
9445 	for_each_vi(pi, v, vi) {
9446 		ifp = vi->ifp;
9447 		if (ifp == NULL)
9448 			continue;
9449 
9450 		lc = &pi->link_cfg;
9451 		if (lc->link_ok) {
9452 			ifp->if_baudrate = IF_Mbps(lc->speed);
9453 			if_link_state_change(ifp, LINK_STATE_UP);
9454 		} else {
9455 			if_link_state_change(ifp, LINK_STATE_DOWN);
9456 		}
9457 	}
9458 }
9459 
9460 void
9461 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9462 {
9463 	struct adapter *sc;
9464 
9465 	sx_slock(&t4_list_lock);
9466 	SLIST_FOREACH(sc, &t4_list, link) {
9467 		/*
9468 		 * func should not make any assumptions about what state sc is
9469 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9470 		 */
9471 		func(sc, arg);
9472 	}
9473 	sx_sunlock(&t4_list_lock);
9474 }
9475 
9476 static int
9477 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9478     struct thread *td)
9479 {
9480 	int rc;
9481 	struct adapter *sc = dev->si_drv1;
9482 
9483 	rc = priv_check(td, PRIV_DRIVER);
9484 	if (rc != 0)
9485 		return (rc);
9486 
9487 	switch (cmd) {
9488 	case CHELSIO_T4_GETREG: {
9489 		struct t4_reg *edata = (struct t4_reg *)data;
9490 
9491 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9492 			return (EFAULT);
9493 
9494 		if (edata->size == 4)
9495 			edata->val = t4_read_reg(sc, edata->addr);
9496 		else if (edata->size == 8)
9497 			edata->val = t4_read_reg64(sc, edata->addr);
9498 		else
9499 			return (EINVAL);
9500 
9501 		break;
9502 	}
9503 	case CHELSIO_T4_SETREG: {
9504 		struct t4_reg *edata = (struct t4_reg *)data;
9505 
9506 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9507 			return (EFAULT);
9508 
9509 		if (edata->size == 4) {
9510 			if (edata->val & 0xffffffff00000000)
9511 				return (EINVAL);
9512 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9513 		} else if (edata->size == 8)
9514 			t4_write_reg64(sc, edata->addr, edata->val);
9515 		else
9516 			return (EINVAL);
9517 		break;
9518 	}
9519 	case CHELSIO_T4_REGDUMP: {
9520 		struct t4_regdump *regs = (struct t4_regdump *)data;
9521 		int reglen = t4_get_regs_len(sc);
9522 		uint8_t *buf;
9523 
9524 		if (regs->len < reglen) {
9525 			regs->len = reglen; /* hint to the caller */
9526 			return (ENOBUFS);
9527 		}
9528 
9529 		regs->len = reglen;
9530 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9531 		get_regs(sc, regs, buf);
9532 		rc = copyout(buf, regs->data, reglen);
9533 		free(buf, M_CXGBE);
9534 		break;
9535 	}
9536 	case CHELSIO_T4_GET_FILTER_MODE:
9537 		rc = get_filter_mode(sc, (uint32_t *)data);
9538 		break;
9539 	case CHELSIO_T4_SET_FILTER_MODE:
9540 		rc = set_filter_mode(sc, *(uint32_t *)data);
9541 		break;
9542 	case CHELSIO_T4_GET_FILTER:
9543 		rc = get_filter(sc, (struct t4_filter *)data);
9544 		break;
9545 	case CHELSIO_T4_SET_FILTER:
9546 		rc = set_filter(sc, (struct t4_filter *)data);
9547 		break;
9548 	case CHELSIO_T4_DEL_FILTER:
9549 		rc = del_filter(sc, (struct t4_filter *)data);
9550 		break;
9551 	case CHELSIO_T4_GET_SGE_CONTEXT:
9552 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9553 		break;
9554 	case CHELSIO_T4_LOAD_FW:
9555 		rc = load_fw(sc, (struct t4_data *)data);
9556 		break;
9557 	case CHELSIO_T4_GET_MEM:
9558 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9559 		break;
9560 	case CHELSIO_T4_GET_I2C:
9561 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9562 		break;
9563 	case CHELSIO_T4_CLEAR_STATS: {
9564 		int i, v, bg_map;
9565 		u_int port_id = *(uint32_t *)data;
9566 		struct port_info *pi;
9567 		struct vi_info *vi;
9568 
9569 		if (port_id >= sc->params.nports)
9570 			return (EINVAL);
9571 		pi = sc->port[port_id];
9572 		if (pi == NULL)
9573 			return (EIO);
9574 
9575 		/* MAC stats */
9576 		t4_clr_port_stats(sc, pi->tx_chan);
9577 		pi->tx_parse_error = 0;
9578 		pi->tnl_cong_drops = 0;
9579 		mtx_lock(&sc->reg_lock);
9580 		for_each_vi(pi, v, vi) {
9581 			if (vi->flags & VI_INIT_DONE)
9582 				t4_clr_vi_stats(sc, vi->viid);
9583 		}
9584 		bg_map = pi->mps_bg_map;
9585 		v = 0;	/* reuse */
9586 		while (bg_map) {
9587 			i = ffs(bg_map) - 1;
9588 			t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
9589 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
9590 			bg_map &= ~(1 << i);
9591 		}
9592 		mtx_unlock(&sc->reg_lock);
9593 
9594 		/*
9595 		 * Since this command accepts a port, clear stats for
9596 		 * all VIs on this port.
