xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision 2284664ef9fcb0baaf59f1ef7df877c0b0f2b187)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011 Chelsio Communications, Inc.
5  * All rights reserved.
6  * Written by: Navdeep Parhar <np@FreeBSD.org>
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_ratelimit.h"
37 #include "opt_rss.h"
38 
39 #include <sys/param.h>
40 #include <sys/conf.h>
41 #include <sys/priv.h>
42 #include <sys/kernel.h>
43 #include <sys/bus.h>
44 #include <sys/module.h>
45 #include <sys/malloc.h>
46 #include <sys/queue.h>
47 #include <sys/taskqueue.h>
48 #include <sys/pciio.h>
49 #include <dev/pci/pcireg.h>
50 #include <dev/pci/pcivar.h>
51 #include <dev/pci/pci_private.h>
52 #include <sys/firmware.h>
53 #include <sys/sbuf.h>
54 #include <sys/smp.h>
55 #include <sys/socket.h>
56 #include <sys/sockio.h>
57 #include <sys/sysctl.h>
58 #include <net/ethernet.h>
59 #include <net/if.h>
60 #include <net/if_types.h>
61 #include <net/if_dl.h>
62 #include <net/if_vlan_var.h>
63 #ifdef RSS
64 #include <net/rss_config.h>
65 #endif
66 #if defined(__i386__) || defined(__amd64__)
67 #include <machine/md_var.h>
68 #include <machine/cputypes.h>
69 #include <vm/vm.h>
70 #include <vm/pmap.h>
71 #endif
72 #include <crypto/rijndael/rijndael.h>
73 #ifdef DDB
74 #include <ddb/ddb.h>
75 #include <ddb/db_lex.h>
76 #endif
77 
78 #include "common/common.h"
79 #include "common/t4_msg.h"
80 #include "common/t4_regs.h"
81 #include "common/t4_regs_values.h"
82 #include "cudbg/cudbg.h"
83 #include "t4_ioctl.h"
84 #include "t4_l2t.h"
85 #include "t4_mp_ring.h"
86 #include "t4_if.h"
87 #include "t4_smt.h"
88 
89 /* T4 bus driver interface */
90 static int t4_probe(device_t);
91 static int t4_attach(device_t);
92 static int t4_detach(device_t);
93 static int t4_ready(device_t);
94 static int t4_read_port_device(device_t, int, device_t *);
95 static device_method_t t4_methods[] = {
96 	DEVMETHOD(device_probe,		t4_probe),
97 	DEVMETHOD(device_attach,	t4_attach),
98 	DEVMETHOD(device_detach,	t4_detach),
99 
100 	DEVMETHOD(t4_is_main_ready,	t4_ready),
101 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
102 
103 	DEVMETHOD_END
104 };
105 static driver_t t4_driver = {
106 	"t4nex",
107 	t4_methods,
108 	sizeof(struct adapter)
109 };
110 
111 
112 /* T4 port (cxgbe) interface */
113 static int cxgbe_probe(device_t);
114 static int cxgbe_attach(device_t);
115 static int cxgbe_detach(device_t);
116 device_method_t cxgbe_methods[] = {
117 	DEVMETHOD(device_probe,		cxgbe_probe),
118 	DEVMETHOD(device_attach,	cxgbe_attach),
119 	DEVMETHOD(device_detach,	cxgbe_detach),
120 	{ 0, 0 }
121 };
122 static driver_t cxgbe_driver = {
123 	"cxgbe",
124 	cxgbe_methods,
125 	sizeof(struct port_info)
126 };
127 
128 /* T4 VI (vcxgbe) interface */
129 static int vcxgbe_probe(device_t);
130 static int vcxgbe_attach(device_t);
131 static int vcxgbe_detach(device_t);
132 static device_method_t vcxgbe_methods[] = {
133 	DEVMETHOD(device_probe,		vcxgbe_probe),
134 	DEVMETHOD(device_attach,	vcxgbe_attach),
135 	DEVMETHOD(device_detach,	vcxgbe_detach),
136 	{ 0, 0 }
137 };
138 static driver_t vcxgbe_driver = {
139 	"vcxgbe",
140 	vcxgbe_methods,
141 	sizeof(struct vi_info)
142 };
143 
144 static d_ioctl_t t4_ioctl;
145 
146 static struct cdevsw t4_cdevsw = {
147        .d_version = D_VERSION,
148        .d_ioctl = t4_ioctl,
149        .d_name = "t4nex",
150 };
151 
152 /* T5 bus driver interface */
153 static int t5_probe(device_t);
154 static device_method_t t5_methods[] = {
155 	DEVMETHOD(device_probe,		t5_probe),
156 	DEVMETHOD(device_attach,	t4_attach),
157 	DEVMETHOD(device_detach,	t4_detach),
158 
159 	DEVMETHOD(t4_is_main_ready,	t4_ready),
160 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
161 
162 	DEVMETHOD_END
163 };
164 static driver_t t5_driver = {
165 	"t5nex",
166 	t5_methods,
167 	sizeof(struct adapter)
168 };
169 
170 
171 /* T5 port (cxl) interface */
172 static driver_t cxl_driver = {
173 	"cxl",
174 	cxgbe_methods,
175 	sizeof(struct port_info)
176 };
177 
178 /* T5 VI (vcxl) interface */
179 static driver_t vcxl_driver = {
180 	"vcxl",
181 	vcxgbe_methods,
182 	sizeof(struct vi_info)
183 };
184 
185 /* T6 bus driver interface */
186 static int t6_probe(device_t);
187 static device_method_t t6_methods[] = {
188 	DEVMETHOD(device_probe,		t6_probe),
189 	DEVMETHOD(device_attach,	t4_attach),
190 	DEVMETHOD(device_detach,	t4_detach),
191 
192 	DEVMETHOD(t4_is_main_ready,	t4_ready),
193 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
194 
195 	DEVMETHOD_END
196 };
197 static driver_t t6_driver = {
198 	"t6nex",
199 	t6_methods,
200 	sizeof(struct adapter)
201 };
202 
203 
204 /* T6 port (cc) interface */
205 static driver_t cc_driver = {
206 	"cc",
207 	cxgbe_methods,
208 	sizeof(struct port_info)
209 };
210 
211 /* T6 VI (vcc) interface */
212 static driver_t vcc_driver = {
213 	"vcc",
214 	vcxgbe_methods,
215 	sizeof(struct vi_info)
216 };
217 
218 /* ifnet interface */
219 static void cxgbe_init(void *);
220 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
221 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
222 static void cxgbe_qflush(struct ifnet *);
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 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
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.  All the _CF names here are special.
383  */
384 #define DEFAULT_CF	"default"
385 #define BUILTIN_CF	"built-in"
386 #define FLASH_CF	"flash"
387 #define UWIRE_CF	"uwire"
388 #define FPGA_CF		"fpga"
389 static char t4_cfg_file[32] = DEFAULT_CF;
390 TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
391 
392 /*
393  * PAUSE settings (bit 0, 1, 2 = rx_pause, tx_pause, pause_autoneg respectively).
394  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
395  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
396  *            mark or when signalled to do so, 0 to never emit PAUSE.
397  * pause_autoneg = 1 means PAUSE will be negotiated if possible and the
398  *                 negotiated settings will override rx_pause/tx_pause.
399  *                 Otherwise rx_pause/tx_pause are applied forcibly.
400  */
401 static int t4_pause_settings = PAUSE_RX | PAUSE_TX | PAUSE_AUTONEG;
402 TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
403 
404 /*
405  * Forward Error Correction settings (bit 0, 1 = RS, BASER respectively).
406  * -1 to run with the firmware default.  Same as FEC_AUTO (bit 5)
407  *  0 to disable FEC.
408  */
409 static int t4_fec = -1;
410 TUNABLE_INT("hw.cxgbe.fec", &t4_fec);
411 
412 /*
413  * Link autonegotiation.
414  * -1 to run with the firmware default.
415  *  0 to disable.
416  *  1 to enable.
417  */
418 static int t4_autoneg = -1;
419 TUNABLE_INT("hw.cxgbe.autoneg", &t4_autoneg);
420 
421 /*
422  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
423  * encouraged respectively).
424  */
425 static unsigned int t4_fw_install = 1;
426 TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
427 
428 /*
429  * ASIC features that will be used.  Disable the ones you don't want so that the
430  * chip resources aren't wasted on features that will not be used.
431  */
432 static int t4_nbmcaps_allowed = 0;
433 TUNABLE_INT("hw.cxgbe.nbmcaps_allowed", &t4_nbmcaps_allowed);
434 
435 static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
436 TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
437 
438 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
439     FW_CAPS_CONFIG_SWITCH_EGRESS;
440 TUNABLE_INT("hw.cxgbe.switchcaps_allowed", &t4_switchcaps_allowed);
441 
442 #ifdef RATELIMIT
443 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
444 	FW_CAPS_CONFIG_NIC_HASHFILTER | FW_CAPS_CONFIG_NIC_ETHOFLD;
445 #else
446 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
447 	FW_CAPS_CONFIG_NIC_HASHFILTER;
448 #endif
449 TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
450 
451 static int t4_toecaps_allowed = -1;
452 TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
453 
454 static int t4_rdmacaps_allowed = -1;
455 TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
456 
457 static int t4_cryptocaps_allowed = -1;
458 TUNABLE_INT("hw.cxgbe.cryptocaps_allowed", &t4_cryptocaps_allowed);
459 
460 static int t4_iscsicaps_allowed = -1;
461 TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
462 
463 static int t4_fcoecaps_allowed = 0;
464 TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
465 
466 static int t5_write_combine = 0;
467 TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
468 
469 static int t4_num_vis = 1;
470 TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
471 /*
472  * PCIe Relaxed Ordering.
473  * -1: driver should figure out a good value.
474  * 0: disable RO.
475  * 1: enable RO.
476  * 2: leave RO alone.
477  */
478 static int pcie_relaxed_ordering = -1;
479 TUNABLE_INT("hw.cxgbe.pcie_relaxed_ordering", &pcie_relaxed_ordering);
480 
481 static int t4_panic_on_fatal_err = 0;
482 TUNABLE_INT("hw.cxgbe.panic_on_fatal_err", &t4_panic_on_fatal_err);
483 
484 #ifdef TCP_OFFLOAD
485 /*
486  * TOE tunables.
487  */
488 static int t4_cop_managed_offloading = 0;
489 TUNABLE_INT("hw.cxgbe.cop_managed_offloading", &t4_cop_managed_offloading);
490 #endif
491 
492 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
493 static int vi_mac_funcs[] = {
494 	FW_VI_FUNC_ETH,
495 	FW_VI_FUNC_OFLD,
496 	FW_VI_FUNC_IWARP,
497 	FW_VI_FUNC_OPENISCSI,
498 	FW_VI_FUNC_OPENFCOE,
499 	FW_VI_FUNC_FOISCSI,
500 	FW_VI_FUNC_FOFCOE,
501 };
502 
503 struct intrs_and_queues {
504 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
505 	uint16_t num_vis;	/* number of VIs for each port */
506 	uint16_t nirq;		/* Total # of vectors */
507 	uint16_t ntxq;		/* # of NIC txq's for each port */
508 	uint16_t nrxq;		/* # of NIC rxq's for each port */
509 	uint16_t nofldtxq;	/* # of TOE/ETHOFLD txq's for each port */
510 	uint16_t nofldrxq;	/* # of TOE rxq's for each port */
511 
512 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
513 	uint16_t ntxq_vi;	/* # of NIC txq's */
514 	uint16_t nrxq_vi;	/* # of NIC rxq's */
515 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
516 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
517 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
518 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
519 };
520 
521 static void setup_memwin(struct adapter *);
522 static void position_memwin(struct adapter *, int, uint32_t);
523 static int validate_mem_range(struct adapter *, uint32_t, int);
524 static int fwmtype_to_hwmtype(int);
525 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
526     uint32_t *);
527 static int fixup_devlog_params(struct adapter *);
528 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
529 static int prep_firmware(struct adapter *);
530 static int partition_resources(struct adapter *, const struct firmware *,
531     const char *);
532 static int get_params__pre_init(struct adapter *);
533 static int get_params__post_init(struct adapter *);
534 static int set_params__post_init(struct adapter *);
535 static void t4_set_desc(struct adapter *);
536 static bool fixed_ifmedia(struct port_info *);
537 static void build_medialist(struct port_info *);
538 static void init_link_config(struct port_info *);
539 static int fixup_link_config(struct port_info *);
540 static int apply_link_config(struct port_info *);
541 static int cxgbe_init_synchronized(struct vi_info *);
542 static int cxgbe_uninit_synchronized(struct vi_info *);
543 static void quiesce_txq(struct adapter *, struct sge_txq *);
544 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
545 static void quiesce_iq(struct adapter *, struct sge_iq *);
546 static void quiesce_fl(struct adapter *, struct sge_fl *);
547 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
548     driver_intr_t *, void *, char *);
549 static int t4_free_irq(struct adapter *, struct irq *);
550 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
551 static void vi_refresh_stats(struct adapter *, struct vi_info *);
552 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
553 static void cxgbe_tick(void *);
554 static void cxgbe_sysctls(struct port_info *);
555 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
556 static int sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS);
557 static int sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS);
558 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
559 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
560 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
561 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
562 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
563 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
564 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
565 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
566 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
567 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
568 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
569 static int sysctl_loadavg(SYSCTL_HANDLER_ARGS);
570 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
571 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
572 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
573 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
574 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
575 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
576 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
577 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
578 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
579 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
580 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
581 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
582 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
583 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
584 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
585 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
586 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
587 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
588 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
589 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
590 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
591 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
592 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
593 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
594 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
595 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
596 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
597 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
598 static int sysctl_cpus(SYSCTL_HANDLER_ARGS);
599 #ifdef TCP_OFFLOAD
600 static int sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS);
601 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
602 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
603 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
604 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
605 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
606 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
607 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
608 #endif
609 static int get_sge_context(struct adapter *, struct t4_sge_context *);
610 static int load_fw(struct adapter *, struct t4_data *);
611 static int load_cfg(struct adapter *, struct t4_data *);
612 static int load_boot(struct adapter *, struct t4_bootrom *);
613 static int load_bootcfg(struct adapter *, struct t4_data *);
614 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
615 static void free_offload_policy(struct t4_offload_policy *);
616 static int set_offload_policy(struct adapter *, struct t4_offload_policy *);
617 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
618 static int read_i2c(struct adapter *, struct t4_i2c_data *);
619 #ifdef TCP_OFFLOAD
620 static int toe_capability(struct vi_info *, int);
621 #endif
622 static int mod_event(module_t, int, void *);
623 static int notify_siblings(device_t, int);
624 
625 struct {
626 	uint16_t device;
627 	char *desc;
628 } t4_pciids[] = {
629 	{0xa000, "Chelsio Terminator 4 FPGA"},
630 	{0x4400, "Chelsio T440-dbg"},
631 	{0x4401, "Chelsio T420-CR"},
632 	{0x4402, "Chelsio T422-CR"},
633 	{0x4403, "Chelsio T440-CR"},
634 	{0x4404, "Chelsio T420-BCH"},
635 	{0x4405, "Chelsio T440-BCH"},
636 	{0x4406, "Chelsio T440-CH"},
637 	{0x4407, "Chelsio T420-SO"},
638 	{0x4408, "Chelsio T420-CX"},
639 	{0x4409, "Chelsio T420-BT"},
640 	{0x440a, "Chelsio T404-BT"},
641 	{0x440e, "Chelsio T440-LP-CR"},
642 }, t5_pciids[] = {
643 	{0xb000, "Chelsio Terminator 5 FPGA"},
644 	{0x5400, "Chelsio T580-dbg"},
645 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
646 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
647 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
648 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
649 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
650 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
651 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
652 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
653 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
654 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
655 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
656 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
657 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
658 	{0x5418,  "Chelsio T540-BT"},		/* 4 x 10GBaseT */
659 	{0x5419,  "Chelsio T540-LP-BT"},	/* 4 x 10GBaseT */
660 	{0x541a,  "Chelsio T540-SO-BT"},	/* 4 x 10GBaseT, nomem */
661 	{0x541b,  "Chelsio T540-SO-CR"},	/* 4 x 10G, nomem */
662 }, t6_pciids[] = {
663 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
664 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
665 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
666 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
667 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
668 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
669 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
670 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
671 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
672 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
673 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
674 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
675 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
676 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
677 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
678 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
679 
680 	/* Custom */
681 	{0x6480, "Custom T6225-CR"},
682 	{0x6481, "Custom T62100-CR"},
683 	{0x6482, "Custom T6225-CR"},
684 	{0x6483, "Custom T62100-CR"},
685 	{0x6484, "Custom T64100-CR"},
686 	{0x6485, "Custom T6240-SO"},
687 	{0x6486, "Custom T6225-SO-CR"},
688 	{0x6487, "Custom T6225-CR"},
689 };
690 
691 #ifdef TCP_OFFLOAD
692 /*
693  * service_iq_fl() has an iq and needs the fl.  Offset of fl from the iq should
694  * be exactly the same for both rxq and ofld_rxq.
695  */
696 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
697 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
698 #endif
699 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
700 
701 static int
702 t4_probe(device_t dev)
703 {
704 	int i;
705 	uint16_t v = pci_get_vendor(dev);
706 	uint16_t d = pci_get_device(dev);
707 	uint8_t f = pci_get_function(dev);
708 
709 	if (v != PCI_VENDOR_ID_CHELSIO)
710 		return (ENXIO);
711 
712 	/* Attach only to PF0 of the FPGA */
713 	if (d == 0xa000 && f != 0)
714 		return (ENXIO);
715 
716 	for (i = 0; i < nitems(t4_pciids); i++) {
717 		if (d == t4_pciids[i].device) {
718 			device_set_desc(dev, t4_pciids[i].desc);
719 			return (BUS_PROBE_DEFAULT);
720 		}
721 	}
722 
723 	return (ENXIO);
724 }
725 
726 static int
727 t5_probe(device_t dev)
728 {
729 	int i;
730 	uint16_t v = pci_get_vendor(dev);
731 	uint16_t d = pci_get_device(dev);
732 	uint8_t f = pci_get_function(dev);
733 
734 	if (v != PCI_VENDOR_ID_CHELSIO)
735 		return (ENXIO);
736 
737 	/* Attach only to PF0 of the FPGA */
738 	if (d == 0xb000 && f != 0)
739 		return (ENXIO);
740 
741 	for (i = 0; i < nitems(t5_pciids); i++) {
742 		if (d == t5_pciids[i].device) {
743 			device_set_desc(dev, t5_pciids[i].desc);
744 			return (BUS_PROBE_DEFAULT);
745 		}
746 	}
747 
748 	return (ENXIO);
749 }
750 
751 static int
752 t6_probe(device_t dev)
753 {
754 	int i;
755 	uint16_t v = pci_get_vendor(dev);
756 	uint16_t d = pci_get_device(dev);
757 
758 	if (v != PCI_VENDOR_ID_CHELSIO)
759 		return (ENXIO);
760 
761 	for (i = 0; i < nitems(t6_pciids); i++) {
762 		if (d == t6_pciids[i].device) {
763 			device_set_desc(dev, t6_pciids[i].desc);
764 			return (BUS_PROBE_DEFAULT);
765 		}
766 	}
767 
768 	return (ENXIO);
769 }
770 
771 static void
772 t5_attribute_workaround(device_t dev)
773 {
774 	device_t root_port;
775 	uint32_t v;
776 
777 	/*
778 	 * The T5 chips do not properly echo the No Snoop and Relaxed
779 	 * Ordering attributes when replying to a TLP from a Root
780 	 * Port.  As a workaround, find the parent Root Port and
781 	 * disable No Snoop and Relaxed Ordering.  Note that this
782 	 * affects all devices under this root port.
783 	 */
784 	root_port = pci_find_pcie_root_port(dev);
785 	if (root_port == NULL) {
786 		device_printf(dev, "Unable to find parent root port\n");
787 		return;
788 	}
789 
790 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
791 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
792 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
793 	    0)
794 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
795 		    device_get_nameunit(root_port));
796 }
797 
798 static const struct devnames devnames[] = {
799 	{
800 		.nexus_name = "t4nex",
801 		.ifnet_name = "cxgbe",
802 		.vi_ifnet_name = "vcxgbe",
803 		.pf03_drv_name = "t4iov",
804 		.vf_nexus_name = "t4vf",
805 		.vf_ifnet_name = "cxgbev"
806 	}, {
807 		.nexus_name = "t5nex",
808 		.ifnet_name = "cxl",
809 		.vi_ifnet_name = "vcxl",
810 		.pf03_drv_name = "t5iov",
811 		.vf_nexus_name = "t5vf",
812 		.vf_ifnet_name = "cxlv"
813 	}, {
814 		.nexus_name = "t6nex",
815 		.ifnet_name = "cc",
816 		.vi_ifnet_name = "vcc",
817 		.pf03_drv_name = "t6iov",
818 		.vf_nexus_name = "t6vf",
819 		.vf_ifnet_name = "ccv"
820 	}
821 };
822 
823 void
824 t4_init_devnames(struct adapter *sc)
825 {
826 	int id;
827 
828 	id = chip_id(sc);
829 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
830 		sc->names = &devnames[id - CHELSIO_T4];
831 	else {
832 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
833 		sc->names = NULL;
834 	}
835 }
836 
837 static int
838 t4_attach(device_t dev)
839 {
840 	struct adapter *sc;
841 	int rc = 0, i, j, rqidx, tqidx, nports;
842 	struct make_dev_args mda;
843 	struct intrs_and_queues iaq;
844 	struct sge *s;
845 	uint32_t *buf;
846 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
847 	int ofld_tqidx;
848 #endif
849 #ifdef TCP_OFFLOAD
850 	int ofld_rqidx;
851 #endif
852 #ifdef DEV_NETMAP
853 	int nm_rqidx, nm_tqidx;
854 #endif
855 	int num_vis;
856 
857 	sc = device_get_softc(dev);
858 	sc->dev = dev;
859 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
860 
861 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
862 		t5_attribute_workaround(dev);
863 	pci_enable_busmaster(dev);
864 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
865 		uint32_t v;
866 
867 		pci_set_max_read_req(dev, 4096);
868 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
869 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
870 		if (pcie_relaxed_ordering == 0 &&
871 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
872 			v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
873 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
874 		} else if (pcie_relaxed_ordering == 1 &&
875 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
876 			v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
877 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
878 		}
879 	}
880 
881 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
882 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
883 	sc->traceq = -1;
884 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
885 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
886 	    device_get_nameunit(dev));
887 
888 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
889 	    device_get_nameunit(dev));
890 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
891 	t4_add_adapter(sc);
892 
893 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
894 	TAILQ_INIT(&sc->sfl);
895 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
896 
897 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
898 
899 	sc->policy = NULL;
900 	rw_init(&sc->policy_lock, "connection offload policy");
901 
902 	rc = t4_map_bars_0_and_4(sc);
903 	if (rc != 0)
904 		goto done; /* error message displayed already */
905 
906 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
907 
908 	/* Prepare the adapter for operation. */
909 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
910 	rc = -t4_prep_adapter(sc, buf);
911 	free(buf, M_CXGBE);
912 	if (rc != 0) {
913 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
914 		goto done;
915 	}
916 
917 	/*
918 	 * This is the real PF# to which we're attaching.  Works from within PCI
919 	 * passthrough environments too, where pci_get_function() could return a
920 	 * different PF# depending on the passthrough configuration.  We need to
921 	 * use the real PF# in all our communication with the firmware.
