xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision 39afde0dcc44e0033fabdbcc61e28d0358da1ba2)
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_rss.h"
37 
38 #include <sys/param.h>
39 #include <sys/conf.h>
40 #include <sys/priv.h>
41 #include <sys/kernel.h>
42 #include <sys/bus.h>
43 #include <sys/module.h>
44 #include <sys/malloc.h>
45 #include <sys/queue.h>
46 #include <sys/taskqueue.h>
47 #include <sys/pciio.h>
48 #include <dev/pci/pcireg.h>
49 #include <dev/pci/pcivar.h>
50 #include <dev/pci/pci_private.h>
51 #include <sys/firmware.h>
52 #include <sys/sbuf.h>
53 #include <sys/smp.h>
54 #include <sys/socket.h>
55 #include <sys/sockio.h>
56 #include <sys/sysctl.h>
57 #include <net/ethernet.h>
58 #include <net/if.h>
59 #include <net/if_types.h>
60 #include <net/if_dl.h>
61 #include <net/if_vlan_var.h>
62 #ifdef RSS
63 #include <net/rss_config.h>
64 #endif
65 #if defined(__i386__) || defined(__amd64__)
66 #include <machine/md_var.h>
67 #include <machine/cputypes.h>
68 #include <vm/vm.h>
69 #include <vm/pmap.h>
70 #endif
71 #ifdef DDB
72 #include <ddb/ddb.h>
73 #include <ddb/db_lex.h>
74 #endif
75 
76 #include "common/common.h"
77 #include "common/t4_msg.h"
78 #include "common/t4_regs.h"
79 #include "common/t4_regs_values.h"
80 #include "cudbg/cudbg.h"
81 #include "t4_ioctl.h"
82 #include "t4_l2t.h"
83 #include "t4_mp_ring.h"
84 #include "t4_if.h"
85 
86 /* T4 bus driver interface */
87 static int t4_probe(device_t);
88 static int t4_attach(device_t);
89 static int t4_detach(device_t);
90 static int t4_ready(device_t);
91 static int t4_read_port_device(device_t, int, device_t *);
92 static device_method_t t4_methods[] = {
93 	DEVMETHOD(device_probe,		t4_probe),
94 	DEVMETHOD(device_attach,	t4_attach),
95 	DEVMETHOD(device_detach,	t4_detach),
96 
97 	DEVMETHOD(t4_is_main_ready,	t4_ready),
98 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
99 
100 	DEVMETHOD_END
101 };
102 static driver_t t4_driver = {
103 	"t4nex",
104 	t4_methods,
105 	sizeof(struct adapter)
106 };
107 
108 
109 /* T4 port (cxgbe) interface */
110 static int cxgbe_probe(device_t);
111 static int cxgbe_attach(device_t);
112 static int cxgbe_detach(device_t);
113 device_method_t cxgbe_methods[] = {
114 	DEVMETHOD(device_probe,		cxgbe_probe),
115 	DEVMETHOD(device_attach,	cxgbe_attach),
116 	DEVMETHOD(device_detach,	cxgbe_detach),
117 	{ 0, 0 }
118 };
119 static driver_t cxgbe_driver = {
120 	"cxgbe",
121 	cxgbe_methods,
122 	sizeof(struct port_info)
123 };
124 
125 /* T4 VI (vcxgbe) interface */
126 static int vcxgbe_probe(device_t);
127 static int vcxgbe_attach(device_t);
128 static int vcxgbe_detach(device_t);
129 static device_method_t vcxgbe_methods[] = {
130 	DEVMETHOD(device_probe,		vcxgbe_probe),
131 	DEVMETHOD(device_attach,	vcxgbe_attach),
132 	DEVMETHOD(device_detach,	vcxgbe_detach),
133 	{ 0, 0 }
134 };
135 static driver_t vcxgbe_driver = {
136 	"vcxgbe",
137 	vcxgbe_methods,
138 	sizeof(struct vi_info)
139 };
140 
141 static d_ioctl_t t4_ioctl;
142 
143 static struct cdevsw t4_cdevsw = {
144        .d_version = D_VERSION,
145        .d_ioctl = t4_ioctl,
146        .d_name = "t4nex",
147 };
148 
149 /* T5 bus driver interface */
150 static int t5_probe(device_t);
151 static device_method_t t5_methods[] = {
152 	DEVMETHOD(device_probe,		t5_probe),
153 	DEVMETHOD(device_attach,	t4_attach),
154 	DEVMETHOD(device_detach,	t4_detach),
155 
156 	DEVMETHOD(t4_is_main_ready,	t4_ready),
157 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
158 
159 	DEVMETHOD_END
160 };
161 static driver_t t5_driver = {
162 	"t5nex",
163 	t5_methods,
164 	sizeof(struct adapter)
165 };
166 
167 
168 /* T5 port (cxl) interface */
169 static driver_t cxl_driver = {
170 	"cxl",
171 	cxgbe_methods,
172 	sizeof(struct port_info)
173 };
174 
175 /* T5 VI (vcxl) interface */
176 static driver_t vcxl_driver = {
177 	"vcxl",
178 	vcxgbe_methods,
179 	sizeof(struct vi_info)
180 };
181 
182 /* T6 bus driver interface */
183 static int t6_probe(device_t);
184 static device_method_t t6_methods[] = {
185 	DEVMETHOD(device_probe,		t6_probe),
186 	DEVMETHOD(device_attach,	t4_attach),
187 	DEVMETHOD(device_detach,	t4_detach),
188 
189 	DEVMETHOD(t4_is_main_ready,	t4_ready),
190 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
191 
192 	DEVMETHOD_END
193 };
194 static driver_t t6_driver = {
195 	"t6nex",
196 	t6_methods,
197 	sizeof(struct adapter)
198 };
199 
200 
201 /* T6 port (cc) interface */
202 static driver_t cc_driver = {
203 	"cc",
204 	cxgbe_methods,
205 	sizeof(struct port_info)
206 };
207 
208 /* T6 VI (vcc) interface */
209 static driver_t vcc_driver = {
210 	"vcc",
211 	vcxgbe_methods,
212 	sizeof(struct vi_info)
213 };
214 
215 /* ifnet + media interface */
216 static void cxgbe_init(void *);
217 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
218 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
219 static void cxgbe_qflush(struct ifnet *);
220 static int cxgbe_media_change(struct ifnet *);
221 static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
222 
223 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
224 
225 /*
226  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
227  * then ADAPTER_LOCK, then t4_uld_list_lock.
228  */
229 static struct sx t4_list_lock;
230 SLIST_HEAD(, adapter) t4_list;
231 #ifdef TCP_OFFLOAD
232 static struct sx t4_uld_list_lock;
233 SLIST_HEAD(, uld_info) t4_uld_list;
234 #endif
235 
236 /*
237  * Tunables.  See tweak_tunables() too.
238  *
239  * Each tunable is set to a default value here if it's known at compile-time.
240  * Otherwise it is set to -n as an indication to tweak_tunables() that it should
241  * provide a reasonable default (upto n) when the driver is loaded.
242  *
243  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
244  * T5 are under hw.cxl.
245  */
246 
247 /*
248  * Number of queues for tx and rx, NIC and offload.
249  */
250 #define NTXQ 16
251 int t4_ntxq = -NTXQ;
252 TUNABLE_INT("hw.cxgbe.ntxq", &t4_ntxq);
253 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq);	/* Old name, undocumented */
254 
255 #define NRXQ 8
256 int t4_nrxq = -NRXQ;
257 TUNABLE_INT("hw.cxgbe.nrxq", &t4_nrxq);
258 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq);	/* Old name, undocumented */
259 
260 #define NTXQ_VI 1
261 static int t4_ntxq_vi = -NTXQ_VI;
262 TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
263 
264 #define NRXQ_VI 1
265 static int t4_nrxq_vi = -NRXQ_VI;
266 TUNABLE_INT("hw.cxgbe.nrxq_vi", &t4_nrxq_vi);
267 
268 static int t4_rsrv_noflowq = 0;
269 TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
270 
271 #ifdef TCP_OFFLOAD
272 #define NOFLDTXQ 8
273 static int t4_nofldtxq = -NOFLDTXQ;
274 TUNABLE_INT("hw.cxgbe.nofldtxq", &t4_nofldtxq);
275 
276 #define NOFLDRXQ 2
277 static int t4_nofldrxq = -NOFLDRXQ;
278 TUNABLE_INT("hw.cxgbe.nofldrxq", &t4_nofldrxq);
279 
280 #define NOFLDTXQ_VI 1
281 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
282 TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
283 
284 #define NOFLDRXQ_VI 1
285 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
286 TUNABLE_INT("hw.cxgbe.nofldrxq_vi", &t4_nofldrxq_vi);
287 
288 #define TMR_IDX_OFLD 1
289 int t4_tmr_idx_ofld = TMR_IDX_OFLD;
290 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_ofld", &t4_tmr_idx_ofld);
291 
292 #define PKTC_IDX_OFLD (-1)
293 int t4_pktc_idx_ofld = PKTC_IDX_OFLD;
294 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_ofld", &t4_pktc_idx_ofld);
295 
296 /* 0 means chip/fw default, non-zero number is value in microseconds */
297 static u_long t4_toe_keepalive_idle = 0;
298 TUNABLE_ULONG("hw.cxgbe.toe.keepalive_idle", &t4_toe_keepalive_idle);
299 
300 /* 0 means chip/fw default, non-zero number is value in microseconds */
301 static u_long t4_toe_keepalive_interval = 0;
302 TUNABLE_ULONG("hw.cxgbe.toe.keepalive_interval", &t4_toe_keepalive_interval);
303 
304 /* 0 means chip/fw default, non-zero number is # of keepalives before abort */
305 static int t4_toe_keepalive_count = 0;
306 TUNABLE_INT("hw.cxgbe.toe.keepalive_count", &t4_toe_keepalive_count);
307 
308 /* 0 means chip/fw default, non-zero number is value in microseconds */
309 static u_long t4_toe_rexmt_min = 0;
310 TUNABLE_ULONG("hw.cxgbe.toe.rexmt_min", &t4_toe_rexmt_min);
311 
312 /* 0 means chip/fw default, non-zero number is value in microseconds */
313 static u_long t4_toe_rexmt_max = 0;
314 TUNABLE_ULONG("hw.cxgbe.toe.rexmt_max", &t4_toe_rexmt_max);
315 
316 /* 0 means chip/fw default, non-zero number is # of rexmt before abort */
317 static int t4_toe_rexmt_count = 0;
318 TUNABLE_INT("hw.cxgbe.toe.rexmt_count", &t4_toe_rexmt_count);
319 
320 /* -1 means chip/fw default, other values are raw backoff values to use */
321 static int t4_toe_rexmt_backoff[16] = {
322 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
323 };
324 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.0", &t4_toe_rexmt_backoff[0]);
325 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.1", &t4_toe_rexmt_backoff[1]);
326 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.2", &t4_toe_rexmt_backoff[2]);
327 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.3", &t4_toe_rexmt_backoff[3]);
328 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.4", &t4_toe_rexmt_backoff[4]);
329 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.5", &t4_toe_rexmt_backoff[5]);
330 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.6", &t4_toe_rexmt_backoff[6]);
331 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.7", &t4_toe_rexmt_backoff[7]);
332 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.8", &t4_toe_rexmt_backoff[8]);
333 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.9", &t4_toe_rexmt_backoff[9]);
334 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.10", &t4_toe_rexmt_backoff[10]);
335 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.11", &t4_toe_rexmt_backoff[11]);
336 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.12", &t4_toe_rexmt_backoff[12]);
337 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.13", &t4_toe_rexmt_backoff[13]);
338 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.14", &t4_toe_rexmt_backoff[14]);
339 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.15", &t4_toe_rexmt_backoff[15]);
340 #endif
341 
342 #ifdef DEV_NETMAP
343 #define NNMTXQ_VI 2
344 static int t4_nnmtxq_vi = -NNMTXQ_VI;
345 TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
346 
347 #define NNMRXQ_VI 2
348 static int t4_nnmrxq_vi = -NNMRXQ_VI;
349 TUNABLE_INT("hw.cxgbe.nnmrxq_vi", &t4_nnmrxq_vi);
350 #endif
351 
352 /*
353  * Holdoff parameters for ports.
354  */
355 #define TMR_IDX 1
356 int t4_tmr_idx = TMR_IDX;
357 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx", &t4_tmr_idx);
358 
359 #define PKTC_IDX (-1)
360 int t4_pktc_idx = PKTC_IDX;
361 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx", &t4_pktc_idx);
362 
363 /*
364  * Size (# of entries) of each tx and rx queue.
365  */
366 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
367 TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
368 
369 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
370 TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
371 
372 /*
373  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
374  */
375 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
376 TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
377 
378 /*
379  * Configuration file.
380  */
381 #define DEFAULT_CF	"default"
382 #define FLASH_CF	"flash"
383 #define UWIRE_CF	"uwire"
384 #define FPGA_CF		"fpga"
385 static char t4_cfg_file[32] = DEFAULT_CF;
386 TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
387 
388 /*
389  * PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
390  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
391  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
392  *            mark or when signalled to do so, 0 to never emit PAUSE.
393  */
394 static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
395 TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
396 
397 /*
398  * Forward Error Correction settings (bit 0, 1, 2 = FEC_RS, FEC_BASER_RS,
399  * FEC_RESERVED respectively).
400  * -1 to run with the firmware default.
401  *  0 to disable FEC.
402  */
403 static int t4_fec = -1;
404 TUNABLE_INT("hw.cxgbe.fec", &t4_fec);
405 
406 /*
407  * Link autonegotiation.
408  * -1 to run with the firmware default.
409  *  0 to disable.
410  *  1 to enable.
411  */
412 static int t4_autoneg = -1;
413 TUNABLE_INT("hw.cxgbe.autoneg", &t4_autoneg);
414 
415 /*
416  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
417  * encouraged respectively).
418  */
419 static unsigned int t4_fw_install = 1;
420 TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
421 
422 /*
423  * ASIC features that will be used.  Disable the ones you don't want so that the
424  * chip resources aren't wasted on features that will not be used.
425  */
426 static int t4_nbmcaps_allowed = 0;
427 TUNABLE_INT("hw.cxgbe.nbmcaps_allowed", &t4_nbmcaps_allowed);
428 
429 static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
430 TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
431 
432 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
433     FW_CAPS_CONFIG_SWITCH_EGRESS;
434 TUNABLE_INT("hw.cxgbe.switchcaps_allowed", &t4_switchcaps_allowed);
435 
436 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC;
437 TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
438 
439 static int t4_toecaps_allowed = -1;
440 TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
441 
442 static int t4_rdmacaps_allowed = -1;
443 TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
444 
445 static int t4_cryptocaps_allowed = -1;
446 TUNABLE_INT("hw.cxgbe.cryptocaps_allowed", &t4_cryptocaps_allowed);
447 
448 static int t4_iscsicaps_allowed = -1;
449 TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
450 
451 static int t4_fcoecaps_allowed = 0;
452 TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
453 
454 static int t5_write_combine = 1;
455 TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
456 
457 static int t4_num_vis = 1;
458 TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
459 /*
460  * PCIe Relaxed Ordering.
461  * -1: driver should figure out a good value.
462  * 0: disable RO.
463  * 1: enable RO.
464  * 2: leave RO alone.
465  */
466 static int pcie_relaxed_ordering = -1;
467 TUNABLE_INT("hw.cxgbe.pcie_relaxed_ordering", &pcie_relaxed_ordering);
468 
469 
470 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
471 static int vi_mac_funcs[] = {
472 	FW_VI_FUNC_ETH,
473 	FW_VI_FUNC_OFLD,
474 	FW_VI_FUNC_IWARP,
475 	FW_VI_FUNC_OPENISCSI,
476 	FW_VI_FUNC_OPENFCOE,
477 	FW_VI_FUNC_FOISCSI,
478 	FW_VI_FUNC_FOFCOE,
479 };
480 
481 struct intrs_and_queues {
482 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
483 	uint16_t num_vis;	/* number of VIs for each port */
484 	uint16_t nirq;		/* Total # of vectors */
485 	uint16_t ntxq;		/* # of NIC txq's for each port */
486 	uint16_t nrxq;		/* # of NIC rxq's for each port */
487 	uint16_t nofldtxq;	/* # of TOE txq's for each port */
488 	uint16_t nofldrxq;	/* # of TOE rxq's for each port */
489 
490 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
491 	uint16_t ntxq_vi;	/* # of NIC txq's */
492 	uint16_t nrxq_vi;	/* # of NIC rxq's */
493 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
494 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
495 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
496 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
497 };
498 
499 struct filter_entry {
500         uint32_t valid:1;	/* filter allocated and valid */
501         uint32_t locked:1;	/* filter is administratively locked */
502         uint32_t pending:1;	/* filter action is pending firmware reply */
503 	uint32_t smtidx:8;	/* Source MAC Table index for smac */
504 	struct l2t_entry *l2t;	/* Layer Two Table entry for dmac */
505 
506         struct t4_filter_specification fs;
507 };
508 
509 static void setup_memwin(struct adapter *);
510 static void position_memwin(struct adapter *, int, uint32_t);
511 static int rw_via_memwin(struct adapter *, int, uint32_t, uint32_t *, int, int);
512 static inline int read_via_memwin(struct adapter *, int, uint32_t, uint32_t *,
513     int);
514 static inline int write_via_memwin(struct adapter *, int, uint32_t,
515     const uint32_t *, int);
516 static int validate_mem_range(struct adapter *, uint32_t, int);
517 static int fwmtype_to_hwmtype(int);
518 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
519     uint32_t *);
520 static int fixup_devlog_params(struct adapter *);
521 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
522 static int prep_firmware(struct adapter *);
523 static int partition_resources(struct adapter *, const struct firmware *,
524     const char *);
525 static int get_params__pre_init(struct adapter *);
526 static int get_params__post_init(struct adapter *);
527 static int set_params__post_init(struct adapter *);
528 static void t4_set_desc(struct adapter *);
529 static void build_medialist(struct port_info *, struct ifmedia *);
530 static void init_l1cfg(struct port_info *);
531 static int cxgbe_init_synchronized(struct vi_info *);
532 static int cxgbe_uninit_synchronized(struct vi_info *);
533 static void quiesce_txq(struct adapter *, struct sge_txq *);
534 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
535 static void quiesce_iq(struct adapter *, struct sge_iq *);
536 static void quiesce_fl(struct adapter *, struct sge_fl *);
537 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
538     driver_intr_t *, void *, char *);
539 static int t4_free_irq(struct adapter *, struct irq *);
540 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
541 static void vi_refresh_stats(struct adapter *, struct vi_info *);
542 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
543 static void cxgbe_tick(void *);
544 static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
545 static void cxgbe_sysctls(struct port_info *);
546 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
547 static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
548 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
549 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
550 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
551 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
552 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
553 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
554 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
555 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
556 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
557 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
558 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
559 #ifdef SBUF_DRAIN
560 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
561 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
562 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
563 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
564 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
565 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
566 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
567 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
568 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
569 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
570 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
571 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
572 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
573 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
574 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
575 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
576 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
577 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
578 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
579 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
580 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
581 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
582 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
583 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
584 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
585 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
586 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
587 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
588 static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
589 #endif
590 #ifdef TCP_OFFLOAD
591 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
592 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
593 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
594 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
595 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
596 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
597 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
598 #endif
599 static uint32_t fconf_iconf_to_mode(uint32_t, uint32_t);
600 static uint32_t mode_to_fconf(uint32_t);
601 static uint32_t mode_to_iconf(uint32_t);
602 static int check_fspec_against_fconf_iconf(struct adapter *,
603     struct t4_filter_specification *);
604 static int get_filter_mode(struct adapter *, uint32_t *);
605 static int set_filter_mode(struct adapter *, uint32_t);
606 static inline uint64_t get_filter_hits(struct adapter *, uint32_t);
607 static int get_filter(struct adapter *, struct t4_filter *);
608 static int set_filter(struct adapter *, struct t4_filter *);
609 static int del_filter(struct adapter *, struct t4_filter *);
610 static void clear_filter(struct filter_entry *);
611 static int set_filter_wr(struct adapter *, int);
612 static int del_filter_wr(struct adapter *, int);
613 static int set_tcb_rpl(struct sge_iq *, const struct rss_header *,
614     struct mbuf *);
615 static int get_sge_context(struct adapter *, struct t4_sge_context *);
616 static int load_fw(struct adapter *, struct t4_data *);
617 static int load_cfg(struct adapter *, struct t4_data *);
618 static int load_boot(struct adapter *, struct t4_bootrom *);
619 static int load_bootcfg(struct adapter *, struct t4_data *);
620 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
621 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
622 static int read_i2c(struct adapter *, struct t4_i2c_data *);
623 #ifdef TCP_OFFLOAD
624 static int toe_capability(struct vi_info *, int);
625 #endif
626 static int mod_event(module_t, int, void *);
627 static int notify_siblings(device_t, int);
628 
629 struct {
630 	uint16_t device;
631 	char *desc;
632 } t4_pciids[] = {
633 	{0xa000, "Chelsio Terminator 4 FPGA"},
634 	{0x4400, "Chelsio T440-dbg"},
635 	{0x4401, "Chelsio T420-CR"},
636 	{0x4402, "Chelsio T422-CR"},
637 	{0x4403, "Chelsio T440-CR"},
638 	{0x4404, "Chelsio T420-BCH"},
639 	{0x4405, "Chelsio T440-BCH"},
640 	{0x4406, "Chelsio T440-CH"},
641 	{0x4407, "Chelsio T420-SO"},
642 	{0x4408, "Chelsio T420-CX"},
643 	{0x4409, "Chelsio T420-BT"},
644 	{0x440a, "Chelsio T404-BT"},
645 	{0x440e, "Chelsio T440-LP-CR"},
646 }, t5_pciids[] = {
647 	{0xb000, "Chelsio Terminator 5 FPGA"},
648 	{0x5400, "Chelsio T580-dbg"},
649 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
650 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
651 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
652 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
653 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
654 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
655 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
656 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
657 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
658 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
659 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
660 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
661 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
662 #ifdef notyet
663 	{0x5404,  "Chelsio T520-BCH"},
664 	{0x5405,  "Chelsio T540-BCH"},
665 	{0x5406,  "Chelsio T540-CH"},
666 	{0x5408,  "Chelsio T520-CX"},
667 	{0x540b,  "Chelsio B520-SR"},
668 	{0x540c,  "Chelsio B504-BT"},
669 	{0x540f,  "Chelsio Amsterdam"},
670 	{0x5413,  "Chelsio T580-CHR"},
671 #endif
672 }, t6_pciids[] = {
673 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
674 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
675 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
676 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
677 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
678 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
679 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
680 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
681 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
682 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
683 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
684 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
685 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
686 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
687 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
688 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
689 
690 	/* Custom */
691 	{0x6480, "Chelsio T6225 80"},
692 	{0x6481, "Chelsio T62100 81"},
693 	{0x6484, "Chelsio T62100 84"},
694 };
695 
696 #ifdef TCP_OFFLOAD
697 /*
698  * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
699  * exactly the same for both rxq and ofld_rxq.
700  */
701 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
702 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
703 #endif
704 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
705 
706 static int
707 t4_probe(device_t dev)
708 {
709 	int i;
710 	uint16_t v = pci_get_vendor(dev);
711 	uint16_t d = pci_get_device(dev);
712 	uint8_t f = pci_get_function(dev);
713 
714 	if (v != PCI_VENDOR_ID_CHELSIO)
715 		return (ENXIO);
716 
717 	/* Attach only to PF0 of the FPGA */
718 	if (d == 0xa000 && f != 0)
719 		return (ENXIO);
720 
721 	for (i = 0; i < nitems(t4_pciids); i++) {
722 		if (d == t4_pciids[i].device) {
723 			device_set_desc(dev, t4_pciids[i].desc);
724 			return (BUS_PROBE_DEFAULT);
725 		}
726 	}
727 
728 	return (ENXIO);
729 }
730 
731 static int
732 t5_probe(device_t dev)
733 {
734 	int i;
735 	uint16_t v = pci_get_vendor(dev);
736 	uint16_t d = pci_get_device(dev);
737 	uint8_t f = pci_get_function(dev);
738 
739 	if (v != PCI_VENDOR_ID_CHELSIO)
740 		return (ENXIO);
741 
742 	/* Attach only to PF0 of the FPGA */
743 	if (d == 0xb000 && f != 0)
744 		return (ENXIO);
745 
746 	for (i = 0; i < nitems(t5_pciids); i++) {
747 		if (d == t5_pciids[i].device) {
748 			device_set_desc(dev, t5_pciids[i].desc);
749 			return (BUS_PROBE_DEFAULT);
750 		}
751 	}
752 
753 	return (ENXIO);
754 }
755 
756 static int
757 t6_probe(device_t dev)
758 {
759 	int i;
760 	uint16_t v = pci_get_vendor(dev);
761 	uint16_t d = pci_get_device(dev);
762 
763 	if (v != PCI_VENDOR_ID_CHELSIO)
764 		return (ENXIO);
765 
766 	for (i = 0; i < nitems(t6_pciids); i++) {
767 		if (d == t6_pciids[i].device) {
768 			device_set_desc(dev, t6_pciids[i].desc);
769 			return (BUS_PROBE_DEFAULT);
770 		}
771 	}
772 
773 	return (ENXIO);
774 }
775 
776 static void
777 t5_attribute_workaround(device_t dev)
778 {
779 	device_t root_port;
780 	uint32_t v;
781 
782 	/*
783 	 * The T5 chips do not properly echo the No Snoop and Relaxed
784 	 * Ordering attributes when replying to a TLP from a Root
785 	 * Port.  As a workaround, find the parent Root Port and
786 	 * disable No Snoop and Relaxed Ordering.  Note that this
787 	 * affects all devices under this root port.
788 	 */
789 	root_port = pci_find_pcie_root_port(dev);
790 	if (root_port == NULL) {
791 		device_printf(dev, "Unable to find parent root port\n");
792 		return;
793 	}
794 
795 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
796 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
797 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
798 	    0)
799 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
800 		    device_get_nameunit(root_port));
801 }
802 
803 static const struct devnames devnames[] = {
804 	{
805 		.nexus_name = "t4nex",
806 		.ifnet_name = "cxgbe",
807 		.vi_ifnet_name = "vcxgbe",
808 		.pf03_drv_name = "t4iov",
809 		.vf_nexus_name = "t4vf",
810 		.vf_ifnet_name = "cxgbev"
811 	}, {
812 		.nexus_name = "t5nex",
813 		.ifnet_name = "cxl",
814 		.vi_ifnet_name = "vcxl",
815 		.pf03_drv_name = "t5iov",
816 		.vf_nexus_name = "t5vf",
817 		.vf_ifnet_name = "cxlv"
818 	}, {
819 		.nexus_name = "t6nex",
820 		.ifnet_name = "cc",
821 		.vi_ifnet_name = "vcc",
822 		.pf03_drv_name = "t6iov",
823 		.vf_nexus_name = "t6vf",
824 		.vf_ifnet_name = "ccv"
825 	}
826 };
827 
828 void
829 t4_init_devnames(struct adapter *sc)
830 {
831 	int id;
832 
833 	id = chip_id(sc);
834 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
835 		sc->names = &devnames[id - CHELSIO_T4];
836 	else {
837 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
838 		sc->names = NULL;
839 	}
840 }
841 
842 static int
843 t4_attach(device_t dev)
844 {
845 	struct adapter *sc;
846 	int rc = 0, i, j, rqidx, tqidx, nports;
847 	struct make_dev_args mda;
848 	struct intrs_and_queues iaq;
849 	struct sge *s;
850 	uint32_t *buf;
851 #ifdef TCP_OFFLOAD
852 	int ofld_rqidx, ofld_tqidx;
853 #endif
854 #ifdef DEV_NETMAP
855 	int nm_rqidx, nm_tqidx;
856 #endif
857 	int num_vis;
858 
859 	sc = device_get_softc(dev);
860 	sc->dev = dev;
861 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
862 
863 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
864 		t5_attribute_workaround(dev);
865 	pci_enable_busmaster(dev);
866 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
867 		uint32_t v;
868 
869 		pci_set_max_read_req(dev, 4096);
870 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
871 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
872 		if (pcie_relaxed_ordering == 0 &&
873 		    (v | PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
874 			v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
875 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
876 		} else if (pcie_relaxed_ordering == 1 &&
877 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
878 			v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
879 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
880 		}
881 	}
882 
883 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
884 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
885 	sc->traceq = -1;
886 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
887 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
888 	    device_get_nameunit(dev));
889 
890 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
891 	    device_get_nameunit(dev));
892 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
893 	t4_add_adapter(sc);
894 
895 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
896 	TAILQ_INIT(&sc->sfl);
897 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
898 
899 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
900 
901 	rc = t4_map_bars_0_and_4(sc);
902 	if (rc != 0)
903 		goto done; /* error message displayed already */
904 
905 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
906 
907 	/* Prepare the adapter for operation. */
908 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
909 	rc = -t4_prep_adapter(sc, buf);
910 	free(buf, M_CXGBE);
911 	if (rc != 0) {
912 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
913 		goto done;
914 	}
915 
916 	/*
917 	 * This is the real PF# to which we're attaching.  Works from within PCI
918 	 * passthrough environments too, where pci_get_function() could return a
919 	 * different PF# depending on the passthrough configuration.  We need to
920 	 * use the real PF# in all our communication with the firmware.
921 	 */
922 	j = t4_read_reg(sc, A_PL_WHOAMI);
923 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
924 	sc->mbox = sc->pf;
925 
926 	t4_init_devnames(sc);
927 	if (sc->names == NULL) {
928 		rc = ENOTSUP;
929 		goto done; /* error message displayed already */
930 	}
931 
932 	/*
933 	 * Do this really early, with the memory windows set up even before the
934 	 * character device.  The userland tool's register i/o and mem read
935 	 * will work even in "recovery mode".
936 	 */
937 	setup_memwin(sc);
938 	if (t4_init_devlog_params(sc, 0) == 0)
939 		fixup_devlog_params(sc);
940 	make_dev_args_init(&mda);
941 	mda.mda_devsw = &t4_cdevsw;
942 	mda.mda_uid = UID_ROOT;
943 	mda.mda_gid = GID_WHEEL;
944 	mda.mda_mode = 0600;
945 	mda.mda_si_drv1 = sc;
946 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
947 	if (rc != 0)
948 		device_printf(dev, "failed to create nexus char device: %d.\n",
949 		    rc);
950 
951 	/* Go no further if recovery mode has been requested. */
952 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
953 		device_printf(dev, "recovery mode.\n");
954 		goto done;
955 	}
956 
957 #if defined(__i386__)
958 	if ((cpu_feature & CPUID_CX8) == 0) {
959 		device_printf(dev, "64 bit atomics not available.\n");
960 		rc = ENOTSUP;
961 		goto done;
962 	}
963 #endif
964 
965 	/* Prepare the firmware for operation */
966 	rc = prep_firmware(sc);
967 	if (rc != 0)
968 		goto done; /* error message displayed already */
969 
970 	rc = get_params__post_init(sc);
971 	if (rc != 0)
972 		goto done; /* error message displayed already */
973 
974 	rc = set_params__post_init(sc);
975 	if (rc != 0)
976 		goto done; /* error message displayed already */
977 
978 	rc = t4_map_bar_2(sc);
979 	if (rc != 0)
980 		goto done; /* error message displayed already */
981 
982 	rc = t4_create_dma_tag(sc);
983 	if (rc != 0)
984 		goto done; /* error message displayed already */
985 
986 	/*
987 	 * First pass over all the ports - allocate VIs and initialize some
988 	 * basic parameters like mac address, port type, etc.
989 	 */
990 	for_each_port(sc, i) {
991 		struct port_info *pi;
992 
993 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
994 		sc->port[i] = pi;
995 
996 		/* These must be set before t4_port_init */
997 		pi->adapter = sc;
998 		pi->port_id = i;
999 		/*
1000 		 * XXX: vi[0] is special so we can't delay this allocation until
1001 		 * pi->nvi's final value is known.
1002 		 */
1003 		pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
1004 		    M_ZERO | M_WAITOK);
1005 
1006 		/*
1007 		 * Allocate the "main" VI and initialize parameters
1008 		 * like mac addr.
1009 		 */
1010 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1011 		if (rc != 0) {
1012 			device_printf(dev, "unable to initialize port %d: %d\n",
1013 			    i, rc);
1014 			free(pi->vi, M_CXGBE);
1015 			free(pi, M_CXGBE);
1016 			sc->port[i] = NULL;
1017 			goto done;
1018 		}
1019 
1020 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1021 		    device_get_nameunit(dev), i);
1022 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1023 		sc->chan_map[pi->tx_chan] = i;
1024 
1025 		/* All VIs on this port share this media. */
1026 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1027 		    cxgbe_media_status);
1028 
1029 		pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
1030 		if (pi->dev == NULL) {
1031 			device_printf(dev,
1032 			    "failed to add device for port %d.\n", i);
1033 			rc = ENXIO;
1034 			goto done;
1035 		}
1036 		pi->vi[0].dev = pi->dev;
1037 		device_set_softc(pi->dev, pi);
1038 	}
1039 
1040 	/*
1041 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1042 	 */
1043 	nports = sc->params.nports;
1044 	rc = cfg_itype_and_nqueues(sc, &iaq);
1045 	if (rc != 0)
1046 		goto done; /* error message displayed already */
1047 
1048 	num_vis = iaq.num_vis;
1049 	sc->intr_type = iaq.intr_type;
1050 	sc->intr_count = iaq.nirq;
1051 
1052 	s = &sc->sge;
1053 	s->nrxq = nports * iaq.nrxq;
1054 	s->ntxq = nports * iaq.ntxq;
1055 	if (num_vis > 1) {
1056 		s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1057 		s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1058 	}
1059 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1060 	s->neq += nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
1061 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1062 #ifdef TCP_OFFLOAD
1063 	if (is_offload(sc)) {
1064 		s->nofldrxq = nports * iaq.nofldrxq;
1065 		s->nofldtxq = nports * iaq.nofldtxq;
1066 		if (num_vis > 1) {
1067 			s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1068 			s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1069 		}
1070 		s->neq += s->nofldtxq + s->nofldrxq;
1071 		s->niq += s->nofldrxq;
1072 
1073 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1074 		    M_CXGBE, M_ZERO | M_WAITOK);
1075 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
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_tx_sched(sc);
1109 
1110 	/*
1111 	 * Second pass over the ports.  This time we know the number of rx and
1112 	 * tx queues that each port should get.
1113 	 */
1114 	rqidx = tqidx = 0;
1115 #ifdef TCP_OFFLOAD
1116 	ofld_rqidx = ofld_tqidx = 0;
1117 #endif
1118 #ifdef DEV_NETMAP
1119 	nm_rqidx = nm_tqidx = 0;
1120 #endif
1121 	for_each_port(sc, i) {
1122 		struct port_info *pi = sc->port[i];
1123 		struct vi_info *vi;
1124 
1125 		if (pi == NULL)
1126 			continue;
1127 
1128 		pi->nvi = num_vis;
1129 		for_each_vi(pi, j, vi) {
1130 			vi->pi = pi;
1131 			vi->qsize_rxq = t4_qsize_rxq;
1132 			vi->qsize_txq = t4_qsize_txq;
1133 
1134 			vi->first_rxq = rqidx;
1135 			vi->first_txq = tqidx;
1136 			vi->tmr_idx = t4_tmr_idx;
1137 			vi->pktc_idx = t4_pktc_idx;
1138 			vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1139 			vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1140 
1141 			rqidx += vi->nrxq;
1142 			tqidx += vi->ntxq;
1143 
1144 			if (j == 0 && vi->ntxq > 1)
1145 				vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1146 			else
1147 				vi->rsrv_noflowq = 0;
1148 
1149 #ifdef TCP_OFFLOAD
1150 			vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1151 			vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1152 			vi->first_ofld_rxq = ofld_rqidx;
1153 			vi->first_ofld_txq = ofld_tqidx;
1154 			vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1155 			vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1156 
1157 			ofld_rqidx += vi->nofldrxq;
1158 			ofld_tqidx += vi->nofldtxq;
1159 #endif
1160 #ifdef DEV_NETMAP
1161 			if (j > 0) {
1162 				vi->first_nm_rxq = nm_rqidx;
1163 				vi->first_nm_txq = nm_tqidx;
1164 				vi->nnmrxq = iaq.nnmrxq_vi;
1165 				vi->nnmtxq = iaq.nnmtxq_vi;
1166 				nm_rqidx += vi->nnmrxq;
1167 				nm_tqidx += vi->nnmtxq;
1168 			}
1169 #endif
1170 		}
1171 	}
1172 
1173 	rc = t4_setup_intr_handlers(sc);
1174 	if (rc != 0) {
1175 		device_printf(dev,
1176 		    "failed to setup interrupt handlers: %d\n", rc);
1177 		goto done;
1178 	}
1179 
1180 	rc = bus_generic_probe(dev);
1181 	if (rc != 0) {
1182 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1183 		goto done;
1184 	}
1185 
1186 	/*
1187 	 * Ensure thread-safe mailbox access (in debug builds).
1188 	 *
1189 	 * So far this was the only thread accessing the mailbox but various
1190 	 * ifnets and sysctls are about to be created and their handlers/ioctls
1191 	 * will access the mailbox from different threads.
