xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision 74fe6c29fb7eef3418d7919dcd41dc1a04a982a1)
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 TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
441 
442 static int t4_toecaps_allowed = -1;
443 TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
444 
445 static int t4_rdmacaps_allowed = -1;
446 TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
447 
448 static int t4_cryptocaps_allowed = -1;
449 TUNABLE_INT("hw.cxgbe.cryptocaps_allowed", &t4_cryptocaps_allowed);
450 
451 static int t4_iscsicaps_allowed = -1;
452 TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
453 
454 static int t4_fcoecaps_allowed = 0;
455 TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
456 
457 static int t5_write_combine = 1;
458 TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
459 
460 static int t4_num_vis = 1;
461 TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
462 /*
463  * PCIe Relaxed Ordering.
464  * -1: driver should figure out a good value.
465  * 0: disable RO.
466  * 1: enable RO.
467  * 2: leave RO alone.
468  */
469 static int pcie_relaxed_ordering = -1;
470 TUNABLE_INT("hw.cxgbe.pcie_relaxed_ordering", &pcie_relaxed_ordering);
471 
472 
473 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
474 static int vi_mac_funcs[] = {
475 	FW_VI_FUNC_ETH,
476 	FW_VI_FUNC_OFLD,
477 	FW_VI_FUNC_IWARP,
478 	FW_VI_FUNC_OPENISCSI,
479 	FW_VI_FUNC_OPENFCOE,
480 	FW_VI_FUNC_FOISCSI,
481 	FW_VI_FUNC_FOFCOE,
482 };
483 
484 struct intrs_and_queues {
485 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
486 	uint16_t num_vis;	/* number of VIs for each port */
487 	uint16_t nirq;		/* Total # of vectors */
488 	uint16_t ntxq;		/* # of NIC txq's for each port */
489 	uint16_t nrxq;		/* # of NIC rxq's for each port */
490 	uint16_t nofldtxq;	/* # of TOE txq's for each port */
491 	uint16_t nofldrxq;	/* # of TOE rxq's for each port */
492 
493 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
494 	uint16_t ntxq_vi;	/* # of NIC txq's */
495 	uint16_t nrxq_vi;	/* # of NIC rxq's */
496 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
497 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
498 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
499 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
500 };
501 
502 struct filter_entry {
503         uint32_t valid:1;	/* filter allocated and valid */
504         uint32_t locked:1;	/* filter is administratively locked */
505         uint32_t pending:1;	/* filter action is pending firmware reply */
506 	uint32_t smtidx:8;	/* Source MAC Table index for smac */
507 	struct l2t_entry *l2t;	/* Layer Two Table entry for dmac */
508 
509         struct t4_filter_specification fs;
510 };
511 
512 static void setup_memwin(struct adapter *);
513 static void position_memwin(struct adapter *, int, uint32_t);
514 static int rw_via_memwin(struct adapter *, int, uint32_t, uint32_t *, int, int);
515 static inline int read_via_memwin(struct adapter *, int, uint32_t, uint32_t *,
516     int);
517 static inline int write_via_memwin(struct adapter *, int, uint32_t,
518     const uint32_t *, int);
519 static int validate_mem_range(struct adapter *, uint32_t, int);
520 static int fwmtype_to_hwmtype(int);
521 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
522     uint32_t *);
523 static int fixup_devlog_params(struct adapter *);
524 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
525 static int prep_firmware(struct adapter *);
526 static int partition_resources(struct adapter *, const struct firmware *,
527     const char *);
528 static int get_params__pre_init(struct adapter *);
529 static int get_params__post_init(struct adapter *);
530 static int set_params__post_init(struct adapter *);
531 static void t4_set_desc(struct adapter *);
532 static void build_medialist(struct port_info *, struct ifmedia *);
533 static void init_l1cfg(struct port_info *);
534 static int cxgbe_init_synchronized(struct vi_info *);
535 static int cxgbe_uninit_synchronized(struct vi_info *);
536 static void quiesce_txq(struct adapter *, struct sge_txq *);
537 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
538 static void quiesce_iq(struct adapter *, struct sge_iq *);
539 static void quiesce_fl(struct adapter *, struct sge_fl *);
540 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
541     driver_intr_t *, void *, char *);
542 static int t4_free_irq(struct adapter *, struct irq *);
543 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
544 static void vi_refresh_stats(struct adapter *, struct vi_info *);
545 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
546 static void cxgbe_tick(void *);
547 static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
548 static void cxgbe_sysctls(struct port_info *);
549 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
550 static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
551 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
552 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
553 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
554 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
555 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
556 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
557 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
558 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
559 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
560 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
561 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
562 #ifdef SBUF_DRAIN
563 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
564 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
565 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
566 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
567 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
568 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
569 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
570 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
571 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
572 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
573 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
574 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
575 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
576 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
577 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
578 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
579 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
580 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
581 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
582 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
583 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
584 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
585 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
586 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
587 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
588 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
589 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
590 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
591 static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
592 #endif
593 #ifdef TCP_OFFLOAD
594 static int sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS);
595 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
596 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
597 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
598 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
599 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
600 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
601 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
602 #endif
603 static uint32_t fconf_iconf_to_mode(uint32_t, uint32_t);
604 static uint32_t mode_to_fconf(uint32_t);
605 static uint32_t mode_to_iconf(uint32_t);
606 static int check_fspec_against_fconf_iconf(struct adapter *,
607     struct t4_filter_specification *);
608 static int get_filter_mode(struct adapter *, uint32_t *);
609 static int set_filter_mode(struct adapter *, uint32_t);
610 static inline uint64_t get_filter_hits(struct adapter *, uint32_t);
611 static int get_filter(struct adapter *, struct t4_filter *);
612 static int set_filter(struct adapter *, struct t4_filter *);
613 static int del_filter(struct adapter *, struct t4_filter *);
614 static void clear_filter(struct filter_entry *);
615 static int set_filter_wr(struct adapter *, int);
616 static int del_filter_wr(struct adapter *, int);
617 static int set_tcb_rpl(struct sge_iq *, const struct rss_header *,
618     struct mbuf *);
619 static int get_sge_context(struct adapter *, struct t4_sge_context *);
620 static int load_fw(struct adapter *, struct t4_data *);
621 static int load_cfg(struct adapter *, struct t4_data *);
622 static int load_boot(struct adapter *, struct t4_bootrom *);
623 static int load_bootcfg(struct adapter *, struct t4_data *);
624 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
625 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
626 static int read_i2c(struct adapter *, struct t4_i2c_data *);
627 #ifdef TCP_OFFLOAD
628 static int toe_capability(struct vi_info *, int);
629 #endif
630 static int mod_event(module_t, int, void *);
631 static int notify_siblings(device_t, int);
632 
633 struct {
634 	uint16_t device;
635 	char *desc;
636 } t4_pciids[] = {
637 	{0xa000, "Chelsio Terminator 4 FPGA"},
638 	{0x4400, "Chelsio T440-dbg"},
639 	{0x4401, "Chelsio T420-CR"},
640 	{0x4402, "Chelsio T422-CR"},
641 	{0x4403, "Chelsio T440-CR"},
642 	{0x4404, "Chelsio T420-BCH"},
643 	{0x4405, "Chelsio T440-BCH"},
644 	{0x4406, "Chelsio T440-CH"},
645 	{0x4407, "Chelsio T420-SO"},
646 	{0x4408, "Chelsio T420-CX"},
647 	{0x4409, "Chelsio T420-BT"},
648 	{0x440a, "Chelsio T404-BT"},
649 	{0x440e, "Chelsio T440-LP-CR"},
650 }, t5_pciids[] = {
651 	{0xb000, "Chelsio Terminator 5 FPGA"},
652 	{0x5400, "Chelsio T580-dbg"},
653 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
654 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
655 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
656 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
657 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
658 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
659 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
660 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
661 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
662 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
663 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
664 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
665 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
666 #ifdef notyet
667 	{0x5404,  "Chelsio T520-BCH"},
668 	{0x5405,  "Chelsio T540-BCH"},
669 	{0x5406,  "Chelsio T540-CH"},
670 	{0x5408,  "Chelsio T520-CX"},
671 	{0x540b,  "Chelsio B520-SR"},
672 	{0x540c,  "Chelsio B504-BT"},
673 	{0x540f,  "Chelsio Amsterdam"},
674 	{0x5413,  "Chelsio T580-CHR"},
675 #endif
676 }, t6_pciids[] = {
677 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
678 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
679 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
680 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
681 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
682 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
683 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
684 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
685 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
686 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
687 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
688 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
689 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
690 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
691 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
692 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
693 
694 	/* Custom */
695 	{0x6480, "Chelsio T6225 80"},
696 	{0x6481, "Chelsio T62100 81"},
697 	{0x6484, "Chelsio T62100 84"},
698 };
699 
700 #ifdef TCP_OFFLOAD
701 /*
702  * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
703  * exactly the same for both rxq and ofld_rxq.
704  */
705 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
706 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
707 #endif
708 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
709 
710 static int
711 t4_probe(device_t dev)
712 {
713 	int i;
714 	uint16_t v = pci_get_vendor(dev);
715 	uint16_t d = pci_get_device(dev);
716 	uint8_t f = pci_get_function(dev);
717 
718 	if (v != PCI_VENDOR_ID_CHELSIO)
719 		return (ENXIO);
720 
721 	/* Attach only to PF0 of the FPGA */
722 	if (d == 0xa000 && f != 0)
723 		return (ENXIO);
724 
725 	for (i = 0; i < nitems(t4_pciids); i++) {
726 		if (d == t4_pciids[i].device) {
727 			device_set_desc(dev, t4_pciids[i].desc);
728 			return (BUS_PROBE_DEFAULT);
729 		}
730 	}
731 
732 	return (ENXIO);
733 }
734 
735 static int
736 t5_probe(device_t dev)
737 {
738 	int i;
739 	uint16_t v = pci_get_vendor(dev);
740 	uint16_t d = pci_get_device(dev);
741 	uint8_t f = pci_get_function(dev);
742 
743 	if (v != PCI_VENDOR_ID_CHELSIO)
744 		return (ENXIO);
745 
746 	/* Attach only to PF0 of the FPGA */
747 	if (d == 0xb000 && f != 0)
748 		return (ENXIO);
749 
750 	for (i = 0; i < nitems(t5_pciids); i++) {
751 		if (d == t5_pciids[i].device) {
752 			device_set_desc(dev, t5_pciids[i].desc);
753 			return (BUS_PROBE_DEFAULT);
754 		}
755 	}
756 
757 	return (ENXIO);
758 }
759 
760 static int
761 t6_probe(device_t dev)
762 {
763 	int i;
764 	uint16_t v = pci_get_vendor(dev);
765 	uint16_t d = pci_get_device(dev);
766 
767 	if (v != PCI_VENDOR_ID_CHELSIO)
768 		return (ENXIO);
769 
770 	for (i = 0; i < nitems(t6_pciids); i++) {
771 		if (d == t6_pciids[i].device) {
772 			device_set_desc(dev, t6_pciids[i].desc);
773 			return (BUS_PROBE_DEFAULT);
774 		}
775 	}
776 
777 	return (ENXIO);
778 }
779 
780 static void
781 t5_attribute_workaround(device_t dev)
782 {
783 	device_t root_port;
784 	uint32_t v;
785 
786 	/*
787 	 * The T5 chips do not properly echo the No Snoop and Relaxed
788 	 * Ordering attributes when replying to a TLP from a Root
789 	 * Port.  As a workaround, find the parent Root Port and
790 	 * disable No Snoop and Relaxed Ordering.  Note that this
791 	 * affects all devices under this root port.
792 	 */
793 	root_port = pci_find_pcie_root_port(dev);
794 	if (root_port == NULL) {
795 		device_printf(dev, "Unable to find parent root port\n");
796 		return;
797 	}
798 
799 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
800 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
801 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
802 	    0)
803 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
804 		    device_get_nameunit(root_port));
805 }
806 
807 static const struct devnames devnames[] = {
808 	{
809 		.nexus_name = "t4nex",
810 		.ifnet_name = "cxgbe",
811 		.vi_ifnet_name = "vcxgbe",
812 		.pf03_drv_name = "t4iov",
813 		.vf_nexus_name = "t4vf",
814 		.vf_ifnet_name = "cxgbev"
815 	}, {
816 		.nexus_name = "t5nex",
817 		.ifnet_name = "cxl",
818 		.vi_ifnet_name = "vcxl",
819 		.pf03_drv_name = "t5iov",
820 		.vf_nexus_name = "t5vf",
821 		.vf_ifnet_name = "cxlv"
822 	}, {
823 		.nexus_name = "t6nex",
824 		.ifnet_name = "cc",
825 		.vi_ifnet_name = "vcc",
826 		.pf03_drv_name = "t6iov",
827 		.vf_nexus_name = "t6vf",
828 		.vf_ifnet_name = "ccv"
829 	}
830 };
831 
832 void
833 t4_init_devnames(struct adapter *sc)
834 {
835 	int id;
836 
837 	id = chip_id(sc);
838 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
839 		sc->names = &devnames[id - CHELSIO_T4];
840 	else {
841 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
842 		sc->names = NULL;
843 	}
844 }
845 
846 static int
847 t4_attach(device_t dev)
848 {
849 	struct adapter *sc;
850 	int rc = 0, i, j, rqidx, tqidx, nports;
851 	struct make_dev_args mda;
852 	struct intrs_and_queues iaq;
853 	struct sge *s;
854 	uint32_t *buf;
855 #ifdef TCP_OFFLOAD
856 	int ofld_rqidx, ofld_tqidx;
857 #endif
858 #ifdef DEV_NETMAP
859 	int nm_rqidx, nm_tqidx;
860 #endif
861 	int num_vis;
862 
863 	sc = device_get_softc(dev);
864 	sc->dev = dev;
865 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
866 
867 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
868 		t5_attribute_workaround(dev);
869 	pci_enable_busmaster(dev);
870 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
871 		uint32_t v;
872 
873 		pci_set_max_read_req(dev, 4096);
874 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
875 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
876 		if (pcie_relaxed_ordering == 0 &&
877 		    (v | PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
878 			v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
879 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
880 		} else if (pcie_relaxed_ordering == 1 &&
881 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
882 			v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
883 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
884 		}
885 	}
886 
887 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
888 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
889 	sc->traceq = -1;
890 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
891 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
892 	    device_get_nameunit(dev));
893 
894 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
895 	    device_get_nameunit(dev));
896 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
897 	t4_add_adapter(sc);
898 
899 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
900 	TAILQ_INIT(&sc->sfl);
901 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
902 
903 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
904 
905 	rc = t4_map_bars_0_and_4(sc);
906 	if (rc != 0)
907 		goto done; /* error message displayed already */
908 
909 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
910 
911 	/* Prepare the adapter for operation. */
912 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
913 	rc = -t4_prep_adapter(sc, buf);
914 	free(buf, M_CXGBE);
915 	if (rc != 0) {
916 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
917 		goto done;
918 	}
919 
920 	/*
921 	 * This is the real PF# to which we're attaching.  Works from within PCI
922 	 * passthrough environments too, where pci_get_function() could return a
923 	 * different PF# depending on the passthrough configuration.  We need to
924 	 * use the real PF# in all our communication with the firmware.
925 	 */
926 	j = t4_read_reg(sc, A_PL_WHOAMI);
927 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
928 	sc->mbox = sc->pf;
929 
930 	t4_init_devnames(sc);
931 	if (sc->names == NULL) {
932 		rc = ENOTSUP;
933 		goto done; /* error message displayed already */
934 	}
935 
936 	/*
937 	 * Do this really early, with the memory windows set up even before the
938 	 * character device.  The userland tool's register i/o and mem read
939 	 * will work even in "recovery mode".
940 	 */
941 	setup_memwin(sc);
942 	if (t4_init_devlog_params(sc, 0) == 0)
943 		fixup_devlog_params(sc);
944 	make_dev_args_init(&mda);
945 	mda.mda_devsw = &t4_cdevsw;
946 	mda.mda_uid = UID_ROOT;
947 	mda.mda_gid = GID_WHEEL;
948 	mda.mda_mode = 0600;
949 	mda.mda_si_drv1 = sc;
950 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
951 	if (rc != 0)
952 		device_printf(dev, "failed to create nexus char device: %d.\n",
953 		    rc);
954 
955 	/* Go no further if recovery mode has been requested. */
956 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
957 		device_printf(dev, "recovery mode.\n");
958 		goto done;
959 	}
960 
961 #if defined(__i386__)
962 	if ((cpu_feature & CPUID_CX8) == 0) {
963 		device_printf(dev, "64 bit atomics not available.\n");
964 		rc = ENOTSUP;
965 		goto done;
966 	}
967 #endif
968 
969 	/* Prepare the firmware for operation */
970 	rc = prep_firmware(sc);
971 	if (rc != 0)
972 		goto done; /* error message displayed already */
973 
974 	rc = get_params__post_init(sc);
975 	if (rc != 0)
976 		goto done; /* error message displayed already */
977 
978 	rc = set_params__post_init(sc);
979 	if (rc != 0)
980 		goto done; /* error message displayed already */
981 
982 	rc = t4_map_bar_2(sc);
983 	if (rc != 0)
984 		goto done; /* error message displayed already */
985 
986 	rc = t4_create_dma_tag(sc);
987 	if (rc != 0)
988 		goto done; /* error message displayed already */
989 
990 	/*
991 	 * First pass over all the ports - allocate VIs and initialize some
992 	 * basic parameters like mac address, port type, etc.
993 	 */
994 	for_each_port(sc, i) {
995 		struct port_info *pi;
996 
997 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
998 		sc->port[i] = pi;
999 
1000 		/* These must be set before t4_port_init */
1001 		pi->adapter = sc;
1002 		pi->port_id = i;
1003 		/*
1004 		 * XXX: vi[0] is special so we can't delay this allocation until
1005 		 * pi->nvi's final value is known.
1006 		 */
1007 		pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
1008 		    M_ZERO | M_WAITOK);
1009 
1010 		/*
1011 		 * Allocate the "main" VI and initialize parameters
1012 		 * like mac addr.
1013 		 */
1014 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1015 		if (rc != 0) {
1016 			device_printf(dev, "unable to initialize port %d: %d\n",
1017 			    i, rc);
1018 			free(pi->vi, M_CXGBE);
1019 			free(pi, M_CXGBE);
1020 			sc->port[i] = NULL;
1021 			goto done;
1022 		}
1023 
1024 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1025 		    device_get_nameunit(dev), i);
1026 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1027 		sc->chan_map[pi->tx_chan] = i;
1028 
1029 		/* All VIs on this port share this media. */
1030 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1031 		    cxgbe_media_status);
1032 
1033 		pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
1034 		if (pi->dev == NULL) {
1035 			device_printf(dev,
1036 			    "failed to add device for port %d.\n", i);
1037 			rc = ENXIO;
1038 			goto done;
1039 		}
1040 		pi->vi[0].dev = pi->dev;
1041 		device_set_softc(pi->dev, pi);
1042 	}
1043 
1044 	/*
1045 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1046 	 */
1047 	nports = sc->params.nports;
1048 	rc = cfg_itype_and_nqueues(sc, &iaq);
1049 	if (rc != 0)
1050 		goto done; /* error message displayed already */
1051 
1052 	num_vis = iaq.num_vis;
1053 	sc->intr_type = iaq.intr_type;
1054 	sc->intr_count = iaq.nirq;
1055 
1056 	s = &sc->sge;
1057 	s->nrxq = nports * iaq.nrxq;
1058 	s->ntxq = nports * iaq.ntxq;
1059 	if (num_vis > 1) {
1060 		s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1061 		s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1062 	}
1063 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1064 	s->neq += nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
1065 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1066 #ifdef TCP_OFFLOAD
1067 	if (is_offload(sc)) {
1068 		s->nofldrxq = nports * iaq.nofldrxq;
1069 		s->nofldtxq = nports * iaq.nofldtxq;
1070 		if (num_vis > 1) {
1071 			s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1072 			s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1073 		}
1074 		s->neq += s->nofldtxq + s->nofldrxq;
1075 		s->niq += s->nofldrxq;
1076 
1077 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1078 		    M_CXGBE, M_ZERO | M_WAITOK);
1079 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1080 		    M_CXGBE, M_ZERO | M_WAITOK);
1081 	}
1082 #endif
1083 #ifdef DEV_NETMAP
1084 	if (num_vis > 1) {
1085 		s->nnmrxq = nports * (num_vis - 1) * iaq.nnmrxq_vi;
1086 		s->nnmtxq = nports * (num_vis - 1) * iaq.nnmtxq_vi;
1087 	}
1088 	s->neq += s->nnmtxq + s->nnmrxq;
1089 	s->niq += s->nnmrxq;
1090 
1091 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1092 	    M_CXGBE, M_ZERO | M_WAITOK);
1093 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1094 	    M_CXGBE, M_ZERO | M_WAITOK);
1095 #endif
1096 
1097 	s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1098 	    M_ZERO | M_WAITOK);
1099 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1100 	    M_ZERO | M_WAITOK);
1101 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1102 	    M_ZERO | M_WAITOK);
1103 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1104 	    M_ZERO | M_WAITOK);
1105 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1106 	    M_ZERO | M_WAITOK);
1107 
1108 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1109 	    M_ZERO | M_WAITOK);
1110 
1111 	t4_init_l2t(sc, M_WAITOK);
1112 	t4_init_tx_sched(sc);
1113 
1114 	/*
1115 	 * Second pass over the ports.  This time we know the number of rx and
1116 	 * tx queues that each port should get.
1117 	 */
1118 	rqidx = tqidx = 0;
1119 #ifdef TCP_OFFLOAD
1120 	ofld_rqidx = ofld_tqidx = 0;
1121 #endif
1122 #ifdef DEV_NETMAP
1123 	nm_rqidx = nm_tqidx = 0;
1124 #endif
1125 	for_each_port(sc, i) {
1126 		struct port_info *pi = sc->port[i];
1127 		struct vi_info *vi;
1128 
1129 		if (pi == NULL)
1130 			continue;
1131 
1132 		pi->nvi = num_vis;
1133 		for_each_vi(pi, j, vi) {
1134 			vi->pi = pi;
1135 			vi->qsize_rxq = t4_qsize_rxq;
1136 			vi->qsize_txq = t4_qsize_txq;
1137 
1138 			vi->first_rxq = rqidx;
1139 			vi->first_txq = tqidx;
1140 			vi->tmr_idx = t4_tmr_idx;
1141 			vi->pktc_idx = t4_pktc_idx;
1142 			vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1143 			vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1144 
1145 			rqidx += vi->nrxq;
1146 			tqidx += vi->ntxq;
1147 
1148 			if (j == 0 && vi->ntxq > 1)
1149 				vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1150 			else
1151 				vi->rsrv_noflowq = 0;
1152 
1153 #ifdef TCP_OFFLOAD
1154 			vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1155 			vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1156 			vi->first_ofld_rxq = ofld_rqidx;
1157 			vi->first_ofld_txq = ofld_tqidx;
1158 			vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1159 			vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1160 
1161 			ofld_rqidx += vi->nofldrxq;
1162 			ofld_tqidx += vi->nofldtxq;
1163 #endif
1164 #ifdef DEV_NETMAP
1165 			if (j > 0) {
1166 				vi->first_nm_rxq = nm_rqidx;
1167 				vi->first_nm_txq = nm_tqidx;
1168 				vi->nnmrxq = iaq.nnmrxq_vi;
1169 				vi->nnmtxq = iaq.nnmtxq_vi;
1170 				nm_rqidx += vi->nnmrxq;
1171 				nm_tqidx += vi->nnmtxq;
1172 			}
1173 #endif
1174 		}
1175 	}
1176 
1177 	rc = t4_setup_intr_handlers(sc);
1178 	if (rc != 0) {
1179 		device_printf(dev,
1180 		    "failed to setup interrupt handlers: %d\n", rc);
1181 		goto done;
1182 	}
1183 
1184 	rc = bus_generic_probe(dev);
1185 	if (rc != 0) {
1186 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1187 		goto done;
1188 	}
1189 
1190 	/*
1191 	 * Ensure thread-safe mailbox access (in debug builds).
1192 	 *
1193 	 * So far this was the only thread accessing the mailbox but various
1194 	 * ifnets and sysctls are about to be created and their handlers/ioctls
1195 	 * will access the mailbox from different threads.
1196 	 */
1197 	sc->flags |= CHK_MBOX_ACCESS;
1198 
1199 	rc = bus_generic_attach(dev);
1200 	if (rc != 0) {
1201 		device_printf(dev,
1202 		    "failed to attach all child ports: %d\n", rc);
1203 		goto done;
1204 	}
1205 
1206 	device_printf(dev,
1207 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1208 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1209 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1210 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1211 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1212 
1213 	t4_set_desc(sc);
1214 
1215 	notify_siblings(dev, 0);
1216 
1217 done:
1218 	if (rc != 0 && sc->cdev) {
1219 		/* cdev was created and so cxgbetool works; recover that way. */
1220 		device_printf(dev,
1221 		    "error during attach, adapter is now in recovery mode.\n");
1222 		rc = 0;
1223 	}
1224 
1225 	if (rc != 0)
1226 		t4_detach_common(dev);
1227 	else
1228 		t4_sysctls(sc);
1229 
1230 	return (rc);
1231 }
1232 
1233 static int
1234 t4_ready(device_t dev)
1235 {
1236 	struct adapter *sc;
1237 
1238 	sc = device_get_softc(dev);
1239 	if (sc->flags & FW_OK)
1240 		return (0);
1241 	return (ENXIO);
1242 }
1243 
1244 static int
1245 t4_read_port_device(device_t dev, int port, device_t *child)
1246 {
1247 	struct adapter *sc;
1248 	struct port_info *pi;
1249 
1250 	sc = device_get_softc(dev);
1251 	if (port < 0 || port >= MAX_NPORTS)
1252 		return (EINVAL);
1253 	pi = sc->port[port];
1254 	if (pi == NULL || pi->dev == NULL)
1255 		return (ENXIO);
1256 	*child = pi->dev;
1257 	return (0);
1258 }
1259 
1260 static int
1261 notify_siblings(device_t dev, int detaching)
1262 {
1263 	device_t sibling;
1264 	int error, i;
1265 
1266 	error = 0;
1267 	for (i = 0; i < PCI_FUNCMAX; i++) {
1268 		if (i == pci_get_function(dev))
1269 			continue;
1270 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1271 		    pci_get_slot(dev), i);
1272 		if (sibling == NULL || !device_is_attached(sibling))
1273 			continue;
1274 		if (detaching)
1275 			error = T4_DETACH_CHILD(sibling);
1276 		else
1277 			(void)T4_ATTACH_CHILD(sibling);
1278 		if (error)
1279 			break;
1280 	}
1281 	return (error);
1282 }
1283 
1284 /*
1285  * Idempotent
1286  */
1287 static int
1288 t4_detach(device_t dev)
1289 {
1290 	struct adapter *sc;
1291 	int rc;
1292 
1293 	sc = device_get_softc(dev);
1294 
1295 	rc = notify_siblings(dev, 1);
1296 	if (rc) {
1297 		device_printf(dev,
1298 		    "failed to detach sibling devices: %d\n", rc);
1299 		return (rc);
1300 	}
1301 
1302 	return (t4_detach_common(dev));
1303 }
1304 
1305 int
1306 t4_detach_common(device_t dev)
1307 {
1308 	struct adapter *sc;
1309 	struct port_info *pi;
1310 	int i, rc;
1311 
1312 	sc = device_get_softc(dev);
1313 
1314 	sc->flags &= ~CHK_MBOX_ACCESS;
1315 	if (sc->flags & FULL_INIT_DONE) {
1316 		if (!(sc->flags & IS_VF))
1317 			t4_intr_disable(sc);
1318 	}
1319 
1320 	if (sc->cdev) {
1321 		destroy_dev(sc->cdev);
1322 		sc->cdev = NULL;
1323 	}
1324 
1325 	if (device_is_attached(dev)) {
1326 		rc = bus_generic_detach(dev);
1327 		if (rc) {
1328 			device_printf(dev,
1329 			    "failed to detach child devices: %d\n", rc);
1330 			return (rc);
1331 		}
1332 	}
1333 
1334 	for (i = 0; i < sc->intr_count; i++)
1335 		t4_free_irq(sc, &sc->irq[i]);
1336 
1337 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1338 		t4_free_tx_sched(sc);
1339 
1340 	for (i = 0; i < MAX_NPORTS; i++) {
1341 		pi = sc->port[i];
1342 		if (pi) {
1343 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1344 			if (pi->dev)
1345 				device_delete_child(dev, pi->dev);
1346 
1347 			mtx_destroy(&pi->pi_lock);
1348 			free(pi->vi, M_CXGBE);
1349 			free(pi, M_CXGBE);
1350 		}
1351 	}
1352 
1353 	device_delete_children(dev);
1354 
1355 	if (sc->flags & FULL_INIT_DONE)
1356 		adapter_full_uninit(sc);
1357 
1358 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1359 		t4_fw_bye(sc, sc->mbox);
1360 
1361 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1362 		pci_release_msi(dev);
1363 
1364 	if (sc->regs_res)
1365 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1366 		    sc->regs_res);
1367 
1368 	if (sc->udbs_res)
1369 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1370 		    sc->udbs_res);
1371 
1372 	if (sc->msix_res)
1373 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1374 		    sc->msix_res);
1375 
1376 	if (sc->l2t)
1377 		t4_free_l2t(sc->l2t);
1378 
1379 #ifdef TCP_OFFLOAD
1380 	free(sc->sge.ofld_rxq, M_CXGBE);
1381 	free(sc->sge.ofld_txq, M_CXGBE);
1382 #endif
1383 #ifdef DEV_NETMAP
1384 	free(sc->sge.nm_rxq, M_CXGBE);
1385 	free(sc->sge.nm_txq, M_CXGBE);
1386 #endif
1387 	free(sc->irq, M_CXGBE);
1388 	free(sc->sge.rxq, M_CXGBE);
1389 	free(sc->sge.txq, M_CXGBE);
1390 	free(sc->sge.ctrlq, M_CXGBE);
1391 	free(sc->sge.iqmap, M_CXGBE);
1392 	free(sc->sge.eqmap, M_CXGBE);
1393 	free(sc->tids.ftid_tab, M_CXGBE);
1394 	free(sc->tt.tls_rx_ports, M_CXGBE);
1395 	t4_destroy_dma_tag(sc);
1396 	if (mtx_initialized(&sc->sc_lock)) {
1397 		sx_xlock(&t4_list_lock);
1398 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1399 		sx_xunlock(&t4_list_lock);
1400 		mtx_destroy(&sc->sc_lock);
1401 	}
1402 
1403 	callout_drain(&sc->sfl_callout);
1404 	if (mtx_initialized(&sc->tids.ftid_lock))
1405 		mtx_destroy(&sc->tids.ftid_lock);
1406 	if (mtx_initialized(&sc->sfl_lock))
1407 		mtx_destroy(&sc->sfl_lock);
1408 	if (mtx_initialized(&sc->ifp_lock))
1409 		mtx_destroy(&sc->ifp_lock);
1410 	if (mtx_initialized(&sc->reg_lock))
1411 		mtx_destroy(&sc->reg_lock);
1412 
1413 	for (i = 0; i < NUM_MEMWIN; i++) {
1414 		struct memwin *mw = &sc->memwin[i];
1415 
1416 		if (rw_initialized(&mw->mw_lock))
1417 			rw_destroy(&mw->mw_lock);
1418 	}
1419 
1420 	bzero(sc, sizeof(*sc));
1421 
1422 	return (0);
1423 }
1424 
1425 static int
1426 cxgbe_probe(device_t dev)
1427 {
1428 	char buf[128];
1429 	struct port_info *pi = device_get_softc(dev);
1430 
1431 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1432 	device_set_desc_copy(dev, buf);
1433 
1434 	return (BUS_PROBE_DEFAULT);
1435 }
1436 
1437 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1438     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1439     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1440 #define T4_CAP_ENABLE (T4_CAP)
1441 
1442 static int
1443 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1444 {
1445 	struct ifnet *ifp;
1446 	struct sbuf *sb;
1447 
1448 	vi->xact_addr_filt = -1;
1449 	callout_init(&vi->tick, 1);
1450 
1451 	/* Allocate an ifnet and set it up */
1452 	ifp = if_alloc(IFT_ETHER);
1453 	if (ifp == NULL) {
1454 		device_printf(dev, "Cannot allocate ifnet\n");
1455 		return (ENOMEM);
1456 	}
1457 	vi->ifp = ifp;
1458 	ifp->if_softc = vi;
1459 
1460 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1461 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1462 
1463 	ifp->if_init = cxgbe_init;
1464 	ifp->if_ioctl = cxgbe_ioctl;
1465 	ifp->if_transmit = cxgbe_transmit;
1466 	ifp->if_qflush = cxgbe_qflush;
1467 	ifp->if_get_counter = cxgbe_get_counter;
1468 
1469 	ifp->if_capabilities = T4_CAP;
1470 #ifdef TCP_OFFLOAD
1471 	if (vi->nofldrxq != 0)
1472 		ifp->if_capabilities |= IFCAP_TOE;
1473 #endif
1474 #ifdef DEV_NETMAP
1475 	if (vi->nnmrxq != 0)
1476 		ifp->if_capabilities |= IFCAP_NETMAP;
1477 #endif
1478 	ifp->if_capenable = T4_CAP_ENABLE;
1479 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1480 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1481 
1482 	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1483 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1484 	ifp->if_hw_tsomaxsegsize = 65536;
1485 
1486 	vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1487 	    EVENTHANDLER_PRI_ANY);
1488 
1489 	ether_ifattach(ifp, vi->hw_addr);
1490 #ifdef DEV_NETMAP
1491 	if (ifp->if_capabilities & IFCAP_NETMAP)
1492 		cxgbe_nm_attach(vi);
1493 #endif
1494 	sb = sbuf_new_auto();
1495 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1496 #ifdef TCP_OFFLOAD
1497 	if (ifp->if_capabilities & IFCAP_TOE)
1498 		sbuf_printf(sb, "; %d txq, %d rxq (TOE)",
1499 		    vi->nofldtxq, vi->nofldrxq);
1500 #endif
1501 #ifdef DEV_NETMAP
1502 	if (ifp->if_capabilities & IFCAP_NETMAP)
1503 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1504 		    vi->nnmtxq, vi->nnmrxq);
1505 #endif
1506 	sbuf_finish(sb);
1507 	device_printf(dev, "%s\n", sbuf_data(sb));
1508 	sbuf_delete(sb);
1509 
1510 	vi_sysctls(vi);
1511 
1512 	return (0);
1513 }
1514 
1515 static int
1516 cxgbe_attach(device_t dev)
1517 {
1518 	struct port_info *pi = device_get_softc(dev);
1519 	struct adapter *sc = pi->adapter;
1520 	struct vi_info *vi;
1521 	int i, rc;
1522 
1523 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1524 
1525 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1526 	if (rc)
1527 		return (rc);
1528 
1529 	for_each_vi(pi, i, vi) {
1530 		if (i == 0)
1531 			continue;
1532 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1533 		if (vi->dev == NULL) {
1534 			device_printf(dev, "failed to add VI %d\n", i);
1535 			continue;
1536 		}
1537 		device_set_softc(vi->dev, vi);
1538 	}
1539 
1540 	cxgbe_sysctls(pi);
1541 
1542 	bus_generic_attach(dev);
1543 
1544 	return (0);
1545 }
1546 
1547 static void
1548 cxgbe_vi_detach(struct vi_info *vi)
1549 {
1550 	struct ifnet *ifp = vi->ifp;
1551 
1552 	ether_ifdetach(ifp);
1553 
1554 	if (vi->vlan_c)
1555 		EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1556 
1557 	/* Let detach proceed even if these fail. */
1558 #ifdef DEV_NETMAP
1559 	if (ifp->if_capabilities & IFCAP_NETMAP)
1560 		cxgbe_nm_detach(vi);
1561 #endif
1562 	cxgbe_uninit_synchronized(vi);
1563 	callout_drain(&vi->tick);
1564 	vi_full_uninit(vi);
1565 
1566 	if_free(vi->ifp);
1567 	vi->ifp = NULL;
1568 }
1569 
1570 static int
1571 cxgbe_detach(device_t dev)
1572 {
1573 	struct port_info *pi = device_get_softc(dev);
1574 	struct adapter *sc = pi->adapter;
1575 	int rc;
1576 
1577 	/* Detach the extra VIs first. */
1578 	rc = bus_generic_detach(dev);
1579 	if (rc)
1580 		return (rc);
1581 	device_delete_children(dev);
1582 
1583 	doom_vi(sc, &pi->vi[0]);
1584 
1585 	if (pi->flags & HAS_TRACEQ) {
1586 		sc->traceq = -1;	/* cloner should not create ifnet */
1587 		t4_tracer_port_detach(sc);
1588 	}
1589 
1590 	cxgbe_vi_detach(&pi->vi[0]);
1591 	callout_drain(&pi->tick);
1592 	ifmedia_removeall(&pi->media);
1593 
1594 	end_synchronized_op(sc, 0);
1595 
1596 	return (0);
1597 }
1598 
1599 static void
1600 cxgbe_init(void *arg)
1601 {
1602 	struct vi_info *vi = arg;
1603 	struct adapter *sc = vi->pi->adapter;
1604 
1605 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1606 		return;
1607 	cxgbe_init_synchronized(vi);
1608 	end_synchronized_op(sc, 0);
1609 }
1610 
1611 static int
1612 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1613 {
1614 	int rc = 0, mtu, flags, can_sleep;
1615 	struct vi_info *vi = ifp->if_softc;
1616 	struct port_info *pi = vi->pi;
1617 	struct adapter *sc = pi->adapter;
1618 	struct ifreq *ifr = (struct ifreq *)data;
1619 	uint32_t mask;
1620 
1621 	switch (cmd) {
1622 	case SIOCSIFMTU:
1623 		mtu = ifr->ifr_mtu;
1624 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1625 			return (EINVAL);
1626 
1627 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1628 		if (rc)
1629 			return (rc);
1630 		ifp->if_mtu = mtu;
1631 		if (vi->flags & VI_INIT_DONE) {
1632 			t4_update_fl_bufsize(ifp);
1633 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1634 				rc = update_mac_settings(ifp, XGMAC_MTU);
1635 		}
1636 		end_synchronized_op(sc, 0);
1637 		break;
1638 
1639 	case SIOCSIFFLAGS:
1640 		can_sleep = 0;
1641 redo_sifflags:
1642 		rc = begin_synchronized_op(sc, vi,
1643 		    can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1644 		if (rc)
1645 			return (rc);
1646 
1647 		if (ifp->if_flags & IFF_UP) {
1648 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1649 				flags = vi->if_flags;
1650 				if ((ifp->if_flags ^ flags) &
1651 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1652 					if (can_sleep == 1) {
1653 						end_synchronized_op(sc, 0);
1654 						can_sleep = 0;
1655 						goto redo_sifflags;
1656 					}
1657 					rc = update_mac_settings(ifp,
1658 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1659 				}
1660 			} else {
1661 				if (can_sleep == 0) {
1662 					end_synchronized_op(sc, LOCK_HELD);
1663 					can_sleep = 1;
1664 					goto redo_sifflags;
1665 				}
1666 				rc = cxgbe_init_synchronized(vi);
1667 			}
1668 			vi->if_flags = ifp->if_flags;
1669 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1670 			if (can_sleep == 0) {
1671 				end_synchronized_op(sc, LOCK_HELD);
1672 				can_sleep = 1;
1673 				goto redo_sifflags;
1674 			}
1675 			rc = cxgbe_uninit_synchronized(vi);
1676 		}
1677 		end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1678 		break;
1679 
1680 	case SIOCADDMULTI:
1681 	case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1682 		rc = begin_synchronized_op(sc, vi, HOLD_LOCK, "t4multi");
1683 		if (rc)
1684 			return (rc);
1685 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1686 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1687 		end_synchronized_op(sc, LOCK_HELD);
1688 		break;
1689 
1690 	case SIOCSIFCAP:
1691 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1692 		if (rc)
1693 			return (rc);
1694 
1695 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1696 		if (mask & IFCAP_TXCSUM) {
1697 			ifp->if_capenable ^= IFCAP_TXCSUM;
1698 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1699 
1700 			if (IFCAP_TSO4 & ifp->if_capenable &&
1701 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1702 				ifp->if_capenable &= ~IFCAP_TSO4;
1703 				if_printf(ifp,
1704 				    "tso4 disabled due to -txcsum.\n");
1705 			}
1706 		}
1707 		if (mask & IFCAP_TXCSUM_IPV6) {
1708 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1709 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1710 
1711 			if (IFCAP_TSO6 & ifp->if_capenable &&
1712 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1713 				ifp->if_capenable &= ~IFCAP_TSO6;
1714 				if_printf(ifp,
1715 				    "tso6 disabled due to -txcsum6.\n");
1716 			}
1717 		}
1718 		if (mask & IFCAP_RXCSUM)
1719 			ifp->if_capenable ^= IFCAP_RXCSUM;
1720 		if (mask & IFCAP_RXCSUM_IPV6)
1721 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1722 
1723 		/*
1724 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1725 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1726 		 * sending a TSO request our way, so it's sufficient to toggle
1727 		 * IFCAP_TSOx only.