922 	 */
923 	j = t4_read_reg(sc, A_PL_WHOAMI);
924 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
925 	sc->mbox = sc->pf;
926 
927 	t4_init_devnames(sc);
928 	if (sc->names == NULL) {
929 		rc = ENOTSUP;
930 		goto done; /* error message displayed already */
931 	}
932 
933 	/*
934 	 * Do this really early, with the memory windows set up even before the
935 	 * character device.  The userland tool's register i/o and mem read
936 	 * will work even in "recovery mode".
937 	 */
938 	setup_memwin(sc);
939 	if (t4_init_devlog_params(sc, 0) == 0)
940 		fixup_devlog_params(sc);
941 	make_dev_args_init(&mda);
942 	mda.mda_devsw = &t4_cdevsw;
943 	mda.mda_uid = UID_ROOT;
944 	mda.mda_gid = GID_WHEEL;
945 	mda.mda_mode = 0600;
946 	mda.mda_si_drv1 = sc;
947 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
948 	if (rc != 0)
949 		device_printf(dev, "failed to create nexus char device: %d.\n",
950 		    rc);
951 
952 	/* Go no further if recovery mode has been requested. */
953 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
954 		device_printf(dev, "recovery mode.\n");
955 		goto done;
956 	}
957 
958 #if defined(__i386__)
959 	if ((cpu_feature & CPUID_CX8) == 0) {
960 		device_printf(dev, "64 bit atomics not available.\n");
961 		rc = ENOTSUP;
962 		goto done;
963 	}
964 #endif
965 
966 	/* Prepare the firmware for operation */
967 	rc = prep_firmware(sc);
968 	if (rc != 0)
969 		goto done; /* error message displayed already */
970 
971 	rc = get_params__post_init(sc);
972 	if (rc != 0)
973 		goto done; /* error message displayed already */
974 
975 	rc = set_params__post_init(sc);
976 	if (rc != 0)
977 		goto done; /* error message displayed already */
978 
979 	rc = t4_map_bar_2(sc);
980 	if (rc != 0)
981 		goto done; /* error message displayed already */
982 
983 	rc = t4_create_dma_tag(sc);
984 	if (rc != 0)
985 		goto done; /* error message displayed already */
986 
987 	/*
988 	 * First pass over all the ports - allocate VIs and initialize some
989 	 * basic parameters like mac address, port type, etc.
990 	 */
991 	for_each_port(sc, i) {
992 		struct port_info *pi;
993 
994 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
995 		sc->port[i] = pi;
996 
997 		/* These must be set before t4_port_init */
998 		pi->adapter = sc;
999 		pi->port_id = i;
1000 		/*
1001 		 * XXX: vi[0] is special so we can't delay this allocation until
1002 		 * pi->nvi's final value is known.
1003 		 */
1004 		pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
1005 		    M_ZERO | M_WAITOK);
1006 
1007 		/*
1008 		 * Allocate the "main" VI and initialize parameters
1009 		 * like mac addr.
1010 		 */
1011 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1012 		if (rc != 0) {
1013 			device_printf(dev, "unable to initialize port %d: %d\n",
1014 			    i, rc);
1015 			free(pi->vi, M_CXGBE);
1016 			free(pi, M_CXGBE);
1017 			sc->port[i] = NULL;
1018 			goto done;
1019 		}
1020 
1021 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1022 		    device_get_nameunit(dev), i);
1023 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1024 		sc->chan_map[pi->tx_chan] = i;
1025 
1026 		/* All VIs on this port share this media. */
1027 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1028 		    cxgbe_media_status);
1029 
1030 		PORT_LOCK(pi);
1031 		init_link_config(pi);
1032 		fixup_link_config(pi);
1033 		build_medialist(pi);
1034 		if (fixed_ifmedia(pi))
1035 			pi->flags |= FIXED_IFMEDIA;
1036 		PORT_UNLOCK(pi);
1037 
1038 		pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
1039 		if (pi->dev == NULL) {
1040 			device_printf(dev,
1041 			    "failed to add device for port %d.\n", i);
1042 			rc = ENXIO;
1043 			goto done;
1044 		}
1045 		pi->vi[0].dev = pi->dev;
1046 		device_set_softc(pi->dev, pi);
1047 	}
1048 
1049 	/*
1050 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1051 	 */
1052 	nports = sc->params.nports;
1053 	rc = cfg_itype_and_nqueues(sc, &iaq);
1054 	if (rc != 0)
1055 		goto done; /* error message displayed already */
1056 
1057 	num_vis = iaq.num_vis;
1058 	sc->intr_type = iaq.intr_type;
1059 	sc->intr_count = iaq.nirq;
1060 
1061 	s = &sc->sge;
1062 	s->nrxq = nports * iaq.nrxq;
1063 	s->ntxq = nports * iaq.ntxq;
1064 	if (num_vis > 1) {
1065 		s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1066 		s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1067 	}
1068 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1069 	s->neq += nports;		/* ctrl queues: 1 per port */
1070 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1071 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1072 	if (is_offload(sc) || is_ethoffload(sc)) {
1073 		s->nofldtxq = nports * iaq.nofldtxq;
1074 		if (num_vis > 1)
1075 			s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1076 		s->neq += s->nofldtxq;
1077 
1078 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1079 		    M_CXGBE, M_ZERO | M_WAITOK);
1080 	}
1081 #endif
1082 #ifdef TCP_OFFLOAD
1083 	if (is_offload(sc)) {
1084 		s->nofldrxq = nports * iaq.nofldrxq;
1085 		if (num_vis > 1)
1086 			s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1087 		s->neq += s->nofldrxq;	/* free list */
1088 		s->niq += s->nofldrxq;
1089 
1090 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1091 		    M_CXGBE, M_ZERO | M_WAITOK);
1092 	}
1093 #endif
1094 #ifdef DEV_NETMAP
1095 	if (num_vis > 1) {
1096 		s->nnmrxq = nports * (num_vis - 1) * iaq.nnmrxq_vi;
1097 		s->nnmtxq = nports * (num_vis - 1) * iaq.nnmtxq_vi;
1098 	}
1099 	s->neq += s->nnmtxq + s->nnmrxq;
1100 	s->niq += s->nnmrxq;
1101 
1102 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1103 	    M_CXGBE, M_ZERO | M_WAITOK);
1104 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1105 	    M_CXGBE, M_ZERO | M_WAITOK);
1106 #endif
1107 
1108 	s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1109 	    M_ZERO | M_WAITOK);
1110 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1111 	    M_ZERO | M_WAITOK);
1112 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1113 	    M_ZERO | M_WAITOK);
1114 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1115 	    M_ZERO | M_WAITOK);
1116 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1117 	    M_ZERO | M_WAITOK);
1118 
1119 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1120 	    M_ZERO | M_WAITOK);
1121 
1122 	t4_init_l2t(sc, M_WAITOK);
1123 	t4_init_smt(sc, M_WAITOK);
1124 	t4_init_tx_sched(sc);
1125 #ifdef RATELIMIT
1126 	t4_init_etid_table(sc);
1127 #endif
1128 
1129 	/*
1130 	 * Second pass over the ports.  This time we know the number of rx and
1131 	 * tx queues that each port should get.
1132 	 */
1133 	rqidx = tqidx = 0;
1134 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1135 	ofld_tqidx = 0;
1136 #endif
1137 #ifdef TCP_OFFLOAD
1138 	ofld_rqidx = 0;
1139 #endif
1140 #ifdef DEV_NETMAP
1141 	nm_rqidx = nm_tqidx = 0;
1142 #endif
1143 	for_each_port(sc, i) {
1144 		struct port_info *pi = sc->port[i];
1145 		struct vi_info *vi;
1146 
1147 		if (pi == NULL)
1148 			continue;
1149 
1150 		pi->nvi = num_vis;
1151 		for_each_vi(pi, j, vi) {
1152 			vi->pi = pi;
1153 			vi->qsize_rxq = t4_qsize_rxq;
1154 			vi->qsize_txq = t4_qsize_txq;
1155 
1156 			vi->first_rxq = rqidx;
1157 			vi->first_txq = tqidx;
1158 			vi->tmr_idx = t4_tmr_idx;
1159 			vi->pktc_idx = t4_pktc_idx;
1160 			vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1161 			vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1162 
1163 			rqidx += vi->nrxq;
1164 			tqidx += vi->ntxq;
1165 
1166 			if (j == 0 && vi->ntxq > 1)
1167 				vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1168 			else
1169 				vi->rsrv_noflowq = 0;
1170 
1171 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1172 			vi->first_ofld_txq = ofld_tqidx;
1173 			vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1174 			ofld_tqidx += vi->nofldtxq;
1175 #endif
1176 #ifdef TCP_OFFLOAD
1177 			vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1178 			vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1179 			vi->first_ofld_rxq = ofld_rqidx;
1180 			vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1181 
1182 			ofld_rqidx += vi->nofldrxq;
1183 #endif
1184 #ifdef DEV_NETMAP
1185 			if (j > 0) {
1186 				vi->first_nm_rxq = nm_rqidx;
1187 				vi->first_nm_txq = nm_tqidx;
1188 				vi->nnmrxq = iaq.nnmrxq_vi;
1189 				vi->nnmtxq = iaq.nnmtxq_vi;
1190 				nm_rqidx += vi->nnmrxq;
1191 				nm_tqidx += vi->nnmtxq;
1192 			}
1193 #endif
1194 		}
1195 	}
1196 
1197 	rc = t4_setup_intr_handlers(sc);
1198 	if (rc != 0) {
1199 		device_printf(dev,
1200 		    "failed to setup interrupt handlers: %d\n", rc);
1201 		goto done;
1202 	}
1203 
1204 	rc = bus_generic_probe(dev);
1205 	if (rc != 0) {
1206 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1207 		goto done;
1208 	}
1209 
1210 	/*
1211 	 * Ensure thread-safe mailbox access (in debug builds).
1212 	 *
1213 	 * So far this was the only thread accessing the mailbox but various
1214 	 * ifnets and sysctls are about to be created and their handlers/ioctls
1215 	 * will access the mailbox from different threads.
1216 	 */
1217 	sc->flags |= CHK_MBOX_ACCESS;
1218 
1219 	rc = bus_generic_attach(dev);
1220 	if (rc != 0) {
1221 		device_printf(dev,
1222 		    "failed to attach all child ports: %d\n", rc);
1223 		goto done;
1224 	}
1225 
1226 	device_printf(dev,
1227 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1228 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1229 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1230 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1231 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1232 
1233 	t4_set_desc(sc);
1234 
1235 	notify_siblings(dev, 0);
1236 
1237 done:
1238 	if (rc != 0 && sc->cdev) {
1239 		/* cdev was created and so cxgbetool works; recover that way. */
1240 		device_printf(dev,
1241 		    "error during attach, adapter is now in recovery mode.\n");
1242 		rc = 0;
1243 	}
1244 
1245 	if (rc != 0)
1246 		t4_detach_common(dev);
1247 	else
1248 		t4_sysctls(sc);
1249 
1250 	return (rc);
1251 }
1252 
1253 static int
1254 t4_ready(device_t dev)
1255 {
1256 	struct adapter *sc;
1257 
1258 	sc = device_get_softc(dev);
1259 	if (sc->flags & FW_OK)
1260 		return (0);
1261 	return (ENXIO);
1262 }
1263 
1264 static int
1265 t4_read_port_device(device_t dev, int port, device_t *child)
1266 {
1267 	struct adapter *sc;
1268 	struct port_info *pi;
1269 
1270 	sc = device_get_softc(dev);
1271 	if (port < 0 || port >= MAX_NPORTS)
1272 		return (EINVAL);
1273 	pi = sc->port[port];
1274 	if (pi == NULL || pi->dev == NULL)
1275 		return (ENXIO);
1276 	*child = pi->dev;
1277 	return (0);
1278 }
1279 
1280 static int
1281 notify_siblings(device_t dev, int detaching)
1282 {
1283 	device_t sibling;
1284 	int error, i;
1285 
1286 	error = 0;
1287 	for (i = 0; i < PCI_FUNCMAX; i++) {
1288 		if (i == pci_get_function(dev))
1289 			continue;
1290 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1291 		    pci_get_slot(dev), i);
1292 		if (sibling == NULL || !device_is_attached(sibling))
1293 			continue;
1294 		if (detaching)
1295 			error = T4_DETACH_CHILD(sibling);
1296 		else
1297 			(void)T4_ATTACH_CHILD(sibling);
1298 		if (error)
1299 			break;
1300 	}
1301 	return (error);
1302 }
1303 
1304 /*
1305  * Idempotent
1306  */
1307 static int
1308 t4_detach(device_t dev)
1309 {
1310 	struct adapter *sc;
1311 	int rc;
1312 
1313 	sc = device_get_softc(dev);
1314 
1315 	rc = notify_siblings(dev, 1);
1316 	if (rc) {
1317 		device_printf(dev,
1318 		    "failed to detach sibling devices: %d\n", rc);
1319 		return (rc);
1320 	}
1321 
1322 	return (t4_detach_common(dev));
1323 }
1324 
1325 int
1326 t4_detach_common(device_t dev)
1327 {
1328 	struct adapter *sc;
1329 	struct port_info *pi;
1330 	int i, rc;
1331 
1332 	sc = device_get_softc(dev);
1333 
1334 	if (sc->cdev) {
1335 		destroy_dev(sc->cdev);
1336 		sc->cdev = NULL;
1337 	}
1338 
1339 	sc->flags &= ~CHK_MBOX_ACCESS;
1340 	if (sc->flags & FULL_INIT_DONE) {
1341 		if (!(sc->flags & IS_VF))
1342 			t4_intr_disable(sc);
1343 	}
1344 
1345 	if (device_is_attached(dev)) {
1346 		rc = bus_generic_detach(dev);
1347 		if (rc) {
1348 			device_printf(dev,
1349 			    "failed to detach child devices: %d\n", rc);
1350 			return (rc);
1351 		}
1352 	}
1353 
1354 	for (i = 0; i < sc->intr_count; i++)
1355 		t4_free_irq(sc, &sc->irq[i]);
1356 
1357 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1358 		t4_free_tx_sched(sc);
1359 
1360 	for (i = 0; i < MAX_NPORTS; i++) {
1361 		pi = sc->port[i];
1362 		if (pi) {
1363 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1364 			if (pi->dev)
1365 				device_delete_child(dev, pi->dev);
1366 
1367 			mtx_destroy(&pi->pi_lock);
1368 			free(pi->vi, M_CXGBE);
1369 			free(pi, M_CXGBE);
1370 		}
1371 	}
1372 
1373 	device_delete_children(dev);
1374 
1375 	if (sc->flags & FULL_INIT_DONE)
1376 		adapter_full_uninit(sc);
1377 
1378 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1379 		t4_fw_bye(sc, sc->mbox);
1380 
1381 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1382 		pci_release_msi(dev);
1383 
1384 	if (sc->regs_res)
1385 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1386 		    sc->regs_res);
1387 
1388 	if (sc->udbs_res)
1389 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1390 		    sc->udbs_res);
1391 
1392 	if (sc->msix_res)
1393 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1394 		    sc->msix_res);
1395 
1396 	if (sc->l2t)
1397 		t4_free_l2t(sc->l2t);
1398 	if (sc->smt)
1399 		t4_free_smt(sc->smt);
1400 #ifdef RATELIMIT
1401 	t4_free_etid_table(sc);
1402 #endif
1403 
1404 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1405 	free(sc->sge.ofld_txq, M_CXGBE);
1406 #endif
1407 #ifdef TCP_OFFLOAD
1408 	free(sc->sge.ofld_rxq, M_CXGBE);
1409 #endif
1410 #ifdef DEV_NETMAP
1411 	free(sc->sge.nm_rxq, M_CXGBE);
1412 	free(sc->sge.nm_txq, M_CXGBE);
1413 #endif
1414 	free(sc->irq, M_CXGBE);
1415 	free(sc->sge.rxq, M_CXGBE);
1416 	free(sc->sge.txq, M_CXGBE);
1417 	free(sc->sge.ctrlq, M_CXGBE);
1418 	free(sc->sge.iqmap, M_CXGBE);
1419 	free(sc->sge.eqmap, M_CXGBE);
1420 	free(sc->tids.ftid_tab, M_CXGBE);
1421 	free(sc->tids.hpftid_tab, M_CXGBE);
1422 	free_hftid_hash(&sc->tids);
1423 	free(sc->tids.atid_tab, M_CXGBE);
1424 	free(sc->tids.tid_tab, M_CXGBE);
1425 	free(sc->tt.tls_rx_ports, M_CXGBE);
1426 	t4_destroy_dma_tag(sc);
1427 	if (mtx_initialized(&sc->sc_lock)) {
1428 		sx_xlock(&t4_list_lock);
1429 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1430 		sx_xunlock(&t4_list_lock);
1431 		mtx_destroy(&sc->sc_lock);
1432 	}
1433 
1434 	callout_drain(&sc->sfl_callout);
1435 	if (mtx_initialized(&sc->tids.ftid_lock)) {
1436 		mtx_destroy(&sc->tids.ftid_lock);
1437 		cv_destroy(&sc->tids.ftid_cv);
1438 	}
1439 	if (mtx_initialized(&sc->tids.atid_lock))
1440 		mtx_destroy(&sc->tids.atid_lock);
1441 	if (mtx_initialized(&sc->sfl_lock))
1442 		mtx_destroy(&sc->sfl_lock);
1443 	if (mtx_initialized(&sc->ifp_lock))
1444 		mtx_destroy(&sc->ifp_lock);
1445 	if (mtx_initialized(&sc->reg_lock))
1446 		mtx_destroy(&sc->reg_lock);
1447 
1448 	if (rw_initialized(&sc->policy_lock)) {
1449 		rw_destroy(&sc->policy_lock);
1450 #ifdef TCP_OFFLOAD
1451 		if (sc->policy != NULL)
1452 			free_offload_policy(sc->policy);
1453 #endif
1454 	}
1455 
1456 	for (i = 0; i < NUM_MEMWIN; i++) {
1457 		struct memwin *mw = &sc->memwin[i];
1458 
1459 		if (rw_initialized(&mw->mw_lock))
1460 			rw_destroy(&mw->mw_lock);
1461 	}
1462 
1463 	bzero(sc, sizeof(*sc));
1464 
1465 	return (0);
1466 }
1467 
1468 static int
1469 cxgbe_probe(device_t dev)
1470 {
1471 	char buf[128];
1472 	struct port_info *pi = device_get_softc(dev);
1473 
1474 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1475 	device_set_desc_copy(dev, buf);
1476 
1477 	return (BUS_PROBE_DEFAULT);
1478 }
1479 
1480 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1481     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1482     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS | \
1483     IFCAP_HWRXTSTMP)
1484 #define T4_CAP_ENABLE (T4_CAP)
1485 
1486 static int
1487 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1488 {
1489 	struct ifnet *ifp;
1490 	struct sbuf *sb;
1491 
1492 	vi->xact_addr_filt = -1;
1493 	callout_init(&vi->tick, 1);
1494 
1495 	/* Allocate an ifnet and set it up */
1496 	ifp = if_alloc(IFT_ETHER);
1497 	if (ifp == NULL) {
1498 		device_printf(dev, "Cannot allocate ifnet\n");
1499 		return (ENOMEM);
1500 	}
1501 	vi->ifp = ifp;
1502 	ifp->if_softc = vi;
1503 
1504 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1505 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1506 
1507 	ifp->if_init = cxgbe_init;
1508 	ifp->if_ioctl = cxgbe_ioctl;
1509 	ifp->if_transmit = cxgbe_transmit;
1510 	ifp->if_qflush = cxgbe_qflush;
1511 	ifp->if_get_counter = cxgbe_get_counter;
1512 #ifdef RATELIMIT
1513 	ifp->if_snd_tag_alloc = cxgbe_snd_tag_alloc;
1514 	ifp->if_snd_tag_modify = cxgbe_snd_tag_modify;
1515 	ifp->if_snd_tag_query = cxgbe_snd_tag_query;
1516 	ifp->if_snd_tag_free = cxgbe_snd_tag_free;
1517 #endif
1518 
1519 	ifp->if_capabilities = T4_CAP;
1520 	ifp->if_capenable = T4_CAP_ENABLE;
1521 #ifdef TCP_OFFLOAD
1522 	if (vi->nofldrxq != 0)
1523 		ifp->if_capabilities |= IFCAP_TOE;