1192 	 */
1193 	sc->flags |= CHK_MBOX_ACCESS;
1194 
1195 	rc = bus_generic_attach(dev);
1196 	if (rc != 0) {
1197 		device_printf(dev,
1198 		    "failed to attach all child ports: %d\n", rc);
1199 		goto done;
1200 	}
1201 
1202 	device_printf(dev,
1203 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1204 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1205 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1206 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1207 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1208 
1209 	t4_set_desc(sc);
1210 
1211 	notify_siblings(dev, 0);
1212 
1213 done:
1214 	if (rc != 0 && sc->cdev) {
1215 		/* cdev was created and so cxgbetool works; recover that way. */
1216 		device_printf(dev,
1217 		    "error during attach, adapter is now in recovery mode.\n");
1218 		rc = 0;
1219 	}
1220 
1221 	if (rc != 0)
1222 		t4_detach_common(dev);
1223 	else
1224 		t4_sysctls(sc);
1225 
1226 	return (rc);
1227 }
1228 
1229 static int
1230 t4_ready(device_t dev)
1231 {
1232 	struct adapter *sc;
1233 
1234 	sc = device_get_softc(dev);
1235 	if (sc->flags & FW_OK)
1236 		return (0);
1237 	return (ENXIO);
1238 }
1239 
1240 static int
1241 t4_read_port_device(device_t dev, int port, device_t *child)
1242 {
1243 	struct adapter *sc;
1244 	struct port_info *pi;
1245 
1246 	sc = device_get_softc(dev);
1247 	if (port < 0 || port >= MAX_NPORTS)
1248 		return (EINVAL);
1249 	pi = sc->port[port];
1250 	if (pi == NULL || pi->dev == NULL)
1251 		return (ENXIO);
1252 	*child = pi->dev;
1253 	return (0);
1254 }
1255 
1256 static int
1257 notify_siblings(device_t dev, int detaching)
1258 {
1259 	device_t sibling;
1260 	int error, i;
1261 
1262 	error = 0;
1263 	for (i = 0; i < PCI_FUNCMAX; i++) {
1264 		if (i == pci_get_function(dev))
1265 			continue;
1266 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1267 		    pci_get_slot(dev), i);
1268 		if (sibling == NULL || !device_is_attached(sibling))
1269 			continue;
1270 		if (detaching)
1271 			error = T4_DETACH_CHILD(sibling);
1272 		else
1273 			(void)T4_ATTACH_CHILD(sibling);
1274 		if (error)
1275 			break;
1276 	}
1277 	return (error);
1278 }
1279 
1280 /*
1281  * Idempotent
1282  */
1283 static int
1284 t4_detach(device_t dev)
1285 {
1286 	struct adapter *sc;
1287 	int rc;
1288 
1289 	sc = device_get_softc(dev);
1290 
1291 	rc = notify_siblings(dev, 1);
1292 	if (rc) {
1293 		device_printf(dev,
1294 		    "failed to detach sibling devices: %d\n", rc);
1295 		return (rc);
1296 	}
1297 
1298 	return (t4_detach_common(dev));
1299 }
1300 
1301 int
1302 t4_detach_common(device_t dev)
1303 {
1304 	struct adapter *sc;
1305 	struct port_info *pi;
1306 	int i, rc;
1307 
1308 	sc = device_get_softc(dev);
1309 
1310 	sc->flags &= ~CHK_MBOX_ACCESS;
1311 	if (sc->flags & FULL_INIT_DONE) {
1312 		if (!(sc->flags & IS_VF))
1313 			t4_intr_disable(sc);
1314 	}
1315 
1316 	if (sc->cdev) {
1317 		destroy_dev(sc->cdev);
1318 		sc->cdev = NULL;
1319 	}
1320 
1321 	if (device_is_attached(dev)) {
1322 		rc = bus_generic_detach(dev);
1323 		if (rc) {
1324 			device_printf(dev,
1325 			    "failed to detach child devices: %d\n", rc);
1326 			return (rc);
1327 		}
1328 	}
1329 
1330 	for (i = 0; i < sc->intr_count; i++)
1331 		t4_free_irq(sc, &sc->irq[i]);
1332 
1333 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1334 		t4_free_tx_sched(sc);
1335 
1336 	for (i = 0; i < MAX_NPORTS; i++) {
1337 		pi = sc->port[i];
1338 		if (pi) {
1339 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1340 			if (pi->dev)
1341 				device_delete_child(dev, pi->dev);
1342 
1343 			mtx_destroy(&pi->pi_lock);
1344 			free(pi->vi, M_CXGBE);
1345 			free(pi, M_CXGBE);
1346 		}
1347 	}
1348 
1349 	device_delete_children(dev);
1350 
1351 	if (sc->flags & FULL_INIT_DONE)
1352 		adapter_full_uninit(sc);
1353 
1354 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1355 		t4_fw_bye(sc, sc->mbox);
1356 
1357 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1358 		pci_release_msi(dev);
1359 
1360 	if (sc->regs_res)
1361 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1362 		    sc->regs_res);
1363 
1364 	if (sc->udbs_res)
1365 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1366 		    sc->udbs_res);
1367 
1368 	if (sc->msix_res)
1369 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1370 		    sc->msix_res);
1371 
1372 	if (sc->l2t)
1373 		t4_free_l2t(sc->l2t);
1374 
1375 #ifdef TCP_OFFLOAD
1376 	free(sc->sge.ofld_rxq, M_CXGBE);
1377 	free(sc->sge.ofld_txq, M_CXGBE);
1378 #endif
1379 #ifdef DEV_NETMAP
1380 	free(sc->sge.nm_rxq, M_CXGBE);
1381 	free(sc->sge.nm_txq, M_CXGBE);
1382 #endif
1383 	free(sc->irq, M_CXGBE);
1384 	free(sc->sge.rxq, M_CXGBE);
1385 	free(sc->sge.txq, M_CXGBE);
1386 	free(sc->sge.ctrlq, M_CXGBE);
1387 	free(sc->sge.iqmap, M_CXGBE);
1388 	free(sc->sge.eqmap, M_CXGBE);
1389 	free(sc->tids.ftid_tab, M_CXGBE);
1390 	t4_destroy_dma_tag(sc);
1391 	if (mtx_initialized(&sc->sc_lock)) {
1392 		sx_xlock(&t4_list_lock);
1393 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1394 		sx_xunlock(&t4_list_lock);
1395 		mtx_destroy(&sc->sc_lock);
1396 	}
1397 
1398 	callout_drain(&sc->sfl_callout);
1399 	if (mtx_initialized(&sc->tids.ftid_lock))
1400 		mtx_destroy(&sc->tids.ftid_lock);
1401 	if (mtx_initialized(&sc->sfl_lock))
1402 		mtx_destroy(&sc->sfl_lock);
1403 	if (mtx_initialized(&sc->ifp_lock))
1404 		mtx_destroy(&sc->ifp_lock);
1405 	if (mtx_initialized(&sc->reg_lock))
1406 		mtx_destroy(&sc->reg_lock);
1407 
1408 	for (i = 0; i < NUM_MEMWIN; i++) {
1409 		struct memwin *mw = &sc->memwin[i];
1410 
1411 		if (rw_initialized(&mw->mw_lock))
1412 			rw_destroy(&mw->mw_lock);
1413 	}
1414 
1415 	bzero(sc, sizeof(*sc));
1416 
1417 	return (0);
1418 }
1419 
1420 static int
1421 cxgbe_probe(device_t dev)
1422 {
1423 	char buf[128];
1424 	struct port_info *pi = device_get_softc(dev);
1425 
1426 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1427 	device_set_desc_copy(dev, buf);
1428 
1429 	return (BUS_PROBE_DEFAULT);
1430 }
1431 
1432 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1433     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1434     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1435 #define T4_CAP_ENABLE (T4_CAP)
1436 
1437 static int
1438 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1439 {
1440 	struct ifnet *ifp;
1441 	struct sbuf *sb;
1442 
1443 	vi->xact_addr_filt = -1;
1444 	callout_init(&vi->tick, 1);
1445 
1446 	/* Allocate an ifnet and set it up */
1447 	ifp = if_alloc(IFT_ETHER);
1448 	if (ifp == NULL) {
1449 		device_printf(dev, "Cannot allocate ifnet\n");
1450 		return (ENOMEM);
1451 	}
1452 	vi->ifp = ifp;
1453 	ifp->if_softc = vi;
1454 
1455 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1456 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1457 
1458 	ifp->if_init = cxgbe_init;
1459 	ifp->if_ioctl = cxgbe_ioctl;
1460 	ifp->if_transmit = cxgbe_transmit;
1461 	ifp->if_qflush = cxgbe_qflush;
1462 	ifp->if_get_counter = cxgbe_get_counter;
1463 
1464 	ifp->if_capabilities = T4_CAP;
1465 #ifdef TCP_OFFLOAD
1466 	if (vi->nofldrxq != 0)
1467 		ifp->if_capabilities |= IFCAP_TOE;
1468 #endif
1469 #ifdef DEV_NETMAP
1470 	if (vi->nnmrxq != 0)
1471 		ifp->if_capabilities |= IFCAP_NETMAP;
1472 #endif
1473 	ifp->if_capenable = T4_CAP_ENABLE;
1474 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1475 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1476 
1477 	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1478 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1479 	ifp->if_hw_tsomaxsegsize = 65536;
1480 
1481 	vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1482 	    EVENTHANDLER_PRI_ANY);
1483 
1484 	ether_ifattach(ifp, vi->hw_addr);
1485 #ifdef DEV_NETMAP
1486 	if (ifp->if_capabilities & IFCAP_NETMAP)
1487 		cxgbe_nm_attach(vi);
1488 #endif
1489 	sb = sbuf_new_auto();
1490 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1491 #ifdef TCP_OFFLOAD
1492 	if (ifp->if_capabilities & IFCAP_TOE)
1493 		sbuf_printf(sb, "; %d txq, %d rxq (TOE)",
1494 		    vi->nofldtxq, vi->nofldrxq);
1495 #endif
1496 #ifdef DEV_NETMAP
1497 	if (ifp->if_capabilities & IFCAP_NETMAP)
1498 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1499 		    vi->nnmtxq, vi->nnmrxq);
1500 #endif
1501 	sbuf_finish(sb);
1502 	device_printf(dev, "%s\n", sbuf_data(sb));
1503 	sbuf_delete(sb);
1504 
1505 	vi_sysctls(vi);
1506 
1507 	return (0);
1508 }
1509 
1510 static int
1511 cxgbe_attach(device_t dev)
1512 {
1513 	struct port_info *pi = device_get_softc(dev);
1514 	struct adapter *sc = pi->adapter;
1515 	struct vi_info *vi;
1516 	int i, rc;
1517 
1518 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1519 
1520 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1521 	if (rc)
1522 		return (rc);
1523 
1524 	for_each_vi(pi, i, vi) {
1525 		if (i == 0)
1526 			continue;
1527 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1528 		if (vi->dev == NULL) {
1529 			device_printf(dev, "failed to add VI %d\n", i);
1530 			continue;
1531 		}
1532 		device_set_softc(vi->dev, vi);
1533 	}
1534 
1535 	cxgbe_sysctls(pi);
1536 
1537 	bus_generic_attach(dev);
1538 
1539 	return (0);
1540 }
1541 
1542 static void
1543 cxgbe_vi_detach(struct vi_info *vi)
1544 {
1545 	struct ifnet *ifp = vi->ifp;
1546 
1547 	ether_ifdetach(ifp);
1548 
1549 	if (vi->vlan_c)
1550 		EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1551 
1552 	/* Let detach proceed even if these fail. */
1553 #ifdef DEV_NETMAP
1554 	if (ifp->if_capabilities & IFCAP_NETMAP)
1555 		cxgbe_nm_detach(vi);
1556 #endif
1557 	cxgbe_uninit_synchronized(vi);
1558 	callout_drain(&vi->tick);
1559 	vi_full_uninit(vi);
1560 
1561 	if_free(vi->ifp);
1562 	vi->ifp = NULL;
1563 }
1564 
1565 static int
1566 cxgbe_detach(device_t dev)
1567 {
1568 	struct port_info *pi = device_get_softc(dev);
1569 	struct adapter *sc = pi->adapter;
1570 	int rc;
1571 
1572 	/* Detach the extra VIs first. */
1573 	rc = bus_generic_detach(dev);
1574 	if (rc)
1575 		return (rc);
1576 	device_delete_children(dev);
1577 
1578 	doom_vi(sc, &pi->vi[0]);
1579 
1580 	if (pi->flags & HAS_TRACEQ) {
1581 		sc->traceq = -1;	/* cloner should not create ifnet */
1582 		t4_tracer_port_detach(sc);
1583 	}
1584 
1585 	cxgbe_vi_detach(&pi->vi[0]);
1586 	callout_drain(&pi->tick);
1587 	ifmedia_removeall(&pi->media);
1588 
1589 	end_synchronized_op(sc, 0);
1590 
1591 	return (0);
1592 }
1593 
1594 static void
1595 cxgbe_init(void *arg)
1596 {
1597 	struct vi_info *vi = arg;
1598 	struct adapter *sc = vi->pi->adapter;
1599 
1600 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1601 		return;
1602 	cxgbe_init_synchronized(vi);
1603 	end_synchronized_op(sc, 0);
1604 }
1605 
1606 static int
1607 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1608 {
1609 	int rc = 0, mtu, flags, can_sleep;
1610 	struct vi_info *vi = ifp->if_softc;
1611 	struct port_info *pi = vi->pi;
1612 	struct adapter *sc = pi->adapter;
1613 	struct ifreq *ifr = (struct ifreq *)data;
1614 	uint32_t mask;
1615 
1616 	switch (cmd) {
1617 	case SIOCSIFMTU:
1618 		mtu = ifr->ifr_mtu;
1619 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1620 			return (EINVAL);
1621 
1622 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1623 		if (rc)
1624 			return (rc);
1625 		ifp->if_mtu = mtu;
1626 		if (vi->flags & VI_INIT_DONE) {
1627 			t4_update_fl_bufsize(ifp);
1628 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1629 				rc = update_mac_settings(ifp, XGMAC_MTU);
1630 		}
1631 		end_synchronized_op(sc, 0);
1632 		break;
1633 
1634 	case SIOCSIFFLAGS:
1635 		can_sleep = 0;
1636 redo_sifflags:
1637 		rc = begin_synchronized_op(sc, vi,
1638 		    can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1639 		if (rc)
1640 			return (rc);
1641 
1642 		if (ifp->if_flags & IFF_UP) {
1643 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1644 				flags = vi->if_flags;
1645 				if ((ifp->if_flags ^ flags) &
1646 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1647 					if (can_sleep == 1) {
1648 						end_synchronized_op(sc, 0);
1649 						can_sleep = 0;
1650 						goto redo_sifflags;
1651 					}
1652 					rc = update_mac_settings(ifp,
1653 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1654 				}
1655 			} else {
1656 				if (can_sleep == 0) {
1657 					end_synchronized_op(sc, LOCK_HELD);
1658 					can_sleep = 1;
1659 					goto redo_sifflags;
1660 				}
1661 				rc = cxgbe_init_synchronized(vi);
1662 			}
1663 			vi->if_flags = ifp->if_flags;
1664 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1665 			if (can_sleep == 0) {
1666 				end_synchronized_op(sc, LOCK_HELD);
1667 				can_sleep = 1;
1668 				goto redo_sifflags;
1669 			}
1670 			rc = cxgbe_uninit_synchronized(vi);
1671 		}
1672 		end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1673 		break;
1674 
1675 	case SIOCADDMULTI:
1676 	case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1677 		rc = begin_synchronized_op(sc, vi, HOLD_LOCK, "t4multi");
1678 		if (rc)
1679 			return (rc);
1680 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1681 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1682 		end_synchronized_op(sc, LOCK_HELD);
1683 		break;
1684 
1685 	case SIOCSIFCAP:
1686 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1687 		if (rc)
1688 			return (rc);
1689 
1690 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1691 		if (mask & IFCAP_TXCSUM) {
1692 			ifp->if_capenable ^= IFCAP_TXCSUM;
1693 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1694 
1695 			if (IFCAP_TSO4 & ifp->if_capenable &&
1696 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1697 				ifp->if_capenable &= ~IFCAP_TSO4;
1698 				if_printf(ifp,
1699 				    "tso4 disabled due to -txcsum.\n");
1700 			}
1701 		}
1702 		if (mask & IFCAP_TXCSUM_IPV6) {
1703 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1704 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1705 
1706 			if (IFCAP_TSO6 & ifp->if_capenable &&
1707 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1708 				ifp->if_capenable &= ~IFCAP_TSO6;
1709 				if_printf(ifp,
1710 				    "tso6 disabled due to -txcsum6.\n");
1711 			}
1712 		}
1713 		if (mask & IFCAP_RXCSUM)
1714 			ifp->if_capenable ^= IFCAP_RXCSUM;
1715 		if (mask & IFCAP_RXCSUM_IPV6)
1716 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1717 
1718 		/*
1719 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1720 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1721 		 * sending a TSO request our way, so it's sufficient to toggle
1722 		 * IFCAP_TSOx only.
1723 		 */
1724 		if (mask & IFCAP_TSO4) {
1725 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1726 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1727 				if_printf(ifp, "enable txcsum first.\n");
1728 				rc = EAGAIN;
1729 				goto fail;
1730 			}
1731 			ifp->if_capenable ^= IFCAP_TSO4;
1732 		}
1733 		if (mask & IFCAP_TSO6) {
1734 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1735 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1736 				if_printf(ifp, "enable txcsum6 first.\n");
1737 				rc = EAGAIN;
1738 				goto fail;
1739 			}
1740 			ifp->if_capenable ^= IFCAP_TSO6;
1741 		}
1742 		if (mask & IFCAP_LRO) {
1743 #if defined(INET) || defined(INET6)
1744 			int i;
1745 			struct sge_rxq *rxq;
1746 
1747 			ifp->if_capenable ^= IFCAP_LRO;
1748 			for_each_rxq(vi, i, rxq) {
1749 				if (ifp->if_capenable & IFCAP_LRO)
1750 					rxq->iq.flags |= IQ_LRO_ENABLED;
1751 				else
1752 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1753 			}
1754 #endif
1755 		}
1756 #ifdef TCP_OFFLOAD
1757 		if (mask & IFCAP_TOE) {
1758 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1759 
1760 			rc = toe_capability(vi, enable);
1761 			if (rc != 0)
1762 				goto fail;
1763 
1764 			ifp->if_capenable ^= mask;
1765 		}
1766 #endif
1767 		if (mask & IFCAP_VLAN_HWTAGGING) {
1768 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1769 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1770 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1771 		}
1772 		if (mask & IFCAP_VLAN_MTU) {
1773 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1774 
1775 			/* Need to find out how to disable auto-mtu-inflation */
1776 		}
1777 		if (mask & IFCAP_VLAN_HWTSO)
1778 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1779 		if (mask & IFCAP_VLAN_HWCSUM)
1780 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1781 
1782 #ifdef VLAN_CAPABILITIES
1783 		VLAN_CAPABILITIES(ifp);
1784 #endif
1785 fail:
1786 		end_synchronized_op(sc, 0);
1787 		break;
1788 
1789 	case SIOCSIFMEDIA:
1790 	case SIOCGIFMEDIA:
1791 	case SIOCGIFXMEDIA:
1792 		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1793 		break;
1794 
1795 	case SIOCGI2C: {
1796 		struct ifi2creq i2c;
1797 
1798 		rc = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
1799 		if (rc != 0)
1800 			break;
1801 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1802 			rc = EPERM;
1803 			break;
1804 		}
1805 		if (i2c.len > sizeof(i2c.data)) {
1806 			rc = EINVAL;
1807 			break;
1808 		}
1809 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1810 		if (rc)
1811 			return (rc);
1812 		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
1813 		    i2c.offset, i2c.len, &i2c.data[0]);
1814 		end_synchronized_op(sc, 0);
1815 		if (rc == 0)
1816 			rc = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
1817 		break;
1818 	}
1819 
1820 	default:
1821 		rc = ether_ioctl(ifp, cmd, data);
1822 	}
1823 
1824 	return (rc);
1825 }
1826 
1827 static int
1828 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1829 {
1830 	struct vi_info *vi = ifp->if_softc;
1831 	struct port_info *pi = vi->pi;
1832 	struct adapter *sc = pi->adapter;
1833 	struct sge_txq *txq;
1834 	void *items[1];
1835 	int rc;
1836 
1837 	M_ASSERTPKTHDR(m);
1838 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1839 
1840 	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1841 		m_freem(m);
1842 		return (ENETDOWN);
1843 	}
1844 
1845 	rc = parse_pkt(sc, &m);
1846 	if (__predict_false(rc != 0)) {
1847 		MPASS(m == NULL);			/* was freed already */
1848 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1849 		return (rc);
1850 	}
1851 
1852 	/* Select a txq. */
1853 	txq = &sc->sge.txq[vi->first_txq];
1854 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1855 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1856 		    vi->rsrv_noflowq);
1857 
1858 	items[0] = m;
1859 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1860 	if (__predict_false(rc != 0))
1861 		m_freem(m);
1862 
1863 	return (rc);
1864 }
1865 
1866 static void
1867 cxgbe_qflush(struct ifnet *ifp)
1868 {
1869 	struct vi_info *vi = ifp->if_softc;
1870 	struct sge_txq *txq;
1871 	int i;
1872 
1873 	/* queues do not exist if !VI_INIT_DONE. */
1874 	if (vi->flags & VI_INIT_DONE) {
1875 		for_each_txq(vi, i, txq) {
1876 			TXQ_LOCK(txq);
1877 			txq->eq.flags |= EQ_QFLUSH;
1878 			TXQ_UNLOCK(txq);
1879 			while (!mp_ring_is_idle(txq->r)) {
1880 				mp_ring_check_drainage(txq->r, 0);
1881 				pause("qflush", 1);
1882 			}
1883 			TXQ_LOCK(txq);
1884 			txq->eq.flags &= ~EQ_QFLUSH;
1885 			TXQ_UNLOCK(txq);
1886 		}
1887 	}
1888 	if_qflush(ifp);
1889 }
1890 
1891 static uint64_t
1892 vi_get_counter(struct ifnet *ifp, ift_counter c)
1893 {
1894 	struct vi_info *vi = ifp->if_softc;
1895 	struct fw_vi_stats_vf *s = &vi->stats;
1896 
1897 	vi_refresh_stats(vi->pi->adapter, vi);
1898 
1899 	switch (c) {
1900 	case IFCOUNTER_IPACKETS:
1901 		return (s->rx_bcast_frames + s->rx_mcast_frames +
1902 		    s->rx_ucast_frames);
1903 	case IFCOUNTER_IERRORS:
1904 		return (s->rx_err_frames);
1905 	case IFCOUNTER_OPACKETS:
1906 		return (s->tx_bcast_frames + s->tx_mcast_frames +
1907 		    s->tx_ucast_frames + s->tx_offload_frames);
1908 	case IFCOUNTER_OERRORS:
1909 		return (s->tx_drop_frames);
1910 	case IFCOUNTER_IBYTES:
1911 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
1912 		    s->rx_ucast_bytes);
1913 	case IFCOUNTER_OBYTES:
1914 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
1915 		    s->tx_ucast_bytes + s->tx_offload_bytes);
1916 	case IFCOUNTER_IMCASTS:
1917 		return (s->rx_mcast_frames);
1918 	case IFCOUNTER_OMCASTS:
1919 		return (s->tx_mcast_frames);
1920 	case IFCOUNTER_OQDROPS: {
1921 		uint64_t drops;
1922 
1923 		drops = 0;
1924 		if (vi->flags & VI_INIT_DONE) {
1925 			int i;
1926 			struct sge_txq *txq;
1927 
1928 			for_each_txq(vi, i, txq)
1929 				drops += counter_u64_fetch(txq->r->drops);
1930 		}
1931 
1932 		return (drops);
1933 
1934 	}
1935 
1936 	default:
1937 		return (if_get_counter_default(ifp, c));
1938 	}
1939 }
1940 
1941 uint64_t
1942 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
1943 {
1944 	struct vi_info *vi = ifp->if_softc;
1945 	struct port_info *pi = vi->pi;
1946 	struct adapter *sc = pi->adapter;
1947 	struct port_stats *s = &pi->stats;
1948 
1949 	if (pi->nvi > 1 || sc->flags & IS_VF)
1950 		return (vi_get_counter(ifp, c));
1951 
1952 	cxgbe_refresh_stats(sc, pi);
1953 
1954 	switch (c) {
1955 	case IFCOUNTER_IPACKETS:
1956 		return (s->rx_frames);
1957 
1958 	case IFCOUNTER_IERRORS:
1959 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
1960 		    s->rx_fcs_err + s->rx_len_err);
1961 
1962 	case IFCOUNTER_OPACKETS:
1963 		return (s->tx_frames);
1964 
1965 	case IFCOUNTER_OERRORS:
1966 		return (s->tx_error_frames);
1967 
1968 	case IFCOUNTER_IBYTES:
1969 		return (s->rx_octets);
1970 
1971 	case IFCOUNTER_OBYTES:
1972 		return (s->tx_octets);
1973 
1974 	case IFCOUNTER_IMCASTS:
1975 		return (s->rx_mcast_frames);
1976 
1977 	case IFCOUNTER_OMCASTS:
1978 		return (s->tx_mcast_frames);
1979 
1980 	case IFCOUNTER_IQDROPS:
1981 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
1982 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
1983 		    s->rx_trunc3 + pi->tnl_cong_drops);
1984 
1985 	case IFCOUNTER_OQDROPS: {
1986 		uint64_t drops;
1987 
1988 		drops = s->tx_drop;
1989 		if (vi->flags & VI_INIT_DONE) {
1990 			int i;
1991 			struct sge_txq *txq;
1992 
1993 			for_each_txq(vi, i, txq)
1994 				drops += counter_u64_fetch(txq->r->drops);
1995 		}
1996 
1997 		return (drops);
1998 
1999 	}
2000 
2001 	default:
2002 		return (if_get_counter_default(ifp, c));
2003 	}
2004 }
2005 
2006 static int
2007 cxgbe_media_change(struct ifnet *ifp)
2008 {
2009 	struct vi_info *vi = ifp->if_softc;
2010 
2011 	device_printf(vi->dev, "%s unimplemented.\n", __func__);
2012 
2013 	return (EOPNOTSUPP);
2014 }
2015 
2016 static void
2017 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2018 {
2019 	struct vi_info *vi = ifp->if_softc;
2020 	struct port_info *pi = vi->pi;
2021 	struct ifmedia_entry *cur;
2022 	struct link_config *lc = &pi->link_cfg;
2023 
2024 	/*
2025 	 * If all the interfaces are administratively down the firmware does not
2026 	 * report transceiver changes.  Refresh port info here so that ifconfig
2027 	 * displays accurate information at all times.
2028 	 */
2029 	if (begin_synchronized_op(pi->adapter, NULL, SLEEP_OK | INTR_OK,
2030 	    "t4med") == 0) {
2031 		PORT_LOCK(pi);
2032 		if (pi->up_vis == 0) {
2033 			t4_update_port_info(pi);
2034 			build_medialist(pi, &pi->media);
2035 		}
2036 		PORT_UNLOCK(pi);
2037 		end_synchronized_op(pi->adapter, 0);
2038 	}
2039 
2040 	ifmr->ifm_status = IFM_AVALID;
2041 	if (lc->link_ok == 0)
2042 		return;
2043 
2044 	ifmr->ifm_status |= IFM_ACTIVE;
2045 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2046 	if (lc->fc & PAUSE_RX)
2047 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2048 	if (lc->fc & PAUSE_TX)
2049 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2050 
2051 	/* active and current will differ iff current media is autoselect. */
2052 	cur = pi->media.ifm_cur;
2053 	if (cur != NULL && IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
2054 		return;
2055 
2056 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2057 	if (lc->fc & PAUSE_RX)
2058 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2059 	if (lc->fc & PAUSE_TX)
2060 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2061 	switch (lc->speed) {
2062 	case 10000:
2063 		ifmr->ifm_active |= IFM_10G_T;
2064 		break;
2065 	case 1000:
2066 		ifmr->ifm_active |= IFM_1000_T;
2067 		break;
2068 	case 100:
2069 		ifmr->ifm_active |= IFM_100_TX;
2070 		break;
2071 	case 10:
2072 		ifmr->ifm_active |= IFM_10_T;
2073 		break;
2074 	default:
2075 		device_printf(vi->dev, "link up but speed unknown (%u)\n",
2076 		    lc->speed);
2077 	}
2078 }
2079 
2080 static int
2081 vcxgbe_probe(device_t dev)
2082 {
2083 	char buf[128];
2084 	struct vi_info *vi = device_get_softc(dev);
2085 
2086 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2087 	    vi - vi->pi->vi);
2088 	device_set_desc_copy(dev, buf);
2089 
2090 	return (BUS_PROBE_DEFAULT);
2091 }
2092 
2093 static int
2094 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2095 {
2096 	int func, index, rc;
2097 	uint32_t param, val;
2098 
2099 	ASSERT_SYNCHRONIZED_OP(sc);
2100 
2101 	index = vi - pi->vi;
2102 	MPASS(index > 0);	/* This function deals with _extra_ VIs only */
2103 	KASSERT(index < nitems(vi_mac_funcs),
2104 	    ("%s: VI %s doesn't have a MAC func", __func__,
2105 	    device_get_nameunit(vi->dev)));
2106 	func = vi_mac_funcs[index];
2107 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2108 	    vi->hw_addr, &vi->rss_size, func, 0);
2109 	if (rc < 0) {
2110 		device_printf(vi->dev, "failed to allocate virtual interface %d"
2111 		    "for port %d: %d\n", index, pi->port_id, -rc);
2112 		return (-rc);
2113 	}
2114 	vi->viid = rc;
2115 	if (chip_id(sc) <= CHELSIO_T5)
2116 		vi->smt_idx = (rc & 0x7f) << 1;
2117 	else
2118 		vi->smt_idx = (rc & 0x7f);
2119 
2120 	if (vi->rss_size == 1) {
2121 		/*
2122 		 * This VI didn't get a slice of the RSS table.  Reduce the
2123 		 * number of VIs being created (hw.cxgbe.num_vis) or modify the
2124 		 * configuration file (nvi, rssnvi for this PF) if this is a
2125 		 * problem.
2126 		 */
2127 		device_printf(vi->dev, "RSS table not available.\n");
2128 		vi->rss_base = 0xffff;
2129 
2130 		return (0);
2131 	}
2132 
2133 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2134 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2135 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2136 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2137 	if (rc)
2138 		vi->rss_base = 0xffff;
2139 	else {
2140 		MPASS((val >> 16) == vi->rss_size);
2141 		vi->rss_base = val & 0xffff;
2142 	}
2143 
2144 	return (0);
2145 }
2146 
2147 static int
2148 vcxgbe_attach(device_t dev)
2149 {
2150 	struct vi_info *vi;
2151 	struct port_info *pi;
2152 	struct adapter *sc;
2153 	int rc;
2154 
2155 	vi = device_get_softc(dev);
2156 	pi = vi->pi;
2157 	sc = pi->adapter;
2158 
2159 	rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2160 	if (rc)
2161 		return (rc);
2162 	rc = alloc_extra_vi(sc, pi, vi);
2163 	end_synchronized_op(sc, 0);
2164 	if (rc)
2165 		return (rc);
2166 
2167 	rc = cxgbe_vi_attach(dev, vi);
2168 	if (rc) {
2169 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2170 		return (rc);
2171 	}
2172 	return (0);
2173 }
2174 
2175 static int
2176 vcxgbe_detach(device_t dev)
2177 {
2178 	struct vi_info *vi;
2179 	struct adapter *sc;
2180 
2181 	vi = device_get_softc(dev);
2182 	sc = vi->pi->adapter;
2183 
2184 	doom_vi(sc, vi);
2185 
2186 	cxgbe_vi_detach(vi);
2187 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2188 
2189 	end_synchronized_op(sc, 0);
2190 
2191 	return (0);
2192 }
2193 
2194 void
2195 t4_fatal_err(struct adapter *sc)
2196 {
2197 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2198 	t4_intr_disable(sc);
2199 	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2200 	    device_get_nameunit(sc->dev));
2201 }
2202 
2203 void
2204 t4_add_adapter(struct adapter *sc)
2205 {
2206 	sx_xlock(&t4_list_lock);
2207 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2208 	sx_xunlock(&t4_list_lock);
2209 }
2210 
2211 int
2212 t4_map_bars_0_and_4(struct adapter *sc)
2213 {
2214 	sc->regs_rid = PCIR_BAR(0);
2215 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2216 	    &sc->regs_rid, RF_ACTIVE);
2217 	if (sc->regs_res == NULL) {
2218 		device_printf(sc->dev, "cannot map registers.\n");
2219 		return (ENXIO);
2220 	}
2221 	sc->bt = rman_get_bustag(sc->regs_res);
2222 	sc->bh = rman_get_bushandle(sc->regs_res);
2223 	sc->mmio_len = rman_get_size(sc->regs_res);
2224 	setbit(&sc->doorbells, DOORBELL_KDB);
2225 
2226 	sc->msix_rid = PCIR_BAR(4);
2227 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2228 	    &sc->msix_rid, RF_ACTIVE);
2229 	if (sc->msix_res == NULL) {
2230 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2231 		return (ENXIO);
2232 	}
2233 
2234 	return (0);
2235 }
2236 
2237 int
2238 t4_map_bar_2(struct adapter *sc)
2239 {
2240 
2241 	/*
2242 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2243 	 * to map it if RDMA is disabled.
2244 	 */
2245 	if (is_t4(sc) && sc->rdmacaps == 0)
2246 		return (0);
2247 
2248 	sc->udbs_rid = PCIR_BAR(2);
2249 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2250 	    &sc->udbs_rid, RF_ACTIVE);
2251 	if (sc->udbs_res == NULL) {
2252 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2253 		return (ENXIO);
2254 	}
2255 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2256 
2257 	if (chip_id(sc) >= CHELSIO_T5) {
2258 		setbit(&sc->doorbells, DOORBELL_UDB);
2259 #if defined(__i386__) || defined(__amd64__)
2260 		if (t5_write_combine) {
2261 			int rc, mode;
2262 
2263 			/*
2264 			 * Enable write combining on BAR2.  This is the
2265 			 * userspace doorbell BAR and is split into 128B
2266 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2267 			 * with an egress queue.  The first 64B has the doorbell
2268 			 * and the second 64B can be used to submit a tx work
2269 			 * request with an implicit doorbell.
2270 			 */
2271 
2272 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2273 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2274 			if (rc == 0) {
2275 				clrbit(&sc->doorbells, DOORBELL_UDB);
2276 				setbit(&sc->doorbells, DOORBELL_WCWR);
2277 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2278 			} else {
2279 				t5_write_combine = 0;
2280 				device_printf(sc->dev,
2281 				    "couldn't enable write combining: %d\n",
2282 				    rc);
2283 			}
2284 
2285 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2286 			t4_write_reg(sc, A_SGE_STAT_CFG,
2287 			    V_STATSOURCE_T5(7) | mode);
2288 		}
2289 #else
2290 		t5_write_combine = 0;
2291 #endif
2292 		sc->iwt.wc_en = t5_write_combine;
2293 	}
2294 
2295 	return (0);
2296 }
2297 
2298 struct memwin_init {
2299 	uint32_t base;
2300 	uint32_t aperture;
2301 };
2302 
2303 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2304 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2305 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2306 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2307 };
2308 
2309 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2310 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2311 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2312 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2313 };
2314 
2315 static void
2316 setup_memwin(struct adapter *sc)
2317 {
2318 	const struct memwin_init *mw_init;
2319 	struct memwin *mw;
2320 	int i;
2321 	uint32_t bar0;
2322 
2323 	if (is_t4(sc)) {
2324 		/*
2325 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2326 		 * mechanism.  Works from within PCI passthrough environments
2327 		 * too, where rman_get_start() can return a different value.  We
2328 		 * need to program the T4 memory window decoders with the actual
2329 		 * addresses that will be coming across the PCIe link.
2330 		 */
2331 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2332 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2333 
2334 		mw_init = &t4_memwin[0];
2335 	} else {
2336 		/* T5+ use the relative offset inside the PCIe BAR */
2337 		bar0 = 0;
2338 
2339 		mw_init = &t5_memwin[0];
2340 	}
2341 
2342 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2343 		rw_init(&mw->mw_lock, "memory window access");
2344 		mw->mw_base = mw_init->base;
2345 		mw->mw_aperture = mw_init->aperture;
2346 		mw->mw_curpos = 0;
2347 		t4_write_reg(sc,
2348 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2349 		    (mw->mw_base + bar0) | V_BIR(0) |
2350 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2351 		rw_wlock(&mw->mw_lock);
2352 		position_memwin(sc, i, 0);
2353 		rw_wunlock(&mw->mw_lock);
2354 	}
2355 
2356 	/* flush */
2357 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2358 }
2359 
2360 /*
2361  * Positions the memory window at the given address in the card's address space.
2362  * There are some alignment requirements and the actual position may be at an
2363  * address prior to the requested address.  mw->mw_curpos always has the actual
2364  * position of the window.
2365  */
2366 static void
2367 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2368 {
2369 	struct memwin *mw;
2370 	uint32_t pf;
2371 	uint32_t reg;
2372 
2373 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2374 	mw = &sc->memwin[idx];
2375 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2376 
2377 	if (is_t4(sc)) {
2378 		pf = 0;
2379 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2380 	} else {
2381 		pf = V_PFNUM(sc->pf);
2382 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2383 	}
2384 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2385 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2386 	t4_read_reg(sc, reg);	/* flush */
2387 }
2388 
2389 static int
2390 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2391     int len, int rw)
2392 {
2393 	struct memwin *mw;
2394 	uint32_t mw_end, v;
2395 
2396 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2397 
2398 	/* Memory can only be accessed in naturally aligned 4 byte units */
2399 	if (addr & 3 || len & 3 || len <= 0)
2400 		return (EINVAL);
2401 
2402 	mw = &sc->memwin[idx];
2403 	while (len > 0) {
2404 		rw_rlock(&mw->mw_lock);
2405 		mw_end = mw->mw_curpos + mw->mw_aperture;
2406 		if (addr >= mw_end || addr < mw->mw_curpos) {
2407 			/* Will need to reposition the window */
2408 			if (!rw_try_upgrade(&mw->mw_lock)) {
2409 				rw_runlock(&mw->mw_lock);
2410 				rw_wlock(&mw->mw_lock);
2411 			}
2412 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2413 			position_memwin(sc, idx, addr);
2414 			rw_downgrade(&mw->mw_lock);
2415 			mw_end = mw->mw_curpos + mw->mw_aperture;
2416 		}
2417 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2418 		while (addr < mw_end && len > 0) {
2419 			if (rw == 0) {
2420 				v = t4_read_reg(sc, mw->mw_base + addr -
2421 				    mw->mw_curpos);
2422 				*val++ = le32toh(v);
2423 			} else {
2424 				v = *val++;
2425 				t4_write_reg(sc, mw->mw_base + addr -
2426 				    mw->mw_curpos, htole32(v));
2427 			}
2428 			addr += 4;
2429 			len -= 4;
2430 		}
2431 		rw_runlock(&mw->mw_lock);
2432 	}
2433 
2434 	return (0);
2435 }
2436 
2437 static inline int
2438 read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2439     int len)
2440 {
2441 
2442 	return (rw_via_memwin(sc, idx, addr, val, len, 0));
2443 }
2444 
2445 static inline int
2446 write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
2447     const uint32_t *val, int len)
2448 {
2449 
2450 	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
2451 }
2452 
2453 static int
2454 t4_range_cmp(const void *a, const void *b)
2455 {
2456 	return ((const struct t4_range *)a)->start -
2457 	       ((const struct t4_range *)b)->start;
2458 }
2459 
2460 /*
2461  * Verify that the memory range specified by the addr/len pair is valid within
2462  * the card's address space.