9597 		 */
9598 		for_each_vi(pi, v, vi) {
9599 			if (vi->flags & VI_INIT_DONE) {
9600 				struct sge_rxq *rxq;
9601 				struct sge_txq *txq;
9602 				struct sge_wrq *wrq;
9603 
9604 				for_each_rxq(vi, i, rxq) {
9605 #if defined(INET) || defined(INET6)
9606 					rxq->lro.lro_queued = 0;
9607 					rxq->lro.lro_flushed = 0;
9608 #endif
9609 					rxq->rxcsum = 0;
9610 					rxq->vlan_extraction = 0;
9611 				}
9612 
9613 				for_each_txq(vi, i, txq) {
9614 					txq->txcsum = 0;
9615 					txq->tso_wrs = 0;
9616 					txq->vlan_insertion = 0;
9617 					txq->imm_wrs = 0;
9618 					txq->sgl_wrs = 0;
9619 					txq->txpkt_wrs = 0;
9620 					txq->txpkts0_wrs = 0;
9621 					txq->txpkts1_wrs = 0;
9622 					txq->txpkts0_pkts = 0;
9623 					txq->txpkts1_pkts = 0;
9624 					mp_ring_reset_stats(txq->r);
9625 				}
9626 
9627 #ifdef TCP_OFFLOAD
9628 				/* nothing to clear for each ofld_rxq */
9629 
9630 				for_each_ofld_txq(vi, i, wrq) {
9631 					wrq->tx_wrs_direct = 0;
9632 					wrq->tx_wrs_copied = 0;
9633 				}
9634 #endif
9635 
9636 				if (IS_MAIN_VI(vi)) {
9637 					wrq = &sc->sge.ctrlq[pi->port_id];
9638 					wrq->tx_wrs_direct = 0;
9639 					wrq->tx_wrs_copied = 0;
9640 				}
9641 			}
9642 		}
9643 		break;
9644 	}
9645 	case CHELSIO_T4_SCHED_CLASS:
9646 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9647 		break;
9648 	case CHELSIO_T4_SCHED_QUEUE:
9649 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9650 		break;
9651 	case CHELSIO_T4_GET_TRACER:
9652 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9653 		break;
9654 	case CHELSIO_T4_SET_TRACER:
9655 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9656 		break;
9657 	case CHELSIO_T4_LOAD_CFG:
9658 		rc = load_cfg(sc, (struct t4_data *)data);
9659 		break;
9660 	case CHELSIO_T4_LOAD_BOOT:
9661 		rc = load_boot(sc, (struct t4_bootrom *)data);
9662 		break;
9663 	case CHELSIO_T4_LOAD_BOOTCFG:
9664 		rc = load_bootcfg(sc, (struct t4_data *)data);
9665 		break;
9666 	case CHELSIO_T4_CUDBG_DUMP:
9667 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
9668 		break;
9669 	case CHELSIO_T4_SET_OFLD_POLICY:
9670 		rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
9671 		break;
9672 	default:
9673 		rc = ENOTTY;
9674 	}
9675 
9676 	return (rc);
9677 }
9678 
9679 void
9680 t4_db_full(struct adapter *sc)
9681 {
9682 
9683 	CXGBE_UNIMPLEMENTED(__func__);
9684 }
9685 
9686 void
9687 t4_db_dropped(struct adapter *sc)
9688 {
9689 
9690 	CXGBE_UNIMPLEMENTED(__func__);
9691 }
9692 
9693 #ifdef TCP_OFFLOAD
9694 static int
9695 toe_capability(struct vi_info *vi, int enable)
9696 {
9697 	int rc;
9698 	struct port_info *pi = vi->pi;
9699 	struct adapter *sc = pi->adapter;
9700 
9701 	ASSERT_SYNCHRONIZED_OP(sc);
9702 
9703 	if (!is_offload(sc))
9704 		return (ENODEV);
9705 
9706 	if (enable) {
9707 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9708 			/* TOE is already enabled. */
9709 			return (0);
9710 		}
9711 
9712 		/*
9713 		 * We need the port's queues around so that we're able to send
9714 		 * and receive CPLs to/from the TOE even if the ifnet for this
9715 		 * port has never been UP'd administratively.