1728 		 */
1729 		if (mask & IFCAP_TSO4) {
1730 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1731 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1732 				if_printf(ifp, "enable txcsum first.\n");
1733 				rc = EAGAIN;
1734 				goto fail;
1735 			}
1736 			ifp->if_capenable ^= IFCAP_TSO4;
1737 		}
1738 		if (mask & IFCAP_TSO6) {
1739 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1740 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1741 				if_printf(ifp, "enable txcsum6 first.\n");
1742 				rc = EAGAIN;
1743 				goto fail;
1744 			}
1745 			ifp->if_capenable ^= IFCAP_TSO6;
1746 		}
1747 		if (mask & IFCAP_LRO) {
1748 #if defined(INET) || defined(INET6)
1749 			int i;
1750 			struct sge_rxq *rxq;
1751 
1752 			ifp->if_capenable ^= IFCAP_LRO;
1753 			for_each_rxq(vi, i, rxq) {
1754 				if (ifp->if_capenable & IFCAP_LRO)
1755 					rxq->iq.flags |= IQ_LRO_ENABLED;
1756 				else
1757 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1758 			}
1759 #endif
1760 		}
1761 #ifdef TCP_OFFLOAD
1762 		if (mask & IFCAP_TOE) {
1763 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1764 
1765 			rc = toe_capability(vi, enable);
1766 			if (rc != 0)
1767 				goto fail;
1768 
1769 			ifp->if_capenable ^= mask;
1770 		}
1771 #endif
1772 		if (mask & IFCAP_VLAN_HWTAGGING) {
1773 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1774 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1775 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1776 		}
1777 		if (mask & IFCAP_VLAN_MTU) {
1778 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1779 
1780 			/* Need to find out how to disable auto-mtu-inflation */
1781 		}
1782 		if (mask & IFCAP_VLAN_HWTSO)
1783 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1784 		if (mask & IFCAP_VLAN_HWCSUM)
1785 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1786 
1787 #ifdef VLAN_CAPABILITIES
1788 		VLAN_CAPABILITIES(ifp);
1789 #endif
1790 fail:
1791 		end_synchronized_op(sc, 0);
1792 		break;
1793 
1794 	case SIOCSIFMEDIA:
1795 	case SIOCGIFMEDIA:
1796 	case SIOCGIFXMEDIA:
1797 		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1798 		break;
1799 
1800 	case SIOCGI2C: {
1801 		struct ifi2creq i2c;
1802 
1803 		rc = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
1804 		if (rc != 0)
1805 			break;
1806 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1807 			rc = EPERM;
1808 			break;
1809 		}
1810 		if (i2c.len > sizeof(i2c.data)) {
1811 			rc = EINVAL;
1812 			break;
1813 		}
1814 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1815 		if (rc)
1816 			return (rc);
1817 		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
1818 		    i2c.offset, i2c.len, &i2c.data[0]);
1819 		end_synchronized_op(sc, 0);
1820 		if (rc == 0)
1821 			rc = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
1822 		break;
1823 	}
1824 
1825 	default:
1826 		rc = ether_ioctl(ifp, cmd, data);
1827 	}
1828 
1829 	return (rc);
1830 }
1831 
1832 static int
1833 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1834 {
1835 	struct vi_info *vi = ifp->if_softc;
1836 	struct port_info *pi = vi->pi;
1837 	struct adapter *sc = pi->adapter;
1838 	struct sge_txq *txq;
1839 	void *items[1];
1840 	int rc;
1841 
1842 	M_ASSERTPKTHDR(m);
1843 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1844 
1845 	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1846 		m_freem(m);
1847 		return (ENETDOWN);
1848 	}
1849 
1850 	rc = parse_pkt(sc, &m);
1851 	if (__predict_false(rc != 0)) {
1852 		MPASS(m == NULL);			/* was freed already */
1853 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1854 		return (rc);
1855 	}
1856 
1857 	/* Select a txq. */
1858 	txq = &sc->sge.txq[vi->first_txq];
1859 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1860 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1861 		    vi->rsrv_noflowq);
1862 
1863 	items[0] = m;
1864 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1865 	if (__predict_false(rc != 0))
1866 		m_freem(m);
1867 
1868 	return (rc);
1869 }
1870 
1871 static void
1872 cxgbe_qflush(struct ifnet *ifp)
1873 {
1874 	struct vi_info *vi = ifp->if_softc;
1875 	struct sge_txq *txq;
1876 	int i;
1877 
1878 	/* queues do not exist if !VI_INIT_DONE. */
1879 	if (vi->flags & VI_INIT_DONE) {
1880 		for_each_txq(vi, i, txq) {
1881 			TXQ_LOCK(txq);
1882 			txq->eq.flags |= EQ_QFLUSH;
1883 			TXQ_UNLOCK(txq);
1884 			while (!mp_ring_is_idle(txq->r)) {
1885 				mp_ring_check_drainage(txq->r, 0);
1886 				pause("qflush", 1);
1887 			}
1888 			TXQ_LOCK(txq);
1889 			txq->eq.flags &= ~EQ_QFLUSH;
1890 			TXQ_UNLOCK(txq);
1891 		}
1892 	}
1893 	if_qflush(ifp);
1894 }
1895 
1896 static uint64_t
1897 vi_get_counter(struct ifnet *ifp, ift_counter c)
1898 {
1899 	struct vi_info *vi = ifp->if_softc;
1900 	struct fw_vi_stats_vf *s = &vi->stats;
1901 
1902 	vi_refresh_stats(vi->pi->adapter, vi);
1903 
1904 	switch (c) {
1905 	case IFCOUNTER_IPACKETS:
1906 		return (s->rx_bcast_frames + s->rx_mcast_frames +
1907 		    s->rx_ucast_frames);
1908 	case IFCOUNTER_IERRORS:
1909 		return (s->rx_err_frames);
1910 	case IFCOUNTER_OPACKETS:
1911 		return (s->tx_bcast_frames + s->tx_mcast_frames +
1912 		    s->tx_ucast_frames + s->tx_offload_frames);
1913 	case IFCOUNTER_OERRORS:
1914 		return (s->tx_drop_frames);
1915 	case IFCOUNTER_IBYTES:
1916 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
1917 		    s->rx_ucast_bytes);
1918 	case IFCOUNTER_OBYTES:
1919 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
1920 		    s->tx_ucast_bytes + s->tx_offload_bytes);
1921 	case IFCOUNTER_IMCASTS:
1922 		return (s->rx_mcast_frames);
1923 	case IFCOUNTER_OMCASTS:
1924 		return (s->tx_mcast_frames);
1925 	case IFCOUNTER_OQDROPS: {
1926 		uint64_t drops;
1927 
1928 		drops = 0;
1929 		if (vi->flags & VI_INIT_DONE) {
1930 			int i;
1931 			struct sge_txq *txq;
1932 
1933 			for_each_txq(vi, i, txq)
1934 				drops += counter_u64_fetch(txq->r->drops);
1935 		}
1936 
1937 		return (drops);
1938 
1939 	}
1940 
1941 	default:
1942 		return (if_get_counter_default(ifp, c));
1943 	}
1944 }
1945 
1946 uint64_t
1947 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
1948 {
1949 	struct vi_info *vi = ifp->if_softc;
1950 	struct port_info *pi = vi->pi;
1951 	struct adapter *sc = pi->adapter;
1952 	struct port_stats *s = &pi->stats;
1953 
1954 	if (pi->nvi > 1 || sc->flags & IS_VF)
1955 		return (vi_get_counter(ifp, c));
1956 
1957 	cxgbe_refresh_stats(sc, pi);
1958 
1959 	switch (c) {
1960 	case IFCOUNTER_IPACKETS:
1961 		return (s->rx_frames);
1962 
1963 	case IFCOUNTER_IERRORS:
1964 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
1965 		    s->rx_fcs_err + s->rx_len_err);
1966 
1967 	case IFCOUNTER_OPACKETS:
1968 		return (s->tx_frames);
1969 
1970 	case IFCOUNTER_OERRORS:
1971 		return (s->tx_error_frames);
1972 
1973 	case IFCOUNTER_IBYTES:
1974 		return (s->rx_octets);
1975 
1976 	case IFCOUNTER_OBYTES:
1977 		return (s->tx_octets);
1978 
1979 	case IFCOUNTER_IMCASTS:
1980 		return (s->rx_mcast_frames);
1981 
1982 	case IFCOUNTER_OMCASTS:
1983 		return (s->tx_mcast_frames);
1984 
1985 	case IFCOUNTER_IQDROPS:
1986 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
1987 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
1988 		    s->rx_trunc3 + pi->tnl_cong_drops);
1989 
1990 	case IFCOUNTER_OQDROPS: {
1991 		uint64_t drops;
1992 
1993 		drops = s->tx_drop;
1994 		if (vi->flags & VI_INIT_DONE) {
1995 			int i;
1996 			struct sge_txq *txq;
1997 
1998 			for_each_txq(vi, i, txq)
1999 				drops += counter_u64_fetch(txq->r->drops);
2000 		}
2001 
2002 		return (drops);
2003 
2004 	}
2005 
2006 	default:
2007 		return (if_get_counter_default(ifp, c));
2008 	}
2009 }
2010 
2011 static int
2012 cxgbe_media_change(struct ifnet *ifp)
2013 {
2014 	struct vi_info *vi = ifp->if_softc;
2015 
2016 	device_printf(vi->dev, "%s unimplemented.\n", __func__);
2017 
2018 	return (EOPNOTSUPP);
2019 }
2020 
2021 static void
2022 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2023 {
2024 	struct vi_info *vi = ifp->if_softc;
2025 	struct port_info *pi = vi->pi;
2026 	struct ifmedia_entry *cur;
2027 	struct link_config *lc = &pi->link_cfg;
2028 
2029 	/*
2030 	 * If all the interfaces are administratively down the firmware does not
2031 	 * report transceiver changes.  Refresh port info here so that ifconfig
2032 	 * displays accurate information at all times.
2033 	 */
2034 	if (begin_synchronized_op(pi->adapter, NULL, SLEEP_OK | INTR_OK,
2035 	    "t4med") == 0) {
2036 		PORT_LOCK(pi);
2037 		if (pi->up_vis == 0) {
2038 			t4_update_port_info(pi);
2039 			build_medialist(pi, &pi->media);
2040 		}
2041 		PORT_UNLOCK(pi);
2042 		end_synchronized_op(pi->adapter, 0);
2043 	}
2044 
2045 	ifmr->ifm_status = IFM_AVALID;
2046 	if (lc->link_ok == 0)
2047 		return;
2048 
2049 	ifmr->ifm_status |= IFM_ACTIVE;
2050 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2051 	if (lc->fc & PAUSE_RX)
2052 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2053 	if (lc->fc & PAUSE_TX)
2054 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2055 
2056 	/* active and current will differ iff current media is autoselect. */
2057 	cur = pi->media.ifm_cur;
2058 	if (cur != NULL && IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
2059 		return;
2060 
2061 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2062 	if (lc->fc & PAUSE_RX)
2063 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2064 	if (lc->fc & PAUSE_TX)
2065 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2066 	switch (lc->speed) {
2067 	case 10000:
2068 		ifmr->ifm_active |= IFM_10G_T;
2069 		break;
2070 	case 1000:
2071 		ifmr->ifm_active |= IFM_1000_T;
2072 		break;
2073 	case 100:
2074 		ifmr->ifm_active |= IFM_100_TX;
2075 		break;
2076 	case 10:
2077 		ifmr->ifm_active |= IFM_10_T;
2078 		break;
2079 	default:
2080 		device_printf(vi->dev, "link up but speed unknown (%u)\n",
2081 		    lc->speed);
2082 	}
2083 }
2084 
2085 static int
2086 vcxgbe_probe(device_t dev)
2087 {
2088 	char buf[128];
2089 	struct vi_info *vi = device_get_softc(dev);
2090 
2091 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2092 	    vi - vi->pi->vi);
2093 	device_set_desc_copy(dev, buf);
2094 
2095 	return (BUS_PROBE_DEFAULT);
2096 }
2097 
2098 static int
2099 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2100 {
2101 	int func, index, rc;
2102 	uint32_t param, val;
2103 
2104 	ASSERT_SYNCHRONIZED_OP(sc);
2105 
2106 	index = vi - pi->vi;
2107 	MPASS(index > 0);	/* This function deals with _extra_ VIs only */
2108 	KASSERT(index < nitems(vi_mac_funcs),
2109 	    ("%s: VI %s doesn't have a MAC func", __func__,
2110 	    device_get_nameunit(vi->dev)));
2111 	func = vi_mac_funcs[index];
2112 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2113 	    vi->hw_addr, &vi->rss_size, func, 0);
2114 	if (rc < 0) {
2115 		device_printf(vi->dev, "failed to allocate virtual interface %d"
2116 		    "for port %d: %d\n", index, pi->port_id, -rc);
2117 		return (-rc);
2118 	}
2119 	vi->viid = rc;
2120 	if (chip_id(sc) <= CHELSIO_T5)
2121 		vi->smt_idx = (rc & 0x7f) << 1;
2122 	else
2123 		vi->smt_idx = (rc & 0x7f);
2124 
2125 	if (vi->rss_size == 1) {
2126 		/*
2127 		 * This VI didn't get a slice of the RSS table.  Reduce the
2128 		 * number of VIs being created (hw.cxgbe.num_vis) or modify the
2129 		 * configuration file (nvi, rssnvi for this PF) if this is a
2130 		 * problem.
2131 		 */
2132 		device_printf(vi->dev, "RSS table not available.\n");
2133 		vi->rss_base = 0xffff;
2134 
2135 		return (0);
2136 	}
2137 
2138 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2139 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2140 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2141 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2142 	if (rc)
2143 		vi->rss_base = 0xffff;
2144 	else {
2145 		MPASS((val >> 16) == vi->rss_size);
2146 		vi->rss_base = val & 0xffff;
2147 	}
2148 
2149 	return (0);
2150 }
2151 
2152 static int
2153 vcxgbe_attach(device_t dev)
2154 {
2155 	struct vi_info *vi;
2156 	struct port_info *pi;
2157 	struct adapter *sc;
2158 	int rc;
2159 
2160 	vi = device_get_softc(dev);
2161 	pi = vi->pi;
2162 	sc = pi->adapter;
2163 
2164 	rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2165 	if (rc)
2166 		return (rc);
2167 	rc = alloc_extra_vi(sc, pi, vi);
2168 	end_synchronized_op(sc, 0);
2169 	if (rc)
2170 		return (rc);
2171 
2172 	rc = cxgbe_vi_attach(dev, vi);
2173 	if (rc) {
2174 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2175 		return (rc);
2176 	}
2177 	return (0);
2178 }
2179 
2180 static int
2181 vcxgbe_detach(device_t dev)
2182 {
2183 	struct vi_info *vi;
2184 	struct adapter *sc;
2185 
2186 	vi = device_get_softc(dev);
2187 	sc = vi->pi->adapter;
2188 
2189 	doom_vi(sc, vi);
2190 
2191 	cxgbe_vi_detach(vi);
2192 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2193 
2194 	end_synchronized_op(sc, 0);
2195 
2196 	return (0);
2197 }
2198 
2199 void
2200 t4_fatal_err(struct adapter *sc)
2201 {
2202 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2203 	t4_intr_disable(sc);
2204 	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2205 	    device_get_nameunit(sc->dev));
2206 }
2207 
2208 void
2209 t4_add_adapter(struct adapter *sc)
2210 {
2211 	sx_xlock(&t4_list_lock);
2212 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2213 	sx_xunlock(&t4_list_lock);
2214 }
2215 
2216 int
2217 t4_map_bars_0_and_4(struct adapter *sc)
2218 {
2219 	sc->regs_rid = PCIR_BAR(0);
2220 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2221 	    &sc->regs_rid, RF_ACTIVE);
2222 	if (sc->regs_res == NULL) {
2223 		device_printf(sc->dev, "cannot map registers.\n");
2224 		return (ENXIO);
2225 	}
2226 	sc->bt = rman_get_bustag(sc->regs_res);
2227 	sc->bh = rman_get_bushandle(sc->regs_res);
2228 	sc->mmio_len = rman_get_size(sc->regs_res);
2229 	setbit(&sc->doorbells, DOORBELL_KDB);
2230 
2231 	sc->msix_rid = PCIR_BAR(4);
2232 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2233 	    &sc->msix_rid, RF_ACTIVE);
2234 	if (sc->msix_res == NULL) {
2235 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2236 		return (ENXIO);
2237 	}
2238 
2239 	return (0);
2240 }
2241 
2242 int
2243 t4_map_bar_2(struct adapter *sc)
2244 {
2245 
2246 	/*
2247 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2248 	 * to map it if RDMA is disabled.
2249 	 */
2250 	if (is_t4(sc) && sc->rdmacaps == 0)
2251 		return (0);
2252 
2253 	sc->udbs_rid = PCIR_BAR(2);
2254 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2255 	    &sc->udbs_rid, RF_ACTIVE);
2256 	if (sc->udbs_res == NULL) {
2257 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2258 		return (ENXIO);
2259 	}
2260 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2261 
2262 	if (chip_id(sc) >= CHELSIO_T5) {
2263 		setbit(&sc->doorbells, DOORBELL_UDB);
2264 #if defined(__i386__) || defined(__amd64__)
2265 		if (t5_write_combine) {
2266 			int rc, mode;
2267 
2268 			/*
2269 			 * Enable write combining on BAR2.  This is the
2270 			 * userspace doorbell BAR and is split into 128B
2271 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2272 			 * with an egress queue.  The first 64B has the doorbell
2273 			 * and the second 64B can be used to submit a tx work
2274 			 * request with an implicit doorbell.
2275 			 */
2276 
2277 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2278 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2279 			if (rc == 0) {
2280 				clrbit(&sc->doorbells, DOORBELL_UDB);
2281 				setbit(&sc->doorbells, DOORBELL_WCWR);
2282 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2283 			} else {
2284 				t5_write_combine = 0;
2285 				device_printf(sc->dev,
2286 				    "couldn't enable write combining: %d\n",
2287 				    rc);
2288 			}
2289 
2290 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2291 			t4_write_reg(sc, A_SGE_STAT_CFG,
2292 			    V_STATSOURCE_T5(7) | mode);
2293 		}
2294 #else
2295 		t5_write_combine = 0;
2296 #endif
2297 		sc->iwt.wc_en = t5_write_combine;
2298 	}
2299 
2300 	return (0);
2301 }
2302 
2303 struct memwin_init {
2304 	uint32_t base;
2305 	uint32_t aperture;
2306 };
2307 
2308 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2309 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2310 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2311 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2312 };
2313 
2314 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2315 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2316 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2317 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2318 };
2319 
2320 static void
2321 setup_memwin(struct adapter *sc)
2322 {
2323 	const struct memwin_init *mw_init;
2324 	struct memwin *mw;
2325 	int i;
2326 	uint32_t bar0;
2327 
2328 	if (is_t4(sc)) {
2329 		/*
2330 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2331 		 * mechanism.  Works from within PCI passthrough environments
2332 		 * too, where rman_get_start() can return a different value.  We
2333 		 * need to program the T4 memory window decoders with the actual
2334 		 * addresses that will be coming across the PCIe link.
2335 		 */
2336 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2337 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2338 
2339 		mw_init = &t4_memwin[0];
2340 	} else {
2341 		/* T5+ use the relative offset inside the PCIe BAR */
2342 		bar0 = 0;
2343 
2344 		mw_init = &t5_memwin[0];
2345 	}
2346 
2347 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2348 		rw_init(&mw->mw_lock, "memory window access");
2349 		mw->mw_base = mw_init->base;
2350 		mw->mw_aperture = mw_init->aperture;
2351 		mw->mw_curpos = 0;
2352 		t4_write_reg(sc,
2353 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2354 		    (mw->mw_base + bar0) | V_BIR(0) |
2355 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2356 		rw_wlock(&mw->mw_lock);
2357 		position_memwin(sc, i, 0);
2358 		rw_wunlock(&mw->mw_lock);
2359 	}
2360 
2361 	/* flush */
2362 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2363 }
2364 
2365 /*
2366  * Positions the memory window at the given address in the card's address space.
2367  * There are some alignment requirements and the actual position may be at an
2368  * address prior to the requested address.  mw->mw_curpos always has the actual
2369  * position of the window.
2370  */
2371 static void
2372 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2373 {
2374 	struct memwin *mw;
2375 	uint32_t pf;
2376 	uint32_t reg;
2377 
2378 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2379 	mw = &sc->memwin[idx];
2380 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2381 
2382 	if (is_t4(sc)) {
2383 		pf = 0;
2384 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2385 	} else {
2386 		pf = V_PFNUM(sc->pf);
2387 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2388 	}
2389 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2390 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2391 	t4_read_reg(sc, reg);	/* flush */
2392 }
2393 
2394 static int
2395 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2396     int len, int rw)
2397 {
2398 	struct memwin *mw;
2399 	uint32_t mw_end, v;
2400 
2401 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2402 
2403 	/* Memory can only be accessed in naturally aligned 4 byte units */
2404 	if (addr & 3 || len & 3 || len <= 0)
2405 		return (EINVAL);
2406 
2407 	mw = &sc->memwin[idx];
2408 	while (len > 0) {
2409 		rw_rlock(&mw->mw_lock);
2410 		mw_end = mw->mw_curpos + mw->mw_aperture;
2411 		if (addr >= mw_end || addr < mw->mw_curpos) {
2412 			/* Will need to reposition the window */
2413 			if (!rw_try_upgrade(&mw->mw_lock)) {
2414 				rw_runlock(&mw->mw_lock);
2415 				rw_wlock(&mw->mw_lock);
2416 			}
2417 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2418 			position_memwin(sc, idx, addr);
2419 			rw_downgrade(&mw->mw_lock);
2420 			mw_end = mw->mw_curpos + mw->mw_aperture;
2421 		}
2422 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2423 		while (addr < mw_end && len > 0) {
2424 			if (rw == 0) {
2425 				v = t4_read_reg(sc, mw->mw_base + addr -
2426 				    mw->mw_curpos);
2427 				*val++ = le32toh(v);
2428 			} else {
2429 				v = *val++;
2430 				t4_write_reg(sc, mw->mw_base + addr -
2431 				    mw->mw_curpos, htole32(v));
2432 			}
2433 			addr += 4;
2434 			len -= 4;
2435 		}
2436 		rw_runlock(&mw->mw_lock);
2437 	}
2438 
2439 	return (0);
2440 }
2441 
2442 static inline int
2443 read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2444     int len)
2445 {
2446 
2447 	return (rw_via_memwin(sc, idx, addr, val, len, 0));
2448 }
2449 
2450 static inline int
2451 write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
2452     const uint32_t *val, int len)
2453 {
2454 
2455 	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
2456 }
2457 
2458 static int
2459 t4_range_cmp(const void *a, const void *b)
2460 {
2461 	return ((const struct t4_range *)a)->start -
2462 	       ((const struct t4_range *)b)->start;
2463 }
2464 
2465 /*
2466  * Verify that the memory range specified by the addr/len pair is valid within
2467  * the card's address space.
2468  */
2469 static int
2470 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2471 {
2472 	struct t4_range mem_ranges[4], *r, *next;
2473 	uint32_t em, addr_len;
2474 	int i, n, remaining;
2475 
2476 	/* Memory can only be accessed in naturally aligned 4 byte units */
2477 	if (addr & 3 || len & 3 || len <= 0)
2478 		return (EINVAL);
2479 
2480 	/* Enabled memories */
2481 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2482 
2483 	r = &mem_ranges[0];
2484 	n = 0;
2485 	bzero(r, sizeof(mem_ranges));
2486 	if (em & F_EDRAM0_ENABLE) {
2487 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2488 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2489 		if (r->size > 0) {
2490 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2491 			if (addr >= r->start &&
2492 			    addr + len <= r->start + r->size)
2493 				return (0);
2494 			r++;
2495 			n++;
2496 		}
2497 	}
2498 	if (em & F_EDRAM1_ENABLE) {
2499 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2500 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2501 		if (r->size > 0) {
2502 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2503 			if (addr >= r->start &&
2504 			    addr + len <= r->start + r->size)
2505 				return (0);
2506 			r++;
2507 			n++;
2508 		}
2509 	}
2510 	if (em & F_EXT_MEM_ENABLE) {
2511 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2512 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2513 		if (r->size > 0) {
2514 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2515 			if (addr >= r->start &&
2516 			    addr + len <= r->start + r->size)
2517 				return (0);
2518 			r++;
2519 			n++;
2520 		}
2521 	}
2522 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2523 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2524 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2525 		if (r->size > 0) {
2526 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2527 			if (addr >= r->start &&
2528 			    addr + len <= r->start + r->size)
2529 				return (0);
2530 			r++;
2531 			n++;
2532 		}
2533 	}
2534 	MPASS(n <= nitems(mem_ranges));
2535 
2536 	if (n > 1) {
2537 		/* Sort and merge the ranges. */
2538 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2539 
2540 		/* Start from index 0 and examine the next n - 1 entries. */
2541 		r = &mem_ranges[0];
2542 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2543 
2544 			MPASS(r->size > 0);	/* r is a valid entry. */
2545 			next = r + 1;
2546 			MPASS(next->size > 0);	/* and so is the next one. */
2547 
2548 			while (r->start + r->size >= next->start) {
2549 				/* Merge the next one into the current entry. */
2550 				r->size = max(r->start + r->size,
2551 				    next->start + next->size) - r->start;
2552 				n--;	/* One fewer entry in total. */
2553 				if (--remaining == 0)
2554 					goto done;	/* short circuit */
2555 				next++;
2556 			}
2557 			if (next != r + 1) {
2558 				/*
2559 				 * Some entries were merged into r and next
2560 				 * points to the first valid entry that couldn't
2561 				 * be merged.
2562 				 */
2563 				MPASS(next->size > 0);	/* must be valid */
2564 				memcpy(r + 1, next, remaining * sizeof(*r));
2565 #ifdef INVARIANTS
2566 				/*
2567 				 * This so that the foo->size assertion in the
2568 				 * next iteration of the loop do the right
2569 				 * thing for entries that were pulled up and are
2570 				 * no longer valid.
2571 				 */
2572 				MPASS(n < nitems(mem_ranges));
2573 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2574 				    sizeof(struct t4_range));
2575 #endif
2576 			}
2577 		}
2578 done:
2579 		/* Done merging the ranges. */
2580 		MPASS(n > 0);
2581 		r = &mem_ranges[0];
2582 		for (i = 0; i < n; i++, r++) {
2583 			if (addr >= r->start &&
2584 			    addr + len <= r->start + r->size)
2585 				return (0);
2586 		}
2587 	}
2588 
2589 	return (EFAULT);
2590 }
2591 
2592 static int
2593 fwmtype_to_hwmtype(int mtype)
2594 {
2595 
2596 	switch (mtype) {
2597 	case FW_MEMTYPE_EDC0:
2598 		return (MEM_EDC0);
2599 	case FW_MEMTYPE_EDC1:
2600 		return (MEM_EDC1);
2601 	case FW_MEMTYPE_EXTMEM:
2602 		return (MEM_MC0);
2603 	case FW_MEMTYPE_EXTMEM1:
2604 		return (MEM_MC1);
2605 	default:
2606 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2607 	}
2608 }
2609 
2610 /*
2611  * Verify that the memory range specified by the memtype/offset/len pair is
2612  * valid and lies entirely within the memtype specified.  The global address of
2613  * the start of the range is returned in addr.
2614  */
2615 static int
2616 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2617     uint32_t *addr)
2618 {
2619 	uint32_t em, addr_len, maddr;
2620 
2621 	/* Memory can only be accessed in naturally aligned 4 byte units */
2622 	if (off & 3 || len & 3 || len == 0)
2623 		return (EINVAL);
2624 
2625 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2626 	switch (fwmtype_to_hwmtype(mtype)) {
2627 	case MEM_EDC0:
2628 		if (!(em & F_EDRAM0_ENABLE))
2629 			return (EINVAL);
2630 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2631 		maddr = G_EDRAM0_BASE(addr_len) << 20;
2632 		break;
2633 	case MEM_EDC1:
2634 		if (!(em & F_EDRAM1_ENABLE))
2635 			return (EINVAL);
2636 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2637 		maddr = G_EDRAM1_BASE(addr_len) << 20;
2638 		break;
2639 	case MEM_MC:
2640 		if (!(em & F_EXT_MEM_ENABLE))
2641 			return (EINVAL);
2642 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2643 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2644 		break;
2645 	case MEM_MC1:
2646 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2647 			return (EINVAL);
2648 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2649 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2650 		break;
2651 	default:
2652 		return (EINVAL);
2653 	}
2654 
2655 	*addr = maddr + off;	/* global address */
2656 	return (validate_mem_range(sc, *addr, len));
2657 }
2658 
2659 static int
2660 fixup_devlog_params(struct adapter *sc)
2661 {
2662 	struct devlog_params *dparams = &sc->params.devlog;
2663 	int rc;
2664 
2665 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2666 	    dparams->size, &dparams->addr);
2667 
2668 	return (rc);
2669 }
2670 
2671 static void
2672 update_nirq(struct intrs_and_queues *iaq, int nports)
2673 {
2674 	int extra = T4_EXTRA_INTR;
2675 
2676 	iaq->nirq = extra;
2677 	iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
2678 	iaq->nirq += nports * (iaq->num_vis - 1) *
2679 	    max(iaq->nrxq_vi, iaq->nnmrxq_vi);
2680 	iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
2681 }
2682 
2683 /*
2684  * Adjust requirements to fit the number of interrupts available.
2685  */
2686 static void
2687 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
2688     int navail)
2689 {
2690 	int old_nirq;
2691 	const int nports = sc->params.nports;
2692 
2693 	MPASS(nports > 0);
2694 	MPASS(navail > 0);
2695 
2696 	bzero(iaq, sizeof(*iaq));
2697 	iaq->intr_type = itype;
2698 	iaq->num_vis = t4_num_vis;
2699 	iaq->ntxq = t4_ntxq;
2700 	iaq->ntxq_vi = t4_ntxq_vi;
2701 	iaq->nrxq = t4_nrxq;
2702 	iaq->nrxq_vi = t4_nrxq_vi;
2703 #ifdef TCP_OFFLOAD
2704 	if (is_offload(sc)) {
2705 		iaq->nofldtxq = t4_nofldtxq;
2706 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
2707 		iaq->nofldrxq = t4_nofldrxq;
2708 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
2709 	}
2710 #endif
2711 #ifdef DEV_NETMAP
2712 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
2713 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
2714 #endif
2715 
2716 	update_nirq(iaq, nports);
2717 	if (iaq->nirq <= navail &&
2718 	    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2719 		/*
2720 		 * This is the normal case -- there are enough interrupts for
2721 		 * everything.
2722 		 */
2723 		goto done;
2724 	}
2725 
2726 	/*
2727 	 * If extra VIs have been configured try reducing their count and see if
2728 	 * that works.
2729 	 */
2730 	while (iaq->num_vis > 1) {
2731 		iaq->num_vis--;
2732 		update_nirq(iaq, nports);
2733 		if (iaq->nirq <= navail &&
2734 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2735 			device_printf(sc->dev, "virtual interfaces per port "
2736 			    "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
2737 			    "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
2738 			    "itype %d, navail %u, nirq %d.\n",
2739 			    iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
2740 			    iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
2741 			    itype, navail, iaq->nirq);
2742 			goto done;
2743 		}
2744 	}
2745 
2746 	/*
2747 	 * Extra VIs will not be created.  Log a message if they were requested.
2748 	 */
2749 	MPASS(iaq->num_vis == 1);
2750 	iaq->ntxq_vi = iaq->nrxq_vi = 0;
2751 	iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
2752 	iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
2753 	if (iaq->num_vis != t4_num_vis) {
2754 		device_printf(sc->dev, "extra virtual interfaces disabled.  "
2755 		    "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
2756 		    "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
2757 		    iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
2758 		    iaq->nnmrxq_vi, itype, navail, iaq->nirq);
2759 	}
2760 
2761 	/*
2762 	 * Keep reducing the number of NIC rx queues to the next lower power of
2763 	 * 2 (for even RSS distribution) and halving the TOE rx queues and see
2764 	 * if that works.
2765 	 */
2766 	do {
2767 		if (iaq->nrxq > 1) {
2768 			do {
2769 				iaq->nrxq--;
2770 			} while (!powerof2(iaq->nrxq));
2771 		}
2772 		if (iaq->nofldrxq > 1)
2773 			iaq->nofldrxq >>= 1;
2774 
2775 		old_nirq = iaq->nirq;
2776 		update_nirq(iaq, nports);
2777 		if (iaq->nirq <= navail &&
2778 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2779 			device_printf(sc->dev, "running with reduced number of "
2780 			    "rx queues because of shortage of interrupts.  "
2781 			    "nrxq=%u, nofldrxq=%u.  "
2782 			    "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
2783 			    iaq->nofldrxq, itype, navail, iaq->nirq);
2784 			goto done;
2785 		}
2786 	} while (old_nirq != iaq->nirq);
2787 
2788 	/* One interrupt for everything.  Ugh. */
2789 	device_printf(sc->dev, "running with minimal number of queues.  "
2790 	    "itype %d, navail %u.\n", itype, navail);
2791 	iaq->nirq = 1;
2792 	MPASS(iaq->nrxq == 1);
2793 	iaq->ntxq = 1;
2794 	if (iaq->nofldrxq > 1)
2795 		iaq->nofldtxq = 1;
2796 done:
2797 	MPASS(iaq->num_vis > 0);
2798 	if (iaq->num_vis > 1) {
2799 		MPASS(iaq->nrxq_vi > 0);
2800 		MPASS(iaq->ntxq_vi > 0);
2801 	}
2802 	MPASS(iaq->nirq > 0);
2803 	MPASS(iaq->nrxq > 0);
2804 	MPASS(iaq->ntxq > 0);
2805 	if (itype == INTR_MSI) {
2806 		MPASS(powerof2(iaq->nirq));
2807 	}
2808 }
2809 
2810 static int
2811 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
2812 {
2813 	int rc, itype, navail, nalloc;
2814 
2815 	for (itype = INTR_MSIX; itype; itype >>= 1) {
2816 
2817 		if ((itype & t4_intr_types) == 0)
2818 			continue;	/* not allowed */
2819 
2820 		if (itype == INTR_MSIX)
2821 			navail = pci_msix_count(sc->dev);
2822 		else if (itype == INTR_MSI)
2823 			navail = pci_msi_count(sc->dev);
2824 		else
2825 			navail = 1;
2826 restart:
2827 		if (navail == 0)
2828 			continue;
2829 
2830 		calculate_iaq(sc, iaq, itype, navail);
2831 		nalloc = iaq->nirq;
2832 		rc = 0;
2833 		if (itype == INTR_MSIX)
2834 			rc = pci_alloc_msix(sc->dev, &nalloc);
2835 		else if (itype == INTR_MSI)
2836 			rc = pci_alloc_msi(sc->dev, &nalloc);
2837 
2838 		if (rc == 0 && nalloc > 0) {
2839 			if (nalloc == iaq->nirq)
2840 				return (0);
2841 
2842 			/*
2843 			 * Didn't get the number requested.  Use whatever number
2844 			 * the kernel is willing to allocate.
