xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision 4f52dfbb8d6c4d446500c5b097e3806ec219fbd4)
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 static int t4_panic_on_fatal_err = 0;
473 TUNABLE_INT("hw.cxgbe.panic_on_fatal_err", &t4_panic_on_fatal_err);
474 
475 #ifdef TCP_OFFLOAD
476 /*
477  * TOE tunables.
478  */
479 static int t4_cop_managed_offloading = 0;
480 TUNABLE_INT("hw.cxgbe.cop_managed_offloading", &t4_cop_managed_offloading);
481 #endif
482 
483 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
484 static int vi_mac_funcs[] = {
485 	FW_VI_FUNC_ETH,
486 	FW_VI_FUNC_OFLD,
487 	FW_VI_FUNC_IWARP,
488 	FW_VI_FUNC_OPENISCSI,
489 	FW_VI_FUNC_OPENFCOE,
490 	FW_VI_FUNC_FOISCSI,
491 	FW_VI_FUNC_FOFCOE,
492 };
493 
494 struct intrs_and_queues {
495 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
496 	uint16_t num_vis;	/* number of VIs for each port */
497 	uint16_t nirq;		/* Total # of vectors */
498 	uint16_t ntxq;		/* # of NIC txq's for each port */
499 	uint16_t nrxq;		/* # of NIC rxq's for each port */
500 	uint16_t nofldtxq;	/* # of TOE txq's for each port */
501 	uint16_t nofldrxq;	/* # of TOE rxq's for each port */
502 
503 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
504 	uint16_t ntxq_vi;	/* # of NIC txq's */
505 	uint16_t nrxq_vi;	/* # of NIC rxq's */
506 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
507 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
508 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
509 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
510 };
511 
512 static void setup_memwin(struct adapter *);
513 static void position_memwin(struct adapter *, int, uint32_t);
514 static int validate_mem_range(struct adapter *, uint32_t, int);
515 static int fwmtype_to_hwmtype(int);
516 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
517     uint32_t *);
518 static int fixup_devlog_params(struct adapter *);
519 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
520 static int prep_firmware(struct adapter *);
521 static int partition_resources(struct adapter *, const struct firmware *,
522     const char *);
523 static int get_params__pre_init(struct adapter *);
524 static int get_params__post_init(struct adapter *);
525 static int set_params__post_init(struct adapter *);
526 static void t4_set_desc(struct adapter *);
527 static void build_medialist(struct port_info *, struct ifmedia *);
528 static void init_l1cfg(struct port_info *);
529 static int cxgbe_init_synchronized(struct vi_info *);
530 static int cxgbe_uninit_synchronized(struct vi_info *);
531 static void quiesce_txq(struct adapter *, struct sge_txq *);
532 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
533 static void quiesce_iq(struct adapter *, struct sge_iq *);
534 static void quiesce_fl(struct adapter *, struct sge_fl *);
535 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
536     driver_intr_t *, void *, char *);
537 static int t4_free_irq(struct adapter *, struct irq *);
538 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
539 static void vi_refresh_stats(struct adapter *, struct vi_info *);
540 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
541 static void cxgbe_tick(void *);
542 static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
543 static void cxgbe_sysctls(struct port_info *);
544 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
545 static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
546 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
547 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
548 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
549 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
550 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
551 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
552 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
553 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
554 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
555 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
556 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
557 #ifdef SBUF_DRAIN
558 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
559 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
560 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
561 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
562 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
563 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
564 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
565 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
566 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
567 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
568 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
569 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
570 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
571 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
572 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
573 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
574 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
575 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
576 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
577 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
578 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
579 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
580 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
581 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
582 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
583 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
584 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
585 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
586 static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
587 #endif
588 #ifdef TCP_OFFLOAD
589 static int sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS);
590 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
591 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
592 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
593 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
594 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
595 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
596 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
597 #endif
598 static int get_sge_context(struct adapter *, struct t4_sge_context *);
599 static int load_fw(struct adapter *, struct t4_data *);
600 static int load_cfg(struct adapter *, struct t4_data *);
601 static int load_boot(struct adapter *, struct t4_bootrom *);
602 static int load_bootcfg(struct adapter *, struct t4_data *);
603 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
604 static void free_offload_policy(struct t4_offload_policy *);
605 static int set_offload_policy(struct adapter *, struct t4_offload_policy *);
606 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
607 static int read_i2c(struct adapter *, struct t4_i2c_data *);
608 #ifdef TCP_OFFLOAD
609 static int toe_capability(struct vi_info *, int);
610 #endif
611 static int mod_event(module_t, int, void *);
612 static int notify_siblings(device_t, int);
613 
614 struct {
615 	uint16_t device;
616 	char *desc;
617 } t4_pciids[] = {
618 	{0xa000, "Chelsio Terminator 4 FPGA"},
619 	{0x4400, "Chelsio T440-dbg"},
620 	{0x4401, "Chelsio T420-CR"},
621 	{0x4402, "Chelsio T422-CR"},
622 	{0x4403, "Chelsio T440-CR"},
623 	{0x4404, "Chelsio T420-BCH"},
624 	{0x4405, "Chelsio T440-BCH"},
625 	{0x4406, "Chelsio T440-CH"},
626 	{0x4407, "Chelsio T420-SO"},
627 	{0x4408, "Chelsio T420-CX"},
628 	{0x4409, "Chelsio T420-BT"},
629 	{0x440a, "Chelsio T404-BT"},
630 	{0x440e, "Chelsio T440-LP-CR"},
631 }, t5_pciids[] = {
632 	{0xb000, "Chelsio Terminator 5 FPGA"},
633 	{0x5400, "Chelsio T580-dbg"},
634 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
635 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
636 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
637 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
638 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
639 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
640 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
641 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
642 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
643 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
644 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
645 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
646 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
647 #ifdef notyet
648 	{0x5404,  "Chelsio T520-BCH"},
649 	{0x5405,  "Chelsio T540-BCH"},
650 	{0x5406,  "Chelsio T540-CH"},
651 	{0x5408,  "Chelsio T520-CX"},
652 	{0x540b,  "Chelsio B520-SR"},
653 	{0x540c,  "Chelsio B504-BT"},
654 	{0x540f,  "Chelsio Amsterdam"},
655 	{0x5413,  "Chelsio T580-CHR"},
656 #endif
657 }, t6_pciids[] = {
658 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
659 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
660 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
661 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
662 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
663 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
664 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
665 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
666 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
667 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
668 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
669 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
670 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
671 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
672 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
673 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
674 
675 	/* Custom */
676 	{0x6480, "Chelsio T6225 80"},
677 	{0x6481, "Chelsio T62100 81"},
678 	{0x6484, "Chelsio T62100 84"},
679 };
680 
681 #ifdef TCP_OFFLOAD
682 /*
683  * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
684  * exactly the same for both rxq and ofld_rxq.
685  */
686 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
687 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
688 #endif
689 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
690 
691 static int
692 t4_probe(device_t dev)
693 {
694 	int i;
695 	uint16_t v = pci_get_vendor(dev);
696 	uint16_t d = pci_get_device(dev);
697 	uint8_t f = pci_get_function(dev);
698 
699 	if (v != PCI_VENDOR_ID_CHELSIO)
700 		return (ENXIO);
701 
702 	/* Attach only to PF0 of the FPGA */
703 	if (d == 0xa000 && f != 0)
704 		return (ENXIO);
705 
706 	for (i = 0; i < nitems(t4_pciids); i++) {
707 		if (d == t4_pciids[i].device) {
708 			device_set_desc(dev, t4_pciids[i].desc);
709 			return (BUS_PROBE_DEFAULT);
710 		}
711 	}
712 
713 	return (ENXIO);
714 }
715 
716 static int
717 t5_probe(device_t dev)
718 {
719 	int i;
720 	uint16_t v = pci_get_vendor(dev);
721 	uint16_t d = pci_get_device(dev);
722 	uint8_t f = pci_get_function(dev);
723 
724 	if (v != PCI_VENDOR_ID_CHELSIO)
725 		return (ENXIO);
726 
727 	/* Attach only to PF0 of the FPGA */
728 	if (d == 0xb000 && f != 0)
729 		return (ENXIO);
730 
731 	for (i = 0; i < nitems(t5_pciids); i++) {
732 		if (d == t5_pciids[i].device) {
733 			device_set_desc(dev, t5_pciids[i].desc);
734 			return (BUS_PROBE_DEFAULT);
735 		}
736 	}
737 
738 	return (ENXIO);
739 }
740 
741 static int
742 t6_probe(device_t dev)
743 {
744 	int i;
745 	uint16_t v = pci_get_vendor(dev);
746 	uint16_t d = pci_get_device(dev);
747 
748 	if (v != PCI_VENDOR_ID_CHELSIO)
749 		return (ENXIO);
750 
751 	for (i = 0; i < nitems(t6_pciids); i++) {
752 		if (d == t6_pciids[i].device) {
753 			device_set_desc(dev, t6_pciids[i].desc);
754 			return (BUS_PROBE_DEFAULT);
755 		}
756 	}
757 
758 	return (ENXIO);
759 }
760 
761 static void
762 t5_attribute_workaround(device_t dev)
763 {
764 	device_t root_port;
765 	uint32_t v;
766 
767 	/*
768 	 * The T5 chips do not properly echo the No Snoop and Relaxed
769 	 * Ordering attributes when replying to a TLP from a Root
770 	 * Port.  As a workaround, find the parent Root Port and
771 	 * disable No Snoop and Relaxed Ordering.  Note that this
772 	 * affects all devices under this root port.
773 	 */
774 	root_port = pci_find_pcie_root_port(dev);
775 	if (root_port == NULL) {
776 		device_printf(dev, "Unable to find parent root port\n");
777 		return;
778 	}
779 
780 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
781 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
782 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
783 	    0)
784 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
785 		    device_get_nameunit(root_port));
786 }
787 
788 static const struct devnames devnames[] = {
789 	{
790 		.nexus_name = "t4nex",
791 		.ifnet_name = "cxgbe",
792 		.vi_ifnet_name = "vcxgbe",
793 		.pf03_drv_name = "t4iov",
794 		.vf_nexus_name = "t4vf",
795 		.vf_ifnet_name = "cxgbev"
796 	}, {
797 		.nexus_name = "t5nex",
798 		.ifnet_name = "cxl",
799 		.vi_ifnet_name = "vcxl",
800 		.pf03_drv_name = "t5iov",
801 		.vf_nexus_name = "t5vf",
802 		.vf_ifnet_name = "cxlv"
803 	}, {
804 		.nexus_name = "t6nex",
805 		.ifnet_name = "cc",
806 		.vi_ifnet_name = "vcc",
807 		.pf03_drv_name = "t6iov",
808 		.vf_nexus_name = "t6vf",
809 		.vf_ifnet_name = "ccv"
810 	}
811 };
812 
813 void
814 t4_init_devnames(struct adapter *sc)
815 {
816 	int id;
817 
818 	id = chip_id(sc);
819 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
820 		sc->names = &devnames[id - CHELSIO_T4];
821 	else {
822 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
823 		sc->names = NULL;
824 	}
825 }
826 
827 static int
828 t4_attach(device_t dev)
829 {
830 	struct adapter *sc;
831 	int rc = 0, i, j, rqidx, tqidx, nports;
832 	struct make_dev_args mda;
833 	struct intrs_and_queues iaq;
834 	struct sge *s;
835 	uint32_t *buf;
836 #ifdef TCP_OFFLOAD
837 	int ofld_rqidx, ofld_tqidx;
838 #endif
839 #ifdef DEV_NETMAP
840 	int nm_rqidx, nm_tqidx;
841 #endif
842 	int num_vis;
843 
844 	sc = device_get_softc(dev);
845 	sc->dev = dev;
846 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
847 
848 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
849 		t5_attribute_workaround(dev);
850 	pci_enable_busmaster(dev);
851 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
852 		uint32_t v;
853 
854 		pci_set_max_read_req(dev, 4096);
855 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
856 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
857 		if (pcie_relaxed_ordering == 0 &&
858 		    (v | PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
859 			v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
860 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
861 		} else if (pcie_relaxed_ordering == 1 &&
862 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
863 			v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
864 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
865 		}
866 	}
867 
868 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
869 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
870 	sc->traceq = -1;
871 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
872 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
873 	    device_get_nameunit(dev));
874 
875 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
876 	    device_get_nameunit(dev));
877 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
878 	t4_add_adapter(sc);
879 
880 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
881 	TAILQ_INIT(&sc->sfl);
882 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
883 
884 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
885 
886 	sc->policy = NULL;
887 	rw_init(&sc->policy_lock, "connection offload policy");
888 
889 	rc = t4_map_bars_0_and_4(sc);
890 	if (rc != 0)
891 		goto done; /* error message displayed already */
892 
893 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
894 
895 	/* Prepare the adapter for operation. */
896 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
897 	rc = -t4_prep_adapter(sc, buf);
898 	free(buf, M_CXGBE);
899 	if (rc != 0) {
900 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
901 		goto done;
902 	}
903 
904 	/*
905 	 * This is the real PF# to which we're attaching.  Works from within PCI
906 	 * passthrough environments too, where pci_get_function() could return a
907 	 * different PF# depending on the passthrough configuration.  We need to
908 	 * use the real PF# in all our communication with the firmware.
909 	 */
910 	j = t4_read_reg(sc, A_PL_WHOAMI);
911 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
912 	sc->mbox = sc->pf;
913 
914 	t4_init_devnames(sc);
915 	if (sc->names == NULL) {
916 		rc = ENOTSUP;
917 		goto done; /* error message displayed already */
918 	}
919 
920 	/*
921 	 * Do this really early, with the memory windows set up even before the
922 	 * character device.  The userland tool's register i/o and mem read
923 	 * will work even in "recovery mode".
924 	 */
925 	setup_memwin(sc);
926 	if (t4_init_devlog_params(sc, 0) == 0)
927 		fixup_devlog_params(sc);
928 	make_dev_args_init(&mda);
929 	mda.mda_devsw = &t4_cdevsw;
930 	mda.mda_uid = UID_ROOT;
931 	mda.mda_gid = GID_WHEEL;
932 	mda.mda_mode = 0600;
933 	mda.mda_si_drv1 = sc;
934 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
935 	if (rc != 0)
936 		device_printf(dev, "failed to create nexus char device: %d.\n",
937 		    rc);
938 
939 	/* Go no further if recovery mode has been requested. */
940 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
941 		device_printf(dev, "recovery mode.\n");
942 		goto done;
943 	}
944 
945 #if defined(__i386__)
946 	if ((cpu_feature & CPUID_CX8) == 0) {
947 		device_printf(dev, "64 bit atomics not available.\n");
948 		rc = ENOTSUP;
949 		goto done;
950 	}
951 #endif
952 
953 	/* Prepare the firmware for operation */
954 	rc = prep_firmware(sc);
955 	if (rc != 0)
956 		goto done; /* error message displayed already */
957 
958 	rc = get_params__post_init(sc);
959 	if (rc != 0)
960 		goto done; /* error message displayed already */
961 
962 	rc = set_params__post_init(sc);
963 	if (rc != 0)
964 		goto done; /* error message displayed already */
965 
966 	rc = t4_map_bar_2(sc);
967 	if (rc != 0)
968 		goto done; /* error message displayed already */
969 
970 	rc = t4_create_dma_tag(sc);
971 	if (rc != 0)
972 		goto done; /* error message displayed already */
973 
974 	/*
975 	 * First pass over all the ports - allocate VIs and initialize some
976 	 * basic parameters like mac address, port type, etc.
977 	 */
978 	for_each_port(sc, i) {
979 		struct port_info *pi;
980 
981 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
982 		sc->port[i] = pi;
983 
984 		/* These must be set before t4_port_init */
985 		pi->adapter = sc;
986 		pi->port_id = i;
987 		/*
988 		 * XXX: vi[0] is special so we can't delay this allocation until
989 		 * pi->nvi's final value is known.
990 		 */
991 		pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
992 		    M_ZERO | M_WAITOK);
993 
994 		/*
995 		 * Allocate the "main" VI and initialize parameters
996 		 * like mac addr.
997 		 */
998 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
999 		if (rc != 0) {
1000 			device_printf(dev, "unable to initialize port %d: %d\n",
1001 			    i, rc);
1002 			free(pi->vi, M_CXGBE);
1003 			free(pi, M_CXGBE);
1004 			sc->port[i] = NULL;
1005 			goto done;
1006 		}
1007 
1008 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1009 		    device_get_nameunit(dev), i);
1010 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1011 		sc->chan_map[pi->tx_chan] = i;
1012 
1013 		/* All VIs on this port share this media. */
1014 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1015 		    cxgbe_media_status);
1016 
1017 		pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
1018 		if (pi->dev == NULL) {
1019 			device_printf(dev,
1020 			    "failed to add device for port %d.\n", i);
1021 			rc = ENXIO;
1022 			goto done;
1023 		}
1024 		pi->vi[0].dev = pi->dev;
1025 		device_set_softc(pi->dev, pi);
1026 	}
1027 
1028 	/*
1029 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1030 	 */
1031 	nports = sc->params.nports;
1032 	rc = cfg_itype_and_nqueues(sc, &iaq);
1033 	if (rc != 0)
1034 		goto done; /* error message displayed already */
1035 
1036 	num_vis = iaq.num_vis;
1037 	sc->intr_type = iaq.intr_type;
1038 	sc->intr_count = iaq.nirq;
1039 
1040 	s = &sc->sge;
1041 	s->nrxq = nports * iaq.nrxq;
1042 	s->ntxq = nports * iaq.ntxq;
1043 	if (num_vis > 1) {
1044 		s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1045 		s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1046 	}
1047 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1048 	s->neq += nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
1049 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1050 #ifdef TCP_OFFLOAD
1051 	if (is_offload(sc)) {
1052 		s->nofldrxq = nports * iaq.nofldrxq;
1053 		s->nofldtxq = nports * iaq.nofldtxq;
1054 		if (num_vis > 1) {
1055 			s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1056 			s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1057 		}
1058 		s->neq += s->nofldtxq + s->nofldrxq;
1059 		s->niq += s->nofldrxq;
1060 
1061 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1062 		    M_CXGBE, M_ZERO | M_WAITOK);
1063 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1064 		    M_CXGBE, M_ZERO | M_WAITOK);
1065 	}
1066 #endif
1067 #ifdef DEV_NETMAP
1068 	if (num_vis > 1) {
1069 		s->nnmrxq = nports * (num_vis - 1) * iaq.nnmrxq_vi;
1070 		s->nnmtxq = nports * (num_vis - 1) * iaq.nnmtxq_vi;
1071 	}
1072 	s->neq += s->nnmtxq + s->nnmrxq;
1073 	s->niq += s->nnmrxq;
1074 
1075 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1076 	    M_CXGBE, M_ZERO | M_WAITOK);
1077 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1078 	    M_CXGBE, M_ZERO | M_WAITOK);
1079 #endif
1080 
1081 	s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1082 	    M_ZERO | M_WAITOK);
1083 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1084 	    M_ZERO | M_WAITOK);
1085 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1086 	    M_ZERO | M_WAITOK);
1087 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1088 	    M_ZERO | M_WAITOK);
1089 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1090 	    M_ZERO | M_WAITOK);
1091 
1092 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1093 	    M_ZERO | M_WAITOK);
1094 
1095 	t4_init_l2t(sc, M_WAITOK);
1096 	t4_init_tx_sched(sc);
1097 
1098 	/*
1099 	 * Second pass over the ports.  This time we know the number of rx and
1100 	 * tx queues that each port should get.
1101 	 */
1102 	rqidx = tqidx = 0;
1103 #ifdef TCP_OFFLOAD
1104 	ofld_rqidx = ofld_tqidx = 0;
1105 #endif
1106 #ifdef DEV_NETMAP
1107 	nm_rqidx = nm_tqidx = 0;
1108 #endif
1109 	for_each_port(sc, i) {
1110 		struct port_info *pi = sc->port[i];
1111 		struct vi_info *vi;
1112 
1113 		if (pi == NULL)
1114 			continue;
1115 
1116 		pi->nvi = num_vis;
1117 		for_each_vi(pi, j, vi) {
1118 			vi->pi = pi;
1119 			vi->qsize_rxq = t4_qsize_rxq;
1120 			vi->qsize_txq = t4_qsize_txq;
1121 
1122 			vi->first_rxq = rqidx;
1123 			vi->first_txq = tqidx;
1124 			vi->tmr_idx = t4_tmr_idx;
1125 			vi->pktc_idx = t4_pktc_idx;
1126 			vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1127 			vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1128 
1129 			rqidx += vi->nrxq;
1130 			tqidx += vi->ntxq;
1131 
1132 			if (j == 0 && vi->ntxq > 1)
1133 				vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1134 			else
1135 				vi->rsrv_noflowq = 0;
1136 
1137 #ifdef TCP_OFFLOAD
1138 			vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1139 			vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1140 			vi->first_ofld_rxq = ofld_rqidx;
1141 			vi->first_ofld_txq = ofld_tqidx;
1142 			vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1143 			vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1144 
1145 			ofld_rqidx += vi->nofldrxq;
1146 			ofld_tqidx += vi->nofldtxq;
1147 #endif
1148 #ifdef DEV_NETMAP
1149 			if (j > 0) {
1150 				vi->first_nm_rxq = nm_rqidx;
1151 				vi->first_nm_txq = nm_tqidx;
1152 				vi->nnmrxq = iaq.nnmrxq_vi;
1153 				vi->nnmtxq = iaq.nnmtxq_vi;
1154 				nm_rqidx += vi->nnmrxq;
1155 				nm_tqidx += vi->nnmtxq;
1156 			}
1157 #endif
1158 		}
1159 	}
1160 
1161 	rc = t4_setup_intr_handlers(sc);
1162 	if (rc != 0) {
1163 		device_printf(dev,
1164 		    "failed to setup interrupt handlers: %d\n", rc);
1165 		goto done;
1166 	}
1167 
1168 	rc = bus_generic_probe(dev);
1169 	if (rc != 0) {
1170 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1171 		goto done;
1172 	}
1173 
1174 	/*
1175 	 * Ensure thread-safe mailbox access (in debug builds).
1176 	 *
1177 	 * So far this was the only thread accessing the mailbox but various
1178 	 * ifnets and sysctls are about to be created and their handlers/ioctls
1179 	 * will access the mailbox from different threads.
1180 	 */
1181 	sc->flags |= CHK_MBOX_ACCESS;
1182 
1183 	rc = bus_generic_attach(dev);
1184 	if (rc != 0) {
1185 		device_printf(dev,
1186 		    "failed to attach all child ports: %d\n", rc);
1187 		goto done;
1188 	}
1189 
1190 	device_printf(dev,
1191 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1192 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1193 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1194 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1195 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1196 
1197 	t4_set_desc(sc);
1198 
1199 	notify_siblings(dev, 0);
1200 
1201 done:
1202 	if (rc != 0 && sc->cdev) {
1203 		/* cdev was created and so cxgbetool works; recover that way. */
1204 		device_printf(dev,
1205 		    "error during attach, adapter is now in recovery mode.\n");
1206 		rc = 0;
1207 	}
1208 
1209 	if (rc != 0)
1210 		t4_detach_common(dev);
1211 	else
1212 		t4_sysctls(sc);
1213 
1214 	return (rc);
1215 }
1216 
1217 static int
1218 t4_ready(device_t dev)
1219 {
1220 	struct adapter *sc;
1221 
1222 	sc = device_get_softc(dev);
1223 	if (sc->flags & FW_OK)
1224 		return (0);
1225 	return (ENXIO);
1226 }
1227 
1228 static int
1229 t4_read_port_device(device_t dev, int port, device_t *child)
1230 {
1231 	struct adapter *sc;
1232 	struct port_info *pi;
1233 
1234 	sc = device_get_softc(dev);
1235 	if (port < 0 || port >= MAX_NPORTS)
1236 		return (EINVAL);
1237 	pi = sc->port[port];
1238 	if (pi == NULL || pi->dev == NULL)
1239 		return (ENXIO);
1240 	*child = pi->dev;
1241 	return (0);
1242 }
1243 
1244 static int
1245 notify_siblings(device_t dev, int detaching)
1246 {
1247 	device_t sibling;
1248 	int error, i;
1249 
1250 	error = 0;
1251 	for (i = 0; i < PCI_FUNCMAX; i++) {
1252 		if (i == pci_get_function(dev))
1253 			continue;
1254 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1255 		    pci_get_slot(dev), i);
1256 		if (sibling == NULL || !device_is_attached(sibling))
1257 			continue;
1258 		if (detaching)
1259 			error = T4_DETACH_CHILD(sibling);
1260 		else
1261 			(void)T4_ATTACH_CHILD(sibling);
1262 		if (error)
1263 			break;
1264 	}
1265 	return (error);
1266 }
1267 
1268 /*
1269  * Idempotent
1270  */
1271 static int
1272 t4_detach(device_t dev)
1273 {
1274 	struct adapter *sc;
1275 	int rc;
1276 
1277 	sc = device_get_softc(dev);
1278 
1279 	rc = notify_siblings(dev, 1);
1280 	if (rc) {
1281 		device_printf(dev,
1282 		    "failed to detach sibling devices: %d\n", rc);
1283 		return (rc);
1284 	}
1285 
1286 	return (t4_detach_common(dev));
1287 }
1288 
1289 int
1290 t4_detach_common(device_t dev)
1291 {
1292 	struct adapter *sc;
1293 	struct port_info *pi;
1294 	int i, rc;
1295 
1296 	sc = device_get_softc(dev);
1297 
1298 	if (sc->cdev) {
1299 		destroy_dev(sc->cdev);
1300 		sc->cdev = NULL;
1301 	}
1302 
1303 	sc->flags &= ~CHK_MBOX_ACCESS;
1304 	if (sc->flags & FULL_INIT_DONE) {
1305 		if (!(sc->flags & IS_VF))
1306 			t4_intr_disable(sc);
1307 	}
1308 
1309 	if (device_is_attached(dev)) {
1310 		rc = bus_generic_detach(dev);
1311 		if (rc) {
1312 			device_printf(dev,
1313 			    "failed to detach child devices: %d\n", rc);
1314 			return (rc);
1315 		}
1316 	}
1317 
1318 	for (i = 0; i < sc->intr_count; i++)
1319 		t4_free_irq(sc, &sc->irq[i]);
1320 
1321 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1322 		t4_free_tx_sched(sc);
1323 
1324 	for (i = 0; i < MAX_NPORTS; i++) {
1325 		pi = sc->port[i];
1326 		if (pi) {
1327 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1328 			if (pi->dev)
1329 				device_delete_child(dev, pi->dev);
1330 
1331 			mtx_destroy(&pi->pi_lock);
1332 			free(pi->vi, M_CXGBE);
1333 			free(pi, M_CXGBE);
1334 		}
1335 	}
1336 
1337 	device_delete_children(dev);
1338 
1339 	if (sc->flags & FULL_INIT_DONE)
1340 		adapter_full_uninit(sc);
1341 
1342 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1343 		t4_fw_bye(sc, sc->mbox);
1344 
1345 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1346 		pci_release_msi(dev);
1347 
1348 	if (sc->regs_res)
1349 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1350 		    sc->regs_res);
1351 
1352 	if (sc->udbs_res)
1353 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1354 		    sc->udbs_res);
1355 
1356 	if (sc->msix_res)
1357 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1358 		    sc->msix_res);
1359 
1360 	if (sc->l2t)
1361 		t4_free_l2t(sc->l2t);
1362 
1363 #ifdef TCP_OFFLOAD
1364 	free(sc->sge.ofld_rxq, M_CXGBE);
1365 	free(sc->sge.ofld_txq, M_CXGBE);
1366 #endif
1367 #ifdef DEV_NETMAP
1368 	free(sc->sge.nm_rxq, M_CXGBE);
1369 	free(sc->sge.nm_txq, M_CXGBE);
1370 #endif
1371 	free(sc->irq, M_CXGBE);
1372 	free(sc->sge.rxq, M_CXGBE);
1373 	free(sc->sge.txq, M_CXGBE);
1374 	free(sc->sge.ctrlq, M_CXGBE);
1375 	free(sc->sge.iqmap, M_CXGBE);
1376 	free(sc->sge.eqmap, M_CXGBE);
1377 	free(sc->tids.ftid_tab, M_CXGBE);
1378 	free(sc->tt.tls_rx_ports, M_CXGBE);
1379 	t4_destroy_dma_tag(sc);
1380 	if (mtx_initialized(&sc->sc_lock)) {
1381 		sx_xlock(&t4_list_lock);
1382 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1383 		sx_xunlock(&t4_list_lock);
1384 		mtx_destroy(&sc->sc_lock);
1385 	}
1386 
1387 	callout_drain(&sc->sfl_callout);
1388 	if (mtx_initialized(&sc->tids.ftid_lock))
1389 		mtx_destroy(&sc->tids.ftid_lock);
1390 	if (mtx_initialized(&sc->sfl_lock))
1391 		mtx_destroy(&sc->sfl_lock);
1392 	if (mtx_initialized(&sc->ifp_lock))
1393 		mtx_destroy(&sc->ifp_lock);
1394 	if (mtx_initialized(&sc->reg_lock))
1395 		mtx_destroy(&sc->reg_lock);
1396 
1397 	if (rw_initialized(&sc->policy_lock)) {
1398 		rw_destroy(&sc->policy_lock);
1399 #ifdef TCP_OFFLOAD
1400 		if (sc->policy != NULL)
1401 			free_offload_policy(sc->policy);
1402 #endif
1403 	}
1404 
1405 	for (i = 0; i < NUM_MEMWIN; i++) {
1406 		struct memwin *mw = &sc->memwin[i];
1407 
1408 		if (rw_initialized(&mw->mw_lock))
1409 			rw_destroy(&mw->mw_lock);
1410 	}
1411 
1412 	bzero(sc, sizeof(*sc));
1413 
1414 	return (0);
1415 }
1416 
1417 static int
1418 cxgbe_probe(device_t dev)
1419 {
1420 	char buf[128];
1421 	struct port_info *pi = device_get_softc(dev);
1422 
1423 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1424 	device_set_desc_copy(dev, buf);
1425 
1426 	return (BUS_PROBE_DEFAULT);
1427 }
1428 
1429 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1430     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1431     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1432 #define T4_CAP_ENABLE (T4_CAP)
1433 
1434 static int
1435 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1436 {
1437 	struct ifnet *ifp;
1438 	struct sbuf *sb;
1439 
1440 	vi->xact_addr_filt = -1;
1441 	callout_init(&vi->tick, 1);
1442 
1443 	/* Allocate an ifnet and set it up */
1444 	ifp = if_alloc(IFT_ETHER);
1445 	if (ifp == NULL) {
1446 		device_printf(dev, "Cannot allocate ifnet\n");
1447 		return (ENOMEM);
1448 	}
1449 	vi->ifp = ifp;
1450 	ifp->if_softc = vi;
1451 
1452 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1453 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1454 
1455 	ifp->if_init = cxgbe_init;
1456 	ifp->if_ioctl = cxgbe_ioctl;
1457 	ifp->if_transmit = cxgbe_transmit;
1458 	ifp->if_qflush = cxgbe_qflush;
1459 	ifp->if_get_counter = cxgbe_get_counter;
1460 
1461 	ifp->if_capabilities = T4_CAP;
1462 #ifdef TCP_OFFLOAD
1463 	if (vi->nofldrxq != 0)
1464 		ifp->if_capabilities |= IFCAP_TOE;
1465 #endif
1466 #ifdef DEV_NETMAP
1467 	if (vi->nnmrxq != 0)
1468 		ifp->if_capabilities |= IFCAP_NETMAP;
1469 #endif
1470 	ifp->if_capenable = T4_CAP_ENABLE;
1471 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1472 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1473 
1474 	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1475 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1476 	ifp->if_hw_tsomaxsegsize = 65536;
1477 
1478 	vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1479 	    EVENTHANDLER_PRI_ANY);
1480 
1481 	ether_ifattach(ifp, vi->hw_addr);
1482 #ifdef DEV_NETMAP
1483 	if (ifp->if_capabilities & IFCAP_NETMAP)
1484 		cxgbe_nm_attach(vi);
1485 #endif
1486 	sb = sbuf_new_auto();
1487 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1488 #ifdef TCP_OFFLOAD
1489 	if (ifp->if_capabilities & IFCAP_TOE)
1490 		sbuf_printf(sb, "; %d txq, %d rxq (TOE)",
1491 		    vi->nofldtxq, vi->nofldrxq);
1492 #endif
1493 #ifdef DEV_NETMAP
1494 	if (ifp->if_capabilities & IFCAP_NETMAP)
1495 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1496 		    vi->nnmtxq, vi->nnmrxq);
1497 #endif
1498 	sbuf_finish(sb);
1499 	device_printf(dev, "%s\n", sbuf_data(sb));
1500 	sbuf_delete(sb);
1501 
1502 	vi_sysctls(vi);
1503 
1504 	return (0);
1505 }
1506 
1507 static int
1508 cxgbe_attach(device_t dev)
1509 {
1510 	struct port_info *pi = device_get_softc(dev);
1511 	struct adapter *sc = pi->adapter;
1512 	struct vi_info *vi;
1513 	int i, rc;
1514 
1515 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1516 
1517 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1518 	if (rc)
1519 		return (rc);
1520 
1521 	for_each_vi(pi, i, vi) {
1522 		if (i == 0)
1523 			continue;
1524 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1525 		if (vi->dev == NULL) {
1526 			device_printf(dev, "failed to add VI %d\n", i);
1527 			continue;
1528 		}
1529 		device_set_softc(vi->dev, vi);
1530 	}
1531 
1532 	cxgbe_sysctls(pi);
1533 
1534 	bus_generic_attach(dev);
1535 
1536 	return (0);
1537 }
1538 
1539 static void
1540 cxgbe_vi_detach(struct vi_info *vi)
1541 {
1542 	struct ifnet *ifp = vi->ifp;
1543 
1544 	ether_ifdetach(ifp);
1545 
1546 	if (vi->vlan_c)
1547 		EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1548 
1549 	/* Let detach proceed even if these fail. */
1550 #ifdef DEV_NETMAP
1551 	if (ifp->if_capabilities & IFCAP_NETMAP)
1552 		cxgbe_nm_detach(vi);
1553 #endif
1554 	cxgbe_uninit_synchronized(vi);
1555 	callout_drain(&vi->tick);
1556 	vi_full_uninit(vi);
1557 
1558 	if_free(vi->ifp);
1559 	vi->ifp = NULL;
1560 }
1561 
1562 static int
1563 cxgbe_detach(device_t dev)
1564 {
1565 	struct port_info *pi = device_get_softc(dev);
1566 	struct adapter *sc = pi->adapter;
1567 	int rc;
1568 
1569 	/* Detach the extra VIs first. */
1570 	rc = bus_generic_detach(dev);
1571 	if (rc)
1572 		return (rc);
1573 	device_delete_children(dev);
1574 
1575 	doom_vi(sc, &pi->vi[0]);
1576 
1577 	if (pi->flags & HAS_TRACEQ) {
1578 		sc->traceq = -1;	/* cloner should not create ifnet */
1579 		t4_tracer_port_detach(sc);
1580 	}
1581 
1582 	cxgbe_vi_detach(&pi->vi[0]);
1583 	callout_drain(&pi->tick);
1584 	ifmedia_removeall(&pi->media);
1585 
1586 	end_synchronized_op(sc, 0);
1587 
1588 	return (0);
1589 }
1590 
1591 static void
1592 cxgbe_init(void *arg)
1593 {
1594 	struct vi_info *vi = arg;
1595 	struct adapter *sc = vi->pi->adapter;
1596 
1597 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1598 		return;
1599 	cxgbe_init_synchronized(vi);
1600 	end_synchronized_op(sc, 0);
1601 }
1602 
1603 static int
1604 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1605 {
1606 	int rc = 0, mtu, flags, can_sleep;
1607 	struct vi_info *vi = ifp->if_softc;
1608 	struct port_info *pi = vi->pi;
1609 	struct adapter *sc = pi->adapter;
1610 	struct ifreq *ifr = (struct ifreq *)data;
1611 	uint32_t mask;
1612 
1613 	switch (cmd) {
1614 	case SIOCSIFMTU:
1615 		mtu = ifr->ifr_mtu;
1616 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1617 			return (EINVAL);
1618 
1619 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1620 		if (rc)
1621 			return (rc);
1622 		ifp->if_mtu = mtu;
1623 		if (vi->flags & VI_INIT_DONE) {
1624 			t4_update_fl_bufsize(ifp);
1625 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1626 				rc = update_mac_settings(ifp, XGMAC_MTU);
1627 		}
1628 		end_synchronized_op(sc, 0);
1629 		break;
1630 
1631 	case SIOCSIFFLAGS:
1632 		can_sleep = 0;
1633 redo_sifflags:
1634 		rc = begin_synchronized_op(sc, vi,
1635 		    can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1636 		if (rc) {
1637 			if_printf(ifp, "%ssleepable synch operation failed: %d."