2463  */
2464 static int
2465 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2466 {
2467 	struct t4_range mem_ranges[4], *r, *next;
2468 	uint32_t em, addr_len;
2469 	int i, n, remaining;
2470 
2471 	/* Memory can only be accessed in naturally aligned 4 byte units */
2472 	if (addr & 3 || len & 3 || len <= 0)
2473 		return (EINVAL);
2474 
2475 	/* Enabled memories */
2476 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2477 
2478 	r = &mem_ranges[0];
2479 	n = 0;
2480 	bzero(r, sizeof(mem_ranges));
2481 	if (em & F_EDRAM0_ENABLE) {
2482 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2483 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2484 		if (r->size > 0) {
2485 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2486 			if (addr >= r->start &&
2487 			    addr + len <= r->start + r->size)
2488 				return (0);
2489 			r++;
2490 			n++;
2491 		}
2492 	}
2493 	if (em & F_EDRAM1_ENABLE) {
2494 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2495 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2496 		if (r->size > 0) {
2497 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2498 			if (addr >= r->start &&
2499 			    addr + len <= r->start + r->size)
2500 				return (0);
2501 			r++;
2502 			n++;
2503 		}
2504 	}
2505 	if (em & F_EXT_MEM_ENABLE) {
2506 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2507 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2508 		if (r->size > 0) {
2509 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2510 			if (addr >= r->start &&
2511 			    addr + len <= r->start + r->size)
2512 				return (0);
2513 			r++;
2514 			n++;
2515 		}
2516 	}
2517 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2518 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2519 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2520 		if (r->size > 0) {
2521 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2522 			if (addr >= r->start &&
2523 			    addr + len <= r->start + r->size)
2524 				return (0);
2525 			r++;
2526 			n++;
2527 		}
2528 	}
2529 	MPASS(n <= nitems(mem_ranges));
2530 
2531 	if (n > 1) {
2532 		/* Sort and merge the ranges. */
2533 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2534 
2535 		/* Start from index 0 and examine the next n - 1 entries. */
2536 		r = &mem_ranges[0];
2537 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2538 
2539 			MPASS(r->size > 0);	/* r is a valid entry. */
2540 			next = r + 1;
2541 			MPASS(next->size > 0);	/* and so is the next one. */
2542 
2543 			while (r->start + r->size >= next->start) {
2544 				/* Merge the next one into the current entry. */
2545 				r->size = max(r->start + r->size,
2546 				    next->start + next->size) - r->start;
2547 				n--;	/* One fewer entry in total. */
2548 				if (--remaining == 0)
2549 					goto done;	/* short circuit */
2550 				next++;
2551 			}
2552 			if (next != r + 1) {
2553 				/*
2554 				 * Some entries were merged into r and next
2555 				 * points to the first valid entry that couldn't
2556 				 * be merged.
2557 				 */
2558 				MPASS(next->size > 0);	/* must be valid */
2559 				memcpy(r + 1, next, remaining * sizeof(*r));
2560 #ifdef INVARIANTS
2561 				/*
2562 				 * This so that the foo->size assertion in the
2563 				 * next iteration of the loop do the right
2564 				 * thing for entries that were pulled up and are
2565 				 * no longer valid.
2566 				 */
2567 				MPASS(n < nitems(mem_ranges));
2568 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2569 				    sizeof(struct t4_range));
2570 #endif
2571 			}
2572 		}
2573 done:
2574 		/* Done merging the ranges. */
2575 		MPASS(n > 0);
2576 		r = &mem_ranges[0];
2577 		for (i = 0; i < n; i++, r++) {
2578 			if (addr >= r->start &&
2579 			    addr + len <= r->start + r->size)
2580 				return (0);
2581 		}
2582 	}
2583 
2584 	return (EFAULT);
2585 }
2586 
2587 static int
2588 fwmtype_to_hwmtype(int mtype)
2589 {
2590 
2591 	switch (mtype) {
2592 	case FW_MEMTYPE_EDC0:
2593 		return (MEM_EDC0);
2594 	case FW_MEMTYPE_EDC1:
2595 		return (MEM_EDC1);
2596 	case FW_MEMTYPE_EXTMEM:
2597 		return (MEM_MC0);
2598 	case FW_MEMTYPE_EXTMEM1:
2599 		return (MEM_MC1);
2600 	default:
2601 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2602 	}
2603 }
2604 
2605 /*
2606  * Verify that the memory range specified by the memtype/offset/len pair is
2607  * valid and lies entirely within the memtype specified.  The global address of
2608  * the start of the range is returned in addr.
2609  */
2610 static int
2611 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2612     uint32_t *addr)
2613 {
2614 	uint32_t em, addr_len, maddr;
2615 
2616 	/* Memory can only be accessed in naturally aligned 4 byte units */
2617 	if (off & 3 || len & 3 || len == 0)
2618 		return (EINVAL);
2619 
2620 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2621 	switch (fwmtype_to_hwmtype(mtype)) {
2622 	case MEM_EDC0:
2623 		if (!(em & F_EDRAM0_ENABLE))
2624 			return (EINVAL);
2625 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2626 		maddr = G_EDRAM0_BASE(addr_len) << 20;
2627 		break;
2628 	case MEM_EDC1:
2629 		if (!(em & F_EDRAM1_ENABLE))
2630 			return (EINVAL);
2631 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2632 		maddr = G_EDRAM1_BASE(addr_len) << 20;
2633 		break;
2634 	case MEM_MC:
2635 		if (!(em & F_EXT_MEM_ENABLE))
2636 			return (EINVAL);
2637 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2638 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2639 		break;
2640 	case MEM_MC1:
2641 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2642 			return (EINVAL);
2643 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2644 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2645 		break;
2646 	default:
2647 		return (EINVAL);
2648 	}
2649 
2650 	*addr = maddr + off;	/* global address */
2651 	return (validate_mem_range(sc, *addr, len));
2652 }
2653 
2654 static int
2655 fixup_devlog_params(struct adapter *sc)
2656 {
2657 	struct devlog_params *dparams = &sc->params.devlog;
2658 	int rc;
2659 
2660 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2661 	    dparams->size, &dparams->addr);
2662 
2663 	return (rc);
2664 }
2665 
2666 static void
2667 update_nirq(struct intrs_and_queues *iaq, int nports)
2668 {
2669 	int extra = T4_EXTRA_INTR;
2670 
2671 	iaq->nirq = extra;
2672 	iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
2673 	iaq->nirq += nports * (iaq->num_vis - 1) *
2674 	    max(iaq->nrxq_vi, iaq->nnmrxq_vi);
2675 	iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
2676 }
2677 
2678 /*
2679  * Adjust requirements to fit the number of interrupts available.
2680  */
2681 static void
2682 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
2683     int navail)
2684 {
2685 	int old_nirq;
2686 	const int nports = sc->params.nports;
2687 
2688 	MPASS(nports > 0);
2689 	MPASS(navail > 0);
2690 
2691 	bzero(iaq, sizeof(*iaq));
2692 	iaq->intr_type = itype;
2693 	iaq->num_vis = t4_num_vis;
2694 	iaq->ntxq = t4_ntxq;
2695 	iaq->ntxq_vi = t4_ntxq_vi;
2696 	iaq->nrxq = t4_nrxq;
2697 	iaq->nrxq_vi = t4_nrxq_vi;
2698 #ifdef TCP_OFFLOAD
2699 	if (is_offload(sc)) {
2700 		iaq->nofldtxq = t4_nofldtxq;
2701 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
2702 		iaq->nofldrxq = t4_nofldrxq;
2703 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
2704 	}
2705 #endif
2706 #ifdef DEV_NETMAP
2707 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
2708 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
2709 #endif
2710 
2711 	update_nirq(iaq, nports);
2712 	if (iaq->nirq <= navail &&
2713 	    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2714 		/*
2715 		 * This is the normal case -- there are enough interrupts for
2716 		 * everything.
2717 		 */
2718 		goto done;
2719 	}
2720 
2721 	/*
2722 	 * If extra VIs have been configured try reducing their count and see if
2723 	 * that works.
2724 	 */
2725 	while (iaq->num_vis > 1) {
2726 		iaq->num_vis--;
2727 		update_nirq(iaq, nports);
2728 		if (iaq->nirq <= navail &&
2729 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2730 			device_printf(sc->dev, "virtual interfaces per port "
2731 			    "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
2732 			    "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
2733 			    "itype %d, navail %u, nirq %d.\n",
2734 			    iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
2735 			    iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
2736 			    itype, navail, iaq->nirq);
2737 			goto done;
2738 		}
2739 	}
2740 
2741 	/*
2742 	 * Extra VIs will not be created.  Log a message if they were requested.
2743 	 */
2744 	MPASS(iaq->num_vis == 1);
2745 	iaq->ntxq_vi = iaq->nrxq_vi = 0;
2746 	iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
2747 	iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
2748 	if (iaq->num_vis != t4_num_vis) {
2749 		device_printf(sc->dev, "extra virtual interfaces disabled.  "
2750 		    "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
2751 		    "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
2752 		    iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
2753 		    iaq->nnmrxq_vi, itype, navail, iaq->nirq);
2754 	}
2755 
2756 	/*
2757 	 * Keep reducing the number of NIC rx queues to the next lower power of
2758 	 * 2 (for even RSS distribution) and halving the TOE rx queues and see
2759 	 * if that works.
2760 	 */
2761 	do {
2762 		if (iaq->nrxq > 1) {
2763 			do {
2764 				iaq->nrxq--;
2765 			} while (!powerof2(iaq->nrxq));
2766 		}
2767 		if (iaq->nofldrxq > 1)
2768 			iaq->nofldrxq >>= 1;
2769 
2770 		old_nirq = iaq->nirq;
2771 		update_nirq(iaq, nports);
2772 		if (iaq->nirq <= navail &&
2773 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2774 			device_printf(sc->dev, "running with reduced number of "
2775 			    "rx queues because of shortage of interrupts.  "
2776 			    "nrxq=%u, nofldrxq=%u.  "
2777 			    "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
2778 			    iaq->nofldrxq, itype, navail, iaq->nirq);
2779 			goto done;
2780 		}
2781 	} while (old_nirq != iaq->nirq);
2782 
2783 	/* One interrupt for everything.  Ugh. */
2784 	device_printf(sc->dev, "running with minimal number of queues.  "
2785 	    "itype %d, navail %u.\n", itype, navail);
2786 	iaq->nirq = 1;
2787 	MPASS(iaq->nrxq == 1);
2788 	iaq->ntxq = 1;
2789 	if (iaq->nofldrxq > 1)
2790 		iaq->nofldtxq = 1;
2791 done:
2792 	MPASS(iaq->num_vis > 0);
2793 	if (iaq->num_vis > 1) {
2794 		MPASS(iaq->nrxq_vi > 0);
2795 		MPASS(iaq->ntxq_vi > 0);
2796 	}
2797 	MPASS(iaq->nirq > 0);
2798 	MPASS(iaq->nrxq > 0);
2799 	MPASS(iaq->ntxq > 0);
2800 	if (itype == INTR_MSI) {
2801 		MPASS(powerof2(iaq->nirq));
2802 	}
2803 }
2804 
2805 static int
2806 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
2807 {
2808 	int rc, itype, navail, nalloc;
2809 
2810 	for (itype = INTR_MSIX; itype; itype >>= 1) {
2811 
2812 		if ((itype & t4_intr_types) == 0)
2813 			continue;	/* not allowed */
2814 
2815 		if (itype == INTR_MSIX)
2816 			navail = pci_msix_count(sc->dev);
2817 		else if (itype == INTR_MSI)
2818 			navail = pci_msi_count(sc->dev);
2819 		else
2820 			navail = 1;
2821 restart:
2822 		if (navail == 0)
2823 			continue;
2824 
2825 		calculate_iaq(sc, iaq, itype, navail);
2826 		nalloc = iaq->nirq;
2827 		rc = 0;
2828 		if (itype == INTR_MSIX)
2829 			rc = pci_alloc_msix(sc->dev, &nalloc);
2830 		else if (itype == INTR_MSI)
2831 			rc = pci_alloc_msi(sc->dev, &nalloc);
2832 
2833 		if (rc == 0 && nalloc > 0) {
2834 			if (nalloc == iaq->nirq)
2835 				return (0);
2836 
2837 			/*
2838 			 * Didn't get the number requested.  Use whatever number
2839 			 * the kernel is willing to allocate.
2840 			 */
2841 			device_printf(sc->dev, "fewer vectors than requested, "
2842 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2843 			    itype, iaq->nirq, nalloc);
2844 			pci_release_msi(sc->dev);
2845 			navail = nalloc;
2846 			goto restart;
2847 		}
2848 
2849 		device_printf(sc->dev,
2850 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2851 		    itype, rc, iaq->nirq, nalloc);
2852 	}
2853 
2854 	device_printf(sc->dev,
2855 	    "failed to find a usable interrupt type.  "
2856 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2857 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2858 
2859 	return (ENXIO);
2860 }
2861 
2862 #define FW_VERSION(chip) ( \
2863     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2864     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2865     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2866     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2867 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2868 
2869 struct fw_info {
2870 	uint8_t chip;
2871 	char *kld_name;
2872 	char *fw_mod_name;
2873 	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
2874 } fw_info[] = {
2875 	{
2876 		.chip = CHELSIO_T4,
2877 		.kld_name = "t4fw_cfg",
2878 		.fw_mod_name = "t4fw",
2879 		.fw_hdr = {
2880 			.chip = FW_HDR_CHIP_T4,
2881 			.fw_ver = htobe32_const(FW_VERSION(T4)),
2882 			.intfver_nic = FW_INTFVER(T4, NIC),
2883 			.intfver_vnic = FW_INTFVER(T4, VNIC),
2884 			.intfver_ofld = FW_INTFVER(T4, OFLD),
2885 			.intfver_ri = FW_INTFVER(T4, RI),
2886 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2887 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
2888 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2889 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
2890 		},
2891 	}, {
2892 		.chip = CHELSIO_T5,
2893 		.kld_name = "t5fw_cfg",
2894 		.fw_mod_name = "t5fw",
2895 		.fw_hdr = {
2896 			.chip = FW_HDR_CHIP_T5,
2897 			.fw_ver = htobe32_const(FW_VERSION(T5)),
2898 			.intfver_nic = FW_INTFVER(T5, NIC),
2899 			.intfver_vnic = FW_INTFVER(T5, VNIC),
2900 			.intfver_ofld = FW_INTFVER(T5, OFLD),
2901 			.intfver_ri = FW_INTFVER(T5, RI),
2902 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
2903 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
2904 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
2905 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
2906 		},
2907 	}, {
2908 		.chip = CHELSIO_T6,
2909 		.kld_name = "t6fw_cfg",
2910 		.fw_mod_name = "t6fw",
2911 		.fw_hdr = {
2912 			.chip = FW_HDR_CHIP_T6,
2913 			.fw_ver = htobe32_const(FW_VERSION(T6)),
2914 			.intfver_nic = FW_INTFVER(T6, NIC),
2915 			.intfver_vnic = FW_INTFVER(T6, VNIC),
2916 			.intfver_ofld = FW_INTFVER(T6, OFLD),
2917 			.intfver_ri = FW_INTFVER(T6, RI),
2918 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
2919 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
2920 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
2921 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
2922 		},
2923 	}
2924 };
2925 
2926 static struct fw_info *
2927 find_fw_info(int chip)
2928 {
2929 	int i;
2930 
2931 	for (i = 0; i < nitems(fw_info); i++) {
2932 		if (fw_info[i].chip == chip)
2933 			return (&fw_info[i]);
2934 	}
2935 	return (NULL);
2936 }
2937 
2938 /*
2939  * Is the given firmware API compatible with the one the driver was compiled
2940  * with?
2941  */
2942 static int
2943 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
2944 {
2945 
2946 	/* short circuit if it's the exact same firmware version */
2947 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
2948 		return (1);
2949 
2950 	/*
2951 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
2952 	 * features that are supported in the driver.
2953 	 */
2954 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
2955 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
2956 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
2957 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
2958 		return (1);
2959 #undef SAME_INTF
2960 
2961 	return (0);
2962 }
2963 
2964 /*
2965  * The firmware in the KLD is usable, but should it be installed?  This routine
2966  * explains itself in detail if it indicates the KLD firmware should be
2967  * installed.
2968  */
2969 static int
2970 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
2971 {
2972 	const char *reason;
2973 
2974 	if (!card_fw_usable) {
2975 		reason = "incompatible or unusable";
2976 		goto install;
2977 	}
2978 
2979 	if (k > c) {
2980 		reason = "older than the version bundled with this driver";
2981 		goto install;
2982 	}
2983 
2984 	if (t4_fw_install == 2 && k != c) {
2985 		reason = "different than the version bundled with this driver";
2986 		goto install;
2987 	}
2988 
2989 	return (0);
2990 
2991 install:
2992 	if (t4_fw_install == 0) {
2993 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2994 		    "but the driver is prohibited from installing a different "
2995 		    "firmware on the card.\n",
2996 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2997 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
2998 
2999 		return (0);
3000 	}
3001 
3002 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3003 	    "installing firmware %u.%u.%u.%u on card.\n",
3004 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3005 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3006 	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3007 	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3008 
3009 	return (1);
3010 }
3011 
3012 /*
3013  * Establish contact with the firmware and determine if we are the master driver
3014  * or not, and whether we are responsible for chip initialization.
3015  */
3016 static int
3017 prep_firmware(struct adapter *sc)
3018 {
3019 	const struct firmware *fw = NULL, *default_cfg;
3020 	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
3021 	enum dev_state state;
3022 	struct fw_info *fw_info;
3023 	struct fw_hdr *card_fw;		/* fw on the card */
3024 	const struct fw_hdr *kld_fw;	/* fw in the KLD */
3025 	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
3026 					   against */
3027 
3028 	/* This is the firmware whose headers the driver was compiled against */
3029 	fw_info = find_fw_info(chip_id(sc));
3030 	if (fw_info == NULL) {
3031 		device_printf(sc->dev,
3032 		    "unable to look up firmware information for chip %d.\n",
3033 		    chip_id(sc));
3034 		return (EINVAL);
3035 	}
3036 	drv_fw = &fw_info->fw_hdr;
3037 
3038 	/*
3039 	 * The firmware KLD contains many modules.  The KLD name is also the
3040 	 * name of the module that contains the default config file.
3041 	 */
3042 	default_cfg = firmware_get(fw_info->kld_name);
3043 
3044 	/* This is the firmware in the KLD */
3045 	fw = firmware_get(fw_info->fw_mod_name);
3046 	if (fw != NULL) {
3047 		kld_fw = (const void *)fw->data;
3048 		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3049 	} else {
3050 		kld_fw = NULL;
3051 		kld_fw_usable = 0;
3052 	}
3053 
3054 	/* Read the header of the firmware on the card */
3055 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3056 	rc = -t4_read_flash(sc, FLASH_FW_START,
3057 	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
3058 	if (rc == 0) {
3059 		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
3060 		if (card_fw->fw_ver == be32toh(0xffffffff)) {
3061 			uint32_t d = be32toh(kld_fw->fw_ver);
3062 
3063 			if (!kld_fw_usable) {
3064 				device_printf(sc->dev,
3065 				    "no firmware on the card and no usable "
3066 				    "firmware bundled with the driver.\n");
3067 				rc = EIO;
3068 				goto done;
3069 			} else if (t4_fw_install == 0) {
3070 				device_printf(sc->dev,
3071 				    "no firmware on the card and the driver "
3072 				    "is prohibited from installing new "
3073 				    "firmware.\n");
3074 				rc = EIO;
3075 				goto done;
3076 			}
3077 
3078 			device_printf(sc->dev, "no firmware on the card, "
3079 			    "installing firmware %d.%d.%d.%d\n",
3080 			    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3081 			    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3082 			rc = t4_fw_forceinstall(sc, fw->data, fw->datasize);
3083 			if (rc < 0) {
3084 				rc = -rc;
3085 				device_printf(sc->dev,
3086 				    "firmware install failed: %d.\n", rc);
3087 				goto done;
3088 			}
3089 			memcpy(card_fw, kld_fw, sizeof(*card_fw));
3090 			card_fw_usable = 1;
3091 			need_fw_reset = 0;
3092 		}
3093 	} else {
3094 		device_printf(sc->dev,
3095 		    "Unable to read card's firmware header: %d\n", rc);
3096 		card_fw_usable = 0;
3097 	}
3098 
3099 	/* Contact firmware. */
3100 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3101 	if (rc < 0 || state == DEV_STATE_ERR) {
3102 		rc = -rc;
3103 		device_printf(sc->dev,
3104 		    "failed to connect to the firmware: %d, %d.\n", rc, state);
3105 		goto done;
3106 	}
3107 	pf = rc;
3108 	if (pf == sc->mbox)
3109 		sc->flags |= MASTER_PF;
3110 	else if (state == DEV_STATE_UNINIT) {
3111 		/*
3112 		 * We didn't get to be the master so we definitely won't be
3113 		 * configuring the chip.  It's a bug if someone else hasn't
3114 		 * configured it already.
3115 		 */
3116 		device_printf(sc->dev, "couldn't be master(%d), "
3117 		    "device not already initialized either(%d).\n", rc, state);
3118 		rc = EPROTO;
3119 		goto done;
3120 	}
3121 
3122 	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3123 	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3124 		/*
3125 		 * Common case: the firmware on the card is an exact match and
3126 		 * the KLD is an exact match too, or the KLD is
3127 		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
3128 		 * here -- use cxgbetool loadfw if you want to reinstall the
3129 		 * same firmware as the one on the card.
3130 		 */
3131 	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3132 	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3133 	    be32toh(card_fw->fw_ver))) {
3134 
3135 		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3136 		if (rc != 0) {
3137 			device_printf(sc->dev,
3138 			    "failed to install firmware: %d\n", rc);
3139 			goto done;
3140 		}
3141 
3142 		/* Installed successfully, update the cached header too. */
3143 		memcpy(card_fw, kld_fw, sizeof(*card_fw));
3144 		card_fw_usable = 1;
3145 		need_fw_reset = 0;	/* already reset as part of load_fw */
3146 	}
3147 
3148 	if (!card_fw_usable) {
3149 		uint32_t d, c, k;
3150 
3151 		d = ntohl(drv_fw->fw_ver);
3152 		c = ntohl(card_fw->fw_ver);
3153 		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3154 
3155 		device_printf(sc->dev, "Cannot find a usable firmware: "
3156 		    "fw_install %d, chip state %d, "
3157 		    "driver compiled with %d.%d.%d.%d, "
3158 		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3159 		    t4_fw_install, state,
3160 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3161 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3162 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3163 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3164 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3165 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3166 		rc = EINVAL;
3167 		goto done;
3168 	}
3169 
3170 	/* Reset device */
3171 	if (need_fw_reset &&
3172 	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3173 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3174 		if (rc != ETIMEDOUT && rc != EIO)
3175 			t4_fw_bye(sc, sc->mbox);
3176 		goto done;
3177 	}
3178 	sc->flags |= FW_OK;
3179 
3180 	rc = get_params__pre_init(sc);
3181 	if (rc != 0)
3182 		goto done; /* error message displayed already */
3183 
3184 	/* Partition adapter resources as specified in the config file. */
3185 	if (state == DEV_STATE_UNINIT) {
3186 
3187 		KASSERT(sc->flags & MASTER_PF,
3188 		    ("%s: trying to change chip settings when not master.",
3189 		    __func__));
3190 
3191 		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3192 		if (rc != 0)
3193 			goto done;	/* error message displayed already */
3194 
3195 		t4_tweak_chip_settings(sc);
3196 
3197 		/* get basic stuff going */
3198 		rc = -t4_fw_initialize(sc, sc->mbox);
3199 		if (rc != 0) {
3200 			device_printf(sc->dev, "fw init failed: %d.\n", rc);
3201 			goto done;
3202 		}
3203 	} else {
3204 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3205 		sc->cfcsum = 0;
3206 	}
3207 
3208 done:
3209 	free(card_fw, M_CXGBE);
3210 	if (fw != NULL)
3211 		firmware_put(fw, FIRMWARE_UNLOAD);
3212 	if (default_cfg != NULL)
3213 		firmware_put(default_cfg, FIRMWARE_UNLOAD);
3214 
3215 	return (rc);
3216 }
3217 
3218 #define FW_PARAM_DEV(param) \
3219 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3220 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3221 #define FW_PARAM_PFVF(param) \
3222 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3223 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3224 
3225 /*
3226  * Partition chip resources for use between various PFs, VFs, etc.
3227  */
3228 static int
3229 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3230     const char *name_prefix)
3231 {
3232 	const struct firmware *cfg = NULL;
3233 	int rc = 0;
3234 	struct fw_caps_config_cmd caps;
3235 	uint32_t mtype, moff, finicsum, cfcsum;
3236 
3237 	/*
3238 	 * Figure out what configuration file to use.  Pick the default config
3239 	 * file for the card if the user hasn't specified one explicitly.
3240 	 */
3241 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3242 	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3243 		/* Card specific overrides go here. */
3244 		if (pci_get_device(sc->dev) == 0x440a)
3245 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3246 		if (is_fpga(sc))
3247 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3248 	}
3249 
3250 	/*
3251 	 * We need to load another module if the profile is anything except
3252 	 * "default" or "flash".
3253 	 */
3254 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3255 	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3256 		char s[32];
3257 
3258 		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3259 		cfg = firmware_get(s);
3260 		if (cfg == NULL) {
3261 			if (default_cfg != NULL) {
3262 				device_printf(sc->dev,
3263 				    "unable to load module \"%s\" for "
3264 				    "configuration profile \"%s\", will use "
3265 				    "the default config file instead.\n",
3266 				    s, sc->cfg_file);
3267 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3268 				    "%s", DEFAULT_CF);
3269 			} else {
3270 				device_printf(sc->dev,
3271 				    "unable to load module \"%s\" for "
3272 				    "configuration profile \"%s\", will use "
3273 				    "the config file on the card's flash "
3274 				    "instead.\n", s, sc->cfg_file);
3275 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3276 				    "%s", FLASH_CF);
3277 			}
3278 		}
3279 	}
3280 
3281 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3282 	    default_cfg == NULL) {
3283 		device_printf(sc->dev,
3284 		    "default config file not available, will use the config "
3285 		    "file on the card's flash instead.\n");
3286 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3287 	}
3288 
3289 	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3290 		u_int cflen;
3291 		const uint32_t *cfdata;
3292 		uint32_t param, val, addr;
3293 
3294 		KASSERT(cfg != NULL || default_cfg != NULL,
3295 		    ("%s: no config to upload", __func__));
3296 
3297 		/*
3298 		 * Ask the firmware where it wants us to upload the config file.
3299 		 */
3300 		param = FW_PARAM_DEV(CF);
3301 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3302 		if (rc != 0) {
3303 			/* No support for config file?  Shouldn't happen. */
3304 			device_printf(sc->dev,
3305 			    "failed to query config file location: %d.\n", rc);
3306 			goto done;
3307 		}
3308 		mtype = G_FW_PARAMS_PARAM_Y(val);
3309 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3310 
3311 		/*
3312 		 * XXX: sheer laziness.  We deliberately added 4 bytes of
3313 		 * useless stuffing/comments at the end of the config file so
3314 		 * it's ok to simply throw away the last remaining bytes when
3315 		 * the config file is not an exact multiple of 4.  This also
3316 		 * helps with the validate_mt_off_len check.
3317 		 */
3318 		if (cfg != NULL) {
3319 			cflen = cfg->datasize & ~3;
3320 			cfdata = cfg->data;
3321 		} else {
3322 			cflen = default_cfg->datasize & ~3;
3323 			cfdata = default_cfg->data;
3324 		}
3325 
3326 		if (cflen > FLASH_CFG_MAX_SIZE) {
3327 			device_printf(sc->dev,
3328 			    "config file too long (%d, max allowed is %d).  "
3329 			    "Will try to use the config on the card, if any.\n",
3330 			    cflen, FLASH_CFG_MAX_SIZE);
3331 			goto use_config_on_flash;
3332 		}
3333 
3334 		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3335 		if (rc != 0) {
3336 			device_printf(sc->dev,
3337 			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3338 			    "Will try to use the config on the card, if any.\n",
3339 			    __func__, mtype, moff, cflen, rc);
3340 			goto use_config_on_flash;
3341 		}
3342 		write_via_memwin(sc, 2, addr, cfdata, cflen);
3343 	} else {
3344 use_config_on_flash:
3345 		mtype = FW_MEMTYPE_FLASH;
3346 		moff = t4_flash_cfg_addr(sc);
3347 	}
3348 
3349 	bzero(&caps, sizeof(caps));
3350 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3351 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3352 	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3353 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3354 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3355 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3356 	if (rc != 0) {
3357 		device_printf(sc->dev,
3358 		    "failed to pre-process config file: %d "
3359 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3360 		goto done;
3361 	}
3362 
3363 	finicsum = be32toh(caps.finicsum);
3364 	cfcsum = be32toh(caps.cfcsum);
3365 	if (finicsum != cfcsum) {
3366 		device_printf(sc->dev,
3367 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3368 		    finicsum, cfcsum);
3369 	}
3370 	sc->cfcsum = cfcsum;
3371 
3372 #define LIMIT_CAPS(x) do { \
3373 	caps.x &= htobe16(t4_##x##_allowed); \
3374 } while (0)
3375 
3376 	/*
3377 	 * Let the firmware know what features will (not) be used so it can tune
3378 	 * things accordingly.
3379 	 */
3380 	LIMIT_CAPS(nbmcaps);
3381 	LIMIT_CAPS(linkcaps);
3382 	LIMIT_CAPS(switchcaps);
3383 	LIMIT_CAPS(niccaps);
3384 	LIMIT_CAPS(toecaps);
3385 	LIMIT_CAPS(rdmacaps);
3386 	LIMIT_CAPS(cryptocaps);
3387 	LIMIT_CAPS(iscsicaps);
3388 	LIMIT_CAPS(fcoecaps);
3389 #undef LIMIT_CAPS
3390 
3391 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3392 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3393 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3394 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3395 	if (rc != 0) {
3396 		device_printf(sc->dev,
3397 		    "failed to process config file: %d.\n", rc);
3398 	}
3399 done:
3400 	if (cfg != NULL)
3401 		firmware_put(cfg, FIRMWARE_UNLOAD);
3402 	return (rc);
3403 }
3404 
3405 /*
3406  * Retrieve parameters that are needed (or nice to have) very early.
3407  */
3408 static int
3409 get_params__pre_init(struct adapter *sc)
3410 {
3411 	int rc;
3412 	uint32_t param[2], val[2];
3413 
3414 	t4_get_version_info(sc);
3415 
3416 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3417 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3418 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3419 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3420 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3421 
3422 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3423 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3424 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3425 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3426 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3427 
3428 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3429 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3430 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3431 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3432 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3433 
3434 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3435 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3436 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3437 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3438 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3439 
3440 	param[0] = FW_PARAM_DEV(PORTVEC);
3441 	param[1] = FW_PARAM_DEV(CCLK);
3442 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3443 	if (rc != 0) {
3444 		device_printf(sc->dev,
3445 		    "failed to query parameters (pre_init): %d.\n", rc);
3446 		return (rc);
3447 	}
3448 
3449 	sc->params.portvec = val[0];
3450 	sc->params.nports = bitcount32(val[0]);
3451 	sc->params.vpd.cclk = val[1];
3452 
3453 	/* Read device log parameters. */
3454 	rc = -t4_init_devlog_params(sc, 1);
3455 	if (rc == 0)
3456 		fixup_devlog_params(sc);
3457 	else {
3458 		device_printf(sc->dev,
3459 		    "failed to get devlog parameters: %d.\n", rc);
3460 		rc = 0;	/* devlog isn't critical for device operation */
3461 	}
3462 
3463 	return (rc);
3464 }
3465 
3466 /*
3467  * Retrieve various parameters that are of interest to the driver.  The device
3468  * has been initialized by the firmware at this point.
3469  */
3470 static int
3471 get_params__post_init(struct adapter *sc)
3472 {
3473 	int rc;
3474 	uint32_t param[7], val[7];
3475 	struct fw_caps_config_cmd caps;
3476 
3477 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3478 	param[1] = FW_PARAM_PFVF(EQ_START);
3479 	param[2] = FW_PARAM_PFVF(FILTER_START);
3480 	param[3] = FW_PARAM_PFVF(FILTER_END);
3481 	param[4] = FW_PARAM_PFVF(L2T_START);
3482 	param[5] = FW_PARAM_PFVF(L2T_END);
3483 	param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
3484 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
3485 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
3486 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
3487 	if (rc != 0) {
3488 		device_printf(sc->dev,
3489 		    "failed to query parameters (post_init): %d.\n", rc);
3490 		return (rc);
3491 	}
3492 
3493 	sc->sge.iq_start = val[0];
3494 	sc->sge.eq_start = val[1];
3495 	sc->tids.ftid_base = val[2];
3496 	sc->tids.nftids = val[3] - val[2] + 1;
3497 	sc->params.ftid_min = val[2];
3498 	sc->params.ftid_max = val[3];
3499 	sc->vres.l2t.start = val[4];
3500 	sc->vres.l2t.size = val[5] - val[4] + 1;
3501 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3502 	    ("%s: L2 table size (%u) larger than expected (%u)",
3503 	    __func__, sc->vres.l2t.size, L2T_SIZE));
3504 	sc->params.core_vdd = val[6];
3505 
3506 	/*
3507 	 * MPSBGMAP is queried separately because only recent firmwares support
3508 	 * it as a parameter and we don't want the compound query above to fail
3509 	 * on older firmwares.
3510 	 */
3511 	param[0] = FW_PARAM_DEV(MPSBGMAP);
3512 	val[0] = 0;
3513 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3514 	if (rc == 0)
3515 		sc->params.mps_bg_map = val[0];
3516 	else
3517 		sc->params.mps_bg_map = 0;
3518 
3519 	/* get capabilites */
3520 	bzero(&caps, sizeof(caps));
3521 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3522 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3523 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3524 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3525 	if (rc != 0) {
3526 		device_printf(sc->dev,
3527 		    "failed to get card capabilities: %d.\n", rc);
3528 		return (rc);
3529 	}
3530 
3531 #define READ_CAPS(x) do { \
3532 	sc->x = htobe16(caps.x); \
3533 } while (0)
3534 	READ_CAPS(nbmcaps);
3535 	READ_CAPS(linkcaps);
3536 	READ_CAPS(switchcaps);
3537 	READ_CAPS(niccaps);
3538 	READ_CAPS(toecaps);
3539 	READ_CAPS(rdmacaps);
3540 	READ_CAPS(cryptocaps);
3541 	READ_CAPS(iscsicaps);
3542 	READ_CAPS(fcoecaps);
3543 
3544 	/*
3545 	 * The firmware attempts memfree TOE configuration for -SO cards and
3546 	 * will report toecaps=0 if it runs out of resources (this depends on
3547 	 * the config file).  It may not report 0 for other capabilities
3548 	 * dependent on the TOE in this case.  Set them to 0 here so that the
3549 	 * driver doesn't bother tracking resources that will never be used.
3550 	 */
3551 	if (sc->toecaps == 0) {
3552 		sc->iscsicaps = 0;
3553 		sc->rdmacaps = 0;
3554 	}
3555 
3556 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3557 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3558 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3559 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3560 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3561 		if (rc != 0) {
3562 			device_printf(sc->dev,
3563 			    "failed to query NIC parameters: %d.\n", rc);
3564 			return (rc);
3565 		}
3566 		sc->tids.etid_base = val[0];
3567 		sc->params.etid_min = val[0];
3568 		sc->tids.netids = val[1] - val[0] + 1;
3569 		sc->params.netids = sc->tids.netids;
3570 		sc->params.eo_wr_cred = val[2];
3571 		sc->params.ethoffload = 1;
3572 	}
3573 
3574 	if (sc->toecaps) {
3575 		/* query offload-related parameters */
3576 		param[0] = FW_PARAM_DEV(NTID);
3577 		param[1] = FW_PARAM_PFVF(SERVER_START);
3578 		param[2] = FW_PARAM_PFVF(SERVER_END);
3579 		param[3] = FW_PARAM_PFVF(TDDP_START);
3580 		param[4] = FW_PARAM_PFVF(TDDP_END);
3581 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3582 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3583 		if (rc != 0) {
3584 			device_printf(sc->dev,
3585 			    "failed to query TOE parameters: %d.\n", rc);
3586 			return (rc);
3587 		}
3588 		sc->tids.ntids = val[0];
3589 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3590 		sc->tids.stid_base = val[1];
3591 		sc->tids.nstids = val[2] - val[1] + 1;
3592 		sc->vres.ddp.start = val[3];
3593 		sc->vres.ddp.size = val[4] - val[3] + 1;
3594 		sc->params.ofldq_wr_cred = val[5];
3595 		sc->params.offload = 1;
3596 	}
3597 	if (sc->rdmacaps) {
3598 		param[0] = FW_PARAM_PFVF(STAG_START);
3599 		param[1] = FW_PARAM_PFVF(STAG_END);
3600 		param[2] = FW_PARAM_PFVF(RQ_START);
3601 		param[3] = FW_PARAM_PFVF(RQ_END);
3602 		param[4] = FW_PARAM_PFVF(PBL_START);
3603 		param[5] = FW_PARAM_PFVF(PBL_END);
3604 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3605 		if (rc != 0) {
3606 			device_printf(sc->dev,
3607 			    "failed to query RDMA parameters(1): %d.\n", rc);
3608 			return (rc);
3609 		}
3610 		sc->vres.stag.start = val[0];
3611 		sc->vres.stag.size = val[1] - val[0] + 1;
3612 		sc->vres.rq.start = val[2];
3613 		sc->vres.rq.size = val[3] - val[2] + 1;
3614 		sc->vres.pbl.start = val[4];
3615 		sc->vres.pbl.size = val[5] - val[4] + 1;
3616 
3617 		param[0] = FW_PARAM_PFVF(SQRQ_START);
3618 		param[1] = FW_PARAM_PFVF(SQRQ_END);
3619 		param[2] = FW_PARAM_PFVF(CQ_START);
3620 		param[3] = FW_PARAM_PFVF(CQ_END);
3621 		param[4] = FW_PARAM_PFVF(OCQ_START);
3622 		param[5] = FW_PARAM_PFVF(OCQ_END);
3623 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3624 		if (rc != 0) {
3625 			device_printf(sc->dev,
3626 			    "failed to query RDMA parameters(2): %d.\n", rc);
3627 			return (rc);
3628 		}
3629 		sc->vres.qp.start = val[0];
3630 		sc->vres.qp.size = val[1] - val[0] + 1;
3631 		sc->vres.cq.start = val[2];
3632 		sc->vres.cq.size = val[3] - val[2] + 1;
3633 		sc->vres.ocq.start = val[4];
3634 		sc->vres.ocq.size = val[5] - val[4] + 1;
3635 
3636 		param[0] = FW_PARAM_PFVF(SRQ_START);
3637 		param[1] = FW_PARAM_PFVF(SRQ_END);
3638 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
3639 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
3640 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
3641 		if (rc != 0) {
3642 			device_printf(sc->dev,
3643 			    "failed to query RDMA parameters(3): %d.\n", rc);
3644 			return (rc);
3645 		}
3646 		sc->vres.srq.start = val[0];
3647 		sc->vres.srq.size = val[1] - val[0] + 1;
3648 		sc->params.max_ordird_qp = val[2];
3649 		sc->params.max_ird_adapter = val[3];
3650 	}
3651 	if (sc->iscsicaps) {
3652 		param[0] = FW_PARAM_PFVF(ISCSI_START);
3653 		param[1] = FW_PARAM_PFVF(ISCSI_END);
3654 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3655 		if (rc != 0) {
3656 			device_printf(sc->dev,
3657 			    "failed to query iSCSI parameters: %d.\n", rc);
3658 			return (rc);
3659 		}
3660 		sc->vres.iscsi.start = val[0];
3661 		sc->vres.iscsi.size = val[1] - val[0] + 1;
3662 	}
3663 
3664 	t4_init_sge_params(sc);
3665 
3666 	/*
3667 	 * We've got the params we wanted to query via the firmware.  Now grab
3668 	 * some others directly from the chip.