2845 			 */
2846 			device_printf(sc->dev, "fewer vectors than requested, "
2847 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2848 			    itype, iaq->nirq, nalloc);
2849 			pci_release_msi(sc->dev);
2850 			navail = nalloc;
2851 			goto restart;
2852 		}
2853 
2854 		device_printf(sc->dev,
2855 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2856 		    itype, rc, iaq->nirq, nalloc);
2857 	}
2858 
2859 	device_printf(sc->dev,
2860 	    "failed to find a usable interrupt type.  "
2861 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2862 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2863 
2864 	return (ENXIO);
2865 }
2866 
2867 #define FW_VERSION(chip) ( \
2868     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2869     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2870     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2871     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2872 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2873 
2874 struct fw_info {
2875 	uint8_t chip;
2876 	char *kld_name;
2877 	char *fw_mod_name;
2878 	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
2879 } fw_info[] = {
2880 	{
2881 		.chip = CHELSIO_T4,
2882 		.kld_name = "t4fw_cfg",
2883 		.fw_mod_name = "t4fw",
2884 		.fw_hdr = {
2885 			.chip = FW_HDR_CHIP_T4,
2886 			.fw_ver = htobe32_const(FW_VERSION(T4)),
2887 			.intfver_nic = FW_INTFVER(T4, NIC),
2888 			.intfver_vnic = FW_INTFVER(T4, VNIC),
2889 			.intfver_ofld = FW_INTFVER(T4, OFLD),
2890 			.intfver_ri = FW_INTFVER(T4, RI),
2891 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2892 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
2893 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2894 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
2895 		},
2896 	}, {
2897 		.chip = CHELSIO_T5,
2898 		.kld_name = "t5fw_cfg",
2899 		.fw_mod_name = "t5fw",
2900 		.fw_hdr = {
2901 			.chip = FW_HDR_CHIP_T5,
2902 			.fw_ver = htobe32_const(FW_VERSION(T5)),
2903 			.intfver_nic = FW_INTFVER(T5, NIC),
2904 			.intfver_vnic = FW_INTFVER(T5, VNIC),
2905 			.intfver_ofld = FW_INTFVER(T5, OFLD),
2906 			.intfver_ri = FW_INTFVER(T5, RI),
2907 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
2908 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
2909 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
2910 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
2911 		},
2912 	}, {
2913 		.chip = CHELSIO_T6,
2914 		.kld_name = "t6fw_cfg",
2915 		.fw_mod_name = "t6fw",
2916 		.fw_hdr = {
2917 			.chip = FW_HDR_CHIP_T6,
2918 			.fw_ver = htobe32_const(FW_VERSION(T6)),
2919 			.intfver_nic = FW_INTFVER(T6, NIC),
2920 			.intfver_vnic = FW_INTFVER(T6, VNIC),
2921 			.intfver_ofld = FW_INTFVER(T6, OFLD),
2922 			.intfver_ri = FW_INTFVER(T6, RI),
2923 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
2924 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
2925 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
2926 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
2927 		},
2928 	}
2929 };
2930 
2931 static struct fw_info *
2932 find_fw_info(int chip)
2933 {
2934 	int i;
2935 
2936 	for (i = 0; i < nitems(fw_info); i++) {
2937 		if (fw_info[i].chip == chip)
2938 			return (&fw_info[i]);
2939 	}
2940 	return (NULL);
2941 }
2942 
2943 /*
2944  * Is the given firmware API compatible with the one the driver was compiled
2945  * with?
2946  */
2947 static int
2948 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
2949 {
2950 
2951 	/* short circuit if it's the exact same firmware version */
2952 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
2953 		return (1);
2954 
2955 	/*
2956 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
2957 	 * features that are supported in the driver.
2958 	 */
2959 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
2960 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
2961 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
2962 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
2963 		return (1);
2964 #undef SAME_INTF
2965 
2966 	return (0);
2967 }
2968 
2969 /*
2970  * The firmware in the KLD is usable, but should it be installed?  This routine
2971  * explains itself in detail if it indicates the KLD firmware should be
2972  * installed.
2973  */
2974 static int
2975 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
2976 {
2977 	const char *reason;
2978 
2979 	if (!card_fw_usable) {
2980 		reason = "incompatible or unusable";
2981 		goto install;
2982 	}
2983 
2984 	if (k > c) {
2985 		reason = "older than the version bundled with this driver";
2986 		goto install;
2987 	}
2988 
2989 	if (t4_fw_install == 2 && k != c) {
2990 		reason = "different than the version bundled with this driver";
2991 		goto install;
2992 	}
2993 
2994 	return (0);
2995 
2996 install:
2997 	if (t4_fw_install == 0) {
2998 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2999 		    "but the driver is prohibited from installing a different "
3000 		    "firmware on the card.\n",
3001 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3002 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3003 
3004 		return (0);
3005 	}
3006 
3007 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3008 	    "installing firmware %u.%u.%u.%u on card.\n",
3009 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3010 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3011 	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3012 	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3013 
3014 	return (1);
3015 }
3016 
3017 /*
3018  * Establish contact with the firmware and determine if we are the master driver
3019  * or not, and whether we are responsible for chip initialization.
3020  */
3021 static int
3022 prep_firmware(struct adapter *sc)
3023 {
3024 	const struct firmware *fw = NULL, *default_cfg;
3025 	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
3026 	enum dev_state state;
3027 	struct fw_info *fw_info;
3028 	struct fw_hdr *card_fw;		/* fw on the card */
3029 	const struct fw_hdr *kld_fw;	/* fw in the KLD */
3030 	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
3031 					   against */
3032 
3033 	/* This is the firmware whose headers the driver was compiled against */
3034 	fw_info = find_fw_info(chip_id(sc));
3035 	if (fw_info == NULL) {
3036 		device_printf(sc->dev,
3037 		    "unable to look up firmware information for chip %d.\n",
3038 		    chip_id(sc));
3039 		return (EINVAL);
3040 	}
3041 	drv_fw = &fw_info->fw_hdr;
3042 
3043 	/*
3044 	 * The firmware KLD contains many modules.  The KLD name is also the
3045 	 * name of the module that contains the default config file.
3046 	 */
3047 	default_cfg = firmware_get(fw_info->kld_name);
3048 
3049 	/* This is the firmware in the KLD */
3050 	fw = firmware_get(fw_info->fw_mod_name);
3051 	if (fw != NULL) {
3052 		kld_fw = (const void *)fw->data;
3053 		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3054 	} else {
3055 		kld_fw = NULL;
3056 		kld_fw_usable = 0;
3057 	}
3058 
3059 	/* Read the header of the firmware on the card */
3060 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3061 	rc = -t4_read_flash(sc, FLASH_FW_START,
3062 	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
3063 	if (rc == 0) {
3064 		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
3065 		if (card_fw->fw_ver == be32toh(0xffffffff)) {
3066 			uint32_t d = be32toh(kld_fw->fw_ver);
3067 
3068 			if (!kld_fw_usable) {
3069 				device_printf(sc->dev,
3070 				    "no firmware on the card and no usable "
3071 				    "firmware bundled with the driver.\n");
3072 				rc = EIO;
3073 				goto done;
3074 			} else if (t4_fw_install == 0) {
3075 				device_printf(sc->dev,
3076 				    "no firmware on the card and the driver "
3077 				    "is prohibited from installing new "
3078 				    "firmware.\n");
3079 				rc = EIO;
3080 				goto done;
3081 			}
3082 
3083 			device_printf(sc->dev, "no firmware on the card, "
3084 			    "installing firmware %d.%d.%d.%d\n",
3085 			    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3086 			    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3087 			rc = t4_fw_forceinstall(sc, fw->data, fw->datasize);
3088 			if (rc < 0) {
3089 				rc = -rc;
3090 				device_printf(sc->dev,
3091 				    "firmware install failed: %d.\n", rc);
3092 				goto done;
3093 			}
3094 			memcpy(card_fw, kld_fw, sizeof(*card_fw));
3095 			card_fw_usable = 1;
3096 			need_fw_reset = 0;
3097 		}
3098 	} else {
3099 		device_printf(sc->dev,
3100 		    "Unable to read card's firmware header: %d\n", rc);
3101 		card_fw_usable = 0;
3102 	}
3103 
3104 	/* Contact firmware. */
3105 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3106 	if (rc < 0 || state == DEV_STATE_ERR) {
3107 		rc = -rc;
3108 		device_printf(sc->dev,
3109 		    "failed to connect to the firmware: %d, %d.\n", rc, state);
3110 		goto done;
3111 	}
3112 	pf = rc;
3113 	if (pf == sc->mbox)
3114 		sc->flags |= MASTER_PF;
3115 	else if (state == DEV_STATE_UNINIT) {
3116 		/*
3117 		 * We didn't get to be the master so we definitely won't be
3118 		 * configuring the chip.  It's a bug if someone else hasn't
3119 		 * configured it already.
3120 		 */
3121 		device_printf(sc->dev, "couldn't be master(%d), "
3122 		    "device not already initialized either(%d).\n", rc, state);
3123 		rc = EPROTO;
3124 		goto done;
3125 	}
3126 
3127 	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3128 	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3129 		/*
3130 		 * Common case: the firmware on the card is an exact match and
3131 		 * the KLD is an exact match too, or the KLD is
3132 		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
3133 		 * here -- use cxgbetool loadfw if you want to reinstall the
3134 		 * same firmware as the one on the card.
3135 		 */
3136 	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3137 	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3138 	    be32toh(card_fw->fw_ver))) {
3139 
3140 		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3141 		if (rc != 0) {
3142 			device_printf(sc->dev,
3143 			    "failed to install firmware: %d\n", rc);
3144 			goto done;
3145 		}
3146 
3147 		/* Installed successfully, update the cached header too. */
3148 		memcpy(card_fw, kld_fw, sizeof(*card_fw));
3149 		card_fw_usable = 1;
3150 		need_fw_reset = 0;	/* already reset as part of load_fw */
3151 	}
3152 
3153 	if (!card_fw_usable) {
3154 		uint32_t d, c, k;
3155 
3156 		d = ntohl(drv_fw->fw_ver);
3157 		c = ntohl(card_fw->fw_ver);
3158 		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3159 
3160 		device_printf(sc->dev, "Cannot find a usable firmware: "
3161 		    "fw_install %d, chip state %d, "
3162 		    "driver compiled with %d.%d.%d.%d, "
3163 		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3164 		    t4_fw_install, state,
3165 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3166 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3167 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3168 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3169 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3170 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3171 		rc = EINVAL;
3172 		goto done;
3173 	}
3174 
3175 	/* Reset device */
3176 	if (need_fw_reset &&
3177 	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3178 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3179 		if (rc != ETIMEDOUT && rc != EIO)
3180 			t4_fw_bye(sc, sc->mbox);
3181 		goto done;
3182 	}
3183 	sc->flags |= FW_OK;
3184 
3185 	rc = get_params__pre_init(sc);
3186 	if (rc != 0)
3187 		goto done; /* error message displayed already */
3188 
3189 	/* Partition adapter resources as specified in the config file. */
3190 	if (state == DEV_STATE_UNINIT) {
3191 
3192 		KASSERT(sc->flags & MASTER_PF,
3193 		    ("%s: trying to change chip settings when not master.",
3194 		    __func__));
3195 
3196 		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3197 		if (rc != 0)
3198 			goto done;	/* error message displayed already */
3199 
3200 		t4_tweak_chip_settings(sc);
3201 
3202 		/* get basic stuff going */
3203 		rc = -t4_fw_initialize(sc, sc->mbox);
3204 		if (rc != 0) {
3205 			device_printf(sc->dev, "fw init failed: %d.\n", rc);
3206 			goto done;
3207 		}
3208 	} else {
3209 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3210 		sc->cfcsum = 0;
3211 	}
3212 
3213 done:
3214 	free(card_fw, M_CXGBE);
3215 	if (fw != NULL)
3216 		firmware_put(fw, FIRMWARE_UNLOAD);
3217 	if (default_cfg != NULL)
3218 		firmware_put(default_cfg, FIRMWARE_UNLOAD);
3219 
3220 	return (rc);
3221 }
3222 
3223 #define FW_PARAM_DEV(param) \
3224 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3225 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3226 #define FW_PARAM_PFVF(param) \
3227 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3228 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3229 
3230 /*
3231  * Partition chip resources for use between various PFs, VFs, etc.
3232  */
3233 static int
3234 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3235     const char *name_prefix)
3236 {
3237 	const struct firmware *cfg = NULL;
3238 	int rc = 0;
3239 	struct fw_caps_config_cmd caps;
3240 	uint32_t mtype, moff, finicsum, cfcsum;
3241 
3242 	/*
3243 	 * Figure out what configuration file to use.  Pick the default config
3244 	 * file for the card if the user hasn't specified one explicitly.
3245 	 */
3246 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3247 	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3248 		/* Card specific overrides go here. */
3249 		if (pci_get_device(sc->dev) == 0x440a)
3250 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3251 		if (is_fpga(sc))
3252 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3253 	}
3254 
3255 	/*
3256 	 * We need to load another module if the profile is anything except
3257 	 * "default" or "flash".
3258 	 */
3259 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3260 	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3261 		char s[32];
3262 
3263 		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3264 		cfg = firmware_get(s);
3265 		if (cfg == NULL) {
3266 			if (default_cfg != NULL) {
3267 				device_printf(sc->dev,
3268 				    "unable to load module \"%s\" for "
3269 				    "configuration profile \"%s\", will use "
3270 				    "the default config file instead.\n",
3271 				    s, sc->cfg_file);
3272 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3273 				    "%s", DEFAULT_CF);
3274 			} else {
3275 				device_printf(sc->dev,
3276 				    "unable to load module \"%s\" for "
3277 				    "configuration profile \"%s\", will use "
3278 				    "the config file on the card's flash "
3279 				    "instead.\n", s, sc->cfg_file);
3280 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3281 				    "%s", FLASH_CF);
3282 			}
3283 		}
3284 	}
3285 
3286 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3287 	    default_cfg == NULL) {
3288 		device_printf(sc->dev,
3289 		    "default config file not available, will use the config "
3290 		    "file on the card's flash instead.\n");
3291 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3292 	}
3293 
3294 	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3295 		u_int cflen;
3296 		const uint32_t *cfdata;
3297 		uint32_t param, val, addr;
3298 
3299 		KASSERT(cfg != NULL || default_cfg != NULL,
3300 		    ("%s: no config to upload", __func__));
3301 
3302 		/*
3303 		 * Ask the firmware where it wants us to upload the config file.
3304 		 */
3305 		param = FW_PARAM_DEV(CF);
3306 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3307 		if (rc != 0) {
3308 			/* No support for config file?  Shouldn't happen. */
3309 			device_printf(sc->dev,
3310 			    "failed to query config file location: %d.\n", rc);
3311 			goto done;
3312 		}
3313 		mtype = G_FW_PARAMS_PARAM_Y(val);
3314 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3315 
3316 		/*
3317 		 * XXX: sheer laziness.  We deliberately added 4 bytes of
3318 		 * useless stuffing/comments at the end of the config file so
3319 		 * it's ok to simply throw away the last remaining bytes when
3320 		 * the config file is not an exact multiple of 4.  This also
3321 		 * helps with the validate_mt_off_len check.
3322 		 */
3323 		if (cfg != NULL) {
3324 			cflen = cfg->datasize & ~3;
3325 			cfdata = cfg->data;
3326 		} else {
3327 			cflen = default_cfg->datasize & ~3;
3328 			cfdata = default_cfg->data;
3329 		}
3330 
3331 		if (cflen > FLASH_CFG_MAX_SIZE) {
3332 			device_printf(sc->dev,
3333 			    "config file too long (%d, max allowed is %d).  "
3334 			    "Will try to use the config on the card, if any.\n",
3335 			    cflen, FLASH_CFG_MAX_SIZE);
3336 			goto use_config_on_flash;
3337 		}
3338 
3339 		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3340 		if (rc != 0) {
3341 			device_printf(sc->dev,
3342 			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3343 			    "Will try to use the config on the card, if any.\n",
3344 			    __func__, mtype, moff, cflen, rc);
3345 			goto use_config_on_flash;
3346 		}
3347 		write_via_memwin(sc, 2, addr, cfdata, cflen);
3348 	} else {
3349 use_config_on_flash:
3350 		mtype = FW_MEMTYPE_FLASH;
3351 		moff = t4_flash_cfg_addr(sc);
3352 	}
3353 
3354 	bzero(&caps, sizeof(caps));
3355 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3356 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3357 	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3358 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3359 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3360 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3361 	if (rc != 0) {
3362 		device_printf(sc->dev,
3363 		    "failed to pre-process config file: %d "
3364 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3365 		goto done;
3366 	}
3367 
3368 	finicsum = be32toh(caps.finicsum);
3369 	cfcsum = be32toh(caps.cfcsum);
3370 	if (finicsum != cfcsum) {
3371 		device_printf(sc->dev,
3372 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3373 		    finicsum, cfcsum);
3374 	}
3375 	sc->cfcsum = cfcsum;
3376 
3377 #define LIMIT_CAPS(x) do { \
3378 	caps.x &= htobe16(t4_##x##_allowed); \
3379 } while (0)
3380 
3381 	/*
3382 	 * Let the firmware know what features will (not) be used so it can tune
3383 	 * things accordingly.
3384 	 */
3385 	LIMIT_CAPS(nbmcaps);
3386 	LIMIT_CAPS(linkcaps);
3387 	LIMIT_CAPS(switchcaps);
3388 	LIMIT_CAPS(niccaps);
3389 	LIMIT_CAPS(toecaps);
3390 	LIMIT_CAPS(rdmacaps);
3391 	LIMIT_CAPS(cryptocaps);
3392 	LIMIT_CAPS(iscsicaps);
3393 	LIMIT_CAPS(fcoecaps);
3394 #undef LIMIT_CAPS
3395 
3396 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3397 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3398 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3399 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3400 	if (rc != 0) {
3401 		device_printf(sc->dev,
3402 		    "failed to process config file: %d.\n", rc);
3403 	}
3404 done:
3405 	if (cfg != NULL)
3406 		firmware_put(cfg, FIRMWARE_UNLOAD);
3407 	return (rc);
3408 }
3409 
3410 /*
3411  * Retrieve parameters that are needed (or nice to have) very early.
3412  */
3413 static int
3414 get_params__pre_init(struct adapter *sc)
3415 {
3416 	int rc;
3417 	uint32_t param[2], val[2];
3418 
3419 	t4_get_version_info(sc);
3420 
3421 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3422 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3423 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3424 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3425 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3426 
3427 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3428 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3429 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3430 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3431 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3432 
3433 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3434 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3435 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3436 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3437 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3438 
3439 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3440 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3441 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3442 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3443 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3444 
3445 	param[0] = FW_PARAM_DEV(PORTVEC);
3446 	param[1] = FW_PARAM_DEV(CCLK);
3447 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3448 	if (rc != 0) {
3449 		device_printf(sc->dev,
3450 		    "failed to query parameters (pre_init): %d.\n", rc);
3451 		return (rc);
3452 	}
3453 
3454 	sc->params.portvec = val[0];
3455 	sc->params.nports = bitcount32(val[0]);
3456 	sc->params.vpd.cclk = val[1];
3457 
3458 	/* Read device log parameters. */
3459 	rc = -t4_init_devlog_params(sc, 1);
3460 	if (rc == 0)
3461 		fixup_devlog_params(sc);
3462 	else {
3463 		device_printf(sc->dev,
3464 		    "failed to get devlog parameters: %d.\n", rc);
3465 		rc = 0;	/* devlog isn't critical for device operation */
3466 	}
3467 
3468 	return (rc);
3469 }
3470 
3471 /*
3472  * Retrieve various parameters that are of interest to the driver.  The device
3473  * has been initialized by the firmware at this point.
3474  */
3475 static int
3476 get_params__post_init(struct adapter *sc)
3477 {
3478 	int rc;
3479 	uint32_t param[7], val[7];
3480 	struct fw_caps_config_cmd caps;
3481 
3482 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3483 	param[1] = FW_PARAM_PFVF(EQ_START);
3484 	param[2] = FW_PARAM_PFVF(FILTER_START);
3485 	param[3] = FW_PARAM_PFVF(FILTER_END);
3486 	param[4] = FW_PARAM_PFVF(L2T_START);
3487 	param[5] = FW_PARAM_PFVF(L2T_END);
3488 	param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
3489 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
3490 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
3491 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
3492 	if (rc != 0) {
3493 		device_printf(sc->dev,
3494 		    "failed to query parameters (post_init): %d.\n", rc);
3495 		return (rc);
3496 	}
3497 
3498 	sc->sge.iq_start = val[0];
3499 	sc->sge.eq_start = val[1];
3500 	sc->tids.ftid_base = val[2];
3501 	sc->tids.nftids = val[3] - val[2] + 1;
3502 	sc->params.ftid_min = val[2];
3503 	sc->params.ftid_max = val[3];
3504 	sc->vres.l2t.start = val[4];
3505 	sc->vres.l2t.size = val[5] - val[4] + 1;
3506 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3507 	    ("%s: L2 table size (%u) larger than expected (%u)",
3508 	    __func__, sc->vres.l2t.size, L2T_SIZE));
3509 	sc->params.core_vdd = val[6];
3510 
3511 	/*
3512 	 * MPSBGMAP is queried separately because only recent firmwares support
3513 	 * it as a parameter and we don't want the compound query above to fail
3514 	 * on older firmwares.
3515 	 */
3516 	param[0] = FW_PARAM_DEV(MPSBGMAP);
3517 	val[0] = 0;
3518 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3519 	if (rc == 0)
3520 		sc->params.mps_bg_map = val[0];
3521 	else
3522 		sc->params.mps_bg_map = 0;
3523 
3524 	/* get capabilites */
3525 	bzero(&caps, sizeof(caps));
3526 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3527 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3528 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3529 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3530 	if (rc != 0) {
3531 		device_printf(sc->dev,
3532 		    "failed to get card capabilities: %d.\n", rc);
3533 		return (rc);
3534 	}
3535 
3536 #define READ_CAPS(x) do { \
3537 	sc->x = htobe16(caps.x); \
3538 } while (0)
3539 	READ_CAPS(nbmcaps);
3540 	READ_CAPS(linkcaps);
3541 	READ_CAPS(switchcaps);
3542 	READ_CAPS(niccaps);
3543 	READ_CAPS(toecaps);
3544 	READ_CAPS(rdmacaps);
3545 	READ_CAPS(cryptocaps);
3546 	READ_CAPS(iscsicaps);
3547 	READ_CAPS(fcoecaps);
3548 
3549 	/*
3550 	 * The firmware attempts memfree TOE configuration for -SO cards and
3551 	 * will report toecaps=0 if it runs out of resources (this depends on
3552 	 * the config file).  It may not report 0 for other capabilities
3553 	 * dependent on the TOE in this case.  Set them to 0 here so that the
3554 	 * driver doesn't bother tracking resources that will never be used.
3555 	 */
3556 	if (sc->toecaps == 0) {
3557 		sc->iscsicaps = 0;
3558 		sc->rdmacaps = 0;
3559 	}
3560 
3561 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3562 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3563 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3564 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3565 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3566 		if (rc != 0) {
3567 			device_printf(sc->dev,
3568 			    "failed to query NIC parameters: %d.\n", rc);
3569 			return (rc);
3570 		}
3571 		sc->tids.etid_base = val[0];
3572 		sc->params.etid_min = val[0];
3573 		sc->tids.netids = val[1] - val[0] + 1;
3574 		sc->params.netids = sc->tids.netids;
3575 		sc->params.eo_wr_cred = val[2];
3576 		sc->params.ethoffload = 1;
3577 	}
3578 
3579 	if (sc->toecaps) {
3580 		/* query offload-related parameters */
3581 		param[0] = FW_PARAM_DEV(NTID);
3582 		param[1] = FW_PARAM_PFVF(SERVER_START);
3583 		param[2] = FW_PARAM_PFVF(SERVER_END);
3584 		param[3] = FW_PARAM_PFVF(TDDP_START);
3585 		param[4] = FW_PARAM_PFVF(TDDP_END);
3586 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3587 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3588 		if (rc != 0) {
3589 			device_printf(sc->dev,
3590 			    "failed to query TOE parameters: %d.\n", rc);
3591 			return (rc);
3592 		}
3593 		sc->tids.ntids = val[0];
3594 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3595 		sc->tids.stid_base = val[1];
3596 		sc->tids.nstids = val[2] - val[1] + 1;
3597 		sc->vres.ddp.start = val[3];
3598 		sc->vres.ddp.size = val[4] - val[3] + 1;
3599 		sc->params.ofldq_wr_cred = val[5];
3600 		sc->params.offload = 1;
3601 	}
3602 	if (sc->rdmacaps) {
3603 		param[0] = FW_PARAM_PFVF(STAG_START);
3604 		param[1] = FW_PARAM_PFVF(STAG_END);
3605 		param[2] = FW_PARAM_PFVF(RQ_START);
3606 		param[3] = FW_PARAM_PFVF(RQ_END);
3607 		param[4] = FW_PARAM_PFVF(PBL_START);
3608 		param[5] = FW_PARAM_PFVF(PBL_END);
3609 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3610 		if (rc != 0) {
3611 			device_printf(sc->dev,
3612 			    "failed to query RDMA parameters(1): %d.\n", rc);
3613 			return (rc);
3614 		}
3615 		sc->vres.stag.start = val[0];
3616 		sc->vres.stag.size = val[1] - val[0] + 1;
3617 		sc->vres.rq.start = val[2];
3618 		sc->vres.rq.size = val[3] - val[2] + 1;
3619 		sc->vres.pbl.start = val[4];
3620 		sc->vres.pbl.size = val[5] - val[4] + 1;
3621 
3622 		param[0] = FW_PARAM_PFVF(SQRQ_START);
3623 		param[1] = FW_PARAM_PFVF(SQRQ_END);
3624 		param[2] = FW_PARAM_PFVF(CQ_START);
3625 		param[3] = FW_PARAM_PFVF(CQ_END);
3626 		param[4] = FW_PARAM_PFVF(OCQ_START);
3627 		param[5] = FW_PARAM_PFVF(OCQ_END);
3628 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3629 		if (rc != 0) {
3630 			device_printf(sc->dev,
3631 			    "failed to query RDMA parameters(2): %d.\n", rc);
3632 			return (rc);
3633 		}
3634 		sc->vres.qp.start = val[0];
3635 		sc->vres.qp.size = val[1] - val[0] + 1;
3636 		sc->vres.cq.start = val[2];
3637 		sc->vres.cq.size = val[3] - val[2] + 1;
3638 		sc->vres.ocq.start = val[4];
3639 		sc->vres.ocq.size = val[5] - val[4] + 1;
3640 
3641 		param[0] = FW_PARAM_PFVF(SRQ_START);
3642 		param[1] = FW_PARAM_PFVF(SRQ_END);
3643 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
3644 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
3645 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
3646 		if (rc != 0) {
3647 			device_printf(sc->dev,
3648 			    "failed to query RDMA parameters(3): %d.\n", rc);
3649 			return (rc);
3650 		}
3651 		sc->vres.srq.start = val[0];
3652 		sc->vres.srq.size = val[1] - val[0] + 1;
3653 		sc->params.max_ordird_qp = val[2];
3654 		sc->params.max_ird_adapter = val[3];
3655 	}
3656 	if (sc->iscsicaps) {
3657 		param[0] = FW_PARAM_PFVF(ISCSI_START);
3658 		param[1] = FW_PARAM_PFVF(ISCSI_END);
3659 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3660 		if (rc != 0) {
3661 			device_printf(sc->dev,
3662 			    "failed to query iSCSI parameters: %d.\n", rc);
3663 			return (rc);
3664 		}
3665 		sc->vres.iscsi.start = val[0];
3666 		sc->vres.iscsi.size = val[1] - val[0] + 1;
3667 	}
3668 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
3669 		param[0] = FW_PARAM_PFVF(TLS_START);
3670 		param[1] = FW_PARAM_PFVF(TLS_END);
3671 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3672 		if (rc != 0) {
3673 			device_printf(sc->dev,
3674 			    "failed to query TLS parameters: %d.\n", rc);
3675 			return (rc);
3676 		}
3677 		sc->vres.key.start = val[0];
3678 		sc->vres.key.size = val[1] - val[0] + 1;
3679 	}
3680 
3681 	t4_init_sge_params(sc);
3682 
3683 	/*
3684 	 * We've got the params we wanted to query via the firmware.  Now grab
3685 	 * some others directly from the chip.
3686 	 */
3687 	rc = t4_read_chip_settings(sc);
3688 
3689 	return (rc);
3690 }
3691 
3692 static int
3693 set_params__post_init(struct adapter *sc)
3694 {
3695 	uint32_t param, val;
3696 #ifdef TCP_OFFLOAD
3697 	int i, v, shift;
3698 #endif
3699 
3700 	/* ask for encapsulated CPLs */
3701 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3702 	val = 1;
3703 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3704 
3705 #ifdef TCP_OFFLOAD
3706 	/*
3707 	 * Override the TOE timers with user provided tunables.  This is not the
3708 	 * recommended way to change the timers (the firmware config file is) so
3709 	 * these tunables are not documented.
3710 	 *
3711 	 * All the timer tunables are in microseconds.
3712 	 */
3713 	if (t4_toe_keepalive_idle != 0) {
3714 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
3715 		v &= M_KEEPALIVEIDLE;
3716 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
3717 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
3718 	}
3719 	if (t4_toe_keepalive_interval != 0) {
3720 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
3721 		v &= M_KEEPALIVEINTVL;
3722 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
3723 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
3724 	}
3725 	if (t4_toe_keepalive_count != 0) {
3726 		v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
3727 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
3728 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
3729 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
3730 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
3731 	}
3732 	if (t4_toe_rexmt_min != 0) {
3733 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
3734 		v &= M_RXTMIN;
3735 		t4_set_reg_field(sc, A_TP_RXT_MIN,
3736 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
3737 	}
3738 	if (t4_toe_rexmt_max != 0) {
3739 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
3740 		v &= M_RXTMAX;
3741 		t4_set_reg_field(sc, A_TP_RXT_MAX,
3742 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
3743 	}
3744 	if (t4_toe_rexmt_count != 0) {
3745 		v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
3746 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
3747 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
3748 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
3749 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
3750 	}
3751 	for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
3752 		if (t4_toe_rexmt_backoff[i] != -1) {
3753 			v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
3754 			shift = (i & 3) << 3;
3755 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
3756 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
3757 		}
3758 	}
3759 #endif
3760 	return (0);
3761 }
3762 
3763 #undef FW_PARAM_PFVF
3764 #undef FW_PARAM_DEV
3765 
3766 static void
3767 t4_set_desc(struct adapter *sc)
3768 {
3769 	char buf[128];
3770 	struct adapter_params *p = &sc->params;
3771 
3772 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
3773 
3774 	device_set_desc_copy(sc->dev, buf);
3775 }
3776 
3777 static void
3778 build_medialist(struct port_info *pi, struct ifmedia *media)
3779 {
3780 	int m;
3781 
3782 	PORT_LOCK_ASSERT_OWNED(pi);
3783 
3784 	ifmedia_removeall(media);
3785 
3786 	/*
3787 	 * XXX: Would it be better to ifmedia_add all 4 combinations of pause
3788 	 * settings for every speed instead of just txpause|rxpause?  ifconfig
3789 	 * media display looks much better if autoselect is the only case where
3790 	 * ifm_current is different from ifm_active.  If the user picks anything
3791 	 * except txpause|rxpause the display is ugly.
3792 	 */
3793 	m = IFM_ETHER | IFM_FDX | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
3794 
3795 	switch(pi->port_type) {
3796 	case FW_PORT_TYPE_BT_XFI:
3797 	case FW_PORT_TYPE_BT_XAUI:
3798 		ifmedia_add(media, m | IFM_10G_T, 0, NULL);
3799 		/* fall through */
3800 
3801 	case FW_PORT_TYPE_BT_SGMII:
3802 		ifmedia_add(media, m | IFM_1000_T, 0, NULL);
3803 		ifmedia_add(media, m | IFM_100_TX, 0, NULL);
3804 		ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL);
3805 		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
3806 		break;
3807 
3808 	case FW_PORT_TYPE_CX4:
3809 		ifmedia_add(media, m | IFM_10G_CX4, 0, NULL);
3810 		ifmedia_set(media, m | IFM_10G_CX4);
3811 		break;
3812 
3813 	case FW_PORT_TYPE_QSFP_10G:
3814 	case FW_PORT_TYPE_SFP:
3815 	case FW_PORT_TYPE_FIBER_XFI:
3816 	case FW_PORT_TYPE_FIBER_XAUI:
3817 		switch (pi->mod_type) {
3818 
3819 		case FW_PORT_MOD_TYPE_LR:
3820 			ifmedia_add(media, m | IFM_10G_LR, 0, NULL);
3821 			ifmedia_set(media, m | IFM_10G_LR);
3822 			break;
3823 
3824 		case FW_PORT_MOD_TYPE_SR:
3825 			ifmedia_add(media, m | IFM_10G_SR, 0, NULL);
3826 			ifmedia_set(media, m | IFM_10G_SR);
3827 			break;
3828 
3829 		case FW_PORT_MOD_TYPE_LRM:
3830 			ifmedia_add(media, m | IFM_10G_LRM, 0, NULL);
3831 			ifmedia_set(media, m | IFM_10G_LRM);
3832 			break;
3833 
3834 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3835 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3836 			ifmedia_add(media, m | IFM_10G_TWINAX, 0, NULL);
3837 			ifmedia_set(media, m | IFM_10G_TWINAX);
3838 			break;
3839 
3840 		case FW_PORT_MOD_TYPE_NONE:
3841 			m &= ~IFM_FDX;
3842 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3843 			ifmedia_set(media, m | IFM_NONE);
3844 			break;
3845 
3846 		case FW_PORT_MOD_TYPE_NA:
3847 		case FW_PORT_MOD_TYPE_ER:
3848 		default:
3849 			device_printf(pi->dev,
3850 			    "unknown port_type (%d), mod_type (%d)\n",
3851 			    pi->port_type, pi->mod_type);
3852 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3853 			ifmedia_set(media, m | IFM_UNKNOWN);
3854 			break;
3855 		}
3856 		break;
3857 
3858 	case FW_PORT_TYPE_CR_QSFP:
3859 	case FW_PORT_TYPE_SFP28:
3860 	case FW_PORT_TYPE_KR_SFP28:
3861 		switch (pi->mod_type) {
3862 
3863 		case FW_PORT_MOD_TYPE_SR:
3864 			ifmedia_add(media, m | IFM_25G_SR, 0, NULL);
3865 			ifmedia_set(media, m | IFM_25G_SR);
3866 			break;
3867 
3868 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3869 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3870 			ifmedia_add(media, m | IFM_25G_CR, 0, NULL);
3871 			ifmedia_set(media, m | IFM_25G_CR);
3872 			break;
3873 
3874 		case FW_PORT_MOD_TYPE_NONE:
3875 			m &= ~IFM_FDX;
3876 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3877 			ifmedia_set(media, m | IFM_NONE);
3878 			break;
3879 
3880 		default:
3881 			device_printf(pi->dev,
3882 			    "unknown port_type (%d), mod_type (%d)\n",
3883 			    pi->port_type, pi->mod_type);
3884 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3885 			ifmedia_set(media, m | IFM_UNKNOWN);
3886 			break;
3887 		}
3888 		break;
3889 
3890 	case FW_PORT_TYPE_QSFP:
3891 		switch (pi->mod_type) {
3892 
3893 		case FW_PORT_MOD_TYPE_LR:
3894 			ifmedia_add(media, m | IFM_40G_LR4, 0, NULL);
3895 			ifmedia_set(media, m | IFM_40G_LR4);
3896 			break;
3897 
3898 		case FW_PORT_MOD_TYPE_SR:
3899 			ifmedia_add(media, m | IFM_40G_SR4, 0, NULL);
3900 			ifmedia_set(media, m | IFM_40G_SR4);
3901 			break;
3902 
3903 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3904 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3905 			ifmedia_add(media, m | IFM_40G_CR4, 0, NULL);
3906 			ifmedia_set(media, m | IFM_40G_CR4);
3907 			break;
3908 
3909 		case FW_PORT_MOD_TYPE_NONE:
3910 			m &= ~IFM_FDX;
3911 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3912 			ifmedia_set(media, m | IFM_NONE);
3913 			break;
3914 
3915 		default:
3916 			device_printf(pi->dev,
3917 			    "unknown port_type (%d), mod_type (%d)\n",
3918 			    pi->port_type, pi->mod_type);
3919 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3920 			ifmedia_set(media, m | IFM_UNKNOWN);
3921 			break;
3922 		}
3923 		break;
3924 
3925 	case FW_PORT_TYPE_KR4_100G:
3926 	case FW_PORT_TYPE_CR4_QSFP:
3927 		switch (pi->mod_type) {
3928 
3929 		case FW_PORT_MOD_TYPE_LR:
3930 			ifmedia_add(media, m | IFM_100G_LR4, 0, NULL);
3931 			ifmedia_set(media, m | IFM_100G_LR4);
3932 			break;
3933 
3934 		case FW_PORT_MOD_TYPE_SR:
3935 			ifmedia_add(media, m | IFM_100G_SR4, 0, NULL);
3936 			ifmedia_set(media, m | IFM_100G_SR4);
3937 			break;
3938 
3939 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3940 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3941 			ifmedia_add(media, m | IFM_100G_CR4, 0, NULL);
3942 			ifmedia_set(media, m | IFM_100G_CR4);
3943 			break;
3944 
3945 		case FW_PORT_MOD_TYPE_NONE:
3946 			m &= ~IFM_FDX;
3947 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3948 			ifmedia_set(media, m | IFM_NONE);
3949 			break;
3950 
3951 		default:
3952 			device_printf(pi->dev,
3953 			    "unknown port_type (%d), mod_type (%d)\n",
3954 			    pi->port_type, pi->mod_type);
3955 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3956 			ifmedia_set(media, m | IFM_UNKNOWN);
3957 			break;
3958 		}
3959 		break;
3960 
3961 	default:
3962 		device_printf(pi->dev,
3963 		    "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
3964 		    pi->mod_type);
3965 		ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3966 		ifmedia_set(media, m | IFM_UNKNOWN);
3967 		break;
3968 	}
3969 }
3970 
3971 /*
3972  * Update all the requested_* fields in the link config and then send a mailbox
3973  * command to apply the settings.