1638 			    "  if_flags 0x%08x, if_drv_flags 0x%08x\n",
1639 			    can_sleep ? "" : "non-", rc, ifp->if_flags,
1640 			    ifp->if_drv_flags);
1641 			return (rc);
1642 		}
1643 
1644 		if (ifp->if_flags & IFF_UP) {
1645 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1646 				flags = vi->if_flags;
1647 				if ((ifp->if_flags ^ flags) &
1648 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1649 					if (can_sleep == 1) {
1650 						end_synchronized_op(sc, 0);
1651 						can_sleep = 0;
1652 						goto redo_sifflags;
1653 					}
1654 					rc = update_mac_settings(ifp,
1655 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1656 				}
1657 			} else {
1658 				if (can_sleep == 0) {
1659 					end_synchronized_op(sc, LOCK_HELD);
1660 					can_sleep = 1;
1661 					goto redo_sifflags;
1662 				}
1663 				rc = cxgbe_init_synchronized(vi);
1664 			}
1665 			vi->if_flags = ifp->if_flags;
1666 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1667 			if (can_sleep == 0) {
1668 				end_synchronized_op(sc, LOCK_HELD);
1669 				can_sleep = 1;
1670 				goto redo_sifflags;
1671 			}
1672 			rc = cxgbe_uninit_synchronized(vi);
1673 		}
1674 		end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1675 		break;
1676 
1677 	case SIOCADDMULTI:
1678 	case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1679 		rc = begin_synchronized_op(sc, vi, HOLD_LOCK, "t4multi");
1680 		if (rc)
1681 			return (rc);
1682 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1683 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1684 		end_synchronized_op(sc, LOCK_HELD);
1685 		break;
1686 
1687 	case SIOCSIFCAP:
1688 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1689 		if (rc)
1690 			return (rc);
1691 
1692 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1693 		if (mask & IFCAP_TXCSUM) {
1694 			ifp->if_capenable ^= IFCAP_TXCSUM;
1695 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1696 
1697 			if (IFCAP_TSO4 & ifp->if_capenable &&
1698 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1699 				ifp->if_capenable &= ~IFCAP_TSO4;
1700 				if_printf(ifp,
1701 				    "tso4 disabled due to -txcsum.\n");
1702 			}
1703 		}
1704 		if (mask & IFCAP_TXCSUM_IPV6) {
1705 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1706 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1707 
1708 			if (IFCAP_TSO6 & ifp->if_capenable &&
1709 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1710 				ifp->if_capenable &= ~IFCAP_TSO6;
1711 				if_printf(ifp,
1712 				    "tso6 disabled due to -txcsum6.\n");
1713 			}
1714 		}
1715 		if (mask & IFCAP_RXCSUM)
1716 			ifp->if_capenable ^= IFCAP_RXCSUM;
1717 		if (mask & IFCAP_RXCSUM_IPV6)
1718 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1719 
1720 		/*
1721 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1722 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1723 		 * sending a TSO request our way, so it's sufficient to toggle
1724 		 * IFCAP_TSOx only.
1725 		 */
1726 		if (mask & IFCAP_TSO4) {
1727 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1728 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1729 				if_printf(ifp, "enable txcsum first.\n");
1730 				rc = EAGAIN;
1731 				goto fail;
1732 			}
1733 			ifp->if_capenable ^= IFCAP_TSO4;
1734 		}
1735 		if (mask & IFCAP_TSO6) {
1736 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1737 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1738 				if_printf(ifp, "enable txcsum6 first.\n");
1739 				rc = EAGAIN;
1740 				goto fail;
1741 			}
1742 			ifp->if_capenable ^= IFCAP_TSO6;
1743 		}
1744 		if (mask & IFCAP_LRO) {
1745 #if defined(INET) || defined(INET6)
1746 			int i;
1747 			struct sge_rxq *rxq;
1748 
1749 			ifp->if_capenable ^= IFCAP_LRO;
1750 			for_each_rxq(vi, i, rxq) {
1751 				if (ifp->if_capenable & IFCAP_LRO)
1752 					rxq->iq.flags |= IQ_LRO_ENABLED;
1753 				else
1754 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1755 			}
1756 #endif
1757 		}
1758 #ifdef TCP_OFFLOAD
1759 		if (mask & IFCAP_TOE) {
1760 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1761 
1762 			rc = toe_capability(vi, enable);
1763 			if (rc != 0)
1764 				goto fail;
1765 
1766 			ifp->if_capenable ^= mask;
1767 		}
1768 #endif
1769 		if (mask & IFCAP_VLAN_HWTAGGING) {
1770 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1771 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1772 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1773 		}
1774 		if (mask & IFCAP_VLAN_MTU) {
1775 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1776 
1777 			/* Need to find out how to disable auto-mtu-inflation */
1778 		}
1779 		if (mask & IFCAP_VLAN_HWTSO)
1780 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1781 		if (mask & IFCAP_VLAN_HWCSUM)
1782 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1783 
1784 #ifdef VLAN_CAPABILITIES
1785 		VLAN_CAPABILITIES(ifp);
1786 #endif
1787 fail:
1788 		end_synchronized_op(sc, 0);
1789 		break;
1790 
1791 	case SIOCSIFMEDIA:
1792 	case SIOCGIFMEDIA:
1793 	case SIOCGIFXMEDIA:
1794 		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1795 		break;
1796 
1797 	case SIOCGI2C: {
1798 		struct ifi2creq i2c;
1799 
1800 		rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
1801 		if (rc != 0)
1802 			break;
1803 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1804 			rc = EPERM;
1805 			break;
1806 		}
1807 		if (i2c.len > sizeof(i2c.data)) {
1808 			rc = EINVAL;
1809 			break;
1810 		}
1811 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1812 		if (rc)
1813 			return (rc);
1814 		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
1815 		    i2c.offset, i2c.len, &i2c.data[0]);
1816 		end_synchronized_op(sc, 0);
1817 		if (rc == 0)
1818 			rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
1819 		break;
1820 	}
1821 
1822 	default:
1823 		rc = ether_ioctl(ifp, cmd, data);
1824 	}
1825 
1826 	return (rc);
1827 }
1828 
1829 static int
1830 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1831 {
1832 	struct vi_info *vi = ifp->if_softc;
1833 	struct port_info *pi = vi->pi;
1834 	struct adapter *sc = pi->adapter;
1835 	struct sge_txq *txq;
1836 	void *items[1];
1837 	int rc;
1838 
1839 	M_ASSERTPKTHDR(m);
1840 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1841 
1842 	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1843 		m_freem(m);
1844 		return (ENETDOWN);
1845 	}
1846 
1847 	rc = parse_pkt(sc, &m);
1848 	if (__predict_false(rc != 0)) {
1849 		MPASS(m == NULL);			/* was freed already */
1850 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1851 		return (rc);
1852 	}
1853 
1854 	/* Select a txq. */
1855 	txq = &sc->sge.txq[vi->first_txq];
1856 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1857 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1858 		    vi->rsrv_noflowq);
1859 
1860 	items[0] = m;
1861 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1862 	if (__predict_false(rc != 0))
1863 		m_freem(m);
1864 
1865 	return (rc);
1866 }
1867 
1868 static void
1869 cxgbe_qflush(struct ifnet *ifp)
1870 {
1871 	struct vi_info *vi = ifp->if_softc;
1872 	struct sge_txq *txq;
1873 	int i;
1874 
1875 	/* queues do not exist if !VI_INIT_DONE. */
1876 	if (vi->flags & VI_INIT_DONE) {
1877 		for_each_txq(vi, i, txq) {
1878 			TXQ_LOCK(txq);
1879 			txq->eq.flags |= EQ_QFLUSH;
1880 			TXQ_UNLOCK(txq);
1881 			while (!mp_ring_is_idle(txq->r)) {
1882 				mp_ring_check_drainage(txq->r, 0);
1883 				pause("qflush", 1);
1884 			}
1885 			TXQ_LOCK(txq);
1886 			txq->eq.flags &= ~EQ_QFLUSH;
1887 			TXQ_UNLOCK(txq);
1888 		}
1889 	}
1890 	if_qflush(ifp);
1891 }
1892 
1893 static uint64_t
1894 vi_get_counter(struct ifnet *ifp, ift_counter c)
1895 {
1896 	struct vi_info *vi = ifp->if_softc;
1897 	struct fw_vi_stats_vf *s = &vi->stats;
1898 
1899 	vi_refresh_stats(vi->pi->adapter, vi);
1900 
1901 	switch (c) {
1902 	case IFCOUNTER_IPACKETS:
1903 		return (s->rx_bcast_frames + s->rx_mcast_frames +
1904 		    s->rx_ucast_frames);
1905 	case IFCOUNTER_IERRORS:
1906 		return (s->rx_err_frames);
1907 	case IFCOUNTER_OPACKETS:
1908 		return (s->tx_bcast_frames + s->tx_mcast_frames +
1909 		    s->tx_ucast_frames + s->tx_offload_frames);
1910 	case IFCOUNTER_OERRORS:
1911 		return (s->tx_drop_frames);
1912 	case IFCOUNTER_IBYTES:
1913 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
1914 		    s->rx_ucast_bytes);
1915 	case IFCOUNTER_OBYTES:
1916 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
1917 		    s->tx_ucast_bytes + s->tx_offload_bytes);
1918 	case IFCOUNTER_IMCASTS:
1919 		return (s->rx_mcast_frames);
1920 	case IFCOUNTER_OMCASTS:
1921 		return (s->tx_mcast_frames);
1922 	case IFCOUNTER_OQDROPS: {
1923 		uint64_t drops;
1924 
1925 		drops = 0;
1926 		if (vi->flags & VI_INIT_DONE) {
1927 			int i;
1928 			struct sge_txq *txq;
1929 
1930 			for_each_txq(vi, i, txq)
1931 				drops += counter_u64_fetch(txq->r->drops);
1932 		}
1933 
1934 		return (drops);
1935 
1936 	}
1937 
1938 	default:
1939 		return (if_get_counter_default(ifp, c));
1940 	}
1941 }
1942 
1943 uint64_t
1944 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
1945 {
1946 	struct vi_info *vi = ifp->if_softc;
1947 	struct port_info *pi = vi->pi;
1948 	struct adapter *sc = pi->adapter;
1949 	struct port_stats *s = &pi->stats;
1950 
1951 	if (pi->nvi > 1 || sc->flags & IS_VF)
1952 		return (vi_get_counter(ifp, c));
1953 
1954 	cxgbe_refresh_stats(sc, pi);
1955 
1956 	switch (c) {
1957 	case IFCOUNTER_IPACKETS:
1958 		return (s->rx_frames);
1959 
1960 	case IFCOUNTER_IERRORS:
1961 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
1962 		    s->rx_fcs_err + s->rx_len_err);
1963 
1964 	case IFCOUNTER_OPACKETS:
1965 		return (s->tx_frames);
1966 
1967 	case IFCOUNTER_OERRORS:
1968 		return (s->tx_error_frames);
1969 
1970 	case IFCOUNTER_IBYTES:
1971 		return (s->rx_octets);
1972 
1973 	case IFCOUNTER_OBYTES:
1974 		return (s->tx_octets);
1975 
1976 	case IFCOUNTER_IMCASTS:
1977 		return (s->rx_mcast_frames);
1978 
1979 	case IFCOUNTER_OMCASTS:
1980 		return (s->tx_mcast_frames);
1981 
1982 	case IFCOUNTER_IQDROPS:
1983 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
1984 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
1985 		    s->rx_trunc3 + pi->tnl_cong_drops);
1986 
1987 	case IFCOUNTER_OQDROPS: {
1988 		uint64_t drops;
1989 
1990 		drops = s->tx_drop;
1991 		if (vi->flags & VI_INIT_DONE) {
1992 			int i;
1993 			struct sge_txq *txq;
1994 
1995 			for_each_txq(vi, i, txq)
1996 				drops += counter_u64_fetch(txq->r->drops);
1997 		}
1998 
1999 		return (drops);
2000 
2001 	}
2002 
2003 	default:
2004 		return (if_get_counter_default(ifp, c));
2005 	}
2006 }
2007 
2008 static int
2009 cxgbe_media_change(struct ifnet *ifp)
2010 {
2011 	struct vi_info *vi = ifp->if_softc;
2012 
2013 	device_printf(vi->dev, "%s unimplemented.\n", __func__);
2014 
2015 	return (EOPNOTSUPP);
2016 }
2017 
2018 static void
2019 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2020 {
2021 	struct vi_info *vi = ifp->if_softc;
2022 	struct port_info *pi = vi->pi;
2023 	struct ifmedia_entry *cur;
2024 	struct link_config *lc = &pi->link_cfg;
2025 
2026 	/*
2027 	 * If all the interfaces are administratively down the firmware does not
2028 	 * report transceiver changes.  Refresh port info here so that ifconfig
2029 	 * displays accurate information at all times.
2030 	 */
2031 	if (begin_synchronized_op(pi->adapter, NULL, SLEEP_OK | INTR_OK,
2032 	    "t4med") == 0) {
2033 		PORT_LOCK(pi);
2034 		if (pi->up_vis == 0) {
2035 			t4_update_port_info(pi);
2036 			build_medialist(pi, &pi->media);
2037 		}
2038 		PORT_UNLOCK(pi);
2039 		end_synchronized_op(pi->adapter, 0);
2040 	}
2041 
2042 	ifmr->ifm_status = IFM_AVALID;
2043 	if (lc->link_ok == 0)
2044 		return;
2045 
2046 	ifmr->ifm_status |= IFM_ACTIVE;
2047 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2048 	if (lc->fc & PAUSE_RX)
2049 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2050 	if (lc->fc & PAUSE_TX)
2051 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2052 
2053 	/* active and current will differ iff current media is autoselect. */
2054 	cur = pi->media.ifm_cur;
2055 	if (cur != NULL && IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
2056 		return;
2057 
2058 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2059 	if (lc->fc & PAUSE_RX)
2060 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2061 	if (lc->fc & PAUSE_TX)
2062 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2063 	switch (lc->speed) {
2064 	case 10000:
2065 		ifmr->ifm_active |= IFM_10G_T;
2066 		break;
2067 	case 1000:
2068 		ifmr->ifm_active |= IFM_1000_T;
2069 		break;
2070 	case 100:
2071 		ifmr->ifm_active |= IFM_100_TX;
2072 		break;
2073 	case 10:
2074 		ifmr->ifm_active |= IFM_10_T;
2075 		break;
2076 	default:
2077 		device_printf(vi->dev, "link up but speed unknown (%u)\n",
2078 		    lc->speed);
2079 	}
2080 }
2081 
2082 static int
2083 vcxgbe_probe(device_t dev)
2084 {
2085 	char buf[128];
2086 	struct vi_info *vi = device_get_softc(dev);
2087 
2088 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2089 	    vi - vi->pi->vi);
2090 	device_set_desc_copy(dev, buf);
2091 
2092 	return (BUS_PROBE_DEFAULT);
2093 }
2094 
2095 static int
2096 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2097 {
2098 	int func, index, rc;
2099 	uint32_t param, val;
2100 
2101 	ASSERT_SYNCHRONIZED_OP(sc);
2102 
2103 	index = vi - pi->vi;
2104 	MPASS(index > 0);	/* This function deals with _extra_ VIs only */
2105 	KASSERT(index < nitems(vi_mac_funcs),
2106 	    ("%s: VI %s doesn't have a MAC func", __func__,
2107 	    device_get_nameunit(vi->dev)));
2108 	func = vi_mac_funcs[index];
2109 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2110 	    vi->hw_addr, &vi->rss_size, func, 0);
2111 	if (rc < 0) {
2112 		device_printf(vi->dev, "failed to allocate virtual interface %d"
2113 		    "for port %d: %d\n", index, pi->port_id, -rc);
2114 		return (-rc);
2115 	}
2116 	vi->viid = rc;
2117 	if (chip_id(sc) <= CHELSIO_T5)
2118 		vi->smt_idx = (rc & 0x7f) << 1;
2119 	else
2120 		vi->smt_idx = (rc & 0x7f);
2121 
2122 	if (vi->rss_size == 1) {
2123 		/*
2124 		 * This VI didn't get a slice of the RSS table.  Reduce the
2125 		 * number of VIs being created (hw.cxgbe.num_vis) or modify the
2126 		 * configuration file (nvi, rssnvi for this PF) if this is a
2127 		 * problem.
2128 		 */
2129 		device_printf(vi->dev, "RSS table not available.\n");
2130 		vi->rss_base = 0xffff;
2131 
2132 		return (0);
2133 	}
2134 
2135 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2136 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2137 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2138 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2139 	if (rc)
2140 		vi->rss_base = 0xffff;
2141 	else {
2142 		MPASS((val >> 16) == vi->rss_size);
2143 		vi->rss_base = val & 0xffff;
2144 	}
2145 
2146 	return (0);
2147 }
2148 
2149 static int
2150 vcxgbe_attach(device_t dev)
2151 {
2152 	struct vi_info *vi;
2153 	struct port_info *pi;
2154 	struct adapter *sc;
2155 	int rc;
2156 
2157 	vi = device_get_softc(dev);
2158 	pi = vi->pi;
2159 	sc = pi->adapter;
2160 
2161 	rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2162 	if (rc)
2163 		return (rc);
2164 	rc = alloc_extra_vi(sc, pi, vi);
2165 	end_synchronized_op(sc, 0);
2166 	if (rc)
2167 		return (rc);
2168 
2169 	rc = cxgbe_vi_attach(dev, vi);
2170 	if (rc) {
2171 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2172 		return (rc);
2173 	}
2174 	return (0);
2175 }
2176 
2177 static int
2178 vcxgbe_detach(device_t dev)
2179 {
2180 	struct vi_info *vi;
2181 	struct adapter *sc;
2182 
2183 	vi = device_get_softc(dev);
2184 	sc = vi->pi->adapter;
2185 
2186 	doom_vi(sc, vi);
2187 
2188 	cxgbe_vi_detach(vi);
2189 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2190 
2191 	end_synchronized_op(sc, 0);
2192 
2193 	return (0);
2194 }
2195 
2196 void
2197 t4_fatal_err(struct adapter *sc)
2198 {
2199 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2200 	t4_intr_disable(sc);
2201 	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2202 	    device_get_nameunit(sc->dev));
2203 	if (t4_panic_on_fatal_err)
2204 		panic("panic requested on fatal error");
2205 }
2206 
2207 void
2208 t4_add_adapter(struct adapter *sc)
2209 {
2210 	sx_xlock(&t4_list_lock);
2211 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2212 	sx_xunlock(&t4_list_lock);
2213 }
2214 
2215 int
2216 t4_map_bars_0_and_4(struct adapter *sc)
2217 {
2218 	sc->regs_rid = PCIR_BAR(0);
2219 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2220 	    &sc->regs_rid, RF_ACTIVE);
2221 	if (sc->regs_res == NULL) {
2222 		device_printf(sc->dev, "cannot map registers.\n");
2223 		return (ENXIO);
2224 	}
2225 	sc->bt = rman_get_bustag(sc->regs_res);
2226 	sc->bh = rman_get_bushandle(sc->regs_res);
2227 	sc->mmio_len = rman_get_size(sc->regs_res);
2228 	setbit(&sc->doorbells, DOORBELL_KDB);
2229 
2230 	sc->msix_rid = PCIR_BAR(4);
2231 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2232 	    &sc->msix_rid, RF_ACTIVE);
2233 	if (sc->msix_res == NULL) {
2234 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2235 		return (ENXIO);
2236 	}
2237 
2238 	return (0);
2239 }
2240 
2241 int
2242 t4_map_bar_2(struct adapter *sc)
2243 {
2244 
2245 	/*
2246 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2247 	 * to map it if RDMA is disabled.
2248 	 */
2249 	if (is_t4(sc) && sc->rdmacaps == 0)
2250 		return (0);
2251 
2252 	sc->udbs_rid = PCIR_BAR(2);
2253 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2254 	    &sc->udbs_rid, RF_ACTIVE);
2255 	if (sc->udbs_res == NULL) {
2256 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2257 		return (ENXIO);
2258 	}
2259 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2260 
2261 	if (chip_id(sc) >= CHELSIO_T5) {
2262 		setbit(&sc->doorbells, DOORBELL_UDB);
2263 #if defined(__i386__) || defined(__amd64__)
2264 		if (t5_write_combine) {
2265 			int rc, mode;
2266 
2267 			/*
2268 			 * Enable write combining on BAR2.  This is the
2269 			 * userspace doorbell BAR and is split into 128B
2270 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2271 			 * with an egress queue.  The first 64B has the doorbell
2272 			 * and the second 64B can be used to submit a tx work
2273 			 * request with an implicit doorbell.
2274 			 */
2275 
2276 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2277 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2278 			if (rc == 0) {
2279 				clrbit(&sc->doorbells, DOORBELL_UDB);
2280 				setbit(&sc->doorbells, DOORBELL_WCWR);
2281 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2282 			} else {
2283 				t5_write_combine = 0;
2284 				device_printf(sc->dev,
2285 				    "couldn't enable write combining: %d\n",
2286 				    rc);
2287 			}
2288 
2289 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2290 			t4_write_reg(sc, A_SGE_STAT_CFG,
2291 			    V_STATSOURCE_T5(7) | mode);
2292 		}
2293 #else
2294 		t5_write_combine = 0;
2295 #endif
2296 		sc->iwt.wc_en = t5_write_combine;
2297 	}
2298 
2299 	return (0);
2300 }
2301 
2302 struct memwin_init {
2303 	uint32_t base;
2304 	uint32_t aperture;
2305 };
2306 
2307 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2308 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2309 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2310 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2311 };
2312 
2313 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2314 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2315 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2316 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2317 };
2318 
2319 static void
2320 setup_memwin(struct adapter *sc)
2321 {
2322 	const struct memwin_init *mw_init;
2323 	struct memwin *mw;
2324 	int i;
2325 	uint32_t bar0;
2326 
2327 	if (is_t4(sc)) {
2328 		/*
2329 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2330 		 * mechanism.  Works from within PCI passthrough environments
2331 		 * too, where rman_get_start() can return a different value.  We
2332 		 * need to program the T4 memory window decoders with the actual
2333 		 * addresses that will be coming across the PCIe link.
2334 		 */
2335 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2336 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2337 
2338 		mw_init = &t4_memwin[0];
2339 	} else {
2340 		/* T5+ use the relative offset inside the PCIe BAR */
2341 		bar0 = 0;
2342 
2343 		mw_init = &t5_memwin[0];
2344 	}
2345 
2346 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2347 		rw_init(&mw->mw_lock, "memory window access");
2348 		mw->mw_base = mw_init->base;
2349 		mw->mw_aperture = mw_init->aperture;
2350 		mw->mw_curpos = 0;
2351 		t4_write_reg(sc,
2352 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2353 		    (mw->mw_base + bar0) | V_BIR(0) |
2354 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2355 		rw_wlock(&mw->mw_lock);
2356 		position_memwin(sc, i, 0);
2357 		rw_wunlock(&mw->mw_lock);
2358 	}
2359 
2360 	/* flush */
2361 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2362 }
2363 
2364 /*
2365  * Positions the memory window at the given address in the card's address space.
2366  * There are some alignment requirements and the actual position may be at an
2367  * address prior to the requested address.  mw->mw_curpos always has the actual
2368  * position of the window.
2369  */
2370 static void
2371 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2372 {
2373 	struct memwin *mw;
2374 	uint32_t pf;
2375 	uint32_t reg;
2376 
2377 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2378 	mw = &sc->memwin[idx];
2379 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2380 
2381 	if (is_t4(sc)) {
2382 		pf = 0;
2383 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2384 	} else {
2385 		pf = V_PFNUM(sc->pf);
2386 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2387 	}
2388 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2389 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2390 	t4_read_reg(sc, reg);	/* flush */
2391 }
2392 
2393 int
2394 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2395     int len, int rw)
2396 {
2397 	struct memwin *mw;
2398 	uint32_t mw_end, v;
2399 
2400 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2401 
2402 	/* Memory can only be accessed in naturally aligned 4 byte units */
2403 	if (addr & 3 || len & 3 || len <= 0)
2404 		return (EINVAL);
2405 
2406 	mw = &sc->memwin[idx];
2407 	while (len > 0) {
2408 		rw_rlock(&mw->mw_lock);
2409 		mw_end = mw->mw_curpos + mw->mw_aperture;
2410 		if (addr >= mw_end || addr < mw->mw_curpos) {
2411 			/* Will need to reposition the window */
2412 			if (!rw_try_upgrade(&mw->mw_lock)) {
2413 				rw_runlock(&mw->mw_lock);
2414 				rw_wlock(&mw->mw_lock);
2415 			}
2416 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2417 			position_memwin(sc, idx, addr);
2418 			rw_downgrade(&mw->mw_lock);
2419 			mw_end = mw->mw_curpos + mw->mw_aperture;
2420 		}
2421 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2422 		while (addr < mw_end && len > 0) {
2423 			if (rw == 0) {
2424 				v = t4_read_reg(sc, mw->mw_base + addr -
2425 				    mw->mw_curpos);
2426 				*val++ = le32toh(v);
2427 			} else {
2428 				v = *val++;
2429 				t4_write_reg(sc, mw->mw_base + addr -
2430 				    mw->mw_curpos, htole32(v));
2431 			}
2432 			addr += 4;
2433 			len -= 4;
2434 		}
2435 		rw_runlock(&mw->mw_lock);
2436 	}
2437 
2438 	return (0);
2439 }
2440 
2441 int
2442 alloc_atid_tab(struct tid_info *t, int flags)
2443 {
2444 	int i;
2445 
2446 	MPASS(t->natids > 0);
2447 	MPASS(t->atid_tab == NULL);
2448 
2449 	t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
2450 	    M_ZERO | flags);
2451 	if (t->atid_tab == NULL)
2452 		return (ENOMEM);
2453 	mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
2454 	t->afree = t->atid_tab;
2455 	t->atids_in_use = 0;
2456 	for (i = 1; i < t->natids; i++)
2457 		t->atid_tab[i - 1].next = &t->atid_tab[i];
2458 	t->atid_tab[t->natids - 1].next = NULL;
2459 
2460 	return (0);
2461 }
2462 
2463 void
2464 free_atid_tab(struct tid_info *t)
2465 {
2466 
2467 	KASSERT(t->atids_in_use == 0,
2468 	    ("%s: %d atids still in use.", __func__, t->atids_in_use));
2469 
2470 	if (mtx_initialized(&t->atid_lock))
2471 		mtx_destroy(&t->atid_lock);
2472 	free(t->atid_tab, M_CXGBE);
2473 	t->atid_tab = NULL;
2474 }
2475 
2476 int
2477 alloc_atid(struct adapter *sc, void *ctx)
2478 {
2479 	struct tid_info *t = &sc->tids;
2480 	int atid = -1;
2481 
2482 	mtx_lock(&t->atid_lock);
2483 	if (t->afree) {
2484 		union aopen_entry *p = t->afree;
2485 
2486 		atid = p - t->atid_tab;
2487 		MPASS(atid <= M_TID_TID);
2488 		t->afree = p->next;
2489 		p->data = ctx;
2490 		t->atids_in_use++;
2491 	}
2492 	mtx_unlock(&t->atid_lock);
2493 	return (atid);
2494 }
2495 
2496 void *
2497 lookup_atid(struct adapter *sc, int atid)
2498 {
2499 	struct tid_info *t = &sc->tids;
2500 
2501 	return (t->atid_tab[atid].data);
2502 }
2503 
2504 void
2505 free_atid(struct adapter *sc, int atid)
2506 {
2507 	struct tid_info *t = &sc->tids;
2508 	union aopen_entry *p = &t->atid_tab[atid];
2509 
2510 	mtx_lock(&t->atid_lock);
2511 	p->next = t->afree;
2512 	t->afree = p;
2513 	t->atids_in_use--;
2514 	mtx_unlock(&t->atid_lock);
2515 }
2516 
2517 static void
2518 queue_tid_release(struct adapter *sc, int tid)
2519 {
2520 
2521 	CXGBE_UNIMPLEMENTED("deferred tid release");
2522 }
2523 
2524 void
2525 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
2526 {
2527 	struct wrqe *wr;
2528 	struct cpl_tid_release *req;
2529 
2530 	wr = alloc_wrqe(sizeof(*req), ctrlq);
2531 	if (wr == NULL) {
2532 		queue_tid_release(sc, tid);	/* defer */
2533 		return;
2534 	}
2535 	req = wrtod(wr);
2536 
2537 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
2538 
2539 	t4_wrq_tx(sc, wr);
2540 }
2541 
2542 static int
2543 t4_range_cmp(const void *a, const void *b)
2544 {
2545 	return ((const struct t4_range *)a)->start -
2546 	       ((const struct t4_range *)b)->start;
2547 }
2548 
2549 /*
2550  * Verify that the memory range specified by the addr/len pair is valid within
2551  * the card's address space.
2552  */
2553 static int
2554 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2555 {
2556 	struct t4_range mem_ranges[4], *r, *next;
2557 	uint32_t em, addr_len;
2558 	int i, n, remaining;
2559 
2560 	/* Memory can only be accessed in naturally aligned 4 byte units */
2561 	if (addr & 3 || len & 3 || len <= 0)
2562 		return (EINVAL);
2563 
2564 	/* Enabled memories */
2565 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2566 
2567 	r = &mem_ranges[0];
2568 	n = 0;
2569 	bzero(r, sizeof(mem_ranges));
2570 	if (em & F_EDRAM0_ENABLE) {
2571 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2572 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2573 		if (r->size > 0) {
2574 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2575 			if (addr >= r->start &&
2576 			    addr + len <= r->start + r->size)
2577 				return (0);
2578 			r++;
2579 			n++;
2580 		}
2581 	}
2582 	if (em & F_EDRAM1_ENABLE) {
2583 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2584 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2585 		if (r->size > 0) {
2586 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2587 			if (addr >= r->start &&
2588 			    addr + len <= r->start + r->size)
2589 				return (0);
2590 			r++;
2591 			n++;
2592 		}
2593 	}
2594 	if (em & F_EXT_MEM_ENABLE) {
2595 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2596 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2597 		if (r->size > 0) {
2598 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2599 			if (addr >= r->start &&
2600 			    addr + len <= r->start + r->size)
2601 				return (0);
2602 			r++;
2603 			n++;
2604 		}
2605 	}
2606 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2607 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2608 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2609 		if (r->size > 0) {
2610 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2611 			if (addr >= r->start &&
2612 			    addr + len <= r->start + r->size)
2613 				return (0);
2614 			r++;
2615 			n++;
2616 		}
2617 	}
2618 	MPASS(n <= nitems(mem_ranges));
2619 
2620 	if (n > 1) {
2621 		/* Sort and merge the ranges. */
2622 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2623 
2624 		/* Start from index 0 and examine the next n - 1 entries. */
2625 		r = &mem_ranges[0];
2626 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2627 
2628 			MPASS(r->size > 0);	/* r is a valid entry. */
2629 			next = r + 1;
2630 			MPASS(next->size > 0);	/* and so is the next one. */
2631 
2632 			while (r->start + r->size >= next->start) {
2633 				/* Merge the next one into the current entry. */
2634 				r->size = max(r->start + r->size,
2635 				    next->start + next->size) - r->start;
2636 				n--;	/* One fewer entry in total. */
2637 				if (--remaining == 0)
2638 					goto done;	/* short circuit */
2639 				next++;
2640 			}
2641 			if (next != r + 1) {
2642 				/*
2643 				 * Some entries were merged into r and next
2644 				 * points to the first valid entry that couldn't
2645 				 * be merged.
2646 				 */
2647 				MPASS(next->size > 0);	/* must be valid */
2648 				memcpy(r + 1, next, remaining * sizeof(*r));
2649 #ifdef INVARIANTS
2650 				/*
2651 				 * This so that the foo->size assertion in the
2652 				 * next iteration of the loop do the right
2653 				 * thing for entries that were pulled up and are
2654 				 * no longer valid.
2655 				 */
2656 				MPASS(n < nitems(mem_ranges));
2657 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2658 				    sizeof(struct t4_range));
2659 #endif
2660 			}
2661 		}
2662 done:
2663 		/* Done merging the ranges. */
2664 		MPASS(n > 0);
2665 		r = &mem_ranges[0];
2666 		for (i = 0; i < n; i++, r++) {
2667 			if (addr >= r->start &&
2668 			    addr + len <= r->start + r->size)
2669 				return (0);
2670 		}
2671 	}
2672 
2673 	return (EFAULT);
2674 }
2675 
2676 static int
2677 fwmtype_to_hwmtype(int mtype)
2678 {
2679 
2680 	switch (mtype) {
2681 	case FW_MEMTYPE_EDC0:
2682 		return (MEM_EDC0);
2683 	case FW_MEMTYPE_EDC1:
2684 		return (MEM_EDC1);
2685 	case FW_MEMTYPE_EXTMEM:
2686 		return (MEM_MC0);
2687 	case FW_MEMTYPE_EXTMEM1:
2688 		return (MEM_MC1);
2689 	default:
2690 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2691 	}
2692 }
2693 
2694 /*
2695  * Verify that the memory range specified by the memtype/offset/len pair is
2696  * valid and lies entirely within the memtype specified.  The global address of
2697  * the start of the range is returned in addr.
2698  */
2699 static int
2700 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2701     uint32_t *addr)
2702 {
2703 	uint32_t em, addr_len, maddr;
2704 
2705 	/* Memory can only be accessed in naturally aligned 4 byte units */
2706 	if (off & 3 || len & 3 || len == 0)
2707 		return (EINVAL);
2708 
2709 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2710 	switch (fwmtype_to_hwmtype(mtype)) {
2711 	case MEM_EDC0:
2712 		if (!(em & F_EDRAM0_ENABLE))
2713 			return (EINVAL);
2714 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2715 		maddr = G_EDRAM0_BASE(addr_len) << 20;
2716 		break;
2717 	case MEM_EDC1:
2718 		if (!(em & F_EDRAM1_ENABLE))
2719 			return (EINVAL);
2720 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2721 		maddr = G_EDRAM1_BASE(addr_len) << 20;
2722 		break;
2723 	case MEM_MC:
2724 		if (!(em & F_EXT_MEM_ENABLE))
2725 			return (EINVAL);
2726 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2727 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2728 		break;
2729 	case MEM_MC1:
2730 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2731 			return (EINVAL);
2732 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2733 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2734 		break;
2735 	default:
2736 		return (EINVAL);
2737 	}
2738 
2739 	*addr = maddr + off;	/* global address */
2740 	return (validate_mem_range(sc, *addr, len));
2741 }
2742 
2743 static int
2744 fixup_devlog_params(struct adapter *sc)
2745 {
2746 	struct devlog_params *dparams = &sc->params.devlog;
2747 	int rc;
2748 
2749 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2750 	    dparams->size, &dparams->addr);
2751 
2752 	return (rc);
2753 }
2754 
2755 static void
2756 update_nirq(struct intrs_and_queues *iaq, int nports)
2757 {
2758 	int extra = T4_EXTRA_INTR;
2759 
2760 	iaq->nirq = extra;
2761 	iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
2762 	iaq->nirq += nports * (iaq->num_vis - 1) *
2763 	    max(iaq->nrxq_vi, iaq->nnmrxq_vi);
2764 	iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
2765 }
2766 
2767 /*
2768  * Adjust requirements to fit the number of interrupts available.