1524 #endif
1525 #ifdef DEV_NETMAP
1526 	if (vi->nnmrxq != 0)
1527 		ifp->if_capabilities |= IFCAP_NETMAP;
1528 #endif
1529 #ifdef RATELIMIT
1530 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0) {
1531 		ifp->if_capabilities |= IFCAP_TXRTLMT;
1532 		ifp->if_capenable |= IFCAP_TXRTLMT;
1533 	}
1534 #endif
1535 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1536 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1537 
1538 	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1539 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1540 	ifp->if_hw_tsomaxsegsize = 65536;
1541 
1542 	ether_ifattach(ifp, vi->hw_addr);
1543 #ifdef DEV_NETMAP
1544 	if (ifp->if_capabilities & IFCAP_NETMAP)
1545 		cxgbe_nm_attach(vi);
1546 #endif
1547 	sb = sbuf_new_auto();
1548 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1549 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1550 	switch (ifp->if_capabilities & (IFCAP_TOE | IFCAP_TXRTLMT)) {
1551 	case IFCAP_TOE:
1552 		sbuf_printf(sb, "; %d txq (TOE)", vi->nofldtxq);
1553 		break;
1554 	case IFCAP_TOE | IFCAP_TXRTLMT:
1555 		sbuf_printf(sb, "; %d txq (TOE/ETHOFLD)", vi->nofldtxq);
1556 		break;
1557 	case IFCAP_TXRTLMT:
1558 		sbuf_printf(sb, "; %d txq (ETHOFLD)", vi->nofldtxq);
1559 		break;
1560 	}
1561 #endif
1562 #ifdef TCP_OFFLOAD
1563 	if (ifp->if_capabilities & IFCAP_TOE)
1564 		sbuf_printf(sb, ", %d rxq (TOE)", vi->nofldrxq);
1565 #endif
1566 #ifdef DEV_NETMAP
1567 	if (ifp->if_capabilities & IFCAP_NETMAP)
1568 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1569 		    vi->nnmtxq, vi->nnmrxq);
1570 #endif
1571 	sbuf_finish(sb);
1572 	device_printf(dev, "%s\n", sbuf_data(sb));
1573 	sbuf_delete(sb);
1574 
1575 	vi_sysctls(vi);
1576 
1577 	return (0);
1578 }
1579 
1580 static int
1581 cxgbe_attach(device_t dev)
1582 {
1583 	struct port_info *pi = device_get_softc(dev);
1584 	struct adapter *sc = pi->adapter;
1585 	struct vi_info *vi;
1586 	int i, rc;
1587 
1588 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1589 
1590 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1591 	if (rc)
1592 		return (rc);
1593 
1594 	for_each_vi(pi, i, vi) {
1595 		if (i == 0)
1596 			continue;
1597 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1598 		if (vi->dev == NULL) {
1599 			device_printf(dev, "failed to add VI %d\n", i);
1600 			continue;
1601 		}
1602 		device_set_softc(vi->dev, vi);
1603 	}
1604 
1605 	cxgbe_sysctls(pi);
1606 
1607 	bus_generic_attach(dev);
1608 
1609 	return (0);
1610 }
1611 
1612 static void
1613 cxgbe_vi_detach(struct vi_info *vi)
1614 {
1615 	struct ifnet *ifp = vi->ifp;
1616 
1617 	ether_ifdetach(ifp);
1618 
1619 	/* Let detach proceed even if these fail. */
1620 #ifdef DEV_NETMAP
1621 	if (ifp->if_capabilities & IFCAP_NETMAP)
1622 		cxgbe_nm_detach(vi);
1623 #endif
1624 	cxgbe_uninit_synchronized(vi);
1625 	callout_drain(&vi->tick);
1626 	vi_full_uninit(vi);
1627 
1628 	if_free(vi->ifp);
1629 	vi->ifp = NULL;
1630 }
1631 
1632 static int
1633 cxgbe_detach(device_t dev)
1634 {
1635 	struct port_info *pi = device_get_softc(dev);
1636 	struct adapter *sc = pi->adapter;
1637 	int rc;
1638 
1639 	/* Detach the extra VIs first. */
1640 	rc = bus_generic_detach(dev);
1641 	if (rc)
1642 		return (rc);
1643 	device_delete_children(dev);
1644 
1645 	doom_vi(sc, &pi->vi[0]);
1646 
1647 	if (pi->flags & HAS_TRACEQ) {
1648 		sc->traceq = -1;	/* cloner should not create ifnet */
1649 		t4_tracer_port_detach(sc);
1650 	}
1651 
1652 	cxgbe_vi_detach(&pi->vi[0]);
1653 	callout_drain(&pi->tick);
1654 	ifmedia_removeall(&pi->media);
1655 
1656 	end_synchronized_op(sc, 0);
1657 
1658 	return (0);
1659 }
1660 
1661 static void
1662 cxgbe_init(void *arg)
1663 {
1664 	struct vi_info *vi = arg;
1665 	struct adapter *sc = vi->pi->adapter;
1666 
1667 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1668 		return;
1669 	cxgbe_init_synchronized(vi);
1670 	end_synchronized_op(sc, 0);
1671 }
1672 
1673 static int
1674 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1675 {
1676 	int rc = 0, mtu, flags;
1677 	struct vi_info *vi = ifp->if_softc;
1678 	struct port_info *pi = vi->pi;
1679 	struct adapter *sc = pi->adapter;
1680 	struct ifreq *ifr = (struct ifreq *)data;
1681 	uint32_t mask;
1682 
1683 	switch (cmd) {
1684 	case SIOCSIFMTU:
1685 		mtu = ifr->ifr_mtu;
1686 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1687 			return (EINVAL);
1688 
1689 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1690 		if (rc)
1691 			return (rc);
1692 		ifp->if_mtu = mtu;
1693 		if (vi->flags & VI_INIT_DONE) {
1694 			t4_update_fl_bufsize(ifp);
1695 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1696 				rc = update_mac_settings(ifp, XGMAC_MTU);
1697 		}
1698 		end_synchronized_op(sc, 0);
1699 		break;
1700 
1701 	case SIOCSIFFLAGS:
1702 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4flg");
1703 		if (rc)
1704 			return (rc);
1705 
1706 		if (ifp->if_flags & IFF_UP) {
1707 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1708 				flags = vi->if_flags;
1709 				if ((ifp->if_flags ^ flags) &
1710 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1711 					rc = update_mac_settings(ifp,
1712 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1713 				}
1714 			} else {
1715 				rc = cxgbe_init_synchronized(vi);
1716 			}
1717 			vi->if_flags = ifp->if_flags;
1718 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1719 			rc = cxgbe_uninit_synchronized(vi);
1720 		}
1721 		end_synchronized_op(sc, 0);
1722 		break;
1723 
1724 	case SIOCADDMULTI:
1725 	case SIOCDELMULTI:
1726 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4multi");
1727 		if (rc)
1728 			return (rc);
1729 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1730 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1731 		end_synchronized_op(sc, 0);
1732 		break;
1733 
1734 	case SIOCSIFCAP:
1735 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1736 		if (rc)
1737 			return (rc);
1738 
1739 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1740 		if (mask & IFCAP_TXCSUM) {
1741 			ifp->if_capenable ^= IFCAP_TXCSUM;
1742 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1743 
1744 			if (IFCAP_TSO4 & ifp->if_capenable &&
1745 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1746 				ifp->if_capenable &= ~IFCAP_TSO4;
1747 				if_printf(ifp,
1748 				    "tso4 disabled due to -txcsum.\n");
1749 			}
1750 		}
1751 		if (mask & IFCAP_TXCSUM_IPV6) {
1752 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1753 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1754 
1755 			if (IFCAP_TSO6 & ifp->if_capenable &&
1756 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1757 				ifp->if_capenable &= ~IFCAP_TSO6;
1758 				if_printf(ifp,
1759 				    "tso6 disabled due to -txcsum6.\n");
1760 			}
1761 		}
1762 		if (mask & IFCAP_RXCSUM)
1763 			ifp->if_capenable ^= IFCAP_RXCSUM;
1764 		if (mask & IFCAP_RXCSUM_IPV6)
1765 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1766 
1767 		/*
1768 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1769 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1770 		 * sending a TSO request our way, so it's sufficient to toggle
1771 		 * IFCAP_TSOx only.
1772 		 */
1773 		if (mask & IFCAP_TSO4) {
1774 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1775 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1776 				if_printf(ifp, "enable txcsum first.\n");
1777 				rc = EAGAIN;
1778 				goto fail;
1779 			}
1780 			ifp->if_capenable ^= IFCAP_TSO4;
1781 		}
1782 		if (mask & IFCAP_TSO6) {
1783 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1784 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1785 				if_printf(ifp, "enable txcsum6 first.\n");
1786 				rc = EAGAIN;
1787 				goto fail;
1788 			}
1789 			ifp->if_capenable ^= IFCAP_TSO6;
1790 		}
1791 		if (mask & IFCAP_LRO) {
1792 #if defined(INET) || defined(INET6)
1793 			int i;
1794 			struct sge_rxq *rxq;
1795 
1796 			ifp->if_capenable ^= IFCAP_LRO;
1797 			for_each_rxq(vi, i, rxq) {
1798 				if (ifp->if_capenable & IFCAP_LRO)
1799 					rxq->iq.flags |= IQ_LRO_ENABLED;
1800 				else
1801 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1802 			}
1803 #endif
1804 		}
1805 #ifdef TCP_OFFLOAD
1806 		if (mask & IFCAP_TOE) {
1807 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1808 
1809 			rc = toe_capability(vi, enable);
1810 			if (rc != 0)
1811 				goto fail;
1812 
1813 			ifp->if_capenable ^= mask;
1814 		}
1815 #endif
1816 		if (mask & IFCAP_VLAN_HWTAGGING) {
1817 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1818 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1819 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1820 		}
1821 		if (mask & IFCAP_VLAN_MTU) {
1822 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1823 
1824 			/* Need to find out how to disable auto-mtu-inflation */
1825 		}
1826 		if (mask & IFCAP_VLAN_HWTSO)
1827 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1828 		if (mask & IFCAP_VLAN_HWCSUM)
1829 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1830 #ifdef RATELIMIT
1831 		if (mask & IFCAP_TXRTLMT)
1832 			ifp->if_capenable ^= IFCAP_TXRTLMT;
1833 #endif
1834 		if (mask & IFCAP_HWRXTSTMP) {
1835 			int i;
1836 			struct sge_rxq *rxq;
1837 
1838 			ifp->if_capenable ^= IFCAP_HWRXTSTMP;
1839 			for_each_rxq(vi, i, rxq) {
1840 				if (ifp->if_capenable & IFCAP_HWRXTSTMP)
1841 					rxq->iq.flags |= IQ_RX_TIMESTAMP;
1842 				else
1843 					rxq->iq.flags &= ~IQ_RX_TIMESTAMP;
1844 			}
1845 		}
1846 
1847 #ifdef VLAN_CAPABILITIES
1848 		VLAN_CAPABILITIES(ifp);
1849 #endif
1850 fail:
1851 		end_synchronized_op(sc, 0);
1852 		break;
1853 
1854 	case SIOCSIFMEDIA:
1855 	case SIOCGIFMEDIA:
1856 	case SIOCGIFXMEDIA:
1857 		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1858 		break;
1859 
1860 	case SIOCGI2C: {
1861 		struct ifi2creq i2c;
1862 
1863 		rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
1864 		if (rc != 0)
1865 			break;
1866 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1867 			rc = EPERM;
1868 			break;
1869 		}
1870 		if (i2c.len > sizeof(i2c.data)) {
1871 			rc = EINVAL;
1872 			break;
1873 		}
1874 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1875 		if (rc)
1876 			return (rc);
1877 		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
1878 		    i2c.offset, i2c.len, &i2c.data[0]);
1879 		end_synchronized_op(sc, 0);
1880 		if (rc == 0)
1881 			rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
1882 		break;
1883 	}
1884 
1885 	default:
1886 		rc = ether_ioctl(ifp, cmd, data);
1887 	}
1888 
1889 	return (rc);
1890 }
1891 
1892 static int
1893 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1894 {
1895 	struct vi_info *vi = ifp->if_softc;
1896 	struct port_info *pi = vi->pi;
1897 	struct adapter *sc = pi->adapter;
1898 	struct sge_txq *txq;
1899 	void *items[1];
1900 	int rc;
1901 
1902 	M_ASSERTPKTHDR(m);
1903 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1904 
1905 	if (__predict_false(pi->link_cfg.link_ok == false)) {
1906 		m_freem(m);
1907 		return (ENETDOWN);
1908 	}
1909 
1910 	rc = parse_pkt(sc, &m);
1911 	if (__predict_false(rc != 0)) {
1912 		MPASS(m == NULL);			/* was freed already */
1913 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1914 		return (rc);
1915 	}
1916 #ifdef RATELIMIT
1917 	if (m->m_pkthdr.snd_tag != NULL) {
1918 		/* EAGAIN tells the stack we are not the correct interface. */
1919 		if (__predict_false(ifp != m->m_pkthdr.snd_tag->ifp)) {
1920 			m_freem(m);
1921 			return (EAGAIN);
1922 		}
1923 
1924 		return (ethofld_transmit(ifp, m));
1925 	}
1926 #endif
1927 
1928 	/* Select a txq. */
1929 	txq = &sc->sge.txq[vi->first_txq];
1930 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1931 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1932 		    vi->rsrv_noflowq);
1933 
1934 	items[0] = m;
1935 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1936 	if (__predict_false(rc != 0))
1937 		m_freem(m);
1938 
1939 	return (rc);
1940 }
1941 
1942 static void
1943 cxgbe_qflush(struct ifnet *ifp)
1944 {
1945 	struct vi_info *vi = ifp->if_softc;
1946 	struct sge_txq *txq;
1947 	int i;
1948 
1949 	/* queues do not exist if !VI_INIT_DONE. */
1950 	if (vi->flags & VI_INIT_DONE) {
1951 		for_each_txq(vi, i, txq) {
1952 			TXQ_LOCK(txq);
1953 			txq->eq.flags |= EQ_QFLUSH;
1954 			TXQ_UNLOCK(txq);
1955 			while (!mp_ring_is_idle(txq->r)) {
1956 				mp_ring_check_drainage(txq->r, 0);
1957 				pause("qflush", 1);
1958 			}
1959 			TXQ_LOCK(txq);
1960 			txq->eq.flags &= ~EQ_QFLUSH;
1961 			TXQ_UNLOCK(txq);
1962 		}
1963 	}
1964 	if_qflush(ifp);
1965 }
1966 
1967 static uint64_t
1968 vi_get_counter(struct ifnet *ifp, ift_counter c)
1969 {
1970 	struct vi_info *vi = ifp->if_softc;
1971 	struct fw_vi_stats_vf *s = &vi->stats;
1972 
1973 	vi_refresh_stats(vi->pi->adapter, vi);
1974 
1975 	switch (c) {
1976 	case IFCOUNTER_IPACKETS:
1977 		return (s->rx_bcast_frames + s->rx_mcast_frames +
1978 		    s->rx_ucast_frames);
1979 	case IFCOUNTER_IERRORS:
1980 		return (s->rx_err_frames);
1981 	case IFCOUNTER_OPACKETS:
1982 		return (s->tx_bcast_frames + s->tx_mcast_frames +
1983 		    s->tx_ucast_frames + s->tx_offload_frames);
1984 	case IFCOUNTER_OERRORS:
1985 		return (s->tx_drop_frames);
1986 	case IFCOUNTER_IBYTES:
1987 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
1988 		    s->rx_ucast_bytes);
1989 	case IFCOUNTER_OBYTES:
1990 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
1991 		    s->tx_ucast_bytes + s->tx_offload_bytes);
1992 	case IFCOUNTER_IMCASTS:
1993 		return (s->rx_mcast_frames);
1994 	case IFCOUNTER_OMCASTS:
1995 		return (s->tx_mcast_frames);
1996 	case IFCOUNTER_OQDROPS: {
1997 		uint64_t drops;
1998 
1999 		drops = 0;
2000 		if (vi->flags & VI_INIT_DONE) {
2001 			int i;
2002 			struct sge_txq *txq;
2003 
2004 			for_each_txq(vi, i, txq)
2005 				drops += counter_u64_fetch(txq->r->drops);
2006 		}
2007 
2008 		return (drops);
2009 
2010 	}
2011 
2012 	default:
2013 		return (if_get_counter_default(ifp, c));
2014 	}
2015 }
2016 
2017 uint64_t
2018 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
2019 {
2020 	struct vi_info *vi = ifp->if_softc;
2021 	struct port_info *pi = vi->pi;
2022 	struct adapter *sc = pi->adapter;
2023 	struct port_stats *s = &pi->stats;
2024 
2025 	if (pi->nvi > 1 || sc->flags & IS_VF)
2026 		return (vi_get_counter(ifp, c));
2027 
2028 	cxgbe_refresh_stats(sc, pi);
2029 
2030 	switch (c) {
2031 	case IFCOUNTER_IPACKETS:
2032 		return (s->rx_frames);
2033 
2034 	case IFCOUNTER_IERRORS:
2035 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
2036 		    s->rx_fcs_err + s->rx_len_err);
2037 
2038 	case IFCOUNTER_OPACKETS:
2039 		return (s->tx_frames);
2040 
2041 	case IFCOUNTER_OERRORS:
2042 		return (s->tx_error_frames);
2043 
2044 	case IFCOUNTER_IBYTES:
2045 		return (s->rx_octets);
2046 
2047 	case IFCOUNTER_OBYTES:
2048 		return (s->tx_octets);
2049 
2050 	case IFCOUNTER_IMCASTS:
2051 		return (s->rx_mcast_frames);
2052 
2053 	case IFCOUNTER_OMCASTS:
2054 		return (s->tx_mcast_frames);
2055 
2056 	case IFCOUNTER_IQDROPS:
2057 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
2058 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
2059 		    s->rx_trunc3 + pi->tnl_cong_drops);
2060 
2061 	case IFCOUNTER_OQDROPS: {
2062 		uint64_t drops;
2063 
2064 		drops = s->tx_drop;
2065 		if (vi->flags & VI_INIT_DONE) {
2066 			int i;
2067 			struct sge_txq *txq;
2068 
2069 			for_each_txq(vi, i, txq)
2070 				drops += counter_u64_fetch(txq->r->drops);
2071 		}
2072 
2073 		return (drops);
2074 
2075 	}
2076 
2077 	default:
2078 		return (if_get_counter_default(ifp, c));
2079 	}
2080 }
2081 
2082 /*
2083  * The kernel picks a media from the list we had provided but we still validate
2084  * the requeste.
2085  */
2086 int
2087 cxgbe_media_change(struct ifnet *ifp)
2088 {
2089 	struct vi_info *vi = ifp->if_softc;
2090 	struct port_info *pi = vi->pi;
2091 	struct ifmedia *ifm = &pi->media;
2092 	struct link_config *lc = &pi->link_cfg;
2093 	struct adapter *sc = pi->adapter;
2094 	int rc;
2095 
2096 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mec");
2097 	if (rc != 0)
2098 		return (rc);
2099 	PORT_LOCK(pi);
2100 	if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
2101 		/* ifconfig .. media autoselect */
2102 		if (!(lc->supported & FW_PORT_CAP32_ANEG)) {
2103 			rc = ENOTSUP; /* AN not supported by transceiver */
2104 			goto done;
2105 		}
2106 		lc->requested_aneg = AUTONEG_ENABLE;
2107 		lc->requested_speed = 0;
2108 		lc->requested_fc |= PAUSE_AUTONEG;
2109 	} else {
2110 		lc->requested_aneg = AUTONEG_DISABLE;
2111 		lc->requested_speed =
2112 		    ifmedia_baudrate(ifm->ifm_media) / 1000000;
2113 		lc->requested_fc = 0;
2114 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
2115 			lc->requested_fc |= PAUSE_RX;
2116 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
2117 			lc->requested_fc |= PAUSE_TX;
2118 	}
2119 	if (pi->up_vis > 0) {
2120 		fixup_link_config(pi);
2121 		rc = apply_link_config(pi);
2122 	}
2123 done:
2124 	PORT_UNLOCK(pi);
2125 	end_synchronized_op(sc, 0);
2126 	return (rc);
2127 }
2128 
2129 /*
2130  * Base media word (without ETHER, pause, link active, etc.) for the port at the
2131  * given speed.
