xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision 48d41ef0fb2a73cd29a75544e164a59829c29351)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011 Chelsio Communications, Inc.
5  * All rights reserved.
6  * Written by: Navdeep Parhar <np@FreeBSD.org>
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_rss.h"
37 
38 #include <sys/param.h>
39 #include <sys/conf.h>
40 #include <sys/priv.h>
41 #include <sys/kernel.h>
42 #include <sys/bus.h>
43 #include <sys/module.h>
44 #include <sys/malloc.h>
45 #include <sys/queue.h>
46 #include <sys/taskqueue.h>
47 #include <sys/pciio.h>
48 #include <dev/pci/pcireg.h>
49 #include <dev/pci/pcivar.h>
50 #include <dev/pci/pci_private.h>
51 #include <sys/firmware.h>
52 #include <sys/sbuf.h>
53 #include <sys/smp.h>
54 #include <sys/socket.h>
55 #include <sys/sockio.h>
56 #include <sys/sysctl.h>
57 #include <net/ethernet.h>
58 #include <net/if.h>
59 #include <net/if_types.h>
60 #include <net/if_dl.h>
61 #include <net/if_vlan_var.h>
62 #ifdef RSS
63 #include <net/rss_config.h>
64 #endif
65 #if defined(__i386__) || defined(__amd64__)
66 #include <machine/md_var.h>
67 #include <machine/cputypes.h>
68 #include <vm/vm.h>
69 #include <vm/pmap.h>
70 #endif
71 #include <crypto/rijndael/rijndael.h>
72 #ifdef DDB
73 #include <ddb/ddb.h>
74 #include <ddb/db_lex.h>
75 #endif
76 
77 #include "common/common.h"
78 #include "common/t4_msg.h"
79 #include "common/t4_regs.h"
80 #include "common/t4_regs_values.h"
81 #include "cudbg/cudbg.h"
82 #include "t4_ioctl.h"
83 #include "t4_l2t.h"
84 #include "t4_mp_ring.h"
85 #include "t4_if.h"
86 
87 /* T4 bus driver interface */
88 static int t4_probe(device_t);
89 static int t4_attach(device_t);
90 static int t4_detach(device_t);
91 static int t4_ready(device_t);
92 static int t4_read_port_device(device_t, int, device_t *);
93 static device_method_t t4_methods[] = {
94 	DEVMETHOD(device_probe,		t4_probe),
95 	DEVMETHOD(device_attach,	t4_attach),
96 	DEVMETHOD(device_detach,	t4_detach),
97 
98 	DEVMETHOD(t4_is_main_ready,	t4_ready),
99 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
100 
101 	DEVMETHOD_END
102 };
103 static driver_t t4_driver = {
104 	"t4nex",
105 	t4_methods,
106 	sizeof(struct adapter)
107 };
108 
109 
110 /* T4 port (cxgbe) interface */
111 static int cxgbe_probe(device_t);
112 static int cxgbe_attach(device_t);
113 static int cxgbe_detach(device_t);
114 device_method_t cxgbe_methods[] = {
115 	DEVMETHOD(device_probe,		cxgbe_probe),
116 	DEVMETHOD(device_attach,	cxgbe_attach),
117 	DEVMETHOD(device_detach,	cxgbe_detach),
118 	{ 0, 0 }
119 };
120 static driver_t cxgbe_driver = {
121 	"cxgbe",
122 	cxgbe_methods,
123 	sizeof(struct port_info)
124 };
125 
126 /* T4 VI (vcxgbe) interface */
127 static int vcxgbe_probe(device_t);
128 static int vcxgbe_attach(device_t);
129 static int vcxgbe_detach(device_t);
130 static device_method_t vcxgbe_methods[] = {
131 	DEVMETHOD(device_probe,		vcxgbe_probe),
132 	DEVMETHOD(device_attach,	vcxgbe_attach),
133 	DEVMETHOD(device_detach,	vcxgbe_detach),
134 	{ 0, 0 }
135 };
136 static driver_t vcxgbe_driver = {
137 	"vcxgbe",
138 	vcxgbe_methods,
139 	sizeof(struct vi_info)
140 };
141 
142 static d_ioctl_t t4_ioctl;
143 
144 static struct cdevsw t4_cdevsw = {
145        .d_version = D_VERSION,
146        .d_ioctl = t4_ioctl,
147        .d_name = "t4nex",
148 };
149 
150 /* T5 bus driver interface */
151 static int t5_probe(device_t);
152 static device_method_t t5_methods[] = {
153 	DEVMETHOD(device_probe,		t5_probe),
154 	DEVMETHOD(device_attach,	t4_attach),
155 	DEVMETHOD(device_detach,	t4_detach),
156 
157 	DEVMETHOD(t4_is_main_ready,	t4_ready),
158 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
159 
160 	DEVMETHOD_END
161 };
162 static driver_t t5_driver = {
163 	"t5nex",
164 	t5_methods,
165 	sizeof(struct adapter)
166 };
167 
168 
169 /* T5 port (cxl) interface */
170 static driver_t cxl_driver = {
171 	"cxl",
172 	cxgbe_methods,
173 	sizeof(struct port_info)
174 };
175 
176 /* T5 VI (vcxl) interface */
177 static driver_t vcxl_driver = {
178 	"vcxl",
179 	vcxgbe_methods,
180 	sizeof(struct vi_info)
181 };
182 
183 /* T6 bus driver interface */
184 static int t6_probe(device_t);
185 static device_method_t t6_methods[] = {
186 	DEVMETHOD(device_probe,		t6_probe),
187 	DEVMETHOD(device_attach,	t4_attach),
188 	DEVMETHOD(device_detach,	t4_detach),
189 
190 	DEVMETHOD(t4_is_main_ready,	t4_ready),
191 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
192 
193 	DEVMETHOD_END
194 };
195 static driver_t t6_driver = {
196 	"t6nex",
197 	t6_methods,
198 	sizeof(struct adapter)
199 };
200 
201 
202 /* T6 port (cc) interface */
203 static driver_t cc_driver = {
204 	"cc",
205 	cxgbe_methods,
206 	sizeof(struct port_info)
207 };
208 
209 /* T6 VI (vcc) interface */
210 static driver_t vcc_driver = {
211 	"vcc",
212 	vcxgbe_methods,
213 	sizeof(struct vi_info)
214 };
215 
216 /* ifnet + media interface */
217 static void cxgbe_init(void *);
218 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
219 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
220 static void cxgbe_qflush(struct ifnet *);
221 static int cxgbe_media_change(struct ifnet *);
222 static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
223 
224 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
225 
226 /*
227  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
228  * then ADAPTER_LOCK, then t4_uld_list_lock.
229  */
230 static struct sx t4_list_lock;
231 SLIST_HEAD(, adapter) t4_list;
232 #ifdef TCP_OFFLOAD
233 static struct sx t4_uld_list_lock;
234 SLIST_HEAD(, uld_info) t4_uld_list;
235 #endif
236 
237 /*
238  * Tunables.  See tweak_tunables() too.
239  *
240  * Each tunable is set to a default value here if it's known at compile-time.
241  * Otherwise it is set to -n as an indication to tweak_tunables() that it should
242  * provide a reasonable default (upto n) when the driver is loaded.
243  *
244  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
245  * T5 are under hw.cxl.
246  */
247 
248 /*
249  * Number of queues for tx and rx, NIC and offload.
250  */
251 #define NTXQ 16
252 int t4_ntxq = -NTXQ;
253 TUNABLE_INT("hw.cxgbe.ntxq", &t4_ntxq);
254 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq);	/* Old name, undocumented */
255 
256 #define NRXQ 8
257 int t4_nrxq = -NRXQ;
258 TUNABLE_INT("hw.cxgbe.nrxq", &t4_nrxq);
259 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq);	/* Old name, undocumented */
260 
261 #define NTXQ_VI 1
262 static int t4_ntxq_vi = -NTXQ_VI;
263 TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
264 
265 #define NRXQ_VI 1
266 static int t4_nrxq_vi = -NRXQ_VI;
267 TUNABLE_INT("hw.cxgbe.nrxq_vi", &t4_nrxq_vi);
268 
269 static int t4_rsrv_noflowq = 0;
270 TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
271 
272 #ifdef TCP_OFFLOAD
273 #define NOFLDTXQ 8
274 static int t4_nofldtxq = -NOFLDTXQ;
275 TUNABLE_INT("hw.cxgbe.nofldtxq", &t4_nofldtxq);
276 
277 #define NOFLDRXQ 2
278 static int t4_nofldrxq = -NOFLDRXQ;
279 TUNABLE_INT("hw.cxgbe.nofldrxq", &t4_nofldrxq);
280 
281 #define NOFLDTXQ_VI 1
282 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
283 TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
284 
285 #define NOFLDRXQ_VI 1
286 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
287 TUNABLE_INT("hw.cxgbe.nofldrxq_vi", &t4_nofldrxq_vi);
288 
289 #define TMR_IDX_OFLD 1
290 int t4_tmr_idx_ofld = TMR_IDX_OFLD;
291 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_ofld", &t4_tmr_idx_ofld);
292 
293 #define PKTC_IDX_OFLD (-1)
294 int t4_pktc_idx_ofld = PKTC_IDX_OFLD;
295 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_ofld", &t4_pktc_idx_ofld);
296 
297 /* 0 means chip/fw default, non-zero number is value in microseconds */
298 static u_long t4_toe_keepalive_idle = 0;
299 TUNABLE_ULONG("hw.cxgbe.toe.keepalive_idle", &t4_toe_keepalive_idle);
300 
301 /* 0 means chip/fw default, non-zero number is value in microseconds */
302 static u_long t4_toe_keepalive_interval = 0;
303 TUNABLE_ULONG("hw.cxgbe.toe.keepalive_interval", &t4_toe_keepalive_interval);
304 
305 /* 0 means chip/fw default, non-zero number is # of keepalives before abort */
306 static int t4_toe_keepalive_count = 0;
307 TUNABLE_INT("hw.cxgbe.toe.keepalive_count", &t4_toe_keepalive_count);
308 
309 /* 0 means chip/fw default, non-zero number is value in microseconds */
310 static u_long t4_toe_rexmt_min = 0;
311 TUNABLE_ULONG("hw.cxgbe.toe.rexmt_min", &t4_toe_rexmt_min);
312 
313 /* 0 means chip/fw default, non-zero number is value in microseconds */
314 static u_long t4_toe_rexmt_max = 0;
315 TUNABLE_ULONG("hw.cxgbe.toe.rexmt_max", &t4_toe_rexmt_max);
316 
317 /* 0 means chip/fw default, non-zero number is # of rexmt before abort */
318 static int t4_toe_rexmt_count = 0;
319 TUNABLE_INT("hw.cxgbe.toe.rexmt_count", &t4_toe_rexmt_count);
320 
321 /* -1 means chip/fw default, other values are raw backoff values to use */
322 static int t4_toe_rexmt_backoff[16] = {
323 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
324 };
325 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.0", &t4_toe_rexmt_backoff[0]);
326 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.1", &t4_toe_rexmt_backoff[1]);
327 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.2", &t4_toe_rexmt_backoff[2]);
328 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.3", &t4_toe_rexmt_backoff[3]);
329 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.4", &t4_toe_rexmt_backoff[4]);
330 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.5", &t4_toe_rexmt_backoff[5]);
331 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.6", &t4_toe_rexmt_backoff[6]);
332 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.7", &t4_toe_rexmt_backoff[7]);
333 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.8", &t4_toe_rexmt_backoff[8]);
334 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.9", &t4_toe_rexmt_backoff[9]);
335 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.10", &t4_toe_rexmt_backoff[10]);
336 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.11", &t4_toe_rexmt_backoff[11]);
337 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.12", &t4_toe_rexmt_backoff[12]);
338 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.13", &t4_toe_rexmt_backoff[13]);
339 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.14", &t4_toe_rexmt_backoff[14]);
340 TUNABLE_INT("hw.cxgbe.toe.rexmt_backoff.15", &t4_toe_rexmt_backoff[15]);
341 #endif
342 
343 #ifdef DEV_NETMAP
344 #define NNMTXQ_VI 2
345 static int t4_nnmtxq_vi = -NNMTXQ_VI;
346 TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
347 
348 #define NNMRXQ_VI 2
349 static int t4_nnmrxq_vi = -NNMRXQ_VI;
350 TUNABLE_INT("hw.cxgbe.nnmrxq_vi", &t4_nnmrxq_vi);
351 #endif
352 
353 /*
354  * Holdoff parameters for ports.
355  */
356 #define TMR_IDX 1
357 int t4_tmr_idx = TMR_IDX;
358 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx", &t4_tmr_idx);
359 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx);	/* Old name */
360 
361 #define PKTC_IDX (-1)
362 int t4_pktc_idx = PKTC_IDX;
363 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx", &t4_pktc_idx);
364 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx);	/* Old name */
365 
366 /*
367  * Size (# of entries) of each tx and rx queue.
368  */
369 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
370 TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
371 
372 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
373 TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
374 
375 /*
376  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
377  */
378 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
379 TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
380 
381 /*
382  * Configuration file.
383  */
384 #define DEFAULT_CF	"default"
385 #define FLASH_CF	"flash"
386 #define UWIRE_CF	"uwire"
387 #define FPGA_CF		"fpga"
388 static char t4_cfg_file[32] = DEFAULT_CF;
389 TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
390 
391 /*
392  * PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
393  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
394  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
395  *            mark or when signalled to do so, 0 to never emit PAUSE.
396  */
397 static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
398 TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
399 
400 /*
401  * Forward Error Correction settings (bit 0, 1, 2 = FEC_RS, FEC_BASER_RS,
402  * FEC_RESERVED respectively).
403  * -1 to run with the firmware default.
404  *  0 to disable FEC.
405  */
406 static int t4_fec = -1;
407 TUNABLE_INT("hw.cxgbe.fec", &t4_fec);
408 
409 /*
410  * Link autonegotiation.
411  * -1 to run with the firmware default.
412  *  0 to disable.
413  *  1 to enable.
414  */
415 static int t4_autoneg = -1;
416 TUNABLE_INT("hw.cxgbe.autoneg", &t4_autoneg);
417 
418 /*
419  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
420  * encouraged respectively).
421  */
422 static unsigned int t4_fw_install = 1;
423 TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
424 
425 /*
426  * ASIC features that will be used.  Disable the ones you don't want so that the
427  * chip resources aren't wasted on features that will not be used.
428  */
429 static int t4_nbmcaps_allowed = 0;
430 TUNABLE_INT("hw.cxgbe.nbmcaps_allowed", &t4_nbmcaps_allowed);
431 
432 static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
433 TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
434 
435 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
436     FW_CAPS_CONFIG_SWITCH_EGRESS;
437 TUNABLE_INT("hw.cxgbe.switchcaps_allowed", &t4_switchcaps_allowed);
438 
439 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
440 	FW_CAPS_CONFIG_NIC_HASHFILTER;
441 TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
442 
443 static int t4_toecaps_allowed = -1;
444 TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
445 
446 static int t4_rdmacaps_allowed = -1;
447 TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
448 
449 static int t4_cryptocaps_allowed = -1;
450 TUNABLE_INT("hw.cxgbe.cryptocaps_allowed", &t4_cryptocaps_allowed);
451 
452 static int t4_iscsicaps_allowed = -1;
453 TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
454 
455 static int t4_fcoecaps_allowed = 0;
456 TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
457 
458 static int t5_write_combine = 0;
459 TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
460 
461 static int t4_num_vis = 1;
462 TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
463 /*
464  * PCIe Relaxed Ordering.
465  * -1: driver should figure out a good value.
466  * 0: disable RO.
467  * 1: enable RO.
468  * 2: leave RO alone.
469  */
470 static int pcie_relaxed_ordering = -1;
471 TUNABLE_INT("hw.cxgbe.pcie_relaxed_ordering", &pcie_relaxed_ordering);
472 
473 static int t4_panic_on_fatal_err = 0;
474 TUNABLE_INT("hw.cxgbe.panic_on_fatal_err", &t4_panic_on_fatal_err);
475 
476 #ifdef TCP_OFFLOAD
477 /*
478  * TOE tunables.
479  */
480 static int t4_cop_managed_offloading = 0;
481 TUNABLE_INT("hw.cxgbe.cop_managed_offloading", &t4_cop_managed_offloading);
482 #endif
483 
484 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
485 static int vi_mac_funcs[] = {
486 	FW_VI_FUNC_ETH,
487 	FW_VI_FUNC_OFLD,
488 	FW_VI_FUNC_IWARP,
489 	FW_VI_FUNC_OPENISCSI,
490 	FW_VI_FUNC_OPENFCOE,
491 	FW_VI_FUNC_FOISCSI,
492 	FW_VI_FUNC_FOFCOE,
493 };
494 
495 struct intrs_and_queues {
496 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
497 	uint16_t num_vis;	/* number of VIs for each port */
498 	uint16_t nirq;		/* Total # of vectors */
499 	uint16_t ntxq;		/* # of NIC txq's for each port */
500 	uint16_t nrxq;		/* # of NIC rxq's for each port */
501 	uint16_t nofldtxq;	/* # of TOE txq's for each port */
502 	uint16_t nofldrxq;	/* # of TOE rxq's for each port */
503 
504 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
505 	uint16_t ntxq_vi;	/* # of NIC txq's */
506 	uint16_t nrxq_vi;	/* # of NIC rxq's */
507 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
508 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
509 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
510 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
511 };
512 
513 static void setup_memwin(struct adapter *);
514 static void position_memwin(struct adapter *, int, uint32_t);
515 static int validate_mem_range(struct adapter *, uint32_t, int);
516 static int fwmtype_to_hwmtype(int);
517 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
518     uint32_t *);
519 static int fixup_devlog_params(struct adapter *);
520 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
521 static int prep_firmware(struct adapter *);
522 static int partition_resources(struct adapter *, const struct firmware *,
523     const char *);
524 static int get_params__pre_init(struct adapter *);
525 static int get_params__post_init(struct adapter *);
526 static int set_params__post_init(struct adapter *);
527 static void t4_set_desc(struct adapter *);
528 static void build_medialist(struct port_info *, struct ifmedia *);
529 static void init_l1cfg(struct port_info *);
530 static int cxgbe_init_synchronized(struct vi_info *);
531 static int cxgbe_uninit_synchronized(struct vi_info *);
532 static void quiesce_txq(struct adapter *, struct sge_txq *);
533 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
534 static void quiesce_iq(struct adapter *, struct sge_iq *);
535 static void quiesce_fl(struct adapter *, struct sge_fl *);
536 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
537     driver_intr_t *, void *, char *);
538 static int t4_free_irq(struct adapter *, struct irq *);
539 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
540 static void vi_refresh_stats(struct adapter *, struct vi_info *);
541 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
542 static void cxgbe_tick(void *);
543 static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
544 static void cxgbe_sysctls(struct port_info *);
545 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
546 static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
547 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
548 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
549 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
550 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
551 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
552 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
553 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
554 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
555 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
556 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
557 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
558 #ifdef SBUF_DRAIN
559 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
560 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
561 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
562 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
563 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
564 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
565 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
566 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
567 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
568 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
569 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
570 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
571 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
572 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
573 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
574 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
575 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
576 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
577 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
578 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
579 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
580 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
581 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
582 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
583 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
584 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
585 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
586 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
587 static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
588 #endif
589 #ifdef TCP_OFFLOAD
590 static int sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS);
591 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
592 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
593 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
594 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
595 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
596 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
597 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
598 #endif
599 static int get_sge_context(struct adapter *, struct t4_sge_context *);
600 static int load_fw(struct adapter *, struct t4_data *);
601 static int load_cfg(struct adapter *, struct t4_data *);
602 static int load_boot(struct adapter *, struct t4_bootrom *);
603 static int load_bootcfg(struct adapter *, struct t4_data *);
604 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
605 static void free_offload_policy(struct t4_offload_policy *);
606 static int set_offload_policy(struct adapter *, struct t4_offload_policy *);
607 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
608 static int read_i2c(struct adapter *, struct t4_i2c_data *);
609 #ifdef TCP_OFFLOAD
610 static int toe_capability(struct vi_info *, int);
611 #endif
612 static int mod_event(module_t, int, void *);
613 static int notify_siblings(device_t, int);
614 
615 struct {
616 	uint16_t device;
617 	char *desc;
618 } t4_pciids[] = {
619 	{0xa000, "Chelsio Terminator 4 FPGA"},
620 	{0x4400, "Chelsio T440-dbg"},
621 	{0x4401, "Chelsio T420-CR"},
622 	{0x4402, "Chelsio T422-CR"},
623 	{0x4403, "Chelsio T440-CR"},
624 	{0x4404, "Chelsio T420-BCH"},
625 	{0x4405, "Chelsio T440-BCH"},
626 	{0x4406, "Chelsio T440-CH"},
627 	{0x4407, "Chelsio T420-SO"},
628 	{0x4408, "Chelsio T420-CX"},
629 	{0x4409, "Chelsio T420-BT"},
630 	{0x440a, "Chelsio T404-BT"},
631 	{0x440e, "Chelsio T440-LP-CR"},
632 }, t5_pciids[] = {
633 	{0xb000, "Chelsio Terminator 5 FPGA"},
634 	{0x5400, "Chelsio T580-dbg"},
635 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
636 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
637 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
638 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
639 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
640 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
641 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
642 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
643 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
644 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
645 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
646 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
647 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
648 #ifdef notyet
649 	{0x5404,  "Chelsio T520-BCH"},
650 	{0x5405,  "Chelsio T540-BCH"},
651 	{0x5406,  "Chelsio T540-CH"},
652 	{0x5408,  "Chelsio T520-CX"},
653 	{0x540b,  "Chelsio B520-SR"},
654 	{0x540c,  "Chelsio B504-BT"},
655 	{0x540f,  "Chelsio Amsterdam"},
656 	{0x5413,  "Chelsio T580-CHR"},
657 #endif
658 }, t6_pciids[] = {
659 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
660 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
661 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
662 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
663 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
664 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
665 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
666 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
667 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
668 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
669 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
670 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
671 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
672 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
673 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
674 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
675 
676 	/* Custom */
677 	{0x6480, "Chelsio T6225 80"},
678 	{0x6481, "Chelsio T62100 81"},
679 	{0x6484, "Chelsio T62100 84"},
680 };
681 
682 #ifdef TCP_OFFLOAD
683 /*
684  * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
685  * exactly the same for both rxq and ofld_rxq.
686  */
687 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
688 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
689 #endif
690 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
691 
692 static int
693 t4_probe(device_t dev)
694 {
695 	int i;
696 	uint16_t v = pci_get_vendor(dev);
697 	uint16_t d = pci_get_device(dev);
698 	uint8_t f = pci_get_function(dev);
699 
700 	if (v != PCI_VENDOR_ID_CHELSIO)
701 		return (ENXIO);
702 
703 	/* Attach only to PF0 of the FPGA */
704 	if (d == 0xa000 && f != 0)
705 		return (ENXIO);
706 
707 	for (i = 0; i < nitems(t4_pciids); i++) {
708 		if (d == t4_pciids[i].device) {
709 			device_set_desc(dev, t4_pciids[i].desc);
710 			return (BUS_PROBE_DEFAULT);
711 		}
712 	}
713 
714 	return (ENXIO);
715 }
716 
717 static int
718 t5_probe(device_t dev)
719 {
720 	int i;
721 	uint16_t v = pci_get_vendor(dev);
722 	uint16_t d = pci_get_device(dev);
723 	uint8_t f = pci_get_function(dev);
724 
725 	if (v != PCI_VENDOR_ID_CHELSIO)
726 		return (ENXIO);
727 
728 	/* Attach only to PF0 of the FPGA */
729 	if (d == 0xb000 && f != 0)
730 		return (ENXIO);
731 
732 	for (i = 0; i < nitems(t5_pciids); i++) {
733 		if (d == t5_pciids[i].device) {
734 			device_set_desc(dev, t5_pciids[i].desc);
735 			return (BUS_PROBE_DEFAULT);
736 		}
737 	}
738 
739 	return (ENXIO);
740 }
741 
742 static int
743 t6_probe(device_t dev)
744 {
745 	int i;
746 	uint16_t v = pci_get_vendor(dev);
747 	uint16_t d = pci_get_device(dev);
748 
749 	if (v != PCI_VENDOR_ID_CHELSIO)
750 		return (ENXIO);
751 
752 	for (i = 0; i < nitems(t6_pciids); i++) {
753 		if (d == t6_pciids[i].device) {
754 			device_set_desc(dev, t6_pciids[i].desc);
755 			return (BUS_PROBE_DEFAULT);
756 		}
757 	}
758 
759 	return (ENXIO);
760 }
761 
762 static void
763 t5_attribute_workaround(device_t dev)
764 {
765 	device_t root_port;
766 	uint32_t v;
767 
768 	/*
769 	 * The T5 chips do not properly echo the No Snoop and Relaxed
770 	 * Ordering attributes when replying to a TLP from a Root
771 	 * Port.  As a workaround, find the parent Root Port and
772 	 * disable No Snoop and Relaxed Ordering.  Note that this
773 	 * affects all devices under this root port.
774 	 */
775 	root_port = pci_find_pcie_root_port(dev);
776 	if (root_port == NULL) {
777 		device_printf(dev, "Unable to find parent root port\n");
778 		return;
779 	}
780 
781 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
782 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
783 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
784 	    0)
785 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
786 		    device_get_nameunit(root_port));
787 }
788 
789 static const struct devnames devnames[] = {
790 	{
791 		.nexus_name = "t4nex",
792 		.ifnet_name = "cxgbe",
793 		.vi_ifnet_name = "vcxgbe",
794 		.pf03_drv_name = "t4iov",
795 		.vf_nexus_name = "t4vf",
796 		.vf_ifnet_name = "cxgbev"
797 	}, {
798 		.nexus_name = "t5nex",
799 		.ifnet_name = "cxl",
800 		.vi_ifnet_name = "vcxl",
801 		.pf03_drv_name = "t5iov",
802 		.vf_nexus_name = "t5vf",
803 		.vf_ifnet_name = "cxlv"
804 	}, {
805 		.nexus_name = "t6nex",
806 		.ifnet_name = "cc",
807 		.vi_ifnet_name = "vcc",
808 		.pf03_drv_name = "t6iov",
809 		.vf_nexus_name = "t6vf",
810 		.vf_ifnet_name = "ccv"
811 	}
812 };
813 
814 void
815 t4_init_devnames(struct adapter *sc)
816 {
817 	int id;
818 
819 	id = chip_id(sc);
820 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
821 		sc->names = &devnames[id - CHELSIO_T4];
822 	else {
823 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
824 		sc->names = NULL;
825 	}
826 }
827 
828 static int
829 t4_attach(device_t dev)
830 {
831 	struct adapter *sc;
832 	int rc = 0, i, j, rqidx, tqidx, nports;
833 	struct make_dev_args mda;
834 	struct intrs_and_queues iaq;
835 	struct sge *s;
836 	uint32_t *buf;
837 #ifdef TCP_OFFLOAD
838 	int ofld_rqidx, ofld_tqidx;
839 #endif
840 #ifdef DEV_NETMAP
841 	int nm_rqidx, nm_tqidx;
842 #endif
843 	int num_vis;
844 
845 	sc = device_get_softc(dev);
846 	sc->dev = dev;
847 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
848 
849 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
850 		t5_attribute_workaround(dev);
851 	pci_enable_busmaster(dev);
852 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
853 		uint32_t v;
854 
855 		pci_set_max_read_req(dev, 4096);
856 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
857 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
858 		if (pcie_relaxed_ordering == 0 &&
859 		    (v | PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
860 			v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
861 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
862 		} else if (pcie_relaxed_ordering == 1 &&
863 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
864 			v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
865 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
866 		}
867 	}
868 
869 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
870 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
871 	sc->traceq = -1;
872 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
873 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
874 	    device_get_nameunit(dev));
875 
876 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
877 	    device_get_nameunit(dev));
878 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
879 	t4_add_adapter(sc);
880 
881 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
882 	TAILQ_INIT(&sc->sfl);
883 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
884 
885 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
886 
887 	sc->policy = NULL;
888 	rw_init(&sc->policy_lock, "connection offload policy");
889 
890 	rc = t4_map_bars_0_and_4(sc);
891 	if (rc != 0)
892 		goto done; /* error message displayed already */
893 
894 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
895 
896 	/* Prepare the adapter for operation. */
897 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
898 	rc = -t4_prep_adapter(sc, buf);
899 	free(buf, M_CXGBE);
900 	if (rc != 0) {
901 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
902 		goto done;
903 	}
904 
905 	/*
906 	 * This is the real PF# to which we're attaching.  Works from within PCI
907 	 * passthrough environments too, where pci_get_function() could return a
908 	 * different PF# depending on the passthrough configuration.  We need to
909 	 * use the real PF# in all our communication with the firmware.
910 	 */
911 	j = t4_read_reg(sc, A_PL_WHOAMI);
912 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
913 	sc->mbox = sc->pf;
914 
915 	t4_init_devnames(sc);
916 	if (sc->names == NULL) {
917 		rc = ENOTSUP;
918 		goto done; /* error message displayed already */
919 	}
920 
921 	/*
922 	 * Do this really early, with the memory windows set up even before the
923 	 * character device.  The userland tool's register i/o and mem read
924 	 * will work even in "recovery mode".
925 	 */
926 	setup_memwin(sc);
927 	if (t4_init_devlog_params(sc, 0) == 0)
928 		fixup_devlog_params(sc);
929 	make_dev_args_init(&mda);
930 	mda.mda_devsw = &t4_cdevsw;
931 	mda.mda_uid = UID_ROOT;
932 	mda.mda_gid = GID_WHEEL;
933 	mda.mda_mode = 0600;
934 	mda.mda_si_drv1 = sc;
935 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
936 	if (rc != 0)
937 		device_printf(dev, "failed to create nexus char device: %d.\n",
938 		    rc);
939 
940 	/* Go no further if recovery mode has been requested. */
941 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
942 		device_printf(dev, "recovery mode.\n");
943 		goto done;
944 	}
945 
946 #if defined(__i386__)
947 	if ((cpu_feature & CPUID_CX8) == 0) {
948 		device_printf(dev, "64 bit atomics not available.\n");
949 		rc = ENOTSUP;
950 		goto done;
951 	}
952 #endif
953 
954 	/* Prepare the firmware for operation */
955 	rc = prep_firmware(sc);
956 	if (rc != 0)
957 		goto done; /* error message displayed already */
958 
959 	rc = get_params__post_init(sc);
960 	if (rc != 0)
961 		goto done; /* error message displayed already */
962 
963 	rc = set_params__post_init(sc);
964 	if (rc != 0)
965 		goto done; /* error message displayed already */
966 
967 	rc = t4_map_bar_2(sc);
968 	if (rc != 0)
969 		goto done; /* error message displayed already */
970 
971 	rc = t4_create_dma_tag(sc);
972 	if (rc != 0)
973 		goto done; /* error message displayed already */
974 
975 	/*
976 	 * First pass over all the ports - allocate VIs and initialize some
977 	 * basic parameters like mac address, port type, etc.
978 	 */
979 	for_each_port(sc, i) {
980 		struct port_info *pi;
981 
982 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
983 		sc->port[i] = pi;
984 
985 		/* These must be set before t4_port_init */
986 		pi->adapter = sc;
987 		pi->port_id = i;
988 		/*
989 		 * XXX: vi[0] is special so we can't delay this allocation until
990 		 * pi->nvi's final value is known.
991 		 */
992 		pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
993 		    M_ZERO | M_WAITOK);
994 
995 		/*
996 		 * Allocate the "main" VI and initialize parameters
997 		 * like mac addr.
998 		 */
999 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1000 		if (rc != 0) {
1001 			device_printf(dev, "unable to initialize port %d: %d\n",
1002 			    i, rc);
1003 			free(pi->vi, M_CXGBE);
1004 			free(pi, M_CXGBE);
1005 			sc->port[i] = NULL;
1006 			goto done;
1007 		}
1008 
1009 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1010 		    device_get_nameunit(dev), i);
1011 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1012 		sc->chan_map[pi->tx_chan] = i;
1013 
1014 		/* All VIs on this port share this media. */
1015 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1016 		    cxgbe_media_status);
1017 
1018 		pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
1019 		if (pi->dev == NULL) {
1020 			device_printf(dev,
1021 			    "failed to add device for port %d.\n", i);
1022 			rc = ENXIO;
1023 			goto done;
1024 		}
1025 		pi->vi[0].dev = pi->dev;
1026 		device_set_softc(pi->dev, pi);
1027 	}
1028 
1029 	/*
1030 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1031 	 */
1032 	nports = sc->params.nports;
1033 	rc = cfg_itype_and_nqueues(sc, &iaq);
1034 	if (rc != 0)
1035 		goto done; /* error message displayed already */
1036 
1037 	num_vis = iaq.num_vis;
1038 	sc->intr_type = iaq.intr_type;
1039 	sc->intr_count = iaq.nirq;
1040 
1041 	s = &sc->sge;
1042 	s->nrxq = nports * iaq.nrxq;
1043 	s->ntxq = nports * iaq.ntxq;
1044 	if (num_vis > 1) {
1045 		s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1046 		s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1047 	}
1048 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1049 	s->neq += nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
1050 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1051 #ifdef TCP_OFFLOAD
1052 	if (is_offload(sc)) {
1053 		s->nofldrxq = nports * iaq.nofldrxq;
1054 		s->nofldtxq = nports * iaq.nofldtxq;
1055 		if (num_vis > 1) {
1056 			s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1057 			s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1058 		}
1059 		s->neq += s->nofldtxq + s->nofldrxq;
1060 		s->niq += s->nofldrxq;
1061 
1062 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1063 		    M_CXGBE, M_ZERO | M_WAITOK);
1064 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1065 		    M_CXGBE, M_ZERO | M_WAITOK);
1066 	}
1067 #endif
1068 #ifdef DEV_NETMAP
1069 	if (num_vis > 1) {
1070 		s->nnmrxq = nports * (num_vis - 1) * iaq.nnmrxq_vi;
1071 		s->nnmtxq = nports * (num_vis - 1) * iaq.nnmtxq_vi;
1072 	}
1073 	s->neq += s->nnmtxq + s->nnmrxq;
1074 	s->niq += s->nnmrxq;
1075 
1076 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1077 	    M_CXGBE, M_ZERO | M_WAITOK);
1078 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1079 	    M_CXGBE, M_ZERO | M_WAITOK);
1080 #endif
1081 
1082 	s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1083 	    M_ZERO | M_WAITOK);
1084 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1085 	    M_ZERO | M_WAITOK);
1086 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1087 	    M_ZERO | M_WAITOK);
1088 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1089 	    M_ZERO | M_WAITOK);
1090 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1091 	    M_ZERO | M_WAITOK);
1092 
1093 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1094 	    M_ZERO | M_WAITOK);
1095 
1096 	t4_init_l2t(sc, M_WAITOK);
1097 	t4_init_tx_sched(sc);
1098 
1099 	/*
1100 	 * Second pass over the ports.  This time we know the number of rx and
1101 	 * tx queues that each port should get.
1102 	 */
1103 	rqidx = tqidx = 0;
1104 #ifdef TCP_OFFLOAD
1105 	ofld_rqidx = ofld_tqidx = 0;
1106 #endif
1107 #ifdef DEV_NETMAP
1108 	nm_rqidx = nm_tqidx = 0;
1109 #endif
1110 	for_each_port(sc, i) {
1111 		struct port_info *pi = sc->port[i];
1112 		struct vi_info *vi;
1113 
1114 		if (pi == NULL)
1115 			continue;
1116 
1117 		pi->nvi = num_vis;
1118 		for_each_vi(pi, j, vi) {
1119 			vi->pi = pi;
1120 			vi->qsize_rxq = t4_qsize_rxq;
1121 			vi->qsize_txq = t4_qsize_txq;
1122 
1123 			vi->first_rxq = rqidx;
1124 			vi->first_txq = tqidx;
1125 			vi->tmr_idx = t4_tmr_idx;
1126 			vi->pktc_idx = t4_pktc_idx;
1127 			vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1128 			vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1129 
1130 			rqidx += vi->nrxq;
1131 			tqidx += vi->ntxq;
1132 
1133 			if (j == 0 && vi->ntxq > 1)
1134 				vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1135 			else
1136 				vi->rsrv_noflowq = 0;
1137 
1138 #ifdef TCP_OFFLOAD
1139 			vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1140 			vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1141 			vi->first_ofld_rxq = ofld_rqidx;
1142 			vi->first_ofld_txq = ofld_tqidx;
1143 			vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1144 			vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1145 
1146 			ofld_rqidx += vi->nofldrxq;
1147 			ofld_tqidx += vi->nofldtxq;
1148 #endif
1149 #ifdef DEV_NETMAP
1150 			if (j > 0) {
1151 				vi->first_nm_rxq = nm_rqidx;
1152 				vi->first_nm_txq = nm_tqidx;
1153 				vi->nnmrxq = iaq.nnmrxq_vi;
1154 				vi->nnmtxq = iaq.nnmtxq_vi;
1155 				nm_rqidx += vi->nnmrxq;
1156 				nm_tqidx += vi->nnmtxq;
1157 			}
1158 #endif
1159 		}
1160 	}
1161 
1162 	rc = t4_setup_intr_handlers(sc);
1163 	if (rc != 0) {
1164 		device_printf(dev,
1165 		    "failed to setup interrupt handlers: %d\n", rc);
1166 		goto done;
1167 	}
1168 
1169 	rc = bus_generic_probe(dev);
1170 	if (rc != 0) {
1171 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1172 		goto done;
1173 	}
1174 
1175 	/*
1176 	 * Ensure thread-safe mailbox access (in debug builds).
1177 	 *
1178 	 * So far this was the only thread accessing the mailbox but various
1179 	 * ifnets and sysctls are about to be created and their handlers/ioctls
1180 	 * will access the mailbox from different threads.