3669 	 */
3670 	rc = t4_read_chip_settings(sc);
3671 
3672 	return (rc);
3673 }
3674 
3675 static int
3676 set_params__post_init(struct adapter *sc)
3677 {
3678 	uint32_t param, val;
3679 #ifdef TCP_OFFLOAD
3680 	int i, v, shift;
3681 #endif
3682 
3683 	/* ask for encapsulated CPLs */
3684 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3685 	val = 1;
3686 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3687 
3688 #ifdef TCP_OFFLOAD
3689 	/*
3690 	 * Override the TOE timers with user provided tunables.  This is not the
3691 	 * recommended way to change the timers (the firmware config file is) so
3692 	 * these tunables are not documented.
3693 	 *
3694 	 * All the timer tunables are in microseconds.
3695 	 */
3696 	if (t4_toe_keepalive_idle != 0) {
3697 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
3698 		v &= M_KEEPALIVEIDLE;
3699 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
3700 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
3701 	}
3702 	if (t4_toe_keepalive_interval != 0) {
3703 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
3704 		v &= M_KEEPALIVEINTVL;
3705 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
3706 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
3707 	}
3708 	if (t4_toe_keepalive_count != 0) {
3709 		v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
3710 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
3711 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
3712 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
3713 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
3714 	}
3715 	if (t4_toe_rexmt_min != 0) {
3716 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
3717 		v &= M_RXTMIN;
3718 		t4_set_reg_field(sc, A_TP_RXT_MIN,
3719 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
3720 	}
3721 	if (t4_toe_rexmt_max != 0) {
3722 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
3723 		v &= M_RXTMAX;
3724 		t4_set_reg_field(sc, A_TP_RXT_MAX,
3725 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
3726 	}
3727 	if (t4_toe_rexmt_count != 0) {
3728 		v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
3729 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
3730 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
3731 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
3732 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
3733 	}
3734 	for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
3735 		if (t4_toe_rexmt_backoff[i] != -1) {
3736 			v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
3737 			shift = (i & 3) << 3;
3738 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
3739 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
3740 		}
3741 	}
3742 #endif
3743 	return (0);
3744 }
3745 
3746 #undef FW_PARAM_PFVF
3747 #undef FW_PARAM_DEV
3748 
3749 static void
3750 t4_set_desc(struct adapter *sc)
3751 {
3752 	char buf[128];
3753 	struct adapter_params *p = &sc->params;
3754 
3755 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
3756 
3757 	device_set_desc_copy(sc->dev, buf);
3758 }
3759 
3760 static void
3761 build_medialist(struct port_info *pi, struct ifmedia *media)
3762 {
3763 	int m;
3764 
3765 	PORT_LOCK_ASSERT_OWNED(pi);
3766 
3767 	ifmedia_removeall(media);
3768 
3769 	/*
3770 	 * XXX: Would it be better to ifmedia_add all 4 combinations of pause
3771 	 * settings for every speed instead of just txpause|rxpause?  ifconfig
3772 	 * media display looks much better if autoselect is the only case where
3773 	 * ifm_current is different from ifm_active.  If the user picks anything
3774 	 * except txpause|rxpause the display is ugly.
3775 	 */
3776 	m = IFM_ETHER | IFM_FDX | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
3777 
3778 	switch(pi->port_type) {
3779 	case FW_PORT_TYPE_BT_XFI:
3780 	case FW_PORT_TYPE_BT_XAUI:
3781 		ifmedia_add(media, m | IFM_10G_T, 0, NULL);
3782 		/* fall through */
3783 
3784 	case FW_PORT_TYPE_BT_SGMII:
3785 		ifmedia_add(media, m | IFM_1000_T, 0, NULL);
3786 		ifmedia_add(media, m | IFM_100_TX, 0, NULL);
3787 		ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL);
3788 		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
3789 		break;
3790 
3791 	case FW_PORT_TYPE_CX4:
3792 		ifmedia_add(media, m | IFM_10G_CX4, 0, NULL);
3793 		ifmedia_set(media, m | IFM_10G_CX4);
3794 		break;
3795 
3796 	case FW_PORT_TYPE_QSFP_10G:
3797 	case FW_PORT_TYPE_SFP:
3798 	case FW_PORT_TYPE_FIBER_XFI:
3799 	case FW_PORT_TYPE_FIBER_XAUI:
3800 		switch (pi->mod_type) {
3801 
3802 		case FW_PORT_MOD_TYPE_LR:
3803 			ifmedia_add(media, m | IFM_10G_LR, 0, NULL);
3804 			ifmedia_set(media, m | IFM_10G_LR);
3805 			break;
3806 
3807 		case FW_PORT_MOD_TYPE_SR:
3808 			ifmedia_add(media, m | IFM_10G_SR, 0, NULL);
3809 			ifmedia_set(media, m | IFM_10G_SR);
3810 			break;
3811 
3812 		case FW_PORT_MOD_TYPE_LRM:
3813 			ifmedia_add(media, m | IFM_10G_LRM, 0, NULL);
3814 			ifmedia_set(media, m | IFM_10G_LRM);
3815 			break;
3816 
3817 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3818 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3819 			ifmedia_add(media, m | IFM_10G_TWINAX, 0, NULL);
3820 			ifmedia_set(media, m | IFM_10G_TWINAX);
3821 			break;
3822 
3823 		case FW_PORT_MOD_TYPE_NONE:
3824 			m &= ~IFM_FDX;
3825 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3826 			ifmedia_set(media, m | IFM_NONE);
3827 			break;
3828 
3829 		case FW_PORT_MOD_TYPE_NA:
3830 		case FW_PORT_MOD_TYPE_ER:
3831 		default:
3832 			device_printf(pi->dev,
3833 			    "unknown port_type (%d), mod_type (%d)\n",
3834 			    pi->port_type, pi->mod_type);
3835 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3836 			ifmedia_set(media, m | IFM_UNKNOWN);
3837 			break;
3838 		}
3839 		break;
3840 
3841 	case FW_PORT_TYPE_CR_QSFP:
3842 	case FW_PORT_TYPE_SFP28:
3843 	case FW_PORT_TYPE_KR_SFP28:
3844 		switch (pi->mod_type) {
3845 
3846 		case FW_PORT_MOD_TYPE_SR:
3847 			ifmedia_add(media, m | IFM_25G_SR, 0, NULL);
3848 			ifmedia_set(media, m | IFM_25G_SR);
3849 			break;
3850 
3851 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3852 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3853 			ifmedia_add(media, m | IFM_25G_CR, 0, NULL);
3854 			ifmedia_set(media, m | IFM_25G_CR);
3855 			break;
3856 
3857 		case FW_PORT_MOD_TYPE_NONE:
3858 			m &= ~IFM_FDX;
3859 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3860 			ifmedia_set(media, m | IFM_NONE);
3861 			break;
3862 
3863 		default:
3864 			device_printf(pi->dev,
3865 			    "unknown port_type (%d), mod_type (%d)\n",
3866 			    pi->port_type, pi->mod_type);
3867 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3868 			ifmedia_set(media, m | IFM_UNKNOWN);
3869 			break;
3870 		}
3871 		break;
3872 
3873 	case FW_PORT_TYPE_QSFP:
3874 		switch (pi->mod_type) {
3875 
3876 		case FW_PORT_MOD_TYPE_LR:
3877 			ifmedia_add(media, m | IFM_40G_LR4, 0, NULL);
3878 			ifmedia_set(media, m | IFM_40G_LR4);
3879 			break;
3880 
3881 		case FW_PORT_MOD_TYPE_SR:
3882 			ifmedia_add(media, m | IFM_40G_SR4, 0, NULL);
3883 			ifmedia_set(media, m | IFM_40G_SR4);
3884 			break;
3885 
3886 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3887 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3888 			ifmedia_add(media, m | IFM_40G_CR4, 0, NULL);
3889 			ifmedia_set(media, m | IFM_40G_CR4);
3890 			break;
3891 
3892 		case FW_PORT_MOD_TYPE_NONE:
3893 			m &= ~IFM_FDX;
3894 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3895 			ifmedia_set(media, m | IFM_NONE);
3896 			break;
3897 
3898 		default:
3899 			device_printf(pi->dev,
3900 			    "unknown port_type (%d), mod_type (%d)\n",
3901 			    pi->port_type, pi->mod_type);
3902 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3903 			ifmedia_set(media, m | IFM_UNKNOWN);
3904 			break;
3905 		}
3906 		break;
3907 
3908 	case FW_PORT_TYPE_KR4_100G:
3909 	case FW_PORT_TYPE_CR4_QSFP:
3910 		switch (pi->mod_type) {
3911 
3912 		case FW_PORT_MOD_TYPE_LR:
3913 			ifmedia_add(media, m | IFM_100G_LR4, 0, NULL);
3914 			ifmedia_set(media, m | IFM_100G_LR4);
3915 			break;
3916 
3917 		case FW_PORT_MOD_TYPE_SR:
3918 			ifmedia_add(media, m | IFM_100G_SR4, 0, NULL);
3919 			ifmedia_set(media, m | IFM_100G_SR4);
3920 			break;
3921 
3922 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3923 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3924 			ifmedia_add(media, m | IFM_100G_CR4, 0, NULL);
3925 			ifmedia_set(media, m | IFM_100G_CR4);
3926 			break;
3927 
3928 		case FW_PORT_MOD_TYPE_NONE:
3929 			m &= ~IFM_FDX;
3930 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3931 			ifmedia_set(media, m | IFM_NONE);
3932 			break;
3933 
3934 		default:
3935 			device_printf(pi->dev,
3936 			    "unknown port_type (%d), mod_type (%d)\n",
3937 			    pi->port_type, pi->mod_type);
3938 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3939 			ifmedia_set(media, m | IFM_UNKNOWN);
3940 			break;
3941 		}
3942 		break;
3943 
3944 	default:
3945 		device_printf(pi->dev,
3946 		    "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
3947 		    pi->mod_type);
3948 		ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3949 		ifmedia_set(media, m | IFM_UNKNOWN);
3950 		break;
3951 	}
3952 }
3953 
3954 /*
3955  * Update all the requested_* fields in the link config and then send a mailbox
3956  * command to apply the settings.
3957  */
3958 static void
3959 init_l1cfg(struct port_info *pi)
3960 {
3961 	struct adapter *sc = pi->adapter;
3962 	struct link_config *lc = &pi->link_cfg;
3963 	int rc;
3964 
3965 	ASSERT_SYNCHRONIZED_OP(sc);
3966 
3967 	if (t4_autoneg != 0 && lc->supported & FW_PORT_CAP_ANEG) {
3968 		lc->requested_aneg = AUTONEG_ENABLE;
3969 		lc->requested_speed = 0;
3970 	} else {
3971 		lc->requested_aneg = AUTONEG_DISABLE;
3972 		lc->requested_speed = port_top_speed(pi);	/* in Gbps */
3973 	}
3974 
3975 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX);
3976 
3977 	if (t4_fec != -1) {
3978 		lc->requested_fec = t4_fec & (FEC_RS | FEC_BASER_RS |
3979 		    FEC_RESERVED);
3980 	} else {
3981 		/* Use the suggested value provided by the firmware in acaps */
3982 		if (lc->advertising & FW_PORT_CAP_FEC_RS)
3983 			lc->requested_fec = FEC_RS;
3984 		else if (lc->advertising & FW_PORT_CAP_FEC_BASER_RS)
3985 			lc->requested_fec = FEC_BASER_RS;
3986 		else if (lc->advertising & FW_PORT_CAP_FEC_RESERVED)
3987 			lc->requested_fec = FEC_RESERVED;
3988 		else
3989 			lc->requested_fec = 0;
3990 	}
3991 
3992 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
3993 	if (rc != 0) {
3994 		device_printf(pi->dev, "l1cfg failed: %d\n", rc);
3995 	} else {
3996 		lc->fc = lc->requested_fc;
3997 		lc->fec = lc->requested_fec;
3998 	}
3999 }
4000 
4001 #define FW_MAC_EXACT_CHUNK	7
4002 
4003 /*
4004  * Program the port's XGMAC based on parameters in ifnet.  The caller also
4005  * indicates which parameters should be programmed (the rest are left alone).
4006  */
4007 int
4008 update_mac_settings(struct ifnet *ifp, int flags)
4009 {
4010 	int rc = 0;
4011 	struct vi_info *vi = ifp->if_softc;
4012 	struct port_info *pi = vi->pi;
4013 	struct adapter *sc = pi->adapter;
4014 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4015 
4016 	ASSERT_SYNCHRONIZED_OP(sc);
4017 	KASSERT(flags, ("%s: not told what to update.", __func__));
4018 
4019 	if (flags & XGMAC_MTU)
4020 		mtu = ifp->if_mtu;
4021 
4022 	if (flags & XGMAC_PROMISC)
4023 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4024 
4025 	if (flags & XGMAC_ALLMULTI)
4026 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4027 
4028 	if (flags & XGMAC_VLANEX)
4029 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4030 
4031 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4032 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4033 		    allmulti, 1, vlanex, false);
4034 		if (rc) {
4035 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4036 			    rc);
4037 			return (rc);
4038 		}
4039 	}
4040 
4041 	if (flags & XGMAC_UCADDR) {
4042 		uint8_t ucaddr[ETHER_ADDR_LEN];
4043 
4044 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4045 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4046 		    ucaddr, true, true);
4047 		if (rc < 0) {
4048 			rc = -rc;
4049 			if_printf(ifp, "change_mac failed: %d\n", rc);
4050 			return (rc);
4051 		} else {
4052 			vi->xact_addr_filt = rc;
4053 			rc = 0;
4054 		}
4055 	}
4056 
4057 	if (flags & XGMAC_MCADDRS) {
4058 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4059 		int del = 1;
4060 		uint64_t hash = 0;
4061 		struct ifmultiaddr *ifma;
4062 		int i = 0, j;
4063 
4064 		if_maddr_rlock(ifp);
4065 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4066 			if (ifma->ifma_addr->sa_family != AF_LINK)
4067 				continue;
4068 			mcaddr[i] =
4069 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4070 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4071 			i++;
4072 
4073 			if (i == FW_MAC_EXACT_CHUNK) {
4074 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4075 				    del, i, mcaddr, NULL, &hash, 0);
4076 				if (rc < 0) {
4077 					rc = -rc;
4078 					for (j = 0; j < i; j++) {
4079 						if_printf(ifp,
4080 						    "failed to add mc address"
4081 						    " %02x:%02x:%02x:"
4082 						    "%02x:%02x:%02x rc=%d\n",
4083 						    mcaddr[j][0], mcaddr[j][1],
4084 						    mcaddr[j][2], mcaddr[j][3],
4085 						    mcaddr[j][4], mcaddr[j][5],
4086 						    rc);
4087 					}
4088 					goto mcfail;
4089 				}
4090 				del = 0;
4091 				i = 0;
4092 			}
4093 		}
4094 		if (i > 0) {
4095 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4096 			    mcaddr, NULL, &hash, 0);
4097 			if (rc < 0) {
4098 				rc = -rc;
4099 				for (j = 0; j < i; j++) {
4100 					if_printf(ifp,
4101 					    "failed to add mc address"
4102 					    " %02x:%02x:%02x:"
4103 					    "%02x:%02x:%02x rc=%d\n",
4104 					    mcaddr[j][0], mcaddr[j][1],
4105 					    mcaddr[j][2], mcaddr[j][3],
4106 					    mcaddr[j][4], mcaddr[j][5],
4107 					    rc);
4108 				}
4109 				goto mcfail;
4110 			}
4111 		}
4112 
4113 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4114 		if (rc != 0)
4115 			if_printf(ifp, "failed to set mc address hash: %d", rc);
4116 mcfail:
4117 		if_maddr_runlock(ifp);
4118 	}
4119 
4120 	return (rc);
4121 }
4122 
4123 /*
4124  * {begin|end}_synchronized_op must be called from the same thread.
4125  */
4126 int
4127 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4128     char *wmesg)
4129 {
4130 	int rc, pri;
4131 
4132 #ifdef WITNESS
4133 	/* the caller thinks it's ok to sleep, but is it really? */
4134 	if (flags & SLEEP_OK)
4135 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4136 		    "begin_synchronized_op");
4137 #endif
4138 
4139 	if (INTR_OK)
4140 		pri = PCATCH;
4141 	else
4142 		pri = 0;
4143 
4144 	ADAPTER_LOCK(sc);
4145 	for (;;) {
4146 
4147 		if (vi && IS_DOOMED(vi)) {
4148 			rc = ENXIO;
4149 			goto done;
4150 		}
4151 
4152 		if (!IS_BUSY(sc)) {
4153 			rc = 0;
4154 			break;
4155 		}
4156 
4157 		if (!(flags & SLEEP_OK)) {
4158 			rc = EBUSY;
4159 			goto done;
4160 		}
4161 
4162 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4163 			rc = EINTR;
4164 			goto done;
4165 		}
4166 	}
4167 
4168 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4169 	SET_BUSY(sc);
4170 #ifdef INVARIANTS
4171 	sc->last_op = wmesg;
4172 	sc->last_op_thr = curthread;
4173 	sc->last_op_flags = flags;
4174 #endif
4175 
4176 done:
4177 	if (!(flags & HOLD_LOCK) || rc)
4178 		ADAPTER_UNLOCK(sc);
4179 
4180 	return (rc);
4181 }
4182 
4183 /*
4184  * Tell if_ioctl and if_init that the VI is going away.  This is
4185  * special variant of begin_synchronized_op and must be paired with a
4186  * call to end_synchronized_op.
4187  */
4188 void
4189 doom_vi(struct adapter *sc, struct vi_info *vi)
4190 {
4191 
4192 	ADAPTER_LOCK(sc);
4193 	SET_DOOMED(vi);
4194 	wakeup(&sc->flags);
4195 	while (IS_BUSY(sc))
4196 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4197 	SET_BUSY(sc);
4198 #ifdef INVARIANTS
4199 	sc->last_op = "t4detach";
4200 	sc->last_op_thr = curthread;
4201 	sc->last_op_flags = 0;
4202 #endif
4203 	ADAPTER_UNLOCK(sc);
4204 }
4205 
4206 /*
4207  * {begin|end}_synchronized_op must be called from the same thread.
4208  */
4209 void
4210 end_synchronized_op(struct adapter *sc, int flags)
4211 {
4212 
4213 	if (flags & LOCK_HELD)
4214 		ADAPTER_LOCK_ASSERT_OWNED(sc);
4215 	else
4216 		ADAPTER_LOCK(sc);
4217 
4218 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
4219 	CLR_BUSY(sc);
4220 	wakeup(&sc->flags);
4221 	ADAPTER_UNLOCK(sc);
4222 }
4223 
4224 static int
4225 cxgbe_init_synchronized(struct vi_info *vi)
4226 {
4227 	struct port_info *pi = vi->pi;
4228 	struct adapter *sc = pi->adapter;
4229 	struct ifnet *ifp = vi->ifp;
4230 	int rc = 0, i;
4231 	struct sge_txq *txq;
4232 
4233 	ASSERT_SYNCHRONIZED_OP(sc);
4234 
4235 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
4236 		return (0);	/* already running */
4237 
4238 	if (!(sc->flags & FULL_INIT_DONE) &&
4239 	    ((rc = adapter_full_init(sc)) != 0))
4240 		return (rc);	/* error message displayed already */
4241 
4242 	if (!(vi->flags & VI_INIT_DONE) &&
4243 	    ((rc = vi_full_init(vi)) != 0))
4244 		return (rc); /* error message displayed already */
4245 
4246 	rc = update_mac_settings(ifp, XGMAC_ALL);
4247 	if (rc)
4248 		goto done;	/* error message displayed already */
4249 
4250 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
4251 	if (rc != 0) {
4252 		if_printf(ifp, "enable_vi failed: %d\n", rc);
4253 		goto done;
4254 	}
4255 
4256 	/*
4257 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
4258 	 * if this changes.
4259 	 */
4260 
4261 	for_each_txq(vi, i, txq) {
4262 		TXQ_LOCK(txq);
4263 		txq->eq.flags |= EQ_ENABLED;
4264 		TXQ_UNLOCK(txq);
4265 	}
4266 
4267 	/*
4268 	 * The first iq of the first port to come up is used for tracing.
4269 	 */
4270 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4271 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4272 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4273 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4274 		    V_QUEUENUMBER(sc->traceq));
4275 		pi->flags |= HAS_TRACEQ;
4276 	}
4277 
4278 	/* all ok */
4279 	PORT_LOCK(pi);
4280 	if (pi->up_vis++ == 0) {
4281 		t4_update_port_info(pi);
4282 		build_medialist(pi, &pi->media);
4283 		init_l1cfg(pi);
4284 	}
4285 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
4286 
4287 	if (pi->nvi > 1 || sc->flags & IS_VF)
4288 		callout_reset(&vi->tick, hz, vi_tick, vi);
4289 	else
4290 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4291 	PORT_UNLOCK(pi);
4292 done:
4293 	if (rc != 0)
4294 		cxgbe_uninit_synchronized(vi);
4295 
4296 	return (rc);
4297 }
4298 
4299 /*
4300  * Idempotent.
4301  */
4302 static int
4303 cxgbe_uninit_synchronized(struct vi_info *vi)
4304 {
4305 	struct port_info *pi = vi->pi;
4306 	struct adapter *sc = pi->adapter;
4307 	struct ifnet *ifp = vi->ifp;
4308 	int rc, i;
4309 	struct sge_txq *txq;
4310 
4311 	ASSERT_SYNCHRONIZED_OP(sc);
4312 
4313 	if (!(vi->flags & VI_INIT_DONE)) {
4314 		KASSERT(!(ifp->if_drv_flags & IFF_DRV_RUNNING),
4315 		    ("uninited VI is running"));
4316 		return (0);
4317 	}
4318 
4319 	/*
4320 	 * Disable the VI so that all its data in either direction is discarded
4321 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4322 	 * tick) intact as the TP can deliver negative advice or data that it's
4323 	 * holding in its RAM (for an offloaded connection) even after the VI is
4324 	 * disabled.
4325 	 */
4326 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4327 	if (rc) {
4328 		if_printf(ifp, "disable_vi failed: %d\n", rc);
4329 		return (rc);
4330 	}
4331 
4332 	for_each_txq(vi, i, txq) {
4333 		TXQ_LOCK(txq);
4334 		txq->eq.flags &= ~EQ_ENABLED;
4335 		TXQ_UNLOCK(txq);
4336 	}
4337 
4338 	PORT_LOCK(pi);
4339 	if (pi->nvi > 1 || sc->flags & IS_VF)
4340 		callout_stop(&vi->tick);
4341 	else
4342 		callout_stop(&pi->tick);
4343 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4344 		PORT_UNLOCK(pi);
4345 		return (0);
4346 	}
4347 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4348 	pi->up_vis--;
4349 	if (pi->up_vis > 0) {
4350 		PORT_UNLOCK(pi);
4351 		return (0);
4352 	}
4353 	PORT_UNLOCK(pi);
4354 
4355 	pi->link_cfg.link_ok = 0;
4356 	pi->link_cfg.speed = 0;
4357 	pi->link_cfg.link_down_rc = 255;
4358 	t4_os_link_changed(pi);
4359 	pi->old_link_cfg = pi->link_cfg;
4360 
4361 	return (0);
4362 }
4363 
4364 /*
4365  * It is ok for this function to fail midway and return right away.  t4_detach
4366  * will walk the entire sc->irq list and clean up whatever is valid.
4367  */
4368 int
4369 t4_setup_intr_handlers(struct adapter *sc)
4370 {
4371 	int rc, rid, p, q, v;
4372 	char s[8];
4373 	struct irq *irq;
4374 	struct port_info *pi;
4375 	struct vi_info *vi;
4376 	struct sge *sge = &sc->sge;
4377 	struct sge_rxq *rxq;
4378 #ifdef TCP_OFFLOAD
4379 	struct sge_ofld_rxq *ofld_rxq;
4380 #endif
4381 #ifdef DEV_NETMAP
4382 	struct sge_nm_rxq *nm_rxq;
4383 #endif
4384 #ifdef RSS
4385 	int nbuckets = rss_getnumbuckets();
4386 #endif
4387 
4388 	/*
4389 	 * Setup interrupts.
4390 	 */
4391 	irq = &sc->irq[0];
4392 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
4393 	if (forwarding_intr_to_fwq(sc))
4394 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4395 
4396 	/* Multiple interrupts. */
4397 	if (sc->flags & IS_VF)
4398 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4399 		    ("%s: too few intr.", __func__));
4400 	else
4401 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4402 		    ("%s: too few intr.", __func__));
4403 
4404 	/* The first one is always error intr on PFs */
4405 	if (!(sc->flags & IS_VF)) {
4406 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4407 		if (rc != 0)
4408 			return (rc);
4409 		irq++;
4410 		rid++;
4411 	}
4412 
4413 	/* The second one is always the firmware event queue (first on VFs) */
4414 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4415 	if (rc != 0)
4416 		return (rc);
4417 	irq++;
4418 	rid++;
4419 
4420 	for_each_port(sc, p) {
4421 		pi = sc->port[p];
4422 		for_each_vi(pi, v, vi) {
4423 			vi->first_intr = rid - 1;
4424 
4425 			if (vi->nnmrxq > 0) {
4426 				int n = max(vi->nrxq, vi->nnmrxq);
4427 
4428 				rxq = &sge->rxq[vi->first_rxq];
4429 #ifdef DEV_NETMAP
4430 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4431 #endif
4432 				for (q = 0; q < n; q++) {
4433 					snprintf(s, sizeof(s), "%x%c%x", p,
4434 					    'a' + v, q);
4435 					if (q < vi->nrxq)
4436 						irq->rxq = rxq++;
4437 #ifdef DEV_NETMAP
4438 					if (q < vi->nnmrxq)
4439 						irq->nm_rxq = nm_rxq++;
4440 #endif
4441 					rc = t4_alloc_irq(sc, irq, rid,
4442 					    t4_vi_intr, irq, s);
4443 					if (rc != 0)
4444 						return (rc);
4445 #ifdef RSS
4446 					if (q < vi->nrxq) {
4447 						bus_bind_intr(sc->dev, irq->res,
4448 						    rss_getcpu(q % nbuckets));
4449 					}
4450 #endif
4451 					irq++;
4452 					rid++;
4453 					vi->nintr++;
4454 				}
4455 			} else {
4456 				for_each_rxq(vi, q, rxq) {
4457 					snprintf(s, sizeof(s), "%x%c%x", p,
4458 					    'a' + v, q);
4459 					rc = t4_alloc_irq(sc, irq, rid,
4460 					    t4_intr, rxq, s);
4461 					if (rc != 0)
4462 						return (rc);
4463 #ifdef RSS
4464 					bus_bind_intr(sc->dev, irq->res,
4465 					    rss_getcpu(q % nbuckets));
4466 #endif
4467 					irq++;
4468 					rid++;
4469 					vi->nintr++;
4470 				}
4471 			}
4472 #ifdef TCP_OFFLOAD
4473 			for_each_ofld_rxq(vi, q, ofld_rxq) {
4474 				snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
4475 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
4476 				    ofld_rxq, s);
4477 				if (rc != 0)
4478 					return (rc);
4479 				irq++;
4480 				rid++;
4481 				vi->nintr++;
4482 			}
4483 #endif
4484 		}
4485 	}
4486 	MPASS(irq == &sc->irq[sc->intr_count]);
4487 
4488 	return (0);
4489 }
4490 
4491 int
4492 adapter_full_init(struct adapter *sc)
4493 {
4494 	int rc, i;
4495 #ifdef RSS
4496 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4497 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4498 #endif
4499 
4500 	ASSERT_SYNCHRONIZED_OP(sc);
4501 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4502 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
4503 	    ("%s: FULL_INIT_DONE already", __func__));
4504 
4505 	/*
4506 	 * queues that belong to the adapter (not any particular port).
4507 	 */
4508 	rc = t4_setup_adapter_queues(sc);
4509 	if (rc != 0)
4510 		goto done;
4511 
4512 	for (i = 0; i < nitems(sc->tq); i++) {
4513 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
4514 		    taskqueue_thread_enqueue, &sc->tq[i]);
4515 		if (sc->tq[i] == NULL) {
4516 			device_printf(sc->dev,
4517 			    "failed to allocate task queue %d\n", i);
4518 			rc = ENOMEM;
4519 			goto done;
4520 		}
4521 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
4522 		    device_get_nameunit(sc->dev), i);
4523 	}
4524 #ifdef RSS
4525 	MPASS(RSS_KEYSIZE == 40);
4526 	rss_getkey((void *)&raw_rss_key[0]);
4527 	for (i = 0; i < nitems(rss_key); i++) {
4528 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
4529 	}
4530 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
4531 #endif
4532 
4533 	if (!(sc->flags & IS_VF))
4534 		t4_intr_enable(sc);
4535 	sc->flags |= FULL_INIT_DONE;
4536 done:
4537 	if (rc != 0)
4538 		adapter_full_uninit(sc);
4539 
4540 	return (rc);
4541 }
4542 
4543 int
4544 adapter_full_uninit(struct adapter *sc)
4545 {
4546 	int i;
4547 
4548 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4549 
4550 	t4_teardown_adapter_queues(sc);
4551 
4552 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
4553 		taskqueue_free(sc->tq[i]);
4554 		sc->tq[i] = NULL;
4555 	}
4556 
4557 	sc->flags &= ~FULL_INIT_DONE;
4558 
4559 	return (0);
4560 }
4561 
4562 #ifdef RSS
4563 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
4564     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
4565     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
4566     RSS_HASHTYPE_RSS_UDP_IPV6)
4567 
4568 /* Translates kernel hash types to hardware. */
4569 static int
4570 hashconfig_to_hashen(int hashconfig)
4571 {
4572 	int hashen = 0;
4573 
4574 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
4575 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
4576 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
4577 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
4578 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
4579 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4580 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4581 	}
4582 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
4583 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4584 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4585 	}
4586 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
4587 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4588 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
4589 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4590 
4591 	return (hashen);
4592 }
4593 
4594 /* Translates hardware hash types to kernel. */
4595 static int
4596 hashen_to_hashconfig(int hashen)
4597 {
4598 	int hashconfig = 0;
4599 
4600 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
4601 		/*
4602 		 * If UDP hashing was enabled it must have been enabled for
4603 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
4604 		 * enabling any 4-tuple hash is nonsense configuration.
4605 		 */
4606 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4607 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
4608 
4609 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4610 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
4611 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4612 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
4613 	}
4614 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4615 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
4616 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4617 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
4618 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
4619 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
4620 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
4621 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
4622 
4623 	return (hashconfig);
4624 }
4625 #endif
4626 
4627 int
4628 vi_full_init(struct vi_info *vi)
4629 {
4630 	struct adapter *sc = vi->pi->adapter;
4631 	struct ifnet *ifp = vi->ifp;
4632 	uint16_t *rss;
4633 	struct sge_rxq *rxq;
4634 	int rc, i, j, hashen;
4635 #ifdef RSS
4636 	int nbuckets = rss_getnumbuckets();
4637 	int hashconfig = rss_gethashconfig();
4638 	int extra;
4639 #endif
4640 
4641 	ASSERT_SYNCHRONIZED_OP(sc);
4642 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
4643 	    ("%s: VI_INIT_DONE already", __func__));
4644 
4645 	sysctl_ctx_init(&vi->ctx);
4646 	vi->flags |= VI_SYSCTL_CTX;
4647 
4648 	/*
4649 	 * Allocate tx/rx/fl queues for this VI.
4650 	 */
4651 	rc = t4_setup_vi_queues(vi);
4652 	if (rc != 0)
4653 		goto done;	/* error message displayed already */
4654 
4655 	/*
4656 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
4657 	 */
4658 	if (vi->nrxq > vi->rss_size) {
4659 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
4660 		    "some queues will never receive traffic.\n", vi->nrxq,
4661 		    vi->rss_size);
4662 	} else if (vi->rss_size % vi->nrxq) {
4663 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
4664 		    "expect uneven traffic distribution.\n", vi->nrxq,
4665 		    vi->rss_size);
4666 	}
4667 #ifdef RSS
4668 	if (vi->nrxq != nbuckets) {
4669 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
4670 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
4671 	}
4672 #endif
4673 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
4674 	for (i = 0; i < vi->rss_size;) {
4675 #ifdef RSS
4676 		j = rss_get_indirection_to_bucket(i);
4677 		j %= vi->nrxq;
4678 		rxq = &sc->sge.rxq[vi->first_rxq + j];
4679 		rss[i++] = rxq->iq.abs_id;
4680 #else
4681 		for_each_rxq(vi, j, rxq) {
4682 			rss[i++] = rxq->iq.abs_id;
4683 			if (i == vi->rss_size)
4684 				break;
4685 		}
4686 #endif
4687 	}
4688 
4689 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
4690 	    vi->rss_size);
4691 	if (rc != 0) {
4692 		if_printf(ifp, "rss_config failed: %d\n", rc);
4693 		goto done;
4694 	}
4695 
4696 #ifdef RSS
4697 	hashen = hashconfig_to_hashen(hashconfig);
4698 
4699 	/*
4700 	 * We may have had to enable some hashes even though the global config
4701 	 * wants them disabled.  This is a potential problem that must be
4702 	 * reported to the user.
4703 	 */
4704 	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
4705 
4706 	/*
4707 	 * If we consider only the supported hash types, then the enabled hashes
4708 	 * are a superset of the requested hashes.  In other words, there cannot
4709 	 * be any supported hash that was requested but not enabled, but there
4710 	 * can be hashes that were not requested but had to be enabled.
4711 	 */
4712 	extra &= SUPPORTED_RSS_HASHTYPES;
4713 	MPASS((extra & hashconfig) == 0);
4714 
4715 	if (extra) {
4716 		if_printf(ifp,
4717 		    "global RSS config (0x%x) cannot be accommodated.\n",
4718 		    hashconfig);
4719 	}
4720 	if (extra & RSS_HASHTYPE_RSS_IPV4)
4721 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
4722 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
4723 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
4724 	if (extra & RSS_HASHTYPE_RSS_IPV6)
4725 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
4726 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
4727 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
4728 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
4729 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
4730 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
4731 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
4732 #else
4733 	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
4734 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
4735 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4736 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
4737 #endif
4738 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
4739 	if (rc != 0) {
4740 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
4741 		goto done;
4742 	}
4743 
4744 	vi->rss = rss;
4745 	vi->flags |= VI_INIT_DONE;
4746 done:
4747 	if (rc != 0)
4748 		vi_full_uninit(vi);
4749 
4750 	return (rc);
4751 }
4752 
4753 /*
4754  * Idempotent.