2132  */
2133 static int
2134 port_mword(struct port_info *pi, uint32_t speed)
2135 {
2136 
2137 	MPASS(speed & M_FW_PORT_CAP32_SPEED);
2138 	MPASS(powerof2(speed));
2139 
2140 	switch(pi->port_type) {
2141 	case FW_PORT_TYPE_BT_SGMII:
2142 	case FW_PORT_TYPE_BT_XFI:
2143 	case FW_PORT_TYPE_BT_XAUI:
2144 		/* BaseT */
2145 		switch (speed) {
2146 		case FW_PORT_CAP32_SPEED_100M:
2147 			return (IFM_100_T);
2148 		case FW_PORT_CAP32_SPEED_1G:
2149 			return (IFM_1000_T);
2150 		case FW_PORT_CAP32_SPEED_10G:
2151 			return (IFM_10G_T);
2152 		}
2153 		break;
2154 	case FW_PORT_TYPE_KX4:
2155 		if (speed == FW_PORT_CAP32_SPEED_10G)
2156 			return (IFM_10G_KX4);
2157 		break;
2158 	case FW_PORT_TYPE_CX4:
2159 		if (speed == FW_PORT_CAP32_SPEED_10G)
2160 			return (IFM_10G_CX4);
2161 		break;
2162 	case FW_PORT_TYPE_KX:
2163 		if (speed == FW_PORT_CAP32_SPEED_1G)
2164 			return (IFM_1000_KX);
2165 		break;
2166 	case FW_PORT_TYPE_KR:
2167 	case FW_PORT_TYPE_BP_AP:
2168 	case FW_PORT_TYPE_BP4_AP:
2169 	case FW_PORT_TYPE_BP40_BA:
2170 	case FW_PORT_TYPE_KR4_100G:
2171 	case FW_PORT_TYPE_KR_SFP28:
2172 	case FW_PORT_TYPE_KR_XLAUI:
2173 		switch (speed) {
2174 		case FW_PORT_CAP32_SPEED_1G:
2175 			return (IFM_1000_KX);
2176 		case FW_PORT_CAP32_SPEED_10G:
2177 			return (IFM_10G_KR);
2178 		case FW_PORT_CAP32_SPEED_25G:
2179 			return (IFM_25G_KR);
2180 		case FW_PORT_CAP32_SPEED_40G:
2181 			return (IFM_40G_KR4);
2182 		case FW_PORT_CAP32_SPEED_50G:
2183 			return (IFM_50G_KR2);
2184 		case FW_PORT_CAP32_SPEED_100G:
2185 			return (IFM_100G_KR4);
2186 		}
2187 		break;
2188 	case FW_PORT_TYPE_FIBER_XFI:
2189 	case FW_PORT_TYPE_FIBER_XAUI:
2190 	case FW_PORT_TYPE_SFP:
2191 	case FW_PORT_TYPE_QSFP_10G:
2192 	case FW_PORT_TYPE_QSA:
2193 	case FW_PORT_TYPE_QSFP:
2194 	case FW_PORT_TYPE_CR4_QSFP:
2195 	case FW_PORT_TYPE_CR_QSFP:
2196 	case FW_PORT_TYPE_CR2_QSFP:
2197 	case FW_PORT_TYPE_SFP28:
2198 		/* Pluggable transceiver */
2199 		switch (pi->mod_type) {
2200 		case FW_PORT_MOD_TYPE_LR:
2201 			switch (speed) {
2202 			case FW_PORT_CAP32_SPEED_1G:
2203 				return (IFM_1000_LX);
2204 			case FW_PORT_CAP32_SPEED_10G:
2205 				return (IFM_10G_LR);
2206 			case FW_PORT_CAP32_SPEED_25G:
2207 				return (IFM_25G_LR);
2208 			case FW_PORT_CAP32_SPEED_40G:
2209 				return (IFM_40G_LR4);
2210 			case FW_PORT_CAP32_SPEED_50G:
2211 				return (IFM_50G_LR2);
2212 			case FW_PORT_CAP32_SPEED_100G:
2213 				return (IFM_100G_LR4);
2214 			}
2215 			break;
2216 		case FW_PORT_MOD_TYPE_SR:
2217 			switch (speed) {
2218 			case FW_PORT_CAP32_SPEED_1G:
2219 				return (IFM_1000_SX);
2220 			case FW_PORT_CAP32_SPEED_10G:
2221 				return (IFM_10G_SR);
2222 			case FW_PORT_CAP32_SPEED_25G:
2223 				return (IFM_25G_SR);
2224 			case FW_PORT_CAP32_SPEED_40G:
2225 				return (IFM_40G_SR4);
2226 			case FW_PORT_CAP32_SPEED_50G:
2227 				return (IFM_50G_SR2);
2228 			case FW_PORT_CAP32_SPEED_100G:
2229 				return (IFM_100G_SR4);
2230 			}
2231 			break;
2232 		case FW_PORT_MOD_TYPE_ER:
2233 			if (speed == FW_PORT_CAP32_SPEED_10G)
2234 				return (IFM_10G_ER);
2235 			break;
2236 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2237 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2238 			switch (speed) {
2239 			case FW_PORT_CAP32_SPEED_1G:
2240 				return (IFM_1000_CX);
2241 			case FW_PORT_CAP32_SPEED_10G:
2242 				return (IFM_10G_TWINAX);
2243 			case FW_PORT_CAP32_SPEED_25G:
2244 				return (IFM_25G_CR);
2245 			case FW_PORT_CAP32_SPEED_40G:
2246 				return (IFM_40G_CR4);
2247 			case FW_PORT_CAP32_SPEED_50G:
2248 				return (IFM_50G_CR2);
2249 			case FW_PORT_CAP32_SPEED_100G:
2250 				return (IFM_100G_CR4);
2251 			}
2252 			break;
2253 		case FW_PORT_MOD_TYPE_LRM:
2254 			if (speed == FW_PORT_CAP32_SPEED_10G)
2255 				return (IFM_10G_LRM);
2256 			break;
2257 		case FW_PORT_MOD_TYPE_NA:
2258 			MPASS(0);	/* Not pluggable? */
2259 			/* fall throough */
2260 		case FW_PORT_MOD_TYPE_ERROR:
2261 		case FW_PORT_MOD_TYPE_UNKNOWN:
2262 		case FW_PORT_MOD_TYPE_NOTSUPPORTED:
2263 			break;
2264 		case FW_PORT_MOD_TYPE_NONE:
2265 			return (IFM_NONE);
2266 		}
2267 		break;
2268 	case FW_PORT_TYPE_NONE:
2269 		return (IFM_NONE);
2270 	}
2271 
2272 	return (IFM_UNKNOWN);
2273 }
2274 
2275 void
2276 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2277 {
2278 	struct vi_info *vi = ifp->if_softc;
2279 	struct port_info *pi = vi->pi;
2280 	struct adapter *sc = pi->adapter;
2281 	struct link_config *lc = &pi->link_cfg;
2282 
2283 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4med") != 0)
2284 		return;
2285 	PORT_LOCK(pi);
2286 
2287 	if (pi->up_vis == 0) {
2288 		/*
2289 		 * If all the interfaces are administratively down the firmware
2290 		 * does not report transceiver changes.  Refresh port info here
2291 		 * so that ifconfig displays accurate ifmedia at all times.
2292 		 * This is the only reason we have a synchronized op in this
2293 		 * function.  Just PORT_LOCK would have been enough otherwise.
2294 		 */
2295 		t4_update_port_info(pi);
2296 		build_medialist(pi);
2297 	}
2298 
2299 	/* ifm_status */
2300 	ifmr->ifm_status = IFM_AVALID;
2301 	if (lc->link_ok == false)
2302 		goto done;
2303 	ifmr->ifm_status |= IFM_ACTIVE;
2304 
2305 	/* ifm_active */
2306 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2307 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2308 	if (lc->fc & PAUSE_RX)
2309 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2310 	if (lc->fc & PAUSE_TX)
2311 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2312 	ifmr->ifm_active |= port_mword(pi, speed_to_fwcap(lc->speed));
2313 done:
2314 	PORT_UNLOCK(pi);
2315 	end_synchronized_op(sc, 0);
2316 }
2317 
2318 static int
2319 vcxgbe_probe(device_t dev)
2320 {
2321 	char buf[128];
2322 	struct vi_info *vi = device_get_softc(dev);
2323 
2324 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2325 	    vi - vi->pi->vi);
2326 	device_set_desc_copy(dev, buf);
2327 
2328 	return (BUS_PROBE_DEFAULT);
2329 }
2330 
2331 static int
2332 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2333 {
2334 	int func, index, rc;
2335 	uint32_t param, val;
2336 
2337 	ASSERT_SYNCHRONIZED_OP(sc);
2338 
2339 	index = vi - pi->vi;
2340 	MPASS(index > 0);	/* This function deals with _extra_ VIs only */
2341 	KASSERT(index < nitems(vi_mac_funcs),
2342 	    ("%s: VI %s doesn't have a MAC func", __func__,
2343 	    device_get_nameunit(vi->dev)));
2344 	func = vi_mac_funcs[index];
2345 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2346 	    vi->hw_addr, &vi->rss_size, func, 0);
2347 	if (rc < 0) {
2348 		device_printf(vi->dev, "failed to allocate virtual interface %d"
2349 		    "for port %d: %d\n", index, pi->port_id, -rc);
2350 		return (-rc);
2351 	}
2352 	vi->viid = rc;
2353 	if (chip_id(sc) <= CHELSIO_T5)
2354 		vi->smt_idx = (rc & 0x7f) << 1;
2355 	else
2356 		vi->smt_idx = (rc & 0x7f);
2357 
2358 	if (vi->rss_size == 1) {
2359 		/*
2360 		 * This VI didn't get a slice of the RSS table.  Reduce the
2361 		 * number of VIs being created (hw.cxgbe.num_vis) or modify the
2362 		 * configuration file (nvi, rssnvi for this PF) if this is a
2363 		 * problem.
2364 		 */
2365 		device_printf(vi->dev, "RSS table not available.\n");
2366 		vi->rss_base = 0xffff;
2367 
2368 		return (0);
2369 	}
2370 
2371 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2372 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2373 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2374 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2375 	if (rc)
2376 		vi->rss_base = 0xffff;
2377 	else {
2378 		MPASS((val >> 16) == vi->rss_size);
2379 		vi->rss_base = val & 0xffff;
2380 	}
2381 
2382 	return (0);
2383 }
2384 
2385 static int
2386 vcxgbe_attach(device_t dev)
2387 {
2388 	struct vi_info *vi;
2389 	struct port_info *pi;
2390 	struct adapter *sc;
2391 	int rc;
2392 
2393 	vi = device_get_softc(dev);
2394 	pi = vi->pi;
2395 	sc = pi->adapter;
2396 
2397 	rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2398 	if (rc)
2399 		return (rc);
2400 	rc = alloc_extra_vi(sc, pi, vi);
2401 	end_synchronized_op(sc, 0);
2402 	if (rc)
2403 		return (rc);
2404 
2405 	rc = cxgbe_vi_attach(dev, vi);
2406 	if (rc) {
2407 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2408 		return (rc);
2409 	}
2410 	return (0);
2411 }
2412 
2413 static int
2414 vcxgbe_detach(device_t dev)
2415 {
2416 	struct vi_info *vi;
2417 	struct adapter *sc;
2418 
2419 	vi = device_get_softc(dev);
2420 	sc = vi->pi->adapter;
2421 
2422 	doom_vi(sc, vi);
2423 
2424 	cxgbe_vi_detach(vi);
2425 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2426 
2427 	end_synchronized_op(sc, 0);
2428 
2429 	return (0);
2430 }
2431 
2432 void
2433 t4_fatal_err(struct adapter *sc)
2434 {
2435 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2436 	t4_intr_disable(sc);
2437 	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2438 	    device_get_nameunit(sc->dev));
2439 	if (t4_panic_on_fatal_err)
2440 		panic("panic requested on fatal error");
2441 }
2442 
2443 void
2444 t4_add_adapter(struct adapter *sc)
2445 {
2446 	sx_xlock(&t4_list_lock);
2447 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2448 	sx_xunlock(&t4_list_lock);
2449 }
2450 
2451 int
2452 t4_map_bars_0_and_4(struct adapter *sc)
2453 {
2454 	sc->regs_rid = PCIR_BAR(0);
2455 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2456 	    &sc->regs_rid, RF_ACTIVE);
2457 	if (sc->regs_res == NULL) {
2458 		device_printf(sc->dev, "cannot map registers.\n");
2459 		return (ENXIO);
2460 	}
2461 	sc->bt = rman_get_bustag(sc->regs_res);
2462 	sc->bh = rman_get_bushandle(sc->regs_res);
2463 	sc->mmio_len = rman_get_size(sc->regs_res);
2464 	setbit(&sc->doorbells, DOORBELL_KDB);
2465 
2466 	sc->msix_rid = PCIR_BAR(4);
2467 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2468 	    &sc->msix_rid, RF_ACTIVE);
2469 	if (sc->msix_res == NULL) {
2470 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2471 		return (ENXIO);
2472 	}
2473 
2474 	return (0);
2475 }
2476 
2477 int
2478 t4_map_bar_2(struct adapter *sc)
2479 {
2480 
2481 	/*
2482 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2483 	 * to map it if RDMA is disabled.
2484 	 */
2485 	if (is_t4(sc) && sc->rdmacaps == 0)
2486 		return (0);
2487 
2488 	sc->udbs_rid = PCIR_BAR(2);
2489 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2490 	    &sc->udbs_rid, RF_ACTIVE);
2491 	if (sc->udbs_res == NULL) {
2492 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2493 		return (ENXIO);
2494 	}
2495 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2496 
2497 	if (chip_id(sc) >= CHELSIO_T5) {
2498 		setbit(&sc->doorbells, DOORBELL_UDB);
2499 #if defined(__i386__) || defined(__amd64__)
2500 		if (t5_write_combine) {
2501 			int rc, mode;
2502 
2503 			/*
2504 			 * Enable write combining on BAR2.  This is the
2505 			 * userspace doorbell BAR and is split into 128B
2506 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2507 			 * with an egress queue.  The first 64B has the doorbell
2508 			 * and the second 64B can be used to submit a tx work
2509 			 * request with an implicit doorbell.
2510 			 */
2511 
2512 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2513 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2514 			if (rc == 0) {
2515 				clrbit(&sc->doorbells, DOORBELL_UDB);
2516 				setbit(&sc->doorbells, DOORBELL_WCWR);
2517 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2518 			} else {
2519 				device_printf(sc->dev,
2520 				    "couldn't enable write combining: %d\n",
2521 				    rc);
2522 			}
2523 
2524 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2525 			t4_write_reg(sc, A_SGE_STAT_CFG,
2526 			    V_STATSOURCE_T5(7) | mode);
2527 		}
2528 #endif
2529 	}
2530 	sc->iwt.wc_en = isset(&sc->doorbells, DOORBELL_UDBWC) ? 1 : 0;
2531 
2532 	return (0);
2533 }
2534 
2535 struct memwin_init {
2536 	uint32_t base;
2537 	uint32_t aperture;
2538 };
2539 
2540 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2541 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2542 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2543 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2544 };
2545 
2546 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2547 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2548 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2549 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2550 };
2551 
2552 static void
2553 setup_memwin(struct adapter *sc)
2554 {
2555 	const struct memwin_init *mw_init;
2556 	struct memwin *mw;
2557 	int i;
2558 	uint32_t bar0;
2559 
2560 	if (is_t4(sc)) {
2561 		/*
2562 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2563 		 * mechanism.  Works from within PCI passthrough environments
2564 		 * too, where rman_get_start() can return a different value.  We
2565 		 * need to program the T4 memory window decoders with the actual
2566 		 * addresses that will be coming across the PCIe link.
2567 		 */
2568 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2569 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2570 
2571 		mw_init = &t4_memwin[0];
2572 	} else {
2573 		/* T5+ use the relative offset inside the PCIe BAR */
2574 		bar0 = 0;
2575 
2576 		mw_init = &t5_memwin[0];
2577 	}
2578 
2579 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2580 		rw_init(&mw->mw_lock, "memory window access");
2581 		mw->mw_base = mw_init->base;
2582 		mw->mw_aperture = mw_init->aperture;
2583 		mw->mw_curpos = 0;
2584 		t4_write_reg(sc,
2585 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2586 		    (mw->mw_base + bar0) | V_BIR(0) |
2587 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2588 		rw_wlock(&mw->mw_lock);
2589 		position_memwin(sc, i, 0);
2590 		rw_wunlock(&mw->mw_lock);
2591 	}
2592 
2593 	/* flush */
2594 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2595 }
2596 
2597 /*
2598  * Positions the memory window at the given address in the card's address space.
2599  * There are some alignment requirements and the actual position may be at an
2600  * address prior to the requested address.  mw->mw_curpos always has the actual
2601  * position of the window.
2602  */
2603 static void
2604 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2605 {
2606 	struct memwin *mw;
2607 	uint32_t pf;
2608 	uint32_t reg;
2609 
2610 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2611 	mw = &sc->memwin[idx];
2612 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2613 
2614 	if (is_t4(sc)) {
2615 		pf = 0;
2616 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2617 	} else {
2618 		pf = V_PFNUM(sc->pf);
2619 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2620 	}
2621 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2622 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2623 	t4_read_reg(sc, reg);	/* flush */
2624 }
2625 
2626 int
2627 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2628     int len, int rw)
2629 {
2630 	struct memwin *mw;
2631 	uint32_t mw_end, v;
2632 
2633 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2634 
2635 	/* Memory can only be accessed in naturally aligned 4 byte units */
2636 	if (addr & 3 || len & 3 || len <= 0)
2637 		return (EINVAL);
2638 
2639 	mw = &sc->memwin[idx];
2640 	while (len > 0) {
2641 		rw_rlock(&mw->mw_lock);
2642 		mw_end = mw->mw_curpos + mw->mw_aperture;
2643 		if (addr >= mw_end || addr < mw->mw_curpos) {
2644 			/* Will need to reposition the window */
2645 			if (!rw_try_upgrade(&mw->mw_lock)) {
2646 				rw_runlock(&mw->mw_lock);
2647 				rw_wlock(&mw->mw_lock);
2648 			}
2649 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2650 			position_memwin(sc, idx, addr);
2651 			rw_downgrade(&mw->mw_lock);
2652 			mw_end = mw->mw_curpos + mw->mw_aperture;
2653 		}
2654 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2655 		while (addr < mw_end && len > 0) {
2656 			if (rw == 0) {
2657 				v = t4_read_reg(sc, mw->mw_base + addr -
2658 				    mw->mw_curpos);
2659 				*val++ = le32toh(v);
2660 			} else {
2661 				v = *val++;
2662 				t4_write_reg(sc, mw->mw_base + addr -
2663 				    mw->mw_curpos, htole32(v));
2664 			}
2665 			addr += 4;
2666 			len -= 4;
2667 		}
2668 		rw_runlock(&mw->mw_lock);
2669 	}
2670 
2671 	return (0);
2672 }
2673 
2674 int
2675 alloc_atid_tab(struct tid_info *t, int flags)
2676 {
2677 	int i;
2678 
2679 	MPASS(t->natids > 0);
2680 	MPASS(t->atid_tab == NULL);
2681 
2682 	t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
2683 	    M_ZERO | flags);
2684 	if (t->atid_tab == NULL)
2685 		return (ENOMEM);
2686 	mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
2687 	t->afree = t->atid_tab;
2688 	t->atids_in_use = 0;
2689 	for (i = 1; i < t->natids; i++)
2690 		t->atid_tab[i - 1].next = &t->atid_tab[i];
2691 	t->atid_tab[t->natids - 1].next = NULL;
2692 
2693 	return (0);
2694 }
2695 
2696 void
2697 free_atid_tab(struct tid_info *t)
2698 {
2699 
2700 	KASSERT(t->atids_in_use == 0,
2701 	    ("%s: %d atids still in use.", __func__, t->atids_in_use));
2702 
2703 	if (mtx_initialized(&t->atid_lock))
2704 		mtx_destroy(&t->atid_lock);
2705 	free(t->atid_tab, M_CXGBE);
2706 	t->atid_tab = NULL;
2707 }
2708 
2709 int
2710 alloc_atid(struct adapter *sc, void *ctx)
2711 {
2712 	struct tid_info *t = &sc->tids;
2713 	int atid = -1;
2714 
2715 	mtx_lock(&t->atid_lock);
2716 	if (t->afree) {
2717 		union aopen_entry *p = t->afree;
2718 
2719 		atid = p - t->atid_tab;
2720 		MPASS(atid <= M_TID_TID);
2721 		t->afree = p->next;
2722 		p->data = ctx;
2723 		t->atids_in_use++;
2724 	}
2725 	mtx_unlock(&t->atid_lock);
2726 	return (atid);
2727 }
2728 
2729 void *
2730 lookup_atid(struct adapter *sc, int atid)
2731 {
2732 	struct tid_info *t = &sc->tids;
2733 
2734 	return (t->atid_tab[atid].data);
2735 }
2736 
2737 void
2738 free_atid(struct adapter *sc, int atid)
2739 {
2740 	struct tid_info *t = &sc->tids;
2741 	union aopen_entry *p = &t->atid_tab[atid];
2742 
2743 	mtx_lock(&t->atid_lock);
2744 	p->next = t->afree;
2745 	t->afree = p;
2746 	t->atids_in_use--;
2747 	mtx_unlock(&t->atid_lock);
2748 }
2749 
2750 static void
2751 queue_tid_release(struct adapter *sc, int tid)
2752 {
2753 
2754 	CXGBE_UNIMPLEMENTED("deferred tid release");
2755 }
2756 
2757 void
2758 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
2759 {
2760 	struct wrqe *wr;
2761 	struct cpl_tid_release *req;
2762 
2763 	wr = alloc_wrqe(sizeof(*req), ctrlq);
2764 	if (wr == NULL) {
2765 		queue_tid_release(sc, tid);	/* defer */
2766 		return;
2767 	}
2768 	req = wrtod(wr);
2769 
2770 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
2771 
2772 	t4_wrq_tx(sc, wr);
2773 }
2774 
2775 static int
2776 t4_range_cmp(const void *a, const void *b)
2777 {
2778 	return ((const struct t4_range *)a)->start -
2779 	       ((const struct t4_range *)b)->start;
2780 }
2781 
2782 /*
2783  * Verify that the memory range specified by the addr/len pair is valid within
2784  * the card's address space.
2785  */
2786 static int
2787 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2788 {
2789 	struct t4_range mem_ranges[4], *r, *next;
2790 	uint32_t em, addr_len;
2791 	int i, n, remaining;
2792 
2793 	/* Memory can only be accessed in naturally aligned 4 byte units */
2794 	if (addr & 3 || len & 3 || len <= 0)
2795 		return (EINVAL);
2796 
2797 	/* Enabled memories */
2798 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2799 
2800 	r = &mem_ranges[0];
2801 	n = 0;
2802 	bzero(r, sizeof(mem_ranges));
2803 	if (em & F_EDRAM0_ENABLE) {
2804 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2805 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2806 		if (r->size > 0) {
2807 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2808 			if (addr >= r->start &&
2809 			    addr + len <= r->start + r->size)
2810 				return (0);
2811 			r++;
2812 			n++;
2813 		}
2814 	}
2815 	if (em & F_EDRAM1_ENABLE) {
2816 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2817 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2818 		if (r->size > 0) {
2819 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2820 			if (addr >= r->start &&
2821 			    addr + len <= r->start + r->size)
2822 				return (0);
2823 			r++;
2824 			n++;
2825 		}
2826 	}
2827 	if (em & F_EXT_MEM_ENABLE) {
2828 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2829 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2830 		if (r->size > 0) {
2831 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2832 			if (addr >= r->start &&
2833 			    addr + len <= r->start + r->size)
2834 				return (0);
2835 			r++;
2836 			n++;
2837 		}
2838 	}
2839 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2840 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2841 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2842 		if (r->size > 0) {
2843 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2844 			if (addr >= r->start &&
2845 			    addr + len <= r->start + r->size)
2846 				return (0);
2847 			r++;
2848 			n++;
2849 		}
2850 	}
2851 	MPASS(n <= nitems(mem_ranges));
2852 
2853 	if (n > 1) {
2854 		/* Sort and merge the ranges. */
2855 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2856 
2857 		/* Start from index 0 and examine the next n - 1 entries. */
2858 		r = &mem_ranges[0];
2859 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2860 
2861 			MPASS(r->size > 0);	/* r is a valid entry. */
2862 			next = r + 1;
2863 			MPASS(next->size > 0);	/* and so is the next one. */
2864 
2865 			while (r->start + r->size >= next->start) {
2866 				/* Merge the next one into the current entry. */
2867 				r->size = max(r->start + r->size,
2868 				    next->start + next->size) - r->start;
2869 				n--;	/* One fewer entry in total. */
2870 				if (--remaining == 0)
2871 					goto done;	/* short circuit */
2872 				next++;
2873 			}
2874 			if (next != r + 1) {
2875 				/*
2876 				 * Some entries were merged into r and next
2877 				 * points to the first valid entry that couldn't
2878 				 * be merged.
2879 				 */
2880 				MPASS(next->size > 0);	/* must be valid */
2881 				memcpy(r + 1, next, remaining * sizeof(*r));
2882 #ifdef INVARIANTS
2883 				/*
2884 				 * This so that the foo->size assertion in the
2885 				 * next iteration of the loop do the right
2886 				 * thing for entries that were pulled up and are
2887 				 * no longer valid.