1181 	 */
1182 	sc->flags |= CHK_MBOX_ACCESS;
1183 
1184 	rc = bus_generic_attach(dev);
1185 	if (rc != 0) {
1186 		device_printf(dev,
1187 		    "failed to attach all child ports: %d\n", rc);
1188 		goto done;
1189 	}
1190 
1191 	device_printf(dev,
1192 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1193 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1194 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1195 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1196 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1197 
1198 	t4_set_desc(sc);
1199 
1200 	notify_siblings(dev, 0);
1201 
1202 done:
1203 	if (rc != 0 && sc->cdev) {
1204 		/* cdev was created and so cxgbetool works; recover that way. */
1205 		device_printf(dev,
1206 		    "error during attach, adapter is now in recovery mode.\n");
1207 		rc = 0;
1208 	}
1209 
1210 	if (rc != 0)
1211 		t4_detach_common(dev);
1212 	else
1213 		t4_sysctls(sc);
1214 
1215 	return (rc);
1216 }
1217 
1218 static int
1219 t4_ready(device_t dev)
1220 {
1221 	struct adapter *sc;
1222 
1223 	sc = device_get_softc(dev);
1224 	if (sc->flags & FW_OK)
1225 		return (0);
1226 	return (ENXIO);
1227 }
1228 
1229 static int
1230 t4_read_port_device(device_t dev, int port, device_t *child)
1231 {
1232 	struct adapter *sc;
1233 	struct port_info *pi;
1234 
1235 	sc = device_get_softc(dev);
1236 	if (port < 0 || port >= MAX_NPORTS)
1237 		return (EINVAL);
1238 	pi = sc->port[port];
1239 	if (pi == NULL || pi->dev == NULL)
1240 		return (ENXIO);
1241 	*child = pi->dev;
1242 	return (0);
1243 }
1244 
1245 static int
1246 notify_siblings(device_t dev, int detaching)
1247 {
1248 	device_t sibling;
1249 	int error, i;
1250 
1251 	error = 0;
1252 	for (i = 0; i < PCI_FUNCMAX; i++) {
1253 		if (i == pci_get_function(dev))
1254 			continue;
1255 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1256 		    pci_get_slot(dev), i);
1257 		if (sibling == NULL || !device_is_attached(sibling))
1258 			continue;
1259 		if (detaching)
1260 			error = T4_DETACH_CHILD(sibling);
1261 		else
1262 			(void)T4_ATTACH_CHILD(sibling);
1263 		if (error)
1264 			break;
1265 	}
1266 	return (error);
1267 }
1268 
1269 /*
1270  * Idempotent
1271  */
1272 static int
1273 t4_detach(device_t dev)
1274 {
1275 	struct adapter *sc;
1276 	int rc;
1277 
1278 	sc = device_get_softc(dev);
1279 
1280 	rc = notify_siblings(dev, 1);
1281 	if (rc) {
1282 		device_printf(dev,
1283 		    "failed to detach sibling devices: %d\n", rc);
1284 		return (rc);
1285 	}
1286 
1287 	return (t4_detach_common(dev));
1288 }
1289 
1290 int
1291 t4_detach_common(device_t dev)
1292 {
1293 	struct adapter *sc;
1294 	struct port_info *pi;
1295 	int i, rc;
1296 
1297 	sc = device_get_softc(dev);
1298 
1299 	if (sc->cdev) {
1300 		destroy_dev(sc->cdev);
1301 		sc->cdev = NULL;
1302 	}
1303 
1304 	sc->flags &= ~CHK_MBOX_ACCESS;
1305 	if (sc->flags & FULL_INIT_DONE) {
1306 		if (!(sc->flags & IS_VF))
1307 			t4_intr_disable(sc);
1308 	}
1309 
1310 	if (device_is_attached(dev)) {
1311 		rc = bus_generic_detach(dev);
1312 		if (rc) {
1313 			device_printf(dev,
1314 			    "failed to detach child devices: %d\n", rc);
1315 			return (rc);
1316 		}
1317 	}
1318 
1319 	for (i = 0; i < sc->intr_count; i++)
1320 		t4_free_irq(sc, &sc->irq[i]);
1321 
1322 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1323 		t4_free_tx_sched(sc);
1324 
1325 	for (i = 0; i < MAX_NPORTS; i++) {
1326 		pi = sc->port[i];
1327 		if (pi) {
1328 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1329 			if (pi->dev)
1330 				device_delete_child(dev, pi->dev);
1331 
1332 			mtx_destroy(&pi->pi_lock);
1333 			free(pi->vi, M_CXGBE);
1334 			free(pi, M_CXGBE);
1335 		}
1336 	}
1337 
1338 	device_delete_children(dev);
1339 
1340 	if (sc->flags & FULL_INIT_DONE)
1341 		adapter_full_uninit(sc);
1342 
1343 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1344 		t4_fw_bye(sc, sc->mbox);
1345 
1346 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1347 		pci_release_msi(dev);
1348 
1349 	if (sc->regs_res)
1350 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1351 		    sc->regs_res);
1352 
1353 	if (sc->udbs_res)
1354 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1355 		    sc->udbs_res);
1356 
1357 	if (sc->msix_res)
1358 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1359 		    sc->msix_res);
1360 
1361 	if (sc->l2t)
1362 		t4_free_l2t(sc->l2t);
1363 
1364 #ifdef TCP_OFFLOAD
1365 	free(sc->sge.ofld_rxq, M_CXGBE);
1366 	free(sc->sge.ofld_txq, M_CXGBE);
1367 #endif
1368 #ifdef DEV_NETMAP
1369 	free(sc->sge.nm_rxq, M_CXGBE);
1370 	free(sc->sge.nm_txq, M_CXGBE);
1371 #endif
1372 	free(sc->irq, M_CXGBE);
1373 	free(sc->sge.rxq, M_CXGBE);
1374 	free(sc->sge.txq, M_CXGBE);
1375 	free(sc->sge.ctrlq, M_CXGBE);
1376 	free(sc->sge.iqmap, M_CXGBE);
1377 	free(sc->sge.eqmap, M_CXGBE);
1378 	free(sc->tids.ftid_tab, M_CXGBE);
1379 	free(sc->tids.hftid_tab, M_CXGBE);
1380 	free(sc->tids.atid_tab, M_CXGBE);
1381 	free(sc->tids.tid_tab, M_CXGBE);
1382 	free(sc->tt.tls_rx_ports, M_CXGBE);
1383 	t4_destroy_dma_tag(sc);
1384 	if (mtx_initialized(&sc->sc_lock)) {
1385 		sx_xlock(&t4_list_lock);
1386 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1387 		sx_xunlock(&t4_list_lock);
1388 		mtx_destroy(&sc->sc_lock);
1389 	}
1390 
1391 	callout_drain(&sc->sfl_callout);
1392 	if (mtx_initialized(&sc->tids.ftid_lock)) {
1393 		mtx_destroy(&sc->tids.ftid_lock);
1394 		cv_destroy(&sc->tids.ftid_cv);
1395 	}
1396 	if (mtx_initialized(&sc->tids.hftid_lock)) {
1397 		mtx_destroy(&sc->tids.hftid_lock);
1398 		cv_destroy(&sc->tids.hftid_cv);
1399 	}
1400 	if (mtx_initialized(&sc->tids.atid_lock))
1401 		mtx_destroy(&sc->tids.atid_lock);
1402 	if (mtx_initialized(&sc->sfl_lock))
1403 		mtx_destroy(&sc->sfl_lock);
1404 	if (mtx_initialized(&sc->ifp_lock))
1405 		mtx_destroy(&sc->ifp_lock);
1406 	if (mtx_initialized(&sc->reg_lock))
1407 		mtx_destroy(&sc->reg_lock);
1408 
1409 	if (rw_initialized(&sc->policy_lock)) {
1410 		rw_destroy(&sc->policy_lock);
1411 #ifdef TCP_OFFLOAD
1412 		if (sc->policy != NULL)
1413 			free_offload_policy(sc->policy);
1414 #endif
1415 	}
1416 
1417 	for (i = 0; i < NUM_MEMWIN; i++) {
1418 		struct memwin *mw = &sc->memwin[i];
1419 
1420 		if (rw_initialized(&mw->mw_lock))
1421 			rw_destroy(&mw->mw_lock);
1422 	}
1423 
1424 	bzero(sc, sizeof(*sc));
1425 
1426 	return (0);
1427 }
1428 
1429 static int
1430 cxgbe_probe(device_t dev)
1431 {
1432 	char buf[128];
1433 	struct port_info *pi = device_get_softc(dev);
1434 
1435 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1436 	device_set_desc_copy(dev, buf);
1437 
1438 	return (BUS_PROBE_DEFAULT);
1439 }
1440 
1441 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1442     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1443     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1444 #define T4_CAP_ENABLE (T4_CAP)
1445 
1446 static int
1447 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1448 {
1449 	struct ifnet *ifp;
1450 	struct sbuf *sb;
1451 
1452 	vi->xact_addr_filt = -1;
1453 	callout_init(&vi->tick, 1);
1454 
1455 	/* Allocate an ifnet and set it up */
1456 	ifp = if_alloc(IFT_ETHER);
1457 	if (ifp == NULL) {
1458 		device_printf(dev, "Cannot allocate ifnet\n");
1459 		return (ENOMEM);
1460 	}
1461 	vi->ifp = ifp;
1462 	ifp->if_softc = vi;
1463 
1464 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1465 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1466 
1467 	ifp->if_init = cxgbe_init;
1468 	ifp->if_ioctl = cxgbe_ioctl;
1469 	ifp->if_transmit = cxgbe_transmit;
1470 	ifp->if_qflush = cxgbe_qflush;
1471 	ifp->if_get_counter = cxgbe_get_counter;
1472 
1473 	ifp->if_capabilities = T4_CAP;
1474 #ifdef TCP_OFFLOAD
1475 	if (vi->nofldrxq != 0)
1476 		ifp->if_capabilities |= IFCAP_TOE;
1477 #endif
1478 #ifdef DEV_NETMAP
1479 	if (vi->nnmrxq != 0)
1480 		ifp->if_capabilities |= IFCAP_NETMAP;
1481 #endif
1482 	ifp->if_capenable = T4_CAP_ENABLE;
1483 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1484 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1485 
1486 	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1487 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1488 	ifp->if_hw_tsomaxsegsize = 65536;
1489 
1490 	vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1491 	    EVENTHANDLER_PRI_ANY);
1492 
1493 	ether_ifattach(ifp, vi->hw_addr);
1494 #ifdef DEV_NETMAP
1495 	if (ifp->if_capabilities & IFCAP_NETMAP)
1496 		cxgbe_nm_attach(vi);
1497 #endif
1498 	sb = sbuf_new_auto();
1499 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1500 #ifdef TCP_OFFLOAD
1501 	if (ifp->if_capabilities & IFCAP_TOE)
1502 		sbuf_printf(sb, "; %d txq, %d rxq (TOE)",
1503 		    vi->nofldtxq, vi->nofldrxq);
1504 #endif
1505 #ifdef DEV_NETMAP
1506 	if (ifp->if_capabilities & IFCAP_NETMAP)
1507 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1508 		    vi->nnmtxq, vi->nnmrxq);
1509 #endif
1510 	sbuf_finish(sb);
1511 	device_printf(dev, "%s\n", sbuf_data(sb));
1512 	sbuf_delete(sb);
1513 
1514 	vi_sysctls(vi);
1515 
1516 	return (0);
1517 }
1518 
1519 static int
1520 cxgbe_attach(device_t dev)
1521 {
1522 	struct port_info *pi = device_get_softc(dev);
1523 	struct adapter *sc = pi->adapter;
1524 	struct vi_info *vi;
1525 	int i, rc;
1526 
1527 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1528 
1529 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1530 	if (rc)
1531 		return (rc);
1532 
1533 	for_each_vi(pi, i, vi) {
1534 		if (i == 0)
1535 			continue;
1536 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1537 		if (vi->dev == NULL) {
1538 			device_printf(dev, "failed to add VI %d\n", i);
1539 			continue;
1540 		}
1541 		device_set_softc(vi->dev, vi);
1542 	}
1543 
1544 	cxgbe_sysctls(pi);
1545 
1546 	bus_generic_attach(dev);
1547 
1548 	return (0);
1549 }
1550 
1551 static void
1552 cxgbe_vi_detach(struct vi_info *vi)
1553 {
1554 	struct ifnet *ifp = vi->ifp;
1555 
1556 	ether_ifdetach(ifp);
1557 
1558 	if (vi->vlan_c)
1559 		EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1560 
1561 	/* Let detach proceed even if these fail. */
1562 #ifdef DEV_NETMAP
1563 	if (ifp->if_capabilities & IFCAP_NETMAP)
1564 		cxgbe_nm_detach(vi);
1565 #endif
1566 	cxgbe_uninit_synchronized(vi);
1567 	callout_drain(&vi->tick);
1568 	vi_full_uninit(vi);
1569 
1570 	if_free(vi->ifp);
1571 	vi->ifp = NULL;
1572 }
1573 
1574 static int
1575 cxgbe_detach(device_t dev)
1576 {
1577 	struct port_info *pi = device_get_softc(dev);
1578 	struct adapter *sc = pi->adapter;
1579 	int rc;
1580 
1581 	/* Detach the extra VIs first. */
1582 	rc = bus_generic_detach(dev);
1583 	if (rc)
1584 		return (rc);
1585 	device_delete_children(dev);
1586 
1587 	doom_vi(sc, &pi->vi[0]);
1588 
1589 	if (pi->flags & HAS_TRACEQ) {
1590 		sc->traceq = -1;	/* cloner should not create ifnet */
1591 		t4_tracer_port_detach(sc);
1592 	}
1593 
1594 	cxgbe_vi_detach(&pi->vi[0]);
1595 	callout_drain(&pi->tick);
1596 	ifmedia_removeall(&pi->media);
1597 
1598 	end_synchronized_op(sc, 0);
1599 
1600 	return (0);
1601 }
1602 
1603 static void
1604 cxgbe_init(void *arg)
1605 {
1606 	struct vi_info *vi = arg;
1607 	struct adapter *sc = vi->pi->adapter;
1608 
1609 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1610 		return;
1611 	cxgbe_init_synchronized(vi);
1612 	end_synchronized_op(sc, 0);
1613 }
1614 
1615 static int
1616 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1617 {
1618 	int rc = 0, mtu, flags, can_sleep;
1619 	struct vi_info *vi = ifp->if_softc;
1620 	struct port_info *pi = vi->pi;
1621 	struct adapter *sc = pi->adapter;
1622 	struct ifreq *ifr = (struct ifreq *)data;
1623 	uint32_t mask;
1624 
1625 	switch (cmd) {
1626 	case SIOCSIFMTU:
1627 		mtu = ifr->ifr_mtu;
1628 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1629 			return (EINVAL);
1630 
1631 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1632 		if (rc)
1633 			return (rc);
1634 		ifp->if_mtu = mtu;
1635 		if (vi->flags & VI_INIT_DONE) {
1636 			t4_update_fl_bufsize(ifp);
1637 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1638 				rc = update_mac_settings(ifp, XGMAC_MTU);
1639 		}
1640 		end_synchronized_op(sc, 0);
1641 		break;
1642 
1643 	case SIOCSIFFLAGS:
1644 		can_sleep = 0;
1645 redo_sifflags:
1646 		rc = begin_synchronized_op(sc, vi,
1647 		    can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1648 		if (rc) {
1649 			if_printf(ifp, "%ssleepable synch operation failed: %d."
1650 			    "  if_flags 0x%08x, if_drv_flags 0x%08x\n",
1651 			    can_sleep ? "" : "non-", rc, ifp->if_flags,
1652 			    ifp->if_drv_flags);
1653 			return (rc);
1654 		}
1655 
1656 		if (ifp->if_flags & IFF_UP) {
1657 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1658 				flags = vi->if_flags;
1659 				if ((ifp->if_flags ^ flags) &
1660 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1661 					if (can_sleep == 1) {
1662 						end_synchronized_op(sc, 0);
1663 						can_sleep = 0;
1664 						goto redo_sifflags;
1665 					}
1666 					rc = update_mac_settings(ifp,
1667 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1668 				}
1669 			} else {
1670 				if (can_sleep == 0) {
1671 					end_synchronized_op(sc, LOCK_HELD);
1672 					can_sleep = 1;
1673 					goto redo_sifflags;
1674 				}
1675 				rc = cxgbe_init_synchronized(vi);
1676 			}
1677 			vi->if_flags = ifp->if_flags;
1678 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1679 			if (can_sleep == 0) {
1680 				end_synchronized_op(sc, LOCK_HELD);
1681 				can_sleep = 1;
1682 				goto redo_sifflags;
1683 			}
1684 			rc = cxgbe_uninit_synchronized(vi);
1685 		}
1686 		end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1687 		break;
1688 
1689 	case SIOCADDMULTI:
1690 	case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1691 		rc = begin_synchronized_op(sc, vi, HOLD_LOCK, "t4multi");
1692 		if (rc)
1693 			return (rc);
1694 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1695 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1696 		end_synchronized_op(sc, LOCK_HELD);
1697 		break;
1698 
1699 	case SIOCSIFCAP:
1700 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1701 		if (rc)
1702 			return (rc);
1703 
1704 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1705 		if (mask & IFCAP_TXCSUM) {
1706 			ifp->if_capenable ^= IFCAP_TXCSUM;
1707 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1708 
1709 			if (IFCAP_TSO4 & ifp->if_capenable &&
1710 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1711 				ifp->if_capenable &= ~IFCAP_TSO4;
1712 				if_printf(ifp,
1713 				    "tso4 disabled due to -txcsum.\n");
1714 			}
1715 		}
1716 		if (mask & IFCAP_TXCSUM_IPV6) {
1717 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1718 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1719 
1720 			if (IFCAP_TSO6 & ifp->if_capenable &&
1721 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1722 				ifp->if_capenable &= ~IFCAP_TSO6;
1723 				if_printf(ifp,
1724 				    "tso6 disabled due to -txcsum6.\n");
1725 			}
1726 		}
1727 		if (mask & IFCAP_RXCSUM)
1728 			ifp->if_capenable ^= IFCAP_RXCSUM;
1729 		if (mask & IFCAP_RXCSUM_IPV6)
1730 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1731 
1732 		/*
1733 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1734 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1735 		 * sending a TSO request our way, so it's sufficient to toggle
1736 		 * IFCAP_TSOx only.
1737 		 */
1738 		if (mask & IFCAP_TSO4) {
1739 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1740 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1741 				if_printf(ifp, "enable txcsum first.\n");
1742 				rc = EAGAIN;
1743 				goto fail;
1744 			}
1745 			ifp->if_capenable ^= IFCAP_TSO4;
1746 		}
1747 		if (mask & IFCAP_TSO6) {
1748 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1749 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1750 				if_printf(ifp, "enable txcsum6 first.\n");
1751 				rc = EAGAIN;
1752 				goto fail;
1753 			}
1754 			ifp->if_capenable ^= IFCAP_TSO6;
1755 		}
1756 		if (mask & IFCAP_LRO) {
1757 #if defined(INET) || defined(INET6)
1758 			int i;
1759 			struct sge_rxq *rxq;
1760 
1761 			ifp->if_capenable ^= IFCAP_LRO;
1762 			for_each_rxq(vi, i, rxq) {
1763 				if (ifp->if_capenable & IFCAP_LRO)
1764 					rxq->iq.flags |= IQ_LRO_ENABLED;
1765 				else
1766 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1767 			}
1768 #endif
1769 		}
1770 #ifdef TCP_OFFLOAD
1771 		if (mask & IFCAP_TOE) {
1772 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1773 
1774 			rc = toe_capability(vi, enable);
1775 			if (rc != 0)
1776 				goto fail;
1777 
1778 			ifp->if_capenable ^= mask;
1779 		}
1780 #endif
1781 		if (mask & IFCAP_VLAN_HWTAGGING) {
1782 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1783 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1784 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1785 		}
1786 		if (mask & IFCAP_VLAN_MTU) {
1787 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1788 
1789 			/* Need to find out how to disable auto-mtu-inflation */
1790 		}
1791 		if (mask & IFCAP_VLAN_HWTSO)
1792 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1793 		if (mask & IFCAP_VLAN_HWCSUM)
1794 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1795 
1796 #ifdef VLAN_CAPABILITIES
1797 		VLAN_CAPABILITIES(ifp);
1798 #endif
1799 fail:
1800 		end_synchronized_op(sc, 0);
1801 		break;
1802 
1803 	case SIOCSIFMEDIA:
1804 	case SIOCGIFMEDIA:
1805 	case SIOCGIFXMEDIA:
1806 		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1807 		break;
1808 
1809 	case SIOCGI2C: {
1810 		struct ifi2creq i2c;
1811 
1812 		rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
1813 		if (rc != 0)
1814 			break;
1815 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1816 			rc = EPERM;
1817 			break;
1818 		}
1819 		if (i2c.len > sizeof(i2c.data)) {
1820 			rc = EINVAL;
1821 			break;
1822 		}
1823 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1824 		if (rc)
1825 			return (rc);
1826 		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
1827 		    i2c.offset, i2c.len, &i2c.data[0]);
1828 		end_synchronized_op(sc, 0);
1829 		if (rc == 0)
1830 			rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
1831 		break;
1832 	}
1833 
1834 	default:
1835 		rc = ether_ioctl(ifp, cmd, data);
1836 	}
1837 
1838 	return (rc);
1839 }
1840 
1841 static int
1842 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1843 {
1844 	struct vi_info *vi = ifp->if_softc;
1845 	struct port_info *pi = vi->pi;
1846 	struct adapter *sc = pi->adapter;
1847 	struct sge_txq *txq;
1848 	void *items[1];
1849 	int rc;
1850 
1851 	M_ASSERTPKTHDR(m);
1852 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1853 
1854 	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1855 		m_freem(m);
1856 		return (ENETDOWN);
1857 	}
1858 
1859 	rc = parse_pkt(sc, &m);
1860 	if (__predict_false(rc != 0)) {
1861 		MPASS(m == NULL);			/* was freed already */
1862 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1863 		return (rc);
1864 	}
1865 
1866 	/* Select a txq. */
1867 	txq = &sc->sge.txq[vi->first_txq];
1868 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1869 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1870 		    vi->rsrv_noflowq);
1871 
1872 	items[0] = m;
1873 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1874 	if (__predict_false(rc != 0))
1875 		m_freem(m);
1876 
1877 	return (rc);
1878 }
1879 
1880 static void
1881 cxgbe_qflush(struct ifnet *ifp)
1882 {
1883 	struct vi_info *vi = ifp->if_softc;
1884 	struct sge_txq *txq;
1885 	int i;
1886 
1887 	/* queues do not exist if !VI_INIT_DONE. */
1888 	if (vi->flags & VI_INIT_DONE) {
1889 		for_each_txq(vi, i, txq) {
1890 			TXQ_LOCK(txq);
1891 			txq->eq.flags |= EQ_QFLUSH;
1892 			TXQ_UNLOCK(txq);
1893 			while (!mp_ring_is_idle(txq->r)) {
1894 				mp_ring_check_drainage(txq->r, 0);
1895 				pause("qflush", 1);
1896 			}
1897 			TXQ_LOCK(txq);
1898 			txq->eq.flags &= ~EQ_QFLUSH;
1899 			TXQ_UNLOCK(txq);
1900 		}
1901 	}
1902 	if_qflush(ifp);
1903 }
1904 
1905 static uint64_t
1906 vi_get_counter(struct ifnet *ifp, ift_counter c)
1907 {
1908 	struct vi_info *vi = ifp->if_softc;
1909 	struct fw_vi_stats_vf *s = &vi->stats;
1910 
1911 	vi_refresh_stats(vi->pi->adapter, vi);
1912 
1913 	switch (c) {
1914 	case IFCOUNTER_IPACKETS:
1915 		return (s->rx_bcast_frames + s->rx_mcast_frames +
1916 		    s->rx_ucast_frames);
1917 	case IFCOUNTER_IERRORS:
1918 		return (s->rx_err_frames);
1919 	case IFCOUNTER_OPACKETS:
1920 		return (s->tx_bcast_frames + s->tx_mcast_frames +
1921 		    s->tx_ucast_frames + s->tx_offload_frames);
1922 	case IFCOUNTER_OERRORS:
1923 		return (s->tx_drop_frames);
1924 	case IFCOUNTER_IBYTES:
1925 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
1926 		    s->rx_ucast_bytes);
1927 	case IFCOUNTER_OBYTES:
1928 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
1929 		    s->tx_ucast_bytes + s->tx_offload_bytes);
1930 	case IFCOUNTER_IMCASTS:
1931 		return (s->rx_mcast_frames);
1932 	case IFCOUNTER_OMCASTS:
1933 		return (s->tx_mcast_frames);
1934 	case IFCOUNTER_OQDROPS: {
1935 		uint64_t drops;
1936 
1937 		drops = 0;
1938 		if (vi->flags & VI_INIT_DONE) {
1939 			int i;
1940 			struct sge_txq *txq;
1941 
1942 			for_each_txq(vi, i, txq)
1943 				drops += counter_u64_fetch(txq->r->drops);
1944 		}
1945 
1946 		return (drops);
1947 
1948 	}
1949 
1950 	default:
1951 		return (if_get_counter_default(ifp, c));
1952 	}
1953 }
1954 
1955 uint64_t
1956 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
1957 {
1958 	struct vi_info *vi = ifp->if_softc;
1959 	struct port_info *pi = vi->pi;
1960 	struct adapter *sc = pi->adapter;
1961 	struct port_stats *s = &pi->stats;
1962 
1963 	if (pi->nvi > 1 || sc->flags & IS_VF)
1964 		return (vi_get_counter(ifp, c));
1965 
1966 	cxgbe_refresh_stats(sc, pi);
1967 
1968 	switch (c) {
1969 	case IFCOUNTER_IPACKETS:
1970 		return (s->rx_frames);
1971 
1972 	case IFCOUNTER_IERRORS:
1973 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
1974 		    s->rx_fcs_err + s->rx_len_err);
1975 
1976 	case IFCOUNTER_OPACKETS:
1977 		return (s->tx_frames);
1978 
1979 	case IFCOUNTER_OERRORS:
1980 		return (s->tx_error_frames);
1981 
1982 	case IFCOUNTER_IBYTES:
1983 		return (s->rx_octets);
1984 
1985 	case IFCOUNTER_OBYTES:
1986 		return (s->tx_octets);
1987 
1988 	case IFCOUNTER_IMCASTS:
1989 		return (s->rx_mcast_frames);
1990 
1991 	case IFCOUNTER_OMCASTS:
1992 		return (s->tx_mcast_frames);
1993 
1994 	case IFCOUNTER_IQDROPS:
1995 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
1996 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
1997 		    s->rx_trunc3 + pi->tnl_cong_drops);
1998 
1999 	case IFCOUNTER_OQDROPS: {
2000 		uint64_t drops;
2001 
2002 		drops = s->tx_drop;
2003 		if (vi->flags & VI_INIT_DONE) {
2004 			int i;
2005 			struct sge_txq *txq;
2006 
2007 			for_each_txq(vi, i, txq)
2008 				drops += counter_u64_fetch(txq->r->drops);
2009 		}
2010 
2011 		return (drops);
2012 
2013 	}
2014 
2015 	default:
2016 		return (if_get_counter_default(ifp, c));
2017 	}
2018 }
2019 
2020 static int
2021 cxgbe_media_change(struct ifnet *ifp)
2022 {
2023 	struct vi_info *vi = ifp->if_softc;
2024 
2025 	device_printf(vi->dev, "%s unimplemented.\n", __func__);
2026 
2027 	return (EOPNOTSUPP);
2028 }
2029 
2030 static void
2031 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2032 {
2033 	struct vi_info *vi = ifp->if_softc;
2034 	struct port_info *pi = vi->pi;
2035 	struct ifmedia_entry *cur;
2036 	struct link_config *lc = &pi->link_cfg;
2037 
2038 	/*
2039 	 * If all the interfaces are administratively down the firmware does not
2040 	 * report transceiver changes.  Refresh port info here so that ifconfig
2041 	 * displays accurate information at all times.
2042 	 */
2043 	if (begin_synchronized_op(pi->adapter, NULL, SLEEP_OK | INTR_OK,
2044 	    "t4med") == 0) {
2045 		PORT_LOCK(pi);
2046 		if (pi->up_vis == 0) {
2047 			t4_update_port_info(pi);
2048 			build_medialist(pi, &pi->media);
2049 		}
2050 		PORT_UNLOCK(pi);
2051 		end_synchronized_op(pi->adapter, 0);
2052 	}
2053 
2054 	ifmr->ifm_status = IFM_AVALID;
2055 	if (lc->link_ok == 0)
2056 		return;
2057 
2058 	ifmr->ifm_status |= IFM_ACTIVE;
2059 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2060 	if (lc->fc & PAUSE_RX)
2061 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2062 	if (lc->fc & PAUSE_TX)
2063 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2064 
2065 	/* active and current will differ iff current media is autoselect. */
2066 	cur = pi->media.ifm_cur;
2067 	if (cur != NULL && IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
2068 		return;
2069 
2070 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2071 	if (lc->fc & PAUSE_RX)
2072 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2073 	if (lc->fc & PAUSE_TX)
2074 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2075 	switch (lc->speed) {
2076 	case 10000:
2077 		ifmr->ifm_active |= IFM_10G_T;
2078 		break;
2079 	case 1000:
2080 		ifmr->ifm_active |= IFM_1000_T;
2081 		break;
2082 	case 100:
2083 		ifmr->ifm_active |= IFM_100_TX;
2084 		break;
2085 	case 10:
2086 		ifmr->ifm_active |= IFM_10_T;
2087 		break;
2088 	default:
2089 		device_printf(vi->dev, "link up but speed unknown (%u)\n",
2090 		    lc->speed);
2091 	}
2092 }
2093 
2094 static int
2095 vcxgbe_probe(device_t dev)
2096 {
2097 	char buf[128];
2098 	struct vi_info *vi = device_get_softc(dev);
2099 
2100 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2101 	    vi - vi->pi->vi);
2102 	device_set_desc_copy(dev, buf);
2103 
2104 	return (BUS_PROBE_DEFAULT);
2105 }
2106 
2107 static int
2108 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2109 {
2110 	int func, index, rc;
2111 	uint32_t param, val;
2112 
2113 	ASSERT_SYNCHRONIZED_OP(sc);
2114 
2115 	index = vi - pi->vi;
2116 	MPASS(index > 0);	/* This function deals with _extra_ VIs only */
2117 	KASSERT(index < nitems(vi_mac_funcs),
2118 	    ("%s: VI %s doesn't have a MAC func", __func__,
2119 	    device_get_nameunit(vi->dev)));
2120 	func = vi_mac_funcs[index];
2121 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2122 	    vi->hw_addr, &vi->rss_size, func, 0);
2123 	if (rc < 0) {
2124 		device_printf(vi->dev, "failed to allocate virtual interface %d"
2125 		    "for port %d: %d\n", index, pi->port_id, -rc);
2126 		return (-rc);
2127 	}
2128 	vi->viid = rc;
2129 	if (chip_id(sc) <= CHELSIO_T5)
2130 		vi->smt_idx = (rc & 0x7f) << 1;
2131 	else
2132 		vi->smt_idx = (rc & 0x7f);
2133 
2134 	if (vi->rss_size == 1) {
2135 		/*
2136 		 * This VI didn't get a slice of the RSS table.  Reduce the
2137 		 * number of VIs being created (hw.cxgbe.num_vis) or modify the
2138 		 * configuration file (nvi, rssnvi for this PF) if this is a
2139 		 * problem.
2140 		 */
2141 		device_printf(vi->dev, "RSS table not available.\n");
2142 		vi->rss_base = 0xffff;
2143 
2144 		return (0);
2145 	}
2146 
2147 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2148 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2149 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2150 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2151 	if (rc)
2152 		vi->rss_base = 0xffff;
2153 	else {
2154 		MPASS((val >> 16) == vi->rss_size);
2155 		vi->rss_base = val & 0xffff;
2156 	}
2157 
2158 	return (0);
2159 }
2160 
2161 static int
2162 vcxgbe_attach(device_t dev)
2163 {
2164 	struct vi_info *vi;
2165 	struct port_info *pi;
2166 	struct adapter *sc;
2167 	int rc;
2168 
2169 	vi = device_get_softc(dev);
2170 	pi = vi->pi;
2171 	sc = pi->adapter;
2172 
2173 	rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2174 	if (rc)
2175 		return (rc);
2176 	rc = alloc_extra_vi(sc, pi, vi);
2177 	end_synchronized_op(sc, 0);
2178 	if (rc)
2179 		return (rc);
2180 
2181 	rc = cxgbe_vi_attach(dev, vi);
2182 	if (rc) {
2183 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2184 		return (rc);
2185 	}
2186 	return (0);
2187 }
2188 
2189 static int
2190 vcxgbe_detach(device_t dev)
2191 {
2192 	struct vi_info *vi;
2193 	struct adapter *sc;
2194 
2195 	vi = device_get_softc(dev);
2196 	sc = vi->pi->adapter;
2197 
2198 	doom_vi(sc, vi);
2199 
2200 	cxgbe_vi_detach(vi);
2201 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2202 
2203 	end_synchronized_op(sc, 0);
2204 
2205 	return (0);
2206 }
2207 
2208 void
2209 t4_fatal_err(struct adapter *sc)
2210 {
2211 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2212 	t4_intr_disable(sc);
2213 	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2214 	    device_get_nameunit(sc->dev));
2215 	if (t4_panic_on_fatal_err)
2216 		panic("panic requested on fatal error");
2217 }
2218 
2219 void
2220 t4_add_adapter(struct adapter *sc)
2221 {
2222 	sx_xlock(&t4_list_lock);
2223 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2224 	sx_xunlock(&t4_list_lock);
2225 }
2226 
2227 int
2228 t4_map_bars_0_and_4(struct adapter *sc)
2229 {
2230 	sc->regs_rid = PCIR_BAR(0);
2231 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2232 	    &sc->regs_rid, RF_ACTIVE);
2233 	if (sc->regs_res == NULL) {
2234 		device_printf(sc->dev, "cannot map registers.\n");
2235 		return (ENXIO);
2236 	}
2237 	sc->bt = rman_get_bustag(sc->regs_res);
2238 	sc->bh = rman_get_bushandle(sc->regs_res);
2239 	sc->mmio_len = rman_get_size(sc->regs_res);
2240 	setbit(&sc->doorbells, DOORBELL_KDB);
2241 
2242 	sc->msix_rid = PCIR_BAR(4);
2243 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2244 	    &sc->msix_rid, RF_ACTIVE);
2245 	if (sc->msix_res == NULL) {
2246 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2247 		return (ENXIO);
2248 	}
2249 
2250 	return (0);
2251 }
2252 
2253 int
2254 t4_map_bar_2(struct adapter *sc)
2255 {
2256 
2257 	/*
2258 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2259 	 * to map it if RDMA is disabled.
2260 	 */
2261 	if (is_t4(sc) && sc->rdmacaps == 0)
2262 		return (0);
2263 
2264 	sc->udbs_rid = PCIR_BAR(2);
2265 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2266 	    &sc->udbs_rid, RF_ACTIVE);
2267 	if (sc->udbs_res == NULL) {
2268 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2269 		return (ENXIO);
2270 	}
2271 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2272 
2273 	if (chip_id(sc) >= CHELSIO_T5) {
2274 		setbit(&sc->doorbells, DOORBELL_UDB);
2275 #if defined(__i386__) || defined(__amd64__)
2276 		if (t5_write_combine) {
2277 			int rc, mode;
2278 
2279 			/*
2280 			 * Enable write combining on BAR2.  This is the
2281 			 * userspace doorbell BAR and is split into 128B
2282 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2283 			 * with an egress queue.  The first 64B has the doorbell
2284 			 * and the second 64B can be used to submit a tx work
2285 			 * request with an implicit doorbell.
2286 			 */
2287 
2288 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2289 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2290 			if (rc == 0) {
2291 				clrbit(&sc->doorbells, DOORBELL_UDB);
2292 				setbit(&sc->doorbells, DOORBELL_WCWR);
2293 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2294 			} else {
2295 				device_printf(sc->dev,
2296 				    "couldn't enable write combining: %d\n",
2297 				    rc);
2298 			}
2299 
2300 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2301 			t4_write_reg(sc, A_SGE_STAT_CFG,
2302 			    V_STATSOURCE_T5(7) | mode);
2303 		}
2304 #endif
2305 	}
2306 	sc->iwt.wc_en = isset(&sc->doorbells, DOORBELL_UDBWC) ? 1 : 0;
2307 
2308 	return (0);
2309 }
2310 
2311 struct memwin_init {
2312 	uint32_t base;
2313 	uint32_t aperture;
2314 };
2315 
2316 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2317 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2318 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2319 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2320 };
2321 
2322 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2323 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2324 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2325 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2326 };
2327 
2328 static void
2329 setup_memwin(struct adapter *sc)
2330 {
2331 	const struct memwin_init *mw_init;
2332 	struct memwin *mw;
2333 	int i;
2334 	uint32_t bar0;
2335 
2336 	if (is_t4(sc)) {
2337 		/*
2338 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2339 		 * mechanism.  Works from within PCI passthrough environments
2340 		 * too, where rman_get_start() can return a different value.  We
2341 		 * need to program the T4 memory window decoders with the actual
2342 		 * addresses that will be coming across the PCIe link.
2343 		 */
2344 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2345 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2346 
2347 		mw_init = &t4_memwin[0];
2348 	} else {
2349 		/* T5+ use the relative offset inside the PCIe BAR */
2350 		bar0 = 0;
2351 
2352 		mw_init = &t5_memwin[0];
2353 	}
2354 
2355 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2356 		rw_init(&mw->mw_lock, "memory window access");
2357 		mw->mw_base = mw_init->base;
2358 		mw->mw_aperture = mw_init->aperture;
2359 		mw->mw_curpos = 0;
2360 		t4_write_reg(sc,
2361 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2362 		    (mw->mw_base + bar0) | V_BIR(0) |
2363 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2364 		rw_wlock(&mw->mw_lock);
2365 		position_memwin(sc, i, 0);
2366 		rw_wunlock(&mw->mw_lock);
2367 	}
2368 
2369 	/* flush */
2370 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2371 }
2372 
2373 /*
2374  * Positions the memory window at the given address in the card's address space.
2375  * There are some alignment requirements and the actual position may be at an
2376  * address prior to the requested address.  mw->mw_curpos always has the actual
2377  * position of the window.
