xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision 752d135e0dacd9a463d24ffb89779b67ce0a7ea0)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011 Chelsio Communications, Inc.
5  * All rights reserved.
6  * Written by: Navdeep Parhar <np@FreeBSD.org>
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_ratelimit.h"
37 #include "opt_rss.h"
38 
39 #include <sys/param.h>
40 #include <sys/conf.h>
41 #include <sys/priv.h>
42 #include <sys/kernel.h>
43 #include <sys/bus.h>
44 #include <sys/module.h>
45 #include <sys/malloc.h>
46 #include <sys/queue.h>
47 #include <sys/taskqueue.h>
48 #include <sys/pciio.h>
49 #include <dev/pci/pcireg.h>
50 #include <dev/pci/pcivar.h>
51 #include <dev/pci/pci_private.h>
52 #include <sys/firmware.h>
53 #include <sys/sbuf.h>
54 #include <sys/smp.h>
55 #include <sys/socket.h>
56 #include <sys/sockio.h>
57 #include <sys/sysctl.h>
58 #include <net/ethernet.h>
59 #include <net/if.h>
60 #include <net/if_types.h>
61 #include <net/if_dl.h>
62 #include <net/if_vlan_var.h>
63 #ifdef RSS
64 #include <net/rss_config.h>
65 #endif
66 #include <netinet/in.h>
67 #include <netinet/ip.h>
68 #if defined(__i386__) || defined(__amd64__)
69 #include <machine/md_var.h>
70 #include <machine/cputypes.h>
71 #include <vm/vm.h>
72 #include <vm/pmap.h>
73 #endif
74 #include <crypto/rijndael/rijndael.h>
75 #ifdef DDB
76 #include <ddb/ddb.h>
77 #include <ddb/db_lex.h>
78 #endif
79 
80 #include "common/common.h"
81 #include "common/t4_msg.h"
82 #include "common/t4_regs.h"
83 #include "common/t4_regs_values.h"
84 #include "cudbg/cudbg.h"
85 #include "t4_ioctl.h"
86 #include "t4_l2t.h"
87 #include "t4_mp_ring.h"
88 #include "t4_if.h"
89 #include "t4_smt.h"
90 
91 /* T4 bus driver interface */
92 static int t4_probe(device_t);
93 static int t4_attach(device_t);
94 static int t4_detach(device_t);
95 static int t4_child_location_str(device_t, device_t, char *, size_t);
96 static int t4_ready(device_t);
97 static int t4_read_port_device(device_t, int, device_t *);
98 static device_method_t t4_methods[] = {
99 	DEVMETHOD(device_probe,		t4_probe),
100 	DEVMETHOD(device_attach,	t4_attach),
101 	DEVMETHOD(device_detach,	t4_detach),
102 
103 	DEVMETHOD(bus_child_location_str, t4_child_location_str),
104 
105 	DEVMETHOD(t4_is_main_ready,	t4_ready),
106 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
107 
108 	DEVMETHOD_END
109 };
110 static driver_t t4_driver = {
111 	"t4nex",
112 	t4_methods,
113 	sizeof(struct adapter)
114 };
115 
116 
117 /* T4 port (cxgbe) interface */
118 static int cxgbe_probe(device_t);
119 static int cxgbe_attach(device_t);
120 static int cxgbe_detach(device_t);
121 device_method_t cxgbe_methods[] = {
122 	DEVMETHOD(device_probe,		cxgbe_probe),
123 	DEVMETHOD(device_attach,	cxgbe_attach),
124 	DEVMETHOD(device_detach,	cxgbe_detach),
125 	{ 0, 0 }
126 };
127 static driver_t cxgbe_driver = {
128 	"cxgbe",
129 	cxgbe_methods,
130 	sizeof(struct port_info)
131 };
132 
133 /* T4 VI (vcxgbe) interface */
134 static int vcxgbe_probe(device_t);
135 static int vcxgbe_attach(device_t);
136 static int vcxgbe_detach(device_t);
137 static device_method_t vcxgbe_methods[] = {
138 	DEVMETHOD(device_probe,		vcxgbe_probe),
139 	DEVMETHOD(device_attach,	vcxgbe_attach),
140 	DEVMETHOD(device_detach,	vcxgbe_detach),
141 	{ 0, 0 }
142 };
143 static driver_t vcxgbe_driver = {
144 	"vcxgbe",
145 	vcxgbe_methods,
146 	sizeof(struct vi_info)
147 };
148 
149 static d_ioctl_t t4_ioctl;
150 
151 static struct cdevsw t4_cdevsw = {
152        .d_version = D_VERSION,
153        .d_ioctl = t4_ioctl,
154        .d_name = "t4nex",
155 };
156 
157 /* T5 bus driver interface */
158 static int t5_probe(device_t);
159 static device_method_t t5_methods[] = {
160 	DEVMETHOD(device_probe,		t5_probe),
161 	DEVMETHOD(device_attach,	t4_attach),
162 	DEVMETHOD(device_detach,	t4_detach),
163 
164 	DEVMETHOD(bus_child_location_str, t4_child_location_str),
165 
166 	DEVMETHOD(t4_is_main_ready,	t4_ready),
167 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
168 
169 	DEVMETHOD_END
170 };
171 static driver_t t5_driver = {
172 	"t5nex",
173 	t5_methods,
174 	sizeof(struct adapter)
175 };
176 
177 
178 /* T5 port (cxl) interface */
179 static driver_t cxl_driver = {
180 	"cxl",
181 	cxgbe_methods,
182 	sizeof(struct port_info)
183 };
184 
185 /* T5 VI (vcxl) interface */
186 static driver_t vcxl_driver = {
187 	"vcxl",
188 	vcxgbe_methods,
189 	sizeof(struct vi_info)
190 };
191 
192 /* T6 bus driver interface */
193 static int t6_probe(device_t);
194 static device_method_t t6_methods[] = {
195 	DEVMETHOD(device_probe,		t6_probe),
196 	DEVMETHOD(device_attach,	t4_attach),
197 	DEVMETHOD(device_detach,	t4_detach),
198 
199 	DEVMETHOD(bus_child_location_str, t4_child_location_str),
200 
201 	DEVMETHOD(t4_is_main_ready,	t4_ready),
202 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
203 
204 	DEVMETHOD_END
205 };
206 static driver_t t6_driver = {
207 	"t6nex",
208 	t6_methods,
209 	sizeof(struct adapter)
210 };
211 
212 
213 /* T6 port (cc) interface */
214 static driver_t cc_driver = {
215 	"cc",
216 	cxgbe_methods,
217 	sizeof(struct port_info)
218 };
219 
220 /* T6 VI (vcc) interface */
221 static driver_t vcc_driver = {
222 	"vcc",
223 	vcxgbe_methods,
224 	sizeof(struct vi_info)
225 };
226 
227 /* ifnet interface */
228 static void cxgbe_init(void *);
229 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
230 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
231 static void cxgbe_qflush(struct ifnet *);
232 
233 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
234 
235 /*
236  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
237  * then ADAPTER_LOCK, then t4_uld_list_lock.
238  */
239 static struct sx t4_list_lock;
240 SLIST_HEAD(, adapter) t4_list;
241 #ifdef TCP_OFFLOAD
242 static struct sx t4_uld_list_lock;
243 SLIST_HEAD(, uld_info) t4_uld_list;
244 #endif
245 
246 /*
247  * Tunables.  See tweak_tunables() too.
248  *
249  * Each tunable is set to a default value here if it's known at compile-time.
250  * Otherwise it is set to -n as an indication to tweak_tunables() that it should
251  * provide a reasonable default (upto n) when the driver is loaded.
252  *
253  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
254  * T5 are under hw.cxl.
255  */
256 
257 /*
258  * Number of queues for tx and rx, NIC and offload.
259  */
260 #define NTXQ 16
261 int t4_ntxq = -NTXQ;
262 TUNABLE_INT("hw.cxgbe.ntxq", &t4_ntxq);
263 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq);	/* Old name, undocumented */
264 
265 #define NRXQ 8
266 int t4_nrxq = -NRXQ;
267 TUNABLE_INT("hw.cxgbe.nrxq", &t4_nrxq);
268 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq);	/* Old name, undocumented */
269 
270 #define NTXQ_VI 1
271 static int t4_ntxq_vi = -NTXQ_VI;
272 TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
273 
274 #define NRXQ_VI 1
275 static int t4_nrxq_vi = -NRXQ_VI;
276 TUNABLE_INT("hw.cxgbe.nrxq_vi", &t4_nrxq_vi);
277 
278 static int t4_rsrv_noflowq = 0;
279 TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
280 
281 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
282 #define NOFLDTXQ 8
283 static int t4_nofldtxq = -NOFLDTXQ;
284 TUNABLE_INT("hw.cxgbe.nofldtxq", &t4_nofldtxq);
285 
286 #define NOFLDRXQ 2
287 static int t4_nofldrxq = -NOFLDRXQ;
288 TUNABLE_INT("hw.cxgbe.nofldrxq", &t4_nofldrxq);
289 
290 #define NOFLDTXQ_VI 1
291 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
292 TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
293 
294 #define NOFLDRXQ_VI 1
295 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
296 TUNABLE_INT("hw.cxgbe.nofldrxq_vi", &t4_nofldrxq_vi);
297 
298 #define TMR_IDX_OFLD 1
299 int t4_tmr_idx_ofld = TMR_IDX_OFLD;
300 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_ofld", &t4_tmr_idx_ofld);
301 
302 #define PKTC_IDX_OFLD (-1)
303 int t4_pktc_idx_ofld = PKTC_IDX_OFLD;
304 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_ofld", &t4_pktc_idx_ofld);
305 
306 /* 0 means chip/fw default, non-zero number is value in microseconds */
307 static u_long t4_toe_keepalive_idle = 0;
308 TUNABLE_ULONG("hw.cxgbe.toe.keepalive_idle", &t4_toe_keepalive_idle);
309 
310 /* 0 means chip/fw default, non-zero number is value in microseconds */
311 static u_long t4_toe_keepalive_interval = 0;
312 TUNABLE_ULONG("hw.cxgbe.toe.keepalive_interval", &t4_toe_keepalive_interval);
313 
314 /* 0 means chip/fw default, non-zero number is # of keepalives before abort */
315 static int t4_toe_keepalive_count = 0;
316 TUNABLE_INT("hw.cxgbe.toe.keepalive_count", &t4_toe_keepalive_count);
317 
318 /* 0 means chip/fw default, non-zero number is value in microseconds */
319 static u_long t4_toe_rexmt_min = 0;
320 TUNABLE_ULONG("hw.cxgbe.toe.rexmt_min", &t4_toe_rexmt_min);
321 
322 /* 0 means chip/fw default, non-zero number is value in microseconds */
323 static u_long t4_toe_rexmt_max = 0;
324 TUNABLE_ULONG("hw.cxgbe.toe.rexmt_max", &t4_toe_rexmt_max);
325 
326 /* 0 means chip/fw default, non-zero number is # of rexmt before abort */
327 static int t4_toe_rexmt_count = 0;
328 TUNABLE_INT("hw.cxgbe.toe.rexmt_count", &t4_toe_rexmt_count);
329 
330 /* -1 means chip/fw default, other values are raw backoff values to use */
331 static int t4_toe_rexmt_backoff[16] = {
332 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
333 };
334 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.0", &t4_toe_rexmt_backoff[0]);
335 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.1", &t4_toe_rexmt_backoff[1]);
336 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.2", &t4_toe_rexmt_backoff[2]);
337 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.3", &t4_toe_rexmt_backoff[3]);
338 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.4", &t4_toe_rexmt_backoff[4]);
339 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.5", &t4_toe_rexmt_backoff[5]);
340 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.6", &t4_toe_rexmt_backoff[6]);
341 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.7", &t4_toe_rexmt_backoff[7]);
342 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.8", &t4_toe_rexmt_backoff[8]);
343 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.9", &t4_toe_rexmt_backoff[9]);
344 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.10", &t4_toe_rexmt_backoff[10]);
345 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.11", &t4_toe_rexmt_backoff[11]);
346 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.12", &t4_toe_rexmt_backoff[12]);
347 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.13", &t4_toe_rexmt_backoff[13]);
348 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.14", &t4_toe_rexmt_backoff[14]);
349 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.15", &t4_toe_rexmt_backoff[15]);
350 #endif
351 
352 #ifdef DEV_NETMAP
353 #define NNMTXQ_VI 2
354 static int t4_nnmtxq_vi = -NNMTXQ_VI;
355 TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
356 
357 #define NNMRXQ_VI 2
358 static int t4_nnmrxq_vi = -NNMRXQ_VI;
359 TUNABLE_INT("hw.cxgbe.nnmrxq_vi", &t4_nnmrxq_vi);
360 #endif
361 
362 /*
363  * Holdoff parameters for ports.
364  */
365 #define TMR_IDX 1
366 int t4_tmr_idx = TMR_IDX;
367 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx", &t4_tmr_idx);
368 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx);	/* Old name */
369 
370 #define PKTC_IDX (-1)
371 int t4_pktc_idx = PKTC_IDX;
372 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx", &t4_pktc_idx);
373 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx);	/* Old name */
374 
375 /*
376  * Size (# of entries) of each tx and rx queue.
377  */
378 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
379 TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
380 
381 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
382 TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
383 
384 /*
385  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
386  */
387 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
388 TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
389 
390 /*
391  * Configuration file.  All the _CF names here are special.
392  */
393 #define DEFAULT_CF	"default"
394 #define BUILTIN_CF	"built-in"
395 #define FLASH_CF	"flash"
396 #define UWIRE_CF	"uwire"
397 #define FPGA_CF		"fpga"
398 static char t4_cfg_file[32] = DEFAULT_CF;
399 TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
400 
401 /*
402  * PAUSE settings (bit 0, 1, 2 = rx_pause, tx_pause, pause_autoneg respectively).
403  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
404  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
405  *            mark or when signalled to do so, 0 to never emit PAUSE.
406  * pause_autoneg = 1 means PAUSE will be negotiated if possible and the
407  *                 negotiated settings will override rx_pause/tx_pause.
408  *                 Otherwise rx_pause/tx_pause are applied forcibly.
409  */
410 static int t4_pause_settings = PAUSE_RX | PAUSE_TX | PAUSE_AUTONEG;
411 TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
412 
413 /*
414  * Forward Error Correction settings (bit 0, 1 = RS, BASER respectively).
415  * -1 to run with the firmware default.  Same as FEC_AUTO (bit 5)
416  *  0 to disable FEC.
417  */
418 static int t4_fec = -1;
419 TUNABLE_INT("hw.cxgbe.fec", &t4_fec);
420 
421 /*
422  * Link autonegotiation.
423  * -1 to run with the firmware default.
424  *  0 to disable.
425  *  1 to enable.
426  */
427 static int t4_autoneg = -1;
428 TUNABLE_INT("hw.cxgbe.autoneg", &t4_autoneg);
429 
430 /*
431  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
432  * encouraged respectively).
433  */
434 static unsigned int t4_fw_install = 1;
435 TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
436 
437 /*
438  * ASIC features that will be used.  Disable the ones you don't want so that the
439  * chip resources aren't wasted on features that will not be used.
440  */
441 static int t4_nbmcaps_allowed = 0;
442 TUNABLE_INT("hw.cxgbe.nbmcaps_allowed", &t4_nbmcaps_allowed);
443 
444 static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
445 TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
446 
447 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
448     FW_CAPS_CONFIG_SWITCH_EGRESS;
449 TUNABLE_INT("hw.cxgbe.switchcaps_allowed", &t4_switchcaps_allowed);
450 
451 #ifdef RATELIMIT
452 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
453 	FW_CAPS_CONFIG_NIC_HASHFILTER | FW_CAPS_CONFIG_NIC_ETHOFLD;
454 #else
455 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
456 	FW_CAPS_CONFIG_NIC_HASHFILTER;
457 #endif
458 TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
459 
460 static int t4_toecaps_allowed = -1;
461 TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
462 
463 static int t4_rdmacaps_allowed = -1;
464 TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
465 
466 static int t4_cryptocaps_allowed = -1;
467 TUNABLE_INT("hw.cxgbe.cryptocaps_allowed", &t4_cryptocaps_allowed);
468 
469 static int t4_iscsicaps_allowed = -1;
470 TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
471 
472 static int t4_fcoecaps_allowed = 0;
473 TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
474 
475 static int t5_write_combine = 0;
476 TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
477 
478 static int t4_num_vis = 1;
479 TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
480 /*
481  * PCIe Relaxed Ordering.
482  * -1: driver should figure out a good value.
483  * 0: disable RO.
484  * 1: enable RO.
485  * 2: leave RO alone.
486  */
487 static int pcie_relaxed_ordering = -1;
488 TUNABLE_INT("hw.cxgbe.pcie_relaxed_ordering", &pcie_relaxed_ordering);
489 
490 static int t4_panic_on_fatal_err = 0;
491 TUNABLE_INT("hw.cxgbe.panic_on_fatal_err", &t4_panic_on_fatal_err);
492 
493 #ifdef TCP_OFFLOAD
494 /*
495  * TOE tunables.
496  */
497 static int t4_cop_managed_offloading = 0;
498 TUNABLE_INT("hw.cxgbe.cop_managed_offloading", &t4_cop_managed_offloading);
499 #endif
500 
501 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
502 static int vi_mac_funcs[] = {
503 	FW_VI_FUNC_ETH,
504 	FW_VI_FUNC_OFLD,
505 	FW_VI_FUNC_IWARP,
506 	FW_VI_FUNC_OPENISCSI,
507 	FW_VI_FUNC_OPENFCOE,
508 	FW_VI_FUNC_FOISCSI,
509 	FW_VI_FUNC_FOFCOE,
510 };
511 
512 struct intrs_and_queues {
513 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
514 	uint16_t num_vis;	/* number of VIs for each port */
515 	uint16_t nirq;		/* Total # of vectors */
516 	uint16_t ntxq;		/* # of NIC txq's for each port */
517 	uint16_t nrxq;		/* # of NIC rxq's for each port */
518 	uint16_t nofldtxq;	/* # of TOE/ETHOFLD txq's for each port */
519 	uint16_t nofldrxq;	/* # of TOE rxq's for each port */
520 
521 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
522 	uint16_t ntxq_vi;	/* # of NIC txq's */
523 	uint16_t nrxq_vi;	/* # of NIC rxq's */
524 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
525 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
526 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
527 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
528 };
529 
530 static void setup_memwin(struct adapter *);
531 static void position_memwin(struct adapter *, int, uint32_t);
532 static int validate_mem_range(struct adapter *, uint32_t, uint32_t);
533 static int fwmtype_to_hwmtype(int);
534 static int validate_mt_off_len(struct adapter *, int, uint32_t, uint32_t,
535     uint32_t *);
536 static int fixup_devlog_params(struct adapter *);
537 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
538 static int prep_firmware(struct adapter *);
539 static int partition_resources(struct adapter *, const struct firmware *,
540     const char *);
541 static int get_params__pre_init(struct adapter *);
542 static int get_params__post_init(struct adapter *);
543 static int set_params__post_init(struct adapter *);
544 static void t4_set_desc(struct adapter *);
545 static bool fixed_ifmedia(struct port_info *);
546 static void build_medialist(struct port_info *);
547 static void init_link_config(struct port_info *);
548 static int fixup_link_config(struct port_info *);
549 static int apply_link_config(struct port_info *);
550 static int cxgbe_init_synchronized(struct vi_info *);
551 static int cxgbe_uninit_synchronized(struct vi_info *);
552 static void quiesce_txq(struct adapter *, struct sge_txq *);
553 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
554 static void quiesce_iq(struct adapter *, struct sge_iq *);
555 static void quiesce_fl(struct adapter *, struct sge_fl *);
556 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
557     driver_intr_t *, void *, char *);
558 static int t4_free_irq(struct adapter *, struct irq *);
559 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
560 static void vi_refresh_stats(struct adapter *, struct vi_info *);
561 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
562 static void cxgbe_tick(void *);
563 static void cxgbe_sysctls(struct port_info *);
564 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
565 static int sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS);
566 static int sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS);
567 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
568 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
569 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
570 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
571 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
572 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
573 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
574 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
575 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
576 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
577 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
578 static int sysctl_loadavg(SYSCTL_HANDLER_ARGS);
579 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
580 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
581 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
582 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
583 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
584 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
585 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
586 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
587 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
588 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
589 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
590 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
591 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
592 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
593 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
594 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
595 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
596 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
597 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
598 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
599 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
600 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
601 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
602 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
603 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
604 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
605 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
606 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
607 static int sysctl_cpus(SYSCTL_HANDLER_ARGS);
608 #ifdef TCP_OFFLOAD
609 static int sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS);
610 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
611 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
612 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
613 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
614 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
615 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
616 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
617 #endif
618 static int get_sge_context(struct adapter *, struct t4_sge_context *);
619 static int load_fw(struct adapter *, struct t4_data *);
620 static int load_cfg(struct adapter *, struct t4_data *);
621 static int load_boot(struct adapter *, struct t4_bootrom *);
622 static int load_bootcfg(struct adapter *, struct t4_data *);
623 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
624 static void free_offload_policy(struct t4_offload_policy *);
625 static int set_offload_policy(struct adapter *, struct t4_offload_policy *);
626 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
627 static int read_i2c(struct adapter *, struct t4_i2c_data *);
628 #ifdef TCP_OFFLOAD
629 static int toe_capability(struct vi_info *, int);
630 #endif
631 static int mod_event(module_t, int, void *);
632 static int notify_siblings(device_t, int);
633 
634 struct {
635 	uint16_t device;
636 	char *desc;
637 } t4_pciids[] = {
638 	{0xa000, "Chelsio Terminator 4 FPGA"},
639 	{0x4400, "Chelsio T440-dbg"},
640 	{0x4401, "Chelsio T420-CR"},
641 	{0x4402, "Chelsio T422-CR"},
642 	{0x4403, "Chelsio T440-CR"},
643 	{0x4404, "Chelsio T420-BCH"},
644 	{0x4405, "Chelsio T440-BCH"},
645 	{0x4406, "Chelsio T440-CH"},
646 	{0x4407, "Chelsio T420-SO"},
647 	{0x4408, "Chelsio T420-CX"},
648 	{0x4409, "Chelsio T420-BT"},
649 	{0x440a, "Chelsio T404-BT"},
650 	{0x440e, "Chelsio T440-LP-CR"},
651 }, t5_pciids[] = {
652 	{0xb000, "Chelsio Terminator 5 FPGA"},
653 	{0x5400, "Chelsio T580-dbg"},
654 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
655 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
656 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
657 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
658 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
659 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
660 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
661 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
662 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
663 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
664 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
665 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
666 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
667 	{0x5418,  "Chelsio T540-BT"},		/* 4 x 10GBaseT */
668 	{0x5419,  "Chelsio T540-LP-BT"},	/* 4 x 10GBaseT */
669 	{0x541a,  "Chelsio T540-SO-BT"},	/* 4 x 10GBaseT, nomem */
670 	{0x541b,  "Chelsio T540-SO-CR"},	/* 4 x 10G, nomem */
671 }, t6_pciids[] = {
672 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
673 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
674 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
675 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
676 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
677 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
678 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
679 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
680 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
681 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
682 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
683 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
684 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
685 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
686 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
687 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
688 
689 	/* Custom */
690 	{0x6480, "Custom T6225-CR"},
691 	{0x6481, "Custom T62100-CR"},
692 	{0x6482, "Custom T6225-CR"},
693 	{0x6483, "Custom T62100-CR"},
694 	{0x6484, "Custom T64100-CR"},
695 	{0x6485, "Custom T6240-SO"},
696 	{0x6486, "Custom T6225-SO-CR"},
697 	{0x6487, "Custom T6225-CR"},
698 };
699 
700 #ifdef TCP_OFFLOAD
701 /*
702  * service_iq_fl() has an iq and needs the fl.  Offset of fl from the iq should
703  * be exactly the same for both rxq and ofld_rxq.
704  */
705 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
706 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
707 #endif
708 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
709 
710 static int
711 t4_probe(device_t dev)
712 {
713 	int i;
714 	uint16_t v = pci_get_vendor(dev);
715 	uint16_t d = pci_get_device(dev);
716 	uint8_t f = pci_get_function(dev);
717 
718 	if (v != PCI_VENDOR_ID_CHELSIO)
719 		return (ENXIO);
720 
721 	/* Attach only to PF0 of the FPGA */
722 	if (d == 0xa000 && f != 0)
723 		return (ENXIO);
724 
725 	for (i = 0; i < nitems(t4_pciids); i++) {
726 		if (d == t4_pciids[i].device) {
727 			device_set_desc(dev, t4_pciids[i].desc);
728 			return (BUS_PROBE_DEFAULT);
729 		}
730 	}
731 
732 	return (ENXIO);
733 }
734 
735 static int
736 t5_probe(device_t dev)
737 {
738 	int i;
739 	uint16_t v = pci_get_vendor(dev);
740 	uint16_t d = pci_get_device(dev);
741 	uint8_t f = pci_get_function(dev);
742 
743 	if (v != PCI_VENDOR_ID_CHELSIO)
744 		return (ENXIO);
745 
746 	/* Attach only to PF0 of the FPGA */
747 	if (d == 0xb000 && f != 0)
748 		return (ENXIO);
749 
750 	for (i = 0; i < nitems(t5_pciids); i++) {
751 		if (d == t5_pciids[i].device) {
752 			device_set_desc(dev, t5_pciids[i].desc);
753 			return (BUS_PROBE_DEFAULT);
754 		}
755 	}
756 
757 	return (ENXIO);
758 }
759 
760 static int
761 t6_probe(device_t dev)
762 {
763 	int i;
764 	uint16_t v = pci_get_vendor(dev);
765 	uint16_t d = pci_get_device(dev);
766 
767 	if (v != PCI_VENDOR_ID_CHELSIO)
768 		return (ENXIO);
769 
770 	for (i = 0; i < nitems(t6_pciids); i++) {
771 		if (d == t6_pciids[i].device) {
772 			device_set_desc(dev, t6_pciids[i].desc);
773 			return (BUS_PROBE_DEFAULT);
774 		}
775 	}
776 
777 	return (ENXIO);
778 }
779 
780 static void
781 t5_attribute_workaround(device_t dev)
782 {
783 	device_t root_port;
784 	uint32_t v;
785 
786 	/*
787 	 * The T5 chips do not properly echo the No Snoop and Relaxed
788 	 * Ordering attributes when replying to a TLP from a Root
789 	 * Port.  As a workaround, find the parent Root Port and
790 	 * disable No Snoop and Relaxed Ordering.  Note that this
791 	 * affects all devices under this root port.
792 	 */
793 	root_port = pci_find_pcie_root_port(dev);
794 	if (root_port == NULL) {
795 		device_printf(dev, "Unable to find parent root port\n");
796 		return;
797 	}
798 
799 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
800 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
801 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
802 	    0)
803 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
804 		    device_get_nameunit(root_port));
805 }
806 
807 static const struct devnames devnames[] = {
808 	{
809 		.nexus_name = "t4nex",
810 		.ifnet_name = "cxgbe",
811 		.vi_ifnet_name = "vcxgbe",
812 		.pf03_drv_name = "t4iov",
813 		.vf_nexus_name = "t4vf",
814 		.vf_ifnet_name = "cxgbev"
815 	}, {
816 		.nexus_name = "t5nex",
817 		.ifnet_name = "cxl",
818 		.vi_ifnet_name = "vcxl",
819 		.pf03_drv_name = "t5iov",
820 		.vf_nexus_name = "t5vf",
821 		.vf_ifnet_name = "cxlv"
822 	}, {
823 		.nexus_name = "t6nex",
824 		.ifnet_name = "cc",
825 		.vi_ifnet_name = "vcc",
826 		.pf03_drv_name = "t6iov",
827 		.vf_nexus_name = "t6vf",
828 		.vf_ifnet_name = "ccv"
829 	}
830 };
831 
832 void
833 t4_init_devnames(struct adapter *sc)
834 {
835 	int id;
836 
837 	id = chip_id(sc);
838 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
839 		sc->names = &devnames[id - CHELSIO_T4];
840 	else {
841 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
842 		sc->names = NULL;
843 	}
844 }
845 
846 static int
847 t4_ifnet_unit(struct adapter *sc, struct port_info *pi)
848 {
849 	const char *parent, *name;
850 	long value;
851 	int line, unit;
852 
853 	line = 0;
854 	parent = device_get_nameunit(sc->dev);
855 	name = sc->names->ifnet_name;
856 	while (resource_find_dev(&line, name, &unit, "at", parent) == 0) {
857 		if (resource_long_value(name, unit, "port", &value) == 0 &&
858 		    value == pi->port_id)
859 			return (unit);
860 	}
861 	return (-1);
862 }
863 
864 static int
865 t4_attach(device_t dev)
866 {
867 	struct adapter *sc;
868 	int rc = 0, i, j, rqidx, tqidx, nports;
869 	struct make_dev_args mda;
870 	struct intrs_and_queues iaq;
871 	struct sge *s;
872 	uint32_t *buf;
873 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
874 	int ofld_tqidx;
875 #endif
876 #ifdef TCP_OFFLOAD
877 	int ofld_rqidx;
878 #endif
879 #ifdef DEV_NETMAP
880 	int nm_rqidx, nm_tqidx;
881 #endif
882 	int num_vis;
883 
884 	sc = device_get_softc(dev);
885 	sc->dev = dev;
886 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
887 
888 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
889 		t5_attribute_workaround(dev);
890 	pci_enable_busmaster(dev);
891 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
892 		uint32_t v;
893 
894 		pci_set_max_read_req(dev, 4096);
895 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
896 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
897 		if (pcie_relaxed_ordering == 0 &&
898 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
899 			v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
900 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
901 		} else if (pcie_relaxed_ordering == 1 &&
902 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
903 			v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
904 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
905 		}
906 	}
907 
908 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
909 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
910 	sc->traceq = -1;
911 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
912 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
913 	    device_get_nameunit(dev));
914 
915 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
916 	    device_get_nameunit(dev));
917 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
918 	t4_add_adapter(sc);
919 
920 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
921 	TAILQ_INIT(&sc->sfl);
922 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
923 
924 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
925 
926 	sc->policy = NULL;
927 	rw_init(&sc->policy_lock, "connection offload policy");
928 
929 	rc = t4_map_bars_0_and_4(sc);
930 	if (rc != 0)
931 		goto done; /* error message displayed already */
932 
933 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
934 
935 	/* Prepare the adapter for operation. */
936 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
937 	rc = -t4_prep_adapter(sc, buf);
938 	free(buf, M_CXGBE);
939 	if (rc != 0) {
940 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
941 		goto done;
942 	}
943 
944 	/*
945 	 * This is the real PF# to which we're attaching.  Works from within PCI
946 	 * passthrough environments too, where pci_get_function() could return a
947 	 * different PF# depending on the passthrough configuration.  We need to
948 	 * use the real PF# in all our communication with the firmware.
949 	 */
950 	j = t4_read_reg(sc, A_PL_WHOAMI);
951 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
952 	sc->mbox = sc->pf;
953 
954 	t4_init_devnames(sc);
955 	if (sc->names == NULL) {
956 		rc = ENOTSUP;
957 		goto done; /* error message displayed already */
958 	}
959 
960 	/*
961 	 * Do this really early, with the memory windows set up even before the
962 	 * character device.  The userland tool's register i/o and mem read
963 	 * will work even in "recovery mode".
964 	 */
965 	setup_memwin(sc);
966 	if (t4_init_devlog_params(sc, 0) == 0)
967 		fixup_devlog_params(sc);
968 	make_dev_args_init(&mda);
969 	mda.mda_devsw = &t4_cdevsw;
970 	mda.mda_uid = UID_ROOT;
971 	mda.mda_gid = GID_WHEEL;
972 	mda.mda_mode = 0600;
973 	mda.mda_si_drv1 = sc;
974 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
975 	if (rc != 0)
976 		device_printf(dev, "failed to create nexus char device: %d.\n",
977 		    rc);
978 
979 	/* Go no further if recovery mode has been requested. */
980 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
981 		device_printf(dev, "recovery mode.\n");
982 		goto done;
983 	}
984 
985 #if defined(__i386__)
986 	if ((cpu_feature & CPUID_CX8) == 0) {
987 		device_printf(dev, "64 bit atomics not available.\n");
988 		rc = ENOTSUP;
989 		goto done;
990 	}
991 #endif
992 
993 	/* Prepare the firmware for operation */
994 	rc = prep_firmware(sc);
995 	if (rc != 0)
996 		goto done; /* error message displayed already */
997 
998 	rc = get_params__post_init(sc);
999 	if (rc != 0)
1000 		goto done; /* error message displayed already */
1001 
1002 	rc = set_params__post_init(sc);
1003 	if (rc != 0)
1004 		goto done; /* error message displayed already */
1005 
1006 	rc = t4_map_bar_2(sc);
1007 	if (rc != 0)
1008 		goto done; /* error message displayed already */
1009 
1010 	rc = t4_create_dma_tag(sc);
1011 	if (rc != 0)
1012 		goto done; /* error message displayed already */
1013 
1014 	/*
1015 	 * First pass over all the ports - allocate VIs and initialize some
1016 	 * basic parameters like mac address, port type, etc.
1017 	 */
1018 	for_each_port(sc, i) {
1019 		struct port_info *pi;
1020 
1021 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
1022 		sc->port[i] = pi;
1023 
1024 		/* These must be set before t4_port_init */
1025 		pi->adapter = sc;
1026 		pi->port_id = i;
1027 		/*
1028 		 * XXX: vi[0] is special so we can't delay this allocation until
1029 		 * pi->nvi's final value is known.
1030 		 */
1031 		pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
1032 		    M_ZERO | M_WAITOK);
1033 
1034 		/*
1035 		 * Allocate the "main" VI and initialize parameters
1036 		 * like mac addr.
1037 		 */
1038 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1039 		if (rc != 0) {
1040 			device_printf(dev, "unable to initialize port %d: %d\n",
1041 			    i, rc);
1042 			free(pi->vi, M_CXGBE);
1043 			free(pi, M_CXGBE);
1044 			sc->port[i] = NULL;
1045 			goto done;
1046 		}
1047 
1048 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1049 		    device_get_nameunit(dev), i);
1050 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1051 		sc->chan_map[pi->tx_chan] = i;
1052 
1053 		/* All VIs on this port share this media. */
1054 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1055 		    cxgbe_media_status);
1056 
1057 		PORT_LOCK(pi);
1058 		init_link_config(pi);
1059 		fixup_link_config(pi);
1060 		build_medialist(pi);
1061 		if (fixed_ifmedia(pi))
1062 			pi->flags |= FIXED_IFMEDIA;
1063 		PORT_UNLOCK(pi);
1064 
1065 		pi->dev = device_add_child(dev, sc->names->ifnet_name,
1066 		    t4_ifnet_unit(sc, pi));
1067 		if (pi->dev == NULL) {
1068 			device_printf(dev,
1069 			    "failed to add device for port %d.\n", i);
1070 			rc = ENXIO;
1071 			goto done;
1072 		}
1073 		pi->vi[0].dev = pi->dev;
1074 		device_set_softc(pi->dev, pi);
1075 	}
1076 
1077 	/*
1078 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1079 	 */
1080 	nports = sc->params.nports;
1081 	rc = cfg_itype_and_nqueues(sc, &iaq);
1082 	if (rc != 0)
1083 		goto done; /* error message displayed already */
1084 
1085 	num_vis = iaq.num_vis;
1086 	sc->intr_type = iaq.intr_type;
1087 	sc->intr_count = iaq.nirq;
1088 
1089 	s = &sc->sge;
1090 	s->nrxq = nports * iaq.nrxq;
1091 	s->ntxq = nports * iaq.ntxq;
1092 	if (num_vis > 1) {
1093 		s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1094 		s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1095 	}
1096 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1097 	s->neq += nports;		/* ctrl queues: 1 per port */
1098 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1099 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1100 	if (is_offload(sc) || is_ethoffload(sc)) {
1101 		s->nofldtxq = nports * iaq.nofldtxq;
1102 		if (num_vis > 1)
1103 			s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1104 		s->neq += s->nofldtxq;
1105 
1106 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1107 		    M_CXGBE, M_ZERO | M_WAITOK);
1108 	}
1109 #endif
1110 #ifdef TCP_OFFLOAD
1111 	if (is_offload(sc)) {
1112 		s->nofldrxq = nports * iaq.nofldrxq;
1113 		if (num_vis > 1)
1114 			s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1115 		s->neq += s->nofldrxq;	/* free list */
1116 		s->niq += s->nofldrxq;
1117 
1118 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1119 		    M_CXGBE, M_ZERO | M_WAITOK);
1120 	}
1121 #endif
1122 #ifdef DEV_NETMAP
1123 	if (num_vis > 1) {
1124 		s->nnmrxq = nports * (num_vis - 1) * iaq.nnmrxq_vi;
1125 		s->nnmtxq = nports * (num_vis - 1) * iaq.nnmtxq_vi;
1126 	}
1127 	s->neq += s->nnmtxq + s->nnmrxq;
1128 	s->niq += s->nnmrxq;
1129 
1130 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1131 	    M_CXGBE, M_ZERO | M_WAITOK);
1132 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1133 	    M_CXGBE, M_ZERO | M_WAITOK);
1134 #endif
1135 
1136 	s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1137 	    M_ZERO | M_WAITOK);
1138 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1139 	    M_ZERO | M_WAITOK);
1140 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1141 	    M_ZERO | M_WAITOK);
1142 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1143 	    M_ZERO | M_WAITOK);
1144 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1145 	    M_ZERO | M_WAITOK);
1146 
1147 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1148 	    M_ZERO | M_WAITOK);
1149 
1150 	t4_init_l2t(sc, M_WAITOK);
1151 	t4_init_smt(sc, M_WAITOK);
1152 	t4_init_tx_sched(sc);
1153 #ifdef RATELIMIT
1154 	t4_init_etid_table(sc);
1155 #endif
1156 
1157 	/*
1158 	 * Second pass over the ports.  This time we know the number of rx and
1159 	 * tx queues that each port should get.
1160 	 */
1161 	rqidx = tqidx = 0;
1162 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1163 	ofld_tqidx = 0;
1164 #endif
1165 #ifdef TCP_OFFLOAD
1166 	ofld_rqidx = 0;
1167 #endif
1168 #ifdef DEV_NETMAP
1169 	nm_rqidx = nm_tqidx = 0;
1170 #endif
1171 	for_each_port(sc, i) {
1172 		struct port_info *pi = sc->port[i];
1173 		struct vi_info *vi;
1174 
1175 		if (pi == NULL)
1176 			continue;
1177 
1178 		pi->nvi = num_vis;
1179 		for_each_vi(pi, j, vi) {
1180 			vi->pi = pi;
1181 			vi->qsize_rxq = t4_qsize_rxq;
1182 			vi->qsize_txq = t4_qsize_txq;
1183 
1184 			vi->first_rxq = rqidx;
1185 			vi->first_txq = tqidx;
1186 			vi->tmr_idx = t4_tmr_idx;
1187 			vi->pktc_idx = t4_pktc_idx;
1188 			vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1189 			vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1190 
1191 			rqidx += vi->nrxq;
1192 			tqidx += vi->ntxq;
1193 
1194 			if (j == 0 && vi->ntxq > 1)
1195 				vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1196 			else
1197 				vi->rsrv_noflowq = 0;
1198 
1199 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1200 			vi->first_ofld_txq = ofld_tqidx;
1201 			vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1202 			ofld_tqidx += vi->nofldtxq;
1203 #endif
1204 #ifdef TCP_OFFLOAD
1205 			vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1206 			vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1207 			vi->first_ofld_rxq = ofld_rqidx;
1208 			vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1209 
1210 			ofld_rqidx += vi->nofldrxq;
1211 #endif
1212 #ifdef DEV_NETMAP
1213 			if (j > 0) {
1214 				vi->first_nm_rxq = nm_rqidx;
1215 				vi->first_nm_txq = nm_tqidx;
1216 				vi->nnmrxq = iaq.nnmrxq_vi;
1217 				vi->nnmtxq = iaq.nnmtxq_vi;
1218 				nm_rqidx += vi->nnmrxq;
1219 				nm_tqidx += vi->nnmtxq;
1220 			}
1221 #endif
1222 		}
1223 	}
1224 
1225 	rc = t4_setup_intr_handlers(sc);
1226 	if (rc != 0) {
1227 		device_printf(dev,
1228 		    "failed to setup interrupt handlers: %d\n", rc);
1229 		goto done;
1230 	}
1231 
1232 	rc = bus_generic_probe(dev);
1233 	if (rc != 0) {
1234 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1235 		goto done;
1236 	}
1237 
1238 	/*
1239 	 * Ensure thread-safe mailbox access (in debug builds).
1240 	 *
1241 	 * So far this was the only thread accessing the mailbox but various
1242 	 * ifnets and sysctls are about to be created and their handlers/ioctls
1243 	 * will access the mailbox from different threads.