2888 				 */
2889 				MPASS(n < nitems(mem_ranges));
2890 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2891 				    sizeof(struct t4_range));
2892 #endif
2893 			}
2894 		}
2895 done:
2896 		/* Done merging the ranges. */
2897 		MPASS(n > 0);
2898 		r = &mem_ranges[0];
2899 		for (i = 0; i < n; i++, r++) {
2900 			if (addr >= r->start &&
2901 			    addr + len <= r->start + r->size)
2902 				return (0);
2903 		}
2904 	}
2905 
2906 	return (EFAULT);
2907 }
2908 
2909 static int
2910 fwmtype_to_hwmtype(int mtype)
2911 {
2912 
2913 	switch (mtype) {
2914 	case FW_MEMTYPE_EDC0:
2915 		return (MEM_EDC0);
2916 	case FW_MEMTYPE_EDC1:
2917 		return (MEM_EDC1);
2918 	case FW_MEMTYPE_EXTMEM:
2919 		return (MEM_MC0);
2920 	case FW_MEMTYPE_EXTMEM1:
2921 		return (MEM_MC1);
2922 	default:
2923 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2924 	}
2925 }
2926 
2927 /*
2928  * Verify that the memory range specified by the memtype/offset/len pair is
2929  * valid and lies entirely within the memtype specified.  The global address of
2930  * the start of the range is returned in addr.
2931  */
2932 static int
2933 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2934     uint32_t *addr)
2935 {
2936 	uint32_t em, addr_len, maddr;
2937 
2938 	/* Memory can only be accessed in naturally aligned 4 byte units */
2939 	if (off & 3 || len & 3 || len == 0)
2940 		return (EINVAL);
2941 
2942 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2943 	switch (fwmtype_to_hwmtype(mtype)) {
2944 	case MEM_EDC0:
2945 		if (!(em & F_EDRAM0_ENABLE))
2946 			return (EINVAL);
2947 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2948 		maddr = G_EDRAM0_BASE(addr_len) << 20;
2949 		break;
2950 	case MEM_EDC1:
2951 		if (!(em & F_EDRAM1_ENABLE))
2952 			return (EINVAL);
2953 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2954 		maddr = G_EDRAM1_BASE(addr_len) << 20;
2955 		break;
2956 	case MEM_MC:
2957 		if (!(em & F_EXT_MEM_ENABLE))
2958 			return (EINVAL);
2959 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2960 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2961 		break;
2962 	case MEM_MC1:
2963 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2964 			return (EINVAL);
2965 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2966 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2967 		break;
2968 	default:
2969 		return (EINVAL);
2970 	}
2971 
2972 	*addr = maddr + off;	/* global address */
2973 	return (validate_mem_range(sc, *addr, len));
2974 }
2975 
2976 static int
2977 fixup_devlog_params(struct adapter *sc)
2978 {
2979 	struct devlog_params *dparams = &sc->params.devlog;
2980 	int rc;
2981 
2982 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2983 	    dparams->size, &dparams->addr);
2984 
2985 	return (rc);
2986 }
2987 
2988 static void
2989 update_nirq(struct intrs_and_queues *iaq, int nports)
2990 {
2991 	int extra = T4_EXTRA_INTR;
2992 
2993 	iaq->nirq = extra;
2994 	iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
2995 	iaq->nirq += nports * (iaq->num_vis - 1) *
2996 	    max(iaq->nrxq_vi, iaq->nnmrxq_vi);
2997 	iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
2998 }
2999 
3000 /*
3001  * Adjust requirements to fit the number of interrupts available.
3002  */
3003 static void
3004 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
3005     int navail)
3006 {
3007 	int old_nirq;
3008 	const int nports = sc->params.nports;
3009 
3010 	MPASS(nports > 0);
3011 	MPASS(navail > 0);
3012 
3013 	bzero(iaq, sizeof(*iaq));
3014 	iaq->intr_type = itype;
3015 	iaq->num_vis = t4_num_vis;
3016 	iaq->ntxq = t4_ntxq;
3017 	iaq->ntxq_vi = t4_ntxq_vi;
3018 	iaq->nrxq = t4_nrxq;
3019 	iaq->nrxq_vi = t4_nrxq_vi;
3020 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
3021 	if (is_offload(sc) || is_ethoffload(sc)) {
3022 		iaq->nofldtxq = t4_nofldtxq;
3023 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
3024 	}
3025 #endif
3026 #ifdef TCP_OFFLOAD
3027 	if (is_offload(sc)) {
3028 		iaq->nofldrxq = t4_nofldrxq;
3029 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
3030 	}
3031 #endif
3032 #ifdef DEV_NETMAP
3033 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
3034 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
3035 #endif
3036 
3037 	update_nirq(iaq, nports);
3038 	if (iaq->nirq <= navail &&
3039 	    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3040 		/*
3041 		 * This is the normal case -- there are enough interrupts for
3042 		 * everything.
3043 		 */
3044 		goto done;
3045 	}
3046 
3047 	/*
3048 	 * If extra VIs have been configured try reducing their count and see if
3049 	 * that works.
3050 	 */
3051 	while (iaq->num_vis > 1) {
3052 		iaq->num_vis--;
3053 		update_nirq(iaq, nports);
3054 		if (iaq->nirq <= navail &&
3055 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3056 			device_printf(sc->dev, "virtual interfaces per port "
3057 			    "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
3058 			    "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
3059 			    "itype %d, navail %u, nirq %d.\n",
3060 			    iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
3061 			    iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
3062 			    itype, navail, iaq->nirq);
3063 			goto done;
3064 		}
3065 	}
3066 
3067 	/*
3068 	 * Extra VIs will not be created.  Log a message if they were requested.
3069 	 */
3070 	MPASS(iaq->num_vis == 1);
3071 	iaq->ntxq_vi = iaq->nrxq_vi = 0;
3072 	iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
3073 	iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
3074 	if (iaq->num_vis != t4_num_vis) {
3075 		device_printf(sc->dev, "extra virtual interfaces disabled.  "
3076 		    "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
3077 		    "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
3078 		    iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
3079 		    iaq->nnmrxq_vi, itype, navail, iaq->nirq);
3080 	}
3081 
3082 	/*
3083 	 * Keep reducing the number of NIC rx queues to the next lower power of
3084 	 * 2 (for even RSS distribution) and halving the TOE rx queues and see
3085 	 * if that works.
3086 	 */
3087 	do {
3088 		if (iaq->nrxq > 1) {
3089 			do {
3090 				iaq->nrxq--;
3091 			} while (!powerof2(iaq->nrxq));
3092 		}
3093 		if (iaq->nofldrxq > 1)
3094 			iaq->nofldrxq >>= 1;
3095 
3096 		old_nirq = iaq->nirq;
3097 		update_nirq(iaq, nports);
3098 		if (iaq->nirq <= navail &&
3099 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3100 			device_printf(sc->dev, "running with reduced number of "
3101 			    "rx queues because of shortage of interrupts.  "
3102 			    "nrxq=%u, nofldrxq=%u.  "
3103 			    "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
3104 			    iaq->nofldrxq, itype, navail, iaq->nirq);
3105 			goto done;
3106 		}
3107 	} while (old_nirq != iaq->nirq);
3108 
3109 	/* One interrupt for everything.  Ugh. */
3110 	device_printf(sc->dev, "running with minimal number of queues.  "
3111 	    "itype %d, navail %u.\n", itype, navail);
3112 	iaq->nirq = 1;
3113 	MPASS(iaq->nrxq == 1);
3114 	iaq->ntxq = 1;
3115 	if (iaq->nofldrxq > 1)
3116 		iaq->nofldtxq = 1;
3117 done:
3118 	MPASS(iaq->num_vis > 0);
3119 	if (iaq->num_vis > 1) {
3120 		MPASS(iaq->nrxq_vi > 0);
3121 		MPASS(iaq->ntxq_vi > 0);
3122 	}
3123 	MPASS(iaq->nirq > 0);
3124 	MPASS(iaq->nrxq > 0);
3125 	MPASS(iaq->ntxq > 0);
3126 	if (itype == INTR_MSI) {
3127 		MPASS(powerof2(iaq->nirq));
3128 	}
3129 }
3130 
3131 static int
3132 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
3133 {
3134 	int rc, itype, navail, nalloc;
3135 
3136 	for (itype = INTR_MSIX; itype; itype >>= 1) {
3137 
3138 		if ((itype & t4_intr_types) == 0)
3139 			continue;	/* not allowed */
3140 
3141 		if (itype == INTR_MSIX)
3142 			navail = pci_msix_count(sc->dev);
3143 		else if (itype == INTR_MSI)
3144 			navail = pci_msi_count(sc->dev);
3145 		else
3146 			navail = 1;
3147 restart:
3148 		if (navail == 0)
3149 			continue;
3150 
3151 		calculate_iaq(sc, iaq, itype, navail);
3152 		nalloc = iaq->nirq;
3153 		rc = 0;
3154 		if (itype == INTR_MSIX)
3155 			rc = pci_alloc_msix(sc->dev, &nalloc);
3156 		else if (itype == INTR_MSI)
3157 			rc = pci_alloc_msi(sc->dev, &nalloc);
3158 
3159 		if (rc == 0 && nalloc > 0) {
3160 			if (nalloc == iaq->nirq)
3161 				return (0);
3162 
3163 			/*
3164 			 * Didn't get the number requested.  Use whatever number
3165 			 * the kernel is willing to allocate.
3166 			 */
3167 			device_printf(sc->dev, "fewer vectors than requested, "
3168 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
3169 			    itype, iaq->nirq, nalloc);
3170 			pci_release_msi(sc->dev);
3171 			navail = nalloc;
3172 			goto restart;
3173 		}
3174 
3175 		device_printf(sc->dev,
3176 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
3177 		    itype, rc, iaq->nirq, nalloc);
3178 	}
3179 
3180 	device_printf(sc->dev,
3181 	    "failed to find a usable interrupt type.  "
3182 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
3183 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
3184 
3185 	return (ENXIO);
3186 }
3187 
3188 #define FW_VERSION(chip) ( \
3189     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
3190     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
3191     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
3192     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
3193 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
3194 
3195 struct fw_info {
3196 	uint8_t chip;
3197 	char *kld_name;
3198 	char *fw_mod_name;
3199 	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
3200 } fw_info[] = {
3201 	{
3202 		.chip = CHELSIO_T4,
3203 		.kld_name = "t4fw_cfg",
3204 		.fw_mod_name = "t4fw",
3205 		.fw_hdr = {
3206 			.chip = FW_HDR_CHIP_T4,
3207 			.fw_ver = htobe32(FW_VERSION(T4)),
3208 			.intfver_nic = FW_INTFVER(T4, NIC),
3209 			.intfver_vnic = FW_INTFVER(T4, VNIC),
3210 			.intfver_ofld = FW_INTFVER(T4, OFLD),
3211 			.intfver_ri = FW_INTFVER(T4, RI),
3212 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
3213 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
3214 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
3215 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
3216 		},
3217 	}, {
3218 		.chip = CHELSIO_T5,
3219 		.kld_name = "t5fw_cfg",
3220 		.fw_mod_name = "t5fw",
3221 		.fw_hdr = {
3222 			.chip = FW_HDR_CHIP_T5,
3223 			.fw_ver = htobe32(FW_VERSION(T5)),
3224 			.intfver_nic = FW_INTFVER(T5, NIC),
3225 			.intfver_vnic = FW_INTFVER(T5, VNIC),
3226 			.intfver_ofld = FW_INTFVER(T5, OFLD),
3227 			.intfver_ri = FW_INTFVER(T5, RI),
3228 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
3229 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
3230 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
3231 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
3232 		},
3233 	}, {
3234 		.chip = CHELSIO_T6,
3235 		.kld_name = "t6fw_cfg",
3236 		.fw_mod_name = "t6fw",
3237 		.fw_hdr = {
3238 			.chip = FW_HDR_CHIP_T6,
3239 			.fw_ver = htobe32(FW_VERSION(T6)),
3240 			.intfver_nic = FW_INTFVER(T6, NIC),
3241 			.intfver_vnic = FW_INTFVER(T6, VNIC),
3242 			.intfver_ofld = FW_INTFVER(T6, OFLD),
3243 			.intfver_ri = FW_INTFVER(T6, RI),
3244 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3245 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
3246 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3247 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
3248 		},
3249 	}
3250 };
3251 
3252 static struct fw_info *
3253 find_fw_info(int chip)
3254 {
3255 	int i;
3256 
3257 	for (i = 0; i < nitems(fw_info); i++) {
3258 		if (fw_info[i].chip == chip)
3259 			return (&fw_info[i]);
3260 	}
3261 	return (NULL);
3262 }
3263 
3264 /*
3265  * Is the given firmware API compatible with the one the driver was compiled
3266  * with?
3267  */
3268 static int
3269 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
3270 {
3271 
3272 	/* short circuit if it's the exact same firmware version */
3273 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
3274 		return (1);
3275 
3276 	/*
3277 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
3278 	 * features that are supported in the driver.
3279 	 */
3280 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
3281 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
3282 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
3283 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
3284 		return (1);
3285 #undef SAME_INTF
3286 
3287 	return (0);
3288 }
3289 
3290 /*
3291  * The firmware in the KLD is usable, but should it be installed?  This routine
3292  * explains itself in detail if it indicates the KLD firmware should be
3293  * installed.
3294  */
3295 static int
3296 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
3297 {
3298 	const char *reason;
3299 
3300 	if (!card_fw_usable) {
3301 		reason = "incompatible or unusable";
3302 		goto install;
3303 	}
3304 
3305 	if (k > c) {
3306 		reason = "older than the version bundled with this driver";
3307 		goto install;
3308 	}
3309 
3310 	if (t4_fw_install == 2 && k != c) {
3311 		reason = "different than the version bundled with this driver";
3312 		goto install;
3313 	}
3314 
3315 	return (0);
3316 
3317 install:
3318 	if (t4_fw_install == 0) {
3319 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3320 		    "but the driver is prohibited from installing a different "
3321 		    "firmware on the card.\n",
3322 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3323 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3324 
3325 		return (0);
3326 	}
3327 
3328 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3329 	    "installing firmware %u.%u.%u.%u on card.\n",
3330 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3331 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3332 	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3333 	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3334 
3335 	return (1);
3336 }
3337 
3338 /*
3339  * Establish contact with the firmware and determine if we are the master driver
3340  * or not, and whether we are responsible for chip initialization.
3341  */
3342 static int
3343 prep_firmware(struct adapter *sc)
3344 {
3345 	const struct firmware *fw = NULL, *default_cfg;
3346 	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
3347 	enum dev_state state;
3348 	struct fw_info *fw_info;
3349 	struct fw_hdr *card_fw;		/* fw on the card */
3350 	const struct fw_hdr *kld_fw;	/* fw in the KLD */
3351 	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
3352 					   against */
3353 
3354 	/* This is the firmware whose headers the driver was compiled against */
3355 	fw_info = find_fw_info(chip_id(sc));
3356 	if (fw_info == NULL) {
3357 		device_printf(sc->dev,
3358 		    "unable to look up firmware information for chip %d.\n",
3359 		    chip_id(sc));
3360 		return (EINVAL);
3361 	}
3362 	drv_fw = &fw_info->fw_hdr;
3363 
3364 	/*
3365 	 * The firmware KLD contains many modules.  The KLD name is also the
3366 	 * name of the module that contains the default config file.
3367 	 */
3368 	default_cfg = firmware_get(fw_info->kld_name);
3369 
3370 	/* This is the firmware in the KLD */
3371 	fw = firmware_get(fw_info->fw_mod_name);
3372 	if (fw != NULL) {
3373 		kld_fw = (const void *)fw->data;
3374 		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3375 	} else {
3376 		kld_fw = NULL;
3377 		kld_fw_usable = 0;
3378 	}
3379 
3380 	/* Read the header of the firmware on the card */
3381 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3382 	rc = -t4_read_flash(sc, FLASH_FW_START,
3383 	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
3384 	if (rc == 0) {
3385 		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
3386 		if (card_fw->fw_ver == be32toh(0xffffffff)) {
3387 			uint32_t d = be32toh(kld_fw->fw_ver);
3388 
3389 			if (!kld_fw_usable) {
3390 				device_printf(sc->dev,
3391 				    "no firmware on the card and no usable "
3392 				    "firmware bundled with the driver.\n");
3393 				rc = EIO;
3394 				goto done;
3395 			} else if (t4_fw_install == 0) {
3396 				device_printf(sc->dev,
3397 				    "no firmware on the card and the driver "
3398 				    "is prohibited from installing new "
3399 				    "firmware.\n");
3400 				rc = EIO;
3401 				goto done;
3402 			}
3403 
3404 			device_printf(sc->dev, "no firmware on the card, "
3405 			    "installing firmware %d.%d.%d.%d\n",
3406 			    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3407 			    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3408 			rc = t4_fw_forceinstall(sc, fw->data, fw->datasize);
3409 			if (rc < 0) {
3410 				rc = -rc;
3411 				device_printf(sc->dev,
3412 				    "firmware install failed: %d.\n", rc);
3413 				goto done;
3414 			}
3415 			memcpy(card_fw, kld_fw, sizeof(*card_fw));
3416 			card_fw_usable = 1;
3417 			need_fw_reset = 0;
3418 		}
3419 	} else {
3420 		device_printf(sc->dev,
3421 		    "Unable to read card's firmware header: %d\n", rc);
3422 		card_fw_usable = 0;
3423 	}
3424 
3425 	/* Contact firmware. */
3426 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3427 	if (rc < 0 || state == DEV_STATE_ERR) {
3428 		rc = -rc;
3429 		device_printf(sc->dev,
3430 		    "failed to connect to the firmware: %d, %d.\n", rc, state);
3431 		goto done;
3432 	}
3433 	pf = rc;
3434 	if (pf == sc->mbox)
3435 		sc->flags |= MASTER_PF;
3436 	else if (state == DEV_STATE_UNINIT) {
3437 		/*
3438 		 * We didn't get to be the master so we definitely won't be
3439 		 * configuring the chip.  It's a bug if someone else hasn't
3440 		 * configured it already.
3441 		 */
3442 		device_printf(sc->dev, "couldn't be master(%d), "
3443 		    "device not already initialized either(%d).\n", rc, state);
3444 		rc = EPROTO;
3445 		goto done;
3446 	}
3447 
3448 	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3449 	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3450 		/*
3451 		 * Common case: the firmware on the card is an exact match and
3452 		 * the KLD is an exact match too, or the KLD is
3453 		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
3454 		 * here -- use cxgbetool loadfw if you want to reinstall the
3455 		 * same firmware as the one on the card.
3456 		 */
3457 	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3458 	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3459 	    be32toh(card_fw->fw_ver))) {
3460 
3461 		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3462 		if (rc != 0) {
3463 			device_printf(sc->dev,
3464 			    "failed to install firmware: %d\n", rc);
3465 			goto done;
3466 		}
3467 
3468 		/* Installed successfully, update the cached header too. */
3469 		memcpy(card_fw, kld_fw, sizeof(*card_fw));
3470 		card_fw_usable = 1;
3471 		need_fw_reset = 0;	/* already reset as part of load_fw */
3472 	}
3473 
3474 	if (!card_fw_usable) {
3475 		uint32_t d, c, k;
3476 
3477 		d = ntohl(drv_fw->fw_ver);
3478 		c = ntohl(card_fw->fw_ver);
3479 		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3480 
3481 		device_printf(sc->dev, "Cannot find a usable firmware: "
3482 		    "fw_install %d, chip state %d, "
3483 		    "driver compiled with %d.%d.%d.%d, "
3484 		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3485 		    t4_fw_install, state,
3486 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3487 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3488 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3489 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3490 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3491 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3492 		rc = EINVAL;
3493 		goto done;
3494 	}
3495 
3496 	/* Reset device */
3497 	if (need_fw_reset &&
3498 	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3499 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3500 		if (rc != ETIMEDOUT && rc != EIO)
3501 			t4_fw_bye(sc, sc->mbox);
3502 		goto done;
3503 	}
3504 	sc->flags |= FW_OK;
3505 
3506 	rc = get_params__pre_init(sc);
3507 	if (rc != 0)
3508 		goto done; /* error message displayed already */
3509 
3510 	/* Partition adapter resources as specified in the config file. */
3511 	if (state == DEV_STATE_UNINIT) {
3512 
3513 		KASSERT(sc->flags & MASTER_PF,
3514 		    ("%s: trying to change chip settings when not master.",
3515 		    __func__));
3516 
3517 		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3518 		if (rc != 0)
3519 			goto done;	/* error message displayed already */
3520 
3521 		t4_tweak_chip_settings(sc);
3522 
3523 		/* get basic stuff going */
3524 		rc = -t4_fw_initialize(sc, sc->mbox);
3525 		if (rc != 0) {
3526 			device_printf(sc->dev, "fw init failed: %d.\n", rc);
3527 			goto done;
3528 		}
3529 	} else {
3530 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3531 		sc->cfcsum = 0;
3532 	}
3533 
3534 done:
3535 	free(card_fw, M_CXGBE);
3536 	if (fw != NULL)
3537 		firmware_put(fw, FIRMWARE_UNLOAD);
3538 	if (default_cfg != NULL)
3539 		firmware_put(default_cfg, FIRMWARE_UNLOAD);
3540 
3541 	return (rc);
3542 }
3543 
3544 #define FW_PARAM_DEV(param) \
3545 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3546 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3547 #define FW_PARAM_PFVF(param) \
3548 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3549 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3550 
3551 /*
3552  * Partition chip resources for use between various PFs, VFs, etc.
3553  */
3554 static int
3555 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3556     const char *name_prefix)
3557 {
3558 	const struct firmware *cfg = NULL;
3559 	int rc = 0;
3560 	struct fw_caps_config_cmd caps;
3561 	uint32_t mtype, moff, finicsum, cfcsum;
3562 
3563 	/*
3564 	 * Figure out what configuration file to use.  Pick the default config
3565 	 * file for the card if the user hasn't specified one explicitly.
3566 	 */
3567 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3568 	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3569 		/* Card specific overrides go here. */
3570 		if (pci_get_device(sc->dev) == 0x440a)
3571 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3572 		if (is_fpga(sc))
3573 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3574 	} else if (strncmp(t4_cfg_file, BUILTIN_CF, sizeof(t4_cfg_file)) == 0)
3575 		goto use_built_in_config;	/* go straight to config. */
3576 
3577 	/*
3578 	 * We need to load another module if the profile is anything except
3579 	 * "default" or "flash".