2378  */
2379 static void
2380 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2381 {
2382 	struct memwin *mw;
2383 	uint32_t pf;
2384 	uint32_t reg;
2385 
2386 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2387 	mw = &sc->memwin[idx];
2388 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2389 
2390 	if (is_t4(sc)) {
2391 		pf = 0;
2392 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2393 	} else {
2394 		pf = V_PFNUM(sc->pf);
2395 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2396 	}
2397 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2398 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2399 	t4_read_reg(sc, reg);	/* flush */
2400 }
2401 
2402 int
2403 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2404     int len, int rw)
2405 {
2406 	struct memwin *mw;
2407 	uint32_t mw_end, v;
2408 
2409 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2410 
2411 	/* Memory can only be accessed in naturally aligned 4 byte units */
2412 	if (addr & 3 || len & 3 || len <= 0)
2413 		return (EINVAL);
2414 
2415 	mw = &sc->memwin[idx];
2416 	while (len > 0) {
2417 		rw_rlock(&mw->mw_lock);
2418 		mw_end = mw->mw_curpos + mw->mw_aperture;
2419 		if (addr >= mw_end || addr < mw->mw_curpos) {
2420 			/* Will need to reposition the window */
2421 			if (!rw_try_upgrade(&mw->mw_lock)) {
2422 				rw_runlock(&mw->mw_lock);
2423 				rw_wlock(&mw->mw_lock);
2424 			}
2425 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2426 			position_memwin(sc, idx, addr);
2427 			rw_downgrade(&mw->mw_lock);
2428 			mw_end = mw->mw_curpos + mw->mw_aperture;
2429 		}
2430 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2431 		while (addr < mw_end && len > 0) {
2432 			if (rw == 0) {
2433 				v = t4_read_reg(sc, mw->mw_base + addr -
2434 				    mw->mw_curpos);
2435 				*val++ = le32toh(v);
2436 			} else {
2437 				v = *val++;
2438 				t4_write_reg(sc, mw->mw_base + addr -
2439 				    mw->mw_curpos, htole32(v));
2440 			}
2441 			addr += 4;
2442 			len -= 4;
2443 		}
2444 		rw_runlock(&mw->mw_lock);
2445 	}
2446 
2447 	return (0);
2448 }
2449 
2450 int
2451 alloc_atid_tab(struct tid_info *t, int flags)
2452 {
2453 	int i;
2454 
2455 	MPASS(t->natids > 0);
2456 	MPASS(t->atid_tab == NULL);
2457 
2458 	t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
2459 	    M_ZERO | flags);
2460 	if (t->atid_tab == NULL)
2461 		return (ENOMEM);
2462 	mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
2463 	t->afree = t->atid_tab;
2464 	t->atids_in_use = 0;
2465 	for (i = 1; i < t->natids; i++)
2466 		t->atid_tab[i - 1].next = &t->atid_tab[i];
2467 	t->atid_tab[t->natids - 1].next = NULL;
2468 
2469 	return (0);
2470 }
2471 
2472 void
2473 free_atid_tab(struct tid_info *t)
2474 {
2475 
2476 	KASSERT(t->atids_in_use == 0,
2477 	    ("%s: %d atids still in use.", __func__, t->atids_in_use));
2478 
2479 	if (mtx_initialized(&t->atid_lock))
2480 		mtx_destroy(&t->atid_lock);
2481 	free(t->atid_tab, M_CXGBE);
2482 	t->atid_tab = NULL;
2483 }
2484 
2485 int
2486 alloc_atid(struct adapter *sc, void *ctx)
2487 {
2488 	struct tid_info *t = &sc->tids;
2489 	int atid = -1;
2490 
2491 	mtx_lock(&t->atid_lock);
2492 	if (t->afree) {
2493 		union aopen_entry *p = t->afree;
2494 
2495 		atid = p - t->atid_tab;
2496 		MPASS(atid <= M_TID_TID);
2497 		t->afree = p->next;
2498 		p->data = ctx;
2499 		t->atids_in_use++;
2500 	}
2501 	mtx_unlock(&t->atid_lock);
2502 	return (atid);
2503 }
2504 
2505 void *
2506 lookup_atid(struct adapter *sc, int atid)
2507 {
2508 	struct tid_info *t = &sc->tids;
2509 
2510 	return (t->atid_tab[atid].data);
2511 }
2512 
2513 void
2514 free_atid(struct adapter *sc, int atid)
2515 {
2516 	struct tid_info *t = &sc->tids;
2517 	union aopen_entry *p = &t->atid_tab[atid];
2518 
2519 	mtx_lock(&t->atid_lock);
2520 	p->next = t->afree;
2521 	t->afree = p;
2522 	t->atids_in_use--;
2523 	mtx_unlock(&t->atid_lock);
2524 }
2525 
2526 static void
2527 queue_tid_release(struct adapter *sc, int tid)
2528 {
2529 
2530 	CXGBE_UNIMPLEMENTED("deferred tid release");
2531 }
2532 
2533 void
2534 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
2535 {
2536 	struct wrqe *wr;
2537 	struct cpl_tid_release *req;
2538 
2539 	wr = alloc_wrqe(sizeof(*req), ctrlq);
2540 	if (wr == NULL) {
2541 		queue_tid_release(sc, tid);	/* defer */
2542 		return;
2543 	}
2544 	req = wrtod(wr);
2545 
2546 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
2547 
2548 	t4_wrq_tx(sc, wr);
2549 }
2550 
2551 static int
2552 t4_range_cmp(const void *a, const void *b)
2553 {
2554 	return ((const struct t4_range *)a)->start -
2555 	       ((const struct t4_range *)b)->start;
2556 }
2557 
2558 /*
2559  * Verify that the memory range specified by the addr/len pair is valid within
2560  * the card's address space.
2561  */
2562 static int
2563 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2564 {
2565 	struct t4_range mem_ranges[4], *r, *next;
2566 	uint32_t em, addr_len;
2567 	int i, n, remaining;
2568 
2569 	/* Memory can only be accessed in naturally aligned 4 byte units */
2570 	if (addr & 3 || len & 3 || len <= 0)
2571 		return (EINVAL);
2572 
2573 	/* Enabled memories */
2574 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2575 
2576 	r = &mem_ranges[0];
2577 	n = 0;
2578 	bzero(r, sizeof(mem_ranges));
2579 	if (em & F_EDRAM0_ENABLE) {
2580 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2581 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2582 		if (r->size > 0) {
2583 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2584 			if (addr >= r->start &&
2585 			    addr + len <= r->start + r->size)
2586 				return (0);
2587 			r++;
2588 			n++;
2589 		}
2590 	}
2591 	if (em & F_EDRAM1_ENABLE) {
2592 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2593 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2594 		if (r->size > 0) {
2595 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2596 			if (addr >= r->start &&
2597 			    addr + len <= r->start + r->size)
2598 				return (0);
2599 			r++;
2600 			n++;
2601 		}
2602 	}
2603 	if (em & F_EXT_MEM_ENABLE) {
2604 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2605 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2606 		if (r->size > 0) {
2607 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2608 			if (addr >= r->start &&
2609 			    addr + len <= r->start + r->size)
2610 				return (0);
2611 			r++;
2612 			n++;
2613 		}
2614 	}
2615 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2616 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2617 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2618 		if (r->size > 0) {
2619 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2620 			if (addr >= r->start &&
2621 			    addr + len <= r->start + r->size)
2622 				return (0);
2623 			r++;
2624 			n++;
2625 		}
2626 	}
2627 	MPASS(n <= nitems(mem_ranges));
2628 
2629 	if (n > 1) {
2630 		/* Sort and merge the ranges. */
2631 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2632 
2633 		/* Start from index 0 and examine the next n - 1 entries. */
2634 		r = &mem_ranges[0];
2635 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2636 
2637 			MPASS(r->size > 0);	/* r is a valid entry. */
2638 			next = r + 1;
2639 			MPASS(next->size > 0);	/* and so is the next one. */
2640 
2641 			while (r->start + r->size >= next->start) {
2642 				/* Merge the next one into the current entry. */
2643 				r->size = max(r->start + r->size,
2644 				    next->start + next->size) - r->start;
2645 				n--;	/* One fewer entry in total. */
2646 				if (--remaining == 0)
2647 					goto done;	/* short circuit */
2648 				next++;
2649 			}
2650 			if (next != r + 1) {
2651 				/*
2652 				 * Some entries were merged into r and next
2653 				 * points to the first valid entry that couldn't
2654 				 * be merged.
2655 				 */
2656 				MPASS(next->size > 0);	/* must be valid */
2657 				memcpy(r + 1, next, remaining * sizeof(*r));
2658 #ifdef INVARIANTS
2659 				/*
2660 				 * This so that the foo->size assertion in the
2661 				 * next iteration of the loop do the right
2662 				 * thing for entries that were pulled up and are
2663 				 * no longer valid.
2664 				 */
2665 				MPASS(n < nitems(mem_ranges));
2666 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2667 				    sizeof(struct t4_range));
2668 #endif
2669 			}
2670 		}
2671 done:
2672 		/* Done merging the ranges. */
2673 		MPASS(n > 0);
2674 		r = &mem_ranges[0];
2675 		for (i = 0; i < n; i++, r++) {
2676 			if (addr >= r->start &&
2677 			    addr + len <= r->start + r->size)
2678 				return (0);
2679 		}
2680 	}
2681 
2682 	return (EFAULT);
2683 }
2684 
2685 static int
2686 fwmtype_to_hwmtype(int mtype)
2687 {
2688 
2689 	switch (mtype) {
2690 	case FW_MEMTYPE_EDC0:
2691 		return (MEM_EDC0);
2692 	case FW_MEMTYPE_EDC1:
2693 		return (MEM_EDC1);
2694 	case FW_MEMTYPE_EXTMEM:
2695 		return (MEM_MC0);
2696 	case FW_MEMTYPE_EXTMEM1:
2697 		return (MEM_MC1);
2698 	default:
2699 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2700 	}
2701 }
2702 
2703 /*
2704  * Verify that the memory range specified by the memtype/offset/len pair is
2705  * valid and lies entirely within the memtype specified.  The global address of
2706  * the start of the range is returned in addr.
2707  */
2708 static int
2709 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2710     uint32_t *addr)
2711 {
2712 	uint32_t em, addr_len, maddr;
2713 
2714 	/* Memory can only be accessed in naturally aligned 4 byte units */
2715 	if (off & 3 || len & 3 || len == 0)
2716 		return (EINVAL);
2717 
2718 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2719 	switch (fwmtype_to_hwmtype(mtype)) {
2720 	case MEM_EDC0:
2721 		if (!(em & F_EDRAM0_ENABLE))
2722 			return (EINVAL);
2723 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2724 		maddr = G_EDRAM0_BASE(addr_len) << 20;
2725 		break;
2726 	case MEM_EDC1:
2727 		if (!(em & F_EDRAM1_ENABLE))
2728 			return (EINVAL);
2729 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2730 		maddr = G_EDRAM1_BASE(addr_len) << 20;
2731 		break;
2732 	case MEM_MC:
2733 		if (!(em & F_EXT_MEM_ENABLE))
2734 			return (EINVAL);
2735 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2736 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2737 		break;
2738 	case MEM_MC1:
2739 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2740 			return (EINVAL);
2741 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2742 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2743 		break;
2744 	default:
2745 		return (EINVAL);
2746 	}
2747 
2748 	*addr = maddr + off;	/* global address */
2749 	return (validate_mem_range(sc, *addr, len));
2750 }
2751 
2752 static int
2753 fixup_devlog_params(struct adapter *sc)
2754 {
2755 	struct devlog_params *dparams = &sc->params.devlog;
2756 	int rc;
2757 
2758 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2759 	    dparams->size, &dparams->addr);
2760 
2761 	return (rc);
2762 }
2763 
2764 static void
2765 update_nirq(struct intrs_and_queues *iaq, int nports)
2766 {
2767 	int extra = T4_EXTRA_INTR;
2768 
2769 	iaq->nirq = extra;
2770 	iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
2771 	iaq->nirq += nports * (iaq->num_vis - 1) *
2772 	    max(iaq->nrxq_vi, iaq->nnmrxq_vi);
2773 	iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
2774 }
2775 
2776 /*
2777  * Adjust requirements to fit the number of interrupts available.
2778  */
2779 static void
2780 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
2781     int navail)
2782 {
2783 	int old_nirq;
2784 	const int nports = sc->params.nports;
2785 
2786 	MPASS(nports > 0);
2787 	MPASS(navail > 0);
2788 
2789 	bzero(iaq, sizeof(*iaq));
2790 	iaq->intr_type = itype;
2791 	iaq->num_vis = t4_num_vis;
2792 	iaq->ntxq = t4_ntxq;
2793 	iaq->ntxq_vi = t4_ntxq_vi;
2794 	iaq->nrxq = t4_nrxq;
2795 	iaq->nrxq_vi = t4_nrxq_vi;
2796 #ifdef TCP_OFFLOAD
2797 	if (is_offload(sc)) {
2798 		iaq->nofldtxq = t4_nofldtxq;
2799 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
2800 		iaq->nofldrxq = t4_nofldrxq;
2801 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
2802 	}
2803 #endif
2804 #ifdef DEV_NETMAP
2805 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
2806 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
2807 #endif
2808 
2809 	update_nirq(iaq, nports);
2810 	if (iaq->nirq <= navail &&
2811 	    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2812 		/*
2813 		 * This is the normal case -- there are enough interrupts for
2814 		 * everything.
2815 		 */
2816 		goto done;
2817 	}
2818 
2819 	/*
2820 	 * If extra VIs have been configured try reducing their count and see if
2821 	 * that works.
2822 	 */
2823 	while (iaq->num_vis > 1) {
2824 		iaq->num_vis--;
2825 		update_nirq(iaq, nports);
2826 		if (iaq->nirq <= navail &&
2827 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2828 			device_printf(sc->dev, "virtual interfaces per port "
2829 			    "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
2830 			    "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
2831 			    "itype %d, navail %u, nirq %d.\n",
2832 			    iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
2833 			    iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
2834 			    itype, navail, iaq->nirq);
2835 			goto done;
2836 		}
2837 	}
2838 
2839 	/*
2840 	 * Extra VIs will not be created.  Log a message if they were requested.
2841 	 */
2842 	MPASS(iaq->num_vis == 1);
2843 	iaq->ntxq_vi = iaq->nrxq_vi = 0;
2844 	iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
2845 	iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
2846 	if (iaq->num_vis != t4_num_vis) {
2847 		device_printf(sc->dev, "extra virtual interfaces disabled.  "
2848 		    "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
2849 		    "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
2850 		    iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
2851 		    iaq->nnmrxq_vi, itype, navail, iaq->nirq);
2852 	}
2853 
2854 	/*
2855 	 * Keep reducing the number of NIC rx queues to the next lower power of
2856 	 * 2 (for even RSS distribution) and halving the TOE rx queues and see
2857 	 * if that works.
2858 	 */
2859 	do {
2860 		if (iaq->nrxq > 1) {
2861 			do {
2862 				iaq->nrxq--;
2863 			} while (!powerof2(iaq->nrxq));
2864 		}
2865 		if (iaq->nofldrxq > 1)
2866 			iaq->nofldrxq >>= 1;
2867 
2868 		old_nirq = iaq->nirq;
2869 		update_nirq(iaq, nports);
2870 		if (iaq->nirq <= navail &&
2871 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2872 			device_printf(sc->dev, "running with reduced number of "
2873 			    "rx queues because of shortage of interrupts.  "
2874 			    "nrxq=%u, nofldrxq=%u.  "
2875 			    "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
2876 			    iaq->nofldrxq, itype, navail, iaq->nirq);
2877 			goto done;
2878 		}
2879 	} while (old_nirq != iaq->nirq);
2880 
2881 	/* One interrupt for everything.  Ugh. */
2882 	device_printf(sc->dev, "running with minimal number of queues.  "
2883 	    "itype %d, navail %u.\n", itype, navail);
2884 	iaq->nirq = 1;
2885 	MPASS(iaq->nrxq == 1);
2886 	iaq->ntxq = 1;
2887 	if (iaq->nofldrxq > 1)
2888 		iaq->nofldtxq = 1;
2889 done:
2890 	MPASS(iaq->num_vis > 0);
2891 	if (iaq->num_vis > 1) {
2892 		MPASS(iaq->nrxq_vi > 0);
2893 		MPASS(iaq->ntxq_vi > 0);
2894 	}
2895 	MPASS(iaq->nirq > 0);
2896 	MPASS(iaq->nrxq > 0);
2897 	MPASS(iaq->ntxq > 0);
2898 	if (itype == INTR_MSI) {
2899 		MPASS(powerof2(iaq->nirq));
2900 	}
2901 }
2902 
2903 static int
2904 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
2905 {
2906 	int rc, itype, navail, nalloc;
2907 
2908 	for (itype = INTR_MSIX; itype; itype >>= 1) {
2909 
2910 		if ((itype & t4_intr_types) == 0)
2911 			continue;	/* not allowed */
2912 
2913 		if (itype == INTR_MSIX)
2914 			navail = pci_msix_count(sc->dev);
2915 		else if (itype == INTR_MSI)
2916 			navail = pci_msi_count(sc->dev);
2917 		else
2918 			navail = 1;
2919 restart:
2920 		if (navail == 0)
2921 			continue;
2922 
2923 		calculate_iaq(sc, iaq, itype, navail);
2924 		nalloc = iaq->nirq;
2925 		rc = 0;
2926 		if (itype == INTR_MSIX)
2927 			rc = pci_alloc_msix(sc->dev, &nalloc);
2928 		else if (itype == INTR_MSI)
2929 			rc = pci_alloc_msi(sc->dev, &nalloc);
2930 
2931 		if (rc == 0 && nalloc > 0) {
2932 			if (nalloc == iaq->nirq)
2933 				return (0);
2934 
2935 			/*
2936 			 * Didn't get the number requested.  Use whatever number
2937 			 * the kernel is willing to allocate.
2938 			 */
2939 			device_printf(sc->dev, "fewer vectors than requested, "
2940 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2941 			    itype, iaq->nirq, nalloc);
2942 			pci_release_msi(sc->dev);
2943 			navail = nalloc;
2944 			goto restart;
2945 		}
2946 
2947 		device_printf(sc->dev,
2948 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2949 		    itype, rc, iaq->nirq, nalloc);
2950 	}
2951 
2952 	device_printf(sc->dev,
2953 	    "failed to find a usable interrupt type.  "
2954 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2955 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2956 
2957 	return (ENXIO);
2958 }
2959 
2960 #define FW_VERSION(chip) ( \
2961     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2962     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2963     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2964     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2965 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2966 
2967 struct fw_info {
2968 	uint8_t chip;
2969 	char *kld_name;
2970 	char *fw_mod_name;
2971 	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
2972 } fw_info[] = {
2973 	{
2974 		.chip = CHELSIO_T4,
2975 		.kld_name = "t4fw_cfg",
2976 		.fw_mod_name = "t4fw",
2977 		.fw_hdr = {
2978 			.chip = FW_HDR_CHIP_T4,
2979 			.fw_ver = htobe32_const(FW_VERSION(T4)),
2980 			.intfver_nic = FW_INTFVER(T4, NIC),
2981 			.intfver_vnic = FW_INTFVER(T4, VNIC),
2982 			.intfver_ofld = FW_INTFVER(T4, OFLD),
2983 			.intfver_ri = FW_INTFVER(T4, RI),
2984 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2985 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
2986 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2987 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
2988 		},
2989 	}, {
2990 		.chip = CHELSIO_T5,
2991 		.kld_name = "t5fw_cfg",
2992 		.fw_mod_name = "t5fw",
2993 		.fw_hdr = {
2994 			.chip = FW_HDR_CHIP_T5,
2995 			.fw_ver = htobe32_const(FW_VERSION(T5)),
2996 			.intfver_nic = FW_INTFVER(T5, NIC),
2997 			.intfver_vnic = FW_INTFVER(T5, VNIC),
2998 			.intfver_ofld = FW_INTFVER(T5, OFLD),
2999 			.intfver_ri = FW_INTFVER(T5, RI),
3000 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
3001 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
3002 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
3003 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
3004 		},
3005 	}, {
3006 		.chip = CHELSIO_T6,
3007 		.kld_name = "t6fw_cfg",
3008 		.fw_mod_name = "t6fw",
3009 		.fw_hdr = {
3010 			.chip = FW_HDR_CHIP_T6,
3011 			.fw_ver = htobe32_const(FW_VERSION(T6)),
3012 			.intfver_nic = FW_INTFVER(T6, NIC),
3013 			.intfver_vnic = FW_INTFVER(T6, VNIC),
3014 			.intfver_ofld = FW_INTFVER(T6, OFLD),
3015 			.intfver_ri = FW_INTFVER(T6, RI),
3016 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3017 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
3018 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3019 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
3020 		},
3021 	}
3022 };
3023 
3024 static struct fw_info *
3025 find_fw_info(int chip)
3026 {
3027 	int i;
3028 
3029 	for (i = 0; i < nitems(fw_info); i++) {
3030 		if (fw_info[i].chip == chip)
3031 			return (&fw_info[i]);
3032 	}
3033 	return (NULL);
3034 }
3035 
3036 /*
3037  * Is the given firmware API compatible with the one the driver was compiled
3038  * with?
3039  */
3040 static int
3041 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
3042 {
3043 
3044 	/* short circuit if it's the exact same firmware version */
3045 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
3046 		return (1);
3047 
3048 	/*
3049 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
3050 	 * features that are supported in the driver.
3051 	 */
3052 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
3053 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
3054 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
3055 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
3056 		return (1);
3057 #undef SAME_INTF
3058 
3059 	return (0);
3060 }
3061 
3062 /*
3063  * The firmware in the KLD is usable, but should it be installed?  This routine
3064  * explains itself in detail if it indicates the KLD firmware should be
3065  * installed.
3066  */
3067 static int
3068 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
3069 {
3070 	const char *reason;
3071 
3072 	if (!card_fw_usable) {
3073 		reason = "incompatible or unusable";
3074 		goto install;
3075 	}
3076 
3077 	if (k > c) {
3078 		reason = "older than the version bundled with this driver";
3079 		goto install;
3080 	}
3081 
3082 	if (t4_fw_install == 2 && k != c) {
3083 		reason = "different than the version bundled with this driver";
3084 		goto install;
3085 	}
3086 
3087 	return (0);
3088 
3089 install:
3090 	if (t4_fw_install == 0) {
3091 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3092 		    "but the driver is prohibited from installing a different "
3093 		    "firmware on the card.\n",
3094 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3095 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3096 
3097 		return (0);
3098 	}
3099 
3100 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3101 	    "installing firmware %u.%u.%u.%u on card.\n",
3102 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3103 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3104 	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3105 	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3106 
3107 	return (1);
3108 }
3109 
3110 /*
3111  * Establish contact with the firmware and determine if we are the master driver
3112  * or not, and whether we are responsible for chip initialization.
3113  */
3114 static int
3115 prep_firmware(struct adapter *sc)
3116 {
3117 	const struct firmware *fw = NULL, *default_cfg;
3118 	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
3119 	enum dev_state state;
3120 	struct fw_info *fw_info;
3121 	struct fw_hdr *card_fw;		/* fw on the card */
3122 	const struct fw_hdr *kld_fw;	/* fw in the KLD */
3123 	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
3124 					   against */
3125 
3126 	/* This is the firmware whose headers the driver was compiled against */
3127 	fw_info = find_fw_info(chip_id(sc));
3128 	if (fw_info == NULL) {
3129 		device_printf(sc->dev,
3130 		    "unable to look up firmware information for chip %d.\n",
3131 		    chip_id(sc));
3132 		return (EINVAL);
3133 	}
3134 	drv_fw = &fw_info->fw_hdr;
3135 
3136 	/*
3137 	 * The firmware KLD contains many modules.  The KLD name is also the
3138 	 * name of the module that contains the default config file.
3139 	 */
3140 	default_cfg = firmware_get(fw_info->kld_name);
3141 
3142 	/* This is the firmware in the KLD */
3143 	fw = firmware_get(fw_info->fw_mod_name);
3144 	if (fw != NULL) {
3145 		kld_fw = (const void *)fw->data;
3146 		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3147 	} else {
3148 		kld_fw = NULL;
3149 		kld_fw_usable = 0;
3150 	}
3151 
3152 	/* Read the header of the firmware on the card */
3153 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3154 	rc = -t4_read_flash(sc, FLASH_FW_START,
3155 	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
3156 	if (rc == 0) {
3157 		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
3158 		if (card_fw->fw_ver == be32toh(0xffffffff)) {
3159 			uint32_t d = be32toh(kld_fw->fw_ver);
3160 
3161 			if (!kld_fw_usable) {
3162 				device_printf(sc->dev,
3163 				    "no firmware on the card and no usable "
3164 				    "firmware bundled with the driver.\n");
3165 				rc = EIO;
3166 				goto done;
3167 			} else if (t4_fw_install == 0) {
3168 				device_printf(sc->dev,
3169 				    "no firmware on the card and the driver "
3170 				    "is prohibited from installing new "
3171 				    "firmware.\n");
3172 				rc = EIO;
3173 				goto done;
3174 			}
3175 
3176 			device_printf(sc->dev, "no firmware on the card, "
3177 			    "installing firmware %d.%d.%d.%d\n",
3178 			    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3179 			    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3180 			rc = t4_fw_forceinstall(sc, fw->data, fw->datasize);
3181 			if (rc < 0) {
3182 				rc = -rc;
3183 				device_printf(sc->dev,
3184 				    "firmware install failed: %d.\n", rc);
3185 				goto done;
3186 			}
3187 			memcpy(card_fw, kld_fw, sizeof(*card_fw));
3188 			card_fw_usable = 1;
3189 			need_fw_reset = 0;
3190 		}
3191 	} else {
3192 		device_printf(sc->dev,
3193 		    "Unable to read card's firmware header: %d\n", rc);
3194 		card_fw_usable = 0;
3195 	}
3196 
3197 	/* Contact firmware. */
3198 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3199 	if (rc < 0 || state == DEV_STATE_ERR) {
3200 		rc = -rc;
3201 		device_printf(sc->dev,
3202 		    "failed to connect to the firmware: %d, %d.\n", rc, state);
3203 		goto done;
3204 	}
3205 	pf = rc;
3206 	if (pf == sc->mbox)
3207 		sc->flags |= MASTER_PF;
3208 	else if (state == DEV_STATE_UNINIT) {
3209 		/*
3210 		 * We didn't get to be the master so we definitely won't be
3211 		 * configuring the chip.  It's a bug if someone else hasn't
3212 		 * configured it already.
3213 		 */
3214 		device_printf(sc->dev, "couldn't be master(%d), "
3215 		    "device not already initialized either(%d).\n", rc, state);
3216 		rc = EPROTO;
3217 		goto done;
3218 	}
3219 
3220 	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3221 	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3222 		/*
3223 		 * Common case: the firmware on the card is an exact match and
3224 		 * the KLD is an exact match too, or the KLD is
3225 		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
3226 		 * here -- use cxgbetool loadfw if you want to reinstall the
3227 		 * same firmware as the one on the card.
3228 		 */
3229 	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3230 	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3231 	    be32toh(card_fw->fw_ver))) {
3232 
3233 		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3234 		if (rc != 0) {
3235 			device_printf(sc->dev,
3236 			    "failed to install firmware: %d\n", rc);
3237 			goto done;
3238 		}
3239 
3240 		/* Installed successfully, update the cached header too. */
3241 		memcpy(card_fw, kld_fw, sizeof(*card_fw));
3242 		card_fw_usable = 1;
3243 		need_fw_reset = 0;	/* already reset as part of load_fw */
3244 	}
3245 
3246 	if (!card_fw_usable) {
3247 		uint32_t d, c, k;
3248 
3249 		d = ntohl(drv_fw->fw_ver);
3250 		c = ntohl(card_fw->fw_ver);
3251 		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3252 
3253 		device_printf(sc->dev, "Cannot find a usable firmware: "
3254 		    "fw_install %d, chip state %d, "
3255 		    "driver compiled with %d.%d.%d.%d, "
3256 		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3257 		    t4_fw_install, state,
3258 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3259 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3260 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3261 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3262 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3263 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3264 		rc = EINVAL;
3265 		goto done;
3266 	}
3267 
3268 	/* Reset device */
3269 	if (need_fw_reset &&
3270 	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3271 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3272 		if (rc != ETIMEDOUT && rc != EIO)
3273 			t4_fw_bye(sc, sc->mbox);
3274 		goto done;
3275 	}
3276 	sc->flags |= FW_OK;
3277 
3278 	rc = get_params__pre_init(sc);
3279 	if (rc != 0)
3280 		goto done; /* error message displayed already */
3281 
3282 	/* Partition adapter resources as specified in the config file. */
3283 	if (state == DEV_STATE_UNINIT) {
3284 
3285 		KASSERT(sc->flags & MASTER_PF,
3286 		    ("%s: trying to change chip settings when not master.",
3287 		    __func__));
3288 
3289 		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3290 		if (rc != 0)
3291 			goto done;	/* error message displayed already */
3292 
3293 		t4_tweak_chip_settings(sc);
3294 
3295 		/* get basic stuff going */
3296 		rc = -t4_fw_initialize(sc, sc->mbox);
3297 		if (rc != 0) {
3298 			device_printf(sc->dev, "fw init failed: %d.\n", rc);
3299 			goto done;
3300 		}
3301 	} else {
3302 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3303 		sc->cfcsum = 0;
3304 	}
3305 
3306 done:
3307 	free(card_fw, M_CXGBE);
3308 	if (fw != NULL)
3309 		firmware_put(fw, FIRMWARE_UNLOAD);
3310 	if (default_cfg != NULL)
3311 		firmware_put(default_cfg, FIRMWARE_UNLOAD);
3312 
3313 	return (rc);
3314 }
3315 
3316 #define FW_PARAM_DEV(param) \
3317 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3318 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3319 #define FW_PARAM_PFVF(param) \
3320 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3321 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3322 
3323 /*
3324  * Partition chip resources for use between various PFs, VFs, etc.
3325  */
3326 static int
3327 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3328     const char *name_prefix)
3329 {
3330 	const struct firmware *cfg = NULL;
3331 	int rc = 0;
3332 	struct fw_caps_config_cmd caps;
3333 	uint32_t mtype, moff, finicsum, cfcsum;
3334 
3335 	/*
3336 	 * Figure out what configuration file to use.  Pick the default config
3337 	 * file for the card if the user hasn't specified one explicitly.
3338 	 */
3339 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3340 	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3341 		/* Card specific overrides go here. */
3342 		if (pci_get_device(sc->dev) == 0x440a)
3343 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3344 		if (is_fpga(sc))
3345 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3346 	}
3347 
3348 	/*
3349 	 * We need to load another module if the profile is anything except
3350 	 * "default" or "flash".
3351 	 */
3352 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3353 	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3354 		char s[32];
3355 
3356 		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3357 		cfg = firmware_get(s);
3358 		if (cfg == NULL) {
3359 			if (default_cfg != NULL) {
3360 				device_printf(sc->dev,
3361 				    "unable to load module \"%s\" for "
3362 				    "configuration profile \"%s\", will use "
3363 				    "the default config file instead.\n",
3364 				    s, sc->cfg_file);
3365 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3366 				    "%s", DEFAULT_CF);
3367 			} else {
3368 				device_printf(sc->dev,
3369 				    "unable to load module \"%s\" for "
3370 				    "configuration profile \"%s\", will use "
3371 				    "the config file on the card's flash "
3372 				    "instead.\n", s, sc->cfg_file);
3373 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3374 				    "%s", FLASH_CF);
3375 			}
3376 		}
3377 	}
3378 
3379 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3380 	    default_cfg == NULL) {
3381 		device_printf(sc->dev,
3382 		    "default config file not available, will use the config "
3383 		    "file on the card's flash instead.\n");
3384 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3385 	}
3386 
3387 	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3388 		u_int cflen;
3389 		const uint32_t *cfdata;
3390 		uint32_t param, val, addr;
3391 
3392 		KASSERT(cfg != NULL || default_cfg != NULL,
3393 		    ("%s: no config to upload", __func__));
3394 
3395 		/*
3396 		 * Ask the firmware where it wants us to upload the config file.
3397 		 */
3398 		param = FW_PARAM_DEV(CF);
3399 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3400 		if (rc != 0) {
3401 			/* No support for config file?  Shouldn't happen. */
3402 			device_printf(sc->dev,
3403 			    "failed to query config file location: %d.\n", rc);
3404 			goto done;
3405 		}
3406 		mtype = G_FW_PARAMS_PARAM_Y(val);
3407 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3408 
3409 		/*
3410 		 * XXX: sheer laziness.  We deliberately added 4 bytes of
3411 		 * useless stuffing/comments at the end of the config file so
3412 		 * it's ok to simply throw away the last remaining bytes when
3413 		 * the config file is not an exact multiple of 4.  This also
3414 		 * helps with the validate_mt_off_len check.
3415 		 */
3416 		if (cfg != NULL) {
3417 			cflen = cfg->datasize & ~3;
3418 			cfdata = cfg->data;
3419 		} else {
3420 			cflen = default_cfg->datasize & ~3;
3421 			cfdata = default_cfg->data;
3422 		}
3423 
3424 		if (cflen > FLASH_CFG_MAX_SIZE) {
3425 			device_printf(sc->dev,
3426 			    "config file too long (%d, max allowed is %d).  "
3427 			    "Will try to use the config on the card, if any.\n",
3428 			    cflen, FLASH_CFG_MAX_SIZE);
3429 			goto use_config_on_flash;
3430 		}
3431 
3432 		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3433 		if (rc != 0) {
3434 			device_printf(sc->dev,
3435 			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3436 			    "Will try to use the config on the card, if any.\n",
3437 			    __func__, mtype, moff, cflen, rc);
3438 			goto use_config_on_flash;
3439 		}
3440 		write_via_memwin(sc, 2, addr, cfdata, cflen);
3441 	} else {
3442 use_config_on_flash:
3443 		mtype = FW_MEMTYPE_FLASH;
3444 		moff = t4_flash_cfg_addr(sc);
3445 	}
3446 
3447 	bzero(&caps, sizeof(caps));
3448 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3449 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3450 	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3451 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3452 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3453 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3454 	if (rc != 0) {
3455 		device_printf(sc->dev,
3456 		    "failed to pre-process config file: %d "
3457 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3458 		goto done;
3459 	}
3460 
3461 	finicsum = be32toh(caps.finicsum);
3462 	cfcsum = be32toh(caps.cfcsum);
3463 	if (finicsum != cfcsum) {
3464 		device_printf(sc->dev,
3465 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3466 		    finicsum, cfcsum);
3467 	}
3468 	sc->cfcsum = cfcsum;
3469 
3470 #define LIMIT_CAPS(x) do { \
3471 	caps.x &= htobe16(t4_##x##_allowed); \
3472 } while (0)
3473 
3474 	/*
3475 	 * Let the firmware know what features will (not) be used so it can tune
3476 	 * things accordingly.
3477 	 */
3478 	LIMIT_CAPS(nbmcaps);
3479 	LIMIT_CAPS(linkcaps);
3480 	LIMIT_CAPS(switchcaps);
3481 	LIMIT_CAPS(niccaps);
3482 	LIMIT_CAPS(toecaps);
3483 	LIMIT_CAPS(rdmacaps);
3484 	LIMIT_CAPS(cryptocaps);
3485 	LIMIT_CAPS(iscsicaps);
3486 	LIMIT_CAPS(fcoecaps);
3487 #undef LIMIT_CAPS
3488 
3489 	if (caps.niccaps & htobe16(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
3490 		/*
3491 		 * TOE and hashfilters are mutually exclusive.  It is a config
3492 		 * file or firmware bug if both are reported as available.  Try
3493 		 * to cope with the situation in non-debug builds by disabling
3494 		 * TOE.
3495 		 */
3496 		MPASS(caps.toecaps == 0);
3497 
3498 		caps.toecaps = 0;
3499 		caps.rdmacaps = 0;
3500 		caps.iscsicaps = 0;
3501 	}
3502 
3503 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3504 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3505 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3506 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3507 	if (rc != 0) {
3508 		device_printf(sc->dev,
3509 		    "failed to process config file: %d.\n", rc);
3510 	}
3511 done:
3512 	if (cfg != NULL)
3513 		firmware_put(cfg, FIRMWARE_UNLOAD);
3514 	return (rc);
3515 }
3516 
3517 /*
3518  * Retrieve parameters that are needed (or nice to have) very early.
3519  */
3520 static int
3521 get_params__pre_init(struct adapter *sc)
3522 {
3523 	int rc;
3524 	uint32_t param[2], val[2];
3525 
3526 	t4_get_version_info(sc);
3527 
3528 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3529 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3530 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3531 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3532 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3533 
3534 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3535 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3536 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3537 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3538 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3539 
3540 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3541 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3542 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3543 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3544 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3545 
3546 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3547 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3548 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3549 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3550 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3551 
3552 	param[0] = FW_PARAM_DEV(PORTVEC);
3553 	param[1] = FW_PARAM_DEV(CCLK);
3554 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3555 	if (rc != 0) {
3556 		device_printf(sc->dev,
3557 		    "failed to query parameters (pre_init): %d.\n", rc);
3558 		return (rc);
3559 	}
3560 
3561 	sc->params.portvec = val[0];
3562 	sc->params.nports = bitcount32(val[0]);
3563 	sc->params.vpd.cclk = val[1];
3564 
3565 	/* Read device log parameters. */
3566 	rc = -t4_init_devlog_params(sc, 1);
3567 	if (rc == 0)
3568 		fixup_devlog_params(sc);
3569 	else {
3570 		device_printf(sc->dev,
3571 		    "failed to get devlog parameters: %d.\n", rc);
3572 		rc = 0;	/* devlog isn't critical for device operation */
3573 	}
3574 
3575 	return (rc);
3576 }
3577 
3578 /*
3579  * Retrieve various parameters that are of interest to the driver.  The device
3580  * has been initialized by the firmware at this point.
3581  */
3582 static int
3583 get_params__post_init(struct adapter *sc)
3584 {
3585 	int rc;
3586 	uint32_t param[7], val[7];
3587 	struct fw_caps_config_cmd caps;
3588 
3589 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3590 	param[1] = FW_PARAM_PFVF(EQ_START);
3591 	param[2] = FW_PARAM_PFVF(FILTER_START);
3592 	param[3] = FW_PARAM_PFVF(FILTER_END);
3593 	param[4] = FW_PARAM_PFVF(L2T_START);
3594 	param[5] = FW_PARAM_PFVF(L2T_END);
3595 	param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
3596 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
3597 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
3598 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
3599 	if (rc != 0) {
3600 		device_printf(sc->dev,
3601 		    "failed to query parameters (post_init): %d.\n", rc);
3602 		return (rc);
3603 	}
3604 
3605 	sc->sge.iq_start = val[0];
3606 	sc->sge.eq_start = val[1];
3607 	sc->tids.ftid_base = val[2];
3608 	sc->tids.nftids = val[3] - val[2] + 1;
3609 	sc->params.ftid_min = val[2];
3610 	sc->params.ftid_max = val[3];
3611 	sc->vres.l2t.start = val[4];
3612 	sc->vres.l2t.size = val[5] - val[4] + 1;
3613 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3614 	    ("%s: L2 table size (%u) larger than expected (%u)",
3615 	    __func__, sc->vres.l2t.size, L2T_SIZE));
3616 	sc->params.core_vdd = val[6];
3617 
3618 	/*
3619 	 * MPSBGMAP is queried separately because only recent firmwares support
3620 	 * it as a parameter and we don't want the compound query above to fail
3621 	 * on older firmwares.