4755  */
4756 int
4757 vi_full_uninit(struct vi_info *vi)
4758 {
4759 	struct port_info *pi = vi->pi;
4760 	struct adapter *sc = pi->adapter;
4761 	int i;
4762 	struct sge_rxq *rxq;
4763 	struct sge_txq *txq;
4764 #ifdef TCP_OFFLOAD
4765 	struct sge_ofld_rxq *ofld_rxq;
4766 	struct sge_wrq *ofld_txq;
4767 #endif
4768 
4769 	if (vi->flags & VI_INIT_DONE) {
4770 
4771 		/* Need to quiesce queues.  */
4772 
4773 		/* XXX: Only for the first VI? */
4774 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
4775 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
4776 
4777 		for_each_txq(vi, i, txq) {
4778 			quiesce_txq(sc, txq);
4779 		}
4780 
4781 #ifdef TCP_OFFLOAD
4782 		for_each_ofld_txq(vi, i, ofld_txq) {
4783 			quiesce_wrq(sc, ofld_txq);
4784 		}
4785 #endif
4786 
4787 		for_each_rxq(vi, i, rxq) {
4788 			quiesce_iq(sc, &rxq->iq);
4789 			quiesce_fl(sc, &rxq->fl);
4790 		}
4791 
4792 #ifdef TCP_OFFLOAD
4793 		for_each_ofld_rxq(vi, i, ofld_rxq) {
4794 			quiesce_iq(sc, &ofld_rxq->iq);
4795 			quiesce_fl(sc, &ofld_rxq->fl);
4796 		}
4797 #endif
4798 		free(vi->rss, M_CXGBE);
4799 		free(vi->nm_rss, M_CXGBE);
4800 	}
4801 
4802 	t4_teardown_vi_queues(vi);
4803 	vi->flags &= ~VI_INIT_DONE;
4804 
4805 	return (0);
4806 }
4807 
4808 static void
4809 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
4810 {
4811 	struct sge_eq *eq = &txq->eq;
4812 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
4813 
4814 	(void) sc;	/* unused */
4815 
4816 #ifdef INVARIANTS
4817 	TXQ_LOCK(txq);
4818 	MPASS((eq->flags & EQ_ENABLED) == 0);
4819 	TXQ_UNLOCK(txq);
4820 #endif
4821 
4822 	/* Wait for the mp_ring to empty. */
4823 	while (!mp_ring_is_idle(txq->r)) {
4824 		mp_ring_check_drainage(txq->r, 0);
4825 		pause("rquiesce", 1);
4826 	}
4827 
4828 	/* Then wait for the hardware to finish. */
4829 	while (spg->cidx != htobe16(eq->pidx))
4830 		pause("equiesce", 1);
4831 
4832 	/* Finally, wait for the driver to reclaim all descriptors. */
4833 	while (eq->cidx != eq->pidx)
4834 		pause("dquiesce", 1);
4835 }
4836 
4837 static void
4838 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
4839 {
4840 
4841 	/* XXXTX */
4842 }
4843 
4844 static void
4845 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
4846 {
4847 	(void) sc;	/* unused */
4848 
4849 	/* Synchronize with the interrupt handler */
4850 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
4851 		pause("iqfree", 1);
4852 }
4853 
4854 static void
4855 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
4856 {
4857 	mtx_lock(&sc->sfl_lock);
4858 	FL_LOCK(fl);
4859 	fl->flags |= FL_DOOMED;
4860 	FL_UNLOCK(fl);
4861 	callout_stop(&sc->sfl_callout);
4862 	mtx_unlock(&sc->sfl_lock);
4863 
4864 	KASSERT((fl->flags & FL_STARVING) == 0,
4865 	    ("%s: still starving", __func__));
4866 }
4867 
4868 static int
4869 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
4870     driver_intr_t *handler, void *arg, char *name)
4871 {
4872 	int rc;
4873 
4874 	irq->rid = rid;
4875 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
4876 	    RF_SHAREABLE | RF_ACTIVE);
4877 	if (irq->res == NULL) {
4878 		device_printf(sc->dev,
4879 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
4880 		return (ENOMEM);
4881 	}
4882 
4883 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
4884 	    NULL, handler, arg, &irq->tag);
4885 	if (rc != 0) {
4886 		device_printf(sc->dev,
4887 		    "failed to setup interrupt for rid %d, name %s: %d\n",
4888 		    rid, name, rc);
4889 	} else if (name)
4890 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
4891 
4892 	return (rc);
4893 }
4894 
4895 static int
4896 t4_free_irq(struct adapter *sc, struct irq *irq)
4897 {
4898 	if (irq->tag)
4899 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
4900 	if (irq->res)
4901 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
4902 
4903 	bzero(irq, sizeof(*irq));
4904 
4905 	return (0);
4906 }
4907 
4908 static void
4909 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
4910 {
4911 
4912 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
4913 	t4_get_regs(sc, buf, regs->len);
4914 }
4915 
4916 #define	A_PL_INDIR_CMD	0x1f8
4917 
4918 #define	S_PL_AUTOINC	31
4919 #define	M_PL_AUTOINC	0x1U
4920 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
4921 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
4922 
4923 #define	S_PL_VFID	20
4924 #define	M_PL_VFID	0xffU
4925 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
4926 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
4927 
4928 #define	S_PL_ADDR	0
4929 #define	M_PL_ADDR	0xfffffU
4930 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
4931 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
4932 
4933 #define	A_PL_INDIR_DATA	0x1fc
4934 
4935 static uint64_t
4936 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
4937 {
4938 	u32 stats[2];
4939 
4940 	mtx_assert(&sc->reg_lock, MA_OWNED);
4941 	if (sc->flags & IS_VF) {
4942 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
4943 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
4944 	} else {
4945 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4946 		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4947 		    V_PL_ADDR(VF_MPS_REG(reg)));
4948 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
4949 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
4950 	}
4951 	return (((uint64_t)stats[1]) << 32 | stats[0]);
4952 }
4953 
4954 static void
4955 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
4956     struct fw_vi_stats_vf *stats)
4957 {
4958 
4959 #define GET_STAT(name) \
4960 	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
4961 
4962 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
4963 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
4964 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
4965 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
4966 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
4967 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
4968 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
4969 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
4970 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
4971 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
4972 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
4973 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
4974 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
4975 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
4976 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
4977 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
4978 
4979 #undef GET_STAT
4980 }
4981 
4982 static void
4983 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
4984 {
4985 	int reg;
4986 
4987 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4988 	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4989 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
4990 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
4991 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
4992 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
4993 }
4994 
4995 static void
4996 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
4997 {
4998 	struct timeval tv;
4999 	const struct timeval interval = {0, 250000};	/* 250ms */
5000 
5001 	if (!(vi->flags & VI_INIT_DONE))
5002 		return;
5003 
5004 	getmicrotime(&tv);
5005 	timevalsub(&tv, &interval);
5006 	if (timevalcmp(&tv, &vi->last_refreshed, <))
5007 		return;
5008 
5009 	mtx_lock(&sc->reg_lock);
5010 	t4_get_vi_stats(sc, vi->viid, &vi->stats);
5011 	getmicrotime(&vi->last_refreshed);
5012 	mtx_unlock(&sc->reg_lock);
5013 }
5014 
5015 static void
5016 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5017 {
5018 	u_int i, v, tnl_cong_drops, bg_map;
5019 	struct timeval tv;
5020 	const struct timeval interval = {0, 250000};	/* 250ms */
5021 
5022 	getmicrotime(&tv);
5023 	timevalsub(&tv, &interval);
5024 	if (timevalcmp(&tv, &pi->last_refreshed, <))
5025 		return;
5026 
5027 	tnl_cong_drops = 0;
5028 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5029 	bg_map = pi->mps_bg_map;
5030 	while (bg_map) {
5031 		i = ffs(bg_map) - 1;
5032 		mtx_lock(&sc->reg_lock);
5033 		t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5034 		    A_TP_MIB_TNL_CNG_DROP_0 + i);
5035 		mtx_unlock(&sc->reg_lock);
5036 		tnl_cong_drops += v;
5037 		bg_map &= ~(1 << i);
5038 	}
5039 	pi->tnl_cong_drops = tnl_cong_drops;
5040 	getmicrotime(&pi->last_refreshed);
5041 }
5042 
5043 static void
5044 cxgbe_tick(void *arg)
5045 {
5046 	struct port_info *pi = arg;
5047 	struct adapter *sc = pi->adapter;
5048 
5049 	PORT_LOCK_ASSERT_OWNED(pi);
5050 	cxgbe_refresh_stats(sc, pi);
5051 
5052 	callout_schedule(&pi->tick, hz);
5053 }
5054 
5055 void
5056 vi_tick(void *arg)
5057 {
5058 	struct vi_info *vi = arg;
5059 	struct adapter *sc = vi->pi->adapter;
5060 
5061 	vi_refresh_stats(sc, vi);
5062 
5063 	callout_schedule(&vi->tick, hz);
5064 }
5065 
5066 static void
5067 cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
5068 {
5069 	struct ifnet *vlan;
5070 
5071 	if (arg != ifp || ifp->if_type != IFT_ETHER)
5072 		return;
5073 
5074 	vlan = VLAN_DEVAT(ifp, vid);
5075 	VLAN_SETCOOKIE(vlan, ifp);
5076 }
5077 
5078 /*
5079  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5080  */
5081 static char *caps_decoder[] = {
5082 	"\20\001IPMI\002NCSI",				/* 0: NBM */
5083 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
5084 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
5085 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
5086 	    "\006HASHFILTER\007ETHOFLD",
5087 	"\20\001TOE",					/* 4: TOE */
5088 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
5089 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
5090 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5091 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5092 	    "\007T10DIF"
5093 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5094 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
5095 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
5096 		    "\004PO_INITIATOR\005PO_TARGET",
5097 };
5098 
5099 void
5100 t4_sysctls(struct adapter *sc)
5101 {
5102 	struct sysctl_ctx_list *ctx;
5103 	struct sysctl_oid *oid;
5104 	struct sysctl_oid_list *children, *c0;
5105 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5106 
5107 	ctx = device_get_sysctl_ctx(sc->dev);
5108 
5109 	/*
5110 	 * dev.t4nex.X.
5111 	 */
5112 	oid = device_get_sysctl_tree(sc->dev);
5113 	c0 = children = SYSCTL_CHILDREN(oid);
5114 
5115 	sc->sc_do_rxcopy = 1;
5116 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5117 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5118 
5119 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5120 	    sc->params.nports, "# of ports");
5121 
5122 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5123 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
5124 	    sysctl_bitfield, "A", "available doorbells");
5125 
5126 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5127 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
5128 
5129 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5130 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5131 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5132 	    "interrupt holdoff timer values (us)");
5133 
5134 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5135 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5136 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5137 	    "interrupt holdoff packet counter values");
5138 
5139 	t4_sge_sysctls(sc, ctx, children);
5140 
5141 	sc->lro_timeout = 100;
5142 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5143 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5144 
5145 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5146 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
5147 
5148 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5149 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5150 
5151 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5152 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5153 
5154 	if (sc->flags & IS_VF)
5155 		return;
5156 
5157 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5158 	    NULL, chip_rev(sc), "chip hardware revision");
5159 
5160 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5161 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5162 
5163 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5164 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5165 
5166 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5167 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5168 
5169 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5170 	    CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5171 
5172 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5173 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5174 
5175 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5176 	    sc->er_version, 0, "expansion ROM version");
5177 
5178 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5179 	    sc->bs_version, 0, "bootstrap firmware version");
5180 
5181 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5182 	    NULL, sc->params.scfg_vers, "serial config version");
5183 
5184 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5185 	    NULL, sc->params.vpd_vers, "VPD version");
5186 
5187 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5188 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5189 
5190 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5191 	    sc->cfcsum, "config file checksum");
5192 
5193 #define SYSCTL_CAP(name, n, text) \
5194 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5195 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
5196 	    sysctl_bitfield, "A", "available " text " capabilities")
5197 
5198 	SYSCTL_CAP(nbmcaps, 0, "NBM");
5199 	SYSCTL_CAP(linkcaps, 1, "link");
5200 	SYSCTL_CAP(switchcaps, 2, "switch");
5201 	SYSCTL_CAP(niccaps, 3, "NIC");
5202 	SYSCTL_CAP(toecaps, 4, "TCP offload");
5203 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
5204 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5205 	SYSCTL_CAP(cryptocaps, 7, "crypto");
5206 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
5207 #undef SYSCTL_CAP
5208 
5209 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5210 	    NULL, sc->tids.nftids, "number of filters");
5211 
5212 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5213 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5214 	    "chip temperature (in Celsius)");
5215 
5216 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
5217 	    &sc->params.core_vdd, 0, "core Vdd (in mV)");
5218 
5219 #ifdef SBUF_DRAIN
5220 	/*
5221 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
5222 	 */
5223 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5224 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5225 	    "logs and miscellaneous information");
5226 	children = SYSCTL_CHILDREN(oid);
5227 
5228 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
5229 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5230 	    sysctl_cctrl, "A", "congestion control");
5231 
5232 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
5233 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5234 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
5235 
5236 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
5237 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
5238 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
5239 
5240 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
5241 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
5242 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
5243 
5244 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
5245 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
5246 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
5247 
5248 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
5249 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
5250 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
5251 
5252 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
5253 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
5254 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
5255 
5256 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
5257 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5258 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
5259 	    "A", "CIM logic analyzer");
5260 
5261 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
5262 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5263 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
5264 
5265 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
5266 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
5267 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
5268 
5269 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
5270 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
5271 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
5272 
5273 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
5274 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
5275 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
5276 
5277 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5278 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5279 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5280 
5281 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5282 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5283 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5284 
5285 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5286 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5287 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5288 
5289 	if (chip_id(sc) > CHELSIO_T4) {
5290 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5291 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5292 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5293 
5294 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5295 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5296 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5297 	}
5298 
5299 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5300 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5301 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5302 
5303 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5304 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5305 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
5306 
5307 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5308 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5309 	    sysctl_cpl_stats, "A", "CPL statistics");
5310 
5311 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5312 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5313 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5314 
5315 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5316 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5317 	    sysctl_devlog, "A", "firmware's device log");
5318 
5319 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5320 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5321 	    sysctl_fcoe_stats, "A", "FCoE statistics");
5322 
5323 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5324 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5325 	    sysctl_hw_sched, "A", "hardware scheduler ");
5326 
5327 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5328 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5329 	    sysctl_l2t, "A", "hardware L2 table");
5330 
5331 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5332 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5333 	    sysctl_lb_stats, "A", "loopback statistics");
5334 
5335 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5336 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5337 	    sysctl_meminfo, "A", "memory regions");
5338 
5339 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5340 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5341 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5342 	    "A", "MPS TCAM entries");
5343 
5344 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5345 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5346 	    sysctl_path_mtus, "A", "path MTUs");
5347 
5348 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5349 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5350 	    sysctl_pm_stats, "A", "PM statistics");
5351 
5352 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5353 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5354 	    sysctl_rdma_stats, "A", "RDMA statistics");
5355 
5356 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5357 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5358 	    sysctl_tcp_stats, "A", "TCP statistics");
5359 
5360 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5361 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5362 	    sysctl_tids, "A", "TID information");
5363 
5364 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5365 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5366 	    sysctl_tp_err_stats, "A", "TP error statistics");
5367 
5368 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5369 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5370 	    "TP logic analyzer event capture mask");
5371 
5372 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5373 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5374 	    sysctl_tp_la, "A", "TP logic analyzer");
5375 
5376 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5377 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5378 	    sysctl_tx_rate, "A", "Tx rate");
5379 
5380 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5381 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5382 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5383 
5384 	if (chip_id(sc) >= CHELSIO_T5) {
5385 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5386 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5387 		    sysctl_wcwr_stats, "A", "write combined work requests");
5388 	}
5389 #endif
5390 
5391 #ifdef TCP_OFFLOAD
5392 	if (is_offload(sc)) {
5393 		int i;
5394 		char s[4];
5395 
5396 		/*
5397 		 * dev.t4nex.X.toe.
5398 		 */
5399 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5400 		    NULL, "TOE parameters");
5401 		children = SYSCTL_CHILDREN(oid);
5402 
5403 		sc->tt.cong_algorithm = -1;
5404 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
5405 		    CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
5406 		    "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
5407 		    "3 = highspeed)");
5408 
5409 		sc->tt.sndbuf = 256 * 1024;
5410 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5411 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
5412 
5413 		sc->tt.ddp = 0;
5414 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5415 		    &sc->tt.ddp, 0, "DDP allowed");
5416 
5417 		sc->tt.rx_coalesce = 1;
5418 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5419 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5420 
5421 		sc->tt.tx_align = 1;
5422 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5423 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5424 
5425 		sc->tt.tx_zcopy = 0;
5426 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
5427 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
5428 		    "Enable zero-copy aio_write(2)");
5429 
5430 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
5431 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
5432 		    "TP timer tick (us)");
5433 
5434 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
5435 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
5436 		    "TCP timestamp tick (us)");
5437 
5438 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5439 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5440 		    "DACK tick (us)");
5441 
5442 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5443 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5444 		    "IU", "DACK timer (us)");
5445 
5446 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5447 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5448 		    sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
5449 
5450 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5451 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5452 		    sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
5453 
5454 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5455 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5456 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
5457 
5458 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5459 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5460 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
5461 
5462 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5463 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5464 		    sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
5465 
5466 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
5467 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
5468 		    sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
5469 
5470 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
5471 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
5472 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
5473 
5474 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
5475 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
5476 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
5477 
5478 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
5479 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
5480 		    sysctl_tp_shift_cnt, "IU",
5481 		    "Number of SYN retransmissions before abort");
5482 
5483 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
5484 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
5485 		    sysctl_tp_shift_cnt, "IU",
5486 		    "Number of retransmissions before abort");
5487 
5488 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
5489 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
5490 		    sysctl_tp_shift_cnt, "IU",
5491 		    "Number of keepalive probes before abort");
5492 
5493 		oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
5494 		    CTLFLAG_RD, NULL, "TOE retransmit backoffs");
5495 		children = SYSCTL_CHILDREN(oid);
5496 		for (i = 0; i < 16; i++) {
5497 			snprintf(s, sizeof(s), "%u", i);
5498 			SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
5499 			    CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
5500 			    "IU", "TOE retransmit backoff");
5501 		}
5502 	}
5503 #endif
5504 }
5505 
5506 void
5507 vi_sysctls(struct vi_info *vi)
5508 {
5509 	struct sysctl_ctx_list *ctx;
5510 	struct sysctl_oid *oid;
5511 	struct sysctl_oid_list *children;
5512 
5513 	ctx = device_get_sysctl_ctx(vi->dev);
5514 
5515 	/*
5516 	 * dev.v?(cxgbe|cxl).X.
5517 	 */
5518 	oid = device_get_sysctl_tree(vi->dev);
5519 	children = SYSCTL_CHILDREN(oid);
5520 
5521 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
5522 	    vi->viid, "VI identifer");
5523 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
5524 	    &vi->nrxq, 0, "# of rx queues");
5525 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5526 	    &vi->ntxq, 0, "# of tx queues");
5527 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5528 	    &vi->first_rxq, 0, "index of first rx queue");
5529 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5530 	    &vi->first_txq, 0, "index of first tx queue");
5531 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
5532 	    vi->rss_size, "size of RSS indirection table");
5533 
5534 	if (IS_MAIN_VI(vi)) {
5535 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
5536 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
5537 		    "Reserve queue 0 for non-flowid packets");
5538 	}
5539 
5540 #ifdef TCP_OFFLOAD
5541 	if (vi->nofldrxq != 0) {
5542 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5543 		    &vi->nofldrxq, 0,
5544 		    "# of rx queues for offloaded TCP connections");
5545 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5546 		    &vi->nofldtxq, 0,
5547 		    "# of tx queues for offloaded TCP connections");
5548 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5549 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
5550 		    "index of first TOE rx queue");
5551 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5552 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
5553 		    "index of first TOE tx queue");
5554 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
5555 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5556 		    sysctl_holdoff_tmr_idx_ofld, "I",
5557 		    "holdoff timer index for TOE queues");
5558 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
5559 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5560 		    sysctl_holdoff_pktc_idx_ofld, "I",
5561 		    "holdoff packet counter index for TOE queues");
5562 	}
5563 #endif
5564 #ifdef DEV_NETMAP
5565 	if (vi->nnmrxq != 0) {
5566 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5567 		    &vi->nnmrxq, 0, "# of netmap rx queues");
5568 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5569 		    &vi->nnmtxq, 0, "# of netmap tx queues");
5570 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5571 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
5572 		    "index of first netmap rx queue");
5573 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
5574 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
5575 		    "index of first netmap tx queue");
5576 	}
5577 #endif
5578 
5579 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
5580 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
5581 	    "holdoff timer index");
5582 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
5583 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
5584 	    "holdoff packet counter index");
5585 
5586 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
5587 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
5588 	    "rx queue size");
5589 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
5590 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
5591 	    "tx queue size");
5592 }
5593 
5594 static void
5595 cxgbe_sysctls(struct port_info *pi)
5596 {
5597 	struct sysctl_ctx_list *ctx;
5598 	struct sysctl_oid *oid;
5599 	struct sysctl_oid_list *children, *children2;
5600 	struct adapter *sc = pi->adapter;
5601 	int i;
5602 	char name[16];
5603 
5604 	ctx = device_get_sysctl_ctx(pi->dev);
5605 
5606 	/*
5607 	 * dev.cxgbe.X.
5608 	 */
5609 	oid = device_get_sysctl_tree(pi->dev);
5610 	children = SYSCTL_CHILDREN(oid);
5611 
5612 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
5613 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
5614 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
5615 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
5616 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
5617 		    "PHY temperature (in Celsius)");
5618 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
5619 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
5620 		    "PHY firmware version");
5621 	}
5622 
5623 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
5624 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
5625 	    "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
5626 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
5627 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
5628 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
5629 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
5630 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
5631 	    "autonegotiation (-1 = not supported)");
5632 
5633 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
5634 	    port_top_speed(pi), "max speed (in Gbps)");
5635 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
5636 	    pi->mps_bg_map, "MPS buffer group map");
5637 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
5638 	    NULL, pi->rx_e_chan_map, "TP rx e-channel map");
5639 
5640 	if (sc->flags & IS_VF)
5641 		return;
5642 
5643 	/*
5644 	 * dev.(cxgbe|cxl).X.tc.
5645 	 */
5646 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
5647 	    "Tx scheduler traffic classes (cl_rl)");
5648 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
5649 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
5650 
5651 		snprintf(name, sizeof(name), "%d", i);
5652 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
5653 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
5654 		    "traffic class"));
5655 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
5656 		    &tc->flags, 0, "flags");
5657 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
5658 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
5659 #ifdef SBUF_DRAIN
5660 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
5661 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
5662 		    sysctl_tc_params, "A", "traffic class parameters");
5663 #endif
5664 	}
5665 
5666 	/*
5667 	 * dev.cxgbe.X.stats.
5668 	 */
5669 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
5670 	    NULL, "port statistics");
5671 	children = SYSCTL_CHILDREN(oid);
5672 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
5673 	    &pi->tx_parse_error, 0,
5674 	    "# of tx packets with invalid length or # of segments");
5675 
5676 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
5677 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
5678 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
5679 	    sysctl_handle_t4_reg64, "QU", desc)
5680 
5681 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
5682 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
5683 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
5684 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
5685 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
5686 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
5687 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
5688 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
5689 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
5690 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
5691 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
5692 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
5693 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
5694 	    "# of tx frames in this range",
5695 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
5696 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
5697 	    "# of tx frames in this range",
5698 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
5699 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
5700 	    "# of tx frames in this range",
5701 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
5702 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
5703 	    "# of tx frames in this range",
5704 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
5705 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
5706 	    "# of tx frames in this range",
5707 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
5708 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
5709 	    "# of tx frames in this range",
5710 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
5711 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
5712 	    "# of tx frames in this range",
5713 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
5714 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
5715 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
5716 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
5717 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
5718 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
5719 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
5720 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5721 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5722 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5723 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5724 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5725 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5726 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5727 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5728 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5729 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5730 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5731 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5732 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5733 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5734 
5735 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5736 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5737 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5738 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5739 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5740 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5741 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5742 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5743 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5744 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5745 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5746 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5747 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5748 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5749 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5750 	    "# of frames received with bad FCS",
5751 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5752 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5753 	    "# of frames received with length error",
5754 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5755 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5756 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5757 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5758 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5759 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5760 	    "# of rx frames in this range",
5761 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5762 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5763 	    "# of rx frames in this range",
5764 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5765 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5766 	    "# of rx frames in this range",
5767 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5768 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5769 	    "# of rx frames in this range",
5770 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5771 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5772 	    "# of rx frames in this range",
5773 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5774 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5775 	    "# of rx frames in this range",
5776 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5777 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5778 	    "# of rx frames in this range",
5779 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5780 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5781 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5782 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5783 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5784 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5785 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5786 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5787 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5788 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5789 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5790 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5791 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5792 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5793 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5794 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5795 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5796 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5797 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5798 
5799 #undef SYSCTL_ADD_T4_REG64
5800 
5801 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5802 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5803 	    &pi->stats.name, desc)
5804 
5805 	/* We get these from port_stats and they may be stale by up to 1s */
5806 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5807 	    "# drops due to buffer-group 0 overflows");
5808 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5809 	    "# drops due to buffer-group 1 overflows");
5810 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5811 	    "# drops due to buffer-group 2 overflows");
5812 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5813 	    "# drops due to buffer-group 3 overflows");
5814 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5815 	    "# of buffer-group 0 truncated packets");
5816 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5817 	    "# of buffer-group 1 truncated packets");
5818 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5819 	    "# of buffer-group 2 truncated packets");
5820 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5821 	    "# of buffer-group 3 truncated packets");
5822 
5823 #undef SYSCTL_ADD_T4_PORTSTAT
5824 }
5825 
5826 static int
5827 sysctl_int_array(SYSCTL_HANDLER_ARGS)
5828 {
5829 	int rc, *i, space = 0;
5830 	struct sbuf sb;
5831 
5832 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
5833 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
5834 		if (space)
5835 			sbuf_printf(&sb, " ");
5836 		sbuf_printf(&sb, "%d", *i);
5837 		space = 1;
5838 	}
5839 	rc = sbuf_finish(&sb);
5840 	sbuf_delete(&sb);
5841 	return (rc);
5842 }
5843 
5844 static int
5845 sysctl_bitfield(SYSCTL_HANDLER_ARGS)
5846 {
5847 	int rc;
5848 	struct sbuf *sb;
5849 
5850 	rc = sysctl_wire_old_buffer(req, 0);
5851 	if (rc != 0)
5852 		return(rc);
5853 
5854 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5855 	if (sb == NULL)
5856 		return (ENOMEM);
5857 
5858 	sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
5859 	rc = sbuf_finish(sb);
5860 	sbuf_delete(sb);
5861 
5862 	return (rc);
5863 }
5864 
5865 static int
5866 sysctl_btphy(SYSCTL_HANDLER_ARGS)
5867 {
5868 	struct port_info *pi = arg1;
5869 	int op = arg2;
5870 	struct adapter *sc = pi->adapter;
5871 	u_int v;
5872 	int rc;
5873 
5874 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
5875 	if (rc)
5876 		return (rc);
5877 	/* XXX: magic numbers */
5878 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
5879 	    &v);
5880 	end_synchronized_op(sc, 0);
5881 	if (rc)
5882 		return (rc);
5883 	if (op == 0)
5884 		v /= 256;
5885 
5886 	rc = sysctl_handle_int(oidp, &v, 0, req);
5887 	return (rc);
5888 }
5889 
5890 static int
5891 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
5892 {
5893 	struct vi_info *vi = arg1;
5894 	int rc, val;
5895 
5896 	val = vi->rsrv_noflowq;
5897 	rc = sysctl_handle_int(oidp, &val, 0, req);
5898 	if (rc != 0 || req->newptr == NULL)
5899 		return (rc);
5900 
5901 	if ((val >= 1) && (vi->ntxq > 1))
5902 		vi->rsrv_noflowq = 1;
5903 	else
5904 		vi->rsrv_noflowq = 0;
5905 
5906 	return (rc);
5907 }
5908 
5909 static int
5910 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
5911 {
5912 	struct vi_info *vi = arg1;
5913 	struct adapter *sc = vi->pi->adapter;
5914 	int idx, rc, i;
5915 	struct sge_rxq *rxq;
5916 	uint8_t v;
5917 
5918 	idx = vi->tmr_idx;
5919 
5920 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5921 	if (rc != 0 || req->newptr == NULL)
5922 		return (rc);
5923 
5924 	if (idx < 0 || idx >= SGE_NTIMERS)
5925 		return (EINVAL);
5926 
5927 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5928 	    "t4tmr");
5929 	if (rc)
5930 		return (rc);
5931 
5932 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
5933 	for_each_rxq(vi, i, rxq) {
5934 #ifdef atomic_store_rel_8
5935 		atomic_store_rel_8(&rxq->iq.intr_params, v);
5936 #else
5937 		rxq->iq.intr_params = v;
5938 #endif
5939 	}
5940 	vi->tmr_idx = idx;
5941 
5942 	end_synchronized_op(sc, LOCK_HELD);
5943 	return (0);
5944 }
5945 
5946 static int
5947 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
5948 {
5949 	struct vi_info *vi = arg1;
5950 	struct adapter *sc = vi->pi->adapter;
5951 	int idx, rc;
5952 
5953 	idx = vi->pktc_idx;
5954 
5955 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5956 	if (rc != 0 || req->newptr == NULL)
5957 		return (rc);
5958 
5959 	if (idx < -1 || idx >= SGE_NCOUNTERS)
5960 		return (EINVAL);
5961 
5962 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5963 	    "t4pktc");
5964 	if (rc)
5965 		return (rc);
5966 
5967 	if (vi->flags & VI_INIT_DONE)
5968 		rc = EBUSY; /* cannot be changed once the queues are created */
5969 	else
5970 		vi->pktc_idx = idx;
5971 
5972 	end_synchronized_op(sc, LOCK_HELD);
5973 	return (rc);
5974 }
5975 
5976 static int
5977 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
5978 {
5979 	struct vi_info *vi = arg1;
5980 	struct adapter *sc = vi->pi->adapter;
5981 	int qsize, rc;
5982 
5983 	qsize = vi->qsize_rxq;
5984 
5985 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5986 	if (rc != 0 || req->newptr == NULL)
5987 		return (rc);
5988 
5989 	if (qsize < 128 || (qsize & 7))
5990 		return (EINVAL);
5991 
5992 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5993 	    "t4rxqs");
5994 	if (rc)
5995 		return (rc);
5996 
5997 	if (vi->flags & VI_INIT_DONE)
5998 		rc = EBUSY; /* cannot be changed once the queues are created */
5999 	else
6000 		vi->qsize_rxq = qsize;
6001 
6002 	end_synchronized_op(sc, LOCK_HELD);
6003 	return (rc);
6004 }
6005 
6006 static int
6007 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
6008 {
6009 	struct vi_info *vi = arg1;
6010 	struct adapter *sc = vi->pi->adapter;
6011 	int qsize, rc;
6012 
6013 	qsize = vi->qsize_txq;
6014 
6015 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6016 	if (rc != 0 || req->newptr == NULL)
6017 		return (rc);
6018 
6019 	if (qsize < 128 || qsize > 65536)
6020 		return (EINVAL);
6021 
6022 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6023 	    "t4txqs");
6024 	if (rc)
6025 		return (rc);
6026 
6027 	if (vi->flags & VI_INIT_DONE)
6028 		rc = EBUSY; /* cannot be changed once the queues are created */
6029 	else
6030 		vi->qsize_txq = qsize;
6031 
6032 	end_synchronized_op(sc, LOCK_HELD);
6033 	return (rc);
6034 }
6035 
6036 static int
6037 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
6038 {
6039 	struct port_info *pi = arg1;
6040 	struct adapter *sc = pi->adapter;
6041 	struct link_config *lc = &pi->link_cfg;
6042 	int rc;
6043 
6044 	if (req->newptr == NULL) {
6045 		struct sbuf *sb;
6046 		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
6047 
6048 		rc = sysctl_wire_old_buffer(req, 0);
6049 		if (rc != 0)
6050 			return(rc);
6051 
6052 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6053 		if (sb == NULL)
6054 			return (ENOMEM);
6055 
6056 		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
6057 		rc = sbuf_finish(sb);
6058 		sbuf_delete(sb);
6059 	} else {
6060 		char s[2];
6061 		int n;
6062 
6063 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
6064 		s[1] = 0;
6065 
6066 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6067 		if (rc != 0)
6068 			return(rc);
6069 
6070 		if (s[1] != 0)
6071 			return (EINVAL);
6072 		if (s[0] < '0' || s[0] > '9')
6073 			return (EINVAL);	/* not a number */
6074 		n = s[0] - '0';
6075 		if (n & ~(PAUSE_TX | PAUSE_RX))
6076 			return (EINVAL);	/* some other bit is set too */
6077 
6078 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6079 		    "t4PAUSE");
6080 		if (rc)
6081 			return (rc);
6082 		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
6083 			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
6084 			lc->requested_fc |= n;
6085 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6086 			if (rc == 0) {
6087 				lc->fc = lc->requested_fc;
6088 			}
6089 		}
6090 		end_synchronized_op(sc, 0);
6091 	}
6092 
6093 	return (rc);
6094 }
6095 
6096 static int
6097 sysctl_fec(SYSCTL_HANDLER_ARGS)
6098 {
6099 	struct port_info *pi = arg1;
6100 	struct adapter *sc = pi->adapter;
6101 	struct link_config *lc = &pi->link_cfg;
6102 	int rc;
6103 
6104 	if (req->newptr == NULL) {
6105 		struct sbuf *sb;
6106 		static char *bits = "\20\1RS\2BASER_RS\3RESERVED";
6107 
6108 		rc = sysctl_wire_old_buffer(req, 0);
6109 		if (rc != 0)
6110 			return(rc);
6111 
6112 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6113 		if (sb == NULL)
6114 			return (ENOMEM);
6115 
6116 		sbuf_printf(sb, "%b", lc->fec & M_FW_PORT_CAP_FEC, bits);
6117 		rc = sbuf_finish(sb);
6118 		sbuf_delete(sb);
6119 	} else {
6120 		char s[2];
6121 		int n;
6122 
6123 		s[0] = '0' + (lc->requested_fec & M_FW_PORT_CAP_FEC);
6124 		s[1] = 0;
6125 
6126 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6127 		if (rc != 0)
6128 			return(rc);
6129 
6130 		if (s[1] != 0)
6131 			return (EINVAL);
6132 		if (s[0] < '0' || s[0] > '9')
6133 			return (EINVAL);	/* not a number */
6134 		n = s[0] - '0';
6135 		if (n & ~M_FW_PORT_CAP_FEC)
6136 			return (EINVAL);	/* some other bit is set too */
6137 
6138 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6139 		    "t4fec");
6140 		if (rc)
6141 			return (rc);
6142 		if ((lc->requested_fec & M_FW_PORT_CAP_FEC) != n) {
6143 			lc->requested_fec = n &
6144 			    G_FW_PORT_CAP_FEC(lc->supported);
6145 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6146 			if (rc == 0) {
6147 				lc->fec = lc->requested_fec;
6148 			}
6149 		}
6150 		end_synchronized_op(sc, 0);
6151 	}
6152 
6153 	return (rc);
6154 }
6155 
6156 static int
6157 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
6158 {
6159 	struct port_info *pi = arg1;
6160 	struct adapter *sc = pi->adapter;
6161 	struct link_config *lc = &pi->link_cfg;
6162 	int rc, val, old;
6163 
6164 	if (lc->supported & FW_PORT_CAP_ANEG)
6165 		val = lc->requested_aneg == AUTONEG_ENABLE ? 1 : 0;
6166 	else
6167 		val = -1;
6168 	rc = sysctl_handle_int(oidp, &val, 0, req);
6169 	if (rc != 0 || req->newptr == NULL)
6170 		return (rc);
6171 	if ((lc->supported & FW_PORT_CAP_ANEG) == 0)
6172 		return (ENOTSUP);
6173 
6174 	if (val == 0)
6175 		val = AUTONEG_DISABLE;
6176 	else if (val == 1)
6177 		val = AUTONEG_ENABLE;
6178 	else
6179 		return (EINVAL);
6180 	if (lc->requested_aneg == val)
6181 		return (0);	/* no change */
6182 
6183 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6184 	    "t4aneg");
6185 	if (rc)
6186 		return (rc);
6187 	old = lc->requested_aneg;
6188 	lc->requested_aneg = val;
6189 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6190 	if (rc != 0)
6191 		lc->requested_aneg = old;
6192 	end_synchronized_op(sc, 0);
6193 	return (rc);
6194 }
6195 
6196 static int
6197 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
6198 {
6199 	struct adapter *sc = arg1;
6200 	int reg = arg2;
6201 	uint64_t val;
6202 
6203 	val = t4_read_reg64(sc, reg);
6204 
6205 	return (sysctl_handle_64(oidp, &val, 0, req));
6206 }
6207 
6208 static int
6209 sysctl_temperature(SYSCTL_HANDLER_ARGS)
6210 {
6211 	struct adapter *sc = arg1;
6212 	int rc, t;
6213 	uint32_t param, val;
6214 
6215 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
6216 	if (rc)
6217 		return (rc);
6218 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6219 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
6220 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
6221 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6222 	end_synchronized_op(sc, 0);
6223 	if (rc)
6224 		return (rc);
6225 
6226 	/* unknown is returned as 0 but we display -1 in that case */
6227 	t = val == 0 ? -1 : val;
6228 
6229 	rc = sysctl_handle_int(oidp, &t, 0, req);
6230 	return (rc);
6231 }
6232 
6233 #ifdef SBUF_DRAIN
6234 static int
6235 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
6236 {
6237 	struct adapter *sc = arg1;
6238 	struct sbuf *sb;
6239 	int rc, i;
6240 	uint16_t incr[NMTUS][NCCTRL_WIN];
6241 	static const char *dec_fac[] = {
6242 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
6243 		"0.9375"
6244 	};
6245 
6246 	rc = sysctl_wire_old_buffer(req, 0);
6247 	if (rc != 0)
6248 		return (rc);
6249 
6250 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6251 	if (sb == NULL)
6252 		return (ENOMEM);
6253 
6254 	t4_read_cong_tbl(sc, incr);
6255 
6256 	for (i = 0; i < NCCTRL_WIN; ++i) {
6257 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
6258 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
6259 		    incr[5][i], incr[6][i], incr[7][i]);
6260 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
6261 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
6262 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
6263 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
6264 	}
6265 
6266 	rc = sbuf_finish(sb);
6267 	sbuf_delete(sb);
6268 
6269 	return (rc);
6270 }
6271 
6272 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
6273 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
6274 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
6275 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
6276 };
6277 
6278 static int
6279 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
6280 {
6281 	struct adapter *sc = arg1;
6282 	struct sbuf *sb;
6283 	int rc, i, n, qid = arg2;
6284 	uint32_t *buf, *p;
6285 	char *qtype;
6286 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
6287 
6288 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
6289 	    ("%s: bad qid %d\n", __func__, qid));
6290 
6291 	if (qid < CIM_NUM_IBQ) {
6292 		/* inbound queue */
6293 		qtype = "IBQ";
6294 		n = 4 * CIM_IBQ_SIZE;
6295 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6296 		rc = t4_read_cim_ibq(sc, qid, buf, n);
6297 	} else {
6298 		/* outbound queue */
6299 		qtype = "OBQ";
6300 		qid -= CIM_NUM_IBQ;
6301 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
6302 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6303 		rc = t4_read_cim_obq(sc, qid, buf, n);
6304 	}
6305 
6306 	if (rc < 0) {
6307 		rc = -rc;
6308 		goto done;
6309 	}
6310 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
6311 
6312 	rc = sysctl_wire_old_buffer(req, 0);
6313 	if (rc != 0)
6314 		goto done;
6315 
6316 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6317 	if (sb == NULL) {
6318 		rc = ENOMEM;
6319 		goto done;
6320 	}
6321 
6322 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
6323 	for (i = 0, p = buf; i < n; i += 16, p += 4)
6324 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
6325 		    p[2], p[3]);
6326 
6327 	rc = sbuf_finish(sb);
6328 	sbuf_delete(sb);
6329 done:
6330 	free(buf, M_CXGBE);
6331 	return (rc);
6332 }
6333 
6334 static int
6335 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
6336 {
6337 	struct adapter *sc = arg1;
6338 	u_int cfg;
6339 	struct sbuf *sb;
6340 	uint32_t *buf, *p;
6341 	int rc;
6342 
6343 	MPASS(chip_id(sc) <= CHELSIO_T5);
6344 
6345 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6346 	if (rc != 0)
6347 		return (rc);
6348 
6349 	rc = sysctl_wire_old_buffer(req, 0);
6350 	if (rc != 0)
6351 		return (rc);
6352 
6353 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6354 	if (sb == NULL)
6355 		return (ENOMEM);
6356 
6357 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6358 	    M_ZERO | M_WAITOK);
6359 
6360 	rc = -t4_cim_read_la(sc, buf, NULL);
6361 	if (rc != 0)
6362 		goto done;
6363 
6364 	sbuf_printf(sb, "Status   Data      PC%s",
6365 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6366 	    "     LS0Stat  LS0Addr             LS0Data");
6367 
6368 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
6369 		if (cfg & F_UPDBGLACAPTPCONLY) {
6370 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
6371 			    p[6], p[7]);
6372 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
6373 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
6374 			    p[4] & 0xff, p[5] >> 8);
6375 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
6376 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6377 			    p[1] & 0xf, p[2] >> 4);
6378 		} else {
6379 			sbuf_printf(sb,
6380 			    "\n  %02x   %x%07x %x%07x %08x %08x "
6381 			    "%08x%08x%08x%08x",
6382 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6383 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
6384 			    p[6], p[7]);
6385 		}
6386 	}
6387 
6388 	rc = sbuf_finish(sb);
6389 	sbuf_delete(sb);
6390 done:
6391 	free(buf, M_CXGBE);
6392 	return (rc);
6393 }
6394 
6395 static int
6396 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
6397 {
6398 	struct adapter *sc = arg1;
6399 	u_int cfg;
6400 	struct sbuf *sb;
6401 	uint32_t *buf, *p;
6402 	int rc;
6403 
6404 	MPASS(chip_id(sc) > CHELSIO_T5);
6405 
6406 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6407 	if (rc != 0)
6408 		return (rc);
6409 
6410 	rc = sysctl_wire_old_buffer(req, 0);
6411 	if (rc != 0)
6412 		return (rc);
6413 
6414 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6415 	if (sb == NULL)
6416 		return (ENOMEM);
6417 
6418 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6419 	    M_ZERO | M_WAITOK);
6420 
6421 	rc = -t4_cim_read_la(sc, buf, NULL);
6422 	if (rc != 0)
6423 		goto done;
6424 
6425 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
6426 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6427 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
6428 
6429 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
6430 		if (cfg & F_UPDBGLACAPTPCONLY) {
6431 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
6432 			    p[3] & 0xff, p[2], p[1], p[0]);
6433 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
6434 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
6435 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
6436 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
6437 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
6438 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
6439 			    p[6] >> 16);
6440 		} else {
6441 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
6442 			    "%08x %08x %08x %08x %08x %08x",
6443 			    (p[9] >> 16) & 0xff,
6444 			    p[9] & 0xffff, p[8] >> 16,
6445 			    p[8] & 0xffff, p[7] >> 16,
6446 			    p[7] & 0xffff, p[6] >> 16,
6447 			    p[2], p[1], p[0], p[5], p[4], p[3]);
6448 		}
6449 	}
6450 
6451 	rc = sbuf_finish(sb);
6452 	sbuf_delete(sb);
6453 done:
6454 	free(buf, M_CXGBE);
6455 	return (rc);
6456 }
6457 
6458 static int
6459 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
6460 {
6461 	struct adapter *sc = arg1;
6462 	u_int i;
6463 	struct sbuf *sb;
6464 	uint32_t *buf, *p;
6465 	int rc;
6466 
6467 	rc = sysctl_wire_old_buffer(req, 0);
6468 	if (rc != 0)
6469 		return (rc);
6470 
6471 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6472 	if (sb == NULL)
6473 		return (ENOMEM);
6474 
6475 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
6476 	    M_ZERO | M_WAITOK);
6477 
6478 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
6479 	p = buf;
6480 
6481 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6482 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
6483 		    p[1], p[0]);
6484 	}
6485 
6486 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
6487 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6488 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
6489 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
6490 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
6491 		    (p[1] >> 2) | ((p[2] & 3) << 30),
6492 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
6493 		    p[0] & 1);
6494 	}
6495 
6496 	rc = sbuf_finish(sb);
6497 	sbuf_delete(sb);
6498 	free(buf, M_CXGBE);
6499 	return (rc);
6500 }
6501 
6502 static int
6503 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
6504 {
6505 	struct adapter *sc = arg1;
6506 	u_int i;
6507 	struct sbuf *sb;
6508 	uint32_t *buf, *p;
6509 	int rc;
6510 
6511 	rc = sysctl_wire_old_buffer(req, 0);
6512 	if (rc != 0)
6513 		return (rc);
6514 
6515 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6516 	if (sb == NULL)
6517 		return (ENOMEM);
6518 
6519 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
6520 	    M_ZERO | M_WAITOK);
6521 
6522 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
6523 	p = buf;
6524 
6525 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
6526 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6527 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
6528 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
6529 		    p[4], p[3], p[2], p[1], p[0]);
6530 	}
6531 
6532 	sbuf_printf(sb, "\n\nCntl ID               Data");
6533 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6534 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
6535 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
6536 	}
6537 
6538 	rc = sbuf_finish(sb);
6539 	sbuf_delete(sb);
6540 	free(buf, M_CXGBE);
6541 	return (rc);
6542 }
6543 
6544 static int
6545 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
6546 {
6547 	struct adapter *sc = arg1;
6548 	struct sbuf *sb;
6549 	int rc, i;
6550 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6551 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6552 	uint16_t thres[CIM_NUM_IBQ];
6553 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
6554 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
6555 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
6556 
6557 	cim_num_obq = sc->chip_params->cim_num_obq;
6558 	if (is_t4(sc)) {
6559 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
6560 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
6561 	} else {
6562 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
6563 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
6564 	}
6565 	nq = CIM_NUM_IBQ + cim_num_obq;
6566 
6567 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
6568 	if (rc == 0)
6569 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
6570 	if (rc != 0)
6571 		return (rc);
6572 
6573 	t4_read_cimq_cfg(sc, base, size, thres);
6574 
6575 	rc = sysctl_wire_old_buffer(req, 0);
6576 	if (rc != 0)
6577 		return (rc);
6578 
6579 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6580 	if (sb == NULL)
6581 		return (ENOMEM);
6582 
6583 	sbuf_printf(sb,
6584 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
6585 
6586 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
6587 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
6588 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
6589 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6590 		    G_QUEREMFLITS(p[2]) * 16);
6591 	for ( ; i < nq; i++, p += 4, wr += 2)
6592 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
6593 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
6594 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6595 		    G_QUEREMFLITS(p[2]) * 16);
6596 
6597 	rc = sbuf_finish(sb);
6598 	sbuf_delete(sb);
6599 
6600 	return (rc);
6601 }
6602 
6603 static int
6604 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
6605 {
6606 	struct adapter *sc = arg1;
6607 	struct sbuf *sb;
6608 	int rc;
6609 	struct tp_cpl_stats stats;
6610 
6611 	rc = sysctl_wire_old_buffer(req, 0);
6612 	if (rc != 0)
6613 		return (rc);
6614 
6615 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6616 	if (sb == NULL)
6617 		return (ENOMEM);
6618 
6619 	mtx_lock(&sc->reg_lock);
6620 	t4_tp_get_cpl_stats(sc, &stats, 0);
6621 	mtx_unlock(&sc->reg_lock);
6622 
6623 	if (sc->chip_params->nchan > 2) {
6624 		sbuf_printf(sb, "                 channel 0  channel 1"
6625 		    "  channel 2  channel 3");
6626 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
6627 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
6628 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
6629 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
6630 	} else {
6631 		sbuf_printf(sb, "                 channel 0  channel 1");
6632 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
6633 		    stats.req[0], stats.req[1]);
6634 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
6635 		    stats.rsp[0], stats.rsp[1]);
6636 	}
6637 
6638 	rc = sbuf_finish(sb);
6639 	sbuf_delete(sb);
6640 
6641 	return (rc);
6642 }
6643 
6644 static int
6645 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
6646 {
6647 	struct adapter *sc = arg1;
6648 	struct sbuf *sb;
6649 	int rc;
6650 	struct tp_usm_stats stats;
6651 
6652 	rc = sysctl_wire_old_buffer(req, 0);
6653 	if (rc != 0)
6654 		return(rc);
6655 
6656 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6657 	if (sb == NULL)
6658 		return (ENOMEM);
6659 
6660 	t4_get_usm_stats(sc, &stats, 1);
6661 
6662 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
6663 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
6664 	sbuf_printf(sb, "Drops:  %u", stats.drops);
6665 
6666 	rc = sbuf_finish(sb);
6667 	sbuf_delete(sb);
6668 
6669 	return (rc);
6670 }
6671 
6672 static const char * const devlog_level_strings[] = {
6673 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
6674 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
6675 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
6676 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
6677 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
6678 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
6679 };
6680 
6681 static const char * const devlog_facility_strings[] = {
6682 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
6683 	[FW_DEVLOG_FACILITY_CF]		= "CF",
6684 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
6685 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
6686 	[FW_DEVLOG_FACILITY_RES]	= "RES",
6687 	[FW_DEVLOG_FACILITY_HW]		= "HW",
6688 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
6689 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
6690 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
6691 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
6692 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
6693 	[FW_DEVLOG_FACILITY_VI]		= "VI",
6694 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
6695 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
6696 	[FW_DEVLOG_FACILITY_TM]		= "TM",
6697 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
6698 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
6699 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
6700 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
6701 	[FW_DEVLOG_FACILITY_RI]		= "RI",
6702 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
6703 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
6704 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
6705 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
6706 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
6707 };
6708 
6709 static int
6710 sysctl_devlog(SYSCTL_HANDLER_ARGS)
6711 {
6712 	struct adapter *sc = arg1;
6713 	struct devlog_params *dparams = &sc->params.devlog;
6714 	struct fw_devlog_e *buf, *e;
6715 	int i, j, rc, nentries, first = 0;
6716 	struct sbuf *sb;
6717 	uint64_t ftstamp = UINT64_MAX;
6718 
6719 	if (dparams->addr == 0)
6720 		return (ENXIO);
6721 
6722 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
6723 	if (buf == NULL)
6724 		return (ENOMEM);
6725 
6726 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
6727 	if (rc != 0)
6728 		goto done;
6729 
6730 	nentries = dparams->size / sizeof(struct fw_devlog_e);
6731 	for (i = 0; i < nentries; i++) {
6732 		e = &buf[i];
6733 
6734 		if (e->timestamp == 0)
6735 			break;	/* end */
6736 
6737 		e->timestamp = be64toh(e->timestamp);
6738 		e->seqno = be32toh(e->seqno);
6739 		for (j = 0; j < 8; j++)
6740 			e->params[j] = be32toh(e->params[j]);
6741 
6742 		if (e->timestamp < ftstamp) {
6743 			ftstamp = e->timestamp;
6744 			first = i;
6745 		}
6746 	}
6747 
6748 	if (buf[first].timestamp == 0)
6749 		goto done;	/* nothing in the log */
6750 
6751 	rc = sysctl_wire_old_buffer(req, 0);
6752 	if (rc != 0)
6753 		goto done;
6754 
6755 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6756 	if (sb == NULL) {
6757 		rc = ENOMEM;
6758 		goto done;
6759 	}
6760 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
6761 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
6762 
6763 	i = first;
6764 	do {
6765 		e = &buf[i];
6766 		if (e->timestamp == 0)
6767 			break;	/* end */
6768 
6769 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
6770 		    e->seqno, e->timestamp,
6771 		    (e->level < nitems(devlog_level_strings) ?