3580 	 */
3581 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3582 	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3583 		char s[32];
3584 
3585 		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3586 		cfg = firmware_get(s);
3587 		if (cfg == NULL) {
3588 			if (default_cfg != NULL) {
3589 				device_printf(sc->dev,
3590 				    "unable to load module \"%s\" for "
3591 				    "configuration profile \"%s\", will use "
3592 				    "the default config file instead.\n",
3593 				    s, sc->cfg_file);
3594 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3595 				    "%s", DEFAULT_CF);
3596 			} else {
3597 				device_printf(sc->dev,
3598 				    "unable to load module \"%s\" for "
3599 				    "configuration profile \"%s\", will use "
3600 				    "the config file on the card's flash "
3601 				    "instead.\n", s, sc->cfg_file);
3602 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3603 				    "%s", FLASH_CF);
3604 			}
3605 		}
3606 	}
3607 
3608 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3609 	    default_cfg == NULL) {
3610 		device_printf(sc->dev,
3611 		    "default config file not available, will use the config "
3612 		    "file on the card's flash instead.\n");
3613 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3614 	}
3615 
3616 	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3617 		u_int cflen;
3618 		const uint32_t *cfdata;
3619 		uint32_t param, val, addr;
3620 
3621 		KASSERT(cfg != NULL || default_cfg != NULL,
3622 		    ("%s: no config to upload", __func__));
3623 
3624 		/*
3625 		 * Ask the firmware where it wants us to upload the config file.
3626 		 */
3627 		param = FW_PARAM_DEV(CF);
3628 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3629 		if (rc != 0) {
3630 			/* No support for config file?  Shouldn't happen. */
3631 			device_printf(sc->dev,
3632 			    "failed to query config file location: %d.\n", rc);
3633 			goto done;
3634 		}
3635 		mtype = G_FW_PARAMS_PARAM_Y(val);
3636 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3637 
3638 		/*
3639 		 * XXX: sheer laziness.  We deliberately added 4 bytes of
3640 		 * useless stuffing/comments at the end of the config file so
3641 		 * it's ok to simply throw away the last remaining bytes when
3642 		 * the config file is not an exact multiple of 4.  This also
3643 		 * helps with the validate_mt_off_len check.
3644 		 */
3645 		if (cfg != NULL) {
3646 			cflen = cfg->datasize & ~3;
3647 			cfdata = cfg->data;
3648 		} else {
3649 			cflen = default_cfg->datasize & ~3;
3650 			cfdata = default_cfg->data;
3651 		}
3652 
3653 		if (cflen > FLASH_CFG_MAX_SIZE) {
3654 			device_printf(sc->dev,
3655 			    "config file too long (%d, max allowed is %d).  "
3656 			    "Will try to use the config on the card, if any.\n",
3657 			    cflen, FLASH_CFG_MAX_SIZE);
3658 			goto use_config_on_flash;
3659 		}
3660 
3661 		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3662 		if (rc != 0) {
3663 			device_printf(sc->dev,
3664 			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3665 			    "Will try to use the config on the card, if any.\n",
3666 			    __func__, mtype, moff, cflen, rc);
3667 			goto use_config_on_flash;
3668 		}
3669 		write_via_memwin(sc, 2, addr, cfdata, cflen);
3670 	} else {
3671 use_config_on_flash:
3672 		mtype = FW_MEMTYPE_FLASH;
3673 		moff = t4_flash_cfg_addr(sc);
3674 	}
3675 
3676 	bzero(&caps, sizeof(caps));
3677 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3678 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3679 	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3680 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3681 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3682 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3683 	if (rc != 0) {
3684 		device_printf(sc->dev,
3685 		    "failed to pre-process config file: %d "
3686 		    "(mtype %d, moff 0x%x).  Will reset the firmware and retry "
3687 		    "with the built-in configuration.\n", rc, mtype, moff);
3688 
3689 	    	rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
3690 		if (rc != 0) {
3691 			device_printf(sc->dev,
3692 			    "firmware reset failed: %d.\n", rc);
3693 			if (rc != ETIMEDOUT && rc != EIO) {
3694 				t4_fw_bye(sc, sc->mbox);
3695 				sc->flags &= ~FW_OK;
3696 			}
3697 			goto done;
3698 		}
3699 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", "built-in");
3700 use_built_in_config:
3701 		bzero(&caps, sizeof(caps));
3702 		caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3703 		    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3704 		caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3705 		rc = t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3706 		if (rc != 0) {
3707 			device_printf(sc->dev,
3708 			    "built-in configuration failed: %d.\n", rc);
3709 			goto done;
3710 		}
3711 	}
3712 
3713 	finicsum = be32toh(caps.finicsum);
3714 	cfcsum = be32toh(caps.cfcsum);
3715 	if (finicsum != cfcsum) {
3716 		device_printf(sc->dev,
3717 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3718 		    finicsum, cfcsum);
3719 	}
3720 	sc->cfcsum = cfcsum;
3721 
3722 #define LIMIT_CAPS(x) do { \
3723 	caps.x &= htobe16(t4_##x##_allowed); \
3724 } while (0)
3725 
3726 	/*
3727 	 * Let the firmware know what features will (not) be used so it can tune
3728 	 * things accordingly.
3729 	 */
3730 	LIMIT_CAPS(nbmcaps);
3731 	LIMIT_CAPS(linkcaps);
3732 	LIMIT_CAPS(switchcaps);
3733 	LIMIT_CAPS(niccaps);
3734 	LIMIT_CAPS(toecaps);
3735 	LIMIT_CAPS(rdmacaps);
3736 	LIMIT_CAPS(cryptocaps);
3737 	LIMIT_CAPS(iscsicaps);
3738 	LIMIT_CAPS(fcoecaps);
3739 #undef LIMIT_CAPS
3740 
3741 	if (caps.niccaps & htobe16(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
3742 		/*
3743 		 * TOE and hashfilters are mutually exclusive.  It is a config
3744 		 * file or firmware bug if both are reported as available.  Try
3745 		 * to cope with the situation in non-debug builds by disabling
3746 		 * TOE.
3747 		 */
3748 		MPASS(caps.toecaps == 0);
3749 
3750 		caps.toecaps = 0;
3751 		caps.rdmacaps = 0;
3752 		caps.iscsicaps = 0;
3753 	}
3754 
3755 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3756 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3757 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3758 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3759 	if (rc != 0) {
3760 		device_printf(sc->dev,
3761 		    "failed to process config file: %d.\n", rc);
3762 	}
3763 done:
3764 	if (cfg != NULL)
3765 		firmware_put(cfg, FIRMWARE_UNLOAD);
3766 	return (rc);
3767 }
3768 
3769 /*
3770  * Retrieve parameters that are needed (or nice to have) very early.
3771  */
3772 static int
3773 get_params__pre_init(struct adapter *sc)
3774 {
3775 	int rc;
3776 	uint32_t param[2], val[2];
3777 
3778 	t4_get_version_info(sc);
3779 
3780 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3781 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3782 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3783 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3784 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3785 
3786 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3787 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3788 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3789 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3790 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3791 
3792 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3793 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3794 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3795 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3796 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3797 
3798 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3799 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3800 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3801 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3802 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3803 
3804 	param[0] = FW_PARAM_DEV(PORTVEC);
3805 	param[1] = FW_PARAM_DEV(CCLK);
3806 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3807 	if (rc != 0) {
3808 		device_printf(sc->dev,
3809 		    "failed to query parameters (pre_init): %d.\n", rc);
3810 		return (rc);
3811 	}
3812 
3813 	sc->params.portvec = val[0];
3814 	sc->params.nports = bitcount32(val[0]);
3815 	sc->params.vpd.cclk = val[1];
3816 
3817 	/* Read device log parameters. */
3818 	rc = -t4_init_devlog_params(sc, 1);
3819 	if (rc == 0)
3820 		fixup_devlog_params(sc);
3821 	else {
3822 		device_printf(sc->dev,
3823 		    "failed to get devlog parameters: %d.\n", rc);
3824 		rc = 0;	/* devlog isn't critical for device operation */
3825 	}
3826 
3827 	return (rc);
3828 }
3829 
3830 /*
3831  * Retrieve various parameters that are of interest to the driver.  The device
3832  * has been initialized by the firmware at this point.
3833  */
3834 static int
3835 get_params__post_init(struct adapter *sc)
3836 {
3837 	int rc;
3838 	uint32_t param[7], val[7];
3839 	struct fw_caps_config_cmd caps;
3840 
3841 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3842 	param[1] = FW_PARAM_PFVF(EQ_START);
3843 	param[2] = FW_PARAM_PFVF(FILTER_START);
3844 	param[3] = FW_PARAM_PFVF(FILTER_END);
3845 	param[4] = FW_PARAM_PFVF(L2T_START);
3846 	param[5] = FW_PARAM_PFVF(L2T_END);
3847 	param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
3848 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
3849 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
3850 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
3851 	if (rc != 0) {
3852 		device_printf(sc->dev,
3853 		    "failed to query parameters (post_init): %d.\n", rc);
3854 		return (rc);
3855 	}
3856 
3857 	sc->sge.iq_start = val[0];
3858 	sc->sge.eq_start = val[1];
3859 	if ((int)val[3] > (int)val[2]) {
3860 		sc->tids.ftid_base = val[2];
3861 		sc->tids.ftid_end = val[3];
3862 		sc->tids.nftids = val[3] - val[2] + 1;
3863 	}
3864 	sc->vres.l2t.start = val[4];
3865 	sc->vres.l2t.size = val[5] - val[4] + 1;
3866 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3867 	    ("%s: L2 table size (%u) larger than expected (%u)",
3868 	    __func__, sc->vres.l2t.size, L2T_SIZE));
3869 	sc->params.core_vdd = val[6];
3870 
3871 	if (chip_id(sc) >= CHELSIO_T6) {
3872 
3873 #ifdef INVARIANTS
3874 		if (sc->params.fw_vers >=
3875 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
3876 		    V_FW_HDR_FW_VER_MICRO(1) | V_FW_HDR_FW_VER_BUILD(0))) {
3877 			/*
3878 			 * Note that the code to enable the region should run
3879 			 * before t4_fw_initialize and not here.  This is just a
3880 			 * reminder to add said code.
3881 			 */
3882 			device_printf(sc->dev,
3883 			    "hpfilter region not enabled.\n");
3884 		}
3885 #endif
3886 
3887 		sc->tids.tid_base = t4_read_reg(sc,
3888 		    A_LE_DB_ACTIVE_TABLE_START_INDEX);
3889 
3890 		param[0] = FW_PARAM_PFVF(HPFILTER_START);
3891 		param[1] = FW_PARAM_PFVF(HPFILTER_END);
3892 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3893 		if (rc != 0) {
3894 			device_printf(sc->dev,
3895 			   "failed to query hpfilter parameters: %d.\n", rc);
3896 			return (rc);
3897 		}
3898 		if ((int)val[1] > (int)val[0]) {
3899 			sc->tids.hpftid_base = val[0];
3900 			sc->tids.hpftid_end = val[1];
3901 			sc->tids.nhpftids = val[1] - val[0] + 1;
3902 
3903 			/*
3904 			 * These should go off if the layout changes and the
3905 			 * driver needs to catch up.
3906 			 */
3907 			MPASS(sc->tids.hpftid_base == 0);
3908 			MPASS(sc->tids.tid_base == sc->tids.nhpftids);
3909 		}
3910 	}
3911 
3912 	/*
3913 	 * MPSBGMAP is queried separately because only recent firmwares support
3914 	 * it as a parameter and we don't want the compound query above to fail
3915 	 * on older firmwares.
3916 	 */
3917 	param[0] = FW_PARAM_DEV(MPSBGMAP);
3918 	val[0] = 0;
3919 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3920 	if (rc == 0)
3921 		sc->params.mps_bg_map = val[0];
3922 	else
3923 		sc->params.mps_bg_map = 0;
3924 
3925 	/*
3926 	 * Determine whether the firmware supports the filter2 work request.
3927 	 * This is queried separately for the same reason as MPSBGMAP above.
3928 	 */
3929 	param[0] = FW_PARAM_DEV(FILTER2_WR);
3930 	val[0] = 0;
3931 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3932 	if (rc == 0)
3933 		sc->params.filter2_wr_support = val[0] != 0;
3934 	else
3935 		sc->params.filter2_wr_support = 0;
3936 
3937 	/* get capabilites */
3938 	bzero(&caps, sizeof(caps));
3939 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3940 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3941 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3942 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3943 	if (rc != 0) {
3944 		device_printf(sc->dev,
3945 		    "failed to get card capabilities: %d.\n", rc);
3946 		return (rc);
3947 	}
3948 
3949 #define READ_CAPS(x) do { \
3950 	sc->x = htobe16(caps.x); \
3951 } while (0)
3952 	READ_CAPS(nbmcaps);
3953 	READ_CAPS(linkcaps);
3954 	READ_CAPS(switchcaps);
3955 	READ_CAPS(niccaps);
3956 	READ_CAPS(toecaps);
3957 	READ_CAPS(rdmacaps);
3958 	READ_CAPS(cryptocaps);
3959 	READ_CAPS(iscsicaps);
3960 	READ_CAPS(fcoecaps);
3961 
3962 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_HASHFILTER) {
3963 		MPASS(chip_id(sc) > CHELSIO_T4);
3964 		MPASS(sc->toecaps == 0);
3965 		sc->toecaps = 0;
3966 
3967 		param[0] = FW_PARAM_DEV(NTID);
3968 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
3969 		if (rc != 0) {
3970 			device_printf(sc->dev,
3971 			    "failed to query HASHFILTER parameters: %d.\n", rc);
3972 			return (rc);
3973 		}
3974 		sc->tids.ntids = val[0];
3975 		if (sc->params.fw_vers <
3976 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
3977 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
3978 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
3979 			sc->tids.ntids -= sc->tids.nhpftids;
3980 		}
3981 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3982 		sc->params.hash_filter = 1;
3983 	}
3984 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3985 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3986 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3987 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3988 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3989 		if (rc != 0) {
3990 			device_printf(sc->dev,
3991 			    "failed to query NIC parameters: %d.\n", rc);
3992 			return (rc);
3993 		}
3994 		if ((int)val[1] > (int)val[0]) {
3995 			sc->tids.etid_base = val[0];
3996 			sc->tids.etid_end = val[1];
3997 			sc->tids.netids = val[1] - val[0] + 1;
3998 			sc->params.eo_wr_cred = val[2];
3999 			sc->params.ethoffload = 1;
4000 		}
4001 	}
4002 	if (sc->toecaps) {
4003 		/* query offload-related parameters */
4004 		param[0] = FW_PARAM_DEV(NTID);
4005 		param[1] = FW_PARAM_PFVF(SERVER_START);
4006 		param[2] = FW_PARAM_PFVF(SERVER_END);
4007 		param[3] = FW_PARAM_PFVF(TDDP_START);
4008 		param[4] = FW_PARAM_PFVF(TDDP_END);
4009 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4010 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4011 		if (rc != 0) {
4012 			device_printf(sc->dev,
4013 			    "failed to query TOE parameters: %d.\n", rc);
4014 			return (rc);
4015 		}
4016 		sc->tids.ntids = val[0];
4017 		if (sc->params.fw_vers <
4018 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4019 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4020 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4021 			sc->tids.ntids -= sc->tids.nhpftids;
4022 		}
4023 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4024 		if ((int)val[2] > (int)val[1]) {
4025 			sc->tids.stid_base = val[1];
4026 			sc->tids.nstids = val[2] - val[1] + 1;
4027 		}
4028 		sc->vres.ddp.start = val[3];
4029 		sc->vres.ddp.size = val[4] - val[3] + 1;
4030 		sc->params.ofldq_wr_cred = val[5];
4031 		sc->params.offload = 1;
4032 	} else {
4033 		/*
4034 		 * The firmware attempts memfree TOE configuration for -SO cards
4035 		 * and will report toecaps=0 if it runs out of resources (this
4036 		 * depends on the config file).  It may not report 0 for other
4037 		 * capabilities dependent on the TOE in this case.  Set them to
4038 		 * 0 here so that the driver doesn't bother tracking resources
4039 		 * that will never be used.
4040 		 */
4041 		sc->iscsicaps = 0;
4042 		sc->rdmacaps = 0;
4043 	}
4044 	if (sc->rdmacaps) {
4045 		param[0] = FW_PARAM_PFVF(STAG_START);
4046 		param[1] = FW_PARAM_PFVF(STAG_END);
4047 		param[2] = FW_PARAM_PFVF(RQ_START);
4048 		param[3] = FW_PARAM_PFVF(RQ_END);
4049 		param[4] = FW_PARAM_PFVF(PBL_START);
4050 		param[5] = FW_PARAM_PFVF(PBL_END);
4051 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4052 		if (rc != 0) {
4053 			device_printf(sc->dev,
4054 			    "failed to query RDMA parameters(1): %d.\n", rc);
4055 			return (rc);
4056 		}
4057 		sc->vres.stag.start = val[0];
4058 		sc->vres.stag.size = val[1] - val[0] + 1;
4059 		sc->vres.rq.start = val[2];
4060 		sc->vres.rq.size = val[3] - val[2] + 1;
4061 		sc->vres.pbl.start = val[4];
4062 		sc->vres.pbl.size = val[5] - val[4] + 1;
4063 
4064 		param[0] = FW_PARAM_PFVF(SQRQ_START);
4065 		param[1] = FW_PARAM_PFVF(SQRQ_END);
4066 		param[2] = FW_PARAM_PFVF(CQ_START);
4067 		param[3] = FW_PARAM_PFVF(CQ_END);
4068 		param[4] = FW_PARAM_PFVF(OCQ_START);
4069 		param[5] = FW_PARAM_PFVF(OCQ_END);
4070 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4071 		if (rc != 0) {
4072 			device_printf(sc->dev,
4073 			    "failed to query RDMA parameters(2): %d.\n", rc);
4074 			return (rc);
4075 		}
4076 		sc->vres.qp.start = val[0];
4077 		sc->vres.qp.size = val[1] - val[0] + 1;
4078 		sc->vres.cq.start = val[2];
4079 		sc->vres.cq.size = val[3] - val[2] + 1;
4080 		sc->vres.ocq.start = val[4];
4081 		sc->vres.ocq.size = val[5] - val[4] + 1;
4082 
4083 		param[0] = FW_PARAM_PFVF(SRQ_START);
4084 		param[1] = FW_PARAM_PFVF(SRQ_END);
4085 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
4086 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4087 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
4088 		if (rc != 0) {
4089 			device_printf(sc->dev,
4090 			    "failed to query RDMA parameters(3): %d.\n", rc);
4091 			return (rc);
4092 		}
4093 		sc->vres.srq.start = val[0];
4094 		sc->vres.srq.size = val[1] - val[0] + 1;
4095 		sc->params.max_ordird_qp = val[2];
4096 		sc->params.max_ird_adapter = val[3];
4097 	}
4098 	if (sc->iscsicaps) {
4099 		param[0] = FW_PARAM_PFVF(ISCSI_START);
4100 		param[1] = FW_PARAM_PFVF(ISCSI_END);
4101 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4102 		if (rc != 0) {
4103 			device_printf(sc->dev,
4104 			    "failed to query iSCSI parameters: %d.\n", rc);
4105 			return (rc);
4106 		}
4107 		sc->vres.iscsi.start = val[0];
4108 		sc->vres.iscsi.size = val[1] - val[0] + 1;
4109 	}
4110 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
4111 		param[0] = FW_PARAM_PFVF(TLS_START);
4112 		param[1] = FW_PARAM_PFVF(TLS_END);
4113 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4114 		if (rc != 0) {
4115 			device_printf(sc->dev,
4116 			    "failed to query TLS parameters: %d.\n", rc);
4117 			return (rc);
4118 		}
4119 		sc->vres.key.start = val[0];
4120 		sc->vres.key.size = val[1] - val[0] + 1;
4121 	}
4122 
4123 	t4_init_sge_params(sc);
4124 
4125 	/*
4126 	 * We've got the params we wanted to query via the firmware.  Now grab
4127 	 * some others directly from the chip.
4128 	 */
4129 	rc = t4_read_chip_settings(sc);
4130 
4131 	return (rc);
4132 }
4133 
4134 static int
4135 set_params__post_init(struct adapter *sc)
4136 {
4137 	uint32_t param, val;
4138 #ifdef TCP_OFFLOAD
4139 	int i, v, shift;
4140 #endif
4141 
4142 	/* ask for encapsulated CPLs */
4143 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4144 	val = 1;
4145 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4146 
4147 	/* Enable 32b port caps if the firmware supports it. */
4148 	param = FW_PARAM_PFVF(PORT_CAPS32);
4149 	val = 1;
4150 	if (t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val) == 0)
4151 		sc->params.port_caps32 = 1;
4152 
4153 #ifdef TCP_OFFLOAD
4154 	/*
4155 	 * Override the TOE timers with user provided tunables.  This is not the
4156 	 * recommended way to change the timers (the firmware config file is) so
4157 	 * these tunables are not documented.
4158 	 *
4159 	 * All the timer tunables are in microseconds.
4160 	 */
4161 	if (t4_toe_keepalive_idle != 0) {
4162 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
4163 		v &= M_KEEPALIVEIDLE;
4164 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
4165 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
4166 	}
4167 	if (t4_toe_keepalive_interval != 0) {
4168 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
4169 		v &= M_KEEPALIVEINTVL;
4170 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
4171 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
4172 	}
4173 	if (t4_toe_keepalive_count != 0) {
4174 		v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
4175 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4176 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
4177 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
4178 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
4179 	}
4180 	if (t4_toe_rexmt_min != 0) {
4181 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
4182 		v &= M_RXTMIN;
4183 		t4_set_reg_field(sc, A_TP_RXT_MIN,
4184 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
4185 	}
4186 	if (t4_toe_rexmt_max != 0) {
4187 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
4188 		v &= M_RXTMAX;
4189 		t4_set_reg_field(sc, A_TP_RXT_MAX,
4190 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
4191 	}
4192 	if (t4_toe_rexmt_count != 0) {
4193 		v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
4194 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4195 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
4196 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
4197 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
4198 	}
4199 	for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
4200 		if (t4_toe_rexmt_backoff[i] != -1) {
4201 			v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
4202 			shift = (i & 3) << 3;
4203 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
4204 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
4205 		}
4206 	}
4207 #endif
4208 	return (0);
4209 }
4210 
4211 #undef FW_PARAM_PFVF
4212 #undef FW_PARAM_DEV
4213 
4214 static void
4215 t4_set_desc(struct adapter *sc)
4216 {
4217 	char buf[128];
4218 	struct adapter_params *p = &sc->params;
4219 
4220 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
4221 
4222 	device_set_desc_copy(sc->dev, buf);
4223 }
4224 
4225 static inline void
4226 ifmedia_add4(struct ifmedia *ifm, int m)
4227 {
4228 
4229 	ifmedia_add(ifm, m, 0, NULL);
4230 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE, 0, NULL);
4231 	ifmedia_add(ifm, m | IFM_ETH_RXPAUSE, 0, NULL);
4232 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE, 0, NULL);
4233 }
4234 
4235 /*
4236  * This is the selected media, which is not quite the same as the active media.