3622 	 */
3623 	param[0] = FW_PARAM_DEV(MPSBGMAP);
3624 	val[0] = 0;
3625 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3626 	if (rc == 0)
3627 		sc->params.mps_bg_map = val[0];
3628 	else
3629 		sc->params.mps_bg_map = 0;
3630 
3631 	/*
3632 	 * Determine whether the firmware supports the filter2 work request.
3633 	 * This is queried separately for the same reason as MPSBGMAP above.
3634 	 */
3635 	param[0] = FW_PARAM_DEV(FILTER2_WR);
3636 	val[0] = 0;
3637 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3638 	if (rc == 0)
3639 		sc->params.filter2_wr_support = val[0] != 0;
3640 	else
3641 		sc->params.filter2_wr_support = 0;
3642 
3643 	/* get capabilites */
3644 	bzero(&caps, sizeof(caps));
3645 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3646 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3647 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3648 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3649 	if (rc != 0) {
3650 		device_printf(sc->dev,
3651 		    "failed to get card capabilities: %d.\n", rc);
3652 		return (rc);
3653 	}
3654 
3655 #define READ_CAPS(x) do { \
3656 	sc->x = htobe16(caps.x); \
3657 } while (0)
3658 	READ_CAPS(nbmcaps);
3659 	READ_CAPS(linkcaps);
3660 	READ_CAPS(switchcaps);
3661 	READ_CAPS(niccaps);
3662 	READ_CAPS(toecaps);
3663 	READ_CAPS(rdmacaps);
3664 	READ_CAPS(cryptocaps);
3665 	READ_CAPS(iscsicaps);
3666 	READ_CAPS(fcoecaps);
3667 
3668 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_HASHFILTER) {
3669 		MPASS(chip_id(sc) > CHELSIO_T4);
3670 		MPASS(sc->toecaps == 0);
3671 		sc->toecaps = 0;
3672 
3673 		param[0] = FW_PARAM_DEV(NTID);
3674 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3675 		if (rc != 0) {
3676 			device_printf(sc->dev,
3677 			    "failed to query HASHFILTER parameters: %d.\n", rc);
3678 			return (rc);
3679 		}
3680 		sc->tids.ntids = val[0];
3681 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3682 		sc->params.hash_filter = 1;
3683 	}
3684 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3685 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3686 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3687 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3688 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3689 		if (rc != 0) {
3690 			device_printf(sc->dev,
3691 			    "failed to query NIC parameters: %d.\n", rc);
3692 			return (rc);
3693 		}
3694 		sc->tids.etid_base = val[0];
3695 		sc->params.etid_min = val[0];
3696 		sc->tids.netids = val[1] - val[0] + 1;
3697 		sc->params.netids = sc->tids.netids;
3698 		sc->params.eo_wr_cred = val[2];
3699 		sc->params.ethoffload = 1;
3700 	}
3701 	if (sc->toecaps) {
3702 		/* query offload-related parameters */
3703 		param[0] = FW_PARAM_DEV(NTID);
3704 		param[1] = FW_PARAM_PFVF(SERVER_START);
3705 		param[2] = FW_PARAM_PFVF(SERVER_END);
3706 		param[3] = FW_PARAM_PFVF(TDDP_START);
3707 		param[4] = FW_PARAM_PFVF(TDDP_END);
3708 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3709 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3710 		if (rc != 0) {
3711 			device_printf(sc->dev,
3712 			    "failed to query TOE parameters: %d.\n", rc);
3713 			return (rc);
3714 		}
3715 		sc->tids.ntids = val[0];
3716 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3717 		sc->tids.stid_base = val[1];
3718 		sc->tids.nstids = val[2] - val[1] + 1;
3719 		sc->vres.ddp.start = val[3];
3720 		sc->vres.ddp.size = val[4] - val[3] + 1;
3721 		sc->params.ofldq_wr_cred = val[5];
3722 		sc->params.offload = 1;
3723 	} else {
3724 		/*
3725 		 * The firmware attempts memfree TOE configuration for -SO cards
3726 		 * and will report toecaps=0 if it runs out of resources (this
3727 		 * depends on the config file).  It may not report 0 for other
3728 		 * capabilities dependent on the TOE in this case.  Set them to
3729 		 * 0 here so that the driver doesn't bother tracking resources
3730 		 * that will never be used.
3731 		 */
3732 		sc->iscsicaps = 0;
3733 		sc->rdmacaps = 0;
3734 	}
3735 	if (sc->rdmacaps) {
3736 		param[0] = FW_PARAM_PFVF(STAG_START);
3737 		param[1] = FW_PARAM_PFVF(STAG_END);
3738 		param[2] = FW_PARAM_PFVF(RQ_START);
3739 		param[3] = FW_PARAM_PFVF(RQ_END);
3740 		param[4] = FW_PARAM_PFVF(PBL_START);
3741 		param[5] = FW_PARAM_PFVF(PBL_END);
3742 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3743 		if (rc != 0) {
3744 			device_printf(sc->dev,
3745 			    "failed to query RDMA parameters(1): %d.\n", rc);
3746 			return (rc);
3747 		}
3748 		sc->vres.stag.start = val[0];
3749 		sc->vres.stag.size = val[1] - val[0] + 1;
3750 		sc->vres.rq.start = val[2];
3751 		sc->vres.rq.size = val[3] - val[2] + 1;
3752 		sc->vres.pbl.start = val[4];
3753 		sc->vres.pbl.size = val[5] - val[4] + 1;
3754 
3755 		param[0] = FW_PARAM_PFVF(SQRQ_START);
3756 		param[1] = FW_PARAM_PFVF(SQRQ_END);
3757 		param[2] = FW_PARAM_PFVF(CQ_START);
3758 		param[3] = FW_PARAM_PFVF(CQ_END);
3759 		param[4] = FW_PARAM_PFVF(OCQ_START);
3760 		param[5] = FW_PARAM_PFVF(OCQ_END);
3761 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3762 		if (rc != 0) {
3763 			device_printf(sc->dev,
3764 			    "failed to query RDMA parameters(2): %d.\n", rc);
3765 			return (rc);
3766 		}
3767 		sc->vres.qp.start = val[0];
3768 		sc->vres.qp.size = val[1] - val[0] + 1;
3769 		sc->vres.cq.start = val[2];
3770 		sc->vres.cq.size = val[3] - val[2] + 1;
3771 		sc->vres.ocq.start = val[4];
3772 		sc->vres.ocq.size = val[5] - val[4] + 1;
3773 
3774 		param[0] = FW_PARAM_PFVF(SRQ_START);
3775 		param[1] = FW_PARAM_PFVF(SRQ_END);
3776 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
3777 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
3778 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
3779 		if (rc != 0) {
3780 			device_printf(sc->dev,
3781 			    "failed to query RDMA parameters(3): %d.\n", rc);
3782 			return (rc);
3783 		}
3784 		sc->vres.srq.start = val[0];
3785 		sc->vres.srq.size = val[1] - val[0] + 1;
3786 		sc->params.max_ordird_qp = val[2];
3787 		sc->params.max_ird_adapter = val[3];
3788 	}
3789 	if (sc->iscsicaps) {
3790 		param[0] = FW_PARAM_PFVF(ISCSI_START);
3791 		param[1] = FW_PARAM_PFVF(ISCSI_END);
3792 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3793 		if (rc != 0) {
3794 			device_printf(sc->dev,
3795 			    "failed to query iSCSI parameters: %d.\n", rc);
3796 			return (rc);
3797 		}
3798 		sc->vres.iscsi.start = val[0];
3799 		sc->vres.iscsi.size = val[1] - val[0] + 1;
3800 	}
3801 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
3802 		param[0] = FW_PARAM_PFVF(TLS_START);
3803 		param[1] = FW_PARAM_PFVF(TLS_END);
3804 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3805 		if (rc != 0) {
3806 			device_printf(sc->dev,
3807 			    "failed to query TLS parameters: %d.\n", rc);
3808 			return (rc);
3809 		}
3810 		sc->vres.key.start = val[0];
3811 		sc->vres.key.size = val[1] - val[0] + 1;
3812 	}
3813 
3814 	t4_init_sge_params(sc);
3815 
3816 	/*
3817 	 * We've got the params we wanted to query via the firmware.  Now grab
3818 	 * some others directly from the chip.
3819 	 */
3820 	rc = t4_read_chip_settings(sc);
3821 
3822 	return (rc);
3823 }
3824 
3825 static int
3826 set_params__post_init(struct adapter *sc)
3827 {
3828 	uint32_t param, val;
3829 #ifdef TCP_OFFLOAD
3830 	int i, v, shift;
3831 #endif
3832 
3833 	/* ask for encapsulated CPLs */
3834 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3835 	val = 1;
3836 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3837 
3838 #ifdef TCP_OFFLOAD
3839 	/*
3840 	 * Override the TOE timers with user provided tunables.  This is not the
3841 	 * recommended way to change the timers (the firmware config file is) so
3842 	 * these tunables are not documented.
3843 	 *
3844 	 * All the timer tunables are in microseconds.
3845 	 */
3846 	if (t4_toe_keepalive_idle != 0) {
3847 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
3848 		v &= M_KEEPALIVEIDLE;
3849 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
3850 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
3851 	}
3852 	if (t4_toe_keepalive_interval != 0) {
3853 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
3854 		v &= M_KEEPALIVEINTVL;
3855 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
3856 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
3857 	}
3858 	if (t4_toe_keepalive_count != 0) {
3859 		v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
3860 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
3861 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
3862 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
3863 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
3864 	}
3865 	if (t4_toe_rexmt_min != 0) {
3866 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
3867 		v &= M_RXTMIN;
3868 		t4_set_reg_field(sc, A_TP_RXT_MIN,
3869 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
3870 	}
3871 	if (t4_toe_rexmt_max != 0) {
3872 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
3873 		v &= M_RXTMAX;
3874 		t4_set_reg_field(sc, A_TP_RXT_MAX,
3875 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
3876 	}
3877 	if (t4_toe_rexmt_count != 0) {
3878 		v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
3879 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
3880 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
3881 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
3882 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
3883 	}
3884 	for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
3885 		if (t4_toe_rexmt_backoff[i] != -1) {
3886 			v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
3887 			shift = (i & 3) << 3;
3888 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
3889 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
3890 		}
3891 	}
3892 #endif
3893 	return (0);
3894 }
3895 
3896 #undef FW_PARAM_PFVF
3897 #undef FW_PARAM_DEV
3898 
3899 static void
3900 t4_set_desc(struct adapter *sc)
3901 {
3902 	char buf[128];
3903 	struct adapter_params *p = &sc->params;
3904 
3905 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
3906 
3907 	device_set_desc_copy(sc->dev, buf);
3908 }
3909 
3910 static void
3911 build_medialist(struct port_info *pi, struct ifmedia *media)
3912 {
3913 	int m;
3914 
3915 	PORT_LOCK_ASSERT_OWNED(pi);
3916 
3917 	ifmedia_removeall(media);
3918 
3919 	/*
3920 	 * XXX: Would it be better to ifmedia_add all 4 combinations of pause
3921 	 * settings for every speed instead of just txpause|rxpause?  ifconfig
3922 	 * media display looks much better if autoselect is the only case where
3923 	 * ifm_current is different from ifm_active.  If the user picks anything
3924 	 * except txpause|rxpause the display is ugly.
3925 	 */
3926 	m = IFM_ETHER | IFM_FDX | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
3927 
3928 	switch(pi->port_type) {
3929 	case FW_PORT_TYPE_BT_XFI:
3930 	case FW_PORT_TYPE_BT_XAUI:
3931 		ifmedia_add(media, m | IFM_10G_T, 0, NULL);
3932 		/* fall through */
3933 
3934 	case FW_PORT_TYPE_BT_SGMII:
3935 		ifmedia_add(media, m | IFM_1000_T, 0, NULL);
3936 		ifmedia_add(media, m | IFM_100_TX, 0, NULL);
3937 		ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL);
3938 		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
3939 		break;
3940 
3941 	case FW_PORT_TYPE_CX4:
3942 		ifmedia_add(media, m | IFM_10G_CX4, 0, NULL);
3943 		ifmedia_set(media, m | IFM_10G_CX4);
3944 		break;
3945 
3946 	case FW_PORT_TYPE_QSFP_10G:
3947 	case FW_PORT_TYPE_SFP:
3948 	case FW_PORT_TYPE_FIBER_XFI:
3949 	case FW_PORT_TYPE_FIBER_XAUI:
3950 		switch (pi->mod_type) {
3951 
3952 		case FW_PORT_MOD_TYPE_LR:
3953 			ifmedia_add(media, m | IFM_10G_LR, 0, NULL);
3954 			ifmedia_set(media, m | IFM_10G_LR);
3955 			break;
3956 
3957 		case FW_PORT_MOD_TYPE_SR:
3958 			ifmedia_add(media, m | IFM_10G_SR, 0, NULL);
3959 			ifmedia_set(media, m | IFM_10G_SR);
3960 			break;
3961 
3962 		case FW_PORT_MOD_TYPE_LRM:
3963 			ifmedia_add(media, m | IFM_10G_LRM, 0, NULL);
3964 			ifmedia_set(media, m | IFM_10G_LRM);
3965 			break;
3966 
3967 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3968 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3969 			ifmedia_add(media, m | IFM_10G_TWINAX, 0, NULL);
3970 			ifmedia_set(media, m | IFM_10G_TWINAX);
3971 			break;
3972 
3973 		case FW_PORT_MOD_TYPE_NONE:
3974 			m &= ~IFM_FDX;
3975 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3976 			ifmedia_set(media, m | IFM_NONE);
3977 			break;
3978 
3979 		case FW_PORT_MOD_TYPE_NA:
3980 		case FW_PORT_MOD_TYPE_ER:
3981 		default:
3982 			device_printf(pi->dev,
3983 			    "unknown port_type (%d), mod_type (%d)\n",
3984 			    pi->port_type, pi->mod_type);
3985 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3986 			ifmedia_set(media, m | IFM_UNKNOWN);
3987 			break;
3988 		}
3989 		break;
3990 
3991 	case FW_PORT_TYPE_CR_QSFP:
3992 	case FW_PORT_TYPE_SFP28:
3993 	case FW_PORT_TYPE_KR_SFP28:
3994 		switch (pi->mod_type) {
3995 
3996 		case FW_PORT_MOD_TYPE_SR:
3997 			ifmedia_add(media, m | IFM_25G_SR, 0, NULL);
3998 			ifmedia_set(media, m | IFM_25G_SR);
3999 			break;
4000 
4001 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
4002 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
4003 			ifmedia_add(media, m | IFM_25G_CR, 0, NULL);
4004 			ifmedia_set(media, m | IFM_25G_CR);
4005 			break;
4006 
4007 		case FW_PORT_MOD_TYPE_NONE:
4008 			m &= ~IFM_FDX;
4009 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
4010 			ifmedia_set(media, m | IFM_NONE);
4011 			break;
4012 
4013 		default:
4014 			device_printf(pi->dev,
4015 			    "unknown port_type (%d), mod_type (%d)\n",
4016 			    pi->port_type, pi->mod_type);
4017 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
4018 			ifmedia_set(media, m | IFM_UNKNOWN);
4019 			break;
4020 		}
4021 		break;
4022 
4023 	case FW_PORT_TYPE_QSFP:
4024 		switch (pi->mod_type) {
4025 
4026 		case FW_PORT_MOD_TYPE_LR:
4027 			ifmedia_add(media, m | IFM_40G_LR4, 0, NULL);
4028 			ifmedia_set(media, m | IFM_40G_LR4);
4029 			break;
4030 
4031 		case FW_PORT_MOD_TYPE_SR:
4032 			ifmedia_add(media, m | IFM_40G_SR4, 0, NULL);
4033 			ifmedia_set(media, m | IFM_40G_SR4);
4034 			break;
4035 
4036 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
4037 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
4038 			ifmedia_add(media, m | IFM_40G_CR4, 0, NULL);
4039 			ifmedia_set(media, m | IFM_40G_CR4);
4040 			break;
4041 
4042 		case FW_PORT_MOD_TYPE_NONE:
4043 			m &= ~IFM_FDX;
4044 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
4045 			ifmedia_set(media, m | IFM_NONE);
4046 			break;
4047 
4048 		default:
4049 			device_printf(pi->dev,
4050 			    "unknown port_type (%d), mod_type (%d)\n",
4051 			    pi->port_type, pi->mod_type);
4052 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
4053 			ifmedia_set(media, m | IFM_UNKNOWN);
4054 			break;
4055 		}
4056 		break;
4057 
4058 	case FW_PORT_TYPE_KR4_100G:
4059 	case FW_PORT_TYPE_CR4_QSFP:
4060 		switch (pi->mod_type) {
4061 
4062 		case FW_PORT_MOD_TYPE_LR:
4063 			ifmedia_add(media, m | IFM_100G_LR4, 0, NULL);
4064 			ifmedia_set(media, m | IFM_100G_LR4);
4065 			break;
4066 
4067 		case FW_PORT_MOD_TYPE_SR:
4068 			ifmedia_add(media, m | IFM_100G_SR4, 0, NULL);
4069 			ifmedia_set(media, m | IFM_100G_SR4);
4070 			break;
4071 
4072 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
4073 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
4074 			ifmedia_add(media, m | IFM_100G_CR4, 0, NULL);
4075 			ifmedia_set(media, m | IFM_100G_CR4);
4076 			break;
4077 
4078 		case FW_PORT_MOD_TYPE_NONE:
4079 			m &= ~IFM_FDX;
4080 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
4081 			ifmedia_set(media, m | IFM_NONE);
4082 			break;
4083 
4084 		default:
4085 			device_printf(pi->dev,
4086 			    "unknown port_type (%d), mod_type (%d)\n",
4087 			    pi->port_type, pi->mod_type);
4088 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
4089 			ifmedia_set(media, m | IFM_UNKNOWN);
4090 			break;
4091 		}
4092 		break;
4093 
4094 	default:
4095 		device_printf(pi->dev,
4096 		    "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
4097 		    pi->mod_type);
4098 		ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
4099 		ifmedia_set(media, m | IFM_UNKNOWN);
4100 		break;
4101 	}
4102 }
4103 
4104 /*
4105  * Update all the requested_* fields in the link config and then send a mailbox
4106  * command to apply the settings.
4107  */
4108 static void
4109 init_l1cfg(struct port_info *pi)
4110 {
4111 	struct adapter *sc = pi->adapter;
4112 	struct link_config *lc = &pi->link_cfg;
4113 	int rc;
4114 
4115 	ASSERT_SYNCHRONIZED_OP(sc);
4116 
4117 	lc->requested_speed = port_top_speed(pi);	/* in Gbps */
4118 	if (t4_autoneg != 0 && lc->supported & FW_PORT_CAP_ANEG) {
4119 		lc->requested_aneg = AUTONEG_ENABLE;
4120 	} else {
4121 		lc->requested_aneg = AUTONEG_DISABLE;
4122 	}
4123 
4124 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX);
4125 
4126 	if (t4_fec != -1) {
4127 		lc->requested_fec = t4_fec & (FEC_RS | FEC_BASER_RS |
4128 		    FEC_RESERVED);
4129 	} else {
4130 		/* Use the suggested value provided by the firmware in acaps */
4131 		if (lc->advertising & FW_PORT_CAP_FEC_RS)
4132 			lc->requested_fec = FEC_RS;
4133 		else if (lc->advertising & FW_PORT_CAP_FEC_BASER_RS)
4134 			lc->requested_fec = FEC_BASER_RS;
4135 		else
4136 			lc->requested_fec = 0;
4137 	}
4138 
4139 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
4140 	if (rc != 0) {
4141 		device_printf(pi->dev, "l1cfg failed: %d\n", rc);
4142 	} else {
4143 		lc->fc = lc->requested_fc;
4144 		lc->fec = lc->requested_fec;
4145 	}
4146 }
4147 
4148 #define FW_MAC_EXACT_CHUNK	7
4149 
4150 /*
4151  * Program the port's XGMAC based on parameters in ifnet.  The caller also
4152  * indicates which parameters should be programmed (the rest are left alone).
4153  */
4154 int
4155 update_mac_settings(struct ifnet *ifp, int flags)
4156 {
4157 	int rc = 0;
4158 	struct vi_info *vi = ifp->if_softc;
4159 	struct port_info *pi = vi->pi;
4160 	struct adapter *sc = pi->adapter;
4161 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4162 
4163 	ASSERT_SYNCHRONIZED_OP(sc);
4164 	KASSERT(flags, ("%s: not told what to update.", __func__));
4165 
4166 	if (flags & XGMAC_MTU)
4167 		mtu = ifp->if_mtu;
4168 
4169 	if (flags & XGMAC_PROMISC)
4170 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4171 
4172 	if (flags & XGMAC_ALLMULTI)
4173 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4174 
4175 	if (flags & XGMAC_VLANEX)
4176 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4177 
4178 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4179 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4180 		    allmulti, 1, vlanex, false);
4181 		if (rc) {
4182 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4183 			    rc);
4184 			return (rc);
4185 		}
4186 	}
4187 
4188 	if (flags & XGMAC_UCADDR) {
4189 		uint8_t ucaddr[ETHER_ADDR_LEN];
4190 
4191 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4192 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4193 		    ucaddr, true, true);
4194 		if (rc < 0) {
4195 			rc = -rc;
4196 			if_printf(ifp, "change_mac failed: %d\n", rc);
4197 			return (rc);
4198 		} else {
4199 			vi->xact_addr_filt = rc;
4200 			rc = 0;
4201 		}
4202 	}
4203 
4204 	if (flags & XGMAC_MCADDRS) {
4205 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4206 		int del = 1;
4207 		uint64_t hash = 0;
4208 		struct ifmultiaddr *ifma;
4209 		int i = 0, j;
4210 
4211 		if_maddr_rlock(ifp);
4212 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4213 			if (ifma->ifma_addr->sa_family != AF_LINK)
4214 				continue;
4215 			mcaddr[i] =
4216 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4217 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4218 			i++;
4219 
4220 			if (i == FW_MAC_EXACT_CHUNK) {
4221 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4222 				    del, i, mcaddr, NULL, &hash, 0);
4223 				if (rc < 0) {
4224 					rc = -rc;
4225 					for (j = 0; j < i; j++) {
4226 						if_printf(ifp,
4227 						    "failed to add mc address"
4228 						    " %02x:%02x:%02x:"
4229 						    "%02x:%02x:%02x rc=%d\n",
4230 						    mcaddr[j][0], mcaddr[j][1],
4231 						    mcaddr[j][2], mcaddr[j][3],
4232 						    mcaddr[j][4], mcaddr[j][5],
4233 						    rc);
4234 					}
4235 					goto mcfail;
4236 				}
4237 				del = 0;
4238 				i = 0;
4239 			}
4240 		}
4241 		if (i > 0) {
4242 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4243 			    mcaddr, NULL, &hash, 0);
4244 			if (rc < 0) {
4245 				rc = -rc;
4246 				for (j = 0; j < i; j++) {
4247 					if_printf(ifp,
4248 					    "failed to add mc address"
4249 					    " %02x:%02x:%02x:"
4250 					    "%02x:%02x:%02x rc=%d\n",
4251 					    mcaddr[j][0], mcaddr[j][1],
4252 					    mcaddr[j][2], mcaddr[j][3],
4253 					    mcaddr[j][4], mcaddr[j][5],
4254 					    rc);
4255 				}
4256 				goto mcfail;
4257 			}
4258 		}
4259 
4260 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4261 		if (rc != 0)
4262 			if_printf(ifp, "failed to set mc address hash: %d", rc);
4263 mcfail:
4264 		if_maddr_runlock(ifp);
4265 	}
4266 
4267 	return (rc);
4268 }
4269 
4270 /*
4271  * {begin|end}_synchronized_op must be called from the same thread.
4272  */
4273 int
4274 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4275     char *wmesg)
4276 {
4277 	int rc, pri;
4278 
4279 #ifdef WITNESS
4280 	/* the caller thinks it's ok to sleep, but is it really? */
4281 	if (flags & SLEEP_OK)
4282 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4283 		    "begin_synchronized_op");
4284 #endif
4285 
4286 	if (INTR_OK)
4287 		pri = PCATCH;
4288 	else
4289 		pri = 0;
4290 
4291 	ADAPTER_LOCK(sc);
4292 	for (;;) {
4293 
4294 		if (vi && IS_DOOMED(vi)) {
4295 			rc = ENXIO;
4296 			goto done;
4297 		}
4298 
4299 		if (!IS_BUSY(sc)) {
4300 			rc = 0;
4301 			break;
4302 		}
4303 
4304 		if (!(flags & SLEEP_OK)) {
4305 			rc = EBUSY;
4306 			goto done;
4307 		}
4308 
4309 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4310 			rc = EINTR;
4311 			goto done;
4312 		}
4313 	}
4314 
4315 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4316 	SET_BUSY(sc);
4317 #ifdef INVARIANTS
4318 	sc->last_op = wmesg;
4319 	sc->last_op_thr = curthread;
4320 	sc->last_op_flags = flags;
4321 #endif
4322 
4323 done:
4324 	if (!(flags & HOLD_LOCK) || rc)
4325 		ADAPTER_UNLOCK(sc);
4326 
4327 	return (rc);
4328 }
4329 
4330 /*
4331  * Tell if_ioctl and if_init that the VI is going away.  This is
4332  * special variant of begin_synchronized_op and must be paired with a
4333  * call to end_synchronized_op.
4334  */
4335 void
4336 doom_vi(struct adapter *sc, struct vi_info *vi)
4337 {
4338 
4339 	ADAPTER_LOCK(sc);
4340 	SET_DOOMED(vi);
4341 	wakeup(&sc->flags);
4342 	while (IS_BUSY(sc))
4343 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4344 	SET_BUSY(sc);
4345 #ifdef INVARIANTS
4346 	sc->last_op = "t4detach";
4347 	sc->last_op_thr = curthread;
4348 	sc->last_op_flags = 0;
4349 #endif
4350 	ADAPTER_UNLOCK(sc);
4351 }
4352 
4353 /*
4354  * {begin|end}_synchronized_op must be called from the same thread.
4355  */
4356 void
4357 end_synchronized_op(struct adapter *sc, int flags)
4358 {
4359 
4360 	if (flags & LOCK_HELD)
4361 		ADAPTER_LOCK_ASSERT_OWNED(sc);
4362 	else
4363 		ADAPTER_LOCK(sc);
4364 
4365 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
4366 	CLR_BUSY(sc);
4367 	wakeup(&sc->flags);
4368 	ADAPTER_UNLOCK(sc);
4369 }
4370 
4371 static int
4372 cxgbe_init_synchronized(struct vi_info *vi)
4373 {
4374 	struct port_info *pi = vi->pi;
4375 	struct adapter *sc = pi->adapter;
4376 	struct ifnet *ifp = vi->ifp;
4377 	int rc = 0, i;
4378 	struct sge_txq *txq;
4379 
4380 	ASSERT_SYNCHRONIZED_OP(sc);
4381 
4382 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
4383 		return (0);	/* already running */
4384 
4385 	if (!(sc->flags & FULL_INIT_DONE) &&
4386 	    ((rc = adapter_full_init(sc)) != 0))
4387 		return (rc);	/* error message displayed already */
4388 
4389 	if (!(vi->flags & VI_INIT_DONE) &&
4390 	    ((rc = vi_full_init(vi)) != 0))
4391 		return (rc); /* error message displayed already */
4392 
4393 	rc = update_mac_settings(ifp, XGMAC_ALL);
4394 	if (rc)
4395 		goto done;	/* error message displayed already */
4396 
4397 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
4398 	if (rc != 0) {
4399 		if_printf(ifp, "enable_vi failed: %d\n", rc);
4400 		goto done;
4401 	}
4402 
4403 	/*
4404 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
4405 	 * if this changes.
4406 	 */
4407 
4408 	for_each_txq(vi, i, txq) {
4409 		TXQ_LOCK(txq);
4410 		txq->eq.flags |= EQ_ENABLED;
4411 		TXQ_UNLOCK(txq);
4412 	}
4413 
4414 	/*
4415 	 * The first iq of the first port to come up is used for tracing.
4416 	 */
4417 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4418 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4419 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4420 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4421 		    V_QUEUENUMBER(sc->traceq));
4422 		pi->flags |= HAS_TRACEQ;
4423 	}
4424 
4425 	/* all ok */
4426 	PORT_LOCK(pi);
4427 	if (pi->up_vis++ == 0) {
4428 		t4_update_port_info(pi);
4429 		build_medialist(pi, &pi->media);
4430 		init_l1cfg(pi);
4431 	}
4432 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
4433 
4434 	if (pi->nvi > 1 || sc->flags & IS_VF)
4435 		callout_reset(&vi->tick, hz, vi_tick, vi);
4436 	else
4437 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4438 	PORT_UNLOCK(pi);
4439 done:
4440 	if (rc != 0)
4441 		cxgbe_uninit_synchronized(vi);
4442 
4443 	return (rc);
4444 }
4445 
4446 /*
4447  * Idempotent.
4448  */
4449 static int
4450 cxgbe_uninit_synchronized(struct vi_info *vi)
4451 {
4452 	struct port_info *pi = vi->pi;
4453 	struct adapter *sc = pi->adapter;
4454 	struct ifnet *ifp = vi->ifp;
4455 	int rc, i;
4456 	struct sge_txq *txq;
4457 
4458 	ASSERT_SYNCHRONIZED_OP(sc);
4459 
4460 	if (!(vi->flags & VI_INIT_DONE)) {
4461 		if (__predict_false(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4462 			KASSERT(0, ("uninited VI is running"));
4463 			if_printf(ifp, "uninited VI with running ifnet.  "
4464 			    "vi->flags 0x%016lx, if_flags 0x%08x, "
4465 			    "if_drv_flags 0x%08x\n", vi->flags, ifp->if_flags,
4466 			    ifp->if_drv_flags);
4467 		}
4468 		return (0);
4469 	}
4470 
4471 	/*
4472 	 * Disable the VI so that all its data in either direction is discarded
4473 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4474 	 * tick) intact as the TP can deliver negative advice or data that it's
4475 	 * holding in its RAM (for an offloaded connection) even after the VI is
4476 	 * disabled.
4477 	 */
4478 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4479 	if (rc) {
4480 		if_printf(ifp, "disable_vi failed: %d\n", rc);
4481 		return (rc);
4482 	}
4483 
4484 	for_each_txq(vi, i, txq) {
4485 		TXQ_LOCK(txq);
4486 		txq->eq.flags &= ~EQ_ENABLED;
4487 		TXQ_UNLOCK(txq);
4488 	}
4489 
4490 	PORT_LOCK(pi);
4491 	if (pi->nvi > 1 || sc->flags & IS_VF)
4492 		callout_stop(&vi->tick);
4493 	else
4494 		callout_stop(&pi->tick);
4495 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4496 		PORT_UNLOCK(pi);
4497 		return (0);
4498 	}
4499 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4500 	pi->up_vis--;
4501 	if (pi->up_vis > 0) {
4502 		PORT_UNLOCK(pi);
4503 		return (0);
4504 	}
4505 	PORT_UNLOCK(pi);
4506 
4507 	pi->link_cfg.link_ok = 0;
4508 	pi->link_cfg.speed = 0;
4509 	pi->link_cfg.link_down_rc = 255;
4510 	t4_os_link_changed(pi);
4511 	pi->old_link_cfg = pi->link_cfg;
4512 
4513 	return (0);
4514 }
4515 
4516 /*
4517  * It is ok for this function to fail midway and return right away.  t4_detach
4518  * will walk the entire sc->irq list and clean up whatever is valid.
4519  */
4520 int
4521 t4_setup_intr_handlers(struct adapter *sc)
4522 {
4523 	int rc, rid, p, q, v;
4524 	char s[8];
4525 	struct irq *irq;
4526 	struct port_info *pi;
4527 	struct vi_info *vi;
4528 	struct sge *sge = &sc->sge;
4529 	struct sge_rxq *rxq;
4530 #ifdef TCP_OFFLOAD
4531 	struct sge_ofld_rxq *ofld_rxq;
4532 #endif
4533 #ifdef DEV_NETMAP
4534 	struct sge_nm_rxq *nm_rxq;
4535 #endif
4536 #ifdef RSS
4537 	int nbuckets = rss_getnumbuckets();
4538 #endif
4539 
4540 	/*
4541 	 * Setup interrupts.
4542 	 */
4543 	irq = &sc->irq[0];
4544 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
4545 	if (forwarding_intr_to_fwq(sc))
4546 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4547 
4548 	/* Multiple interrupts. */
4549 	if (sc->flags & IS_VF)
4550 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4551 		    ("%s: too few intr.", __func__));
4552 	else
4553 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4554 		    ("%s: too few intr.", __func__));
4555 
4556 	/* The first one is always error intr on PFs */
4557 	if (!(sc->flags & IS_VF)) {
4558 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4559 		if (rc != 0)
4560 			return (rc);
4561 		irq++;
4562 		rid++;
4563 	}
4564 
4565 	/* The second one is always the firmware event queue (first on VFs) */
4566 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4567 	if (rc != 0)
4568 		return (rc);
4569 	irq++;
4570 	rid++;
4571 
4572 	for_each_port(sc, p) {
4573 		pi = sc->port[p];
4574 		for_each_vi(pi, v, vi) {
4575 			vi->first_intr = rid - 1;
4576 
4577 			if (vi->nnmrxq > 0) {
4578 				int n = max(vi->nrxq, vi->nnmrxq);
4579 
4580 				rxq = &sge->rxq[vi->first_rxq];
4581 #ifdef DEV_NETMAP
4582 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4583 #endif
4584 				for (q = 0; q < n; q++) {
4585 					snprintf(s, sizeof(s), "%x%c%x", p,
4586 					    'a' + v, q);
4587 					if (q < vi->nrxq)
4588 						irq->rxq = rxq++;
4589 #ifdef DEV_NETMAP
4590 					if (q < vi->nnmrxq)
4591 						irq->nm_rxq = nm_rxq++;
4592 #endif
4593 					rc = t4_alloc_irq(sc, irq, rid,
4594 					    t4_vi_intr, irq, s);
4595 					if (rc != 0)
4596 						return (rc);
4597 #ifdef RSS
4598 					if (q < vi->nrxq) {
4599 						bus_bind_intr(sc->dev, irq->res,
4600 						    rss_getcpu(q % nbuckets));
4601 					}
4602 #endif
4603 					irq++;
4604 					rid++;
4605 					vi->nintr++;
4606 				}
4607 			} else {
4608 				for_each_rxq(vi, q, rxq) {
4609 					snprintf(s, sizeof(s), "%x%c%x", p,
4610 					    'a' + v, q);
4611 					rc = t4_alloc_irq(sc, irq, rid,
4612 					    t4_intr, rxq, s);
4613 					if (rc != 0)
4614 						return (rc);
4615 #ifdef RSS
4616 					bus_bind_intr(sc->dev, irq->res,
4617 					    rss_getcpu(q % nbuckets));
4618 #endif
4619 					irq++;
4620 					rid++;
4621 					vi->nintr++;
4622 				}
4623 			}
4624 #ifdef TCP_OFFLOAD
4625 			for_each_ofld_rxq(vi, q, ofld_rxq) {
4626 				snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
4627 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
4628 				    ofld_rxq, s);
4629 				if (rc != 0)
4630 					return (rc);
4631 				irq++;
4632 				rid++;
4633 				vi->nintr++;
4634 			}
4635 #endif
4636 		}
4637 	}
4638 	MPASS(irq == &sc->irq[sc->intr_count]);
4639 
4640 	return (0);
4641 }
4642 
4643 int
4644 adapter_full_init(struct adapter *sc)
4645 {
4646 	int rc, i;
4647 #ifdef RSS
4648 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4649 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4650 #endif
4651 
4652 	ASSERT_SYNCHRONIZED_OP(sc);
4653 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4654 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
4655 	    ("%s: FULL_INIT_DONE already", __func__));
4656 
4657 	/*
4658 	 * queues that belong to the adapter (not any particular port).
4659 	 */
4660 	rc = t4_setup_adapter_queues(sc);
4661 	if (rc != 0)
4662 		goto done;
4663 
4664 	for (i = 0; i < nitems(sc->tq); i++) {
4665 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
4666 		    taskqueue_thread_enqueue, &sc->tq[i]);
4667 		if (sc->tq[i] == NULL) {
4668 			device_printf(sc->dev,
4669 			    "failed to allocate task queue %d\n", i);
4670 			rc = ENOMEM;
4671 			goto done;
4672 		}
4673 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
4674 		    device_get_nameunit(sc->dev), i);
4675 	}
4676 #ifdef RSS
4677 	MPASS(RSS_KEYSIZE == 40);
4678 	rss_getkey((void *)&raw_rss_key[0]);
4679 	for (i = 0; i < nitems(rss_key); i++) {
4680 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
4681 	}
4682 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
4683 #endif
4684 
4685 	if (!(sc->flags & IS_VF))
4686 		t4_intr_enable(sc);
4687 	sc->flags |= FULL_INIT_DONE;
4688 done:
4689 	if (rc != 0)
4690 		adapter_full_uninit(sc);
4691 
4692 	return (rc);
4693 }
4694 
4695 int
4696 adapter_full_uninit(struct adapter *sc)
4697 {
4698 	int i;
4699 
4700 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4701 
4702 	t4_teardown_adapter_queues(sc);
4703 
4704 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
4705 		taskqueue_free(sc->tq[i]);
4706 		sc->tq[i] = NULL;
4707 	}
4708 
4709 	sc->flags &= ~FULL_INIT_DONE;
4710 
4711 	return (0);
4712 }
4713 
4714 #ifdef RSS
4715 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
4716     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
4717     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
4718     RSS_HASHTYPE_RSS_UDP_IPV6)
4719 
4720 /* Translates kernel hash types to hardware. */
4721 static int
4722 hashconfig_to_hashen(int hashconfig)
4723 {
4724 	int hashen = 0;
4725 
4726 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
4727 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
4728 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
4729 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
4730 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
4731 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4732 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4733 	}
4734 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
4735 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4736 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4737 	}
4738 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
4739 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4740 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
4741 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4742 
4743 	return (hashen);
4744 }
4745 
4746 /* Translates hardware hash types to kernel. */
4747 static int
4748 hashen_to_hashconfig(int hashen)
4749 {
4750 	int hashconfig = 0;
4751 
4752 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
4753 		/*
4754 		 * If UDP hashing was enabled it must have been enabled for
4755 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
4756 		 * enabling any 4-tuple hash is nonsense configuration.