6772 			devlog_level_strings[e->level] : "UNKNOWN"),
6773 		    (e->facility < nitems(devlog_facility_strings) ?
6774 			devlog_facility_strings[e->facility] : "UNKNOWN"));
6775 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
6776 		    e->params[2], e->params[3], e->params[4],
6777 		    e->params[5], e->params[6], e->params[7]);
6778 
6779 		if (++i == nentries)
6780 			i = 0;
6781 	} while (i != first);
6782 
6783 	rc = sbuf_finish(sb);
6784 	sbuf_delete(sb);
6785 done:
6786 	free(buf, M_CXGBE);
6787 	return (rc);
6788 }
6789 
6790 static int
6791 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
6792 {
6793 	struct adapter *sc = arg1;
6794 	struct sbuf *sb;
6795 	int rc;
6796 	struct tp_fcoe_stats stats[MAX_NCHAN];
6797 	int i, nchan = sc->chip_params->nchan;
6798 
6799 	rc = sysctl_wire_old_buffer(req, 0);
6800 	if (rc != 0)
6801 		return (rc);
6802 
6803 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6804 	if (sb == NULL)
6805 		return (ENOMEM);
6806 
6807 	for (i = 0; i < nchan; i++)
6808 		t4_get_fcoe_stats(sc, i, &stats[i], 1);
6809 
6810 	if (nchan > 2) {
6811 		sbuf_printf(sb, "                   channel 0        channel 1"
6812 		    "        channel 2        channel 3");
6813 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
6814 		    stats[0].octets_ddp, stats[1].octets_ddp,
6815 		    stats[2].octets_ddp, stats[3].octets_ddp);
6816 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
6817 		    stats[0].frames_ddp, stats[1].frames_ddp,
6818 		    stats[2].frames_ddp, stats[3].frames_ddp);
6819 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
6820 		    stats[0].frames_drop, stats[1].frames_drop,
6821 		    stats[2].frames_drop, stats[3].frames_drop);
6822 	} else {
6823 		sbuf_printf(sb, "                   channel 0        channel 1");
6824 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
6825 		    stats[0].octets_ddp, stats[1].octets_ddp);
6826 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
6827 		    stats[0].frames_ddp, stats[1].frames_ddp);
6828 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
6829 		    stats[0].frames_drop, stats[1].frames_drop);
6830 	}
6831 
6832 	rc = sbuf_finish(sb);
6833 	sbuf_delete(sb);
6834 
6835 	return (rc);
6836 }
6837 
6838 static int
6839 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
6840 {
6841 	struct adapter *sc = arg1;
6842 	struct sbuf *sb;
6843 	int rc, i;
6844 	unsigned int map, kbps, ipg, mode;
6845 	unsigned int pace_tab[NTX_SCHED];
6846 
6847 	rc = sysctl_wire_old_buffer(req, 0);
6848 	if (rc != 0)
6849 		return (rc);
6850 
6851 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6852 	if (sb == NULL)
6853 		return (ENOMEM);
6854 
6855 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
6856 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
6857 	t4_read_pace_tbl(sc, pace_tab);
6858 
6859 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
6860 	    "Class IPG (0.1 ns)   Flow IPG (us)");
6861 
6862 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
6863 		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
6864 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
6865 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
6866 		if (kbps)
6867 			sbuf_printf(sb, "%9u     ", kbps);
6868 		else
6869 			sbuf_printf(sb, " disabled     ");
6870 
6871 		if (ipg)
6872 			sbuf_printf(sb, "%13u        ", ipg);
6873 		else
6874 			sbuf_printf(sb, "     disabled        ");
6875 
6876 		if (pace_tab[i])
6877 			sbuf_printf(sb, "%10u", pace_tab[i]);
6878 		else
6879 			sbuf_printf(sb, "  disabled");
6880 	}
6881 
6882 	rc = sbuf_finish(sb);
6883 	sbuf_delete(sb);
6884 
6885 	return (rc);
6886 }
6887 
6888 static int
6889 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
6890 {
6891 	struct adapter *sc = arg1;
6892 	struct sbuf *sb;
6893 	int rc, i, j;
6894 	uint64_t *p0, *p1;
6895 	struct lb_port_stats s[2];
6896 	static const char *stat_name[] = {
6897 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
6898 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
6899 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
6900 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
6901 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
6902 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
6903 		"BG2FramesTrunc:", "BG3FramesTrunc:"
6904 	};
6905 
6906 	rc = sysctl_wire_old_buffer(req, 0);
6907 	if (rc != 0)
6908 		return (rc);
6909 
6910 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6911 	if (sb == NULL)
6912 		return (ENOMEM);
6913 
6914 	memset(s, 0, sizeof(s));
6915 
6916 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
6917 		t4_get_lb_stats(sc, i, &s[0]);
6918 		t4_get_lb_stats(sc, i + 1, &s[1]);
6919 
6920 		p0 = &s[0].octets;
6921 		p1 = &s[1].octets;
6922 		sbuf_printf(sb, "%s                       Loopback %u"
6923 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
6924 
6925 		for (j = 0; j < nitems(stat_name); j++)
6926 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
6927 				   *p0++, *p1++);
6928 	}
6929 
6930 	rc = sbuf_finish(sb);
6931 	sbuf_delete(sb);
6932 
6933 	return (rc);
6934 }
6935 
6936 static int
6937 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
6938 {
6939 	int rc = 0;
6940 	struct port_info *pi = arg1;
6941 	struct link_config *lc = &pi->link_cfg;
6942 	struct sbuf *sb;
6943 
6944 	rc = sysctl_wire_old_buffer(req, 0);
6945 	if (rc != 0)
6946 		return(rc);
6947 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
6948 	if (sb == NULL)
6949 		return (ENOMEM);
6950 
6951 	if (lc->link_ok || lc->link_down_rc == 255)
6952 		sbuf_printf(sb, "n/a");
6953 	else
6954 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
6955 
6956 	rc = sbuf_finish(sb);
6957 	sbuf_delete(sb);
6958 
6959 	return (rc);
6960 }
6961 
6962 struct mem_desc {
6963 	unsigned int base;
6964 	unsigned int limit;
6965 	unsigned int idx;
6966 };
6967 
6968 static int
6969 mem_desc_cmp(const void *a, const void *b)
6970 {
6971 	return ((const struct mem_desc *)a)->base -
6972 	       ((const struct mem_desc *)b)->base;
6973 }
6974 
6975 static void
6976 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
6977     unsigned int to)
6978 {
6979 	unsigned int size;
6980 
6981 	if (from == to)
6982 		return;
6983 
6984 	size = to - from + 1;
6985 	if (size == 0)
6986 		return;
6987 
6988 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
6989 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
6990 }
6991 
6992 static int
6993 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
6994 {
6995 	struct adapter *sc = arg1;
6996 	struct sbuf *sb;
6997 	int rc, i, n;
6998 	uint32_t lo, hi, used, alloc;
6999 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
7000 	static const char *region[] = {
7001 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
7002 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
7003 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
7004 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
7005 		"RQUDP region:", "PBL region:", "TXPBL region:",
7006 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
7007 		"On-chip queues:"
7008 	};
7009 	struct mem_desc avail[4];
7010 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
7011 	struct mem_desc *md = mem;
7012 
7013 	rc = sysctl_wire_old_buffer(req, 0);
7014 	if (rc != 0)
7015 		return (rc);
7016 
7017 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7018 	if (sb == NULL)
7019 		return (ENOMEM);
7020 
7021 	for (i = 0; i < nitems(mem); i++) {
7022 		mem[i].limit = 0;
7023 		mem[i].idx = i;
7024 	}
7025 
7026 	/* Find and sort the populated memory ranges */
7027 	i = 0;
7028 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
7029 	if (lo & F_EDRAM0_ENABLE) {
7030 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
7031 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
7032 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
7033 		avail[i].idx = 0;
7034 		i++;
7035 	}
7036 	if (lo & F_EDRAM1_ENABLE) {
7037 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
7038 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
7039 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
7040 		avail[i].idx = 1;
7041 		i++;
7042 	}
7043 	if (lo & F_EXT_MEM_ENABLE) {
7044 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
7045 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
7046 		avail[i].limit = avail[i].base +
7047 		    (G_EXT_MEM_SIZE(hi) << 20);
7048 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
7049 		i++;
7050 	}
7051 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
7052 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
7053 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
7054 		avail[i].limit = avail[i].base +
7055 		    (G_EXT_MEM1_SIZE(hi) << 20);
7056 		avail[i].idx = 4;
7057 		i++;
7058 	}
7059 	if (!i)                                    /* no memory available */
7060 		return 0;
7061 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
7062 
7063 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
7064 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
7065 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
7066 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
7067 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
7068 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
7069 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
7070 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
7071 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
7072 
7073 	/* the next few have explicit upper bounds */
7074 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
7075 	md->limit = md->base - 1 +
7076 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
7077 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
7078 	md++;
7079 
7080 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
7081 	md->limit = md->base - 1 +
7082 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
7083 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
7084 	md++;
7085 
7086 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7087 		if (chip_id(sc) <= CHELSIO_T5)
7088 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
7089 		else
7090 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
7091 		md->limit = 0;
7092 	} else {
7093 		md->base = 0;
7094 		md->idx = nitems(region);  /* hide it */
7095 	}
7096 	md++;
7097 
7098 #define ulp_region(reg) \
7099 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
7100 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
7101 
7102 	ulp_region(RX_ISCSI);
7103 	ulp_region(RX_TDDP);
7104 	ulp_region(TX_TPT);
7105 	ulp_region(RX_STAG);
7106 	ulp_region(RX_RQ);
7107 	ulp_region(RX_RQUDP);
7108 	ulp_region(RX_PBL);
7109 	ulp_region(TX_PBL);
7110 #undef ulp_region
7111 
7112 	md->base = 0;
7113 	md->idx = nitems(region);
7114 	if (!is_t4(sc)) {
7115 		uint32_t size = 0;
7116 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
7117 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
7118 
7119 		if (is_t5(sc)) {
7120 			if (sge_ctrl & F_VFIFO_ENABLE)
7121 				size = G_DBVFIFO_SIZE(fifo_size);
7122 		} else
7123 			size = G_T6_DBVFIFO_SIZE(fifo_size);
7124 
7125 		if (size) {
7126 			md->base = G_BASEADDR(t4_read_reg(sc,
7127 			    A_SGE_DBVFIFO_BADDR));
7128 			md->limit = md->base + (size << 2) - 1;
7129 		}
7130 	}
7131 	md++;
7132 
7133 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
7134 	md->limit = 0;
7135 	md++;
7136 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
7137 	md->limit = 0;
7138 	md++;
7139 
7140 	md->base = sc->vres.ocq.start;
7141 	if (sc->vres.ocq.size)
7142 		md->limit = md->base + sc->vres.ocq.size - 1;
7143 	else
7144 		md->idx = nitems(region);  /* hide it */
7145 	md++;
7146 
7147 	/* add any address-space holes, there can be up to 3 */
7148 	for (n = 0; n < i - 1; n++)
7149 		if (avail[n].limit < avail[n + 1].base)
7150 			(md++)->base = avail[n].limit;
7151 	if (avail[n].limit)
7152 		(md++)->base = avail[n].limit;
7153 
7154 	n = md - mem;
7155 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
7156 
7157 	for (lo = 0; lo < i; lo++)
7158 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
7159 				avail[lo].limit - 1);
7160 
7161 	sbuf_printf(sb, "\n");
7162 	for (i = 0; i < n; i++) {
7163 		if (mem[i].idx >= nitems(region))
7164 			continue;                        /* skip holes */
7165 		if (!mem[i].limit)
7166 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
7167 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
7168 				mem[i].limit);
7169 	}
7170 
7171 	sbuf_printf(sb, "\n");
7172 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
7173 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
7174 	mem_region_show(sb, "uP RAM:", lo, hi);
7175 
7176 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
7177 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
7178 	mem_region_show(sb, "uP Extmem2:", lo, hi);
7179 
7180 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
7181 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
7182 		   G_PMRXMAXPAGE(lo),
7183 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
7184 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
7185 
7186 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
7187 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
7188 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
7189 		   G_PMTXMAXPAGE(lo),
7190 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
7191 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
7192 	sbuf_printf(sb, "%u p-structs\n",
7193 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
7194 
7195 	for (i = 0; i < 4; i++) {
7196 		if (chip_id(sc) > CHELSIO_T5)
7197 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
7198 		else
7199 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
7200 		if (is_t5(sc)) {
7201 			used = G_T5_USED(lo);
7202 			alloc = G_T5_ALLOC(lo);
7203 		} else {
7204 			used = G_USED(lo);
7205 			alloc = G_ALLOC(lo);
7206 		}
7207 		/* For T6 these are MAC buffer groups */
7208 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
7209 		    i, used, alloc);
7210 	}
7211 	for (i = 0; i < sc->chip_params->nchan; i++) {
7212 		if (chip_id(sc) > CHELSIO_T5)
7213 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
7214 		else
7215 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
7216 		if (is_t5(sc)) {
7217 			used = G_T5_USED(lo);
7218 			alloc = G_T5_ALLOC(lo);
7219 		} else {
7220 			used = G_USED(lo);
7221 			alloc = G_ALLOC(lo);
7222 		}
7223 		/* For T6 these are MAC buffer groups */
7224 		sbuf_printf(sb,
7225 		    "\nLoopback %d using %u pages out of %u allocated",
7226 		    i, used, alloc);
7227 	}
7228 
7229 	rc = sbuf_finish(sb);
7230 	sbuf_delete(sb);
7231 
7232 	return (rc);
7233 }
7234 
7235 static inline void
7236 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
7237 {
7238 	*mask = x | y;
7239 	y = htobe64(y);
7240 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
7241 }
7242 
7243 static int
7244 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
7245 {
7246 	struct adapter *sc = arg1;
7247 	struct sbuf *sb;
7248 	int rc, i;
7249 
7250 	MPASS(chip_id(sc) <= CHELSIO_T5);
7251 
7252 	rc = sysctl_wire_old_buffer(req, 0);
7253 	if (rc != 0)
7254 		return (rc);
7255 
7256 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7257 	if (sb == NULL)
7258 		return (ENOMEM);
7259 
7260 	sbuf_printf(sb,
7261 	    "Idx  Ethernet address     Mask     Vld Ports PF"
7262 	    "  VF              Replication             P0 P1 P2 P3  ML");
7263 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7264 		uint64_t tcamx, tcamy, mask;
7265 		uint32_t cls_lo, cls_hi;
7266 		uint8_t addr[ETHER_ADDR_LEN];
7267 
7268 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
7269 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
7270 		if (tcamx & tcamy)
7271 			continue;
7272 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7273 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7274 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7275 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
7276 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
7277 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
7278 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
7279 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
7280 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
7281 
7282 		if (cls_lo & F_REPLICATE) {
7283 			struct fw_ldst_cmd ldst_cmd;
7284 
7285 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7286 			ldst_cmd.op_to_addrspace =
7287 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7288 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7289 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7290 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7291 			ldst_cmd.u.mps.rplc.fid_idx =
7292 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7293 				V_FW_LDST_CMD_IDX(i));
7294 
7295 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7296 			    "t4mps");
7297 			if (rc)
7298 				break;
7299 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7300 			    sizeof(ldst_cmd), &ldst_cmd);
7301 			end_synchronized_op(sc, 0);
7302 
7303 			if (rc != 0) {
7304 				sbuf_printf(sb, "%36d", rc);
7305 				rc = 0;
7306 			} else {
7307 				sbuf_printf(sb, " %08x %08x %08x %08x",
7308 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7309 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7310 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7311 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7312 			}
7313 		} else
7314 			sbuf_printf(sb, "%36s", "");
7315 
7316 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
7317 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
7318 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
7319 	}
7320 
7321 	if (rc)
7322 		(void) sbuf_finish(sb);
7323 	else
7324 		rc = sbuf_finish(sb);
7325 	sbuf_delete(sb);
7326 
7327 	return (rc);
7328 }
7329 
7330 static int
7331 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
7332 {
7333 	struct adapter *sc = arg1;
7334 	struct sbuf *sb;
7335 	int rc, i;
7336 
7337 	MPASS(chip_id(sc) > CHELSIO_T5);
7338 
7339 	rc = sysctl_wire_old_buffer(req, 0);
7340 	if (rc != 0)
7341 		return (rc);
7342 
7343 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7344 	if (sb == NULL)
7345 		return (ENOMEM);
7346 
7347 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
7348 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
7349 	    "                           Replication"
7350 	    "                                    P0 P1 P2 P3  ML\n");
7351 
7352 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7353 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
7354 		uint16_t ivlan;
7355 		uint64_t tcamx, tcamy, val, mask;
7356 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
7357 		uint8_t addr[ETHER_ADDR_LEN];
7358 
7359 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
7360 		if (i < 256)
7361 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
7362 		else
7363 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
7364 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7365 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7366 		tcamy = G_DMACH(val) << 32;
7367 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7368 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7369 		lookup_type = G_DATALKPTYPE(data2);
7370 		port_num = G_DATAPORTNUM(data2);
7371 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7372 			/* Inner header VNI */
7373 			vniy = ((data2 & F_DATAVIDH2) << 23) |
7374 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7375 			dip_hit = data2 & F_DATADIPHIT;
7376 			vlan_vld = 0;
7377 		} else {
7378 			vniy = 0;
7379 			dip_hit = 0;
7380 			vlan_vld = data2 & F_DATAVIDH2;
7381 			ivlan = G_VIDL(val);
7382 		}
7383 
7384 		ctl |= V_CTLXYBITSEL(1);
7385 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7386 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7387 		tcamx = G_DMACH(val) << 32;
7388 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7389 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7390 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7391 			/* Inner header VNI mask */
7392 			vnix = ((data2 & F_DATAVIDH2) << 23) |
7393 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7394 		} else
7395 			vnix = 0;
7396 
7397 		if (tcamx & tcamy)
7398 			continue;
7399 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7400 
7401 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7402 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7403 
7404 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7405 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7406 			    "%012jx %06x %06x    -    -   %3c"
7407 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
7408 			    addr[1], addr[2], addr[3], addr[4], addr[5],
7409 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
7410 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7411 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7412 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7413 		} else {
7414 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7415 			    "%012jx    -       -   ", i, addr[0], addr[1],
7416 			    addr[2], addr[3], addr[4], addr[5],
7417 			    (uintmax_t)mask);
7418 
7419 			if (vlan_vld)
7420 				sbuf_printf(sb, "%4u   Y     ", ivlan);
7421 			else
7422 				sbuf_printf(sb, "  -    N     ");
7423 
7424 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
7425 			    lookup_type ? 'I' : 'O', port_num,
7426 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7427 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7428 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7429 		}
7430 
7431 
7432 		if (cls_lo & F_T6_REPLICATE) {
7433 			struct fw_ldst_cmd ldst_cmd;
7434 
7435 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7436 			ldst_cmd.op_to_addrspace =
7437 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7438 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7439 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7440 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7441 			ldst_cmd.u.mps.rplc.fid_idx =
7442 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7443 				V_FW_LDST_CMD_IDX(i));
7444 
7445 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7446 			    "t6mps");
7447 			if (rc)
7448 				break;
7449 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7450 			    sizeof(ldst_cmd), &ldst_cmd);
7451 			end_synchronized_op(sc, 0);
7452 
7453 			if (rc != 0) {
7454 				sbuf_printf(sb, "%72d", rc);
7455 				rc = 0;
7456 			} else {
7457 				sbuf_printf(sb, " %08x %08x %08x %08x"
7458 				    " %08x %08x %08x %08x",
7459 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
7460 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
7461 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
7462 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
7463 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7464 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7465 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7466 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7467 			}
7468 		} else
7469 			sbuf_printf(sb, "%72s", "");
7470 
7471 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
7472 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
7473 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
7474 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
7475 	}
7476 
7477 	if (rc)
7478 		(void) sbuf_finish(sb);
7479 	else
7480 		rc = sbuf_finish(sb);
7481 	sbuf_delete(sb);
7482 
7483 	return (rc);
7484 }
7485 
7486 static int
7487 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
7488 {
7489 	struct adapter *sc = arg1;
7490 	struct sbuf *sb;
7491 	int rc;
7492 	uint16_t mtus[NMTUS];
7493 
7494 	rc = sysctl_wire_old_buffer(req, 0);
7495 	if (rc != 0)
7496 		return (rc);
7497 
7498 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7499 	if (sb == NULL)
7500 		return (ENOMEM);
7501 
7502 	t4_read_mtu_tbl(sc, mtus, NULL);
7503 
7504 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
7505 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
7506 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
7507 	    mtus[14], mtus[15]);
7508 
7509 	rc = sbuf_finish(sb);
7510 	sbuf_delete(sb);
7511 
7512 	return (rc);
7513 }
7514 
7515 static int
7516 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
7517 {
7518 	struct adapter *sc = arg1;
7519 	struct sbuf *sb;
7520 	int rc, i;
7521 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
7522 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
7523 	static const char *tx_stats[MAX_PM_NSTATS] = {
7524 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
7525 		"Tx FIFO wait", NULL, "Tx latency"
7526 	};
7527 	static const char *rx_stats[MAX_PM_NSTATS] = {
7528 		"Read:", "Write bypass:", "Write mem:", "Flush:",
7529 		"Rx FIFO wait", NULL, "Rx latency"
7530 	};
7531 
7532 	rc = sysctl_wire_old_buffer(req, 0);
7533 	if (rc != 0)
7534 		return (rc);
7535 
7536 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7537 	if (sb == NULL)
7538 		return (ENOMEM);
7539 
7540 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
7541 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
7542 
7543 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
7544 	for (i = 0; i < 4; i++) {
7545 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7546 		    tx_cyc[i]);
7547 	}
7548 
7549 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
7550 	for (i = 0; i < 4; i++) {
7551 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7552 		    rx_cyc[i]);
7553 	}
7554 
7555 	if (chip_id(sc) > CHELSIO_T5) {
7556 		sbuf_printf(sb,
7557 		    "\n              Total wait      Total occupancy");
7558 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7559 		    tx_cyc[i]);
7560 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7561 		    rx_cyc[i]);
7562 
7563 		i += 2;
7564 		MPASS(i < nitems(tx_stats));
7565 
7566 		sbuf_printf(sb,
7567 		    "\n                   Reads           Total wait");
7568 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7569 		    tx_cyc[i]);
7570 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7571 		    rx_cyc[i]);
7572 	}
7573 
7574 	rc = sbuf_finish(sb);
7575 	sbuf_delete(sb);
7576 
7577 	return (rc);
7578 }
7579 
7580 static int
7581 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
7582 {
7583 	struct adapter *sc = arg1;
7584 	struct sbuf *sb;
7585 	int rc;
7586 	struct tp_rdma_stats stats;
7587 
7588 	rc = sysctl_wire_old_buffer(req, 0);
7589 	if (rc != 0)
7590 		return (rc);
7591 
7592 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7593 	if (sb == NULL)
7594 		return (ENOMEM);
7595 
7596 	mtx_lock(&sc->reg_lock);
7597 	t4_tp_get_rdma_stats(sc, &stats, 0);
7598 	mtx_unlock(&sc->reg_lock);
7599 
7600 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
7601 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
7602 
7603 	rc = sbuf_finish(sb);
7604 	sbuf_delete(sb);
7605 
7606 	return (rc);
7607 }
7608 
7609 static int
7610 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
7611 {
7612 	struct adapter *sc = arg1;
7613 	struct sbuf *sb;
7614 	int rc;
7615 	struct tp_tcp_stats v4, v6;
7616 
7617 	rc = sysctl_wire_old_buffer(req, 0);
7618 	if (rc != 0)
7619 		return (rc);
7620 
7621 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7622 	if (sb == NULL)
7623 		return (ENOMEM);
7624 
7625 	mtx_lock(&sc->reg_lock);
7626 	t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
7627 	mtx_unlock(&sc->reg_lock);
7628 
7629 	sbuf_printf(sb,
7630 	    "                                IP                 IPv6\n");
7631 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
7632 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
7633 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
7634 	    v4.tcp_in_segs, v6.tcp_in_segs);
7635 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
7636 	    v4.tcp_out_segs, v6.tcp_out_segs);
7637 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
7638 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
7639 
7640 	rc = sbuf_finish(sb);
7641 	sbuf_delete(sb);
7642 
7643 	return (rc);
7644 }
7645 
7646 static int
7647 sysctl_tids(SYSCTL_HANDLER_ARGS)
7648 {
7649 	struct adapter *sc = arg1;
7650 	struct sbuf *sb;
7651 	int rc;
7652 	struct tid_info *t = &sc->tids;
7653 
7654 	rc = sysctl_wire_old_buffer(req, 0);
7655 	if (rc != 0)
7656 		return (rc);
7657 
7658 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7659 	if (sb == NULL)
7660 		return (ENOMEM);
7661 
7662 	if (t->natids) {
7663 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
7664 		    t->atids_in_use);
7665 	}
7666 
7667 	if (t->ntids) {
7668 		sbuf_printf(sb, "TID range: ");
7669 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7670 			uint32_t b, hb;
7671 
7672 			if (chip_id(sc) <= CHELSIO_T5) {
7673 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
7674 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
7675 			} else {
7676 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
7677 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
7678 			}
7679 
7680 			if (b)
7681 				sbuf_printf(sb, "0-%u, ", b - 1);
7682 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
7683 		} else
7684 			sbuf_printf(sb, "0-%u", t->ntids - 1);
7685 		sbuf_printf(sb, ", in use: %u\n",
7686 		    atomic_load_acq_int(&t->tids_in_use));
7687 	}
7688 
7689 	if (t->nstids) {
7690 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
7691 		    t->stid_base + t->nstids - 1, t->stids_in_use);
7692 	}
7693 
7694 	if (t->nftids) {
7695 		sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
7696 		    t->ftid_base + t->nftids - 1);
7697 	}
7698 
7699 	if (t->netids) {
7700 		sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
7701 		    t->etid_base + t->netids - 1);
7702 	}
7703 
7704 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
7705 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
7706 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
7707 
7708 	rc = sbuf_finish(sb);
7709 	sbuf_delete(sb);
7710 
7711 	return (rc);
7712 }
7713 
7714 static int
7715 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
7716 {
7717 	struct adapter *sc = arg1;
7718 	struct sbuf *sb;
7719 	int rc;
7720 	struct tp_err_stats stats;
7721 
7722 	rc = sysctl_wire_old_buffer(req, 0);
7723 	if (rc != 0)
7724 		return (rc);
7725 
7726 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7727 	if (sb == NULL)
7728 		return (ENOMEM);
7729 
7730 	mtx_lock(&sc->reg_lock);
7731 	t4_tp_get_err_stats(sc, &stats, 0);
7732 	mtx_unlock(&sc->reg_lock);
7733 
7734 	if (sc->chip_params->nchan > 2) {
7735 		sbuf_printf(sb, "                 channel 0  channel 1"
7736 		    "  channel 2  channel 3\n");
7737 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
7738 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
7739 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
7740 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
7741 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
7742 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
7743 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
7744 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
7745 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
7746 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
7747 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
7748 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
7749 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
7750 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
7751 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
7752 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
7753 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
7754 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
7755 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
7756 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
7757 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
7758 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
7759 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
7760 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
7761 	} else {
7762 		sbuf_printf(sb, "                 channel 0  channel 1\n");
7763 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
7764 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
7765 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
7766 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
7767 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
7768 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
7769 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
7770 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
7771 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
7772 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
7773 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
7774 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
7775 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
7776 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
7777 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
7778 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
7779 	}
7780 
7781 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
7782 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
7783 
7784 	rc = sbuf_finish(sb);
7785 	sbuf_delete(sb);
7786 
7787 	return (rc);
7788 }
7789 
7790 static int
7791 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
7792 {
7793 	struct adapter *sc = arg1;
7794 	struct tp_params *tpp = &sc->params.tp;
7795 	u_int mask;
7796 	int rc;
7797 
7798 	mask = tpp->la_mask >> 16;
7799 	rc = sysctl_handle_int(oidp, &mask, 0, req);
7800 	if (rc != 0 || req->newptr == NULL)
7801 		return (rc);
7802 	if (mask > 0xffff)
7803 		return (EINVAL);
7804 	tpp->la_mask = mask << 16;
7805 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
7806 
7807 	return (0);
7808 }
7809 
7810 struct field_desc {
7811 	const char *name;
7812 	u_int start;
7813 	u_int width;
7814 };
7815 
7816 static void
7817 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
7818 {
7819 	char buf[32];
7820 	int line_size = 0;
7821 
7822 	while (f->name) {
7823 		uint64_t mask = (1ULL << f->width) - 1;
7824 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