3974  */
3975 static void
3976 init_l1cfg(struct port_info *pi)
3977 {
3978 	struct adapter *sc = pi->adapter;
3979 	struct link_config *lc = &pi->link_cfg;
3980 	int rc;
3981 
3982 	ASSERT_SYNCHRONIZED_OP(sc);
3983 
3984 	if (t4_autoneg != 0 && lc->supported & FW_PORT_CAP_ANEG) {
3985 		lc->requested_aneg = AUTONEG_ENABLE;
3986 		lc->requested_speed = 0;
3987 	} else {
3988 		lc->requested_aneg = AUTONEG_DISABLE;
3989 		lc->requested_speed = port_top_speed(pi);	/* in Gbps */
3990 	}
3991 
3992 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX);
3993 
3994 	if (t4_fec != -1) {
3995 		lc->requested_fec = t4_fec & (FEC_RS | FEC_BASER_RS |
3996 		    FEC_RESERVED);
3997 	} else {
3998 		/* Use the suggested value provided by the firmware in acaps */
3999 		if (lc->advertising & FW_PORT_CAP_FEC_RS)
4000 			lc->requested_fec = FEC_RS;
4001 		else if (lc->advertising & FW_PORT_CAP_FEC_BASER_RS)
4002 			lc->requested_fec = FEC_BASER_RS;
4003 		else if (lc->advertising & FW_PORT_CAP_FEC_RESERVED)
4004 			lc->requested_fec = FEC_RESERVED;
4005 		else
4006 			lc->requested_fec = 0;
4007 	}
4008 
4009 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
4010 	if (rc != 0) {
4011 		device_printf(pi->dev, "l1cfg failed: %d\n", rc);
4012 	} else {
4013 		lc->fc = lc->requested_fc;
4014 		lc->fec = lc->requested_fec;
4015 	}
4016 }
4017 
4018 #define FW_MAC_EXACT_CHUNK	7
4019 
4020 /*
4021  * Program the port's XGMAC based on parameters in ifnet.  The caller also
4022  * indicates which parameters should be programmed (the rest are left alone).
4023  */
4024 int
4025 update_mac_settings(struct ifnet *ifp, int flags)
4026 {
4027 	int rc = 0;
4028 	struct vi_info *vi = ifp->if_softc;
4029 	struct port_info *pi = vi->pi;
4030 	struct adapter *sc = pi->adapter;
4031 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4032 
4033 	ASSERT_SYNCHRONIZED_OP(sc);
4034 	KASSERT(flags, ("%s: not told what to update.", __func__));
4035 
4036 	if (flags & XGMAC_MTU)
4037 		mtu = ifp->if_mtu;
4038 
4039 	if (flags & XGMAC_PROMISC)
4040 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4041 
4042 	if (flags & XGMAC_ALLMULTI)
4043 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4044 
4045 	if (flags & XGMAC_VLANEX)
4046 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4047 
4048 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4049 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4050 		    allmulti, 1, vlanex, false);
4051 		if (rc) {
4052 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4053 			    rc);
4054 			return (rc);
4055 		}
4056 	}
4057 
4058 	if (flags & XGMAC_UCADDR) {
4059 		uint8_t ucaddr[ETHER_ADDR_LEN];
4060 
4061 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4062 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4063 		    ucaddr, true, true);
4064 		if (rc < 0) {
4065 			rc = -rc;
4066 			if_printf(ifp, "change_mac failed: %d\n", rc);
4067 			return (rc);
4068 		} else {
4069 			vi->xact_addr_filt = rc;
4070 			rc = 0;
4071 		}
4072 	}
4073 
4074 	if (flags & XGMAC_MCADDRS) {
4075 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4076 		int del = 1;
4077 		uint64_t hash = 0;
4078 		struct ifmultiaddr *ifma;
4079 		int i = 0, j;
4080 
4081 		if_maddr_rlock(ifp);
4082 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4083 			if (ifma->ifma_addr->sa_family != AF_LINK)
4084 				continue;
4085 			mcaddr[i] =
4086 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4087 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4088 			i++;
4089 
4090 			if (i == FW_MAC_EXACT_CHUNK) {
4091 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4092 				    del, i, mcaddr, NULL, &hash, 0);
4093 				if (rc < 0) {
4094 					rc = -rc;
4095 					for (j = 0; j < i; j++) {
4096 						if_printf(ifp,
4097 						    "failed to add mc address"
4098 						    " %02x:%02x:%02x:"
4099 						    "%02x:%02x:%02x rc=%d\n",
4100 						    mcaddr[j][0], mcaddr[j][1],
4101 						    mcaddr[j][2], mcaddr[j][3],
4102 						    mcaddr[j][4], mcaddr[j][5],
4103 						    rc);
4104 					}
4105 					goto mcfail;
4106 				}
4107 				del = 0;
4108 				i = 0;
4109 			}
4110 		}
4111 		if (i > 0) {
4112 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4113 			    mcaddr, NULL, &hash, 0);
4114 			if (rc < 0) {
4115 				rc = -rc;
4116 				for (j = 0; j < i; j++) {
4117 					if_printf(ifp,
4118 					    "failed to add mc address"
4119 					    " %02x:%02x:%02x:"
4120 					    "%02x:%02x:%02x rc=%d\n",
4121 					    mcaddr[j][0], mcaddr[j][1],
4122 					    mcaddr[j][2], mcaddr[j][3],
4123 					    mcaddr[j][4], mcaddr[j][5],
4124 					    rc);
4125 				}
4126 				goto mcfail;
4127 			}
4128 		}
4129 
4130 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4131 		if (rc != 0)
4132 			if_printf(ifp, "failed to set mc address hash: %d", rc);
4133 mcfail:
4134 		if_maddr_runlock(ifp);
4135 	}
4136 
4137 	return (rc);
4138 }
4139 
4140 /*
4141  * {begin|end}_synchronized_op must be called from the same thread.
4142  */
4143 int
4144 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4145     char *wmesg)
4146 {
4147 	int rc, pri;
4148 
4149 #ifdef WITNESS
4150 	/* the caller thinks it's ok to sleep, but is it really? */
4151 	if (flags & SLEEP_OK)
4152 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4153 		    "begin_synchronized_op");
4154 #endif
4155 
4156 	if (INTR_OK)
4157 		pri = PCATCH;
4158 	else
4159 		pri = 0;
4160 
4161 	ADAPTER_LOCK(sc);
4162 	for (;;) {
4163 
4164 		if (vi && IS_DOOMED(vi)) {
4165 			rc = ENXIO;
4166 			goto done;
4167 		}
4168 
4169 		if (!IS_BUSY(sc)) {
4170 			rc = 0;
4171 			break;
4172 		}
4173 
4174 		if (!(flags & SLEEP_OK)) {
4175 			rc = EBUSY;
4176 			goto done;
4177 		}
4178 
4179 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4180 			rc = EINTR;
4181 			goto done;
4182 		}
4183 	}
4184 
4185 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4186 	SET_BUSY(sc);
4187 #ifdef INVARIANTS
4188 	sc->last_op = wmesg;
4189 	sc->last_op_thr = curthread;
4190 	sc->last_op_flags = flags;
4191 #endif
4192 
4193 done:
4194 	if (!(flags & HOLD_LOCK) || rc)
4195 		ADAPTER_UNLOCK(sc);
4196 
4197 	return (rc);
4198 }
4199 
4200 /*
4201  * Tell if_ioctl and if_init that the VI is going away.  This is
4202  * special variant of begin_synchronized_op and must be paired with a
4203  * call to end_synchronized_op.
4204  */
4205 void
4206 doom_vi(struct adapter *sc, struct vi_info *vi)
4207 {
4208 
4209 	ADAPTER_LOCK(sc);
4210 	SET_DOOMED(vi);
4211 	wakeup(&sc->flags);
4212 	while (IS_BUSY(sc))
4213 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4214 	SET_BUSY(sc);
4215 #ifdef INVARIANTS
4216 	sc->last_op = "t4detach";
4217 	sc->last_op_thr = curthread;
4218 	sc->last_op_flags = 0;
4219 #endif
4220 	ADAPTER_UNLOCK(sc);
4221 }
4222 
4223 /*
4224  * {begin|end}_synchronized_op must be called from the same thread.
4225  */
4226 void
4227 end_synchronized_op(struct adapter *sc, int flags)
4228 {
4229 
4230 	if (flags & LOCK_HELD)
4231 		ADAPTER_LOCK_ASSERT_OWNED(sc);
4232 	else
4233 		ADAPTER_LOCK(sc);
4234 
4235 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
4236 	CLR_BUSY(sc);
4237 	wakeup(&sc->flags);
4238 	ADAPTER_UNLOCK(sc);
4239 }
4240 
4241 static int
4242 cxgbe_init_synchronized(struct vi_info *vi)
4243 {
4244 	struct port_info *pi = vi->pi;
4245 	struct adapter *sc = pi->adapter;
4246 	struct ifnet *ifp = vi->ifp;
4247 	int rc = 0, i;
4248 	struct sge_txq *txq;
4249 
4250 	ASSERT_SYNCHRONIZED_OP(sc);
4251 
4252 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
4253 		return (0);	/* already running */
4254 
4255 	if (!(sc->flags & FULL_INIT_DONE) &&
4256 	    ((rc = adapter_full_init(sc)) != 0))
4257 		return (rc);	/* error message displayed already */
4258 
4259 	if (!(vi->flags & VI_INIT_DONE) &&
4260 	    ((rc = vi_full_init(vi)) != 0))
4261 		return (rc); /* error message displayed already */
4262 
4263 	rc = update_mac_settings(ifp, XGMAC_ALL);
4264 	if (rc)
4265 		goto done;	/* error message displayed already */
4266 
4267 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
4268 	if (rc != 0) {
4269 		if_printf(ifp, "enable_vi failed: %d\n", rc);
4270 		goto done;
4271 	}
4272 
4273 	/*
4274 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
4275 	 * if this changes.
4276 	 */
4277 
4278 	for_each_txq(vi, i, txq) {
4279 		TXQ_LOCK(txq);
4280 		txq->eq.flags |= EQ_ENABLED;
4281 		TXQ_UNLOCK(txq);
4282 	}
4283 
4284 	/*
4285 	 * The first iq of the first port to come up is used for tracing.
4286 	 */
4287 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4288 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4289 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4290 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4291 		    V_QUEUENUMBER(sc->traceq));
4292 		pi->flags |= HAS_TRACEQ;
4293 	}
4294 
4295 	/* all ok */
4296 	PORT_LOCK(pi);
4297 	if (pi->up_vis++ == 0) {
4298 		t4_update_port_info(pi);
4299 		build_medialist(pi, &pi->media);
4300 		init_l1cfg(pi);
4301 	}
4302 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
4303 
4304 	if (pi->nvi > 1 || sc->flags & IS_VF)
4305 		callout_reset(&vi->tick, hz, vi_tick, vi);
4306 	else
4307 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4308 	PORT_UNLOCK(pi);
4309 done:
4310 	if (rc != 0)
4311 		cxgbe_uninit_synchronized(vi);
4312 
4313 	return (rc);
4314 }
4315 
4316 /*
4317  * Idempotent.
4318  */
4319 static int
4320 cxgbe_uninit_synchronized(struct vi_info *vi)
4321 {
4322 	struct port_info *pi = vi->pi;
4323 	struct adapter *sc = pi->adapter;
4324 	struct ifnet *ifp = vi->ifp;
4325 	int rc, i;
4326 	struct sge_txq *txq;
4327 
4328 	ASSERT_SYNCHRONIZED_OP(sc);
4329 
4330 	if (!(vi->flags & VI_INIT_DONE)) {
4331 		KASSERT(!(ifp->if_drv_flags & IFF_DRV_RUNNING),
4332 		    ("uninited VI is running"));
4333 		return (0);
4334 	}
4335 
4336 	/*
4337 	 * Disable the VI so that all its data in either direction is discarded
4338 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4339 	 * tick) intact as the TP can deliver negative advice or data that it's
4340 	 * holding in its RAM (for an offloaded connection) even after the VI is
4341 	 * disabled.
4342 	 */
4343 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4344 	if (rc) {
4345 		if_printf(ifp, "disable_vi failed: %d\n", rc);
4346 		return (rc);
4347 	}
4348 
4349 	for_each_txq(vi, i, txq) {
4350 		TXQ_LOCK(txq);
4351 		txq->eq.flags &= ~EQ_ENABLED;
4352 		TXQ_UNLOCK(txq);
4353 	}
4354 
4355 	PORT_LOCK(pi);
4356 	if (pi->nvi > 1 || sc->flags & IS_VF)
4357 		callout_stop(&vi->tick);
4358 	else
4359 		callout_stop(&pi->tick);
4360 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4361 		PORT_UNLOCK(pi);
4362 		return (0);
4363 	}
4364 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4365 	pi->up_vis--;
4366 	if (pi->up_vis > 0) {
4367 		PORT_UNLOCK(pi);
4368 		return (0);
4369 	}
4370 	PORT_UNLOCK(pi);
4371 
4372 	pi->link_cfg.link_ok = 0;
4373 	pi->link_cfg.speed = 0;
4374 	pi->link_cfg.link_down_rc = 255;
4375 	t4_os_link_changed(pi);
4376 	pi->old_link_cfg = pi->link_cfg;
4377 
4378 	return (0);
4379 }
4380 
4381 /*
4382  * It is ok for this function to fail midway and return right away.  t4_detach
4383  * will walk the entire sc->irq list and clean up whatever is valid.
4384  */
4385 int
4386 t4_setup_intr_handlers(struct adapter *sc)
4387 {
4388 	int rc, rid, p, q, v;
4389 	char s[8];
4390 	struct irq *irq;
4391 	struct port_info *pi;
4392 	struct vi_info *vi;
4393 	struct sge *sge = &sc->sge;
4394 	struct sge_rxq *rxq;
4395 #ifdef TCP_OFFLOAD
4396 	struct sge_ofld_rxq *ofld_rxq;
4397 #endif
4398 #ifdef DEV_NETMAP
4399 	struct sge_nm_rxq *nm_rxq;
4400 #endif
4401 #ifdef RSS
4402 	int nbuckets = rss_getnumbuckets();
4403 #endif
4404 
4405 	/*
4406 	 * Setup interrupts.
4407 	 */
4408 	irq = &sc->irq[0];
4409 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
4410 	if (forwarding_intr_to_fwq(sc))
4411 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4412 
4413 	/* Multiple interrupts. */
4414 	if (sc->flags & IS_VF)
4415 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4416 		    ("%s: too few intr.", __func__));
4417 	else
4418 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4419 		    ("%s: too few intr.", __func__));
4420 
4421 	/* The first one is always error intr on PFs */
4422 	if (!(sc->flags & IS_VF)) {
4423 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4424 		if (rc != 0)
4425 			return (rc);
4426 		irq++;
4427 		rid++;
4428 	}
4429 
4430 	/* The second one is always the firmware event queue (first on VFs) */
4431 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4432 	if (rc != 0)
4433 		return (rc);
4434 	irq++;
4435 	rid++;
4436 
4437 	for_each_port(sc, p) {
4438 		pi = sc->port[p];
4439 		for_each_vi(pi, v, vi) {
4440 			vi->first_intr = rid - 1;
4441 
4442 			if (vi->nnmrxq > 0) {
4443 				int n = max(vi->nrxq, vi->nnmrxq);
4444 
4445 				rxq = &sge->rxq[vi->first_rxq];
4446 #ifdef DEV_NETMAP
4447 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4448 #endif
4449 				for (q = 0; q < n; q++) {
4450 					snprintf(s, sizeof(s), "%x%c%x", p,
4451 					    'a' + v, q);
4452 					if (q < vi->nrxq)
4453 						irq->rxq = rxq++;
4454 #ifdef DEV_NETMAP
4455 					if (q < vi->nnmrxq)
4456 						irq->nm_rxq = nm_rxq++;
4457 #endif
4458 					rc = t4_alloc_irq(sc, irq, rid,
4459 					    t4_vi_intr, irq, s);
4460 					if (rc != 0)
4461 						return (rc);
4462 #ifdef RSS
4463 					if (q < vi->nrxq) {
4464 						bus_bind_intr(sc->dev, irq->res,
4465 						    rss_getcpu(q % nbuckets));
4466 					}
4467 #endif
4468 					irq++;
4469 					rid++;
4470 					vi->nintr++;
4471 				}
4472 			} else {
4473 				for_each_rxq(vi, q, rxq) {
4474 					snprintf(s, sizeof(s), "%x%c%x", p,
4475 					    'a' + v, q);
4476 					rc = t4_alloc_irq(sc, irq, rid,
4477 					    t4_intr, rxq, s);
4478 					if (rc != 0)
4479 						return (rc);
4480 #ifdef RSS
4481 					bus_bind_intr(sc->dev, irq->res,
4482 					    rss_getcpu(q % nbuckets));
4483 #endif
4484 					irq++;
4485 					rid++;
4486 					vi->nintr++;
4487 				}
4488 			}
4489 #ifdef TCP_OFFLOAD
4490 			for_each_ofld_rxq(vi, q, ofld_rxq) {
4491 				snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
4492 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
4493 				    ofld_rxq, s);
4494 				if (rc != 0)
4495 					return (rc);
4496 				irq++;
4497 				rid++;
4498 				vi->nintr++;
4499 			}
4500 #endif
4501 		}
4502 	}
4503 	MPASS(irq == &sc->irq[sc->intr_count]);
4504 
4505 	return (0);
4506 }
4507 
4508 int
4509 adapter_full_init(struct adapter *sc)
4510 {
4511 	int rc, i;
4512 #ifdef RSS
4513 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4514 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4515 #endif
4516 
4517 	ASSERT_SYNCHRONIZED_OP(sc);
4518 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4519 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
4520 	    ("%s: FULL_INIT_DONE already", __func__));
4521 
4522 	/*
4523 	 * queues that belong to the adapter (not any particular port).
4524 	 */
4525 	rc = t4_setup_adapter_queues(sc);
4526 	if (rc != 0)
4527 		goto done;
4528 
4529 	for (i = 0; i < nitems(sc->tq); i++) {
4530 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
4531 		    taskqueue_thread_enqueue, &sc->tq[i]);
4532 		if (sc->tq[i] == NULL) {
4533 			device_printf(sc->dev,
4534 			    "failed to allocate task queue %d\n", i);
4535 			rc = ENOMEM;
4536 			goto done;
4537 		}
4538 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
4539 		    device_get_nameunit(sc->dev), i);
4540 	}
4541 #ifdef RSS
4542 	MPASS(RSS_KEYSIZE == 40);
4543 	rss_getkey((void *)&raw_rss_key[0]);
4544 	for (i = 0; i < nitems(rss_key); i++) {
4545 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
4546 	}
4547 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
4548 #endif
4549 
4550 	if (!(sc->flags & IS_VF))
4551 		t4_intr_enable(sc);
4552 	sc->flags |= FULL_INIT_DONE;
4553 done:
4554 	if (rc != 0)
4555 		adapter_full_uninit(sc);
4556 
4557 	return (rc);
4558 }
4559 
4560 int
4561 adapter_full_uninit(struct adapter *sc)
4562 {
4563 	int i;
4564 
4565 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4566 
4567 	t4_teardown_adapter_queues(sc);
4568 
4569 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
4570 		taskqueue_free(sc->tq[i]);
4571 		sc->tq[i] = NULL;
4572 	}
4573 
4574 	sc->flags &= ~FULL_INIT_DONE;
4575 
4576 	return (0);
4577 }
4578 
4579 #ifdef RSS
4580 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
4581     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
4582     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
4583     RSS_HASHTYPE_RSS_UDP_IPV6)
4584 
4585 /* Translates kernel hash types to hardware. */
4586 static int
4587 hashconfig_to_hashen(int hashconfig)
4588 {
4589 	int hashen = 0;
4590 
4591 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
4592 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
4593 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
4594 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
4595 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
4596 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4597 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4598 	}
4599 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
4600 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4601 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4602 	}
4603 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
4604 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4605 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
4606 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4607 
4608 	return (hashen);
4609 }
4610 
4611 /* Translates hardware hash types to kernel. */
4612 static int
4613 hashen_to_hashconfig(int hashen)
4614 {
4615 	int hashconfig = 0;
4616 
4617 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
4618 		/*
4619 		 * If UDP hashing was enabled it must have been enabled for
4620 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
4621 		 * enabling any 4-tuple hash is nonsense configuration.
4622 		 */
4623 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4624 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
4625 
4626 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4627 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
4628 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4629 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
4630 	}
4631 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4632 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
4633 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4634 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
4635 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
4636 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
4637 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
4638 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
4639 
4640 	return (hashconfig);
4641 }
4642 #endif
4643 
4644 int
4645 vi_full_init(struct vi_info *vi)
4646 {
4647 	struct adapter *sc = vi->pi->adapter;
4648 	struct ifnet *ifp = vi->ifp;
4649 	uint16_t *rss;
4650 	struct sge_rxq *rxq;
4651 	int rc, i, j, hashen;
4652 #ifdef RSS
4653 	int nbuckets = rss_getnumbuckets();
4654 	int hashconfig = rss_gethashconfig();
4655 	int extra;
4656 #endif
4657 
4658 	ASSERT_SYNCHRONIZED_OP(sc);
4659 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
4660 	    ("%s: VI_INIT_DONE already", __func__));
4661 
4662 	sysctl_ctx_init(&vi->ctx);
4663 	vi->flags |= VI_SYSCTL_CTX;
4664 
4665 	/*
4666 	 * Allocate tx/rx/fl queues for this VI.
4667 	 */
4668 	rc = t4_setup_vi_queues(vi);
4669 	if (rc != 0)
4670 		goto done;	/* error message displayed already */
4671 
4672 	/*
4673 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
4674 	 */
4675 	if (vi->nrxq > vi->rss_size) {
4676 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
4677 		    "some queues will never receive traffic.\n", vi->nrxq,
4678 		    vi->rss_size);
4679 	} else if (vi->rss_size % vi->nrxq) {
4680 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
4681 		    "expect uneven traffic distribution.\n", vi->nrxq,
4682 		    vi->rss_size);
4683 	}
4684 #ifdef RSS
4685 	if (vi->nrxq != nbuckets) {
4686 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
4687 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
4688 	}
4689 #endif
4690 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
4691 	for (i = 0; i < vi->rss_size;) {
4692 #ifdef RSS
4693 		j = rss_get_indirection_to_bucket(i);
4694 		j %= vi->nrxq;
4695 		rxq = &sc->sge.rxq[vi->first_rxq + j];
4696 		rss[i++] = rxq->iq.abs_id;
4697 #else
4698 		for_each_rxq(vi, j, rxq) {
4699 			rss[i++] = rxq->iq.abs_id;
4700 			if (i == vi->rss_size)
4701 				break;
4702 		}
4703 #endif
4704 	}
4705 
4706 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
4707 	    vi->rss_size);
4708 	if (rc != 0) {
4709 		if_printf(ifp, "rss_config failed: %d\n", rc);
4710 		goto done;
4711 	}
4712 
4713 #ifdef RSS
4714 	hashen = hashconfig_to_hashen(hashconfig);
4715 
4716 	/*
4717 	 * We may have had to enable some hashes even though the global config
4718 	 * wants them disabled.  This is a potential problem that must be
4719 	 * reported to the user.
4720 	 */
4721 	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
4722 
4723 	/*
4724 	 * If we consider only the supported hash types, then the enabled hashes
4725 	 * are a superset of the requested hashes.  In other words, there cannot
4726 	 * be any supported hash that was requested but not enabled, but there
4727 	 * can be hashes that were not requested but had to be enabled.
4728 	 */
4729 	extra &= SUPPORTED_RSS_HASHTYPES;
4730 	MPASS((extra & hashconfig) == 0);
4731 
4732 	if (extra) {
4733 		if_printf(ifp,
4734 		    "global RSS config (0x%x) cannot be accommodated.\n",
4735 		    hashconfig);
4736 	}
4737 	if (extra & RSS_HASHTYPE_RSS_IPV4)
4738 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
4739 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
4740 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
4741 	if (extra & RSS_HASHTYPE_RSS_IPV6)
4742 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
4743 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
4744 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
4745 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
4746 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
4747 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
4748 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
4749 #else
4750 	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
4751 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
4752 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4753 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
4754 #endif
4755 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
4756 	if (rc != 0) {
4757 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
4758 		goto done;
4759 	}
4760 
4761 	vi->rss = rss;
4762 	vi->flags |= VI_INIT_DONE;
4763 done:
4764 	if (rc != 0)
4765 		vi_full_uninit(vi);
4766 
4767 	return (rc);
4768 }
4769 
4770 /*
4771  * Idempotent.
4772  */
4773 int
4774 vi_full_uninit(struct vi_info *vi)
4775 {
4776 	struct port_info *pi = vi->pi;
4777 	struct adapter *sc = pi->adapter;
4778 	int i;
4779 	struct sge_rxq *rxq;
4780 	struct sge_txq *txq;
4781 #ifdef TCP_OFFLOAD
4782 	struct sge_ofld_rxq *ofld_rxq;
4783 	struct sge_wrq *ofld_txq;
4784 #endif
4785 
4786 	if (vi->flags & VI_INIT_DONE) {
4787 
4788 		/* Need to quiesce queues.  */
4789 
4790 		/* XXX: Only for the first VI? */
4791 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
4792 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
4793 
4794 		for_each_txq(vi, i, txq) {
4795 			quiesce_txq(sc, txq);
4796 		}
4797 
4798 #ifdef TCP_OFFLOAD
4799 		for_each_ofld_txq(vi, i, ofld_txq) {
4800 			quiesce_wrq(sc, ofld_txq);
4801 		}
4802 #endif
4803 
4804 		for_each_rxq(vi, i, rxq) {
4805 			quiesce_iq(sc, &rxq->iq);
4806 			quiesce_fl(sc, &rxq->fl);
4807 		}
4808 
4809 #ifdef TCP_OFFLOAD
4810 		for_each_ofld_rxq(vi, i, ofld_rxq) {
4811 			quiesce_iq(sc, &ofld_rxq->iq);
4812 			quiesce_fl(sc, &ofld_rxq->fl);
4813 		}
4814 #endif
4815 		free(vi->rss, M_CXGBE);
4816 		free(vi->nm_rss, M_CXGBE);
4817 	}
4818 
4819 	t4_teardown_vi_queues(vi);
4820 	vi->flags &= ~VI_INIT_DONE;
4821 
4822 	return (0);
4823 }
4824 
4825 static void
4826 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
4827 {
4828 	struct sge_eq *eq = &txq->eq;
4829 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
4830 
4831 	(void) sc;	/* unused */
4832 
4833 #ifdef INVARIANTS
4834 	TXQ_LOCK(txq);
4835 	MPASS((eq->flags & EQ_ENABLED) == 0);
4836 	TXQ_UNLOCK(txq);
4837 #endif
4838 
4839 	/* Wait for the mp_ring to empty. */
4840 	while (!mp_ring_is_idle(txq->r)) {
4841 		mp_ring_check_drainage(txq->r, 0);
4842 		pause("rquiesce", 1);
4843 	}
4844 
4845 	/* Then wait for the hardware to finish. */
4846 	while (spg->cidx != htobe16(eq->pidx))
4847 		pause("equiesce", 1);
4848 
4849 	/* Finally, wait for the driver to reclaim all descriptors. */
4850 	while (eq->cidx != eq->pidx)
4851 		pause("dquiesce", 1);
4852 }
4853 
4854 static void
4855 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
4856 {
4857 
4858 	/* XXXTX */
4859 }
4860 
4861 static void
4862 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
4863 {
4864 	(void) sc;	/* unused */
4865 
4866 	/* Synchronize with the interrupt handler */
4867 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
4868 		pause("iqfree", 1);
4869 }
4870 
4871 static void
4872 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
4873 {
4874 	mtx_lock(&sc->sfl_lock);
4875 	FL_LOCK(fl);
4876 	fl->flags |= FL_DOOMED;
4877 	FL_UNLOCK(fl);
4878 	callout_stop(&sc->sfl_callout);
4879 	mtx_unlock(&sc->sfl_lock);
4880 
4881 	KASSERT((fl->flags & FL_STARVING) == 0,
4882 	    ("%s: still starving", __func__));
4883 }
4884 
4885 static int
4886 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
4887     driver_intr_t *handler, void *arg, char *name)
4888 {
4889 	int rc;
4890 
4891 	irq->rid = rid;
4892 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
4893 	    RF_SHAREABLE | RF_ACTIVE);
4894 	if (irq->res == NULL) {
4895 		device_printf(sc->dev,
4896 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
4897 		return (ENOMEM);
4898 	}
4899 
4900 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
4901 	    NULL, handler, arg, &irq->tag);
4902 	if (rc != 0) {
4903 		device_printf(sc->dev,
4904 		    "failed to setup interrupt for rid %d, name %s: %d\n",
4905 		    rid, name, rc);
4906 	} else if (name)
4907 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
4908 
4909 	return (rc);
4910 }
4911 
4912 static int
4913 t4_free_irq(struct adapter *sc, struct irq *irq)
4914 {
4915 	if (irq->tag)
4916 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
4917 	if (irq->res)
4918 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
4919 
4920 	bzero(irq, sizeof(*irq));
4921 
4922 	return (0);
4923 }
4924 
4925 static void
4926 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
4927 {
4928 
4929 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
4930 	t4_get_regs(sc, buf, regs->len);
4931 }
4932 
4933 #define	A_PL_INDIR_CMD	0x1f8
4934 
4935 #define	S_PL_AUTOINC	31
4936 #define	M_PL_AUTOINC	0x1U
4937 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
4938 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
4939 
4940 #define	S_PL_VFID	20
4941 #define	M_PL_VFID	0xffU
4942 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
4943 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
4944 
4945 #define	S_PL_ADDR	0
4946 #define	M_PL_ADDR	0xfffffU
4947 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
4948 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
4949 
4950 #define	A_PL_INDIR_DATA	0x1fc
4951 
4952 static uint64_t
4953 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
4954 {
4955 	u32 stats[2];
4956 
4957 	mtx_assert(&sc->reg_lock, MA_OWNED);
4958 	if (sc->flags & IS_VF) {
4959 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
4960 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
4961 	} else {
4962 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4963 		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4964 		    V_PL_ADDR(VF_MPS_REG(reg)));
4965 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
4966 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
4967 	}
4968 	return (((uint64_t)stats[1]) << 32 | stats[0]);
4969 }
4970 
4971 static void
4972 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
4973     struct fw_vi_stats_vf *stats)
4974 {
4975 
4976 #define GET_STAT(name) \
4977 	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
4978 
4979 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
4980 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
4981 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
4982 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
4983 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
4984 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
4985 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
4986 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
4987 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
4988 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
4989 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
4990 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
4991 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
4992 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
4993 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
4994 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
4995 
4996 #undef GET_STAT
4997 }
4998 
4999 static void
5000 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
5001 {
5002 	int reg;
5003 
5004 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5005 	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5006 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
5007 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
5008 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
5009 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
5010 }
5011 
5012 static void
5013 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
5014 {
5015 	struct timeval tv;
5016 	const struct timeval interval = {0, 250000};	/* 250ms */
5017 
5018 	if (!(vi->flags & VI_INIT_DONE))
5019 		return;
5020 
5021 	getmicrotime(&tv);
5022 	timevalsub(&tv, &interval);
5023 	if (timevalcmp(&tv, &vi->last_refreshed, <))
5024 		return;
5025 
5026 	mtx_lock(&sc->reg_lock);
5027 	t4_get_vi_stats(sc, vi->viid, &vi->stats);
5028 	getmicrotime(&vi->last_refreshed);
5029 	mtx_unlock(&sc->reg_lock);
5030 }
5031 
5032 static void
5033 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5034 {
5035 	u_int i, v, tnl_cong_drops, bg_map;
5036 	struct timeval tv;
5037 	const struct timeval interval = {0, 250000};	/* 250ms */
5038 
5039 	getmicrotime(&tv);
5040 	timevalsub(&tv, &interval);
5041 	if (timevalcmp(&tv, &pi->last_refreshed, <))
5042 		return;
5043 
5044 	tnl_cong_drops = 0;
5045 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5046 	bg_map = pi->mps_bg_map;
5047 	while (bg_map) {
5048 		i = ffs(bg_map) - 1;
5049 		mtx_lock(&sc->reg_lock);
5050 		t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5051 		    A_TP_MIB_TNL_CNG_DROP_0 + i);
5052 		mtx_unlock(&sc->reg_lock);
5053 		tnl_cong_drops += v;
5054 		bg_map &= ~(1 << i);
5055 	}
5056 	pi->tnl_cong_drops = tnl_cong_drops;
5057 	getmicrotime(&pi->last_refreshed);
5058 }
5059 
5060 static void
5061 cxgbe_tick(void *arg)
5062 {
5063 	struct port_info *pi = arg;
5064 	struct adapter *sc = pi->adapter;
5065 
5066 	PORT_LOCK_ASSERT_OWNED(pi);
5067 	cxgbe_refresh_stats(sc, pi);
5068 
5069 	callout_schedule(&pi->tick, hz);
5070 }
5071 
5072 void
5073 vi_tick(void *arg)
5074 {
5075 	struct vi_info *vi = arg;
5076 	struct adapter *sc = vi->pi->adapter;
5077 
5078 	vi_refresh_stats(sc, vi);
5079 
5080 	callout_schedule(&vi->tick, hz);
5081 }
5082 
5083 static void
5084 cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
5085 {
5086 	struct ifnet *vlan;
5087 
5088 	if (arg != ifp || ifp->if_type != IFT_ETHER)
5089 		return;
5090 
5091 	vlan = VLAN_DEVAT(ifp, vid);
5092 	VLAN_SETCOOKIE(vlan, ifp);
5093 }
5094 
5095 /*
5096  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5097  */
5098 static char *caps_decoder[] = {
5099 	"\20\001IPMI\002NCSI",				/* 0: NBM */
5100 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
5101 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
5102 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
5103 	    "\006HASHFILTER\007ETHOFLD",
5104 	"\20\001TOE",					/* 4: TOE */
5105 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
5106 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
5107 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5108 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5109 	    "\007T10DIF"
5110 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5111 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
5112 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
5113 		    "\004PO_INITIATOR\005PO_TARGET",
5114 };
5115 
5116 void
5117 t4_sysctls(struct adapter *sc)
5118 {
5119 	struct sysctl_ctx_list *ctx;
5120 	struct sysctl_oid *oid;
5121 	struct sysctl_oid_list *children, *c0;
5122 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5123 
5124 	ctx = device_get_sysctl_ctx(sc->dev);
5125 
5126 	/*
5127 	 * dev.t4nex.X.
5128 	 */
5129 	oid = device_get_sysctl_tree(sc->dev);
5130 	c0 = children = SYSCTL_CHILDREN(oid);
5131 
5132 	sc->sc_do_rxcopy = 1;
5133 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5134 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5135 
5136 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5137 	    sc->params.nports, "# of ports");
5138 
5139 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5140 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
5141 	    sysctl_bitfield, "A", "available doorbells");
5142 
5143 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5144 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
5145 
5146 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5147 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5148 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5149 	    "interrupt holdoff timer values (us)");
5150 
5151 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5152 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5153 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5154 	    "interrupt holdoff packet counter values");
5155 
5156 	t4_sge_sysctls(sc, ctx, children);
5157 
5158 	sc->lro_timeout = 100;
5159 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5160 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5161 
5162 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5163 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
5164 
5165 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5166 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5167 
5168 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5169 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5170 
5171 	if (sc->flags & IS_VF)
5172 		return;
5173 
5174 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5175 	    NULL, chip_rev(sc), "chip hardware revision");
5176 
5177 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5178 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5179 
5180 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5181 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5182 
5183 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5184 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5185 
5186 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5187 	    CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5188 
5189 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5190 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5191 
5192 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5193 	    sc->er_version, 0, "expansion ROM version");
5194 
5195 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5196 	    sc->bs_version, 0, "bootstrap firmware version");
5197 
5198 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5199 	    NULL, sc->params.scfg_vers, "serial config version");
5200 
5201 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5202 	    NULL, sc->params.vpd_vers, "VPD version");
5203 
5204 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5205 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5206 
5207 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5208 	    sc->cfcsum, "config file checksum");
5209 
5210 #define SYSCTL_CAP(name, n, text) \
5211 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5212 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
5213 	    sysctl_bitfield, "A", "available " text " capabilities")
5214 
5215 	SYSCTL_CAP(nbmcaps, 0, "NBM");
5216 	SYSCTL_CAP(linkcaps, 1, "link");
5217 	SYSCTL_CAP(switchcaps, 2, "switch");
5218 	SYSCTL_CAP(niccaps, 3, "NIC");
5219 	SYSCTL_CAP(toecaps, 4, "TCP offload");
5220 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
5221 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5222 	SYSCTL_CAP(cryptocaps, 7, "crypto");
5223 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
5224 #undef SYSCTL_CAP
5225 
5226 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5227 	    NULL, sc->tids.nftids, "number of filters");
5228 
5229 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5230 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5231 	    "chip temperature (in Celsius)");
5232 
5233 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
5234 	    &sc->params.core_vdd, 0, "core Vdd (in mV)");
5235 
5236 #ifdef SBUF_DRAIN
5237 	/*
5238 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
5239 	 */
5240 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5241 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5242 	    "logs and miscellaneous information");
5243 	children = SYSCTL_CHILDREN(oid);
5244 
5245 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
5246 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5247 	    sysctl_cctrl, "A", "congestion control");
5248 
5249 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
5250 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5251 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
5252 
5253 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
5254 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
5255 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
5256 
5257 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
5258 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
5259 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
5260 
5261 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
5262 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
5263 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
5264 
5265 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
5266 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
5267 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
5268 
5269 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
5270 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
5271 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
5272 
5273 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
5274 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5275 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
5276 	    "A", "CIM logic analyzer");
5277 
5278 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
5279 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5280 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
5281 
5282 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
5283 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
5284 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
5285 
5286 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
5287 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
5288 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
5289 
5290 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
5291 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
5292 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
5293 
5294 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5295 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5296 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5297 
5298 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5299 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5300 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5301 
5302 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5303 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5304 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5305 
5306 	if (chip_id(sc) > CHELSIO_T4) {
5307 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5308 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5309 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5310 
5311 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5312 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5313 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5314 	}
5315 
5316 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5317 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5318 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5319 
5320 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5321 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5322 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
5323 
5324 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5325 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5326 	    sysctl_cpl_stats, "A", "CPL statistics");
5327 
5328 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5329 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5330 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5331 
5332 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5333 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5334 	    sysctl_devlog, "A", "firmware's device log");
5335 
5336 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5337 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5338 	    sysctl_fcoe_stats, "A", "FCoE statistics");
5339 
5340 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5341 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5342 	    sysctl_hw_sched, "A", "hardware scheduler ");
5343 
5344 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5345 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5346 	    sysctl_l2t, "A", "hardware L2 table");
5347 
5348 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5349 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5350 	    sysctl_lb_stats, "A", "loopback statistics");
5351 
5352 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5353 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5354 	    sysctl_meminfo, "A", "memory regions");
5355 
5356 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5357 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5358 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5359 	    "A", "MPS TCAM entries");
5360 
5361 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5362 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5363 	    sysctl_path_mtus, "A", "path MTUs");
5364 
5365 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5366 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5367 	    sysctl_pm_stats, "A", "PM statistics");
5368 
5369 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5370 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5371 	    sysctl_rdma_stats, "A", "RDMA statistics");
5372 
5373 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5374 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5375 	    sysctl_tcp_stats, "A", "TCP statistics");
5376 
5377 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5378 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5379 	    sysctl_tids, "A", "TID information");
5380 
5381 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5382 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5383 	    sysctl_tp_err_stats, "A", "TP error statistics");
5384 
5385 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5386 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5387 	    "TP logic analyzer event capture mask");
5388 
5389 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5390 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5391 	    sysctl_tp_la, "A", "TP logic analyzer");
5392 
5393 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5394 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5395 	    sysctl_tx_rate, "A", "Tx rate");
5396 
5397 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5398 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5399 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5400 
5401 	if (chip_id(sc) >= CHELSIO_T5) {
5402 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5403 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5404 		    sysctl_wcwr_stats, "A", "write combined work requests");
5405 	}
5406 #endif
5407 
5408 #ifdef TCP_OFFLOAD
5409 	if (is_offload(sc)) {
5410 		int i;
5411 		char s[4];
5412 
5413 		/*
5414 		 * dev.t4nex.X.toe.