4757 		 */
4758 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4759 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
4760 
4761 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4762 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
4763 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4764 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
4765 	}
4766 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4767 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
4768 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4769 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
4770 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
4771 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
4772 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
4773 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
4774 
4775 	return (hashconfig);
4776 }
4777 #endif
4778 
4779 int
4780 vi_full_init(struct vi_info *vi)
4781 {
4782 	struct adapter *sc = vi->pi->adapter;
4783 	struct ifnet *ifp = vi->ifp;
4784 	uint16_t *rss;
4785 	struct sge_rxq *rxq;
4786 	int rc, i, j, hashen;
4787 #ifdef RSS
4788 	int nbuckets = rss_getnumbuckets();
4789 	int hashconfig = rss_gethashconfig();
4790 	int extra;
4791 #endif
4792 
4793 	ASSERT_SYNCHRONIZED_OP(sc);
4794 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
4795 	    ("%s: VI_INIT_DONE already", __func__));
4796 
4797 	sysctl_ctx_init(&vi->ctx);
4798 	vi->flags |= VI_SYSCTL_CTX;
4799 
4800 	/*
4801 	 * Allocate tx/rx/fl queues for this VI.
4802 	 */
4803 	rc = t4_setup_vi_queues(vi);
4804 	if (rc != 0)
4805 		goto done;	/* error message displayed already */
4806 
4807 	/*
4808 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
4809 	 */
4810 	if (vi->nrxq > vi->rss_size) {
4811 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
4812 		    "some queues will never receive traffic.\n", vi->nrxq,
4813 		    vi->rss_size);
4814 	} else if (vi->rss_size % vi->nrxq) {
4815 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
4816 		    "expect uneven traffic distribution.\n", vi->nrxq,
4817 		    vi->rss_size);
4818 	}
4819 #ifdef RSS
4820 	if (vi->nrxq != nbuckets) {
4821 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
4822 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
4823 	}
4824 #endif
4825 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
4826 	for (i = 0; i < vi->rss_size;) {
4827 #ifdef RSS
4828 		j = rss_get_indirection_to_bucket(i);
4829 		j %= vi->nrxq;
4830 		rxq = &sc->sge.rxq[vi->first_rxq + j];
4831 		rss[i++] = rxq->iq.abs_id;
4832 #else
4833 		for_each_rxq(vi, j, rxq) {
4834 			rss[i++] = rxq->iq.abs_id;
4835 			if (i == vi->rss_size)
4836 				break;
4837 		}
4838 #endif
4839 	}
4840 
4841 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
4842 	    vi->rss_size);
4843 	if (rc != 0) {
4844 		if_printf(ifp, "rss_config failed: %d\n", rc);
4845 		goto done;
4846 	}
4847 
4848 #ifdef RSS
4849 	hashen = hashconfig_to_hashen(hashconfig);
4850 
4851 	/*
4852 	 * We may have had to enable some hashes even though the global config
4853 	 * wants them disabled.  This is a potential problem that must be
4854 	 * reported to the user.
4855 	 */
4856 	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
4857 
4858 	/*
4859 	 * If we consider only the supported hash types, then the enabled hashes
4860 	 * are a superset of the requested hashes.  In other words, there cannot
4861 	 * be any supported hash that was requested but not enabled, but there
4862 	 * can be hashes that were not requested but had to be enabled.
4863 	 */
4864 	extra &= SUPPORTED_RSS_HASHTYPES;
4865 	MPASS((extra & hashconfig) == 0);
4866 
4867 	if (extra) {
4868 		if_printf(ifp,
4869 		    "global RSS config (0x%x) cannot be accommodated.\n",
4870 		    hashconfig);
4871 	}
4872 	if (extra & RSS_HASHTYPE_RSS_IPV4)
4873 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
4874 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
4875 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
4876 	if (extra & RSS_HASHTYPE_RSS_IPV6)
4877 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
4878 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
4879 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
4880 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
4881 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
4882 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
4883 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
4884 #else
4885 	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
4886 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
4887 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4888 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
4889 #endif
4890 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
4891 	if (rc != 0) {
4892 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
4893 		goto done;
4894 	}
4895 
4896 	vi->rss = rss;
4897 	vi->flags |= VI_INIT_DONE;
4898 done:
4899 	if (rc != 0)
4900 		vi_full_uninit(vi);
4901 
4902 	return (rc);
4903 }
4904 
4905 /*
4906  * Idempotent.
4907  */
4908 int
4909 vi_full_uninit(struct vi_info *vi)
4910 {
4911 	struct port_info *pi = vi->pi;
4912 	struct adapter *sc = pi->adapter;
4913 	int i;
4914 	struct sge_rxq *rxq;
4915 	struct sge_txq *txq;
4916 #ifdef TCP_OFFLOAD
4917 	struct sge_ofld_rxq *ofld_rxq;
4918 	struct sge_wrq *ofld_txq;
4919 #endif
4920 
4921 	if (vi->flags & VI_INIT_DONE) {
4922 
4923 		/* Need to quiesce queues.  */
4924 
4925 		/* XXX: Only for the first VI? */
4926 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
4927 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
4928 
4929 		for_each_txq(vi, i, txq) {
4930 			quiesce_txq(sc, txq);
4931 		}
4932 
4933 #ifdef TCP_OFFLOAD
4934 		for_each_ofld_txq(vi, i, ofld_txq) {
4935 			quiesce_wrq(sc, ofld_txq);
4936 		}
4937 #endif
4938 
4939 		for_each_rxq(vi, i, rxq) {
4940 			quiesce_iq(sc, &rxq->iq);
4941 			quiesce_fl(sc, &rxq->fl);
4942 		}
4943 
4944 #ifdef TCP_OFFLOAD
4945 		for_each_ofld_rxq(vi, i, ofld_rxq) {
4946 			quiesce_iq(sc, &ofld_rxq->iq);
4947 			quiesce_fl(sc, &ofld_rxq->fl);
4948 		}
4949 #endif
4950 		free(vi->rss, M_CXGBE);
4951 		free(vi->nm_rss, M_CXGBE);
4952 	}
4953 
4954 	t4_teardown_vi_queues(vi);
4955 	vi->flags &= ~VI_INIT_DONE;
4956 
4957 	return (0);
4958 }
4959 
4960 static void
4961 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
4962 {
4963 	struct sge_eq *eq = &txq->eq;
4964 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
4965 
4966 	(void) sc;	/* unused */
4967 
4968 #ifdef INVARIANTS
4969 	TXQ_LOCK(txq);
4970 	MPASS((eq->flags & EQ_ENABLED) == 0);
4971 	TXQ_UNLOCK(txq);
4972 #endif
4973 
4974 	/* Wait for the mp_ring to empty. */
4975 	while (!mp_ring_is_idle(txq->r)) {
4976 		mp_ring_check_drainage(txq->r, 0);
4977 		pause("rquiesce", 1);
4978 	}
4979 
4980 	/* Then wait for the hardware to finish. */
4981 	while (spg->cidx != htobe16(eq->pidx))
4982 		pause("equiesce", 1);
4983 
4984 	/* Finally, wait for the driver to reclaim all descriptors. */
4985 	while (eq->cidx != eq->pidx)
4986 		pause("dquiesce", 1);
4987 }
4988 
4989 static void
4990 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
4991 {
4992 
4993 	/* XXXTX */
4994 }
4995 
4996 static void
4997 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
4998 {
4999 	(void) sc;	/* unused */
5000 
5001 	/* Synchronize with the interrupt handler */
5002 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
5003 		pause("iqfree", 1);
5004 }
5005 
5006 static void
5007 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
5008 {
5009 	mtx_lock(&sc->sfl_lock);
5010 	FL_LOCK(fl);
5011 	fl->flags |= FL_DOOMED;
5012 	FL_UNLOCK(fl);
5013 	callout_stop(&sc->sfl_callout);
5014 	mtx_unlock(&sc->sfl_lock);
5015 
5016 	KASSERT((fl->flags & FL_STARVING) == 0,
5017 	    ("%s: still starving", __func__));
5018 }
5019 
5020 static int
5021 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
5022     driver_intr_t *handler, void *arg, char *name)
5023 {
5024 	int rc;
5025 
5026 	irq->rid = rid;
5027 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
5028 	    RF_SHAREABLE | RF_ACTIVE);
5029 	if (irq->res == NULL) {
5030 		device_printf(sc->dev,
5031 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
5032 		return (ENOMEM);
5033 	}
5034 
5035 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
5036 	    NULL, handler, arg, &irq->tag);
5037 	if (rc != 0) {
5038 		device_printf(sc->dev,
5039 		    "failed to setup interrupt for rid %d, name %s: %d\n",
5040 		    rid, name, rc);
5041 	} else if (name)
5042 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
5043 
5044 	return (rc);
5045 }
5046 
5047 static int
5048 t4_free_irq(struct adapter *sc, struct irq *irq)
5049 {
5050 	if (irq->tag)
5051 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
5052 	if (irq->res)
5053 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
5054 
5055 	bzero(irq, sizeof(*irq));
5056 
5057 	return (0);
5058 }
5059 
5060 static void
5061 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
5062 {
5063 
5064 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
5065 	t4_get_regs(sc, buf, regs->len);
5066 }
5067 
5068 #define	A_PL_INDIR_CMD	0x1f8
5069 
5070 #define	S_PL_AUTOINC	31
5071 #define	M_PL_AUTOINC	0x1U
5072 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
5073 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
5074 
5075 #define	S_PL_VFID	20
5076 #define	M_PL_VFID	0xffU
5077 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
5078 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
5079 
5080 #define	S_PL_ADDR	0
5081 #define	M_PL_ADDR	0xfffffU
5082 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
5083 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
5084 
5085 #define	A_PL_INDIR_DATA	0x1fc
5086 
5087 static uint64_t
5088 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
5089 {
5090 	u32 stats[2];
5091 
5092 	mtx_assert(&sc->reg_lock, MA_OWNED);
5093 	if (sc->flags & IS_VF) {
5094 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
5095 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
5096 	} else {
5097 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5098 		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5099 		    V_PL_ADDR(VF_MPS_REG(reg)));
5100 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
5101 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
5102 	}
5103 	return (((uint64_t)stats[1]) << 32 | stats[0]);
5104 }
5105 
5106 static void
5107 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
5108     struct fw_vi_stats_vf *stats)
5109 {
5110 
5111 #define GET_STAT(name) \
5112 	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
5113 
5114 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
5115 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
5116 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
5117 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
5118 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
5119 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
5120 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
5121 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
5122 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
5123 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
5124 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
5125 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
5126 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
5127 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
5128 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
5129 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
5130 
5131 #undef GET_STAT
5132 }
5133 
5134 static void
5135 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
5136 {
5137 	int reg;
5138 
5139 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5140 	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5141 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
5142 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
5143 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
5144 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
5145 }
5146 
5147 static void
5148 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
5149 {
5150 	struct timeval tv;
5151 	const struct timeval interval = {0, 250000};	/* 250ms */
5152 
5153 	if (!(vi->flags & VI_INIT_DONE))
5154 		return;
5155 
5156 	getmicrotime(&tv);
5157 	timevalsub(&tv, &interval);
5158 	if (timevalcmp(&tv, &vi->last_refreshed, <))
5159 		return;
5160 
5161 	mtx_lock(&sc->reg_lock);
5162 	t4_get_vi_stats(sc, vi->viid, &vi->stats);
5163 	getmicrotime(&vi->last_refreshed);
5164 	mtx_unlock(&sc->reg_lock);
5165 }
5166 
5167 static void
5168 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5169 {
5170 	u_int i, v, tnl_cong_drops, bg_map;
5171 	struct timeval tv;
5172 	const struct timeval interval = {0, 250000};	/* 250ms */
5173 
5174 	getmicrotime(&tv);
5175 	timevalsub(&tv, &interval);
5176 	if (timevalcmp(&tv, &pi->last_refreshed, <))
5177 		return;
5178 
5179 	tnl_cong_drops = 0;
5180 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5181 	bg_map = pi->mps_bg_map;
5182 	while (bg_map) {
5183 		i = ffs(bg_map) - 1;
5184 		mtx_lock(&sc->reg_lock);
5185 		t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5186 		    A_TP_MIB_TNL_CNG_DROP_0 + i);
5187 		mtx_unlock(&sc->reg_lock);
5188 		tnl_cong_drops += v;
5189 		bg_map &= ~(1 << i);
5190 	}
5191 	pi->tnl_cong_drops = tnl_cong_drops;
5192 	getmicrotime(&pi->last_refreshed);
5193 }
5194 
5195 static void
5196 cxgbe_tick(void *arg)
5197 {
5198 	struct port_info *pi = arg;
5199 	struct adapter *sc = pi->adapter;
5200 
5201 	PORT_LOCK_ASSERT_OWNED(pi);
5202 	cxgbe_refresh_stats(sc, pi);
5203 
5204 	callout_schedule(&pi->tick, hz);
5205 }
5206 
5207 void
5208 vi_tick(void *arg)
5209 {
5210 	struct vi_info *vi = arg;
5211 	struct adapter *sc = vi->pi->adapter;
5212 
5213 	vi_refresh_stats(sc, vi);
5214 
5215 	callout_schedule(&vi->tick, hz);
5216 }
5217 
5218 static void
5219 cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
5220 {
5221 	struct ifnet *vlan;
5222 
5223 	if (arg != ifp || ifp->if_type != IFT_ETHER)
5224 		return;
5225 
5226 	vlan = VLAN_DEVAT(ifp, vid);
5227 	VLAN_SETCOOKIE(vlan, ifp);
5228 }
5229 
5230 /*
5231  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5232  */
5233 static char *caps_decoder[] = {
5234 	"\20\001IPMI\002NCSI",				/* 0: NBM */
5235 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
5236 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
5237 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
5238 	    "\006HASHFILTER\007ETHOFLD",
5239 	"\20\001TOE",					/* 4: TOE */
5240 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
5241 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
5242 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5243 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5244 	    "\007T10DIF"
5245 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5246 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
5247 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
5248 		    "\004PO_INITIATOR\005PO_TARGET",
5249 };
5250 
5251 void
5252 t4_sysctls(struct adapter *sc)
5253 {
5254 	struct sysctl_ctx_list *ctx;
5255 	struct sysctl_oid *oid;
5256 	struct sysctl_oid_list *children, *c0;
5257 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5258 
5259 	ctx = device_get_sysctl_ctx(sc->dev);
5260 
5261 	/*
5262 	 * dev.t4nex.X.
5263 	 */
5264 	oid = device_get_sysctl_tree(sc->dev);
5265 	c0 = children = SYSCTL_CHILDREN(oid);
5266 
5267 	sc->sc_do_rxcopy = 1;
5268 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5269 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5270 
5271 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5272 	    sc->params.nports, "# of ports");
5273 
5274 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5275 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
5276 	    sysctl_bitfield, "A", "available doorbells");
5277 
5278 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5279 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
5280 
5281 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5282 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5283 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5284 	    "interrupt holdoff timer values (us)");
5285 
5286 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5287 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5288 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5289 	    "interrupt holdoff packet counter values");
5290 
5291 	t4_sge_sysctls(sc, ctx, children);
5292 
5293 	sc->lro_timeout = 100;
5294 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5295 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5296 
5297 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5298 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
5299 
5300 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5301 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5302 
5303 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5304 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5305 
5306 	if (sc->flags & IS_VF)
5307 		return;
5308 
5309 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5310 	    NULL, chip_rev(sc), "chip hardware revision");
5311 
5312 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5313 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5314 
5315 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5316 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5317 
5318 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5319 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5320 
5321 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5322 	    CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5323 
5324 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5325 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5326 
5327 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5328 	    sc->er_version, 0, "expansion ROM version");
5329 
5330 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5331 	    sc->bs_version, 0, "bootstrap firmware version");
5332 
5333 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5334 	    NULL, sc->params.scfg_vers, "serial config version");
5335 
5336 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5337 	    NULL, sc->params.vpd_vers, "VPD version");
5338 
5339 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5340 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5341 
5342 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5343 	    sc->cfcsum, "config file checksum");
5344 
5345 #define SYSCTL_CAP(name, n, text) \
5346 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5347 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
5348 	    sysctl_bitfield, "A", "available " text " capabilities")
5349 
5350 	SYSCTL_CAP(nbmcaps, 0, "NBM");
5351 	SYSCTL_CAP(linkcaps, 1, "link");
5352 	SYSCTL_CAP(switchcaps, 2, "switch");
5353 	SYSCTL_CAP(niccaps, 3, "NIC");
5354 	SYSCTL_CAP(toecaps, 4, "TCP offload");
5355 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
5356 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5357 	SYSCTL_CAP(cryptocaps, 7, "crypto");
5358 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
5359 #undef SYSCTL_CAP
5360 
5361 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5362 	    NULL, sc->tids.nftids, "number of filters");
5363 
5364 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5365 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5366 	    "chip temperature (in Celsius)");
5367 
5368 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
5369 	    &sc->params.core_vdd, 0, "core Vdd (in mV)");
5370 
5371 #ifdef SBUF_DRAIN
5372 	/*
5373 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
5374 	 */
5375 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5376 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5377 	    "logs and miscellaneous information");
5378 	children = SYSCTL_CHILDREN(oid);
5379 
5380 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
5381 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5382 	    sysctl_cctrl, "A", "congestion control");
5383 
5384 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
5385 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5386 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
5387 
5388 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
5389 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
5390 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
5391 
5392 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
5393 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
5394 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
5395 
5396 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
5397 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
5398 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
5399 
5400 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
5401 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
5402 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
5403 
5404 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
5405 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
5406 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
5407 
5408 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
5409 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5410 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
5411 	    "A", "CIM logic analyzer");
5412 
5413 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
5414 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5415 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
5416 
5417 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
5418 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
5419 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
5420 
5421 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
5422 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
5423 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
5424 
5425 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
5426 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
5427 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
5428 
5429 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5430 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5431 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5432 
5433 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5434 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5435 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5436 
5437 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5438 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5439 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5440 
5441 	if (chip_id(sc) > CHELSIO_T4) {
5442 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5443 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5444 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5445 
5446 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5447 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5448 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5449 	}
5450 
5451 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5452 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5453 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5454 
5455 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5456 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5457 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
5458 
5459 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5460 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5461 	    sysctl_cpl_stats, "A", "CPL statistics");
5462 
5463 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5464 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5465 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5466 
5467 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5468 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5469 	    sysctl_devlog, "A", "firmware's device log");
5470 
5471 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5472 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5473 	    sysctl_fcoe_stats, "A", "FCoE statistics");
5474 
5475 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5476 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5477 	    sysctl_hw_sched, "A", "hardware scheduler ");
5478 
5479 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5480 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5481 	    sysctl_l2t, "A", "hardware L2 table");
5482 
5483 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5484 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5485 	    sysctl_lb_stats, "A", "loopback statistics");
5486 
5487 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5488 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5489 	    sysctl_meminfo, "A", "memory regions");
5490 
5491 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5492 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5493 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5494 	    "A", "MPS TCAM entries");
5495 
5496 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5497 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5498 	    sysctl_path_mtus, "A", "path MTUs");
5499 
5500 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5501 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5502 	    sysctl_pm_stats, "A", "PM statistics");
5503 
5504 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5505 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5506 	    sysctl_rdma_stats, "A", "RDMA statistics");
5507 
5508 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5509 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5510 	    sysctl_tcp_stats, "A", "TCP statistics");
5511 
5512 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5513 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5514 	    sysctl_tids, "A", "TID information");
5515 
5516 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5517 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5518 	    sysctl_tp_err_stats, "A", "TP error statistics");
5519 
5520 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5521 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5522 	    "TP logic analyzer event capture mask");
5523 
5524 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5525 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5526 	    sysctl_tp_la, "A", "TP logic analyzer");
5527 
5528 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5529 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5530 	    sysctl_tx_rate, "A", "Tx rate");
5531 
5532 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5533 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5534 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5535 
5536 	if (chip_id(sc) >= CHELSIO_T5) {
5537 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5538 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5539 		    sysctl_wcwr_stats, "A", "write combined work requests");
5540 	}
5541 #endif
5542 
5543 #ifdef TCP_OFFLOAD
5544 	if (is_offload(sc)) {
5545 		int i;
5546 		char s[4];
5547 
5548 		/*
5549 		 * dev.t4nex.X.toe.
5550 		 */
5551 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5552 		    NULL, "TOE parameters");
5553 		children = SYSCTL_CHILDREN(oid);
5554 
5555 		sc->tt.cong_algorithm = -1;
5556 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
5557 		    CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
5558 		    "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
5559 		    "3 = highspeed)");
5560 
5561 		sc->tt.sndbuf = 256 * 1024;
5562 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5563 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
5564 
5565 		sc->tt.ddp = 0;
5566 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5567 		    &sc->tt.ddp, 0, "DDP allowed");
5568 
5569 		sc->tt.rx_coalesce = 1;
5570 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5571 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5572 
5573 		sc->tt.tls = 0;
5574 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tls", CTLFLAG_RW,
5575 		    &sc->tt.tls, 0, "Inline TLS allowed");
5576 
5577 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls_rx_ports",
5578 		    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tls_rx_ports,
5579 		    "I", "TCP ports that use inline TLS+TOE RX");
5580 
5581 		sc->tt.tx_align = 1;
5582 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5583 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5584 
5585 		sc->tt.tx_zcopy = 0;
5586 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
5587 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
5588 		    "Enable zero-copy aio_write(2)");
5589 
5590 		sc->tt.cop_managed_offloading = !!t4_cop_managed_offloading;
5591 		SYSCTL_ADD_INT(ctx, children, OID_AUTO,
5592 		    "cop_managed_offloading", CTLFLAG_RW,
5593 		    &sc->tt.cop_managed_offloading, 0,
5594 		    "COP (Connection Offload Policy) controls all TOE offload");
5595 
5596 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
5597 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
5598 		    "TP timer tick (us)");
5599 
5600 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
5601 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
5602 		    "TCP timestamp tick (us)");
5603 
5604 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5605 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5606 		    "DACK tick (us)");
5607 
5608 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5609 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5610 		    "IU", "DACK timer (us)");
5611 
5612 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5613 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5614 		    sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
5615 
5616 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5617 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5618 		    sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
5619 
5620 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5621 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5622 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
5623 
5624 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5625 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5626 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
5627 
5628 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5629 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5630 		    sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
5631 
5632 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
5633 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
5634 		    sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
5635 
5636 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
5637 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
5638 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
5639 
5640 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
5641 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
5642 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
5643 
5644 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
5645 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
5646 		    sysctl_tp_shift_cnt, "IU",
5647 		    "Number of SYN retransmissions before abort");
5648 
5649 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
5650 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
5651 		    sysctl_tp_shift_cnt, "IU",
5652 		    "Number of retransmissions before abort");
5653 
5654 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
5655 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
5656 		    sysctl_tp_shift_cnt, "IU",
5657 		    "Number of keepalive probes before abort");
5658 
5659 		oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
5660 		    CTLFLAG_RD, NULL, "TOE retransmit backoffs");
5661 		children = SYSCTL_CHILDREN(oid);
5662 		for (i = 0; i < 16; i++) {
5663 			snprintf(s, sizeof(s), "%u", i);
5664 			SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
5665 			    CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
5666 			    "IU", "TOE retransmit backoff");
5667 		}
5668 	}
5669 #endif
5670 }
5671 
5672 void
5673 vi_sysctls(struct vi_info *vi)
5674 {
5675 	struct sysctl_ctx_list *ctx;
5676 	struct sysctl_oid *oid;
5677 	struct sysctl_oid_list *children;
5678 
5679 	ctx = device_get_sysctl_ctx(vi->dev);
5680 
5681 	/*
5682 	 * dev.v?(cxgbe|cxl).X.
5683 	 */
5684 	oid = device_get_sysctl_tree(vi->dev);
5685 	children = SYSCTL_CHILDREN(oid);
5686 
5687 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
5688 	    vi->viid, "VI identifer");
5689 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
5690 	    &vi->nrxq, 0, "# of rx queues");
5691 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5692 	    &vi->ntxq, 0, "# of tx queues");
5693 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5694 	    &vi->first_rxq, 0, "index of first rx queue");
5695 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5696 	    &vi->first_txq, 0, "index of first tx queue");
5697 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
5698 	    vi->rss_size, "size of RSS indirection table");
5699 
5700 	if (IS_MAIN_VI(vi)) {
5701 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
5702 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
5703 		    "Reserve queue 0 for non-flowid packets");
5704 	}
5705 
5706 #ifdef TCP_OFFLOAD
5707 	if (vi->nofldrxq != 0) {
5708 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5709 		    &vi->nofldrxq, 0,
5710 		    "# of rx queues for offloaded TCP connections");
5711 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5712 		    &vi->nofldtxq, 0,
5713 		    "# of tx queues for offloaded TCP connections");
5714 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5715 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
5716 		    "index of first TOE rx queue");
5717 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5718 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
5719 		    "index of first TOE tx queue");
5720 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
5721 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5722 		    sysctl_holdoff_tmr_idx_ofld, "I",
5723 		    "holdoff timer index for TOE queues");
5724 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
5725 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5726 		    sysctl_holdoff_pktc_idx_ofld, "I",
5727 		    "holdoff packet counter index for TOE queues");
5728 	}
5729 #endif
5730 #ifdef DEV_NETMAP
5731 	if (vi->nnmrxq != 0) {
5732 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5733 		    &vi->nnmrxq, 0, "# of netmap rx queues");
5734 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5735 		    &vi->nnmtxq, 0, "# of netmap tx queues");
5736 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5737 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
5738 		    "index of first netmap rx queue");
5739 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
5740 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
5741 		    "index of first netmap tx queue");
5742 	}
5743 #endif
5744 
5745 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
5746 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
5747 	    "holdoff timer index");
5748 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
5749 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
5750 	    "holdoff packet counter index");
5751 
5752 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
5753 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
5754 	    "rx queue size");
5755 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
5756 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
5757 	    "tx queue size");
5758 }
5759 
5760 static void
5761 cxgbe_sysctls(struct port_info *pi)
5762 {
5763 	struct sysctl_ctx_list *ctx;
5764 	struct sysctl_oid *oid;
5765 	struct sysctl_oid_list *children, *children2;
5766 	struct adapter *sc = pi->adapter;
5767 	int i;
5768 	char name[16];
5769 
5770 	ctx = device_get_sysctl_ctx(pi->dev);
5771 
5772 	/*
5773 	 * dev.cxgbe.X.
5774 	 */
5775 	oid = device_get_sysctl_tree(pi->dev);
5776 	children = SYSCTL_CHILDREN(oid);
5777 
5778 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
5779 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
5780 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
5781 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
5782 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
5783 		    "PHY temperature (in Celsius)");
5784 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
5785 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
5786 		    "PHY firmware version");
5787 	}
5788 
5789 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
5790 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
5791 	    "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
5792 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
5793 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
5794 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
5795 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
5796 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
5797 	    "autonegotiation (-1 = not supported)");
5798 
5799 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
5800 	    port_top_speed(pi), "max speed (in Gbps)");
5801 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
5802 	    pi->mps_bg_map, "MPS buffer group map");
5803 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
5804 	    NULL, pi->rx_e_chan_map, "TP rx e-channel map");
5805 
5806 	if (sc->flags & IS_VF)
5807 		return;
5808 
5809 	/*
5810 	 * dev.(cxgbe|cxl).X.tc.
5811 	 */
5812 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
5813 	    "Tx scheduler traffic classes (cl_rl)");
5814 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
5815 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
5816 
5817 		snprintf(name, sizeof(name), "%d", i);
5818 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
5819 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
5820 		    "traffic class"));
5821 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
5822 		    &tc->flags, 0, "flags");
5823 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
5824 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
5825 #ifdef SBUF_DRAIN
5826 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
5827 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
5828 		    sysctl_tc_params, "A", "traffic class parameters");
5829 #endif
5830 	}
5831 
5832 	/*
5833 	 * dev.cxgbe.X.stats.