4237  * The media line in ifconfig is "media: Ethernet selected (active)" if selected
4238  * and active are not the same, and "media: Ethernet selected" otherwise.
4239  */
4240 static void
4241 set_current_media(struct port_info *pi)
4242 {
4243 	struct link_config *lc;
4244 	struct ifmedia *ifm;
4245 	int mword;
4246 	u_int speed;
4247 
4248 	PORT_LOCK_ASSERT_OWNED(pi);
4249 
4250 	/* Leave current media alone if it's already set to IFM_NONE. */
4251 	ifm = &pi->media;
4252 	if (ifm->ifm_cur != NULL &&
4253 	    IFM_SUBTYPE(ifm->ifm_cur->ifm_media) == IFM_NONE)
4254 		return;
4255 
4256 	lc = &pi->link_cfg;
4257 	if (lc->requested_aneg != AUTONEG_DISABLE &&
4258 	    lc->supported & FW_PORT_CAP32_ANEG) {
4259 		ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
4260 		return;
4261 	}
4262 	mword = IFM_ETHER | IFM_FDX;
4263 	if (lc->requested_fc & PAUSE_TX)
4264 		mword |= IFM_ETH_TXPAUSE;
4265 	if (lc->requested_fc & PAUSE_RX)
4266 		mword |= IFM_ETH_RXPAUSE;
4267 	if (lc->requested_speed == 0)
4268 		speed = port_top_speed(pi) * 1000;	/* Gbps -> Mbps */
4269 	else
4270 		speed = lc->requested_speed;
4271 	mword |= port_mword(pi, speed_to_fwcap(speed));
4272 	ifmedia_set(ifm, mword);
4273 }
4274 
4275 /*
4276  * Returns true if the ifmedia list for the port cannot change.
4277  */
4278 static bool
4279 fixed_ifmedia(struct port_info *pi)
4280 {
4281 
4282 	return (pi->port_type == FW_PORT_TYPE_BT_SGMII ||
4283 	    pi->port_type == FW_PORT_TYPE_BT_XFI ||
4284 	    pi->port_type == FW_PORT_TYPE_BT_XAUI ||
4285 	    pi->port_type == FW_PORT_TYPE_KX4 ||
4286 	    pi->port_type == FW_PORT_TYPE_KX ||
4287 	    pi->port_type == FW_PORT_TYPE_KR ||
4288 	    pi->port_type == FW_PORT_TYPE_BP_AP ||
4289 	    pi->port_type == FW_PORT_TYPE_BP4_AP ||
4290 	    pi->port_type == FW_PORT_TYPE_BP40_BA ||
4291 	    pi->port_type == FW_PORT_TYPE_KR4_100G ||
4292 	    pi->port_type == FW_PORT_TYPE_KR_SFP28 ||
4293 	    pi->port_type == FW_PORT_TYPE_KR_XLAUI);
4294 }
4295 
4296 static void
4297 build_medialist(struct port_info *pi)
4298 {
4299 	uint32_t ss, speed;
4300 	int unknown, mword, bit;
4301 	struct link_config *lc;
4302 	struct ifmedia *ifm;
4303 
4304 	PORT_LOCK_ASSERT_OWNED(pi);
4305 
4306 	if (pi->flags & FIXED_IFMEDIA)
4307 		return;
4308 
4309 	/*
4310 	 * Rebuild the ifmedia list.
4311 	 */
4312 	ifm = &pi->media;
4313 	ifmedia_removeall(ifm);
4314 	lc = &pi->link_cfg;
4315 	ss = G_FW_PORT_CAP32_SPEED(lc->supported); /* Supported Speeds */
4316 	if (__predict_false(ss == 0)) {	/* not supposed to happen. */
4317 		MPASS(ss != 0);
4318 no_media:
4319 		MPASS(LIST_EMPTY(&ifm->ifm_list));
4320 		ifmedia_add(ifm, IFM_ETHER | IFM_NONE, 0, NULL);
4321 		ifmedia_set(ifm, IFM_ETHER | IFM_NONE);
4322 		return;
4323 	}
4324 
4325 	unknown = 0;
4326 	for (bit = S_FW_PORT_CAP32_SPEED; bit < fls(ss); bit++) {
4327 		speed = 1 << bit;
4328 		MPASS(speed & M_FW_PORT_CAP32_SPEED);
4329 		if (ss & speed) {
4330 			mword = port_mword(pi, speed);
4331 			if (mword == IFM_NONE) {
4332 				goto no_media;
4333 			} else if (mword == IFM_UNKNOWN)
4334 				unknown++;
4335 			else
4336 				ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | mword);
4337 		}
4338 	}
4339 	if (unknown > 0) /* Add one unknown for all unknown media types. */
4340 		ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | IFM_UNKNOWN);
4341 	if (lc->supported & FW_PORT_CAP32_ANEG)
4342 		ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
4343 
4344 	set_current_media(pi);
4345 }
4346 
4347 /*
4348  * Initialize the requested fields in the link config based on driver tunables.
4349  */
4350 static void
4351 init_link_config(struct port_info *pi)
4352 {
4353 	struct link_config *lc = &pi->link_cfg;
4354 
4355 	PORT_LOCK_ASSERT_OWNED(pi);
4356 
4357 	lc->requested_speed = 0;
4358 
4359 	if (t4_autoneg == 0)
4360 		lc->requested_aneg = AUTONEG_DISABLE;
4361 	else if (t4_autoneg == 1)
4362 		lc->requested_aneg = AUTONEG_ENABLE;
4363 	else
4364 		lc->requested_aneg = AUTONEG_AUTO;
4365 
4366 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX |
4367 	    PAUSE_AUTONEG);
4368 
4369 	if (t4_fec == -1 || t4_fec & FEC_AUTO)
4370 		lc->requested_fec = FEC_AUTO;
4371 	else {
4372 		lc->requested_fec = FEC_NONE;
4373 		if (t4_fec & FEC_RS)
4374 			lc->requested_fec |= FEC_RS;
4375 		if (t4_fec & FEC_BASER_RS)
4376 			lc->requested_fec |= FEC_BASER_RS;
4377 	}
4378 }
4379 
4380 /*
4381  * Makes sure that all requested settings comply with what's supported by the
4382  * port.  Returns the number of settings that were invalid and had to be fixed.
4383  */
4384 static int
4385 fixup_link_config(struct port_info *pi)
4386 {
4387 	int n = 0;
4388 	struct link_config *lc = &pi->link_cfg;
4389 	uint32_t fwspeed;
4390 
4391 	PORT_LOCK_ASSERT_OWNED(pi);
4392 
4393 	/* Speed (when not autonegotiating) */
4394 	if (lc->requested_speed != 0) {
4395 		fwspeed = speed_to_fwcap(lc->requested_speed);
4396 		if ((fwspeed & lc->supported) == 0) {
4397 			n++;
4398 			lc->requested_speed = 0;
4399 		}
4400 	}
4401 
4402 	/* Link autonegotiation */
4403 	MPASS(lc->requested_aneg == AUTONEG_ENABLE ||
4404 	    lc->requested_aneg == AUTONEG_DISABLE ||
4405 	    lc->requested_aneg == AUTONEG_AUTO);
4406 	if (lc->requested_aneg == AUTONEG_ENABLE &&
4407 	    !(lc->supported & FW_PORT_CAP32_ANEG)) {
4408 		n++;
4409 		lc->requested_aneg = AUTONEG_AUTO;
4410 	}
4411 
4412 	/* Flow control */
4413 	MPASS((lc->requested_fc & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG)) == 0);
4414 	if (lc->requested_fc & PAUSE_TX &&
4415 	    !(lc->supported & FW_PORT_CAP32_FC_TX)) {
4416 		n++;
4417 		lc->requested_fc &= ~PAUSE_TX;
4418 	}
4419 	if (lc->requested_fc & PAUSE_RX &&
4420 	    !(lc->supported & FW_PORT_CAP32_FC_RX)) {
4421 		n++;
4422 		lc->requested_fc &= ~PAUSE_RX;
4423 	}
4424 	if (!(lc->requested_fc & PAUSE_AUTONEG) &&
4425 	    !(lc->supported & FW_PORT_CAP32_FORCE_PAUSE)) {
4426 		n++;
4427 		lc->requested_fc |= PAUSE_AUTONEG;
4428 	}
4429 
4430 	/* FEC */
4431 	if ((lc->requested_fec & FEC_RS &&
4432 	    !(lc->supported & FW_PORT_CAP32_FEC_RS)) ||
4433 	    (lc->requested_fec & FEC_BASER_RS &&
4434 	    !(lc->supported & FW_PORT_CAP32_FEC_BASER_RS))) {
4435 		n++;
4436 		lc->requested_fec = FEC_AUTO;
4437 	}
4438 
4439 	return (n);
4440 }
4441 
4442 /*
4443  * Apply the requested L1 settings, which are expected to be valid, to the
4444  * hardware.
4445  */
4446 static int
4447 apply_link_config(struct port_info *pi)
4448 {
4449 	struct adapter *sc = pi->adapter;
4450 	struct link_config *lc = &pi->link_cfg;
4451 	int rc;
4452 
4453 #ifdef INVARIANTS
4454 	ASSERT_SYNCHRONIZED_OP(sc);
4455 	PORT_LOCK_ASSERT_OWNED(pi);
4456 
4457 	if (lc->requested_aneg == AUTONEG_ENABLE)
4458 		MPASS(lc->supported & FW_PORT_CAP32_ANEG);
4459 	if (!(lc->requested_fc & PAUSE_AUTONEG))
4460 		MPASS(lc->supported & FW_PORT_CAP32_FORCE_PAUSE);
4461 	if (lc->requested_fc & PAUSE_TX)
4462 		MPASS(lc->supported & FW_PORT_CAP32_FC_TX);
4463 	if (lc->requested_fc & PAUSE_RX)
4464 		MPASS(lc->supported & FW_PORT_CAP32_FC_RX);
4465 	if (lc->requested_fec & FEC_RS)
4466 		MPASS(lc->supported & FW_PORT_CAP32_FEC_RS);
4467 	if (lc->requested_fec & FEC_BASER_RS)
4468 		MPASS(lc->supported & FW_PORT_CAP32_FEC_BASER_RS);
4469 #endif
4470 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
4471 	if (rc != 0) {
4472 		/* Don't complain if the VF driver gets back an EPERM. */
4473 		if (!(sc->flags & IS_VF) || rc != FW_EPERM)
4474 			device_printf(pi->dev, "l1cfg failed: %d\n", rc);
4475 	} else {
4476 		/*
4477 		 * An L1_CFG will almost always result in a link-change event if
4478 		 * the link is up, and the driver will refresh the actual
4479 		 * fec/fc/etc. when the notification is processed.  If the link
4480 		 * is down then the actual settings are meaningless.
4481 		 *
4482 		 * This takes care of the case where a change in the L1 settings
4483 		 * may not result in a notification.
4484 		 */
4485 		if (lc->link_ok && !(lc->requested_fc & PAUSE_AUTONEG))
4486 			lc->fc = lc->requested_fc & (PAUSE_TX | PAUSE_RX);
4487 	}
4488 	return (rc);
4489 }
4490 
4491 #define FW_MAC_EXACT_CHUNK	7
4492 
4493 /*
4494  * Program the port's XGMAC based on parameters in ifnet.  The caller also
4495  * indicates which parameters should be programmed (the rest are left alone).
4496  */
4497 int
4498 update_mac_settings(struct ifnet *ifp, int flags)
4499 {
4500 	int rc = 0;
4501 	struct vi_info *vi = ifp->if_softc;
4502 	struct port_info *pi = vi->pi;
4503 	struct adapter *sc = pi->adapter;
4504 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4505 
4506 	ASSERT_SYNCHRONIZED_OP(sc);
4507 	KASSERT(flags, ("%s: not told what to update.", __func__));
4508 
4509 	if (flags & XGMAC_MTU)
4510 		mtu = ifp->if_mtu;
4511 
4512 	if (flags & XGMAC_PROMISC)
4513 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4514 
4515 	if (flags & XGMAC_ALLMULTI)
4516 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4517 
4518 	if (flags & XGMAC_VLANEX)
4519 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4520 
4521 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4522 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4523 		    allmulti, 1, vlanex, false);
4524 		if (rc) {
4525 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4526 			    rc);
4527 			return (rc);
4528 		}
4529 	}
4530 
4531 	if (flags & XGMAC_UCADDR) {
4532 		uint8_t ucaddr[ETHER_ADDR_LEN];
4533 
4534 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4535 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4536 		    ucaddr, true, true);
4537 		if (rc < 0) {
4538 			rc = -rc;
4539 			if_printf(ifp, "change_mac failed: %d\n", rc);
4540 			return (rc);
4541 		} else {
4542 			vi->xact_addr_filt = rc;
4543 			rc = 0;
4544 		}
4545 	}
4546 
4547 	if (flags & XGMAC_MCADDRS) {
4548 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4549 		int del = 1;
4550 		uint64_t hash = 0;
4551 		struct ifmultiaddr *ifma;
4552 		int i = 0, j;
4553 
4554 		if_maddr_rlock(ifp);
4555 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4556 			if (ifma->ifma_addr->sa_family != AF_LINK)
4557 				continue;
4558 			mcaddr[i] =
4559 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4560 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4561 			i++;
4562 
4563 			if (i == FW_MAC_EXACT_CHUNK) {
4564 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4565 				    del, i, mcaddr, NULL, &hash, 0);
4566 				if (rc < 0) {
4567 					rc = -rc;
4568 					for (j = 0; j < i; j++) {
4569 						if_printf(ifp,
4570 						    "failed to add mc address"
4571 						    " %02x:%02x:%02x:"
4572 						    "%02x:%02x:%02x rc=%d\n",
4573 						    mcaddr[j][0], mcaddr[j][1],
4574 						    mcaddr[j][2], mcaddr[j][3],
4575 						    mcaddr[j][4], mcaddr[j][5],
4576 						    rc);
4577 					}
4578 					goto mcfail;
4579 				}
4580 				del = 0;
4581 				i = 0;
4582 			}
4583 		}
4584 		if (i > 0) {
4585 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4586 			    mcaddr, NULL, &hash, 0);
4587 			if (rc < 0) {
4588 				rc = -rc;
4589 				for (j = 0; j < i; j++) {
4590 					if_printf(ifp,
4591 					    "failed to add mc address"
4592 					    " %02x:%02x:%02x:"
4593 					    "%02x:%02x:%02x rc=%d\n",
4594 					    mcaddr[j][0], mcaddr[j][1],
4595 					    mcaddr[j][2], mcaddr[j][3],
4596 					    mcaddr[j][4], mcaddr[j][5],
4597 					    rc);
4598 				}
4599 				goto mcfail;
4600 			}
4601 		}
4602 
4603 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4604 		if (rc != 0)
4605 			if_printf(ifp, "failed to set mc address hash: %d", rc);
4606 mcfail:
4607 		if_maddr_runlock(ifp);
4608 	}
4609 
4610 	return (rc);
4611 }
4612 
4613 /*
4614  * {begin|end}_synchronized_op must be called from the same thread.
4615  */
4616 int
4617 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4618     char *wmesg)
4619 {
4620 	int rc, pri;
4621 
4622 #ifdef WITNESS
4623 	/* the caller thinks it's ok to sleep, but is it really? */
4624 	if (flags & SLEEP_OK)
4625 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4626 		    "begin_synchronized_op");
4627 #endif
4628 
4629 	if (INTR_OK)
4630 		pri = PCATCH;
4631 	else
4632 		pri = 0;
4633 
4634 	ADAPTER_LOCK(sc);
4635 	for (;;) {
4636 
4637 		if (vi && IS_DOOMED(vi)) {
4638 			rc = ENXIO;
4639 			goto done;
4640 		}
4641 
4642 		if (!IS_BUSY(sc)) {
4643 			rc = 0;
4644 			break;
4645 		}
4646 
4647 		if (!(flags & SLEEP_OK)) {
4648 			rc = EBUSY;
4649 			goto done;
4650 		}
4651 
4652 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4653 			rc = EINTR;
4654 			goto done;
4655 		}
4656 	}
4657 
4658 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4659 	SET_BUSY(sc);
4660 #ifdef INVARIANTS
4661 	sc->last_op = wmesg;
4662 	sc->last_op_thr = curthread;
4663 	sc->last_op_flags = flags;
4664 #endif
4665 
4666 done:
4667 	if (!(flags & HOLD_LOCK) || rc)
4668 		ADAPTER_UNLOCK(sc);
4669 
4670 	return (rc);
4671 }
4672 
4673 /*
4674  * Tell if_ioctl and if_init that the VI is going away.  This is
4675  * special variant of begin_synchronized_op and must be paired with a
4676  * call to end_synchronized_op.
4677  */
4678 void
4679 doom_vi(struct adapter *sc, struct vi_info *vi)
4680 {
4681 
4682 	ADAPTER_LOCK(sc);
4683 	SET_DOOMED(vi);
4684 	wakeup(&sc->flags);
4685 	while (IS_BUSY(sc))
4686 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4687 	SET_BUSY(sc);
4688 #ifdef INVARIANTS
4689 	sc->last_op = "t4detach";
4690 	sc->last_op_thr = curthread;
4691 	sc->last_op_flags = 0;
4692 #endif
4693 	ADAPTER_UNLOCK(sc);
4694 }
4695 
4696 /*
4697  * {begin|end}_synchronized_op must be called from the same thread.
4698  */
4699 void
4700 end_synchronized_op(struct adapter *sc, int flags)
4701 {
4702 
4703 	if (flags & LOCK_HELD)
4704 		ADAPTER_LOCK_ASSERT_OWNED(sc);
4705 	else
4706 		ADAPTER_LOCK(sc);
4707 
4708 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
4709 	CLR_BUSY(sc);
4710 	wakeup(&sc->flags);
4711 	ADAPTER_UNLOCK(sc);
4712 }
4713 
4714 static int
4715 cxgbe_init_synchronized(struct vi_info *vi)
4716 {
4717 	struct port_info *pi = vi->pi;
4718 	struct adapter *sc = pi->adapter;
4719 	struct ifnet *ifp = vi->ifp;
4720 	int rc = 0, i;
4721 	struct sge_txq *txq;
4722 
4723 	ASSERT_SYNCHRONIZED_OP(sc);
4724 
4725 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
4726 		return (0);	/* already running */
4727 
4728 	if (!(sc->flags & FULL_INIT_DONE) &&
4729 	    ((rc = adapter_full_init(sc)) != 0))
4730 		return (rc);	/* error message displayed already */
4731 
4732 	if (!(vi->flags & VI_INIT_DONE) &&
4733 	    ((rc = vi_full_init(vi)) != 0))
4734 		return (rc); /* error message displayed already */
4735 
4736 	rc = update_mac_settings(ifp, XGMAC_ALL);
4737 	if (rc)
4738 		goto done;	/* error message displayed already */
4739 
4740 	PORT_LOCK(pi);
4741 	if (pi->up_vis == 0) {
4742 		t4_update_port_info(pi);
4743 		fixup_link_config(pi);
4744 		build_medialist(pi);
4745 		apply_link_config(pi);
4746 	}
4747 
4748 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
4749 	if (rc != 0) {
4750 		if_printf(ifp, "enable_vi failed: %d\n", rc);
4751 		PORT_UNLOCK(pi);
4752 		goto done;
4753 	}
4754 
4755 	/*
4756 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
4757 	 * if this changes.
4758 	 */
4759 
4760 	for_each_txq(vi, i, txq) {
4761 		TXQ_LOCK(txq);
4762 		txq->eq.flags |= EQ_ENABLED;
4763 		TXQ_UNLOCK(txq);
4764 	}
4765 
4766 	/*
4767 	 * The first iq of the first port to come up is used for tracing.
4768 	 */
4769 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4770 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4771 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4772 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4773 		    V_QUEUENUMBER(sc->traceq));
4774 		pi->flags |= HAS_TRACEQ;
4775 	}
4776 
4777 	/* all ok */
4778 	pi->up_vis++;
4779 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
4780 
4781 	if (pi->nvi > 1 || sc->flags & IS_VF)
4782 		callout_reset(&vi->tick, hz, vi_tick, vi);
4783 	else
4784 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4785 	PORT_UNLOCK(pi);
4786 done:
4787 	if (rc != 0)
4788 		cxgbe_uninit_synchronized(vi);
4789 
4790 	return (rc);
4791 }
4792 
4793 /*
4794  * Idempotent.
4795  */
4796 static int
4797 cxgbe_uninit_synchronized(struct vi_info *vi)
4798 {
4799 	struct port_info *pi = vi->pi;
4800 	struct adapter *sc = pi->adapter;
4801 	struct ifnet *ifp = vi->ifp;
4802 	int rc, i;
4803 	struct sge_txq *txq;
4804 
4805 	ASSERT_SYNCHRONIZED_OP(sc);
4806 
4807 	if (!(vi->flags & VI_INIT_DONE)) {
4808 		if (__predict_false(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4809 			KASSERT(0, ("uninited VI is running"));
4810 			if_printf(ifp, "uninited VI with running ifnet.  "
4811 			    "vi->flags 0x%016lx, if_flags 0x%08x, "
4812 			    "if_drv_flags 0x%08x\n", vi->flags, ifp->if_flags,
4813 			    ifp->if_drv_flags);
4814 		}
4815 		return (0);
4816 	}
4817 
4818 	/*
4819 	 * Disable the VI so that all its data in either direction is discarded
4820 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4821 	 * tick) intact as the TP can deliver negative advice or data that it's
4822 	 * holding in its RAM (for an offloaded connection) even after the VI is
4823 	 * disabled.