1244 	 */
1245 	sc->flags |= CHK_MBOX_ACCESS;
1246 
1247 	rc = bus_generic_attach(dev);
1248 	if (rc != 0) {
1249 		device_printf(dev,
1250 		    "failed to attach all child ports: %d\n", rc);
1251 		goto done;
1252 	}
1253 
1254 	device_printf(dev,
1255 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1256 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1257 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1258 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1259 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1260 
1261 	t4_set_desc(sc);
1262 
1263 	notify_siblings(dev, 0);
1264 
1265 done:
1266 	if (rc != 0 && sc->cdev) {
1267 		/* cdev was created and so cxgbetool works; recover that way. */
1268 		device_printf(dev,
1269 		    "error during attach, adapter is now in recovery mode.\n");
1270 		rc = 0;
1271 	}
1272 
1273 	if (rc != 0)
1274 		t4_detach_common(dev);
1275 	else
1276 		t4_sysctls(sc);
1277 
1278 	return (rc);
1279 }
1280 
1281 static int
1282 t4_child_location_str(device_t bus, device_t dev, char *buf, size_t buflen)
1283 {
1284 	struct port_info *pi;
1285 
1286 	pi = device_get_softc(dev);
1287 	snprintf(buf, buflen, "port=%d", pi->port_id);
1288 	return (0);
1289 }
1290 
1291 static int
1292 t4_ready(device_t dev)
1293 {
1294 	struct adapter *sc;
1295 
1296 	sc = device_get_softc(dev);
1297 	if (sc->flags & FW_OK)
1298 		return (0);
1299 	return (ENXIO);
1300 }
1301 
1302 static int
1303 t4_read_port_device(device_t dev, int port, device_t *child)
1304 {
1305 	struct adapter *sc;
1306 	struct port_info *pi;
1307 
1308 	sc = device_get_softc(dev);
1309 	if (port < 0 || port >= MAX_NPORTS)
1310 		return (EINVAL);
1311 	pi = sc->port[port];
1312 	if (pi == NULL || pi->dev == NULL)
1313 		return (ENXIO);
1314 	*child = pi->dev;
1315 	return (0);
1316 }
1317 
1318 static int
1319 notify_siblings(device_t dev, int detaching)
1320 {
1321 	device_t sibling;
1322 	int error, i;
1323 
1324 	error = 0;
1325 	for (i = 0; i < PCI_FUNCMAX; i++) {
1326 		if (i == pci_get_function(dev))
1327 			continue;
1328 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1329 		    pci_get_slot(dev), i);
1330 		if (sibling == NULL || !device_is_attached(sibling))
1331 			continue;
1332 		if (detaching)
1333 			error = T4_DETACH_CHILD(sibling);
1334 		else
1335 			(void)T4_ATTACH_CHILD(sibling);
1336 		if (error)
1337 			break;
1338 	}
1339 	return (error);
1340 }
1341 
1342 /*
1343  * Idempotent
1344  */
1345 static int
1346 t4_detach(device_t dev)
1347 {
1348 	struct adapter *sc;
1349 	int rc;
1350 
1351 	sc = device_get_softc(dev);
1352 
1353 	rc = notify_siblings(dev, 1);
1354 	if (rc) {
1355 		device_printf(dev,
1356 		    "failed to detach sibling devices: %d\n", rc);
1357 		return (rc);
1358 	}
1359 
1360 	return (t4_detach_common(dev));
1361 }
1362 
1363 int
1364 t4_detach_common(device_t dev)
1365 {
1366 	struct adapter *sc;
1367 	struct port_info *pi;
1368 	int i, rc;
1369 
1370 	sc = device_get_softc(dev);
1371 
1372 	if (sc->cdev) {
1373 		destroy_dev(sc->cdev);
1374 		sc->cdev = NULL;
1375 	}
1376 
1377 	sc->flags &= ~CHK_MBOX_ACCESS;
1378 	if (sc->flags & FULL_INIT_DONE) {
1379 		if (!(sc->flags & IS_VF))
1380 			t4_intr_disable(sc);
1381 	}
1382 
1383 	if (device_is_attached(dev)) {
1384 		rc = bus_generic_detach(dev);
1385 		if (rc) {
1386 			device_printf(dev,
1387 			    "failed to detach child devices: %d\n", rc);
1388 			return (rc);
1389 		}
1390 	}
1391 
1392 	for (i = 0; i < sc->intr_count; i++)
1393 		t4_free_irq(sc, &sc->irq[i]);
1394 
1395 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1396 		t4_free_tx_sched(sc);
1397 
1398 	for (i = 0; i < MAX_NPORTS; i++) {
1399 		pi = sc->port[i];
1400 		if (pi) {
1401 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1402 			if (pi->dev)
1403 				device_delete_child(dev, pi->dev);
1404 
1405 			mtx_destroy(&pi->pi_lock);
1406 			free(pi->vi, M_CXGBE);
1407 			free(pi, M_CXGBE);
1408 		}
1409 	}
1410 
1411 	device_delete_children(dev);
1412 
1413 	if (sc->flags & FULL_INIT_DONE)
1414 		adapter_full_uninit(sc);
1415 
1416 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1417 		t4_fw_bye(sc, sc->mbox);
1418 
1419 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1420 		pci_release_msi(dev);
1421 
1422 	if (sc->regs_res)
1423 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1424 		    sc->regs_res);
1425 
1426 	if (sc->udbs_res)
1427 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1428 		    sc->udbs_res);
1429 
1430 	if (sc->msix_res)
1431 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1432 		    sc->msix_res);
1433 
1434 	if (sc->l2t)
1435 		t4_free_l2t(sc->l2t);
1436 	if (sc->smt)
1437 		t4_free_smt(sc->smt);
1438 #ifdef RATELIMIT
1439 	t4_free_etid_table(sc);
1440 #endif
1441 
1442 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1443 	free(sc->sge.ofld_txq, M_CXGBE);
1444 #endif
1445 #ifdef TCP_OFFLOAD
1446 	free(sc->sge.ofld_rxq, M_CXGBE);
1447 #endif
1448 #ifdef DEV_NETMAP
1449 	free(sc->sge.nm_rxq, M_CXGBE);
1450 	free(sc->sge.nm_txq, M_CXGBE);
1451 #endif
1452 	free(sc->irq, M_CXGBE);
1453 	free(sc->sge.rxq, M_CXGBE);
1454 	free(sc->sge.txq, M_CXGBE);
1455 	free(sc->sge.ctrlq, M_CXGBE);
1456 	free(sc->sge.iqmap, M_CXGBE);
1457 	free(sc->sge.eqmap, M_CXGBE);
1458 	free(sc->tids.ftid_tab, M_CXGBE);
1459 	free(sc->tids.hpftid_tab, M_CXGBE);
1460 	free_hftid_hash(&sc->tids);
1461 	free(sc->tids.atid_tab, M_CXGBE);
1462 	free(sc->tids.tid_tab, M_CXGBE);
1463 	free(sc->tt.tls_rx_ports, M_CXGBE);
1464 	t4_destroy_dma_tag(sc);
1465 	if (mtx_initialized(&sc->sc_lock)) {
1466 		sx_xlock(&t4_list_lock);
1467 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1468 		sx_xunlock(&t4_list_lock);
1469 		mtx_destroy(&sc->sc_lock);
1470 	}
1471 
1472 	callout_drain(&sc->sfl_callout);
1473 	if (mtx_initialized(&sc->tids.ftid_lock)) {
1474 		mtx_destroy(&sc->tids.ftid_lock);
1475 		cv_destroy(&sc->tids.ftid_cv);
1476 	}
1477 	if (mtx_initialized(&sc->tids.atid_lock))
1478 		mtx_destroy(&sc->tids.atid_lock);
1479 	if (mtx_initialized(&sc->sfl_lock))
1480 		mtx_destroy(&sc->sfl_lock);
1481 	if (mtx_initialized(&sc->ifp_lock))
1482 		mtx_destroy(&sc->ifp_lock);
1483 	if (mtx_initialized(&sc->reg_lock))
1484 		mtx_destroy(&sc->reg_lock);
1485 
1486 	if (rw_initialized(&sc->policy_lock)) {
1487 		rw_destroy(&sc->policy_lock);
1488 #ifdef TCP_OFFLOAD
1489 		if (sc->policy != NULL)
1490 			free_offload_policy(sc->policy);
1491 #endif
1492 	}
1493 
1494 	for (i = 0; i < NUM_MEMWIN; i++) {
1495 		struct memwin *mw = &sc->memwin[i];
1496 
1497 		if (rw_initialized(&mw->mw_lock))
1498 			rw_destroy(&mw->mw_lock);
1499 	}
1500 
1501 	bzero(sc, sizeof(*sc));
1502 
1503 	return (0);
1504 }
1505 
1506 static int
1507 cxgbe_probe(device_t dev)
1508 {
1509 	char buf[128];
1510 	struct port_info *pi = device_get_softc(dev);
1511 
1512 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1513 	device_set_desc_copy(dev, buf);
1514 
1515 	return (BUS_PROBE_DEFAULT);
1516 }
1517 
1518 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1519     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1520     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS | \
1521     IFCAP_HWRXTSTMP)
1522 #define T4_CAP_ENABLE (T4_CAP)
1523 
1524 static int
1525 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1526 {
1527 	struct ifnet *ifp;
1528 	struct sbuf *sb;
1529 
1530 	vi->xact_addr_filt = -1;
1531 	callout_init(&vi->tick, 1);
1532 
1533 	/* Allocate an ifnet and set it up */
1534 	ifp = if_alloc(IFT_ETHER);
1535 	if (ifp == NULL) {
1536 		device_printf(dev, "Cannot allocate ifnet\n");
1537 		return (ENOMEM);
1538 	}
1539 	vi->ifp = ifp;
1540 	ifp->if_softc = vi;
1541 
1542 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1543 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1544 
1545 	ifp->if_init = cxgbe_init;
1546 	ifp->if_ioctl = cxgbe_ioctl;
1547 	ifp->if_transmit = cxgbe_transmit;
1548 	ifp->if_qflush = cxgbe_qflush;
1549 	ifp->if_get_counter = cxgbe_get_counter;
1550 #ifdef RATELIMIT
1551 	ifp->if_snd_tag_alloc = cxgbe_snd_tag_alloc;
1552 	ifp->if_snd_tag_modify = cxgbe_snd_tag_modify;
1553 	ifp->if_snd_tag_query = cxgbe_snd_tag_query;
1554 	ifp->if_snd_tag_free = cxgbe_snd_tag_free;
1555 #endif
1556 
1557 	ifp->if_capabilities = T4_CAP;
1558 	ifp->if_capenable = T4_CAP_ENABLE;
1559 #ifdef TCP_OFFLOAD
1560 	if (vi->nofldrxq != 0)
1561 		ifp->if_capabilities |= IFCAP_TOE;
1562 #endif
1563 #ifdef DEV_NETMAP
1564 	if (vi->nnmrxq != 0)
1565 		ifp->if_capabilities |= IFCAP_NETMAP;
1566 #endif
1567 #ifdef RATELIMIT
1568 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0) {
1569 		ifp->if_capabilities |= IFCAP_TXRTLMT;
1570 		ifp->if_capenable |= IFCAP_TXRTLMT;
1571 	}
1572 #endif
1573 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1574 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1575 
1576 	ifp->if_hw_tsomax = IP_MAXPACKET;
1577 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_TSO;
1578 #ifdef RATELIMIT
1579 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0)
1580 		ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_EO_TSO;
1581 #endif
1582 	ifp->if_hw_tsomaxsegsize = 65536;
1583 
1584 	ether_ifattach(ifp, vi->hw_addr);
1585 #ifdef DEV_NETMAP
1586 	if (ifp->if_capabilities & IFCAP_NETMAP)
1587 		cxgbe_nm_attach(vi);
1588 #endif
1589 	sb = sbuf_new_auto();
1590 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1591 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1592 	switch (ifp->if_capabilities & (IFCAP_TOE | IFCAP_TXRTLMT)) {
1593 	case IFCAP_TOE:
1594 		sbuf_printf(sb, "; %d txq (TOE)", vi->nofldtxq);
1595 		break;
1596 	case IFCAP_TOE | IFCAP_TXRTLMT:
1597 		sbuf_printf(sb, "; %d txq (TOE/ETHOFLD)", vi->nofldtxq);
1598 		break;
1599 	case IFCAP_TXRTLMT:
1600 		sbuf_printf(sb, "; %d txq (ETHOFLD)", vi->nofldtxq);
1601 		break;
1602 	}
1603 #endif
1604 #ifdef TCP_OFFLOAD
1605 	if (ifp->if_capabilities & IFCAP_TOE)
1606 		sbuf_printf(sb, ", %d rxq (TOE)", vi->nofldrxq);
1607 #endif
1608 #ifdef DEV_NETMAP
1609 	if (ifp->if_capabilities & IFCAP_NETMAP)
1610 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1611 		    vi->nnmtxq, vi->nnmrxq);
1612 #endif
1613 	sbuf_finish(sb);
1614 	device_printf(dev, "%s\n", sbuf_data(sb));
1615 	sbuf_delete(sb);
1616 
1617 	vi_sysctls(vi);
1618 
1619 	return (0);
1620 }
1621 
1622 static int
1623 cxgbe_attach(device_t dev)
1624 {
1625 	struct port_info *pi = device_get_softc(dev);
1626 	struct adapter *sc = pi->adapter;
1627 	struct vi_info *vi;
1628 	int i, rc;
1629 
1630 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1631 
1632 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1633 	if (rc)
1634 		return (rc);
1635 
1636 	for_each_vi(pi, i, vi) {
1637 		if (i == 0)
1638 			continue;
1639 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1640 		if (vi->dev == NULL) {
1641 			device_printf(dev, "failed to add VI %d\n", i);
1642 			continue;
1643 		}
1644 		device_set_softc(vi->dev, vi);
1645 	}
1646 
1647 	cxgbe_sysctls(pi);
1648 
1649 	bus_generic_attach(dev);
1650 
1651 	return (0);
1652 }
1653 
1654 static void
1655 cxgbe_vi_detach(struct vi_info *vi)
1656 {
1657 	struct ifnet *ifp = vi->ifp;
1658 
1659 	ether_ifdetach(ifp);
1660 
1661 	/* Let detach proceed even if these fail. */
1662 #ifdef DEV_NETMAP
1663 	if (ifp->if_capabilities & IFCAP_NETMAP)
1664 		cxgbe_nm_detach(vi);
1665 #endif
1666 	cxgbe_uninit_synchronized(vi);
1667 	callout_drain(&vi->tick);
1668 	vi_full_uninit(vi);
1669 
1670 	if_free(vi->ifp);
1671 	vi->ifp = NULL;
1672 }
1673 
1674 static int
1675 cxgbe_detach(device_t dev)
1676 {
1677 	struct port_info *pi = device_get_softc(dev);
1678 	struct adapter *sc = pi->adapter;
1679 	int rc;
1680 
1681 	/* Detach the extra VIs first. */
1682 	rc = bus_generic_detach(dev);
1683 	if (rc)
1684 		return (rc);
1685 	device_delete_children(dev);
1686 
1687 	doom_vi(sc, &pi->vi[0]);
1688 
1689 	if (pi->flags & HAS_TRACEQ) {
1690 		sc->traceq = -1;	/* cloner should not create ifnet */
1691 		t4_tracer_port_detach(sc);
1692 	}
1693 
1694 	cxgbe_vi_detach(&pi->vi[0]);
1695 	callout_drain(&pi->tick);
1696 	ifmedia_removeall(&pi->media);
1697 
1698 	end_synchronized_op(sc, 0);
1699 
1700 	return (0);
1701 }
1702 
1703 static void
1704 cxgbe_init(void *arg)
1705 {
1706 	struct vi_info *vi = arg;
1707 	struct adapter *sc = vi->pi->adapter;
1708 
1709 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1710 		return;
1711 	cxgbe_init_synchronized(vi);
1712 	end_synchronized_op(sc, 0);
1713 }
1714 
1715 static int
1716 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1717 {
1718 	int rc = 0, mtu, flags;
1719 	struct vi_info *vi = ifp->if_softc;
1720 	struct port_info *pi = vi->pi;
1721 	struct adapter *sc = pi->adapter;
1722 	struct ifreq *ifr = (struct ifreq *)data;
1723 	uint32_t mask;
1724 
1725 	switch (cmd) {
1726 	case SIOCSIFMTU:
1727 		mtu = ifr->ifr_mtu;
1728 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1729 			return (EINVAL);
1730 
1731 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1732 		if (rc)
1733 			return (rc);
1734 		ifp->if_mtu = mtu;
1735 		if (vi->flags & VI_INIT_DONE) {
1736 			t4_update_fl_bufsize(ifp);
1737 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1738 				rc = update_mac_settings(ifp, XGMAC_MTU);
1739 		}
1740 		end_synchronized_op(sc, 0);
1741 		break;
1742 
1743 	case SIOCSIFFLAGS:
1744 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4flg");
1745 		if (rc)
1746 			return (rc);
1747 
1748 		if (ifp->if_flags & IFF_UP) {
1749 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1750 				flags = vi->if_flags;
1751 				if ((ifp->if_flags ^ flags) &
1752 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1753 					rc = update_mac_settings(ifp,
1754 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1755 				}
1756 			} else {
1757 				rc = cxgbe_init_synchronized(vi);
1758 			}
1759 			vi->if_flags = ifp->if_flags;
1760 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1761 			rc = cxgbe_uninit_synchronized(vi);
1762 		}
1763 		end_synchronized_op(sc, 0);
1764 		break;
1765 
1766 	case SIOCADDMULTI:
1767 	case SIOCDELMULTI:
1768 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4multi");
1769 		if (rc)
1770 			return (rc);
1771 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1772 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1773 		end_synchronized_op(sc, 0);
1774 		break;
1775 
1776 	case SIOCSIFCAP:
1777 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1778 		if (rc)
1779 			return (rc);
1780 
1781 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1782 		if (mask & IFCAP_TXCSUM) {
1783 			ifp->if_capenable ^= IFCAP_TXCSUM;
1784 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1785 
1786 			if (IFCAP_TSO4 & ifp->if_capenable &&
1787 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1788 				ifp->if_capenable &= ~IFCAP_TSO4;
1789 				if_printf(ifp,
1790 				    "tso4 disabled due to -txcsum.\n");
1791 			}
1792 		}
1793 		if (mask & IFCAP_TXCSUM_IPV6) {
1794 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1795 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1796 
1797 			if (IFCAP_TSO6 & ifp->if_capenable &&
1798 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1799 				ifp->if_capenable &= ~IFCAP_TSO6;
1800 				if_printf(ifp,
1801 				    "tso6 disabled due to -txcsum6.\n");
1802 			}
1803 		}
1804 		if (mask & IFCAP_RXCSUM)
1805 			ifp->if_capenable ^= IFCAP_RXCSUM;
1806 		if (mask & IFCAP_RXCSUM_IPV6)
1807 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1808 
1809 		/*
1810 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1811 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1812 		 * sending a TSO request our way, so it's sufficient to toggle
1813 		 * IFCAP_TSOx only.
1814 		 */
1815 		if (mask & IFCAP_TSO4) {
1816 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1817 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1818 				if_printf(ifp, "enable txcsum first.\n");
1819 				rc = EAGAIN;
1820 				goto fail;
1821 			}
1822 			ifp->if_capenable ^= IFCAP_TSO4;
1823 		}
1824 		if (mask & IFCAP_TSO6) {
1825 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1826 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1827 				if_printf(ifp, "enable txcsum6 first.\n");
1828 				rc = EAGAIN;
1829 				goto fail;
1830 			}
1831 			ifp->if_capenable ^= IFCAP_TSO6;
1832 		}
1833 		if (mask & IFCAP_LRO) {
1834 #if defined(INET) || defined(INET6)
1835 			int i;
1836 			struct sge_rxq *rxq;
1837 
1838 			ifp->if_capenable ^= IFCAP_LRO;
1839 			for_each_rxq(vi, i, rxq) {
1840 				if (ifp->if_capenable & IFCAP_LRO)
1841 					rxq->iq.flags |= IQ_LRO_ENABLED;
1842 				else
1843 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1844 			}
1845 #endif
1846 		}
1847 #ifdef TCP_OFFLOAD
1848 		if (mask & IFCAP_TOE) {
1849 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1850 
1851 			rc = toe_capability(vi, enable);
1852 			if (rc != 0)
1853 				goto fail;
1854 
1855 			ifp->if_capenable ^= mask;
1856 		}
1857 #endif
1858 		if (mask & IFCAP_VLAN_HWTAGGING) {
1859 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1860 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1861 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1862 		}
1863 		if (mask & IFCAP_VLAN_MTU) {
1864 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1865 
1866 			/* Need to find out how to disable auto-mtu-inflation */
1867 		}
1868 		if (mask & IFCAP_VLAN_HWTSO)
1869 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1870 		if (mask & IFCAP_VLAN_HWCSUM)
1871 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1872 #ifdef RATELIMIT
1873 		if (mask & IFCAP_TXRTLMT)
1874 			ifp->if_capenable ^= IFCAP_TXRTLMT;
1875 #endif
1876 		if (mask & IFCAP_HWRXTSTMP) {
1877 			int i;
1878 			struct sge_rxq *rxq;
1879 
1880 			ifp->if_capenable ^= IFCAP_HWRXTSTMP;
1881 			for_each_rxq(vi, i, rxq) {
1882 				if (ifp->if_capenable & IFCAP_HWRXTSTMP)
1883 					rxq->iq.flags |= IQ_RX_TIMESTAMP;
1884 				else
1885 					rxq->iq.flags &= ~IQ_RX_TIMESTAMP;
1886 			}
1887 		}
1888 
1889 #ifdef VLAN_CAPABILITIES
1890 		VLAN_CAPABILITIES(ifp);
1891 #endif
1892 fail:
1893 		end_synchronized_op(sc, 0);
1894 		break;
1895 
1896 	case SIOCSIFMEDIA:
1897 	case SIOCGIFMEDIA:
1898 	case SIOCGIFXMEDIA:
1899 		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1900 		break;
1901 
1902 	case SIOCGI2C: {
1903 		struct ifi2creq i2c;
1904 
1905 		rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
1906 		if (rc != 0)
1907 			break;
1908 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1909 			rc = EPERM;
1910 			break;
1911 		}
1912 		if (i2c.len > sizeof(i2c.data)) {
1913 			rc = EINVAL;
1914 			break;
1915 		}
1916 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1917 		if (rc)
1918 			return (rc);
1919 		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
1920 		    i2c.offset, i2c.len, &i2c.data[0]);
1921 		end_synchronized_op(sc, 0);
1922 		if (rc == 0)
1923 			rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
1924 		break;
1925 	}
1926 
1927 	default:
1928 		rc = ether_ioctl(ifp, cmd, data);
1929 	}
1930 
1931 	return (rc);
1932 }
1933 
1934 static int
1935 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1936 {
1937 	struct vi_info *vi = ifp->if_softc;
1938 	struct port_info *pi = vi->pi;
1939 	struct adapter *sc = pi->adapter;
1940 	struct sge_txq *txq;
1941 	void *items[1];
1942 	int rc;
1943 
1944 	M_ASSERTPKTHDR(m);
1945 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1946 
1947 	if (__predict_false(pi->link_cfg.link_ok == false)) {
1948 		m_freem(m);
1949 		return (ENETDOWN);
1950 	}
1951 
1952 	rc = parse_pkt(sc, &m);
1953 	if (__predict_false(rc != 0)) {
1954 		MPASS(m == NULL);			/* was freed already */
1955 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1956 		return (rc);
1957 	}
1958 #ifdef RATELIMIT
1959 	if (m->m_pkthdr.snd_tag != NULL) {
1960 		/* EAGAIN tells the stack we are not the correct interface. */
1961 		if (__predict_false(ifp != m->m_pkthdr.snd_tag->ifp)) {
1962 			m_freem(m);
1963 			return (EAGAIN);
1964 		}
1965 
1966 		return (ethofld_transmit(ifp, m));
1967 	}
1968 #endif
1969 
1970 	/* Select a txq. */
1971 	txq = &sc->sge.txq[vi->first_txq];
1972 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1973 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1974 		    vi->rsrv_noflowq);
1975 
1976 	items[0] = m;
1977 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1978 	if (__predict_false(rc != 0))
1979 		m_freem(m);
1980 
1981 	return (rc);
1982 }
1983 
1984 static void
1985 cxgbe_qflush(struct ifnet *ifp)
1986 {
1987 	struct vi_info *vi = ifp->if_softc;
1988 	struct sge_txq *txq;
1989 	int i;
1990 
1991 	/* queues do not exist if !VI_INIT_DONE. */
1992 	if (vi->flags & VI_INIT_DONE) {
1993 		for_each_txq(vi, i, txq) {
1994 			TXQ_LOCK(txq);
1995 			txq->eq.flags |= EQ_QFLUSH;
1996 			TXQ_UNLOCK(txq);
1997 			while (!mp_ring_is_idle(txq->r)) {
1998 				mp_ring_check_drainage(txq->r, 0);
1999 				pause("qflush", 1);
2000 			}
2001 			TXQ_LOCK(txq);
2002 			txq->eq.flags &= ~EQ_QFLUSH;
2003 			TXQ_UNLOCK(txq);
2004 		}
2005 	}
2006 	if_qflush(ifp);
2007 }
2008 
2009 static uint64_t
2010 vi_get_counter(struct ifnet *ifp, ift_counter c)
2011 {
2012 	struct vi_info *vi = ifp->if_softc;
2013 	struct fw_vi_stats_vf *s = &vi->stats;
2014 
2015 	vi_refresh_stats(vi->pi->adapter, vi);
2016 
2017 	switch (c) {
2018 	case IFCOUNTER_IPACKETS:
2019 		return (s->rx_bcast_frames + s->rx_mcast_frames +
2020 		    s->rx_ucast_frames);
2021 	case IFCOUNTER_IERRORS:
2022 		return (s->rx_err_frames);
2023 	case IFCOUNTER_OPACKETS:
2024 		return (s->tx_bcast_frames + s->tx_mcast_frames +
2025 		    s->tx_ucast_frames + s->tx_offload_frames);
2026 	case IFCOUNTER_OERRORS:
2027 		return (s->tx_drop_frames);
2028 	case IFCOUNTER_IBYTES:
2029 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
2030 		    s->rx_ucast_bytes);
2031 	case IFCOUNTER_OBYTES:
2032 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
2033 		    s->tx_ucast_bytes + s->tx_offload_bytes);
2034 	case IFCOUNTER_IMCASTS:
2035 		return (s->rx_mcast_frames);
2036 	case IFCOUNTER_OMCASTS:
2037 		return (s->tx_mcast_frames);
2038 	case IFCOUNTER_OQDROPS: {
2039 		uint64_t drops;
2040 
2041 		drops = 0;
2042 		if (vi->flags & VI_INIT_DONE) {
2043 			int i;
2044 			struct sge_txq *txq;
2045 
2046 			for_each_txq(vi, i, txq)
2047 				drops += counter_u64_fetch(txq->r->drops);
2048 		}
2049 
2050 		return (drops);
2051 
2052 	}
2053 
2054 	default:
2055 		return (if_get_counter_default(ifp, c));
2056 	}
2057 }
2058 
2059 uint64_t
2060 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
2061 {
2062 	struct vi_info *vi = ifp->if_softc;
2063 	struct port_info *pi = vi->pi;
2064 	struct adapter *sc = pi->adapter;
2065 	struct port_stats *s = &pi->stats;
2066 
2067 	if (pi->nvi > 1 || sc->flags & IS_VF)
2068 		return (vi_get_counter(ifp, c));
2069 
2070 	cxgbe_refresh_stats(sc, pi);
2071 
2072 	switch (c) {
2073 	case IFCOUNTER_IPACKETS:
2074 		return (s->rx_frames);
2075 
2076 	case IFCOUNTER_IERRORS:
2077 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
2078 		    s->rx_fcs_err + s->rx_len_err);
2079 
2080 	case IFCOUNTER_OPACKETS:
2081 		return (s->tx_frames);
2082 
2083 	case IFCOUNTER_OERRORS:
2084 		return (s->tx_error_frames);
2085 
2086 	case IFCOUNTER_IBYTES:
2087 		return (s->rx_octets);
2088 
2089 	case IFCOUNTER_OBYTES:
2090 		return (s->tx_octets);
2091 
2092 	case IFCOUNTER_IMCASTS:
2093 		return (s->rx_mcast_frames);
2094 
2095 	case IFCOUNTER_OMCASTS:
2096 		return (s->tx_mcast_frames);
2097 
2098 	case IFCOUNTER_IQDROPS:
2099 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
2100 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
2101 		    s->rx_trunc3 + pi->tnl_cong_drops);
2102 
2103 	case IFCOUNTER_OQDROPS: {
2104 		uint64_t drops;
2105 
2106 		drops = s->tx_drop;
2107 		if (vi->flags & VI_INIT_DONE) {
2108 			int i;
2109 			struct sge_txq *txq;
2110 
2111 			for_each_txq(vi, i, txq)
2112 				drops += counter_u64_fetch(txq->r->drops);
2113 		}
2114 
2115 		return (drops);
2116 
2117 	}
2118 
2119 	default:
2120 		return (if_get_counter_default(ifp, c));
2121 	}
2122 }
2123 
2124 /*
2125  * The kernel picks a media from the list we had provided but we still validate
2126  * the requeste.
2127  */
2128 int
2129 cxgbe_media_change(struct ifnet *ifp)
2130 {
2131 	struct vi_info *vi = ifp->if_softc;
2132 	struct port_info *pi = vi->pi;
2133 	struct ifmedia *ifm = &pi->media;
2134 	struct link_config *lc = &pi->link_cfg;
2135 	struct adapter *sc = pi->adapter;
2136 	int rc;
2137 
2138 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mec");
2139 	if (rc != 0)
2140 		return (rc);
2141 	PORT_LOCK(pi);
2142 	if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
2143 		/* ifconfig .. media autoselect */
2144 		if (!(lc->supported & FW_PORT_CAP32_ANEG)) {
2145 			rc = ENOTSUP; /* AN not supported by transceiver */
2146 			goto done;
2147 		}
2148 		lc->requested_aneg = AUTONEG_ENABLE;
2149 		lc->requested_speed = 0;
2150 		lc->requested_fc |= PAUSE_AUTONEG;
2151 	} else {
2152 		lc->requested_aneg = AUTONEG_DISABLE;
2153 		lc->requested_speed =
2154 		    ifmedia_baudrate(ifm->ifm_media) / 1000000;
2155 		lc->requested_fc = 0;
2156 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
2157 			lc->requested_fc |= PAUSE_RX;
2158 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
2159 			lc->requested_fc |= PAUSE_TX;
2160 	}
2161 	if (pi->up_vis > 0) {
2162 		fixup_link_config(pi);
2163 		rc = apply_link_config(pi);
2164 	}
2165 done:
2166 	PORT_UNLOCK(pi);
2167 	end_synchronized_op(sc, 0);
2168 	return (rc);
2169 }
2170 
2171 /*
2172  * Base media word (without ETHER, pause, link active, etc.) for the port at the
2173  * given speed.
2174  */
2175 static int
2176 port_mword(struct port_info *pi, uint32_t speed)
2177 {
2178 
2179 	MPASS(speed & M_FW_PORT_CAP32_SPEED);
2180 	MPASS(powerof2(speed));
2181 
2182 	switch(pi->port_type) {
2183 	case FW_PORT_TYPE_BT_SGMII:
2184 	case FW_PORT_TYPE_BT_XFI:
2185 	case FW_PORT_TYPE_BT_XAUI:
2186 		/* BaseT */
2187 		switch (speed) {
2188 		case FW_PORT_CAP32_SPEED_100M:
2189 			return (IFM_100_T);
2190 		case FW_PORT_CAP32_SPEED_1G:
2191 			return (IFM_1000_T);
2192 		case FW_PORT_CAP32_SPEED_10G:
2193 			return (IFM_10G_T);
2194 		}
2195 		break;
2196 	case FW_PORT_TYPE_KX4:
2197 		if (speed == FW_PORT_CAP32_SPEED_10G)
2198 			return (IFM_10G_KX4);
2199 		break;
2200 	case FW_PORT_TYPE_CX4:
2201 		if (speed == FW_PORT_CAP32_SPEED_10G)
2202 			return (IFM_10G_CX4);
2203 		break;
2204 	case FW_PORT_TYPE_KX:
2205 		if (speed == FW_PORT_CAP32_SPEED_1G)
2206 			return (IFM_1000_KX);
2207 		break;
2208 	case FW_PORT_TYPE_KR:
2209 	case FW_PORT_TYPE_BP_AP:
2210 	case FW_PORT_TYPE_BP4_AP:
2211 	case FW_PORT_TYPE_BP40_BA:
2212 	case FW_PORT_TYPE_KR4_100G:
2213 	case FW_PORT_TYPE_KR_SFP28:
2214 	case FW_PORT_TYPE_KR_XLAUI:
2215 		switch (speed) {
2216 		case FW_PORT_CAP32_SPEED_1G:
2217 			return (IFM_1000_KX);
2218 		case FW_PORT_CAP32_SPEED_10G:
2219 			return (IFM_10G_KR);
2220 		case FW_PORT_CAP32_SPEED_25G:
2221 			return (IFM_25G_KR);
2222 		case FW_PORT_CAP32_SPEED_40G:
2223 			return (IFM_40G_KR4);
2224 		case FW_PORT_CAP32_SPEED_50G:
2225 			return (IFM_50G_KR2);
2226 		case FW_PORT_CAP32_SPEED_100G:
2227 			return (IFM_100G_KR4);
2228 		}
2229 		break;
2230 	case FW_PORT_TYPE_FIBER_XFI:
2231 	case FW_PORT_TYPE_FIBER_XAUI:
2232 	case FW_PORT_TYPE_SFP:
2233 	case FW_PORT_TYPE_QSFP_10G:
2234 	case FW_PORT_TYPE_QSA:
2235 	case FW_PORT_TYPE_QSFP:
2236 	case FW_PORT_TYPE_CR4_QSFP:
2237 	case FW_PORT_TYPE_CR_QSFP:
2238 	case FW_PORT_TYPE_CR2_QSFP:
2239 	case FW_PORT_TYPE_SFP28:
2240 		/* Pluggable transceiver */
2241 		switch (pi->mod_type) {
2242 		case FW_PORT_MOD_TYPE_LR:
2243 			switch (speed) {
2244 			case FW_PORT_CAP32_SPEED_1G:
2245 				return (IFM_1000_LX);
2246 			case FW_PORT_CAP32_SPEED_10G:
2247 				return (IFM_10G_LR);
2248 			case FW_PORT_CAP32_SPEED_25G:
2249 				return (IFM_25G_LR);
2250 			case FW_PORT_CAP32_SPEED_40G:
2251 				return (IFM_40G_LR4);
2252 			case FW_PORT_CAP32_SPEED_50G:
2253 				return (IFM_50G_LR2);
2254 			case FW_PORT_CAP32_SPEED_100G:
2255 				return (IFM_100G_LR4);
2256 			}
2257 			break;
2258 		case FW_PORT_MOD_TYPE_SR:
2259 			switch (speed) {
2260 			case FW_PORT_CAP32_SPEED_1G:
2261 				return (IFM_1000_SX);
2262 			case FW_PORT_CAP32_SPEED_10G:
2263 				return (IFM_10G_SR);
2264 			case FW_PORT_CAP32_SPEED_25G:
2265 				return (IFM_25G_SR);
2266 			case FW_PORT_CAP32_SPEED_40G:
2267 				return (IFM_40G_SR4);
2268 			case FW_PORT_CAP32_SPEED_50G:
2269 				return (IFM_50G_SR2);
2270 			case FW_PORT_CAP32_SPEED_100G:
2271 				return (IFM_100G_SR4);
2272 			}
2273 			break;
2274 		case FW_PORT_MOD_TYPE_ER:
2275 			if (speed == FW_PORT_CAP32_SPEED_10G)
2276 				return (IFM_10G_ER);
2277 			break;
2278 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2279 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2280 			switch (speed) {
2281 			case FW_PORT_CAP32_SPEED_1G:
2282 				return (IFM_1000_CX);
2283 			case FW_PORT_CAP32_SPEED_10G:
2284 				return (IFM_10G_TWINAX);
2285 			case FW_PORT_CAP32_SPEED_25G:
2286 				return (IFM_25G_CR);
2287 			case FW_PORT_CAP32_SPEED_40G:
2288 				return (IFM_40G_CR4);
2289 			case FW_PORT_CAP32_SPEED_50G:
2290 				return (IFM_50G_CR2);
2291 			case FW_PORT_CAP32_SPEED_100G:
2292 				return (IFM_100G_CR4);
2293 			}
2294 			break;
2295 		case FW_PORT_MOD_TYPE_LRM:
2296 			if (speed == FW_PORT_CAP32_SPEED_10G)
2297 				return (IFM_10G_LRM);
2298 			break;
2299 		case FW_PORT_MOD_TYPE_NA:
2300 			MPASS(0);	/* Not pluggable? */
2301 			/* fall throough */
2302 		case FW_PORT_MOD_TYPE_ERROR:
2303 		case FW_PORT_MOD_TYPE_UNKNOWN:
2304 		case FW_PORT_MOD_TYPE_NOTSUPPORTED:
2305 			break;
2306 		case FW_PORT_MOD_TYPE_NONE:
2307 			return (IFM_NONE);
2308 		}
2309 		break;
2310 	case FW_PORT_TYPE_NONE:
2311 		return (IFM_NONE);
2312 	}
2313 
2314 	return (IFM_UNKNOWN);
2315 }
2316 
2317 void
2318 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2319 {
2320 	struct vi_info *vi = ifp->if_softc;
2321 	struct port_info *pi = vi->pi;
2322 	struct adapter *sc = pi->adapter;
2323 	struct link_config *lc = &pi->link_cfg;
2324 
2325 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4med") != 0)
2326 		return;
2327 	PORT_LOCK(pi);
2328 
2329 	if (pi->up_vis == 0) {
2330 		/*
2331 		 * If all the interfaces are administratively down the firmware
2332 		 * does not report transceiver changes.  Refresh port info here
2333 		 * so that ifconfig displays accurate ifmedia at all times.
2334 		 * This is the only reason we have a synchronized op in this
2335 		 * function.  Just PORT_LOCK would have been enough otherwise.
2336 		 */
2337 		t4_update_port_info(pi);
2338 		build_medialist(pi);
2339 	}
2340 
2341 	/* ifm_status */
2342 	ifmr->ifm_status = IFM_AVALID;
2343 	if (lc->link_ok == false)
2344 		goto done;
2345 	ifmr->ifm_status |= IFM_ACTIVE;
2346 
2347 	/* ifm_active */
2348 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2349 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2350 	if (lc->fc & PAUSE_RX)
2351 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2352 	if (lc->fc & PAUSE_TX)
2353 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2354 	ifmr->ifm_active |= port_mword(pi, speed_to_fwcap(lc->speed));
2355 done:
2356 	PORT_UNLOCK(pi);
2357 	end_synchronized_op(sc, 0);
2358 }
2359 
2360 static int
2361 vcxgbe_probe(device_t dev)
2362 {
2363 	char buf[128];
2364 	struct vi_info *vi = device_get_softc(dev);
2365 
2366 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2367 	    vi - vi->pi->vi);
2368 	device_set_desc_copy(dev, buf);
2369 
2370 	return (BUS_PROBE_DEFAULT);
2371 }
2372 
2373 static int
2374 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2375 {
2376 	int func, index, rc;
2377 	uint32_t param, val;
2378 
2379 	ASSERT_SYNCHRONIZED_OP(sc);
2380 
2381 	index = vi - pi->vi;
2382 	MPASS(index > 0);	/* This function deals with _extra_ VIs only */
2383 	KASSERT(index < nitems(vi_mac_funcs),
2384 	    ("%s: VI %s doesn't have a MAC func", __func__,
2385 	    device_get_nameunit(vi->dev)));
2386 	func = vi_mac_funcs[index];
2387 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2388 	    vi->hw_addr, &vi->rss_size, func, 0);
2389 	if (rc < 0) {
2390 		device_printf(vi->dev, "failed to allocate virtual interface %d"
2391 		    "for port %d: %d\n", index, pi->port_id, -rc);
2392 		return (-rc);
2393 	}
2394 	vi->viid = rc;
2395 	if (chip_id(sc) <= CHELSIO_T5)
2396 		vi->smt_idx = (rc & 0x7f) << 1;
2397 	else
2398 		vi->smt_idx = (rc & 0x7f);
2399 
2400 	if (vi->rss_size == 1) {
2401 		/*
2402 		 * This VI didn't get a slice of the RSS table.  Reduce the
2403 		 * number of VIs being created (hw.cxgbe.num_vis) or modify the
2404 		 * configuration file (nvi, rssnvi for this PF) if this is a
2405 		 * problem.