2769  */
2770 static void
2771 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
2772     int navail)
2773 {
2774 	int old_nirq;
2775 	const int nports = sc->params.nports;
2776 
2777 	MPASS(nports > 0);
2778 	MPASS(navail > 0);
2779 
2780 	bzero(iaq, sizeof(*iaq));
2781 	iaq->intr_type = itype;
2782 	iaq->num_vis = t4_num_vis;
2783 	iaq->ntxq = t4_ntxq;
2784 	iaq->ntxq_vi = t4_ntxq_vi;
2785 	iaq->nrxq = t4_nrxq;
2786 	iaq->nrxq_vi = t4_nrxq_vi;
2787 #ifdef TCP_OFFLOAD
2788 	if (is_offload(sc)) {
2789 		iaq->nofldtxq = t4_nofldtxq;
2790 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
2791 		iaq->nofldrxq = t4_nofldrxq;
2792 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
2793 	}
2794 #endif
2795 #ifdef DEV_NETMAP
2796 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
2797 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
2798 #endif
2799 
2800 	update_nirq(iaq, nports);
2801 	if (iaq->nirq <= navail &&
2802 	    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2803 		/*
2804 		 * This is the normal case -- there are enough interrupts for
2805 		 * everything.
2806 		 */
2807 		goto done;
2808 	}
2809 
2810 	/*
2811 	 * If extra VIs have been configured try reducing their count and see if
2812 	 * that works.
2813 	 */
2814 	while (iaq->num_vis > 1) {
2815 		iaq->num_vis--;
2816 		update_nirq(iaq, nports);
2817 		if (iaq->nirq <= navail &&
2818 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2819 			device_printf(sc->dev, "virtual interfaces per port "
2820 			    "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
2821 			    "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
2822 			    "itype %d, navail %u, nirq %d.\n",
2823 			    iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
2824 			    iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
2825 			    itype, navail, iaq->nirq);
2826 			goto done;
2827 		}
2828 	}
2829 
2830 	/*
2831 	 * Extra VIs will not be created.  Log a message if they were requested.
2832 	 */
2833 	MPASS(iaq->num_vis == 1);
2834 	iaq->ntxq_vi = iaq->nrxq_vi = 0;
2835 	iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
2836 	iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
2837 	if (iaq->num_vis != t4_num_vis) {
2838 		device_printf(sc->dev, "extra virtual interfaces disabled.  "
2839 		    "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
2840 		    "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
2841 		    iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
2842 		    iaq->nnmrxq_vi, itype, navail, iaq->nirq);
2843 	}
2844 
2845 	/*
2846 	 * Keep reducing the number of NIC rx queues to the next lower power of
2847 	 * 2 (for even RSS distribution) and halving the TOE rx queues and see
2848 	 * if that works.
2849 	 */
2850 	do {
2851 		if (iaq->nrxq > 1) {
2852 			do {
2853 				iaq->nrxq--;
2854 			} while (!powerof2(iaq->nrxq));
2855 		}
2856 		if (iaq->nofldrxq > 1)
2857 			iaq->nofldrxq >>= 1;
2858 
2859 		old_nirq = iaq->nirq;
2860 		update_nirq(iaq, nports);
2861 		if (iaq->nirq <= navail &&
2862 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2863 			device_printf(sc->dev, "running with reduced number of "
2864 			    "rx queues because of shortage of interrupts.  "
2865 			    "nrxq=%u, nofldrxq=%u.  "
2866 			    "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
2867 			    iaq->nofldrxq, itype, navail, iaq->nirq);
2868 			goto done;
2869 		}
2870 	} while (old_nirq != iaq->nirq);
2871 
2872 	/* One interrupt for everything.  Ugh. */
2873 	device_printf(sc->dev, "running with minimal number of queues.  "
2874 	    "itype %d, navail %u.\n", itype, navail);
2875 	iaq->nirq = 1;
2876 	MPASS(iaq->nrxq == 1);
2877 	iaq->ntxq = 1;
2878 	if (iaq->nofldrxq > 1)
2879 		iaq->nofldtxq = 1;
2880 done:
2881 	MPASS(iaq->num_vis > 0);
2882 	if (iaq->num_vis > 1) {
2883 		MPASS(iaq->nrxq_vi > 0);
2884 		MPASS(iaq->ntxq_vi > 0);
2885 	}
2886 	MPASS(iaq->nirq > 0);
2887 	MPASS(iaq->nrxq > 0);
2888 	MPASS(iaq->ntxq > 0);
2889 	if (itype == INTR_MSI) {
2890 		MPASS(powerof2(iaq->nirq));
2891 	}
2892 }
2893 
2894 static int
2895 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
2896 {
2897 	int rc, itype, navail, nalloc;
2898 
2899 	for (itype = INTR_MSIX; itype; itype >>= 1) {
2900 
2901 		if ((itype & t4_intr_types) == 0)
2902 			continue;	/* not allowed */
2903 
2904 		if (itype == INTR_MSIX)
2905 			navail = pci_msix_count(sc->dev);
2906 		else if (itype == INTR_MSI)
2907 			navail = pci_msi_count(sc->dev);
2908 		else
2909 			navail = 1;
2910 restart:
2911 		if (navail == 0)
2912 			continue;
2913 
2914 		calculate_iaq(sc, iaq, itype, navail);
2915 		nalloc = iaq->nirq;
2916 		rc = 0;
2917 		if (itype == INTR_MSIX)
2918 			rc = pci_alloc_msix(sc->dev, &nalloc);
2919 		else if (itype == INTR_MSI)
2920 			rc = pci_alloc_msi(sc->dev, &nalloc);
2921 
2922 		if (rc == 0 && nalloc > 0) {
2923 			if (nalloc == iaq->nirq)
2924 				return (0);
2925 
2926 			/*
2927 			 * Didn't get the number requested.  Use whatever number
2928 			 * the kernel is willing to allocate.
2929 			 */
2930 			device_printf(sc->dev, "fewer vectors than requested, "
2931 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2932 			    itype, iaq->nirq, nalloc);
2933 			pci_release_msi(sc->dev);
2934 			navail = nalloc;
2935 			goto restart;
2936 		}
2937 
2938 		device_printf(sc->dev,
2939 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2940 		    itype, rc, iaq->nirq, nalloc);
2941 	}
2942 
2943 	device_printf(sc->dev,
2944 	    "failed to find a usable interrupt type.  "
2945 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2946 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2947 
2948 	return (ENXIO);
2949 }
2950 
2951 #define FW_VERSION(chip) ( \
2952     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2953     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2954     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2955     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2956 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2957 
2958 struct fw_info {
2959 	uint8_t chip;
2960 	char *kld_name;
2961 	char *fw_mod_name;
2962 	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
2963 } fw_info[] = {
2964 	{
2965 		.chip = CHELSIO_T4,
2966 		.kld_name = "t4fw_cfg",
2967 		.fw_mod_name = "t4fw",
2968 		.fw_hdr = {
2969 			.chip = FW_HDR_CHIP_T4,
2970 			.fw_ver = htobe32_const(FW_VERSION(T4)),
2971 			.intfver_nic = FW_INTFVER(T4, NIC),
2972 			.intfver_vnic = FW_INTFVER(T4, VNIC),
2973 			.intfver_ofld = FW_INTFVER(T4, OFLD),
2974 			.intfver_ri = FW_INTFVER(T4, RI),
2975 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2976 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
2977 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2978 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
2979 		},
2980 	}, {
2981 		.chip = CHELSIO_T5,
2982 		.kld_name = "t5fw_cfg",
2983 		.fw_mod_name = "t5fw",
2984 		.fw_hdr = {
2985 			.chip = FW_HDR_CHIP_T5,
2986 			.fw_ver = htobe32_const(FW_VERSION(T5)),
2987 			.intfver_nic = FW_INTFVER(T5, NIC),
2988 			.intfver_vnic = FW_INTFVER(T5, VNIC),
2989 			.intfver_ofld = FW_INTFVER(T5, OFLD),
2990 			.intfver_ri = FW_INTFVER(T5, RI),
2991 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
2992 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
2993 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
2994 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
2995 		},
2996 	}, {
2997 		.chip = CHELSIO_T6,
2998 		.kld_name = "t6fw_cfg",
2999 		.fw_mod_name = "t6fw",
3000 		.fw_hdr = {
3001 			.chip = FW_HDR_CHIP_T6,
3002 			.fw_ver = htobe32_const(FW_VERSION(T6)),
3003 			.intfver_nic = FW_INTFVER(T6, NIC),
3004 			.intfver_vnic = FW_INTFVER(T6, VNIC),
3005 			.intfver_ofld = FW_INTFVER(T6, OFLD),
3006 			.intfver_ri = FW_INTFVER(T6, RI),
3007 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3008 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
3009 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3010 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
3011 		},
3012 	}
3013 };
3014 
3015 static struct fw_info *
3016 find_fw_info(int chip)
3017 {
3018 	int i;
3019 
3020 	for (i = 0; i < nitems(fw_info); i++) {
3021 		if (fw_info[i].chip == chip)
3022 			return (&fw_info[i]);
3023 	}
3024 	return (NULL);
3025 }
3026 
3027 /*
3028  * Is the given firmware API compatible with the one the driver was compiled
3029  * with?
3030  */
3031 static int
3032 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
3033 {
3034 
3035 	/* short circuit if it's the exact same firmware version */
3036 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
3037 		return (1);
3038 
3039 	/*
3040 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
3041 	 * features that are supported in the driver.
3042 	 */
3043 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
3044 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
3045 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
3046 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
3047 		return (1);
3048 #undef SAME_INTF
3049 
3050 	return (0);
3051 }
3052 
3053 /*
3054  * The firmware in the KLD is usable, but should it be installed?  This routine
3055  * explains itself in detail if it indicates the KLD firmware should be
3056  * installed.
3057  */
3058 static int
3059 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
3060 {
3061 	const char *reason;
3062 
3063 	if (!card_fw_usable) {
3064 		reason = "incompatible or unusable";
3065 		goto install;
3066 	}
3067 
3068 	if (k > c) {
3069 		reason = "older than the version bundled with this driver";
3070 		goto install;
3071 	}
3072 
3073 	if (t4_fw_install == 2 && k != c) {
3074 		reason = "different than the version bundled with this driver";
3075 		goto install;
3076 	}
3077 
3078 	return (0);
3079 
3080 install:
3081 	if (t4_fw_install == 0) {
3082 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3083 		    "but the driver is prohibited from installing a different "
3084 		    "firmware on the card.\n",
3085 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3086 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3087 
3088 		return (0);
3089 	}
3090 
3091 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3092 	    "installing firmware %u.%u.%u.%u on card.\n",
3093 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3094 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3095 	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3096 	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3097 
3098 	return (1);
3099 }
3100 
3101 /*
3102  * Establish contact with the firmware and determine if we are the master driver
3103  * or not, and whether we are responsible for chip initialization.
3104  */
3105 static int
3106 prep_firmware(struct adapter *sc)
3107 {
3108 	const struct firmware *fw = NULL, *default_cfg;
3109 	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
3110 	enum dev_state state;
3111 	struct fw_info *fw_info;
3112 	struct fw_hdr *card_fw;		/* fw on the card */
3113 	const struct fw_hdr *kld_fw;	/* fw in the KLD */
3114 	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
3115 					   against */
3116 
3117 	/* This is the firmware whose headers the driver was compiled against */
3118 	fw_info = find_fw_info(chip_id(sc));
3119 	if (fw_info == NULL) {
3120 		device_printf(sc->dev,
3121 		    "unable to look up firmware information for chip %d.\n",
3122 		    chip_id(sc));
3123 		return (EINVAL);
3124 	}
3125 	drv_fw = &fw_info->fw_hdr;
3126 
3127 	/*
3128 	 * The firmware KLD contains many modules.  The KLD name is also the
3129 	 * name of the module that contains the default config file.
3130 	 */
3131 	default_cfg = firmware_get(fw_info->kld_name);
3132 
3133 	/* This is the firmware in the KLD */
3134 	fw = firmware_get(fw_info->fw_mod_name);
3135 	if (fw != NULL) {
3136 		kld_fw = (const void *)fw->data;
3137 		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3138 	} else {
3139 		kld_fw = NULL;
3140 		kld_fw_usable = 0;
3141 	}
3142 
3143 	/* Read the header of the firmware on the card */
3144 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3145 	rc = -t4_read_flash(sc, FLASH_FW_START,
3146 	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
3147 	if (rc == 0) {
3148 		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
3149 		if (card_fw->fw_ver == be32toh(0xffffffff)) {
3150 			uint32_t d = be32toh(kld_fw->fw_ver);
3151 
3152 			if (!kld_fw_usable) {
3153 				device_printf(sc->dev,
3154 				    "no firmware on the card and no usable "
3155 				    "firmware bundled with the driver.\n");
3156 				rc = EIO;
3157 				goto done;
3158 			} else if (t4_fw_install == 0) {
3159 				device_printf(sc->dev,
3160 				    "no firmware on the card and the driver "
3161 				    "is prohibited from installing new "
3162 				    "firmware.\n");
3163 				rc = EIO;
3164 				goto done;
3165 			}
3166 
3167 			device_printf(sc->dev, "no firmware on the card, "
3168 			    "installing firmware %d.%d.%d.%d\n",
3169 			    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3170 			    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3171 			rc = t4_fw_forceinstall(sc, fw->data, fw->datasize);
3172 			if (rc < 0) {
3173 				rc = -rc;
3174 				device_printf(sc->dev,
3175 				    "firmware install failed: %d.\n", rc);
3176 				goto done;
3177 			}
3178 			memcpy(card_fw, kld_fw, sizeof(*card_fw));
3179 			card_fw_usable = 1;
3180 			need_fw_reset = 0;
3181 		}
3182 	} else {
3183 		device_printf(sc->dev,
3184 		    "Unable to read card's firmware header: %d\n", rc);
3185 		card_fw_usable = 0;
3186 	}
3187 
3188 	/* Contact firmware. */
3189 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3190 	if (rc < 0 || state == DEV_STATE_ERR) {
3191 		rc = -rc;
3192 		device_printf(sc->dev,
3193 		    "failed to connect to the firmware: %d, %d.\n", rc, state);
3194 		goto done;
3195 	}
3196 	pf = rc;
3197 	if (pf == sc->mbox)
3198 		sc->flags |= MASTER_PF;
3199 	else if (state == DEV_STATE_UNINIT) {
3200 		/*
3201 		 * We didn't get to be the master so we definitely won't be
3202 		 * configuring the chip.  It's a bug if someone else hasn't
3203 		 * configured it already.
3204 		 */
3205 		device_printf(sc->dev, "couldn't be master(%d), "
3206 		    "device not already initialized either(%d).\n", rc, state);
3207 		rc = EPROTO;
3208 		goto done;
3209 	}
3210 
3211 	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3212 	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3213 		/*
3214 		 * Common case: the firmware on the card is an exact match and
3215 		 * the KLD is an exact match too, or the KLD is
3216 		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
3217 		 * here -- use cxgbetool loadfw if you want to reinstall the
3218 		 * same firmware as the one on the card.
3219 		 */
3220 	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3221 	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3222 	    be32toh(card_fw->fw_ver))) {
3223 
3224 		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3225 		if (rc != 0) {
3226 			device_printf(sc->dev,
3227 			    "failed to install firmware: %d\n", rc);
3228 			goto done;
3229 		}
3230 
3231 		/* Installed successfully, update the cached header too. */
3232 		memcpy(card_fw, kld_fw, sizeof(*card_fw));
3233 		card_fw_usable = 1;
3234 		need_fw_reset = 0;	/* already reset as part of load_fw */
3235 	}
3236 
3237 	if (!card_fw_usable) {
3238 		uint32_t d, c, k;
3239 
3240 		d = ntohl(drv_fw->fw_ver);
3241 		c = ntohl(card_fw->fw_ver);
3242 		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3243 
3244 		device_printf(sc->dev, "Cannot find a usable firmware: "
3245 		    "fw_install %d, chip state %d, "
3246 		    "driver compiled with %d.%d.%d.%d, "
3247 		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3248 		    t4_fw_install, state,
3249 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3250 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3251 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3252 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3253 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3254 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3255 		rc = EINVAL;
3256 		goto done;
3257 	}
3258 
3259 	/* Reset device */
3260 	if (need_fw_reset &&
3261 	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3262 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3263 		if (rc != ETIMEDOUT && rc != EIO)
3264 			t4_fw_bye(sc, sc->mbox);
3265 		goto done;
3266 	}
3267 	sc->flags |= FW_OK;
3268 
3269 	rc = get_params__pre_init(sc);
3270 	if (rc != 0)
3271 		goto done; /* error message displayed already */
3272 
3273 	/* Partition adapter resources as specified in the config file. */
3274 	if (state == DEV_STATE_UNINIT) {
3275 
3276 		KASSERT(sc->flags & MASTER_PF,
3277 		    ("%s: trying to change chip settings when not master.",
3278 		    __func__));
3279 
3280 		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3281 		if (rc != 0)
3282 			goto done;	/* error message displayed already */
3283 
3284 		t4_tweak_chip_settings(sc);
3285 
3286 		/* get basic stuff going */
3287 		rc = -t4_fw_initialize(sc, sc->mbox);
3288 		if (rc != 0) {
3289 			device_printf(sc->dev, "fw init failed: %d.\n", rc);
3290 			goto done;
3291 		}
3292 	} else {
3293 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3294 		sc->cfcsum = 0;
3295 	}
3296 
3297 done:
3298 	free(card_fw, M_CXGBE);
3299 	if (fw != NULL)
3300 		firmware_put(fw, FIRMWARE_UNLOAD);
3301 	if (default_cfg != NULL)
3302 		firmware_put(default_cfg, FIRMWARE_UNLOAD);
3303 
3304 	return (rc);
3305 }
3306 
3307 #define FW_PARAM_DEV(param) \
3308 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3309 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3310 #define FW_PARAM_PFVF(param) \
3311 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3312 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3313 
3314 /*
3315  * Partition chip resources for use between various PFs, VFs, etc.
3316  */
3317 static int
3318 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3319     const char *name_prefix)
3320 {
3321 	const struct firmware *cfg = NULL;
3322 	int rc = 0;
3323 	struct fw_caps_config_cmd caps;
3324 	uint32_t mtype, moff, finicsum, cfcsum;
3325 
3326 	/*
3327 	 * Figure out what configuration file to use.  Pick the default config
3328 	 * file for the card if the user hasn't specified one explicitly.
3329 	 */
3330 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3331 	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3332 		/* Card specific overrides go here. */
3333 		if (pci_get_device(sc->dev) == 0x440a)
3334 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3335 		if (is_fpga(sc))
3336 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3337 	}
3338 
3339 	/*
3340 	 * We need to load another module if the profile is anything except
3341 	 * "default" or "flash".
3342 	 */
3343 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3344 	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3345 		char s[32];
3346 
3347 		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3348 		cfg = firmware_get(s);
3349 		if (cfg == NULL) {
3350 			if (default_cfg != NULL) {
3351 				device_printf(sc->dev,
3352 				    "unable to load module \"%s\" for "
3353 				    "configuration profile \"%s\", will use "
3354 				    "the default config file instead.\n",
3355 				    s, sc->cfg_file);
3356 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3357 				    "%s", DEFAULT_CF);
3358 			} else {
3359 				device_printf(sc->dev,
3360 				    "unable to load module \"%s\" for "
3361 				    "configuration profile \"%s\", will use "
3362 				    "the config file on the card's flash "
3363 				    "instead.\n", s, sc->cfg_file);
3364 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3365 				    "%s", FLASH_CF);
3366 			}
3367 		}
3368 	}
3369 
3370 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3371 	    default_cfg == NULL) {
3372 		device_printf(sc->dev,
3373 		    "default config file not available, will use the config "
3374 		    "file on the card's flash instead.\n");
3375 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3376 	}
3377 
3378 	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3379 		u_int cflen;
3380 		const uint32_t *cfdata;
3381 		uint32_t param, val, addr;
3382 
3383 		KASSERT(cfg != NULL || default_cfg != NULL,
3384 		    ("%s: no config to upload", __func__));
3385 
3386 		/*
3387 		 * Ask the firmware where it wants us to upload the config file.
3388 		 */
3389 		param = FW_PARAM_DEV(CF);
3390 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3391 		if (rc != 0) {
3392 			/* No support for config file?  Shouldn't happen. */
3393 			device_printf(sc->dev,
3394 			    "failed to query config file location: %d.\n", rc);
3395 			goto done;
3396 		}
3397 		mtype = G_FW_PARAMS_PARAM_Y(val);
3398 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3399 
3400 		/*
3401 		 * XXX: sheer laziness.  We deliberately added 4 bytes of
3402 		 * useless stuffing/comments at the end of the config file so
3403 		 * it's ok to simply throw away the last remaining bytes when
3404 		 * the config file is not an exact multiple of 4.  This also
3405 		 * helps with the validate_mt_off_len check.
3406 		 */
3407 		if (cfg != NULL) {
3408 			cflen = cfg->datasize & ~3;
3409 			cfdata = cfg->data;
3410 		} else {
3411 			cflen = default_cfg->datasize & ~3;
3412 			cfdata = default_cfg->data;
3413 		}
3414 
3415 		if (cflen > FLASH_CFG_MAX_SIZE) {
3416 			device_printf(sc->dev,
3417 			    "config file too long (%d, max allowed is %d).  "
3418 			    "Will try to use the config on the card, if any.\n",
3419 			    cflen, FLASH_CFG_MAX_SIZE);
3420 			goto use_config_on_flash;
3421 		}
3422 
3423 		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3424 		if (rc != 0) {
3425 			device_printf(sc->dev,
3426 			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3427 			    "Will try to use the config on the card, if any.\n",
3428 			    __func__, mtype, moff, cflen, rc);
3429 			goto use_config_on_flash;
3430 		}
3431 		write_via_memwin(sc, 2, addr, cfdata, cflen);
3432 	} else {
3433 use_config_on_flash:
3434 		mtype = FW_MEMTYPE_FLASH;
3435 		moff = t4_flash_cfg_addr(sc);
3436 	}
3437 
3438 	bzero(&caps, sizeof(caps));
3439 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3440 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3441 	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3442 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3443 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3444 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3445 	if (rc != 0) {
3446 		device_printf(sc->dev,
3447 		    "failed to pre-process config file: %d "
3448 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3449 		goto done;
3450 	}
3451 
3452 	finicsum = be32toh(caps.finicsum);
3453 	cfcsum = be32toh(caps.cfcsum);
3454 	if (finicsum != cfcsum) {
3455 		device_printf(sc->dev,
3456 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3457 		    finicsum, cfcsum);
3458 	}
3459 	sc->cfcsum = cfcsum;
3460 
3461 #define LIMIT_CAPS(x) do { \
3462 	caps.x &= htobe16(t4_##x##_allowed); \
3463 } while (0)
3464 
3465 	/*
3466 	 * Let the firmware know what features will (not) be used so it can tune
3467 	 * things accordingly.
3468 	 */
3469 	LIMIT_CAPS(nbmcaps);
3470 	LIMIT_CAPS(linkcaps);
3471 	LIMIT_CAPS(switchcaps);
3472 	LIMIT_CAPS(niccaps);
3473 	LIMIT_CAPS(toecaps);
3474 	LIMIT_CAPS(rdmacaps);
3475 	LIMIT_CAPS(cryptocaps);
3476 	LIMIT_CAPS(iscsicaps);
3477 	LIMIT_CAPS(fcoecaps);
3478 #undef LIMIT_CAPS
3479 
3480 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3481 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3482 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3483 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3484 	if (rc != 0) {
3485 		device_printf(sc->dev,
3486 		    "failed to process config file: %d.\n", rc);
3487 	}
3488 done:
3489 	if (cfg != NULL)
3490 		firmware_put(cfg, FIRMWARE_UNLOAD);
3491 	return (rc);
3492 }
3493 
3494 /*
3495  * Retrieve parameters that are needed (or nice to have) very early.
3496  */
3497 static int
3498 get_params__pre_init(struct adapter *sc)
3499 {
3500 	int rc;
3501 	uint32_t param[2], val[2];
3502 
3503 	t4_get_version_info(sc);
3504 
3505 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3506 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3507 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3508 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3509 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3510 
3511 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3512 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3513 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3514 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3515 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3516 
3517 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3518 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3519 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3520 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3521 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3522 
3523 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3524 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3525 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3526 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3527 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3528 
3529 	param[0] = FW_PARAM_DEV(PORTVEC);
3530 	param[1] = FW_PARAM_DEV(CCLK);
3531 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3532 	if (rc != 0) {
3533 		device_printf(sc->dev,
3534 		    "failed to query parameters (pre_init): %d.\n", rc);
3535 		return (rc);
3536 	}
3537 
3538 	sc->params.portvec = val[0];
3539 	sc->params.nports = bitcount32(val[0]);
3540 	sc->params.vpd.cclk = val[1];
3541 
3542 	/* Read device log parameters. */
3543 	rc = -t4_init_devlog_params(sc, 1);
3544 	if (rc == 0)
3545 		fixup_devlog_params(sc);
3546 	else {
3547 		device_printf(sc->dev,
3548 		    "failed to get devlog parameters: %d.\n", rc);
3549 		rc = 0;	/* devlog isn't critical for device operation */
3550 	}
3551 
3552 	return (rc);
3553 }
3554 
3555 /*
3556  * Retrieve various parameters that are of interest to the driver.  The device
3557  * has been initialized by the firmware at this point.
3558  */
3559 static int
3560 get_params__post_init(struct adapter *sc)
3561 {
3562 	int rc;
3563 	uint32_t param[7], val[7];
3564 	struct fw_caps_config_cmd caps;
3565 
3566 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3567 	param[1] = FW_PARAM_PFVF(EQ_START);
3568 	param[2] = FW_PARAM_PFVF(FILTER_START);
3569 	param[3] = FW_PARAM_PFVF(FILTER_END);
3570 	param[4] = FW_PARAM_PFVF(L2T_START);
3571 	param[5] = FW_PARAM_PFVF(L2T_END);
3572 	param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
3573 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
3574 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
3575 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
3576 	if (rc != 0) {
3577 		device_printf(sc->dev,
3578 		    "failed to query parameters (post_init): %d.\n", rc);
3579 		return (rc);
3580 	}
3581 
3582 	sc->sge.iq_start = val[0];
3583 	sc->sge.eq_start = val[1];
3584 	sc->tids.ftid_base = val[2];
3585 	sc->tids.nftids = val[3] - val[2] + 1;
3586 	sc->params.ftid_min = val[2];
3587 	sc->params.ftid_max = val[3];
3588 	sc->vres.l2t.start = val[4];
3589 	sc->vres.l2t.size = val[5] - val[4] + 1;
3590 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3591 	    ("%s: L2 table size (%u) larger than expected (%u)",
3592 	    __func__, sc->vres.l2t.size, L2T_SIZE));
3593 	sc->params.core_vdd = val[6];
3594 
3595 	/*
3596 	 * MPSBGMAP is queried separately because only recent firmwares support
3597 	 * it as a parameter and we don't want the compound query above to fail
3598 	 * on older firmwares.
3599 	 */
3600 	param[0] = FW_PARAM_DEV(MPSBGMAP);
3601 	val[0] = 0;
3602 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3603 	if (rc == 0)
3604 		sc->params.mps_bg_map = val[0];
3605 	else
3606 		sc->params.mps_bg_map = 0;
3607 
3608 	/* get capabilites */
3609 	bzero(&caps, sizeof(caps));
3610 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3611 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3612 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3613 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3614 	if (rc != 0) {
3615 		device_printf(sc->dev,
3616 		    "failed to get card capabilities: %d.\n", rc);
3617 		return (rc);
3618 	}
3619 
3620 #define READ_CAPS(x) do { \
3621 	sc->x = htobe16(caps.x); \
3622 } while (0)
3623 	READ_CAPS(nbmcaps);
3624 	READ_CAPS(linkcaps);
3625 	READ_CAPS(switchcaps);
3626 	READ_CAPS(niccaps);
3627 	READ_CAPS(toecaps);
3628 	READ_CAPS(rdmacaps);
3629 	READ_CAPS(cryptocaps);
3630 	READ_CAPS(iscsicaps);
3631 	READ_CAPS(fcoecaps);
3632 
3633 	/*
3634 	 * The firmware attempts memfree TOE configuration for -SO cards and
3635 	 * will report toecaps=0 if it runs out of resources (this depends on
3636 	 * the config file).  It may not report 0 for other capabilities
3637 	 * dependent on the TOE in this case.  Set them to 0 here so that the
3638 	 * driver doesn't bother tracking resources that will never be used.
3639 	 */
3640 	if (sc->toecaps == 0) {
3641 		sc->iscsicaps = 0;
3642 		sc->rdmacaps = 0;
3643 	}
3644 
3645 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3646 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3647 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3648 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3649 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3650 		if (rc != 0) {
3651 			device_printf(sc->dev,
3652 			    "failed to query NIC parameters: %d.\n", rc);
3653 			return (rc);
3654 		}
3655 		sc->tids.etid_base = val[0];
3656 		sc->params.etid_min = val[0];
3657 		sc->tids.netids = val[1] - val[0] + 1;
3658 		sc->params.netids = sc->tids.netids;
3659 		sc->params.eo_wr_cred = val[2];
3660 		sc->params.ethoffload = 1;
3661 	}
3662 
3663 	if (sc->toecaps) {
3664 		/* query offload-related parameters */
3665 		param[0] = FW_PARAM_DEV(NTID);
3666 		param[1] = FW_PARAM_PFVF(SERVER_START);
3667 		param[2] = FW_PARAM_PFVF(SERVER_END);
3668 		param[3] = FW_PARAM_PFVF(TDDP_START);
3669 		param[4] = FW_PARAM_PFVF(TDDP_END);
3670 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3671 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3672 		if (rc != 0) {
3673 			device_printf(sc->dev,
3674 			    "failed to query TOE parameters: %d.\n", rc);
3675 			return (rc);
3676 		}
3677 		sc->tids.ntids = val[0];
3678 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3679 		sc->tids.stid_base = val[1];
3680 		sc->tids.nstids = val[2] - val[1] + 1;
3681 		sc->vres.ddp.start = val[3];
3682 		sc->vres.ddp.size = val[4] - val[3] + 1;
3683 		sc->params.ofldq_wr_cred = val[5];
3684 		sc->params.offload = 1;
3685 	}
3686 	if (sc->rdmacaps) {
3687 		param[0] = FW_PARAM_PFVF(STAG_START);
3688 		param[1] = FW_PARAM_PFVF(STAG_END);
3689 		param[2] = FW_PARAM_PFVF(RQ_START);
3690 		param[3] = FW_PARAM_PFVF(RQ_END);
3691 		param[4] = FW_PARAM_PFVF(PBL_START);
3692 		param[5] = FW_PARAM_PFVF(PBL_END);
3693 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3694 		if (rc != 0) {
3695 			device_printf(sc->dev,
3696 			    "failed to query RDMA parameters(1): %d.\n", rc);
3697 			return (rc);
3698 		}
3699 		sc->vres.stag.start = val[0];
3700 		sc->vres.stag.size = val[1] - val[0] + 1;
3701 		sc->vres.rq.start = val[2];
3702 		sc->vres.rq.size = val[3] - val[2] + 1;
3703 		sc->vres.pbl.start = val[4];
3704 		sc->vres.pbl.size = val[5] - val[4] + 1;
3705 
3706 		param[0] = FW_PARAM_PFVF(SQRQ_START);
3707 		param[1] = FW_PARAM_PFVF(SQRQ_END);
3708 		param[2] = FW_PARAM_PFVF(CQ_START);
3709 		param[3] = FW_PARAM_PFVF(CQ_END);
3710 		param[4] = FW_PARAM_PFVF(OCQ_START);
3711 		param[5] = FW_PARAM_PFVF(OCQ_END);
3712 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3713 		if (rc != 0) {
3714 			device_printf(sc->dev,
3715 			    "failed to query RDMA parameters(2): %d.\n", rc);
3716 			return (rc);
3717 		}
3718 		sc->vres.qp.start = val[0];
3719 		sc->vres.qp.size = val[1] - val[0] + 1;
3720 		sc->vres.cq.start = val[2];
3721 		sc->vres.cq.size = val[3] - val[2] + 1;
3722 		sc->vres.ocq.start = val[4];
3723 		sc->vres.ocq.size = val[5] - val[4] + 1;
3724 
3725 		param[0] = FW_PARAM_PFVF(SRQ_START);
3726 		param[1] = FW_PARAM_PFVF(SRQ_END);
3727 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
3728 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
3729 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
3730 		if (rc != 0) {
3731 			device_printf(sc->dev,
3732 			    "failed to query RDMA parameters(3): %d.\n", rc);
3733 			return (rc);
3734 		}
3735 		sc->vres.srq.start = val[0];
3736 		sc->vres.srq.size = val[1] - val[0] + 1;
3737 		sc->params.max_ordird_qp = val[2];
3738 		sc->params.max_ird_adapter = val[3];
3739 	}
3740 	if (sc->iscsicaps) {
3741 		param[0] = FW_PARAM_PFVF(ISCSI_START);
3742 		param[1] = FW_PARAM_PFVF(ISCSI_END);
3743 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3744 		if (rc != 0) {
3745 			device_printf(sc->dev,
3746 			    "failed to query iSCSI parameters: %d.\n", rc);
3747 			return (rc);
3748 		}
3749 		sc->vres.iscsi.start = val[0];
3750 		sc->vres.iscsi.size = val[1] - val[0] + 1;
3751 	}
3752 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
3753 		param[0] = FW_PARAM_PFVF(TLS_START);
3754 		param[1] = FW_PARAM_PFVF(TLS_END);
3755 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3756 		if (rc != 0) {
3757 			device_printf(sc->dev,
3758 			    "failed to query TLS parameters: %d.\n", rc);
3759 			return (rc);
3760 		}
3761 		sc->vres.key.start = val[0];
3762 		sc->vres.key.size = val[1] - val[0] + 1;
3763 	}
3764 
3765 	t4_init_sge_params(sc);
3766 
3767 	/*
3768 	 * We've got the params we wanted to query via the firmware.  Now grab
3769 	 * some others directly from the chip.
3770 	 */
3771 	rc = t4_read_chip_settings(sc);
3772 
3773 	return (rc);
3774 }
3775 
3776 static int
3777 set_params__post_init(struct adapter *sc)
3778 {
3779 	uint32_t param, val;
3780 #ifdef TCP_OFFLOAD
3781 	int i, v, shift;
3782 #endif
3783 
3784 	/* ask for encapsulated CPLs */
3785 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3786 	val = 1;
3787 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3788 
3789 #ifdef TCP_OFFLOAD
3790 	/*
3791 	 * Override the TOE timers with user provided tunables.  This is not the
3792 	 * recommended way to change the timers (the firmware config file is) so
3793 	 * these tunables are not documented.
3794 	 *
3795 	 * All the timer tunables are in microseconds.
3796 	 */
3797 	if (t4_toe_keepalive_idle != 0) {
3798 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
3799 		v &= M_KEEPALIVEIDLE;
3800 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
3801 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
3802 	}
3803 	if (t4_toe_keepalive_interval != 0) {
3804 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
3805 		v &= M_KEEPALIVEINTVL;
3806 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
3807 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
3808 	}
3809 	if (t4_toe_keepalive_count != 0) {
3810 		v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
3811 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
3812 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
3813 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
3814 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
3815 	}
3816 	if (t4_toe_rexmt_min != 0) {
3817 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
3818 		v &= M_RXTMIN;
3819 		t4_set_reg_field(sc, A_TP_RXT_MIN,
3820 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
3821 	}
3822 	if (t4_toe_rexmt_max != 0) {
3823 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
3824 		v &= M_RXTMAX;
3825 		t4_set_reg_field(sc, A_TP_RXT_MAX,
3826 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
3827 	}
3828 	if (t4_toe_rexmt_count != 0) {
3829 		v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
3830 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
3831 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
3832 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
3833 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
3834 	}
3835 	for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
3836 		if (t4_toe_rexmt_backoff[i] != -1) {
3837 			v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
3838 			shift = (i & 3) << 3;
3839 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
3840 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
3841 		}
3842 	}
3843 #endif
3844 	return (0);
3845 }
3846 
3847 #undef FW_PARAM_PFVF
3848 #undef FW_PARAM_DEV
3849 
3850 static void
3851 t4_set_desc(struct adapter *sc)
3852 {
3853 	char buf[128];
3854 	struct adapter_params *p = &sc->params;
3855 
3856 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
3857 
3858 	device_set_desc_copy(sc->dev, buf);
3859 }
3860 
3861 static void
3862 build_medialist(struct port_info *pi, struct ifmedia *media)
3863 {
3864 	int m;
3865 
3866 	PORT_LOCK_ASSERT_OWNED(pi);
3867 
3868 	ifmedia_removeall(media);
3869 
3870 	/*
3871 	 * XXX: Would it be better to ifmedia_add all 4 combinations of pause
3872 	 * settings for every speed instead of just txpause|rxpause?  ifconfig
3873 	 * media display looks much better if autoselect is the only case where
3874 	 * ifm_current is different from ifm_active.  If the user picks anything
3875 	 * except txpause|rxpause the display is ugly.