9716 		 */
9717 		if (!(vi->flags & VI_INIT_DONE)) {
9718 			rc = vi_full_init(vi);
9719 			if (rc)
9720 				return (rc);
9721 		}
9722 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9723 			rc = vi_full_init(&pi->vi[0]);
9724 			if (rc)
9725 				return (rc);
9726 		}
9727 
9728 		if (isset(&sc->offload_map, pi->port_id)) {
9729 			/* TOE is enabled on another VI of this port. */
9730 			pi->uld_vis++;
9731 			return (0);
9732 		}
9733 
9734 		if (!uld_active(sc, ULD_TOM)) {
9735 			rc = t4_activate_uld(sc, ULD_TOM);
9736 			if (rc == EAGAIN) {
9737 				log(LOG_WARNING,
9738 				    "You must kldload t4_tom.ko before trying "
9739 				    "to enable TOE on a cxgbe interface.\n");
9740 			}
9741 			if (rc != 0)
9742 				return (rc);
9743 			KASSERT(sc->tom_softc != NULL,
9744 			    ("%s: TOM activated but softc NULL", __func__));
9745 			KASSERT(uld_active(sc, ULD_TOM),
9746 			    ("%s: TOM activated but flag not set", __func__));
9747 		}
9748 
9749 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
9750 		if (!uld_active(sc, ULD_IWARP))
9751 			(void) t4_activate_uld(sc, ULD_IWARP);
9752 		if (!uld_active(sc, ULD_ISCSI))
9753 			(void) t4_activate_uld(sc, ULD_ISCSI);
9754 
9755 		pi->uld_vis++;
9756 		setbit(&sc->offload_map, pi->port_id);
9757 	} else {
9758 		pi->uld_vis--;
9759 
9760 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9761 			return (0);
9762 
9763 		KASSERT(uld_active(sc, ULD_TOM),
9764 		    ("%s: TOM never initialized?", __func__));
9765 		clrbit(&sc->offload_map, pi->port_id);
9766 	}
9767 
9768 	return (0);
9769 }
9770 
9771 /*
9772  * Add an upper layer driver to the global list.
9773  */
9774 int
9775 t4_register_uld(struct uld_info *ui)
9776 {
9777 	int rc = 0;
9778 	struct uld_info *u;
9779 
9780 	sx_xlock(&t4_uld_list_lock);
9781 	SLIST_FOREACH(u, &t4_uld_list, link) {
9782 	    if (u->uld_id == ui->uld_id) {
9783 		    rc = EEXIST;
9784 		    goto done;
9785 	    }
9786 	}
9787 
9788 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9789 	ui->refcount = 0;
9790 done:
9791 	sx_xunlock(&t4_uld_list_lock);
9792 	return (rc);
9793 }
9794 
9795 int
9796 t4_unregister_uld(struct uld_info *ui)
9797 {
9798 	int rc = EINVAL;
9799 	struct uld_info *u;
9800 
9801 	sx_xlock(&t4_uld_list_lock);
9802 
9803 	SLIST_FOREACH(u, &t4_uld_list, link) {
9804 	    if (u == ui) {
9805 		    if (ui->refcount > 0) {
9806 			    rc = EBUSY;
9807 			    goto done;
9808 		    }
9809 
9810 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9811 		    rc = 0;
9812 		    goto done;
9813 	    }
9814 	}
9815 done:
9816 	sx_xunlock(&t4_uld_list_lock);
9817 	return (rc);
9818 }
9819 
9820 int
9821 t4_activate_uld(struct adapter *sc, int id)
9822 {
9823 	int rc;
9824 	struct uld_info *ui;
9825 
9826 	ASSERT_SYNCHRONIZED_OP(sc);
9827 
9828 	if (id < 0 || id > ULD_MAX)
9829 		return (EINVAL);
9830 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
9831 
9832 	sx_slock(&t4_uld_list_lock);
9833 
9834 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9835 		if (ui->uld_id == id) {
9836 			if (!(sc->flags & FULL_INIT_DONE)) {
9837 				rc = adapter_full_init(sc);
9838 				if (rc != 0)
9839 					break;
9840 			}
9841 
9842 			rc = ui->activate(sc);
9843 			if (rc == 0) {
9844 				setbit(&sc->active_ulds, id);
9845 				ui->refcount++;
9846 			}
9847 			break;
9848 		}
9849 	}
9850 
9851 	sx_sunlock(&t4_uld_list_lock);
9852 
9853 	return (rc);
9854 }
9855 
9856 int
9857 t4_deactivate_uld(struct adapter *sc, int id)
9858 {
9859 	int rc;
9860 	struct uld_info *ui;
9861 
9862 	ASSERT_SYNCHRONIZED_OP(sc);
9863 
9864 	if (id < 0 || id > ULD_MAX)
9865 		return (EINVAL);
9866 	rc = ENXIO;
9867 
9868 	sx_slock(&t4_uld_list_lock);
9869 