7825 		    ((uintmax_t)v >> f->start) & mask);
7826 
7827 		if (line_size + len >= 79) {
7828 			line_size = 8;
7829 			sbuf_printf(sb, "\n        ");
7830 		}
7831 		sbuf_printf(sb, "%s ", buf);
7832 		line_size += len + 1;
7833 		f++;
7834 	}
7835 	sbuf_printf(sb, "\n");
7836 }
7837 
7838 static const struct field_desc tp_la0[] = {
7839 	{ "RcfOpCodeOut", 60, 4 },
7840 	{ "State", 56, 4 },
7841 	{ "WcfState", 52, 4 },
7842 	{ "RcfOpcSrcOut", 50, 2 },
7843 	{ "CRxError", 49, 1 },
7844 	{ "ERxError", 48, 1 },
7845 	{ "SanityFailed", 47, 1 },
7846 	{ "SpuriousMsg", 46, 1 },
7847 	{ "FlushInputMsg", 45, 1 },
7848 	{ "FlushInputCpl", 44, 1 },
7849 	{ "RssUpBit", 43, 1 },
7850 	{ "RssFilterHit", 42, 1 },
7851 	{ "Tid", 32, 10 },
7852 	{ "InitTcb", 31, 1 },
7853 	{ "LineNumber", 24, 7 },
7854 	{ "Emsg", 23, 1 },
7855 	{ "EdataOut", 22, 1 },
7856 	{ "Cmsg", 21, 1 },
7857 	{ "CdataOut", 20, 1 },
7858 	{ "EreadPdu", 19, 1 },
7859 	{ "CreadPdu", 18, 1 },
7860 	{ "TunnelPkt", 17, 1 },
7861 	{ "RcfPeerFin", 16, 1 },
7862 	{ "RcfReasonOut", 12, 4 },
7863 	{ "TxCchannel", 10, 2 },
7864 	{ "RcfTxChannel", 8, 2 },
7865 	{ "RxEchannel", 6, 2 },
7866 	{ "RcfRxChannel", 5, 1 },
7867 	{ "RcfDataOutSrdy", 4, 1 },
7868 	{ "RxDvld", 3, 1 },
7869 	{ "RxOoDvld", 2, 1 },
7870 	{ "RxCongestion", 1, 1 },
7871 	{ "TxCongestion", 0, 1 },
7872 	{ NULL }
7873 };
7874 
7875 static const struct field_desc tp_la1[] = {
7876 	{ "CplCmdIn", 56, 8 },
7877 	{ "CplCmdOut", 48, 8 },
7878 	{ "ESynOut", 47, 1 },
7879 	{ "EAckOut", 46, 1 },
7880 	{ "EFinOut", 45, 1 },
7881 	{ "ERstOut", 44, 1 },
7882 	{ "SynIn", 43, 1 },
7883 	{ "AckIn", 42, 1 },
7884 	{ "FinIn", 41, 1 },
7885 	{ "RstIn", 40, 1 },
7886 	{ "DataIn", 39, 1 },
7887 	{ "DataInVld", 38, 1 },
7888 	{ "PadIn", 37, 1 },
7889 	{ "RxBufEmpty", 36, 1 },
7890 	{ "RxDdp", 35, 1 },
7891 	{ "RxFbCongestion", 34, 1 },
7892 	{ "TxFbCongestion", 33, 1 },
7893 	{ "TxPktSumSrdy", 32, 1 },
7894 	{ "RcfUlpType", 28, 4 },
7895 	{ "Eread", 27, 1 },
7896 	{ "Ebypass", 26, 1 },
7897 	{ "Esave", 25, 1 },
7898 	{ "Static0", 24, 1 },
7899 	{ "Cread", 23, 1 },
7900 	{ "Cbypass", 22, 1 },
7901 	{ "Csave", 21, 1 },
7902 	{ "CPktOut", 20, 1 },
7903 	{ "RxPagePoolFull", 18, 2 },
7904 	{ "RxLpbkPkt", 17, 1 },
7905 	{ "TxLpbkPkt", 16, 1 },
7906 	{ "RxVfValid", 15, 1 },
7907 	{ "SynLearned", 14, 1 },
7908 	{ "SetDelEntry", 13, 1 },
7909 	{ "SetInvEntry", 12, 1 },
7910 	{ "CpcmdDvld", 11, 1 },
7911 	{ "CpcmdSave", 10, 1 },
7912 	{ "RxPstructsFull", 8, 2 },
7913 	{ "EpcmdDvld", 7, 1 },
7914 	{ "EpcmdFlush", 6, 1 },
7915 	{ "EpcmdTrimPrefix", 5, 1 },
7916 	{ "EpcmdTrimPostfix", 4, 1 },
7917 	{ "ERssIp4Pkt", 3, 1 },
7918 	{ "ERssIp6Pkt", 2, 1 },
7919 	{ "ERssTcpUdpPkt", 1, 1 },
7920 	{ "ERssFceFipPkt", 0, 1 },
7921 	{ NULL }
7922 };
7923 
7924 static const struct field_desc tp_la2[] = {
7925 	{ "CplCmdIn", 56, 8 },
7926 	{ "MpsVfVld", 55, 1 },
7927 	{ "MpsPf", 52, 3 },
7928 	{ "MpsVf", 44, 8 },
7929 	{ "SynIn", 43, 1 },
7930 	{ "AckIn", 42, 1 },
7931 	{ "FinIn", 41, 1 },
7932 	{ "RstIn", 40, 1 },
7933 	{ "DataIn", 39, 1 },
7934 	{ "DataInVld", 38, 1 },
7935 	{ "PadIn", 37, 1 },
7936 	{ "RxBufEmpty", 36, 1 },
7937 	{ "RxDdp", 35, 1 },
7938 	{ "RxFbCongestion", 34, 1 },
7939 	{ "TxFbCongestion", 33, 1 },
7940 	{ "TxPktSumSrdy", 32, 1 },
7941 	{ "RcfUlpType", 28, 4 },
7942 	{ "Eread", 27, 1 },
7943 	{ "Ebypass", 26, 1 },
7944 	{ "Esave", 25, 1 },
7945 	{ "Static0", 24, 1 },
7946 	{ "Cread", 23, 1 },
7947 	{ "Cbypass", 22, 1 },
7948 	{ "Csave", 21, 1 },
7949 	{ "CPktOut", 20, 1 },
7950 	{ "RxPagePoolFull", 18, 2 },
7951 	{ "RxLpbkPkt", 17, 1 },
7952 	{ "TxLpbkPkt", 16, 1 },
7953 	{ "RxVfValid", 15, 1 },
7954 	{ "SynLearned", 14, 1 },
7955 	{ "SetDelEntry", 13, 1 },
7956 	{ "SetInvEntry", 12, 1 },
7957 	{ "CpcmdDvld", 11, 1 },
7958 	{ "CpcmdSave", 10, 1 },
7959 	{ "RxPstructsFull", 8, 2 },
7960 	{ "EpcmdDvld", 7, 1 },
7961 	{ "EpcmdFlush", 6, 1 },
7962 	{ "EpcmdTrimPrefix", 5, 1 },
7963 	{ "EpcmdTrimPostfix", 4, 1 },
7964 	{ "ERssIp4Pkt", 3, 1 },
7965 	{ "ERssIp6Pkt", 2, 1 },
7966 	{ "ERssTcpUdpPkt", 1, 1 },
7967 	{ "ERssFceFipPkt", 0, 1 },
7968 	{ NULL }
7969 };
7970 
7971 static void
7972 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
7973 {
7974 
7975 	field_desc_show(sb, *p, tp_la0);
7976 }
7977 
7978 static void
7979 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
7980 {
7981 
7982 	if (idx)
7983 		sbuf_printf(sb, "\n");
7984 	field_desc_show(sb, p[0], tp_la0);
7985 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7986 		field_desc_show(sb, p[1], tp_la0);
7987 }
7988 
7989 static void
7990 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
7991 {
7992 
7993 	if (idx)
7994 		sbuf_printf(sb, "\n");
7995 	field_desc_show(sb, p[0], tp_la0);
7996 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7997 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
7998 }
7999 
8000 static int
8001 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
8002 {
8003 	struct adapter *sc = arg1;
8004 	struct sbuf *sb;
8005 	uint64_t *buf, *p;
8006 	int rc;
8007 	u_int i, inc;
8008 	void (*show_func)(struct sbuf *, uint64_t *, int);
8009 
8010 	rc = sysctl_wire_old_buffer(req, 0);
8011 	if (rc != 0)
8012 		return (rc);
8013 
8014 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8015 	if (sb == NULL)
8016 		return (ENOMEM);
8017 
8018 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
8019 
8020 	t4_tp_read_la(sc, buf, NULL);
8021 	p = buf;
8022 
8023 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
8024 	case 2:
8025 		inc = 2;
8026 		show_func = tp_la_show2;
8027 		break;
8028 	case 3:
8029 		inc = 2;
8030 		show_func = tp_la_show3;
8031 		break;
8032 	default:
8033 		inc = 1;
8034 		show_func = tp_la_show;
8035 	}
8036 
8037 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
8038 		(*show_func)(sb, p, i);
8039 
8040 	rc = sbuf_finish(sb);
8041 	sbuf_delete(sb);
8042 	free(buf, M_CXGBE);
8043 	return (rc);
8044 }
8045 
8046 static int
8047 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
8048 {
8049 	struct adapter *sc = arg1;
8050 	struct sbuf *sb;
8051 	int rc;
8052 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
8053 
8054 	rc = sysctl_wire_old_buffer(req, 0);
8055 	if (rc != 0)
8056 		return (rc);
8057 
8058 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8059 	if (sb == NULL)
8060 		return (ENOMEM);
8061 
8062 	t4_get_chan_txrate(sc, nrate, orate);
8063 
8064 	if (sc->chip_params->nchan > 2) {
8065 		sbuf_printf(sb, "              channel 0   channel 1"
8066 		    "   channel 2   channel 3\n");
8067 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
8068 		    nrate[0], nrate[1], nrate[2], nrate[3]);
8069 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
8070 		    orate[0], orate[1], orate[2], orate[3]);
8071 	} else {
8072 		sbuf_printf(sb, "              channel 0   channel 1\n");
8073 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
8074 		    nrate[0], nrate[1]);
8075 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
8076 		    orate[0], orate[1]);
8077 	}
8078 
8079 	rc = sbuf_finish(sb);
8080 	sbuf_delete(sb);
8081 
8082 	return (rc);
8083 }
8084 
8085 static int
8086 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
8087 {
8088 	struct adapter *sc = arg1;
8089 	struct sbuf *sb;
8090 	uint32_t *buf, *p;
8091 	int rc, i;
8092 
8093 	rc = sysctl_wire_old_buffer(req, 0);
8094 	if (rc != 0)
8095 		return (rc);
8096 
8097 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8098 	if (sb == NULL)
8099 		return (ENOMEM);
8100 
8101 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
8102 	    M_ZERO | M_WAITOK);
8103 
8104 	t4_ulprx_read_la(sc, buf);
8105 	p = buf;
8106 
8107 	sbuf_printf(sb, "      Pcmd        Type   Message"
8108 	    "                Data");
8109 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
8110 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
8111 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
8112 	}
8113 
8114 	rc = sbuf_finish(sb);
8115 	sbuf_delete(sb);
8116 	free(buf, M_CXGBE);
8117 	return (rc);
8118 }
8119 
8120 static int
8121 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
8122 {
8123 	struct adapter *sc = arg1;
8124 	struct sbuf *sb;
8125 	int rc, v;
8126 
8127 	MPASS(chip_id(sc) >= CHELSIO_T5);
8128 
8129 	rc = sysctl_wire_old_buffer(req, 0);
8130 	if (rc != 0)
8131 		return (rc);
8132 
8133 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8134 	if (sb == NULL)
8135 		return (ENOMEM);
8136 
8137 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
8138 	if (G_STATSOURCE_T5(v) == 7) {
8139 		int mode;
8140 
8141 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
8142 		if (mode == 0) {
8143 			sbuf_printf(sb, "total %d, incomplete %d",
8144 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8145 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8146 		} else if (mode == 1) {
8147 			sbuf_printf(sb, "total %d, data overflow %d",
8148 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8149 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8150 		} else {
8151 			sbuf_printf(sb, "unknown mode %d", mode);
8152 		}
8153 	}
8154 	rc = sbuf_finish(sb);
8155 	sbuf_delete(sb);
8156 
8157 	return (rc);
8158 }
8159 
8160 static int
8161 sysctl_tc_params(SYSCTL_HANDLER_ARGS)
8162 {
8163 	struct adapter *sc = arg1;
8164 	struct tx_cl_rl_params tc;
8165 	struct sbuf *sb;
8166 	int i, rc, port_id, mbps, gbps;
8167 
8168 	rc = sysctl_wire_old_buffer(req, 0);
8169 	if (rc != 0)
8170 		return (rc);
8171 
8172 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8173 	if (sb == NULL)
8174 		return (ENOMEM);
8175 
8176 	port_id = arg2 >> 16;
8177 	MPASS(port_id < sc->params.nports);
8178 	MPASS(sc->port[port_id] != NULL);
8179 	i = arg2 & 0xffff;
8180 	MPASS(i < sc->chip_params->nsched_cls);
8181 
8182 	mtx_lock(&sc->tc_lock);
8183 	tc = sc->port[port_id]->sched_params->cl_rl[i];
8184 	mtx_unlock(&sc->tc_lock);
8185 
8186 	if (tc.flags & TX_CLRL_ERROR) {
8187 		sbuf_printf(sb, "error");
8188 		goto done;
8189 	}
8190 
8191 	if (tc.ratemode == SCHED_CLASS_RATEMODE_REL) {
8192 		/* XXX: top speed or actual link speed? */
8193 		gbps = port_top_speed(sc->port[port_id]);
8194 		sbuf_printf(sb, " %u%% of %uGbps", tc.maxrate, gbps);
8195 	} else if (tc.ratemode == SCHED_CLASS_RATEMODE_ABS) {
8196 		switch (tc.rateunit) {
8197 		case SCHED_CLASS_RATEUNIT_BITS:
8198 			mbps = tc.maxrate / 1000;
8199 			gbps = tc.maxrate / 1000000;
8200 			if (tc.maxrate == gbps * 1000000)
8201 				sbuf_printf(sb, " %uGbps", gbps);
8202 			else if (tc.maxrate == mbps * 1000)
8203 				sbuf_printf(sb, " %uMbps", mbps);
8204 			else
8205 				sbuf_printf(sb, " %uKbps", tc.maxrate);
8206 			break;
8207 		case SCHED_CLASS_RATEUNIT_PKTS:
8208 			sbuf_printf(sb, " %upps", tc.maxrate);
8209 			break;
8210 		default:
8211 			rc = ENXIO;
8212 			goto done;
8213 		}
8214 	}
8215 
8216 	switch (tc.mode) {
8217 	case SCHED_CLASS_MODE_CLASS:
8218 		sbuf_printf(sb, " aggregate");
8219 		break;
8220 	case SCHED_CLASS_MODE_FLOW:
8221 		sbuf_printf(sb, " per-flow");
8222 		break;
8223 	default:
8224 		rc = ENXIO;
8225 		goto done;
8226 	}
8227 
8228 done:
8229 	if (rc == 0)
8230 		rc = sbuf_finish(sb);
8231 	sbuf_delete(sb);
8232 
8233 	return (rc);
8234 }
8235 #endif
8236 
8237 #ifdef TCP_OFFLOAD
8238 static void
8239 unit_conv(char *buf, size_t len, u_int val, u_int factor)
8240 {
8241 	u_int rem = val % factor;
8242 
8243 	if (rem == 0)
8244 		snprintf(buf, len, "%u", val / factor);
8245 	else {
8246 		while (rem % 10 == 0)
8247 			rem /= 10;
8248 		snprintf(buf, len, "%u.%u", val / factor, rem);
8249 	}
8250 }
8251 
8252 static int
8253 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
8254 {
8255 	struct adapter *sc = arg1;
8256 	char buf[16];
8257 	u_int res, re;
8258 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8259 
8260 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8261 	switch (arg2) {
8262 	case 0:
8263 		/* timer_tick */
8264 		re = G_TIMERRESOLUTION(res);
8265 		break;
8266 	case 1:
8267 		/* TCP timestamp tick */
8268 		re = G_TIMESTAMPRESOLUTION(res);
8269 		break;
8270 	case 2:
8271 		/* DACK tick */
8272 		re = G_DELAYEDACKRESOLUTION(res);
8273 		break;
8274 	default:
8275 		return (EDOOFUS);
8276 	}
8277 
8278 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
8279 
8280 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
8281 }
8282 
8283 static int
8284 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
8285 {
8286 	struct adapter *sc = arg1;
8287 	u_int res, dack_re, v;
8288 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8289 
8290 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8291 	dack_re = G_DELAYEDACKRESOLUTION(res);
8292 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
8293 
8294 	return (sysctl_handle_int(oidp, &v, 0, req));
8295 }
8296 
8297 static int
8298 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
8299 {
8300 	struct adapter *sc = arg1;
8301 	int reg = arg2;
8302 	u_int tre;
8303 	u_long tp_tick_us, v;
8304 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8305 
8306 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
8307 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
8308 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
8309 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
8310 
8311 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
8312 	tp_tick_us = (cclk_ps << tre) / 1000000;
8313 
8314 	if (reg == A_TP_INIT_SRTT)
8315 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
8316 	else
8317 		v = tp_tick_us * t4_read_reg(sc, reg);
8318 
8319 	return (sysctl_handle_long(oidp, &v, 0, req));
8320 }
8321 
8322 /*
8323  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
8324  * passed to this function.
8325  */
8326 static int
8327 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
8328 {
8329 	struct adapter *sc = arg1;
8330 	int idx = arg2;
8331 	u_int v;
8332 
8333 	MPASS(idx >= 0 && idx <= 24);
8334 
8335 	v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
8336 
8337 	return (sysctl_handle_int(oidp, &v, 0, req));
8338 }
8339 
8340 static int
8341 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
8342 {
8343 	struct adapter *sc = arg1;
8344 	int idx = arg2;
8345 	u_int shift, v, r;
8346 
8347 	MPASS(idx >= 0 && idx < 16);
8348 
8349 	r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
8350 	shift = (idx & 3) << 3;
8351 	v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
8352 
8353 	return (sysctl_handle_int(oidp, &v, 0, req));
8354 }
8355 
8356 static int
8357 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
8358 {
8359 	struct vi_info *vi = arg1;
8360 	struct adapter *sc = vi->pi->adapter;
8361 	int idx, rc, i;
8362 	struct sge_ofld_rxq *ofld_rxq;
8363 	uint8_t v;
8364 
8365 	idx = vi->ofld_tmr_idx;
8366 
8367 	rc = sysctl_handle_int(oidp, &idx, 0, req);
8368 	if (rc != 0 || req->newptr == NULL)
8369 		return (rc);
8370 
8371 	if (idx < 0 || idx >= SGE_NTIMERS)
8372 		return (EINVAL);
8373 
8374 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8375 	    "t4otmr");
8376 	if (rc)
8377 		return (rc);
8378 
8379 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
8380 	for_each_ofld_rxq(vi, i, ofld_rxq) {
8381 #ifdef atomic_store_rel_8
8382 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
8383 #else
8384 		ofld_rxq->iq.intr_params = v;
8385 #endif
8386 	}
8387 	vi->ofld_tmr_idx = idx;
8388 
8389 	end_synchronized_op(sc, LOCK_HELD);
8390 	return (0);
8391 }
8392 
8393 static int
8394 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
8395 {
8396 	struct vi_info *vi = arg1;
8397 	struct adapter *sc = vi->pi->adapter;
8398 	int idx, rc;
8399 
8400 	idx = vi->ofld_pktc_idx;
8401 
8402 	rc = sysctl_handle_int(oidp, &idx, 0, req);
8403 	if (rc != 0 || req->newptr == NULL)
8404 		return (rc);
8405 
8406 	if (idx < -1 || idx >= SGE_NCOUNTERS)
8407 		return (EINVAL);
8408 
8409 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8410 	    "t4opktc");
8411 	if (rc)
8412 		return (rc);
8413 
8414 	if (vi->flags & VI_INIT_DONE)
8415 		rc = EBUSY; /* cannot be changed once the queues are created */
8416 	else
8417 		vi->ofld_pktc_idx = idx;
8418 
8419 	end_synchronized_op(sc, LOCK_HELD);
8420 	return (rc);
8421 }
8422 #endif
8423 
8424 static uint32_t
8425 fconf_iconf_to_mode(uint32_t fconf, uint32_t iconf)
8426 {
8427 	uint32_t mode;
8428 
8429 	mode = T4_FILTER_IPv4 | T4_FILTER_IPv6 | T4_FILTER_IP_SADDR |
8430 	    T4_FILTER_IP_DADDR | T4_FILTER_IP_SPORT | T4_FILTER_IP_DPORT;
8431 
8432 	if (fconf & F_FRAGMENTATION)
8433 		mode |= T4_FILTER_IP_FRAGMENT;
8434 
8435 	if (fconf & F_MPSHITTYPE)
8436 		mode |= T4_FILTER_MPS_HIT_TYPE;
8437 
8438 	if (fconf & F_MACMATCH)
8439 		mode |= T4_FILTER_MAC_IDX;
8440 
8441 	if (fconf & F_ETHERTYPE)
8442 		mode |= T4_FILTER_ETH_TYPE;
8443 
8444 	if (fconf & F_PROTOCOL)
8445 		mode |= T4_FILTER_IP_PROTO;
8446 
8447 	if (fconf & F_TOS)
8448 		mode |= T4_FILTER_IP_TOS;
8449 
8450 	if (fconf & F_VLAN)
8451 		mode |= T4_FILTER_VLAN;
8452 
8453 	if (fconf & F_VNIC_ID) {
8454 		mode |= T4_FILTER_VNIC;
8455 		if (iconf & F_VNIC)
8456 			mode |= T4_FILTER_IC_VNIC;
8457 	}
8458 
8459 	if (fconf & F_PORT)
8460 		mode |= T4_FILTER_PORT;
8461 
8462 	if (fconf & F_FCOE)
8463 		mode |= T4_FILTER_FCoE;
8464 
8465 	return (mode);
8466 }
8467 
8468 static uint32_t
8469 mode_to_fconf(uint32_t mode)
8470 {
8471 	uint32_t fconf = 0;
8472 
8473 	if (mode & T4_FILTER_IP_FRAGMENT)
8474 		fconf |= F_FRAGMENTATION;
8475 
8476 	if (mode & T4_FILTER_MPS_HIT_TYPE)
8477 		fconf |= F_MPSHITTYPE;
8478 
8479 	if (mode & T4_FILTER_MAC_IDX)
8480 		fconf |= F_MACMATCH;
8481 
8482 	if (mode & T4_FILTER_ETH_TYPE)
8483 		fconf |= F_ETHERTYPE;
8484 
8485 	if (mode & T4_FILTER_IP_PROTO)
8486 		fconf |= F_PROTOCOL;
8487 
8488 	if (mode & T4_FILTER_IP_TOS)
8489 		fconf |= F_TOS;
8490 
8491 	if (mode & T4_FILTER_VLAN)
8492 		fconf |= F_VLAN;
8493 
8494 	if (mode & T4_FILTER_VNIC)
8495 		fconf |= F_VNIC_ID;
8496 
8497 	if (mode & T4_FILTER_PORT)
8498 		fconf |= F_PORT;
8499 
8500 	if (mode & T4_FILTER_FCoE)
8501 		fconf |= F_FCOE;
8502 
8503 	return (fconf);
8504 }
8505 
8506 static uint32_t
8507 mode_to_iconf(uint32_t mode)
8508 {
8509 
8510 	if (mode & T4_FILTER_IC_VNIC)
8511 		return (F_VNIC);
8512 	return (0);
8513 }
8514 
8515 static int check_fspec_against_fconf_iconf(struct adapter *sc,
8516     struct t4_filter_specification *fs)
8517 {
8518 	struct tp_params *tpp = &sc->params.tp;
8519 	uint32_t fconf = 0;
8520 
8521 	if (fs->val.frag || fs->mask.frag)
8522 		fconf |= F_FRAGMENTATION;
8523 
8524 	if (fs->val.matchtype || fs->mask.matchtype)
8525 		fconf |= F_MPSHITTYPE;
8526 
8527 	if (fs->val.macidx || fs->mask.macidx)
8528 		fconf |= F_MACMATCH;
8529 
8530 	if (fs->val.ethtype || fs->mask.ethtype)
8531 		fconf |= F_ETHERTYPE;
8532 
8533 	if (fs->val.proto || fs->mask.proto)
8534 		fconf |= F_PROTOCOL;
8535 
8536 	if (fs->val.tos || fs->mask.tos)
8537 		fconf |= F_TOS;
8538 
8539 	if (fs->val.vlan_vld || fs->mask.vlan_vld)
8540 		fconf |= F_VLAN;
8541 
8542 	if (fs->val.ovlan_vld || fs->mask.ovlan_vld) {
8543 		fconf |= F_VNIC_ID;
8544 		if (tpp->ingress_config & F_VNIC)
8545 			return (EINVAL);
8546 	}
8547 
8548 	if (fs->val.pfvf_vld || fs->mask.pfvf_vld) {
8549 		fconf |= F_VNIC_ID;
8550 		if ((tpp->ingress_config & F_VNIC) == 0)
8551 			return (EINVAL);
8552 	}
8553 
8554 	if (fs->val.iport || fs->mask.iport)
8555 		fconf |= F_PORT;
8556 
8557 	if (fs->val.fcoe || fs->mask.fcoe)
8558 		fconf |= F_FCOE;
8559 
8560 	if ((tpp->vlan_pri_map | fconf) != tpp->vlan_pri_map)
8561 		return (E2BIG);
8562 
8563 	return (0);
8564 }
8565 
8566 static int
8567 get_filter_mode(struct adapter *sc, uint32_t *mode)
8568 {
8569 	struct tp_params *tpp = &sc->params.tp;
8570 
8571 	/*
8572 	 * We trust the cached values of the relevant TP registers.  This means
8573 	 * things work reliably only if writes to those registers are always via
8574 	 * t4_set_filter_mode.
8575 	 */
8576 	*mode = fconf_iconf_to_mode(tpp->vlan_pri_map, tpp->ingress_config);
8577 
8578 	return (0);
8579 }
8580 
8581 static int
8582 set_filter_mode(struct adapter *sc, uint32_t mode)
8583 {
8584 	struct tp_params *tpp = &sc->params.tp;
8585 	uint32_t fconf, iconf;
8586 	int rc;
8587 
8588 	iconf = mode_to_iconf(mode);
8589 	if ((iconf ^ tpp->ingress_config) & F_VNIC) {
8590 		/*
8591 		 * For now we just complain if A_TP_INGRESS_CONFIG is not
8592 		 * already set to the correct value for the requested filter
8593 		 * mode.  It's not clear if it's safe to write to this register
8594 		 * on the fly.  (And we trust the cached value of the register).
8595 		 */
8596 		return (EBUSY);
8597 	}
8598 
8599 	fconf = mode_to_fconf(mode);
8600 
8601 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8602 	    "t4setfm");
8603 	if (rc)
8604 		return (rc);
8605 
8606 	if (sc->tids.ftids_in_use > 0) {
8607 		rc = EBUSY;
8608 		goto done;
8609 	}
8610 
8611 #ifdef TCP_OFFLOAD
8612 	if (uld_active(sc, ULD_TOM)) {
8613 		rc = EBUSY;
8614 		goto done;
8615 	}
8616 #endif
8617 
8618 	rc = -t4_set_filter_mode(sc, fconf, true);
8619 done:
8620 	end_synchronized_op(sc, LOCK_HELD);
8621 	return (rc);
8622 }
8623 
8624 static inline uint64_t
8625 get_filter_hits(struct adapter *sc, uint32_t fid)
8626 {
8627 	uint32_t tcb_addr;
8628 
8629 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE) +
8630 	    (fid + sc->tids.ftid_base) * TCB_SIZE;
8631 
8632 	if (is_t4(sc)) {
8633 		uint64_t hits;
8634 
8635 		read_via_memwin(sc, 0, tcb_addr + 16, (uint32_t *)&hits, 8);
8636 		return (be64toh(hits));
8637 	} else {
8638 		uint32_t hits;
8639 
8640 		read_via_memwin(sc, 0, tcb_addr + 24, &hits, 4);
8641 		return (be32toh(hits));
8642 	}
8643 }
8644 
8645 static int
8646 get_filter(struct adapter *sc, struct t4_filter *t)
8647 {
8648 	int i, rc, nfilters = sc->tids.nftids;
8649 	struct filter_entry *f;
8650 
8651 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8652 	    "t4getf");
8653 	if (rc)
8654 		return (rc);
8655 
8656 	if (sc->tids.ftids_in_use == 0 || sc->tids.ftid_tab == NULL ||
8657 	    t->idx >= nfilters) {
8658 		t->idx = 0xffffffff;
8659 		goto done;
8660 	}
8661 
8662 	f = &sc->tids.ftid_tab[t->idx];
8663 	for (i = t->idx; i < nfilters; i++, f++) {
8664 		if (f->valid) {
8665 			t->idx = i;
8666 			t->l2tidx = f->l2t ? f->l2t->idx : 0;
8667 			t->smtidx = f->smtidx;
8668 			if (f->fs.hitcnts)
8669 				t->hits = get_filter_hits(sc, t->idx);
8670 			else
8671 				t->hits = UINT64_MAX;
8672 			t->fs = f->fs;
8673 
8674 			goto done;
8675 		}
8676 	}
8677 
8678 	t->idx = 0xffffffff;
8679 done:
8680 	end_synchronized_op(sc, LOCK_HELD);
8681 	return (0);
8682 }
8683 
8684 static int
8685 set_filter(struct adapter *sc, struct t4_filter *t)
8686 {
8687 	unsigned int nfilters, nports;
8688 	struct filter_entry *f;
8689 	int i, rc;
8690 
8691 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setf");
8692 	if (rc)
8693 		return (rc);
8694 
8695 	nfilters = sc->tids.nftids;
8696 	nports = sc->params.nports;
8697 
8698 	if (nfilters == 0) {
8699 		rc = ENOTSUP;
8700 		goto done;
8701 	}
8702 
8703 	if (t->idx >= nfilters) {
8704 		rc = EINVAL;
8705 		goto done;
8706 	}
8707 
8708 	/* Validate against the global filter mode and ingress config */
8709 	rc = check_fspec_against_fconf_iconf(sc, &t->fs);
8710 	if (rc != 0)
8711 		goto done;
8712 
8713 	if (t->fs.action == FILTER_SWITCH && t->fs.eport >= nports) {
8714 		rc = EINVAL;
8715 		goto done;
8716 	}
8717 
8718 	if (t->fs.val.iport >= nports) {
8719 		rc = EINVAL;
8720 		goto done;
8721 	}
8722 
8723 	/* Can't specify an iq if not steering to it */
8724 	if (!t->fs.dirsteer && t->fs.iq) {
8725 		rc = EINVAL;
8726 		goto done;
8727 	}
8728 
8729 	/* IPv6 filter idx must be 4 aligned */
8730 	if (t->fs.type == 1 &&
8731 	    ((t->idx & 0x3) || t->idx + 4 >= nfilters)) {
8732 		rc = EINVAL;
8733 		goto done;
8734 	}
8735 
8736 	if (!(sc->flags & FULL_INIT_DONE) &&
8737 	    ((rc = adapter_full_init(sc)) != 0))
8738 		goto done;
8739 
8740 	if (sc->tids.ftid_tab == NULL) {
8741 		KASSERT(sc->tids.ftids_in_use == 0,
8742 		    ("%s: no memory allocated but filters_in_use > 0",
8743 		    __func__));
8744 
8745 		sc->tids.ftid_tab = malloc(sizeof (struct filter_entry) *
8746 		    nfilters, M_CXGBE, M_NOWAIT | M_ZERO);
8747 		if (sc->tids.ftid_tab == NULL) {
8748 			rc = ENOMEM;
8749 			goto done;
8750 		}
8751 		mtx_init(&sc->tids.ftid_lock, "T4 filters", 0, MTX_DEF);
8752 	}
8753 
8754 	for (i = 0; i < 4; i++) {
8755 		f = &sc->tids.ftid_tab[t->idx + i];
8756 
8757 		if (f->pending || f->valid) {
8758 			rc = EBUSY;
8759 			goto done;
8760 		}
8761 		if (f->locked) {
8762 			rc = EPERM;
8763 			goto done;
8764 		}
8765 
8766 		if (t->fs.type == 0)
8767 			break;
8768 	}
8769 
8770 	f = &sc->tids.ftid_tab[t->idx];
8771 	f->fs = t->fs;
8772 
8773 	rc = set_filter_wr(sc, t->idx);
8774 done:
8775 	end_synchronized_op(sc, 0);
8776 
8777 	if (rc == 0) {
8778 		mtx_lock(&sc->tids.ftid_lock);
8779 		for (;;) {
8780 			if (f->pending == 0) {
8781 				rc = f->valid ? 0 : EIO;
8782 				break;
8783 			}
8784 
8785 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8786 			    PCATCH, "t4setfw", 0)) {
8787 				rc = EINPROGRESS;
8788 				break;
8789 			}
8790 		}
8791 		mtx_unlock(&sc->tids.ftid_lock);
8792 	}
8793 	return (rc);
8794 }
8795 
8796 static int
8797 del_filter(struct adapter *sc, struct t4_filter *t)
8798 {
8799 	unsigned int nfilters;
8800 	struct filter_entry *f;
8801 	int rc;
8802 
8803 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4delf");
8804 	if (rc)
8805 		return (rc);
8806 
8807 	nfilters = sc->tids.nftids;
8808 
8809 	if (nfilters == 0) {
8810 		rc = ENOTSUP;
8811 		goto done;
8812 	}
8813 
8814 	if (sc->tids.ftid_tab == NULL || sc->tids.ftids_in_use == 0 ||
8815 	    t->idx >= nfilters) {
8816 		rc = EINVAL;
8817 		goto done;
8818 	}
8819 
8820 	if (!(sc->flags & FULL_INIT_DONE)) {
8821 		rc = EAGAIN;
8822 		goto done;
8823 	}
8824 
8825 	f = &sc->tids.ftid_tab[t->idx];
8826 
8827 	if (f->pending) {
8828 		rc = EBUSY;
8829 		goto done;
8830 	}
8831 	if (f->locked) {
8832 		rc = EPERM;
8833 		goto done;
8834 	}
8835 
8836 	if (f->valid) {
8837 		t->fs = f->fs;	/* extra info for the caller */
8838 		rc = del_filter_wr(sc, t->idx);
8839 	}
8840 
8841 done:
8842 	end_synchronized_op(sc, 0);
8843 
8844 	if (rc == 0) {
8845 		mtx_lock(&sc->tids.ftid_lock);
8846 		for (;;) {
8847 			if (f->pending == 0) {
8848 				rc = f->valid ? EIO : 0;
8849 				break;
8850 			}
8851 
8852 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8853 			    PCATCH, "t4delfw", 0)) {
8854 				rc = EINPROGRESS;
8855 				break;
8856 			}
8857 		}
8858 		mtx_unlock(&sc->tids.ftid_lock);
8859 	}
8860 
8861 	return (rc);
8862 }
8863 
8864 static void
8865 clear_filter(struct filter_entry *f)
8866 {
8867 	if (f->l2t)
8868 		t4_l2t_release(f->l2t);
8869 
8870 	bzero(f, sizeof (*f));
8871 }
8872 
8873 static int
8874 set_filter_wr(struct adapter *sc, int fidx)
8875 {
8876 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8877 	struct fw_filter_wr *fwr;
8878 	unsigned int ftid, vnic_vld, vnic_vld_mask;
8879 	struct wrq_cookie cookie;
8880 
8881 	ASSERT_SYNCHRONIZED_OP(sc);
8882 
8883 	if (f->fs.newdmac || f->fs.newvlan) {
8884 		/* This filter needs an L2T entry; allocate one. */
8885 		f->l2t = t4_l2t_alloc_switching(sc->l2t);
8886 		if (f->l2t == NULL)
8887 			return (EAGAIN);
8888 		if (t4_l2t_set_switching(sc, f->l2t, f->fs.vlan, f->fs.eport,
8889 		    f->fs.dmac)) {
8890 			t4_l2t_release(f->l2t);
8891 			f->l2t = NULL;
8892 			return (ENOMEM);
8893 		}
8894 	}
8895 
8896 	/* Already validated against fconf, iconf */
8897 	MPASS((f->fs.val.pfvf_vld & f->fs.val.ovlan_vld) == 0);
8898 	MPASS((f->fs.mask.pfvf_vld & f->fs.mask.ovlan_vld) == 0);
8899 	if (f->fs.val.pfvf_vld || f->fs.val.ovlan_vld)
8900 		vnic_vld = 1;
8901 	else
8902 		vnic_vld = 0;
8903 	if (f->fs.mask.pfvf_vld || f->fs.mask.ovlan_vld)
8904 		vnic_vld_mask = 1;
8905 	else
8906 		vnic_vld_mask = 0;
8907 
8908 	ftid = sc->tids.ftid_base + fidx;
8909 
8910 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8911 	if (fwr == NULL)
8912 		return (ENOMEM);
8913 	bzero(fwr, sizeof(*fwr));
8914 
8915 	fwr->op_pkd = htobe32(V_FW_WR_OP(FW_FILTER_WR));
8916 	fwr->len16_pkd = htobe32(FW_LEN16(*fwr));
8917 	fwr->tid_to_iq =
8918 	    htobe32(V_FW_FILTER_WR_TID(ftid) |
8919 		V_FW_FILTER_WR_RQTYPE(f->fs.type) |
8920 		V_FW_FILTER_WR_NOREPLY(0) |
8921 		V_FW_FILTER_WR_IQ(f->fs.iq));
8922 	fwr->del_filter_to_l2tix =
8923 	    htobe32(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
8924 		V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
8925 		V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
8926 		V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
8927 		V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
8928 		V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
8929 		V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
8930 		V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
8931 		V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
8932 		    f->fs.newvlan == VLAN_REWRITE) |
8933 		V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
8934 		    f->fs.newvlan == VLAN_REWRITE) |
8935 		V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
8936 		V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
8937 		V_FW_FILTER_WR_PRIO(f->fs.prio) |
8938 		V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
8939 	fwr->ethtype = htobe16(f->fs.val.ethtype);
8940 	fwr->ethtypem = htobe16(f->fs.mask.ethtype);
8941 	fwr->frag_to_ovlan_vldm =
8942 	    (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
8943 		V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
8944 		V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.vlan_vld) |
8945 		V_FW_FILTER_WR_OVLAN_VLD(vnic_vld) |
8946 		V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.vlan_vld) |
8947 		V_FW_FILTER_WR_OVLAN_VLDM(vnic_vld_mask));
8948 	fwr->smac_sel = 0;
8949 	fwr->rx_chan_rx_rpl_iq = htobe16(V_FW_FILTER_WR_RX_CHAN(0) |
8950 	    V_FW_FILTER_WR_RX_RPL_IQ(sc->sge.fwq.abs_id));
8951 	fwr->maci_to_matchtypem =
8952 	    htobe32(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
8953 		V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
8954 		V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
8955 		V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
8956 		V_FW_FILTER_WR_PORT(f->fs.val.iport) |
8957 		V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
8958 		V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
8959 		V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
8960 	fwr->ptcl = f->fs.val.proto;
8961 	fwr->ptclm = f->fs.mask.proto;
8962 	fwr->ttyp = f->fs.val.tos;
8963 	fwr->ttypm = f->fs.mask.tos;
8964 	fwr->ivlan = htobe16(f->fs.val.vlan);
8965 	fwr->ivlanm = htobe16(f->fs.mask.vlan);
8966 	fwr->ovlan = htobe16(f->fs.val.vnic);
8967 	fwr->ovlanm = htobe16(f->fs.mask.vnic);
8968 	bcopy(f->fs.val.dip, fwr->lip, sizeof (fwr->lip));
8969 	bcopy(f->fs.mask.dip, fwr->lipm, sizeof (fwr->lipm));
8970 	bcopy(f->fs.val.sip, fwr->fip, sizeof (fwr->fip));
8971 	bcopy(f->fs.mask.sip, fwr->fipm, sizeof (fwr->fipm));
8972 	fwr->lp = htobe16(f->fs.val.dport);
8973 	fwr->lpm = htobe16(f->fs.mask.dport);
8974 	fwr->fp = htobe16(f->fs.val.sport);
8975 	fwr->fpm = htobe16(f->fs.mask.sport);
8976 	if (f->fs.newsmac)
8977 		bcopy(f->fs.smac, fwr->sma, sizeof (fwr->sma));
8978 
8979 	f->pending = 1;
8980 	sc->tids.ftids_in_use++;
8981 
8982 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8983 	return (0);
8984 }
8985 
8986 static int
8987 del_filter_wr(struct adapter *sc, int fidx)
8988 {
8989 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8990 	struct fw_filter_wr *fwr;
8991 	unsigned int ftid;
8992 	struct wrq_cookie cookie;
8993 
8994 	ftid = sc->tids.ftid_base + fidx;
8995 
8996 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8997 	if (fwr == NULL)
8998 		return (ENOMEM);
8999 	bzero(fwr, sizeof (*fwr));
9000 
9001 	t4_mk_filtdelwr(ftid, fwr, sc->sge.fwq.abs_id);
9002 
9003 	f->pending = 1;
9004 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
9005 	return (0);
9006 }
9007 
9008 int
9009 t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
9010 {
9011 	struct adapter *sc = iq->adapter;
9012 	const struct cpl_set_tcb_rpl *rpl = (const void *)(rss + 1);
9013 	unsigned int idx = GET_TID(rpl);
9014 	unsigned int rc;
9015 	struct filter_entry *f;
9016 
9017 	KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
9018 	    rss->opcode));
9019 	MPASS(iq == &sc->sge.fwq);
9020 	MPASS(is_ftid(sc, idx));
9021 
9022 	idx -= sc->tids.ftid_base;
9023 	f = &sc->tids.ftid_tab[idx];
9024 	rc = G_COOKIE(rpl->cookie);
9025 
9026 	mtx_lock(&sc->tids.ftid_lock);
9027 	if (rc == FW_FILTER_WR_FLT_ADDED) {
9028 		KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
9029 		    __func__, idx));
9030 		f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
9031 		f->pending = 0;  /* asynchronous setup completed */
9032 		f->valid = 1;
9033 	} else {
9034 		if (rc != FW_FILTER_WR_FLT_DELETED) {
9035 			/* Add or delete failed, display an error */
9036 			log(LOG_ERR,
9037 			    "filter %u setup failed with error %u\n",
9038 			    idx, rc);
9039 		}
9040 
9041 		clear_filter(f);
9042 		sc->tids.ftids_in_use--;
9043 	}
9044 	wakeup(&sc->tids.ftid_tab);
9045 	mtx_unlock(&sc->tids.ftid_lock);
9046 
9047 	return (0);
9048 }
9049 
9050 static int
9051 set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
9052 {
9053 
9054 	MPASS(iq->set_tcb_rpl != NULL);
9055 	return (iq->set_tcb_rpl(iq, rss, m));
9056 }
9057 
9058 static int
9059 l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
9060 {
9061 
9062 	MPASS(iq->l2t_write_rpl != NULL);
9063 	return (iq->l2t_write_rpl(iq, rss, m));
9064 }
9065 
9066 static int
9067 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
9068 {
9069 	int rc;
9070 
9071 	if (cntxt->cid > M_CTXTQID)
9072 		return (EINVAL);
9073 
9074 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
9075 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
9076 		return (EINVAL);
9077 
9078 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
9079 	if (rc)
9080 		return (rc);
9081 
9082 	if (sc->flags & FW_OK) {
9083 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
9084 		    &cntxt->data[0]);
9085 		if (rc == 0)
9086 			goto done;
9087 	}
9088 
9089 	/*
9090 	 * Read via firmware failed or wasn't even attempted.  Read directly via
9091 	 * the backdoor.