2406 		 */
2407 		device_printf(vi->dev, "RSS table not available.\n");
2408 		vi->rss_base = 0xffff;
2409 
2410 		return (0);
2411 	}
2412 
2413 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2414 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2415 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2416 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2417 	if (rc)
2418 		vi->rss_base = 0xffff;
2419 	else {
2420 		MPASS((val >> 16) == vi->rss_size);
2421 		vi->rss_base = val & 0xffff;
2422 	}
2423 
2424 	return (0);
2425 }
2426 
2427 static int
2428 vcxgbe_attach(device_t dev)
2429 {
2430 	struct vi_info *vi;
2431 	struct port_info *pi;
2432 	struct adapter *sc;
2433 	int rc;
2434 
2435 	vi = device_get_softc(dev);
2436 	pi = vi->pi;
2437 	sc = pi->adapter;
2438 
2439 	rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2440 	if (rc)
2441 		return (rc);
2442 	rc = alloc_extra_vi(sc, pi, vi);
2443 	end_synchronized_op(sc, 0);
2444 	if (rc)
2445 		return (rc);
2446 
2447 	rc = cxgbe_vi_attach(dev, vi);
2448 	if (rc) {
2449 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2450 		return (rc);
2451 	}
2452 	return (0);
2453 }
2454 
2455 static int
2456 vcxgbe_detach(device_t dev)
2457 {
2458 	struct vi_info *vi;
2459 	struct adapter *sc;
2460 
2461 	vi = device_get_softc(dev);
2462 	sc = vi->pi->adapter;
2463 
2464 	doom_vi(sc, vi);
2465 
2466 	cxgbe_vi_detach(vi);
2467 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2468 
2469 	end_synchronized_op(sc, 0);
2470 
2471 	return (0);
2472 }
2473 
2474 void
2475 t4_fatal_err(struct adapter *sc)
2476 {
2477 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2478 	t4_intr_disable(sc);
2479 	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2480 	    device_get_nameunit(sc->dev));
2481 	if (t4_panic_on_fatal_err)
2482 		panic("panic requested on fatal error");
2483 }
2484 
2485 void
2486 t4_add_adapter(struct adapter *sc)
2487 {
2488 	sx_xlock(&t4_list_lock);
2489 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2490 	sx_xunlock(&t4_list_lock);
2491 }
2492 
2493 int
2494 t4_map_bars_0_and_4(struct adapter *sc)
2495 {
2496 	sc->regs_rid = PCIR_BAR(0);
2497 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2498 	    &sc->regs_rid, RF_ACTIVE);
2499 	if (sc->regs_res == NULL) {
2500 		device_printf(sc->dev, "cannot map registers.\n");
2501 		return (ENXIO);
2502 	}
2503 	sc->bt = rman_get_bustag(sc->regs_res);
2504 	sc->bh = rman_get_bushandle(sc->regs_res);
2505 	sc->mmio_len = rman_get_size(sc->regs_res);
2506 	setbit(&sc->doorbells, DOORBELL_KDB);
2507 
2508 	sc->msix_rid = PCIR_BAR(4);
2509 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2510 	    &sc->msix_rid, RF_ACTIVE);
2511 	if (sc->msix_res == NULL) {
2512 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2513 		return (ENXIO);
2514 	}
2515 
2516 	return (0);
2517 }
2518 
2519 int
2520 t4_map_bar_2(struct adapter *sc)
2521 {
2522 
2523 	/*
2524 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2525 	 * to map it if RDMA is disabled.
2526 	 */
2527 	if (is_t4(sc) && sc->rdmacaps == 0)
2528 		return (0);
2529 
2530 	sc->udbs_rid = PCIR_BAR(2);
2531 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2532 	    &sc->udbs_rid, RF_ACTIVE);
2533 	if (sc->udbs_res == NULL) {
2534 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2535 		return (ENXIO);
2536 	}
2537 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2538 
2539 	if (chip_id(sc) >= CHELSIO_T5) {
2540 		setbit(&sc->doorbells, DOORBELL_UDB);
2541 #if defined(__i386__) || defined(__amd64__)
2542 		if (t5_write_combine) {
2543 			int rc, mode;
2544 
2545 			/*
2546 			 * Enable write combining on BAR2.  This is the
2547 			 * userspace doorbell BAR and is split into 128B
2548 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2549 			 * with an egress queue.  The first 64B has the doorbell
2550 			 * and the second 64B can be used to submit a tx work
2551 			 * request with an implicit doorbell.
2552 			 */
2553 
2554 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2555 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2556 			if (rc == 0) {
2557 				clrbit(&sc->doorbells, DOORBELL_UDB);
2558 				setbit(&sc->doorbells, DOORBELL_WCWR);
2559 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2560 			} else {
2561 				device_printf(sc->dev,
2562 				    "couldn't enable write combining: %d\n",
2563 				    rc);
2564 			}
2565 
2566 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2567 			t4_write_reg(sc, A_SGE_STAT_CFG,
2568 			    V_STATSOURCE_T5(7) | mode);
2569 		}
2570 #endif
2571 	}
2572 	sc->iwt.wc_en = isset(&sc->doorbells, DOORBELL_UDBWC) ? 1 : 0;
2573 
2574 	return (0);
2575 }
2576 
2577 struct memwin_init {
2578 	uint32_t base;
2579 	uint32_t aperture;
2580 };
2581 
2582 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2583 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2584 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2585 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2586 };
2587 
2588 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2589 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2590 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2591 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2592 };
2593 
2594 static void
2595 setup_memwin(struct adapter *sc)
2596 {
2597 	const struct memwin_init *mw_init;
2598 	struct memwin *mw;
2599 	int i;
2600 	uint32_t bar0;
2601 
2602 	if (is_t4(sc)) {
2603 		/*
2604 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2605 		 * mechanism.  Works from within PCI passthrough environments
2606 		 * too, where rman_get_start() can return a different value.  We
2607 		 * need to program the T4 memory window decoders with the actual
2608 		 * addresses that will be coming across the PCIe link.
2609 		 */
2610 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2611 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2612 
2613 		mw_init = &t4_memwin[0];
2614 	} else {
2615 		/* T5+ use the relative offset inside the PCIe BAR */
2616 		bar0 = 0;
2617 
2618 		mw_init = &t5_memwin[0];
2619 	}
2620 
2621 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2622 		rw_init(&mw->mw_lock, "memory window access");
2623 		mw->mw_base = mw_init->base;
2624 		mw->mw_aperture = mw_init->aperture;
2625 		mw->mw_curpos = 0;
2626 		t4_write_reg(sc,
2627 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2628 		    (mw->mw_base + bar0) | V_BIR(0) |
2629 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2630 		rw_wlock(&mw->mw_lock);
2631 		position_memwin(sc, i, 0);
2632 		rw_wunlock(&mw->mw_lock);
2633 	}
2634 
2635 	/* flush */
2636 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2637 }
2638 
2639 /*
2640  * Positions the memory window at the given address in the card's address space.
2641  * There are some alignment requirements and the actual position may be at an
2642  * address prior to the requested address.  mw->mw_curpos always has the actual
2643  * position of the window.
2644  */
2645 static void
2646 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2647 {
2648 	struct memwin *mw;
2649 	uint32_t pf;
2650 	uint32_t reg;
2651 
2652 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2653 	mw = &sc->memwin[idx];
2654 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2655 
2656 	if (is_t4(sc)) {
2657 		pf = 0;
2658 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2659 	} else {
2660 		pf = V_PFNUM(sc->pf);
2661 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2662 	}
2663 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2664 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2665 	t4_read_reg(sc, reg);	/* flush */
2666 }
2667 
2668 int
2669 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2670     int len, int rw)
2671 {
2672 	struct memwin *mw;
2673 	uint32_t mw_end, v;
2674 
2675 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2676 
2677 	/* Memory can only be accessed in naturally aligned 4 byte units */
2678 	if (addr & 3 || len & 3 || len <= 0)
2679 		return (EINVAL);
2680 
2681 	mw = &sc->memwin[idx];
2682 	while (len > 0) {
2683 		rw_rlock(&mw->mw_lock);
2684 		mw_end = mw->mw_curpos + mw->mw_aperture;
2685 		if (addr >= mw_end || addr < mw->mw_curpos) {
2686 			/* Will need to reposition the window */
2687 			if (!rw_try_upgrade(&mw->mw_lock)) {
2688 				rw_runlock(&mw->mw_lock);
2689 				rw_wlock(&mw->mw_lock);
2690 			}
2691 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2692 			position_memwin(sc, idx, addr);
2693 			rw_downgrade(&mw->mw_lock);
2694 			mw_end = mw->mw_curpos + mw->mw_aperture;
2695 		}
2696 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2697 		while (addr < mw_end && len > 0) {
2698 			if (rw == 0) {
2699 				v = t4_read_reg(sc, mw->mw_base + addr -
2700 				    mw->mw_curpos);
2701 				*val++ = le32toh(v);
2702 			} else {
2703 				v = *val++;
2704 				t4_write_reg(sc, mw->mw_base + addr -
2705 				    mw->mw_curpos, htole32(v));
2706 			}
2707 			addr += 4;
2708 			len -= 4;
2709 		}
2710 		rw_runlock(&mw->mw_lock);
2711 	}
2712 
2713 	return (0);
2714 }
2715 
2716 int
2717 alloc_atid_tab(struct tid_info *t, int flags)
2718 {
2719 	int i;
2720 
2721 	MPASS(t->natids > 0);
2722 	MPASS(t->atid_tab == NULL);
2723 
2724 	t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
2725 	    M_ZERO | flags);
2726 	if (t->atid_tab == NULL)
2727 		return (ENOMEM);
2728 	mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
2729 	t->afree = t->atid_tab;
2730 	t->atids_in_use = 0;
2731 	for (i = 1; i < t->natids; i++)
2732 		t->atid_tab[i - 1].next = &t->atid_tab[i];
2733 	t->atid_tab[t->natids - 1].next = NULL;
2734 
2735 	return (0);
2736 }
2737 
2738 void
2739 free_atid_tab(struct tid_info *t)
2740 {
2741 
2742 	KASSERT(t->atids_in_use == 0,
2743 	    ("%s: %d atids still in use.", __func__, t->atids_in_use));
2744 
2745 	if (mtx_initialized(&t->atid_lock))
2746 		mtx_destroy(&t->atid_lock);
2747 	free(t->atid_tab, M_CXGBE);
2748 	t->atid_tab = NULL;
2749 }
2750 
2751 int
2752 alloc_atid(struct adapter *sc, void *ctx)
2753 {
2754 	struct tid_info *t = &sc->tids;
2755 	int atid = -1;
2756 
2757 	mtx_lock(&t->atid_lock);
2758 	if (t->afree) {
2759 		union aopen_entry *p = t->afree;
2760 
2761 		atid = p - t->atid_tab;
2762 		MPASS(atid <= M_TID_TID);
2763 		t->afree = p->next;
2764 		p->data = ctx;
2765 		t->atids_in_use++;
2766 	}
2767 	mtx_unlock(&t->atid_lock);
2768 	return (atid);
2769 }
2770 
2771 void *
2772 lookup_atid(struct adapter *sc, int atid)
2773 {
2774 	struct tid_info *t = &sc->tids;
2775 
2776 	return (t->atid_tab[atid].data);
2777 }
2778 
2779 void
2780 free_atid(struct adapter *sc, int atid)
2781 {
2782 	struct tid_info *t = &sc->tids;
2783 	union aopen_entry *p = &t->atid_tab[atid];
2784 
2785 	mtx_lock(&t->atid_lock);
2786 	p->next = t->afree;
2787 	t->afree = p;
2788 	t->atids_in_use--;
2789 	mtx_unlock(&t->atid_lock);
2790 }
2791 
2792 static void
2793 queue_tid_release(struct adapter *sc, int tid)
2794 {
2795 
2796 	CXGBE_UNIMPLEMENTED("deferred tid release");
2797 }
2798 
2799 void
2800 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
2801 {
2802 	struct wrqe *wr;
2803 	struct cpl_tid_release *req;
2804 
2805 	wr = alloc_wrqe(sizeof(*req), ctrlq);
2806 	if (wr == NULL) {
2807 		queue_tid_release(sc, tid);	/* defer */
2808 		return;
2809 	}
2810 	req = wrtod(wr);
2811 
2812 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
2813 
2814 	t4_wrq_tx(sc, wr);
2815 }
2816 
2817 static int
2818 t4_range_cmp(const void *a, const void *b)
2819 {
2820 	return ((const struct t4_range *)a)->start -
2821 	       ((const struct t4_range *)b)->start;
2822 }
2823 
2824 /*
2825  * Verify that the memory range specified by the addr/len pair is valid within
2826  * the card's address space.
2827  */
2828 static int
2829 validate_mem_range(struct adapter *sc, uint32_t addr, uint32_t len)
2830 {
2831 	struct t4_range mem_ranges[4], *r, *next;
2832 	uint32_t em, addr_len;
2833 	int i, n, remaining;
2834 
2835 	/* Memory can only be accessed in naturally aligned 4 byte units */
2836 	if (addr & 3 || len & 3 || len == 0)
2837 		return (EINVAL);
2838 
2839 	/* Enabled memories */
2840 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2841 
2842 	r = &mem_ranges[0];
2843 	n = 0;
2844 	bzero(r, sizeof(mem_ranges));
2845 	if (em & F_EDRAM0_ENABLE) {
2846 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2847 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2848 		if (r->size > 0) {
2849 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2850 			if (addr >= r->start &&
2851 			    addr + len <= r->start + r->size)
2852 				return (0);
2853 			r++;
2854 			n++;
2855 		}
2856 	}
2857 	if (em & F_EDRAM1_ENABLE) {
2858 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2859 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2860 		if (r->size > 0) {
2861 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2862 			if (addr >= r->start &&
2863 			    addr + len <= r->start + r->size)
2864 				return (0);
2865 			r++;
2866 			n++;
2867 		}
2868 	}
2869 	if (em & F_EXT_MEM_ENABLE) {
2870 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2871 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2872 		if (r->size > 0) {
2873 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2874 			if (addr >= r->start &&
2875 			    addr + len <= r->start + r->size)
2876 				return (0);
2877 			r++;
2878 			n++;
2879 		}
2880 	}
2881 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2882 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2883 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2884 		if (r->size > 0) {
2885 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2886 			if (addr >= r->start &&
2887 			    addr + len <= r->start + r->size)
2888 				return (0);
2889 			r++;
2890 			n++;
2891 		}
2892 	}
2893 	MPASS(n <= nitems(mem_ranges));
2894 
2895 	if (n > 1) {
2896 		/* Sort and merge the ranges. */
2897 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2898 
2899 		/* Start from index 0 and examine the next n - 1 entries. */
2900 		r = &mem_ranges[0];
2901 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2902 
2903 			MPASS(r->size > 0);	/* r is a valid entry. */
2904 			next = r + 1;
2905 			MPASS(next->size > 0);	/* and so is the next one. */
2906 
2907 			while (r->start + r->size >= next->start) {
2908 				/* Merge the next one into the current entry. */
2909 				r->size = max(r->start + r->size,
2910 				    next->start + next->size) - r->start;
2911 				n--;	/* One fewer entry in total. */
2912 				if (--remaining == 0)
2913 					goto done;	/* short circuit */
2914 				next++;
2915 			}
2916 			if (next != r + 1) {
2917 				/*
2918 				 * Some entries were merged into r and next
2919 				 * points to the first valid entry that couldn't
2920 				 * be merged.
2921 				 */
2922 				MPASS(next->size > 0);	/* must be valid */
2923 				memcpy(r + 1, next, remaining * sizeof(*r));
2924 #ifdef INVARIANTS
2925 				/*
2926 				 * This so that the foo->size assertion in the
2927 				 * next iteration of the loop do the right
2928 				 * thing for entries that were pulled up and are
2929 				 * no longer valid.
2930 				 */
2931 				MPASS(n < nitems(mem_ranges));
2932 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2933 				    sizeof(struct t4_range));
2934 #endif
2935 			}
2936 		}
2937 done:
2938 		/* Done merging the ranges. */
2939 		MPASS(n > 0);
2940 		r = &mem_ranges[0];
2941 		for (i = 0; i < n; i++, r++) {
2942 			if (addr >= r->start &&
2943 			    addr + len <= r->start + r->size)
2944 				return (0);
2945 		}
2946 	}
2947 
2948 	return (EFAULT);
2949 }
2950 
2951 static int
2952 fwmtype_to_hwmtype(int mtype)
2953 {
2954 
2955 	switch (mtype) {
2956 	case FW_MEMTYPE_EDC0:
2957 		return (MEM_EDC0);
2958 	case FW_MEMTYPE_EDC1:
2959 		return (MEM_EDC1);
2960 	case FW_MEMTYPE_EXTMEM:
2961 		return (MEM_MC0);
2962 	case FW_MEMTYPE_EXTMEM1:
2963 		return (MEM_MC1);
2964 	default:
2965 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2966 	}
2967 }
2968 
2969 /*
2970  * Verify that the memory range specified by the memtype/offset/len pair is
2971  * valid and lies entirely within the memtype specified.  The global address of
2972  * the start of the range is returned in addr.
2973  */
2974 static int
2975 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, uint32_t len,
2976     uint32_t *addr)
2977 {
2978 	uint32_t em, addr_len, maddr;
2979 
2980 	/* Memory can only be accessed in naturally aligned 4 byte units */
2981 	if (off & 3 || len & 3 || len == 0)
2982 		return (EINVAL);
2983 
2984 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2985 	switch (fwmtype_to_hwmtype(mtype)) {
2986 	case MEM_EDC0:
2987 		if (!(em & F_EDRAM0_ENABLE))
2988 			return (EINVAL);
2989 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2990 		maddr = G_EDRAM0_BASE(addr_len) << 20;
2991 		break;
2992 	case MEM_EDC1:
2993 		if (!(em & F_EDRAM1_ENABLE))
2994 			return (EINVAL);
2995 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2996 		maddr = G_EDRAM1_BASE(addr_len) << 20;
2997 		break;
2998 	case MEM_MC:
2999 		if (!(em & F_EXT_MEM_ENABLE))
3000 			return (EINVAL);
3001 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
3002 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
3003 		break;
3004 	case MEM_MC1:
3005 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
3006 			return (EINVAL);
3007 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
3008 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
3009 		break;
3010 	default:
3011 		return (EINVAL);
3012 	}
3013 
3014 	*addr = maddr + off;	/* global address */
3015 	return (validate_mem_range(sc, *addr, len));
3016 }
3017 
3018 static int
3019 fixup_devlog_params(struct adapter *sc)
3020 {
3021 	struct devlog_params *dparams = &sc->params.devlog;
3022 	int rc;
3023 
3024 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
3025 	    dparams->size, &dparams->addr);
3026 
3027 	return (rc);
3028 }
3029 
3030 static void
3031 update_nirq(struct intrs_and_queues *iaq, int nports)
3032 {
3033 	int extra = T4_EXTRA_INTR;
3034 
3035 	iaq->nirq = extra;
3036 	iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
3037 	iaq->nirq += nports * (iaq->num_vis - 1) *
3038 	    max(iaq->nrxq_vi, iaq->nnmrxq_vi);
3039 	iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
3040 }
3041 
3042 /*
3043  * Adjust requirements to fit the number of interrupts available.
3044  */
3045 static void
3046 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
3047     int navail)
3048 {
3049 	int old_nirq;
3050 	const int nports = sc->params.nports;
3051 
3052 	MPASS(nports > 0);
3053 	MPASS(navail > 0);
3054 
3055 	bzero(iaq, sizeof(*iaq));
3056 	iaq->intr_type = itype;
3057 	iaq->num_vis = t4_num_vis;
3058 	iaq->ntxq = t4_ntxq;
3059 	iaq->ntxq_vi = t4_ntxq_vi;
3060 	iaq->nrxq = t4_nrxq;
3061 	iaq->nrxq_vi = t4_nrxq_vi;
3062 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
3063 	if (is_offload(sc) || is_ethoffload(sc)) {
3064 		iaq->nofldtxq = t4_nofldtxq;
3065 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
3066 	}
3067 #endif
3068 #ifdef TCP_OFFLOAD
3069 	if (is_offload(sc)) {
3070 		iaq->nofldrxq = t4_nofldrxq;
3071 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
3072 	}
3073 #endif
3074 #ifdef DEV_NETMAP
3075 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
3076 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
3077 #endif
3078 
3079 	update_nirq(iaq, nports);
3080 	if (iaq->nirq <= navail &&
3081 	    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3082 		/*
3083 		 * This is the normal case -- there are enough interrupts for
3084 		 * everything.
3085 		 */
3086 		goto done;
3087 	}
3088 
3089 	/*
3090 	 * If extra VIs have been configured try reducing their count and see if
3091 	 * that works.
3092 	 */
3093 	while (iaq->num_vis > 1) {
3094 		iaq->num_vis--;
3095 		update_nirq(iaq, nports);
3096 		if (iaq->nirq <= navail &&
3097 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3098 			device_printf(sc->dev, "virtual interfaces per port "
3099 			    "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
3100 			    "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
3101 			    "itype %d, navail %u, nirq %d.\n",
3102 			    iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
3103 			    iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
3104 			    itype, navail, iaq->nirq);
3105 			goto done;
3106 		}
3107 	}
3108 
3109 	/*
3110 	 * Extra VIs will not be created.  Log a message if they were requested.
3111 	 */
3112 	MPASS(iaq->num_vis == 1);
3113 	iaq->ntxq_vi = iaq->nrxq_vi = 0;
3114 	iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
3115 	iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
3116 	if (iaq->num_vis != t4_num_vis) {
3117 		device_printf(sc->dev, "extra virtual interfaces disabled.  "
3118 		    "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
3119 		    "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
3120 		    iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
3121 		    iaq->nnmrxq_vi, itype, navail, iaq->nirq);
3122 	}
3123 
3124 	/*
3125 	 * Keep reducing the number of NIC rx queues to the next lower power of
3126 	 * 2 (for even RSS distribution) and halving the TOE rx queues and see
3127 	 * if that works.
3128 	 */
3129 	do {
3130 		if (iaq->nrxq > 1) {
3131 			do {
3132 				iaq->nrxq--;
3133 			} while (!powerof2(iaq->nrxq));
3134 		}
3135 		if (iaq->nofldrxq > 1)
3136 			iaq->nofldrxq >>= 1;
3137 
3138 		old_nirq = iaq->nirq;
3139 		update_nirq(iaq, nports);
3140 		if (iaq->nirq <= navail &&
3141 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3142 			device_printf(sc->dev, "running with reduced number of "
3143 			    "rx queues because of shortage of interrupts.  "
3144 			    "nrxq=%u, nofldrxq=%u.  "
3145 			    "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
3146 			    iaq->nofldrxq, itype, navail, iaq->nirq);
3147 			goto done;
3148 		}
3149 	} while (old_nirq != iaq->nirq);
3150 
3151 	/* One interrupt for everything.  Ugh. */
3152 	device_printf(sc->dev, "running with minimal number of queues.  "
3153 	    "itype %d, navail %u.\n", itype, navail);
3154 	iaq->nirq = 1;
3155 	MPASS(iaq->nrxq == 1);
3156 	iaq->ntxq = 1;
3157 	if (iaq->nofldrxq > 1)
3158 		iaq->nofldtxq = 1;
3159 done:
3160 	MPASS(iaq->num_vis > 0);
3161 	if (iaq->num_vis > 1) {
3162 		MPASS(iaq->nrxq_vi > 0);
3163 		MPASS(iaq->ntxq_vi > 0);
3164 	}
3165 	MPASS(iaq->nirq > 0);
3166 	MPASS(iaq->nrxq > 0);
3167 	MPASS(iaq->ntxq > 0);
3168 	if (itype == INTR_MSI) {
3169 		MPASS(powerof2(iaq->nirq));
3170 	}
3171 }
3172 
3173 static int
3174 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
3175 {
3176 	int rc, itype, navail, nalloc;
3177 
3178 	for (itype = INTR_MSIX; itype; itype >>= 1) {
3179 
3180 		if ((itype & t4_intr_types) == 0)
3181 			continue;	/* not allowed */
3182 
3183 		if (itype == INTR_MSIX)
3184 			navail = pci_msix_count(sc->dev);
3185 		else if (itype == INTR_MSI)
3186 			navail = pci_msi_count(sc->dev);
3187 		else
3188 			navail = 1;
3189 restart:
3190 		if (navail == 0)
3191 			continue;
3192 
3193 		calculate_iaq(sc, iaq, itype, navail);
3194 		nalloc = iaq->nirq;
3195 		rc = 0;
3196 		if (itype == INTR_MSIX)
3197 			rc = pci_alloc_msix(sc->dev, &nalloc);
3198 		else if (itype == INTR_MSI)
3199 			rc = pci_alloc_msi(sc->dev, &nalloc);
3200 
3201 		if (rc == 0 && nalloc > 0) {
3202 			if (nalloc == iaq->nirq)
3203 				return (0);
3204 
3205 			/*
3206 			 * Didn't get the number requested.  Use whatever number
3207 			 * the kernel is willing to allocate.
3208 			 */
3209 			device_printf(sc->dev, "fewer vectors than requested, "
3210 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
3211 			    itype, iaq->nirq, nalloc);
3212 			pci_release_msi(sc->dev);
3213 			navail = nalloc;
3214 			goto restart;
3215 		}
3216 
3217 		device_printf(sc->dev,
3218 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
3219 		    itype, rc, iaq->nirq, nalloc);
3220 	}
3221 
3222 	device_printf(sc->dev,
3223 	    "failed to find a usable interrupt type.  "
3224 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
3225 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
3226 
3227 	return (ENXIO);
3228 }
3229 
3230 #define FW_VERSION(chip) ( \
3231     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
3232     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
3233     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
3234     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
3235 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
3236 
3237 struct fw_info {
3238 	uint8_t chip;
3239 	char *kld_name;
3240 	char *fw_mod_name;
3241 	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
3242 } fw_info[] = {
3243 	{
3244 		.chip = CHELSIO_T4,
3245 		.kld_name = "t4fw_cfg",
3246 		.fw_mod_name = "t4fw",
3247 		.fw_hdr = {
3248 			.chip = FW_HDR_CHIP_T4,
3249 			.fw_ver = htobe32(FW_VERSION(T4)),
3250 			.intfver_nic = FW_INTFVER(T4, NIC),
3251 			.intfver_vnic = FW_INTFVER(T4, VNIC),
3252 			.intfver_ofld = FW_INTFVER(T4, OFLD),
3253 			.intfver_ri = FW_INTFVER(T4, RI),
3254 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
3255 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
3256 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
3257 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
3258 		},
3259 	}, {
3260 		.chip = CHELSIO_T5,
3261 		.kld_name = "t5fw_cfg",
3262 		.fw_mod_name = "t5fw",
3263 		.fw_hdr = {
3264 			.chip = FW_HDR_CHIP_T5,
3265 			.fw_ver = htobe32(FW_VERSION(T5)),
3266 			.intfver_nic = FW_INTFVER(T5, NIC),
3267 			.intfver_vnic = FW_INTFVER(T5, VNIC),
3268 			.intfver_ofld = FW_INTFVER(T5, OFLD),
3269 			.intfver_ri = FW_INTFVER(T5, RI),
3270 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
3271 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
3272 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
3273 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
3274 		},
3275 	}, {
3276 		.chip = CHELSIO_T6,
3277 		.kld_name = "t6fw_cfg",
3278 		.fw_mod_name = "t6fw",
3279 		.fw_hdr = {
3280 			.chip = FW_HDR_CHIP_T6,
3281 			.fw_ver = htobe32(FW_VERSION(T6)),
3282 			.intfver_nic = FW_INTFVER(T6, NIC),
3283 			.intfver_vnic = FW_INTFVER(T6, VNIC),
3284 			.intfver_ofld = FW_INTFVER(T6, OFLD),
3285 			.intfver_ri = FW_INTFVER(T6, RI),
3286 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3287 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
3288 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3289 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
3290 		},
3291 	}
3292 };
3293 
3294 static struct fw_info *
3295 find_fw_info(int chip)
3296 {
3297 	int i;
3298 
3299 	for (i = 0; i < nitems(fw_info); i++) {
3300 		if (fw_info[i].chip == chip)
3301 			return (&fw_info[i]);
3302 	}
3303 	return (NULL);
3304 }
3305 
3306 /*
3307  * Is the given firmware API compatible with the one the driver was compiled
3308  * with?
3309  */
3310 static int
3311 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
3312 {
3313 
3314 	/* short circuit if it's the exact same firmware version */
3315 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
3316 		return (1);
3317 
3318 	/*
3319 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
3320 	 * features that are supported in the driver.
3321 	 */
3322 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
3323 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
3324 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
3325 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
3326 		return (1);
3327 #undef SAME_INTF
3328 
3329 	return (0);
3330 }
3331 
3332 /*
3333  * The firmware in the KLD is usable, but should it be installed?  This routine
3334  * explains itself in detail if it indicates the KLD firmware should be
3335  * installed.
3336  */
3337 static int
3338 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
3339 {
3340 	const char *reason;
3341 
3342 	if (!card_fw_usable) {
3343 		reason = "incompatible or unusable";
3344 		goto install;
3345 	}
3346 
3347 	if (k > c) {
3348 		reason = "older than the version bundled with this driver";
3349 		goto install;
3350 	}
3351 
3352 	if (t4_fw_install == 2 && k != c) {
3353 		reason = "different than the version bundled with this driver";
3354 		goto install;
3355 	}
3356 
3357 	return (0);
3358 
3359 install:
3360 	if (t4_fw_install == 0) {
3361 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3362 		    "but the driver is prohibited from installing a different "
3363 		    "firmware on the card.\n",
3364 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3365 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3366 
3367 		return (0);
3368 	}
3369 
3370 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3371 	    "installing firmware %u.%u.%u.%u on card.\n",
3372 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3373 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3374 	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3375 	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3376 
3377 	return (1);
3378 }
3379 
3380 /*
3381  * Establish contact with the firmware and determine if we are the master driver
3382  * or not, and whether we are responsible for chip initialization.
3383  */
3384 static int
3385 prep_firmware(struct adapter *sc)
3386 {
3387 	const struct firmware *fw = NULL, *default_cfg;
3388 	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
3389 	enum dev_state state;
3390 	struct fw_info *fw_info;
3391 	struct fw_hdr *card_fw;		/* fw on the card */
3392 	const struct fw_hdr *kld_fw;	/* fw in the KLD */
3393 	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
3394 					   against */
3395 
3396 	/* This is the firmware whose headers the driver was compiled against */
3397 	fw_info = find_fw_info(chip_id(sc));
3398 	if (fw_info == NULL) {
3399 		device_printf(sc->dev,
3400 		    "unable to look up firmware information for chip %d.\n",
3401 		    chip_id(sc));
3402 		return (EINVAL);
3403 	}
3404 	drv_fw = &fw_info->fw_hdr;
3405 
3406 	/*
3407 	 * The firmware KLD contains many modules.  The KLD name is also the
3408 	 * name of the module that contains the default config file.
3409 	 */
3410 	default_cfg = firmware_get(fw_info->kld_name);
3411 
3412 	/* This is the firmware in the KLD */
3413 	fw = firmware_get(fw_info->fw_mod_name);
3414 	if (fw != NULL) {
3415 		kld_fw = (const void *)fw->data;
3416 		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3417 	} else {
3418 		kld_fw = NULL;
3419 		kld_fw_usable = 0;
3420 	}
3421 
3422 	/* Read the header of the firmware on the card */
3423 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3424 	rc = -t4_read_flash(sc, FLASH_FW_START,
3425 	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
3426 	if (rc == 0) {
3427 		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
3428 		if (card_fw->fw_ver == be32toh(0xffffffff)) {
3429 			uint32_t d = be32toh(kld_fw->fw_ver);
3430 
3431 			if (!kld_fw_usable) {
3432 				device_printf(sc->dev,
3433 				    "no firmware on the card and no usable "
3434 				    "firmware bundled with the driver.\n");
3435 				rc = EIO;
3436 				goto done;
3437 			} else if (t4_fw_install == 0) {
3438 				device_printf(sc->dev,
3439 				    "no firmware on the card and the driver "
3440 				    "is prohibited from installing new "
3441 				    "firmware.\n");
3442 				rc = EIO;
3443 				goto done;
3444 			}
3445 
3446 			device_printf(sc->dev, "no firmware on the card, "
3447 			    "installing firmware %d.%d.%d.%d\n",
3448 			    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3449 			    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3450 			rc = t4_fw_forceinstall(sc, fw->data, fw->datasize);
3451 			if (rc < 0) {
3452 				rc = -rc;
3453 				device_printf(sc->dev,
3454 				    "firmware install failed: %d.\n", rc);
3455 				goto done;
3456 			}
3457 			memcpy(card_fw, kld_fw, sizeof(*card_fw));
3458 			card_fw_usable = 1;
3459 			need_fw_reset = 0;
3460 		}
3461 	} else {
3462 		device_printf(sc->dev,
3463 		    "Unable to read card's firmware header: %d\n", rc);
3464 		card_fw_usable = 0;
3465 	}
3466 
3467 	/* Contact firmware. */
3468 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3469 	if (rc < 0 || state == DEV_STATE_ERR) {
3470 		rc = -rc;
3471 		device_printf(sc->dev,
3472 		    "failed to connect to the firmware: %d, %d.\n", rc, state);
3473 		goto done;
3474 	}
3475 	pf = rc;
3476 	if (pf == sc->mbox)
3477 		sc->flags |= MASTER_PF;
3478 	else if (state == DEV_STATE_UNINIT) {
3479 		/*
3480 		 * We didn't get to be the master so we definitely won't be
3481 		 * configuring the chip.  It's a bug if someone else hasn't
3482 		 * configured it already.
3483 		 */
3484 		device_printf(sc->dev, "couldn't be master(%d), "
3485 		    "device not already initialized either(%d).\n", rc, state);
3486 		rc = EPROTO;
3487 		goto done;
3488 	}
3489 
3490 	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3491 	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3492 		/*
3493 		 * Common case: the firmware on the card is an exact match and
3494 		 * the KLD is an exact match too, or the KLD is
3495 		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
3496 		 * here -- use cxgbetool loadfw if you want to reinstall the
3497 		 * same firmware as the one on the card.
3498 		 */
3499 	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3500 	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3501 	    be32toh(card_fw->fw_ver))) {
3502 
3503 		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3504 		if (rc != 0) {
3505 			device_printf(sc->dev,
3506 			    "failed to install firmware: %d\n", rc);
3507 			goto done;
3508 		}
3509 
3510 		/* Installed successfully, update the cached header too. */
3511 		memcpy(card_fw, kld_fw, sizeof(*card_fw));
3512 		card_fw_usable = 1;
3513 		need_fw_reset = 0;	/* already reset as part of load_fw */
3514 	}
3515 
3516 	if (!card_fw_usable) {
3517 		uint32_t d, c, k;
3518 
3519 		d = ntohl(drv_fw->fw_ver);
3520 		c = ntohl(card_fw->fw_ver);
3521 		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3522 
3523 		device_printf(sc->dev, "Cannot find a usable firmware: "
3524 		    "fw_install %d, chip state %d, "
3525 		    "driver compiled with %d.%d.%d.%d, "
3526 		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3527 		    t4_fw_install, state,
3528 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3529 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3530 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3531 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3532 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3533 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3534 		rc = EINVAL;
3535 		goto done;
3536 	}
3537 
3538 	/* Reset device */
3539 	if (need_fw_reset &&
3540 	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3541 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3542 		if (rc != ETIMEDOUT && rc != EIO)
3543 			t4_fw_bye(sc, sc->mbox);
3544 		goto done;
3545 	}
3546 	sc->flags |= FW_OK;
3547 
3548 	rc = get_params__pre_init(sc);
3549 	if (rc != 0)
3550 		goto done; /* error message displayed already */
3551 
3552 	/* Partition adapter resources as specified in the config file. */
3553 	if (state == DEV_STATE_UNINIT) {
3554 
3555 		KASSERT(sc->flags & MASTER_PF,
3556 		    ("%s: trying to change chip settings when not master.",
3557 		    __func__));
3558 
3559 		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3560 		if (rc != 0)
3561 			goto done;	/* error message displayed already */
3562 
3563 		t4_tweak_chip_settings(sc);
3564 
3565 		/* get basic stuff going */
3566 		rc = -t4_fw_initialize(sc, sc->mbox);
3567 		if (rc != 0) {
3568 			device_printf(sc->dev, "fw init failed: %d.\n", rc);
3569 			goto done;
3570 		}
3571 	} else {
3572 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3573 		sc->cfcsum = 0;
3574 	}
3575 
3576 done:
3577 	free(card_fw, M_CXGBE);
3578 	if (fw != NULL)
3579 		firmware_put(fw, FIRMWARE_UNLOAD);
3580 	if (default_cfg != NULL)
3581 		firmware_put(default_cfg, FIRMWARE_UNLOAD);
3582 
3583 	return (rc);
3584 }
3585 
3586 #define FW_PARAM_DEV(param) \
3587 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3588 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3589 #define FW_PARAM_PFVF(param) \
3590 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3591 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3592 
3593 /*
3594  * Partition chip resources for use between various PFs, VFs, etc.
3595  */
3596 static int
3597 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3598     const char *name_prefix)
3599 {
3600 	const struct firmware *cfg = NULL;
3601 	int rc = 0;
3602 	struct fw_caps_config_cmd caps;
3603 	uint32_t mtype, moff, finicsum, cfcsum;
3604 
3605 	/*
3606 	 * Figure out what configuration file to use.  Pick the default config
3607 	 * file for the card if the user hasn't specified one explicitly.
3608 	 */
3609 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3610 	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3611 		/* Card specific overrides go here. */
3612 		if (pci_get_device(sc->dev) == 0x440a)
3613 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3614 		if (is_fpga(sc))
3615 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3616 	} else if (strncmp(t4_cfg_file, BUILTIN_CF, sizeof(t4_cfg_file)) == 0)
3617 		goto use_built_in_config;	/* go straight to config. */
3618 
3619 	/*
3620 	 * We need to load another module if the profile is anything except
3621 	 * "default" or "flash".
3622 	 */
3623 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3624 	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3625 		char s[32];
3626 
3627 		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3628 		cfg = firmware_get(s);
3629 		if (cfg == NULL) {
3630 			if (default_cfg != NULL) {
3631 				device_printf(sc->dev,
3632 				    "unable to load module \"%s\" for "
3633 				    "configuration profile \"%s\", will use "
3634 				    "the default config file instead.\n",
3635 				    s, sc->cfg_file);
3636 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3637 				    "%s", DEFAULT_CF);
3638 			} else {
3639 				device_printf(sc->dev,
3640 				    "unable to load module \"%s\" for "
3641 				    "configuration profile \"%s\", will use "
3642 				    "the config file on the card's flash "
3643 				    "instead.\n", s, sc->cfg_file);
3644 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3645 				    "%s", FLASH_CF);
3646 			}
3647 		}
3648 	}
3649 
3650 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3651 	    default_cfg == NULL) {
3652 		device_printf(sc->dev,
3653 		    "default config file not available, will use the config "
3654 		    "file on the card's flash instead.\n");
3655 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3656 	}
3657 
3658 	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3659 		u_int cflen;
3660 		const uint32_t *cfdata;
3661 		uint32_t param, val, addr;
3662 
3663 		KASSERT(cfg != NULL || default_cfg != NULL,
3664 		    ("%s: no config to upload", __func__));
3665 
3666 		/*
3667 		 * Ask the firmware where it wants us to upload the config file.
3668 		 */
3669 		param = FW_PARAM_DEV(CF);
3670 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3671 		if (rc != 0) {
3672 			/* No support for config file?  Shouldn't happen. */
3673 			device_printf(sc->dev,
3674 			    "failed to query config file location: %d.\n", rc);
3675 			goto done;
3676 		}
3677 		mtype = G_FW_PARAMS_PARAM_Y(val);
3678 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3679 
3680 		/*
3681 		 * XXX: sheer laziness.  We deliberately added 4 bytes of
3682 		 * useless stuffing/comments at the end of the config file so
3683 		 * it's ok to simply throw away the last remaining bytes when
3684 		 * the config file is not an exact multiple of 4.  This also
3685 		 * helps with the validate_mt_off_len check.
3686 		 */
3687 		if (cfg != NULL) {
3688 			cflen = cfg->datasize & ~3;
3689 			cfdata = cfg->data;
3690 		} else {
3691 			cflen = default_cfg->datasize & ~3;
3692 			cfdata = default_cfg->data;
3693 		}
3694 
3695 		if (cflen > FLASH_CFG_MAX_SIZE) {
3696 			device_printf(sc->dev,
3697 			    "config file too long (%d, max allowed is %d).  "
3698 			    "Will try to use the config on the card, if any.\n",
3699 			    cflen, FLASH_CFG_MAX_SIZE);
3700 			goto use_config_on_flash;
3701 		}
3702 
3703 		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3704 		if (rc != 0) {
3705 			device_printf(sc->dev,
3706 			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3707 			    "Will try to use the config on the card, if any.\n",
3708 			    __func__, mtype, moff, cflen, rc);
3709 			goto use_config_on_flash;
3710 		}
3711 		write_via_memwin(sc, 2, addr, cfdata, cflen);
3712 	} else {
3713 use_config_on_flash:
3714 		mtype = FW_MEMTYPE_FLASH;
3715 		moff = t4_flash_cfg_addr(sc);
3716 	}
3717 
3718 	bzero(&caps, sizeof(caps));
3719 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3720 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3721 	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3722 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3723 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3724 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3725 	if (rc != 0) {
3726 		device_printf(sc->dev,
3727 		    "failed to pre-process config file: %d "
3728 		    "(mtype %d, moff 0x%x).  Will reset the firmware and retry "
3729 		    "with the built-in configuration.\n", rc, mtype, moff);
3730 
3731 	    	rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
3732 		if (rc != 0) {
3733 			device_printf(sc->dev,
3734 			    "firmware reset failed: %d.\n", rc);
3735 			if (rc != ETIMEDOUT && rc != EIO) {
3736 				t4_fw_bye(sc, sc->mbox);
3737 				sc->flags &= ~FW_OK;
3738 			}
3739 			goto done;
3740 		}
3741 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", "built-in");
3742 use_built_in_config:
3743 		bzero(&caps, sizeof(caps));
3744 		caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3745 		    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3746 		caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3747 		rc = t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3748 		if (rc != 0) {
3749 			device_printf(sc->dev,
3750 			    "built-in configuration failed: %d.\n", rc);
3751 			goto done;
3752 		}
3753 	}
3754 
3755 	finicsum = be32toh(caps.finicsum);
3756 	cfcsum = be32toh(caps.cfcsum);
3757 	if (finicsum != cfcsum) {
3758 		device_printf(sc->dev,
3759 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3760 		    finicsum, cfcsum);
3761 	}
3762 	sc->cfcsum = cfcsum;
3763 
3764 #define LIMIT_CAPS(x) do { \
3765 	caps.x &= htobe16(t4_##x##_allowed); \
3766 } while (0)
3767 
3768 	/*
3769 	 * Let the firmware know what features will (not) be used so it can tune
3770 	 * things accordingly.