3876 	 */
3877 	m = IFM_ETHER | IFM_FDX | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
3878 
3879 	switch(pi->port_type) {
3880 	case FW_PORT_TYPE_BT_XFI:
3881 	case FW_PORT_TYPE_BT_XAUI:
3882 		ifmedia_add(media, m | IFM_10G_T, 0, NULL);
3883 		/* fall through */
3884 
3885 	case FW_PORT_TYPE_BT_SGMII:
3886 		ifmedia_add(media, m | IFM_1000_T, 0, NULL);
3887 		ifmedia_add(media, m | IFM_100_TX, 0, NULL);
3888 		ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL);
3889 		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
3890 		break;
3891 
3892 	case FW_PORT_TYPE_CX4:
3893 		ifmedia_add(media, m | IFM_10G_CX4, 0, NULL);
3894 		ifmedia_set(media, m | IFM_10G_CX4);
3895 		break;
3896 
3897 	case FW_PORT_TYPE_QSFP_10G:
3898 	case FW_PORT_TYPE_SFP:
3899 	case FW_PORT_TYPE_FIBER_XFI:
3900 	case FW_PORT_TYPE_FIBER_XAUI:
3901 		switch (pi->mod_type) {
3902 
3903 		case FW_PORT_MOD_TYPE_LR:
3904 			ifmedia_add(media, m | IFM_10G_LR, 0, NULL);
3905 			ifmedia_set(media, m | IFM_10G_LR);
3906 			break;
3907 
3908 		case FW_PORT_MOD_TYPE_SR:
3909 			ifmedia_add(media, m | IFM_10G_SR, 0, NULL);
3910 			ifmedia_set(media, m | IFM_10G_SR);
3911 			break;
3912 
3913 		case FW_PORT_MOD_TYPE_LRM:
3914 			ifmedia_add(media, m | IFM_10G_LRM, 0, NULL);
3915 			ifmedia_set(media, m | IFM_10G_LRM);
3916 			break;
3917 
3918 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3919 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3920 			ifmedia_add(media, m | IFM_10G_TWINAX, 0, NULL);
3921 			ifmedia_set(media, m | IFM_10G_TWINAX);
3922 			break;
3923 
3924 		case FW_PORT_MOD_TYPE_NONE:
3925 			m &= ~IFM_FDX;
3926 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3927 			ifmedia_set(media, m | IFM_NONE);
3928 			break;
3929 
3930 		case FW_PORT_MOD_TYPE_NA:
3931 		case FW_PORT_MOD_TYPE_ER:
3932 		default:
3933 			device_printf(pi->dev,
3934 			    "unknown port_type (%d), mod_type (%d)\n",
3935 			    pi->port_type, pi->mod_type);
3936 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3937 			ifmedia_set(media, m | IFM_UNKNOWN);
3938 			break;
3939 		}
3940 		break;
3941 
3942 	case FW_PORT_TYPE_CR_QSFP:
3943 	case FW_PORT_TYPE_SFP28:
3944 	case FW_PORT_TYPE_KR_SFP28:
3945 		switch (pi->mod_type) {
3946 
3947 		case FW_PORT_MOD_TYPE_SR:
3948 			ifmedia_add(media, m | IFM_25G_SR, 0, NULL);
3949 			ifmedia_set(media, m | IFM_25G_SR);
3950 			break;
3951 
3952 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3953 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3954 			ifmedia_add(media, m | IFM_25G_CR, 0, NULL);
3955 			ifmedia_set(media, m | IFM_25G_CR);
3956 			break;
3957 
3958 		case FW_PORT_MOD_TYPE_NONE:
3959 			m &= ~IFM_FDX;
3960 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3961 			ifmedia_set(media, m | IFM_NONE);
3962 			break;
3963 
3964 		default:
3965 			device_printf(pi->dev,
3966 			    "unknown port_type (%d), mod_type (%d)\n",
3967 			    pi->port_type, pi->mod_type);
3968 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3969 			ifmedia_set(media, m | IFM_UNKNOWN);
3970 			break;
3971 		}
3972 		break;
3973 
3974 	case FW_PORT_TYPE_QSFP:
3975 		switch (pi->mod_type) {
3976 
3977 		case FW_PORT_MOD_TYPE_LR:
3978 			ifmedia_add(media, m | IFM_40G_LR4, 0, NULL);
3979 			ifmedia_set(media, m | IFM_40G_LR4);
3980 			break;
3981 
3982 		case FW_PORT_MOD_TYPE_SR:
3983 			ifmedia_add(media, m | IFM_40G_SR4, 0, NULL);
3984 			ifmedia_set(media, m | IFM_40G_SR4);
3985 			break;
3986 
3987 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3988 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3989 			ifmedia_add(media, m | IFM_40G_CR4, 0, NULL);
3990 			ifmedia_set(media, m | IFM_40G_CR4);
3991 			break;
3992 
3993 		case FW_PORT_MOD_TYPE_NONE:
3994 			m &= ~IFM_FDX;
3995 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3996 			ifmedia_set(media, m | IFM_NONE);
3997 			break;
3998 
3999 		default:
4000 			device_printf(pi->dev,
4001 			    "unknown port_type (%d), mod_type (%d)\n",
4002 			    pi->port_type, pi->mod_type);
4003 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
4004 			ifmedia_set(media, m | IFM_UNKNOWN);
4005 			break;
4006 		}
4007 		break;
4008 
4009 	case FW_PORT_TYPE_KR4_100G:
4010 	case FW_PORT_TYPE_CR4_QSFP:
4011 		switch (pi->mod_type) {
4012 
4013 		case FW_PORT_MOD_TYPE_LR:
4014 			ifmedia_add(media, m | IFM_100G_LR4, 0, NULL);
4015 			ifmedia_set(media, m | IFM_100G_LR4);
4016 			break;
4017 
4018 		case FW_PORT_MOD_TYPE_SR:
4019 			ifmedia_add(media, m | IFM_100G_SR4, 0, NULL);
4020 			ifmedia_set(media, m | IFM_100G_SR4);
4021 			break;
4022 
4023 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
4024 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
4025 			ifmedia_add(media, m | IFM_100G_CR4, 0, NULL);
4026 			ifmedia_set(media, m | IFM_100G_CR4);
4027 			break;
4028 
4029 		case FW_PORT_MOD_TYPE_NONE:
4030 			m &= ~IFM_FDX;
4031 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
4032 			ifmedia_set(media, m | IFM_NONE);
4033 			break;
4034 
4035 		default:
4036 			device_printf(pi->dev,
4037 			    "unknown port_type (%d), mod_type (%d)\n",
4038 			    pi->port_type, pi->mod_type);
4039 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
4040 			ifmedia_set(media, m | IFM_UNKNOWN);
4041 			break;
4042 		}
4043 		break;
4044 
4045 	default:
4046 		device_printf(pi->dev,
4047 		    "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
4048 		    pi->mod_type);
4049 		ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
4050 		ifmedia_set(media, m | IFM_UNKNOWN);
4051 		break;
4052 	}
4053 }
4054 
4055 /*
4056  * Update all the requested_* fields in the link config and then send a mailbox
4057  * command to apply the settings.
4058  */
4059 static void
4060 init_l1cfg(struct port_info *pi)
4061 {
4062 	struct adapter *sc = pi->adapter;
4063 	struct link_config *lc = &pi->link_cfg;
4064 	int rc;
4065 
4066 	ASSERT_SYNCHRONIZED_OP(sc);
4067 
4068 	lc->requested_speed = port_top_speed(pi);	/* in Gbps */
4069 	if (t4_autoneg != 0 && lc->supported & FW_PORT_CAP_ANEG) {
4070 		lc->requested_aneg = AUTONEG_ENABLE;
4071 	} else {
4072 		lc->requested_aneg = AUTONEG_DISABLE;
4073 	}
4074 
4075 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX);
4076 
4077 	if (t4_fec != -1) {
4078 		lc->requested_fec = t4_fec & (FEC_RS | FEC_BASER_RS |
4079 		    FEC_RESERVED);
4080 	} else {
4081 		/* Use the suggested value provided by the firmware in acaps */
4082 		if (lc->advertising & FW_PORT_CAP_FEC_RS)
4083 			lc->requested_fec = FEC_RS;
4084 		else if (lc->advertising & FW_PORT_CAP_FEC_BASER_RS)
4085 			lc->requested_fec = FEC_BASER_RS;
4086 		else
4087 			lc->requested_fec = 0;
4088 	}
4089 
4090 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
4091 	if (rc != 0) {
4092 		device_printf(pi->dev, "l1cfg failed: %d\n", rc);
4093 	} else {
4094 		lc->fc = lc->requested_fc;
4095 		lc->fec = lc->requested_fec;
4096 	}
4097 }
4098 
4099 #define FW_MAC_EXACT_CHUNK	7
4100 
4101 /*
4102  * Program the port's XGMAC based on parameters in ifnet.  The caller also
4103  * indicates which parameters should be programmed (the rest are left alone).
4104  */
4105 int
4106 update_mac_settings(struct ifnet *ifp, int flags)
4107 {
4108 	int rc = 0;
4109 	struct vi_info *vi = ifp->if_softc;
4110 	struct port_info *pi = vi->pi;
4111 	struct adapter *sc = pi->adapter;
4112 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4113 
4114 	ASSERT_SYNCHRONIZED_OP(sc);
4115 	KASSERT(flags, ("%s: not told what to update.", __func__));
4116 
4117 	if (flags & XGMAC_MTU)
4118 		mtu = ifp->if_mtu;
4119 
4120 	if (flags & XGMAC_PROMISC)
4121 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4122 
4123 	if (flags & XGMAC_ALLMULTI)
4124 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4125 
4126 	if (flags & XGMAC_VLANEX)
4127 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4128 
4129 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4130 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4131 		    allmulti, 1, vlanex, false);
4132 		if (rc) {
4133 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4134 			    rc);
4135 			return (rc);
4136 		}
4137 	}
4138 
4139 	if (flags & XGMAC_UCADDR) {
4140 		uint8_t ucaddr[ETHER_ADDR_LEN];
4141 
4142 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4143 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4144 		    ucaddr, true, true);
4145 		if (rc < 0) {
4146 			rc = -rc;
4147 			if_printf(ifp, "change_mac failed: %d\n", rc);
4148 			return (rc);
4149 		} else {
4150 			vi->xact_addr_filt = rc;
4151 			rc = 0;
4152 		}
4153 	}
4154 
4155 	if (flags & XGMAC_MCADDRS) {
4156 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4157 		int del = 1;
4158 		uint64_t hash = 0;
4159 		struct ifmultiaddr *ifma;
4160 		int i = 0, j;
4161 
4162 		if_maddr_rlock(ifp);
4163 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4164 			if (ifma->ifma_addr->sa_family != AF_LINK)
4165 				continue;
4166 			mcaddr[i] =
4167 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4168 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4169 			i++;
4170 
4171 			if (i == FW_MAC_EXACT_CHUNK) {
4172 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4173 				    del, i, mcaddr, NULL, &hash, 0);
4174 				if (rc < 0) {
4175 					rc = -rc;
4176 					for (j = 0; j < i; j++) {
4177 						if_printf(ifp,
4178 						    "failed to add mc address"
4179 						    " %02x:%02x:%02x:"
4180 						    "%02x:%02x:%02x rc=%d\n",
4181 						    mcaddr[j][0], mcaddr[j][1],
4182 						    mcaddr[j][2], mcaddr[j][3],
4183 						    mcaddr[j][4], mcaddr[j][5],
4184 						    rc);
4185 					}
4186 					goto mcfail;
4187 				}
4188 				del = 0;
4189 				i = 0;
4190 			}
4191 		}
4192 		if (i > 0) {
4193 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4194 			    mcaddr, NULL, &hash, 0);
4195 			if (rc < 0) {
4196 				rc = -rc;
4197 				for (j = 0; j < i; j++) {
4198 					if_printf(ifp,
4199 					    "failed to add mc address"
4200 					    " %02x:%02x:%02x:"
4201 					    "%02x:%02x:%02x rc=%d\n",
4202 					    mcaddr[j][0], mcaddr[j][1],
4203 					    mcaddr[j][2], mcaddr[j][3],
4204 					    mcaddr[j][4], mcaddr[j][5],
4205 					    rc);
4206 				}
4207 				goto mcfail;
4208 			}
4209 		}
4210 
4211 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4212 		if (rc != 0)
4213 			if_printf(ifp, "failed to set mc address hash: %d", rc);
4214 mcfail:
4215 		if_maddr_runlock(ifp);
4216 	}
4217 
4218 	return (rc);
4219 }
4220 
4221 /*
4222  * {begin|end}_synchronized_op must be called from the same thread.
4223  */
4224 int
4225 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4226     char *wmesg)
4227 {
4228 	int rc, pri;
4229 
4230 #ifdef WITNESS
4231 	/* the caller thinks it's ok to sleep, but is it really? */
4232 	if (flags & SLEEP_OK)
4233 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4234 		    "begin_synchronized_op");
4235 #endif
4236 
4237 	if (INTR_OK)
4238 		pri = PCATCH;
4239 	else
4240 		pri = 0;
4241 
4242 	ADAPTER_LOCK(sc);
4243 	for (;;) {
4244 
4245 		if (vi && IS_DOOMED(vi)) {
4246 			rc = ENXIO;
4247 			goto done;
4248 		}
4249 
4250 		if (!IS_BUSY(sc)) {
4251 			rc = 0;
4252 			break;
4253 		}
4254 
4255 		if (!(flags & SLEEP_OK)) {
4256 			rc = EBUSY;
4257 			goto done;
4258 		}
4259 
4260 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4261 			rc = EINTR;
4262 			goto done;
4263 		}
4264 	}
4265 
4266 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4267 	SET_BUSY(sc);
4268 #ifdef INVARIANTS
4269 	sc->last_op = wmesg;
4270 	sc->last_op_thr = curthread;
4271 	sc->last_op_flags = flags;
4272 #endif
4273 
4274 done:
4275 	if (!(flags & HOLD_LOCK) || rc)
4276 		ADAPTER_UNLOCK(sc);
4277 
4278 	return (rc);
4279 }
4280 
4281 /*
4282  * Tell if_ioctl and if_init that the VI is going away.  This is
4283  * special variant of begin_synchronized_op and must be paired with a
4284  * call to end_synchronized_op.
4285  */
4286 void
4287 doom_vi(struct adapter *sc, struct vi_info *vi)
4288 {
4289 
4290 	ADAPTER_LOCK(sc);
4291 	SET_DOOMED(vi);
4292 	wakeup(&sc->flags);
4293 	while (IS_BUSY(sc))
4294 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4295 	SET_BUSY(sc);
4296 #ifdef INVARIANTS
4297 	sc->last_op = "t4detach";
4298 	sc->last_op_thr = curthread;
4299 	sc->last_op_flags = 0;
4300 #endif
4301 	ADAPTER_UNLOCK(sc);
4302 }
4303 
4304 /*
4305  * {begin|end}_synchronized_op must be called from the same thread.
4306  */
4307 void
4308 end_synchronized_op(struct adapter *sc, int flags)
4309 {
4310 
4311 	if (flags & LOCK_HELD)
4312 		ADAPTER_LOCK_ASSERT_OWNED(sc);
4313 	else
4314 		ADAPTER_LOCK(sc);
4315 
4316 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
4317 	CLR_BUSY(sc);
4318 	wakeup(&sc->flags);
4319 	ADAPTER_UNLOCK(sc);
4320 }
4321 
4322 static int
4323 cxgbe_init_synchronized(struct vi_info *vi)
4324 {
4325 	struct port_info *pi = vi->pi;
4326 	struct adapter *sc = pi->adapter;
4327 	struct ifnet *ifp = vi->ifp;
4328 	int rc = 0, i;
4329 	struct sge_txq *txq;
4330 
4331 	ASSERT_SYNCHRONIZED_OP(sc);
4332 
4333 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
4334 		return (0);	/* already running */
4335 
4336 	if (!(sc->flags & FULL_INIT_DONE) &&
4337 	    ((rc = adapter_full_init(sc)) != 0))
4338 		return (rc);	/* error message displayed already */
4339 
4340 	if (!(vi->flags & VI_INIT_DONE) &&
4341 	    ((rc = vi_full_init(vi)) != 0))
4342 		return (rc); /* error message displayed already */
4343 
4344 	rc = update_mac_settings(ifp, XGMAC_ALL);
4345 	if (rc)
4346 		goto done;	/* error message displayed already */
4347 
4348 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
4349 	if (rc != 0) {
4350 		if_printf(ifp, "enable_vi failed: %d\n", rc);
4351 		goto done;
4352 	}
4353 
4354 	/*
4355 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
4356 	 * if this changes.
4357 	 */
4358 
4359 	for_each_txq(vi, i, txq) {
4360 		TXQ_LOCK(txq);
4361 		txq->eq.flags |= EQ_ENABLED;
4362 		TXQ_UNLOCK(txq);
4363 	}
4364 
4365 	/*
4366 	 * The first iq of the first port to come up is used for tracing.
4367 	 */
4368 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4369 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4370 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4371 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4372 		    V_QUEUENUMBER(sc->traceq));
4373 		pi->flags |= HAS_TRACEQ;
4374 	}
4375 
4376 	/* all ok */
4377 	PORT_LOCK(pi);
4378 	if (pi->up_vis++ == 0) {
4379 		t4_update_port_info(pi);
4380 		build_medialist(pi, &pi->media);
4381 		init_l1cfg(pi);
4382 	}
4383 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
4384 
4385 	if (pi->nvi > 1 || sc->flags & IS_VF)
4386 		callout_reset(&vi->tick, hz, vi_tick, vi);
4387 	else
4388 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4389 	PORT_UNLOCK(pi);
4390 done:
4391 	if (rc != 0)
4392 		cxgbe_uninit_synchronized(vi);
4393 
4394 	return (rc);
4395 }
4396 
4397 /*
4398  * Idempotent.
4399  */
4400 static int
4401 cxgbe_uninit_synchronized(struct vi_info *vi)
4402 {
4403 	struct port_info *pi = vi->pi;
4404 	struct adapter *sc = pi->adapter;
4405 	struct ifnet *ifp = vi->ifp;
4406 	int rc, i;
4407 	struct sge_txq *txq;
4408 
4409 	ASSERT_SYNCHRONIZED_OP(sc);
4410 
4411 	if (!(vi->flags & VI_INIT_DONE)) {
4412 		if (__predict_false(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4413 			KASSERT(0, ("uninited VI is running"));
4414 			if_printf(ifp, "uninited VI with running ifnet.  "
4415 			    "vi->flags 0x%016lx, if_flags 0x%08x, "
4416 			    "if_drv_flags 0x%08x\n", vi->flags, ifp->if_flags,
4417 			    ifp->if_drv_flags);
4418 		}
4419 		return (0);
4420 	}
4421 
4422 	/*
4423 	 * Disable the VI so that all its data in either direction is discarded
4424 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4425 	 * tick) intact as the TP can deliver negative advice or data that it's
4426 	 * holding in its RAM (for an offloaded connection) even after the VI is
4427 	 * disabled.
4428 	 */
4429 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4430 	if (rc) {
4431 		if_printf(ifp, "disable_vi failed: %d\n", rc);
4432 		return (rc);
4433 	}
4434 
4435 	for_each_txq(vi, i, txq) {
4436 		TXQ_LOCK(txq);
4437 		txq->eq.flags &= ~EQ_ENABLED;
4438 		TXQ_UNLOCK(txq);
4439 	}
4440 
4441 	PORT_LOCK(pi);
4442 	if (pi->nvi > 1 || sc->flags & IS_VF)
4443 		callout_stop(&vi->tick);
4444 	else
4445 		callout_stop(&pi->tick);
4446 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4447 		PORT_UNLOCK(pi);
4448 		return (0);
4449 	}
4450 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4451 	pi->up_vis--;
4452 	if (pi->up_vis > 0) {
4453 		PORT_UNLOCK(pi);
4454 		return (0);
4455 	}
4456 	PORT_UNLOCK(pi);
4457 
4458 	pi->link_cfg.link_ok = 0;
4459 	pi->link_cfg.speed = 0;
4460 	pi->link_cfg.link_down_rc = 255;
4461 	t4_os_link_changed(pi);
4462 	pi->old_link_cfg = pi->link_cfg;
4463 
4464 	return (0);
4465 }
4466 
4467 /*
4468  * It is ok for this function to fail midway and return right away.  t4_detach
4469  * will walk the entire sc->irq list and clean up whatever is valid.
4470  */
4471 int
4472 t4_setup_intr_handlers(struct adapter *sc)
4473 {
4474 	int rc, rid, p, q, v;
4475 	char s[8];
4476 	struct irq *irq;
4477 	struct port_info *pi;
4478 	struct vi_info *vi;
4479 	struct sge *sge = &sc->sge;
4480 	struct sge_rxq *rxq;
4481 #ifdef TCP_OFFLOAD
4482 	struct sge_ofld_rxq *ofld_rxq;
4483 #endif
4484 #ifdef DEV_NETMAP
4485 	struct sge_nm_rxq *nm_rxq;
4486 #endif
4487 #ifdef RSS
4488 	int nbuckets = rss_getnumbuckets();
4489 #endif
4490 
4491 	/*
4492 	 * Setup interrupts.
4493 	 */
4494 	irq = &sc->irq[0];
4495 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
4496 	if (forwarding_intr_to_fwq(sc))
4497 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4498 
4499 	/* Multiple interrupts. */
4500 	if (sc->flags & IS_VF)
4501 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4502 		    ("%s: too few intr.", __func__));
4503 	else
4504 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4505 		    ("%s: too few intr.", __func__));
4506 
4507 	/* The first one is always error intr on PFs */
4508 	if (!(sc->flags & IS_VF)) {
4509 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4510 		if (rc != 0)
4511 			return (rc);
4512 		irq++;
4513 		rid++;
4514 	}
4515 
4516 	/* The second one is always the firmware event queue (first on VFs) */
4517 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4518 	if (rc != 0)
4519 		return (rc);
4520 	irq++;
4521 	rid++;
4522 
4523 	for_each_port(sc, p) {
4524 		pi = sc->port[p];
4525 		for_each_vi(pi, v, vi) {
4526 			vi->first_intr = rid - 1;
4527 
4528 			if (vi->nnmrxq > 0) {
4529 				int n = max(vi->nrxq, vi->nnmrxq);
4530 
4531 				rxq = &sge->rxq[vi->first_rxq];
4532 #ifdef DEV_NETMAP
4533 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4534 #endif
4535 				for (q = 0; q < n; q++) {
4536 					snprintf(s, sizeof(s), "%x%c%x", p,
4537 					    'a' + v, q);
4538 					if (q < vi->nrxq)
4539 						irq->rxq = rxq++;
4540 #ifdef DEV_NETMAP
4541 					if (q < vi->nnmrxq)
4542 						irq->nm_rxq = nm_rxq++;
4543 #endif
4544 					rc = t4_alloc_irq(sc, irq, rid,
4545 					    t4_vi_intr, irq, s);
4546 					if (rc != 0)
4547 						return (rc);
4548 #ifdef RSS
4549 					if (q < vi->nrxq) {
4550 						bus_bind_intr(sc->dev, irq->res,
4551 						    rss_getcpu(q % nbuckets));
4552 					}
4553 #endif
4554 					irq++;
4555 					rid++;
4556 					vi->nintr++;
4557 				}
4558 			} else {
4559 				for_each_rxq(vi, q, rxq) {
4560 					snprintf(s, sizeof(s), "%x%c%x", p,
4561 					    'a' + v, q);
4562 					rc = t4_alloc_irq(sc, irq, rid,
4563 					    t4_intr, rxq, s);
4564 					if (rc != 0)
4565 						return (rc);
4566 #ifdef RSS
4567 					bus_bind_intr(sc->dev, irq->res,
4568 					    rss_getcpu(q % nbuckets));
4569 #endif
4570 					irq++;
4571 					rid++;
4572 					vi->nintr++;
4573 				}
4574 			}
4575 #ifdef TCP_OFFLOAD
4576 			for_each_ofld_rxq(vi, q, ofld_rxq) {
4577 				snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
4578 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
4579 				    ofld_rxq, s);
4580 				if (rc != 0)
4581 					return (rc);
4582 				irq++;
4583 				rid++;
4584 				vi->nintr++;
4585 			}
4586 #endif
4587 		}
4588 	}
4589 	MPASS(irq == &sc->irq[sc->intr_count]);
4590 
4591 	return (0);
4592 }
4593 
4594 int
4595 adapter_full_init(struct adapter *sc)
4596 {
4597 	int rc, i;
4598 #ifdef RSS
4599 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4600 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4601 #endif
4602 
4603 	ASSERT_SYNCHRONIZED_OP(sc);
4604 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4605 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
4606 	    ("%s: FULL_INIT_DONE already", __func__));
4607 
4608 	/*
4609 	 * queues that belong to the adapter (not any particular port).
4610 	 */
4611 	rc = t4_setup_adapter_queues(sc);
4612 	if (rc != 0)
4613 		goto done;
4614 
4615 	for (i = 0; i < nitems(sc->tq); i++) {
4616 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
4617 		    taskqueue_thread_enqueue, &sc->tq[i]);
4618 		if (sc->tq[i] == NULL) {
4619 			device_printf(sc->dev,
4620 			    "failed to allocate task queue %d\n", i);
4621 			rc = ENOMEM;
4622 			goto done;
4623 		}
4624 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
4625 		    device_get_nameunit(sc->dev), i);
4626 	}
4627 #ifdef RSS
4628 	MPASS(RSS_KEYSIZE == 40);
4629 	rss_getkey((void *)&raw_rss_key[0]);
4630 	for (i = 0; i < nitems(rss_key); i++) {
4631 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
4632 	}
4633 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
4634 #endif
4635 
4636 	if (!(sc->flags & IS_VF))
4637 		t4_intr_enable(sc);
4638 	sc->flags |= FULL_INIT_DONE;
4639 done:
4640 	if (rc != 0)
4641 		adapter_full_uninit(sc);
4642 
4643 	return (rc);
4644 }
4645 
4646 int
4647 adapter_full_uninit(struct adapter *sc)
4648 {
4649 	int i;
4650 
4651 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4652 
4653 	t4_teardown_adapter_queues(sc);
4654 
4655 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
4656 		taskqueue_free(sc->tq[i]);
4657 		sc->tq[i] = NULL;
4658 	}
4659 
4660 	sc->flags &= ~FULL_INIT_DONE;
4661 
4662 	return (0);
4663 }
4664 
4665 #ifdef RSS
4666 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
4667     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
4668     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
4669     RSS_HASHTYPE_RSS_UDP_IPV6)
4670 
4671 /* Translates kernel hash types to hardware. */
4672 static int
4673 hashconfig_to_hashen(int hashconfig)
4674 {
4675 	int hashen = 0;
4676 
4677 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
4678 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
4679 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
4680 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
4681 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
4682 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4683 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4684 	}
4685 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
4686 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4687 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4688 	}
4689 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
4690 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4691 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
4692 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4693 
4694 	return (hashen);
4695 }
4696 
4697 /* Translates hardware hash types to kernel. */
4698 static int
4699 hashen_to_hashconfig(int hashen)
4700 {
4701 	int hashconfig = 0;
4702 
4703 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
4704 		/*
4705 		 * If UDP hashing was enabled it must have been enabled for
4706 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
4707 		 * enabling any 4-tuple hash is nonsense configuration.
4708 		 */
4709 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4710 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
4711 
4712 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4713 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
4714 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4715 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
4716 	}
4717 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4718 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
4719 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4720 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
4721 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
4722 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
4723 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
4724 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
4725 
4726 	return (hashconfig);
4727 }
4728 #endif
4729 
4730 int
4731 vi_full_init(struct vi_info *vi)
4732 {
4733 	struct adapter *sc = vi->pi->adapter;
4734 	struct ifnet *ifp = vi->ifp;
4735 	uint16_t *rss;
4736 	struct sge_rxq *rxq;
4737 	int rc, i, j, hashen;
4738 #ifdef RSS
4739 	int nbuckets = rss_getnumbuckets();
4740 	int hashconfig = rss_gethashconfig();
4741 	int extra;
4742 #endif
4743 
4744 	ASSERT_SYNCHRONIZED_OP(sc);
4745 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
4746 	    ("%s: VI_INIT_DONE already", __func__));
4747 
4748 	sysctl_ctx_init(&vi->ctx);
4749 	vi->flags |= VI_SYSCTL_CTX;
4750 
4751 	/*
4752 	 * Allocate tx/rx/fl queues for this VI.
4753 	 */
4754 	rc = t4_setup_vi_queues(vi);
4755 	if (rc != 0)
4756 		goto done;	/* error message displayed already */
4757 
4758 	/*
4759 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
4760 	 */
4761 	if (vi->nrxq > vi->rss_size) {
4762 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
4763 		    "some queues will never receive traffic.\n", vi->nrxq,
4764 		    vi->rss_size);
4765 	} else if (vi->rss_size % vi->nrxq) {
4766 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
4767 		    "expect uneven traffic distribution.\n", vi->nrxq,
4768 		    vi->rss_size);
4769 	}
4770 #ifdef RSS
4771 	if (vi->nrxq != nbuckets) {
4772 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
4773 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
4774 	}
4775 #endif
4776 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
4777 	for (i = 0; i < vi->rss_size;) {
4778 #ifdef RSS
4779 		j = rss_get_indirection_to_bucket(i);
4780 		j %= vi->nrxq;
4781 		rxq = &sc->sge.rxq[vi->first_rxq + j];
4782 		rss[i++] = rxq->iq.abs_id;
4783 #else
4784 		for_each_rxq(vi, j, rxq) {
4785 			rss[i++] = rxq->iq.abs_id;
4786 			if (i == vi->rss_size)
4787 				break;
4788 		}
4789 #endif
4790 	}
4791 
4792 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
4793 	    vi->rss_size);
4794 	if (rc != 0) {
4795 		if_printf(ifp, "rss_config failed: %d\n", rc);
4796 		goto done;
4797 	}
4798 
4799 #ifdef RSS
4800 	hashen = hashconfig_to_hashen(hashconfig);
4801 
4802 	/*
4803 	 * We may have had to enable some hashes even though the global config
4804 	 * wants them disabled.  This is a potential problem that must be
4805 	 * reported to the user.
4806 	 */
4807 	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
4808 
4809 	/*
4810 	 * If we consider only the supported hash types, then the enabled hashes
4811 	 * are a superset of the requested hashes.  In other words, there cannot
4812 	 * be any supported hash that was requested but not enabled, but there
4813 	 * can be hashes that were not requested but had to be enabled.
4814 	 */
4815 	extra &= SUPPORTED_RSS_HASHTYPES;
4816 	MPASS((extra & hashconfig) == 0);
4817 
4818 	if (extra) {
4819 		if_printf(ifp,
4820 		    "global RSS config (0x%x) cannot be accommodated.\n",
4821 		    hashconfig);
4822 	}
4823 	if (extra & RSS_HASHTYPE_RSS_IPV4)
4824 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
4825 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
4826 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
4827 	if (extra & RSS_HASHTYPE_RSS_IPV6)
4828 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
4829 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
4830 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
4831 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
4832 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
4833 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
4834 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
4835 #else
4836 	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
4837 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
4838 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4839 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
4840 #endif
4841 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
4842 	if (rc != 0) {
4843 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
4844 		goto done;
4845 	}
4846 
4847 	vi->rss = rss;
4848 	vi->flags |= VI_INIT_DONE;
4849 done:
4850 	if (rc != 0)
4851 		vi_full_uninit(vi);
4852 
4853 	return (rc);
4854 }
4855 
4856 /*
4857  * Idempotent.