9870 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9871 		if (ui->uld_id == id) {
9872 			rc = ui->deactivate(sc);
9873 			if (rc == 0) {
9874 				clrbit(&sc->active_ulds, id);
9875 				ui->refcount--;
9876 			}
9877 			break;
9878 		}
9879 	}
9880 
9881 	sx_sunlock(&t4_uld_list_lock);
9882 
9883 	return (rc);
9884 }
9885 
9886 int
9887 uld_active(struct adapter *sc, int uld_id)
9888 {
9889 
9890 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9891 
9892 	return (isset(&sc->active_ulds, uld_id));
9893 }
9894 #endif
9895 
9896 /*
9897  * t  = ptr to tunable.
9898  * nc = number of CPUs.
9899  * c  = compiled in default for that tunable.
9900  */
9901 static void
9902 calculate_nqueues(int *t, int nc, const int c)
9903 {
9904 	int nq;
9905 
9906 	if (*t > 0)
9907 		return;
9908 	nq = *t < 0 ? -*t : c;
9909 	*t = min(nc, nq);
9910 }
9911 
9912 /*
9913  * Come up with reasonable defaults for some of the tunables, provided they're
9914  * not set by the user (in which case we'll use the values as is).
9915  */
9916 static void
9917 tweak_tunables(void)
9918 {
9919 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
9920 
9921 	if (t4_ntxq < 1) {
9922 #ifdef RSS
9923 		t4_ntxq = rss_getnumbuckets();
9924 #else
9925 		calculate_nqueues(&t4_ntxq, nc, NTXQ);
9926 #endif
9927 	}
9928 
9929 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
9930 
9931 	if (t4_nrxq < 1) {
9932 #ifdef RSS
9933 		t4_nrxq = rss_getnumbuckets();
9934 #else
9935 		calculate_nqueues(&t4_nrxq, nc, NRXQ);
9936 #endif
9937 	}
9938 
9939 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
9940 
9941 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
9942 	calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
9943 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
9944 #endif
9945 #ifdef TCP_OFFLOAD
9946 	calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
9947 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
9948 
9949 	if (t4_toecaps_allowed == -1)
9950 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9951 
9952 	if (t4_rdmacaps_allowed == -1) {
9953 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9954 		    FW_CAPS_CONFIG_RDMA_RDMAC;
9955 	}
9956 
9957 	if (t4_iscsicaps_allowed == -1) {
9958 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9959 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9960 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
9961 	}
9962 
9963 	if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
9964 		t4_tmr_idx_ofld = TMR_IDX_OFLD;
9965 
9966 	if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
9967 		t4_pktc_idx_ofld = PKTC_IDX_OFLD;
9968 #else
9969 	if (t4_toecaps_allowed == -1)
9970 		t4_toecaps_allowed = 0;
9971 
9972 	if (t4_rdmacaps_allowed == -1)
9973 		t4_rdmacaps_allowed = 0;
9974 
9975 	if (t4_iscsicaps_allowed == -1)
9976 		t4_iscsicaps_allowed = 0;
9977 #endif
9978 
9979 #ifdef DEV_NETMAP
9980 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
9981 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
9982 #endif
9983 
9984 	if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
9985 		t4_tmr_idx = TMR_IDX;
9986 
9987 	if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
9988 		t4_pktc_idx = PKTC_IDX;
9989 
9990 	if (t4_qsize_txq < 128)
9991 		t4_qsize_txq = 128;
9992 
9993 	if (t4_qsize_rxq < 128)
9994 		t4_qsize_rxq = 128;
9995 	while (t4_qsize_rxq & 7)
9996 		t4_qsize_rxq++;
9997 
9998 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9999 
10000 	/*
10001 	 * Number of VIs to create per-port.  The first VI is the "main" regular
10002 	 * VI for the port.  The rest are additional virtual interfaces on the
10003 	 * same physical port.  Note that the main VI does not have native
10004 	 * netmap support but the extra VIs do.