3771 	 */
3772 	LIMIT_CAPS(nbmcaps);
3773 	LIMIT_CAPS(linkcaps);
3774 	LIMIT_CAPS(switchcaps);
3775 	LIMIT_CAPS(niccaps);
3776 	LIMIT_CAPS(toecaps);
3777 	LIMIT_CAPS(rdmacaps);
3778 	LIMIT_CAPS(cryptocaps);
3779 	LIMIT_CAPS(iscsicaps);
3780 	LIMIT_CAPS(fcoecaps);
3781 #undef LIMIT_CAPS
3782 
3783 	if (caps.niccaps & htobe16(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
3784 		/*
3785 		 * TOE and hashfilters are mutually exclusive.  It is a config
3786 		 * file or firmware bug if both are reported as available.  Try
3787 		 * to cope with the situation in non-debug builds by disabling
3788 		 * TOE.
3789 		 */
3790 		MPASS(caps.toecaps == 0);
3791 
3792 		caps.toecaps = 0;
3793 		caps.rdmacaps = 0;
3794 		caps.iscsicaps = 0;
3795 	}
3796 
3797 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3798 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3799 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3800 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3801 	if (rc != 0) {
3802 		device_printf(sc->dev,
3803 		    "failed to process config file: %d.\n", rc);
3804 	}
3805 done:
3806 	if (cfg != NULL)
3807 		firmware_put(cfg, FIRMWARE_UNLOAD);
3808 	return (rc);
3809 }
3810 
3811 /*
3812  * Retrieve parameters that are needed (or nice to have) very early.
3813  */
3814 static int
3815 get_params__pre_init(struct adapter *sc)
3816 {
3817 	int rc;
3818 	uint32_t param[2], val[2];
3819 
3820 	t4_get_version_info(sc);
3821 
3822 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3823 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3824 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3825 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3826 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3827 
3828 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3829 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3830 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3831 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3832 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3833 
3834 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3835 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3836 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3837 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3838 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3839 
3840 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3841 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3842 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3843 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3844 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3845 
3846 	param[0] = FW_PARAM_DEV(PORTVEC);
3847 	param[1] = FW_PARAM_DEV(CCLK);
3848 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3849 	if (rc != 0) {
3850 		device_printf(sc->dev,
3851 		    "failed to query parameters (pre_init): %d.\n", rc);
3852 		return (rc);
3853 	}
3854 
3855 	sc->params.portvec = val[0];
3856 	sc->params.nports = bitcount32(val[0]);
3857 	sc->params.vpd.cclk = val[1];
3858 
3859 	/* Read device log parameters. */
3860 	rc = -t4_init_devlog_params(sc, 1);
3861 	if (rc == 0)
3862 		fixup_devlog_params(sc);
3863 	else {
3864 		device_printf(sc->dev,
3865 		    "failed to get devlog parameters: %d.\n", rc);
3866 		rc = 0;	/* devlog isn't critical for device operation */
3867 	}
3868 
3869 	return (rc);
3870 }
3871 
3872 /*
3873  * Retrieve various parameters that are of interest to the driver.  The device
3874  * has been initialized by the firmware at this point.
3875  */
3876 static int
3877 get_params__post_init(struct adapter *sc)
3878 {
3879 	int rc;
3880 	uint32_t param[7], val[7];
3881 	struct fw_caps_config_cmd caps;
3882 
3883 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3884 	param[1] = FW_PARAM_PFVF(EQ_START);
3885 	param[2] = FW_PARAM_PFVF(FILTER_START);
3886 	param[3] = FW_PARAM_PFVF(FILTER_END);
3887 	param[4] = FW_PARAM_PFVF(L2T_START);
3888 	param[5] = FW_PARAM_PFVF(L2T_END);
3889 	param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
3890 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
3891 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
3892 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
3893 	if (rc != 0) {
3894 		device_printf(sc->dev,
3895 		    "failed to query parameters (post_init): %d.\n", rc);
3896 		return (rc);
3897 	}
3898 
3899 	sc->sge.iq_start = val[0];
3900 	sc->sge.eq_start = val[1];
3901 	if ((int)val[3] > (int)val[2]) {
3902 		sc->tids.ftid_base = val[2];
3903 		sc->tids.ftid_end = val[3];
3904 		sc->tids.nftids = val[3] - val[2] + 1;
3905 	}
3906 	sc->vres.l2t.start = val[4];
3907 	sc->vres.l2t.size = val[5] - val[4] + 1;
3908 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3909 	    ("%s: L2 table size (%u) larger than expected (%u)",
3910 	    __func__, sc->vres.l2t.size, L2T_SIZE));
3911 	sc->params.core_vdd = val[6];
3912 
3913 	if (chip_id(sc) >= CHELSIO_T6) {
3914 
3915 #ifdef INVARIANTS
3916 		if (sc->params.fw_vers >=
3917 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
3918 		    V_FW_HDR_FW_VER_MICRO(1) | V_FW_HDR_FW_VER_BUILD(0))) {
3919 			/*
3920 			 * Note that the code to enable the region should run
3921 			 * before t4_fw_initialize and not here.  This is just a
3922 			 * reminder to add said code.
3923 			 */
3924 			device_printf(sc->dev,
3925 			    "hpfilter region not enabled.\n");
3926 		}
3927 #endif
3928 
3929 		sc->tids.tid_base = t4_read_reg(sc,
3930 		    A_LE_DB_ACTIVE_TABLE_START_INDEX);
3931 
3932 		param[0] = FW_PARAM_PFVF(HPFILTER_START);
3933 		param[1] = FW_PARAM_PFVF(HPFILTER_END);
3934 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3935 		if (rc != 0) {
3936 			device_printf(sc->dev,
3937 			   "failed to query hpfilter parameters: %d.\n", rc);
3938 			return (rc);
3939 		}
3940 		if ((int)val[1] > (int)val[0]) {
3941 			sc->tids.hpftid_base = val[0];
3942 			sc->tids.hpftid_end = val[1];
3943 			sc->tids.nhpftids = val[1] - val[0] + 1;
3944 
3945 			/*
3946 			 * These should go off if the layout changes and the
3947 			 * driver needs to catch up.
3948 			 */
3949 			MPASS(sc->tids.hpftid_base == 0);
3950 			MPASS(sc->tids.tid_base == sc->tids.nhpftids);
3951 		}
3952 	}
3953 
3954 	/*
3955 	 * MPSBGMAP is queried separately because only recent firmwares support
3956 	 * it as a parameter and we don't want the compound query above to fail
3957 	 * on older firmwares.
3958 	 */
3959 	param[0] = FW_PARAM_DEV(MPSBGMAP);
3960 	val[0] = 0;
3961 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3962 	if (rc == 0)
3963 		sc->params.mps_bg_map = val[0];
3964 	else
3965 		sc->params.mps_bg_map = 0;
3966 
3967 	/*
3968 	 * Determine whether the firmware supports the filter2 work request.
3969 	 * This is queried separately for the same reason as MPSBGMAP above.
3970 	 */
3971 	param[0] = FW_PARAM_DEV(FILTER2_WR);
3972 	val[0] = 0;
3973 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3974 	if (rc == 0)
3975 		sc->params.filter2_wr_support = val[0] != 0;
3976 	else
3977 		sc->params.filter2_wr_support = 0;
3978 
3979 	/* get capabilites */
3980 	bzero(&caps, sizeof(caps));
3981 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3982 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3983 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3984 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3985 	if (rc != 0) {
3986 		device_printf(sc->dev,
3987 		    "failed to get card capabilities: %d.\n", rc);
3988 		return (rc);
3989 	}
3990 
3991 #define READ_CAPS(x) do { \
3992 	sc->x = htobe16(caps.x); \
3993 } while (0)
3994 	READ_CAPS(nbmcaps);
3995 	READ_CAPS(linkcaps);
3996 	READ_CAPS(switchcaps);
3997 	READ_CAPS(niccaps);
3998 	READ_CAPS(toecaps);
3999 	READ_CAPS(rdmacaps);
4000 	READ_CAPS(cryptocaps);
4001 	READ_CAPS(iscsicaps);
4002 	READ_CAPS(fcoecaps);
4003 
4004 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_HASHFILTER) {
4005 		MPASS(chip_id(sc) > CHELSIO_T4);
4006 		MPASS(sc->toecaps == 0);
4007 		sc->toecaps = 0;
4008 
4009 		param[0] = FW_PARAM_DEV(NTID);
4010 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4011 		if (rc != 0) {
4012 			device_printf(sc->dev,
4013 			    "failed to query HASHFILTER parameters: %d.\n", rc);
4014 			return (rc);
4015 		}
4016 		sc->tids.ntids = val[0];
4017 		if (sc->params.fw_vers <
4018 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4019 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4020 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4021 			sc->tids.ntids -= sc->tids.nhpftids;
4022 		}
4023 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4024 		sc->params.hash_filter = 1;
4025 	}
4026 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
4027 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
4028 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
4029 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4030 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
4031 		if (rc != 0) {
4032 			device_printf(sc->dev,
4033 			    "failed to query NIC parameters: %d.\n", rc);
4034 			return (rc);
4035 		}
4036 		if ((int)val[1] > (int)val[0]) {
4037 			sc->tids.etid_base = val[0];
4038 			sc->tids.etid_end = val[1];
4039 			sc->tids.netids = val[1] - val[0] + 1;
4040 			sc->params.eo_wr_cred = val[2];
4041 			sc->params.ethoffload = 1;
4042 		}
4043 	}
4044 	if (sc->toecaps) {
4045 		/* query offload-related parameters */
4046 		param[0] = FW_PARAM_DEV(NTID);
4047 		param[1] = FW_PARAM_PFVF(SERVER_START);
4048 		param[2] = FW_PARAM_PFVF(SERVER_END);
4049 		param[3] = FW_PARAM_PFVF(TDDP_START);
4050 		param[4] = FW_PARAM_PFVF(TDDP_END);
4051 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4052 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4053 		if (rc != 0) {
4054 			device_printf(sc->dev,
4055 			    "failed to query TOE parameters: %d.\n", rc);
4056 			return (rc);
4057 		}
4058 		sc->tids.ntids = val[0];
4059 		if (sc->params.fw_vers <
4060 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4061 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4062 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4063 			sc->tids.ntids -= sc->tids.nhpftids;
4064 		}
4065 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4066 		if ((int)val[2] > (int)val[1]) {
4067 			sc->tids.stid_base = val[1];
4068 			sc->tids.nstids = val[2] - val[1] + 1;
4069 		}
4070 		sc->vres.ddp.start = val[3];
4071 		sc->vres.ddp.size = val[4] - val[3] + 1;
4072 		sc->params.ofldq_wr_cred = val[5];
4073 		sc->params.offload = 1;
4074 	} else {
4075 		/*
4076 		 * The firmware attempts memfree TOE configuration for -SO cards
4077 		 * and will report toecaps=0 if it runs out of resources (this
4078 		 * depends on the config file).  It may not report 0 for other
4079 		 * capabilities dependent on the TOE in this case.  Set them to
4080 		 * 0 here so that the driver doesn't bother tracking resources
4081 		 * that will never be used.
4082 		 */
4083 		sc->iscsicaps = 0;
4084 		sc->rdmacaps = 0;
4085 	}
4086 	if (sc->rdmacaps) {
4087 		param[0] = FW_PARAM_PFVF(STAG_START);
4088 		param[1] = FW_PARAM_PFVF(STAG_END);
4089 		param[2] = FW_PARAM_PFVF(RQ_START);
4090 		param[3] = FW_PARAM_PFVF(RQ_END);
4091 		param[4] = FW_PARAM_PFVF(PBL_START);
4092 		param[5] = FW_PARAM_PFVF(PBL_END);
4093 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4094 		if (rc != 0) {
4095 			device_printf(sc->dev,
4096 			    "failed to query RDMA parameters(1): %d.\n", rc);
4097 			return (rc);
4098 		}
4099 		sc->vres.stag.start = val[0];
4100 		sc->vres.stag.size = val[1] - val[0] + 1;
4101 		sc->vres.rq.start = val[2];
4102 		sc->vres.rq.size = val[3] - val[2] + 1;
4103 		sc->vres.pbl.start = val[4];
4104 		sc->vres.pbl.size = val[5] - val[4] + 1;
4105 
4106 		param[0] = FW_PARAM_PFVF(SQRQ_START);
4107 		param[1] = FW_PARAM_PFVF(SQRQ_END);
4108 		param[2] = FW_PARAM_PFVF(CQ_START);
4109 		param[3] = FW_PARAM_PFVF(CQ_END);
4110 		param[4] = FW_PARAM_PFVF(OCQ_START);
4111 		param[5] = FW_PARAM_PFVF(OCQ_END);
4112 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4113 		if (rc != 0) {
4114 			device_printf(sc->dev,
4115 			    "failed to query RDMA parameters(2): %d.\n", rc);
4116 			return (rc);
4117 		}
4118 		sc->vres.qp.start = val[0];
4119 		sc->vres.qp.size = val[1] - val[0] + 1;
4120 		sc->vres.cq.start = val[2];
4121 		sc->vres.cq.size = val[3] - val[2] + 1;
4122 		sc->vres.ocq.start = val[4];
4123 		sc->vres.ocq.size = val[5] - val[4] + 1;
4124 
4125 		param[0] = FW_PARAM_PFVF(SRQ_START);
4126 		param[1] = FW_PARAM_PFVF(SRQ_END);
4127 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
4128 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4129 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
4130 		if (rc != 0) {
4131 			device_printf(sc->dev,
4132 			    "failed to query RDMA parameters(3): %d.\n", rc);
4133 			return (rc);
4134 		}
4135 		sc->vres.srq.start = val[0];
4136 		sc->vres.srq.size = val[1] - val[0] + 1;
4137 		sc->params.max_ordird_qp = val[2];
4138 		sc->params.max_ird_adapter = val[3];
4139 	}
4140 	if (sc->iscsicaps) {
4141 		param[0] = FW_PARAM_PFVF(ISCSI_START);
4142 		param[1] = FW_PARAM_PFVF(ISCSI_END);
4143 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4144 		if (rc != 0) {
4145 			device_printf(sc->dev,
4146 			    "failed to query iSCSI parameters: %d.\n", rc);
4147 			return (rc);
4148 		}
4149 		sc->vres.iscsi.start = val[0];
4150 		sc->vres.iscsi.size = val[1] - val[0] + 1;
4151 	}
4152 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
4153 		param[0] = FW_PARAM_PFVF(TLS_START);
4154 		param[1] = FW_PARAM_PFVF(TLS_END);
4155 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4156 		if (rc != 0) {
4157 			device_printf(sc->dev,
4158 			    "failed to query TLS parameters: %d.\n", rc);
4159 			return (rc);
4160 		}
4161 		sc->vres.key.start = val[0];
4162 		sc->vres.key.size = val[1] - val[0] + 1;
4163 	}
4164 
4165 	t4_init_sge_params(sc);
4166 
4167 	/*
4168 	 * We've got the params we wanted to query via the firmware.  Now grab
4169 	 * some others directly from the chip.
4170 	 */
4171 	rc = t4_read_chip_settings(sc);
4172 
4173 	return (rc);
4174 }
4175 
4176 static int
4177 set_params__post_init(struct adapter *sc)
4178 {
4179 	uint32_t param, val;
4180 #ifdef TCP_OFFLOAD
4181 	int i, v, shift;
4182 #endif
4183 
4184 	/* ask for encapsulated CPLs */
4185 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4186 	val = 1;
4187 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4188 
4189 	/* Enable 32b port caps if the firmware supports it. */
4190 	param = FW_PARAM_PFVF(PORT_CAPS32);
4191 	val = 1;
4192 	if (t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val) == 0)
4193 		sc->params.port_caps32 = 1;
4194 
4195 	/* Let filter + maskhash steer to a part of the VI's RSS region. */
4196 	val = 1 << (G_MASKSIZE(t4_read_reg(sc, A_TP_RSS_CONFIG_TNL)) - 1);
4197 	t4_set_reg_field(sc, A_TP_RSS_CONFIG_TNL, V_MASKFILTER(M_MASKFILTER),
4198 	    V_MASKFILTER(val - 1));
4199 
4200 #ifdef TCP_OFFLOAD
4201 	/*
4202 	 * Override the TOE timers with user provided tunables.  This is not the
4203 	 * recommended way to change the timers (the firmware config file is) so
4204 	 * these tunables are not documented.
4205 	 *
4206 	 * All the timer tunables are in microseconds.
4207 	 */
4208 	if (t4_toe_keepalive_idle != 0) {
4209 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
4210 		v &= M_KEEPALIVEIDLE;
4211 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
4212 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
4213 	}
4214 	if (t4_toe_keepalive_interval != 0) {
4215 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
4216 		v &= M_KEEPALIVEINTVL;
4217 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
4218 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
4219 	}
4220 	if (t4_toe_keepalive_count != 0) {
4221 		v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
4222 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4223 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
4224 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
4225 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
4226 	}
4227 	if (t4_toe_rexmt_min != 0) {
4228 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
4229 		v &= M_RXTMIN;
4230 		t4_set_reg_field(sc, A_TP_RXT_MIN,
4231 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
4232 	}
4233 	if (t4_toe_rexmt_max != 0) {
4234 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
4235 		v &= M_RXTMAX;
4236 		t4_set_reg_field(sc, A_TP_RXT_MAX,
4237 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
4238 	}
4239 	if (t4_toe_rexmt_count != 0) {
4240 		v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
4241 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4242 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
4243 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
4244 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
4245 	}
4246 	for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
4247 		if (t4_toe_rexmt_backoff[i] != -1) {
4248 			v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
4249 			shift = (i & 3) << 3;
4250 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
4251 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
4252 		}
4253 	}
4254 #endif
4255 	return (0);
4256 }
4257 
4258 #undef FW_PARAM_PFVF
4259 #undef FW_PARAM_DEV
4260 
4261 static void
4262 t4_set_desc(struct adapter *sc)
4263 {
4264 	char buf[128];
4265 	struct adapter_params *p = &sc->params;
4266 
4267 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
4268 
4269 	device_set_desc_copy(sc->dev, buf);
4270 }
4271 
4272 static inline void
4273 ifmedia_add4(struct ifmedia *ifm, int m)
4274 {
4275 
4276 	ifmedia_add(ifm, m, 0, NULL);
4277 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE, 0, NULL);
4278 	ifmedia_add(ifm, m | IFM_ETH_RXPAUSE, 0, NULL);
4279 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE, 0, NULL);
4280 }
4281 
4282 /*
4283  * This is the selected media, which is not quite the same as the active media.
4284  * The media line in ifconfig is "media: Ethernet selected (active)" if selected
4285  * and active are not the same, and "media: Ethernet selected" otherwise.
4286  */
4287 static void
4288 set_current_media(struct port_info *pi)
4289 {
4290 	struct link_config *lc;
4291 	struct ifmedia *ifm;
4292 	int mword;
4293 	u_int speed;
4294 
4295 	PORT_LOCK_ASSERT_OWNED(pi);
4296 
4297 	/* Leave current media alone if it's already set to IFM_NONE. */
4298 	ifm = &pi->media;
4299 	if (ifm->ifm_cur != NULL &&
4300 	    IFM_SUBTYPE(ifm->ifm_cur->ifm_media) == IFM_NONE)
4301 		return;
4302 
4303 	lc = &pi->link_cfg;
4304 	if (lc->requested_aneg != AUTONEG_DISABLE &&
4305 	    lc->supported & FW_PORT_CAP32_ANEG) {
4306 		ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
4307 		return;
4308 	}
4309 	mword = IFM_ETHER | IFM_FDX;
4310 	if (lc->requested_fc & PAUSE_TX)
4311 		mword |= IFM_ETH_TXPAUSE;
4312 	if (lc->requested_fc & PAUSE_RX)
4313 		mword |= IFM_ETH_RXPAUSE;
4314 	if (lc->requested_speed == 0)
4315 		speed = port_top_speed(pi) * 1000;	/* Gbps -> Mbps */
4316 	else
4317 		speed = lc->requested_speed;
4318 	mword |= port_mword(pi, speed_to_fwcap(speed));
4319 	ifmedia_set(ifm, mword);
4320 }
4321 
4322 /*
4323  * Returns true if the ifmedia list for the port cannot change.
4324  */
4325 static bool
4326 fixed_ifmedia(struct port_info *pi)
4327 {
4328 
4329 	return (pi->port_type == FW_PORT_TYPE_BT_SGMII ||
4330 	    pi->port_type == FW_PORT_TYPE_BT_XFI ||
4331 	    pi->port_type == FW_PORT_TYPE_BT_XAUI ||
4332 	    pi->port_type == FW_PORT_TYPE_KX4 ||
4333 	    pi->port_type == FW_PORT_TYPE_KX ||
4334 	    pi->port_type == FW_PORT_TYPE_KR ||
4335 	    pi->port_type == FW_PORT_TYPE_BP_AP ||
4336 	    pi->port_type == FW_PORT_TYPE_BP4_AP ||
4337 	    pi->port_type == FW_PORT_TYPE_BP40_BA ||
4338 	    pi->port_type == FW_PORT_TYPE_KR4_100G ||
4339 	    pi->port_type == FW_PORT_TYPE_KR_SFP28 ||
4340 	    pi->port_type == FW_PORT_TYPE_KR_XLAUI);
4341 }
4342 
4343 static void
4344 build_medialist(struct port_info *pi)
4345 {
4346 	uint32_t ss, speed;
4347 	int unknown, mword, bit;
4348 	struct link_config *lc;
4349 	struct ifmedia *ifm;
4350 
4351 	PORT_LOCK_ASSERT_OWNED(pi);
4352 
4353 	if (pi->flags & FIXED_IFMEDIA)
4354 		return;
4355 
4356 	/*
4357 	 * Rebuild the ifmedia list.
4358 	 */
4359 	ifm = &pi->media;
4360 	ifmedia_removeall(ifm);
4361 	lc = &pi->link_cfg;
4362 	ss = G_FW_PORT_CAP32_SPEED(lc->supported); /* Supported Speeds */
4363 	if (__predict_false(ss == 0)) {	/* not supposed to happen. */
4364 		MPASS(ss != 0);
4365 no_media:
4366 		MPASS(LIST_EMPTY(&ifm->ifm_list));
4367 		ifmedia_add(ifm, IFM_ETHER | IFM_NONE, 0, NULL);
4368 		ifmedia_set(ifm, IFM_ETHER | IFM_NONE);
4369 		return;
4370 	}
4371 
4372 	unknown = 0;
4373 	for (bit = S_FW_PORT_CAP32_SPEED; bit < fls(ss); bit++) {
4374 		speed = 1 << bit;
4375 		MPASS(speed & M_FW_PORT_CAP32_SPEED);
4376 		if (ss & speed) {
4377 			mword = port_mword(pi, speed);
4378 			if (mword == IFM_NONE) {
4379 				goto no_media;
4380 			} else if (mword == IFM_UNKNOWN)
4381 				unknown++;
4382 			else
4383 				ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | mword);
4384 		}
4385 	}
4386 	if (unknown > 0) /* Add one unknown for all unknown media types. */
4387 		ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | IFM_UNKNOWN);
4388 	if (lc->supported & FW_PORT_CAP32_ANEG)
4389 		ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
4390 
4391 	set_current_media(pi);
4392 }
4393 
4394 /*
4395  * Initialize the requested fields in the link config based on driver tunables.
4396  */
4397 static void
4398 init_link_config(struct port_info *pi)
4399 {
4400 	struct link_config *lc = &pi->link_cfg;
4401 
4402 	PORT_LOCK_ASSERT_OWNED(pi);
4403 
4404 	lc->requested_speed = 0;
4405 
4406 	if (t4_autoneg == 0)
4407 		lc->requested_aneg = AUTONEG_DISABLE;
4408 	else if (t4_autoneg == 1)
4409 		lc->requested_aneg = AUTONEG_ENABLE;
4410 	else
4411 		lc->requested_aneg = AUTONEG_AUTO;
4412 
4413 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX |
4414 	    PAUSE_AUTONEG);
4415 
4416 	if (t4_fec == -1 || t4_fec & FEC_AUTO)
4417 		lc->requested_fec = FEC_AUTO;
4418 	else {
4419 		lc->requested_fec = FEC_NONE;
4420 		if (t4_fec & FEC_RS)
4421 			lc->requested_fec |= FEC_RS;
4422 		if (t4_fec & FEC_BASER_RS)
4423 			lc->requested_fec |= FEC_BASER_RS;
4424 	}
4425 }
4426 
4427 /*
4428  * Makes sure that all requested settings comply with what's supported by the
4429  * port.  Returns the number of settings that were invalid and had to be fixed.
4430  */
4431 static int
4432 fixup_link_config(struct port_info *pi)
4433 {
4434 	int n = 0;
4435 	struct link_config *lc = &pi->link_cfg;
4436 	uint32_t fwspeed;
4437 
4438 	PORT_LOCK_ASSERT_OWNED(pi);
4439 
4440 	/* Speed (when not autonegotiating) */
4441 	if (lc->requested_speed != 0) {
4442 		fwspeed = speed_to_fwcap(lc->requested_speed);
4443 		if ((fwspeed & lc->supported) == 0) {
4444 			n++;
4445 			lc->requested_speed = 0;
4446 		}
4447 	}
4448 
4449 	/* Link autonegotiation */
4450 	MPASS(lc->requested_aneg == AUTONEG_ENABLE ||
4451 	    lc->requested_aneg == AUTONEG_DISABLE ||
4452 	    lc->requested_aneg == AUTONEG_AUTO);
4453 	if (lc->requested_aneg == AUTONEG_ENABLE &&
4454 	    !(lc->supported & FW_PORT_CAP32_ANEG)) {
4455 		n++;
4456 		lc->requested_aneg = AUTONEG_AUTO;
4457 	}
4458 
4459 	/* Flow control */
4460 	MPASS((lc->requested_fc & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG)) == 0);
4461 	if (lc->requested_fc & PAUSE_TX &&
4462 	    !(lc->supported & FW_PORT_CAP32_FC_TX)) {
4463 		n++;
4464 		lc->requested_fc &= ~PAUSE_TX;
4465 	}
4466 	if (lc->requested_fc & PAUSE_RX &&
4467 	    !(lc->supported & FW_PORT_CAP32_FC_RX)) {
4468 		n++;
4469 		lc->requested_fc &= ~PAUSE_RX;
4470 	}
4471 	if (!(lc->requested_fc & PAUSE_AUTONEG) &&
4472 	    !(lc->supported & FW_PORT_CAP32_FORCE_PAUSE)) {
4473 		n++;
4474 		lc->requested_fc |= PAUSE_AUTONEG;
4475 	}
4476 
4477 	/* FEC */
4478 	if ((lc->requested_fec & FEC_RS &&
4479 	    !(lc->supported & FW_PORT_CAP32_FEC_RS)) ||
4480 	    (lc->requested_fec & FEC_BASER_RS &&
4481 	    !(lc->supported & FW_PORT_CAP32_FEC_BASER_RS))) {
4482 		n++;
4483 		lc->requested_fec = FEC_AUTO;
4484 	}
4485 
4486 	return (n);
4487 }
4488 
4489 /*
4490  * Apply the requested L1 settings, which are expected to be valid, to the
4491  * hardware.
4492  */
4493 static int
4494 apply_link_config(struct port_info *pi)
4495 {
4496 	struct adapter *sc = pi->adapter;
4497 	struct link_config *lc = &pi->link_cfg;
4498 	int rc;
4499 
4500 #ifdef INVARIANTS
4501 	ASSERT_SYNCHRONIZED_OP(sc);
4502 	PORT_LOCK_ASSERT_OWNED(pi);
4503 
4504 	if (lc->requested_aneg == AUTONEG_ENABLE)
4505 		MPASS(lc->supported & FW_PORT_CAP32_ANEG);
4506 	if (!(lc->requested_fc & PAUSE_AUTONEG))
4507 		MPASS(lc->supported & FW_PORT_CAP32_FORCE_PAUSE);
4508 	if (lc->requested_fc & PAUSE_TX)
4509 		MPASS(lc->supported & FW_PORT_CAP32_FC_TX);
4510 	if (lc->requested_fc & PAUSE_RX)
4511 		MPASS(lc->supported & FW_PORT_CAP32_FC_RX);
4512 	if (lc->requested_fec & FEC_RS)
4513 		MPASS(lc->supported & FW_PORT_CAP32_FEC_RS);
4514 	if (lc->requested_fec & FEC_BASER_RS)
4515 		MPASS(lc->supported & FW_PORT_CAP32_FEC_BASER_RS);
4516 #endif
4517 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
4518 	if (rc != 0) {
4519 		/* Don't complain if the VF driver gets back an EPERM. */
4520 		if (!(sc->flags & IS_VF) || rc != FW_EPERM)
4521 			device_printf(pi->dev, "l1cfg failed: %d\n", rc);
4522 	} else {
4523 		/*
4524 		 * An L1_CFG will almost always result in a link-change event if
4525 		 * the link is up, and the driver will refresh the actual
4526 		 * fec/fc/etc. when the notification is processed.  If the link
4527 		 * is down then the actual settings are meaningless.
4528 		 *
4529 		 * This takes care of the case where a change in the L1 settings
4530 		 * may not result in a notification.
4531 		 */
4532 		if (lc->link_ok && !(lc->requested_fc & PAUSE_AUTONEG))
4533 			lc->fc = lc->requested_fc & (PAUSE_TX | PAUSE_RX);
4534 	}
4535 	return (rc);
4536 }
4537 
4538 #define FW_MAC_EXACT_CHUNK	7
4539 
4540 /*
4541  * Program the port's XGMAC based on parameters in ifnet.  The caller also
4542  * indicates which parameters should be programmed (the rest are left alone).
4543  */
4544 int
4545 update_mac_settings(struct ifnet *ifp, int flags)
4546 {
4547 	int rc = 0;
4548 	struct vi_info *vi = ifp->if_softc;
4549 	struct port_info *pi = vi->pi;
4550 	struct adapter *sc = pi->adapter;
4551 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4552 
4553 	ASSERT_SYNCHRONIZED_OP(sc);
4554 	KASSERT(flags, ("%s: not told what to update.", __func__));
4555 
4556 	if (flags & XGMAC_MTU)
4557 		mtu = ifp->if_mtu;
4558 
4559 	if (flags & XGMAC_PROMISC)
4560 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4561 
4562 	if (flags & XGMAC_ALLMULTI)
4563 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4564 
4565 	if (flags & XGMAC_VLANEX)
4566 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4567 
4568 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4569 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4570 		    allmulti, 1, vlanex, false);
4571 		if (rc) {
4572 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4573 			    rc);
4574 			return (rc);
4575 		}
4576 	}
4577 
4578 	if (flags & XGMAC_UCADDR) {
4579 		uint8_t ucaddr[ETHER_ADDR_LEN];
4580 
4581 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4582 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4583 		    ucaddr, true, true);
4584 		if (rc < 0) {
4585 			rc = -rc;
4586 			if_printf(ifp, "change_mac failed: %d\n", rc);
4587 			return (rc);
4588 		} else {
4589 			vi->xact_addr_filt = rc;
4590 			rc = 0;
4591 		}
4592 	}
4593 
4594 	if (flags & XGMAC_MCADDRS) {
4595 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4596 		int del = 1;
4597 		uint64_t hash = 0;
4598 		struct ifmultiaddr *ifma;
4599 		int i = 0, j;
4600 
4601 		if_maddr_rlock(ifp);
4602 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4603 			if (ifma->ifma_addr->sa_family != AF_LINK)
4604 				continue;
4605 			mcaddr[i] =
4606 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4607 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4608 			i++;
4609 
4610 			if (i == FW_MAC_EXACT_CHUNK) {
4611 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4612 				    del, i, mcaddr, NULL, &hash, 0);
4613 				if (rc < 0) {
4614 					rc = -rc;
4615 					for (j = 0; j < i; j++) {
4616 						if_printf(ifp,
4617 						    "failed to add mc address"
4618 						    " %02x:%02x:%02x:"
4619 						    "%02x:%02x:%02x rc=%d\n",
4620 						    mcaddr[j][0], mcaddr[j][1],
4621 						    mcaddr[j][2], mcaddr[j][3],
4622 						    mcaddr[j][4], mcaddr[j][5],
4623 						    rc);
4624 					}
4625 					goto mcfail;
4626 				}
4627 				del = 0;
4628 				i = 0;
4629 			}
4630 		}
4631 		if (i > 0) {
4632 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4633 			    mcaddr, NULL, &hash, 0);
4634 			if (rc < 0) {
4635 				rc = -rc;
4636 				for (j = 0; j < i; j++) {
4637 					if_printf(ifp,
4638 					    "failed to add mc address"
4639 					    " %02x:%02x:%02x:"
4640 					    "%02x:%02x:%02x rc=%d\n",
4641 					    mcaddr[j][0], mcaddr[j][1],
4642 					    mcaddr[j][2], mcaddr[j][3],
4643 					    mcaddr[j][4], mcaddr[j][5],
4644 					    rc);
4645 				}
4646 				goto mcfail;
4647 			}
4648 		}
4649 
4650 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4651 		if (rc != 0)
4652 			if_printf(ifp, "failed to set mc address hash: %d", rc);
4653 mcfail:
4654 		if_maddr_runlock(ifp);
4655 	}
4656 
4657 	return (rc);
4658 }
4659 
4660 /*
4661  * {begin|end}_synchronized_op must be called from the same thread.
4662  */
4663 int
4664 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4665     char *wmesg)
4666 {
4667 	int rc, pri;
4668 
4669 #ifdef WITNESS
4670 	/* the caller thinks it's ok to sleep, but is it really? */
4671 	if (flags & SLEEP_OK)
4672 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4673 		    "begin_synchronized_op");
4674 #endif
4675 
4676 	if (INTR_OK)
4677 		pri = PCATCH;
4678 	else
4679 		pri = 0;
4680 
4681 	ADAPTER_LOCK(sc);
4682 	for (;;) {
4683 
4684 		if (vi && IS_DOOMED(vi)) {
4685 			rc = ENXIO;
4686 			goto done;
4687 		}
4688 
4689 		if (!IS_BUSY(sc)) {
4690 			rc = 0;
4691 			break;
4692 		}
4693 
4694 		if (!(flags & SLEEP_OK)) {
4695 			rc = EBUSY;
4696 			goto done;
4697 		}
4698 
4699 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4700 			rc = EINTR;
4701 			goto done;
4702 		}
4703 	}
4704 
4705 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4706 	SET_BUSY(sc);
4707 #ifdef INVARIANTS
4708 	sc->last_op = wmesg;
4709 	sc->last_op_thr = curthread;
4710 	sc->last_op_flags = flags;
4711 #endif
4712 
4713 done:
4714 	if (!(flags & HOLD_LOCK) || rc)
4715 		ADAPTER_UNLOCK(sc);
4716 
4717 	return (rc);
4718 }
4719 
4720 /*
4721  * Tell if_ioctl and if_init that the VI is going away.  This is
4722  * special variant of begin_synchronized_op and must be paired with a
4723  * call to end_synchronized_op.
4724  */
4725 void
4726 doom_vi(struct adapter *sc, struct vi_info *vi)
4727 {
4728 
4729 	ADAPTER_LOCK(sc);
4730 	SET_DOOMED(vi);
4731 	wakeup(&sc->flags);
4732 	while (IS_BUSY(sc))
4733 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4734 	SET_BUSY(sc);
4735 #ifdef INVARIANTS
4736 	sc->last_op = "t4detach";
4737 	sc->last_op_thr = curthread;
4738 	sc->last_op_flags = 0;
4739 #endif
4740 	ADAPTER_UNLOCK(sc);
4741 }
4742 
4743 /*
4744  * {begin|end}_synchronized_op must be called from the same thread.
4745  */
4746 void
4747 end_synchronized_op(struct adapter *sc, int flags)
4748 {
4749 
4750 	if (flags & LOCK_HELD)
4751 		ADAPTER_LOCK_ASSERT_OWNED(sc);
4752 	else
4753 		ADAPTER_LOCK(sc);
4754 
4755 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
4756 	CLR_BUSY(sc);
4757 	wakeup(&sc->flags);
4758 	ADAPTER_UNLOCK(sc);
4759 }
4760 
4761 static int
4762 cxgbe_init_synchronized(struct vi_info *vi)
4763 {
4764 	struct port_info *pi = vi->pi;
4765 	struct adapter *sc = pi->adapter;
4766 	struct ifnet *ifp = vi->ifp;
4767 	int rc = 0, i;
4768 	struct sge_txq *txq;
4769 
4770 	ASSERT_SYNCHRONIZED_OP(sc);
4771 
4772 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
4773 		return (0);	/* already running */
4774 
4775 	if (!(sc->flags & FULL_INIT_DONE) &&
4776 	    ((rc = adapter_full_init(sc)) != 0))
4777 		return (rc);	/* error message displayed already */
4778 
4779 	if (!(vi->flags & VI_INIT_DONE) &&
4780 	    ((rc = vi_full_init(vi)) != 0))
4781 		return (rc); /* error message displayed already */
4782 
4783 	rc = update_mac_settings(ifp, XGMAC_ALL);
4784 	if (rc)
4785 		goto done;	/* error message displayed already */
4786 
4787 	PORT_LOCK(pi);
4788 	if (pi->up_vis == 0) {
4789 		t4_update_port_info(pi);
4790 		fixup_link_config(pi);
4791 		build_medialist(pi);
4792 		apply_link_config(pi);
4793 	}
4794 
4795 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
4796 	if (rc != 0) {
4797 		if_printf(ifp, "enable_vi failed: %d\n", rc);
4798 		PORT_UNLOCK(pi);
4799 		goto done;
4800 	}
4801 
4802 	/*
4803 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
4804 	 * if this changes.
4805 	 */
4806 
4807 	for_each_txq(vi, i, txq) {
4808 		TXQ_LOCK(txq);
4809 		txq->eq.flags |= EQ_ENABLED;
4810 		TXQ_UNLOCK(txq);
4811 	}
4812 
4813 	/*
4814 	 * The first iq of the first port to come up is used for tracing.
4815 	 */
4816 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4817 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4818 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4819 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4820 		    V_QUEUENUMBER(sc->traceq));
4821 		pi->flags |= HAS_TRACEQ;
4822 	}
4823 
4824 	/* all ok */
4825 	pi->up_vis++;
4826 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
4827 
4828 	if (pi->nvi > 1 || sc->flags & IS_VF)
4829 		callout_reset(&vi->tick, hz, vi_tick, vi);
4830 	else
4831 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4832 	PORT_UNLOCK(pi);
4833 done:
4834 	if (rc != 0)
4835 		cxgbe_uninit_synchronized(vi);
4836 
4837 	return (rc);
4838 }
4839 
4840 /*
4841  * Idempotent.
4842  */
4843 static int
4844 cxgbe_uninit_synchronized(struct vi_info *vi)
4845 {
4846 	struct port_info *pi = vi->pi;
4847 	struct adapter *sc = pi->adapter;
4848 	struct ifnet *ifp = vi->ifp;
4849 	int rc, i;
4850 	struct sge_txq *txq;
4851 
4852 	ASSERT_SYNCHRONIZED_OP(sc);
4853 
4854 	if (!(vi->flags & VI_INIT_DONE)) {
4855 		if (__predict_false(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4856 			KASSERT(0, ("uninited VI is running"));
4857 			if_printf(ifp, "uninited VI with running ifnet.  "
4858 			    "vi->flags 0x%016lx, if_flags 0x%08x, "
4859 			    "if_drv_flags 0x%08x\n", vi->flags, ifp->if_flags,
4860 			    ifp->if_drv_flags);
4861 		}
4862 		return (0);
4863 	}
4864 
4865 	/*
4866 	 * Disable the VI so that all its data in either direction is discarded
4867 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4868 	 * tick) intact as the TP can deliver negative advice or data that it's
4869 	 * holding in its RAM (for an offloaded connection) even after the VI is
4870 	 * disabled.