5415 		 */
5416 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5417 		    NULL, "TOE parameters");
5418 		children = SYSCTL_CHILDREN(oid);
5419 
5420 		sc->tt.cong_algorithm = -1;
5421 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
5422 		    CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
5423 		    "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
5424 		    "3 = highspeed)");
5425 
5426 		sc->tt.sndbuf = 256 * 1024;
5427 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5428 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
5429 
5430 		sc->tt.ddp = 0;
5431 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5432 		    &sc->tt.ddp, 0, "DDP allowed");
5433 
5434 		sc->tt.rx_coalesce = 1;
5435 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5436 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5437 
5438 		sc->tt.tls = 0;
5439 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tls", CTLFLAG_RW,
5440 		    &sc->tt.tls, 0, "Inline TLS allowed");
5441 
5442 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls_rx_ports",
5443 		    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tls_rx_ports,
5444 		    "I", "TCP ports that use inline TLS+TOE RX");
5445 
5446 		sc->tt.tx_align = 1;
5447 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5448 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5449 
5450 		sc->tt.tx_zcopy = 0;
5451 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
5452 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
5453 		    "Enable zero-copy aio_write(2)");
5454 
5455 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
5456 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
5457 		    "TP timer tick (us)");
5458 
5459 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
5460 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
5461 		    "TCP timestamp tick (us)");
5462 
5463 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5464 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5465 		    "DACK tick (us)");
5466 
5467 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5468 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5469 		    "IU", "DACK timer (us)");
5470 
5471 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5472 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5473 		    sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
5474 
5475 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5476 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5477 		    sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
5478 
5479 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5480 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5481 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
5482 
5483 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5484 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5485 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
5486 
5487 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5488 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5489 		    sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
5490 
5491 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
5492 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
5493 		    sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
5494 
5495 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
5496 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
5497 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
5498 
5499 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
5500 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
5501 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
5502 
5503 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
5504 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
5505 		    sysctl_tp_shift_cnt, "IU",
5506 		    "Number of SYN retransmissions before abort");
5507 
5508 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
5509 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
5510 		    sysctl_tp_shift_cnt, "IU",
5511 		    "Number of retransmissions before abort");
5512 
5513 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
5514 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
5515 		    sysctl_tp_shift_cnt, "IU",
5516 		    "Number of keepalive probes before abort");
5517 
5518 		oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
5519 		    CTLFLAG_RD, NULL, "TOE retransmit backoffs");
5520 		children = SYSCTL_CHILDREN(oid);
5521 		for (i = 0; i < 16; i++) {
5522 			snprintf(s, sizeof(s), "%u", i);
5523 			SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
5524 			    CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
5525 			    "IU", "TOE retransmit backoff");
5526 		}
5527 	}
5528 #endif
5529 }
5530 
5531 void
5532 vi_sysctls(struct vi_info *vi)
5533 {
5534 	struct sysctl_ctx_list *ctx;
5535 	struct sysctl_oid *oid;
5536 	struct sysctl_oid_list *children;
5537 
5538 	ctx = device_get_sysctl_ctx(vi->dev);
5539 
5540 	/*
5541 	 * dev.v?(cxgbe|cxl).X.
5542 	 */
5543 	oid = device_get_sysctl_tree(vi->dev);
5544 	children = SYSCTL_CHILDREN(oid);
5545 
5546 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
5547 	    vi->viid, "VI identifer");
5548 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
5549 	    &vi->nrxq, 0, "# of rx queues");
5550 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5551 	    &vi->ntxq, 0, "# of tx queues");
5552 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5553 	    &vi->first_rxq, 0, "index of first rx queue");
5554 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5555 	    &vi->first_txq, 0, "index of first tx queue");
5556 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
5557 	    vi->rss_size, "size of RSS indirection table");
5558 
5559 	if (IS_MAIN_VI(vi)) {
5560 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
5561 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
5562 		    "Reserve queue 0 for non-flowid packets");
5563 	}
5564 
5565 #ifdef TCP_OFFLOAD
5566 	if (vi->nofldrxq != 0) {
5567 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5568 		    &vi->nofldrxq, 0,
5569 		    "# of rx queues for offloaded TCP connections");
5570 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5571 		    &vi->nofldtxq, 0,
5572 		    "# of tx queues for offloaded TCP connections");
5573 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5574 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
5575 		    "index of first TOE rx queue");
5576 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5577 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
5578 		    "index of first TOE tx queue");
5579 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
5580 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5581 		    sysctl_holdoff_tmr_idx_ofld, "I",
5582 		    "holdoff timer index for TOE queues");
5583 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
5584 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5585 		    sysctl_holdoff_pktc_idx_ofld, "I",
5586 		    "holdoff packet counter index for TOE queues");
5587 	}
5588 #endif
5589 #ifdef DEV_NETMAP
5590 	if (vi->nnmrxq != 0) {
5591 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5592 		    &vi->nnmrxq, 0, "# of netmap rx queues");
5593 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5594 		    &vi->nnmtxq, 0, "# of netmap tx queues");
5595 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5596 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
5597 		    "index of first netmap rx queue");
5598 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
5599 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
5600 		    "index of first netmap tx queue");
5601 	}
5602 #endif
5603 
5604 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
5605 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
5606 	    "holdoff timer index");
5607 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
5608 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
5609 	    "holdoff packet counter index");
5610 
5611 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
5612 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
5613 	    "rx queue size");
5614 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
5615 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
5616 	    "tx queue size");
5617 }
5618 
5619 static void
5620 cxgbe_sysctls(struct port_info *pi)
5621 {
5622 	struct sysctl_ctx_list *ctx;
5623 	struct sysctl_oid *oid;
5624 	struct sysctl_oid_list *children, *children2;
5625 	struct adapter *sc = pi->adapter;
5626 	int i;
5627 	char name[16];
5628 
5629 	ctx = device_get_sysctl_ctx(pi->dev);
5630 
5631 	/*
5632 	 * dev.cxgbe.X.
5633 	 */
5634 	oid = device_get_sysctl_tree(pi->dev);
5635 	children = SYSCTL_CHILDREN(oid);
5636 
5637 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
5638 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
5639 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
5640 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
5641 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
5642 		    "PHY temperature (in Celsius)");
5643 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
5644 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
5645 		    "PHY firmware version");
5646 	}
5647 
5648 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
5649 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
5650 	    "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
5651 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
5652 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
5653 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
5654 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
5655 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
5656 	    "autonegotiation (-1 = not supported)");
5657 
5658 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
5659 	    port_top_speed(pi), "max speed (in Gbps)");
5660 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
5661 	    pi->mps_bg_map, "MPS buffer group map");
5662 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
5663 	    NULL, pi->rx_e_chan_map, "TP rx e-channel map");
5664 
5665 	if (sc->flags & IS_VF)
5666 		return;
5667 
5668 	/*
5669 	 * dev.(cxgbe|cxl).X.tc.
5670 	 */
5671 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
5672 	    "Tx scheduler traffic classes (cl_rl)");
5673 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
5674 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
5675 
5676 		snprintf(name, sizeof(name), "%d", i);
5677 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
5678 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
5679 		    "traffic class"));
5680 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
5681 		    &tc->flags, 0, "flags");
5682 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
5683 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
5684 #ifdef SBUF_DRAIN
5685 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
5686 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
5687 		    sysctl_tc_params, "A", "traffic class parameters");
5688 #endif
5689 	}
5690 
5691 	/*
5692 	 * dev.cxgbe.X.stats.
5693 	 */
5694 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
5695 	    NULL, "port statistics");
5696 	children = SYSCTL_CHILDREN(oid);
5697 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
5698 	    &pi->tx_parse_error, 0,
5699 	    "# of tx packets with invalid length or # of segments");
5700 
5701 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
5702 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
5703 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
5704 	    sysctl_handle_t4_reg64, "QU", desc)
5705 
5706 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
5707 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
5708 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
5709 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
5710 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
5711 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
5712 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
5713 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
5714 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
5715 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
5716 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
5717 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
5718 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
5719 	    "# of tx frames in this range",
5720 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
5721 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
5722 	    "# of tx frames in this range",
5723 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
5724 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
5725 	    "# of tx frames in this range",
5726 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
5727 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
5728 	    "# of tx frames in this range",
5729 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
5730 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
5731 	    "# of tx frames in this range",
5732 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
5733 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
5734 	    "# of tx frames in this range",
5735 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
5736 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
5737 	    "# of tx frames in this range",
5738 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
5739 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
5740 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
5741 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
5742 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
5743 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
5744 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
5745 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5746 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5747 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5748 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5749 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5750 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5751 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5752 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5753 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5754 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5755 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5756 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5757 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5758 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5759 
5760 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5761 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5762 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5763 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5764 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5765 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5766 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5767 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5768 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5769 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5770 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5771 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5772 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5773 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5774 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5775 	    "# of frames received with bad FCS",
5776 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5777 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5778 	    "# of frames received with length error",
5779 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5780 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5781 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5782 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5783 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5784 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5785 	    "# of rx frames in this range",
5786 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5787 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5788 	    "# of rx frames in this range",
5789 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5790 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5791 	    "# of rx frames in this range",
5792 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5793 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5794 	    "# of rx frames in this range",
5795 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5796 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5797 	    "# of rx frames in this range",
5798 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5799 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5800 	    "# of rx frames in this range",
5801 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5802 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5803 	    "# of rx frames in this range",
5804 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5805 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5806 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5807 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5808 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5809 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5810 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5811 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5812 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5813 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5814 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5815 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5816 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5817 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5818 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5819 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5820 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5821 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5822 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5823 
5824 #undef SYSCTL_ADD_T4_REG64
5825 
5826 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5827 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5828 	    &pi->stats.name, desc)
5829 
5830 	/* We get these from port_stats and they may be stale by up to 1s */
5831 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5832 	    "# drops due to buffer-group 0 overflows");
5833 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5834 	    "# drops due to buffer-group 1 overflows");
5835 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5836 	    "# drops due to buffer-group 2 overflows");
5837 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5838 	    "# drops due to buffer-group 3 overflows");
5839 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5840 	    "# of buffer-group 0 truncated packets");
5841 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5842 	    "# of buffer-group 1 truncated packets");
5843 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5844 	    "# of buffer-group 2 truncated packets");
5845 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5846 	    "# of buffer-group 3 truncated packets");
5847 
5848 #undef SYSCTL_ADD_T4_PORTSTAT
5849 
5850 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_records",
5851 	    CTLFLAG_RD, &pi->tx_tls_records,
5852 	    "# of TLS records transmitted");
5853 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_octets",
5854 	    CTLFLAG_RD, &pi->tx_tls_octets,
5855 	    "# of payload octets in transmitted TLS records");
5856 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_records",
5857 	    CTLFLAG_RD, &pi->rx_tls_records,
5858 	    "# of TLS records received");
5859 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_octets",
5860 	    CTLFLAG_RD, &pi->rx_tls_octets,
5861 	    "# of payload octets in received TLS records");
5862 }
5863 
5864 static int
5865 sysctl_int_array(SYSCTL_HANDLER_ARGS)
5866 {
5867 	int rc, *i, space = 0;
5868 	struct sbuf sb;
5869 
5870 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
5871 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
5872 		if (space)
5873 			sbuf_printf(&sb, " ");
5874 		sbuf_printf(&sb, "%d", *i);
5875 		space = 1;
5876 	}
5877 	rc = sbuf_finish(&sb);
5878 	sbuf_delete(&sb);
5879 	return (rc);
5880 }
5881 
5882 static int
5883 sysctl_bitfield(SYSCTL_HANDLER_ARGS)
5884 {
5885 	int rc;
5886 	struct sbuf *sb;
5887 
5888 	rc = sysctl_wire_old_buffer(req, 0);
5889 	if (rc != 0)
5890 		return(rc);
5891 
5892 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5893 	if (sb == NULL)
5894 		return (ENOMEM);
5895 
5896 	sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
5897 	rc = sbuf_finish(sb);
5898 	sbuf_delete(sb);
5899 
5900 	return (rc);
5901 }
5902 
5903 static int
5904 sysctl_btphy(SYSCTL_HANDLER_ARGS)
5905 {
5906 	struct port_info *pi = arg1;
5907 	int op = arg2;
5908 	struct adapter *sc = pi->adapter;
5909 	u_int v;
5910 	int rc;
5911 
5912 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
5913 	if (rc)
5914 		return (rc);
5915 	/* XXX: magic numbers */
5916 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
5917 	    &v);
5918 	end_synchronized_op(sc, 0);
5919 	if (rc)
5920 		return (rc);
5921 	if (op == 0)
5922 		v /= 256;
5923 
5924 	rc = sysctl_handle_int(oidp, &v, 0, req);
5925 	return (rc);
5926 }
5927 
5928 static int
5929 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
5930 {
5931 	struct vi_info *vi = arg1;
5932 	int rc, val;
5933 
5934 	val = vi->rsrv_noflowq;
5935 	rc = sysctl_handle_int(oidp, &val, 0, req);
5936 	if (rc != 0 || req->newptr == NULL)
5937 		return (rc);
5938 
5939 	if ((val >= 1) && (vi->ntxq > 1))
5940 		vi->rsrv_noflowq = 1;
5941 	else
5942 		vi->rsrv_noflowq = 0;
5943 
5944 	return (rc);
5945 }
5946 
5947 static int
5948 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
5949 {
5950 	struct vi_info *vi = arg1;
5951 	struct adapter *sc = vi->pi->adapter;
5952 	int idx, rc, i;
5953 	struct sge_rxq *rxq;
5954 	uint8_t v;
5955 
5956 	idx = vi->tmr_idx;
5957 
5958 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5959 	if (rc != 0 || req->newptr == NULL)
5960 		return (rc);
5961 
5962 	if (idx < 0 || idx >= SGE_NTIMERS)
5963 		return (EINVAL);
5964 
5965 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5966 	    "t4tmr");
5967 	if (rc)
5968 		return (rc);
5969 
5970 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
5971 	for_each_rxq(vi, i, rxq) {
5972 #ifdef atomic_store_rel_8
5973 		atomic_store_rel_8(&rxq->iq.intr_params, v);
5974 #else
5975 		rxq->iq.intr_params = v;
5976 #endif
5977 	}
5978 	vi->tmr_idx = idx;
5979 
5980 	end_synchronized_op(sc, LOCK_HELD);
5981 	return (0);
5982 }
5983 
5984 static int
5985 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
5986 {
5987 	struct vi_info *vi = arg1;
5988 	struct adapter *sc = vi->pi->adapter;
5989 	int idx, rc;
5990 
5991 	idx = vi->pktc_idx;
5992 
5993 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5994 	if (rc != 0 || req->newptr == NULL)
5995 		return (rc);
5996 
5997 	if (idx < -1 || idx >= SGE_NCOUNTERS)
5998 		return (EINVAL);
5999 
6000 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6001 	    "t4pktc");
6002 	if (rc)
6003 		return (rc);
6004 
6005 	if (vi->flags & VI_INIT_DONE)
6006 		rc = EBUSY; /* cannot be changed once the queues are created */
6007 	else
6008 		vi->pktc_idx = idx;
6009 
6010 	end_synchronized_op(sc, LOCK_HELD);
6011 	return (rc);
6012 }
6013 
6014 static int
6015 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
6016 {
6017 	struct vi_info *vi = arg1;
6018 	struct adapter *sc = vi->pi->adapter;
6019 	int qsize, rc;
6020 
6021 	qsize = vi->qsize_rxq;
6022 
6023 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6024 	if (rc != 0 || req->newptr == NULL)
6025 		return (rc);
6026 
6027 	if (qsize < 128 || (qsize & 7))
6028 		return (EINVAL);
6029 
6030 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6031 	    "t4rxqs");
6032 	if (rc)
6033 		return (rc);
6034 
6035 	if (vi->flags & VI_INIT_DONE)
6036 		rc = EBUSY; /* cannot be changed once the queues are created */
6037 	else
6038 		vi->qsize_rxq = qsize;
6039 
6040 	end_synchronized_op(sc, LOCK_HELD);
6041 	return (rc);
6042 }
6043 
6044 static int
6045 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
6046 {
6047 	struct vi_info *vi = arg1;
6048 	struct adapter *sc = vi->pi->adapter;
6049 	int qsize, rc;
6050 
6051 	qsize = vi->qsize_txq;
6052 
6053 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6054 	if (rc != 0 || req->newptr == NULL)
6055 		return (rc);
6056 
6057 	if (qsize < 128 || qsize > 65536)
6058 		return (EINVAL);
6059 
6060 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6061 	    "t4txqs");
6062 	if (rc)
6063 		return (rc);
6064 
6065 	if (vi->flags & VI_INIT_DONE)
6066 		rc = EBUSY; /* cannot be changed once the queues are created */
6067 	else
6068 		vi->qsize_txq = qsize;
6069 
6070 	end_synchronized_op(sc, LOCK_HELD);
6071 	return (rc);
6072 }
6073 
6074 static int
6075 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
6076 {
6077 	struct port_info *pi = arg1;
6078 	struct adapter *sc = pi->adapter;
6079 	struct link_config *lc = &pi->link_cfg;
6080 	int rc;
6081 
6082 	if (req->newptr == NULL) {
6083 		struct sbuf *sb;
6084 		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
6085 
6086 		rc = sysctl_wire_old_buffer(req, 0);
6087 		if (rc != 0)
6088 			return(rc);
6089 
6090 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6091 		if (sb == NULL)
6092 			return (ENOMEM);
6093 
6094 		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
6095 		rc = sbuf_finish(sb);
6096 		sbuf_delete(sb);
6097 	} else {
6098 		char s[2];
6099 		int n;
6100 
6101 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
6102 		s[1] = 0;
6103 
6104 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6105 		if (rc != 0)
6106 			return(rc);
6107 
6108 		if (s[1] != 0)
6109 			return (EINVAL);
6110 		if (s[0] < '0' || s[0] > '9')
6111 			return (EINVAL);	/* not a number */
6112 		n = s[0] - '0';
6113 		if (n & ~(PAUSE_TX | PAUSE_RX))
6114 			return (EINVAL);	/* some other bit is set too */
6115 
6116 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6117 		    "t4PAUSE");
6118 		if (rc)
6119 			return (rc);
6120 		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
6121 			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
6122 			lc->requested_fc |= n;
6123 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6124 			if (rc == 0) {
6125 				lc->fc = lc->requested_fc;
6126 			}
6127 		}
6128 		end_synchronized_op(sc, 0);
6129 	}
6130 
6131 	return (rc);
6132 }
6133 
6134 static int
6135 sysctl_fec(SYSCTL_HANDLER_ARGS)
6136 {
6137 	struct port_info *pi = arg1;
6138 	struct adapter *sc = pi->adapter;
6139 	struct link_config *lc = &pi->link_cfg;
6140 	int rc;
6141 
6142 	if (req->newptr == NULL) {
6143 		struct sbuf *sb;
6144 		static char *bits = "\20\1RS\2BASER_RS\3RESERVED";
6145 
6146 		rc = sysctl_wire_old_buffer(req, 0);
6147 		if (rc != 0)
6148 			return(rc);
6149 
6150 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6151 		if (sb == NULL)
6152 			return (ENOMEM);
6153 
6154 		sbuf_printf(sb, "%b", lc->fec & M_FW_PORT_CAP_FEC, bits);
6155 		rc = sbuf_finish(sb);
6156 		sbuf_delete(sb);
6157 	} else {
6158 		char s[2];
6159 		int n;
6160 
6161 		s[0] = '0' + (lc->requested_fec & M_FW_PORT_CAP_FEC);
6162 		s[1] = 0;
6163 
6164 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6165 		if (rc != 0)
6166 			return(rc);
6167 
6168 		if (s[1] != 0)
6169 			return (EINVAL);
6170 		if (s[0] < '0' || s[0] > '9')
6171 			return (EINVAL);	/* not a number */
6172 		n = s[0] - '0';
6173 		if (n & ~M_FW_PORT_CAP_FEC)
6174 			return (EINVAL);	/* some other bit is set too */
6175 
6176 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6177 		    "t4fec");
6178 		if (rc)
6179 			return (rc);
6180 		if ((lc->requested_fec & M_FW_PORT_CAP_FEC) != n) {
6181 			lc->requested_fec = n &
6182 			    G_FW_PORT_CAP_FEC(lc->supported);
6183 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6184 			if (rc == 0) {
6185 				lc->fec = lc->requested_fec;
6186 			}
6187 		}
6188 		end_synchronized_op(sc, 0);
6189 	}
6190 
6191 	return (rc);
6192 }
6193 
6194 static int
6195 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
6196 {
6197 	struct port_info *pi = arg1;
6198 	struct adapter *sc = pi->adapter;
6199 	struct link_config *lc = &pi->link_cfg;
6200 	int rc, val, old;
6201 
6202 	if (lc->supported & FW_PORT_CAP_ANEG)
6203 		val = lc->requested_aneg == AUTONEG_ENABLE ? 1 : 0;
6204 	else
6205 		val = -1;
6206 	rc = sysctl_handle_int(oidp, &val, 0, req);
6207 	if (rc != 0 || req->newptr == NULL)
6208 		return (rc);
6209 	if ((lc->supported & FW_PORT_CAP_ANEG) == 0)
6210 		return (ENOTSUP);
6211 
6212 	if (val == 0)
6213 		val = AUTONEG_DISABLE;
6214 	else if (val == 1)
6215 		val = AUTONEG_ENABLE;
6216 	else
6217 		return (EINVAL);
6218 	if (lc->requested_aneg == val)
6219 		return (0);	/* no change */
6220 
6221 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6222 	    "t4aneg");
6223 	if (rc)
6224 		return (rc);
6225 	old = lc->requested_aneg;
6226 	lc->requested_aneg = val;
6227 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6228 	if (rc != 0)
6229 		lc->requested_aneg = old;
6230 	end_synchronized_op(sc, 0);
6231 	return (rc);
6232 }
6233 
6234 static int
6235 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
6236 {
6237 	struct adapter *sc = arg1;
6238 	int reg = arg2;
6239 	uint64_t val;
6240 
6241 	val = t4_read_reg64(sc, reg);
6242 
6243 	return (sysctl_handle_64(oidp, &val, 0, req));
6244 }
6245 
6246 static int
6247 sysctl_temperature(SYSCTL_HANDLER_ARGS)
6248 {
6249 	struct adapter *sc = arg1;
6250 	int rc, t;
6251 	uint32_t param, val;
6252 
6253 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
6254 	if (rc)
6255 		return (rc);
6256 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6257 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
6258 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
6259 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6260 	end_synchronized_op(sc, 0);
6261 	if (rc)
6262 		return (rc);
6263 
6264 	/* unknown is returned as 0 but we display -1 in that case */
6265 	t = val == 0 ? -1 : val;
6266 
6267 	rc = sysctl_handle_int(oidp, &t, 0, req);
6268 	return (rc);
6269 }
6270 
6271 #ifdef SBUF_DRAIN
6272 static int
6273 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
6274 {
6275 	struct adapter *sc = arg1;
6276 	struct sbuf *sb;
6277 	int rc, i;
6278 	uint16_t incr[NMTUS][NCCTRL_WIN];
6279 	static const char *dec_fac[] = {
6280 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
6281 		"0.9375"
6282 	};
6283 
6284 	rc = sysctl_wire_old_buffer(req, 0);
6285 	if (rc != 0)
6286 		return (rc);
6287 
6288 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6289 	if (sb == NULL)
6290 		return (ENOMEM);
6291 
6292 	t4_read_cong_tbl(sc, incr);
6293 
6294 	for (i = 0; i < NCCTRL_WIN; ++i) {
6295 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
6296 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
6297 		    incr[5][i], incr[6][i], incr[7][i]);
6298 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
6299 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
6300 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
6301 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
6302 	}
6303 
6304 	rc = sbuf_finish(sb);
6305 	sbuf_delete(sb);
6306 
6307 	return (rc);
6308 }
6309 
6310 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
6311 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
6312 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
6313 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
6314 };
6315 
6316 static int
6317 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
6318 {
6319 	struct adapter *sc = arg1;
6320 	struct sbuf *sb;
6321 	int rc, i, n, qid = arg2;
6322 	uint32_t *buf, *p;
6323 	char *qtype;
6324 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
6325 
6326 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
6327 	    ("%s: bad qid %d\n", __func__, qid));
6328 
6329 	if (qid < CIM_NUM_IBQ) {
6330 		/* inbound queue */
6331 		qtype = "IBQ";
6332 		n = 4 * CIM_IBQ_SIZE;
6333 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6334 		rc = t4_read_cim_ibq(sc, qid, buf, n);
6335 	} else {
6336 		/* outbound queue */
6337 		qtype = "OBQ";
6338 		qid -= CIM_NUM_IBQ;
6339 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
6340 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6341 		rc = t4_read_cim_obq(sc, qid, buf, n);
6342 	}
6343 
6344 	if (rc < 0) {
6345 		rc = -rc;
6346 		goto done;
6347 	}
6348 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
6349 
6350 	rc = sysctl_wire_old_buffer(req, 0);
6351 	if (rc != 0)
6352 		goto done;
6353 
6354 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6355 	if (sb == NULL) {
6356 		rc = ENOMEM;
6357 		goto done;
6358 	}
6359 
6360 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
6361 	for (i = 0, p = buf; i < n; i += 16, p += 4)
6362 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
6363 		    p[2], p[3]);
6364 
6365 	rc = sbuf_finish(sb);
6366 	sbuf_delete(sb);
6367 done:
6368 	free(buf, M_CXGBE);
6369 	return (rc);
6370 }
6371 
6372 static int
6373 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
6374 {
6375 	struct adapter *sc = arg1;
6376 	u_int cfg;
6377 	struct sbuf *sb;
6378 	uint32_t *buf, *p;
6379 	int rc;
6380 
6381 	MPASS(chip_id(sc) <= CHELSIO_T5);
6382 
6383 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6384 	if (rc != 0)
6385 		return (rc);
6386 
6387 	rc = sysctl_wire_old_buffer(req, 0);
6388 	if (rc != 0)
6389 		return (rc);
6390 
6391 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6392 	if (sb == NULL)
6393 		return (ENOMEM);
6394 
6395 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6396 	    M_ZERO | M_WAITOK);
6397 
6398 	rc = -t4_cim_read_la(sc, buf, NULL);
6399 	if (rc != 0)
6400 		goto done;
6401 
6402 	sbuf_printf(sb, "Status   Data      PC%s",
6403 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6404 	    "     LS0Stat  LS0Addr             LS0Data");
6405 
6406 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
6407 		if (cfg & F_UPDBGLACAPTPCONLY) {
6408 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
6409 			    p[6], p[7]);
6410 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
6411 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
6412 			    p[4] & 0xff, p[5] >> 8);
6413 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
6414 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6415 			    p[1] & 0xf, p[2] >> 4);
6416 		} else {
6417 			sbuf_printf(sb,
6418 			    "\n  %02x   %x%07x %x%07x %08x %08x "
6419 			    "%08x%08x%08x%08x",
6420 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6421 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
6422 			    p[6], p[7]);
6423 		}
6424 	}
6425 
6426 	rc = sbuf_finish(sb);
6427 	sbuf_delete(sb);
6428 done:
6429 	free(buf, M_CXGBE);
6430 	return (rc);
6431 }
6432 
6433 static int
6434 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
6435 {
6436 	struct adapter *sc = arg1;
6437 	u_int cfg;
6438 	struct sbuf *sb;
6439 	uint32_t *buf, *p;
6440 	int rc;
6441 
6442 	MPASS(chip_id(sc) > CHELSIO_T5);
6443 
6444 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6445 	if (rc != 0)
6446 		return (rc);
6447 
6448 	rc = sysctl_wire_old_buffer(req, 0);
6449 	if (rc != 0)
6450 		return (rc);
6451 
6452 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6453 	if (sb == NULL)
6454 		return (ENOMEM);
6455 
6456 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6457 	    M_ZERO | M_WAITOK);
6458 
6459 	rc = -t4_cim_read_la(sc, buf, NULL);
6460 	if (rc != 0)
6461 		goto done;
6462 
6463 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
6464 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6465 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
6466 
6467 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
6468 		if (cfg & F_UPDBGLACAPTPCONLY) {
6469 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
6470 			    p[3] & 0xff, p[2], p[1], p[0]);
6471 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
6472 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
6473 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
6474 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
6475 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
6476 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
6477 			    p[6] >> 16);
6478 		} else {
6479 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
6480 			    "%08x %08x %08x %08x %08x %08x",
6481 			    (p[9] >> 16) & 0xff,
6482 			    p[9] & 0xffff, p[8] >> 16,
6483 			    p[8] & 0xffff, p[7] >> 16,
6484 			    p[7] & 0xffff, p[6] >> 16,
6485 			    p[2], p[1], p[0], p[5], p[4], p[3]);
6486 		}
6487 	}
6488 
6489 	rc = sbuf_finish(sb);
6490 	sbuf_delete(sb);
6491 done:
6492 	free(buf, M_CXGBE);
6493 	return (rc);
6494 }
6495 
6496 static int
6497 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
6498 {
6499 	struct adapter *sc = arg1;
6500 	u_int i;
6501 	struct sbuf *sb;
6502 	uint32_t *buf, *p;
6503 	int rc;
6504 
6505 	rc = sysctl_wire_old_buffer(req, 0);
6506 	if (rc != 0)
6507 		return (rc);
6508 
6509 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6510 	if (sb == NULL)
6511 		return (ENOMEM);
6512 
6513 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
6514 	    M_ZERO | M_WAITOK);
6515 
6516 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
6517 	p = buf;
6518 
6519 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6520 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
6521 		    p[1], p[0]);
6522 	}
6523 
6524 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
6525 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6526 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
6527 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
6528 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
6529 		    (p[1] >> 2) | ((p[2] & 3) << 30),
6530 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
6531 		    p[0] & 1);
6532 	}
6533 
6534 	rc = sbuf_finish(sb);
6535 	sbuf_delete(sb);
6536 	free(buf, M_CXGBE);
6537 	return (rc);
6538 }
6539 
6540 static int
6541 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
6542 {
6543 	struct adapter *sc = arg1;
6544 	u_int i;
6545 	struct sbuf *sb;
6546 	uint32_t *buf, *p;
6547 	int rc;
6548 
6549 	rc = sysctl_wire_old_buffer(req, 0);
6550 	if (rc != 0)
6551 		return (rc);
6552 
6553 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6554 	if (sb == NULL)
6555 		return (ENOMEM);
6556 
6557 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
6558 	    M_ZERO | M_WAITOK);
6559 
6560 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
6561 	p = buf;
6562 
6563 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
6564 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6565 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
6566 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
6567 		    p[4], p[3], p[2], p[1], p[0]);
6568 	}
6569 
6570 	sbuf_printf(sb, "\n\nCntl ID               Data");
6571 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6572 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
6573 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
6574 	}
6575 
6576 	rc = sbuf_finish(sb);
6577 	sbuf_delete(sb);
6578 	free(buf, M_CXGBE);
6579 	return (rc);
6580 }
6581 
6582 static int
6583 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
6584 {
6585 	struct adapter *sc = arg1;
6586 	struct sbuf *sb;
6587 	int rc, i;
6588 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6589 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6590 	uint16_t thres[CIM_NUM_IBQ];
6591 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
6592 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
6593 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
6594 
6595 	cim_num_obq = sc->chip_params->cim_num_obq;
6596 	if (is_t4(sc)) {
6597 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
6598 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
6599 	} else {
6600 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
6601 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
6602 	}
6603 	nq = CIM_NUM_IBQ + cim_num_obq;
6604 
6605 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
6606 	if (rc == 0)
6607 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
6608 	if (rc != 0)
6609 		return (rc);
6610 
6611 	t4_read_cimq_cfg(sc, base, size, thres);
6612 
6613 	rc = sysctl_wire_old_buffer(req, 0);
6614 	if (rc != 0)
6615 		return (rc);
6616 
6617 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6618 	if (sb == NULL)
6619 		return (ENOMEM);
6620 
6621 	sbuf_printf(sb,
6622 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
6623 
6624 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
6625 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
6626 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
6627 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6628 		    G_QUEREMFLITS(p[2]) * 16);
6629 	for ( ; i < nq; i++, p += 4, wr += 2)
6630 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
6631 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
6632 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6633 		    G_QUEREMFLITS(p[2]) * 16);
6634 
6635 	rc = sbuf_finish(sb);
6636 	sbuf_delete(sb);
6637 
6638 	return (rc);
6639 }
6640 
6641 static int
6642 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
6643 {
6644 	struct adapter *sc = arg1;
6645 	struct sbuf *sb;
6646 	int rc;
6647 	struct tp_cpl_stats stats;
6648 
6649 	rc = sysctl_wire_old_buffer(req, 0);
6650 	if (rc != 0)
6651 		return (rc);
6652 
6653 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6654 	if (sb == NULL)
6655 		return (ENOMEM);
6656 
6657 	mtx_lock(&sc->reg_lock);
6658 	t4_tp_get_cpl_stats(sc, &stats, 0);
6659 	mtx_unlock(&sc->reg_lock);
6660 
6661 	if (sc->chip_params->nchan > 2) {
6662 		sbuf_printf(sb, "                 channel 0  channel 1"
6663 		    "  channel 2  channel 3");
6664 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
6665 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
6666 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
6667 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
6668 	} else {
6669 		sbuf_printf(sb, "                 channel 0  channel 1");
6670 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
6671 		    stats.req[0], stats.req[1]);
6672 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
6673 		    stats.rsp[0], stats.rsp[1]);
6674 	}
6675 
6676 	rc = sbuf_finish(sb);
6677 	sbuf_delete(sb);
6678 
6679 	return (rc);
6680 }
6681 
6682 static int
6683 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
6684 {
6685 	struct adapter *sc = arg1;
6686 	struct sbuf *sb;
6687 	int rc;
6688 	struct tp_usm_stats stats;
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, 256, req);
6695 	if (sb == NULL)
6696 		return (ENOMEM);
6697 
6698 	t4_get_usm_stats(sc, &stats, 1);
6699 
6700 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
6701 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
6702 	sbuf_printf(sb, "Drops:  %u", stats.drops);
6703 
6704 	rc = sbuf_finish(sb);
6705 	sbuf_delete(sb);
6706 
6707 	return (rc);
6708 }
6709 
6710 static const char * const devlog_level_strings[] = {
6711 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
6712 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
6713 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
6714 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
6715 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
6716 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
6717 };
6718 
6719 static const char * const devlog_facility_strings[] = {
6720 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
6721 	[FW_DEVLOG_FACILITY_CF]		= "CF",
6722 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
6723 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
6724 	[FW_DEVLOG_FACILITY_RES]	= "RES",
6725 	[FW_DEVLOG_FACILITY_HW]		= "HW",
6726 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
6727 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
6728 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
6729 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
6730 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
6731 	[FW_DEVLOG_FACILITY_VI]		= "VI",
6732 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
6733 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
6734 	[FW_DEVLOG_FACILITY_TM]		= "TM",
6735 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
6736 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
6737 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
6738 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
6739 	[FW_DEVLOG_FACILITY_RI]		= "RI",
6740 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
6741 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
6742 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
6743 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
6744 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
6745 };
6746 
6747 static int
6748 sysctl_devlog(SYSCTL_HANDLER_ARGS)
6749 {
6750 	struct adapter *sc = arg1;
6751 	struct devlog_params *dparams = &sc->params.devlog;
6752 	struct fw_devlog_e *buf, *e;
6753 	int i, j, rc, nentries, first = 0;
6754 	struct sbuf *sb;
6755 	uint64_t ftstamp = UINT64_MAX;
6756 
6757 	if (dparams->addr == 0)
6758 		return (ENXIO);
6759 
6760 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
6761 	if (buf == NULL)
6762 		return (ENOMEM);
6763 
6764 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
6765 	if (rc != 0)
6766 		goto done;
6767 
6768 	nentries = dparams->size / sizeof(struct fw_devlog_e);
6769 	for (i = 0; i < nentries; i++) {
6770 		e = &buf[i];
6771 
6772 		if (e->timestamp == 0)
6773 			break;	/* end */
6774 
6775 		e->timestamp = be64toh(e->timestamp);
6776 		e->seqno = be32toh(e->seqno);
6777 		for (j = 0; j < 8; j++)
6778 			e->params[j] = be32toh(e->params[j]);
6779 
6780 		if (e->timestamp < ftstamp) {
6781 			ftstamp = e->timestamp;
6782 			first = i;
6783 		}
6784 	}
6785 
6786 	if (buf[first].timestamp == 0)
6787 		goto done;	/* nothing in the log */
6788 
6789 	rc = sysctl_wire_old_buffer(req, 0);
6790 	if (rc != 0)
6791 		goto done;
6792 
6793 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6794 	if (sb == NULL) {
6795 		rc = ENOMEM;
6796 		goto done;
6797 	}
6798 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
6799 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
6800 
6801 	i = first;
6802 	do {
6803 		e = &buf[i];
6804 		if (e->timestamp == 0)
6805 			break;	/* end */
6806 
6807 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
6808 		    e->seqno, e->timestamp,
6809 		    (e->level < nitems(devlog_level_strings) ?