5834 	 */
5835 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
5836 	    NULL, "port statistics");
5837 	children = SYSCTL_CHILDREN(oid);
5838 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
5839 	    &pi->tx_parse_error, 0,
5840 	    "# of tx packets with invalid length or # of segments");
5841 
5842 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
5843 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
5844 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
5845 	    sysctl_handle_t4_reg64, "QU", desc)
5846 
5847 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
5848 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
5849 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
5850 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
5851 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
5852 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
5853 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
5854 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
5855 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
5856 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
5857 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
5858 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
5859 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
5860 	    "# of tx frames in this range",
5861 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
5862 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
5863 	    "# of tx frames in this range",
5864 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
5865 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
5866 	    "# of tx frames in this range",
5867 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
5868 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
5869 	    "# of tx frames in this range",
5870 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
5871 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
5872 	    "# of tx frames in this range",
5873 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
5874 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
5875 	    "# of tx frames in this range",
5876 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
5877 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
5878 	    "# of tx frames in this range",
5879 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
5880 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
5881 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
5882 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
5883 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
5884 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
5885 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
5886 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5887 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5888 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5889 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5890 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5891 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5892 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5893 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5894 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5895 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5896 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5897 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5898 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5899 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5900 
5901 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5902 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5903 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5904 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5905 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5906 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5907 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5908 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5909 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5910 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5911 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5912 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5913 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5914 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5915 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5916 	    "# of frames received with bad FCS",
5917 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5918 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5919 	    "# of frames received with length error",
5920 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5921 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5922 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5923 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5924 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5925 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5926 	    "# of rx frames in this range",
5927 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5928 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5929 	    "# of rx frames in this range",
5930 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5931 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5932 	    "# of rx frames in this range",
5933 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5934 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5935 	    "# of rx frames in this range",
5936 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5937 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5938 	    "# of rx frames in this range",
5939 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5940 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5941 	    "# of rx frames in this range",
5942 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5943 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5944 	    "# of rx frames in this range",
5945 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5946 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5947 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5948 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5949 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5950 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5951 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5952 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5953 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5954 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5955 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5956 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5957 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5958 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5959 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5960 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5961 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5962 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5963 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5964 
5965 #undef SYSCTL_ADD_T4_REG64
5966 
5967 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5968 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5969 	    &pi->stats.name, desc)
5970 
5971 	/* We get these from port_stats and they may be stale by up to 1s */
5972 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5973 	    "# drops due to buffer-group 0 overflows");
5974 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5975 	    "# drops due to buffer-group 1 overflows");
5976 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5977 	    "# drops due to buffer-group 2 overflows");
5978 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5979 	    "# drops due to buffer-group 3 overflows");
5980 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5981 	    "# of buffer-group 0 truncated packets");
5982 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5983 	    "# of buffer-group 1 truncated packets");
5984 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5985 	    "# of buffer-group 2 truncated packets");
5986 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5987 	    "# of buffer-group 3 truncated packets");
5988 
5989 #undef SYSCTL_ADD_T4_PORTSTAT
5990 
5991 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_records",
5992 	    CTLFLAG_RD, &pi->tx_tls_records,
5993 	    "# of TLS records transmitted");
5994 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_octets",
5995 	    CTLFLAG_RD, &pi->tx_tls_octets,
5996 	    "# of payload octets in transmitted TLS records");
5997 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_records",
5998 	    CTLFLAG_RD, &pi->rx_tls_records,
5999 	    "# of TLS records received");
6000 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_octets",
6001 	    CTLFLAG_RD, &pi->rx_tls_octets,
6002 	    "# of payload octets in received TLS records");
6003 }
6004 
6005 static int
6006 sysctl_int_array(SYSCTL_HANDLER_ARGS)
6007 {
6008 	int rc, *i, space = 0;
6009 	struct sbuf sb;
6010 
6011 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
6012 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
6013 		if (space)
6014 			sbuf_printf(&sb, " ");
6015 		sbuf_printf(&sb, "%d", *i);
6016 		space = 1;
6017 	}
6018 	rc = sbuf_finish(&sb);
6019 	sbuf_delete(&sb);
6020 	return (rc);
6021 }
6022 
6023 static int
6024 sysctl_bitfield(SYSCTL_HANDLER_ARGS)
6025 {
6026 	int rc;
6027 	struct sbuf *sb;
6028 
6029 	rc = sysctl_wire_old_buffer(req, 0);
6030 	if (rc != 0)
6031 		return(rc);
6032 
6033 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6034 	if (sb == NULL)
6035 		return (ENOMEM);
6036 
6037 	sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
6038 	rc = sbuf_finish(sb);
6039 	sbuf_delete(sb);
6040 
6041 	return (rc);
6042 }
6043 
6044 static int
6045 sysctl_btphy(SYSCTL_HANDLER_ARGS)
6046 {
6047 	struct port_info *pi = arg1;
6048 	int op = arg2;
6049 	struct adapter *sc = pi->adapter;
6050 	u_int v;
6051 	int rc;
6052 
6053 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
6054 	if (rc)
6055 		return (rc);
6056 	/* XXX: magic numbers */
6057 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
6058 	    &v);
6059 	end_synchronized_op(sc, 0);
6060 	if (rc)
6061 		return (rc);
6062 	if (op == 0)
6063 		v /= 256;
6064 
6065 	rc = sysctl_handle_int(oidp, &v, 0, req);
6066 	return (rc);
6067 }
6068 
6069 static int
6070 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
6071 {
6072 	struct vi_info *vi = arg1;
6073 	int rc, val;
6074 
6075 	val = vi->rsrv_noflowq;
6076 	rc = sysctl_handle_int(oidp, &val, 0, req);
6077 	if (rc != 0 || req->newptr == NULL)
6078 		return (rc);
6079 
6080 	if ((val >= 1) && (vi->ntxq > 1))
6081 		vi->rsrv_noflowq = 1;
6082 	else
6083 		vi->rsrv_noflowq = 0;
6084 
6085 	return (rc);
6086 }
6087 
6088 static int
6089 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
6090 {
6091 	struct vi_info *vi = arg1;
6092 	struct adapter *sc = vi->pi->adapter;
6093 	int idx, rc, i;
6094 	struct sge_rxq *rxq;
6095 	uint8_t v;
6096 
6097 	idx = vi->tmr_idx;
6098 
6099 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6100 	if (rc != 0 || req->newptr == NULL)
6101 		return (rc);
6102 
6103 	if (idx < 0 || idx >= SGE_NTIMERS)
6104 		return (EINVAL);
6105 
6106 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6107 	    "t4tmr");
6108 	if (rc)
6109 		return (rc);
6110 
6111 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
6112 	for_each_rxq(vi, i, rxq) {
6113 #ifdef atomic_store_rel_8
6114 		atomic_store_rel_8(&rxq->iq.intr_params, v);
6115 #else
6116 		rxq->iq.intr_params = v;
6117 #endif
6118 	}
6119 	vi->tmr_idx = idx;
6120 
6121 	end_synchronized_op(sc, LOCK_HELD);
6122 	return (0);
6123 }
6124 
6125 static int
6126 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
6127 {
6128 	struct vi_info *vi = arg1;
6129 	struct adapter *sc = vi->pi->adapter;
6130 	int idx, rc;
6131 
6132 	idx = vi->pktc_idx;
6133 
6134 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6135 	if (rc != 0 || req->newptr == NULL)
6136 		return (rc);
6137 
6138 	if (idx < -1 || idx >= SGE_NCOUNTERS)
6139 		return (EINVAL);
6140 
6141 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6142 	    "t4pktc");
6143 	if (rc)
6144 		return (rc);
6145 
6146 	if (vi->flags & VI_INIT_DONE)
6147 		rc = EBUSY; /* cannot be changed once the queues are created */
6148 	else
6149 		vi->pktc_idx = idx;
6150 
6151 	end_synchronized_op(sc, LOCK_HELD);
6152 	return (rc);
6153 }
6154 
6155 static int
6156 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
6157 {
6158 	struct vi_info *vi = arg1;
6159 	struct adapter *sc = vi->pi->adapter;
6160 	int qsize, rc;
6161 
6162 	qsize = vi->qsize_rxq;
6163 
6164 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6165 	if (rc != 0 || req->newptr == NULL)
6166 		return (rc);
6167 
6168 	if (qsize < 128 || (qsize & 7))
6169 		return (EINVAL);
6170 
6171 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6172 	    "t4rxqs");
6173 	if (rc)
6174 		return (rc);
6175 
6176 	if (vi->flags & VI_INIT_DONE)
6177 		rc = EBUSY; /* cannot be changed once the queues are created */
6178 	else
6179 		vi->qsize_rxq = qsize;
6180 
6181 	end_synchronized_op(sc, LOCK_HELD);
6182 	return (rc);
6183 }
6184 
6185 static int
6186 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
6187 {
6188 	struct vi_info *vi = arg1;
6189 	struct adapter *sc = vi->pi->adapter;
6190 	int qsize, rc;
6191 
6192 	qsize = vi->qsize_txq;
6193 
6194 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6195 	if (rc != 0 || req->newptr == NULL)
6196 		return (rc);
6197 
6198 	if (qsize < 128 || qsize > 65536)
6199 		return (EINVAL);
6200 
6201 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6202 	    "t4txqs");
6203 	if (rc)
6204 		return (rc);
6205 
6206 	if (vi->flags & VI_INIT_DONE)
6207 		rc = EBUSY; /* cannot be changed once the queues are created */
6208 	else
6209 		vi->qsize_txq = qsize;
6210 
6211 	end_synchronized_op(sc, LOCK_HELD);
6212 	return (rc);
6213 }
6214 
6215 static int
6216 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
6217 {
6218 	struct port_info *pi = arg1;
6219 	struct adapter *sc = pi->adapter;
6220 	struct link_config *lc = &pi->link_cfg;
6221 	int rc;
6222 
6223 	if (req->newptr == NULL) {
6224 		struct sbuf *sb;
6225 		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
6226 
6227 		rc = sysctl_wire_old_buffer(req, 0);
6228 		if (rc != 0)
6229 			return(rc);
6230 
6231 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6232 		if (sb == NULL)
6233 			return (ENOMEM);
6234 
6235 		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
6236 		rc = sbuf_finish(sb);
6237 		sbuf_delete(sb);
6238 	} else {
6239 		char s[2];
6240 		int n;
6241 
6242 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
6243 		s[1] = 0;
6244 
6245 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6246 		if (rc != 0)
6247 			return(rc);
6248 
6249 		if (s[1] != 0)
6250 			return (EINVAL);
6251 		if (s[0] < '0' || s[0] > '9')
6252 			return (EINVAL);	/* not a number */
6253 		n = s[0] - '0';
6254 		if (n & ~(PAUSE_TX | PAUSE_RX))
6255 			return (EINVAL);	/* some other bit is set too */
6256 
6257 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6258 		    "t4PAUSE");
6259 		if (rc)
6260 			return (rc);
6261 		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
6262 			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
6263 			lc->requested_fc |= n;
6264 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6265 			if (rc == 0) {
6266 				lc->fc = lc->requested_fc;
6267 			}
6268 		}
6269 		end_synchronized_op(sc, 0);
6270 	}
6271 
6272 	return (rc);
6273 }
6274 
6275 static int
6276 sysctl_fec(SYSCTL_HANDLER_ARGS)
6277 {
6278 	struct port_info *pi = arg1;
6279 	struct adapter *sc = pi->adapter;
6280 	struct link_config *lc = &pi->link_cfg;
6281 	int rc;
6282 
6283 	if (req->newptr == NULL) {
6284 		struct sbuf *sb;
6285 		static char *bits = "\20\1RS\2BASER_RS\3RESERVED";
6286 
6287 		rc = sysctl_wire_old_buffer(req, 0);
6288 		if (rc != 0)
6289 			return(rc);
6290 
6291 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6292 		if (sb == NULL)
6293 			return (ENOMEM);
6294 
6295 		sbuf_printf(sb, "%b", lc->fec & M_FW_PORT_CAP_FEC, bits);
6296 		rc = sbuf_finish(sb);
6297 		sbuf_delete(sb);
6298 	} else {
6299 		char s[2];
6300 		int n;
6301 
6302 		s[0] = '0' + (lc->requested_fec & M_FW_PORT_CAP_FEC);
6303 		s[1] = 0;
6304 
6305 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6306 		if (rc != 0)
6307 			return(rc);
6308 
6309 		if (s[1] != 0)
6310 			return (EINVAL);
6311 		if (s[0] < '0' || s[0] > '9')
6312 			return (EINVAL);	/* not a number */
6313 		n = s[0] - '0';
6314 		if (n & ~M_FW_PORT_CAP_FEC)
6315 			return (EINVAL);	/* some other bit is set too */
6316 
6317 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6318 		    "t4fec");
6319 		if (rc)
6320 			return (rc);
6321 		if ((lc->requested_fec & M_FW_PORT_CAP_FEC) != n) {
6322 			lc->requested_fec = n &
6323 			    G_FW_PORT_CAP_FEC(lc->supported);
6324 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6325 			if (rc == 0) {
6326 				lc->fec = lc->requested_fec;
6327 			}
6328 		}
6329 		end_synchronized_op(sc, 0);
6330 	}
6331 
6332 	return (rc);
6333 }
6334 
6335 static int
6336 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
6337 {
6338 	struct port_info *pi = arg1;
6339 	struct adapter *sc = pi->adapter;
6340 	struct link_config *lc = &pi->link_cfg;
6341 	int rc, val, old;
6342 
6343 	if (lc->supported & FW_PORT_CAP_ANEG)
6344 		val = lc->requested_aneg == AUTONEG_ENABLE ? 1 : 0;
6345 	else
6346 		val = -1;
6347 	rc = sysctl_handle_int(oidp, &val, 0, req);
6348 	if (rc != 0 || req->newptr == NULL)
6349 		return (rc);
6350 	if ((lc->supported & FW_PORT_CAP_ANEG) == 0)
6351 		return (ENOTSUP);
6352 
6353 	if (val == 0)
6354 		val = AUTONEG_DISABLE;
6355 	else if (val == 1)
6356 		val = AUTONEG_ENABLE;
6357 	else
6358 		return (EINVAL);
6359 	if (lc->requested_aneg == val)
6360 		return (0);	/* no change */
6361 
6362 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6363 	    "t4aneg");
6364 	if (rc)
6365 		return (rc);
6366 	old = lc->requested_aneg;
6367 	lc->requested_aneg = val;
6368 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6369 	if (rc != 0)
6370 		lc->requested_aneg = old;
6371 	end_synchronized_op(sc, 0);
6372 	return (rc);
6373 }
6374 
6375 static int
6376 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
6377 {
6378 	struct adapter *sc = arg1;
6379 	int reg = arg2;
6380 	uint64_t val;
6381 
6382 	val = t4_read_reg64(sc, reg);
6383 
6384 	return (sysctl_handle_64(oidp, &val, 0, req));
6385 }
6386 
6387 static int
6388 sysctl_temperature(SYSCTL_HANDLER_ARGS)
6389 {
6390 	struct adapter *sc = arg1;
6391 	int rc, t;
6392 	uint32_t param, val;
6393 
6394 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
6395 	if (rc)
6396 		return (rc);
6397 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6398 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
6399 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
6400 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6401 	end_synchronized_op(sc, 0);
6402 	if (rc)
6403 		return (rc);
6404 
6405 	/* unknown is returned as 0 but we display -1 in that case */
6406 	t = val == 0 ? -1 : val;
6407 
6408 	rc = sysctl_handle_int(oidp, &t, 0, req);
6409 	return (rc);
6410 }
6411 
6412 #ifdef SBUF_DRAIN
6413 static int
6414 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
6415 {
6416 	struct adapter *sc = arg1;
6417 	struct sbuf *sb;
6418 	int rc, i;
6419 	uint16_t incr[NMTUS][NCCTRL_WIN];
6420 	static const char *dec_fac[] = {
6421 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
6422 		"0.9375"
6423 	};
6424 
6425 	rc = sysctl_wire_old_buffer(req, 0);
6426 	if (rc != 0)
6427 		return (rc);
6428 
6429 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6430 	if (sb == NULL)
6431 		return (ENOMEM);
6432 
6433 	t4_read_cong_tbl(sc, incr);
6434 
6435 	for (i = 0; i < NCCTRL_WIN; ++i) {
6436 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
6437 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
6438 		    incr[5][i], incr[6][i], incr[7][i]);
6439 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
6440 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
6441 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
6442 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
6443 	}
6444 
6445 	rc = sbuf_finish(sb);
6446 	sbuf_delete(sb);
6447 
6448 	return (rc);
6449 }
6450 
6451 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
6452 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
6453 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
6454 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
6455 };
6456 
6457 static int
6458 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
6459 {
6460 	struct adapter *sc = arg1;
6461 	struct sbuf *sb;
6462 	int rc, i, n, qid = arg2;
6463 	uint32_t *buf, *p;
6464 	char *qtype;
6465 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
6466 
6467 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
6468 	    ("%s: bad qid %d\n", __func__, qid));
6469 
6470 	if (qid < CIM_NUM_IBQ) {
6471 		/* inbound queue */
6472 		qtype = "IBQ";
6473 		n = 4 * CIM_IBQ_SIZE;
6474 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6475 		rc = t4_read_cim_ibq(sc, qid, buf, n);
6476 	} else {
6477 		/* outbound queue */
6478 		qtype = "OBQ";
6479 		qid -= CIM_NUM_IBQ;
6480 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
6481 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6482 		rc = t4_read_cim_obq(sc, qid, buf, n);
6483 	}
6484 
6485 	if (rc < 0) {
6486 		rc = -rc;
6487 		goto done;
6488 	}
6489 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
6490 
6491 	rc = sysctl_wire_old_buffer(req, 0);
6492 	if (rc != 0)
6493 		goto done;
6494 
6495 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6496 	if (sb == NULL) {
6497 		rc = ENOMEM;
6498 		goto done;
6499 	}
6500 
6501 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
6502 	for (i = 0, p = buf; i < n; i += 16, p += 4)
6503 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
6504 		    p[2], p[3]);
6505 
6506 	rc = sbuf_finish(sb);
6507 	sbuf_delete(sb);
6508 done:
6509 	free(buf, M_CXGBE);
6510 	return (rc);
6511 }
6512 
6513 static int
6514 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
6515 {
6516 	struct adapter *sc = arg1;
6517 	u_int cfg;
6518 	struct sbuf *sb;
6519 	uint32_t *buf, *p;
6520 	int rc;
6521 
6522 	MPASS(chip_id(sc) <= CHELSIO_T5);
6523 
6524 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6525 	if (rc != 0)
6526 		return (rc);
6527 
6528 	rc = sysctl_wire_old_buffer(req, 0);
6529 	if (rc != 0)
6530 		return (rc);
6531 
6532 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6533 	if (sb == NULL)
6534 		return (ENOMEM);
6535 
6536 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6537 	    M_ZERO | M_WAITOK);
6538 
6539 	rc = -t4_cim_read_la(sc, buf, NULL);
6540 	if (rc != 0)
6541 		goto done;
6542 
6543 	sbuf_printf(sb, "Status   Data      PC%s",
6544 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6545 	    "     LS0Stat  LS0Addr             LS0Data");
6546 
6547 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
6548 		if (cfg & F_UPDBGLACAPTPCONLY) {
6549 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
6550 			    p[6], p[7]);
6551 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
6552 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
6553 			    p[4] & 0xff, p[5] >> 8);
6554 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
6555 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6556 			    p[1] & 0xf, p[2] >> 4);
6557 		} else {
6558 			sbuf_printf(sb,
6559 			    "\n  %02x   %x%07x %x%07x %08x %08x "
6560 			    "%08x%08x%08x%08x",
6561 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6562 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
6563 			    p[6], p[7]);
6564 		}
6565 	}
6566 
6567 	rc = sbuf_finish(sb);
6568 	sbuf_delete(sb);
6569 done:
6570 	free(buf, M_CXGBE);
6571 	return (rc);
6572 }
6573 
6574 static int
6575 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
6576 {
6577 	struct adapter *sc = arg1;
6578 	u_int cfg;
6579 	struct sbuf *sb;
6580 	uint32_t *buf, *p;
6581 	int rc;
6582 
6583 	MPASS(chip_id(sc) > CHELSIO_T5);
6584 
6585 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6586 	if (rc != 0)
6587 		return (rc);
6588 
6589 	rc = sysctl_wire_old_buffer(req, 0);
6590 	if (rc != 0)
6591 		return (rc);
6592 
6593 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6594 	if (sb == NULL)
6595 		return (ENOMEM);
6596 
6597 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6598 	    M_ZERO | M_WAITOK);
6599 
6600 	rc = -t4_cim_read_la(sc, buf, NULL);
6601 	if (rc != 0)
6602 		goto done;
6603 
6604 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
6605 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6606 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
6607 
6608 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
6609 		if (cfg & F_UPDBGLACAPTPCONLY) {
6610 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
6611 			    p[3] & 0xff, p[2], p[1], p[0]);
6612 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
6613 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
6614 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
6615 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
6616 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
6617 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
6618 			    p[6] >> 16);
6619 		} else {
6620 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
6621 			    "%08x %08x %08x %08x %08x %08x",
6622 			    (p[9] >> 16) & 0xff,
6623 			    p[9] & 0xffff, p[8] >> 16,
6624 			    p[8] & 0xffff, p[7] >> 16,
6625 			    p[7] & 0xffff, p[6] >> 16,
6626 			    p[2], p[1], p[0], p[5], p[4], p[3]);
6627 		}
6628 	}
6629 
6630 	rc = sbuf_finish(sb);
6631 	sbuf_delete(sb);
6632 done:
6633 	free(buf, M_CXGBE);
6634 	return (rc);
6635 }
6636 
6637 static int
6638 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
6639 {
6640 	struct adapter *sc = arg1;
6641 	u_int i;
6642 	struct sbuf *sb;
6643 	uint32_t *buf, *p;
6644 	int rc;
6645 
6646 	rc = sysctl_wire_old_buffer(req, 0);
6647 	if (rc != 0)
6648 		return (rc);
6649 
6650 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6651 	if (sb == NULL)
6652 		return (ENOMEM);
6653 
6654 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
6655 	    M_ZERO | M_WAITOK);
6656 
6657 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
6658 	p = buf;
6659 
6660 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6661 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
6662 		    p[1], p[0]);
6663 	}
6664 
6665 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
6666 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6667 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
6668 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
6669 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
6670 		    (p[1] >> 2) | ((p[2] & 3) << 30),
6671 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
6672 		    p[0] & 1);
6673 	}
6674 
6675 	rc = sbuf_finish(sb);
6676 	sbuf_delete(sb);
6677 	free(buf, M_CXGBE);
6678 	return (rc);
6679 }
6680 
6681 static int
6682 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
6683 {
6684 	struct adapter *sc = arg1;
6685 	u_int i;
6686 	struct sbuf *sb;
6687 	uint32_t *buf, *p;
6688 	int rc;
6689 
6690 	rc = sysctl_wire_old_buffer(req, 0);
6691 	if (rc != 0)
6692 		return (rc);
6693 
6694 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6695 	if (sb == NULL)
6696 		return (ENOMEM);
6697 
6698 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
6699 	    M_ZERO | M_WAITOK);
6700 
6701 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
6702 	p = buf;
6703 
6704 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
6705 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6706 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
6707 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
6708 		    p[4], p[3], p[2], p[1], p[0]);
6709 	}
6710 
6711 	sbuf_printf(sb, "\n\nCntl ID               Data");
6712 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6713 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
6714 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
6715 	}
6716 
6717 	rc = sbuf_finish(sb);
6718 	sbuf_delete(sb);
6719 	free(buf, M_CXGBE);
6720 	return (rc);
6721 }
6722 
6723 static int
6724 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
6725 {
6726 	struct adapter *sc = arg1;
6727 	struct sbuf *sb;
6728 	int rc, i;
6729 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6730 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6731 	uint16_t thres[CIM_NUM_IBQ];
6732 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
6733 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
6734 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
6735 
6736 	cim_num_obq = sc->chip_params->cim_num_obq;
6737 	if (is_t4(sc)) {
6738 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
6739 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
6740 	} else {
6741 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
6742 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
6743 	}
6744 	nq = CIM_NUM_IBQ + cim_num_obq;
6745 
6746 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
6747 	if (rc == 0)
6748 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
6749 	if (rc != 0)
6750 		return (rc);
6751 
6752 	t4_read_cimq_cfg(sc, base, size, thres);
6753 
6754 	rc = sysctl_wire_old_buffer(req, 0);
6755 	if (rc != 0)
6756 		return (rc);
6757 
6758 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6759 	if (sb == NULL)
6760 		return (ENOMEM);
6761 
6762 	sbuf_printf(sb,
6763 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
6764 
6765 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
6766 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
6767 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
6768 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6769 		    G_QUEREMFLITS(p[2]) * 16);
6770 	for ( ; i < nq; i++, p += 4, wr += 2)
6771 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
6772 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
6773 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6774 		    G_QUEREMFLITS(p[2]) * 16);
6775 
6776 	rc = sbuf_finish(sb);
6777 	sbuf_delete(sb);
6778 
6779 	return (rc);
6780 }
6781 
6782 static int
6783 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
6784 {
6785 	struct adapter *sc = arg1;
6786 	struct sbuf *sb;
6787 	int rc;
6788 	struct tp_cpl_stats stats;
6789 
6790 	rc = sysctl_wire_old_buffer(req, 0);
6791 	if (rc != 0)
6792 		return (rc);
6793 
6794 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6795 	if (sb == NULL)
6796 		return (ENOMEM);
6797 
6798 	mtx_lock(&sc->reg_lock);
6799 	t4_tp_get_cpl_stats(sc, &stats, 0);
6800 	mtx_unlock(&sc->reg_lock);
6801 
6802 	if (sc->chip_params->nchan > 2) {
6803 		sbuf_printf(sb, "                 channel 0  channel 1"
6804 		    "  channel 2  channel 3");
6805 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
6806 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
6807 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
6808 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
6809 	} else {
6810 		sbuf_printf(sb, "                 channel 0  channel 1");
6811 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
6812 		    stats.req[0], stats.req[1]);
6813 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
6814 		    stats.rsp[0], stats.rsp[1]);
6815 	}
6816 
6817 	rc = sbuf_finish(sb);
6818 	sbuf_delete(sb);
6819 
6820 	return (rc);
6821 }
6822 
6823 static int
6824 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
6825 {
6826 	struct adapter *sc = arg1;
6827 	struct sbuf *sb;
6828 	int rc;
6829 	struct tp_usm_stats stats;
6830 
6831 	rc = sysctl_wire_old_buffer(req, 0);
6832 	if (rc != 0)
6833 		return(rc);
6834 
6835 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6836 	if (sb == NULL)
6837 		return (ENOMEM);
6838 
6839 	t4_get_usm_stats(sc, &stats, 1);
6840 
6841 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
6842 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
6843 	sbuf_printf(sb, "Drops:  %u", stats.drops);
6844 
6845 	rc = sbuf_finish(sb);
6846 	sbuf_delete(sb);
6847 
6848 	return (rc);
6849 }
6850 
6851 static const char * const devlog_level_strings[] = {
6852 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
6853 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
6854 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
6855 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
6856 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
6857 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
6858 };
6859 
6860 static const char * const devlog_facility_strings[] = {
6861 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
6862 	[FW_DEVLOG_FACILITY_CF]		= "CF",
6863 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
6864 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
6865 	[FW_DEVLOG_FACILITY_RES]	= "RES",
6866 	[FW_DEVLOG_FACILITY_HW]		= "HW",
6867 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
6868 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
6869 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
6870 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
6871 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
6872 	[FW_DEVLOG_FACILITY_VI]		= "VI",
6873 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
6874 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
6875 	[FW_DEVLOG_FACILITY_TM]		= "TM",
6876 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
6877 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
6878 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
6879 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
6880 	[FW_DEVLOG_FACILITY_RI]		= "RI",
6881 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
6882 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
6883 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
6884 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
6885 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
6886 };
6887 
6888 static int
6889 sysctl_devlog(SYSCTL_HANDLER_ARGS)
6890 {
6891 	struct adapter *sc = arg1;
6892 	struct devlog_params *dparams = &sc->params.devlog;
6893 	struct fw_devlog_e *buf, *e;
6894 	int i, j, rc, nentries, first = 0;
6895 	struct sbuf *sb;
6896 	uint64_t ftstamp = UINT64_MAX;
6897 
6898 	if (dparams->addr == 0)
6899 		return (ENXIO);
6900 
6901 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
6902 	if (buf == NULL)
6903 		return (ENOMEM);
6904 
6905 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
6906 	if (rc != 0)
6907 		goto done;
6908 
6909 	nentries = dparams->size / sizeof(struct fw_devlog_e);
6910 	for (i = 0; i < nentries; i++) {
6911 		e = &buf[i];
6912 
6913 		if (e->timestamp == 0)
6914 			break;	/* end */
6915 
6916 		e->timestamp = be64toh(e->timestamp);
6917 		e->seqno = be32toh(e->seqno);
6918 		for (j = 0; j < 8; j++)
6919 			e->params[j] = be32toh(e->params[j]);
6920 
6921 		if (e->timestamp < ftstamp) {
6922 			ftstamp = e->timestamp;
6923 			first = i;
6924 		}
6925 	}
6926 
6927 	if (buf[first].timestamp == 0)
6928 		goto done;	/* nothing in the log */
6929 
6930 	rc = sysctl_wire_old_buffer(req, 0);
6931 	if (rc != 0)
6932 		goto done;
6933 
6934 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6935 	if (sb == NULL) {
6936 		rc = ENOMEM;
6937 		goto done;
6938 	}
6939 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
6940 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
6941 
6942 	i = first;
6943 	do {
6944 		e = &buf[i];
6945 		if (e->timestamp == 0)
6946 			break;	/* end */
6947 
6948 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
6949 		    e->seqno, e->timestamp,
6950 		    (e->level < nitems(devlog_level_strings) ?
6951 			devlog_level_strings[e->level] : "UNKNOWN"),
6952 		    (e->facility < nitems(devlog_facility_strings) ?
6953 			devlog_facility_strings[e->facility] : "UNKNOWN"));
6954 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
6955 		    e->params[2], e->params[3], e->params[4],
6956 		    e->params[5], e->params[6], e->params[7]);
6957 
6958 		if (++i == nentries)
6959 			i = 0;
6960 	} while (i != first);
6961 
6962 	rc = sbuf_finish(sb);
6963 	sbuf_delete(sb);
6964 done:
6965 	free(buf, M_CXGBE);
6966 	return (rc);
6967 }
6968 
6969 static int
6970 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
6971 {
6972 	struct adapter *sc = arg1;
6973 	struct sbuf *sb;
6974 	int rc;
6975 	struct tp_fcoe_stats stats[MAX_NCHAN];
6976 	int i, nchan = sc->chip_params->nchan;
6977 
6978 	rc = sysctl_wire_old_buffer(req, 0);
6979 	if (rc != 0)
6980 		return (rc);
6981 
6982 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6983 	if (sb == NULL)
6984 		return (ENOMEM);
6985 
6986 	for (i = 0; i < nchan; i++)
6987 		t4_get_fcoe_stats(sc, i, &stats[i], 1);
6988 
6989 	if (nchan > 2) {
6990 		sbuf_printf(sb, "                   channel 0        channel 1"
6991 		    "        channel 2        channel 3");
6992 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
6993 		    stats[0].octets_ddp, stats[1].octets_ddp,
6994 		    stats[2].octets_ddp, stats[3].octets_ddp);
6995 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
6996 		    stats[0].frames_ddp, stats[1].frames_ddp,
6997 		    stats[2].frames_ddp, stats[3].frames_ddp);
6998 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
6999 		    stats[0].frames_drop, stats[1].frames_drop,
7000 		    stats[2].frames_drop, stats[3].frames_drop);
7001 	} else {
7002 		sbuf_printf(sb, "                   channel 0        channel 1");
7003 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
7004 		    stats[0].octets_ddp, stats[1].octets_ddp);
7005 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
7006 		    stats[0].frames_ddp, stats[1].frames_ddp);
7007 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
7008 		    stats[0].frames_drop, stats[1].frames_drop);
7009 	}
7010 
7011 	rc = sbuf_finish(sb);
7012 	sbuf_delete(sb);
7013 
7014 	return (rc);
7015 }
7016 
7017 static int
7018 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
7019 {
7020 	struct adapter *sc = arg1;
7021 	struct sbuf *sb;
7022 	int rc, i;
7023 	unsigned int map, kbps, ipg, mode;
7024 	unsigned int pace_tab[NTX_SCHED];
7025 
7026 	rc = sysctl_wire_old_buffer(req, 0);
7027 	if (rc != 0)
7028 		return (rc);
7029 
7030 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7031 	if (sb == NULL)
7032 		return (ENOMEM);
7033 
7034 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
7035 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
7036 	t4_read_pace_tbl(sc, pace_tab);
7037 
7038 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
7039 	    "Class IPG (0.1 ns)   Flow IPG (us)");
7040 
7041 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
7042 		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
7043 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
7044 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
7045 		if (kbps)
7046 			sbuf_printf(sb, "%9u     ", kbps);
7047 		else
7048 			sbuf_printf(sb, " disabled     ");
7049 
7050 		if (ipg)
7051 			sbuf_printf(sb, "%13u        ", ipg);
7052 		else
7053 			sbuf_printf(sb, "     disabled        ");
7054 
7055 		if (pace_tab[i])
7056 			sbuf_printf(sb, "%10u", pace_tab[i]);
7057 		else
7058 			sbuf_printf(sb, "  disabled");
7059 	}
7060 
7061 	rc = sbuf_finish(sb);
7062 	sbuf_delete(sb);
7063 
7064 	return (rc);
7065 }
7066 
7067 static int
7068 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
7069 {
7070 	struct adapter *sc = arg1;
7071 	struct sbuf *sb;
7072 	int rc, i, j;
7073 	uint64_t *p0, *p1;
7074 	struct lb_port_stats s[2];
7075 	static const char *stat_name[] = {
7076 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
7077 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
7078 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
7079 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
7080 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
7081 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
7082 		"BG2FramesTrunc:", "BG3FramesTrunc:"
7083 	};
7084 
7085 	rc = sysctl_wire_old_buffer(req, 0);
7086 	if (rc != 0)
7087 		return (rc);
7088 
7089 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7090 	if (sb == NULL)
7091 		return (ENOMEM);
7092 
7093 	memset(s, 0, sizeof(s));
7094 
7095 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
7096 		t4_get_lb_stats(sc, i, &s[0]);
7097 		t4_get_lb_stats(sc, i + 1, &s[1]);
7098 
7099 		p0 = &s[0].octets;
7100 		p1 = &s[1].octets;
7101 		sbuf_printf(sb, "%s                       Loopback %u"
7102 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
7103 
7104 		for (j = 0; j < nitems(stat_name); j++)
7105 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
7106 				   *p0++, *p1++);
7107 	}
7108 
7109 	rc = sbuf_finish(sb);
7110 	sbuf_delete(sb);
7111 
7112 	return (rc);
7113 }
7114 
7115 static int
7116 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
7117 {
7118 	int rc = 0;
7119 	struct port_info *pi = arg1;
7120 	struct link_config *lc = &pi->link_cfg;
7121 	struct sbuf *sb;
7122 
7123 	rc = sysctl_wire_old_buffer(req, 0);
7124 	if (rc != 0)
7125 		return(rc);
7126 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
7127 	if (sb == NULL)
7128 		return (ENOMEM);
7129 
7130 	if (lc->link_ok || lc->link_down_rc == 255)
7131 		sbuf_printf(sb, "n/a");
7132 	else
7133 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
7134 
7135 	rc = sbuf_finish(sb);
7136 	sbuf_delete(sb);
7137 
7138 	return (rc);
7139 }
7140 
7141 struct mem_desc {
7142 	unsigned int base;
7143 	unsigned int limit;
7144 	unsigned int idx;
7145 };
7146 
7147 static int
7148 mem_desc_cmp(const void *a, const void *b)
7149 {
7150 	return ((const struct mem_desc *)a)->base -
7151 	       ((const struct mem_desc *)b)->base;
7152 }
7153 
7154 static void
7155 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
7156     unsigned int to)
7157 {
7158 	unsigned int size;
7159 
7160 	if (from == to)
7161 		return;
7162 
7163 	size = to - from + 1;
7164 	if (size == 0)
7165 		return;
7166 
7167 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
7168 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
7169 }
7170 
7171 static int
7172 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
7173 {
7174 	struct adapter *sc = arg1;
7175 	struct sbuf *sb;
7176 	int rc, i, n;
7177 	uint32_t lo, hi, used, alloc;
7178 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
7179 	static const char *region[] = {
7180 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
7181 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
7182 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
7183 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
7184 		"RQUDP region:", "PBL region:", "TXPBL region:",
7185 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
7186 		"On-chip queues:", "TLS keys:",
7187 	};
7188 	struct mem_desc avail[4];
7189 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
7190 	struct mem_desc *md = mem;
7191 
7192 	rc = sysctl_wire_old_buffer(req, 0);
7193 	if (rc != 0)
7194 		return (rc);
7195 
7196 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7197 	if (sb == NULL)
7198 		return (ENOMEM);
7199 
7200 	for (i = 0; i < nitems(mem); i++) {
7201 		mem[i].limit = 0;
7202 		mem[i].idx = i;
7203 	}
7204 
7205 	/* Find and sort the populated memory ranges */
7206 	i = 0;
7207 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
7208 	if (lo & F_EDRAM0_ENABLE) {
7209 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
7210 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
7211 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
7212 		avail[i].idx = 0;
7213 		i++;
7214 	}
7215 	if (lo & F_EDRAM1_ENABLE) {
7216 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
7217 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
7218 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
7219 		avail[i].idx = 1;
7220 		i++;
7221 	}
7222 	if (lo & F_EXT_MEM_ENABLE) {
7223 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
7224 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
7225 		avail[i].limit = avail[i].base +
7226 		    (G_EXT_MEM_SIZE(hi) << 20);
7227 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
7228 		i++;
7229 	}
7230 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
7231 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
7232 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
7233 		avail[i].limit = avail[i].base +
7234 		    (G_EXT_MEM1_SIZE(hi) << 20);
7235 		avail[i].idx = 4;
7236 		i++;
7237 	}
7238 	if (!i)                                    /* no memory available */
7239 		return 0;
7240 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
7241 
7242 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
7243 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
7244 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
7245 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
7246 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
7247 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
7248 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
7249 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
7250 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
7251 
7252 	/* the next few have explicit upper bounds */
7253 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
7254 	md->limit = md->base - 1 +
7255 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
7256 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
7257 	md++;
7258 
7259 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
7260 	md->limit = md->base - 1 +
7261 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
7262 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
7263 	md++;
7264 
7265 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7266 		if (chip_id(sc) <= CHELSIO_T5)
7267 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
7268 		else
7269 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
7270 		md->limit = 0;
7271 	} else {
7272 		md->base = 0;
7273 		md->idx = nitems(region);  /* hide it */
7274 	}
7275 	md++;
7276 
7277 #define ulp_region(reg) \
7278 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
7279 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
7280 
7281 	ulp_region(RX_ISCSI);
7282 	ulp_region(RX_TDDP);
7283 	ulp_region(TX_TPT);
7284 	ulp_region(RX_STAG);
7285 	ulp_region(RX_RQ);
7286 	ulp_region(RX_RQUDP);
7287 	ulp_region(RX_PBL);
7288 	ulp_region(TX_PBL);
7289 #undef ulp_region
7290 
7291 	md->base = 0;
7292 	md->idx = nitems(region);
7293 	if (!is_t4(sc)) {
7294 		uint32_t size = 0;
7295 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
7296 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
7297 
7298 		if (is_t5(sc)) {
7299 			if (sge_ctrl & F_VFIFO_ENABLE)
7300 				size = G_DBVFIFO_SIZE(fifo_size);
7301 		} else
7302 			size = G_T6_DBVFIFO_SIZE(fifo_size);
7303 
7304 		if (size) {
7305 			md->base = G_BASEADDR(t4_read_reg(sc,
7306 			    A_SGE_DBVFIFO_BADDR));
7307 			md->limit = md->base + (size << 2) - 1;
7308 		}
7309 	}
7310 	md++;
7311 
7312 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
7313 	md->limit = 0;
7314 	md++;
7315 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
7316 	md->limit = 0;
7317 	md++;
7318 
7319 	md->base = sc->vres.ocq.start;
7320 	if (sc->vres.ocq.size)
7321 		md->limit = md->base + sc->vres.ocq.size - 1;
7322 	else
7323 		md->idx = nitems(region);  /* hide it */
7324 	md++;
7325 
7326 	md->base = sc->vres.key.start;
7327 	if (sc->vres.key.size)
7328 		md->limit = md->base + sc->vres.key.size - 1;
7329 	else
7330 		md->idx = nitems(region);  /* hide it */
7331 	md++;
7332 
7333 	/* add any address-space holes, there can be up to 3 */
7334 	for (n = 0; n < i - 1; n++)
7335 		if (avail[n].limit < avail[n + 1].base)