4871 	 */
4872 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4873 	if (rc) {
4874 		if_printf(ifp, "disable_vi failed: %d\n", rc);
4875 		return (rc);
4876 	}
4877 
4878 	for_each_txq(vi, i, txq) {
4879 		TXQ_LOCK(txq);
4880 		txq->eq.flags &= ~EQ_ENABLED;
4881 		TXQ_UNLOCK(txq);
4882 	}
4883 
4884 	PORT_LOCK(pi);
4885 	if (pi->nvi > 1 || sc->flags & IS_VF)
4886 		callout_stop(&vi->tick);
4887 	else
4888 		callout_stop(&pi->tick);
4889 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4890 		PORT_UNLOCK(pi);
4891 		return (0);
4892 	}
4893 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4894 	pi->up_vis--;
4895 	if (pi->up_vis > 0) {
4896 		PORT_UNLOCK(pi);
4897 		return (0);
4898 	}
4899 
4900 	pi->link_cfg.link_ok = false;
4901 	pi->link_cfg.speed = 0;
4902 	pi->link_cfg.link_down_rc = 255;
4903 	t4_os_link_changed(pi);
4904 	PORT_UNLOCK(pi);
4905 
4906 	return (0);
4907 }
4908 
4909 /*
4910  * It is ok for this function to fail midway and return right away.  t4_detach
4911  * will walk the entire sc->irq list and clean up whatever is valid.
4912  */
4913 int
4914 t4_setup_intr_handlers(struct adapter *sc)
4915 {
4916 	int rc, rid, p, q, v;
4917 	char s[8];
4918 	struct irq *irq;
4919 	struct port_info *pi;
4920 	struct vi_info *vi;
4921 	struct sge *sge = &sc->sge;
4922 	struct sge_rxq *rxq;
4923 #ifdef TCP_OFFLOAD
4924 	struct sge_ofld_rxq *ofld_rxq;
4925 #endif
4926 #ifdef DEV_NETMAP
4927 	struct sge_nm_rxq *nm_rxq;
4928 #endif
4929 #ifdef RSS
4930 	int nbuckets = rss_getnumbuckets();
4931 #endif
4932 
4933 	/*
4934 	 * Setup interrupts.
4935 	 */
4936 	irq = &sc->irq[0];
4937 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
4938 	if (forwarding_intr_to_fwq(sc))
4939 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4940 
4941 	/* Multiple interrupts. */
4942 	if (sc->flags & IS_VF)
4943 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4944 		    ("%s: too few intr.", __func__));
4945 	else
4946 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4947 		    ("%s: too few intr.", __func__));
4948 
4949 	/* The first one is always error intr on PFs */
4950 	if (!(sc->flags & IS_VF)) {
4951 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4952 		if (rc != 0)
4953 			return (rc);
4954 		irq++;
4955 		rid++;
4956 	}
4957 
4958 	/* The second one is always the firmware event queue (first on VFs) */
4959 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4960 	if (rc != 0)
4961 		return (rc);
4962 	irq++;
4963 	rid++;
4964 
4965 	for_each_port(sc, p) {
4966 		pi = sc->port[p];
4967 		for_each_vi(pi, v, vi) {
4968 			vi->first_intr = rid - 1;
4969 
4970 			if (vi->nnmrxq > 0) {
4971 				int n = max(vi->nrxq, vi->nnmrxq);
4972 
4973 				rxq = &sge->rxq[vi->first_rxq];
4974 #ifdef DEV_NETMAP
4975 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4976 #endif
4977 				for (q = 0; q < n; q++) {
4978 					snprintf(s, sizeof(s), "%x%c%x", p,
4979 					    'a' + v, q);
4980 					if (q < vi->nrxq)
4981 						irq->rxq = rxq++;
4982 #ifdef DEV_NETMAP
4983 					if (q < vi->nnmrxq)
4984 						irq->nm_rxq = nm_rxq++;
4985 
4986 					if (irq->nm_rxq != NULL &&
4987 					    irq->rxq == NULL) {
4988 						/* Netmap rx only */
4989 						rc = t4_alloc_irq(sc, irq, rid,
4990 						    t4_nm_intr, irq->nm_rxq, s);
4991 					}
4992 					if (irq->nm_rxq != NULL &&
4993 					    irq->rxq != NULL) {
4994 						/* NIC and Netmap rx */
4995 						rc = t4_alloc_irq(sc, irq, rid,
4996 						    t4_vi_intr, irq, s);
4997 					}
4998 #endif
4999 					if (irq->rxq != NULL &&
5000 					    irq->nm_rxq == NULL) {
5001 						/* NIC rx only */
5002 						rc = t4_alloc_irq(sc, irq, rid,
5003 						    t4_intr, irq->rxq, s);
5004 					}
5005 					if (rc != 0)
5006 						return (rc);
5007 #ifdef RSS
5008 					if (q < vi->nrxq) {
5009 						bus_bind_intr(sc->dev, irq->res,
5010 						    rss_getcpu(q % nbuckets));
5011 					}
5012 #endif
5013 					irq++;
5014 					rid++;
5015 					vi->nintr++;
5016 				}
5017 			} else {
5018 				for_each_rxq(vi, q, rxq) {
5019 					snprintf(s, sizeof(s), "%x%c%x", p,
5020 					    'a' + v, q);
5021 					rc = t4_alloc_irq(sc, irq, rid,
5022 					    t4_intr, rxq, s);
5023 					if (rc != 0)
5024 						return (rc);
5025 #ifdef RSS
5026 					bus_bind_intr(sc->dev, irq->res,
5027 					    rss_getcpu(q % nbuckets));
5028 #endif
5029 					irq++;
5030 					rid++;
5031 					vi->nintr++;
5032 				}
5033 			}
5034 #ifdef TCP_OFFLOAD
5035 			for_each_ofld_rxq(vi, q, ofld_rxq) {
5036 				snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
5037 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
5038 				    ofld_rxq, s);
5039 				if (rc != 0)
5040 					return (rc);
5041 				irq++;
5042 				rid++;
5043 				vi->nintr++;
5044 			}
5045 #endif
5046 		}
5047 	}
5048 	MPASS(irq == &sc->irq[sc->intr_count]);
5049 
5050 	return (0);
5051 }
5052 
5053 int
5054 adapter_full_init(struct adapter *sc)
5055 {
5056 	int rc, i;
5057 #ifdef RSS
5058 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5059 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5060 #endif
5061 
5062 	ASSERT_SYNCHRONIZED_OP(sc);
5063 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5064 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
5065 	    ("%s: FULL_INIT_DONE already", __func__));
5066 
5067 	/*
5068 	 * queues that belong to the adapter (not any particular port).
5069 	 */
5070 	rc = t4_setup_adapter_queues(sc);
5071 	if (rc != 0)
5072 		goto done;
5073 
5074 	for (i = 0; i < nitems(sc->tq); i++) {
5075 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
5076 		    taskqueue_thread_enqueue, &sc->tq[i]);
5077 		if (sc->tq[i] == NULL) {
5078 			device_printf(sc->dev,
5079 			    "failed to allocate task queue %d\n", i);
5080 			rc = ENOMEM;
5081 			goto done;
5082 		}
5083 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
5084 		    device_get_nameunit(sc->dev), i);
5085 	}
5086 #ifdef RSS
5087 	MPASS(RSS_KEYSIZE == 40);
5088 	rss_getkey((void *)&raw_rss_key[0]);
5089 	for (i = 0; i < nitems(rss_key); i++) {
5090 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
5091 	}
5092 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
5093 #endif
5094 
5095 	if (!(sc->flags & IS_VF))
5096 		t4_intr_enable(sc);
5097 	sc->flags |= FULL_INIT_DONE;
5098 done:
5099 	if (rc != 0)
5100 		adapter_full_uninit(sc);
5101 
5102 	return (rc);
5103 }
5104 
5105 int
5106 adapter_full_uninit(struct adapter *sc)
5107 {
5108 	int i;
5109 
5110 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5111 
5112 	t4_teardown_adapter_queues(sc);
5113 
5114 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
5115 		taskqueue_free(sc->tq[i]);
5116 		sc->tq[i] = NULL;
5117 	}
5118 
5119 	sc->flags &= ~FULL_INIT_DONE;
5120 
5121 	return (0);
5122 }
5123 
5124 #ifdef RSS
5125 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
5126     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
5127     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
5128     RSS_HASHTYPE_RSS_UDP_IPV6)
5129 
5130 /* Translates kernel hash types to hardware. */
5131 static int
5132 hashconfig_to_hashen(int hashconfig)
5133 {
5134 	int hashen = 0;
5135 
5136 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
5137 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
5138 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
5139 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
5140 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
5141 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5142 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5143 	}
5144 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
5145 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5146 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5147 	}
5148 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
5149 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5150 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
5151 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5152 
5153 	return (hashen);
5154 }
5155 
5156 /* Translates hardware hash types to kernel. */
5157 static int
5158 hashen_to_hashconfig(int hashen)
5159 {
5160 	int hashconfig = 0;
5161 
5162 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
5163 		/*
5164 		 * If UDP hashing was enabled it must have been enabled for
5165 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
5166 		 * enabling any 4-tuple hash is nonsense configuration.
5167 		 */
5168 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5169 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
5170 
5171 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5172 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
5173 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5174 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
5175 	}
5176 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5177 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
5178 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5179 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
5180 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
5181 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
5182 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
5183 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
5184 
5185 	return (hashconfig);
5186 }
5187 #endif
5188 
5189 int
5190 vi_full_init(struct vi_info *vi)
5191 {
5192 	struct adapter *sc = vi->pi->adapter;
5193 	struct ifnet *ifp = vi->ifp;
5194 	uint16_t *rss;
5195 	struct sge_rxq *rxq;
5196 	int rc, i, j;
5197 #ifdef RSS
5198 	int nbuckets = rss_getnumbuckets();
5199 	int hashconfig = rss_gethashconfig();
5200 	int extra;
5201 #endif
5202 
5203 	ASSERT_SYNCHRONIZED_OP(sc);
5204 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
5205 	    ("%s: VI_INIT_DONE already", __func__));
5206 
5207 	sysctl_ctx_init(&vi->ctx);
5208 	vi->flags |= VI_SYSCTL_CTX;
5209 
5210 	/*
5211 	 * Allocate tx/rx/fl queues for this VI.
5212 	 */
5213 	rc = t4_setup_vi_queues(vi);
5214 	if (rc != 0)
5215 		goto done;	/* error message displayed already */
5216 
5217 	/*
5218 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
5219 	 */
5220 	if (vi->nrxq > vi->rss_size) {
5221 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
5222 		    "some queues will never receive traffic.\n", vi->nrxq,
5223 		    vi->rss_size);
5224 	} else if (vi->rss_size % vi->nrxq) {
5225 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
5226 		    "expect uneven traffic distribution.\n", vi->nrxq,
5227 		    vi->rss_size);
5228 	}
5229 #ifdef RSS
5230 	if (vi->nrxq != nbuckets) {
5231 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
5232 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
5233 	}
5234 #endif
5235 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
5236 	for (i = 0; i < vi->rss_size;) {
5237 #ifdef RSS
5238 		j = rss_get_indirection_to_bucket(i);
5239 		j %= vi->nrxq;
5240 		rxq = &sc->sge.rxq[vi->first_rxq + j];
5241 		rss[i++] = rxq->iq.abs_id;
5242 #else
5243 		for_each_rxq(vi, j, rxq) {
5244 			rss[i++] = rxq->iq.abs_id;
5245 			if (i == vi->rss_size)
5246 				break;
5247 		}
5248 #endif
5249 	}
5250 
5251 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
5252 	    vi->rss_size);
5253 	if (rc != 0) {
5254 		free(rss, M_CXGBE);
5255 		if_printf(ifp, "rss_config failed: %d\n", rc);
5256 		goto done;
5257 	}
5258 
5259 #ifdef RSS
5260 	vi->hashen = hashconfig_to_hashen(hashconfig);
5261 
5262 	/*
5263 	 * We may have had to enable some hashes even though the global config
5264 	 * wants them disabled.  This is a potential problem that must be
5265 	 * reported to the user.
5266 	 */
5267 	extra = hashen_to_hashconfig(vi->hashen) ^ hashconfig;
5268 
5269 	/*
5270 	 * If we consider only the supported hash types, then the enabled hashes
5271 	 * are a superset of the requested hashes.  In other words, there cannot
5272 	 * be any supported hash that was requested but not enabled, but there
5273 	 * can be hashes that were not requested but had to be enabled.
5274 	 */
5275 	extra &= SUPPORTED_RSS_HASHTYPES;
5276 	MPASS((extra & hashconfig) == 0);
5277 
5278 	if (extra) {
5279 		if_printf(ifp,
5280 		    "global RSS config (0x%x) cannot be accommodated.\n",
5281 		    hashconfig);
5282 	}
5283 	if (extra & RSS_HASHTYPE_RSS_IPV4)
5284 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
5285 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
5286 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
5287 	if (extra & RSS_HASHTYPE_RSS_IPV6)
5288 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
5289 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
5290 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
5291 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
5292 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
5293 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
5294 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
5295 #else
5296 	vi->hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
5297 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
5298 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5299 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
5300 #endif
5301 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, vi->hashen, rss[0], 0, 0);
5302 	if (rc != 0) {
5303 		free(rss, M_CXGBE);
5304 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
5305 		goto done;
5306 	}
5307 
5308 	vi->rss = rss;
5309 	vi->flags |= VI_INIT_DONE;
5310 done:
5311 	if (rc != 0)
5312 		vi_full_uninit(vi);
5313 
5314 	return (rc);
5315 }
5316 
5317 /*
5318  * Idempotent.
5319  */
5320 int
5321 vi_full_uninit(struct vi_info *vi)
5322 {
5323 	struct port_info *pi = vi->pi;
5324 	struct adapter *sc = pi->adapter;
5325 	int i;
5326 	struct sge_rxq *rxq;
5327 	struct sge_txq *txq;
5328 #ifdef TCP_OFFLOAD
5329 	struct sge_ofld_rxq *ofld_rxq;
5330 	struct sge_wrq *ofld_txq;
5331 #endif
5332 
5333 	if (vi->flags & VI_INIT_DONE) {
5334 
5335 		/* Need to quiesce queues.  */
5336 
5337 		/* XXX: Only for the first VI? */
5338 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
5339 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
5340 
5341 		for_each_txq(vi, i, txq) {
5342 			quiesce_txq(sc, txq);
5343 		}
5344 
5345 #ifdef TCP_OFFLOAD
5346 		for_each_ofld_txq(vi, i, ofld_txq) {
5347 			quiesce_wrq(sc, ofld_txq);
5348 		}
5349 #endif
5350 
5351 		for_each_rxq(vi, i, rxq) {
5352 			quiesce_iq(sc, &rxq->iq);
5353 			quiesce_fl(sc, &rxq->fl);
5354 		}
5355 
5356 #ifdef TCP_OFFLOAD
5357 		for_each_ofld_rxq(vi, i, ofld_rxq) {
5358 			quiesce_iq(sc, &ofld_rxq->iq);
5359 			quiesce_fl(sc, &ofld_rxq->fl);
5360 		}
5361 #endif
5362 		free(vi->rss, M_CXGBE);
5363 		free(vi->nm_rss, M_CXGBE);
5364 	}
5365 
5366 	t4_teardown_vi_queues(vi);
5367 	vi->flags &= ~VI_INIT_DONE;
5368 
5369 	return (0);
5370 }
5371 
5372 static void
5373 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
5374 {
5375 	struct sge_eq *eq = &txq->eq;
5376 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
5377 
5378 	(void) sc;	/* unused */
5379 
5380 #ifdef INVARIANTS
5381 	TXQ_LOCK(txq);
5382 	MPASS((eq->flags & EQ_ENABLED) == 0);
5383 	TXQ_UNLOCK(txq);
5384 #endif
5385 
5386 	/* Wait for the mp_ring to empty. */
5387 	while (!mp_ring_is_idle(txq->r)) {
5388 		mp_ring_check_drainage(txq->r, 0);
5389 		pause("rquiesce", 1);
5390 	}
5391 
5392 	/* Then wait for the hardware to finish. */
5393 	while (spg->cidx != htobe16(eq->pidx))
5394 		pause("equiesce", 1);
5395 
5396 	/* Finally, wait for the driver to reclaim all descriptors. */
5397 	while (eq->cidx != eq->pidx)
5398 		pause("dquiesce", 1);
5399 }
5400 
5401 static void
5402 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
5403 {
5404 
5405 	/* XXXTX */
5406 }
5407 
5408 static void
5409 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
5410 {
5411 	(void) sc;	/* unused */
5412 
5413 	/* Synchronize with the interrupt handler */
5414 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
5415 		pause("iqfree", 1);
5416 }
5417 
5418 static void
5419 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
5420 {
5421 	mtx_lock(&sc->sfl_lock);
5422 	FL_LOCK(fl);
5423 	fl->flags |= FL_DOOMED;
5424 	FL_UNLOCK(fl);
5425 	callout_stop(&sc->sfl_callout);
5426 	mtx_unlock(&sc->sfl_lock);
5427 
5428 	KASSERT((fl->flags & FL_STARVING) == 0,
5429 	    ("%s: still starving", __func__));
5430 }
5431 
5432 static int
5433 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
5434     driver_intr_t *handler, void *arg, char *name)
5435 {
5436 	int rc;
5437 
5438 	irq->rid = rid;
5439 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
5440 	    RF_SHAREABLE | RF_ACTIVE);
5441 	if (irq->res == NULL) {
5442 		device_printf(sc->dev,
5443 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
5444 		return (ENOMEM);
5445 	}
5446 
5447 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
5448 	    NULL, handler, arg, &irq->tag);
5449 	if (rc != 0) {
5450 		device_printf(sc->dev,
5451 		    "failed to setup interrupt for rid %d, name %s: %d\n",
5452 		    rid, name, rc);
5453 	} else if (name)
5454 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
5455 
5456 	return (rc);
5457 }
5458 
5459 static int
5460 t4_free_irq(struct adapter *sc, struct irq *irq)
5461 {
5462 	if (irq->tag)
5463 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
5464 	if (irq->res)
5465 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
5466 
5467 	bzero(irq, sizeof(*irq));
5468 
5469 	return (0);
5470 }
5471 
5472 static void
5473 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
5474 {
5475 
5476 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
5477 	t4_get_regs(sc, buf, regs->len);
5478 }
5479 
5480 #define	A_PL_INDIR_CMD	0x1f8
5481 
5482 #define	S_PL_AUTOINC	31
5483 #define	M_PL_AUTOINC	0x1U
5484 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
5485 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
5486 
5487 #define	S_PL_VFID	20
5488 #define	M_PL_VFID	0xffU
5489 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
5490 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
5491 
5492 #define	S_PL_ADDR	0
5493 #define	M_PL_ADDR	0xfffffU
5494 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
5495 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
5496 
5497 #define	A_PL_INDIR_DATA	0x1fc
5498 
5499 static uint64_t
5500 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
5501 {
5502 	u32 stats[2];
5503 
5504 	mtx_assert(&sc->reg_lock, MA_OWNED);
5505 	if (sc->flags & IS_VF) {
5506 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
5507 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
5508 	} else {
5509 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5510 		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5511 		    V_PL_ADDR(VF_MPS_REG(reg)));
5512 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
5513 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
5514 	}
5515 	return (((uint64_t)stats[1]) << 32 | stats[0]);
5516 }
5517 
5518 static void
5519 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
5520     struct fw_vi_stats_vf *stats)
5521 {
5522 
5523 #define GET_STAT(name) \
5524 	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
5525 
5526 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
5527 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
5528 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
5529 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
5530 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
5531 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
5532 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
5533 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
5534 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
5535 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
5536 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
5537 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
5538 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
5539 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
5540 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
5541 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
5542 
5543 #undef GET_STAT
5544 }
5545 
5546 static void
5547 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
5548 {
5549 	int reg;
5550 
5551 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5552 	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5553 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
5554 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
5555 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
5556 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
5557 }
5558 
5559 static void
5560 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
5561 {
5562 	struct timeval tv;
5563 	const struct timeval interval = {0, 250000};	/* 250ms */
5564 
5565 	if (!(vi->flags & VI_INIT_DONE))
5566 		return;
5567 
5568 	getmicrotime(&tv);
5569 	timevalsub(&tv, &interval);
5570 	if (timevalcmp(&tv, &vi->last_refreshed, <))
5571 		return;
5572 
5573 	mtx_lock(&sc->reg_lock);
5574 	t4_get_vi_stats(sc, vi->viid, &vi->stats);
5575 	getmicrotime(&vi->last_refreshed);
5576 	mtx_unlock(&sc->reg_lock);
5577 }
5578 
5579 static void
5580 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5581 {
5582 	u_int i, v, tnl_cong_drops, bg_map;
5583 	struct timeval tv;
5584 	const struct timeval interval = {0, 250000};	/* 250ms */
5585 
5586 	getmicrotime(&tv);
5587 	timevalsub(&tv, &interval);
5588 	if (timevalcmp(&tv, &pi->last_refreshed, <))
5589 		return;
5590 
5591 	tnl_cong_drops = 0;
5592 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5593 	bg_map = pi->mps_bg_map;
5594 	while (bg_map) {
5595 		i = ffs(bg_map) - 1;
5596 		mtx_lock(&sc->reg_lock);
5597 		t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5598 		    A_TP_MIB_TNL_CNG_DROP_0 + i);
5599 		mtx_unlock(&sc->reg_lock);
5600 		tnl_cong_drops += v;
5601 		bg_map &= ~(1 << i);
5602 	}
5603 	pi->tnl_cong_drops = tnl_cong_drops;
5604 	getmicrotime(&pi->last_refreshed);
5605 }
5606 
5607 static void
5608 cxgbe_tick(void *arg)
5609 {
5610 	struct port_info *pi = arg;
5611 	struct adapter *sc = pi->adapter;
5612 
5613 	PORT_LOCK_ASSERT_OWNED(pi);
5614 	cxgbe_refresh_stats(sc, pi);
5615 
5616 	callout_schedule(&pi->tick, hz);
5617 }
5618 
5619 void
5620 vi_tick(void *arg)
5621 {
5622 	struct vi_info *vi = arg;
5623 	struct adapter *sc = vi->pi->adapter;
5624 
5625 	vi_refresh_stats(sc, vi);
5626 
5627 	callout_schedule(&vi->tick, hz);
5628 }
5629 
5630 /*
5631  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5632  */
5633 static char *caps_decoder[] = {
5634 	"\20\001IPMI\002NCSI",				/* 0: NBM */
5635 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
5636 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
5637 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
5638 	    "\006HASHFILTER\007ETHOFLD",
5639 	"\20\001TOE",					/* 4: TOE */
5640 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
5641 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
5642 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5643 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5644 	    "\007T10DIF"
5645 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5646 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
5647 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
5648 		    "\004PO_INITIATOR\005PO_TARGET",
5649 };
5650 
5651 void
5652 t4_sysctls(struct adapter *sc)
5653 {
5654 	struct sysctl_ctx_list *ctx;
5655 	struct sysctl_oid *oid;
5656 	struct sysctl_oid_list *children, *c0;
5657 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5658 
5659 	ctx = device_get_sysctl_ctx(sc->dev);
5660 
5661 	/*
5662 	 * dev.t4nex.X.
5663 	 */
5664 	oid = device_get_sysctl_tree(sc->dev);
5665 	c0 = children = SYSCTL_CHILDREN(oid);
5666 
5667 	sc->sc_do_rxcopy = 1;
5668 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5669 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5670 
5671 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5672 	    sc->params.nports, "# of ports");
5673 
5674 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5675 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, (uintptr_t)&sc->doorbells,
5676 	    sysctl_bitfield_8b, "A", "available doorbells");
5677 
5678 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5679 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
5680 
5681 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5682 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5683 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5684 	    "interrupt holdoff timer values (us)");
5685 
5686 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5687 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5688 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5689 	    "interrupt holdoff packet counter values");
5690 
5691 	t4_sge_sysctls(sc, ctx, children);
5692 
5693 	sc->lro_timeout = 100;
5694 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5695 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5696 
5697 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5698 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
5699 
5700 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5701 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5702 
5703 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5704 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5705 
5706 	if (sc->flags & IS_VF)
5707 		return;
5708 
5709 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5710 	    NULL, chip_rev(sc), "chip hardware revision");
5711 
5712 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5713 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5714 
5715 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5716 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5717 
5718 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5719 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5720 
5721 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5722 	    CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5723 
5724 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5725 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5726 
5727 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5728 	    sc->er_version, 0, "expansion ROM version");
5729 
5730 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5731 	    sc->bs_version, 0, "bootstrap firmware version");
5732 
5733 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5734 	    NULL, sc->params.scfg_vers, "serial config version");
5735 
5736 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5737 	    NULL, sc->params.vpd_vers, "VPD version");
5738 
5739 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5740 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5741 
5742 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5743 	    sc->cfcsum, "config file checksum");
5744 
5745 #define SYSCTL_CAP(name, n, text) \
5746 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5747 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], (uintptr_t)&sc->name, \
5748 	    sysctl_bitfield_16b, "A", "available " text " capabilities")
5749 
5750 	SYSCTL_CAP(nbmcaps, 0, "NBM");
5751 	SYSCTL_CAP(linkcaps, 1, "link");
5752 	SYSCTL_CAP(switchcaps, 2, "switch");
5753 	SYSCTL_CAP(niccaps, 3, "NIC");
5754 	SYSCTL_CAP(toecaps, 4, "TCP offload");
5755 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
5756 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5757 	SYSCTL_CAP(cryptocaps, 7, "crypto");
5758 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
5759 #undef SYSCTL_CAP
5760 
5761 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5762 	    NULL, sc->tids.nftids, "number of filters");
5763 
5764 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5765 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5766 	    "chip temperature (in Celsius)");
5767 
5768 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "loadavg", CTLTYPE_STRING |
5769 	    CTLFLAG_RD, sc, 0, sysctl_loadavg, "A",
5770 	    "microprocessor load averages (debug firmwares only)");
5771 
5772 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
5773 	    &sc->params.core_vdd, 0, "core Vdd (in mV)");
5774 
5775 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "local_cpus",
5776 	    CTLTYPE_STRING | CTLFLAG_RD, sc, LOCAL_CPUS,
5777 	    sysctl_cpus, "A", "local CPUs");
5778 
5779 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "intr_cpus",
5780 	    CTLTYPE_STRING | CTLFLAG_RD, sc, INTR_CPUS,
5781 	    sysctl_cpus, "A", "preferred CPUs for interrupts");
5782 
5783 	/*
5784 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
5785 	 */
5786 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5787 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5788 	    "logs and miscellaneous information");
5789 	children = SYSCTL_CHILDREN(oid);
5790 
5791 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
5792 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5793 	    sysctl_cctrl, "A", "congestion control");
5794 
5795 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
5796 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5797 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
5798 
5799 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
5800 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
5801 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
5802 
5803 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
5804 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
5805 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
5806 
5807 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
5808 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
5809 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
5810 
5811 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
5812 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
5813 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
5814 
5815 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
5816 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
5817 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
5818 
5819 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
5820 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5821 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
5822 	    "A", "CIM logic analyzer");
5823 
5824 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
5825 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5826 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
5827 
5828 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
5829 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
5830 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
5831 
5832 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
5833 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
5834 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
5835 
5836 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
5837 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
5838 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
5839 
5840 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5841 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5842 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5843 
5844 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5845 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5846 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5847 
5848 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5849 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5850 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5851 
5852 	if (chip_id(sc) > CHELSIO_T4) {
5853 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5854 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5855 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5856 
5857 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5858 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5859 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5860 	}
5861 
5862 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5863 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5864 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5865 
5866 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5867 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5868 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
5869 
5870 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5871 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5872 	    sysctl_cpl_stats, "A", "CPL statistics");
5873 
5874 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5875 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5876 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5877 
5878 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5879 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5880 	    sysctl_devlog, "A", "firmware's device log");
5881 
5882 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5883 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5884 	    sysctl_fcoe_stats, "A", "FCoE statistics");
5885 
5886 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5887 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5888 	    sysctl_hw_sched, "A", "hardware scheduler ");
5889 
5890 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5891 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5892 	    sysctl_l2t, "A", "hardware L2 table");
5893 
5894 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "smt",
5895 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5896 	    sysctl_smt, "A", "hardware source MAC table");
5897 
5898 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5899 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5900 	    sysctl_lb_stats, "A", "loopback statistics");
5901 
5902 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5903 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5904 	    sysctl_meminfo, "A", "memory regions");
5905 
5906 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5907 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5908 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5909 	    "A", "MPS TCAM entries");
5910 
5911 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5912 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5913 	    sysctl_path_mtus, "A", "path MTUs");
5914 
5915 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5916 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5917 	    sysctl_pm_stats, "A", "PM statistics");
5918 
5919 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5920 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5921 	    sysctl_rdma_stats, "A", "RDMA statistics");
5922 
5923 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5924 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5925 	    sysctl_tcp_stats, "A", "TCP statistics");
5926 
5927 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5928 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5929 	    sysctl_tids, "A", "TID information");
5930 
5931 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5932 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5933 	    sysctl_tp_err_stats, "A", "TP error statistics");
5934 
5935 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5936 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5937 	    "TP logic analyzer event capture mask");
5938 
5939 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5940 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5941 	    sysctl_tp_la, "A", "TP logic analyzer");
5942 
5943 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5944 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5945 	    sysctl_tx_rate, "A", "Tx rate");
5946 
5947 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5948 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5949 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5950 
5951 	if (chip_id(sc) >= CHELSIO_T5) {
5952 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5953 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5954 		    sysctl_wcwr_stats, "A", "write combined work requests");
5955 	}
5956 
5957 #ifdef TCP_OFFLOAD
5958 	if (is_offload(sc)) {
5959 		int i;
5960 		char s[4];
5961 
5962 		/*
5963 		 * dev.t4nex.X.toe.
5964 		 */
5965 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5966 		    NULL, "TOE parameters");
5967 		children = SYSCTL_CHILDREN(oid);
5968 
5969 		sc->tt.cong_algorithm = -1;
5970 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
5971 		    CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
5972 		    "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
5973 		    "3 = highspeed)");
5974 
5975 		sc->tt.sndbuf = 256 * 1024;
5976 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5977 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
5978 
5979 		sc->tt.ddp = 0;
5980 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5981 		    &sc->tt.ddp, 0, "DDP allowed");
5982 
5983 		sc->tt.rx_coalesce = 1;
5984 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5985 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5986 
5987 		sc->tt.tls = 0;
5988 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tls", CTLFLAG_RW,
5989 		    &sc->tt.tls, 0, "Inline TLS allowed");
5990 
5991 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls_rx_ports",
5992 		    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tls_rx_ports,
5993 		    "I", "TCP ports that use inline TLS+TOE RX");
5994 
5995 		sc->tt.tx_align = 1;
5996 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5997 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5998 
5999 		sc->tt.tx_zcopy = 0;
6000 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
6001 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
6002 		    "Enable zero-copy aio_write(2)");
6003 
6004 		sc->tt.cop_managed_offloading = !!t4_cop_managed_offloading;
6005 		SYSCTL_ADD_INT(ctx, children, OID_AUTO,
6006 		    "cop_managed_offloading", CTLFLAG_RW,
6007 		    &sc->tt.cop_managed_offloading, 0,
6008 		    "COP (Connection Offload Policy) controls all TOE offload");
6009 
6010 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
6011 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
6012 		    "TP timer tick (us)");
6013 
6014 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
6015 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
6016 		    "TCP timestamp tick (us)");
6017 
6018 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
6019 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
6020 		    "DACK tick (us)");
6021 
6022 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
6023 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
6024 		    "IU", "DACK timer (us)");
6025 
6026 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
6027 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
6028 		    sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
6029 
6030 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
6031 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
6032 		    sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
6033 
6034 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
6035 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
6036 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
6037 
6038 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
6039 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
6040 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
6041 
6042 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
6043 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
6044 		    sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
6045 
6046 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
6047 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
6048 		    sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
6049 
6050 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
6051 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
6052 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
6053 
6054 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
6055 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
6056 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
6057 
6058 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
6059 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
6060 		    sysctl_tp_shift_cnt, "IU",
6061 		    "Number of SYN retransmissions before abort");
6062 
6063 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
6064 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
6065 		    sysctl_tp_shift_cnt, "IU",
6066 		    "Number of retransmissions before abort");
6067 
6068 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
6069 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
6070 		    sysctl_tp_shift_cnt, "IU",
6071 		    "Number of keepalive probes before abort");
6072 
6073 		oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
6074 		    CTLFLAG_RD, NULL, "TOE retransmit backoffs");
6075 		children = SYSCTL_CHILDREN(oid);
6076 		for (i = 0; i < 16; i++) {
6077 			snprintf(s, sizeof(s), "%u", i);
6078 			SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
6079 			    CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
6080 			    "IU", "TOE retransmit backoff");
6081 		}
6082 	}
6083 #endif
6084 }
6085 
6086 void
6087 vi_sysctls(struct vi_info *vi)
6088 {
6089 	struct sysctl_ctx_list *ctx;
6090 	struct sysctl_oid *oid;
6091 	struct sysctl_oid_list *children;
6092 
6093 	ctx = device_get_sysctl_ctx(vi->dev);
6094 
6095 	/*
6096 	 * dev.v?(cxgbe|cxl).X.
6097 	 */
6098 	oid = device_get_sysctl_tree(vi->dev);
6099 	children = SYSCTL_CHILDREN(oid);
6100 
6101 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
6102 	    vi->viid, "VI identifer");
6103 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
6104 	    &vi->nrxq, 0, "# of rx queues");
6105 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
6106 	    &vi->ntxq, 0, "# of tx queues");
6107 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
6108 	    &vi->first_rxq, 0, "index of first rx queue");
6109 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
6110 	    &vi->first_txq, 0, "index of first tx queue");
6111 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_base", CTLFLAG_RD, NULL,
6112 	    vi->rss_base, "start of RSS indirection table");
6113 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
6114 	    vi->rss_size, "size of RSS indirection table");
6115 
6116 	if (IS_MAIN_VI(vi)) {
6117 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
6118 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
6119 		    "Reserve queue 0 for non-flowid packets");
6120 	}
6121 
6122 #ifdef TCP_OFFLOAD
6123 	if (vi->nofldrxq != 0) {
6124 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
6125 		    &vi->nofldrxq, 0,
6126 		    "# of rx queues for offloaded TCP connections");
6127 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
6128 		    &vi->nofldtxq, 0,
6129 		    "# of tx queues for offloaded TCP connections");
6130 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
6131 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
6132 		    "index of first TOE rx queue");
6133 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
6134 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
6135 		    "index of first TOE tx queue");
6136 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
6137 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
6138 		    sysctl_holdoff_tmr_idx_ofld, "I",
6139 		    "holdoff timer index for TOE queues");
6140 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
6141 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
6142 		    sysctl_holdoff_pktc_idx_ofld, "I",
6143 		    "holdoff packet counter index for TOE queues");
6144 	}
6145 #endif
6146 #ifdef DEV_NETMAP
6147 	if (vi->nnmrxq != 0) {
6148 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
6149 		    &vi->nnmrxq, 0, "# of netmap rx queues");
6150 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
6151 		    &vi->nnmtxq, 0, "# of netmap tx queues");
6152 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
6153 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
6154 		    "index of first netmap rx queue");
6155 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
6156 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
6157 		    "index of first netmap tx queue");
6158 	}
6159 #endif
6160 
6161 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
6162 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
6163 	    "holdoff timer index");
6164 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
6165 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
6166 	    "holdoff packet counter index");
6167 
6168 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
6169 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
6170 	    "rx queue size");
6171 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
6172 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
6173 	    "tx queue size");
6174 }
6175 
6176 static void
6177 cxgbe_sysctls(struct port_info *pi)
6178 {
6179 	struct sysctl_ctx_list *ctx;
6180 	struct sysctl_oid *oid;
6181 	struct sysctl_oid_list *children, *children2;
6182 	struct adapter *sc = pi->adapter;
6183 	int i;
6184 	char name[16];
6185 	static char *tc_flags = {"\20\1USER\2SYNC\3ASYNC\4ERR"};
6186 
6187 	ctx = device_get_sysctl_ctx(pi->dev);
6188 
6189 	/*
6190 	 * dev.cxgbe.X.
6191 	 */
6192 	oid = device_get_sysctl_tree(pi->dev);
6193 	children = SYSCTL_CHILDREN(oid);
6194 
6195 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
6196 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
6197 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
6198 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
6199 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
6200 		    "PHY temperature (in Celsius)");
6201 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
6202 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
6203 		    "PHY firmware version");
6204 	}
6205 
6206 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
6207 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
6208 	    "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
6209 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
6210 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
6211 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
6212 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
6213 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
6214 	    "autonegotiation (-1 = not supported)");
6215 
6216 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
6217 	    port_top_speed(pi), "max speed (in Gbps)");
6218 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
6219 	    pi->mps_bg_map, "MPS buffer group map");
6220 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
6221 	    NULL, pi->rx_e_chan_map, "TP rx e-channel map");
6222 
6223 	if (sc->flags & IS_VF)
6224 		return;
6225 
6226 	/*
6227 	 * dev.(cxgbe|cxl).X.tc.
6228 	 */
6229 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
6230 	    "Tx scheduler traffic classes (cl_rl)");
6231 	children2 = SYSCTL_CHILDREN(oid);
6232 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "pktsize",
6233 	    CTLFLAG_RW, &pi->sched_params->pktsize, 0,
6234 	    "pktsize for per-flow cl-rl (0 means up to the driver )");
6235 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "burstsize",
6236 	    CTLFLAG_RW, &pi->sched_params->burstsize, 0,
6237 	    "burstsize for per-flow cl-rl (0 means up to the driver)");
6238 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
6239 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
6240 
6241 		snprintf(name, sizeof(name), "%d", i);
6242 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
6243 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
6244 		    "traffic class"));
6245 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "flags",
6246 		    CTLTYPE_STRING | CTLFLAG_RD, tc_flags, (uintptr_t)&tc->flags,
6247 		    sysctl_bitfield_8b, "A", "flags");
6248 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
6249 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
6250 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
6251 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
6252 		    sysctl_tc_params, "A", "traffic class parameters");
6253 	}
6254 
6255 	/*
6256 	 * dev.cxgbe.X.stats.