6810 			devlog_level_strings[e->level] : "UNKNOWN"),
6811 		    (e->facility < nitems(devlog_facility_strings) ?
6812 			devlog_facility_strings[e->facility] : "UNKNOWN"));
6813 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
6814 		    e->params[2], e->params[3], e->params[4],
6815 		    e->params[5], e->params[6], e->params[7]);
6816 
6817 		if (++i == nentries)
6818 			i = 0;
6819 	} while (i != first);
6820 
6821 	rc = sbuf_finish(sb);
6822 	sbuf_delete(sb);
6823 done:
6824 	free(buf, M_CXGBE);
6825 	return (rc);
6826 }
6827 
6828 static int
6829 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
6830 {
6831 	struct adapter *sc = arg1;
6832 	struct sbuf *sb;
6833 	int rc;
6834 	struct tp_fcoe_stats stats[MAX_NCHAN];
6835 	int i, nchan = sc->chip_params->nchan;
6836 
6837 	rc = sysctl_wire_old_buffer(req, 0);
6838 	if (rc != 0)
6839 		return (rc);
6840 
6841 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6842 	if (sb == NULL)
6843 		return (ENOMEM);
6844 
6845 	for (i = 0; i < nchan; i++)
6846 		t4_get_fcoe_stats(sc, i, &stats[i], 1);
6847 
6848 	if (nchan > 2) {
6849 		sbuf_printf(sb, "                   channel 0        channel 1"
6850 		    "        channel 2        channel 3");
6851 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
6852 		    stats[0].octets_ddp, stats[1].octets_ddp,
6853 		    stats[2].octets_ddp, stats[3].octets_ddp);
6854 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
6855 		    stats[0].frames_ddp, stats[1].frames_ddp,
6856 		    stats[2].frames_ddp, stats[3].frames_ddp);
6857 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
6858 		    stats[0].frames_drop, stats[1].frames_drop,
6859 		    stats[2].frames_drop, stats[3].frames_drop);
6860 	} else {
6861 		sbuf_printf(sb, "                   channel 0        channel 1");
6862 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
6863 		    stats[0].octets_ddp, stats[1].octets_ddp);
6864 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
6865 		    stats[0].frames_ddp, stats[1].frames_ddp);
6866 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
6867 		    stats[0].frames_drop, stats[1].frames_drop);
6868 	}
6869 
6870 	rc = sbuf_finish(sb);
6871 	sbuf_delete(sb);
6872 
6873 	return (rc);
6874 }
6875 
6876 static int
6877 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
6878 {
6879 	struct adapter *sc = arg1;
6880 	struct sbuf *sb;
6881 	int rc, i;
6882 	unsigned int map, kbps, ipg, mode;
6883 	unsigned int pace_tab[NTX_SCHED];
6884 
6885 	rc = sysctl_wire_old_buffer(req, 0);
6886 	if (rc != 0)
6887 		return (rc);
6888 
6889 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6890 	if (sb == NULL)
6891 		return (ENOMEM);
6892 
6893 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
6894 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
6895 	t4_read_pace_tbl(sc, pace_tab);
6896 
6897 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
6898 	    "Class IPG (0.1 ns)   Flow IPG (us)");
6899 
6900 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
6901 		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
6902 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
6903 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
6904 		if (kbps)
6905 			sbuf_printf(sb, "%9u     ", kbps);
6906 		else
6907 			sbuf_printf(sb, " disabled     ");
6908 
6909 		if (ipg)
6910 			sbuf_printf(sb, "%13u        ", ipg);
6911 		else
6912 			sbuf_printf(sb, "     disabled        ");
6913 
6914 		if (pace_tab[i])
6915 			sbuf_printf(sb, "%10u", pace_tab[i]);
6916 		else
6917 			sbuf_printf(sb, "  disabled");
6918 	}
6919 
6920 	rc = sbuf_finish(sb);
6921 	sbuf_delete(sb);
6922 
6923 	return (rc);
6924 }
6925 
6926 static int
6927 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
6928 {
6929 	struct adapter *sc = arg1;
6930 	struct sbuf *sb;
6931 	int rc, i, j;
6932 	uint64_t *p0, *p1;
6933 	struct lb_port_stats s[2];
6934 	static const char *stat_name[] = {
6935 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
6936 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
6937 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
6938 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
6939 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
6940 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
6941 		"BG2FramesTrunc:", "BG3FramesTrunc:"
6942 	};
6943 
6944 	rc = sysctl_wire_old_buffer(req, 0);
6945 	if (rc != 0)
6946 		return (rc);
6947 
6948 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6949 	if (sb == NULL)
6950 		return (ENOMEM);
6951 
6952 	memset(s, 0, sizeof(s));
6953 
6954 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
6955 		t4_get_lb_stats(sc, i, &s[0]);
6956 		t4_get_lb_stats(sc, i + 1, &s[1]);
6957 
6958 		p0 = &s[0].octets;
6959 		p1 = &s[1].octets;
6960 		sbuf_printf(sb, "%s                       Loopback %u"
6961 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
6962 
6963 		for (j = 0; j < nitems(stat_name); j++)
6964 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
6965 				   *p0++, *p1++);
6966 	}
6967 
6968 	rc = sbuf_finish(sb);
6969 	sbuf_delete(sb);
6970 
6971 	return (rc);
6972 }
6973 
6974 static int
6975 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
6976 {
6977 	int rc = 0;
6978 	struct port_info *pi = arg1;
6979 	struct link_config *lc = &pi->link_cfg;
6980 	struct sbuf *sb;
6981 
6982 	rc = sysctl_wire_old_buffer(req, 0);
6983 	if (rc != 0)
6984 		return(rc);
6985 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
6986 	if (sb == NULL)
6987 		return (ENOMEM);
6988 
6989 	if (lc->link_ok || lc->link_down_rc == 255)
6990 		sbuf_printf(sb, "n/a");
6991 	else
6992 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
6993 
6994 	rc = sbuf_finish(sb);
6995 	sbuf_delete(sb);
6996 
6997 	return (rc);
6998 }
6999 
7000 struct mem_desc {
7001 	unsigned int base;
7002 	unsigned int limit;
7003 	unsigned int idx;
7004 };
7005 
7006 static int
7007 mem_desc_cmp(const void *a, const void *b)
7008 {
7009 	return ((const struct mem_desc *)a)->base -
7010 	       ((const struct mem_desc *)b)->base;
7011 }
7012 
7013 static void
7014 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
7015     unsigned int to)
7016 {
7017 	unsigned int size;
7018 
7019 	if (from == to)
7020 		return;
7021 
7022 	size = to - from + 1;
7023 	if (size == 0)
7024 		return;
7025 
7026 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
7027 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
7028 }
7029 
7030 static int
7031 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
7032 {
7033 	struct adapter *sc = arg1;
7034 	struct sbuf *sb;
7035 	int rc, i, n;
7036 	uint32_t lo, hi, used, alloc;
7037 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
7038 	static const char *region[] = {
7039 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
7040 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
7041 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
7042 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
7043 		"RQUDP region:", "PBL region:", "TXPBL region:",
7044 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
7045 		"On-chip queues:", "TLS keys:",
7046 	};
7047 	struct mem_desc avail[4];
7048 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
7049 	struct mem_desc *md = mem;
7050 
7051 	rc = sysctl_wire_old_buffer(req, 0);
7052 	if (rc != 0)
7053 		return (rc);
7054 
7055 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7056 	if (sb == NULL)
7057 		return (ENOMEM);
7058 
7059 	for (i = 0; i < nitems(mem); i++) {
7060 		mem[i].limit = 0;
7061 		mem[i].idx = i;
7062 	}
7063 
7064 	/* Find and sort the populated memory ranges */
7065 	i = 0;
7066 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
7067 	if (lo & F_EDRAM0_ENABLE) {
7068 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
7069 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
7070 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
7071 		avail[i].idx = 0;
7072 		i++;
7073 	}
7074 	if (lo & F_EDRAM1_ENABLE) {
7075 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
7076 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
7077 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
7078 		avail[i].idx = 1;
7079 		i++;
7080 	}
7081 	if (lo & F_EXT_MEM_ENABLE) {
7082 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
7083 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
7084 		avail[i].limit = avail[i].base +
7085 		    (G_EXT_MEM_SIZE(hi) << 20);
7086 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
7087 		i++;
7088 	}
7089 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
7090 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
7091 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
7092 		avail[i].limit = avail[i].base +
7093 		    (G_EXT_MEM1_SIZE(hi) << 20);
7094 		avail[i].idx = 4;
7095 		i++;
7096 	}
7097 	if (!i)                                    /* no memory available */
7098 		return 0;
7099 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
7100 
7101 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
7102 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
7103 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
7104 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
7105 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
7106 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
7107 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
7108 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
7109 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
7110 
7111 	/* the next few have explicit upper bounds */
7112 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
7113 	md->limit = md->base - 1 +
7114 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
7115 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
7116 	md++;
7117 
7118 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
7119 	md->limit = md->base - 1 +
7120 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
7121 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
7122 	md++;
7123 
7124 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7125 		if (chip_id(sc) <= CHELSIO_T5)
7126 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
7127 		else
7128 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
7129 		md->limit = 0;
7130 	} else {
7131 		md->base = 0;
7132 		md->idx = nitems(region);  /* hide it */
7133 	}
7134 	md++;
7135 
7136 #define ulp_region(reg) \
7137 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
7138 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
7139 
7140 	ulp_region(RX_ISCSI);
7141 	ulp_region(RX_TDDP);
7142 	ulp_region(TX_TPT);
7143 	ulp_region(RX_STAG);
7144 	ulp_region(RX_RQ);
7145 	ulp_region(RX_RQUDP);
7146 	ulp_region(RX_PBL);
7147 	ulp_region(TX_PBL);
7148 #undef ulp_region
7149 
7150 	md->base = 0;
7151 	md->idx = nitems(region);
7152 	if (!is_t4(sc)) {
7153 		uint32_t size = 0;
7154 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
7155 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
7156 
7157 		if (is_t5(sc)) {
7158 			if (sge_ctrl & F_VFIFO_ENABLE)
7159 				size = G_DBVFIFO_SIZE(fifo_size);
7160 		} else
7161 			size = G_T6_DBVFIFO_SIZE(fifo_size);
7162 
7163 		if (size) {
7164 			md->base = G_BASEADDR(t4_read_reg(sc,
7165 			    A_SGE_DBVFIFO_BADDR));
7166 			md->limit = md->base + (size << 2) - 1;
7167 		}
7168 	}
7169 	md++;
7170 
7171 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
7172 	md->limit = 0;
7173 	md++;
7174 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
7175 	md->limit = 0;
7176 	md++;
7177 
7178 	md->base = sc->vres.ocq.start;
7179 	if (sc->vres.ocq.size)
7180 		md->limit = md->base + sc->vres.ocq.size - 1;
7181 	else
7182 		md->idx = nitems(region);  /* hide it */
7183 	md++;
7184 
7185 	md->base = sc->vres.key.start;
7186 	if (sc->vres.key.size)
7187 		md->limit = md->base + sc->vres.key.size - 1;
7188 	else
7189 		md->idx = nitems(region);  /* hide it */
7190 	md++;
7191 
7192 	/* add any address-space holes, there can be up to 3 */
7193 	for (n = 0; n < i - 1; n++)
7194 		if (avail[n].limit < avail[n + 1].base)
7195 			(md++)->base = avail[n].limit;
7196 	if (avail[n].limit)
7197 		(md++)->base = avail[n].limit;
7198 
7199 	n = md - mem;
7200 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
7201 
7202 	for (lo = 0; lo < i; lo++)
7203 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
7204 				avail[lo].limit - 1);
7205 
7206 	sbuf_printf(sb, "\n");
7207 	for (i = 0; i < n; i++) {
7208 		if (mem[i].idx >= nitems(region))
7209 			continue;                        /* skip holes */
7210 		if (!mem[i].limit)
7211 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
7212 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
7213 				mem[i].limit);
7214 	}
7215 
7216 	sbuf_printf(sb, "\n");
7217 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
7218 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
7219 	mem_region_show(sb, "uP RAM:", lo, hi);
7220 
7221 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
7222 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
7223 	mem_region_show(sb, "uP Extmem2:", lo, hi);
7224 
7225 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
7226 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
7227 		   G_PMRXMAXPAGE(lo),
7228 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
7229 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
7230 
7231 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
7232 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
7233 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
7234 		   G_PMTXMAXPAGE(lo),
7235 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
7236 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
7237 	sbuf_printf(sb, "%u p-structs\n",
7238 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
7239 
7240 	for (i = 0; i < 4; i++) {
7241 		if (chip_id(sc) > CHELSIO_T5)
7242 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
7243 		else
7244 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
7245 		if (is_t5(sc)) {
7246 			used = G_T5_USED(lo);
7247 			alloc = G_T5_ALLOC(lo);
7248 		} else {
7249 			used = G_USED(lo);
7250 			alloc = G_ALLOC(lo);
7251 		}
7252 		/* For T6 these are MAC buffer groups */
7253 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
7254 		    i, used, alloc);
7255 	}
7256 	for (i = 0; i < sc->chip_params->nchan; i++) {
7257 		if (chip_id(sc) > CHELSIO_T5)
7258 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
7259 		else
7260 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
7261 		if (is_t5(sc)) {
7262 			used = G_T5_USED(lo);
7263 			alloc = G_T5_ALLOC(lo);
7264 		} else {
7265 			used = G_USED(lo);
7266 			alloc = G_ALLOC(lo);
7267 		}
7268 		/* For T6 these are MAC buffer groups */
7269 		sbuf_printf(sb,
7270 		    "\nLoopback %d using %u pages out of %u allocated",
7271 		    i, used, alloc);
7272 	}
7273 
7274 	rc = sbuf_finish(sb);
7275 	sbuf_delete(sb);
7276 
7277 	return (rc);
7278 }
7279 
7280 static inline void
7281 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
7282 {
7283 	*mask = x | y;
7284 	y = htobe64(y);
7285 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
7286 }
7287 
7288 static int
7289 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
7290 {
7291 	struct adapter *sc = arg1;
7292 	struct sbuf *sb;
7293 	int rc, i;
7294 
7295 	MPASS(chip_id(sc) <= CHELSIO_T5);
7296 
7297 	rc = sysctl_wire_old_buffer(req, 0);
7298 	if (rc != 0)
7299 		return (rc);
7300 
7301 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7302 	if (sb == NULL)
7303 		return (ENOMEM);
7304 
7305 	sbuf_printf(sb,
7306 	    "Idx  Ethernet address     Mask     Vld Ports PF"
7307 	    "  VF              Replication             P0 P1 P2 P3  ML");
7308 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7309 		uint64_t tcamx, tcamy, mask;
7310 		uint32_t cls_lo, cls_hi;
7311 		uint8_t addr[ETHER_ADDR_LEN];
7312 
7313 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
7314 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
7315 		if (tcamx & tcamy)
7316 			continue;
7317 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7318 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7319 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7320 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
7321 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
7322 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
7323 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
7324 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
7325 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
7326 
7327 		if (cls_lo & F_REPLICATE) {
7328 			struct fw_ldst_cmd ldst_cmd;
7329 
7330 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7331 			ldst_cmd.op_to_addrspace =
7332 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7333 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7334 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7335 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7336 			ldst_cmd.u.mps.rplc.fid_idx =
7337 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7338 				V_FW_LDST_CMD_IDX(i));
7339 
7340 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7341 			    "t4mps");
7342 			if (rc)
7343 				break;
7344 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7345 			    sizeof(ldst_cmd), &ldst_cmd);
7346 			end_synchronized_op(sc, 0);
7347 
7348 			if (rc != 0) {
7349 				sbuf_printf(sb, "%36d", rc);
7350 				rc = 0;
7351 			} else {
7352 				sbuf_printf(sb, " %08x %08x %08x %08x",
7353 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7354 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7355 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7356 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7357 			}
7358 		} else
7359 			sbuf_printf(sb, "%36s", "");
7360 
7361 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
7362 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
7363 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
7364 	}
7365 
7366 	if (rc)
7367 		(void) sbuf_finish(sb);
7368 	else
7369 		rc = sbuf_finish(sb);
7370 	sbuf_delete(sb);
7371 
7372 	return (rc);
7373 }
7374 
7375 static int
7376 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
7377 {
7378 	struct adapter *sc = arg1;
7379 	struct sbuf *sb;
7380 	int rc, i;
7381 
7382 	MPASS(chip_id(sc) > CHELSIO_T5);
7383 
7384 	rc = sysctl_wire_old_buffer(req, 0);
7385 	if (rc != 0)
7386 		return (rc);
7387 
7388 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7389 	if (sb == NULL)
7390 		return (ENOMEM);
7391 
7392 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
7393 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
7394 	    "                           Replication"
7395 	    "                                    P0 P1 P2 P3  ML\n");
7396 
7397 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7398 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
7399 		uint16_t ivlan;
7400 		uint64_t tcamx, tcamy, val, mask;
7401 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
7402 		uint8_t addr[ETHER_ADDR_LEN];
7403 
7404 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
7405 		if (i < 256)
7406 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
7407 		else
7408 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
7409 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7410 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7411 		tcamy = G_DMACH(val) << 32;
7412 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7413 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7414 		lookup_type = G_DATALKPTYPE(data2);
7415 		port_num = G_DATAPORTNUM(data2);
7416 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7417 			/* Inner header VNI */
7418 			vniy = ((data2 & F_DATAVIDH2) << 23) |
7419 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7420 			dip_hit = data2 & F_DATADIPHIT;
7421 			vlan_vld = 0;
7422 		} else {
7423 			vniy = 0;
7424 			dip_hit = 0;
7425 			vlan_vld = data2 & F_DATAVIDH2;
7426 			ivlan = G_VIDL(val);
7427 		}
7428 
7429 		ctl |= V_CTLXYBITSEL(1);
7430 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7431 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7432 		tcamx = G_DMACH(val) << 32;
7433 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7434 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7435 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7436 			/* Inner header VNI mask */
7437 			vnix = ((data2 & F_DATAVIDH2) << 23) |
7438 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7439 		} else
7440 			vnix = 0;
7441 
7442 		if (tcamx & tcamy)
7443 			continue;
7444 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7445 
7446 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7447 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7448 
7449 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7450 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7451 			    "%012jx %06x %06x    -    -   %3c"
7452 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
7453 			    addr[1], addr[2], addr[3], addr[4], addr[5],
7454 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
7455 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7456 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7457 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7458 		} else {
7459 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7460 			    "%012jx    -       -   ", i, addr[0], addr[1],
7461 			    addr[2], addr[3], addr[4], addr[5],
7462 			    (uintmax_t)mask);
7463 
7464 			if (vlan_vld)
7465 				sbuf_printf(sb, "%4u   Y     ", ivlan);
7466 			else
7467 				sbuf_printf(sb, "  -    N     ");
7468 
7469 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
7470 			    lookup_type ? 'I' : 'O', port_num,
7471 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7472 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7473 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7474 		}
7475 
7476 
7477 		if (cls_lo & F_T6_REPLICATE) {
7478 			struct fw_ldst_cmd ldst_cmd;
7479 
7480 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7481 			ldst_cmd.op_to_addrspace =
7482 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7483 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7484 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7485 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7486 			ldst_cmd.u.mps.rplc.fid_idx =
7487 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7488 				V_FW_LDST_CMD_IDX(i));
7489 
7490 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7491 			    "t6mps");
7492 			if (rc)
7493 				break;
7494 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7495 			    sizeof(ldst_cmd), &ldst_cmd);
7496 			end_synchronized_op(sc, 0);
7497 
7498 			if (rc != 0) {
7499 				sbuf_printf(sb, "%72d", rc);
7500 				rc = 0;
7501 			} else {
7502 				sbuf_printf(sb, " %08x %08x %08x %08x"
7503 				    " %08x %08x %08x %08x",
7504 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
7505 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
7506 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
7507 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
7508 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7509 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7510 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7511 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7512 			}
7513 		} else
7514 			sbuf_printf(sb, "%72s", "");
7515 
7516 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
7517 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
7518 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
7519 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
7520 	}
7521 
7522 	if (rc)
7523 		(void) sbuf_finish(sb);
7524 	else
7525 		rc = sbuf_finish(sb);
7526 	sbuf_delete(sb);
7527 
7528 	return (rc);
7529 }
7530 
7531 static int
7532 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
7533 {
7534 	struct adapter *sc = arg1;
7535 	struct sbuf *sb;
7536 	int rc;
7537 	uint16_t mtus[NMTUS];
7538 
7539 	rc = sysctl_wire_old_buffer(req, 0);
7540 	if (rc != 0)
7541 		return (rc);
7542 
7543 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7544 	if (sb == NULL)
7545 		return (ENOMEM);
7546 
7547 	t4_read_mtu_tbl(sc, mtus, NULL);
7548 
7549 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
7550 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
7551 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
7552 	    mtus[14], mtus[15]);
7553 
7554 	rc = sbuf_finish(sb);
7555 	sbuf_delete(sb);
7556 
7557 	return (rc);
7558 }
7559 
7560 static int
7561 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
7562 {
7563 	struct adapter *sc = arg1;
7564 	struct sbuf *sb;
7565 	int rc, i;
7566 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
7567 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
7568 	static const char *tx_stats[MAX_PM_NSTATS] = {
7569 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
7570 		"Tx FIFO wait", NULL, "Tx latency"
7571 	};
7572 	static const char *rx_stats[MAX_PM_NSTATS] = {
7573 		"Read:", "Write bypass:", "Write mem:", "Flush:",
7574 		"Rx FIFO wait", NULL, "Rx latency"
7575 	};
7576 
7577 	rc = sysctl_wire_old_buffer(req, 0);
7578 	if (rc != 0)
7579 		return (rc);
7580 
7581 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7582 	if (sb == NULL)
7583 		return (ENOMEM);
7584 
7585 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
7586 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
7587 
7588 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
7589 	for (i = 0; i < 4; i++) {
7590 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7591 		    tx_cyc[i]);
7592 	}
7593 
7594 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
7595 	for (i = 0; i < 4; i++) {
7596 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7597 		    rx_cyc[i]);
7598 	}
7599 
7600 	if (chip_id(sc) > CHELSIO_T5) {
7601 		sbuf_printf(sb,
7602 		    "\n              Total wait      Total occupancy");
7603 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7604 		    tx_cyc[i]);
7605 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7606 		    rx_cyc[i]);
7607 
7608 		i += 2;
7609 		MPASS(i < nitems(tx_stats));
7610 
7611 		sbuf_printf(sb,
7612 		    "\n                   Reads           Total wait");
7613 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7614 		    tx_cyc[i]);
7615 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7616 		    rx_cyc[i]);
7617 	}
7618 
7619 	rc = sbuf_finish(sb);
7620 	sbuf_delete(sb);
7621 
7622 	return (rc);
7623 }
7624 
7625 static int
7626 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
7627 {
7628 	struct adapter *sc = arg1;
7629 	struct sbuf *sb;
7630 	int rc;
7631 	struct tp_rdma_stats stats;
7632 
7633 	rc = sysctl_wire_old_buffer(req, 0);
7634 	if (rc != 0)
7635 		return (rc);
7636 
7637 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7638 	if (sb == NULL)
7639 		return (ENOMEM);
7640 
7641 	mtx_lock(&sc->reg_lock);
7642 	t4_tp_get_rdma_stats(sc, &stats, 0);
7643 	mtx_unlock(&sc->reg_lock);
7644 
7645 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
7646 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
7647 
7648 	rc = sbuf_finish(sb);
7649 	sbuf_delete(sb);
7650 
7651 	return (rc);
7652 }
7653 
7654 static int
7655 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
7656 {
7657 	struct adapter *sc = arg1;
7658 	struct sbuf *sb;
7659 	int rc;
7660 	struct tp_tcp_stats v4, v6;
7661 
7662 	rc = sysctl_wire_old_buffer(req, 0);
7663 	if (rc != 0)
7664 		return (rc);
7665 
7666 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7667 	if (sb == NULL)
7668 		return (ENOMEM);
7669 
7670 	mtx_lock(&sc->reg_lock);
7671 	t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
7672 	mtx_unlock(&sc->reg_lock);
7673 
7674 	sbuf_printf(sb,
7675 	    "                                IP                 IPv6\n");
7676 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
7677 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
7678 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
7679 	    v4.tcp_in_segs, v6.tcp_in_segs);
7680 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
7681 	    v4.tcp_out_segs, v6.tcp_out_segs);
7682 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
7683 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
7684 
7685 	rc = sbuf_finish(sb);
7686 	sbuf_delete(sb);
7687 
7688 	return (rc);
7689 }
7690 
7691 static int
7692 sysctl_tids(SYSCTL_HANDLER_ARGS)
7693 {
7694 	struct adapter *sc = arg1;
7695 	struct sbuf *sb;
7696 	int rc;
7697 	struct tid_info *t = &sc->tids;
7698 
7699 	rc = sysctl_wire_old_buffer(req, 0);
7700 	if (rc != 0)
7701 		return (rc);
7702 
7703 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7704 	if (sb == NULL)
7705 		return (ENOMEM);
7706 
7707 	if (t->natids) {
7708 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
7709 		    t->atids_in_use);
7710 	}
7711 
7712 	if (t->ntids) {
7713 		sbuf_printf(sb, "TID range: ");
7714 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7715 			uint32_t b, hb;
7716 
7717 			if (chip_id(sc) <= CHELSIO_T5) {
7718 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
7719 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
7720 			} else {
7721 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
7722 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
7723 			}
7724 
7725 			if (b)
7726 				sbuf_printf(sb, "0-%u, ", b - 1);
7727 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
7728 		} else
7729 			sbuf_printf(sb, "0-%u", t->ntids - 1);
7730 		sbuf_printf(sb, ", in use: %u\n",
7731 		    atomic_load_acq_int(&t->tids_in_use));
7732 	}
7733 
7734 	if (t->nstids) {
7735 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
7736 		    t->stid_base + t->nstids - 1, t->stids_in_use);
7737 	}
7738 
7739 	if (t->nftids) {
7740 		sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
7741 		    t->ftid_base + t->nftids - 1);
7742 	}
7743 
7744 	if (t->netids) {
7745 		sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
7746 		    t->etid_base + t->netids - 1);
7747 	}
7748 
7749 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
7750 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
7751 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
7752 
7753 	rc = sbuf_finish(sb);
7754 	sbuf_delete(sb);
7755 
7756 	return (rc);
7757 }
7758 
7759 static int
7760 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
7761 {
7762 	struct adapter *sc = arg1;
7763 	struct sbuf *sb;
7764 	int rc;
7765 	struct tp_err_stats stats;
7766 
7767 	rc = sysctl_wire_old_buffer(req, 0);
7768 	if (rc != 0)
7769 		return (rc);
7770 
7771 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7772 	if (sb == NULL)
7773 		return (ENOMEM);
7774 
7775 	mtx_lock(&sc->reg_lock);
7776 	t4_tp_get_err_stats(sc, &stats, 0);
7777 	mtx_unlock(&sc->reg_lock);
7778 
7779 	if (sc->chip_params->nchan > 2) {
7780 		sbuf_printf(sb, "                 channel 0  channel 1"
7781 		    "  channel 2  channel 3\n");
7782 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
7783 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
7784 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
7785 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
7786 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
7787 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
7788 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
7789 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
7790 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
7791 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
7792 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
7793 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
7794 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
7795 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
7796 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
7797 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
7798 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
7799 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
7800 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
7801 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
7802 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
7803 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
7804 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
7805 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
7806 	} else {
7807 		sbuf_printf(sb, "                 channel 0  channel 1\n");
7808 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
7809 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
7810 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
7811 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
7812 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
7813 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
7814 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
7815 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
7816 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
7817 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
7818 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
7819 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
7820 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
7821 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
7822 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
7823 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
7824 	}
7825 
7826 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
7827 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
7828 
7829 	rc = sbuf_finish(sb);
7830 	sbuf_delete(sb);
7831 
7832 	return (rc);
7833 }
7834 
7835 static int
7836 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
7837 {
7838 	struct adapter *sc = arg1;
7839 	struct tp_params *tpp = &sc->params.tp;
7840 	u_int mask;
7841 	int rc;
7842 
7843 	mask = tpp->la_mask >> 16;
7844 	rc = sysctl_handle_int(oidp, &mask, 0, req);
7845 	if (rc != 0 || req->newptr == NULL)
7846 		return (rc);
7847 	if (mask > 0xffff)
7848 		return (EINVAL);
7849 	tpp->la_mask = mask << 16;
7850 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
7851 
7852 	return (0);
7853 }
7854 
7855 struct field_desc {
7856 	const char *name;
7857 	u_int start;
7858 	u_int width;
7859 };
7860 
7861 static void
7862 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
7863 {
7864 	char buf[32];
7865 	int line_size = 0;
7866 
7867 	while (f->name) {
7868 		uint64_t mask = (1ULL << f->width) - 1;
7869 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
7870 		    ((uintmax_t)v >> f->start) & mask);
7871 
7872 		if (line_size + len >= 79) {
7873 			line_size = 8;
7874 			sbuf_printf(sb, "\n        ");
7875 		}
7876 		sbuf_printf(sb, "%s ", buf);
7877 		line_size += len + 1;
7878 		f++;
7879 	}
7880 	sbuf_printf(sb, "\n");
7881 }
7882 
7883 static const struct field_desc tp_la0[] = {
7884 	{ "RcfOpCodeOut", 60, 4 },
7885 	{ "State", 56, 4 },
7886 	{ "WcfState", 52, 4 },
7887 	{ "RcfOpcSrcOut", 50, 2 },
7888 	{ "CRxError", 49, 1 },
7889 	{ "ERxError", 48, 1 },
7890 	{ "SanityFailed", 47, 1 },
7891 	{ "SpuriousMsg", 46, 1 },
7892 	{ "FlushInputMsg", 45, 1 },
7893 	{ "FlushInputCpl", 44, 1 },
7894 	{ "RssUpBit", 43, 1 },
7895 	{ "RssFilterHit", 42, 1 },
7896 	{ "Tid", 32, 10 },
7897 	{ "InitTcb", 31, 1 },
7898 	{ "LineNumber", 24, 7 },
7899 	{ "Emsg", 23, 1 },
7900 	{ "EdataOut", 22, 1 },