10005 	 *
10006 	 * Limit the number of VIs per port to the number of available
10007 	 * MAC addresses per port.
10008 	 */
10009 	if (t4_num_vis < 1)
10010 		t4_num_vis = 1;
10011 	if (t4_num_vis > nitems(vi_mac_funcs)) {
10012 		t4_num_vis = nitems(vi_mac_funcs);
10013 		printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
10014 	}
10015 
10016 	if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
10017 		pcie_relaxed_ordering = 1;
10018 #if defined(__i386__) || defined(__amd64__)
10019 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
10020 			pcie_relaxed_ordering = 0;
10021 #endif
10022 	}
10023 }
10024 
10025 #ifdef DDB
10026 static void
10027 t4_dump_tcb(struct adapter *sc, int tid)
10028 {
10029 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
10030 
10031 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
10032 	save = t4_read_reg(sc, reg);
10033 	base = sc->memwin[2].mw_base;
10034 
10035 	/* Dump TCB for the tid */
10036 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
10037 	tcb_addr += tid * TCB_SIZE;
10038 
10039 	if (is_t4(sc)) {
10040 		pf = 0;
10041 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
10042 	} else {
10043 		pf = V_PFNUM(sc->pf);
10044 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
10045 	}
10046 	t4_write_reg(sc, reg, win_pos | pf);
10047 	t4_read_reg(sc, reg);
10048 
10049 	off = tcb_addr - win_pos;
10050 	for (i = 0; i < 4; i++) {
10051 		uint32_t buf[8];
10052 		for (j = 0; j < 8; j++, off += 4)
10053 			buf[j] = htonl(t4_read_reg(sc, base + off));
10054 
10055 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
10056 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
10057 		    buf[7]);
10058 	}
10059 
10060 	t4_write_reg(sc, reg, save);
10061 	t4_read_reg(sc, reg);
10062 }
10063 
10064 static void
10065 t4_dump_devlog(struct adapter *sc)
10066 {
10067 	struct devlog_params *dparams = &sc->params.devlog;
10068 	struct fw_devlog_e e;
10069 	int i, first, j, m, nentries, rc;
10070 	uint64_t ftstamp = UINT64_MAX;
10071 
10072 	if (dparams->start == 0) {
10073 		db_printf("devlog params not valid\n");
10074 		return;
10075 	}
10076 
10077 	nentries = dparams->size / sizeof(struct fw_devlog_e);
10078 	m = fwmtype_to_hwmtype(dparams->memtype);
10079 
10080 	/* Find the first entry. */
10081 	first = -1;
10082 	for (i = 0; i < nentries && !db_pager_quit; i++) {
10083 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10084 		    sizeof(e), (void *)&e);
10085 		if (rc != 0)
10086 			break;
10087 
10088 		if (e.timestamp == 0)
10089 			break;
10090 
10091 		e.timestamp = be64toh(e.timestamp);
10092 		if (e.timestamp < ftstamp) {
10093 			ftstamp = e.timestamp;
10094 			first = i;
10095 		}
10096 	}
10097 
10098 	if (first == -1)
10099 		return;
10100 
10101 	i = first;
10102 	do {
10103 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10104 		    sizeof(e), (void *)&e);
10105 		if (rc != 0)
10106 			return;
10107 
10108 		if (e.timestamp == 0)
10109 			return;
10110 
10111 		e.timestamp = be64toh(e.timestamp);
10112 		e.seqno = be32toh(e.seqno);
10113 		for (j = 0; j < 8; j++)
10114 			e.params[j] = be32toh(e.params[j]);
10115 
10116 		db_printf("%10d  %15ju  %8s  %8s  ",
10117 		    e.seqno, e.timestamp,
10118 		    (e.level < nitems(devlog_level_strings) ?