6257 	 */
6258 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
6259 	    NULL, "port statistics");
6260 	children = SYSCTL_CHILDREN(oid);
6261 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
6262 	    &pi->tx_parse_error, 0,
6263 	    "# of tx packets with invalid length or # of segments");
6264 
6265 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
6266 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
6267 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
6268 	    sysctl_handle_t4_reg64, "QU", desc)
6269 
6270 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
6271 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
6272 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
6273 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
6274 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
6275 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
6276 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
6277 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
6278 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
6279 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
6280 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
6281 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
6282 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
6283 	    "# of tx frames in this range",
6284 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
6285 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
6286 	    "# of tx frames in this range",
6287 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
6288 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
6289 	    "# of tx frames in this range",
6290 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
6291 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
6292 	    "# of tx frames in this range",
6293 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
6294 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
6295 	    "# of tx frames in this range",
6296 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
6297 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
6298 	    "# of tx frames in this range",
6299 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
6300 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
6301 	    "# of tx frames in this range",
6302 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
6303 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
6304 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
6305 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
6306 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
6307 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
6308 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
6309 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
6310 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
6311 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
6312 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
6313 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
6314 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
6315 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
6316 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
6317 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
6318 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
6319 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
6320 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
6321 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
6322 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
6323 
6324 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
6325 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
6326 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
6327 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
6328 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
6329 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
6330 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
6331 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
6332 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
6333 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
6334 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
6335 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
6336 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
6337 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
6338 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
6339 	    "# of frames received with bad FCS",
6340 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
6341 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
6342 	    "# of frames received with length error",
6343 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
6344 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
6345 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
6346 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
6347 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
6348 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
6349 	    "# of rx frames in this range",
6350 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
6351 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
6352 	    "# of rx frames in this range",
6353 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
6354 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
6355 	    "# of rx frames in this range",
6356 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
6357 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
6358 	    "# of rx frames in this range",
6359 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
6360 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
6361 	    "# of rx frames in this range",
6362 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
6363 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
6364 	    "# of rx frames in this range",
6365 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
6366 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
6367 	    "# of rx frames in this range",
6368 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
6369 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
6370 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
6371 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
6372 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
6373 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
6374 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
6375 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
6376 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
6377 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
6378 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
6379 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
6380 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
6381 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
6382 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
6383 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
6384 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
6385 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
6386 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
6387 
6388 #undef SYSCTL_ADD_T4_REG64
6389 
6390 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
6391 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
6392 	    &pi->stats.name, desc)
6393 
6394 	/* We get these from port_stats and they may be stale by up to 1s */
6395 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
6396 	    "# drops due to buffer-group 0 overflows");
6397 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
6398 	    "# drops due to buffer-group 1 overflows");
6399 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
6400 	    "# drops due to buffer-group 2 overflows");
6401 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
6402 	    "# drops due to buffer-group 3 overflows");
6403 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
6404 	    "# of buffer-group 0 truncated packets");
6405 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
6406 	    "# of buffer-group 1 truncated packets");
6407 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
6408 	    "# of buffer-group 2 truncated packets");
6409 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
6410 	    "# of buffer-group 3 truncated packets");
6411 
6412 #undef SYSCTL_ADD_T4_PORTSTAT
6413 
6414 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_records",
6415 	    CTLFLAG_RD, &pi->tx_tls_records,
6416 	    "# of TLS records transmitted");
6417 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_octets",
6418 	    CTLFLAG_RD, &pi->tx_tls_octets,
6419 	    "# of payload octets in transmitted TLS records");
6420 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_records",
6421 	    CTLFLAG_RD, &pi->rx_tls_records,
6422 	    "# of TLS records received");
6423 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_octets",
6424 	    CTLFLAG_RD, &pi->rx_tls_octets,
6425 	    "# of payload octets in received TLS records");
6426 }
6427 
6428 static int
6429 sysctl_int_array(SYSCTL_HANDLER_ARGS)
6430 {
6431 	int rc, *i, space = 0;
6432 	struct sbuf sb;
6433 
6434 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
6435 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
6436 		if (space)
6437 			sbuf_printf(&sb, " ");
6438 		sbuf_printf(&sb, "%d", *i);
6439 		space = 1;
6440 	}
6441 	rc = sbuf_finish(&sb);
6442 	sbuf_delete(&sb);
6443 	return (rc);
6444 }
6445 
6446 static int
6447 sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS)
6448 {
6449 	int rc;
6450 	struct sbuf *sb;
6451 
6452 	rc = sysctl_wire_old_buffer(req, 0);
6453 	if (rc != 0)
6454 		return(rc);
6455 
6456 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6457 	if (sb == NULL)
6458 		return (ENOMEM);
6459 
6460 	sbuf_printf(sb, "%b", *(uint8_t *)(uintptr_t)arg2, (char *)arg1);
6461 	rc = sbuf_finish(sb);
6462 	sbuf_delete(sb);
6463 
6464 	return (rc);
6465 }
6466 
6467 static int
6468 sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS)
6469 {
6470 	int rc;
6471 	struct sbuf *sb;
6472 
6473 	rc = sysctl_wire_old_buffer(req, 0);
6474 	if (rc != 0)
6475 		return(rc);
6476 
6477 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6478 	if (sb == NULL)
6479 		return (ENOMEM);
6480 
6481 	sbuf_printf(sb, "%b", *(uint16_t *)(uintptr_t)arg2, (char *)arg1);
6482 	rc = sbuf_finish(sb);
6483 	sbuf_delete(sb);
6484 
6485 	return (rc);
6486 }
6487 
6488 static int
6489 sysctl_btphy(SYSCTL_HANDLER_ARGS)
6490 {
6491 	struct port_info *pi = arg1;
6492 	int op = arg2;
6493 	struct adapter *sc = pi->adapter;
6494 	u_int v;
6495 	int rc;
6496 
6497 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
6498 	if (rc)
6499 		return (rc);
6500 	/* XXX: magic numbers */
6501 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
6502 	    &v);
6503 	end_synchronized_op(sc, 0);
6504 	if (rc)
6505 		return (rc);
6506 	if (op == 0)
6507 		v /= 256;
6508 
6509 	rc = sysctl_handle_int(oidp, &v, 0, req);
6510 	return (rc);
6511 }
6512 
6513 static int
6514 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
6515 {
6516 	struct vi_info *vi = arg1;
6517 	int rc, val;
6518 
6519 	val = vi->rsrv_noflowq;
6520 	rc = sysctl_handle_int(oidp, &val, 0, req);
6521 	if (rc != 0 || req->newptr == NULL)
6522 		return (rc);
6523 
6524 	if ((val >= 1) && (vi->ntxq > 1))
6525 		vi->rsrv_noflowq = 1;
6526 	else
6527 		vi->rsrv_noflowq = 0;
6528 
6529 	return (rc);
6530 }
6531 
6532 static int
6533 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
6534 {
6535 	struct vi_info *vi = arg1;
6536 	struct adapter *sc = vi->pi->adapter;
6537 	int idx, rc, i;
6538 	struct sge_rxq *rxq;
6539 	uint8_t v;
6540 
6541 	idx = vi->tmr_idx;
6542 
6543 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6544 	if (rc != 0 || req->newptr == NULL)
6545 		return (rc);
6546 
6547 	if (idx < 0 || idx >= SGE_NTIMERS)
6548 		return (EINVAL);
6549 
6550 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6551 	    "t4tmr");
6552 	if (rc)
6553 		return (rc);
6554 
6555 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
6556 	for_each_rxq(vi, i, rxq) {
6557 #ifdef atomic_store_rel_8
6558 		atomic_store_rel_8(&rxq->iq.intr_params, v);
6559 #else
6560 		rxq->iq.intr_params = v;
6561 #endif
6562 	}
6563 	vi->tmr_idx = idx;
6564 
6565 	end_synchronized_op(sc, LOCK_HELD);
6566 	return (0);
6567 }
6568 
6569 static int
6570 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
6571 {
6572 	struct vi_info *vi = arg1;
6573 	struct adapter *sc = vi->pi->adapter;
6574 	int idx, rc;
6575 
6576 	idx = vi->pktc_idx;
6577 
6578 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6579 	if (rc != 0 || req->newptr == NULL)
6580 		return (rc);
6581 
6582 	if (idx < -1 || idx >= SGE_NCOUNTERS)
6583 		return (EINVAL);
6584 
6585 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6586 	    "t4pktc");
6587 	if (rc)
6588 		return (rc);
6589 
6590 	if (vi->flags & VI_INIT_DONE)
6591 		rc = EBUSY; /* cannot be changed once the queues are created */
6592 	else
6593 		vi->pktc_idx = idx;
6594 
6595 	end_synchronized_op(sc, LOCK_HELD);
6596 	return (rc);
6597 }
6598 
6599 static int
6600 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
6601 {
6602 	struct vi_info *vi = arg1;
6603 	struct adapter *sc = vi->pi->adapter;
6604 	int qsize, rc;
6605 
6606 	qsize = vi->qsize_rxq;
6607 
6608 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6609 	if (rc != 0 || req->newptr == NULL)
6610 		return (rc);
6611 
6612 	if (qsize < 128 || (qsize & 7))
6613 		return (EINVAL);
6614 
6615 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6616 	    "t4rxqs");
6617 	if (rc)
6618 		return (rc);
6619 
6620 	if (vi->flags & VI_INIT_DONE)
6621 		rc = EBUSY; /* cannot be changed once the queues are created */
6622 	else
6623 		vi->qsize_rxq = qsize;
6624 
6625 	end_synchronized_op(sc, LOCK_HELD);
6626 	return (rc);
6627 }
6628 
6629 static int
6630 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
6631 {
6632 	struct vi_info *vi = arg1;
6633 	struct adapter *sc = vi->pi->adapter;
6634 	int qsize, rc;
6635 
6636 	qsize = vi->qsize_txq;
6637 
6638 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6639 	if (rc != 0 || req->newptr == NULL)
6640 		return (rc);
6641 
6642 	if (qsize < 128 || qsize > 65536)
6643 		return (EINVAL);
6644 
6645 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6646 	    "t4txqs");
6647 	if (rc)
6648 		return (rc);
6649 
6650 	if (vi->flags & VI_INIT_DONE)
6651 		rc = EBUSY; /* cannot be changed once the queues are created */
6652 	else
6653 		vi->qsize_txq = qsize;
6654 
6655 	end_synchronized_op(sc, LOCK_HELD);
6656 	return (rc);
6657 }
6658 
6659 static int
6660 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
6661 {
6662 	struct port_info *pi = arg1;
6663 	struct adapter *sc = pi->adapter;
6664 	struct link_config *lc = &pi->link_cfg;
6665 	int rc;
6666 
6667 	if (req->newptr == NULL) {
6668 		struct sbuf *sb;
6669 		static char *bits = "\20\1RX\2TX\3AUTO";
6670 
6671 		rc = sysctl_wire_old_buffer(req, 0);
6672 		if (rc != 0)
6673 			return(rc);
6674 
6675 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6676 		if (sb == NULL)
6677 			return (ENOMEM);
6678 
6679 		if (lc->link_ok) {
6680 			sbuf_printf(sb, "%b", (lc->fc & (PAUSE_TX | PAUSE_RX)) |
6681 			    (lc->requested_fc & PAUSE_AUTONEG), bits);
6682 		} else {
6683 			sbuf_printf(sb, "%b", lc->requested_fc & (PAUSE_TX |
6684 			    PAUSE_RX | PAUSE_AUTONEG), bits);
6685 		}
6686 		rc = sbuf_finish(sb);
6687 		sbuf_delete(sb);
6688 	} else {
6689 		char s[2];
6690 		int n;
6691 
6692 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX |
6693 		    PAUSE_AUTONEG));
6694 		s[1] = 0;
6695 
6696 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6697 		if (rc != 0)
6698 			return(rc);
6699 
6700 		if (s[1] != 0)
6701 			return (EINVAL);
6702 		if (s[0] < '0' || s[0] > '9')
6703 			return (EINVAL);	/* not a number */
6704 		n = s[0] - '0';
6705 		if (n & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG))
6706 			return (EINVAL);	/* some other bit is set too */
6707 
6708 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6709 		    "t4PAUSE");
6710 		if (rc)
6711 			return (rc);
6712 		PORT_LOCK(pi);
6713 		lc->requested_fc = n;
6714 		fixup_link_config(pi);
6715 		if (pi->up_vis > 0)
6716 			rc = apply_link_config(pi);
6717 		set_current_media(pi);
6718 		PORT_UNLOCK(pi);
6719 		end_synchronized_op(sc, 0);
6720 	}
6721 
6722 	return (rc);
6723 }
6724 
6725 static int
6726 sysctl_fec(SYSCTL_HANDLER_ARGS)
6727 {
6728 	struct port_info *pi = arg1;
6729 	struct adapter *sc = pi->adapter;
6730 	struct link_config *lc = &pi->link_cfg;
6731 	int rc;
6732 	int8_t old;
6733 
6734 	if (req->newptr == NULL) {
6735 		struct sbuf *sb;
6736 		static char *bits = "\20\1RS\2BASE-R\3RSVD1\4RSVD2\5RSVD3\6AUTO";
6737 
6738 		rc = sysctl_wire_old_buffer(req, 0);
6739 		if (rc != 0)
6740 			return(rc);
6741 
6742 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6743 		if (sb == NULL)
6744 			return (ENOMEM);
6745 
6746 		/*
6747 		 * Display the requested_fec when the link is down -- the actual
6748 		 * FEC makes sense only when the link is up.
6749 		 */
6750 		if (lc->link_ok) {
6751 			sbuf_printf(sb, "%b", (lc->fec & M_FW_PORT_CAP32_FEC) |
6752 			    (lc->requested_fec & FEC_AUTO), bits);
6753 		} else {
6754 			sbuf_printf(sb, "%b", lc->requested_fec, bits);
6755 		}
6756 		rc = sbuf_finish(sb);
6757 		sbuf_delete(sb);
6758 	} else {
6759 		char s[3];
6760 		int n;
6761 
6762 		snprintf(s, sizeof(s), "%d",
6763 		    lc->requested_fec == FEC_AUTO ? -1 :
6764 		    lc->requested_fec & M_FW_PORT_CAP32_FEC);
6765 
6766 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6767 		if (rc != 0)
6768 			return(rc);
6769 
6770 		n = strtol(&s[0], NULL, 0);
6771 		if (n < 0 || n & FEC_AUTO)
6772 			n = FEC_AUTO;
6773 		else {
6774 			if (n & ~M_FW_PORT_CAP32_FEC)
6775 				return (EINVAL);/* some other bit is set too */
6776 			if (!powerof2(n))
6777 				return (EINVAL);/* one bit can be set at most */
6778 		}
6779 
6780 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6781 		    "t4fec");
6782 		if (rc)
6783 			return (rc);
6784 		PORT_LOCK(pi);
6785 		old = lc->requested_fec;
6786 		if (n == FEC_AUTO)
6787 			lc->requested_fec = FEC_AUTO;
6788 		else if (n == 0)
6789 			lc->requested_fec = FEC_NONE;
6790 		else {
6791 			if ((lc->supported | V_FW_PORT_CAP32_FEC(n)) !=
6792 			    lc->supported) {
6793 				rc = ENOTSUP;
6794 				goto done;
6795 			}
6796 			lc->requested_fec = n;
6797 		}
6798 		fixup_link_config(pi);
6799 		if (pi->up_vis > 0) {
6800 			rc = apply_link_config(pi);
6801 			if (rc != 0) {
6802 				lc->requested_fec = old;
6803 				if (rc == FW_EPROTO)
6804 					rc = ENOTSUP;
6805 			}
6806 		}
6807 done:
6808 		PORT_UNLOCK(pi);
6809 		end_synchronized_op(sc, 0);
6810 	}
6811 
6812 	return (rc);
6813 }
6814 
6815 static int
6816 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
6817 {
6818 	struct port_info *pi = arg1;
6819 	struct adapter *sc = pi->adapter;
6820 	struct link_config *lc = &pi->link_cfg;
6821 	int rc, val;
6822 
6823 	if (lc->supported & FW_PORT_CAP32_ANEG)
6824 		val = lc->requested_aneg == AUTONEG_DISABLE ? 0 : 1;
6825 	else
6826 		val = -1;
6827 	rc = sysctl_handle_int(oidp, &val, 0, req);
6828 	if (rc != 0 || req->newptr == NULL)
6829 		return (rc);
6830 	if (val == 0)
6831 		val = AUTONEG_DISABLE;
6832 	else if (val == 1)
6833 		val = AUTONEG_ENABLE;
6834 	else
6835 		val = AUTONEG_AUTO;
6836 
6837 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6838 	    "t4aneg");
6839 	if (rc)
6840 		return (rc);
6841 	PORT_LOCK(pi);
6842 	if (val == AUTONEG_ENABLE && !(lc->supported & FW_PORT_CAP32_ANEG)) {
6843 		rc = ENOTSUP;
6844 		goto done;
6845 	}
6846 	lc->requested_aneg = val;
6847 	fixup_link_config(pi);
6848 	if (pi->up_vis > 0)
6849 		rc = apply_link_config(pi);
6850 	set_current_media(pi);
6851 done:
6852 	PORT_UNLOCK(pi);
6853 	end_synchronized_op(sc, 0);
6854 	return (rc);
6855 }
6856 
6857 static int
6858 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
6859 {
6860 	struct adapter *sc = arg1;
6861 	int reg = arg2;
6862 	uint64_t val;
6863 
6864 	val = t4_read_reg64(sc, reg);
6865 
6866 	return (sysctl_handle_64(oidp, &val, 0, req));
6867 }
6868 
6869 static int
6870 sysctl_temperature(SYSCTL_HANDLER_ARGS)
6871 {
6872 	struct adapter *sc = arg1;
6873 	int rc, t;
6874 	uint32_t param, val;
6875 
6876 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
6877 	if (rc)
6878 		return (rc);
6879 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6880 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
6881 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
6882 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6883 	end_synchronized_op(sc, 0);
6884 	if (rc)
6885 		return (rc);
6886 
6887 	/* unknown is returned as 0 but we display -1 in that case */
6888 	t = val == 0 ? -1 : val;
6889 
6890 	rc = sysctl_handle_int(oidp, &t, 0, req);
6891 	return (rc);
6892 }
6893 
6894 static int
6895 sysctl_loadavg(SYSCTL_HANDLER_ARGS)
6896 {
6897 	struct adapter *sc = arg1;
6898 	struct sbuf *sb;
6899 	int rc;
6900 	uint32_t param, val;
6901 
6902 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4lavg");
6903 	if (rc)
6904 		return (rc);
6905 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6906 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_LOAD);
6907 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6908 	end_synchronized_op(sc, 0);
6909 	if (rc)
6910 		return (rc);
6911 
6912 	rc = sysctl_wire_old_buffer(req, 0);
6913 	if (rc != 0)
6914 		return (rc);
6915 
6916 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6917 	if (sb == NULL)
6918 		return (ENOMEM);
6919 
6920 	if (val == 0xffffffff) {
6921 		/* Only debug and custom firmwares report load averages. */
6922 		sbuf_printf(sb, "not available");
6923 	} else {
6924 		sbuf_printf(sb, "%d %d %d", val & 0xff, (val >> 8) & 0xff,
6925 		    (val >> 16) & 0xff);
6926 	}
6927 	rc = sbuf_finish(sb);
6928 	sbuf_delete(sb);
6929 
6930 	return (rc);
6931 }
6932 
6933 static int
6934 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
6935 {
6936 	struct adapter *sc = arg1;
6937 	struct sbuf *sb;
6938 	int rc, i;
6939 	uint16_t incr[NMTUS][NCCTRL_WIN];
6940 	static const char *dec_fac[] = {
6941 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
6942 		"0.9375"
6943 	};
6944 
6945 	rc = sysctl_wire_old_buffer(req, 0);
6946 	if (rc != 0)
6947 		return (rc);
6948 
6949 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6950 	if (sb == NULL)
6951 		return (ENOMEM);
6952 
6953 	t4_read_cong_tbl(sc, incr);
6954 
6955 	for (i = 0; i < NCCTRL_WIN; ++i) {
6956 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
6957 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
6958 		    incr[5][i], incr[6][i], incr[7][i]);
6959 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
6960 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
6961 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
6962 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
6963 	}
6964 
6965 	rc = sbuf_finish(sb);
6966 	sbuf_delete(sb);
6967 
6968 	return (rc);
6969 }
6970 
6971 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
6972 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
6973 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
6974 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
6975 };
6976 
6977 static int
6978 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
6979 {
6980 	struct adapter *sc = arg1;
6981 	struct sbuf *sb;
6982 	int rc, i, n, qid = arg2;
6983 	uint32_t *buf, *p;
6984 	char *qtype;
6985 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
6986 
6987 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
6988 	    ("%s: bad qid %d\n", __func__, qid));
6989 
6990 	if (qid < CIM_NUM_IBQ) {
6991 		/* inbound queue */
6992 		qtype = "IBQ";
6993 		n = 4 * CIM_IBQ_SIZE;
6994 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6995 		rc = t4_read_cim_ibq(sc, qid, buf, n);
6996 	} else {
6997 		/* outbound queue */
6998 		qtype = "OBQ";
6999 		qid -= CIM_NUM_IBQ;
7000 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
7001 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
7002 		rc = t4_read_cim_obq(sc, qid, buf, n);
7003 	}
7004 
7005 	if (rc < 0) {
7006 		rc = -rc;
7007 		goto done;
7008 	}
7009 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
7010 
7011 	rc = sysctl_wire_old_buffer(req, 0);
7012 	if (rc != 0)
7013 		goto done;
7014 
7015 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7016 	if (sb == NULL) {
7017 		rc = ENOMEM;
7018 		goto done;
7019 	}
7020 
7021 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
7022 	for (i = 0, p = buf; i < n; i += 16, p += 4)
7023 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
7024 		    p[2], p[3]);
7025 
7026 	rc = sbuf_finish(sb);
7027 	sbuf_delete(sb);
7028 done:
7029 	free(buf, M_CXGBE);
7030 	return (rc);
7031 }
7032 
7033 static int
7034 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
7035 {
7036 	struct adapter *sc = arg1;
7037 	u_int cfg;
7038 	struct sbuf *sb;
7039 	uint32_t *buf, *p;
7040 	int rc;
7041 
7042 	MPASS(chip_id(sc) <= CHELSIO_T5);
7043 
7044 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
7045 	if (rc != 0)
7046 		return (rc);
7047 
7048 	rc = sysctl_wire_old_buffer(req, 0);
7049 	if (rc != 0)
7050 		return (rc);
7051 
7052 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7053 	if (sb == NULL)
7054 		return (ENOMEM);
7055 
7056 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
7057 	    M_ZERO | M_WAITOK);
7058 
7059 	rc = -t4_cim_read_la(sc, buf, NULL);
7060 	if (rc != 0)
7061 		goto done;
7062 
7063 	sbuf_printf(sb, "Status   Data      PC%s",
7064 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
7065 	    "     LS0Stat  LS0Addr             LS0Data");
7066 
7067 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
7068 		if (cfg & F_UPDBGLACAPTPCONLY) {
7069 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
7070 			    p[6], p[7]);
7071 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
7072 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
7073 			    p[4] & 0xff, p[5] >> 8);
7074 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
7075 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7076 			    p[1] & 0xf, p[2] >> 4);
7077 		} else {
7078 			sbuf_printf(sb,
7079 			    "\n  %02x   %x%07x %x%07x %08x %08x "
7080 			    "%08x%08x%08x%08x",
7081 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7082 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
7083 			    p[6], p[7]);
7084 		}
7085 	}
7086 
7087 	rc = sbuf_finish(sb);
7088 	sbuf_delete(sb);
7089 done:
7090 	free(buf, M_CXGBE);
7091 	return (rc);
7092 }
7093 
7094 static int
7095 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
7096 {
7097 	struct adapter *sc = arg1;
7098 	u_int cfg;
7099 	struct sbuf *sb;
7100 	uint32_t *buf, *p;
7101 	int rc;
7102 
7103 	MPASS(chip_id(sc) > CHELSIO_T5);
7104 
7105 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
7106 	if (rc != 0)
7107 		return (rc);
7108 
7109 	rc = sysctl_wire_old_buffer(req, 0);
7110 	if (rc != 0)
7111 		return (rc);
7112 
7113 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7114 	if (sb == NULL)
7115 		return (ENOMEM);
7116 
7117 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
7118 	    M_ZERO | M_WAITOK);
7119 
7120 	rc = -t4_cim_read_la(sc, buf, NULL);
7121 	if (rc != 0)
7122 		goto done;
7123 
7124 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
7125 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
7126 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
7127 
7128 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
7129 		if (cfg & F_UPDBGLACAPTPCONLY) {
7130 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
7131 			    p[3] & 0xff, p[2], p[1], p[0]);
7132 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
7133 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
7134 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
7135 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
7136 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
7137 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
7138 			    p[6] >> 16);
7139 		} else {
7140 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
7141 			    "%08x %08x %08x %08x %08x %08x",
7142 			    (p[9] >> 16) & 0xff,
7143 			    p[9] & 0xffff, p[8] >> 16,
7144 			    p[8] & 0xffff, p[7] >> 16,
7145 			    p[7] & 0xffff, p[6] >> 16,
7146 			    p[2], p[1], p[0], p[5], p[4], p[3]);
7147 		}
7148 	}
7149 
7150 	rc = sbuf_finish(sb);
7151 	sbuf_delete(sb);
7152 done:
7153 	free(buf, M_CXGBE);
7154 	return (rc);
7155 }
7156 
7157 static int
7158 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
7159 {
7160 	struct adapter *sc = arg1;
7161 	u_int i;
7162 	struct sbuf *sb;
7163 	uint32_t *buf, *p;
7164 	int rc;
7165 
7166 	rc = sysctl_wire_old_buffer(req, 0);
7167 	if (rc != 0)
7168 		return (rc);
7169 
7170 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7171 	if (sb == NULL)
7172 		return (ENOMEM);
7173 
7174 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
7175 	    M_ZERO | M_WAITOK);
7176 
7177 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
7178 	p = buf;
7179 
7180 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7181 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
7182 		    p[1], p[0]);
7183 	}
7184 
7185 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
7186 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7187 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
7188 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
7189 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
7190 		    (p[1] >> 2) | ((p[2] & 3) << 30),
7191 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
7192 		    p[0] & 1);
7193 	}
7194 
7195 	rc = sbuf_finish(sb);
7196 	sbuf_delete(sb);
7197 	free(buf, M_CXGBE);
7198 	return (rc);
7199 }
7200 
7201 static int
7202 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
7203 {
7204 	struct adapter *sc = arg1;
7205 	u_int i;
7206 	struct sbuf *sb;
7207 	uint32_t *buf, *p;
7208 	int rc;
7209 
7210 	rc = sysctl_wire_old_buffer(req, 0);
7211 	if (rc != 0)
7212 		return (rc);
7213 
7214 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7215 	if (sb == NULL)
7216 		return (ENOMEM);
7217 
7218 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
7219 	    M_ZERO | M_WAITOK);
7220 
7221 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
7222 	p = buf;
7223 
7224 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
7225 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7226 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
7227 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
7228 		    p[4], p[3], p[2], p[1], p[0]);
7229 	}
7230 
7231 	sbuf_printf(sb, "\n\nCntl ID               Data");
7232 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7233 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
7234 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
7235 	}
7236 
7237 	rc = sbuf_finish(sb);
7238 	sbuf_delete(sb);
7239 	free(buf, M_CXGBE);
7240 	return (rc);
7241 }
7242 
7243 static int
7244 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
7245 {
7246 	struct adapter *sc = arg1;
7247 	struct sbuf *sb;
7248 	int rc, i;
7249 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7250 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7251 	uint16_t thres[CIM_NUM_IBQ];
7252 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
7253 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
7254 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
7255 
7256 	cim_num_obq = sc->chip_params->cim_num_obq;
7257 	if (is_t4(sc)) {
7258 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
7259 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
7260 	} else {
7261 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
7262 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
7263 	}
7264 	nq = CIM_NUM_IBQ + cim_num_obq;
7265 
7266 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
7267 	if (rc == 0)
7268 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
7269 	if (rc != 0)
7270 		return (rc);
7271 
7272 	t4_read_cimq_cfg(sc, base, size, thres);
7273 
7274 	rc = sysctl_wire_old_buffer(req, 0);
7275 	if (rc != 0)
7276 		return (rc);
7277 
7278 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7279 	if (sb == NULL)
7280 		return (ENOMEM);
7281 
7282 	sbuf_printf(sb,
7283 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
7284 
7285 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
7286 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
7287 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
7288 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7289 		    G_QUEREMFLITS(p[2]) * 16);
7290 	for ( ; i < nq; i++, p += 4, wr += 2)
7291 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
7292 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
7293 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7294 		    G_QUEREMFLITS(p[2]) * 16);
7295 
7296 	rc = sbuf_finish(sb);
7297 	sbuf_delete(sb);
7298 
7299 	return (rc);
7300 }
7301 
7302 static int
7303 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
7304 {
7305 	struct adapter *sc = arg1;
7306 	struct sbuf *sb;
7307 	int rc;
7308 	struct tp_cpl_stats stats;
7309 
7310 	rc = sysctl_wire_old_buffer(req, 0);
7311 	if (rc != 0)
7312 		return (rc);
7313 
7314 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7315 	if (sb == NULL)
7316 		return (ENOMEM);
7317 
7318 	mtx_lock(&sc->reg_lock);
7319 	t4_tp_get_cpl_stats(sc, &stats, 0);
7320 	mtx_unlock(&sc->reg_lock);
7321 
7322 	if (sc->chip_params->nchan > 2) {
7323 		sbuf_printf(sb, "                 channel 0  channel 1"
7324 		    "  channel 2  channel 3");
7325 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
7326 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
7327 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
7328 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
7329 	} else {
7330 		sbuf_printf(sb, "                 channel 0  channel 1");
7331 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
7332 		    stats.req[0], stats.req[1]);
7333 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
7334 		    stats.rsp[0], stats.rsp[1]);
7335 	}
7336 
7337 	rc = sbuf_finish(sb);
7338 	sbuf_delete(sb);
7339 
7340 	return (rc);
7341 }
7342 
7343 static int
7344 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
7345 {
7346 	struct adapter *sc = arg1;
7347 	struct sbuf *sb;
7348 	int rc;
7349 	struct tp_usm_stats stats;
7350 
7351 	rc = sysctl_wire_old_buffer(req, 0);
7352 	if (rc != 0)
7353 		return(rc);
7354 
7355 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7356 	if (sb == NULL)
7357 		return (ENOMEM);
7358 
7359 	t4_get_usm_stats(sc, &stats, 1);
7360 
7361 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
7362 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
7363 	sbuf_printf(sb, "Drops:  %u", stats.drops);
7364 
7365 	rc = sbuf_finish(sb);
7366 	sbuf_delete(sb);
7367 
7368 	return (rc);
7369 }
7370 
7371 static const char * const devlog_level_strings[] = {
7372 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
7373 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
7374 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
7375 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
7376 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
7377 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
7378 };
7379 
7380 static const char * const devlog_facility_strings[] = {
7381 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
7382 	[FW_DEVLOG_FACILITY_CF]		= "CF",
7383 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
7384 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
7385 	[FW_DEVLOG_FACILITY_RES]	= "RES",
7386 	[FW_DEVLOG_FACILITY_HW]		= "HW",
7387 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
7388 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
7389 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
7390 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
7391 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
7392 	[FW_DEVLOG_FACILITY_VI]		= "VI",
7393 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
7394 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
7395 	[FW_DEVLOG_FACILITY_TM]		= "TM",
7396 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
7397 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
7398 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
7399 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
7400 	[FW_DEVLOG_FACILITY_RI]		= "RI",
7401 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
7402 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
7403 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
7404 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
7405 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
7406 };
7407 
7408 static int
7409 sysctl_devlog(SYSCTL_HANDLER_ARGS)
7410 {
7411 	struct adapter *sc = arg1;
7412 	struct devlog_params *dparams = &sc->params.devlog;
7413 	struct fw_devlog_e *buf, *e;
7414 	int i, j, rc, nentries, first = 0;
7415 	struct sbuf *sb;
7416 	uint64_t ftstamp = UINT64_MAX;
7417 
7418 	if (dparams->addr == 0)
7419 		return (ENXIO);
7420 
7421 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
7422 	if (buf == NULL)
7423 		return (ENOMEM);
7424 
7425 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
7426 	if (rc != 0)
7427 		goto done;
7428 
7429 	nentries = dparams->size / sizeof(struct fw_devlog_e);
7430 	for (i = 0; i < nentries; i++) {
7431 		e = &buf[i];
7432 
7433 		if (e->timestamp == 0)
7434 			break;	/* end */
7435 
7436 		e->timestamp = be64toh(e->timestamp);
7437 		e->seqno = be32toh(e->seqno);
7438 		for (j = 0; j < 8; j++)
7439 			e->params[j] = be32toh(e->params[j]);
7440 
7441 		if (e->timestamp < ftstamp) {
7442 			ftstamp = e->timestamp;
7443 			first = i;
7444 		}
7445 	}
7446 
7447 	if (buf[first].timestamp == 0)
7448 		goto done;	/* nothing in the log */
7449 
7450 	rc = sysctl_wire_old_buffer(req, 0);
7451 	if (rc != 0)
7452 		goto done;
7453 
7454 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7455 	if (sb == NULL) {
7456 		rc = ENOMEM;
7457 		goto done;
7458 	}
7459 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
7460 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
7461 
7462 	i = first;
7463 	do {
7464 		e = &buf[i];
7465 		if (e->timestamp == 0)
7466 			break;	/* end */
7467 
7468 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
7469 		    e->seqno, e->timestamp,
7470 		    (e->level < nitems(devlog_level_strings) ?
7471 			devlog_level_strings[e->level] : "UNKNOWN"),
7472 		    (e->facility < nitems(devlog_facility_strings) ?