7336 			(md++)->base = avail[n].limit;
7337 	if (avail[n].limit)
7338 		(md++)->base = avail[n].limit;
7339 
7340 	n = md - mem;
7341 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
7342 
7343 	for (lo = 0; lo < i; lo++)
7344 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
7345 				avail[lo].limit - 1);
7346 
7347 	sbuf_printf(sb, "\n");
7348 	for (i = 0; i < n; i++) {
7349 		if (mem[i].idx >= nitems(region))
7350 			continue;                        /* skip holes */
7351 		if (!mem[i].limit)
7352 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
7353 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
7354 				mem[i].limit);
7355 	}
7356 
7357 	sbuf_printf(sb, "\n");
7358 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
7359 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
7360 	mem_region_show(sb, "uP RAM:", lo, hi);
7361 
7362 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
7363 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
7364 	mem_region_show(sb, "uP Extmem2:", lo, hi);
7365 
7366 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
7367 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
7368 		   G_PMRXMAXPAGE(lo),
7369 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
7370 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
7371 
7372 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
7373 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
7374 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
7375 		   G_PMTXMAXPAGE(lo),
7376 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
7377 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
7378 	sbuf_printf(sb, "%u p-structs\n",
7379 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
7380 
7381 	for (i = 0; i < 4; i++) {
7382 		if (chip_id(sc) > CHELSIO_T5)
7383 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
7384 		else
7385 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
7386 		if (is_t5(sc)) {
7387 			used = G_T5_USED(lo);
7388 			alloc = G_T5_ALLOC(lo);
7389 		} else {
7390 			used = G_USED(lo);
7391 			alloc = G_ALLOC(lo);
7392 		}
7393 		/* For T6 these are MAC buffer groups */
7394 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
7395 		    i, used, alloc);
7396 	}
7397 	for (i = 0; i < sc->chip_params->nchan; i++) {
7398 		if (chip_id(sc) > CHELSIO_T5)
7399 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
7400 		else
7401 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
7402 		if (is_t5(sc)) {
7403 			used = G_T5_USED(lo);
7404 			alloc = G_T5_ALLOC(lo);
7405 		} else {
7406 			used = G_USED(lo);
7407 			alloc = G_ALLOC(lo);
7408 		}
7409 		/* For T6 these are MAC buffer groups */
7410 		sbuf_printf(sb,
7411 		    "\nLoopback %d using %u pages out of %u allocated",
7412 		    i, used, alloc);
7413 	}
7414 
7415 	rc = sbuf_finish(sb);
7416 	sbuf_delete(sb);
7417 
7418 	return (rc);
7419 }
7420 
7421 static inline void
7422 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
7423 {
7424 	*mask = x | y;
7425 	y = htobe64(y);
7426 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
7427 }
7428 
7429 static int
7430 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
7431 {
7432 	struct adapter *sc = arg1;
7433 	struct sbuf *sb;
7434 	int rc, i;
7435 
7436 	MPASS(chip_id(sc) <= CHELSIO_T5);
7437 
7438 	rc = sysctl_wire_old_buffer(req, 0);
7439 	if (rc != 0)
7440 		return (rc);
7441 
7442 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7443 	if (sb == NULL)
7444 		return (ENOMEM);
7445 
7446 	sbuf_printf(sb,
7447 	    "Idx  Ethernet address     Mask     Vld Ports PF"
7448 	    "  VF              Replication             P0 P1 P2 P3  ML");
7449 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7450 		uint64_t tcamx, tcamy, mask;
7451 		uint32_t cls_lo, cls_hi;
7452 		uint8_t addr[ETHER_ADDR_LEN];
7453 
7454 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
7455 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
7456 		if (tcamx & tcamy)
7457 			continue;
7458 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7459 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7460 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7461 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
7462 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
7463 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
7464 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
7465 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
7466 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
7467 
7468 		if (cls_lo & F_REPLICATE) {
7469 			struct fw_ldst_cmd ldst_cmd;
7470 
7471 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7472 			ldst_cmd.op_to_addrspace =
7473 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7474 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7475 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7476 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7477 			ldst_cmd.u.mps.rplc.fid_idx =
7478 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7479 				V_FW_LDST_CMD_IDX(i));
7480 
7481 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7482 			    "t4mps");
7483 			if (rc)
7484 				break;
7485 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7486 			    sizeof(ldst_cmd), &ldst_cmd);
7487 			end_synchronized_op(sc, 0);
7488 
7489 			if (rc != 0) {
7490 				sbuf_printf(sb, "%36d", rc);
7491 				rc = 0;
7492 			} else {
7493 				sbuf_printf(sb, " %08x %08x %08x %08x",
7494 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7495 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7496 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7497 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7498 			}
7499 		} else
7500 			sbuf_printf(sb, "%36s", "");
7501 
7502 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
7503 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
7504 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
7505 	}
7506 
7507 	if (rc)
7508 		(void) sbuf_finish(sb);
7509 	else
7510 		rc = sbuf_finish(sb);
7511 	sbuf_delete(sb);
7512 
7513 	return (rc);
7514 }
7515 
7516 static int
7517 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
7518 {
7519 	struct adapter *sc = arg1;
7520 	struct sbuf *sb;
7521 	int rc, i;
7522 
7523 	MPASS(chip_id(sc) > CHELSIO_T5);
7524 
7525 	rc = sysctl_wire_old_buffer(req, 0);
7526 	if (rc != 0)
7527 		return (rc);
7528 
7529 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7530 	if (sb == NULL)
7531 		return (ENOMEM);
7532 
7533 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
7534 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
7535 	    "                           Replication"
7536 	    "                                    P0 P1 P2 P3  ML\n");
7537 
7538 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7539 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
7540 		uint16_t ivlan;
7541 		uint64_t tcamx, tcamy, val, mask;
7542 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
7543 		uint8_t addr[ETHER_ADDR_LEN];
7544 
7545 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
7546 		if (i < 256)
7547 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
7548 		else
7549 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
7550 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7551 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7552 		tcamy = G_DMACH(val) << 32;
7553 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7554 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7555 		lookup_type = G_DATALKPTYPE(data2);
7556 		port_num = G_DATAPORTNUM(data2);
7557 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7558 			/* Inner header VNI */
7559 			vniy = ((data2 & F_DATAVIDH2) << 23) |
7560 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7561 			dip_hit = data2 & F_DATADIPHIT;
7562 			vlan_vld = 0;
7563 		} else {
7564 			vniy = 0;
7565 			dip_hit = 0;
7566 			vlan_vld = data2 & F_DATAVIDH2;
7567 			ivlan = G_VIDL(val);
7568 		}
7569 
7570 		ctl |= V_CTLXYBITSEL(1);
7571 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7572 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7573 		tcamx = G_DMACH(val) << 32;
7574 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7575 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7576 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7577 			/* Inner header VNI mask */
7578 			vnix = ((data2 & F_DATAVIDH2) << 23) |
7579 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7580 		} else
7581 			vnix = 0;
7582 
7583 		if (tcamx & tcamy)
7584 			continue;
7585 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7586 
7587 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7588 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7589 
7590 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7591 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7592 			    "%012jx %06x %06x    -    -   %3c"
7593 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
7594 			    addr[1], addr[2], addr[3], addr[4], addr[5],
7595 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
7596 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7597 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7598 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7599 		} else {
7600 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7601 			    "%012jx    -       -   ", i, addr[0], addr[1],
7602 			    addr[2], addr[3], addr[4], addr[5],
7603 			    (uintmax_t)mask);
7604 
7605 			if (vlan_vld)
7606 				sbuf_printf(sb, "%4u   Y     ", ivlan);
7607 			else
7608 				sbuf_printf(sb, "  -    N     ");
7609 
7610 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
7611 			    lookup_type ? 'I' : 'O', port_num,
7612 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7613 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7614 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7615 		}
7616 
7617 
7618 		if (cls_lo & F_T6_REPLICATE) {
7619 			struct fw_ldst_cmd ldst_cmd;
7620 
7621 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7622 			ldst_cmd.op_to_addrspace =
7623 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7624 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7625 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7626 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7627 			ldst_cmd.u.mps.rplc.fid_idx =
7628 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7629 				V_FW_LDST_CMD_IDX(i));
7630 
7631 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7632 			    "t6mps");
7633 			if (rc)
7634 				break;
7635 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7636 			    sizeof(ldst_cmd), &ldst_cmd);
7637 			end_synchronized_op(sc, 0);
7638 
7639 			if (rc != 0) {
7640 				sbuf_printf(sb, "%72d", rc);
7641 				rc = 0;
7642 			} else {
7643 				sbuf_printf(sb, " %08x %08x %08x %08x"
7644 				    " %08x %08x %08x %08x",
7645 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
7646 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
7647 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
7648 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
7649 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7650 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7651 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7652 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7653 			}
7654 		} else
7655 			sbuf_printf(sb, "%72s", "");
7656 
7657 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
7658 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
7659 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
7660 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
7661 	}
7662 
7663 	if (rc)
7664 		(void) sbuf_finish(sb);
7665 	else
7666 		rc = sbuf_finish(sb);
7667 	sbuf_delete(sb);
7668 
7669 	return (rc);
7670 }
7671 
7672 static int
7673 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
7674 {
7675 	struct adapter *sc = arg1;
7676 	struct sbuf *sb;
7677 	int rc;
7678 	uint16_t mtus[NMTUS];
7679 
7680 	rc = sysctl_wire_old_buffer(req, 0);
7681 	if (rc != 0)
7682 		return (rc);
7683 
7684 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7685 	if (sb == NULL)
7686 		return (ENOMEM);
7687 
7688 	t4_read_mtu_tbl(sc, mtus, NULL);
7689 
7690 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
7691 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
7692 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
7693 	    mtus[14], mtus[15]);
7694 
7695 	rc = sbuf_finish(sb);
7696 	sbuf_delete(sb);
7697 
7698 	return (rc);
7699 }
7700 
7701 static int
7702 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
7703 {
7704 	struct adapter *sc = arg1;
7705 	struct sbuf *sb;
7706 	int rc, i;
7707 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
7708 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
7709 	static const char *tx_stats[MAX_PM_NSTATS] = {
7710 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
7711 		"Tx FIFO wait", NULL, "Tx latency"
7712 	};
7713 	static const char *rx_stats[MAX_PM_NSTATS] = {
7714 		"Read:", "Write bypass:", "Write mem:", "Flush:",
7715 		"Rx FIFO wait", NULL, "Rx latency"
7716 	};
7717 
7718 	rc = sysctl_wire_old_buffer(req, 0);
7719 	if (rc != 0)
7720 		return (rc);
7721 
7722 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7723 	if (sb == NULL)
7724 		return (ENOMEM);
7725 
7726 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
7727 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
7728 
7729 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
7730 	for (i = 0; i < 4; i++) {
7731 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7732 		    tx_cyc[i]);
7733 	}
7734 
7735 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
7736 	for (i = 0; i < 4; i++) {
7737 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7738 		    rx_cyc[i]);
7739 	}
7740 
7741 	if (chip_id(sc) > CHELSIO_T5) {
7742 		sbuf_printf(sb,
7743 		    "\n              Total wait      Total occupancy");
7744 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7745 		    tx_cyc[i]);
7746 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7747 		    rx_cyc[i]);
7748 
7749 		i += 2;
7750 		MPASS(i < nitems(tx_stats));
7751 
7752 		sbuf_printf(sb,
7753 		    "\n                   Reads           Total wait");
7754 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7755 		    tx_cyc[i]);
7756 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7757 		    rx_cyc[i]);
7758 	}
7759 
7760 	rc = sbuf_finish(sb);
7761 	sbuf_delete(sb);
7762 
7763 	return (rc);
7764 }
7765 
7766 static int
7767 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
7768 {
7769 	struct adapter *sc = arg1;
7770 	struct sbuf *sb;
7771 	int rc;
7772 	struct tp_rdma_stats stats;
7773 
7774 	rc = sysctl_wire_old_buffer(req, 0);
7775 	if (rc != 0)
7776 		return (rc);
7777 
7778 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7779 	if (sb == NULL)
7780 		return (ENOMEM);
7781 
7782 	mtx_lock(&sc->reg_lock);
7783 	t4_tp_get_rdma_stats(sc, &stats, 0);
7784 	mtx_unlock(&sc->reg_lock);
7785 
7786 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
7787 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
7788 
7789 	rc = sbuf_finish(sb);
7790 	sbuf_delete(sb);
7791 
7792 	return (rc);
7793 }
7794 
7795 static int
7796 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
7797 {
7798 	struct adapter *sc = arg1;
7799 	struct sbuf *sb;
7800 	int rc;
7801 	struct tp_tcp_stats v4, v6;
7802 
7803 	rc = sysctl_wire_old_buffer(req, 0);
7804 	if (rc != 0)
7805 		return (rc);
7806 
7807 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7808 	if (sb == NULL)
7809 		return (ENOMEM);
7810 
7811 	mtx_lock(&sc->reg_lock);
7812 	t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
7813 	mtx_unlock(&sc->reg_lock);
7814 
7815 	sbuf_printf(sb,
7816 	    "                                IP                 IPv6\n");
7817 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
7818 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
7819 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
7820 	    v4.tcp_in_segs, v6.tcp_in_segs);
7821 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
7822 	    v4.tcp_out_segs, v6.tcp_out_segs);
7823 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
7824 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
7825 
7826 	rc = sbuf_finish(sb);
7827 	sbuf_delete(sb);
7828 
7829 	return (rc);
7830 }
7831 
7832 static int
7833 sysctl_tids(SYSCTL_HANDLER_ARGS)
7834 {
7835 	struct adapter *sc = arg1;
7836 	struct sbuf *sb;
7837 	int rc;
7838 	struct tid_info *t = &sc->tids;
7839 
7840 	rc = sysctl_wire_old_buffer(req, 0);
7841 	if (rc != 0)
7842 		return (rc);
7843 
7844 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7845 	if (sb == NULL)
7846 		return (ENOMEM);
7847 
7848 	if (t->natids) {
7849 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
7850 		    t->atids_in_use);
7851 	}
7852 
7853 	if (t->ntids) {
7854 		sbuf_printf(sb, "TID range: ");
7855 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7856 			uint32_t b, hb;
7857 
7858 			if (chip_id(sc) <= CHELSIO_T5) {
7859 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
7860 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
7861 			} else {
7862 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
7863 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
7864 			}
7865 
7866 			if (b)
7867 				sbuf_printf(sb, "0-%u, ", b - 1);
7868 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
7869 		} else
7870 			sbuf_printf(sb, "0-%u", t->ntids - 1);
7871 		sbuf_printf(sb, ", in use: %u\n",
7872 		    atomic_load_acq_int(&t->tids_in_use));
7873 	}
7874 
7875 	if (t->nstids) {
7876 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
7877 		    t->stid_base + t->nstids - 1, t->stids_in_use);
7878 	}
7879 
7880 	if (t->nftids) {
7881 		sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
7882 		    t->ftid_base + t->nftids - 1);
7883 	}
7884 
7885 	if (t->netids) {
7886 		sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
7887 		    t->etid_base + t->netids - 1);
7888 	}
7889 
7890 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
7891 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
7892 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
7893 
7894 	rc = sbuf_finish(sb);
7895 	sbuf_delete(sb);
7896 
7897 	return (rc);
7898 }
7899 
7900 static int
7901 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
7902 {
7903 	struct adapter *sc = arg1;
7904 	struct sbuf *sb;
7905 	int rc;
7906 	struct tp_err_stats stats;
7907 
7908 	rc = sysctl_wire_old_buffer(req, 0);
7909 	if (rc != 0)
7910 		return (rc);
7911 
7912 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7913 	if (sb == NULL)
7914 		return (ENOMEM);
7915 
7916 	mtx_lock(&sc->reg_lock);
7917 	t4_tp_get_err_stats(sc, &stats, 0);
7918 	mtx_unlock(&sc->reg_lock);
7919 
7920 	if (sc->chip_params->nchan > 2) {
7921 		sbuf_printf(sb, "                 channel 0  channel 1"
7922 		    "  channel 2  channel 3\n");
7923 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
7924 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
7925 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
7926 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
7927 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
7928 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
7929 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
7930 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
7931 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
7932 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
7933 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
7934 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
7935 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
7936 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
7937 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
7938 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
7939 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
7940 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
7941 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
7942 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
7943 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
7944 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
7945 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
7946 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
7947 	} else {
7948 		sbuf_printf(sb, "                 channel 0  channel 1\n");
7949 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
7950 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
7951 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
7952 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
7953 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
7954 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
7955 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
7956 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
7957 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
7958 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
7959 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
7960 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
7961 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
7962 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
7963 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
7964 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
7965 	}
7966 
7967 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
7968 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
7969 
7970 	rc = sbuf_finish(sb);
7971 	sbuf_delete(sb);
7972 
7973 	return (rc);
7974 }
7975 
7976 static int
7977 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
7978 {
7979 	struct adapter *sc = arg1;
7980 	struct tp_params *tpp = &sc->params.tp;
7981 	u_int mask;
7982 	int rc;
7983 
7984 	mask = tpp->la_mask >> 16;
7985 	rc = sysctl_handle_int(oidp, &mask, 0, req);
7986 	if (rc != 0 || req->newptr == NULL)
7987 		return (rc);
7988 	if (mask > 0xffff)
7989 		return (EINVAL);
7990 	tpp->la_mask = mask << 16;
7991 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
7992 
7993 	return (0);
7994 }
7995 
7996 struct field_desc {
7997 	const char *name;
7998 	u_int start;
7999 	u_int width;
8000 };
8001 
8002 static void
8003 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
8004 {
8005 	char buf[32];
8006 	int line_size = 0;
8007 
8008 	while (f->name) {
8009 		uint64_t mask = (1ULL << f->width) - 1;
8010 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
8011 		    ((uintmax_t)v >> f->start) & mask);
8012 
8013 		if (line_size + len >= 79) {
8014 			line_size = 8;
8015 			sbuf_printf(sb, "\n        ");
8016 		}
8017 		sbuf_printf(sb, "%s ", buf);
8018 		line_size += len + 1;
8019 		f++;
8020 	}
8021 	sbuf_printf(sb, "\n");
8022 }
8023 
8024 static const struct field_desc tp_la0[] = {
8025 	{ "RcfOpCodeOut", 60, 4 },
8026 	{ "State", 56, 4 },
8027 	{ "WcfState", 52, 4 },
8028 	{ "RcfOpcSrcOut", 50, 2 },
8029 	{ "CRxError", 49, 1 },
8030 	{ "ERxError", 48, 1 },
8031 	{ "SanityFailed", 47, 1 },
8032 	{ "SpuriousMsg", 46, 1 },
8033 	{ "FlushInputMsg", 45, 1 },
8034 	{ "FlushInputCpl", 44, 1 },
8035 	{ "RssUpBit", 43, 1 },
8036 	{ "RssFilterHit", 42, 1 },
8037 	{ "Tid", 32, 10 },
8038 	{ "InitTcb", 31, 1 },
8039 	{ "LineNumber", 24, 7 },
8040 	{ "Emsg", 23, 1 },
8041 	{ "EdataOut", 22, 1 },
8042 	{ "Cmsg", 21, 1 },
8043 	{ "CdataOut", 20, 1 },
8044 	{ "EreadPdu", 19, 1 },
8045 	{ "CreadPdu", 18, 1 },
8046 	{ "TunnelPkt", 17, 1 },
8047 	{ "RcfPeerFin", 16, 1 },
8048 	{ "RcfReasonOut", 12, 4 },
8049 	{ "TxCchannel", 10, 2 },
8050 	{ "RcfTxChannel", 8, 2 },
8051 	{ "RxEchannel", 6, 2 },
8052 	{ "RcfRxChannel", 5, 1 },
8053 	{ "RcfDataOutSrdy", 4, 1 },
8054 	{ "RxDvld", 3, 1 },
8055 	{ "RxOoDvld", 2, 1 },
8056 	{ "RxCongestion", 1, 1 },
8057 	{ "TxCongestion", 0, 1 },
8058 	{ NULL }
8059 };
8060 
8061 static const struct field_desc tp_la1[] = {
8062 	{ "CplCmdIn", 56, 8 },
8063 	{ "CplCmdOut", 48, 8 },
8064 	{ "ESynOut", 47, 1 },
8065 	{ "EAckOut", 46, 1 },
8066 	{ "EFinOut", 45, 1 },
8067 	{ "ERstOut", 44, 1 },
8068 	{ "SynIn", 43, 1 },
8069 	{ "AckIn", 42, 1 },
8070 	{ "FinIn", 41, 1 },
8071 	{ "RstIn", 40, 1 },
8072 	{ "DataIn", 39, 1 },
8073 	{ "DataInVld", 38, 1 },
8074 	{ "PadIn", 37, 1 },
8075 	{ "RxBufEmpty", 36, 1 },
8076 	{ "RxDdp", 35, 1 },
8077 	{ "RxFbCongestion", 34, 1 },
8078 	{ "TxFbCongestion", 33, 1 },
8079 	{ "TxPktSumSrdy", 32, 1 },
8080 	{ "RcfUlpType", 28, 4 },
8081 	{ "Eread", 27, 1 },
8082 	{ "Ebypass", 26, 1 },
8083 	{ "Esave", 25, 1 },
8084 	{ "Static0", 24, 1 },
8085 	{ "Cread", 23, 1 },
8086 	{ "Cbypass", 22, 1 },
8087 	{ "Csave", 21, 1 },
8088 	{ "CPktOut", 20, 1 },
8089 	{ "RxPagePoolFull", 18, 2 },
8090 	{ "RxLpbkPkt", 17, 1 },
8091 	{ "TxLpbkPkt", 16, 1 },
8092 	{ "RxVfValid", 15, 1 },
8093 	{ "SynLearned", 14, 1 },
8094 	{ "SetDelEntry", 13, 1 },
8095 	{ "SetInvEntry", 12, 1 },
8096 	{ "CpcmdDvld", 11, 1 },
8097 	{ "CpcmdSave", 10, 1 },
8098 	{ "RxPstructsFull", 8, 2 },
8099 	{ "EpcmdDvld", 7, 1 },
8100 	{ "EpcmdFlush", 6, 1 },
8101 	{ "EpcmdTrimPrefix", 5, 1 },
8102 	{ "EpcmdTrimPostfix", 4, 1 },
8103 	{ "ERssIp4Pkt", 3, 1 },
8104 	{ "ERssIp6Pkt", 2, 1 },
8105 	{ "ERssTcpUdpPkt", 1, 1 },
8106 	{ "ERssFceFipPkt", 0, 1 },
8107 	{ NULL }
8108 };
8109 
8110 static const struct field_desc tp_la2[] = {
8111 	{ "CplCmdIn", 56, 8 },
8112 	{ "MpsVfVld", 55, 1 },
8113 	{ "MpsPf", 52, 3 },
8114 	{ "MpsVf", 44, 8 },
8115 	{ "SynIn", 43, 1 },
8116 	{ "AckIn", 42, 1 },
8117 	{ "FinIn", 41, 1 },
8118 	{ "RstIn", 40, 1 },
8119 	{ "DataIn", 39, 1 },
8120 	{ "DataInVld", 38, 1 },
8121 	{ "PadIn", 37, 1 },
8122 	{ "RxBufEmpty", 36, 1 },
8123 	{ "RxDdp", 35, 1 },
8124 	{ "RxFbCongestion", 34, 1 },
8125 	{ "TxFbCongestion", 33, 1 },
8126 	{ "TxPktSumSrdy", 32, 1 },
8127 	{ "RcfUlpType", 28, 4 },
8128 	{ "Eread", 27, 1 },
8129 	{ "Ebypass", 26, 1 },
8130 	{ "Esave", 25, 1 },
8131 	{ "Static0", 24, 1 },
8132 	{ "Cread", 23, 1 },
8133 	{ "Cbypass", 22, 1 },
8134 	{ "Csave", 21, 1 },
8135 	{ "CPktOut", 20, 1 },
8136 	{ "RxPagePoolFull", 18, 2 },
8137 	{ "RxLpbkPkt", 17, 1 },
8138 	{ "TxLpbkPkt", 16, 1 },
8139 	{ "RxVfValid", 15, 1 },
8140 	{ "SynLearned", 14, 1 },
8141 	{ "SetDelEntry", 13, 1 },
8142 	{ "SetInvEntry", 12, 1 },
8143 	{ "CpcmdDvld", 11, 1 },
8144 	{ "CpcmdSave", 10, 1 },
8145 	{ "RxPstructsFull", 8, 2 },
8146 	{ "EpcmdDvld", 7, 1 },
8147 	{ "EpcmdFlush", 6, 1 },
8148 	{ "EpcmdTrimPrefix", 5, 1 },
8149 	{ "EpcmdTrimPostfix", 4, 1 },
8150 	{ "ERssIp4Pkt", 3, 1 },
8151 	{ "ERssIp6Pkt", 2, 1 },
8152 	{ "ERssTcpUdpPkt", 1, 1 },
8153 	{ "ERssFceFipPkt", 0, 1 },
8154 	{ NULL }
8155 };
8156 
8157 static void
8158 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
8159 {
8160 
8161 	field_desc_show(sb, *p, tp_la0);
8162 }
8163 
8164 static void
8165 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
8166 {
8167 
8168 	if (idx)
8169 		sbuf_printf(sb, "\n");
8170 	field_desc_show(sb, p[0], tp_la0);
8171 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8172 		field_desc_show(sb, p[1], tp_la0);
8173 }
8174 
8175 static void
8176 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
8177 {
8178 
8179 	if (idx)
8180 		sbuf_printf(sb, "\n");
8181 	field_desc_show(sb, p[0], tp_la0);
8182 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8183 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
8184 }
8185 
8186 static int
8187 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
8188 {
8189 	struct adapter *sc = arg1;
8190 	struct sbuf *sb;
8191 	uint64_t *buf, *p;
8192 	int rc;
8193 	u_int i, inc;
8194 	void (*show_func)(struct sbuf *, uint64_t *, int);
8195 
8196 	rc = sysctl_wire_old_buffer(req, 0);
8197 	if (rc != 0)
8198 		return (rc);
8199 
8200 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8201 	if (sb == NULL)
8202 		return (ENOMEM);
8203 
8204 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
8205 
8206 	t4_tp_read_la(sc, buf, NULL);
8207 	p = buf;
8208 
8209 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
8210 	case 2:
8211 		inc = 2;
8212 		show_func = tp_la_show2;
8213 		break;
8214 	case 3:
8215 		inc = 2;
8216 		show_func = tp_la_show3;
8217 		break;
8218 	default:
8219 		inc = 1;
8220 		show_func = tp_la_show;
8221 	}
8222 
8223 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
8224 		(*show_func)(sb, p, i);
8225 
8226 	rc = sbuf_finish(sb);
8227 	sbuf_delete(sb);
8228 	free(buf, M_CXGBE);
8229 	return (rc);
8230 }
8231 
8232 static int
8233 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
8234 {
8235 	struct adapter *sc = arg1;
8236 	struct sbuf *sb;
8237 	int rc;
8238 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
8239 
8240 	rc = sysctl_wire_old_buffer(req, 0);
8241 	if (rc != 0)
8242 		return (rc);
8243 
8244 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8245 	if (sb == NULL)
8246 		return (ENOMEM);
8247 
8248 	t4_get_chan_txrate(sc, nrate, orate);
8249 
8250 	if (sc->chip_params->nchan > 2) {
8251 		sbuf_printf(sb, "              channel 0   channel 1"
8252 		    "   channel 2   channel 3\n");
8253 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
8254 		    nrate[0], nrate[1], nrate[2], nrate[3]);
8255 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
8256 		    orate[0], orate[1], orate[2], orate[3]);
8257 	} else {
8258 		sbuf_printf(sb, "              channel 0   channel 1\n");
8259 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
8260 		    nrate[0], nrate[1]);
8261 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
8262 		    orate[0], orate[1]);
8263 	}
8264 
8265 	rc = sbuf_finish(sb);
8266 	sbuf_delete(sb);
8267 
8268 	return (rc);
8269 }
8270 
8271 static int
8272 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
8273 {
8274 	struct adapter *sc = arg1;
8275 	struct sbuf *sb;
8276 	uint32_t *buf, *p;
8277 	int rc, i;
8278 
8279 	rc = sysctl_wire_old_buffer(req, 0);
8280 	if (rc != 0)
8281 		return (rc);
8282 
8283 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8284 	if (sb == NULL)
8285 		return (ENOMEM);
8286 
8287 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
8288 	    M_ZERO | M_WAITOK);
8289 
8290 	t4_ulprx_read_la(sc, buf);
8291 	p = buf;
8292 
8293 	sbuf_printf(sb, "      Pcmd        Type   Message"
8294 	    "                Data");
8295 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
8296 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
8297 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
8298 	}
8299 
8300 	rc = sbuf_finish(sb);
8301 	sbuf_delete(sb);
8302 	free(buf, M_CXGBE);
8303 	return (rc);
8304 }
8305 
8306 static int
8307 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
8308 {
8309 	struct adapter *sc = arg1;
8310 	struct sbuf *sb;
8311 	int rc, v;
8312 
8313 	MPASS(chip_id(sc) >= CHELSIO_T5);
8314 
8315 	rc = sysctl_wire_old_buffer(req, 0);
8316 	if (rc != 0)
8317 		return (rc);
8318 
8319 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8320 	if (sb == NULL)
8321 		return (ENOMEM);
8322 
8323 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
8324 	if (G_STATSOURCE_T5(v) == 7) {
8325 		int mode;
8326 
8327 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
8328 		if (mode == 0) {
8329 			sbuf_printf(sb, "total %d, incomplete %d",
8330 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8331 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8332 		} else if (mode == 1) {
8333 			sbuf_printf(sb, "total %d, data overflow %d",
8334 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8335 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8336 		} else {
8337 			sbuf_printf(sb, "unknown mode %d", mode);
8338 		}
8339 	}
8340 	rc = sbuf_finish(sb);
8341 	sbuf_delete(sb);
8342 
8343 	return (rc);
8344 }
8345 
8346 static int
8347 sysctl_tc_params(SYSCTL_HANDLER_ARGS)
8348 {
8349 	struct adapter *sc = arg1;
8350 	struct tx_cl_rl_params tc;
8351 	struct sbuf *sb;
8352 	int i, rc, port_id, mbps, gbps;
8353 
8354 	rc = sysctl_wire_old_buffer(req, 0);
8355 	if (rc != 0)
8356 		return (rc);
8357 
8358 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8359 	if (sb == NULL)
8360 		return (ENOMEM);
8361 
8362 	port_id = arg2 >> 16;
8363 	MPASS(port_id < sc->params.nports);
8364 	MPASS(sc->port[port_id] != NULL);
8365 	i = arg2 & 0xffff;
8366 	MPASS(i < sc->chip_params->nsched_cls);
8367 
8368 	mtx_lock(&sc->tc_lock);
8369 	tc = sc->port[port_id]->sched_params->cl_rl[i];
8370 	mtx_unlock(&sc->tc_lock);
8371 
8372 	if (tc.flags & TX_CLRL_ERROR) {
8373 		sbuf_printf(sb, "error");
8374 		goto done;
8375 	}
8376 
8377 	if (tc.ratemode == SCHED_CLASS_RATEMODE_REL) {
8378 		/* XXX: top speed or actual link speed? */
8379 		gbps = port_top_speed(sc->port[port_id]);
8380 		sbuf_printf(sb, " %u%% of %uGbps", tc.maxrate, gbps);
8381 	} else if (tc.ratemode == SCHED_CLASS_RATEMODE_ABS) {
8382 		switch (tc.rateunit) {
8383 		case SCHED_CLASS_RATEUNIT_BITS:
8384 			mbps = tc.maxrate / 1000;
8385 			gbps = tc.maxrate / 1000000;
8386 			if (tc.maxrate == gbps * 1000000)
8387 				sbuf_printf(sb, " %uGbps", gbps);
8388 			else if (tc.maxrate == mbps * 1000)
8389 				sbuf_printf(sb, " %uMbps", mbps);
8390 			else
8391 				sbuf_printf(sb, " %uKbps", tc.maxrate);
8392 			break;
8393 		case SCHED_CLASS_RATEUNIT_PKTS:
8394 			sbuf_printf(sb, " %upps", tc.maxrate);
8395 			break;
8396 		default:
8397 			rc = ENXIO;
8398 			goto done;
8399 		}
8400 	}
8401 
8402 	switch (tc.mode) {
8403 	case SCHED_CLASS_MODE_CLASS:
8404 		sbuf_printf(sb, " aggregate");
8405 		break;
8406 	case SCHED_CLASS_MODE_FLOW:
8407 		sbuf_printf(sb, " per-flow");
8408 		break;
8409 	default:
8410 		rc = ENXIO;
8411 		goto done;
8412 	}
8413 
8414 done:
8415 	if (rc == 0)
8416 		rc = sbuf_finish(sb);
8417 	sbuf_delete(sb);
8418 
8419 	return (rc);
8420 }
8421 #endif
8422 
8423 #ifdef TCP_OFFLOAD
8424 static int
8425 sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS)
8426 {
8427 	struct adapter *sc = arg1;
8428 	int *old_ports, *new_ports;
8429 	int i, new_count, rc;
8430 
8431 	if (req->newptr == NULL && req->oldptr == NULL)
8432 		return (SYSCTL_OUT(req, NULL, imax(sc->tt.num_tls_rx_ports, 1) *
8433 		    sizeof(sc->tt.tls_rx_ports[0])));
8434 
8435 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tlsrx");
8436 	if (rc)
8437 		return (rc);
8438 
8439 	if (sc->tt.num_tls_rx_ports == 0) {
8440 		i = -1;
8441 		rc = SYSCTL_OUT(req, &i, sizeof(i));
8442 	} else
8443 		rc = SYSCTL_OUT(req, sc->tt.tls_rx_ports,
8444 		    sc->tt.num_tls_rx_ports * sizeof(sc->tt.tls_rx_ports[0]));
8445 	if (rc == 0 && req->newptr != NULL) {
8446 		new_count = req->newlen / sizeof(new_ports[0]);
8447 		new_ports = malloc(new_count * sizeof(new_ports[0]), M_CXGBE,
8448 		    M_WAITOK);
8449 		rc = SYSCTL_IN(req, new_ports, new_count *
8450 		    sizeof(new_ports[0]));
8451 		if (rc)
8452 			goto err;
8453 
8454 		/* Allow setting to a single '-1' to clear the list. */
8455 		if (new_count == 1 && new_ports[0] == -1) {
8456 			ADAPTER_LOCK(sc);
8457 			old_ports = sc->tt.tls_rx_ports;
8458 			sc->tt.tls_rx_ports = NULL;
8459 			sc->tt.num_tls_rx_ports = 0;
8460 			ADAPTER_UNLOCK(sc);
8461 			free(old_ports, M_CXGBE);
8462 		} else {
8463 			for (i = 0; i < new_count; i++) {
8464 				if (new_ports[i] < 1 ||
8465 				    new_ports[i] > IPPORT_MAX) {
8466 					rc = EINVAL;
8467 					goto err;
8468 				}
8469 			}
8470 
8471 			ADAPTER_LOCK(sc);
8472 			old_ports = sc->tt.tls_rx_ports;
8473 			sc->tt.tls_rx_ports = new_ports;
8474 			sc->tt.num_tls_rx_ports = new_count;
8475 			ADAPTER_UNLOCK(sc);
8476 			free(old_ports, M_CXGBE);
8477 			new_ports = NULL;
8478 		}
8479 	err:
8480 		free(new_ports, M_CXGBE);
8481 	}
8482 	end_synchronized_op(sc, 0);
8483 	return (rc);
8484 }
8485 
8486 static void
8487 unit_conv(char *buf, size_t len, u_int val, u_int factor)
8488 {
8489 	u_int rem = val % factor;
8490 
8491 	if (rem == 0)
8492 		snprintf(buf, len, "%u", val / factor);
8493 	else {
8494 		while (rem % 10 == 0)
8495 			rem /= 10;
8496 		snprintf(buf, len, "%u.%u", val / factor, rem);
8497 	}
8498 }
8499 
8500 static int
8501 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
8502 {
8503 	struct adapter *sc = arg1;
8504 	char buf[16];
8505 	u_int res, re;
8506 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8507 
8508 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8509 	switch (arg2) {
8510 	case 0:
8511 		/* timer_tick */
8512 		re = G_TIMERRESOLUTION(res);
8513 		break;
8514 	case 1:
8515 		/* TCP timestamp tick */
8516 		re = G_TIMESTAMPRESOLUTION(res);
8517 		break;
8518 	case 2:
8519 		/* DACK tick */
8520 		re = G_DELAYEDACKRESOLUTION(res);
8521 		break;
8522 	default:
8523 		return (EDOOFUS);
8524 	}
8525 
8526 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
8527 
8528 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
8529 }
8530 
8531 static int
8532 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
8533 {
8534 	struct adapter *sc = arg1;
8535 	u_int res, dack_re, v;
8536 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8537 
8538 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8539 	dack_re = G_DELAYEDACKRESOLUTION(res);
8540 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
8541 
8542 	return (sysctl_handle_int(oidp, &v, 0, req));
8543 }
8544 
8545 static int
8546 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
8547 {
8548 	struct adapter *sc = arg1;
8549 	int reg = arg2;
8550 	u_int tre;
8551 	u_long tp_tick_us, v;
8552 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8553 
8554 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
8555 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
8556 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
8557 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
8558 
8559 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
8560 	tp_tick_us = (cclk_ps << tre) / 1000000;
8561 
8562 	if (reg == A_TP_INIT_SRTT)
8563 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
8564 	else
8565 		v = tp_tick_us * t4_read_reg(sc, reg);
8566 
8567 	return (sysctl_handle_long(oidp, &v, 0, req));
8568 }
8569 
8570 /*
8571  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
8572  * passed to this function.
8573  */
8574 static int
8575 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
8576 {
8577 	struct adapter *sc = arg1;
8578 	int idx = arg2;
8579 	u_int v;
8580 
8581 	MPASS(idx >= 0 && idx <= 24);
8582 
8583 	v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
8584 
8585 	return (sysctl_handle_int(oidp, &v, 0, req));
8586 }
8587 
8588 static int
8589 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
8590 {
8591 	struct adapter *sc = arg1;
8592 	int idx = arg2;
8593 	u_int shift, v, r;
8594 
8595 	MPASS(idx >= 0 && idx < 16);
8596 
8597 	r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
8598 	shift = (idx & 3) << 3;
8599 	v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
8600 
8601 	return (sysctl_handle_int(oidp, &v, 0, req));
8602 }
8603 
8604 static int
8605 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
8606 {
8607 	struct vi_info *vi = arg1;
8608 	struct adapter *sc = vi->pi->adapter;
8609 	int idx, rc, i;
8610 	struct sge_ofld_rxq *ofld_rxq;
8611 	uint8_t v;
8612 
8613 	idx = vi->ofld_tmr_idx;
8614 
8615 	rc = sysctl_handle_int(oidp, &idx, 0, req);
8616 	if (rc != 0 || req->newptr == NULL)
8617 		return (rc);
8618 
8619 	if (idx < 0 || idx >= SGE_NTIMERS)
8620 		return (EINVAL);
8621 
8622 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8623 	    "t4otmr");
8624 	if (rc)
8625 		return (rc);
8626 
8627 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
8628 	for_each_ofld_rxq(vi, i, ofld_rxq) {
8629 #ifdef atomic_store_rel_8
8630 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
8631 #else
8632 		ofld_rxq->iq.intr_params = v;
8633 #endif
8634 	}
8635 	vi->ofld_tmr_idx = idx;
8636 
8637 	end_synchronized_op(sc, LOCK_HELD);
8638 	return (0);
8639 }
8640 
8641 static int
8642 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
8643 {
8644 	struct vi_info *vi = arg1;
8645 	struct adapter *sc = vi->pi->adapter;
8646 	int idx, rc;
8647 
8648 	idx = vi->ofld_pktc_idx;
8649 
8650 	rc = sysctl_handle_int(oidp, &idx, 0, req);
8651 	if (rc != 0 || req->newptr == NULL)
8652 		return (rc);
8653 
8654 	if (idx < -1 || idx >= SGE_NCOUNTERS)
8655 		return (EINVAL);
8656 
8657 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8658 	    "t4opktc");
8659 	if (rc)
8660 		return (rc);
8661 
8662 	if (vi->flags & VI_INIT_DONE)
8663 		rc = EBUSY; /* cannot be changed once the queues are created */
8664 	else
8665 		vi->ofld_pktc_idx = idx;
8666 
8667 	end_synchronized_op(sc, LOCK_HELD);
8668 	return (rc);
8669 }
8670 #endif
8671 
8672 static int
8673 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
8674 {
8675 	int rc;
8676 
8677 	if (cntxt->cid > M_CTXTQID)
8678 		return (EINVAL);
8679 
8680 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
8681 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
8682 		return (EINVAL);
8683 
8684 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
8685 	if (rc)
8686 		return (rc);
8687 
8688 	if (sc->flags & FW_OK) {
8689 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
8690 		    &cntxt->data[0]);
8691 		if (rc == 0)
8692 			goto done;
8693 	}
8694 
8695 	/*
8696 	 * Read via firmware failed or wasn't even attempted.  Read directly via
8697 	 * the backdoor.
8698 	 */
8699 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
8700 done:
8701 	end_synchronized_op(sc, 0);
8702 	return (rc);
8703 }
8704 
8705 static int
8706 load_fw(struct adapter *sc, struct t4_data *fw)
8707 {
8708 	int rc;
8709 	uint8_t *fw_data;
8710 
8711 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
8712 	if (rc)
8713 		return (rc);
8714 
8715 	/*
8716 	 * The firmware, with the sole exception of the memory parity error
8717 	 * handler, runs from memory and not flash.  It is almost always safe to
8718 	 * install a new firmware on a running system.  Just set bit 1 in
8719 	 * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
8720 	 */
8721 	if (sc->flags & FULL_INIT_DONE &&
8722 	    (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
8723 		rc = EBUSY;
8724 		goto done;
8725 	}
8726 
8727 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
8728 	if (fw_data == NULL) {
8729 		rc = ENOMEM;
8730 		goto done;
8731 	}
8732 
8733 	rc = copyin(fw->data, fw_data, fw->len);
8734 	if (rc == 0)
8735 		rc = -t4_load_fw(sc, fw_data, fw->len);
8736 
8737 	free(fw_data, M_CXGBE);
8738 done:
8739 	end_synchronized_op(sc, 0);
8740 	return (rc);
8741 }
8742 
8743 static int
8744 load_cfg(struct adapter *sc, struct t4_data *cfg)
8745 {
8746 	int rc;
8747 	uint8_t *cfg_data = NULL;
8748 
8749 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
8750 	if (rc)
8751 		return (rc);
8752 
8753 	if (cfg->len == 0) {
8754 		/* clear */
8755 		rc = -t4_load_cfg(sc, NULL, 0);
8756 		goto done;
8757 	}
8758 
8759 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
8760 	if (cfg_data == NULL) {
8761 		rc = ENOMEM;
8762 		goto done;
8763 	}
8764 
8765 	rc = copyin(cfg->data, cfg_data, cfg->len);
8766 	if (rc == 0)
8767 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
8768 
8769 	free(cfg_data, M_CXGBE);
8770 done:
8771 	end_synchronized_op(sc, 0);
8772 	return (rc);
8773 }
8774 
8775 static int
8776 load_boot(struct adapter *sc, struct t4_bootrom *br)
8777 {
8778 	int rc;
8779 	uint8_t *br_data = NULL;
8780 	u_int offset;
8781 
8782 	if (br->len > 1024 * 1024)
8783 		return (EFBIG);
8784 
8785 	if (br->pf_offset == 0) {
8786 		/* pfidx */
8787 		if (br->pfidx_addr > 7)
8788 			return (EINVAL);
8789 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
8790 		    A_PCIE_PF_EXPROM_OFST)));
8791 	} else if (br->pf_offset == 1) {
8792 		/* offset */
8793 		offset = G_OFFSET(br->pfidx_addr);
8794 	} else {
8795 		return (EINVAL);
8796 	}
8797 
8798 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
8799 	if (rc)
8800 		return (rc);
8801 
8802 	if (br->len == 0) {
8803 		/* clear */
8804 		rc = -t4_load_boot(sc, NULL, offset, 0);
8805 		goto done;
8806 	}
8807 
8808 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
8809 	if (br_data == NULL) {
8810 		rc = ENOMEM;
8811 		goto done;
8812 	}
8813 
8814 	rc = copyin(br->data, br_data, br->len);
8815 	if (rc == 0)
8816 		rc = -t4_load_boot(sc, br_data, offset, br->len);
8817 
8818 	free(br_data, M_CXGBE);
8819 done:
8820 	end_synchronized_op(sc, 0);
8821 	return (rc);
8822 }
8823 
8824 static int
8825 load_bootcfg(struct adapter *sc, struct t4_data *bc)
8826 {
8827 	int rc;
8828 	uint8_t *bc_data = NULL;
8829 
8830 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
8831 	if (rc)
8832 		return (rc);
8833 
8834 	if (bc->len == 0) {
8835 		/* clear */
8836 		rc = -t4_load_bootcfg(sc, NULL, 0);
8837 		goto done;
8838 	}
8839 
8840 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
8841 	if (bc_data == NULL) {
8842 		rc = ENOMEM;
8843 		goto done;
8844 	}
8845 
8846 	rc = copyin(bc->data, bc_data, bc->len);
8847 	if (rc == 0)
8848 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
8849 
8850 	free(bc_data, M_CXGBE);
8851 done:
8852 	end_synchronized_op(sc, 0);
8853 	return (rc);
8854 }
8855 
8856 static int
8857 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
8858 {
8859 	int rc;
8860 	struct cudbg_init *cudbg;
8861 	void *handle, *buf;
8862 
8863 	/* buf is large, don't block if no memory is available */
8864 	buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
8865 	if (buf == NULL)
8866 		return (ENOMEM);
8867 
8868 	handle = cudbg_alloc_handle();
8869 	if (handle == NULL) {
8870 		rc = ENOMEM;
8871 		goto done;
8872 	}
8873 
8874 	cudbg = cudbg_get_init(handle);
8875 	cudbg->adap = sc;
8876 	cudbg->print = (cudbg_print_cb)printf;
8877 
8878 #ifndef notyet
8879 	device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
8880 	    __func__, dump->wr_flash, dump->len, dump->data);
8881 #endif
8882 
8883 	if (dump->wr_flash)
8884 		cudbg->use_flash = 1;
8885 	MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
8886 	memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
8887 
8888 	rc = cudbg_collect(handle, buf, &dump->len);
8889 	if (rc != 0)
8890 		goto done;
8891 
8892 	rc = copyout(buf, dump->data, dump->len);
8893 done:
8894 	cudbg_free_handle(handle);
8895 	free(buf, M_CXGBE);
8896 	return (rc);
8897 }
8898 
8899 static void
8900 free_offload_policy(struct t4_offload_policy *op)
8901 {
8902 	struct offload_rule *r;
8903 	int i;
8904 
8905 	if (op == NULL)
8906 		return;
8907 
8908 	r = &op->rule[0];
8909 	for (i = 0; i < op->nrules; i++, r++) {
8910 		free(r->bpf_prog.bf_insns, M_CXGBE);
8911 	}
8912 	free(op->rule, M_CXGBE);
8913 	free(op, M_CXGBE);
8914 }
8915 
8916 static int
8917 set_offload_policy(struct adapter *sc, struct t4_offload_policy *uop)
8918 {
8919 	int i, rc, len;
8920 	struct t4_offload_policy *op, *old;
8921 	struct bpf_program *bf;
8922 	const struct offload_settings *s;
8923 	struct offload_rule *r;
8924 	void *u;
8925 
8926 	if (!is_offload(sc))
8927 		return (ENODEV);
8928 
8929 	if (uop->nrules == 0) {
8930 		/* Delete installed policies. */
8931 		op = NULL;
8932 		goto set_policy;
8933 	} if (uop->nrules > 256) { /* arbitrary */
8934 		return (E2BIG);
8935 	}
8936 
8937 	/* Copy userspace offload policy to kernel */
8938 	op = malloc(sizeof(*op), M_CXGBE, M_ZERO | M_WAITOK);
8939 	op->nrules = uop->nrules;
8940 	len = op->nrules * sizeof(struct offload_rule);
8941 	op->rule = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
8942 	rc = copyin(uop->rule, op->rule, len);
8943 	if (rc) {
8944 		free(op->rule, M_CXGBE);
8945 		free(op, M_CXGBE);
8946 		return (rc);
8947 	}
8948 
8949 	r = &op->rule[0];
8950 	for (i = 0; i < op->nrules; i++, r++) {
8951 
8952 		/* Validate open_type */
8953 		if (r->open_type != OPEN_TYPE_LISTEN &&
8954 		    r->open_type != OPEN_TYPE_ACTIVE &&
8955 		    r->open_type != OPEN_TYPE_PASSIVE &&
8956 		    r->open_type != OPEN_TYPE_DONTCARE) {
8957 error:
8958 			/*
8959 			 * Rules 0 to i have malloc'd filters that need to be
8960 			 * freed.  Rules i+1 to nrules have userspace pointers
8961 			 * and should be left alone.