4858  */
4859 int
4860 vi_full_uninit(struct vi_info *vi)
4861 {
4862 	struct port_info *pi = vi->pi;
4863 	struct adapter *sc = pi->adapter;
4864 	int i;
4865 	struct sge_rxq *rxq;
4866 	struct sge_txq *txq;
4867 #ifdef TCP_OFFLOAD
4868 	struct sge_ofld_rxq *ofld_rxq;
4869 	struct sge_wrq *ofld_txq;
4870 #endif
4871 
4872 	if (vi->flags & VI_INIT_DONE) {
4873 
4874 		/* Need to quiesce queues.  */
4875 
4876 		/* XXX: Only for the first VI? */
4877 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
4878 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
4879 
4880 		for_each_txq(vi, i, txq) {
4881 			quiesce_txq(sc, txq);
4882 		}
4883 
4884 #ifdef TCP_OFFLOAD
4885 		for_each_ofld_txq(vi, i, ofld_txq) {
4886 			quiesce_wrq(sc, ofld_txq);
4887 		}
4888 #endif
4889 
4890 		for_each_rxq(vi, i, rxq) {
4891 			quiesce_iq(sc, &rxq->iq);
4892 			quiesce_fl(sc, &rxq->fl);
4893 		}
4894 
4895 #ifdef TCP_OFFLOAD
4896 		for_each_ofld_rxq(vi, i, ofld_rxq) {
4897 			quiesce_iq(sc, &ofld_rxq->iq);
4898 			quiesce_fl(sc, &ofld_rxq->fl);
4899 		}
4900 #endif
4901 		free(vi->rss, M_CXGBE);
4902 		free(vi->nm_rss, M_CXGBE);
4903 	}
4904 
4905 	t4_teardown_vi_queues(vi);
4906 	vi->flags &= ~VI_INIT_DONE;
4907 
4908 	return (0);
4909 }
4910 
4911 static void
4912 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
4913 {
4914 	struct sge_eq *eq = &txq->eq;
4915 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
4916 
4917 	(void) sc;	/* unused */
4918 
4919 #ifdef INVARIANTS
4920 	TXQ_LOCK(txq);
4921 	MPASS((eq->flags & EQ_ENABLED) == 0);
4922 	TXQ_UNLOCK(txq);
4923 #endif
4924 
4925 	/* Wait for the mp_ring to empty. */
4926 	while (!mp_ring_is_idle(txq->r)) {
4927 		mp_ring_check_drainage(txq->r, 0);
4928 		pause("rquiesce", 1);
4929 	}
4930 
4931 	/* Then wait for the hardware to finish. */
4932 	while (spg->cidx != htobe16(eq->pidx))
4933 		pause("equiesce", 1);
4934 
4935 	/* Finally, wait for the driver to reclaim all descriptors. */
4936 	while (eq->cidx != eq->pidx)
4937 		pause("dquiesce", 1);
4938 }
4939 
4940 static void
4941 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
4942 {
4943 
4944 	/* XXXTX */
4945 }
4946 
4947 static void
4948 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
4949 {
4950 	(void) sc;	/* unused */
4951 
4952 	/* Synchronize with the interrupt handler */
4953 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
4954 		pause("iqfree", 1);
4955 }
4956 
4957 static void
4958 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
4959 {
4960 	mtx_lock(&sc->sfl_lock);
4961 	FL_LOCK(fl);
4962 	fl->flags |= FL_DOOMED;
4963 	FL_UNLOCK(fl);
4964 	callout_stop(&sc->sfl_callout);
4965 	mtx_unlock(&sc->sfl_lock);
4966 
4967 	KASSERT((fl->flags & FL_STARVING) == 0,
4968 	    ("%s: still starving", __func__));
4969 }
4970 
4971 static int
4972 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
4973     driver_intr_t *handler, void *arg, char *name)
4974 {
4975 	int rc;
4976 
4977 	irq->rid = rid;
4978 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
4979 	    RF_SHAREABLE | RF_ACTIVE);
4980 	if (irq->res == NULL) {
4981 		device_printf(sc->dev,
4982 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
4983 		return (ENOMEM);
4984 	}
4985 
4986 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
4987 	    NULL, handler, arg, &irq->tag);
4988 	if (rc != 0) {
4989 		device_printf(sc->dev,
4990 		    "failed to setup interrupt for rid %d, name %s: %d\n",
4991 		    rid, name, rc);
4992 	} else if (name)
4993 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
4994 
4995 	return (rc);
4996 }
4997 
4998 static int
4999 t4_free_irq(struct adapter *sc, struct irq *irq)
5000 {
5001 	if (irq->tag)
5002 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
5003 	if (irq->res)
5004 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
5005 
5006 	bzero(irq, sizeof(*irq));
5007 
5008 	return (0);
5009 }
5010 
5011 static void
5012 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
5013 {
5014 
5015 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
5016 	t4_get_regs(sc, buf, regs->len);
5017 }
5018 
5019 #define	A_PL_INDIR_CMD	0x1f8
5020 
5021 #define	S_PL_AUTOINC	31
5022 #define	M_PL_AUTOINC	0x1U
5023 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
5024 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
5025 
5026 #define	S_PL_VFID	20
5027 #define	M_PL_VFID	0xffU
5028 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
5029 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
5030 
5031 #define	S_PL_ADDR	0
5032 #define	M_PL_ADDR	0xfffffU
5033 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
5034 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
5035 
5036 #define	A_PL_INDIR_DATA	0x1fc
5037 
5038 static uint64_t
5039 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
5040 {
5041 	u32 stats[2];
5042 
5043 	mtx_assert(&sc->reg_lock, MA_OWNED);
5044 	if (sc->flags & IS_VF) {
5045 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
5046 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
5047 	} else {
5048 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5049 		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5050 		    V_PL_ADDR(VF_MPS_REG(reg)));
5051 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
5052 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
5053 	}
5054 	return (((uint64_t)stats[1]) << 32 | stats[0]);
5055 }
5056 
5057 static void
5058 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
5059     struct fw_vi_stats_vf *stats)
5060 {
5061 
5062 #define GET_STAT(name) \
5063 	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
5064 
5065 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
5066 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
5067 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
5068 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
5069 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
5070 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
5071 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
5072 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
5073 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
5074 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
5075 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
5076 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
5077 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
5078 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
5079 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
5080 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
5081 
5082 #undef GET_STAT
5083 }
5084 
5085 static void
5086 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
5087 {
5088 	int reg;
5089 
5090 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5091 	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5092 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
5093 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
5094 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
5095 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
5096 }
5097 
5098 static void
5099 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
5100 {
5101 	struct timeval tv;
5102 	const struct timeval interval = {0, 250000};	/* 250ms */
5103 
5104 	if (!(vi->flags & VI_INIT_DONE))
5105 		return;
5106 
5107 	getmicrotime(&tv);
5108 	timevalsub(&tv, &interval);
5109 	if (timevalcmp(&tv, &vi->last_refreshed, <))
5110 		return;
5111 
5112 	mtx_lock(&sc->reg_lock);
5113 	t4_get_vi_stats(sc, vi->viid, &vi->stats);
5114 	getmicrotime(&vi->last_refreshed);
5115 	mtx_unlock(&sc->reg_lock);
5116 }
5117 
5118 static void
5119 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5120 {
5121 	u_int i, v, tnl_cong_drops, bg_map;
5122 	struct timeval tv;
5123 	const struct timeval interval = {0, 250000};	/* 250ms */
5124 
5125 	getmicrotime(&tv);
5126 	timevalsub(&tv, &interval);
5127 	if (timevalcmp(&tv, &pi->last_refreshed, <))
5128 		return;
5129 
5130 	tnl_cong_drops = 0;
5131 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5132 	bg_map = pi->mps_bg_map;
5133 	while (bg_map) {
5134 		i = ffs(bg_map) - 1;
5135 		mtx_lock(&sc->reg_lock);
5136 		t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5137 		    A_TP_MIB_TNL_CNG_DROP_0 + i);
5138 		mtx_unlock(&sc->reg_lock);
5139 		tnl_cong_drops += v;
5140 		bg_map &= ~(1 << i);
5141 	}
5142 	pi->tnl_cong_drops = tnl_cong_drops;
5143 	getmicrotime(&pi->last_refreshed);
5144 }
5145 
5146 static void
5147 cxgbe_tick(void *arg)
5148 {
5149 	struct port_info *pi = arg;
5150 	struct adapter *sc = pi->adapter;
5151 
5152 	PORT_LOCK_ASSERT_OWNED(pi);
5153 	cxgbe_refresh_stats(sc, pi);
5154 
5155 	callout_schedule(&pi->tick, hz);
5156 }
5157 
5158 void
5159 vi_tick(void *arg)
5160 {
5161 	struct vi_info *vi = arg;
5162 	struct adapter *sc = vi->pi->adapter;
5163 
5164 	vi_refresh_stats(sc, vi);
5165 
5166 	callout_schedule(&vi->tick, hz);
5167 }
5168 
5169 static void
5170 cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
5171 {
5172 	struct ifnet *vlan;
5173 
5174 	if (arg != ifp || ifp->if_type != IFT_ETHER)
5175 		return;
5176 
5177 	vlan = VLAN_DEVAT(ifp, vid);
5178 	VLAN_SETCOOKIE(vlan, ifp);
5179 }
5180 
5181 /*
5182  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5183  */
5184 static char *caps_decoder[] = {
5185 	"\20\001IPMI\002NCSI",				/* 0: NBM */
5186 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
5187 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
5188 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
5189 	    "\006HASHFILTER\007ETHOFLD",
5190 	"\20\001TOE",					/* 4: TOE */
5191 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
5192 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
5193 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5194 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5195 	    "\007T10DIF"
5196 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5197 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
5198 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
5199 		    "\004PO_INITIATOR\005PO_TARGET",
5200 };
5201 
5202 void
5203 t4_sysctls(struct adapter *sc)
5204 {
5205 	struct sysctl_ctx_list *ctx;
5206 	struct sysctl_oid *oid;
5207 	struct sysctl_oid_list *children, *c0;
5208 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5209 
5210 	ctx = device_get_sysctl_ctx(sc->dev);
5211 
5212 	/*
5213 	 * dev.t4nex.X.
5214 	 */
5215 	oid = device_get_sysctl_tree(sc->dev);
5216 	c0 = children = SYSCTL_CHILDREN(oid);
5217 
5218 	sc->sc_do_rxcopy = 1;
5219 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5220 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5221 
5222 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5223 	    sc->params.nports, "# of ports");
5224 
5225 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5226 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
5227 	    sysctl_bitfield, "A", "available doorbells");
5228 
5229 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5230 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
5231 
5232 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5233 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5234 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5235 	    "interrupt holdoff timer values (us)");
5236 
5237 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5238 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5239 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5240 	    "interrupt holdoff packet counter values");
5241 
5242 	t4_sge_sysctls(sc, ctx, children);
5243 
5244 	sc->lro_timeout = 100;
5245 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5246 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5247 
5248 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5249 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
5250 
5251 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5252 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5253 
5254 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5255 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5256 
5257 	if (sc->flags & IS_VF)
5258 		return;
5259 
5260 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5261 	    NULL, chip_rev(sc), "chip hardware revision");
5262 
5263 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5264 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5265 
5266 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5267 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5268 
5269 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5270 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5271 
5272 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5273 	    CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5274 
5275 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5276 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5277 
5278 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5279 	    sc->er_version, 0, "expansion ROM version");
5280 
5281 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5282 	    sc->bs_version, 0, "bootstrap firmware version");
5283 
5284 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5285 	    NULL, sc->params.scfg_vers, "serial config version");
5286 
5287 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5288 	    NULL, sc->params.vpd_vers, "VPD version");
5289 
5290 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5291 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5292 
5293 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5294 	    sc->cfcsum, "config file checksum");
5295 
5296 #define SYSCTL_CAP(name, n, text) \
5297 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5298 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
5299 	    sysctl_bitfield, "A", "available " text " capabilities")
5300 
5301 	SYSCTL_CAP(nbmcaps, 0, "NBM");
5302 	SYSCTL_CAP(linkcaps, 1, "link");
5303 	SYSCTL_CAP(switchcaps, 2, "switch");
5304 	SYSCTL_CAP(niccaps, 3, "NIC");
5305 	SYSCTL_CAP(toecaps, 4, "TCP offload");
5306 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
5307 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5308 	SYSCTL_CAP(cryptocaps, 7, "crypto");
5309 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
5310 #undef SYSCTL_CAP
5311 
5312 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5313 	    NULL, sc->tids.nftids, "number of filters");
5314 
5315 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5316 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5317 	    "chip temperature (in Celsius)");
5318 
5319 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
5320 	    &sc->params.core_vdd, 0, "core Vdd (in mV)");
5321 
5322 #ifdef SBUF_DRAIN
5323 	/*
5324 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
5325 	 */
5326 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5327 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5328 	    "logs and miscellaneous information");
5329 	children = SYSCTL_CHILDREN(oid);
5330 
5331 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
5332 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5333 	    sysctl_cctrl, "A", "congestion control");
5334 
5335 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
5336 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5337 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
5338 
5339 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
5340 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
5341 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
5342 
5343 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
5344 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
5345 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
5346 
5347 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
5348 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
5349 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
5350 
5351 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
5352 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
5353 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
5354 
5355 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
5356 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
5357 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
5358 
5359 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
5360 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5361 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
5362 	    "A", "CIM logic analyzer");
5363 
5364 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
5365 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5366 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
5367 
5368 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
5369 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
5370 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
5371 
5372 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
5373 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
5374 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
5375 
5376 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
5377 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
5378 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
5379 
5380 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5381 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5382 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5383 
5384 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5385 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5386 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5387 
5388 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5389 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5390 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5391 
5392 	if (chip_id(sc) > CHELSIO_T4) {
5393 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5394 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5395 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5396 
5397 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5398 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5399 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5400 	}
5401 
5402 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5403 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5404 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5405 
5406 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5407 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5408 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
5409 
5410 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5411 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5412 	    sysctl_cpl_stats, "A", "CPL statistics");
5413 
5414 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5415 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5416 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5417 
5418 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5419 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5420 	    sysctl_devlog, "A", "firmware's device log");
5421 
5422 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5423 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5424 	    sysctl_fcoe_stats, "A", "FCoE statistics");
5425 
5426 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5427 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5428 	    sysctl_hw_sched, "A", "hardware scheduler ");
5429 
5430 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5431 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5432 	    sysctl_l2t, "A", "hardware L2 table");
5433 
5434 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5435 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5436 	    sysctl_lb_stats, "A", "loopback statistics");
5437 
5438 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5439 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5440 	    sysctl_meminfo, "A", "memory regions");
5441 
5442 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5443 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5444 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5445 	    "A", "MPS TCAM entries");
5446 
5447 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5448 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5449 	    sysctl_path_mtus, "A", "path MTUs");
5450 
5451 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5452 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5453 	    sysctl_pm_stats, "A", "PM statistics");
5454 
5455 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5456 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5457 	    sysctl_rdma_stats, "A", "RDMA statistics");
5458 
5459 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5460 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5461 	    sysctl_tcp_stats, "A", "TCP statistics");
5462 
5463 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5464 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5465 	    sysctl_tids, "A", "TID information");
5466 
5467 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5468 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5469 	    sysctl_tp_err_stats, "A", "TP error statistics");
5470 
5471 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5472 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5473 	    "TP logic analyzer event capture mask");
5474 
5475 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5476 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5477 	    sysctl_tp_la, "A", "TP logic analyzer");
5478 
5479 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5480 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5481 	    sysctl_tx_rate, "A", "Tx rate");
5482 
5483 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5484 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5485 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5486 
5487 	if (chip_id(sc) >= CHELSIO_T5) {
5488 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5489 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5490 		    sysctl_wcwr_stats, "A", "write combined work requests");
5491 	}
5492 #endif
5493 
5494 #ifdef TCP_OFFLOAD
5495 	if (is_offload(sc)) {
5496 		int i;
5497 		char s[4];
5498 
5499 		/*
5500 		 * dev.t4nex.X.toe.
5501 		 */
5502 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5503 		    NULL, "TOE parameters");
5504 		children = SYSCTL_CHILDREN(oid);
5505 
5506 		sc->tt.cong_algorithm = -1;
5507 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
5508 		    CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
5509 		    "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
5510 		    "3 = highspeed)");
5511 
5512 		sc->tt.sndbuf = 256 * 1024;
5513 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5514 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
5515 
5516 		sc->tt.ddp = 0;
5517 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5518 		    &sc->tt.ddp, 0, "DDP allowed");
5519 
5520 		sc->tt.rx_coalesce = 1;
5521 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5522 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5523 
5524 		sc->tt.tls = 0;
5525 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tls", CTLFLAG_RW,
5526 		    &sc->tt.tls, 0, "Inline TLS allowed");
5527 
5528 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls_rx_ports",
5529 		    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tls_rx_ports,
5530 		    "I", "TCP ports that use inline TLS+TOE RX");
5531 
5532 		sc->tt.tx_align = 1;
5533 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5534 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5535 
5536 		sc->tt.tx_zcopy = 0;
5537 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
5538 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
5539 		    "Enable zero-copy aio_write(2)");
5540 
5541 		sc->tt.cop_managed_offloading = !!t4_cop_managed_offloading;
5542 		SYSCTL_ADD_INT(ctx, children, OID_AUTO,
5543 		    "cop_managed_offloading", CTLFLAG_RW,
5544 		    &sc->tt.cop_managed_offloading, 0,
5545 		    "COP (Connection Offload Policy) controls all TOE offload");
5546 
5547 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
5548 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
5549 		    "TP timer tick (us)");
5550 
5551 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
5552 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
5553 		    "TCP timestamp tick (us)");
5554 
5555 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5556 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5557 		    "DACK tick (us)");
5558 
5559 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5560 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5561 		    "IU", "DACK timer (us)");
5562 
5563 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5564 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5565 		    sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
5566 
5567 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5568 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5569 		    sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
5570 
5571 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5572 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5573 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
5574 
5575 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5576 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5577 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
5578 
5579 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5580 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5581 		    sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
5582 
5583 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
5584 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
5585 		    sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
5586 
5587 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
5588 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
5589 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
5590 
5591 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
5592 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
5593 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
5594 
5595 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
5596 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
5597 		    sysctl_tp_shift_cnt, "IU",
5598 		    "Number of SYN retransmissions before abort");
5599 
5600 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
5601 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
5602 		    sysctl_tp_shift_cnt, "IU",
5603 		    "Number of retransmissions before abort");
5604 
5605 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
5606 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
5607 		    sysctl_tp_shift_cnt, "IU",
5608 		    "Number of keepalive probes before abort");
5609 
5610 		oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
5611 		    CTLFLAG_RD, NULL, "TOE retransmit backoffs");
5612 		children = SYSCTL_CHILDREN(oid);
5613 		for (i = 0; i < 16; i++) {
5614 			snprintf(s, sizeof(s), "%u", i);
5615 			SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
5616 			    CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
5617 			    "IU", "TOE retransmit backoff");
5618 		}
5619 	}
5620 #endif
5621 }
5622 
5623 void
5624 vi_sysctls(struct vi_info *vi)
5625 {
5626 	struct sysctl_ctx_list *ctx;
5627 	struct sysctl_oid *oid;
5628 	struct sysctl_oid_list *children;
5629 
5630 	ctx = device_get_sysctl_ctx(vi->dev);
5631 
5632 	/*
5633 	 * dev.v?(cxgbe|cxl).X.
5634 	 */
5635 	oid = device_get_sysctl_tree(vi->dev);
5636 	children = SYSCTL_CHILDREN(oid);
5637 
5638 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
5639 	    vi->viid, "VI identifer");
5640 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
5641 	    &vi->nrxq, 0, "# of rx queues");
5642 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5643 	    &vi->ntxq, 0, "# of tx queues");
5644 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5645 	    &vi->first_rxq, 0, "index of first rx queue");
5646 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5647 	    &vi->first_txq, 0, "index of first tx queue");
5648 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
5649 	    vi->rss_size, "size of RSS indirection table");
5650 
5651 	if (IS_MAIN_VI(vi)) {
5652 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
5653 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
5654 		    "Reserve queue 0 for non-flowid packets");
5655 	}
5656 
5657 #ifdef TCP_OFFLOAD
5658 	if (vi->nofldrxq != 0) {
5659 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5660 		    &vi->nofldrxq, 0,
5661 		    "# of rx queues for offloaded TCP connections");
5662 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5663 		    &vi->nofldtxq, 0,
5664 		    "# of tx queues for offloaded TCP connections");
5665 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5666 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
5667 		    "index of first TOE rx queue");
5668 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5669 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
5670 		    "index of first TOE tx queue");
5671 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
5672 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5673 		    sysctl_holdoff_tmr_idx_ofld, "I",
5674 		    "holdoff timer index for TOE queues");
5675 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
5676 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
5677 		    sysctl_holdoff_pktc_idx_ofld, "I",
5678 		    "holdoff packet counter index for TOE queues");
5679 	}
5680 #endif
5681 #ifdef DEV_NETMAP
5682 	if (vi->nnmrxq != 0) {
5683 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5684 		    &vi->nnmrxq, 0, "# of netmap rx queues");
5685 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5686 		    &vi->nnmtxq, 0, "# of netmap tx queues");
5687 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5688 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
5689 		    "index of first netmap rx queue");
5690 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
5691 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
5692 		    "index of first netmap tx queue");
5693 	}
5694 #endif
5695 
5696 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
5697 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
5698 	    "holdoff timer index");
5699 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
5700 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
5701 	    "holdoff packet counter index");
5702 
5703 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
5704 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
5705 	    "rx queue size");
5706 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
5707 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
5708 	    "tx queue size");
5709 }
5710 
5711 static void
5712 cxgbe_sysctls(struct port_info *pi)
5713 {
5714 	struct sysctl_ctx_list *ctx;
5715 	struct sysctl_oid *oid;
5716 	struct sysctl_oid_list *children, *children2;
5717 	struct adapter *sc = pi->adapter;
5718 	int i;
5719 	char name[16];
5720 
5721 	ctx = device_get_sysctl_ctx(pi->dev);
5722 
5723 	/*
5724 	 * dev.cxgbe.X.
5725 	 */
5726 	oid = device_get_sysctl_tree(pi->dev);
5727 	children = SYSCTL_CHILDREN(oid);
5728 
5729 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
5730 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
5731 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
5732 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
5733 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
5734 		    "PHY temperature (in Celsius)");
5735 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
5736 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
5737 		    "PHY firmware version");
5738 	}
5739 
5740 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
5741 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
5742 	    "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
5743 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
5744 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
5745 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
5746 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
5747 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
5748 	    "autonegotiation (-1 = not supported)");
5749 
5750 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
5751 	    port_top_speed(pi), "max speed (in Gbps)");
5752 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
5753 	    pi->mps_bg_map, "MPS buffer group map");
5754 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
5755 	    NULL, pi->rx_e_chan_map, "TP rx e-channel map");
5756 
5757 	if (sc->flags & IS_VF)
5758 		return;
5759 
5760 	/*
5761 	 * dev.(cxgbe|cxl).X.tc.
5762 	 */
5763 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
5764 	    "Tx scheduler traffic classes (cl_rl)");
5765 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
5766 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
5767 
5768 		snprintf(name, sizeof(name), "%d", i);
5769 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
5770 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
5771 		    "traffic class"));
5772 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
5773 		    &tc->flags, 0, "flags");
5774 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
5775 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
5776 #ifdef SBUF_DRAIN
5777 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
5778 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
5779 		    sysctl_tc_params, "A", "traffic class parameters");
5780 #endif
5781 	}
5782 
5783 	/*
5784 	 * dev.cxgbe.X.stats.
5785 	 */
5786 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
5787 	    NULL, "port statistics");
5788 	children = SYSCTL_CHILDREN(oid);
5789 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
5790 	    &pi->tx_parse_error, 0,
5791 	    "# of tx packets with invalid length or # of segments");
5792 
5793 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
5794 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
5795 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
5796 	    sysctl_handle_t4_reg64, "QU", desc)
5797 
5798 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
5799 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
5800 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
5801 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
5802 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
5803 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
5804 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
5805 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
5806 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
5807 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
5808 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
5809 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
5810 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
5811 	    "# of tx frames in this range",
5812 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
5813 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
5814 	    "# of tx frames in this range",
5815 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
5816 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
5817 	    "# of tx frames in this range",
5818 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
5819 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
5820 	    "# of tx frames in this range",
5821 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
5822 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
5823 	    "# of tx frames in this range",
5824 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
5825 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
5826 	    "# of tx frames in this range",
5827 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
5828 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
5829 	    "# of tx frames in this range",
5830 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
5831 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
5832 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
5833 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
5834 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
5835 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
5836 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
5837 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5838 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5839 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5840 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5841 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5842 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5843 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5844 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5845 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5846 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5847 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5848 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5849 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5850 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5851 
5852 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5853 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5854 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5855 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5856 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5857 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5858 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5859 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5860 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5861 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5862 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5863 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5864 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5865 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5866 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5867 	    "# of frames received with bad FCS",
5868 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5869 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5870 	    "# of frames received with length error",
5871 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5872 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5873 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5874 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5875 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5876 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5877 	    "# of rx frames in this range",
5878 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5879 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5880 	    "# of rx frames in this range",
5881 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5882 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5883 	    "# of rx frames in this range",
5884 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5885 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5886 	    "# of rx frames in this range",
5887 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5888 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5889 	    "# of rx frames in this range",
5890 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5891 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5892 	    "# of rx frames in this range",
5893 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5894 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5895 	    "# of rx frames in this range",
5896 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5897 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5898 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5899 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5900 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5901 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5902 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5903 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5904 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5905 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5906 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5907 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5908 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5909 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5910 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5911 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5912 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5913 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5914 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5915 
5916 #undef SYSCTL_ADD_T4_REG64
5917 
5918 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5919 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5920 	    &pi->stats.name, desc)
5921 
5922 	/* We get these from port_stats and they may be stale by up to 1s */
5923 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5924 	    "# drops due to buffer-group 0 overflows");
5925 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5926 	    "# drops due to buffer-group 1 overflows");
5927 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5928 	    "# drops due to buffer-group 2 overflows");
5929 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5930 	    "# drops due to buffer-group 3 overflows");
5931 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5932 	    "# of buffer-group 0 truncated packets");
5933 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5934 	    "# of buffer-group 1 truncated packets");
5935 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5936 	    "# of buffer-group 2 truncated packets");
5937 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5938 	    "# of buffer-group 3 truncated packets");
5939 
5940 #undef SYSCTL_ADD_T4_PORTSTAT
5941 
5942 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_records",
5943 	    CTLFLAG_RD, &pi->tx_tls_records,
5944 	    "# of TLS records transmitted");
5945 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_octets",
5946 	    CTLFLAG_RD, &pi->tx_tls_octets,
5947 	    "# of payload octets in transmitted TLS records");
5948 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_records",
5949 	    CTLFLAG_RD, &pi->rx_tls_records,
5950 	    "# of TLS records received");
5951 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_octets",
5952 	    CTLFLAG_RD, &pi->rx_tls_octets,
5953 	    "# of payload octets in received TLS records");
5954 }
5955 
5956 static int
5957 sysctl_int_array(SYSCTL_HANDLER_ARGS)
5958 {
5959 	int rc, *i, space = 0;
5960 	struct sbuf sb;
5961 
5962 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
5963 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
5964 		if (space)
5965 			sbuf_printf(&sb, " ");
5966 		sbuf_printf(&sb, "%d", *i);
5967 		space = 1;
5968 	}
5969 	rc = sbuf_finish(&sb);
5970 	sbuf_delete(&sb);
5971 	return (rc);
5972 }
5973 
5974 static int
5975 sysctl_bitfield(SYSCTL_HANDLER_ARGS)
5976 {
5977 	int rc;
5978 	struct sbuf *sb;
5979 
5980 	rc = sysctl_wire_old_buffer(req, 0);
5981 	if (rc != 0)
5982 		return(rc);
5983 
5984 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5985 	if (sb == NULL)