9092 	 */
9093 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
9094 done:
9095 	end_synchronized_op(sc, 0);
9096 	return (rc);
9097 }
9098 
9099 static int
9100 load_fw(struct adapter *sc, struct t4_data *fw)
9101 {
9102 	int rc;
9103 	uint8_t *fw_data;
9104 
9105 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
9106 	if (rc)
9107 		return (rc);
9108 
9109 	/*
9110 	 * The firmware, with the sole exception of the memory parity error
9111 	 * handler, runs from memory and not flash.  It is almost always safe to
9112 	 * install a new firmware on a running system.  Just set bit 1 in
9113 	 * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
9114 	 */
9115 	if (sc->flags & FULL_INIT_DONE &&
9116 	    (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
9117 		rc = EBUSY;
9118 		goto done;
9119 	}
9120 
9121 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
9122 	if (fw_data == NULL) {
9123 		rc = ENOMEM;
9124 		goto done;
9125 	}
9126 
9127 	rc = copyin(fw->data, fw_data, fw->len);
9128 	if (rc == 0)
9129 		rc = -t4_load_fw(sc, fw_data, fw->len);
9130 
9131 	free(fw_data, M_CXGBE);
9132 done:
9133 	end_synchronized_op(sc, 0);
9134 	return (rc);
9135 }
9136 
9137 static int
9138 load_cfg(struct adapter *sc, struct t4_data *cfg)
9139 {
9140 	int rc;
9141 	uint8_t *cfg_data = NULL;
9142 
9143 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9144 	if (rc)
9145 		return (rc);
9146 
9147 	if (cfg->len == 0) {
9148 		/* clear */
9149 		rc = -t4_load_cfg(sc, NULL, 0);
9150 		goto done;
9151 	}
9152 
9153 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
9154 	if (cfg_data == NULL) {
9155 		rc = ENOMEM;
9156 		goto done;
9157 	}
9158 
9159 	rc = copyin(cfg->data, cfg_data, cfg->len);
9160 	if (rc == 0)
9161 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
9162 
9163 	free(cfg_data, M_CXGBE);
9164 done:
9165 	end_synchronized_op(sc, 0);
9166 	return (rc);
9167 }
9168 
9169 static int
9170 load_boot(struct adapter *sc, struct t4_bootrom *br)
9171 {
9172 	int rc;
9173 	uint8_t *br_data = NULL;
9174 	u_int offset;
9175 
9176 	if (br->len > 1024 * 1024)
9177 		return (EFBIG);
9178 
9179 	if (br->pf_offset == 0) {
9180 		/* pfidx */
9181 		if (br->pfidx_addr > 7)
9182 			return (EINVAL);
9183 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
9184 		    A_PCIE_PF_EXPROM_OFST)));
9185 	} else if (br->pf_offset == 1) {
9186 		/* offset */
9187 		offset = G_OFFSET(br->pfidx_addr);
9188 	} else {
9189 		return (EINVAL);
9190 	}
9191 
9192 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
9193 	if (rc)
9194 		return (rc);
9195 
9196 	if (br->len == 0) {
9197 		/* clear */
9198 		rc = -t4_load_boot(sc, NULL, offset, 0);
9199 		goto done;
9200 	}
9201 
9202 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
9203 	if (br_data == NULL) {
9204 		rc = ENOMEM;
9205 		goto done;
9206 	}
9207 
9208 	rc = copyin(br->data, br_data, br->len);
9209 	if (rc == 0)
9210 		rc = -t4_load_boot(sc, br_data, offset, br->len);
9211 
9212 	free(br_data, M_CXGBE);
9213 done:
9214 	end_synchronized_op(sc, 0);
9215 	return (rc);
9216 }
9217 
9218 static int
9219 load_bootcfg(struct adapter *sc, struct t4_data *bc)
9220 {
9221 	int rc;
9222 	uint8_t *bc_data = NULL;
9223 
9224 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9225 	if (rc)
9226 		return (rc);
9227 
9228 	if (bc->len == 0) {
9229 		/* clear */
9230 		rc = -t4_load_bootcfg(sc, NULL, 0);
9231 		goto done;
9232 	}
9233 
9234 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
9235 	if (bc_data == NULL) {
9236 		rc = ENOMEM;
9237 		goto done;
9238 	}
9239 
9240 	rc = copyin(bc->data, bc_data, bc->len);
9241 	if (rc == 0)
9242 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
9243 
9244 	free(bc_data, M_CXGBE);
9245 done:
9246 	end_synchronized_op(sc, 0);
9247 	return (rc);
9248 }
9249 
9250 static int
9251 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
9252 {
9253 	int rc;
9254 	struct cudbg_init *cudbg;
9255 	void *handle, *buf;
9256 
9257 	/* buf is large, don't block if no memory is available */
9258 	buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
9259 	if (buf == NULL)
9260 		return (ENOMEM);
9261 
9262 	handle = cudbg_alloc_handle();
9263 	if (handle == NULL) {
9264 		rc = ENOMEM;
9265 		goto done;
9266 	}
9267 
9268 	cudbg = cudbg_get_init(handle);
9269 	cudbg->adap = sc;
9270 	cudbg->print = (cudbg_print_cb)printf;
9271 
9272 #ifndef notyet
9273 	device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
9274 	    __func__, dump->wr_flash, dump->len, dump->data);
9275 #endif
9276 
9277 	if (dump->wr_flash)
9278 		cudbg->use_flash = 1;
9279 	MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
9280 	memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
9281 
9282 	rc = cudbg_collect(handle, buf, &dump->len);
9283 	if (rc != 0)
9284 		goto done;
9285 
9286 	rc = copyout(buf, dump->data, dump->len);
9287 done:
9288 	cudbg_free_handle(handle);
9289 	free(buf, M_CXGBE);
9290 	return (rc);
9291 }
9292 
9293 #define MAX_READ_BUF_SIZE (128 * 1024)
9294 static int
9295 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
9296 {
9297 	uint32_t addr, remaining, n;
9298 	uint32_t *buf;
9299 	int rc;
9300 	uint8_t *dst;
9301 
9302 	rc = validate_mem_range(sc, mr->addr, mr->len);
9303 	if (rc != 0)
9304 		return (rc);
9305 
9306 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
9307 	addr = mr->addr;
9308 	remaining = mr->len;
9309 	dst = (void *)mr->data;
9310 
9311 	while (remaining) {
9312 		n = min(remaining, MAX_READ_BUF_SIZE);
9313 		read_via_memwin(sc, 2, addr, buf, n);
9314 
9315 		rc = copyout(buf, dst, n);
9316 		if (rc != 0)
9317 			break;
9318 
9319 		dst += n;
9320 		remaining -= n;
9321 		addr += n;
9322 	}
9323 
9324 	free(buf, M_CXGBE);
9325 	return (rc);
9326 }
9327 #undef MAX_READ_BUF_SIZE
9328 
9329 static int
9330 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
9331 {
9332 	int rc;
9333 
9334 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
9335 		return (EINVAL);
9336 
9337 	if (i2cd->len > sizeof(i2cd->data))
9338 		return (EFBIG);
9339 
9340 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9341 	if (rc)
9342 		return (rc);
9343 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9344 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
9345 	end_synchronized_op(sc, 0);
9346 
9347 	return (rc);
9348 }
9349 
9350 int
9351 t4_os_find_pci_capability(struct adapter *sc, int cap)
9352 {
9353 	int i;
9354 
9355 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9356 }
9357 
9358 int
9359 t4_os_pci_save_state(struct adapter *sc)
9360 {
9361 	device_t dev;
9362 	struct pci_devinfo *dinfo;
9363 
9364 	dev = sc->dev;
9365 	dinfo = device_get_ivars(dev);
9366 
9367 	pci_cfg_save(dev, dinfo, 0);
9368 	return (0);
9369 }
9370 
9371 int
9372 t4_os_pci_restore_state(struct adapter *sc)
9373 {
9374 	device_t dev;
9375 	struct pci_devinfo *dinfo;
9376 
9377 	dev = sc->dev;
9378 	dinfo = device_get_ivars(dev);
9379 
9380 	pci_cfg_restore(dev, dinfo);
9381 	return (0);
9382 }
9383 
9384 void
9385 t4_os_portmod_changed(struct port_info *pi)
9386 {
9387 	struct adapter *sc = pi->adapter;
9388 	struct vi_info *vi;
9389 	struct ifnet *ifp;
9390 	static const char *mod_str[] = {
9391 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9392 	};
9393 
9394 	PORT_LOCK(pi);
9395 	build_medialist(pi, &pi->media);
9396 	PORT_UNLOCK(pi);
9397 	vi = &pi->vi[0];
9398 	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9399 		init_l1cfg(pi);
9400 		end_synchronized_op(sc, LOCK_HELD);
9401 	}
9402 
9403 	ifp = vi->ifp;
9404 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9405 		if_printf(ifp, "transceiver unplugged.\n");
9406 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9407 		if_printf(ifp, "unknown transceiver inserted.\n");
9408 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9409 		if_printf(ifp, "unsupported transceiver inserted.\n");
9410 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9411 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9412 		    port_top_speed(pi), mod_str[pi->mod_type]);
9413 	} else {
9414 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9415 		    pi->mod_type);
9416 	}
9417 }
9418 
9419 void
9420 t4_os_link_changed(struct port_info *pi)
9421 {
9422 	struct vi_info *vi;
9423 	struct ifnet *ifp;
9424 	struct link_config *lc;
9425 	int v;
9426 
9427 	for_each_vi(pi, v, vi) {
9428 		ifp = vi->ifp;
9429 		if (ifp == NULL)
9430 			continue;
9431 
9432 		lc = &pi->link_cfg;
9433 		if (lc->link_ok) {
9434 			ifp->if_baudrate = IF_Mbps(lc->speed);
9435 			if_link_state_change(ifp, LINK_STATE_UP);
9436 		} else {
9437 			if_link_state_change(ifp, LINK_STATE_DOWN);
9438 		}
9439 	}
9440 }
9441 
9442 void
9443 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9444 {
9445 	struct adapter *sc;
9446 
9447 	sx_slock(&t4_list_lock);
9448 	SLIST_FOREACH(sc, &t4_list, link) {
9449 		/*
9450 		 * func should not make any assumptions about what state sc is
9451 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9452 		 */
9453 		func(sc, arg);
9454 	}
9455 	sx_sunlock(&t4_list_lock);
9456 }
9457 
9458 static int
9459 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9460     struct thread *td)
9461 {
9462 	int rc;
9463 	struct adapter *sc = dev->si_drv1;
9464 
9465 	rc = priv_check(td, PRIV_DRIVER);
9466 	if (rc != 0)
9467 		return (rc);
9468 
9469 	switch (cmd) {
9470 	case CHELSIO_T4_GETREG: {
9471 		struct t4_reg *edata = (struct t4_reg *)data;
9472 
9473 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9474 			return (EFAULT);
9475 
9476 		if (edata->size == 4)
9477 			edata->val = t4_read_reg(sc, edata->addr);
9478 		else if (edata->size == 8)
9479 			edata->val = t4_read_reg64(sc, edata->addr);
9480 		else
9481 			return (EINVAL);
9482 
9483 		break;
9484 	}
9485 	case CHELSIO_T4_SETREG: {
9486 		struct t4_reg *edata = (struct t4_reg *)data;
9487 
9488 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9489 			return (EFAULT);
9490 
9491 		if (edata->size == 4) {
9492 			if (edata->val & 0xffffffff00000000)
9493 				return (EINVAL);
9494 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9495 		} else if (edata->size == 8)
9496 			t4_write_reg64(sc, edata->addr, edata->val);
9497 		else
9498 			return (EINVAL);
9499 		break;
9500 	}
9501 	case CHELSIO_T4_REGDUMP: {
9502 		struct t4_regdump *regs = (struct t4_regdump *)data;
9503 		int reglen = t4_get_regs_len(sc);
9504 		uint8_t *buf;
9505 
9506 		if (regs->len < reglen) {
9507 			regs->len = reglen; /* hint to the caller */
9508 			return (ENOBUFS);
9509 		}
9510 
9511 		regs->len = reglen;
9512 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9513 		get_regs(sc, regs, buf);
9514 		rc = copyout(buf, regs->data, reglen);
9515 		free(buf, M_CXGBE);
9516 		break;
9517 	}
9518 	case CHELSIO_T4_GET_FILTER_MODE:
9519 		rc = get_filter_mode(sc, (uint32_t *)data);
9520 		break;
9521 	case CHELSIO_T4_SET_FILTER_MODE:
9522 		rc = set_filter_mode(sc, *(uint32_t *)data);
9523 		break;
9524 	case CHELSIO_T4_GET_FILTER:
9525 		rc = get_filter(sc, (struct t4_filter *)data);
9526 		break;
9527 	case CHELSIO_T4_SET_FILTER:
9528 		rc = set_filter(sc, (struct t4_filter *)data);
9529 		break;
9530 	case CHELSIO_T4_DEL_FILTER:
9531 		rc = del_filter(sc, (struct t4_filter *)data);
9532 		break;
9533 	case CHELSIO_T4_GET_SGE_CONTEXT:
9534 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9535 		break;
9536 	case CHELSIO_T4_LOAD_FW:
9537 		rc = load_fw(sc, (struct t4_data *)data);
9538 		break;
9539 	case CHELSIO_T4_GET_MEM:
9540 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9541 		break;
9542 	case CHELSIO_T4_GET_I2C:
9543 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9544 		break;
9545 	case CHELSIO_T4_CLEAR_STATS: {
9546 		int i, v;
9547 		u_int port_id = *(uint32_t *)data;
9548 		struct port_info *pi;
9549 		struct vi_info *vi;
9550 
9551 		if (port_id >= sc->params.nports)
9552 			return (EINVAL);
9553 		pi = sc->port[port_id];
9554 		if (pi == NULL)
9555 			return (EIO);
9556 
9557 		/* MAC stats */
9558 		t4_clr_port_stats(sc, pi->tx_chan);
9559 		pi->tx_parse_error = 0;
9560 		mtx_lock(&sc->reg_lock);
9561 		for_each_vi(pi, v, vi) {
9562 			if (vi->flags & VI_INIT_DONE)
9563 				t4_clr_vi_stats(sc, vi->viid);
9564 		}
9565 		mtx_unlock(&sc->reg_lock);
9566 
9567 		/*
9568 		 * Since this command accepts a port, clear stats for
9569 		 * all VIs on this port.
9570 		 */
9571 		for_each_vi(pi, v, vi) {
9572 			if (vi->flags & VI_INIT_DONE) {
9573 				struct sge_rxq *rxq;
9574 				struct sge_txq *txq;
9575 				struct sge_wrq *wrq;
9576 
9577 				for_each_rxq(vi, i, rxq) {
9578 #if defined(INET) || defined(INET6)
9579 					rxq->lro.lro_queued = 0;
9580 					rxq->lro.lro_flushed = 0;
9581 #endif
9582 					rxq->rxcsum = 0;
9583 					rxq->vlan_extraction = 0;
9584 				}
9585 
9586 				for_each_txq(vi, i, txq) {
9587 					txq->txcsum = 0;
9588 					txq->tso_wrs = 0;
9589 					txq->vlan_insertion = 0;
9590 					txq->imm_wrs = 0;
9591 					txq->sgl_wrs = 0;
9592 					txq->txpkt_wrs = 0;
9593 					txq->txpkts0_wrs = 0;
9594 					txq->txpkts1_wrs = 0;
9595 					txq->txpkts0_pkts = 0;
9596 					txq->txpkts1_pkts = 0;
9597 					mp_ring_reset_stats(txq->r);
9598 				}
9599 
9600 #ifdef TCP_OFFLOAD
9601 				/* nothing to clear for each ofld_rxq */
9602 
9603 				for_each_ofld_txq(vi, i, wrq) {
9604 					wrq->tx_wrs_direct = 0;
9605 					wrq->tx_wrs_copied = 0;
9606 				}
9607 #endif
9608 
9609 				if (IS_MAIN_VI(vi)) {
9610 					wrq = &sc->sge.ctrlq[pi->port_id];
9611 					wrq->tx_wrs_direct = 0;
9612 					wrq->tx_wrs_copied = 0;
9613 				}
9614 			}
9615 		}
9616 		break;
9617 	}
9618 	case CHELSIO_T4_SCHED_CLASS:
9619 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9620 		break;
9621 	case CHELSIO_T4_SCHED_QUEUE:
9622 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9623 		break;
9624 	case CHELSIO_T4_GET_TRACER:
9625 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9626 		break;
9627 	case CHELSIO_T4_SET_TRACER:
9628 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9629 		break;
9630 	case CHELSIO_T4_LOAD_CFG:
9631 		rc = load_cfg(sc, (struct t4_data *)data);
9632 		break;
9633 	case CHELSIO_T4_LOAD_BOOT:
9634 		rc = load_boot(sc, (struct t4_bootrom *)data);
9635 		break;
9636 	case CHELSIO_T4_LOAD_BOOTCFG:
9637 		rc = load_bootcfg(sc, (struct t4_data *)data);
9638 		break;
9639 	case CHELSIO_T4_CUDBG_DUMP:
9640 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
9641 		break;
9642 	default:
9643 		rc = ENOTTY;
9644 	}
9645 
9646 	return (rc);
9647 }
9648 
9649 void
9650 t4_db_full(struct adapter *sc)
9651 {
9652 
9653 	CXGBE_UNIMPLEMENTED(__func__);
9654 }
9655 
9656 void
9657 t4_db_dropped(struct adapter *sc)
9658 {
9659 
9660 	CXGBE_UNIMPLEMENTED(__func__);
9661 }
9662 
9663 #ifdef TCP_OFFLOAD
9664 static int
9665 toe_capability(struct vi_info *vi, int enable)
9666 {
9667 	int rc;
9668 	struct port_info *pi = vi->pi;
9669 	struct adapter *sc = pi->adapter;
9670 
9671 	ASSERT_SYNCHRONIZED_OP(sc);
9672 
9673 	if (!is_offload(sc))
9674 		return (ENODEV);
9675 
9676 	if (enable) {
9677 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9678 			/* TOE is already enabled. */
9679 			return (0);
9680 		}
9681 
9682 		/*
9683 		 * We need the port's queues around so that we're able to send
9684 		 * and receive CPLs to/from the TOE even if the ifnet for this
9685 		 * port has never been UP'd administratively.
9686 		 */
9687 		if (!(vi->flags & VI_INIT_DONE)) {
9688 			rc = vi_full_init(vi);
9689 			if (rc)
9690 				return (rc);
9691 		}
9692 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9693 			rc = vi_full_init(&pi->vi[0]);
9694 			if (rc)
9695 				return (rc);
9696 		}
9697 
9698 		if (isset(&sc->offload_map, pi->port_id)) {
9699 			/* TOE is enabled on another VI of this port. */
9700 			pi->uld_vis++;
9701 			return (0);
9702 		}
9703 
9704 		if (!uld_active(sc, ULD_TOM)) {
9705 			rc = t4_activate_uld(sc, ULD_TOM);
9706 			if (rc == EAGAIN) {
9707 				log(LOG_WARNING,
9708 				    "You must kldload t4_tom.ko before trying "
9709 				    "to enable TOE on a cxgbe interface.\n");
9710 			}
9711 			if (rc != 0)
9712 				return (rc);
9713 			KASSERT(sc->tom_softc != NULL,
9714 			    ("%s: TOM activated but softc NULL", __func__));
9715 			KASSERT(uld_active(sc, ULD_TOM),
9716 			    ("%s: TOM activated but flag not set", __func__));
9717 		}
9718 
9719 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
9720 		if (!uld_active(sc, ULD_IWARP))
9721 			(void) t4_activate_uld(sc, ULD_IWARP);
9722 		if (!uld_active(sc, ULD_ISCSI))
9723 			(void) t4_activate_uld(sc, ULD_ISCSI);
9724 
9725 		pi->uld_vis++;
9726 		setbit(&sc->offload_map, pi->port_id);
9727 	} else {
9728 		pi->uld_vis--;
9729 
9730 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9731 			return (0);
9732 
9733 		KASSERT(uld_active(sc, ULD_TOM),
9734 		    ("%s: TOM never initialized?", __func__));
9735 		clrbit(&sc->offload_map, pi->port_id);
9736 	}
9737 
9738 	return (0);
9739 }
9740 
9741 /*
9742  * Add an upper layer driver to the global list.
9743  */
9744 int
9745 t4_register_uld(struct uld_info *ui)
9746 {
9747 	int rc = 0;
9748 	struct uld_info *u;
9749 
9750 	sx_xlock(&t4_uld_list_lock);
9751 	SLIST_FOREACH(u, &t4_uld_list, link) {
9752 	    if (u->uld_id == ui->uld_id) {
9753 		    rc = EEXIST;
9754 		    goto done;
9755 	    }
9756 	}
9757 
9758 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9759 	ui->refcount = 0;
9760 done:
9761 	sx_xunlock(&t4_uld_list_lock);
9762 	return (rc);
9763 }
9764 
9765 int
9766 t4_unregister_uld(struct uld_info *ui)
9767 {
9768 	int rc = EINVAL;
9769 	struct uld_info *u;
9770 
9771 	sx_xlock(&t4_uld_list_lock);
9772 
9773 	SLIST_FOREACH(u, &t4_uld_list, link) {
9774 	    if (u == ui) {
9775 		    if (ui->refcount > 0) {
9776 			    rc = EBUSY;
9777 			    goto done;
9778 		    }
9779 
9780 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9781 		    rc = 0;
9782 		    goto done;
9783 	    }
9784 	}
9785 done:
9786 	sx_xunlock(&t4_uld_list_lock);
9787 	return (rc);
9788 }
9789 
9790 int
9791 t4_activate_uld(struct adapter *sc, int id)
9792 {
9793 	int rc;
9794 	struct uld_info *ui;
9795 
9796 	ASSERT_SYNCHRONIZED_OP(sc);
9797 
9798 	if (id < 0 || id > ULD_MAX)
9799 		return (EINVAL);
9800 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
9801 
9802 	sx_slock(&t4_uld_list_lock);
9803 
9804 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9805 		if (ui->uld_id == id) {
9806 			if (!(sc->flags & FULL_INIT_DONE)) {
9807 				rc = adapter_full_init(sc);
9808 				if (rc != 0)
9809 					break;
9810 			}
9811 
9812 			rc = ui->activate(sc);
9813 			if (rc == 0) {
9814 				setbit(&sc->active_ulds, id);
9815 				ui->refcount++;
9816 			}
9817 			break;
9818 		}
9819 	}
9820 
9821 	sx_sunlock(&t4_uld_list_lock);
9822 
9823 	return (rc);
9824 }
9825 
9826 int
9827 t4_deactivate_uld(struct adapter *sc, int id)
9828 {
9829 	int rc;
9830 	struct uld_info *ui;
9831 
9832 	ASSERT_SYNCHRONIZED_OP(sc);
9833 
9834 	if (id < 0 || id > ULD_MAX)
9835 		return (EINVAL);
9836 	rc = ENXIO;
9837 
9838 	sx_slock(&t4_uld_list_lock);
9839 
9840 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9841 		if (ui->uld_id == id) {
9842 			rc = ui->deactivate(sc);
9843 			if (rc == 0) {
9844 				clrbit(&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 uld_active(struct adapter *sc, int uld_id)
9858 {
9859 
9860 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9861 
9862 	return (isset(&sc->active_ulds, uld_id));
9863 }
9864 #endif
9865 
9866 /*
9867  * t  = ptr to tunable.
9868  * nc = number of CPUs.
9869  * c  = compiled in default for that tunable.
9870  */
9871 static void
9872 calculate_nqueues(int *t, int nc, const int c)
9873 {
9874 	int nq;
9875 
9876 	if (*t > 0)
9877 		return;
9878 	nq = *t < 0 ? -*t : c;
9879 	*t = min(nc, nq);
9880 }
9881 
9882 /*
9883  * Come up with reasonable defaults for some of the tunables, provided they're
9884  * not set by the user (in which case we'll use the values as is).
9885  */
9886 static void
9887 tweak_tunables(void)
9888 {
9889 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
9890 
9891 	if (t4_ntxq < 1) {
9892 #ifdef RSS
9893 		t4_ntxq = rss_getnumbuckets();
9894 #else
9895 		calculate_nqueues(&t4_ntxq, nc, NTXQ);
9896 #endif
9897 	}
9898 
9899 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
9900 
9901 	if (t4_nrxq < 1) {
9902 #ifdef RSS
9903 		t4_nrxq = rss_getnumbuckets();
9904 #else
9905 		calculate_nqueues(&t4_nrxq, nc, NRXQ);
9906 #endif
9907 	}
9908 
9909 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
9910 
9911 #ifdef TCP_OFFLOAD
9912 	calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
9913 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
9914 	calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
9915 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
9916 
9917 	if (t4_toecaps_allowed == -1)
9918 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9919 
9920 	if (t4_rdmacaps_allowed == -1) {
9921 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9922 		    FW_CAPS_CONFIG_RDMA_RDMAC;
9923 	}
9924 
9925 	if (t4_iscsicaps_allowed == -1) {
9926 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9927 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9928 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
9929 	}
9930 
9931 	if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
9932 		t4_tmr_idx_ofld = TMR_IDX_OFLD;
9933 
9934 	if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
9935 		t4_pktc_idx_ofld = PKTC_IDX_OFLD;
9936 #else
9937 	if (t4_toecaps_allowed == -1)
9938 		t4_toecaps_allowed = 0;
9939 
9940 	if (t4_rdmacaps_allowed == -1)
9941 		t4_rdmacaps_allowed = 0;
9942 
9943 	if (t4_iscsicaps_allowed == -1)
9944 		t4_iscsicaps_allowed = 0;
9945 #endif
9946 
9947 #ifdef DEV_NETMAP
9948 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
9949 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
9950 #endif
9951 
9952 	if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
9953 		t4_tmr_idx = TMR_IDX;
9954 
9955 	if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
9956 		t4_pktc_idx = PKTC_IDX;
9957 
9958 	if (t4_qsize_txq < 128)
9959 		t4_qsize_txq = 128;
9960 
9961 	if (t4_qsize_rxq < 128)
9962 		t4_qsize_rxq = 128;
9963 	while (t4_qsize_rxq & 7)
9964 		t4_qsize_rxq++;
9965 
9966 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9967 
9968 	/*
9969 	 * Number of VIs to create per-port.  The first VI is the "main" regular
9970 	 * VI for the port.  The rest are additional virtual interfaces on the
9971 	 * same physical port.  Note that the main VI does not have native
9972 	 * netmap support but the extra VIs do.
9973 	 *
9974 	 * Limit the number of VIs per port to the number of available
9975 	 * MAC addresses per port.
9976 	 */
9977 	if (t4_num_vis < 1)
9978 		t4_num_vis = 1;
9979 	if (t4_num_vis > nitems(vi_mac_funcs)) {
9980 		t4_num_vis = nitems(vi_mac_funcs);
9981 		printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
9982 	}
9983 
9984 	if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
9985 		pcie_relaxed_ordering = 1;
9986 #if defined(__i386__) || defined(__amd64__)
9987 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
9988 			pcie_relaxed_ordering = 0;
9989 #endif
9990 	}
9991 }
9992 
9993 #ifdef DDB
9994 static void
9995 t4_dump_tcb(struct adapter *sc, int tid)
9996 {
9997 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
9998 
9999 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
10000 	save = t4_read_reg(sc, reg);
10001 	base = sc->memwin[2].mw_base;
10002 
10003 	/* Dump TCB for the tid */
10004 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
10005 	tcb_addr += tid * TCB_SIZE;
10006 
10007 	if (is_t4(sc)) {
10008 		pf = 0;
10009 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
10010 	} else {
10011 		pf = V_PFNUM(sc->pf);
10012 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
10013 	}
10014 	t4_write_reg(sc, reg, win_pos | pf);
10015 	t4_read_reg(sc, reg);
10016 
10017 	off = tcb_addr - win_pos;
10018 	for (i = 0; i < 4; i++) {
10019 		uint32_t buf[8];
10020 		for (j = 0; j < 8; j++, off += 4)
10021 			buf[j] = htonl(t4_read_reg(sc, base + off));
10022 
10023 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
10024 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
10025 		    buf[7]);
10026 	}
10027 
10028 	t4_write_reg(sc, reg, save);
10029 	t4_read_reg(sc, reg);
10030 }
10031 
10032 static void
10033 t4_dump_devlog(struct adapter *sc)
10034 {
10035 	struct devlog_params *dparams = &sc->params.devlog;
10036 	struct fw_devlog_e e;
10037 	int i, first, j, m, nentries, rc;
10038 	uint64_t ftstamp = UINT64_MAX;
10039 
10040 	if (dparams->start == 0) {
10041 		db_printf("devlog params not valid\n");
10042 		return;
10043 	}
10044 
10045 	nentries = dparams->size / sizeof(struct fw_devlog_e);
10046 	m = fwmtype_to_hwmtype(dparams->memtype);
10047 
10048 	/* Find the first entry. */
10049 	first = -1;
10050 	for (i = 0; i < nentries && !db_pager_quit; i++) {
10051 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10052 		    sizeof(e), (void *)&e);
10053 		if (rc != 0)
10054 			break;
10055 
10056 		if (e.timestamp == 0)
10057 			break;
10058 
10059 		e.timestamp = be64toh(e.timestamp);
10060 		if (e.timestamp < ftstamp) {
10061 			ftstamp = e.timestamp;
10062 			first = i;
10063 		}
10064 	}
10065 
10066 	if (first == -1)
10067 		return;
10068 
10069 	i = first;
10070 	do {
10071 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10072 		    sizeof(e), (void *)&e);
10073 		if (rc != 0)
10074 			return;
10075 
10076 		if (e.timestamp == 0)
10077 			return;
10078 
10079 		e.timestamp = be64toh(e.timestamp);
10080 		e.seqno = be32toh(e.seqno);
10081 		for (j = 0; j < 8; j++)
10082 			e.params[j] = be32toh(e.params[j]);
10083 
10084 		db_printf("%10d  %15ju  %8s  %8s  ",
10085 		    e.seqno, e.timestamp,
10086 		    (e.level < nitems(devlog_level_strings) ?
10087 			devlog_level_strings[e.level] : "UNKNOWN"),
10088 		    (e.facility < nitems(devlog_facility_strings) ?
10089 			devlog_facility_strings[e.facility] : "UNKNOWN"));
10090 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
10091 		    e.params[3], e.params[4], e.params[5], e.params[6],
10092 		    e.params[7]);
10093 
10094 		if (++i == nentries)
10095 			i = 0;
10096 	} while (i != first && !db_pager_quit);
10097 }
10098 
10099 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
10100 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
10101 
10102 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
10103 {
10104 	device_t dev;
10105 	int t;
10106 	bool valid;
10107 
10108 	valid = false;
10109 	t = db_read_token();
10110 	if (t == tIDENT) {
10111 		dev = device_lookup_by_name(db_tok_string);
10112 		valid = true;
10113 	}
10114 	db_skip_to_eol();
10115 	if (!valid) {
10116 		db_printf("usage: show t4 devlog <nexus>\n");
10117 		return;
10118 	}
10119 
10120 	if (dev == NULL) {
10121 		db_printf("device not found\n");
10122 		return;
10123 	}
10124 
10125 	t4_dump_devlog(device_get_softc(dev));
10126 }
10127 
10128 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
10129 {
10130 	device_t dev;
10131 	int radix, tid, t;
10132 	bool valid;
10133 
10134 	valid = false;
10135 	radix = db_radix;
10136 	db_radix = 10;
10137 	t = db_read_token();
10138 	if (t == tIDENT) {
10139 		dev = device_lookup_by_name(db_tok_string);
10140 		t = db_read_token();
10141 		if (t == tNUMBER) {
10142 			tid = db_tok_number;
10143 			valid = true;
10144 		}
10145 	}
10146 	db_radix = radix;
10147 	db_skip_to_eol();
10148 	if (!valid) {
10149 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
10150 		return;
10151 	}
10152 
10153 	if (dev == NULL) {
10154 		db_printf("device not found\n");
10155 		return;
10156 	}
10157 	if (tid < 0) {
10158 		db_printf("invalid tid\n");
10159 		return;
10160 	}
10161 
10162 	t4_dump_tcb(device_get_softc(dev), tid);
10163 }
10164 #endif
10165 
10166 static struct sx mlu;	/* mod load unload */
10167 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
10168 
10169 static int
10170 mod_event(module_t mod, int cmd, void *arg)
10171 {
10172 	int rc = 0;
10173 	static int loaded = 0;
10174 
10175 	switch (cmd) {
10176 	case MOD_LOAD:
10177 		sx_xlock(&mlu);
10178 		if (loaded++ == 0) {
10179 			t4_sge_modload();
10180 			t4_register_cpl_handler(CPL_SET_TCB_RPL, set_tcb_rpl);
10181 			t4_register_cpl_handler(CPL_L2T_WRITE_RPL, l2t_write_rpl);
10182 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
10183 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
10184 			sx_init(&t4_list_lock, "T4/T5 adapters");
10185 			SLIST_INIT(&t4_list);
10186 #ifdef TCP_OFFLOAD
10187 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
10188 			SLIST_INIT(&t4_uld_list);
10189 #endif
10190 			t4_tracer_modload();
10191 			tweak_tunables();
10192 		}
10193 		sx_xunlock(&mlu);
10194 		break;
10195 
10196 	case MOD_UNLOAD:
10197 		sx_xlock(&mlu);
10198 		if (--loaded == 0) {
10199 			int tries;
10200 
10201 			sx_slock(&t4_list_lock);
10202 			if (!SLIST_EMPTY(&t4_list)) {
10203 				rc = EBUSY;
10204 				sx_sunlock(&t4_list_lock);
10205 				goto done_unload;
10206 			}
10207 #ifdef TCP_OFFLOAD
10208 			sx_slock(&t4_uld_list_lock);
10209 			if (!SLIST_EMPTY(&t4_uld_list)) {
10210 				rc = EBUSY;
10211 				sx_sunlock(&t4_uld_list_lock);
10212 				sx_sunlock(&t4_list_lock);
10213 				goto done_unload;
10214 			}
10215 #endif
10216 			tries = 0;
10217 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
10218 				uprintf("%ju clusters with custom free routine "
10219 				    "still is use.\n", t4_sge_extfree_refs());
10220 				pause("t4unload", 2 * hz);
10221 			}
10222 #ifdef TCP_OFFLOAD
10223 			sx_sunlock(&t4_uld_list_lock);
10224 #endif
10225 			sx_sunlock(&t4_list_lock);
10226 
10227 			if (t4_sge_extfree_refs() == 0) {
10228 				t4_tracer_modunload();
10229 #ifdef TCP_OFFLOAD
10230 				sx_destroy(&t4_uld_list_lock);
10231 #endif
10232 				sx_destroy(&t4_list_lock);
10233 				t4_sge_modunload();
10234 				loaded = 0;
10235 			} else {
10236 				rc = EBUSY;
10237 				loaded++;	/* undo earlier decrement */
10238 			}
10239 		}
10240 done_unload:
10241 		sx_xunlock(&mlu);
10242 		break;
10243 	}
10244 
10245 	return (rc);
10246 }
10247 
10248 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10249 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10250 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10251 
10252 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10253 MODULE_VERSION(t4nex, 1);
10254 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10255 #ifdef DEV_NETMAP
10256 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10257 #endif /* DEV_NETMAP */
10258 
10259 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10260 MODULE_VERSION(t5nex, 1);
10261 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10262 #ifdef DEV_NETMAP
10263 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10264 #endif /* DEV_NETMAP */
10265 
10266 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10267 MODULE_VERSION(t6nex, 1);
10268 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10269 #ifdef DEV_NETMAP
10270 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10271 #endif /* DEV_NETMAP */
10272 
10273 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10274 MODULE_VERSION(cxgbe, 1);
10275 
10276 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10277 MODULE_VERSION(cxl, 1);
10278 
10279 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10280 MODULE_VERSION(cc, 1);
10281 
10282 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10283 MODULE_VERSION(vcxgbe, 1);
10284 
10285 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10286 MODULE_VERSION(vcxl, 1);
10287 
10288 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10289 MODULE_VERSION(vcc, 1);
10290