7901 	{ "Cmsg", 21, 1 },
7902 	{ "CdataOut", 20, 1 },
7903 	{ "EreadPdu", 19, 1 },
7904 	{ "CreadPdu", 18, 1 },
7905 	{ "TunnelPkt", 17, 1 },
7906 	{ "RcfPeerFin", 16, 1 },
7907 	{ "RcfReasonOut", 12, 4 },
7908 	{ "TxCchannel", 10, 2 },
7909 	{ "RcfTxChannel", 8, 2 },
7910 	{ "RxEchannel", 6, 2 },
7911 	{ "RcfRxChannel", 5, 1 },
7912 	{ "RcfDataOutSrdy", 4, 1 },
7913 	{ "RxDvld", 3, 1 },
7914 	{ "RxOoDvld", 2, 1 },
7915 	{ "RxCongestion", 1, 1 },
7916 	{ "TxCongestion", 0, 1 },
7917 	{ NULL }
7918 };
7919 
7920 static const struct field_desc tp_la1[] = {
7921 	{ "CplCmdIn", 56, 8 },
7922 	{ "CplCmdOut", 48, 8 },
7923 	{ "ESynOut", 47, 1 },
7924 	{ "EAckOut", 46, 1 },
7925 	{ "EFinOut", 45, 1 },
7926 	{ "ERstOut", 44, 1 },
7927 	{ "SynIn", 43, 1 },
7928 	{ "AckIn", 42, 1 },
7929 	{ "FinIn", 41, 1 },
7930 	{ "RstIn", 40, 1 },
7931 	{ "DataIn", 39, 1 },
7932 	{ "DataInVld", 38, 1 },
7933 	{ "PadIn", 37, 1 },
7934 	{ "RxBufEmpty", 36, 1 },
7935 	{ "RxDdp", 35, 1 },
7936 	{ "RxFbCongestion", 34, 1 },
7937 	{ "TxFbCongestion", 33, 1 },
7938 	{ "TxPktSumSrdy", 32, 1 },
7939 	{ "RcfUlpType", 28, 4 },
7940 	{ "Eread", 27, 1 },
7941 	{ "Ebypass", 26, 1 },
7942 	{ "Esave", 25, 1 },
7943 	{ "Static0", 24, 1 },
7944 	{ "Cread", 23, 1 },
7945 	{ "Cbypass", 22, 1 },
7946 	{ "Csave", 21, 1 },
7947 	{ "CPktOut", 20, 1 },
7948 	{ "RxPagePoolFull", 18, 2 },
7949 	{ "RxLpbkPkt", 17, 1 },
7950 	{ "TxLpbkPkt", 16, 1 },
7951 	{ "RxVfValid", 15, 1 },
7952 	{ "SynLearned", 14, 1 },
7953 	{ "SetDelEntry", 13, 1 },
7954 	{ "SetInvEntry", 12, 1 },
7955 	{ "CpcmdDvld", 11, 1 },
7956 	{ "CpcmdSave", 10, 1 },
7957 	{ "RxPstructsFull", 8, 2 },
7958 	{ "EpcmdDvld", 7, 1 },
7959 	{ "EpcmdFlush", 6, 1 },
7960 	{ "EpcmdTrimPrefix", 5, 1 },
7961 	{ "EpcmdTrimPostfix", 4, 1 },
7962 	{ "ERssIp4Pkt", 3, 1 },
7963 	{ "ERssIp6Pkt", 2, 1 },
7964 	{ "ERssTcpUdpPkt", 1, 1 },
7965 	{ "ERssFceFipPkt", 0, 1 },
7966 	{ NULL }
7967 };
7968 
7969 static const struct field_desc tp_la2[] = {
7970 	{ "CplCmdIn", 56, 8 },
7971 	{ "MpsVfVld", 55, 1 },
7972 	{ "MpsPf", 52, 3 },
7973 	{ "MpsVf", 44, 8 },
7974 	{ "SynIn", 43, 1 },
7975 	{ "AckIn", 42, 1 },
7976 	{ "FinIn", 41, 1 },
7977 	{ "RstIn", 40, 1 },
7978 	{ "DataIn", 39, 1 },
7979 	{ "DataInVld", 38, 1 },
7980 	{ "PadIn", 37, 1 },
7981 	{ "RxBufEmpty", 36, 1 },
7982 	{ "RxDdp", 35, 1 },
7983 	{ "RxFbCongestion", 34, 1 },
7984 	{ "TxFbCongestion", 33, 1 },
7985 	{ "TxPktSumSrdy", 32, 1 },
7986 	{ "RcfUlpType", 28, 4 },
7987 	{ "Eread", 27, 1 },
7988 	{ "Ebypass", 26, 1 },
7989 	{ "Esave", 25, 1 },
7990 	{ "Static0", 24, 1 },
7991 	{ "Cread", 23, 1 },
7992 	{ "Cbypass", 22, 1 },
7993 	{ "Csave", 21, 1 },
7994 	{ "CPktOut", 20, 1 },
7995 	{ "RxPagePoolFull", 18, 2 },
7996 	{ "RxLpbkPkt", 17, 1 },
7997 	{ "TxLpbkPkt", 16, 1 },
7998 	{ "RxVfValid", 15, 1 },
7999 	{ "SynLearned", 14, 1 },
8000 	{ "SetDelEntry", 13, 1 },
8001 	{ "SetInvEntry", 12, 1 },
8002 	{ "CpcmdDvld", 11, 1 },
8003 	{ "CpcmdSave", 10, 1 },
8004 	{ "RxPstructsFull", 8, 2 },
8005 	{ "EpcmdDvld", 7, 1 },
8006 	{ "EpcmdFlush", 6, 1 },
8007 	{ "EpcmdTrimPrefix", 5, 1 },
8008 	{ "EpcmdTrimPostfix", 4, 1 },
8009 	{ "ERssIp4Pkt", 3, 1 },
8010 	{ "ERssIp6Pkt", 2, 1 },
8011 	{ "ERssTcpUdpPkt", 1, 1 },
8012 	{ "ERssFceFipPkt", 0, 1 },
8013 	{ NULL }
8014 };
8015 
8016 static void
8017 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
8018 {
8019 
8020 	field_desc_show(sb, *p, tp_la0);
8021 }
8022 
8023 static void
8024 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
8025 {
8026 
8027 	if (idx)
8028 		sbuf_printf(sb, "\n");
8029 	field_desc_show(sb, p[0], tp_la0);
8030 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8031 		field_desc_show(sb, p[1], tp_la0);
8032 }
8033 
8034 static void
8035 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
8036 {
8037 
8038 	if (idx)
8039 		sbuf_printf(sb, "\n");
8040 	field_desc_show(sb, p[0], tp_la0);
8041 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8042 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
8043 }
8044 
8045 static int
8046 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
8047 {
8048 	struct adapter *sc = arg1;
8049 	struct sbuf *sb;
8050 	uint64_t *buf, *p;
8051 	int rc;
8052 	u_int i, inc;
8053 	void (*show_func)(struct sbuf *, uint64_t *, int);
8054 
8055 	rc = sysctl_wire_old_buffer(req, 0);
8056 	if (rc != 0)
8057 		return (rc);
8058 
8059 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8060 	if (sb == NULL)
8061 		return (ENOMEM);
8062 
8063 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
8064 
8065 	t4_tp_read_la(sc, buf, NULL);
8066 	p = buf;
8067 
8068 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
8069 	case 2:
8070 		inc = 2;
8071 		show_func = tp_la_show2;
8072 		break;
8073 	case 3:
8074 		inc = 2;
8075 		show_func = tp_la_show3;
8076 		break;
8077 	default:
8078 		inc = 1;
8079 		show_func = tp_la_show;
8080 	}
8081 
8082 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
8083 		(*show_func)(sb, p, i);
8084 
8085 	rc = sbuf_finish(sb);
8086 	sbuf_delete(sb);
8087 	free(buf, M_CXGBE);
8088 	return (rc);
8089 }
8090 
8091 static int
8092 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
8093 {
8094 	struct adapter *sc = arg1;
8095 	struct sbuf *sb;
8096 	int rc;
8097 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
8098 
8099 	rc = sysctl_wire_old_buffer(req, 0);
8100 	if (rc != 0)
8101 		return (rc);
8102 
8103 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8104 	if (sb == NULL)
8105 		return (ENOMEM);
8106 
8107 	t4_get_chan_txrate(sc, nrate, orate);
8108 
8109 	if (sc->chip_params->nchan > 2) {
8110 		sbuf_printf(sb, "              channel 0   channel 1"
8111 		    "   channel 2   channel 3\n");
8112 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
8113 		    nrate[0], nrate[1], nrate[2], nrate[3]);
8114 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
8115 		    orate[0], orate[1], orate[2], orate[3]);
8116 	} else {
8117 		sbuf_printf(sb, "              channel 0   channel 1\n");
8118 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
8119 		    nrate[0], nrate[1]);
8120 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
8121 		    orate[0], orate[1]);
8122 	}
8123 
8124 	rc = sbuf_finish(sb);
8125 	sbuf_delete(sb);
8126 
8127 	return (rc);
8128 }
8129 
8130 static int
8131 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
8132 {
8133 	struct adapter *sc = arg1;
8134 	struct sbuf *sb;
8135 	uint32_t *buf, *p;
8136 	int rc, i;
8137 
8138 	rc = sysctl_wire_old_buffer(req, 0);
8139 	if (rc != 0)
8140 		return (rc);
8141 
8142 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8143 	if (sb == NULL)
8144 		return (ENOMEM);
8145 
8146 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
8147 	    M_ZERO | M_WAITOK);
8148 
8149 	t4_ulprx_read_la(sc, buf);
8150 	p = buf;
8151 
8152 	sbuf_printf(sb, "      Pcmd        Type   Message"
8153 	    "                Data");
8154 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
8155 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
8156 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
8157 	}
8158 
8159 	rc = sbuf_finish(sb);
8160 	sbuf_delete(sb);
8161 	free(buf, M_CXGBE);
8162 	return (rc);
8163 }
8164 
8165 static int
8166 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
8167 {
8168 	struct adapter *sc = arg1;
8169 	struct sbuf *sb;
8170 	int rc, v;
8171 
8172 	MPASS(chip_id(sc) >= CHELSIO_T5);
8173 
8174 	rc = sysctl_wire_old_buffer(req, 0);
8175 	if (rc != 0)
8176 		return (rc);
8177 
8178 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8179 	if (sb == NULL)
8180 		return (ENOMEM);
8181 
8182 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
8183 	if (G_STATSOURCE_T5(v) == 7) {
8184 		int mode;
8185 
8186 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
8187 		if (mode == 0) {
8188 			sbuf_printf(sb, "total %d, incomplete %d",
8189 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8190 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8191 		} else if (mode == 1) {
8192 			sbuf_printf(sb, "total %d, data overflow %d",
8193 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8194 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8195 		} else {
8196 			sbuf_printf(sb, "unknown mode %d", mode);
8197 		}
8198 	}
8199 	rc = sbuf_finish(sb);
8200 	sbuf_delete(sb);
8201 
8202 	return (rc);
8203 }
8204 
8205 static int
8206 sysctl_tc_params(SYSCTL_HANDLER_ARGS)
8207 {
8208 	struct adapter *sc = arg1;
8209 	struct tx_cl_rl_params tc;
8210 	struct sbuf *sb;
8211 	int i, rc, port_id, mbps, gbps;
8212 
8213 	rc = sysctl_wire_old_buffer(req, 0);
8214 	if (rc != 0)
8215 		return (rc);
8216 
8217 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8218 	if (sb == NULL)
8219 		return (ENOMEM);
8220 
8221 	port_id = arg2 >> 16;
8222 	MPASS(port_id < sc->params.nports);
8223 	MPASS(sc->port[port_id] != NULL);
8224 	i = arg2 & 0xffff;
8225 	MPASS(i < sc->chip_params->nsched_cls);
8226 
8227 	mtx_lock(&sc->tc_lock);
8228 	tc = sc->port[port_id]->sched_params->cl_rl[i];
8229 	mtx_unlock(&sc->tc_lock);
8230 
8231 	if (tc.flags & TX_CLRL_ERROR) {
8232 		sbuf_printf(sb, "error");
8233 		goto done;
8234 	}
8235 
8236 	if (tc.ratemode == SCHED_CLASS_RATEMODE_REL) {
8237 		/* XXX: top speed or actual link speed? */
8238 		gbps = port_top_speed(sc->port[port_id]);
8239 		sbuf_printf(sb, " %u%% of %uGbps", tc.maxrate, gbps);
8240 	} else if (tc.ratemode == SCHED_CLASS_RATEMODE_ABS) {
8241 		switch (tc.rateunit) {
8242 		case SCHED_CLASS_RATEUNIT_BITS:
8243 			mbps = tc.maxrate / 1000;
8244 			gbps = tc.maxrate / 1000000;
8245 			if (tc.maxrate == gbps * 1000000)
8246 				sbuf_printf(sb, " %uGbps", gbps);
8247 			else if (tc.maxrate == mbps * 1000)
8248 				sbuf_printf(sb, " %uMbps", mbps);
8249 			else
8250 				sbuf_printf(sb, " %uKbps", tc.maxrate);
8251 			break;
8252 		case SCHED_CLASS_RATEUNIT_PKTS:
8253 			sbuf_printf(sb, " %upps", tc.maxrate);
8254 			break;
8255 		default:
8256 			rc = ENXIO;
8257 			goto done;
8258 		}
8259 	}
8260 
8261 	switch (tc.mode) {
8262 	case SCHED_CLASS_MODE_CLASS:
8263 		sbuf_printf(sb, " aggregate");
8264 		break;
8265 	case SCHED_CLASS_MODE_FLOW:
8266 		sbuf_printf(sb, " per-flow");
8267 		break;
8268 	default:
8269 		rc = ENXIO;
8270 		goto done;
8271 	}
8272 
8273 done:
8274 	if (rc == 0)
8275 		rc = sbuf_finish(sb);
8276 	sbuf_delete(sb);
8277 
8278 	return (rc);
8279 }
8280 #endif
8281 
8282 #ifdef TCP_OFFLOAD
8283 static int
8284 sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS)
8285 {
8286 	struct adapter *sc = arg1;
8287 	int *old_ports, *new_ports;
8288 	int i, new_count, rc;
8289 
8290 	if (req->newptr == NULL && req->oldptr == NULL)
8291 		return (SYSCTL_OUT(req, NULL, imax(sc->tt.num_tls_rx_ports, 1) *
8292 		    sizeof(sc->tt.tls_rx_ports[0])));
8293 
8294 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tlsrx");
8295 	if (rc)
8296 		return (rc);
8297 
8298 	if (sc->tt.num_tls_rx_ports == 0) {
8299 		i = -1;
8300 		rc = SYSCTL_OUT(req, &i, sizeof(i));
8301 	} else
8302 		rc = SYSCTL_OUT(req, sc->tt.tls_rx_ports,
8303 		    sc->tt.num_tls_rx_ports * sizeof(sc->tt.tls_rx_ports[0]));
8304 	if (rc == 0 && req->newptr != NULL) {
8305 		new_count = req->newlen / sizeof(new_ports[0]);
8306 		new_ports = malloc(new_count * sizeof(new_ports[0]), M_CXGBE,
8307 		    M_WAITOK);
8308 		rc = SYSCTL_IN(req, new_ports, new_count *
8309 		    sizeof(new_ports[0]));
8310 		if (rc)
8311 			goto err;
8312 
8313 		/* Allow setting to a single '-1' to clear the list. */
8314 		if (new_count == 1 && new_ports[0] == -1) {
8315 			ADAPTER_LOCK(sc);
8316 			old_ports = sc->tt.tls_rx_ports;
8317 			sc->tt.tls_rx_ports = NULL;
8318 			sc->tt.num_tls_rx_ports = 0;
8319 			ADAPTER_UNLOCK(sc);
8320 			free(old_ports, M_CXGBE);
8321 		} else {
8322 			for (i = 0; i < new_count; i++) {
8323 				if (new_ports[i] < 1 ||
8324 				    new_ports[i] > IPPORT_MAX) {
8325 					rc = EINVAL;
8326 					goto err;
8327 				}
8328 			}
8329 
8330 			ADAPTER_LOCK(sc);
8331 			old_ports = sc->tt.tls_rx_ports;
8332 			sc->tt.tls_rx_ports = new_ports;
8333 			sc->tt.num_tls_rx_ports = new_count;
8334 			ADAPTER_UNLOCK(sc);
8335 			free(old_ports, M_CXGBE);
8336 			new_ports = NULL;
8337 		}
8338 	err:
8339 		free(new_ports, M_CXGBE);
8340 	}
8341 	end_synchronized_op(sc, 0);
8342 	return (rc);
8343 }
8344 
8345 static void
8346 unit_conv(char *buf, size_t len, u_int val, u_int factor)
8347 {
8348 	u_int rem = val % factor;
8349 
8350 	if (rem == 0)
8351 		snprintf(buf, len, "%u", val / factor);
8352 	else {
8353 		while (rem % 10 == 0)
8354 			rem /= 10;
8355 		snprintf(buf, len, "%u.%u", val / factor, rem);
8356 	}
8357 }
8358 
8359 static int
8360 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
8361 {
8362 	struct adapter *sc = arg1;
8363 	char buf[16];
8364 	u_int res, re;
8365 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8366 
8367 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8368 	switch (arg2) {
8369 	case 0:
8370 		/* timer_tick */
8371 		re = G_TIMERRESOLUTION(res);
8372 		break;
8373 	case 1:
8374 		/* TCP timestamp tick */
8375 		re = G_TIMESTAMPRESOLUTION(res);
8376 		break;
8377 	case 2:
8378 		/* DACK tick */
8379 		re = G_DELAYEDACKRESOLUTION(res);
8380 		break;
8381 	default:
8382 		return (EDOOFUS);
8383 	}
8384 
8385 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
8386 
8387 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
8388 }
8389 
8390 static int
8391 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
8392 {
8393 	struct adapter *sc = arg1;
8394 	u_int res, dack_re, v;
8395 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8396 
8397 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8398 	dack_re = G_DELAYEDACKRESOLUTION(res);
8399 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
8400 
8401 	return (sysctl_handle_int(oidp, &v, 0, req));
8402 }
8403 
8404 static int
8405 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
8406 {
8407 	struct adapter *sc = arg1;
8408 	int reg = arg2;
8409 	u_int tre;
8410 	u_long tp_tick_us, v;
8411 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8412 
8413 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
8414 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
8415 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
8416 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
8417 
8418 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
8419 	tp_tick_us = (cclk_ps << tre) / 1000000;
8420 
8421 	if (reg == A_TP_INIT_SRTT)
8422 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
8423 	else
8424 		v = tp_tick_us * t4_read_reg(sc, reg);
8425 
8426 	return (sysctl_handle_long(oidp, &v, 0, req));
8427 }
8428 
8429 /*
8430  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
8431  * passed to this function.
8432  */
8433 static int
8434 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
8435 {
8436 	struct adapter *sc = arg1;
8437 	int idx = arg2;
8438 	u_int v;
8439 
8440 	MPASS(idx >= 0 && idx <= 24);
8441 
8442 	v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
8443 
8444 	return (sysctl_handle_int(oidp, &v, 0, req));
8445 }
8446 
8447 static int
8448 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
8449 {
8450 	struct adapter *sc = arg1;
8451 	int idx = arg2;
8452 	u_int shift, v, r;
8453 
8454 	MPASS(idx >= 0 && idx < 16);
8455 
8456 	r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
8457 	shift = (idx & 3) << 3;
8458 	v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
8459 
8460 	return (sysctl_handle_int(oidp, &v, 0, req));
8461 }
8462 
8463 static int
8464 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
8465 {
8466 	struct vi_info *vi = arg1;
8467 	struct adapter *sc = vi->pi->adapter;
8468 	int idx, rc, i;
8469 	struct sge_ofld_rxq *ofld_rxq;
8470 	uint8_t v;
8471 
8472 	idx = vi->ofld_tmr_idx;
8473 
8474 	rc = sysctl_handle_int(oidp, &idx, 0, req);
8475 	if (rc != 0 || req->newptr == NULL)
8476 		return (rc);
8477 
8478 	if (idx < 0 || idx >= SGE_NTIMERS)
8479 		return (EINVAL);
8480 
8481 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8482 	    "t4otmr");
8483 	if (rc)
8484 		return (rc);
8485 
8486 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
8487 	for_each_ofld_rxq(vi, i, ofld_rxq) {
8488 #ifdef atomic_store_rel_8
8489 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
8490 #else
8491 		ofld_rxq->iq.intr_params = v;
8492 #endif
8493 	}
8494 	vi->ofld_tmr_idx = idx;
8495 
8496 	end_synchronized_op(sc, LOCK_HELD);
8497 	return (0);
8498 }
8499 
8500 static int
8501 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
8502 {
8503 	struct vi_info *vi = arg1;
8504 	struct adapter *sc = vi->pi->adapter;
8505 	int idx, rc;
8506 
8507 	idx = vi->ofld_pktc_idx;
8508 
8509 	rc = sysctl_handle_int(oidp, &idx, 0, req);
8510 	if (rc != 0 || req->newptr == NULL)
8511 		return (rc);
8512 
8513 	if (idx < -1 || idx >= SGE_NCOUNTERS)
8514 		return (EINVAL);
8515 
8516 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8517 	    "t4opktc");
8518 	if (rc)
8519 		return (rc);
8520 
8521 	if (vi->flags & VI_INIT_DONE)
8522 		rc = EBUSY; /* cannot be changed once the queues are created */
8523 	else
8524 		vi->ofld_pktc_idx = idx;
8525 
8526 	end_synchronized_op(sc, LOCK_HELD);
8527 	return (rc);
8528 }
8529 #endif
8530 
8531 static uint32_t
8532 fconf_iconf_to_mode(uint32_t fconf, uint32_t iconf)
8533 {
8534 	uint32_t mode;
8535 
8536 	mode = T4_FILTER_IPv4 | T4_FILTER_IPv6 | T4_FILTER_IP_SADDR |
8537 	    T4_FILTER_IP_DADDR | T4_FILTER_IP_SPORT | T4_FILTER_IP_DPORT;
8538 
8539 	if (fconf & F_FRAGMENTATION)
8540 		mode |= T4_FILTER_IP_FRAGMENT;
8541 
8542 	if (fconf & F_MPSHITTYPE)
8543 		mode |= T4_FILTER_MPS_HIT_TYPE;
8544 
8545 	if (fconf & F_MACMATCH)
8546 		mode |= T4_FILTER_MAC_IDX;
8547 
8548 	if (fconf & F_ETHERTYPE)
8549 		mode |= T4_FILTER_ETH_TYPE;
8550 
8551 	if (fconf & F_PROTOCOL)
8552 		mode |= T4_FILTER_IP_PROTO;
8553 
8554 	if (fconf & F_TOS)
8555 		mode |= T4_FILTER_IP_TOS;
8556 
8557 	if (fconf & F_VLAN)
8558 		mode |= T4_FILTER_VLAN;
8559 
8560 	if (fconf & F_VNIC_ID) {
8561 		mode |= T4_FILTER_VNIC;
8562 		if (iconf & F_VNIC)
8563 			mode |= T4_FILTER_IC_VNIC;
8564 	}
8565 
8566 	if (fconf & F_PORT)
8567 		mode |= T4_FILTER_PORT;
8568 
8569 	if (fconf & F_FCOE)
8570 		mode |= T4_FILTER_FCoE;
8571 
8572 	return (mode);
8573 }
8574 
8575 static uint32_t
8576 mode_to_fconf(uint32_t mode)
8577 {
8578 	uint32_t fconf = 0;
8579 
8580 	if (mode & T4_FILTER_IP_FRAGMENT)
8581 		fconf |= F_FRAGMENTATION;
8582 
8583 	if (mode & T4_FILTER_MPS_HIT_TYPE)
8584 		fconf |= F_MPSHITTYPE;
8585 
8586 	if (mode & T4_FILTER_MAC_IDX)
8587 		fconf |= F_MACMATCH;
8588 
8589 	if (mode & T4_FILTER_ETH_TYPE)
8590 		fconf |= F_ETHERTYPE;
8591 
8592 	if (mode & T4_FILTER_IP_PROTO)
8593 		fconf |= F_PROTOCOL;
8594 
8595 	if (mode & T4_FILTER_IP_TOS)
8596 		fconf |= F_TOS;
8597 
8598 	if (mode & T4_FILTER_VLAN)
8599 		fconf |= F_VLAN;
8600 
8601 	if (mode & T4_FILTER_VNIC)
8602 		fconf |= F_VNIC_ID;
8603 
8604 	if (mode & T4_FILTER_PORT)
8605 		fconf |= F_PORT;
8606 
8607 	if (mode & T4_FILTER_FCoE)
8608 		fconf |= F_FCOE;
8609 
8610 	return (fconf);
8611 }
8612 
8613 static uint32_t
8614 mode_to_iconf(uint32_t mode)
8615 {
8616 
8617 	if (mode & T4_FILTER_IC_VNIC)
8618 		return (F_VNIC);
8619 	return (0);
8620 }
8621 
8622 static int check_fspec_against_fconf_iconf(struct adapter *sc,
8623     struct t4_filter_specification *fs)
8624 {
8625 	struct tp_params *tpp = &sc->params.tp;
8626 	uint32_t fconf = 0;
8627 
8628 	if (fs->val.frag || fs->mask.frag)
8629 		fconf |= F_FRAGMENTATION;
8630 
8631 	if (fs->val.matchtype || fs->mask.matchtype)
8632 		fconf |= F_MPSHITTYPE;
8633 
8634 	if (fs->val.macidx || fs->mask.macidx)
8635 		fconf |= F_MACMATCH;
8636 
8637 	if (fs->val.ethtype || fs->mask.ethtype)
8638 		fconf |= F_ETHERTYPE;
8639 
8640 	if (fs->val.proto || fs->mask.proto)
8641 		fconf |= F_PROTOCOL;
8642 
8643 	if (fs->val.tos || fs->mask.tos)
8644 		fconf |= F_TOS;
8645 
8646 	if (fs->val.vlan_vld || fs->mask.vlan_vld)
8647 		fconf |= F_VLAN;
8648 
8649 	if (fs->val.ovlan_vld || fs->mask.ovlan_vld) {
8650 		fconf |= F_VNIC_ID;
8651 		if (tpp->ingress_config & F_VNIC)
8652 			return (EINVAL);
8653 	}
8654 
8655 	if (fs->val.pfvf_vld || fs->mask.pfvf_vld) {
8656 		fconf |= F_VNIC_ID;
8657 		if ((tpp->ingress_config & F_VNIC) == 0)
8658 			return (EINVAL);
8659 	}
8660 
8661 	if (fs->val.iport || fs->mask.iport)
8662 		fconf |= F_PORT;
8663 
8664 	if (fs->val.fcoe || fs->mask.fcoe)
8665 		fconf |= F_FCOE;
8666 
8667 	if ((tpp->vlan_pri_map | fconf) != tpp->vlan_pri_map)
8668 		return (E2BIG);
8669 
8670 	return (0);
8671 }
8672 
8673 static int
8674 get_filter_mode(struct adapter *sc, uint32_t *mode)
8675 {
8676 	struct tp_params *tpp = &sc->params.tp;
8677 
8678 	/*
8679 	 * We trust the cached values of the relevant TP registers.  This means
8680 	 * things work reliably only if writes to those registers are always via
8681 	 * t4_set_filter_mode.
8682 	 */
8683 	*mode = fconf_iconf_to_mode(tpp->vlan_pri_map, tpp->ingress_config);
8684 
8685 	return (0);
8686 }
8687 
8688 static int
8689 set_filter_mode(struct adapter *sc, uint32_t mode)
8690 {
8691 	struct tp_params *tpp = &sc->params.tp;
8692 	uint32_t fconf, iconf;
8693 	int rc;
8694 
8695 	iconf = mode_to_iconf(mode);
8696 	if ((iconf ^ tpp->ingress_config) & F_VNIC) {
8697 		/*
8698 		 * For now we just complain if A_TP_INGRESS_CONFIG is not
8699 		 * already set to the correct value for the requested filter
8700 		 * mode.  It's not clear if it's safe to write to this register
8701 		 * on the fly.  (And we trust the cached value of the register).
8702 		 */
8703 		return (EBUSY);
8704 	}
8705 
8706 	fconf = mode_to_fconf(mode);
8707 
8708 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8709 	    "t4setfm");
8710 	if (rc)
8711 		return (rc);
8712 
8713 	if (sc->tids.ftids_in_use > 0) {
8714 		rc = EBUSY;
8715 		goto done;
8716 	}
8717 
8718 #ifdef TCP_OFFLOAD
8719 	if (uld_active(sc, ULD_TOM)) {
8720 		rc = EBUSY;
8721 		goto done;
8722 	}
8723 #endif
8724 
8725 	rc = -t4_set_filter_mode(sc, fconf, true);
8726 done:
8727 	end_synchronized_op(sc, LOCK_HELD);
8728 	return (rc);
8729 }
8730 
8731 static inline uint64_t
8732 get_filter_hits(struct adapter *sc, uint32_t fid)
8733 {
8734 	uint32_t tcb_addr;
8735 
8736 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE) +
8737 	    (fid + sc->tids.ftid_base) * TCB_SIZE;
8738 
8739 	if (is_t4(sc)) {
8740 		uint64_t hits;
8741 
8742 		read_via_memwin(sc, 0, tcb_addr + 16, (uint32_t *)&hits, 8);
8743 		return (be64toh(hits));
8744 	} else {
8745 		uint32_t hits;
8746 
8747 		read_via_memwin(sc, 0, tcb_addr + 24, &hits, 4);
8748 		return (be32toh(hits));
8749 	}
8750 }
8751 
8752 static int
8753 get_filter(struct adapter *sc, struct t4_filter *t)
8754 {
8755 	int i, rc, nfilters = sc->tids.nftids;
8756 	struct filter_entry *f;
8757 
8758 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8759 	    "t4getf");
8760 	if (rc)
8761 		return (rc);
8762 
8763 	if (sc->tids.ftids_in_use == 0 || sc->tids.ftid_tab == NULL ||
8764 	    t->idx >= nfilters) {
8765 		t->idx = 0xffffffff;
8766 		goto done;
8767 	}
8768 
8769 	f = &sc->tids.ftid_tab[t->idx];
8770 	for (i = t->idx; i < nfilters; i++, f++) {
8771 		if (f->valid) {
8772 			t->idx = i;
8773 			t->l2tidx = f->l2t ? f->l2t->idx : 0;
8774 			t->smtidx = f->smtidx;
8775 			if (f->fs.hitcnts)
8776 				t->hits = get_filter_hits(sc, t->idx);
8777 			else
8778 				t->hits = UINT64_MAX;
8779 			t->fs = f->fs;
8780 
8781 			goto done;
8782 		}
8783 	}
8784 
8785 	t->idx = 0xffffffff;
8786 done:
8787 	end_synchronized_op(sc, LOCK_HELD);
8788 	return (0);
8789 }
8790 
8791 static int
8792 set_filter(struct adapter *sc, struct t4_filter *t)
8793 {
8794 	unsigned int nfilters, nports;
8795 	struct filter_entry *f;
8796 	int i, rc;
8797 
8798 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setf");
8799 	if (rc)
8800 		return (rc);
8801 
8802 	nfilters = sc->tids.nftids;
8803 	nports = sc->params.nports;
8804 
8805 	if (nfilters == 0) {
8806 		rc = ENOTSUP;
8807 		goto done;
8808 	}
8809 
8810 	if (t->idx >= nfilters) {
8811 		rc = EINVAL;
8812 		goto done;
8813 	}
8814 
8815 	/* Validate against the global filter mode and ingress config */
8816 	rc = check_fspec_against_fconf_iconf(sc, &t->fs);
8817 	if (rc != 0)
8818 		goto done;
8819 
8820 	if (t->fs.action == FILTER_SWITCH && t->fs.eport >= nports) {
8821 		rc = EINVAL;
8822 		goto done;
8823 	}
8824 
8825 	if (t->fs.val.iport >= nports) {
8826 		rc = EINVAL;
8827 		goto done;
8828 	}
8829 
8830 	/* Can't specify an iq if not steering to it */
8831 	if (!t->fs.dirsteer && t->fs.iq) {
8832 		rc = EINVAL;
8833 		goto done;
8834 	}
8835 
8836 	/* IPv6 filter idx must be 4 aligned */
8837 	if (t->fs.type == 1 &&
8838 	    ((t->idx & 0x3) || t->idx + 4 >= nfilters)) {
8839 		rc = EINVAL;
8840 		goto done;
8841 	}
8842 
8843 	if (!(sc->flags & FULL_INIT_DONE) &&
8844 	    ((rc = adapter_full_init(sc)) != 0))
8845 		goto done;
8846 
8847 	if (sc->tids.ftid_tab == NULL) {
8848 		KASSERT(sc->tids.ftids_in_use == 0,
8849 		    ("%s: no memory allocated but filters_in_use > 0",
8850 		    __func__));
8851 
8852 		sc->tids.ftid_tab = malloc(sizeof (struct filter_entry) *
8853 		    nfilters, M_CXGBE, M_NOWAIT | M_ZERO);
8854 		if (sc->tids.ftid_tab == NULL) {
8855 			rc = ENOMEM;
8856 			goto done;
8857 		}
8858 		mtx_init(&sc->tids.ftid_lock, "T4 filters", 0, MTX_DEF);
8859 	}
8860 
8861 	for (i = 0; i < 4; i++) {
8862 		f = &sc->tids.ftid_tab[t->idx + i];
8863 
8864 		if (f->pending || f->valid) {
8865 			rc = EBUSY;
8866 			goto done;
8867 		}
8868 		if (f->locked) {
8869 			rc = EPERM;
8870 			goto done;
8871 		}
8872 
8873 		if (t->fs.type == 0)
8874 			break;
8875 	}
8876 
8877 	f = &sc->tids.ftid_tab[t->idx];
8878 	f->fs = t->fs;
8879 
8880 	rc = set_filter_wr(sc, t->idx);
8881 done:
8882 	end_synchronized_op(sc, 0);
8883 
8884 	if (rc == 0) {
8885 		mtx_lock(&sc->tids.ftid_lock);
8886 		for (;;) {
8887 			if (f->pending == 0) {
8888 				rc = f->valid ? 0 : EIO;
8889 				break;
8890 			}
8891 
8892 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8893 			    PCATCH, "t4setfw", 0)) {
8894 				rc = EINPROGRESS;
8895 				break;
8896 			}
8897 		}
8898 		mtx_unlock(&sc->tids.ftid_lock);
8899 	}
8900 	return (rc);
8901 }
8902 
8903 static int
8904 del_filter(struct adapter *sc, struct t4_filter *t)
8905 {
8906 	unsigned int nfilters;
8907 	struct filter_entry *f;
8908 	int rc;
8909 
8910 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4delf");
8911 	if (rc)
8912 		return (rc);
8913 
8914 	nfilters = sc->tids.nftids;
8915 
8916 	if (nfilters == 0) {
8917 		rc = ENOTSUP;
8918 		goto done;
8919 	}
8920 
8921 	if (sc->tids.ftid_tab == NULL || sc->tids.ftids_in_use == 0 ||
8922 	    t->idx >= nfilters) {
8923 		rc = EINVAL;
8924 		goto done;
8925 	}
8926 
8927 	if (!(sc->flags & FULL_INIT_DONE)) {
8928 		rc = EAGAIN;
8929 		goto done;
8930 	}
8931 
8932 	f = &sc->tids.ftid_tab[t->idx];
8933 
8934 	if (f->pending) {
8935 		rc = EBUSY;
8936 		goto done;
8937 	}
8938 	if (f->locked) {
8939 		rc = EPERM;
8940 		goto done;
8941 	}
8942 
8943 	if (f->valid) {
8944 		t->fs = f->fs;	/* extra info for the caller */
8945 		rc = del_filter_wr(sc, t->idx);
8946 	}
8947 
8948 done:
8949 	end_synchronized_op(sc, 0);
8950 
8951 	if (rc == 0) {
8952 		mtx_lock(&sc->tids.ftid_lock);
8953 		for (;;) {
8954 			if (f->pending == 0) {
8955 				rc = f->valid ? EIO : 0;
8956 				break;
8957 			}
8958 
8959 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8960 			    PCATCH, "t4delfw", 0)) {
8961 				rc = EINPROGRESS;
8962 				break;
8963 			}
8964 		}
8965 		mtx_unlock(&sc->tids.ftid_lock);
8966 	}
8967 
8968 	return (rc);
8969 }
8970 
8971 static void
8972 clear_filter(struct filter_entry *f)
8973 {
8974 	if (f->l2t)
8975 		t4_l2t_release(f->l2t);
8976 
8977 	bzero(f, sizeof (*f));
8978 }
8979 
8980 static int
8981 set_filter_wr(struct adapter *sc, int fidx)
8982 {
8983 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8984 	struct fw_filter_wr *fwr;
8985 	unsigned int ftid, vnic_vld, vnic_vld_mask;
8986 	struct wrq_cookie cookie;
8987 
8988 	ASSERT_SYNCHRONIZED_OP(sc);
8989 
8990 	if (f->fs.newdmac || f->fs.newvlan) {
8991 		/* This filter needs an L2T entry; allocate one. */
8992 		f->l2t = t4_l2t_alloc_switching(sc->l2t);
8993 		if (f->l2t == NULL)
8994 			return (EAGAIN);
8995 		if (t4_l2t_set_switching(sc, f->l2t, f->fs.vlan, f->fs.eport,
8996 		    f->fs.dmac)) {
8997 			t4_l2t_release(f->l2t);
8998 			f->l2t = NULL;
8999 			return (ENOMEM);
9000 		}
9001 	}
9002 
9003 	/* Already validated against fconf, iconf */
9004 	MPASS((f->fs.val.pfvf_vld & f->fs.val.ovlan_vld) == 0);
9005 	MPASS((f->fs.mask.pfvf_vld & f->fs.mask.ovlan_vld) == 0);
9006 	if (f->fs.val.pfvf_vld || f->fs.val.ovlan_vld)
9007 		vnic_vld = 1;
9008 	else
9009 		vnic_vld = 0;
9010 	if (f->fs.mask.pfvf_vld || f->fs.mask.ovlan_vld)
9011 		vnic_vld_mask = 1;
9012 	else
9013 		vnic_vld_mask = 0;
9014 
9015 	ftid = sc->tids.ftid_base + fidx;
9016 
9017 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
9018 	if (fwr == NULL)
9019 		return (ENOMEM);
9020 	bzero(fwr, sizeof(*fwr));
9021 
9022 	fwr->op_pkd = htobe32(V_FW_WR_OP(FW_FILTER_WR));
9023 	fwr->len16_pkd = htobe32(FW_LEN16(*fwr));
9024 	fwr->tid_to_iq =
9025 	    htobe32(V_FW_FILTER_WR_TID(ftid) |
9026 		V_FW_FILTER_WR_RQTYPE(f->fs.type) |
9027 		V_FW_FILTER_WR_NOREPLY(0) |
9028 		V_FW_FILTER_WR_IQ(f->fs.iq));
9029 	fwr->del_filter_to_l2tix =
9030 	    htobe32(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
9031 		V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
9032 		V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
9033 		V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
9034 		V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
9035 		V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
9036 		V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
9037 		V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
9038 		V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
9039 		    f->fs.newvlan == VLAN_REWRITE) |
9040 		V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
9041 		    f->fs.newvlan == VLAN_REWRITE) |
9042 		V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
9043 		V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
9044 		V_FW_FILTER_WR_PRIO(f->fs.prio) |
9045 		V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
9046 	fwr->ethtype = htobe16(f->fs.val.ethtype);
9047 	fwr->ethtypem = htobe16(f->fs.mask.ethtype);
9048 	fwr->frag_to_ovlan_vldm =
9049 	    (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
9050 		V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
9051 		V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.vlan_vld) |
9052 		V_FW_FILTER_WR_OVLAN_VLD(vnic_vld) |
9053 		V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.vlan_vld) |
9054 		V_FW_FILTER_WR_OVLAN_VLDM(vnic_vld_mask));
9055 	fwr->smac_sel = 0;
9056 	fwr->rx_chan_rx_rpl_iq = htobe16(V_FW_FILTER_WR_RX_CHAN(0) |
9057 	    V_FW_FILTER_WR_RX_RPL_IQ(sc->sge.fwq.abs_id));
9058 	fwr->maci_to_matchtypem =
9059 	    htobe32(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
9060 		V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
9061 		V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
9062 		V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
9063 		V_FW_FILTER_WR_PORT(f->fs.val.iport) |
9064 		V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
9065 		V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
9066 		V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
9067 	fwr->ptcl = f->fs.val.proto;
9068 	fwr->ptclm = f->fs.mask.proto;
9069 	fwr->ttyp = f->fs.val.tos;
9070 	fwr->ttypm = f->fs.mask.tos;
9071 	fwr->ivlan = htobe16(f->fs.val.vlan);
9072 	fwr->ivlanm = htobe16(f->fs.mask.vlan);
9073 	fwr->ovlan = htobe16(f->fs.val.vnic);
9074 	fwr->ovlanm = htobe16(f->fs.mask.vnic);
9075 	bcopy(f->fs.val.dip, fwr->lip, sizeof (fwr->lip));
9076 	bcopy(f->fs.mask.dip, fwr->lipm, sizeof (fwr->lipm));
9077 	bcopy(f->fs.val.sip, fwr->fip, sizeof (fwr->fip));
9078 	bcopy(f->fs.mask.sip, fwr->fipm, sizeof (fwr->fipm));
9079 	fwr->lp = htobe16(f->fs.val.dport);
9080 	fwr->lpm = htobe16(f->fs.mask.dport);
9081 	fwr->fp = htobe16(f->fs.val.sport);
9082 	fwr->fpm = htobe16(f->fs.mask.sport);
9083 	if (f->fs.newsmac)
9084 		bcopy(f->fs.smac, fwr->sma, sizeof (fwr->sma));
9085 
9086 	f->pending = 1;
9087 	sc->tids.ftids_in_use++;
9088 
9089 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
9090 	return (0);
9091 }
9092 
9093 static int
9094 del_filter_wr(struct adapter *sc, int fidx)
9095 {
9096 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
9097 	struct fw_filter_wr *fwr;
9098 	unsigned int ftid;
9099 	struct wrq_cookie cookie;
9100 
9101 	ftid = sc->tids.ftid_base + fidx;
9102 
9103 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
9104 	if (fwr == NULL)
9105 		return (ENOMEM);
9106 	bzero(fwr, sizeof (*fwr));
9107 
9108 	t4_mk_filtdelwr(ftid, fwr, sc->sge.fwq.abs_id);
9109 
9110 	f->pending = 1;
9111 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
9112 	return (0);
9113 }
9114 
9115 int
9116 t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
9117 {
9118 	struct adapter *sc = iq->adapter;
9119 	const struct cpl_set_tcb_rpl *rpl = (const void *)(rss + 1);
9120 	unsigned int idx = GET_TID(rpl);
9121 	unsigned int rc;
9122 	struct filter_entry *f;
9123 
9124 	KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
9125 	    rss->opcode));
9126 	MPASS(iq == &sc->sge.fwq);
9127 	MPASS(is_ftid(sc, idx));
9128 
9129 	idx -= sc->tids.ftid_base;
9130 	f = &sc->tids.ftid_tab[idx];
9131 	rc = G_COOKIE(rpl->cookie);
9132 
9133 	mtx_lock(&sc->tids.ftid_lock);
9134 	if (rc == FW_FILTER_WR_FLT_ADDED) {
9135 		KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
9136 		    __func__, idx));
9137 		f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
9138 		f->pending = 0;  /* asynchronous setup completed */
9139 		f->valid = 1;
9140 	} else {
9141 		if (rc != FW_FILTER_WR_FLT_DELETED) {
9142 			/* Add or delete failed, display an error */
9143 			log(LOG_ERR,
9144 			    "filter %u setup failed with error %u\n",
9145 			    idx, rc);
9146 		}
9147 
9148 		clear_filter(f);
9149 		sc->tids.ftids_in_use--;
9150 	}
9151 	wakeup(&sc->tids.ftid_tab);
9152 	mtx_unlock(&sc->tids.ftid_lock);
9153 
9154 	return (0);
9155 }
9156 
9157 static int
9158 set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
9159 {
9160 
9161 	MPASS(iq->set_tcb_rpl != NULL);
9162 	return (iq->set_tcb_rpl(iq, rss, m));
9163 }
9164 
9165 static int
9166 l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
9167 {
9168 
9169 	MPASS(iq->l2t_write_rpl != NULL);
9170 	return (iq->l2t_write_rpl(iq, rss, m));
9171 }
9172 
9173 static int
9174 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
9175 {
9176 	int rc;
9177 
9178 	if (cntxt->cid > M_CTXTQID)
9179 		return (EINVAL);
9180 
9181 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
9182 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
9183 		return (EINVAL);
9184 
9185 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
9186 	if (rc)
9187 		return (rc);
9188 
9189 	if (sc->flags & FW_OK) {
9190 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
9191 		    &cntxt->data[0]);
9192 		if (rc == 0)
9193 			goto done;
9194 	}
9195 
9196 	/*
9197 	 * Read via firmware failed or wasn't even attempted.  Read directly via
9198 	 * the backdoor.