10119 			devlog_level_strings[e.level] : "UNKNOWN"),
10120 		    (e.facility < nitems(devlog_facility_strings) ?
10121 			devlog_facility_strings[e.facility] : "UNKNOWN"));
10122 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
10123 		    e.params[3], e.params[4], e.params[5], e.params[6],
10124 		    e.params[7]);
10125 
10126 		if (++i == nentries)
10127 			i = 0;
10128 	} while (i != first && !db_pager_quit);
10129 }
10130 
10131 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
10132 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
10133 
10134 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
10135 {
10136 	device_t dev;
10137 	int t;
10138 	bool valid;
10139 
10140 	valid = false;
10141 	t = db_read_token();
10142 	if (t == tIDENT) {
10143 		dev = device_lookup_by_name(db_tok_string);
10144 		valid = true;
10145 	}
10146 	db_skip_to_eol();
10147 	if (!valid) {
10148 		db_printf("usage: show t4 devlog <nexus>\n");
10149 		return;
10150 	}
10151 
10152 	if (dev == NULL) {
10153 		db_printf("device not found\n");
10154 		return;
10155 	}
10156 
10157 	t4_dump_devlog(device_get_softc(dev));
10158 }
10159 
10160 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
10161 {
10162 	device_t dev;
10163 	int radix, tid, t;
10164 	bool valid;
10165 
10166 	valid = false;
10167 	radix = db_radix;
10168 	db_radix = 10;
10169 	t = db_read_token();
10170 	if (t == tIDENT) {
10171 		dev = device_lookup_by_name(db_tok_string);
10172 		t = db_read_token();
10173 		if (t == tNUMBER) {
10174 			tid = db_tok_number;
10175 			valid = true;
10176 		}
10177 	}
10178 	db_radix = radix;
10179 	db_skip_to_eol();
10180 	if (!valid) {
10181 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
10182 		return;
10183 	}
10184 
10185 	if (dev == NULL) {
10186 		db_printf("device not found\n");
10187 		return;
10188 	}
10189 	if (tid < 0) {
10190 		db_printf("invalid tid\n");
10191 		return;
10192 	}
10193 
10194 	t4_dump_tcb(device_get_softc(dev), tid);
10195 }
10196 #endif
10197 
10198 /*
10199  * Borrowed from cesa_prep_aes_key().
10200  *
10201  * NB: The crypto engine wants the words in the decryption key in reverse
10202  * order.
10203  */
10204 void
10205 t4_aes_getdeckey(void *dec_key, const void *enc_key, unsigned int kbits)
10206 {
10207 	uint32_t ek[4 * (RIJNDAEL_MAXNR + 1)];
10208 	uint32_t *dkey;
10209 	int i;
10210 
10211 	rijndaelKeySetupEnc(ek, enc_key, kbits);
10212 	dkey = dec_key;
10213 	dkey += (kbits / 8) / 4;
10214 
10215 	switch (kbits) {
10216 	case 128:
10217 		for (i = 0; i < 4; i++)
10218 			*--dkey = htobe32(ek[4 * 10 + i]);
10219 		break;
10220 	case 192:
10221 		for (i = 0; i < 2; i++)
10222 			*--dkey = htobe32(ek[4 * 11 + 2 + i]);
10223 		for (i = 0; i < 4; i++)
10224 			*--dkey = htobe32(ek[4 * 12 + i]);
10225 		break;
10226 	case 256:
10227 		for (i = 0; i < 4; i++)
10228 			*--dkey = htobe32(ek[4 * 13 + i]);
10229 		for (i = 0; i < 4; i++)
10230 			*--dkey = htobe32(ek[4 * 14 + i]);
10231 		break;
10232 	}
10233 	MPASS(dkey == dec_key);
10234 }
10235 
10236 static struct sx mlu;	/* mod load unload */
10237 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
10238 
10239 static int
10240 mod_event(module_t mod, int cmd, void *arg)
10241 {
10242 	int rc = 0;
10243 	static int loaded = 0;
10244 
10245 	switch (cmd) {
10246 	case MOD_LOAD:
10247 		sx_xlock(&mlu);
10248 		if (loaded++ == 0) {
10249 			t4_sge_modload();
10250 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10251 			    t4_filter_rpl, CPL_COOKIE_FILTER);
10252 			t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
10253 			    do_l2t_write_rpl, CPL_COOKIE_FILTER);
10254 			t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL,
10255 			    t4_hashfilter_ao_rpl, CPL_COOKIE_HASHFILTER);