5986 		return (ENOMEM);
5987 
5988 	sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
5989 	rc = sbuf_finish(sb);
5990 	sbuf_delete(sb);
5991 
5992 	return (rc);
5993 }
5994 
5995 static int
5996 sysctl_btphy(SYSCTL_HANDLER_ARGS)
5997 {
5998 	struct port_info *pi = arg1;
5999 	int op = arg2;
6000 	struct adapter *sc = pi->adapter;
6001 	u_int v;
6002 	int rc;
6003 
6004 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
6005 	if (rc)
6006 		return (rc);
6007 	/* XXX: magic numbers */
6008 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
6009 	    &v);
6010 	end_synchronized_op(sc, 0);
6011 	if (rc)
6012 		return (rc);
6013 	if (op == 0)
6014 		v /= 256;
6015 
6016 	rc = sysctl_handle_int(oidp, &v, 0, req);
6017 	return (rc);
6018 }
6019 
6020 static int
6021 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
6022 {
6023 	struct vi_info *vi = arg1;
6024 	int rc, val;
6025 
6026 	val = vi->rsrv_noflowq;
6027 	rc = sysctl_handle_int(oidp, &val, 0, req);
6028 	if (rc != 0 || req->newptr == NULL)
6029 		return (rc);
6030 
6031 	if ((val >= 1) && (vi->ntxq > 1))
6032 		vi->rsrv_noflowq = 1;
6033 	else
6034 		vi->rsrv_noflowq = 0;
6035 
6036 	return (rc);
6037 }
6038 
6039 static int
6040 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
6041 {
6042 	struct vi_info *vi = arg1;
6043 	struct adapter *sc = vi->pi->adapter;
6044 	int idx, rc, i;
6045 	struct sge_rxq *rxq;
6046 	uint8_t v;
6047 
6048 	idx = vi->tmr_idx;
6049 
6050 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6051 	if (rc != 0 || req->newptr == NULL)
6052 		return (rc);
6053 
6054 	if (idx < 0 || idx >= SGE_NTIMERS)
6055 		return (EINVAL);
6056 
6057 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6058 	    "t4tmr");
6059 	if (rc)
6060 		return (rc);
6061 
6062 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
6063 	for_each_rxq(vi, i, rxq) {
6064 #ifdef atomic_store_rel_8
6065 		atomic_store_rel_8(&rxq->iq.intr_params, v);
6066 #else
6067 		rxq->iq.intr_params = v;
6068 #endif
6069 	}
6070 	vi->tmr_idx = idx;
6071 
6072 	end_synchronized_op(sc, LOCK_HELD);
6073 	return (0);
6074 }
6075 
6076 static int
6077 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
6078 {
6079 	struct vi_info *vi = arg1;
6080 	struct adapter *sc = vi->pi->adapter;
6081 	int idx, rc;
6082 
6083 	idx = vi->pktc_idx;
6084 
6085 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6086 	if (rc != 0 || req->newptr == NULL)
6087 		return (rc);
6088 
6089 	if (idx < -1 || idx >= SGE_NCOUNTERS)
6090 		return (EINVAL);
6091 
6092 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6093 	    "t4pktc");
6094 	if (rc)
6095 		return (rc);
6096 
6097 	if (vi->flags & VI_INIT_DONE)
6098 		rc = EBUSY; /* cannot be changed once the queues are created */
6099 	else
6100 		vi->pktc_idx = idx;
6101 
6102 	end_synchronized_op(sc, LOCK_HELD);
6103 	return (rc);
6104 }
6105 
6106 static int
6107 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
6108 {
6109 	struct vi_info *vi = arg1;
6110 	struct adapter *sc = vi->pi->adapter;
6111 	int qsize, rc;
6112 
6113 	qsize = vi->qsize_rxq;
6114 
6115 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6116 	if (rc != 0 || req->newptr == NULL)
6117 		return (rc);
6118 
6119 	if (qsize < 128 || (qsize & 7))
6120 		return (EINVAL);
6121 
6122 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6123 	    "t4rxqs");
6124 	if (rc)
6125 		return (rc);
6126 
6127 	if (vi->flags & VI_INIT_DONE)
6128 		rc = EBUSY; /* cannot be changed once the queues are created */
6129 	else
6130 		vi->qsize_rxq = qsize;
6131 
6132 	end_synchronized_op(sc, LOCK_HELD);
6133 	return (rc);
6134 }
6135 
6136 static int
6137 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
6138 {
6139 	struct vi_info *vi = arg1;
6140 	struct adapter *sc = vi->pi->adapter;
6141 	int qsize, rc;
6142 
6143 	qsize = vi->qsize_txq;
6144 
6145 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6146 	if (rc != 0 || req->newptr == NULL)
6147 		return (rc);
6148 
6149 	if (qsize < 128 || qsize > 65536)
6150 		return (EINVAL);
6151 
6152 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6153 	    "t4txqs");
6154 	if (rc)
6155 		return (rc);
6156 
6157 	if (vi->flags & VI_INIT_DONE)
6158 		rc = EBUSY; /* cannot be changed once the queues are created */
6159 	else
6160 		vi->qsize_txq = qsize;
6161 
6162 	end_synchronized_op(sc, LOCK_HELD);
6163 	return (rc);
6164 }
6165 
6166 static int
6167 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
6168 {
6169 	struct port_info *pi = arg1;
6170 	struct adapter *sc = pi->adapter;
6171 	struct link_config *lc = &pi->link_cfg;
6172 	int rc;
6173 
6174 	if (req->newptr == NULL) {
6175 		struct sbuf *sb;
6176 		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
6177 
6178 		rc = sysctl_wire_old_buffer(req, 0);
6179 		if (rc != 0)
6180 			return(rc);
6181 
6182 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6183 		if (sb == NULL)
6184 			return (ENOMEM);
6185 
6186 		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
6187 		rc = sbuf_finish(sb);
6188 		sbuf_delete(sb);
6189 	} else {
6190 		char s[2];
6191 		int n;
6192 
6193 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
6194 		s[1] = 0;
6195 
6196 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6197 		if (rc != 0)
6198 			return(rc);
6199 
6200 		if (s[1] != 0)
6201 			return (EINVAL);
6202 		if (s[0] < '0' || s[0] > '9')
6203 			return (EINVAL);	/* not a number */
6204 		n = s[0] - '0';
6205 		if (n & ~(PAUSE_TX | PAUSE_RX))
6206 			return (EINVAL);	/* some other bit is set too */
6207 
6208 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6209 		    "t4PAUSE");
6210 		if (rc)
6211 			return (rc);
6212 		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
6213 			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
6214 			lc->requested_fc |= n;
6215 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6216 			if (rc == 0) {
6217 				lc->fc = lc->requested_fc;
6218 			}
6219 		}
6220 		end_synchronized_op(sc, 0);
6221 	}
6222 
6223 	return (rc);
6224 }
6225 
6226 static int
6227 sysctl_fec(SYSCTL_HANDLER_ARGS)
6228 {
6229 	struct port_info *pi = arg1;
6230 	struct adapter *sc = pi->adapter;
6231 	struct link_config *lc = &pi->link_cfg;
6232 	int rc;
6233 
6234 	if (req->newptr == NULL) {
6235 		struct sbuf *sb;
6236 		static char *bits = "\20\1RS\2BASER_RS\3RESERVED";
6237 
6238 		rc = sysctl_wire_old_buffer(req, 0);
6239 		if (rc != 0)
6240 			return(rc);
6241 
6242 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6243 		if (sb == NULL)
6244 			return (ENOMEM);
6245 
6246 		sbuf_printf(sb, "%b", lc->fec & M_FW_PORT_CAP_FEC, bits);
6247 		rc = sbuf_finish(sb);
6248 		sbuf_delete(sb);
6249 	} else {
6250 		char s[2];
6251 		int n;
6252 
6253 		s[0] = '0' + (lc->requested_fec & M_FW_PORT_CAP_FEC);
6254 		s[1] = 0;
6255 
6256 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6257 		if (rc != 0)
6258 			return(rc);
6259 
6260 		if (s[1] != 0)
6261 			return (EINVAL);
6262 		if (s[0] < '0' || s[0] > '9')
6263 			return (EINVAL);	/* not a number */
6264 		n = s[0] - '0';
6265 		if (n & ~M_FW_PORT_CAP_FEC)
6266 			return (EINVAL);	/* some other bit is set too */
6267 
6268 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6269 		    "t4fec");
6270 		if (rc)
6271 			return (rc);
6272 		if ((lc->requested_fec & M_FW_PORT_CAP_FEC) != n) {
6273 			lc->requested_fec = n &
6274 			    G_FW_PORT_CAP_FEC(lc->supported);
6275 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6276 			if (rc == 0) {
6277 				lc->fec = lc->requested_fec;
6278 			}
6279 		}
6280 		end_synchronized_op(sc, 0);
6281 	}
6282 
6283 	return (rc);
6284 }
6285 
6286 static int
6287 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
6288 {
6289 	struct port_info *pi = arg1;
6290 	struct adapter *sc = pi->adapter;
6291 	struct link_config *lc = &pi->link_cfg;
6292 	int rc, val, old;
6293 
6294 	if (lc->supported & FW_PORT_CAP_ANEG)
6295 		val = lc->requested_aneg == AUTONEG_ENABLE ? 1 : 0;
6296 	else
6297 		val = -1;
6298 	rc = sysctl_handle_int(oidp, &val, 0, req);
6299 	if (rc != 0 || req->newptr == NULL)
6300 		return (rc);
6301 	if ((lc->supported & FW_PORT_CAP_ANEG) == 0)
6302 		return (ENOTSUP);
6303 
6304 	if (val == 0)
6305 		val = AUTONEG_DISABLE;
6306 	else if (val == 1)
6307 		val = AUTONEG_ENABLE;
6308 	else
6309 		return (EINVAL);
6310 	if (lc->requested_aneg == val)
6311 		return (0);	/* no change */
6312 
6313 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6314 	    "t4aneg");
6315 	if (rc)
6316 		return (rc);
6317 	old = lc->requested_aneg;
6318 	lc->requested_aneg = val;
6319 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6320 	if (rc != 0)
6321 		lc->requested_aneg = old;
6322 	end_synchronized_op(sc, 0);
6323 	return (rc);
6324 }
6325 
6326 static int
6327 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
6328 {
6329 	struct adapter *sc = arg1;
6330 	int reg = arg2;
6331 	uint64_t val;
6332 
6333 	val = t4_read_reg64(sc, reg);
6334 
6335 	return (sysctl_handle_64(oidp, &val, 0, req));
6336 }
6337 
6338 static int
6339 sysctl_temperature(SYSCTL_HANDLER_ARGS)
6340 {
6341 	struct adapter *sc = arg1;
6342 	int rc, t;
6343 	uint32_t param, val;
6344 
6345 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
6346 	if (rc)
6347 		return (rc);
6348 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6349 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
6350 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
6351 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6352 	end_synchronized_op(sc, 0);
6353 	if (rc)
6354 		return (rc);
6355 
6356 	/* unknown is returned as 0 but we display -1 in that case */
6357 	t = val == 0 ? -1 : val;
6358 
6359 	rc = sysctl_handle_int(oidp, &t, 0, req);
6360 	return (rc);
6361 }
6362 
6363 #ifdef SBUF_DRAIN
6364 static int
6365 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
6366 {
6367 	struct adapter *sc = arg1;
6368 	struct sbuf *sb;
6369 	int rc, i;
6370 	uint16_t incr[NMTUS][NCCTRL_WIN];
6371 	static const char *dec_fac[] = {
6372 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
6373 		"0.9375"
6374 	};
6375 
6376 	rc = sysctl_wire_old_buffer(req, 0);
6377 	if (rc != 0)
6378 		return (rc);
6379 
6380 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6381 	if (sb == NULL)
6382 		return (ENOMEM);
6383 
6384 	t4_read_cong_tbl(sc, incr);
6385 
6386 	for (i = 0; i < NCCTRL_WIN; ++i) {
6387 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
6388 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
6389 		    incr[5][i], incr[6][i], incr[7][i]);
6390 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
6391 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
6392 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
6393 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
6394 	}
6395 
6396 	rc = sbuf_finish(sb);
6397 	sbuf_delete(sb);
6398 
6399 	return (rc);
6400 }
6401 
6402 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
6403 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
6404 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
6405 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
6406 };
6407 
6408 static int
6409 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
6410 {
6411 	struct adapter *sc = arg1;
6412 	struct sbuf *sb;
6413 	int rc, i, n, qid = arg2;
6414 	uint32_t *buf, *p;
6415 	char *qtype;
6416 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
6417 
6418 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
6419 	    ("%s: bad qid %d\n", __func__, qid));
6420 
6421 	if (qid < CIM_NUM_IBQ) {
6422 		/* inbound queue */
6423 		qtype = "IBQ";
6424 		n = 4 * CIM_IBQ_SIZE;
6425 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6426 		rc = t4_read_cim_ibq(sc, qid, buf, n);
6427 	} else {
6428 		/* outbound queue */
6429 		qtype = "OBQ";
6430 		qid -= CIM_NUM_IBQ;
6431 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
6432 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6433 		rc = t4_read_cim_obq(sc, qid, buf, n);
6434 	}
6435 
6436 	if (rc < 0) {
6437 		rc = -rc;
6438 		goto done;
6439 	}
6440 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
6441 
6442 	rc = sysctl_wire_old_buffer(req, 0);
6443 	if (rc != 0)
6444 		goto done;
6445 
6446 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6447 	if (sb == NULL) {
6448 		rc = ENOMEM;
6449 		goto done;
6450 	}
6451 
6452 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
6453 	for (i = 0, p = buf; i < n; i += 16, p += 4)
6454 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
6455 		    p[2], p[3]);
6456 
6457 	rc = sbuf_finish(sb);
6458 	sbuf_delete(sb);
6459 done:
6460 	free(buf, M_CXGBE);
6461 	return (rc);
6462 }
6463 
6464 static int
6465 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
6466 {
6467 	struct adapter *sc = arg1;
6468 	u_int cfg;
6469 	struct sbuf *sb;
6470 	uint32_t *buf, *p;
6471 	int rc;
6472 
6473 	MPASS(chip_id(sc) <= CHELSIO_T5);
6474 
6475 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6476 	if (rc != 0)
6477 		return (rc);
6478 
6479 	rc = sysctl_wire_old_buffer(req, 0);
6480 	if (rc != 0)
6481 		return (rc);
6482 
6483 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6484 	if (sb == NULL)
6485 		return (ENOMEM);
6486 
6487 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6488 	    M_ZERO | M_WAITOK);
6489 
6490 	rc = -t4_cim_read_la(sc, buf, NULL);
6491 	if (rc != 0)
6492 		goto done;
6493 
6494 	sbuf_printf(sb, "Status   Data      PC%s",
6495 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6496 	    "     LS0Stat  LS0Addr             LS0Data");
6497 
6498 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
6499 		if (cfg & F_UPDBGLACAPTPCONLY) {
6500 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
6501 			    p[6], p[7]);
6502 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
6503 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
6504 			    p[4] & 0xff, p[5] >> 8);
6505 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
6506 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6507 			    p[1] & 0xf, p[2] >> 4);
6508 		} else {
6509 			sbuf_printf(sb,
6510 			    "\n  %02x   %x%07x %x%07x %08x %08x "
6511 			    "%08x%08x%08x%08x",
6512 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6513 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
6514 			    p[6], p[7]);
6515 		}
6516 	}
6517 
6518 	rc = sbuf_finish(sb);
6519 	sbuf_delete(sb);
6520 done:
6521 	free(buf, M_CXGBE);
6522 	return (rc);
6523 }
6524 
6525 static int
6526 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
6527 {
6528 	struct adapter *sc = arg1;
6529 	u_int cfg;
6530 	struct sbuf *sb;
6531 	uint32_t *buf, *p;
6532 	int rc;
6533 
6534 	MPASS(chip_id(sc) > CHELSIO_T5);
6535 
6536 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6537 	if (rc != 0)
6538 		return (rc);
6539 
6540 	rc = sysctl_wire_old_buffer(req, 0);
6541 	if (rc != 0)
6542 		return (rc);
6543 
6544 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6545 	if (sb == NULL)
6546 		return (ENOMEM);
6547 
6548 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6549 	    M_ZERO | M_WAITOK);
6550 
6551 	rc = -t4_cim_read_la(sc, buf, NULL);
6552 	if (rc != 0)
6553 		goto done;
6554 
6555 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
6556 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6557 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
6558 
6559 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
6560 		if (cfg & F_UPDBGLACAPTPCONLY) {
6561 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
6562 			    p[3] & 0xff, p[2], p[1], p[0]);
6563 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
6564 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
6565 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
6566 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
6567 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
6568 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
6569 			    p[6] >> 16);
6570 		} else {
6571 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
6572 			    "%08x %08x %08x %08x %08x %08x",
6573 			    (p[9] >> 16) & 0xff,
6574 			    p[9] & 0xffff, p[8] >> 16,
6575 			    p[8] & 0xffff, p[7] >> 16,
6576 			    p[7] & 0xffff, p[6] >> 16,
6577 			    p[2], p[1], p[0], p[5], p[4], p[3]);
6578 		}
6579 	}
6580 
6581 	rc = sbuf_finish(sb);
6582 	sbuf_delete(sb);
6583 done:
6584 	free(buf, M_CXGBE);
6585 	return (rc);
6586 }
6587 
6588 static int
6589 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
6590 {
6591 	struct adapter *sc = arg1;
6592 	u_int i;
6593 	struct sbuf *sb;
6594 	uint32_t *buf, *p;
6595 	int rc;
6596 
6597 	rc = sysctl_wire_old_buffer(req, 0);
6598 	if (rc != 0)
6599 		return (rc);
6600 
6601 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6602 	if (sb == NULL)
6603 		return (ENOMEM);
6604 
6605 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
6606 	    M_ZERO | M_WAITOK);
6607 
6608 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
6609 	p = buf;
6610 
6611 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6612 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
6613 		    p[1], p[0]);
6614 	}
6615 
6616 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
6617 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6618 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
6619 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
6620 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
6621 		    (p[1] >> 2) | ((p[2] & 3) << 30),
6622 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
6623 		    p[0] & 1);
6624 	}
6625 
6626 	rc = sbuf_finish(sb);
6627 	sbuf_delete(sb);
6628 	free(buf, M_CXGBE);
6629 	return (rc);
6630 }
6631 
6632 static int
6633 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
6634 {
6635 	struct adapter *sc = arg1;
6636 	u_int i;
6637 	struct sbuf *sb;
6638 	uint32_t *buf, *p;
6639 	int rc;
6640 
6641 	rc = sysctl_wire_old_buffer(req, 0);
6642 	if (rc != 0)
6643 		return (rc);
6644 
6645 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6646 	if (sb == NULL)
6647 		return (ENOMEM);
6648 
6649 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
6650 	    M_ZERO | M_WAITOK);
6651 
6652 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
6653 	p = buf;
6654 
6655 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
6656 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6657 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
6658 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
6659 		    p[4], p[3], p[2], p[1], p[0]);
6660 	}
6661 
6662 	sbuf_printf(sb, "\n\nCntl ID               Data");
6663 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6664 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
6665 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
6666 	}
6667 
6668 	rc = sbuf_finish(sb);
6669 	sbuf_delete(sb);
6670 	free(buf, M_CXGBE);
6671 	return (rc);
6672 }
6673 
6674 static int
6675 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
6676 {
6677 	struct adapter *sc = arg1;
6678 	struct sbuf *sb;
6679 	int rc, i;
6680 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6681 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6682 	uint16_t thres[CIM_NUM_IBQ];
6683 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
6684 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
6685 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
6686 
6687 	cim_num_obq = sc->chip_params->cim_num_obq;
6688 	if (is_t4(sc)) {
6689 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
6690 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
6691 	} else {
6692 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
6693 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
6694 	}
6695 	nq = CIM_NUM_IBQ + cim_num_obq;
6696 
6697 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
6698 	if (rc == 0)
6699 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
6700 	if (rc != 0)
6701 		return (rc);
6702 
6703 	t4_read_cimq_cfg(sc, base, size, thres);
6704 
6705 	rc = sysctl_wire_old_buffer(req, 0);
6706 	if (rc != 0)
6707 		return (rc);
6708 
6709 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6710 	if (sb == NULL)
6711 		return (ENOMEM);
6712 
6713 	sbuf_printf(sb,
6714 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
6715 
6716 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
6717 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
6718 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
6719 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6720 		    G_QUEREMFLITS(p[2]) * 16);
6721 	for ( ; i < nq; i++, p += 4, wr += 2)
6722 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
6723 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
6724 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6725 		    G_QUEREMFLITS(p[2]) * 16);
6726 
6727 	rc = sbuf_finish(sb);
6728 	sbuf_delete(sb);
6729 
6730 	return (rc);
6731 }
6732 
6733 static int
6734 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
6735 {
6736 	struct adapter *sc = arg1;
6737 	struct sbuf *sb;
6738 	int rc;
6739 	struct tp_cpl_stats stats;
6740 
6741 	rc = sysctl_wire_old_buffer(req, 0);
6742 	if (rc != 0)
6743 		return (rc);
6744 
6745 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6746 	if (sb == NULL)
6747 		return (ENOMEM);
6748 
6749 	mtx_lock(&sc->reg_lock);
6750 	t4_tp_get_cpl_stats(sc, &stats, 0);
6751 	mtx_unlock(&sc->reg_lock);
6752 
6753 	if (sc->chip_params->nchan > 2) {
6754 		sbuf_printf(sb, "                 channel 0  channel 1"
6755 		    "  channel 2  channel 3");
6756 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
6757 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
6758 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
6759 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
6760 	} else {
6761 		sbuf_printf(sb, "                 channel 0  channel 1");
6762 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
6763 		    stats.req[0], stats.req[1]);
6764 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
6765 		    stats.rsp[0], stats.rsp[1]);
6766 	}
6767 
6768 	rc = sbuf_finish(sb);
6769 	sbuf_delete(sb);
6770 
6771 	return (rc);
6772 }
6773 
6774 static int
6775 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
6776 {
6777 	struct adapter *sc = arg1;
6778 	struct sbuf *sb;
6779 	int rc;
6780 	struct tp_usm_stats stats;
6781 
6782 	rc = sysctl_wire_old_buffer(req, 0);
6783 	if (rc != 0)
6784 		return(rc);
6785 
6786 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6787 	if (sb == NULL)
6788 		return (ENOMEM);
6789 
6790 	t4_get_usm_stats(sc, &stats, 1);
6791 
6792 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
6793 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
6794 	sbuf_printf(sb, "Drops:  %u", stats.drops);
6795 
6796 	rc = sbuf_finish(sb);
6797 	sbuf_delete(sb);
6798 
6799 	return (rc);
6800 }
6801 
6802 static const char * const devlog_level_strings[] = {
6803 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
6804 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
6805 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
6806 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
6807 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
6808 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
6809 };
6810 
6811 static const char * const devlog_facility_strings[] = {
6812 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
6813 	[FW_DEVLOG_FACILITY_CF]		= "CF",
6814 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
6815 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
6816 	[FW_DEVLOG_FACILITY_RES]	= "RES",
6817 	[FW_DEVLOG_FACILITY_HW]		= "HW",
6818 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
6819 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
6820 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
6821 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
6822 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
6823 	[FW_DEVLOG_FACILITY_VI]		= "VI",
6824 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
6825 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
6826 	[FW_DEVLOG_FACILITY_TM]		= "TM",
6827 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
6828 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
6829 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
6830 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
6831 	[FW_DEVLOG_FACILITY_RI]		= "RI",
6832 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
6833 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
6834 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
6835 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
6836 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
6837 };
6838 
6839 static int
6840 sysctl_devlog(SYSCTL_HANDLER_ARGS)
6841 {
6842 	struct adapter *sc = arg1;
6843 	struct devlog_params *dparams = &sc->params.devlog;
6844 	struct fw_devlog_e *buf, *e;
6845 	int i, j, rc, nentries, first = 0;
6846 	struct sbuf *sb;
6847 	uint64_t ftstamp = UINT64_MAX;
6848 
6849 	if (dparams->addr == 0)
6850 		return (ENXIO);
6851 
6852 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
6853 	if (buf == NULL)
6854 		return (ENOMEM);
6855 
6856 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
6857 	if (rc != 0)
6858 		goto done;
6859 
6860 	nentries = dparams->size / sizeof(struct fw_devlog_e);
6861 	for (i = 0; i < nentries; i++) {
6862 		e = &buf[i];
6863 
6864 		if (e->timestamp == 0)
6865 			break;	/* end */
6866 
6867 		e->timestamp = be64toh(e->timestamp);
6868 		e->seqno = be32toh(e->seqno);
6869 		for (j = 0; j < 8; j++)
6870 			e->params[j] = be32toh(e->params[j]);
6871 
6872 		if (e->timestamp < ftstamp) {
6873 			ftstamp = e->timestamp;
6874 			first = i;
6875 		}
6876 	}
6877 
6878 	if (buf[first].timestamp == 0)
6879 		goto done;	/* nothing in the log */
6880 
6881 	rc = sysctl_wire_old_buffer(req, 0);
6882 	if (rc != 0)
6883 		goto done;
6884 
6885 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6886 	if (sb == NULL) {
6887 		rc = ENOMEM;
6888 		goto done;
6889 	}
6890 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
6891 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
6892 
6893 	i = first;
6894 	do {
6895 		e = &buf[i];
6896 		if (e->timestamp == 0)
6897 			break;	/* end */
6898 
6899 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
6900 		    e->seqno, e->timestamp,
6901 		    (e->level < nitems(devlog_level_strings) ?
6902 			devlog_level_strings[e->level] : "UNKNOWN"),
6903 		    (e->facility < nitems(devlog_facility_strings) ?
6904 			devlog_facility_strings[e->facility] : "UNKNOWN"));
6905 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
6906 		    e->params[2], e->params[3], e->params[4],
6907 		    e->params[5], e->params[6], e->params[7]);
6908 
6909 		if (++i == nentries)
6910 			i = 0;
6911 	} while (i != first);
6912 
6913 	rc = sbuf_finish(sb);
6914 	sbuf_delete(sb);
6915 done:
6916 	free(buf, M_CXGBE);
6917 	return (rc);
6918 }
6919 
6920 static int
6921 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
6922 {
6923 	struct adapter *sc = arg1;
6924 	struct sbuf *sb;
6925 	int rc;
6926 	struct tp_fcoe_stats stats[MAX_NCHAN];
6927 	int i, nchan = sc->chip_params->nchan;
6928 
6929 	rc = sysctl_wire_old_buffer(req, 0);
6930 	if (rc != 0)
6931 		return (rc);
6932 
6933 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6934 	if (sb == NULL)
6935 		return (ENOMEM);
6936 
6937 	for (i = 0; i < nchan; i++)
6938 		t4_get_fcoe_stats(sc, i, &stats[i], 1);
6939 
6940 	if (nchan > 2) {
6941 		sbuf_printf(sb, "                   channel 0        channel 1"
6942 		    "        channel 2        channel 3");
6943 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
6944 		    stats[0].octets_ddp, stats[1].octets_ddp,
6945 		    stats[2].octets_ddp, stats[3].octets_ddp);
6946 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
6947 		    stats[0].frames_ddp, stats[1].frames_ddp,
6948 		    stats[2].frames_ddp, stats[3].frames_ddp);
6949 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
6950 		    stats[0].frames_drop, stats[1].frames_drop,
6951 		    stats[2].frames_drop, stats[3].frames_drop);
6952 	} else {
6953 		sbuf_printf(sb, "                   channel 0        channel 1");
6954 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
6955 		    stats[0].octets_ddp, stats[1].octets_ddp);
6956 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
6957 		    stats[0].frames_ddp, stats[1].frames_ddp);
6958 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
6959 		    stats[0].frames_drop, stats[1].frames_drop);
6960 	}
6961 
6962 	rc = sbuf_finish(sb);
6963 	sbuf_delete(sb);
6964 
6965 	return (rc);
6966 }
6967 
6968 static int
6969 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
6970 {
6971 	struct adapter *sc = arg1;
6972 	struct sbuf *sb;
6973 	int rc, i;
6974 	unsigned int map, kbps, ipg, mode;
6975 	unsigned int pace_tab[NTX_SCHED];
6976 
6977 	rc = sysctl_wire_old_buffer(req, 0);
6978 	if (rc != 0)
6979 		return (rc);
6980 
6981 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6982 	if (sb == NULL)
6983 		return (ENOMEM);
6984 
6985 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
6986 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
6987 	t4_read_pace_tbl(sc, pace_tab);
6988 
6989 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
6990 	    "Class IPG (0.1 ns)   Flow IPG (us)");
6991 
6992 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
6993 		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
6994 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
6995 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
6996 		if (kbps)
6997 			sbuf_printf(sb, "%9u     ", kbps);
6998 		else
6999 			sbuf_printf(sb, " disabled     ");
7000 
7001 		if (ipg)
7002 			sbuf_printf(sb, "%13u        ", ipg);
7003 		else
7004 			sbuf_printf(sb, "     disabled        ");
7005 
7006 		if (pace_tab[i])
7007 			sbuf_printf(sb, "%10u", pace_tab[i]);
7008 		else
7009 			sbuf_printf(sb, "  disabled");
7010 	}
7011 
7012 	rc = sbuf_finish(sb);
7013 	sbuf_delete(sb);
7014 
7015 	return (rc);
7016 }
7017 
7018 static int
7019 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
7020 {
7021 	struct adapter *sc = arg1;
7022 	struct sbuf *sb;
7023 	int rc, i, j;
7024 	uint64_t *p0, *p1;
7025 	struct lb_port_stats s[2];
7026 	static const char *stat_name[] = {
7027 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
7028 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
7029 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
7030 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
7031 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
7032 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
7033 		"BG2FramesTrunc:", "BG3FramesTrunc:"
7034 	};
7035 
7036 	rc = sysctl_wire_old_buffer(req, 0);
7037 	if (rc != 0)
7038 		return (rc);
7039 
7040 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7041 	if (sb == NULL)
7042 		return (ENOMEM);
7043 
7044 	memset(s, 0, sizeof(s));
7045 
7046 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
7047 		t4_get_lb_stats(sc, i, &s[0]);
7048 		t4_get_lb_stats(sc, i + 1, &s[1]);
7049 
7050 		p0 = &s[0].octets;
7051 		p1 = &s[1].octets;
7052 		sbuf_printf(sb, "%s                       Loopback %u"
7053 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
7054 
7055 		for (j = 0; j < nitems(stat_name); j++)
7056 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
7057 				   *p0++, *p1++);
7058 	}
7059 
7060 	rc = sbuf_finish(sb);
7061 	sbuf_delete(sb);
7062 
7063 	return (rc);
7064 }
7065 
7066 static int
7067 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
7068 {
7069 	int rc = 0;
7070 	struct port_info *pi = arg1;
7071 	struct link_config *lc = &pi->link_cfg;
7072 	struct sbuf *sb;
7073 
7074 	rc = sysctl_wire_old_buffer(req, 0);
7075 	if (rc != 0)
7076 		return(rc);
7077 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
7078 	if (sb == NULL)
7079 		return (ENOMEM);
7080 
7081 	if (lc->link_ok || lc->link_down_rc == 255)
7082 		sbuf_printf(sb, "n/a");
7083 	else
7084 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
7085 
7086 	rc = sbuf_finish(sb);
7087 	sbuf_delete(sb);
7088 
7089 	return (rc);
7090 }
7091 
7092 struct mem_desc {
7093 	unsigned int base;
7094 	unsigned int limit;
7095 	unsigned int idx;
7096 };
7097 
7098 static int
7099 mem_desc_cmp(const void *a, const void *b)
7100 {
7101 	return ((const struct mem_desc *)a)->base -
7102 	       ((const struct mem_desc *)b)->base;
7103 }
7104 
7105 static void
7106 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
7107     unsigned int to)
7108 {
7109 	unsigned int size;
7110 
7111 	if (from == to)
7112 		return;
7113 
7114 	size = to - from + 1;
7115 	if (size == 0)
7116 		return;
7117 
7118 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
7119 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
7120 }
7121 
7122 static int
7123 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
7124 {
7125 	struct adapter *sc = arg1;
7126 	struct sbuf *sb;
7127 	int rc, i, n;
7128 	uint32_t lo, hi, used, alloc;
7129 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
7130 	static const char *region[] = {
7131 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
7132 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
7133 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
7134 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
7135 		"RQUDP region:", "PBL region:", "TXPBL region:",
7136 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
7137 		"On-chip queues:", "TLS keys:",
7138 	};
7139 	struct mem_desc avail[4];
7140 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
7141 	struct mem_desc *md = mem;
7142 
7143 	rc = sysctl_wire_old_buffer(req, 0);
7144 	if (rc != 0)
7145 		return (rc);
7146 
7147 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7148 	if (sb == NULL)
7149 		return (ENOMEM);
7150 
7151 	for (i = 0; i < nitems(mem); i++) {
7152 		mem[i].limit = 0;
7153 		mem[i].idx = i;
7154 	}
7155 
7156 	/* Find and sort the populated memory ranges */
7157 	i = 0;
7158 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
7159 	if (lo & F_EDRAM0_ENABLE) {
7160 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
7161 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
7162 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
7163 		avail[i].idx = 0;
7164 		i++;
7165 	}
7166 	if (lo & F_EDRAM1_ENABLE) {
7167 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
7168 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
7169 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
7170 		avail[i].idx = 1;
7171 		i++;
7172 	}
7173 	if (lo & F_EXT_MEM_ENABLE) {
7174 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
7175 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
7176 		avail[i].limit = avail[i].base +
7177 		    (G_EXT_MEM_SIZE(hi) << 20);
7178 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
7179 		i++;
7180 	}
7181 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
7182 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
7183 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
7184 		avail[i].limit = avail[i].base +
7185 		    (G_EXT_MEM1_SIZE(hi) << 20);
7186 		avail[i].idx = 4;
7187 		i++;
7188 	}
7189 	if (!i)                                    /* no memory available */
7190 		return 0;
7191 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
7192 
7193 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
7194 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
7195 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
7196 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
7197 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
7198 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
7199 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
7200 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
7201 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
7202 
7203 	/* the next few have explicit upper bounds */
7204 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
7205 	md->limit = md->base - 1 +
7206 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
7207 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
7208 	md++;
7209 
7210 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
7211 	md->limit = md->base - 1 +
7212 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
7213 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
7214 	md++;
7215 
7216 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7217 		if (chip_id(sc) <= CHELSIO_T5)
7218 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
7219 		else
7220 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
7221 		md->limit = 0;
7222 	} else {
7223 		md->base = 0;
7224 		md->idx = nitems(region);  /* hide it */
7225 	}
7226 	md++;
7227 
7228 #define ulp_region(reg) \
7229 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
7230 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
7231 
7232 	ulp_region(RX_ISCSI);
7233 	ulp_region(RX_TDDP);
7234 	ulp_region(TX_TPT);
7235 	ulp_region(RX_STAG);
7236 	ulp_region(RX_RQ);
7237 	ulp_region(RX_RQUDP);
7238 	ulp_region(RX_PBL);
7239 	ulp_region(TX_PBL);
7240 #undef ulp_region
7241 
7242 	md->base = 0;
7243 	md->idx = nitems(region);
7244 	if (!is_t4(sc)) {
7245 		uint32_t size = 0;
7246 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
7247 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
7248 
7249 		if (is_t5(sc)) {
7250 			if (sge_ctrl & F_VFIFO_ENABLE)
7251 				size = G_DBVFIFO_SIZE(fifo_size);
7252 		} else
7253 			size = G_T6_DBVFIFO_SIZE(fifo_size);
7254 
7255 		if (size) {
7256 			md->base = G_BASEADDR(t4_read_reg(sc,
7257 			    A_SGE_DBVFIFO_BADDR));
7258 			md->limit = md->base + (size << 2) - 1;
7259 		}
7260 	}
7261 	md++;
7262 
7263 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
7264 	md->limit = 0;
7265 	md++;
7266 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
7267 	md->limit = 0;
7268 	md++;
7269 
7270 	md->base = sc->vres.ocq.start;
7271 	if (sc->vres.ocq.size)
7272 		md->limit = md->base + sc->vres.ocq.size - 1;
7273 	else
7274 		md->idx = nitems(region);  /* hide it */
7275 	md++;
7276 
7277 	md->base = sc->vres.key.start;
7278 	if (sc->vres.key.size)
7279 		md->limit = md->base + sc->vres.key.size - 1;
7280 	else
7281 		md->idx = nitems(region);  /* hide it */
7282 	md++;
7283 
7284 	/* add any address-space holes, there can be up to 3 */
7285 	for (n = 0; n < i - 1; n++)
7286 		if (avail[n].limit < avail[n + 1].base)
7287 			(md++)->base = avail[n].limit;
7288 	if (avail[n].limit)
7289 		(md++)->base = avail[n].limit;
7290 
7291 	n = md - mem;
7292 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
7293 
7294 	for (lo = 0; lo < i; lo++)
7295 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
7296 				avail[lo].limit - 1);
7297 
7298 	sbuf_printf(sb, "\n");
7299 	for (i = 0; i < n; i++) {
7300 		if (mem[i].idx >= nitems(region))
7301 			continue;                        /* skip holes */
7302 		if (!mem[i].limit)
7303 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
7304 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
7305 				mem[i].limit);
7306 	}
7307 
7308 	sbuf_printf(sb, "\n");
7309 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
7310 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
7311 	mem_region_show(sb, "uP RAM:", lo, hi);
7312 
7313 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
7314 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
7315 	mem_region_show(sb, "uP Extmem2:", lo, hi);
7316 
7317 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
7318 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
7319 		   G_PMRXMAXPAGE(lo),
7320 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
7321 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
7322 
7323 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
7324 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
7325 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
7326 		   G_PMTXMAXPAGE(lo),
7327 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
7328 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
7329 	sbuf_printf(sb, "%u p-structs\n",
7330 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
7331 
7332 	for (i = 0; i < 4; i++) {
7333 		if (chip_id(sc) > CHELSIO_T5)
7334 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
7335 		else
7336 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
7337 		if (is_t5(sc)) {
7338 			used = G_T5_USED(lo);
7339 			alloc = G_T5_ALLOC(lo);
7340 		} else {
7341 			used = G_USED(lo);
7342 			alloc = G_ALLOC(lo);
7343 		}
7344 		/* For T6 these are MAC buffer groups */
7345 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
7346 		    i, used, alloc);
7347 	}
7348 	for (i = 0; i < sc->chip_params->nchan; i++) {
7349 		if (chip_id(sc) > CHELSIO_T5)
7350 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
7351 		else
7352 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
7353 		if (is_t5(sc)) {
7354 			used = G_T5_USED(lo);
7355 			alloc = G_T5_ALLOC(lo);
7356 		} else {
7357 			used = G_USED(lo);
7358 			alloc = G_ALLOC(lo);
7359 		}
7360 		/* For T6 these are MAC buffer groups */
7361 		sbuf_printf(sb,
7362 		    "\nLoopback %d using %u pages out of %u allocated",
7363 		    i, used, alloc);
7364 	}
7365 
7366 	rc = sbuf_finish(sb);
7367 	sbuf_delete(sb);
7368 
7369 	return (rc);
7370 }
7371 
7372 static inline void
7373 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
7374 {
7375 	*mask = x | y;
7376 	y = htobe64(y);
7377 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
7378 }
7379 
7380 static int
7381 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
7382 {
7383 	struct adapter *sc = arg1;
7384 	struct sbuf *sb;
7385 	int rc, i;
7386 
7387 	MPASS(chip_id(sc) <= CHELSIO_T5);
7388 
7389 	rc = sysctl_wire_old_buffer(req, 0);
7390 	if (rc != 0)
7391 		return (rc);
7392 
7393 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7394 	if (sb == NULL)
7395 		return (ENOMEM);
7396 
7397 	sbuf_printf(sb,
7398 	    "Idx  Ethernet address     Mask     Vld Ports PF"
7399 	    "  VF              Replication             P0 P1 P2 P3  ML");
7400 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7401 		uint64_t tcamx, tcamy, mask;
7402 		uint32_t cls_lo, cls_hi;
7403 		uint8_t addr[ETHER_ADDR_LEN];
7404 
7405 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
7406 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
7407 		if (tcamx & tcamy)
7408 			continue;
7409 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7410 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7411 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7412 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
7413 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
7414 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
7415 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
7416 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
7417 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
7418 
7419 		if (cls_lo & F_REPLICATE) {
7420 			struct fw_ldst_cmd ldst_cmd;
7421 
7422 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7423 			ldst_cmd.op_to_addrspace =