9199 	 */
9200 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
9201 done:
9202 	end_synchronized_op(sc, 0);
9203 	return (rc);
9204 }
9205 
9206 static int
9207 load_fw(struct adapter *sc, struct t4_data *fw)
9208 {
9209 	int rc;
9210 	uint8_t *fw_data;
9211 
9212 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
9213 	if (rc)
9214 		return (rc);
9215 
9216 	/*
9217 	 * The firmware, with the sole exception of the memory parity error
9218 	 * handler, runs from memory and not flash.  It is almost always safe to
9219 	 * install a new firmware on a running system.  Just set bit 1 in
9220 	 * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
9221 	 */
9222 	if (sc->flags & FULL_INIT_DONE &&
9223 	    (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
9224 		rc = EBUSY;
9225 		goto done;
9226 	}
9227 
9228 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
9229 	if (fw_data == NULL) {
9230 		rc = ENOMEM;
9231 		goto done;
9232 	}
9233 
9234 	rc = copyin(fw->data, fw_data, fw->len);
9235 	if (rc == 0)
9236 		rc = -t4_load_fw(sc, fw_data, fw->len);
9237 
9238 	free(fw_data, M_CXGBE);
9239 done:
9240 	end_synchronized_op(sc, 0);
9241 	return (rc);
9242 }
9243 
9244 static int
9245 load_cfg(struct adapter *sc, struct t4_data *cfg)
9246 {
9247 	int rc;
9248 	uint8_t *cfg_data = NULL;
9249 
9250 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9251 	if (rc)
9252 		return (rc);
9253 
9254 	if (cfg->len == 0) {
9255 		/* clear */
9256 		rc = -t4_load_cfg(sc, NULL, 0);
9257 		goto done;
9258 	}
9259 
9260 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
9261 	if (cfg_data == NULL) {
9262 		rc = ENOMEM;
9263 		goto done;
9264 	}
9265 
9266 	rc = copyin(cfg->data, cfg_data, cfg->len);
9267 	if (rc == 0)
9268 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
9269 
9270 	free(cfg_data, M_CXGBE);
9271 done:
9272 	end_synchronized_op(sc, 0);
9273 	return (rc);
9274 }
9275 
9276 static int
9277 load_boot(struct adapter *sc, struct t4_bootrom *br)
9278 {
9279 	int rc;
9280 	uint8_t *br_data = NULL;
9281 	u_int offset;
9282 
9283 	if (br->len > 1024 * 1024)
9284 		return (EFBIG);
9285 
9286 	if (br->pf_offset == 0) {
9287 		/* pfidx */
9288 		if (br->pfidx_addr > 7)
9289 			return (EINVAL);
9290 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
9291 		    A_PCIE_PF_EXPROM_OFST)));
9292 	} else if (br->pf_offset == 1) {
9293 		/* offset */
9294 		offset = G_OFFSET(br->pfidx_addr);
9295 	} else {
9296 		return (EINVAL);
9297 	}
9298 
9299 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
9300 	if (rc)
9301 		return (rc);
9302 
9303 	if (br->len == 0) {
9304 		/* clear */
9305 		rc = -t4_load_boot(sc, NULL, offset, 0);
9306 		goto done;
9307 	}
9308 
9309 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
9310 	if (br_data == NULL) {
9311 		rc = ENOMEM;
9312 		goto done;
9313 	}
9314 
9315 	rc = copyin(br->data, br_data, br->len);
9316 	if (rc == 0)
9317 		rc = -t4_load_boot(sc, br_data, offset, br->len);
9318 
9319 	free(br_data, M_CXGBE);
9320 done:
9321 	end_synchronized_op(sc, 0);
9322 	return (rc);
9323 }
9324 
9325 static int
9326 load_bootcfg(struct adapter *sc, struct t4_data *bc)
9327 {
9328 	int rc;
9329 	uint8_t *bc_data = NULL;
9330 
9331 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9332 	if (rc)
9333 		return (rc);
9334 
9335 	if (bc->len == 0) {
9336 		/* clear */
9337 		rc = -t4_load_bootcfg(sc, NULL, 0);
9338 		goto done;
9339 	}
9340 
9341 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
9342 	if (bc_data == NULL) {
9343 		rc = ENOMEM;
9344 		goto done;
9345 	}
9346 
9347 	rc = copyin(bc->data, bc_data, bc->len);
9348 	if (rc == 0)
9349 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
9350 
9351 	free(bc_data, M_CXGBE);
9352 done:
9353 	end_synchronized_op(sc, 0);
9354 	return (rc);
9355 }
9356 
9357 static int
9358 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
9359 {
9360 	int rc;
9361 	struct cudbg_init *cudbg;
9362 	void *handle, *buf;
9363 
9364 	/* buf is large, don't block if no memory is available */
9365 	buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
9366 	if (buf == NULL)
9367 		return (ENOMEM);
9368 
9369 	handle = cudbg_alloc_handle();
9370 	if (handle == NULL) {
9371 		rc = ENOMEM;
9372 		goto done;
9373 	}
9374 
9375 	cudbg = cudbg_get_init(handle);
9376 	cudbg->adap = sc;
9377 	cudbg->print = (cudbg_print_cb)printf;
9378 
9379 #ifndef notyet
9380 	device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
9381 	    __func__, dump->wr_flash, dump->len, dump->data);
9382 #endif
9383 
9384 	if (dump->wr_flash)
9385 		cudbg->use_flash = 1;
9386 	MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
9387 	memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
9388 
9389 	rc = cudbg_collect(handle, buf, &dump->len);
9390 	if (rc != 0)
9391 		goto done;
9392 
9393 	rc = copyout(buf, dump->data, dump->len);
9394 done:
9395 	cudbg_free_handle(handle);
9396 	free(buf, M_CXGBE);
9397 	return (rc);
9398 }
9399 
9400 #define MAX_READ_BUF_SIZE (128 * 1024)
9401 static int
9402 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
9403 {
9404 	uint32_t addr, remaining, n;
9405 	uint32_t *buf;
9406 	int rc;
9407 	uint8_t *dst;
9408 
9409 	rc = validate_mem_range(sc, mr->addr, mr->len);
9410 	if (rc != 0)
9411 		return (rc);
9412 
9413 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
9414 	addr = mr->addr;
9415 	remaining = mr->len;
9416 	dst = (void *)mr->data;
9417 
9418 	while (remaining) {
9419 		n = min(remaining, MAX_READ_BUF_SIZE);
9420 		read_via_memwin(sc, 2, addr, buf, n);
9421 
9422 		rc = copyout(buf, dst, n);
9423 		if (rc != 0)
9424 			break;
9425 
9426 		dst += n;
9427 		remaining -= n;
9428 		addr += n;
9429 	}
9430 
9431 	free(buf, M_CXGBE);
9432 	return (rc);
9433 }
9434 #undef MAX_READ_BUF_SIZE
9435 
9436 static int
9437 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
9438 {
9439 	int rc;
9440 
9441 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
9442 		return (EINVAL);
9443 
9444 	if (i2cd->len > sizeof(i2cd->data))
9445 		return (EFBIG);
9446 
9447 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9448 	if (rc)
9449 		return (rc);
9450 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9451 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
9452 	end_synchronized_op(sc, 0);
9453 
9454 	return (rc);
9455 }
9456 
9457 int
9458 t4_os_find_pci_capability(struct adapter *sc, int cap)
9459 {
9460 	int i;
9461 
9462 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9463 }
9464 
9465 int
9466 t4_os_pci_save_state(struct adapter *sc)
9467 {
9468 	device_t dev;
9469 	struct pci_devinfo *dinfo;
9470 
9471 	dev = sc->dev;
9472 	dinfo = device_get_ivars(dev);
9473 
9474 	pci_cfg_save(dev, dinfo, 0);
9475 	return (0);
9476 }
9477 
9478 int
9479 t4_os_pci_restore_state(struct adapter *sc)
9480 {
9481 	device_t dev;
9482 	struct pci_devinfo *dinfo;
9483 
9484 	dev = sc->dev;
9485 	dinfo = device_get_ivars(dev);
9486 
9487 	pci_cfg_restore(dev, dinfo);
9488 	return (0);
9489 }
9490 
9491 void
9492 t4_os_portmod_changed(struct port_info *pi)
9493 {
9494 	struct adapter *sc = pi->adapter;
9495 	struct vi_info *vi;
9496 	struct ifnet *ifp;
9497 	static const char *mod_str[] = {
9498 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9499 	};
9500 
9501 	PORT_LOCK(pi);
9502 	build_medialist(pi, &pi->media);
9503 	PORT_UNLOCK(pi);
9504 	vi = &pi->vi[0];
9505 	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9506 		init_l1cfg(pi);
9507 		end_synchronized_op(sc, LOCK_HELD);
9508 	}
9509 
9510 	ifp = vi->ifp;
9511 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9512 		if_printf(ifp, "transceiver unplugged.\n");
9513 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9514 		if_printf(ifp, "unknown transceiver inserted.\n");
9515 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9516 		if_printf(ifp, "unsupported transceiver inserted.\n");
9517 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9518 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9519 		    port_top_speed(pi), mod_str[pi->mod_type]);
9520 	} else {
9521 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9522 		    pi->mod_type);
9523 	}
9524 }
9525 
9526 void
9527 t4_os_link_changed(struct port_info *pi)
9528 {
9529 	struct vi_info *vi;
9530 	struct ifnet *ifp;
9531 	struct link_config *lc;
9532 	int v;
9533 
9534 	for_each_vi(pi, v, vi) {
9535 		ifp = vi->ifp;
9536 		if (ifp == NULL)
9537 			continue;
9538 
9539 		lc = &pi->link_cfg;
9540 		if (lc->link_ok) {
9541 			ifp->if_baudrate = IF_Mbps(lc->speed);
9542 			if_link_state_change(ifp, LINK_STATE_UP);
9543 		} else {
9544 			if_link_state_change(ifp, LINK_STATE_DOWN);
9545 		}
9546 	}
9547 }
9548 
9549 void
9550 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9551 {
9552 	struct adapter *sc;
9553 
9554 	sx_slock(&t4_list_lock);
9555 	SLIST_FOREACH(sc, &t4_list, link) {
9556 		/*
9557 		 * func should not make any assumptions about what state sc is
9558 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9559 		 */
9560 		func(sc, arg);
9561 	}
9562 	sx_sunlock(&t4_list_lock);
9563 }
9564 
9565 static int
9566 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9567     struct thread *td)
9568 {
9569 	int rc;
9570 	struct adapter *sc = dev->si_drv1;
9571 
9572 	rc = priv_check(td, PRIV_DRIVER);
9573 	if (rc != 0)
9574 		return (rc);
9575 
9576 	switch (cmd) {
9577 	case CHELSIO_T4_GETREG: {
9578 		struct t4_reg *edata = (struct t4_reg *)data;
9579 
9580 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9581 			return (EFAULT);
9582 
9583 		if (edata->size == 4)
9584 			edata->val = t4_read_reg(sc, edata->addr);
9585 		else if (edata->size == 8)
9586 			edata->val = t4_read_reg64(sc, edata->addr);
9587 		else
9588 			return (EINVAL);
9589 
9590 		break;
9591 	}
9592 	case CHELSIO_T4_SETREG: {
9593 		struct t4_reg *edata = (struct t4_reg *)data;
9594 
9595 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9596 			return (EFAULT);
9597 
9598 		if (edata->size == 4) {
9599 			if (edata->val & 0xffffffff00000000)
9600 				return (EINVAL);
9601 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9602 		} else if (edata->size == 8)
9603 			t4_write_reg64(sc, edata->addr, edata->val);
9604 		else
9605 			return (EINVAL);
9606 		break;
9607 	}
9608 	case CHELSIO_T4_REGDUMP: {
9609 		struct t4_regdump *regs = (struct t4_regdump *)data;
9610 		int reglen = t4_get_regs_len(sc);
9611 		uint8_t *buf;
9612 
9613 		if (regs->len < reglen) {
9614 			regs->len = reglen; /* hint to the caller */
9615 			return (ENOBUFS);
9616 		}
9617 
9618 		regs->len = reglen;
9619 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9620 		get_regs(sc, regs, buf);
9621 		rc = copyout(buf, regs->data, reglen);
9622 		free(buf, M_CXGBE);
9623 		break;
9624 	}
9625 	case CHELSIO_T4_GET_FILTER_MODE:
9626 		rc = get_filter_mode(sc, (uint32_t *)data);
9627 		break;
9628 	case CHELSIO_T4_SET_FILTER_MODE:
9629 		rc = set_filter_mode(sc, *(uint32_t *)data);
9630 		break;
9631 	case CHELSIO_T4_GET_FILTER:
9632 		rc = get_filter(sc, (struct t4_filter *)data);
9633 		break;
9634 	case CHELSIO_T4_SET_FILTER:
9635 		rc = set_filter(sc, (struct t4_filter *)data);
9636 		break;
9637 	case CHELSIO_T4_DEL_FILTER:
9638 		rc = del_filter(sc, (struct t4_filter *)data);
9639 		break;
9640 	case CHELSIO_T4_GET_SGE_CONTEXT:
9641 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9642 		break;
9643 	case CHELSIO_T4_LOAD_FW:
9644 		rc = load_fw(sc, (struct t4_data *)data);
9645 		break;
9646 	case CHELSIO_T4_GET_MEM:
9647 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9648 		break;
9649 	case CHELSIO_T4_GET_I2C:
9650 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9651 		break;
9652 	case CHELSIO_T4_CLEAR_STATS: {
9653 		int i, v;
9654 		u_int port_id = *(uint32_t *)data;
9655 		struct port_info *pi;
9656 		struct vi_info *vi;
9657 
9658 		if (port_id >= sc->params.nports)
9659 			return (EINVAL);
9660 		pi = sc->port[port_id];
9661 		if (pi == NULL)
9662 			return (EIO);
9663 
9664 		/* MAC stats */
9665 		t4_clr_port_stats(sc, pi->tx_chan);
9666 		pi->tx_parse_error = 0;
9667 		mtx_lock(&sc->reg_lock);
9668 		for_each_vi(pi, v, vi) {
9669 			if (vi->flags & VI_INIT_DONE)
9670 				t4_clr_vi_stats(sc, vi->viid);
9671 		}
9672 		mtx_unlock(&sc->reg_lock);
9673 
9674 		/*
9675 		 * Since this command accepts a port, clear stats for
9676 		 * all VIs on this port.
9677 		 */
9678 		for_each_vi(pi, v, vi) {
9679 			if (vi->flags & VI_INIT_DONE) {
9680 				struct sge_rxq *rxq;
9681 				struct sge_txq *txq;
9682 				struct sge_wrq *wrq;
9683 
9684 				for_each_rxq(vi, i, rxq) {
9685 #if defined(INET) || defined(INET6)
9686 					rxq->lro.lro_queued = 0;
9687 					rxq->lro.lro_flushed = 0;
9688 #endif
9689 					rxq->rxcsum = 0;
9690 					rxq->vlan_extraction = 0;
9691 				}
9692 
9693 				for_each_txq(vi, i, txq) {
9694 					txq->txcsum = 0;
9695 					txq->tso_wrs = 0;
9696 					txq->vlan_insertion = 0;
9697 					txq->imm_wrs = 0;
9698 					txq->sgl_wrs = 0;
9699 					txq->txpkt_wrs = 0;
9700 					txq->txpkts0_wrs = 0;
9701 					txq->txpkts1_wrs = 0;
9702 					txq->txpkts0_pkts = 0;
9703 					txq->txpkts1_pkts = 0;
9704 					mp_ring_reset_stats(txq->r);
9705 				}
9706 
9707 #ifdef TCP_OFFLOAD
9708 				/* nothing to clear for each ofld_rxq */
9709 
9710 				for_each_ofld_txq(vi, i, wrq) {
9711 					wrq->tx_wrs_direct = 0;
9712 					wrq->tx_wrs_copied = 0;
9713 				}
9714 #endif
9715 
9716 				if (IS_MAIN_VI(vi)) {
9717 					wrq = &sc->sge.ctrlq[pi->port_id];
9718 					wrq->tx_wrs_direct = 0;
9719 					wrq->tx_wrs_copied = 0;
9720 				}
9721 			}
9722 		}
9723 		break;
9724 	}
9725 	case CHELSIO_T4_SCHED_CLASS:
9726 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9727 		break;
9728 	case CHELSIO_T4_SCHED_QUEUE:
9729 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9730 		break;
9731 	case CHELSIO_T4_GET_TRACER:
9732 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9733 		break;
9734 	case CHELSIO_T4_SET_TRACER:
9735 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9736 		break;
9737 	case CHELSIO_T4_LOAD_CFG:
9738 		rc = load_cfg(sc, (struct t4_data *)data);
9739 		break;
9740 	case CHELSIO_T4_LOAD_BOOT:
9741 		rc = load_boot(sc, (struct t4_bootrom *)data);
9742 		break;
9743 	case CHELSIO_T4_LOAD_BOOTCFG:
9744 		rc = load_bootcfg(sc, (struct t4_data *)data);
9745 		break;
9746 	case CHELSIO_T4_CUDBG_DUMP:
9747 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
9748 		break;
9749 	default:
9750 		rc = ENOTTY;
9751 	}
9752 
9753 	return (rc);
9754 }
9755 
9756 void
9757 t4_db_full(struct adapter *sc)
9758 {
9759 
9760 	CXGBE_UNIMPLEMENTED(__func__);
9761 }
9762 
9763 void
9764 t4_db_dropped(struct adapter *sc)
9765 {
9766 
9767 	CXGBE_UNIMPLEMENTED(__func__);
9768 }
9769 
9770 #ifdef TCP_OFFLOAD
9771 static int
9772 toe_capability(struct vi_info *vi, int enable)
9773 {
9774 	int rc;
9775 	struct port_info *pi = vi->pi;
9776 	struct adapter *sc = pi->adapter;
9777 
9778 	ASSERT_SYNCHRONIZED_OP(sc);
9779 
9780 	if (!is_offload(sc))
9781 		return (ENODEV);
9782 
9783 	if (enable) {
9784 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9785 			/* TOE is already enabled. */
9786 			return (0);
9787 		}
9788 
9789 		/*
9790 		 * We need the port's queues around so that we're able to send
9791 		 * and receive CPLs to/from the TOE even if the ifnet for this
9792 		 * port has never been UP'd administratively.
9793 		 */
9794 		if (!(vi->flags & VI_INIT_DONE)) {
9795 			rc = vi_full_init(vi);
9796 			if (rc)
9797 				return (rc);
9798 		}
9799 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9800 			rc = vi_full_init(&pi->vi[0]);
9801 			if (rc)
9802 				return (rc);
9803 		}
9804 
9805 		if (isset(&sc->offload_map, pi->port_id)) {
9806 			/* TOE is enabled on another VI of this port. */
9807 			pi->uld_vis++;
9808 			return (0);
9809 		}
9810 
9811 		if (!uld_active(sc, ULD_TOM)) {
9812 			rc = t4_activate_uld(sc, ULD_TOM);
9813 			if (rc == EAGAIN) {
9814 				log(LOG_WARNING,
9815 				    "You must kldload t4_tom.ko before trying "
9816 				    "to enable TOE on a cxgbe interface.\n");
9817 			}
9818 			if (rc != 0)
9819 				return (rc);
9820 			KASSERT(sc->tom_softc != NULL,
9821 			    ("%s: TOM activated but softc NULL", __func__));
9822 			KASSERT(uld_active(sc, ULD_TOM),
9823 			    ("%s: TOM activated but flag not set", __func__));
9824 		}
9825 
9826 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
9827 		if (!uld_active(sc, ULD_IWARP))
9828 			(void) t4_activate_uld(sc, ULD_IWARP);
9829 		if (!uld_active(sc, ULD_ISCSI))
9830 			(void) t4_activate_uld(sc, ULD_ISCSI);
9831 
9832 		pi->uld_vis++;
9833 		setbit(&sc->offload_map, pi->port_id);
9834 	} else {
9835 		pi->uld_vis--;
9836 
9837 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9838 			return (0);
9839 
9840 		KASSERT(uld_active(sc, ULD_TOM),
9841 		    ("%s: TOM never initialized?", __func__));
9842 		clrbit(&sc->offload_map, pi->port_id);
9843 	}
9844 
9845 	return (0);
9846 }
9847 
9848 /*
9849  * Add an upper layer driver to the global list.
9850  */
9851 int
9852 t4_register_uld(struct uld_info *ui)
9853 {
9854 	int rc = 0;
9855 	struct uld_info *u;
9856 
9857 	sx_xlock(&t4_uld_list_lock);
9858 	SLIST_FOREACH(u, &t4_uld_list, link) {
9859 	    if (u->uld_id == ui->uld_id) {
9860 		    rc = EEXIST;
9861 		    goto done;
9862 	    }
9863 	}
9864 
9865 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9866 	ui->refcount = 0;
9867 done:
9868 	sx_xunlock(&t4_uld_list_lock);
9869 	return (rc);
9870 }
9871 
9872 int
9873 t4_unregister_uld(struct uld_info *ui)
9874 {
9875 	int rc = EINVAL;
9876 	struct uld_info *u;
9877 
9878 	sx_xlock(&t4_uld_list_lock);
9879 
9880 	SLIST_FOREACH(u, &t4_uld_list, link) {
9881 	    if (u == ui) {
9882 		    if (ui->refcount > 0) {
9883 			    rc = EBUSY;
9884 			    goto done;
9885 		    }
9886 
9887 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9888 		    rc = 0;
9889 		    goto done;
9890 	    }
9891 	}
9892 done:
9893 	sx_xunlock(&t4_uld_list_lock);
9894 	return (rc);
9895 }
9896 
9897 int
9898 t4_activate_uld(struct adapter *sc, int id)
9899 {
9900 	int rc;
9901 	struct uld_info *ui;
9902 
9903 	ASSERT_SYNCHRONIZED_OP(sc);
9904 
9905 	if (id < 0 || id > ULD_MAX)
9906 		return (EINVAL);
9907 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
9908 
9909 	sx_slock(&t4_uld_list_lock);
9910 
9911 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9912 		if (ui->uld_id == id) {
9913 			if (!(sc->flags & FULL_INIT_DONE)) {
9914 				rc = adapter_full_init(sc);
9915 				if (rc != 0)
9916 					break;
9917 			}
9918 
9919 			rc = ui->activate(sc);
9920 			if (rc == 0) {
9921 				setbit(&sc->active_ulds, id);
9922 				ui->refcount++;
9923 			}
9924 			break;
9925 		}
9926 	}
9927 
9928 	sx_sunlock(&t4_uld_list_lock);
9929 
9930 	return (rc);
9931 }
9932 
9933 int
9934 t4_deactivate_uld(struct adapter *sc, int id)
9935 {
9936 	int rc;
9937 	struct uld_info *ui;
9938 
9939 	ASSERT_SYNCHRONIZED_OP(sc);
9940 
9941 	if (id < 0 || id > ULD_MAX)
9942 		return (EINVAL);
9943 	rc = ENXIO;
9944 
9945 	sx_slock(&t4_uld_list_lock);
9946 
9947 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9948 		if (ui->uld_id == id) {
9949 			rc = ui->deactivate(sc);
9950 			if (rc == 0) {
9951 				clrbit(&sc->active_ulds, id);
9952 				ui->refcount--;
9953 			}
9954 			break;
9955 		}
9956 	}
9957 
9958 	sx_sunlock(&t4_uld_list_lock);
9959 
9960 	return (rc);
9961 }
9962 
9963 int
9964 uld_active(struct adapter *sc, int uld_id)
9965 {
9966 
9967 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9968 
9969 	return (isset(&sc->active_ulds, uld_id));
9970 }
9971 #endif
9972 
9973 /*
9974  * t  = ptr to tunable.
9975  * nc = number of CPUs.
9976  * c  = compiled in default for that tunable.
9977  */
9978 static void
9979 calculate_nqueues(int *t, int nc, const int c)
9980 {
9981 	int nq;
9982 
9983 	if (*t > 0)
9984 		return;
9985 	nq = *t < 0 ? -*t : c;
9986 	*t = min(nc, nq);
9987 }
9988 
9989 /*
9990  * Come up with reasonable defaults for some of the tunables, provided they're
9991  * not set by the user (in which case we'll use the values as is).
9992  */
9993 static void
9994 tweak_tunables(void)
9995 {
9996 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
9997 
9998 	if (t4_ntxq < 1) {
9999 #ifdef RSS
10000 		t4_ntxq = rss_getnumbuckets();
10001 #else
10002 		calculate_nqueues(&t4_ntxq, nc, NTXQ);
10003 #endif
10004 	}
10005 
10006 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
10007 
10008 	if (t4_nrxq < 1) {
10009 #ifdef RSS
10010 		t4_nrxq = rss_getnumbuckets();
10011 #else
10012 		calculate_nqueues(&t4_nrxq, nc, NRXQ);
10013 #endif
10014 	}
10015 
10016 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
10017 
10018 #ifdef TCP_OFFLOAD
10019 	calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
10020 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
10021 	calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
10022 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
10023 
10024 	if (t4_toecaps_allowed == -1)
10025 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
10026 
10027 	if (t4_rdmacaps_allowed == -1) {
10028 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
10029 		    FW_CAPS_CONFIG_RDMA_RDMAC;
10030 	}
10031 
10032 	if (t4_iscsicaps_allowed == -1) {
10033 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
10034 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
10035 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
10036 	}
10037 
10038 	if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
10039 		t4_tmr_idx_ofld = TMR_IDX_OFLD;
10040 
10041 	if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
10042 		t4_pktc_idx_ofld = PKTC_IDX_OFLD;
10043 #else
10044 	if (t4_toecaps_allowed == -1)
10045 		t4_toecaps_allowed = 0;
10046 
10047 	if (t4_rdmacaps_allowed == -1)
10048 		t4_rdmacaps_allowed = 0;
10049 
10050 	if (t4_iscsicaps_allowed == -1)
10051 		t4_iscsicaps_allowed = 0;
10052 #endif
10053 
10054 #ifdef DEV_NETMAP
10055 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
10056 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
10057 #endif
10058 
10059 	if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
10060 		t4_tmr_idx = TMR_IDX;
10061 
10062 	if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
10063 		t4_pktc_idx = PKTC_IDX;
10064 
10065 	if (t4_qsize_txq < 128)
10066 		t4_qsize_txq = 128;
10067 
10068 	if (t4_qsize_rxq < 128)
10069 		t4_qsize_rxq = 128;
10070 	while (t4_qsize_rxq & 7)
10071 		t4_qsize_rxq++;
10072 
10073 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
10074 
10075 	/*
10076 	 * Number of VIs to create per-port.  The first VI is the "main" regular
10077 	 * VI for the port.  The rest are additional virtual interfaces on the
10078 	 * same physical port.  Note that the main VI does not have native
10079 	 * netmap support but the extra VIs do.
10080 	 *
10081 	 * Limit the number of VIs per port to the number of available
10082 	 * MAC addresses per port.
10083 	 */
10084 	if (t4_num_vis < 1)
10085 		t4_num_vis = 1;
10086 	if (t4_num_vis > nitems(vi_mac_funcs)) {
10087 		t4_num_vis = nitems(vi_mac_funcs);
10088 		printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
10089 	}
10090 
10091 	if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
10092 		pcie_relaxed_ordering = 1;
10093 #if defined(__i386__) || defined(__amd64__)
10094 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
10095 			pcie_relaxed_ordering = 0;
10096 #endif
10097 	}
10098 }
10099 
10100 #ifdef DDB
10101 static void
10102 t4_dump_tcb(struct adapter *sc, int tid)
10103 {
10104 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
10105 
10106 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
10107 	save = t4_read_reg(sc, reg);
10108 	base = sc->memwin[2].mw_base;
10109 
10110 	/* Dump TCB for the tid */
10111 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
10112 	tcb_addr += tid * TCB_SIZE;
10113 
10114 	if (is_t4(sc)) {
10115 		pf = 0;
10116 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
10117 	} else {
10118 		pf = V_PFNUM(sc->pf);
10119 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
10120 	}
10121 	t4_write_reg(sc, reg, win_pos | pf);
10122 	t4_read_reg(sc, reg);
10123 
10124 	off = tcb_addr - win_pos;
10125 	for (i = 0; i < 4; i++) {
10126 		uint32_t buf[8];
10127 		for (j = 0; j < 8; j++, off += 4)
10128 			buf[j] = htonl(t4_read_reg(sc, base + off));
10129 
10130 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
10131 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
10132 		    buf[7]);
10133 	}
10134 
10135 	t4_write_reg(sc, reg, save);
10136 	t4_read_reg(sc, reg);
10137 }
10138 
10139 static void
10140 t4_dump_devlog(struct adapter *sc)
10141 {
10142 	struct devlog_params *dparams = &sc->params.devlog;
10143 	struct fw_devlog_e e;
10144 	int i, first, j, m, nentries, rc;
10145 	uint64_t ftstamp = UINT64_MAX;
10146 
10147 	if (dparams->start == 0) {
10148 		db_printf("devlog params not valid\n");
10149 		return;
10150 	}
10151 
10152 	nentries = dparams->size / sizeof(struct fw_devlog_e);
10153 	m = fwmtype_to_hwmtype(dparams->memtype);
10154 
10155 	/* Find the first entry. */
10156 	first = -1;
10157 	for (i = 0; i < nentries && !db_pager_quit; i++) {
10158 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10159 		    sizeof(e), (void *)&e);
10160 		if (rc != 0)
10161 			break;
10162 
10163 		if (e.timestamp == 0)
10164 			break;
10165 
10166 		e.timestamp = be64toh(e.timestamp);
10167 		if (e.timestamp < ftstamp) {
10168 			ftstamp = e.timestamp;
10169 			first = i;
10170 		}
10171 	}
10172 
10173 	if (first == -1)
10174 		return;
10175 
10176 	i = first;
10177 	do {
10178 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10179 		    sizeof(e), (void *)&e);
10180 		if (rc != 0)
10181 			return;
10182 
10183 		if (e.timestamp == 0)
10184 			return;
10185 
10186 		e.timestamp = be64toh(e.timestamp);
10187 		e.seqno = be32toh(e.seqno);
10188 		for (j = 0; j < 8; j++)
10189 			e.params[j] = be32toh(e.params[j]);
10190 
10191 		db_printf("%10d  %15ju  %8s  %8s  ",
10192 		    e.seqno, e.timestamp,
10193 		    (e.level < nitems(devlog_level_strings) ?
10194 			devlog_level_strings[e.level] : "UNKNOWN"),
10195 		    (e.facility < nitems(devlog_facility_strings) ?
10196 			devlog_facility_strings[e.facility] : "UNKNOWN"));
10197 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
10198 		    e.params[3], e.params[4], e.params[5], e.params[6],
10199 		    e.params[7]);
10200 
10201 		if (++i == nentries)
10202 			i = 0;
10203 	} while (i != first && !db_pager_quit);
10204 }
10205 
10206 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
10207 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
10208 
10209 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
10210 {
10211 	device_t dev;
10212 	int t;
10213 	bool valid;
10214 
10215 	valid = false;
10216 	t = db_read_token();
10217 	if (t == tIDENT) {
10218 		dev = device_lookup_by_name(db_tok_string);
10219 		valid = true;
10220 	}
10221 	db_skip_to_eol();
10222 	if (!valid) {
10223 		db_printf("usage: show t4 devlog <nexus>\n");
10224 		return;
10225 	}
10226 
10227 	if (dev == NULL) {
10228 		db_printf("device not found\n");
10229 		return;
10230 	}
10231 
10232 	t4_dump_devlog(device_get_softc(dev));
10233 }
10234 
10235 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
10236 {
10237 	device_t dev;
10238 	int radix, tid, t;
10239 	bool valid;
10240 
10241 	valid = false;
10242 	radix = db_radix;
10243 	db_radix = 10;
10244 	t = db_read_token();
10245 	if (t == tIDENT) {
10246 		dev = device_lookup_by_name(db_tok_string);
10247 		t = db_read_token();
10248 		if (t == tNUMBER) {
10249 			tid = db_tok_number;
10250 			valid = true;
10251 		}
10252 	}
10253 	db_radix = radix;
10254 	db_skip_to_eol();
10255 	if (!valid) {
10256 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
10257 		return;
10258 	}
10259 
10260 	if (dev == NULL) {
10261 		db_printf("device not found\n");
10262 		return;
10263 	}
10264 	if (tid < 0) {
10265 		db_printf("invalid tid\n");
10266 		return;
10267 	}
10268 
10269 	t4_dump_tcb(device_get_softc(dev), tid);
10270 }
10271 #endif
10272 
10273 /*
10274  * Borrowed from cesa_prep_aes_key().
10275  *
10276  * NB: The crypto engine wants the words in the decryption key in reverse
10277  * order.
10278  */
10279 void
10280 t4_aes_getdeckey(void *dec_key, const void *enc_key, unsigned int kbits)
10281 {
10282 	uint32_t ek[4 * (RIJNDAEL_MAXNR + 1)];
10283 	uint32_t *dkey;
10284 	int i;
10285 
10286 	rijndaelKeySetupEnc(ek, enc_key, kbits);
10287 	dkey = dec_key;
10288 	dkey += (kbits / 8) / 4;
10289 
10290 	switch (kbits) {
10291 	case 128:
10292 		for (i = 0; i < 4; i++)
10293 			*--dkey = htobe32(ek[4 * 10 + i]);
10294 		break;
10295 	case 192:
10296 		for (i = 0; i < 2; i++)
10297 			*--dkey = htobe32(ek[4 * 11 + 2 + i]);
10298 		for (i = 0; i < 4; i++)
10299 			*--dkey = htobe32(ek[4 * 12 + i]);
10300 		break;
10301 	case 256:
10302 		for (i = 0; i < 4; i++)
10303 			*--dkey = htobe32(ek[4 * 13 + i]);
10304 		for (i = 0; i < 4; i++)
10305 			*--dkey = htobe32(ek[4 * 14 + i]);
10306 		break;
10307 	}
10308 	MPASS(dkey == dec_key);
10309 }
10310 
10311 static struct sx mlu;	/* mod load unload */
10312 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
10313 
10314 static int
10315 mod_event(module_t mod, int cmd, void *arg)
10316 {
10317 	int rc = 0;
10318 	static int loaded = 0;
10319 
10320 	switch (cmd) {
10321 	case MOD_LOAD:
10322 		sx_xlock(&mlu);
10323 		if (loaded++ == 0) {
10324 			t4_sge_modload();
10325 			t4_register_cpl_handler(CPL_SET_TCB_RPL, set_tcb_rpl);
10326 			t4_register_cpl_handler(CPL_L2T_WRITE_RPL, l2t_write_rpl);
10327 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
10328 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
10329 			sx_init(&t4_list_lock, "T4/T5 adapters");
10330 			SLIST_INIT(&t4_list);
10331 #ifdef TCP_OFFLOAD
10332 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
10333 			SLIST_INIT(&t4_uld_list);
10334 #endif
10335 			t4_tracer_modload();
10336 			tweak_tunables();
10337 		}
10338 		sx_xunlock(&mlu);
10339 		break;
10340 
10341 	case MOD_UNLOAD:
10342 		sx_xlock(&mlu);
10343 		if (--loaded == 0) {
10344 			int tries;
10345 
10346 			sx_slock(&t4_list_lock);
10347 			if (!SLIST_EMPTY(&t4_list)) {
10348 				rc = EBUSY;
10349 				sx_sunlock(&t4_list_lock);
10350 				goto done_unload;
10351 			}
10352 #ifdef TCP_OFFLOAD
10353 			sx_slock(&t4_uld_list_lock);
10354 			if (!SLIST_EMPTY(&t4_uld_list)) {
10355 				rc = EBUSY;
10356 				sx_sunlock(&t4_uld_list_lock);
10357 				sx_sunlock(&t4_list_lock);
10358 				goto done_unload;
10359 			}
10360 #endif
10361 			tries = 0;
10362 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
10363 				uprintf("%ju clusters with custom free routine "
10364 				    "still is use.\n", t4_sge_extfree_refs());
10365 				pause("t4unload", 2 * hz);
10366 			}
10367 #ifdef TCP_OFFLOAD
10368 			sx_sunlock(&t4_uld_list_lock);
10369 #endif
10370 			sx_sunlock(&t4_list_lock);
10371 
10372 			if (t4_sge_extfree_refs() == 0) {
10373 				t4_tracer_modunload();
10374 #ifdef TCP_OFFLOAD
10375 				sx_destroy(&t4_uld_list_lock);
10376 #endif
10377 				sx_destroy(&t4_list_lock);
10378 				t4_sge_modunload();
10379 				loaded = 0;
10380 			} else {
10381 				rc = EBUSY;
10382 				loaded++;	/* undo earlier decrement */
10383 			}
10384 		}
10385 done_unload:
10386 		sx_xunlock(&mlu);
10387 		break;
10388 	}
10389 
10390 	return (rc);
10391 }
10392 
10393 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10394 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10395 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10396 
10397 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10398 MODULE_VERSION(t4nex, 1);
10399 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10400 #ifdef DEV_NETMAP
10401 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10402 #endif /* DEV_NETMAP */
10403 
10404 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10405 MODULE_VERSION(t5nex, 1);
10406 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10407 #ifdef DEV_NETMAP
10408 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10409 #endif /* DEV_NETMAP */
10410 
10411 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10412 MODULE_VERSION(t6nex, 1);
10413 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10414 #ifdef DEV_NETMAP
10415 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10416 #endif /* DEV_NETMAP */
10417 
10418 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10419 MODULE_VERSION(cxgbe, 1);
10420 
10421 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10422 MODULE_VERSION(cxl, 1);
10423 
10424 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10425 MODULE_VERSION(cc, 1);
10426 
10427 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10428 MODULE_VERSION(vcxgbe, 1);
10429 
10430 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10431 MODULE_VERSION(vcxl, 1);
10432 
10433 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10434 MODULE_VERSION(vcc, 1);
10435