10256 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10257 			    t4_hashfilter_tcb_rpl, CPL_COOKIE_HASHFILTER);
10258 			t4_register_shared_cpl_handler(CPL_ABORT_RPL_RSS,
10259 			    t4_del_hashfilter_rpl, CPL_COOKIE_HASHFILTER);
10260 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
10261 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
10262 			t4_register_cpl_handler(CPL_SMT_WRITE_RPL,
10263 			    do_smt_write_rpl);
10264 			sx_init(&t4_list_lock, "T4/T5 adapters");
10265 			SLIST_INIT(&t4_list);
10266 #ifdef TCP_OFFLOAD
10267 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
10268 			SLIST_INIT(&t4_uld_list);
10269 #endif
10270 			t4_tracer_modload();
10271 			tweak_tunables();
10272 		}
10273 		sx_xunlock(&mlu);
10274 		break;
10275 
10276 	case MOD_UNLOAD:
10277 		sx_xlock(&mlu);
10278 		if (--loaded == 0) {
10279 			int tries;
10280 
10281 			sx_slock(&t4_list_lock);
10282 			if (!SLIST_EMPTY(&t4_list)) {
10283 				rc = EBUSY;
10284 				sx_sunlock(&t4_list_lock);
10285 				goto done_unload;
10286 			}
10287 #ifdef TCP_OFFLOAD
10288 			sx_slock(&t4_uld_list_lock);
10289 			if (!SLIST_EMPTY(&t4_uld_list)) {
10290 				rc = EBUSY;
10291 				sx_sunlock(&t4_uld_list_lock);
10292 				sx_sunlock(&t4_list_lock);
10293 				goto done_unload;
10294 			}
10295 #endif
10296 			tries = 0;
10297 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
10298 				uprintf("%ju clusters with custom free routine "
10299 				    "still is use.\n", t4_sge_extfree_refs());
10300 				pause("t4unload", 2 * hz);
10301 			}
10302 #ifdef TCP_OFFLOAD
10303 			sx_sunlock(&t4_uld_list_lock);
10304 #endif
10305 			sx_sunlock(&t4_list_lock);
10306 
10307 			if (t4_sge_extfree_refs() == 0) {
10308 				t4_tracer_modunload();
10309 #ifdef TCP_OFFLOAD
10310 				sx_destroy(&t4_uld_list_lock);
10311 #endif
10312 				sx_destroy(&t4_list_lock);
10313 				t4_sge_modunload();
10314 				loaded = 0;
10315 			} else {
10316 				rc = EBUSY;
10317 				loaded++;	/* undo earlier decrement */
10318 			}
10319 		}
10320 done_unload:
10321 		sx_xunlock(&mlu);
10322 		break;
10323 	}
10324 
10325 	return (rc);
10326 }
10327 
10328 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10329 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10330 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10331 
10332 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10333 MODULE_VERSION(t4nex, 1);
10334 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10335 #ifdef DEV_NETMAP
10336 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10337 #endif /* DEV_NETMAP */
10338 
10339 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10340 MODULE_VERSION(t5nex, 1);
10341 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10342 #ifdef DEV_NETMAP
10343 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10344 #endif /* DEV_NETMAP */
10345 
10346 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10347 MODULE_VERSION(t6nex, 1);
10348 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10349 #ifdef DEV_NETMAP
10350 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10351 #endif /* DEV_NETMAP */
10352 
10353 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10354 MODULE_VERSION(cxgbe, 1);
10355 
10356 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10357 MODULE_VERSION(cxl, 1);
10358 
10359 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10360 MODULE_VERSION(cc, 1);
10361 
10362 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10363 MODULE_VERSION(vcxgbe, 1);
10364 
10365 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10366 MODULE_VERSION(vcxl, 1);
10367 
10368 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10369 MODULE_VERSION(vcc, 1);
10370