7424 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7425 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7426 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7427 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7428 			ldst_cmd.u.mps.rplc.fid_idx =
7429 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7430 				V_FW_LDST_CMD_IDX(i));
7431 
7432 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7433 			    "t4mps");
7434 			if (rc)
7435 				break;
7436 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7437 			    sizeof(ldst_cmd), &ldst_cmd);
7438 			end_synchronized_op(sc, 0);
7439 
7440 			if (rc != 0) {
7441 				sbuf_printf(sb, "%36d", rc);
7442 				rc = 0;
7443 			} else {
7444 				sbuf_printf(sb, " %08x %08x %08x %08x",
7445 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7446 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7447 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7448 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7449 			}
7450 		} else
7451 			sbuf_printf(sb, "%36s", "");
7452 
7453 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
7454 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
7455 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
7456 	}
7457 
7458 	if (rc)
7459 		(void) sbuf_finish(sb);
7460 	else
7461 		rc = sbuf_finish(sb);
7462 	sbuf_delete(sb);
7463 
7464 	return (rc);
7465 }
7466 
7467 static int
7468 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
7469 {
7470 	struct adapter *sc = arg1;
7471 	struct sbuf *sb;
7472 	int rc, i;
7473 
7474 	MPASS(chip_id(sc) > CHELSIO_T5);
7475 
7476 	rc = sysctl_wire_old_buffer(req, 0);
7477 	if (rc != 0)
7478 		return (rc);
7479 
7480 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7481 	if (sb == NULL)
7482 		return (ENOMEM);
7483 
7484 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
7485 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
7486 	    "                           Replication"
7487 	    "                                    P0 P1 P2 P3  ML\n");
7488 
7489 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7490 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
7491 		uint16_t ivlan;
7492 		uint64_t tcamx, tcamy, val, mask;
7493 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
7494 		uint8_t addr[ETHER_ADDR_LEN];
7495 
7496 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
7497 		if (i < 256)
7498 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
7499 		else
7500 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
7501 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7502 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7503 		tcamy = G_DMACH(val) << 32;
7504 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7505 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7506 		lookup_type = G_DATALKPTYPE(data2);
7507 		port_num = G_DATAPORTNUM(data2);
7508 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7509 			/* Inner header VNI */
7510 			vniy = ((data2 & F_DATAVIDH2) << 23) |
7511 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7512 			dip_hit = data2 & F_DATADIPHIT;
7513 			vlan_vld = 0;
7514 		} else {
7515 			vniy = 0;
7516 			dip_hit = 0;
7517 			vlan_vld = data2 & F_DATAVIDH2;
7518 			ivlan = G_VIDL(val);
7519 		}
7520 
7521 		ctl |= V_CTLXYBITSEL(1);
7522 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7523 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7524 		tcamx = G_DMACH(val) << 32;
7525 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7526 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7527 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7528 			/* Inner header VNI mask */
7529 			vnix = ((data2 & F_DATAVIDH2) << 23) |
7530 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7531 		} else
7532 			vnix = 0;
7533 
7534 		if (tcamx & tcamy)
7535 			continue;
7536 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7537 
7538 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7539 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7540 
7541 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7542 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7543 			    "%012jx %06x %06x    -    -   %3c"
7544 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
7545 			    addr[1], addr[2], addr[3], addr[4], addr[5],
7546 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
7547 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7548 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7549 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7550 		} else {
7551 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7552 			    "%012jx    -       -   ", i, addr[0], addr[1],
7553 			    addr[2], addr[3], addr[4], addr[5],
7554 			    (uintmax_t)mask);
7555 
7556 			if (vlan_vld)
7557 				sbuf_printf(sb, "%4u   Y     ", ivlan);
7558 			else
7559 				sbuf_printf(sb, "  -    N     ");
7560 
7561 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
7562 			    lookup_type ? 'I' : 'O', port_num,
7563 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7564 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7565 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7566 		}
7567 
7568 
7569 		if (cls_lo & F_T6_REPLICATE) {
7570 			struct fw_ldst_cmd ldst_cmd;
7571 
7572 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7573 			ldst_cmd.op_to_addrspace =
7574 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7575 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7576 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7577 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7578 			ldst_cmd.u.mps.rplc.fid_idx =
7579 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7580 				V_FW_LDST_CMD_IDX(i));
7581 
7582 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7583 			    "t6mps");
7584 			if (rc)
7585 				break;
7586 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7587 			    sizeof(ldst_cmd), &ldst_cmd);
7588 			end_synchronized_op(sc, 0);
7589 
7590 			if (rc != 0) {
7591 				sbuf_printf(sb, "%72d", rc);
7592 				rc = 0;
7593 			} else {
7594 				sbuf_printf(sb, " %08x %08x %08x %08x"
7595 				    " %08x %08x %08x %08x",
7596 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
7597 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
7598 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
7599 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
7600 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7601 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7602 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7603 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7604 			}
7605 		} else
7606 			sbuf_printf(sb, "%72s", "");
7607 
7608 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
7609 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
7610 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
7611 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
7612 	}
7613 
7614 	if (rc)
7615 		(void) sbuf_finish(sb);
7616 	else
7617 		rc = sbuf_finish(sb);
7618 	sbuf_delete(sb);
7619 
7620 	return (rc);
7621 }
7622 
7623 static int
7624 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
7625 {
7626 	struct adapter *sc = arg1;
7627 	struct sbuf *sb;
7628 	int rc;
7629 	uint16_t mtus[NMTUS];
7630 
7631 	rc = sysctl_wire_old_buffer(req, 0);
7632 	if (rc != 0)
7633 		return (rc);
7634 
7635 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7636 	if (sb == NULL)
7637 		return (ENOMEM);
7638 
7639 	t4_read_mtu_tbl(sc, mtus, NULL);
7640 
7641 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
7642 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
7643 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
7644 	    mtus[14], mtus[15]);
7645 
7646 	rc = sbuf_finish(sb);
7647 	sbuf_delete(sb);
7648 
7649 	return (rc);
7650 }
7651 
7652 static int
7653 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
7654 {
7655 	struct adapter *sc = arg1;
7656 	struct sbuf *sb;
7657 	int rc, i;
7658 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
7659 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
7660 	static const char *tx_stats[MAX_PM_NSTATS] = {
7661 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
7662 		"Tx FIFO wait", NULL, "Tx latency"
7663 	};
7664 	static const char *rx_stats[MAX_PM_NSTATS] = {
7665 		"Read:", "Write bypass:", "Write mem:", "Flush:",
7666 		"Rx FIFO wait", NULL, "Rx latency"
7667 	};
7668 
7669 	rc = sysctl_wire_old_buffer(req, 0);
7670 	if (rc != 0)
7671 		return (rc);
7672 
7673 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7674 	if (sb == NULL)
7675 		return (ENOMEM);
7676 
7677 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
7678 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
7679 
7680 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
7681 	for (i = 0; i < 4; i++) {
7682 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7683 		    tx_cyc[i]);
7684 	}
7685 
7686 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
7687 	for (i = 0; i < 4; i++) {
7688 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7689 		    rx_cyc[i]);
7690 	}
7691 
7692 	if (chip_id(sc) > CHELSIO_T5) {
7693 		sbuf_printf(sb,
7694 		    "\n              Total wait      Total occupancy");
7695 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7696 		    tx_cyc[i]);
7697 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7698 		    rx_cyc[i]);
7699 
7700 		i += 2;
7701 		MPASS(i < nitems(tx_stats));
7702 
7703 		sbuf_printf(sb,
7704 		    "\n                   Reads           Total wait");
7705 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7706 		    tx_cyc[i]);
7707 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7708 		    rx_cyc[i]);
7709 	}
7710 
7711 	rc = sbuf_finish(sb);
7712 	sbuf_delete(sb);
7713 
7714 	return (rc);
7715 }
7716 
7717 static int
7718 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
7719 {
7720 	struct adapter *sc = arg1;
7721 	struct sbuf *sb;
7722 	int rc;
7723 	struct tp_rdma_stats stats;
7724 
7725 	rc = sysctl_wire_old_buffer(req, 0);
7726 	if (rc != 0)
7727 		return (rc);
7728 
7729 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7730 	if (sb == NULL)
7731 		return (ENOMEM);
7732 
7733 	mtx_lock(&sc->reg_lock);
7734 	t4_tp_get_rdma_stats(sc, &stats, 0);
7735 	mtx_unlock(&sc->reg_lock);
7736 
7737 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
7738 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
7739 
7740 	rc = sbuf_finish(sb);
7741 	sbuf_delete(sb);
7742 
7743 	return (rc);
7744 }
7745 
7746 static int
7747 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
7748 {
7749 	struct adapter *sc = arg1;
7750 	struct sbuf *sb;
7751 	int rc;
7752 	struct tp_tcp_stats v4, v6;
7753 
7754 	rc = sysctl_wire_old_buffer(req, 0);
7755 	if (rc != 0)
7756 		return (rc);
7757 
7758 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7759 	if (sb == NULL)
7760 		return (ENOMEM);
7761 
7762 	mtx_lock(&sc->reg_lock);
7763 	t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
7764 	mtx_unlock(&sc->reg_lock);
7765 
7766 	sbuf_printf(sb,
7767 	    "                                IP                 IPv6\n");
7768 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
7769 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
7770 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
7771 	    v4.tcp_in_segs, v6.tcp_in_segs);
7772 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
7773 	    v4.tcp_out_segs, v6.tcp_out_segs);
7774 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
7775 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
7776 
7777 	rc = sbuf_finish(sb);
7778 	sbuf_delete(sb);
7779 
7780 	return (rc);
7781 }
7782 
7783 static int
7784 sysctl_tids(SYSCTL_HANDLER_ARGS)
7785 {
7786 	struct adapter *sc = arg1;
7787 	struct sbuf *sb;
7788 	int rc;
7789 	struct tid_info *t = &sc->tids;
7790 
7791 	rc = sysctl_wire_old_buffer(req, 0);
7792 	if (rc != 0)
7793 		return (rc);
7794 
7795 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7796 	if (sb == NULL)
7797 		return (ENOMEM);
7798 
7799 	if (t->natids) {
7800 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
7801 		    t->atids_in_use);
7802 	}
7803 
7804 	if (t->ntids) {
7805 		sbuf_printf(sb, "TID range: ");
7806 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7807 			uint32_t b, hb;
7808 
7809 			if (chip_id(sc) <= CHELSIO_T5) {
7810 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
7811 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
7812 			} else {
7813 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
7814 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
7815 			}
7816 
7817 			if (b)
7818 				sbuf_printf(sb, "0-%u, ", b - 1);
7819 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
7820 		} else
7821 			sbuf_printf(sb, "0-%u", t->ntids - 1);
7822 		sbuf_printf(sb, ", in use: %u\n",
7823 		    atomic_load_acq_int(&t->tids_in_use));
7824 	}
7825 
7826 	if (t->nstids) {
7827 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
7828 		    t->stid_base + t->nstids - 1, t->stids_in_use);
7829 	}
7830 
7831 	if (t->nftids) {
7832 		sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
7833 		    t->ftid_base + t->nftids - 1);
7834 	}
7835 
7836 	if (t->netids) {
7837 		sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
7838 		    t->etid_base + t->netids - 1);
7839 	}
7840 
7841 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
7842 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
7843 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
7844 
7845 	rc = sbuf_finish(sb);
7846 	sbuf_delete(sb);
7847 
7848 	return (rc);
7849 }
7850 
7851 static int
7852 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
7853 {
7854 	struct adapter *sc = arg1;
7855 	struct sbuf *sb;
7856 	int rc;
7857 	struct tp_err_stats stats;
7858 
7859 	rc = sysctl_wire_old_buffer(req, 0);
7860 	if (rc != 0)
7861 		return (rc);
7862 
7863 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7864 	if (sb == NULL)
7865 		return (ENOMEM);
7866 
7867 	mtx_lock(&sc->reg_lock);
7868 	t4_tp_get_err_stats(sc, &stats, 0);
7869 	mtx_unlock(&sc->reg_lock);
7870 
7871 	if (sc->chip_params->nchan > 2) {
7872 		sbuf_printf(sb, "                 channel 0  channel 1"
7873 		    "  channel 2  channel 3\n");
7874 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
7875 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
7876 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
7877 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
7878 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
7879 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
7880 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
7881 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
7882 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
7883 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
7884 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
7885 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
7886 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
7887 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
7888 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
7889 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
7890 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
7891 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
7892 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
7893 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
7894 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
7895 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
7896 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
7897 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
7898 	} else {
7899 		sbuf_printf(sb, "                 channel 0  channel 1\n");
7900 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
7901 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
7902 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
7903 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
7904 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
7905 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
7906 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
7907 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
7908 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
7909 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
7910 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
7911 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
7912 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
7913 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
7914 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
7915 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
7916 	}
7917 
7918 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
7919 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
7920 
7921 	rc = sbuf_finish(sb);
7922 	sbuf_delete(sb);
7923 
7924 	return (rc);
7925 }
7926 
7927 static int
7928 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
7929 {
7930 	struct adapter *sc = arg1;
7931 	struct tp_params *tpp = &sc->params.tp;
7932 	u_int mask;
7933 	int rc;
7934 
7935 	mask = tpp->la_mask >> 16;
7936 	rc = sysctl_handle_int(oidp, &mask, 0, req);
7937 	if (rc != 0 || req->newptr == NULL)
7938 		return (rc);
7939 	if (mask > 0xffff)
7940 		return (EINVAL);
7941 	tpp->la_mask = mask << 16;
7942 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
7943 
7944 	return (0);
7945 }
7946 
7947 struct field_desc {
7948 	const char *name;
7949 	u_int start;
7950 	u_int width;
7951 };
7952 
7953 static void
7954 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
7955 {
7956 	char buf[32];
7957 	int line_size = 0;
7958 
7959 	while (f->name) {
7960 		uint64_t mask = (1ULL << f->width) - 1;
7961 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
7962 		    ((uintmax_t)v >> f->start) & mask);
7963 
7964 		if (line_size + len >= 79) {
7965 			line_size = 8;
7966 			sbuf_printf(sb, "\n        ");
7967 		}
7968 		sbuf_printf(sb, "%s ", buf);
7969 		line_size += len + 1;
7970 		f++;
7971 	}
7972 	sbuf_printf(sb, "\n");
7973 }
7974 
7975 static const struct field_desc tp_la0[] = {
7976 	{ "RcfOpCodeOut", 60, 4 },
7977 	{ "State", 56, 4 },
7978 	{ "WcfState", 52, 4 },
7979 	{ "RcfOpcSrcOut", 50, 2 },
7980 	{ "CRxError", 49, 1 },
7981 	{ "ERxError", 48, 1 },
7982 	{ "SanityFailed", 47, 1 },
7983 	{ "SpuriousMsg", 46, 1 },
7984 	{ "FlushInputMsg", 45, 1 },
7985 	{ "FlushInputCpl", 44, 1 },
7986 	{ "RssUpBit", 43, 1 },
7987 	{ "RssFilterHit", 42, 1 },
7988 	{ "Tid", 32, 10 },
7989 	{ "InitTcb", 31, 1 },
7990 	{ "LineNumber", 24, 7 },
7991 	{ "Emsg", 23, 1 },
7992 	{ "EdataOut", 22, 1 },
7993 	{ "Cmsg", 21, 1 },
7994 	{ "CdataOut", 20, 1 },
7995 	{ "EreadPdu", 19, 1 },
7996 	{ "CreadPdu", 18, 1 },
7997 	{ "TunnelPkt", 17, 1 },
7998 	{ "RcfPeerFin", 16, 1 },
7999 	{ "RcfReasonOut", 12, 4 },
8000 	{ "TxCchannel", 10, 2 },
8001 	{ "RcfTxChannel", 8, 2 },
8002 	{ "RxEchannel", 6, 2 },
8003 	{ "RcfRxChannel", 5, 1 },
8004 	{ "RcfDataOutSrdy", 4, 1 },
8005 	{ "RxDvld", 3, 1 },
8006 	{ "RxOoDvld", 2, 1 },
8007 	{ "RxCongestion", 1, 1 },
8008 	{ "TxCongestion", 0, 1 },
8009 	{ NULL }
8010 };
8011 
8012 static const struct field_desc tp_la1[] = {
8013 	{ "CplCmdIn", 56, 8 },
8014 	{ "CplCmdOut", 48, 8 },
8015 	{ "ESynOut", 47, 1 },
8016 	{ "EAckOut", 46, 1 },
8017 	{ "EFinOut", 45, 1 },
8018 	{ "ERstOut", 44, 1 },
8019 	{ "SynIn", 43, 1 },
8020 	{ "AckIn", 42, 1 },
8021 	{ "FinIn", 41, 1 },
8022 	{ "RstIn", 40, 1 },
8023 	{ "DataIn", 39, 1 },
8024 	{ "DataInVld", 38, 1 },
8025 	{ "PadIn", 37, 1 },
8026 	{ "RxBufEmpty", 36, 1 },
8027 	{ "RxDdp", 35, 1 },
8028 	{ "RxFbCongestion", 34, 1 },
8029 	{ "TxFbCongestion", 33, 1 },
8030 	{ "TxPktSumSrdy", 32, 1 },
8031 	{ "RcfUlpType", 28, 4 },
8032 	{ "Eread", 27, 1 },
8033 	{ "Ebypass", 26, 1 },
8034 	{ "Esave", 25, 1 },
8035 	{ "Static0", 24, 1 },
8036 	{ "Cread", 23, 1 },
8037 	{ "Cbypass", 22, 1 },
8038 	{ "Csave", 21, 1 },
8039 	{ "CPktOut", 20, 1 },
8040 	{ "RxPagePoolFull", 18, 2 },
8041 	{ "RxLpbkPkt", 17, 1 },
8042 	{ "TxLpbkPkt", 16, 1 },
8043 	{ "RxVfValid", 15, 1 },
8044 	{ "SynLearned", 14, 1 },
8045 	{ "SetDelEntry", 13, 1 },
8046 	{ "SetInvEntry", 12, 1 },
8047 	{ "CpcmdDvld", 11, 1 },
8048 	{ "CpcmdSave", 10, 1 },
8049 	{ "RxPstructsFull", 8, 2 },
8050 	{ "EpcmdDvld", 7, 1 },
8051 	{ "EpcmdFlush", 6, 1 },
8052 	{ "EpcmdTrimPrefix", 5, 1 },
8053 	{ "EpcmdTrimPostfix", 4, 1 },
8054 	{ "ERssIp4Pkt", 3, 1 },
8055 	{ "ERssIp6Pkt", 2, 1 },
8056 	{ "ERssTcpUdpPkt", 1, 1 },
8057 	{ "ERssFceFipPkt", 0, 1 },
8058 	{ NULL }
8059 };
8060 
8061 static const struct field_desc tp_la2[] = {
8062 	{ "CplCmdIn", 56, 8 },
8063 	{ "MpsVfVld", 55, 1 },
8064 	{ "MpsPf", 52, 3 },
8065 	{ "MpsVf", 44, 8 },
8066 	{ "SynIn", 43, 1 },
8067 	{ "AckIn", 42, 1 },
8068 	{ "FinIn", 41, 1 },
8069 	{ "RstIn", 40, 1 },
8070 	{ "DataIn", 39, 1 },
8071 	{ "DataInVld", 38, 1 },
8072 	{ "PadIn", 37, 1 },
8073 	{ "RxBufEmpty", 36, 1 },
8074 	{ "RxDdp", 35, 1 },
8075 	{ "RxFbCongestion", 34, 1 },
8076 	{ "TxFbCongestion", 33, 1 },
8077 	{ "TxPktSumSrdy", 32, 1 },
8078 	{ "RcfUlpType", 28, 4 },
8079 	{ "Eread", 27, 1 },
8080 	{ "Ebypass", 26, 1 },
8081 	{ "Esave", 25, 1 },
8082 	{ "Static0", 24, 1 },
8083 	{ "Cread", 23, 1 },
8084 	{ "Cbypass", 22, 1 },
8085 	{ "Csave", 21, 1 },
8086 	{ "CPktOut", 20, 1 },
8087 	{ "RxPagePoolFull", 18, 2 },
8088 	{ "RxLpbkPkt", 17, 1 },
8089 	{ "TxLpbkPkt", 16, 1 },
8090 	{ "RxVfValid", 15, 1 },
8091 	{ "SynLearned", 14, 1 },
8092 	{ "SetDelEntry", 13, 1 },
8093 	{ "SetInvEntry", 12, 1 },
8094 	{ "CpcmdDvld", 11, 1 },
8095 	{ "CpcmdSave", 10, 1 },
8096 	{ "RxPstructsFull", 8, 2 },
8097 	{ "EpcmdDvld", 7, 1 },
8098 	{ "EpcmdFlush", 6, 1 },
8099 	{ "EpcmdTrimPrefix", 5, 1 },
8100 	{ "EpcmdTrimPostfix", 4, 1 },
8101 	{ "ERssIp4Pkt", 3, 1 },
8102 	{ "ERssIp6Pkt", 2, 1 },
8103 	{ "ERssTcpUdpPkt", 1, 1 },
8104 	{ "ERssFceFipPkt", 0, 1 },
8105 	{ NULL }
8106 };
8107 
8108 static void
8109 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
8110 {
8111 
8112 	field_desc_show(sb, *p, tp_la0);
8113 }
8114 
8115 static void
8116 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
8117 {
8118 
8119 	if (idx)
8120 		sbuf_printf(sb, "\n");
8121 	field_desc_show(sb, p[0], tp_la0);
8122 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8123 		field_desc_show(sb, p[1], tp_la0);
8124 }
8125 
8126 static void
8127 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
8128 {
8129 
8130 	if (idx)
8131 		sbuf_printf(sb, "\n");
8132 	field_desc_show(sb, p[0], tp_la0);
8133 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8134 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
8135 }
8136 
8137 static int
8138 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
8139 {
8140 	struct adapter *sc = arg1;
8141 	struct sbuf *sb;
8142 	uint64_t *buf, *p;
8143 	int rc;
8144 	u_int i, inc;
8145 	void (*show_func)(struct sbuf *, uint64_t *, int);
8146 
8147 	rc = sysctl_wire_old_buffer(req, 0);
8148 	if (rc != 0)
8149 		return (rc);
8150 
8151 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8152 	if (sb == NULL)
8153 		return (ENOMEM);
8154 
8155 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
8156 
8157 	t4_tp_read_la(sc, buf, NULL);
8158 	p = buf;
8159 
8160 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
8161 	case 2:
8162 		inc = 2;
8163 		show_func = tp_la_show2;
8164 		break;
8165 	case 3:
8166 		inc = 2;
8167 		show_func = tp_la_show3;
8168 		break;
8169 	default:
8170 		inc = 1;
8171 		show_func = tp_la_show;
8172 	}
8173 
8174 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
8175 		(*show_func)(sb, p, i);
8176 
8177 	rc = sbuf_finish(sb);
8178 	sbuf_delete(sb);
8179 	free(buf, M_CXGBE);
8180 	return (rc);
8181 }
8182 
8183 static int
8184 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
8185 {
8186 	struct adapter *sc = arg1;
8187 	struct sbuf *sb;
8188 	int rc;
8189 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
8190 
8191 	rc = sysctl_wire_old_buffer(req, 0);
8192 	if (rc != 0)
8193 		return (rc);
8194 
8195 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8196 	if (sb == NULL)
8197 		return (ENOMEM);
8198 
8199 	t4_get_chan_txrate(sc, nrate, orate);
8200 
8201 	if (sc->chip_params->nchan > 2) {
8202 		sbuf_printf(sb, "              channel 0   channel 1"
8203 		    "   channel 2   channel 3\n");
8204 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
8205 		    nrate[0], nrate[1], nrate[2], nrate[3]);
8206 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
8207 		    orate[0], orate[1], orate[2], orate[3]);
8208 	} else {
8209 		sbuf_printf(sb, "              channel 0   channel 1\n");
8210 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
8211 		    nrate[0], nrate[1]);
8212 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
8213 		    orate[0], orate[1]);
8214 	}
8215 
8216 	rc = sbuf_finish(sb);
8217 	sbuf_delete(sb);
8218 
8219 	return (rc);
8220 }
8221 
8222 static int
8223 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
8224 {
8225 	struct adapter *sc = arg1;
8226 	struct sbuf *sb;
8227 	uint32_t *buf, *p;
8228 	int rc, i;
8229 
8230 	rc = sysctl_wire_old_buffer(req, 0);
8231 	if (rc != 0)
8232 		return (rc);
8233 
8234 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8235 	if (sb == NULL)
8236 		return (ENOMEM);
8237 
8238 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
8239 	    M_ZERO | M_WAITOK);
8240 
8241 	t4_ulprx_read_la(sc, buf);
8242 	p = buf;
8243 
8244 	sbuf_printf(sb, "      Pcmd        Type   Message"
8245 	    "                Data");
8246 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
8247 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
8248 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
8249 	}
8250 
8251 	rc = sbuf_finish(sb);
8252 	sbuf_delete(sb);
8253 	free(buf, M_CXGBE);
8254 	return (rc);
8255 }
8256 
8257 static int
8258 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
8259 {
8260 	struct adapter *sc = arg1;
8261 	struct sbuf *sb;
8262 	int rc, v;
8263 
8264 	MPASS(chip_id(sc) >= CHELSIO_T5);
8265 
8266 	rc = sysctl_wire_old_buffer(req, 0);
8267 	if (rc != 0)
8268 		return (rc);
8269 
8270 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8271 	if (sb == NULL)
8272 		return (ENOMEM);
8273 
8274 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
8275 	if (G_STATSOURCE_T5(v) == 7) {
8276 		int mode;
8277 
8278 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
8279 		if (mode == 0) {
8280 			sbuf_printf(sb, "total %d, incomplete %d",
8281 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8282 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8283 		} else if (mode == 1) {
8284 			sbuf_printf(sb, "total %d, data overflow %d",
8285 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8286 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8287 		} else {
8288 			sbuf_printf(sb, "unknown mode %d", mode);
8289 		}
8290 	}
8291 	rc = sbuf_finish(sb);
8292 	sbuf_delete(sb);
8293 
8294 	return (rc);
8295 }
8296 
8297 static int
8298 sysctl_tc_params(SYSCTL_HANDLER_ARGS)
8299 {
8300 	struct adapter *sc = arg1;
8301 	struct tx_cl_rl_params tc;
8302 	struct sbuf *sb;
8303 	int i, rc, port_id, mbps, gbps;
8304 
8305 	rc = sysctl_wire_old_buffer(req, 0);
8306 	if (rc != 0)
8307 		return (rc);
8308 
8309 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8310 	if (sb == NULL)
8311 		return (ENOMEM);
8312 
8313 	port_id = arg2 >> 16;
8314 	MPASS(port_id < sc->params.nports);
8315 	MPASS(sc->port[port_id] != NULL);
8316 	i = arg2 & 0xffff;
8317 	MPASS(i < sc->chip_params->nsched_cls);
8318 
8319 	mtx_lock(&sc->tc_lock);
8320 	tc = sc->port[port_id]->sched_params->cl_rl[i];
8321 	mtx_unlock(&sc->tc_lock);
8322 
8323 	if (tc.flags & TX_CLRL_ERROR) {
8324 		sbuf_printf(sb, "error");
8325 		goto done;
8326 	}
8327 
8328 	if (tc.ratemode == SCHED_CLASS_RATEMODE_REL) {
8329 		/* XXX: top speed or actual link speed? */
8330 		gbps = port_top_speed(sc->port[port_id]);
8331 		sbuf_printf(sb, " %u%% of %uGbps", tc.maxrate, gbps);
8332 	} else if (tc.ratemode == SCHED_CLASS_RATEMODE_ABS) {
8333 		switch (tc.rateunit) {
8334 		case SCHED_CLASS_RATEUNIT_BITS:
8335 			mbps = tc.maxrate / 1000;
8336 			gbps = tc.maxrate / 1000000;
8337 			if (tc.maxrate == gbps * 1000000)
8338 				sbuf_printf(sb, " %uGbps", gbps);
8339 			else if (tc.maxrate == mbps * 1000)
8340 				sbuf_printf(sb, " %uMbps", mbps);
8341 			else
8342 				sbuf_printf(sb, " %uKbps", tc.maxrate);
8343 			break;
8344 		case SCHED_CLASS_RATEUNIT_PKTS:
8345 			sbuf_printf(sb, " %upps", tc.maxrate);
8346 			break;
8347 		default:
8348 			rc = ENXIO;
8349 			goto done;
8350 		}
8351 	}
8352 
8353 	switch (tc.mode) {
8354 	case SCHED_CLASS_MODE_CLASS:
8355 		sbuf_printf(sb, " aggregate");
8356 		break;
8357 	case SCHED_CLASS_MODE_FLOW:
8358 		sbuf_printf(sb, " per-flow");
8359 		break;
8360 	default:
8361 		rc = ENXIO;
8362 		goto done;
8363 	}
8364 
8365 done:
8366 	if (rc == 0)
8367 		rc = sbuf_finish(sb);
8368 	sbuf_delete(sb);
8369 
8370 	return (rc);
8371 }
8372 #endif
8373 
8374 #ifdef TCP_OFFLOAD
8375 static int
8376 sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS)
8377 {
8378 	struct adapter *sc = arg1;
8379 	int *old_ports, *new_ports;
8380 	int i, new_count, rc;
8381 
8382 	if (req->newptr == NULL && req->oldptr == NULL)
8383 		return (SYSCTL_OUT(req, NULL, imax(sc->tt.num_tls_rx_ports, 1) *
8384 		    sizeof(sc->tt.tls_rx_ports[0])));
8385 
8386 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tlsrx");
8387 	if (rc)
8388 		return (rc);
8389 
8390 	if (sc->tt.num_tls_rx_ports == 0) {
8391 		i = -1;
8392 		rc = SYSCTL_OUT(req, &i, sizeof(i));
8393 	} else
8394 		rc = SYSCTL_OUT(req, sc->tt.tls_rx_ports,
8395 		    sc->tt.num_tls_rx_ports * sizeof(sc->tt.tls_rx_ports[0]));
8396 	if (rc == 0 && req->newptr != NULL) {
8397 		new_count = req->newlen / sizeof(new_ports[0]);
8398 		new_ports = malloc(new_count * sizeof(new_ports[0]), M_CXGBE,
8399 		    M_WAITOK);
8400 		rc = SYSCTL_IN(req, new_ports, new_count *
8401 		    sizeof(new_ports[0]));
8402 		if (rc)
8403 			goto err;
8404 
8405 		/* Allow setting to a single '-1' to clear the list. */
8406 		if (new_count == 1 && new_ports[0] == -1) {
8407 			ADAPTER_LOCK(sc);
8408 			old_ports = sc->tt.tls_rx_ports;
8409 			sc->tt.tls_rx_ports = NULL;
8410 			sc->tt.num_tls_rx_ports = 0;
8411 			ADAPTER_UNLOCK(sc);
8412 			free(old_ports, M_CXGBE);
8413 		} else {
8414 			for (i = 0; i < new_count; i++) {
8415 				if (new_ports[i] < 1 ||
8416 				    new_ports[i] > IPPORT_MAX) {
8417 					rc = EINVAL;
8418 					goto err;
8419 				}
8420 			}
8421 
8422 			ADAPTER_LOCK(sc);
8423 			old_ports = sc->tt.tls_rx_ports;
8424 			sc->tt.tls_rx_ports = new_ports;
8425 			sc->tt.num_tls_rx_ports = new_count;
8426 			ADAPTER_UNLOCK(sc);
8427 			free(old_ports, M_CXGBE);
8428 			new_ports = NULL;
8429 		}
8430 	err:
8431 		free(new_ports, M_CXGBE);
8432 	}
8433 	end_synchronized_op(sc, 0);
8434 	return (rc);
8435 }
8436 
8437 static void
8438 unit_conv(char *buf, size_t len, u_int val, u_int factor)
8439 {
8440 	u_int rem = val % factor;
8441 
8442 	if (rem == 0)
8443 		snprintf(buf, len, "%u", val / factor);
8444 	else {
8445 		while (rem % 10 == 0)
8446 			rem /= 10;
8447 		snprintf(buf, len, "%u.%u", val / factor, rem);
8448 	}
8449 }
8450 
8451 static int
8452 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
8453 {
8454 	struct adapter *sc = arg1;
8455 	char buf[16];
8456 	u_int res, re;
8457 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8458 
8459 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8460 	switch (arg2) {
8461 	case 0:
8462 		/* timer_tick */
8463 		re = G_TIMERRESOLUTION(res);
8464 		break;
8465 	case 1:
8466 		/* TCP timestamp tick */
8467 		re = G_TIMESTAMPRESOLUTION(res);
8468 		break;
8469 	case 2:
8470 		/* DACK tick */
8471 		re = G_DELAYEDACKRESOLUTION(res);
8472 		break;
8473 	default:
8474 		return (EDOOFUS);
8475 	}
8476 
8477 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
8478 
8479 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
8480 }
8481 
8482 static int
8483 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
8484 {
8485 	struct adapter *sc = arg1;
8486 	u_int res, dack_re, v;
8487 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8488 
8489 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8490 	dack_re = G_DELAYEDACKRESOLUTION(res);
8491 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
8492 
8493 	return (sysctl_handle_int(oidp, &v, 0, req));
8494 }
8495 
8496 static int
8497 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
8498 {
8499 	struct adapter *sc = arg1;
8500 	int reg = arg2;
8501 	u_int tre;
8502 	u_long tp_tick_us, v;
8503 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8504 
8505 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
8506 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
8507 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
8508 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
8509 
8510 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
8511 	tp_tick_us = (cclk_ps << tre) / 1000000;
8512 
8513 	if (reg == A_TP_INIT_SRTT)
8514 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
8515 	else
8516 		v = tp_tick_us * t4_read_reg(sc, reg);
8517 
8518 	return (sysctl_handle_long(oidp, &v, 0, req));
8519 }
8520 
8521 /*
8522  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
8523  * passed to this function.
8524  */
8525 static int
8526 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
8527 {
8528 	struct adapter *sc = arg1;
8529 	int idx = arg2;
8530 	u_int v;
8531 
8532 	MPASS(idx >= 0 && idx <= 24);
8533 
8534 	v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
8535 
8536 	return (sysctl_handle_int(oidp, &v, 0, req));
8537 }
8538 
8539 static int
8540 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
8541 {
8542 	struct adapter *sc = arg1;
8543 	int idx = arg2;
8544 	u_int shift, v, r;
8545 
8546 	MPASS(idx >= 0 && idx < 16);
8547 
8548 	r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
8549 	shift = (idx & 3) << 3;
8550 	v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
8551 
8552 	return (sysctl_handle_int(oidp, &v, 0, req));
8553 }
8554 
8555 static int
8556 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
8557 {
8558 	struct vi_info *vi = arg1;
8559 	struct adapter *sc = vi->pi->adapter;
8560 	int idx, rc, i;
8561 	struct sge_ofld_rxq *ofld_rxq;
8562 	uint8_t v;
8563 
8564 	idx = vi->ofld_tmr_idx;
8565 
8566 	rc = sysctl_handle_int(oidp, &idx, 0, req);
8567 	if (rc != 0 || req->newptr == NULL)
8568 		return (rc);
8569 
8570 	if (idx < 0 || idx >= SGE_NTIMERS)
8571 		return (EINVAL);
8572 
8573 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8574 	    "t4otmr");
8575 	if (rc)
8576 		return (rc);
8577 
8578 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
8579 	for_each_ofld_rxq(vi, i, ofld_rxq) {
8580 #ifdef atomic_store_rel_8
8581 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
8582 #else
8583 		ofld_rxq->iq.intr_params = v;
8584 #endif
8585 	}
8586 	vi->ofld_tmr_idx = idx;
8587 
8588 	end_synchronized_op(sc, LOCK_HELD);
8589 	return (0);
8590 }
8591 
8592 static int
8593 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
8594 {
8595 	struct vi_info *vi = arg1;
8596 	struct adapter *sc = vi->pi->adapter;
8597 	int idx, rc;
8598 
8599 	idx = vi->ofld_pktc_idx;
8600 
8601 	rc = sysctl_handle_int(oidp, &idx, 0, req);
8602 	if (rc != 0 || req->newptr == NULL)
8603 		return (rc);
8604 
8605 	if (idx < -1 || idx >= SGE_NCOUNTERS)
8606 		return (EINVAL);
8607 
8608 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
8609 	    "t4opktc");
8610 	if (rc)
8611 		return (rc);
8612 
8613 	if (vi->flags & VI_INIT_DONE)
8614 		rc = EBUSY; /* cannot be changed once the queues are created */
8615 	else
8616 		vi->ofld_pktc_idx = idx;
8617 
8618 	end_synchronized_op(sc, LOCK_HELD);
8619 	return (rc);
8620 }
8621 #endif
8622 
8623 static int
8624 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
8625 {
8626 	int rc;
8627 
8628 	if (cntxt->cid > M_CTXTQID)
8629 		return (EINVAL);
8630 
8631 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
8632 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
8633 		return (EINVAL);
8634 
8635 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
8636 	if (rc)
8637 		return (rc);
8638 
8639 	if (sc->flags & FW_OK) {
8640 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
8641 		    &cntxt->data[0]);
8642 		if (rc == 0)
8643 			goto done;
8644 	}
8645 
8646 	/*
8647 	 * Read via firmware failed or wasn't even attempted.  Read directly via
8648 	 * the backdoor.
8649 	 */
8650 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
8651 done:
8652 	end_synchronized_op(sc, 0);
8653 	return (rc);
8654 }
8655 
8656 static int
8657 load_fw(struct adapter *sc, struct t4_data *fw)
8658 {
8659 	int rc;
8660 	uint8_t *fw_data;
8661 
8662 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
8663 	if (rc)
8664 		return (rc);
8665 
8666 	/*
8667 	 * The firmware, with the sole exception of the memory parity error
8668 	 * handler, runs from memory and not flash.  It is almost always safe to
8669 	 * install a new firmware on a running system.  Just set bit 1 in
8670 	 * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
8671 	 */
8672 	if (sc->flags & FULL_INIT_DONE &&
8673 	    (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
8674 		rc = EBUSY;
8675 		goto done;
8676 	}
8677 
8678 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
8679 	if (fw_data == NULL) {
8680 		rc = ENOMEM;
8681 		goto done;
8682 	}
8683 
8684 	rc = copyin(fw->data, fw_data, fw->len);
8685 	if (rc == 0)
8686 		rc = -t4_load_fw(sc, fw_data, fw->len);
8687 
8688 	free(fw_data, M_CXGBE);
8689 done:
8690 	end_synchronized_op(sc, 0);
8691 	return (rc);
8692 }
8693 
8694 static int
8695 load_cfg(struct adapter *sc, struct t4_data *cfg)
8696 {
8697 	int rc;
8698 	uint8_t *cfg_data = NULL;
8699 
8700 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
8701 	if (rc)
8702 		return (rc);
8703 
8704 	if (cfg->len == 0) {
8705 		/* clear */
8706 		rc = -t4_load_cfg(sc, NULL, 0);
8707 		goto done;
8708 	}
8709 
8710 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
8711 	if (cfg_data == NULL) {
8712 		rc = ENOMEM;
8713 		goto done;
8714 	}
8715 
8716 	rc = copyin(cfg->data, cfg_data, cfg->len);
8717 	if (rc == 0)
8718 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
8719 
8720 	free(cfg_data, M_CXGBE);
8721 done:
8722 	end_synchronized_op(sc, 0);
8723 	return (rc);
8724 }
8725 
8726 static int
8727 load_boot(struct adapter *sc, struct t4_bootrom *br)
8728 {
8729 	int rc;
8730 	uint8_t *br_data = NULL;
8731 	u_int offset;
8732 
8733 	if (br->len > 1024 * 1024)
8734 		return (EFBIG);
8735 
8736 	if (br->pf_offset == 0) {
8737 		/* pfidx */
8738 		if (br->pfidx_addr > 7)
8739 			return (EINVAL);
8740 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
8741 		    A_PCIE_PF_EXPROM_OFST)));
8742 	} else if (br->pf_offset == 1) {
8743 		/* offset */
8744 		offset = G_OFFSET(br->pfidx_addr);
8745 	} else {
8746 		return (EINVAL);
8747 	}
8748 
8749 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
8750 	if (rc)
8751 		return (rc);
8752 
8753 	if (br->len == 0) {
8754 		/* clear */
8755 		rc = -t4_load_boot(sc, NULL, offset, 0);
8756 		goto done;
8757 	}
8758 
8759 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
8760 	if (br_data == NULL) {
8761 		rc = ENOMEM;
8762 		goto done;
8763 	}
8764 
8765 	rc = copyin(br->data, br_data, br->len);
8766 	if (rc == 0)
8767 		rc = -t4_load_boot(sc, br_data, offset, br->len);
8768 
8769 	free(br_data, M_CXGBE);
8770 done:
8771 	end_synchronized_op(sc, 0);
8772 	return (rc);
8773 }
8774 
8775 static int
8776 load_bootcfg(struct adapter *sc, struct t4_data *bc)
8777 {
8778 	int rc;
8779 	uint8_t *bc_data = NULL;
8780 
8781 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
8782 	if (rc)
8783 		return (rc);
8784 
8785 	if (bc->len == 0) {
8786 		/* clear */
8787 		rc = -t4_load_bootcfg(sc, NULL, 0);
8788 		goto done;
8789 	}
8790 
8791 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
8792 	if (bc_data == NULL) {
8793 		rc = ENOMEM;
8794 		goto done;
8795 	}
8796 
8797 	rc = copyin(bc->data, bc_data, bc->len);
8798 	if (rc == 0)
8799 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
8800 
8801 	free(bc_data, M_CXGBE);
8802 done:
8803 	end_synchronized_op(sc, 0);
8804 	return (rc);
8805 }
8806 
8807 static int
8808 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
8809 {
8810 	int rc;
8811 	struct cudbg_init *cudbg;
8812 	void *handle, *buf;
8813 
8814 	/* buf is large, don't block if no memory is available */
8815 	buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
8816 	if (buf == NULL)
8817 		return (ENOMEM);
8818 
8819 	handle = cudbg_alloc_handle();
8820 	if (handle == NULL) {
8821 		rc = ENOMEM;
8822 		goto done;
8823 	}
8824 
8825 	cudbg = cudbg_get_init(handle);
8826 	cudbg->adap = sc;
8827 	cudbg->print = (cudbg_print_cb)printf;
8828 
8829 #ifndef notyet
8830 	device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
8831 	    __func__, dump->wr_flash, dump->len, dump->data);
8832 #endif
8833 
8834 	if (dump->wr_flash)
8835 		cudbg->use_flash = 1;
8836 	MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
8837 	memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
8838 
8839 	rc = cudbg_collect(handle, buf, &dump->len);
8840 	if (rc != 0)
8841 		goto done;
8842 
8843 	rc = copyout(buf, dump->data, dump->len);
8844 done:
8845 	cudbg_free_handle(handle);
8846 	free(buf, M_CXGBE);
8847 	return (rc);
8848 }
8849 
8850 static void
8851 free_offload_policy(struct t4_offload_policy *op)
8852 {
8853 	struct offload_rule *r;
8854 	int i;
8855 
8856 	if (op == NULL)
8857 		return;
8858 
8859 	r = &op->rule[0];
8860 	for (i = 0; i < op->nrules; i++, r++) {
8861 		free(r->bpf_prog.bf_insns, M_CXGBE);
8862 	}
8863 	free(op->rule, M_CXGBE);
8864 	free(op, M_CXGBE);
8865 }
8866 
8867 static int
8868 set_offload_policy(struct adapter *sc, struct t4_offload_policy *uop)
8869 {
8870 	int i, rc, len;
8871 	struct t4_offload_policy *op, *old;
8872 	struct bpf_program *bf;
8873 	const struct offload_settings *s;
8874 	struct offload_rule *r;
8875 	void *u;
8876 
8877 	if (!is_offload(sc))
8878 		return (ENODEV);
8879 
8880 	if (uop->nrules == 0) {
8881 		/* Delete installed policies. */
8882 		op = NULL;
8883 		goto set_policy;
8884 	} if (uop->nrules > 256) { /* arbitrary */
8885 		return (E2BIG);
8886 	}
8887 
8888 	/* Copy userspace offload policy to kernel */
8889 	op = malloc(sizeof(*op), M_CXGBE, M_ZERO | M_WAITOK);
8890 	op->nrules = uop->nrules;
8891 	len = op->nrules * sizeof(struct offload_rule);
8892 	op->rule = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
8893 	rc = copyin(uop->rule, op->rule, len);
8894 	if (rc) {
8895 		free(op->rule, M_CXGBE);
8896 		free(op, M_CXGBE);
8897 		return (rc);
8898 	}
8899 
8900 	r = &op->rule[0];
8901 	for (i = 0; i < op->nrules; i++, r++) {
8902 
8903 		/* Validate open_type */
8904 		if (r->open_type != OPEN_TYPE_LISTEN &&
8905 		    r->open_type != OPEN_TYPE_ACTIVE &&
8906 		    r->open_type != OPEN_TYPE_PASSIVE &&
8907 		    r->open_type != OPEN_TYPE_DONTCARE) {
8908 error:
8909 			/*
8910 			 * Rules 0 to i have malloc'd filters that need to be
8911 			 * freed.  Rules i+1 to nrules have userspace pointers
8912 			 * and should be left alone.