7473 			devlog_facility_strings[e->facility] : "UNKNOWN"));
7474 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
7475 		    e->params[2], e->params[3], e->params[4],
7476 		    e->params[5], e->params[6], e->params[7]);
7477 
7478 		if (++i == nentries)
7479 			i = 0;
7480 	} while (i != first);
7481 
7482 	rc = sbuf_finish(sb);
7483 	sbuf_delete(sb);
7484 done:
7485 	free(buf, M_CXGBE);
7486 	return (rc);
7487 }
7488 
7489 static int
7490 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
7491 {
7492 	struct adapter *sc = arg1;
7493 	struct sbuf *sb;
7494 	int rc;
7495 	struct tp_fcoe_stats stats[MAX_NCHAN];
7496 	int i, nchan = sc->chip_params->nchan;
7497 
7498 	rc = sysctl_wire_old_buffer(req, 0);
7499 	if (rc != 0)
7500 		return (rc);
7501 
7502 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7503 	if (sb == NULL)
7504 		return (ENOMEM);
7505 
7506 	for (i = 0; i < nchan; i++)
7507 		t4_get_fcoe_stats(sc, i, &stats[i], 1);
7508 
7509 	if (nchan > 2) {
7510 		sbuf_printf(sb, "                   channel 0        channel 1"
7511 		    "        channel 2        channel 3");
7512 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
7513 		    stats[0].octets_ddp, stats[1].octets_ddp,
7514 		    stats[2].octets_ddp, stats[3].octets_ddp);
7515 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
7516 		    stats[0].frames_ddp, stats[1].frames_ddp,
7517 		    stats[2].frames_ddp, stats[3].frames_ddp);
7518 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
7519 		    stats[0].frames_drop, stats[1].frames_drop,
7520 		    stats[2].frames_drop, stats[3].frames_drop);
7521 	} else {
7522 		sbuf_printf(sb, "                   channel 0        channel 1");
7523 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
7524 		    stats[0].octets_ddp, stats[1].octets_ddp);
7525 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
7526 		    stats[0].frames_ddp, stats[1].frames_ddp);
7527 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
7528 		    stats[0].frames_drop, stats[1].frames_drop);
7529 	}
7530 
7531 	rc = sbuf_finish(sb);
7532 	sbuf_delete(sb);
7533 
7534 	return (rc);
7535 }
7536 
7537 static int
7538 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
7539 {
7540 	struct adapter *sc = arg1;
7541 	struct sbuf *sb;
7542 	int rc, i;
7543 	unsigned int map, kbps, ipg, mode;
7544 	unsigned int pace_tab[NTX_SCHED];
7545 
7546 	rc = sysctl_wire_old_buffer(req, 0);
7547 	if (rc != 0)
7548 		return (rc);
7549 
7550 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7551 	if (sb == NULL)
7552 		return (ENOMEM);
7553 
7554 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
7555 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
7556 	t4_read_pace_tbl(sc, pace_tab);
7557 
7558 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
7559 	    "Class IPG (0.1 ns)   Flow IPG (us)");
7560 
7561 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
7562 		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
7563 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
7564 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
7565 		if (kbps)
7566 			sbuf_printf(sb, "%9u     ", kbps);
7567 		else
7568 			sbuf_printf(sb, " disabled     ");
7569 
7570 		if (ipg)
7571 			sbuf_printf(sb, "%13u        ", ipg);
7572 		else
7573 			sbuf_printf(sb, "     disabled        ");
7574 
7575 		if (pace_tab[i])
7576 			sbuf_printf(sb, "%10u", pace_tab[i]);
7577 		else
7578 			sbuf_printf(sb, "  disabled");
7579 	}
7580 
7581 	rc = sbuf_finish(sb);
7582 	sbuf_delete(sb);
7583 
7584 	return (rc);
7585 }
7586 
7587 static int
7588 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
7589 {
7590 	struct adapter *sc = arg1;
7591 	struct sbuf *sb;
7592 	int rc, i, j;
7593 	uint64_t *p0, *p1;
7594 	struct lb_port_stats s[2];
7595 	static const char *stat_name[] = {
7596 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
7597 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
7598 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
7599 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
7600 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
7601 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
7602 		"BG2FramesTrunc:", "BG3FramesTrunc:"
7603 	};
7604 
7605 	rc = sysctl_wire_old_buffer(req, 0);
7606 	if (rc != 0)
7607 		return (rc);
7608 
7609 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7610 	if (sb == NULL)
7611 		return (ENOMEM);
7612 
7613 	memset(s, 0, sizeof(s));
7614 
7615 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
7616 		t4_get_lb_stats(sc, i, &s[0]);
7617 		t4_get_lb_stats(sc, i + 1, &s[1]);
7618 
7619 		p0 = &s[0].octets;
7620 		p1 = &s[1].octets;
7621 		sbuf_printf(sb, "%s                       Loopback %u"
7622 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
7623 
7624 		for (j = 0; j < nitems(stat_name); j++)
7625 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
7626 				   *p0++, *p1++);
7627 	}
7628 
7629 	rc = sbuf_finish(sb);
7630 	sbuf_delete(sb);
7631 
7632 	return (rc);
7633 }
7634 
7635 static int
7636 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
7637 {
7638 	int rc = 0;
7639 	struct port_info *pi = arg1;
7640 	struct link_config *lc = &pi->link_cfg;
7641 	struct sbuf *sb;
7642 
7643 	rc = sysctl_wire_old_buffer(req, 0);
7644 	if (rc != 0)
7645 		return(rc);
7646 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
7647 	if (sb == NULL)
7648 		return (ENOMEM);
7649 
7650 	if (lc->link_ok || lc->link_down_rc == 255)
7651 		sbuf_printf(sb, "n/a");
7652 	else
7653 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
7654 
7655 	rc = sbuf_finish(sb);
7656 	sbuf_delete(sb);
7657 
7658 	return (rc);
7659 }
7660 
7661 struct mem_desc {
7662 	unsigned int base;
7663 	unsigned int limit;
7664 	unsigned int idx;
7665 };
7666 
7667 static int
7668 mem_desc_cmp(const void *a, const void *b)
7669 {
7670 	return ((const struct mem_desc *)a)->base -
7671 	       ((const struct mem_desc *)b)->base;
7672 }
7673 
7674 static void
7675 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
7676     unsigned int to)
7677 {
7678 	unsigned int size;
7679 
7680 	if (from == to)
7681 		return;
7682 
7683 	size = to - from + 1;
7684 	if (size == 0)
7685 		return;
7686 
7687 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
7688 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
7689 }
7690 
7691 static int
7692 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
7693 {
7694 	struct adapter *sc = arg1;
7695 	struct sbuf *sb;
7696 	int rc, i, n;
7697 	uint32_t lo, hi, used, alloc;
7698 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
7699 	static const char *region[] = {
7700 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
7701 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
7702 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
7703 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
7704 		"RQUDP region:", "PBL region:", "TXPBL region:",
7705 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
7706 		"On-chip queues:", "TLS keys:",
7707 	};
7708 	struct mem_desc avail[4];
7709 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
7710 	struct mem_desc *md = mem;
7711 
7712 	rc = sysctl_wire_old_buffer(req, 0);
7713 	if (rc != 0)
7714 		return (rc);
7715 
7716 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7717 	if (sb == NULL)
7718 		return (ENOMEM);
7719 
7720 	for (i = 0; i < nitems(mem); i++) {
7721 		mem[i].limit = 0;
7722 		mem[i].idx = i;
7723 	}
7724 
7725 	/* Find and sort the populated memory ranges */
7726 	i = 0;
7727 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
7728 	if (lo & F_EDRAM0_ENABLE) {
7729 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
7730 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
7731 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
7732 		avail[i].idx = 0;
7733 		i++;
7734 	}
7735 	if (lo & F_EDRAM1_ENABLE) {
7736 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
7737 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
7738 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
7739 		avail[i].idx = 1;
7740 		i++;
7741 	}
7742 	if (lo & F_EXT_MEM_ENABLE) {
7743 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
7744 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
7745 		avail[i].limit = avail[i].base +
7746 		    (G_EXT_MEM_SIZE(hi) << 20);
7747 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
7748 		i++;
7749 	}
7750 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
7751 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
7752 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
7753 		avail[i].limit = avail[i].base +
7754 		    (G_EXT_MEM1_SIZE(hi) << 20);
7755 		avail[i].idx = 4;
7756 		i++;
7757 	}
7758 	if (!i)                                    /* no memory available */
7759 		return 0;
7760 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
7761 
7762 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
7763 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
7764 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
7765 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
7766 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
7767 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
7768 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
7769 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
7770 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
7771 
7772 	/* the next few have explicit upper bounds */
7773 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
7774 	md->limit = md->base - 1 +
7775 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
7776 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
7777 	md++;
7778 
7779 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
7780 	md->limit = md->base - 1 +
7781 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
7782 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
7783 	md++;
7784 
7785 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7786 		if (chip_id(sc) <= CHELSIO_T5)
7787 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
7788 		else
7789 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
7790 		md->limit = 0;
7791 	} else {
7792 		md->base = 0;
7793 		md->idx = nitems(region);  /* hide it */
7794 	}
7795 	md++;
7796 
7797 #define ulp_region(reg) \
7798 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
7799 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
7800 
7801 	ulp_region(RX_ISCSI);
7802 	ulp_region(RX_TDDP);
7803 	ulp_region(TX_TPT);
7804 	ulp_region(RX_STAG);
7805 	ulp_region(RX_RQ);
7806 	ulp_region(RX_RQUDP);
7807 	ulp_region(RX_PBL);
7808 	ulp_region(TX_PBL);
7809 #undef ulp_region
7810 
7811 	md->base = 0;
7812 	md->idx = nitems(region);
7813 	if (!is_t4(sc)) {
7814 		uint32_t size = 0;
7815 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
7816 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
7817 
7818 		if (is_t5(sc)) {
7819 			if (sge_ctrl & F_VFIFO_ENABLE)
7820 				size = G_DBVFIFO_SIZE(fifo_size);
7821 		} else
7822 			size = G_T6_DBVFIFO_SIZE(fifo_size);
7823 
7824 		if (size) {
7825 			md->base = G_BASEADDR(t4_read_reg(sc,
7826 			    A_SGE_DBVFIFO_BADDR));
7827 			md->limit = md->base + (size << 2) - 1;
7828 		}
7829 	}
7830 	md++;
7831 
7832 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
7833 	md->limit = 0;
7834 	md++;
7835 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
7836 	md->limit = 0;
7837 	md++;
7838 
7839 	md->base = sc->vres.ocq.start;
7840 	if (sc->vres.ocq.size)
7841 		md->limit = md->base + sc->vres.ocq.size - 1;
7842 	else
7843 		md->idx = nitems(region);  /* hide it */
7844 	md++;
7845 
7846 	md->base = sc->vres.key.start;
7847 	if (sc->vres.key.size)
7848 		md->limit = md->base + sc->vres.key.size - 1;
7849 	else
7850 		md->idx = nitems(region);  /* hide it */
7851 	md++;
7852 
7853 	/* add any address-space holes, there can be up to 3 */
7854 	for (n = 0; n < i - 1; n++)
7855 		if (avail[n].limit < avail[n + 1].base)
7856 			(md++)->base = avail[n].limit;
7857 	if (avail[n].limit)
7858 		(md++)->base = avail[n].limit;
7859 
7860 	n = md - mem;
7861 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
7862 
7863 	for (lo = 0; lo < i; lo++)
7864 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
7865 				avail[lo].limit - 1);
7866 
7867 	sbuf_printf(sb, "\n");
7868 	for (i = 0; i < n; i++) {
7869 		if (mem[i].idx >= nitems(region))
7870 			continue;                        /* skip holes */
7871 		if (!mem[i].limit)
7872 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
7873 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
7874 				mem[i].limit);
7875 	}
7876 
7877 	sbuf_printf(sb, "\n");
7878 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
7879 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
7880 	mem_region_show(sb, "uP RAM:", lo, hi);
7881 
7882 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
7883 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
7884 	mem_region_show(sb, "uP Extmem2:", lo, hi);
7885 
7886 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
7887 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
7888 		   G_PMRXMAXPAGE(lo),
7889 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
7890 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
7891 
7892 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
7893 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
7894 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
7895 		   G_PMTXMAXPAGE(lo),
7896 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
7897 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
7898 	sbuf_printf(sb, "%u p-structs\n",
7899 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
7900 
7901 	for (i = 0; i < 4; i++) {
7902 		if (chip_id(sc) > CHELSIO_T5)
7903 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
7904 		else
7905 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
7906 		if (is_t5(sc)) {
7907 			used = G_T5_USED(lo);
7908 			alloc = G_T5_ALLOC(lo);
7909 		} else {
7910 			used = G_USED(lo);
7911 			alloc = G_ALLOC(lo);
7912 		}
7913 		/* For T6 these are MAC buffer groups */
7914 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
7915 		    i, used, alloc);
7916 	}
7917 	for (i = 0; i < sc->chip_params->nchan; i++) {
7918 		if (chip_id(sc) > CHELSIO_T5)
7919 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
7920 		else
7921 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
7922 		if (is_t5(sc)) {
7923 			used = G_T5_USED(lo);
7924 			alloc = G_T5_ALLOC(lo);
7925 		} else {
7926 			used = G_USED(lo);
7927 			alloc = G_ALLOC(lo);
7928 		}
7929 		/* For T6 these are MAC buffer groups */
7930 		sbuf_printf(sb,
7931 		    "\nLoopback %d using %u pages out of %u allocated",
7932 		    i, used, alloc);
7933 	}
7934 
7935 	rc = sbuf_finish(sb);
7936 	sbuf_delete(sb);
7937 
7938 	return (rc);
7939 }
7940 
7941 static inline void
7942 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
7943 {
7944 	*mask = x | y;
7945 	y = htobe64(y);
7946 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
7947 }
7948 
7949 static int
7950 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
7951 {
7952 	struct adapter *sc = arg1;
7953 	struct sbuf *sb;
7954 	int rc, i;
7955 
7956 	MPASS(chip_id(sc) <= CHELSIO_T5);
7957 
7958 	rc = sysctl_wire_old_buffer(req, 0);
7959 	if (rc != 0)
7960 		return (rc);
7961 
7962 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7963 	if (sb == NULL)
7964 		return (ENOMEM);
7965 
7966 	sbuf_printf(sb,
7967 	    "Idx  Ethernet address     Mask     Vld Ports PF"
7968 	    "  VF              Replication             P0 P1 P2 P3  ML");
7969 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7970 		uint64_t tcamx, tcamy, mask;
7971 		uint32_t cls_lo, cls_hi;
7972 		uint8_t addr[ETHER_ADDR_LEN];
7973 
7974 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
7975 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
7976 		if (tcamx & tcamy)
7977 			continue;
7978 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7979 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7980 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7981 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
7982 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
7983 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
7984 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
7985 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
7986 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
7987 
7988 		if (cls_lo & F_REPLICATE) {
7989 			struct fw_ldst_cmd ldst_cmd;
7990 
7991 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7992 			ldst_cmd.op_to_addrspace =
7993 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7994 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7995 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7996 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7997 			ldst_cmd.u.mps.rplc.fid_idx =
7998 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7999 				V_FW_LDST_CMD_IDX(i));
8000 
8001 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8002 			    "t4mps");
8003 			if (rc)
8004 				break;
8005 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
8006 			    sizeof(ldst_cmd), &ldst_cmd);
8007 			end_synchronized_op(sc, 0);
8008 
8009 			if (rc != 0) {
8010 				sbuf_printf(sb, "%36d", rc);
8011 				rc = 0;
8012 			} else {
8013 				sbuf_printf(sb, " %08x %08x %08x %08x",
8014 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
8015 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
8016 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
8017 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
8018 			}
8019 		} else
8020 			sbuf_printf(sb, "%36s", "");
8021 
8022 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
8023 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
8024 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
8025 	}
8026 
8027 	if (rc)
8028 		(void) sbuf_finish(sb);
8029 	else
8030 		rc = sbuf_finish(sb);
8031 	sbuf_delete(sb);
8032 
8033 	return (rc);
8034 }
8035 
8036 static int
8037 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
8038 {
8039 	struct adapter *sc = arg1;
8040 	struct sbuf *sb;
8041 	int rc, i;
8042 
8043 	MPASS(chip_id(sc) > CHELSIO_T5);
8044 
8045 	rc = sysctl_wire_old_buffer(req, 0);
8046 	if (rc != 0)
8047 		return (rc);
8048 
8049 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8050 	if (sb == NULL)
8051 		return (ENOMEM);
8052 
8053 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
8054 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
8055 	    "                           Replication"
8056 	    "                                    P0 P1 P2 P3  ML\n");
8057 
8058 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
8059 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
8060 		uint16_t ivlan;
8061 		uint64_t tcamx, tcamy, val, mask;
8062 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
8063 		uint8_t addr[ETHER_ADDR_LEN];
8064 
8065 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
8066 		if (i < 256)
8067 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
8068 		else
8069 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
8070 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8071 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8072 		tcamy = G_DMACH(val) << 32;
8073 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8074 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8075 		lookup_type = G_DATALKPTYPE(data2);
8076 		port_num = G_DATAPORTNUM(data2);
8077 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8078 			/* Inner header VNI */
8079 			vniy = ((data2 & F_DATAVIDH2) << 23) |
8080 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8081 			dip_hit = data2 & F_DATADIPHIT;
8082 			vlan_vld = 0;
8083 		} else {
8084 			vniy = 0;
8085 			dip_hit = 0;
8086 			vlan_vld = data2 & F_DATAVIDH2;
8087 			ivlan = G_VIDL(val);
8088 		}
8089 
8090 		ctl |= V_CTLXYBITSEL(1);
8091 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8092 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8093 		tcamx = G_DMACH(val) << 32;
8094 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8095 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8096 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8097 			/* Inner header VNI mask */
8098 			vnix = ((data2 & F_DATAVIDH2) << 23) |
8099 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8100 		} else
8101 			vnix = 0;
8102 
8103 		if (tcamx & tcamy)
8104 			continue;
8105 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
8106 
8107 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
8108 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
8109 
8110 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8111 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8112 			    "%012jx %06x %06x    -    -   %3c"
8113 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
8114 			    addr[1], addr[2], addr[3], addr[4], addr[5],
8115 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
8116 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8117 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8118 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8119 		} else {
8120 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8121 			    "%012jx    -       -   ", i, addr[0], addr[1],
8122 			    addr[2], addr[3], addr[4], addr[5],
8123 			    (uintmax_t)mask);
8124 
8125 			if (vlan_vld)
8126 				sbuf_printf(sb, "%4u   Y     ", ivlan);
8127 			else
8128 				sbuf_printf(sb, "  -    N     ");
8129 
8130 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
8131 			    lookup_type ? 'I' : 'O', port_num,
8132 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8133 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8134 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8135 		}
8136 
8137 
8138 		if (cls_lo & F_T6_REPLICATE) {
8139 			struct fw_ldst_cmd ldst_cmd;
8140 
8141 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
8142 			ldst_cmd.op_to_addrspace =
8143 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
8144 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
8145 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
8146 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
8147 			ldst_cmd.u.mps.rplc.fid_idx =
8148 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
8149 				V_FW_LDST_CMD_IDX(i));
8150 
8151 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8152 			    "t6mps");
8153 			if (rc)
8154 				break;
8155 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
8156 			    sizeof(ldst_cmd), &ldst_cmd);
8157 			end_synchronized_op(sc, 0);
8158 
8159 			if (rc != 0) {
8160 				sbuf_printf(sb, "%72d", rc);
8161 				rc = 0;
8162 			} else {
8163 				sbuf_printf(sb, " %08x %08x %08x %08x"
8164 				    " %08x %08x %08x %08x",
8165 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
8166 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
8167 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
8168 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
8169 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
8170 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
8171 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
8172 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
8173 			}
8174 		} else
8175 			sbuf_printf(sb, "%72s", "");
8176 
8177 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
8178 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
8179 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
8180 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
8181 	}
8182 
8183 	if (rc)
8184 		(void) sbuf_finish(sb);
8185 	else
8186 		rc = sbuf_finish(sb);
8187 	sbuf_delete(sb);
8188 
8189 	return (rc);
8190 }
8191 
8192 static int
8193 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
8194 {
8195 	struct adapter *sc = arg1;
8196 	struct sbuf *sb;
8197 	int rc;
8198 	uint16_t mtus[NMTUS];
8199 
8200 	rc = sysctl_wire_old_buffer(req, 0);
8201 	if (rc != 0)
8202 		return (rc);
8203 
8204 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8205 	if (sb == NULL)
8206 		return (ENOMEM);
8207 
8208 	t4_read_mtu_tbl(sc, mtus, NULL);
8209 
8210 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
8211 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
8212 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
8213 	    mtus[14], mtus[15]);
8214 
8215 	rc = sbuf_finish(sb);
8216 	sbuf_delete(sb);
8217 
8218 	return (rc);
8219 }
8220 
8221 static int
8222 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
8223 {
8224 	struct adapter *sc = arg1;
8225 	struct sbuf *sb;
8226 	int rc, i;
8227 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
8228 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
8229 	static const char *tx_stats[MAX_PM_NSTATS] = {
8230 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
8231 		"Tx FIFO wait", NULL, "Tx latency"
8232 	};
8233 	static const char *rx_stats[MAX_PM_NSTATS] = {
8234 		"Read:", "Write bypass:", "Write mem:", "Flush:",
8235 		"Rx FIFO wait", NULL, "Rx latency"
8236 	};
8237 
8238 	rc = sysctl_wire_old_buffer(req, 0);
8239 	if (rc != 0)
8240 		return (rc);
8241 
8242 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8243 	if (sb == NULL)
8244 		return (ENOMEM);
8245 
8246 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
8247 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
8248 
8249 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
8250 	for (i = 0; i < 4; i++) {
8251 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8252 		    tx_cyc[i]);
8253 	}
8254 
8255 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
8256 	for (i = 0; i < 4; i++) {
8257 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8258 		    rx_cyc[i]);
8259 	}
8260 
8261 	if (chip_id(sc) > CHELSIO_T5) {
8262 		sbuf_printf(sb,
8263 		    "\n              Total wait      Total occupancy");
8264 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8265 		    tx_cyc[i]);
8266 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8267 		    rx_cyc[i]);
8268 
8269 		i += 2;
8270 		MPASS(i < nitems(tx_stats));
8271 
8272 		sbuf_printf(sb,
8273 		    "\n                   Reads           Total wait");
8274 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8275 		    tx_cyc[i]);
8276 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8277 		    rx_cyc[i]);
8278 	}
8279 
8280 	rc = sbuf_finish(sb);
8281 	sbuf_delete(sb);
8282 
8283 	return (rc);
8284 }
8285 
8286 static int
8287 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
8288 {
8289 	struct adapter *sc = arg1;
8290 	struct sbuf *sb;
8291 	int rc;
8292 	struct tp_rdma_stats stats;
8293 
8294 	rc = sysctl_wire_old_buffer(req, 0);
8295 	if (rc != 0)
8296 		return (rc);
8297 
8298 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8299 	if (sb == NULL)
8300 		return (ENOMEM);
8301 
8302 	mtx_lock(&sc->reg_lock);
8303 	t4_tp_get_rdma_stats(sc, &stats, 0);
8304 	mtx_unlock(&sc->reg_lock);
8305 
8306 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
8307 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
8308 
8309 	rc = sbuf_finish(sb);
8310 	sbuf_delete(sb);
8311 
8312 	return (rc);
8313 }
8314 
8315 static int
8316 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
8317 {
8318 	struct adapter *sc = arg1;
8319 	struct sbuf *sb;
8320 	int rc;
8321 	struct tp_tcp_stats v4, v6;
8322 
8323 	rc = sysctl_wire_old_buffer(req, 0);
8324 	if (rc != 0)
8325 		return (rc);
8326 
8327 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8328 	if (sb == NULL)
8329 		return (ENOMEM);
8330 
8331 	mtx_lock(&sc->reg_lock);
8332 	t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
8333 	mtx_unlock(&sc->reg_lock);
8334 
8335 	sbuf_printf(sb,
8336 	    "                                IP                 IPv6\n");
8337 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
8338 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
8339 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
8340 	    v4.tcp_in_segs, v6.tcp_in_segs);
8341 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
8342 	    v4.tcp_out_segs, v6.tcp_out_segs);
8343 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
8344 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
8345 
8346 	rc = sbuf_finish(sb);
8347 	sbuf_delete(sb);
8348 
8349 	return (rc);
8350 }
8351 
8352 static int
8353 sysctl_tids(SYSCTL_HANDLER_ARGS)
8354 {
8355 	struct adapter *sc = arg1;
8356 	struct sbuf *sb;
8357 	int rc;
8358 	struct tid_info *t = &sc->tids;
8359 
8360 	rc = sysctl_wire_old_buffer(req, 0);
8361 	if (rc != 0)
8362 		return (rc);
8363 
8364 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8365 	if (sb == NULL)
8366 		return (ENOMEM);
8367 
8368 	if (t->natids) {
8369 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
8370 		    t->atids_in_use);
8371 	}
8372 
8373 	if (t->nhpftids) {
8374 		sbuf_printf(sb, "HPFTID range: %u-%u, in use: %u\n",
8375 		    t->hpftid_base, t->hpftid_end, t->hpftids_in_use);
8376 	}
8377 
8378 	if (t->ntids) {
8379 		sbuf_printf(sb, "TID range: ");
8380 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
8381 			uint32_t b, hb;
8382 
8383 			if (chip_id(sc) <= CHELSIO_T5) {
8384 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
8385 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
8386 			} else {
8387 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
8388 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
8389 			}
8390 
8391 			if (b)
8392 				sbuf_printf(sb, "%u-%u, ", t->tid_base, b - 1);
8393 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
8394 		} else
8395 			sbuf_printf(sb, "%u-%u", t->tid_base, t->ntids - 1);
8396 		sbuf_printf(sb, ", in use: %u\n",
8397 		    atomic_load_acq_int(&t->tids_in_use));
8398 	}
8399 
8400 	if (t->nstids) {
8401 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
8402 		    t->stid_base + t->nstids - 1, t->stids_in_use);
8403 	}
8404 
8405 	if (t->nftids) {
8406 		sbuf_printf(sb, "FTID range: %u-%u, in use: %u\n", t->ftid_base,
8407 		    t->ftid_end, t->ftids_in_use);
8408 	}
8409 
8410 	if (t->netids) {
8411 		sbuf_printf(sb, "ETID range: %u-%u, in use: %u\n", t->etid_base,
8412 		    t->etid_base + t->netids - 1, t->etids_in_use);
8413 	}
8414 
8415 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
8416 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
8417 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
8418 
8419 	rc = sbuf_finish(sb);
8420 	sbuf_delete(sb);
8421 
8422 	return (rc);
8423 }
8424 
8425 static int
8426 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
8427 {
8428 	struct adapter *sc = arg1;
8429 	struct sbuf *sb;
8430 	int rc;
8431 	struct tp_err_stats stats;
8432 
8433 	rc = sysctl_wire_old_buffer(req, 0);
8434 	if (rc != 0)
8435 		return (rc);
8436 
8437 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8438 	if (sb == NULL)
8439 		return (ENOMEM);
8440 
8441 	mtx_lock(&sc->reg_lock);
8442 	t4_tp_get_err_stats(sc, &stats, 0);
8443 	mtx_unlock(&sc->reg_lock);
8444 
8445 	if (sc->chip_params->nchan > 2) {
8446 		sbuf_printf(sb, "                 channel 0  channel 1"
8447 		    "  channel 2  channel 3\n");
8448 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
8449 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
8450 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
8451 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
8452 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
8453 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
8454 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
8455 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
8456 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
8457 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
8458 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
8459 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
8460 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
8461 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
8462 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
8463 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
8464 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
8465 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
8466 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
8467 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
8468 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
8469 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
8470 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
8471 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
8472 	} else {
8473 		sbuf_printf(sb, "                 channel 0  channel 1\n");
8474 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
8475 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
8476 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
8477 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
8478 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
8479 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
8480 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
8481 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
8482 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
8483 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
8484 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
8485 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
8486 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
8487 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
8488 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
8489 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
8490 	}
8491 
8492 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
8493 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
8494 
8495 	rc = sbuf_finish(sb);
8496 	sbuf_delete(sb);
8497 
8498 	return (rc);
8499 }
8500 
8501 static int
8502 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
8503 {
8504 	struct adapter *sc = arg1;
8505 	struct tp_params *tpp = &sc->params.tp;
8506 	u_int mask;
8507 	int rc;
8508 
8509 	mask = tpp->la_mask >> 16;
8510 	rc = sysctl_handle_int(oidp, &mask, 0, req);
8511 	if (rc != 0 || req->newptr == NULL)
8512 		return (rc);
8513 	if (mask > 0xffff)
8514 		return (EINVAL);
8515 	tpp->la_mask = mask << 16;
8516 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
8517 
8518 	return (0);
8519 }
8520 
8521 struct field_desc {
8522 	const char *name;
8523 	u_int start;
8524 	u_int width;
8525 };
8526 
8527 static void
8528 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
8529 {
8530 	char buf[32];
8531 	int line_size = 0;
8532 
8533 	while (f->name) {
8534 		uint64_t mask = (1ULL << f->width) - 1;
8535 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
8536 		    ((uintmax_t)v >> f->start) & mask);
8537 
8538 		if (line_size + len >= 79) {
8539 			line_size = 8;
8540 			sbuf_printf(sb, "\n        ");
8541 		}
8542 		sbuf_printf(sb, "%s ", buf);
8543 		line_size += len + 1;
8544 		f++;
8545 	}
8546 	sbuf_printf(sb, "\n");
8547 }
8548 
8549 static const struct field_desc tp_la0[] = {
8550 	{ "RcfOpCodeOut", 60, 4 },
8551 	{ "State", 56, 4 },
8552 	{ "WcfState", 52, 4 },
8553 	{ "RcfOpcSrcOut", 50, 2 },
8554 	{ "CRxError", 49, 1 },
8555 	{ "ERxError", 48, 1 },
8556 	{ "SanityFailed", 47, 1 },
8557 	{ "SpuriousMsg", 46, 1 },
8558 	{ "FlushInputMsg", 45, 1 },
8559 	{ "FlushInputCpl", 44, 1 },
8560 	{ "RssUpBit", 43, 1 },
8561 	{ "RssFilterHit", 42, 1 },
8562 	{ "Tid", 32, 10 },
8563 	{ "InitTcb", 31, 1 },
8564 	{ "LineNumber", 24, 7 },
8565 	{ "Emsg", 23, 1 },
8566 	{ "EdataOut", 22, 1 },
8567 	{ "Cmsg", 21, 1 },
8568 	{ "CdataOut", 20, 1 },
8569 	{ "EreadPdu", 19, 1 },
8570 	{ "CreadPdu", 18, 1 },
8571 	{ "TunnelPkt", 17, 1 },
8572 	{ "RcfPeerFin", 16, 1 },
8573 	{ "RcfReasonOut", 12, 4 },
8574 	{ "TxCchannel", 10, 2 },
8575 	{ "RcfTxChannel", 8, 2 },
8576 	{ "RxEchannel", 6, 2 },
8577 	{ "RcfRxChannel", 5, 1 },
8578 	{ "RcfDataOutSrdy", 4, 1 },
8579 	{ "RxDvld", 3, 1 },
8580 	{ "RxOoDvld", 2, 1 },
8581 	{ "RxCongestion", 1, 1 },
8582 	{ "TxCongestion", 0, 1 },
8583 	{ NULL }
8584 };
8585 
8586 static const struct field_desc tp_la1[] = {
8587 	{ "CplCmdIn", 56, 8 },
8588 	{ "CplCmdOut", 48, 8 },
8589 	{ "ESynOut", 47, 1 },
8590 	{ "EAckOut", 46, 1 },
8591 	{ "EFinOut", 45, 1 },
8592 	{ "ERstOut", 44, 1 },
8593 	{ "SynIn", 43, 1 },
8594 	{ "AckIn", 42, 1 },
8595 	{ "FinIn", 41, 1 },
8596 	{ "RstIn", 40, 1 },
8597 	{ "DataIn", 39, 1 },
8598 	{ "DataInVld", 38, 1 },
8599 	{ "PadIn", 37, 1 },
8600 	{ "RxBufEmpty", 36, 1 },
8601 	{ "RxDdp", 35, 1 },
8602 	{ "RxFbCongestion", 34, 1 },
8603 	{ "TxFbCongestion", 33, 1 },
8604 	{ "TxPktSumSrdy", 32, 1 },
8605 	{ "RcfUlpType", 28, 4 },
8606 	{ "Eread", 27, 1 },
8607 	{ "Ebypass", 26, 1 },
8608 	{ "Esave", 25, 1 },
8609 	{ "Static0", 24, 1 },
8610 	{ "Cread", 23, 1 },
8611 	{ "Cbypass", 22, 1 },
8612 	{ "Csave", 21, 1 },
8613 	{ "CPktOut", 20, 1 },
8614 	{ "RxPagePoolFull", 18, 2 },
8615 	{ "RxLpbkPkt", 17, 1 },
8616 	{ "TxLpbkPkt", 16, 1 },
8617 	{ "RxVfValid", 15, 1 },
8618 	{ "SynLearned", 14, 1 },
8619 	{ "SetDelEntry", 13, 1 },
8620 	{ "SetInvEntry", 12, 1 },
8621 	{ "CpcmdDvld", 11, 1 },
8622 	{ "CpcmdSave", 10, 1 },
8623 	{ "RxPstructsFull", 8, 2 },
8624 	{ "EpcmdDvld", 7, 1 },
8625 	{ "EpcmdFlush", 6, 1 },
8626 	{ "EpcmdTrimPrefix", 5, 1 },
8627 	{ "EpcmdTrimPostfix", 4, 1 },
8628 	{ "ERssIp4Pkt", 3, 1 },
8629 	{ "ERssIp6Pkt", 2, 1 },
8630 	{ "ERssTcpUdpPkt", 1, 1 },
8631 	{ "ERssFceFipPkt", 0, 1 },
8632 	{ NULL }
8633 };
8634 
8635 static const struct field_desc tp_la2[] = {
8636 	{ "CplCmdIn", 56, 8 },
8637 	{ "MpsVfVld", 55, 1 },
8638 	{ "MpsPf", 52, 3 },
8639 	{ "MpsVf", 44, 8 },
8640 	{ "SynIn", 43, 1 },
8641 	{ "AckIn", 42, 1 },
8642 	{ "FinIn", 41, 1 },
8643 	{ "RstIn", 40, 1 },
8644 	{ "DataIn", 39, 1 },
8645 	{ "DataInVld", 38, 1 },
8646 	{ "PadIn", 37, 1 },
8647 	{ "RxBufEmpty", 36, 1 },
8648 	{ "RxDdp", 35, 1 },
8649 	{ "RxFbCongestion", 34, 1 },
8650 	{ "TxFbCongestion", 33, 1 },
8651 	{ "TxPktSumSrdy", 32, 1 },
8652 	{ "RcfUlpType", 28, 4 },
8653 	{ "Eread", 27, 1 },
8654 	{ "Ebypass", 26, 1 },
8655 	{ "Esave", 25, 1 },
8656 	{ "Static0", 24, 1 },
8657 	{ "Cread", 23, 1 },
8658 	{ "Cbypass", 22, 1 },
8659 	{ "Csave", 21, 1 },
8660 	{ "CPktOut", 20, 1 },
8661 	{ "RxPagePoolFull", 18, 2 },
8662 	{ "RxLpbkPkt", 17, 1 },
8663 	{ "TxLpbkPkt", 16, 1 },
8664 	{ "RxVfValid", 15, 1 },
8665 	{ "SynLearned", 14, 1 },
8666 	{ "SetDelEntry", 13, 1 },
8667 	{ "SetInvEntry", 12, 1 },
8668 	{ "CpcmdDvld", 11, 1 },
8669 	{ "CpcmdSave", 10, 1 },
8670 	{ "RxPstructsFull", 8, 2 },
8671 	{ "EpcmdDvld", 7, 1 },
8672 	{ "EpcmdFlush", 6, 1 },
8673 	{ "EpcmdTrimPrefix", 5, 1 },
8674 	{ "EpcmdTrimPostfix", 4, 1 },
8675 	{ "ERssIp4Pkt", 3, 1 },
8676 	{ "ERssIp6Pkt", 2, 1 },
8677 	{ "ERssTcpUdpPkt", 1, 1 },
8678 	{ "ERssFceFipPkt", 0, 1 },
8679 	{ NULL }
8680 };
8681 
8682 static void
8683 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
8684 {
8685 
8686 	field_desc_show(sb, *p, tp_la0);
8687 }
8688 
8689 static void
8690 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
8691 {
8692 
8693 	if (idx)
8694 		sbuf_printf(sb, "\n");
8695 	field_desc_show(sb, p[0], tp_la0);
8696 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8697 		field_desc_show(sb, p[1], tp_la0);
8698 }
8699 
8700 static void
8701 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
8702 {
8703 
8704 	if (idx)
8705 		sbuf_printf(sb, "\n");
8706 	field_desc_show(sb, p[0], tp_la0);
8707 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8708 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
8709 }
8710 
8711 static int
8712 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
8713 {
8714 	struct adapter *sc = arg1;
8715 	struct sbuf *sb;
8716 	uint64_t *buf, *p;
8717 	int rc;
8718 	u_int i, inc;
8719 	void (*show_func)(struct sbuf *, uint64_t *, int);
8720 
8721 	rc = sysctl_wire_old_buffer(req, 0);
8722 	if (rc != 0)
8723 		return (rc);
8724 
8725 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8726 	if (sb == NULL)
8727 		return (ENOMEM);
8728 
8729 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
8730 
8731 	t4_tp_read_la(sc, buf, NULL);
8732 	p = buf;
8733 
8734 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
8735 	case 2:
8736 		inc = 2;
8737 		show_func = tp_la_show2;
8738 		break;
8739 	case 3:
8740 		inc = 2;
8741 		show_func = tp_la_show3;
8742 		break;
8743 	default:
8744 		inc = 1;
8745 		show_func = tp_la_show;
8746 	}
8747 
8748 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
8749 		(*show_func)(sb, p, i);
8750 
8751 	rc = sbuf_finish(sb);
8752 	sbuf_delete(sb);
8753 	free(buf, M_CXGBE);
8754 	return (rc);
8755 }
8756 
8757 static int
8758 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
8759 {
8760 	struct adapter *sc = arg1;
8761 	struct sbuf *sb;
8762 	int rc;
8763 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
8764 
8765 	rc = sysctl_wire_old_buffer(req, 0);
8766 	if (rc != 0)
8767 		return (rc);
8768 
8769 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8770 	if (sb == NULL)
8771 		return (ENOMEM);
8772 
8773 	t4_get_chan_txrate(sc, nrate, orate);
8774 
8775 	if (sc->chip_params->nchan > 2) {
8776 		sbuf_printf(sb, "              channel 0   channel 1"
8777 		    "   channel 2   channel 3\n");
8778 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
8779 		    nrate[0], nrate[1], nrate[2], nrate[3]);
8780 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
8781 		    orate[0], orate[1], orate[2], orate[3]);
8782 	} else {
8783 		sbuf_printf(sb, "              channel 0   channel 1\n");
8784 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
8785 		    nrate[0], nrate[1]);
8786 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
8787 		    orate[0], orate[1]);
8788 	}
8789 
8790 	rc = sbuf_finish(sb);
8791 	sbuf_delete(sb);
8792 
8793 	return (rc);
8794 }
8795 
8796 static int
8797 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
8798 {
8799 	struct adapter *sc = arg1;
8800 	struct sbuf *sb;
8801 	uint32_t *buf, *p;
8802 	int rc, i;
8803 
8804 	rc = sysctl_wire_old_buffer(req, 0);
8805 	if (rc != 0)
8806 		return (rc);
8807 
8808 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8809 	if (sb == NULL)
8810 		return (ENOMEM);
8811 
8812 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
8813 	    M_ZERO | M_WAITOK);
8814 
8815 	t4_ulprx_read_la(sc, buf);
8816 	p = buf;
8817 
8818 	sbuf_printf(sb, "      Pcmd        Type   Message"
8819 	    "                Data");
8820 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
8821 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
8822 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
8823 	}
8824 
8825 	rc = sbuf_finish(sb);
8826 	sbuf_delete(sb);
8827 	free(buf, M_CXGBE);
8828 	return (rc);
8829 }
8830 
8831 static int
8832 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
8833 {
8834 	struct adapter *sc = arg1;
8835 	struct sbuf *sb;
8836 	int rc, v;
8837 
8838 	MPASS(chip_id(sc) >= CHELSIO_T5);
8839 
8840 	rc = sysctl_wire_old_buffer(req, 0);
8841 	if (rc != 0)
8842 		return (rc);
8843 
8844 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8845 	if (sb == NULL)
8846 		return (ENOMEM);
8847 
8848 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
8849 	if (G_STATSOURCE_T5(v) == 7) {
8850 		int mode;
8851 
8852 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
8853 		if (mode == 0) {
8854 			sbuf_printf(sb, "total %d, incomplete %d",
8855 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8856 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8857 		} else if (mode == 1) {
8858 			sbuf_printf(sb, "total %d, data overflow %d",
8859 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8860 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8861 		} else {
8862 			sbuf_printf(sb, "unknown mode %d", mode);
8863 		}
8864 	}
8865 	rc = sbuf_finish(sb);
8866 	sbuf_delete(sb);
8867 
8868 	return (rc);
8869 }
8870 
8871 static int
8872 sysctl_cpus(SYSCTL_HANDLER_ARGS)
8873 {
8874 	struct adapter *sc = arg1;
8875 	enum cpu_sets op = arg2;
8876 	cpuset_t cpuset;
8877 	struct sbuf *sb;
8878 	int i, rc;
8879 
8880 	MPASS(op == LOCAL_CPUS || op == INTR_CPUS);
8881 
8882 	CPU_ZERO(&cpuset);
8883 	rc = bus_get_cpus(sc->dev, op, sizeof(cpuset), &cpuset);
8884 	if (rc != 0)
8885 		return (rc);
8886 
8887 	rc = sysctl_wire_old_buffer(req, 0);
8888 	if (rc != 0)
8889 		return (rc);
8890 
8891 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8892 	if (sb == NULL)
8893 		return (ENOMEM);
8894 
8895 	CPU_FOREACH(i)
8896 		sbuf_printf(sb, "%d ", i);
8897 	rc = sbuf_finish(sb);
8898 	sbuf_delete(sb);
8899 
8900 	return (rc);
8901 }
8902 
8903 #ifdef TCP_OFFLOAD
8904 static int
8905 sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS)
8906 {
8907 	struct adapter *sc = arg1;
8908 	int *old_ports, *new_ports;
8909 	int i, new_count, rc;
8910 
8911 	if (req->newptr == NULL && req->oldptr == NULL)
8912 		return (SYSCTL_OUT(req, NULL, imax(sc->tt.num_tls_rx_ports, 1) *
8913 		    sizeof(sc->tt.tls_rx_ports[0])));
8914 
8915 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tlsrx");
8916 	if (rc)
8917 		return (rc);
8918 
8919 	if (sc->tt.num_tls_rx_ports == 0) {
8920 		i = -1;
8921 		rc = SYSCTL_OUT(req, &i, sizeof(i));
8922 	} else
8923 		rc = SYSCTL_OUT(req, sc->tt.tls_rx_ports,
8924 		    sc->tt.num_tls_rx_ports * sizeof(sc->tt.tls_rx_ports[0]));
8925 	if (rc == 0 && req->newptr != NULL) {
8926 		new_count = req->newlen / sizeof(new_ports[0]);
8927 		new_ports = malloc(new_count * sizeof(new_ports[0]), M_CXGBE,
8928 		    M_WAITOK);
8929 		rc = SYSCTL_IN(req, new_ports, new_count *
8930 		    sizeof(new_ports[0]));
8931 		if (rc)
8932 			goto err;
8933 
8934 		/* Allow setting to a single '-1' to clear the list. */
8935 		if (new_count == 1 && new_ports[0] == -1) {
8936 			ADAPTER_LOCK(sc);
8937 			old_ports = sc->tt.tls_rx_ports;
8938 			sc->tt.tls_rx_ports = NULL;
8939 			sc->tt.num_tls_rx_ports = 0;
8940 			ADAPTER_UNLOCK(sc);
8941 			free(old_ports, M_CXGBE);
8942 		} else {
8943 			for (i = 0; i < new_count; i++) {
8944 				if (new_ports[i] < 1 ||
8945 				    new_ports[i] > IPPORT_MAX) {
8946 					rc = EINVAL;
8947 					goto err;
8948 				}
8949 			}
8950 
8951 			ADAPTER_LOCK(sc);
8952 			old_ports = sc->tt.tls_rx_ports;
8953 			sc->tt.tls_rx_ports = new_ports;
8954 			sc->tt.num_tls_rx_ports = new_count;
8955 			ADAPTER_UNLOCK(sc);
8956 			free(old_ports, M_CXGBE);
8957 			new_ports = NULL;
8958 		}
8959 	err:
8960 		free(new_ports, M_CXGBE);
8961 	}
8962 	end_synchronized_op(sc, 0);
8963 	return (rc);
8964 }
8965 
8966 static void
8967 unit_conv(char *buf, size_t len, u_int val, u_int factor)
8968 {
8969 	u_int rem = val % factor;
8970 
8971 	if (rem == 0)
8972 		snprintf(buf, len, "%u", val / factor);
8973 	else {
8974 		while (rem % 10 == 0)
8975 			rem /= 10;
8976 		snprintf(buf, len, "%u.%u", val / factor, rem);
8977 	}
8978 }
8979 
8980 static int
8981 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
8982 {
8983 	struct adapter *sc = arg1;
8984 	char buf[16];
8985 	u_int res, re;
8986 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8987 
8988 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8989 	switch (arg2) {
8990 	case 0:
8991 		/* timer_tick */
8992 		re = G_TIMERRESOLUTION(res);
8993 		break;
8994 	case 1:
8995 		/* TCP timestamp tick */
8996 		re = G_TIMESTAMPRESOLUTION(res);
8997 		break;
8998 	case 2:
8999 		/* DACK tick */
9000 		re = G_DELAYEDACKRESOLUTION(res);
9001 		break;
9002 	default:
9003 		return (EDOOFUS);
9004 	}
9005 
9006 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
9007 
9008 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
9009 }
9010 
9011 static int
9012 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
9013 {
9014 	struct adapter *sc = arg1;
9015 	u_int res, dack_re, v;
9016 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9017 
9018 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
9019 	dack_re = G_DELAYEDACKRESOLUTION(res);
9020 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
9021 
9022 	return (sysctl_handle_int(oidp, &v, 0, req));
9023 }
9024 
9025 static int
9026 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
9027 {
9028 	struct adapter *sc = arg1;
9029 	int reg = arg2;
9030 	u_int tre;
9031 	u_long tp_tick_us, v;
9032 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9033 
9034 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
9035 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
9036 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
9037 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
9038 
9039 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
9040 	tp_tick_us = (cclk_ps << tre) / 1000000;
9041 
9042 	if (reg == A_TP_INIT_SRTT)
9043 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
9044 	else
9045 		v = tp_tick_us * t4_read_reg(sc, reg);
9046 
9047 	return (sysctl_handle_long(oidp, &v, 0, req));
9048 }
9049 
9050 /*
9051  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
9052  * passed to this function.
9053  */
9054 static int
9055 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
9056 {
9057 	struct adapter *sc = arg1;
9058 	int idx = arg2;
9059 	u_int v;
9060 
9061 	MPASS(idx >= 0 && idx <= 24);
9062 
9063 	v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
9064 
9065 	return (sysctl_handle_int(oidp, &v, 0, req));
9066 }
9067 
9068 static int
9069 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
9070 {
9071 	struct adapter *sc = arg1;
9072 	int idx = arg2;
9073 	u_int shift, v, r;
9074 
9075 	MPASS(idx >= 0 && idx < 16);
9076 
9077 	r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
9078 	shift = (idx & 3) << 3;
9079 	v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
9080 
9081 	return (sysctl_handle_int(oidp, &v, 0, req));
9082 }
9083 
9084 static int
9085 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
9086 {
9087 	struct vi_info *vi = arg1;
9088 	struct adapter *sc = vi->pi->adapter;
9089 	int idx, rc, i;
9090 	struct sge_ofld_rxq *ofld_rxq;
9091 	uint8_t v;
9092 
9093 	idx = vi->ofld_tmr_idx;
9094 
9095 	rc = sysctl_handle_int(oidp, &idx, 0, req);
9096 	if (rc != 0 || req->newptr == NULL)
9097 		return (rc);
9098 
9099 	if (idx < 0 || idx >= SGE_NTIMERS)
9100 		return (EINVAL);
9101 
9102 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9103 	    "t4otmr");
9104 	if (rc)
9105 		return (rc);
9106 
9107 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
9108 	for_each_ofld_rxq(vi, i, ofld_rxq) {
9109 #ifdef atomic_store_rel_8
9110 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
9111 #else
9112 		ofld_rxq->iq.intr_params = v;
9113 #endif
9114 	}
9115 	vi->ofld_tmr_idx = idx;
9116 
9117 	end_synchronized_op(sc, LOCK_HELD);
9118 	return (0);
9119 }
9120 
9121 static int
9122 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
9123 {
9124 	struct vi_info *vi = arg1;
9125 	struct adapter *sc = vi->pi->adapter;
9126 	int idx, rc;
9127 
9128 	idx = vi->ofld_pktc_idx;
9129 
9130 	rc = sysctl_handle_int(oidp, &idx, 0, req);
9131 	if (rc != 0 || req->newptr == NULL)
9132 		return (rc);
9133 
9134 	if (idx < -1 || idx >= SGE_NCOUNTERS)
9135 		return (EINVAL);
9136 
9137 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9138 	    "t4opktc");
9139 	if (rc)
9140 		return (rc);
9141 
9142 	if (vi->flags & VI_INIT_DONE)
9143 		rc = EBUSY; /* cannot be changed once the queues are created */
9144 	else
9145 		vi->ofld_pktc_idx = idx;
9146 
9147 	end_synchronized_op(sc, LOCK_HELD);
9148 	return (rc);
9149 }
9150 #endif
9151 
9152 static int
9153 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
9154 {
9155 	int rc;
9156 
9157 	if (cntxt->cid > M_CTXTQID)
9158 		return (EINVAL);
9159 
9160 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
9161 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
9162 		return (EINVAL);
9163 
9164 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
9165 	if (rc)
9166 		return (rc);
9167 
9168 	if (sc->flags & FW_OK) {
9169 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
9170 		    &cntxt->data[0]);
9171 		if (rc == 0)
9172 			goto done;
9173 	}
9174 
9175 	/*
9176 	 * Read via firmware failed or wasn't even attempted.  Read directly via
9177 	 * the backdoor.