8962 			 */
8963 			op->nrules = i;
8964 			free_offload_policy(op);
8965 			return (rc);
8966 		}
8967 
8968 		/* Validate settings */
8969 		s = &r->settings;
8970 		if ((s->offload != 0 && s->offload != 1) ||
8971 		    s->cong_algo < -1 || s->cong_algo > CONG_ALG_HIGHSPEED ||
8972 		    s->sched_class < -1 ||
8973 		    s->sched_class >= sc->chip_params->nsched_cls) {
8974 			rc = EINVAL;
8975 			goto error;
8976 		}
8977 
8978 		bf = &r->bpf_prog;
8979 		u = bf->bf_insns;	/* userspace ptr */
8980 		bf->bf_insns = NULL;
8981 		if (bf->bf_len == 0) {
8982 			/* legal, matches everything */
8983 			continue;
8984 		}
8985 		len = bf->bf_len * sizeof(*bf->bf_insns);
8986 		bf->bf_insns = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
8987 		rc = copyin(u, bf->bf_insns, len);
8988 		if (rc != 0)
8989 			goto error;
8990 
8991 		if (!bpf_validate(bf->bf_insns, bf->bf_len)) {
8992 			rc = EINVAL;
8993 			goto error;
8994 		}
8995 	}
8996 set_policy:
8997 	rw_wlock(&sc->policy_lock);
8998 	old = sc->policy;
8999 	sc->policy = op;
9000 	rw_wunlock(&sc->policy_lock);
9001 	free_offload_policy(old);
9002 
9003 	return (0);
9004 }
9005 
9006 #define MAX_READ_BUF_SIZE (128 * 1024)
9007 static int
9008 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
9009 {
9010 	uint32_t addr, remaining, n;
9011 	uint32_t *buf;
9012 	int rc;
9013 	uint8_t *dst;
9014 
9015 	rc = validate_mem_range(sc, mr->addr, mr->len);
9016 	if (rc != 0)
9017 		return (rc);
9018 
9019 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
9020 	addr = mr->addr;
9021 	remaining = mr->len;
9022 	dst = (void *)mr->data;
9023 
9024 	while (remaining) {
9025 		n = min(remaining, MAX_READ_BUF_SIZE);
9026 		read_via_memwin(sc, 2, addr, buf, n);
9027 
9028 		rc = copyout(buf, dst, n);
9029 		if (rc != 0)
9030 			break;
9031 
9032 		dst += n;
9033 		remaining -= n;
9034 		addr += n;
9035 	}
9036 
9037 	free(buf, M_CXGBE);
9038 	return (rc);
9039 }
9040 #undef MAX_READ_BUF_SIZE
9041 
9042 static int
9043 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
9044 {
9045 	int rc;
9046 
9047 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
9048 		return (EINVAL);
9049 
9050 	if (i2cd->len > sizeof(i2cd->data))
9051 		return (EFBIG);
9052 
9053 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9054 	if (rc)
9055 		return (rc);
9056 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9057 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
9058 	end_synchronized_op(sc, 0);
9059 
9060 	return (rc);
9061 }
9062 
9063 int
9064 t4_os_find_pci_capability(struct adapter *sc, int cap)
9065 {
9066 	int i;
9067 
9068 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9069 }
9070 
9071 int
9072 t4_os_pci_save_state(struct adapter *sc)
9073 {
9074 	device_t dev;
9075 	struct pci_devinfo *dinfo;
9076 
9077 	dev = sc->dev;
9078 	dinfo = device_get_ivars(dev);
9079 
9080 	pci_cfg_save(dev, dinfo, 0);
9081 	return (0);
9082 }
9083 
9084 int
9085 t4_os_pci_restore_state(struct adapter *sc)
9086 {
9087 	device_t dev;
9088 	struct pci_devinfo *dinfo;
9089 
9090 	dev = sc->dev;
9091 	dinfo = device_get_ivars(dev);
9092 
9093 	pci_cfg_restore(dev, dinfo);
9094 	return (0);
9095 }
9096 
9097 void
9098 t4_os_portmod_changed(struct port_info *pi)
9099 {
9100 	struct adapter *sc = pi->adapter;
9101 	struct vi_info *vi;
9102 	struct ifnet *ifp;
9103 	static const char *mod_str[] = {
9104 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9105 	};
9106 
9107 	PORT_LOCK(pi);
9108 	build_medialist(pi, &pi->media);
9109 	PORT_UNLOCK(pi);
9110 	vi = &pi->vi[0];
9111 	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9112 		init_l1cfg(pi);
9113 		end_synchronized_op(sc, LOCK_HELD);
9114 	}
9115 
9116 	ifp = vi->ifp;
9117 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9118 		if_printf(ifp, "transceiver unplugged.\n");
9119 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9120 		if_printf(ifp, "unknown transceiver inserted.\n");
9121 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9122 		if_printf(ifp, "unsupported transceiver inserted.\n");
9123 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9124 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9125 		    port_top_speed(pi), mod_str[pi->mod_type]);
9126 	} else {
9127 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9128 		    pi->mod_type);
9129 	}
9130 }
9131 
9132 void
9133 t4_os_link_changed(struct port_info *pi)
9134 {
9135 	struct vi_info *vi;
9136 	struct ifnet *ifp;
9137 	struct link_config *lc;
9138 	int v;
9139 
9140 	for_each_vi(pi, v, vi) {
9141 		ifp = vi->ifp;
9142 		if (ifp == NULL)
9143 			continue;
9144 
9145 		lc = &pi->link_cfg;
9146 		if (lc->link_ok) {
9147 			ifp->if_baudrate = IF_Mbps(lc->speed);
9148 			if_link_state_change(ifp, LINK_STATE_UP);
9149 		} else {
9150 			if_link_state_change(ifp, LINK_STATE_DOWN);
9151 		}
9152 	}
9153 }
9154 
9155 void
9156 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9157 {
9158 	struct adapter *sc;
9159 
9160 	sx_slock(&t4_list_lock);
9161 	SLIST_FOREACH(sc, &t4_list, link) {
9162 		/*
9163 		 * func should not make any assumptions about what state sc is
9164 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9165 		 */
9166 		func(sc, arg);
9167 	}
9168 	sx_sunlock(&t4_list_lock);
9169 }
9170 
9171 static int
9172 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9173     struct thread *td)
9174 {
9175 	int rc;
9176 	struct adapter *sc = dev->si_drv1;
9177 
9178 	rc = priv_check(td, PRIV_DRIVER);
9179 	if (rc != 0)
9180 		return (rc);
9181 
9182 	switch (cmd) {
9183 	case CHELSIO_T4_GETREG: {
9184 		struct t4_reg *edata = (struct t4_reg *)data;
9185 
9186 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9187 			return (EFAULT);
9188 
9189 		if (edata->size == 4)
9190 			edata->val = t4_read_reg(sc, edata->addr);
9191 		else if (edata->size == 8)
9192 			edata->val = t4_read_reg64(sc, edata->addr);
9193 		else
9194 			return (EINVAL);
9195 
9196 		break;
9197 	}
9198 	case CHELSIO_T4_SETREG: {
9199 		struct t4_reg *edata = (struct t4_reg *)data;
9200 
9201 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9202 			return (EFAULT);
9203 
9204 		if (edata->size == 4) {
9205 			if (edata->val & 0xffffffff00000000)
9206 				return (EINVAL);
9207 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9208 		} else if (edata->size == 8)
9209 			t4_write_reg64(sc, edata->addr, edata->val);
9210 		else
9211 			return (EINVAL);
9212 		break;
9213 	}
9214 	case CHELSIO_T4_REGDUMP: {
9215 		struct t4_regdump *regs = (struct t4_regdump *)data;
9216 		int reglen = t4_get_regs_len(sc);
9217 		uint8_t *buf;
9218 
9219 		if (regs->len < reglen) {
9220 			regs->len = reglen; /* hint to the caller */
9221 			return (ENOBUFS);
9222 		}
9223 
9224 		regs->len = reglen;
9225 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9226 		get_regs(sc, regs, buf);
9227 		rc = copyout(buf, regs->data, reglen);
9228 		free(buf, M_CXGBE);
9229 		break;
9230 	}
9231 	case CHELSIO_T4_GET_FILTER_MODE:
9232 		rc = get_filter_mode(sc, (uint32_t *)data);
9233 		break;
9234 	case CHELSIO_T4_SET_FILTER_MODE:
9235 		rc = set_filter_mode(sc, *(uint32_t *)data);
9236 		break;
9237 	case CHELSIO_T4_GET_FILTER:
9238 		rc = get_filter(sc, (struct t4_filter *)data);
9239 		break;
9240 	case CHELSIO_T4_SET_FILTER:
9241 		rc = set_filter(sc, (struct t4_filter *)data);
9242 		break;
9243 	case CHELSIO_T4_DEL_FILTER:
9244 		rc = del_filter(sc, (struct t4_filter *)data);
9245 		break;
9246 	case CHELSIO_T4_GET_SGE_CONTEXT:
9247 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9248 		break;
9249 	case CHELSIO_T4_LOAD_FW:
9250 		rc = load_fw(sc, (struct t4_data *)data);
9251 		break;
9252 	case CHELSIO_T4_GET_MEM:
9253 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9254 		break;
9255 	case CHELSIO_T4_GET_I2C:
9256 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9257 		break;
9258 	case CHELSIO_T4_CLEAR_STATS: {
9259 		int i, v, bg_map;
9260 		u_int port_id = *(uint32_t *)data;
9261 		struct port_info *pi;
9262 		struct vi_info *vi;
9263 
9264 		if (port_id >= sc->params.nports)
9265 			return (EINVAL);
9266 		pi = sc->port[port_id];
9267 		if (pi == NULL)
9268 			return (EIO);
9269 
9270 		/* MAC stats */
9271 		t4_clr_port_stats(sc, pi->tx_chan);
9272 		pi->tx_parse_error = 0;
9273 		pi->tnl_cong_drops = 0;
9274 		mtx_lock(&sc->reg_lock);
9275 		for_each_vi(pi, v, vi) {
9276 			if (vi->flags & VI_INIT_DONE)
9277 				t4_clr_vi_stats(sc, vi->viid);
9278 		}
9279 		bg_map = pi->mps_bg_map;
9280 		v = 0;	/* reuse */
9281 		while (bg_map) {
9282 			i = ffs(bg_map) - 1;
9283 			t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
9284 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
9285 			bg_map &= ~(1 << i);
9286 		}
9287 		mtx_unlock(&sc->reg_lock);
9288 
9289 		/*
9290 		 * Since this command accepts a port, clear stats for
9291 		 * all VIs on this port.
9292 		 */
9293 		for_each_vi(pi, v, vi) {
9294 			if (vi->flags & VI_INIT_DONE) {
9295 				struct sge_rxq *rxq;
9296 				struct sge_txq *txq;
9297 				struct sge_wrq *wrq;
9298 
9299 				for_each_rxq(vi, i, rxq) {
9300 #if defined(INET) || defined(INET6)
9301 					rxq->lro.lro_queued = 0;
9302 					rxq->lro.lro_flushed = 0;
9303 #endif
9304 					rxq->rxcsum = 0;
9305 					rxq->vlan_extraction = 0;
9306 				}
9307 
9308 				for_each_txq(vi, i, txq) {
9309 					txq->txcsum = 0;
9310 					txq->tso_wrs = 0;
9311 					txq->vlan_insertion = 0;
9312 					txq->imm_wrs = 0;
9313 					txq->sgl_wrs = 0;
9314 					txq->txpkt_wrs = 0;
9315 					txq->txpkts0_wrs = 0;
9316 					txq->txpkts1_wrs = 0;
9317 					txq->txpkts0_pkts = 0;
9318 					txq->txpkts1_pkts = 0;
9319 					mp_ring_reset_stats(txq->r);
9320 				}
9321 
9322 #ifdef TCP_OFFLOAD
9323 				/* nothing to clear for each ofld_rxq */
9324 
9325 				for_each_ofld_txq(vi, i, wrq) {
9326 					wrq->tx_wrs_direct = 0;
9327 					wrq->tx_wrs_copied = 0;
9328 				}
9329 #endif
9330 
9331 				if (IS_MAIN_VI(vi)) {
9332 					wrq = &sc->sge.ctrlq[pi->port_id];
9333 					wrq->tx_wrs_direct = 0;
9334 					wrq->tx_wrs_copied = 0;
9335 				}
9336 			}
9337 		}
9338 		break;
9339 	}
9340 	case CHELSIO_T4_SCHED_CLASS:
9341 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9342 		break;
9343 	case CHELSIO_T4_SCHED_QUEUE:
9344 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9345 		break;
9346 	case CHELSIO_T4_GET_TRACER:
9347 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9348 		break;
9349 	case CHELSIO_T4_SET_TRACER:
9350 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9351 		break;
9352 	case CHELSIO_T4_LOAD_CFG:
9353 		rc = load_cfg(sc, (struct t4_data *)data);
9354 		break;
9355 	case CHELSIO_T4_LOAD_BOOT:
9356 		rc = load_boot(sc, (struct t4_bootrom *)data);
9357 		break;
9358 	case CHELSIO_T4_LOAD_BOOTCFG:
9359 		rc = load_bootcfg(sc, (struct t4_data *)data);
9360 		break;
9361 	case CHELSIO_T4_CUDBG_DUMP:
9362 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
9363 		break;
9364 	case CHELSIO_T4_SET_OFLD_POLICY:
9365 		rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
9366 		break;
9367 	default:
9368 		rc = ENOTTY;
9369 	}
9370 
9371 	return (rc);
9372 }
9373 
9374 void
9375 t4_db_full(struct adapter *sc)
9376 {
9377 
9378 	CXGBE_UNIMPLEMENTED(__func__);
9379 }
9380 
9381 void
9382 t4_db_dropped(struct adapter *sc)
9383 {
9384 
9385 	CXGBE_UNIMPLEMENTED(__func__);
9386 }
9387 
9388 #ifdef TCP_OFFLOAD
9389 static int
9390 toe_capability(struct vi_info *vi, int enable)
9391 {
9392 	int rc;
9393 	struct port_info *pi = vi->pi;
9394 	struct adapter *sc = pi->adapter;
9395 
9396 	ASSERT_SYNCHRONIZED_OP(sc);
9397 
9398 	if (!is_offload(sc))
9399 		return (ENODEV);
9400 
9401 	if (enable) {
9402 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9403 			/* TOE is already enabled. */
9404 			return (0);
9405 		}
9406 
9407 		/*
9408 		 * We need the port's queues around so that we're able to send
9409 		 * and receive CPLs to/from the TOE even if the ifnet for this
9410 		 * port has never been UP'd administratively.
9411 		 */
9412 		if (!(vi->flags & VI_INIT_DONE)) {
9413 			rc = vi_full_init(vi);
9414 			if (rc)
9415 				return (rc);
9416 		}
9417 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9418 			rc = vi_full_init(&pi->vi[0]);
9419 			if (rc)
9420 				return (rc);
9421 		}
9422 
9423 		if (isset(&sc->offload_map, pi->port_id)) {
9424 			/* TOE is enabled on another VI of this port. */
9425 			pi->uld_vis++;
9426 			return (0);
9427 		}
9428 
9429 		if (!uld_active(sc, ULD_TOM)) {
9430 			rc = t4_activate_uld(sc, ULD_TOM);
9431 			if (rc == EAGAIN) {
9432 				log(LOG_WARNING,
9433 				    "You must kldload t4_tom.ko before trying "
9434 				    "to enable TOE on a cxgbe interface.\n");
9435 			}
9436 			if (rc != 0)
9437 				return (rc);
9438 			KASSERT(sc->tom_softc != NULL,
9439 			    ("%s: TOM activated but softc NULL", __func__));
9440 			KASSERT(uld_active(sc, ULD_TOM),
9441 			    ("%s: TOM activated but flag not set", __func__));
9442 		}
9443 
9444 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
9445 		if (!uld_active(sc, ULD_IWARP))
9446 			(void) t4_activate_uld(sc, ULD_IWARP);
9447 		if (!uld_active(sc, ULD_ISCSI))
9448 			(void) t4_activate_uld(sc, ULD_ISCSI);
9449 
9450 		pi->uld_vis++;
9451 		setbit(&sc->offload_map, pi->port_id);
9452 	} else {
9453 		pi->uld_vis--;
9454 
9455 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9456 			return (0);
9457 
9458 		KASSERT(uld_active(sc, ULD_TOM),
9459 		    ("%s: TOM never initialized?", __func__));
9460 		clrbit(&sc->offload_map, pi->port_id);
9461 	}
9462 
9463 	return (0);
9464 }
9465 
9466 /*
9467  * Add an upper layer driver to the global list.
9468  */
9469 int
9470 t4_register_uld(struct uld_info *ui)
9471 {
9472 	int rc = 0;
9473 	struct uld_info *u;
9474 
9475 	sx_xlock(&t4_uld_list_lock);
9476 	SLIST_FOREACH(u, &t4_uld_list, link) {
9477 	    if (u->uld_id == ui->uld_id) {
9478 		    rc = EEXIST;
9479 		    goto done;
9480 	    }
9481 	}
9482 
9483 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9484 	ui->refcount = 0;
9485 done:
9486 	sx_xunlock(&t4_uld_list_lock);
9487 	return (rc);
9488 }
9489 
9490 int
9491 t4_unregister_uld(struct uld_info *ui)
9492 {
9493 	int rc = EINVAL;
9494 	struct uld_info *u;
9495 
9496 	sx_xlock(&t4_uld_list_lock);
9497 
9498 	SLIST_FOREACH(u, &t4_uld_list, link) {
9499 	    if (u == ui) {
9500 		    if (ui->refcount > 0) {
9501 			    rc = EBUSY;
9502 			    goto done;
9503 		    }
9504 
9505 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9506 		    rc = 0;
9507 		    goto done;
9508 	    }
9509 	}
9510 done:
9511 	sx_xunlock(&t4_uld_list_lock);
9512 	return (rc);
9513 }
9514 
9515 int
9516 t4_activate_uld(struct adapter *sc, int id)
9517 {
9518 	int rc;
9519 	struct uld_info *ui;
9520 
9521 	ASSERT_SYNCHRONIZED_OP(sc);
9522 
9523 	if (id < 0 || id > ULD_MAX)
9524 		return (EINVAL);
9525 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
9526 
9527 	sx_slock(&t4_uld_list_lock);
9528 
9529 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9530 		if (ui->uld_id == id) {
9531 			if (!(sc->flags & FULL_INIT_DONE)) {
9532 				rc = adapter_full_init(sc);
9533 				if (rc != 0)
9534 					break;
9535 			}
9536 
9537 			rc = ui->activate(sc);
9538 			if (rc == 0) {
9539 				setbit(&sc->active_ulds, id);
9540 				ui->refcount++;
9541 			}
9542 			break;
9543 		}
9544 	}
9545 
9546 	sx_sunlock(&t4_uld_list_lock);
9547 
9548 	return (rc);
9549 }
9550 
9551 int
9552 t4_deactivate_uld(struct adapter *sc, int id)
9553 {
9554 	int rc;
9555 	struct uld_info *ui;
9556 
9557 	ASSERT_SYNCHRONIZED_OP(sc);
9558 
9559 	if (id < 0 || id > ULD_MAX)
9560 		return (EINVAL);
9561 	rc = ENXIO;
9562 
9563 	sx_slock(&t4_uld_list_lock);
9564 
9565 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9566 		if (ui->uld_id == id) {
9567 			rc = ui->deactivate(sc);
9568 			if (rc == 0) {
9569 				clrbit(&sc->active_ulds, id);
9570 				ui->refcount--;
9571 			}
9572 			break;
9573 		}
9574 	}
9575 
9576 	sx_sunlock(&t4_uld_list_lock);
9577 
9578 	return (rc);
9579 }
9580 
9581 int
9582 uld_active(struct adapter *sc, int uld_id)
9583 {
9584 
9585 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9586 
9587 	return (isset(&sc->active_ulds, uld_id));
9588 }
9589 #endif
9590 
9591 /*
9592  * t  = ptr to tunable.
9593  * nc = number of CPUs.
9594  * c  = compiled in default for that tunable.
9595  */
9596 static void
9597 calculate_nqueues(int *t, int nc, const int c)
9598 {
9599 	int nq;
9600 
9601 	if (*t > 0)
9602 		return;
9603 	nq = *t < 0 ? -*t : c;
9604 	*t = min(nc, nq);
9605 }
9606 
9607 /*
9608  * Come up with reasonable defaults for some of the tunables, provided they're
9609  * not set by the user (in which case we'll use the values as is).
9610  */
9611 static void
9612 tweak_tunables(void)
9613 {
9614 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
9615 
9616 	if (t4_ntxq < 1) {
9617 #ifdef RSS
9618 		t4_ntxq = rss_getnumbuckets();
9619 #else
9620 		calculate_nqueues(&t4_ntxq, nc, NTXQ);
9621 #endif
9622 	}
9623 
9624 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
9625 
9626 	if (t4_nrxq < 1) {
9627 #ifdef RSS
9628 		t4_nrxq = rss_getnumbuckets();
9629 #else
9630 		calculate_nqueues(&t4_nrxq, nc, NRXQ);
9631 #endif
9632 	}
9633 
9634 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
9635 
9636 #ifdef TCP_OFFLOAD
9637 	calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
9638 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
9639 	calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
9640 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
9641 
9642 	if (t4_toecaps_allowed == -1)
9643 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9644 
9645 	if (t4_rdmacaps_allowed == -1) {
9646 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9647 		    FW_CAPS_CONFIG_RDMA_RDMAC;
9648 	}
9649 
9650 	if (t4_iscsicaps_allowed == -1) {
9651 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9652 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9653 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
9654 	}
9655 
9656 	if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
9657 		t4_tmr_idx_ofld = TMR_IDX_OFLD;
9658 
9659 	if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
9660 		t4_pktc_idx_ofld = PKTC_IDX_OFLD;
9661 #else
9662 	if (t4_toecaps_allowed == -1)
9663 		t4_toecaps_allowed = 0;
9664 
9665 	if (t4_rdmacaps_allowed == -1)
9666 		t4_rdmacaps_allowed = 0;
9667 
9668 	if (t4_iscsicaps_allowed == -1)
9669 		t4_iscsicaps_allowed = 0;
9670 #endif
9671 
9672 #ifdef DEV_NETMAP
9673 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
9674 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
9675 #endif
9676 
9677 	if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
9678 		t4_tmr_idx = TMR_IDX;
9679 
9680 	if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
9681 		t4_pktc_idx = PKTC_IDX;
9682 
9683 	if (t4_qsize_txq < 128)
9684 		t4_qsize_txq = 128;
9685 
9686 	if (t4_qsize_rxq < 128)
9687 		t4_qsize_rxq = 128;
9688 	while (t4_qsize_rxq & 7)
9689 		t4_qsize_rxq++;
9690 
9691 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9692 
9693 	/*
9694 	 * Number of VIs to create per-port.  The first VI is the "main" regular
9695 	 * VI for the port.  The rest are additional virtual interfaces on the
9696 	 * same physical port.  Note that the main VI does not have native
9697 	 * netmap support but the extra VIs do.
9698 	 *
9699 	 * Limit the number of VIs per port to the number of available
9700 	 * MAC addresses per port.
9701 	 */
9702 	if (t4_num_vis < 1)
9703 		t4_num_vis = 1;
9704 	if (t4_num_vis > nitems(vi_mac_funcs)) {
9705 		t4_num_vis = nitems(vi_mac_funcs);
9706 		printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
9707 	}
9708 
9709 	if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
9710 		pcie_relaxed_ordering = 1;
9711 #if defined(__i386__) || defined(__amd64__)
9712 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
9713 			pcie_relaxed_ordering = 0;
9714 #endif
9715 	}
9716 }
9717 
9718 #ifdef DDB
9719 static void
9720 t4_dump_tcb(struct adapter *sc, int tid)
9721 {
9722 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
9723 
9724 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
9725 	save = t4_read_reg(sc, reg);
9726 	base = sc->memwin[2].mw_base;
9727 
9728 	/* Dump TCB for the tid */
9729 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
9730 	tcb_addr += tid * TCB_SIZE;
9731 
9732 	if (is_t4(sc)) {
9733 		pf = 0;
9734 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
9735 	} else {
9736 		pf = V_PFNUM(sc->pf);
9737 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
9738 	}
9739 	t4_write_reg(sc, reg, win_pos | pf);
9740 	t4_read_reg(sc, reg);
9741 
9742 	off = tcb_addr - win_pos;
9743 	for (i = 0; i < 4; i++) {
9744 		uint32_t buf[8];
9745 		for (j = 0; j < 8; j++, off += 4)
9746 			buf[j] = htonl(t4_read_reg(sc, base + off));
9747 
9748 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
9749 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
9750 		    buf[7]);
9751 	}
9752 
9753 	t4_write_reg(sc, reg, save);
9754 	t4_read_reg(sc, reg);
9755 }
9756 
9757 static void
9758 t4_dump_devlog(struct adapter *sc)
9759 {
9760 	struct devlog_params *dparams = &sc->params.devlog;
9761 	struct fw_devlog_e e;
9762 	int i, first, j, m, nentries, rc;
9763 	uint64_t ftstamp = UINT64_MAX;
9764 
9765 	if (dparams->start == 0) {
9766 		db_printf("devlog params not valid\n");
9767 		return;
9768 	}
9769 
9770 	nentries = dparams->size / sizeof(struct fw_devlog_e);
9771 	m = fwmtype_to_hwmtype(dparams->memtype);
9772 
9773 	/* Find the first entry. */
9774 	first = -1;
9775 	for (i = 0; i < nentries && !db_pager_quit; i++) {
9776 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9777 		    sizeof(e), (void *)&e);
9778 		if (rc != 0)
9779 			break;
9780 
9781 		if (e.timestamp == 0)
9782 			break;
9783 
9784 		e.timestamp = be64toh(e.timestamp);
9785 		if (e.timestamp < ftstamp) {
9786 			ftstamp = e.timestamp;
9787 			first = i;
9788 		}
9789 	}
9790 
9791 	if (first == -1)
9792 		return;
9793 
9794 	i = first;
9795 	do {
9796 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9797 		    sizeof(e), (void *)&e);
9798 		if (rc != 0)
9799 			return;
9800 
9801 		if (e.timestamp == 0)
9802 			return;
9803 
9804 		e.timestamp = be64toh(e.timestamp);
9805 		e.seqno = be32toh(e.seqno);
9806 		for (j = 0; j < 8; j++)
9807 			e.params[j] = be32toh(e.params[j]);
9808 
9809 		db_printf("%10d  %15ju  %8s  %8s  ",
9810 		    e.seqno, e.timestamp,
9811 		    (e.level < nitems(devlog_level_strings) ?
9812 			devlog_level_strings[e.level] : "UNKNOWN"),
9813 		    (e.facility < nitems(devlog_facility_strings) ?
9814 			devlog_facility_strings[e.facility] : "UNKNOWN"));
9815 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
9816 		    e.params[3], e.params[4], e.params[5], e.params[6],
9817 		    e.params[7]);
9818 
9819 		if (++i == nentries)
9820 			i = 0;
9821 	} while (i != first && !db_pager_quit);
9822 }
9823 
9824 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
9825 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
9826 
9827 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
9828 {
9829 	device_t dev;
9830 	int t;
9831 	bool valid;
9832 
9833 	valid = false;
9834 	t = db_read_token();
9835 	if (t == tIDENT) {
9836 		dev = device_lookup_by_name(db_tok_string);
9837 		valid = true;
9838 	}
9839 	db_skip_to_eol();
9840 	if (!valid) {
9841 		db_printf("usage: show t4 devlog <nexus>\n");
9842 		return;
9843 	}
9844 
9845 	if (dev == NULL) {
9846 		db_printf("device not found\n");
9847 		return;
9848 	}
9849 
9850 	t4_dump_devlog(device_get_softc(dev));
9851 }
9852 
9853 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
9854 {
9855 	device_t dev;
9856 	int radix, tid, t;
9857 	bool valid;
9858 
9859 	valid = false;
9860 	radix = db_radix;
9861 	db_radix = 10;
9862 	t = db_read_token();
9863 	if (t == tIDENT) {
9864 		dev = device_lookup_by_name(db_tok_string);
9865 		t = db_read_token();
9866 		if (t == tNUMBER) {
9867 			tid = db_tok_number;
9868 			valid = true;
9869 		}
9870 	}
9871 	db_radix = radix;
9872 	db_skip_to_eol();
9873 	if (!valid) {
9874 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
9875 		return;
9876 	}
9877 
9878 	if (dev == NULL) {
9879 		db_printf("device not found\n");
9880 		return;
9881 	}
9882 	if (tid < 0) {
9883 		db_printf("invalid tid\n");
9884 		return;
9885 	}
9886 
9887 	t4_dump_tcb(device_get_softc(dev), tid);
9888 }
9889 #endif
9890 
9891 /*
9892  * Borrowed from cesa_prep_aes_key().
9893  *
9894  * NB: The crypto engine wants the words in the decryption key in reverse
9895  * order.
9896  */
9897 void
9898 t4_aes_getdeckey(void *dec_key, const void *enc_key, unsigned int kbits)
9899 {
9900 	uint32_t ek[4 * (RIJNDAEL_MAXNR + 1)];
9901 	uint32_t *dkey;
9902 	int i;
9903 
9904 	rijndaelKeySetupEnc(ek, enc_key, kbits);
9905 	dkey = dec_key;
9906 	dkey += (kbits / 8) / 4;
9907 
9908 	switch (kbits) {
9909 	case 128:
9910 		for (i = 0; i < 4; i++)
9911 			*--dkey = htobe32(ek[4 * 10 + i]);
9912 		break;
9913 	case 192:
9914 		for (i = 0; i < 2; i++)
9915 			*--dkey = htobe32(ek[4 * 11 + 2 + i]);
9916 		for (i = 0; i < 4; i++)
9917 			*--dkey = htobe32(ek[4 * 12 + i]);
9918 		break;
9919 	case 256:
9920 		for (i = 0; i < 4; i++)
9921 			*--dkey = htobe32(ek[4 * 13 + i]);
9922 		for (i = 0; i < 4; i++)
9923 			*--dkey = htobe32(ek[4 * 14 + i]);
9924 		break;
9925 	}
9926 	MPASS(dkey == dec_key);
9927 }
9928 
9929 static struct sx mlu;	/* mod load unload */
9930 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
9931 
9932 static int
9933 mod_event(module_t mod, int cmd, void *arg)
9934 {
9935 	int rc = 0;
9936 	static int loaded = 0;
9937 
9938 	switch (cmd) {
9939 	case MOD_LOAD:
9940 		sx_xlock(&mlu);
9941 		if (loaded++ == 0) {
9942 			t4_sge_modload();
9943 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
9944 			    t4_filter_rpl, CPL_COOKIE_FILTER);
9945 			t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
9946 			    do_l2t_write_rpl, CPL_COOKIE_FILTER);
9947 			t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL,
9948 			    t4_hashfilter_ao_rpl, CPL_COOKIE_HASHFILTER);
9949 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
9950 			    t4_hashfilter_tcb_rpl, CPL_COOKIE_HASHFILTER);
9951 			t4_register_shared_cpl_handler(CPL_ABORT_RPL_RSS,
9952 			    t4_del_hashfilter_rpl, CPL_COOKIE_HASHFILTER);
9953 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
9954 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
9955 			sx_init(&t4_list_lock, "T4/T5 adapters");
9956 			SLIST_INIT(&t4_list);
9957 #ifdef TCP_OFFLOAD
9958 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
9959 			SLIST_INIT(&t4_uld_list);
9960 #endif
9961 			t4_tracer_modload();
9962 			tweak_tunables();
9963 		}
9964 		sx_xunlock(&mlu);
9965 		break;
9966 
9967 	case MOD_UNLOAD:
9968 		sx_xlock(&mlu);
9969 		if (--loaded == 0) {
9970 			int tries;
9971 
9972 			sx_slock(&t4_list_lock);
9973 			if (!SLIST_EMPTY(&t4_list)) {
9974 				rc = EBUSY;
9975 				sx_sunlock(&t4_list_lock);
9976 				goto done_unload;
9977 			}
9978 #ifdef TCP_OFFLOAD
9979 			sx_slock(&t4_uld_list_lock);
9980 			if (!SLIST_EMPTY(&t4_uld_list)) {
9981 				rc = EBUSY;
9982 				sx_sunlock(&t4_uld_list_lock);
9983 				sx_sunlock(&t4_list_lock);
9984 				goto done_unload;
9985 			}
9986 #endif
9987 			tries = 0;
9988 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
9989 				uprintf("%ju clusters with custom free routine "
9990 				    "still is use.\n", t4_sge_extfree_refs());
9991 				pause("t4unload", 2 * hz);
9992 			}
9993 #ifdef TCP_OFFLOAD
9994 			sx_sunlock(&t4_uld_list_lock);
9995 #endif
9996 			sx_sunlock(&t4_list_lock);
9997 
9998 			if (t4_sge_extfree_refs() == 0) {
9999 				t4_tracer_modunload();
10000 #ifdef TCP_OFFLOAD
10001 				sx_destroy(&t4_uld_list_lock);
10002 #endif
10003 				sx_destroy(&t4_list_lock);
10004 				t4_sge_modunload();
10005 				loaded = 0;
10006 			} else {
10007 				rc = EBUSY;
10008 				loaded++;	/* undo earlier decrement */
10009 			}
10010 		}
10011 done_unload:
10012 		sx_xunlock(&mlu);
10013 		break;
10014 	}
10015 
10016 	return (rc);
10017 }
10018 
10019 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10020 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10021 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10022 
10023 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10024 MODULE_VERSION(t4nex, 1);
10025 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10026 #ifdef DEV_NETMAP
10027 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10028 #endif /* DEV_NETMAP */
10029 
10030 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10031 MODULE_VERSION(t5nex, 1);
10032 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10033 #ifdef DEV_NETMAP
10034 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10035 #endif /* DEV_NETMAP */
10036 
10037 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10038 MODULE_VERSION(t6nex, 1);
10039 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10040 #ifdef DEV_NETMAP
10041 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10042 #endif /* DEV_NETMAP */
10043 
10044 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10045 MODULE_VERSION(cxgbe, 1);
10046 
10047 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10048 MODULE_VERSION(cxl, 1);
10049 
10050 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10051 MODULE_VERSION(cc, 1);
10052 
10053 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10054 MODULE_VERSION(vcxgbe, 1);
10055 
10056 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10057 MODULE_VERSION(vcxl, 1);
10058 
10059 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10060 MODULE_VERSION(vcc, 1);
10061