8913 			 */
8914 			op->nrules = i;
8915 			free_offload_policy(op);
8916 			return (rc);
8917 		}
8918 
8919 		/* Validate settings */
8920 		s = &r->settings;
8921 		if ((s->offload != 0 && s->offload != 1) ||
8922 		    s->cong_algo < -1 || s->cong_algo > CONG_ALG_HIGHSPEED ||
8923 		    s->sched_class < -1 ||
8924 		    s->sched_class >= sc->chip_params->nsched_cls) {
8925 			rc = EINVAL;
8926 			goto error;
8927 		}
8928 
8929 		bf = &r->bpf_prog;
8930 		u = bf->bf_insns;	/* userspace ptr */
8931 		bf->bf_insns = NULL;
8932 		if (bf->bf_len == 0) {
8933 			/* legal, matches everything */
8934 			continue;
8935 		}
8936 		len = bf->bf_len * sizeof(*bf->bf_insns);
8937 		bf->bf_insns = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
8938 		rc = copyin(u, bf->bf_insns, len);
8939 		if (rc != 0)
8940 			goto error;
8941 
8942 		if (!bpf_validate(bf->bf_insns, bf->bf_len)) {
8943 			rc = EINVAL;
8944 			goto error;
8945 		}
8946 	}
8947 set_policy:
8948 	rw_wlock(&sc->policy_lock);
8949 	old = sc->policy;
8950 	sc->policy = op;
8951 	rw_wunlock(&sc->policy_lock);
8952 	free_offload_policy(old);
8953 
8954 	return (0);
8955 }
8956 
8957 #define MAX_READ_BUF_SIZE (128 * 1024)
8958 static int
8959 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
8960 {
8961 	uint32_t addr, remaining, n;
8962 	uint32_t *buf;
8963 	int rc;
8964 	uint8_t *dst;
8965 
8966 	rc = validate_mem_range(sc, mr->addr, mr->len);
8967 	if (rc != 0)
8968 		return (rc);
8969 
8970 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
8971 	addr = mr->addr;
8972 	remaining = mr->len;
8973 	dst = (void *)mr->data;
8974 
8975 	while (remaining) {
8976 		n = min(remaining, MAX_READ_BUF_SIZE);
8977 		read_via_memwin(sc, 2, addr, buf, n);
8978 
8979 		rc = copyout(buf, dst, n);
8980 		if (rc != 0)
8981 			break;
8982 
8983 		dst += n;
8984 		remaining -= n;
8985 		addr += n;
8986 	}
8987 
8988 	free(buf, M_CXGBE);
8989 	return (rc);
8990 }
8991 #undef MAX_READ_BUF_SIZE
8992 
8993 static int
8994 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
8995 {
8996 	int rc;
8997 
8998 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
8999 		return (EINVAL);
9000 
9001 	if (i2cd->len > sizeof(i2cd->data))
9002 		return (EFBIG);
9003 
9004 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9005 	if (rc)
9006 		return (rc);
9007 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9008 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
9009 	end_synchronized_op(sc, 0);
9010 
9011 	return (rc);
9012 }
9013 
9014 int
9015 t4_os_find_pci_capability(struct adapter *sc, int cap)
9016 {
9017 	int i;
9018 
9019 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9020 }
9021 
9022 int
9023 t4_os_pci_save_state(struct adapter *sc)
9024 {
9025 	device_t dev;
9026 	struct pci_devinfo *dinfo;
9027 
9028 	dev = sc->dev;
9029 	dinfo = device_get_ivars(dev);
9030 
9031 	pci_cfg_save(dev, dinfo, 0);
9032 	return (0);
9033 }
9034 
9035 int
9036 t4_os_pci_restore_state(struct adapter *sc)
9037 {
9038 	device_t dev;
9039 	struct pci_devinfo *dinfo;
9040 
9041 	dev = sc->dev;
9042 	dinfo = device_get_ivars(dev);
9043 
9044 	pci_cfg_restore(dev, dinfo);
9045 	return (0);
9046 }
9047 
9048 void
9049 t4_os_portmod_changed(struct port_info *pi)
9050 {
9051 	struct adapter *sc = pi->adapter;
9052 	struct vi_info *vi;
9053 	struct ifnet *ifp;
9054 	static const char *mod_str[] = {
9055 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9056 	};
9057 
9058 	PORT_LOCK(pi);
9059 	build_medialist(pi, &pi->media);
9060 	PORT_UNLOCK(pi);
9061 	vi = &pi->vi[0];
9062 	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9063 		init_l1cfg(pi);
9064 		end_synchronized_op(sc, LOCK_HELD);
9065 	}
9066 
9067 	ifp = vi->ifp;
9068 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9069 		if_printf(ifp, "transceiver unplugged.\n");
9070 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9071 		if_printf(ifp, "unknown transceiver inserted.\n");
9072 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9073 		if_printf(ifp, "unsupported transceiver inserted.\n");
9074 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9075 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9076 		    port_top_speed(pi), mod_str[pi->mod_type]);
9077 	} else {
9078 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9079 		    pi->mod_type);
9080 	}
9081 }
9082 
9083 void
9084 t4_os_link_changed(struct port_info *pi)
9085 {
9086 	struct vi_info *vi;
9087 	struct ifnet *ifp;
9088 	struct link_config *lc;
9089 	int v;
9090 
9091 	for_each_vi(pi, v, vi) {
9092 		ifp = vi->ifp;
9093 		if (ifp == NULL)
9094 			continue;
9095 
9096 		lc = &pi->link_cfg;
9097 		if (lc->link_ok) {
9098 			ifp->if_baudrate = IF_Mbps(lc->speed);
9099 			if_link_state_change(ifp, LINK_STATE_UP);
9100 		} else {
9101 			if_link_state_change(ifp, LINK_STATE_DOWN);
9102 		}
9103 	}
9104 }
9105 
9106 void
9107 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9108 {
9109 	struct adapter *sc;
9110 
9111 	sx_slock(&t4_list_lock);
9112 	SLIST_FOREACH(sc, &t4_list, link) {
9113 		/*
9114 		 * func should not make any assumptions about what state sc is
9115 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9116 		 */
9117 		func(sc, arg);
9118 	}
9119 	sx_sunlock(&t4_list_lock);
9120 }
9121 
9122 static int
9123 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9124     struct thread *td)
9125 {
9126 	int rc;
9127 	struct adapter *sc = dev->si_drv1;
9128 
9129 	rc = priv_check(td, PRIV_DRIVER);
9130 	if (rc != 0)
9131 		return (rc);
9132 
9133 	switch (cmd) {
9134 	case CHELSIO_T4_GETREG: {
9135 		struct t4_reg *edata = (struct t4_reg *)data;
9136 
9137 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9138 			return (EFAULT);
9139 
9140 		if (edata->size == 4)
9141 			edata->val = t4_read_reg(sc, edata->addr);
9142 		else if (edata->size == 8)
9143 			edata->val = t4_read_reg64(sc, edata->addr);
9144 		else
9145 			return (EINVAL);
9146 
9147 		break;
9148 	}
9149 	case CHELSIO_T4_SETREG: {
9150 		struct t4_reg *edata = (struct t4_reg *)data;
9151 
9152 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9153 			return (EFAULT);
9154 
9155 		if (edata->size == 4) {
9156 			if (edata->val & 0xffffffff00000000)
9157 				return (EINVAL);
9158 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9159 		} else if (edata->size == 8)
9160 			t4_write_reg64(sc, edata->addr, edata->val);
9161 		else
9162 			return (EINVAL);
9163 		break;
9164 	}
9165 	case CHELSIO_T4_REGDUMP: {
9166 		struct t4_regdump *regs = (struct t4_regdump *)data;
9167 		int reglen = t4_get_regs_len(sc);
9168 		uint8_t *buf;
9169 
9170 		if (regs->len < reglen) {
9171 			regs->len = reglen; /* hint to the caller */
9172 			return (ENOBUFS);
9173 		}
9174 
9175 		regs->len = reglen;
9176 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9177 		get_regs(sc, regs, buf);
9178 		rc = copyout(buf, regs->data, reglen);
9179 		free(buf, M_CXGBE);
9180 		break;
9181 	}
9182 	case CHELSIO_T4_GET_FILTER_MODE:
9183 		rc = get_filter_mode(sc, (uint32_t *)data);
9184 		break;
9185 	case CHELSIO_T4_SET_FILTER_MODE:
9186 		rc = set_filter_mode(sc, *(uint32_t *)data);
9187 		break;
9188 	case CHELSIO_T4_GET_FILTER:
9189 		rc = get_filter(sc, (struct t4_filter *)data);
9190 		break;
9191 	case CHELSIO_T4_SET_FILTER:
9192 		rc = set_filter(sc, (struct t4_filter *)data);
9193 		break;
9194 	case CHELSIO_T4_DEL_FILTER:
9195 		rc = del_filter(sc, (struct t4_filter *)data);
9196 		break;
9197 	case CHELSIO_T4_GET_SGE_CONTEXT:
9198 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9199 		break;
9200 	case CHELSIO_T4_LOAD_FW:
9201 		rc = load_fw(sc, (struct t4_data *)data);
9202 		break;
9203 	case CHELSIO_T4_GET_MEM:
9204 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9205 		break;
9206 	case CHELSIO_T4_GET_I2C:
9207 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9208 		break;
9209 	case CHELSIO_T4_CLEAR_STATS: {
9210 		int i, v, bg_map;
9211 		u_int port_id = *(uint32_t *)data;
9212 		struct port_info *pi;
9213 		struct vi_info *vi;
9214 
9215 		if (port_id >= sc->params.nports)
9216 			return (EINVAL);
9217 		pi = sc->port[port_id];
9218 		if (pi == NULL)
9219 			return (EIO);
9220 
9221 		/* MAC stats */
9222 		t4_clr_port_stats(sc, pi->tx_chan);
9223 		pi->tx_parse_error = 0;
9224 		pi->tnl_cong_drops = 0;
9225 		mtx_lock(&sc->reg_lock);
9226 		for_each_vi(pi, v, vi) {
9227 			if (vi->flags & VI_INIT_DONE)
9228 				t4_clr_vi_stats(sc, vi->viid);
9229 		}
9230 		bg_map = pi->mps_bg_map;
9231 		v = 0;	/* reuse */
9232 		while (bg_map) {
9233 			i = ffs(bg_map) - 1;
9234 			t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
9235 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
9236 			bg_map &= ~(1 << i);
9237 		}
9238 		mtx_unlock(&sc->reg_lock);
9239 
9240 		/*
9241 		 * Since this command accepts a port, clear stats for
9242 		 * all VIs on this port.
9243 		 */
9244 		for_each_vi(pi, v, vi) {
9245 			if (vi->flags & VI_INIT_DONE) {
9246 				struct sge_rxq *rxq;
9247 				struct sge_txq *txq;
9248 				struct sge_wrq *wrq;
9249 
9250 				for_each_rxq(vi, i, rxq) {
9251 #if defined(INET) || defined(INET6)
9252 					rxq->lro.lro_queued = 0;
9253 					rxq->lro.lro_flushed = 0;
9254 #endif
9255 					rxq->rxcsum = 0;
9256 					rxq->vlan_extraction = 0;
9257 				}
9258 
9259 				for_each_txq(vi, i, txq) {
9260 					txq->txcsum = 0;
9261 					txq->tso_wrs = 0;
9262 					txq->vlan_insertion = 0;
9263 					txq->imm_wrs = 0;
9264 					txq->sgl_wrs = 0;
9265 					txq->txpkt_wrs = 0;
9266 					txq->txpkts0_wrs = 0;
9267 					txq->txpkts1_wrs = 0;
9268 					txq->txpkts0_pkts = 0;
9269 					txq->txpkts1_pkts = 0;
9270 					mp_ring_reset_stats(txq->r);
9271 				}
9272 
9273 #ifdef TCP_OFFLOAD
9274 				/* nothing to clear for each ofld_rxq */
9275 
9276 				for_each_ofld_txq(vi, i, wrq) {
9277 					wrq->tx_wrs_direct = 0;
9278 					wrq->tx_wrs_copied = 0;
9279 				}
9280 #endif
9281 
9282 				if (IS_MAIN_VI(vi)) {
9283 					wrq = &sc->sge.ctrlq[pi->port_id];
9284 					wrq->tx_wrs_direct = 0;
9285 					wrq->tx_wrs_copied = 0;
9286 				}
9287 			}
9288 		}
9289 		break;
9290 	}
9291 	case CHELSIO_T4_SCHED_CLASS:
9292 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9293 		break;
9294 	case CHELSIO_T4_SCHED_QUEUE:
9295 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9296 		break;
9297 	case CHELSIO_T4_GET_TRACER:
9298 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9299 		break;
9300 	case CHELSIO_T4_SET_TRACER:
9301 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9302 		break;
9303 	case CHELSIO_T4_LOAD_CFG:
9304 		rc = load_cfg(sc, (struct t4_data *)data);
9305 		break;
9306 	case CHELSIO_T4_LOAD_BOOT:
9307 		rc = load_boot(sc, (struct t4_bootrom *)data);
9308 		break;
9309 	case CHELSIO_T4_LOAD_BOOTCFG:
9310 		rc = load_bootcfg(sc, (struct t4_data *)data);
9311 		break;
9312 	case CHELSIO_T4_CUDBG_DUMP:
9313 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
9314 		break;
9315 	case CHELSIO_T4_SET_OFLD_POLICY:
9316 		rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
9317 		break;
9318 	default:
9319 		rc = ENOTTY;
9320 	}
9321 
9322 	return (rc);
9323 }
9324 
9325 void
9326 t4_db_full(struct adapter *sc)
9327 {
9328 
9329 	CXGBE_UNIMPLEMENTED(__func__);
9330 }
9331 
9332 void
9333 t4_db_dropped(struct adapter *sc)
9334 {
9335 
9336 	CXGBE_UNIMPLEMENTED(__func__);
9337 }
9338 
9339 #ifdef TCP_OFFLOAD
9340 static int
9341 toe_capability(struct vi_info *vi, int enable)
9342 {
9343 	int rc;
9344 	struct port_info *pi = vi->pi;
9345 	struct adapter *sc = pi->adapter;
9346 
9347 	ASSERT_SYNCHRONIZED_OP(sc);
9348 
9349 	if (!is_offload(sc))
9350 		return (ENODEV);
9351 
9352 	if (enable) {
9353 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9354 			/* TOE is already enabled. */
9355 			return (0);
9356 		}
9357 
9358 		/*
9359 		 * We need the port's queues around so that we're able to send
9360 		 * and receive CPLs to/from the TOE even if the ifnet for this
9361 		 * port has never been UP'd administratively.
9362 		 */
9363 		if (!(vi->flags & VI_INIT_DONE)) {
9364 			rc = vi_full_init(vi);
9365 			if (rc)
9366 				return (rc);
9367 		}
9368 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9369 			rc = vi_full_init(&pi->vi[0]);
9370 			if (rc)
9371 				return (rc);
9372 		}
9373 
9374 		if (isset(&sc->offload_map, pi->port_id)) {
9375 			/* TOE is enabled on another VI of this port. */
9376 			pi->uld_vis++;
9377 			return (0);
9378 		}
9379 
9380 		if (!uld_active(sc, ULD_TOM)) {
9381 			rc = t4_activate_uld(sc, ULD_TOM);
9382 			if (rc == EAGAIN) {
9383 				log(LOG_WARNING,
9384 				    "You must kldload t4_tom.ko before trying "
9385 				    "to enable TOE on a cxgbe interface.\n");
9386 			}
9387 			if (rc != 0)
9388 				return (rc);
9389 			KASSERT(sc->tom_softc != NULL,
9390 			    ("%s: TOM activated but softc NULL", __func__));
9391 			KASSERT(uld_active(sc, ULD_TOM),
9392 			    ("%s: TOM activated but flag not set", __func__));
9393 		}
9394 
9395 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
9396 		if (!uld_active(sc, ULD_IWARP))
9397 			(void) t4_activate_uld(sc, ULD_IWARP);
9398 		if (!uld_active(sc, ULD_ISCSI))
9399 			(void) t4_activate_uld(sc, ULD_ISCSI);
9400 
9401 		pi->uld_vis++;
9402 		setbit(&sc->offload_map, pi->port_id);
9403 	} else {
9404 		pi->uld_vis--;
9405 
9406 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9407 			return (0);
9408 
9409 		KASSERT(uld_active(sc, ULD_TOM),
9410 		    ("%s: TOM never initialized?", __func__));
9411 		clrbit(&sc->offload_map, pi->port_id);
9412 	}
9413 
9414 	return (0);
9415 }
9416 
9417 /*
9418  * Add an upper layer driver to the global list.
9419  */
9420 int
9421 t4_register_uld(struct uld_info *ui)
9422 {
9423 	int rc = 0;
9424 	struct uld_info *u;
9425 
9426 	sx_xlock(&t4_uld_list_lock);
9427 	SLIST_FOREACH(u, &t4_uld_list, link) {
9428 	    if (u->uld_id == ui->uld_id) {
9429 		    rc = EEXIST;
9430 		    goto done;
9431 	    }
9432 	}
9433 
9434 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9435 	ui->refcount = 0;
9436 done:
9437 	sx_xunlock(&t4_uld_list_lock);
9438 	return (rc);
9439 }
9440 
9441 int
9442 t4_unregister_uld(struct uld_info *ui)
9443 {
9444 	int rc = EINVAL;
9445 	struct uld_info *u;
9446 
9447 	sx_xlock(&t4_uld_list_lock);
9448 
9449 	SLIST_FOREACH(u, &t4_uld_list, link) {
9450 	    if (u == ui) {
9451 		    if (ui->refcount > 0) {
9452 			    rc = EBUSY;
9453 			    goto done;
9454 		    }
9455 
9456 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9457 		    rc = 0;
9458 		    goto done;
9459 	    }
9460 	}
9461 done:
9462 	sx_xunlock(&t4_uld_list_lock);
9463 	return (rc);
9464 }
9465 
9466 int
9467 t4_activate_uld(struct adapter *sc, int id)
9468 {
9469 	int rc;
9470 	struct uld_info *ui;
9471 
9472 	ASSERT_SYNCHRONIZED_OP(sc);
9473 
9474 	if (id < 0 || id > ULD_MAX)
9475 		return (EINVAL);
9476 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
9477 
9478 	sx_slock(&t4_uld_list_lock);
9479 
9480 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9481 		if (ui->uld_id == id) {
9482 			if (!(sc->flags & FULL_INIT_DONE)) {
9483 				rc = adapter_full_init(sc);
9484 				if (rc != 0)
9485 					break;
9486 			}
9487 
9488 			rc = ui->activate(sc);
9489 			if (rc == 0) {
9490 				setbit(&sc->active_ulds, id);
9491 				ui->refcount++;
9492 			}
9493 			break;
9494 		}
9495 	}
9496 
9497 	sx_sunlock(&t4_uld_list_lock);
9498 
9499 	return (rc);
9500 }
9501 
9502 int
9503 t4_deactivate_uld(struct adapter *sc, int id)
9504 {
9505 	int rc;
9506 	struct uld_info *ui;
9507 
9508 	ASSERT_SYNCHRONIZED_OP(sc);
9509 
9510 	if (id < 0 || id > ULD_MAX)
9511 		return (EINVAL);
9512 	rc = ENXIO;
9513 
9514 	sx_slock(&t4_uld_list_lock);
9515 
9516 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9517 		if (ui->uld_id == id) {
9518 			rc = ui->deactivate(sc);
9519 			if (rc == 0) {
9520 				clrbit(&sc->active_ulds, id);
9521 				ui->refcount--;
9522 			}
9523 			break;
9524 		}
9525 	}
9526 
9527 	sx_sunlock(&t4_uld_list_lock);
9528 
9529 	return (rc);
9530 }
9531 
9532 int
9533 uld_active(struct adapter *sc, int uld_id)
9534 {
9535 
9536 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9537 
9538 	return (isset(&sc->active_ulds, uld_id));
9539 }
9540 #endif
9541 
9542 /*
9543  * t  = ptr to tunable.
9544  * nc = number of CPUs.
9545  * c  = compiled in default for that tunable.
9546  */
9547 static void
9548 calculate_nqueues(int *t, int nc, const int c)
9549 {
9550 	int nq;
9551 
9552 	if (*t > 0)
9553 		return;
9554 	nq = *t < 0 ? -*t : c;
9555 	*t = min(nc, nq);
9556 }
9557 
9558 /*
9559  * Come up with reasonable defaults for some of the tunables, provided they're
9560  * not set by the user (in which case we'll use the values as is).
9561  */
9562 static void
9563 tweak_tunables(void)
9564 {
9565 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
9566 
9567 	if (t4_ntxq < 1) {
9568 #ifdef RSS
9569 		t4_ntxq = rss_getnumbuckets();
9570 #else
9571 		calculate_nqueues(&t4_ntxq, nc, NTXQ);
9572 #endif
9573 	}
9574 
9575 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
9576 
9577 	if (t4_nrxq < 1) {
9578 #ifdef RSS
9579 		t4_nrxq = rss_getnumbuckets();
9580 #else
9581 		calculate_nqueues(&t4_nrxq, nc, NRXQ);
9582 #endif
9583 	}
9584 
9585 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
9586 
9587 #ifdef TCP_OFFLOAD
9588 	calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
9589 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
9590 	calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
9591 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
9592 
9593 	if (t4_toecaps_allowed == -1)
9594 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9595 
9596 	if (t4_rdmacaps_allowed == -1) {
9597 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9598 		    FW_CAPS_CONFIG_RDMA_RDMAC;
9599 	}
9600 
9601 	if (t4_iscsicaps_allowed == -1) {
9602 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9603 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9604 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
9605 	}
9606 
9607 	if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
9608 		t4_tmr_idx_ofld = TMR_IDX_OFLD;
9609 
9610 	if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
9611 		t4_pktc_idx_ofld = PKTC_IDX_OFLD;
9612 #else
9613 	if (t4_toecaps_allowed == -1)
9614 		t4_toecaps_allowed = 0;
9615 
9616 	if (t4_rdmacaps_allowed == -1)
9617 		t4_rdmacaps_allowed = 0;
9618 
9619 	if (t4_iscsicaps_allowed == -1)
9620 		t4_iscsicaps_allowed = 0;
9621 #endif
9622 
9623 #ifdef DEV_NETMAP
9624 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
9625 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
9626 #endif
9627 
9628 	if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
9629 		t4_tmr_idx = TMR_IDX;
9630 
9631 	if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
9632 		t4_pktc_idx = PKTC_IDX;
9633 
9634 	if (t4_qsize_txq < 128)
9635 		t4_qsize_txq = 128;
9636 
9637 	if (t4_qsize_rxq < 128)
9638 		t4_qsize_rxq = 128;
9639 	while (t4_qsize_rxq & 7)
9640 		t4_qsize_rxq++;
9641 
9642 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9643 
9644 	/*
9645 	 * Number of VIs to create per-port.  The first VI is the "main" regular
9646 	 * VI for the port.  The rest are additional virtual interfaces on the
9647 	 * same physical port.  Note that the main VI does not have native
9648 	 * netmap support but the extra VIs do.
9649 	 *
9650 	 * Limit the number of VIs per port to the number of available
9651 	 * MAC addresses per port.
9652 	 */
9653 	if (t4_num_vis < 1)
9654 		t4_num_vis = 1;
9655 	if (t4_num_vis > nitems(vi_mac_funcs)) {
9656 		t4_num_vis = nitems(vi_mac_funcs);
9657 		printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
9658 	}
9659 
9660 	if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
9661 		pcie_relaxed_ordering = 1;
9662 #if defined(__i386__) || defined(__amd64__)
9663 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
9664 			pcie_relaxed_ordering = 0;
9665 #endif
9666 	}
9667 }
9668 
9669 #ifdef DDB
9670 static void
9671 t4_dump_tcb(struct adapter *sc, int tid)
9672 {
9673 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
9674 
9675 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
9676 	save = t4_read_reg(sc, reg);
9677 	base = sc->memwin[2].mw_base;
9678 
9679 	/* Dump TCB for the tid */
9680 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
9681 	tcb_addr += tid * TCB_SIZE;
9682 
9683 	if (is_t4(sc)) {
9684 		pf = 0;
9685 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
9686 	} else {
9687 		pf = V_PFNUM(sc->pf);
9688 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
9689 	}
9690 	t4_write_reg(sc, reg, win_pos | pf);
9691 	t4_read_reg(sc, reg);
9692 
9693 	off = tcb_addr - win_pos;
9694 	for (i = 0; i < 4; i++) {
9695 		uint32_t buf[8];
9696 		for (j = 0; j < 8; j++, off += 4)
9697 			buf[j] = htonl(t4_read_reg(sc, base + off));
9698 
9699 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
9700 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
9701 		    buf[7]);
9702 	}
9703 
9704 	t4_write_reg(sc, reg, save);
9705 	t4_read_reg(sc, reg);
9706 }
9707 
9708 static void
9709 t4_dump_devlog(struct adapter *sc)
9710 {
9711 	struct devlog_params *dparams = &sc->params.devlog;
9712 	struct fw_devlog_e e;
9713 	int i, first, j, m, nentries, rc;
9714 	uint64_t ftstamp = UINT64_MAX;
9715 
9716 	if (dparams->start == 0) {
9717 		db_printf("devlog params not valid\n");
9718 		return;
9719 	}
9720 
9721 	nentries = dparams->size / sizeof(struct fw_devlog_e);
9722 	m = fwmtype_to_hwmtype(dparams->memtype);
9723 
9724 	/* Find the first entry. */
9725 	first = -1;
9726 	for (i = 0; i < nentries && !db_pager_quit; i++) {
9727 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9728 		    sizeof(e), (void *)&e);
9729 		if (rc != 0)
9730 			break;
9731 
9732 		if (e.timestamp == 0)
9733 			break;
9734 
9735 		e.timestamp = be64toh(e.timestamp);
9736 		if (e.timestamp < ftstamp) {
9737 			ftstamp = e.timestamp;
9738 			first = i;
9739 		}
9740 	}
9741 
9742 	if (first == -1)
9743 		return;
9744 
9745 	i = first;
9746 	do {
9747 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9748 		    sizeof(e), (void *)&e);
9749 		if (rc != 0)
9750 			return;
9751 
9752 		if (e.timestamp == 0)
9753 			return;
9754 
9755 		e.timestamp = be64toh(e.timestamp);
9756 		e.seqno = be32toh(e.seqno);
9757 		for (j = 0; j < 8; j++)
9758 			e.params[j] = be32toh(e.params[j]);
9759 
9760 		db_printf("%10d  %15ju  %8s  %8s  ",
9761 		    e.seqno, e.timestamp,
9762 		    (e.level < nitems(devlog_level_strings) ?
9763 			devlog_level_strings[e.level] : "UNKNOWN"),
9764 		    (e.facility < nitems(devlog_facility_strings) ?
9765 			devlog_facility_strings[e.facility] : "UNKNOWN"));
9766 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
9767 		    e.params[3], e.params[4], e.params[5], e.params[6],
9768 		    e.params[7]);
9769 
9770 		if (++i == nentries)
9771 			i = 0;
9772 	} while (i != first && !db_pager_quit);
9773 }
9774 
9775 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
9776 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
9777 
9778 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
9779 {
9780 	device_t dev;
9781 	int t;
9782 	bool valid;
9783 
9784 	valid = false;
9785 	t = db_read_token();
9786 	if (t == tIDENT) {
9787 		dev = device_lookup_by_name(db_tok_string);
9788 		valid = true;
9789 	}
9790 	db_skip_to_eol();
9791 	if (!valid) {
9792 		db_printf("usage: show t4 devlog <nexus>\n");
9793 		return;
9794 	}
9795 
9796 	if (dev == NULL) {
9797 		db_printf("device not found\n");
9798 		return;
9799 	}
9800 
9801 	t4_dump_devlog(device_get_softc(dev));
9802 }
9803 
9804 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
9805 {
9806 	device_t dev;
9807 	int radix, tid, t;
9808 	bool valid;
9809 
9810 	valid = false;
9811 	radix = db_radix;
9812 	db_radix = 10;
9813 	t = db_read_token();
9814 	if (t == tIDENT) {
9815 		dev = device_lookup_by_name(db_tok_string);
9816 		t = db_read_token();
9817 		if (t == tNUMBER) {
9818 			tid = db_tok_number;
9819 			valid = true;
9820 		}
9821 	}
9822 	db_radix = radix;
9823 	db_skip_to_eol();
9824 	if (!valid) {
9825 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
9826 		return;
9827 	}
9828 
9829 	if (dev == NULL) {
9830 		db_printf("device not found\n");
9831 		return;
9832 	}
9833 	if (tid < 0) {
9834 		db_printf("invalid tid\n");
9835 		return;
9836 	}
9837 
9838 	t4_dump_tcb(device_get_softc(dev), tid);
9839 }
9840 #endif
9841 
9842 /*
9843  * Borrowed from cesa_prep_aes_key().
9844  *
9845  * NB: The crypto engine wants the words in the decryption key in reverse
9846  * order.
9847  */
9848 void
9849 t4_aes_getdeckey(void *dec_key, const void *enc_key, unsigned int kbits)
9850 {
9851 	uint32_t ek[4 * (RIJNDAEL_MAXNR + 1)];
9852 	uint32_t *dkey;
9853 	int i;
9854 
9855 	rijndaelKeySetupEnc(ek, enc_key, kbits);
9856 	dkey = dec_key;
9857 	dkey += (kbits / 8) / 4;
9858 
9859 	switch (kbits) {
9860 	case 128:
9861 		for (i = 0; i < 4; i++)
9862 			*--dkey = htobe32(ek[4 * 10 + i]);
9863 		break;
9864 	case 192:
9865 		for (i = 0; i < 2; i++)
9866 			*--dkey = htobe32(ek[4 * 11 + 2 + i]);
9867 		for (i = 0; i < 4; i++)
9868 			*--dkey = htobe32(ek[4 * 12 + i]);
9869 		break;
9870 	case 256:
9871 		for (i = 0; i < 4; i++)
9872 			*--dkey = htobe32(ek[4 * 13 + i]);
9873 		for (i = 0; i < 4; i++)
9874 			*--dkey = htobe32(ek[4 * 14 + i]);
9875 		break;
9876 	}
9877 	MPASS(dkey == dec_key);
9878 }
9879 
9880 static struct sx mlu;	/* mod load unload */
9881 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
9882 
9883 static int
9884 mod_event(module_t mod, int cmd, void *arg)
9885 {
9886 	int rc = 0;
9887 	static int loaded = 0;
9888 
9889 	switch (cmd) {
9890 	case MOD_LOAD:
9891 		sx_xlock(&mlu);
9892 		if (loaded++ == 0) {
9893 			t4_sge_modload();
9894 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
9895 			    t4_filter_rpl, CPL_COOKIE_FILTER);
9896 			t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
9897 			    do_l2t_write_rpl, CPL_COOKIE_FILTER);
9898 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
9899 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
9900 			sx_init(&t4_list_lock, "T4/T5 adapters");
9901 			SLIST_INIT(&t4_list);
9902 #ifdef TCP_OFFLOAD
9903 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
9904 			SLIST_INIT(&t4_uld_list);
9905 #endif
9906 			t4_tracer_modload();
9907 			tweak_tunables();
9908 		}
9909 		sx_xunlock(&mlu);
9910 		break;
9911 
9912 	case MOD_UNLOAD:
9913 		sx_xlock(&mlu);
9914 		if (--loaded == 0) {
9915 			int tries;
9916 
9917 			sx_slock(&t4_list_lock);
9918 			if (!SLIST_EMPTY(&t4_list)) {
9919 				rc = EBUSY;
9920 				sx_sunlock(&t4_list_lock);
9921 				goto done_unload;
9922 			}
9923 #ifdef TCP_OFFLOAD
9924 			sx_slock(&t4_uld_list_lock);
9925 			if (!SLIST_EMPTY(&t4_uld_list)) {
9926 				rc = EBUSY;
9927 				sx_sunlock(&t4_uld_list_lock);
9928 				sx_sunlock(&t4_list_lock);
9929 				goto done_unload;
9930 			}
9931 #endif
9932 			tries = 0;
9933 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
9934 				uprintf("%ju clusters with custom free routine "
9935 				    "still is use.\n", t4_sge_extfree_refs());
9936 				pause("t4unload", 2 * hz);
9937 			}
9938 #ifdef TCP_OFFLOAD
9939 			sx_sunlock(&t4_uld_list_lock);
9940 #endif
9941 			sx_sunlock(&t4_list_lock);
9942 
9943 			if (t4_sge_extfree_refs() == 0) {
9944 				t4_tracer_modunload();
9945 #ifdef TCP_OFFLOAD
9946 				sx_destroy(&t4_uld_list_lock);
9947 #endif
9948 				sx_destroy(&t4_list_lock);
9949 				t4_sge_modunload();
9950 				loaded = 0;
9951 			} else {
9952 				rc = EBUSY;
9953 				loaded++;	/* undo earlier decrement */
9954 			}
9955 		}
9956 done_unload:
9957 		sx_xunlock(&mlu);
9958 		break;
9959 	}
9960 
9961 	return (rc);
9962 }
9963 
9964 static devclass_t t4_devclass, t5_devclass, t6_devclass;
9965 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
9966 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
9967 
9968 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
9969 MODULE_VERSION(t4nex, 1);
9970 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
9971 #ifdef DEV_NETMAP
9972 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
9973 #endif /* DEV_NETMAP */
9974 
9975 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
9976 MODULE_VERSION(t5nex, 1);
9977 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
9978 #ifdef DEV_NETMAP
9979 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
9980 #endif /* DEV_NETMAP */
9981 
9982 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
9983 MODULE_VERSION(t6nex, 1);
9984 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
9985 #ifdef DEV_NETMAP
9986 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
9987 #endif /* DEV_NETMAP */
9988 
9989 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
9990 MODULE_VERSION(cxgbe, 1);
9991 
9992 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
9993 MODULE_VERSION(cxl, 1);
9994 
9995 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
9996 MODULE_VERSION(cc, 1);
9997 
9998 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
9999 MODULE_VERSION(vcxgbe, 1);
10000 
10001 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10002 MODULE_VERSION(vcxl, 1);
10003 
10004 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10005 MODULE_VERSION(vcc, 1);
10006