9178 	 */
9179 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
9180 done:
9181 	end_synchronized_op(sc, 0);
9182 	return (rc);
9183 }
9184 
9185 static int
9186 load_fw(struct adapter *sc, struct t4_data *fw)
9187 {
9188 	int rc;
9189 	uint8_t *fw_data;
9190 
9191 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
9192 	if (rc)
9193 		return (rc);
9194 
9195 	/*
9196 	 * The firmware, with the sole exception of the memory parity error
9197 	 * handler, runs from memory and not flash.  It is almost always safe to
9198 	 * install a new firmware on a running system.  Just set bit 1 in
9199 	 * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
9200 	 */
9201 	if (sc->flags & FULL_INIT_DONE &&
9202 	    (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
9203 		rc = EBUSY;
9204 		goto done;
9205 	}
9206 
9207 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
9208 	if (fw_data == NULL) {
9209 		rc = ENOMEM;
9210 		goto done;
9211 	}
9212 
9213 	rc = copyin(fw->data, fw_data, fw->len);
9214 	if (rc == 0)
9215 		rc = -t4_load_fw(sc, fw_data, fw->len);
9216 
9217 	free(fw_data, M_CXGBE);
9218 done:
9219 	end_synchronized_op(sc, 0);
9220 	return (rc);
9221 }
9222 
9223 static int
9224 load_cfg(struct adapter *sc, struct t4_data *cfg)
9225 {
9226 	int rc;
9227 	uint8_t *cfg_data = NULL;
9228 
9229 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9230 	if (rc)
9231 		return (rc);
9232 
9233 	if (cfg->len == 0) {
9234 		/* clear */
9235 		rc = -t4_load_cfg(sc, NULL, 0);
9236 		goto done;
9237 	}
9238 
9239 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
9240 	if (cfg_data == NULL) {
9241 		rc = ENOMEM;
9242 		goto done;
9243 	}
9244 
9245 	rc = copyin(cfg->data, cfg_data, cfg->len);
9246 	if (rc == 0)
9247 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
9248 
9249 	free(cfg_data, M_CXGBE);
9250 done:
9251 	end_synchronized_op(sc, 0);
9252 	return (rc);
9253 }
9254 
9255 static int
9256 load_boot(struct adapter *sc, struct t4_bootrom *br)
9257 {
9258 	int rc;
9259 	uint8_t *br_data = NULL;
9260 	u_int offset;
9261 
9262 	if (br->len > 1024 * 1024)
9263 		return (EFBIG);
9264 
9265 	if (br->pf_offset == 0) {
9266 		/* pfidx */
9267 		if (br->pfidx_addr > 7)
9268 			return (EINVAL);
9269 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
9270 		    A_PCIE_PF_EXPROM_OFST)));
9271 	} else if (br->pf_offset == 1) {
9272 		/* offset */
9273 		offset = G_OFFSET(br->pfidx_addr);
9274 	} else {
9275 		return (EINVAL);
9276 	}
9277 
9278 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
9279 	if (rc)
9280 		return (rc);
9281 
9282 	if (br->len == 0) {
9283 		/* clear */
9284 		rc = -t4_load_boot(sc, NULL, offset, 0);
9285 		goto done;
9286 	}
9287 
9288 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
9289 	if (br_data == NULL) {
9290 		rc = ENOMEM;
9291 		goto done;
9292 	}
9293 
9294 	rc = copyin(br->data, br_data, br->len);
9295 	if (rc == 0)
9296 		rc = -t4_load_boot(sc, br_data, offset, br->len);
9297 
9298 	free(br_data, M_CXGBE);
9299 done:
9300 	end_synchronized_op(sc, 0);
9301 	return (rc);
9302 }
9303 
9304 static int
9305 load_bootcfg(struct adapter *sc, struct t4_data *bc)
9306 {
9307 	int rc;
9308 	uint8_t *bc_data = NULL;
9309 
9310 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9311 	if (rc)
9312 		return (rc);
9313 
9314 	if (bc->len == 0) {
9315 		/* clear */
9316 		rc = -t4_load_bootcfg(sc, NULL, 0);
9317 		goto done;
9318 	}
9319 
9320 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
9321 	if (bc_data == NULL) {
9322 		rc = ENOMEM;
9323 		goto done;
9324 	}
9325 
9326 	rc = copyin(bc->data, bc_data, bc->len);
9327 	if (rc == 0)
9328 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
9329 
9330 	free(bc_data, M_CXGBE);
9331 done:
9332 	end_synchronized_op(sc, 0);
9333 	return (rc);
9334 }
9335 
9336 static int
9337 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
9338 {
9339 	int rc;
9340 	struct cudbg_init *cudbg;
9341 	void *handle, *buf;
9342 
9343 	/* buf is large, don't block if no memory is available */
9344 	buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
9345 	if (buf == NULL)
9346 		return (ENOMEM);
9347 
9348 	handle = cudbg_alloc_handle();
9349 	if (handle == NULL) {
9350 		rc = ENOMEM;
9351 		goto done;
9352 	}
9353 
9354 	cudbg = cudbg_get_init(handle);
9355 	cudbg->adap = sc;
9356 	cudbg->print = (cudbg_print_cb)printf;
9357 
9358 #ifndef notyet
9359 	device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
9360 	    __func__, dump->wr_flash, dump->len, dump->data);
9361 #endif
9362 
9363 	if (dump->wr_flash)
9364 		cudbg->use_flash = 1;
9365 	MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
9366 	memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
9367 
9368 	rc = cudbg_collect(handle, buf, &dump->len);
9369 	if (rc != 0)
9370 		goto done;
9371 
9372 	rc = copyout(buf, dump->data, dump->len);
9373 done:
9374 	cudbg_free_handle(handle);
9375 	free(buf, M_CXGBE);
9376 	return (rc);
9377 }
9378 
9379 static void
9380 free_offload_policy(struct t4_offload_policy *op)
9381 {
9382 	struct offload_rule *r;
9383 	int i;
9384 
9385 	if (op == NULL)
9386 		return;
9387 
9388 	r = &op->rule[0];
9389 	for (i = 0; i < op->nrules; i++, r++) {
9390 		free(r->bpf_prog.bf_insns, M_CXGBE);
9391 	}
9392 	free(op->rule, M_CXGBE);
9393 	free(op, M_CXGBE);
9394 }
9395 
9396 static int
9397 set_offload_policy(struct adapter *sc, struct t4_offload_policy *uop)
9398 {
9399 	int i, rc, len;
9400 	struct t4_offload_policy *op, *old;
9401 	struct bpf_program *bf;
9402 	const struct offload_settings *s;
9403 	struct offload_rule *r;
9404 	void *u;
9405 
9406 	if (!is_offload(sc))
9407 		return (ENODEV);
9408 
9409 	if (uop->nrules == 0) {
9410 		/* Delete installed policies. */
9411 		op = NULL;
9412 		goto set_policy;
9413 	} if (uop->nrules > 256) { /* arbitrary */
9414 		return (E2BIG);
9415 	}
9416 
9417 	/* Copy userspace offload policy to kernel */
9418 	op = malloc(sizeof(*op), M_CXGBE, M_ZERO | M_WAITOK);
9419 	op->nrules = uop->nrules;
9420 	len = op->nrules * sizeof(struct offload_rule);
9421 	op->rule = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9422 	rc = copyin(uop->rule, op->rule, len);
9423 	if (rc) {
9424 		free(op->rule, M_CXGBE);
9425 		free(op, M_CXGBE);
9426 		return (rc);
9427 	}
9428 
9429 	r = &op->rule[0];
9430 	for (i = 0; i < op->nrules; i++, r++) {
9431 
9432 		/* Validate open_type */
9433 		if (r->open_type != OPEN_TYPE_LISTEN &&
9434 		    r->open_type != OPEN_TYPE_ACTIVE &&
9435 		    r->open_type != OPEN_TYPE_PASSIVE &&
9436 		    r->open_type != OPEN_TYPE_DONTCARE) {
9437 error:
9438 			/*
9439 			 * Rules 0 to i have malloc'd filters that need to be
9440 			 * freed.  Rules i+1 to nrules have userspace pointers
9441 			 * and should be left alone.
9442 			 */
9443 			op->nrules = i;
9444 			free_offload_policy(op);
9445 			return (rc);
9446 		}
9447 
9448 		/* Validate settings */
9449 		s = &r->settings;
9450 		if ((s->offload != 0 && s->offload != 1) ||
9451 		    s->cong_algo < -1 || s->cong_algo > CONG_ALG_HIGHSPEED ||
9452 		    s->sched_class < -1 ||
9453 		    s->sched_class >= sc->chip_params->nsched_cls) {
9454 			rc = EINVAL;
9455 			goto error;
9456 		}
9457 
9458 		bf = &r->bpf_prog;
9459 		u = bf->bf_insns;	/* userspace ptr */
9460 		bf->bf_insns = NULL;
9461 		if (bf->bf_len == 0) {
9462 			/* legal, matches everything */
9463 			continue;
9464 		}
9465 		len = bf->bf_len * sizeof(*bf->bf_insns);
9466 		bf->bf_insns = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9467 		rc = copyin(u, bf->bf_insns, len);
9468 		if (rc != 0)
9469 			goto error;
9470 
9471 		if (!bpf_validate(bf->bf_insns, bf->bf_len)) {
9472 			rc = EINVAL;
9473 			goto error;
9474 		}
9475 	}
9476 set_policy:
9477 	rw_wlock(&sc->policy_lock);
9478 	old = sc->policy;
9479 	sc->policy = op;
9480 	rw_wunlock(&sc->policy_lock);
9481 	free_offload_policy(old);
9482 
9483 	return (0);
9484 }
9485 
9486 #define MAX_READ_BUF_SIZE (128 * 1024)
9487 static int
9488 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
9489 {
9490 	uint32_t addr, remaining, n;
9491 	uint32_t *buf;
9492 	int rc;
9493 	uint8_t *dst;
9494 
9495 	rc = validate_mem_range(sc, mr->addr, mr->len);
9496 	if (rc != 0)
9497 		return (rc);
9498 
9499 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
9500 	addr = mr->addr;
9501 	remaining = mr->len;
9502 	dst = (void *)mr->data;
9503 
9504 	while (remaining) {
9505 		n = min(remaining, MAX_READ_BUF_SIZE);
9506 		read_via_memwin(sc, 2, addr, buf, n);
9507 
9508 		rc = copyout(buf, dst, n);
9509 		if (rc != 0)
9510 			break;
9511 
9512 		dst += n;
9513 		remaining -= n;
9514 		addr += n;
9515 	}
9516 
9517 	free(buf, M_CXGBE);
9518 	return (rc);
9519 }
9520 #undef MAX_READ_BUF_SIZE
9521 
9522 static int
9523 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
9524 {
9525 	int rc;
9526 
9527 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
9528 		return (EINVAL);
9529 
9530 	if (i2cd->len > sizeof(i2cd->data))
9531 		return (EFBIG);
9532 
9533 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9534 	if (rc)
9535 		return (rc);
9536 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9537 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
9538 	end_synchronized_op(sc, 0);
9539 
9540 	return (rc);
9541 }
9542 
9543 int
9544 t4_os_find_pci_capability(struct adapter *sc, int cap)
9545 {
9546 	int i;
9547 
9548 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9549 }
9550 
9551 int
9552 t4_os_pci_save_state(struct adapter *sc)
9553 {
9554 	device_t dev;
9555 	struct pci_devinfo *dinfo;
9556 
9557 	dev = sc->dev;
9558 	dinfo = device_get_ivars(dev);
9559 
9560 	pci_cfg_save(dev, dinfo, 0);
9561 	return (0);
9562 }
9563 
9564 int
9565 t4_os_pci_restore_state(struct adapter *sc)
9566 {
9567 	device_t dev;
9568 	struct pci_devinfo *dinfo;
9569 
9570 	dev = sc->dev;
9571 	dinfo = device_get_ivars(dev);
9572 
9573 	pci_cfg_restore(dev, dinfo);
9574 	return (0);
9575 }
9576 
9577 void
9578 t4_os_portmod_changed(struct port_info *pi)
9579 {
9580 	struct adapter *sc = pi->adapter;
9581 	struct vi_info *vi;
9582 	struct ifnet *ifp;
9583 	static const char *mod_str[] = {
9584 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9585 	};
9586 
9587 	KASSERT((pi->flags & FIXED_IFMEDIA) == 0,
9588 	    ("%s: port_type %u", __func__, pi->port_type));
9589 
9590 	vi = &pi->vi[0];
9591 	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9592 		PORT_LOCK(pi);
9593 		build_medialist(pi);
9594 		if (pi->mod_type != FW_PORT_MOD_TYPE_NONE) {
9595 			fixup_link_config(pi);
9596 			apply_link_config(pi);
9597 		}
9598 		PORT_UNLOCK(pi);
9599 		end_synchronized_op(sc, LOCK_HELD);
9600 	}
9601 
9602 	ifp = vi->ifp;
9603 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9604 		if_printf(ifp, "transceiver unplugged.\n");
9605 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9606 		if_printf(ifp, "unknown transceiver inserted.\n");
9607 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9608 		if_printf(ifp, "unsupported transceiver inserted.\n");
9609 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9610 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9611 		    port_top_speed(pi), mod_str[pi->mod_type]);
9612 	} else {
9613 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9614 		    pi->mod_type);
9615 	}
9616 }
9617 
9618 void
9619 t4_os_link_changed(struct port_info *pi)
9620 {
9621 	struct vi_info *vi;
9622 	struct ifnet *ifp;
9623 	struct link_config *lc;
9624 	int v;
9625 
9626 	PORT_LOCK_ASSERT_OWNED(pi);
9627 
9628 	for_each_vi(pi, v, vi) {
9629 		ifp = vi->ifp;
9630 		if (ifp == NULL)
9631 			continue;
9632 
9633 		lc = &pi->link_cfg;
9634 		if (lc->link_ok) {
9635 			ifp->if_baudrate = IF_Mbps(lc->speed);
9636 			if_link_state_change(ifp, LINK_STATE_UP);
9637 		} else {
9638 			if_link_state_change(ifp, LINK_STATE_DOWN);
9639 		}
9640 	}
9641 }
9642 
9643 void
9644 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9645 {
9646 	struct adapter *sc;
9647 
9648 	sx_slock(&t4_list_lock);
9649 	SLIST_FOREACH(sc, &t4_list, link) {
9650 		/*
9651 		 * func should not make any assumptions about what state sc is
9652 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9653 		 */
9654 		func(sc, arg);
9655 	}
9656 	sx_sunlock(&t4_list_lock);
9657 }
9658 
9659 static int
9660 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9661     struct thread *td)
9662 {
9663 	int rc;
9664 	struct adapter *sc = dev->si_drv1;
9665 
9666 	rc = priv_check(td, PRIV_DRIVER);
9667 	if (rc != 0)
9668 		return (rc);
9669 
9670 	switch (cmd) {
9671 	case CHELSIO_T4_GETREG: {
9672 		struct t4_reg *edata = (struct t4_reg *)data;
9673 
9674 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9675 			return (EFAULT);
9676 
9677 		if (edata->size == 4)
9678 			edata->val = t4_read_reg(sc, edata->addr);
9679 		else if (edata->size == 8)
9680 			edata->val = t4_read_reg64(sc, edata->addr);
9681 		else
9682 			return (EINVAL);
9683 
9684 		break;
9685 	}
9686 	case CHELSIO_T4_SETREG: {
9687 		struct t4_reg *edata = (struct t4_reg *)data;
9688 
9689 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9690 			return (EFAULT);
9691 
9692 		if (edata->size == 4) {
9693 			if (edata->val & 0xffffffff00000000)
9694 				return (EINVAL);
9695 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9696 		} else if (edata->size == 8)
9697 			t4_write_reg64(sc, edata->addr, edata->val);
9698 		else
9699 			return (EINVAL);
9700 		break;
9701 	}
9702 	case CHELSIO_T4_REGDUMP: {
9703 		struct t4_regdump *regs = (struct t4_regdump *)data;
9704 		int reglen = t4_get_regs_len(sc);
9705 		uint8_t *buf;
9706 
9707 		if (regs->len < reglen) {
9708 			regs->len = reglen; /* hint to the caller */
9709 			return (ENOBUFS);
9710 		}
9711 
9712 		regs->len = reglen;
9713 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9714 		get_regs(sc, regs, buf);
9715 		rc = copyout(buf, regs->data, reglen);
9716 		free(buf, M_CXGBE);
9717 		break;
9718 	}
9719 	case CHELSIO_T4_GET_FILTER_MODE:
9720 		rc = get_filter_mode(sc, (uint32_t *)data);
9721 		break;
9722 	case CHELSIO_T4_SET_FILTER_MODE:
9723 		rc = set_filter_mode(sc, *(uint32_t *)data);
9724 		break;
9725 	case CHELSIO_T4_GET_FILTER:
9726 		rc = get_filter(sc, (struct t4_filter *)data);
9727 		break;
9728 	case CHELSIO_T4_SET_FILTER:
9729 		rc = set_filter(sc, (struct t4_filter *)data);
9730 		break;
9731 	case CHELSIO_T4_DEL_FILTER:
9732 		rc = del_filter(sc, (struct t4_filter *)data);
9733 		break;
9734 	case CHELSIO_T4_GET_SGE_CONTEXT:
9735 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9736 		break;
9737 	case CHELSIO_T4_LOAD_FW:
9738 		rc = load_fw(sc, (struct t4_data *)data);
9739 		break;
9740 	case CHELSIO_T4_GET_MEM:
9741 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9742 		break;
9743 	case CHELSIO_T4_GET_I2C:
9744 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9745 		break;
9746 	case CHELSIO_T4_CLEAR_STATS: {
9747 		int i, v, bg_map;
9748 		u_int port_id = *(uint32_t *)data;
9749 		struct port_info *pi;
9750 		struct vi_info *vi;
9751 
9752 		if (port_id >= sc->params.nports)
9753 			return (EINVAL);
9754 		pi = sc->port[port_id];
9755 		if (pi == NULL)
9756 			return (EIO);
9757 
9758 		/* MAC stats */
9759 		t4_clr_port_stats(sc, pi->tx_chan);
9760 		pi->tx_parse_error = 0;
9761 		pi->tnl_cong_drops = 0;
9762 		mtx_lock(&sc->reg_lock);
9763 		for_each_vi(pi, v, vi) {
9764 			if (vi->flags & VI_INIT_DONE)
9765 				t4_clr_vi_stats(sc, vi->viid);
9766 		}
9767 		bg_map = pi->mps_bg_map;
9768 		v = 0;	/* reuse */
9769 		while (bg_map) {
9770 			i = ffs(bg_map) - 1;
9771 			t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
9772 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
9773 			bg_map &= ~(1 << i);
9774 		}
9775 		mtx_unlock(&sc->reg_lock);
9776 
9777 		/*
9778 		 * Since this command accepts a port, clear stats for
9779 		 * all VIs on this port.
9780 		 */
9781 		for_each_vi(pi, v, vi) {
9782 			if (vi->flags & VI_INIT_DONE) {
9783 				struct sge_rxq *rxq;
9784 				struct sge_txq *txq;
9785 				struct sge_wrq *wrq;
9786 
9787 				for_each_rxq(vi, i, rxq) {
9788 #if defined(INET) || defined(INET6)
9789 					rxq->lro.lro_queued = 0;
9790 					rxq->lro.lro_flushed = 0;
9791 #endif
9792 					rxq->rxcsum = 0;
9793 					rxq->vlan_extraction = 0;
9794 				}
9795 
9796 				for_each_txq(vi, i, txq) {
9797 					txq->txcsum = 0;
9798 					txq->tso_wrs = 0;
9799 					txq->vlan_insertion = 0;
9800 					txq->imm_wrs = 0;
9801 					txq->sgl_wrs = 0;
9802 					txq->txpkt_wrs = 0;
9803 					txq->txpkts0_wrs = 0;
9804 					txq->txpkts1_wrs = 0;
9805 					txq->txpkts0_pkts = 0;
9806 					txq->txpkts1_pkts = 0;
9807 					txq->raw_wrs = 0;
9808 					mp_ring_reset_stats(txq->r);
9809 				}
9810 
9811 #ifdef TCP_OFFLOAD
9812 				/* nothing to clear for each ofld_rxq */
9813 
9814 				for_each_ofld_txq(vi, i, wrq) {
9815 					wrq->tx_wrs_direct = 0;
9816 					wrq->tx_wrs_copied = 0;
9817 				}
9818 #endif
9819 
9820 				if (IS_MAIN_VI(vi)) {
9821 					wrq = &sc->sge.ctrlq[pi->port_id];
9822 					wrq->tx_wrs_direct = 0;
9823 					wrq->tx_wrs_copied = 0;
9824 				}
9825 			}
9826 		}
9827 		break;
9828 	}
9829 	case CHELSIO_T4_SCHED_CLASS:
9830 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9831 		break;
9832 	case CHELSIO_T4_SCHED_QUEUE:
9833 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9834 		break;
9835 	case CHELSIO_T4_GET_TRACER:
9836 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9837 		break;
9838 	case CHELSIO_T4_SET_TRACER:
9839 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9840 		break;
9841 	case CHELSIO_T4_LOAD_CFG:
9842 		rc = load_cfg(sc, (struct t4_data *)data);
9843 		break;
9844 	case CHELSIO_T4_LOAD_BOOT:
9845 		rc = load_boot(sc, (struct t4_bootrom *)data);
9846 		break;
9847 	case CHELSIO_T4_LOAD_BOOTCFG:
9848 		rc = load_bootcfg(sc, (struct t4_data *)data);
9849 		break;
9850 	case CHELSIO_T4_CUDBG_DUMP:
9851 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
9852 		break;
9853 	case CHELSIO_T4_SET_OFLD_POLICY:
9854 		rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
9855 		break;
9856 	default:
9857 		rc = ENOTTY;
9858 	}
9859 
9860 	return (rc);
9861 }
9862 
9863 void
9864 t4_db_full(struct adapter *sc)
9865 {
9866 
9867 	CXGBE_UNIMPLEMENTED(__func__);
9868 }
9869 
9870 void
9871 t4_db_dropped(struct adapter *sc)
9872 {
9873 
9874 	CXGBE_UNIMPLEMENTED(__func__);
9875 }
9876 
9877 #ifdef TCP_OFFLOAD
9878 static int
9879 toe_capability(struct vi_info *vi, int enable)
9880 {
9881 	int rc;
9882 	struct port_info *pi = vi->pi;
9883 	struct adapter *sc = pi->adapter;
9884 
9885 	ASSERT_SYNCHRONIZED_OP(sc);
9886 
9887 	if (!is_offload(sc))
9888 		return (ENODEV);
9889 
9890 	if (enable) {
9891 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9892 			/* TOE is already enabled. */
9893 			return (0);
9894 		}
9895 
9896 		/*
9897 		 * We need the port's queues around so that we're able to send
9898 		 * and receive CPLs to/from the TOE even if the ifnet for this
9899 		 * port has never been UP'd administratively.
9900 		 */
9901 		if (!(vi->flags & VI_INIT_DONE)) {
9902 			rc = vi_full_init(vi);
9903 			if (rc)
9904 				return (rc);
9905 		}
9906 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9907 			rc = vi_full_init(&pi->vi[0]);
9908 			if (rc)
9909 				return (rc);
9910 		}
9911 
9912 		if (isset(&sc->offload_map, pi->port_id)) {
9913 			/* TOE is enabled on another VI of this port. */
9914 			pi->uld_vis++;
9915 			return (0);
9916 		}
9917 
9918 		if (!uld_active(sc, ULD_TOM)) {
9919 			rc = t4_activate_uld(sc, ULD_TOM);
9920 			if (rc == EAGAIN) {
9921 				log(LOG_WARNING,
9922 				    "You must kldload t4_tom.ko before trying "
9923 				    "to enable TOE on a cxgbe interface.\n");
9924 			}
9925 			if (rc != 0)
9926 				return (rc);
9927 			KASSERT(sc->tom_softc != NULL,
9928 			    ("%s: TOM activated but softc NULL", __func__));
9929 			KASSERT(uld_active(sc, ULD_TOM),
9930 			    ("%s: TOM activated but flag not set", __func__));
9931 		}
9932 
9933 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
9934 		if (!uld_active(sc, ULD_IWARP))
9935 			(void) t4_activate_uld(sc, ULD_IWARP);
9936 		if (!uld_active(sc, ULD_ISCSI))
9937 			(void) t4_activate_uld(sc, ULD_ISCSI);
9938 
9939 		pi->uld_vis++;
9940 		setbit(&sc->offload_map, pi->port_id);
9941 	} else {
9942 		pi->uld_vis--;
9943 
9944 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9945 			return (0);
9946 
9947 		KASSERT(uld_active(sc, ULD_TOM),
9948 		    ("%s: TOM never initialized?", __func__));
9949 		clrbit(&sc->offload_map, pi->port_id);
9950 	}
9951 
9952 	return (0);
9953 }
9954 
9955 /*
9956  * Add an upper layer driver to the global list.
9957  */
9958 int
9959 t4_register_uld(struct uld_info *ui)
9960 {
9961 	int rc = 0;
9962 	struct uld_info *u;
9963 
9964 	sx_xlock(&t4_uld_list_lock);
9965 	SLIST_FOREACH(u, &t4_uld_list, link) {
9966 	    if (u->uld_id == ui->uld_id) {
9967 		    rc = EEXIST;
9968 		    goto done;
9969 	    }
9970 	}
9971 
9972 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9973 	ui->refcount = 0;
9974 done:
9975 	sx_xunlock(&t4_uld_list_lock);
9976 	return (rc);
9977 }
9978 
9979 int
9980 t4_unregister_uld(struct uld_info *ui)
9981 {
9982 	int rc = EINVAL;
9983 	struct uld_info *u;
9984 
9985 	sx_xlock(&t4_uld_list_lock);
9986 
9987 	SLIST_FOREACH(u, &t4_uld_list, link) {
9988 	    if (u == ui) {
9989 		    if (ui->refcount > 0) {
9990 			    rc = EBUSY;
9991 			    goto done;
9992 		    }
9993 
9994 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9995 		    rc = 0;
9996 		    goto done;
9997 	    }
9998 	}
9999 done:
10000 	sx_xunlock(&t4_uld_list_lock);
10001 	return (rc);
10002 }
10003 
10004 int
10005 t4_activate_uld(struct adapter *sc, int id)
10006 {
10007 	int rc;
10008 	struct uld_info *ui;
10009 
10010 	ASSERT_SYNCHRONIZED_OP(sc);
10011 
10012 	if (id < 0 || id > ULD_MAX)
10013 		return (EINVAL);
10014 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
10015 
10016 	sx_slock(&t4_uld_list_lock);
10017 
10018 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10019 		if (ui->uld_id == id) {
10020 			if (!(sc->flags & FULL_INIT_DONE)) {
10021 				rc = adapter_full_init(sc);
10022 				if (rc != 0)
10023 					break;
10024 			}
10025 
10026 			rc = ui->activate(sc);
10027 			if (rc == 0) {
10028 				setbit(&sc->active_ulds, id);
10029 				ui->refcount++;
10030 			}
10031 			break;
10032 		}
10033 	}
10034 
10035 	sx_sunlock(&t4_uld_list_lock);
10036 
10037 	return (rc);
10038 }
10039 
10040 int
10041 t4_deactivate_uld(struct adapter *sc, int id)
10042 {
10043 	int rc;
10044 	struct uld_info *ui;
10045 
10046 	ASSERT_SYNCHRONIZED_OP(sc);
10047 
10048 	if (id < 0 || id > ULD_MAX)
10049 		return (EINVAL);
10050 	rc = ENXIO;
10051 
10052 	sx_slock(&t4_uld_list_lock);
10053 
10054 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10055 		if (ui->uld_id == id) {
10056 			rc = ui->deactivate(sc);
10057 			if (rc == 0) {
10058 				clrbit(&sc->active_ulds, id);
10059 				ui->refcount--;
10060 			}
10061 			break;
10062 		}
10063 	}
10064 
10065 	sx_sunlock(&t4_uld_list_lock);
10066 
10067 	return (rc);
10068 }
10069 
10070 int
10071 uld_active(struct adapter *sc, int uld_id)
10072 {
10073 
10074 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
10075 
10076 	return (isset(&sc->active_ulds, uld_id));
10077 }
10078 #endif
10079 
10080 /*
10081  * t  = ptr to tunable.
10082  * nc = number of CPUs.
10083  * c  = compiled in default for that tunable.
10084  */
10085 static void
10086 calculate_nqueues(int *t, int nc, const int c)
10087 {
10088 	int nq;
10089 
10090 	if (*t > 0)
10091 		return;
10092 	nq = *t < 0 ? -*t : c;
10093 	*t = min(nc, nq);
10094 }
10095 
10096 /*
10097  * Come up with reasonable defaults for some of the tunables, provided they're
10098  * not set by the user (in which case we'll use the values as is).
10099  */
10100 static void
10101 tweak_tunables(void)
10102 {
10103 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
10104 
10105 	if (t4_ntxq < 1) {
10106 #ifdef RSS
10107 		t4_ntxq = rss_getnumbuckets();
10108 #else
10109 		calculate_nqueues(&t4_ntxq, nc, NTXQ);
10110 #endif
10111 	}
10112 
10113 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
10114 
10115 	if (t4_nrxq < 1) {
10116 #ifdef RSS
10117 		t4_nrxq = rss_getnumbuckets();
10118 #else
10119 		calculate_nqueues(&t4_nrxq, nc, NRXQ);
10120 #endif
10121 	}
10122 
10123 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
10124 
10125 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
10126 	calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
10127 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
10128 #endif
10129 #ifdef TCP_OFFLOAD
10130 	calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
10131 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
10132 
10133 	if (t4_toecaps_allowed == -1)
10134 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
10135 
10136 	if (t4_rdmacaps_allowed == -1) {
10137 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
10138 		    FW_CAPS_CONFIG_RDMA_RDMAC;
10139 	}
10140 
10141 	if (t4_iscsicaps_allowed == -1) {
10142 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
10143 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
10144 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
10145 	}
10146 
10147 	if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
10148 		t4_tmr_idx_ofld = TMR_IDX_OFLD;
10149 
10150 	if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
10151 		t4_pktc_idx_ofld = PKTC_IDX_OFLD;
10152 #else
10153 	if (t4_toecaps_allowed == -1)
10154 		t4_toecaps_allowed = 0;
10155 
10156 	if (t4_rdmacaps_allowed == -1)
10157 		t4_rdmacaps_allowed = 0;
10158 
10159 	if (t4_iscsicaps_allowed == -1)
10160 		t4_iscsicaps_allowed = 0;
10161 #endif
10162 
10163 #ifdef DEV_NETMAP
10164 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
10165 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
10166 #endif
10167 
10168 	if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
10169 		t4_tmr_idx = TMR_IDX;
10170 
10171 	if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
10172 		t4_pktc_idx = PKTC_IDX;
10173 
10174 	if (t4_qsize_txq < 128)
10175 		t4_qsize_txq = 128;
10176 
10177 	if (t4_qsize_rxq < 128)
10178 		t4_qsize_rxq = 128;
10179 	while (t4_qsize_rxq & 7)
10180 		t4_qsize_rxq++;
10181 
10182 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
10183 
10184 	/*
10185 	 * Number of VIs to create per-port.  The first VI is the "main" regular
10186 	 * VI for the port.  The rest are additional virtual interfaces on the
10187 	 * same physical port.  Note that the main VI does not have native
10188 	 * netmap support but the extra VIs do.
10189 	 *
10190 	 * Limit the number of VIs per port to the number of available
10191 	 * MAC addresses per port.
10192 	 */
10193 	if (t4_num_vis < 1)
10194 		t4_num_vis = 1;
10195 	if (t4_num_vis > nitems(vi_mac_funcs)) {
10196 		t4_num_vis = nitems(vi_mac_funcs);
10197 		printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
10198 	}
10199 
10200 	if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
10201 		pcie_relaxed_ordering = 1;
10202 #if defined(__i386__) || defined(__amd64__)
10203 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
10204 			pcie_relaxed_ordering = 0;
10205 #endif
10206 	}
10207 }
10208 
10209 #ifdef DDB
10210 static void
10211 t4_dump_tcb(struct adapter *sc, int tid)
10212 {
10213 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
10214 
10215 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
10216 	save = t4_read_reg(sc, reg);
10217 	base = sc->memwin[2].mw_base;
10218 
10219 	/* Dump TCB for the tid */
10220 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
10221 	tcb_addr += tid * TCB_SIZE;
10222 
10223 	if (is_t4(sc)) {
10224 		pf = 0;
10225 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
10226 	} else {
10227 		pf = V_PFNUM(sc->pf);
10228 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
10229 	}
10230 	t4_write_reg(sc, reg, win_pos | pf);
10231 	t4_read_reg(sc, reg);
10232 
10233 	off = tcb_addr - win_pos;
10234 	for (i = 0; i < 4; i++) {
10235 		uint32_t buf[8];
10236 		for (j = 0; j < 8; j++, off += 4)
10237 			buf[j] = htonl(t4_read_reg(sc, base + off));
10238 
10239 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
10240 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
10241 		    buf[7]);
10242 	}
10243 
10244 	t4_write_reg(sc, reg, save);
10245 	t4_read_reg(sc, reg);
10246 }
10247 
10248 static void
10249 t4_dump_devlog(struct adapter *sc)
10250 {
10251 	struct devlog_params *dparams = &sc->params.devlog;
10252 	struct fw_devlog_e e;
10253 	int i, first, j, m, nentries, rc;
10254 	uint64_t ftstamp = UINT64_MAX;
10255 
10256 	if (dparams->start == 0) {
10257 		db_printf("devlog params not valid\n");
10258 		return;
10259 	}
10260 
10261 	nentries = dparams->size / sizeof(struct fw_devlog_e);
10262 	m = fwmtype_to_hwmtype(dparams->memtype);
10263 
10264 	/* Find the first entry. */
10265 	first = -1;
10266 	for (i = 0; i < nentries && !db_pager_quit; i++) {
10267 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10268 		    sizeof(e), (void *)&e);
10269 		if (rc != 0)
10270 			break;
10271 
10272 		if (e.timestamp == 0)
10273 			break;
10274 
10275 		e.timestamp = be64toh(e.timestamp);
10276 		if (e.timestamp < ftstamp) {
10277 			ftstamp = e.timestamp;
10278 			first = i;
10279 		}
10280 	}
10281 
10282 	if (first == -1)
10283 		return;
10284 
10285 	i = first;
10286 	do {
10287 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10288 		    sizeof(e), (void *)&e);
10289 		if (rc != 0)
10290 			return;
10291 
10292 		if (e.timestamp == 0)
10293 			return;
10294 
10295 		e.timestamp = be64toh(e.timestamp);
10296 		e.seqno = be32toh(e.seqno);
10297 		for (j = 0; j < 8; j++)
10298 			e.params[j] = be32toh(e.params[j]);
10299 
10300 		db_printf("%10d  %15ju  %8s  %8s  ",
10301 		    e.seqno, e.timestamp,
10302 		    (e.level < nitems(devlog_level_strings) ?
10303 			devlog_level_strings[e.level] : "UNKNOWN"),
10304 		    (e.facility < nitems(devlog_facility_strings) ?
10305 			devlog_facility_strings[e.facility] : "UNKNOWN"));
10306 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
10307 		    e.params[3], e.params[4], e.params[5], e.params[6],
10308 		    e.params[7]);
10309 
10310 		if (++i == nentries)
10311 			i = 0;
10312 	} while (i != first && !db_pager_quit);
10313 }
10314 
10315 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
10316 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
10317 
10318 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
10319 {
10320 	device_t dev;
10321 	int t;
10322 	bool valid;
10323 
10324 	valid = false;
10325 	t = db_read_token();
10326 	if (t == tIDENT) {
10327 		dev = device_lookup_by_name(db_tok_string);
10328 		valid = true;
10329 	}
10330 	db_skip_to_eol();
10331 	if (!valid) {
10332 		db_printf("usage: show t4 devlog <nexus>\n");
10333 		return;
10334 	}
10335 
10336 	if (dev == NULL) {
10337 		db_printf("device not found\n");
10338 		return;
10339 	}
10340 
10341 	t4_dump_devlog(device_get_softc(dev));
10342 }
10343 
10344 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
10345 {
10346 	device_t dev;
10347 	int radix, tid, t;
10348 	bool valid;
10349 
10350 	valid = false;
10351 	radix = db_radix;
10352 	db_radix = 10;
10353 	t = db_read_token();
10354 	if (t == tIDENT) {
10355 		dev = device_lookup_by_name(db_tok_string);
10356 		t = db_read_token();
10357 		if (t == tNUMBER) {
10358 			tid = db_tok_number;
10359 			valid = true;
10360 		}
10361 	}
10362 	db_radix = radix;
10363 	db_skip_to_eol();
10364 	if (!valid) {
10365 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
10366 		return;
10367 	}
10368 
10369 	if (dev == NULL) {
10370 		db_printf("device not found\n");
10371 		return;
10372 	}
10373 	if (tid < 0) {
10374 		db_printf("invalid tid\n");
10375 		return;
10376 	}
10377 
10378 	t4_dump_tcb(device_get_softc(dev), tid);
10379 }
10380 #endif
10381 
10382 /*
10383  * Borrowed from cesa_prep_aes_key().
10384  *
10385  * NB: The crypto engine wants the words in the decryption key in reverse
10386  * order.
10387  */
10388 void
10389 t4_aes_getdeckey(void *dec_key, const void *enc_key, unsigned int kbits)
10390 {
10391 	uint32_t ek[4 * (RIJNDAEL_MAXNR + 1)];
10392 	uint32_t *dkey;
10393 	int i;
10394 
10395 	rijndaelKeySetupEnc(ek, enc_key, kbits);
10396 	dkey = dec_key;
10397 	dkey += (kbits / 8) / 4;
10398 
10399 	switch (kbits) {
10400 	case 128:
10401 		for (i = 0; i < 4; i++)
10402 			*--dkey = htobe32(ek[4 * 10 + i]);
10403 		break;
10404 	case 192:
10405 		for (i = 0; i < 2; i++)
10406 			*--dkey = htobe32(ek[4 * 11 + 2 + i]);
10407 		for (i = 0; i < 4; i++)
10408 			*--dkey = htobe32(ek[4 * 12 + i]);
10409 		break;
10410 	case 256:
10411 		for (i = 0; i < 4; i++)
10412 			*--dkey = htobe32(ek[4 * 13 + i]);
10413 		for (i = 0; i < 4; i++)
10414 			*--dkey = htobe32(ek[4 * 14 + i]);
10415 		break;
10416 	}
10417 	MPASS(dkey == dec_key);
10418 }
10419 
10420 static struct sx mlu;	/* mod load unload */
10421 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
10422 
10423 static int
10424 mod_event(module_t mod, int cmd, void *arg)
10425 {
10426 	int rc = 0;
10427 	static int loaded = 0;
10428 
10429 	switch (cmd) {
10430 	case MOD_LOAD:
10431 		sx_xlock(&mlu);
10432 		if (loaded++ == 0) {
10433 			t4_sge_modload();
10434 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10435 			    t4_filter_rpl, CPL_COOKIE_FILTER);
10436 			t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
10437 			    do_l2t_write_rpl, CPL_COOKIE_FILTER);
10438 			t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL,
10439 			    t4_hashfilter_ao_rpl, CPL_COOKIE_HASHFILTER);
10440 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10441 			    t4_hashfilter_tcb_rpl, CPL_COOKIE_HASHFILTER);
10442 			t4_register_shared_cpl_handler(CPL_ABORT_RPL_RSS,
10443 			    t4_del_hashfilter_rpl, CPL_COOKIE_HASHFILTER);
10444 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
10445 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
10446 			t4_register_cpl_handler(CPL_SMT_WRITE_RPL,
10447 			    do_smt_write_rpl);
10448 			sx_init(&t4_list_lock, "T4/T5 adapters");
10449 			SLIST_INIT(&t4_list);
10450 #ifdef TCP_OFFLOAD
10451 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
10452 			SLIST_INIT(&t4_uld_list);
10453 #endif
10454 			t4_tracer_modload();
10455 			tweak_tunables();
10456 		}
10457 		sx_xunlock(&mlu);
10458 		break;
10459 
10460 	case MOD_UNLOAD:
10461 		sx_xlock(&mlu);
10462 		if (--loaded == 0) {
10463 			int tries;
10464 
10465 			sx_slock(&t4_list_lock);
10466 			if (!SLIST_EMPTY(&t4_list)) {
10467 				rc = EBUSY;
10468 				sx_sunlock(&t4_list_lock);
10469 				goto done_unload;
10470 			}
10471 #ifdef TCP_OFFLOAD
10472 			sx_slock(&t4_uld_list_lock);
10473 			if (!SLIST_EMPTY(&t4_uld_list)) {
10474 				rc = EBUSY;
10475 				sx_sunlock(&t4_uld_list_lock);
10476 				sx_sunlock(&t4_list_lock);
10477 				goto done_unload;
10478 			}
10479 #endif
10480 			tries = 0;
10481 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
10482 				uprintf("%ju clusters with custom free routine "
10483 				    "still is use.\n", t4_sge_extfree_refs());
10484 				pause("t4unload", 2 * hz);
10485 			}
10486 #ifdef TCP_OFFLOAD
10487 			sx_sunlock(&t4_uld_list_lock);
10488 #endif
10489 			sx_sunlock(&t4_list_lock);
10490 
10491 			if (t4_sge_extfree_refs() == 0) {
10492 				t4_tracer_modunload();
10493 #ifdef TCP_OFFLOAD
10494 				sx_destroy(&t4_uld_list_lock);
10495 #endif
10496 				sx_destroy(&t4_list_lock);
10497 				t4_sge_modunload();
10498 				loaded = 0;
10499 			} else {
10500 				rc = EBUSY;
10501 				loaded++;	/* undo earlier decrement */
10502 			}
10503 		}
10504 done_unload:
10505 		sx_xunlock(&mlu);
10506 		break;
10507 	}
10508 
10509 	return (rc);
10510 }
10511 
10512 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10513 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10514 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10515 
10516 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10517 MODULE_VERSION(t4nex, 1);
10518 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10519 #ifdef DEV_NETMAP
10520 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10521 #endif /* DEV_NETMAP */
10522 
10523 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10524 MODULE_VERSION(t5nex, 1);
10525 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10526 #ifdef DEV_NETMAP
10527 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10528 #endif /* DEV_NETMAP */
10529 
10530 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10531 MODULE_VERSION(t6nex, 1);
10532 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10533 #ifdef DEV_NETMAP
10534 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10535 #endif /* DEV_NETMAP */
10536 
10537 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10538 MODULE_VERSION(cxgbe, 1);
10539 
10540 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10541 MODULE_VERSION(cxl, 1);
10542 
10543 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10544 MODULE_VERSION(cc, 1);
10545 
10546 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10547 MODULE_VERSION(vcxgbe, 1);
10548 
10549 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10550 MODULE_VERSION(vcxl, 1);
10551 
10552 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10553 MODULE_VERSION(vcc, 1);
10554