4824 	 */
4825 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4826 	if (rc) {
4827 		if_printf(ifp, "disable_vi failed: %d\n", rc);
4828 		return (rc);
4829 	}
4830 
4831 	for_each_txq(vi, i, txq) {
4832 		TXQ_LOCK(txq);
4833 		txq->eq.flags &= ~EQ_ENABLED;
4834 		TXQ_UNLOCK(txq);
4835 	}
4836 
4837 	PORT_LOCK(pi);
4838 	if (pi->nvi > 1 || sc->flags & IS_VF)
4839 		callout_stop(&vi->tick);
4840 	else
4841 		callout_stop(&pi->tick);
4842 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4843 		PORT_UNLOCK(pi);
4844 		return (0);
4845 	}
4846 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4847 	pi->up_vis--;
4848 	if (pi->up_vis > 0) {
4849 		PORT_UNLOCK(pi);
4850 		return (0);
4851 	}
4852 
4853 	pi->link_cfg.link_ok = false;
4854 	pi->link_cfg.speed = 0;
4855 	pi->link_cfg.link_down_rc = 255;
4856 	t4_os_link_changed(pi);
4857 	PORT_UNLOCK(pi);
4858 
4859 	return (0);
4860 }
4861 
4862 /*
4863  * It is ok for this function to fail midway and return right away.  t4_detach
4864  * will walk the entire sc->irq list and clean up whatever is valid.
4865  */
4866 int
4867 t4_setup_intr_handlers(struct adapter *sc)
4868 {
4869 	int rc, rid, p, q, v;
4870 	char s[8];
4871 	struct irq *irq;
4872 	struct port_info *pi;
4873 	struct vi_info *vi;
4874 	struct sge *sge = &sc->sge;
4875 	struct sge_rxq *rxq;
4876 #ifdef TCP_OFFLOAD
4877 	struct sge_ofld_rxq *ofld_rxq;
4878 #endif
4879 #ifdef DEV_NETMAP
4880 	struct sge_nm_rxq *nm_rxq;
4881 #endif
4882 #ifdef RSS
4883 	int nbuckets = rss_getnumbuckets();
4884 #endif
4885 
4886 	/*
4887 	 * Setup interrupts.
4888 	 */
4889 	irq = &sc->irq[0];
4890 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
4891 	if (forwarding_intr_to_fwq(sc))
4892 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4893 
4894 	/* Multiple interrupts. */
4895 	if (sc->flags & IS_VF)
4896 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4897 		    ("%s: too few intr.", __func__));
4898 	else
4899 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4900 		    ("%s: too few intr.", __func__));
4901 
4902 	/* The first one is always error intr on PFs */
4903 	if (!(sc->flags & IS_VF)) {
4904 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4905 		if (rc != 0)
4906 			return (rc);
4907 		irq++;
4908 		rid++;
4909 	}
4910 
4911 	/* The second one is always the firmware event queue (first on VFs) */
4912 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4913 	if (rc != 0)
4914 		return (rc);
4915 	irq++;
4916 	rid++;
4917 
4918 	for_each_port(sc, p) {
4919 		pi = sc->port[p];
4920 		for_each_vi(pi, v, vi) {
4921 			vi->first_intr = rid - 1;
4922 
4923 			if (vi->nnmrxq > 0) {
4924 				int n = max(vi->nrxq, vi->nnmrxq);
4925 
4926 				rxq = &sge->rxq[vi->first_rxq];
4927 #ifdef DEV_NETMAP
4928 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4929 #endif
4930 				for (q = 0; q < n; q++) {
4931 					snprintf(s, sizeof(s), "%x%c%x", p,
4932 					    'a' + v, q);
4933 					if (q < vi->nrxq)
4934 						irq->rxq = rxq++;
4935 #ifdef DEV_NETMAP
4936 					if (q < vi->nnmrxq)
4937 						irq->nm_rxq = nm_rxq++;
4938 
4939 					if (irq->nm_rxq != NULL &&
4940 					    irq->rxq == NULL) {
4941 						/* Netmap rx only */
4942 						rc = t4_alloc_irq(sc, irq, rid,
4943 						    t4_nm_intr, irq->nm_rxq, s);
4944 					}
4945 					if (irq->nm_rxq != NULL &&
4946 					    irq->rxq != NULL) {
4947 						/* NIC and Netmap rx */
4948 						rc = t4_alloc_irq(sc, irq, rid,
4949 						    t4_vi_intr, irq, s);
4950 					}
4951 #endif
4952 					if (irq->rxq != NULL &&
4953 					    irq->nm_rxq == NULL) {
4954 						/* NIC rx only */
4955 						rc = t4_alloc_irq(sc, irq, rid,
4956 						    t4_intr, irq->rxq, s);
4957 					}
4958 					if (rc != 0)
4959 						return (rc);
4960 #ifdef RSS
4961 					if (q < vi->nrxq) {
4962 						bus_bind_intr(sc->dev, irq->res,
4963 						    rss_getcpu(q % nbuckets));
4964 					}
4965 #endif
4966 					irq++;
4967 					rid++;
4968 					vi->nintr++;
4969 				}
4970 			} else {
4971 				for_each_rxq(vi, q, rxq) {
4972 					snprintf(s, sizeof(s), "%x%c%x", p,
4973 					    'a' + v, q);
4974 					rc = t4_alloc_irq(sc, irq, rid,
4975 					    t4_intr, rxq, s);
4976 					if (rc != 0)
4977 						return (rc);
4978 #ifdef RSS
4979 					bus_bind_intr(sc->dev, irq->res,
4980 					    rss_getcpu(q % nbuckets));
4981 #endif
4982 					irq++;
4983 					rid++;
4984 					vi->nintr++;
4985 				}
4986 			}
4987 #ifdef TCP_OFFLOAD
4988 			for_each_ofld_rxq(vi, q, ofld_rxq) {
4989 				snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
4990 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
4991 				    ofld_rxq, s);
4992 				if (rc != 0)
4993 					return (rc);
4994 				irq++;
4995 				rid++;
4996 				vi->nintr++;
4997 			}
4998 #endif
4999 		}
5000 	}
5001 	MPASS(irq == &sc->irq[sc->intr_count]);
5002 
5003 	return (0);
5004 }
5005 
5006 int
5007 adapter_full_init(struct adapter *sc)
5008 {
5009 	int rc, i;
5010 #ifdef RSS
5011 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5012 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5013 #endif
5014 
5015 	ASSERT_SYNCHRONIZED_OP(sc);
5016 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5017 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
5018 	    ("%s: FULL_INIT_DONE already", __func__));
5019 
5020 	/*
5021 	 * queues that belong to the adapter (not any particular port).
5022 	 */
5023 	rc = t4_setup_adapter_queues(sc);
5024 	if (rc != 0)
5025 		goto done;
5026 
5027 	for (i = 0; i < nitems(sc->tq); i++) {
5028 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
5029 		    taskqueue_thread_enqueue, &sc->tq[i]);
5030 		if (sc->tq[i] == NULL) {
5031 			device_printf(sc->dev,
5032 			    "failed to allocate task queue %d\n", i);
5033 			rc = ENOMEM;
5034 			goto done;
5035 		}
5036 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
5037 		    device_get_nameunit(sc->dev), i);
5038 	}
5039 #ifdef RSS
5040 	MPASS(RSS_KEYSIZE == 40);
5041 	rss_getkey((void *)&raw_rss_key[0]);
5042 	for (i = 0; i < nitems(rss_key); i++) {
5043 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
5044 	}
5045 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
5046 #endif
5047 
5048 	if (!(sc->flags & IS_VF))
5049 		t4_intr_enable(sc);
5050 	sc->flags |= FULL_INIT_DONE;
5051 done:
5052 	if (rc != 0)
5053 		adapter_full_uninit(sc);
5054 
5055 	return (rc);
5056 }
5057 
5058 int
5059 adapter_full_uninit(struct adapter *sc)
5060 {
5061 	int i;
5062 
5063 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5064 
5065 	t4_teardown_adapter_queues(sc);
5066 
5067 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
5068 		taskqueue_free(sc->tq[i]);
5069 		sc->tq[i] = NULL;
5070 	}
5071 
5072 	sc->flags &= ~FULL_INIT_DONE;
5073 
5074 	return (0);
5075 }
5076 
5077 #ifdef RSS
5078 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
5079     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
5080     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
5081     RSS_HASHTYPE_RSS_UDP_IPV6)
5082 
5083 /* Translates kernel hash types to hardware. */
5084 static int
5085 hashconfig_to_hashen(int hashconfig)
5086 {
5087 	int hashen = 0;
5088 
5089 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
5090 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
5091 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
5092 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
5093 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
5094 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5095 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5096 	}
5097 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
5098 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5099 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5100 	}
5101 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
5102 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5103 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
5104 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5105 
5106 	return (hashen);
5107 }
5108 
5109 /* Translates hardware hash types to kernel. */
5110 static int
5111 hashen_to_hashconfig(int hashen)
5112 {
5113 	int hashconfig = 0;
5114 
5115 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
5116 		/*
5117 		 * If UDP hashing was enabled it must have been enabled for
5118 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
5119 		 * enabling any 4-tuple hash is nonsense configuration.
5120 		 */
5121 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5122 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
5123 
5124 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5125 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
5126 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5127 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
5128 	}
5129 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5130 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
5131 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5132 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
5133 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
5134 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
5135 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
5136 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
5137 
5138 	return (hashconfig);
5139 }
5140 #endif
5141 
5142 int
5143 vi_full_init(struct vi_info *vi)
5144 {
5145 	struct adapter *sc = vi->pi->adapter;
5146 	struct ifnet *ifp = vi->ifp;
5147 	uint16_t *rss;
5148 	struct sge_rxq *rxq;
5149 	int rc, i, j, hashen;
5150 #ifdef RSS
5151 	int nbuckets = rss_getnumbuckets();
5152 	int hashconfig = rss_gethashconfig();
5153 	int extra;
5154 #endif
5155 
5156 	ASSERT_SYNCHRONIZED_OP(sc);
5157 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
5158 	    ("%s: VI_INIT_DONE already", __func__));
5159 
5160 	sysctl_ctx_init(&vi->ctx);
5161 	vi->flags |= VI_SYSCTL_CTX;
5162 
5163 	/*
5164 	 * Allocate tx/rx/fl queues for this VI.
5165 	 */
5166 	rc = t4_setup_vi_queues(vi);
5167 	if (rc != 0)
5168 		goto done;	/* error message displayed already */
5169 
5170 	/*
5171 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
5172 	 */
5173 	if (vi->nrxq > vi->rss_size) {
5174 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
5175 		    "some queues will never receive traffic.\n", vi->nrxq,
5176 		    vi->rss_size);
5177 	} else if (vi->rss_size % vi->nrxq) {
5178 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
5179 		    "expect uneven traffic distribution.\n", vi->nrxq,
5180 		    vi->rss_size);
5181 	}
5182 #ifdef RSS
5183 	if (vi->nrxq != nbuckets) {
5184 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
5185 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
5186 	}
5187 #endif
5188 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
5189 	for (i = 0; i < vi->rss_size;) {
5190 #ifdef RSS
5191 		j = rss_get_indirection_to_bucket(i);
5192 		j %= vi->nrxq;
5193 		rxq = &sc->sge.rxq[vi->first_rxq + j];
5194 		rss[i++] = rxq->iq.abs_id;
5195 #else
5196 		for_each_rxq(vi, j, rxq) {
5197 			rss[i++] = rxq->iq.abs_id;
5198 			if (i == vi->rss_size)
5199 				break;
5200 		}
5201 #endif
5202 	}
5203 
5204 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
5205 	    vi->rss_size);
5206 	if (rc != 0) {
5207 		free(rss, M_CXGBE);
5208 		if_printf(ifp, "rss_config failed: %d\n", rc);
5209 		goto done;
5210 	}
5211 
5212 #ifdef RSS
5213 	hashen = hashconfig_to_hashen(hashconfig);
5214 
5215 	/*
5216 	 * We may have had to enable some hashes even though the global config
5217 	 * wants them disabled.  This is a potential problem that must be
5218 	 * reported to the user.
5219 	 */
5220 	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
5221 
5222 	/*
5223 	 * If we consider only the supported hash types, then the enabled hashes
5224 	 * are a superset of the requested hashes.  In other words, there cannot
5225 	 * be any supported hash that was requested but not enabled, but there
5226 	 * can be hashes that were not requested but had to be enabled.
5227 	 */
5228 	extra &= SUPPORTED_RSS_HASHTYPES;
5229 	MPASS((extra & hashconfig) == 0);
5230 
5231 	if (extra) {
5232 		if_printf(ifp,
5233 		    "global RSS config (0x%x) cannot be accommodated.\n",
5234 		    hashconfig);
5235 	}
5236 	if (extra & RSS_HASHTYPE_RSS_IPV4)
5237 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
5238 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
5239 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
5240 	if (extra & RSS_HASHTYPE_RSS_IPV6)
5241 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
5242 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
5243 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
5244 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
5245 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
5246 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
5247 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
5248 #else
5249 	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
5250 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
5251 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5252 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
5253 #endif
5254 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
5255 	if (rc != 0) {
5256 		free(rss, M_CXGBE);
5257 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
5258 		goto done;
5259 	}
5260 
5261 	vi->rss = rss;
5262 	vi->flags |= VI_INIT_DONE;
5263 done:
5264 	if (rc != 0)
5265 		vi_full_uninit(vi);
5266 
5267 	return (rc);
5268 }
5269 
5270 /*
5271  * Idempotent.
5272  */
5273 int
5274 vi_full_uninit(struct vi_info *vi)
5275 {
5276 	struct port_info *pi = vi->pi;
5277 	struct adapter *sc = pi->adapter;
5278 	int i;
5279 	struct sge_rxq *rxq;
5280 	struct sge_txq *txq;
5281 #ifdef TCP_OFFLOAD
5282 	struct sge_ofld_rxq *ofld_rxq;
5283 	struct sge_wrq *ofld_txq;
5284 #endif
5285 
5286 	if (vi->flags & VI_INIT_DONE) {
5287 
5288 		/* Need to quiesce queues.  */
5289 
5290 		/* XXX: Only for the first VI? */
5291 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
5292 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
5293 
5294 		for_each_txq(vi, i, txq) {
5295 			quiesce_txq(sc, txq);
5296 		}
5297 
5298 #ifdef TCP_OFFLOAD
5299 		for_each_ofld_txq(vi, i, ofld_txq) {
5300 			quiesce_wrq(sc, ofld_txq);
5301 		}
5302 #endif
5303 
5304 		for_each_rxq(vi, i, rxq) {
5305 			quiesce_iq(sc, &rxq->iq);
5306 			quiesce_fl(sc, &rxq->fl);
5307 		}
5308 
5309 #ifdef TCP_OFFLOAD
5310 		for_each_ofld_rxq(vi, i, ofld_rxq) {
5311 			quiesce_iq(sc, &ofld_rxq->iq);
5312 			quiesce_fl(sc, &ofld_rxq->fl);
5313 		}
5314 #endif
5315 		free(vi->rss, M_CXGBE);
5316 		free(vi->nm_rss, M_CXGBE);
5317 	}
5318 
5319 	t4_teardown_vi_queues(vi);
5320 	vi->flags &= ~VI_INIT_DONE;
5321 
5322 	return (0);
5323 }
5324 
5325 static void
5326 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
5327 {
5328 	struct sge_eq *eq = &txq->eq;
5329 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
5330 
5331 	(void) sc;	/* unused */
5332 
5333 #ifdef INVARIANTS
5334 	TXQ_LOCK(txq);
5335 	MPASS((eq->flags & EQ_ENABLED) == 0);
5336 	TXQ_UNLOCK(txq);
5337 #endif
5338 
5339 	/* Wait for the mp_ring to empty. */
5340 	while (!mp_ring_is_idle(txq->r)) {
5341 		mp_ring_check_drainage(txq->r, 0);
5342 		pause("rquiesce", 1);
5343 	}
5344 
5345 	/* Then wait for the hardware to finish. */
5346 	while (spg->cidx != htobe16(eq->pidx))
5347 		pause("equiesce", 1);
5348 
5349 	/* Finally, wait for the driver to reclaim all descriptors. */
5350 	while (eq->cidx != eq->pidx)
5351 		pause("dquiesce", 1);
5352 }
5353 
5354 static void
5355 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
5356 {
5357 
5358 	/* XXXTX */
5359 }
5360 
5361 static void
5362 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
5363 {
5364 	(void) sc;	/* unused */
5365 
5366 	/* Synchronize with the interrupt handler */
5367 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
5368 		pause("iqfree", 1);
5369 }
5370 
5371 static void
5372 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
5373 {
5374 	mtx_lock(&sc->sfl_lock);
5375 	FL_LOCK(fl);
5376 	fl->flags |= FL_DOOMED;
5377 	FL_UNLOCK(fl);
5378 	callout_stop(&sc->sfl_callout);
5379 	mtx_unlock(&sc->sfl_lock);
5380 
5381 	KASSERT((fl->flags & FL_STARVING) == 0,
5382 	    ("%s: still starving", __func__));
5383 }
5384 
5385 static int
5386 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
5387     driver_intr_t *handler, void *arg, char *name)
5388 {
5389 	int rc;
5390 
5391 	irq->rid = rid;
5392 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
5393 	    RF_SHAREABLE | RF_ACTIVE);
5394 	if (irq->res == NULL) {
5395 		device_printf(sc->dev,
5396 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
5397 		return (ENOMEM);
5398 	}
5399 
5400 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
5401 	    NULL, handler, arg, &irq->tag);
5402 	if (rc != 0) {
5403 		device_printf(sc->dev,
5404 		    "failed to setup interrupt for rid %d, name %s: %d\n",
5405 		    rid, name, rc);
5406 	} else if (name)
5407 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
5408 
5409 	return (rc);
5410 }
5411 
5412 static int
5413 t4_free_irq(struct adapter *sc, struct irq *irq)
5414 {
5415 	if (irq->tag)
5416 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
5417 	if (irq->res)
5418 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
5419 
5420 	bzero(irq, sizeof(*irq));
5421 
5422 	return (0);
5423 }
5424 
5425 static void
5426 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
5427 {
5428 
5429 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
5430 	t4_get_regs(sc, buf, regs->len);
5431 }
5432 
5433 #define	A_PL_INDIR_CMD	0x1f8
5434 
5435 #define	S_PL_AUTOINC	31
5436 #define	M_PL_AUTOINC	0x1U
5437 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
5438 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
5439 
5440 #define	S_PL_VFID	20
5441 #define	M_PL_VFID	0xffU
5442 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
5443 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
5444 
5445 #define	S_PL_ADDR	0
5446 #define	M_PL_ADDR	0xfffffU
5447 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
5448 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
5449 
5450 #define	A_PL_INDIR_DATA	0x1fc
5451 
5452 static uint64_t
5453 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
5454 {
5455 	u32 stats[2];
5456 
5457 	mtx_assert(&sc->reg_lock, MA_OWNED);
5458 	if (sc->flags & IS_VF) {
5459 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
5460 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
5461 	} else {
5462 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5463 		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5464 		    V_PL_ADDR(VF_MPS_REG(reg)));
5465 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
5466 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
5467 	}
5468 	return (((uint64_t)stats[1]) << 32 | stats[0]);
5469 }
5470 
5471 static void
5472 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
5473     struct fw_vi_stats_vf *stats)
5474 {
5475 
5476 #define GET_STAT(name) \
5477 	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
5478 
5479 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
5480 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
5481 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
5482 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
5483 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
5484 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
5485 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
5486 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
5487 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
5488 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
5489 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
5490 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
5491 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
5492 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
5493 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
5494 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
5495 
5496 #undef GET_STAT
5497 }
5498 
5499 static void
5500 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
5501 {
5502 	int reg;
5503 
5504 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5505 	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5506 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
5507 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
5508 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
5509 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
5510 }
5511 
5512 static void
5513 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
5514 {
5515 	struct timeval tv;
5516 	const struct timeval interval = {0, 250000};	/* 250ms */
5517 
5518 	if (!(vi->flags & VI_INIT_DONE))
5519 		return;
5520 
5521 	getmicrotime(&tv);
5522 	timevalsub(&tv, &interval);
5523 	if (timevalcmp(&tv, &vi->last_refreshed, <))
5524 		return;
5525 
5526 	mtx_lock(&sc->reg_lock);
5527 	t4_get_vi_stats(sc, vi->viid, &vi->stats);
5528 	getmicrotime(&vi->last_refreshed);
5529 	mtx_unlock(&sc->reg_lock);
5530 }
5531 
5532 static void
5533 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5534 {
5535 	u_int i, v, tnl_cong_drops, bg_map;
5536 	struct timeval tv;
5537 	const struct timeval interval = {0, 250000};	/* 250ms */
5538 
5539 	getmicrotime(&tv);
5540 	timevalsub(&tv, &interval);
5541 	if (timevalcmp(&tv, &pi->last_refreshed, <))
5542 		return;
5543 
5544 	tnl_cong_drops = 0;
5545 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5546 	bg_map = pi->mps_bg_map;
5547 	while (bg_map) {
5548 		i = ffs(bg_map) - 1;
5549 		mtx_lock(&sc->reg_lock);
5550 		t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5551 		    A_TP_MIB_TNL_CNG_DROP_0 + i);
5552 		mtx_unlock(&sc->reg_lock);
5553 		tnl_cong_drops += v;
5554 		bg_map &= ~(1 << i);
5555 	}
5556 	pi->tnl_cong_drops = tnl_cong_drops;
5557 	getmicrotime(&pi->last_refreshed);
5558 }
5559 
5560 static void
5561 cxgbe_tick(void *arg)
5562 {
5563 	struct port_info *pi = arg;
5564 	struct adapter *sc = pi->adapter;
5565 
5566 	PORT_LOCK_ASSERT_OWNED(pi);
5567 	cxgbe_refresh_stats(sc, pi);
5568 
5569 	callout_schedule(&pi->tick, hz);
5570 }
5571 
5572 void
5573 vi_tick(void *arg)
5574 {
5575 	struct vi_info *vi = arg;
5576 	struct adapter *sc = vi->pi->adapter;
5577 
5578 	vi_refresh_stats(sc, vi);
5579 
5580 	callout_schedule(&vi->tick, hz);
5581 }
5582 
5583 /*
5584  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5585  */
5586 static char *caps_decoder[] = {
5587 	"\20\001IPMI\002NCSI",				/* 0: NBM */
5588 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
5589 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
5590 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
5591 	    "\006HASHFILTER\007ETHOFLD",
5592 	"\20\001TOE",					/* 4: TOE */
5593 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
5594 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
5595 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5596 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5597 	    "\007T10DIF"
5598 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5599 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
5600 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
5601 		    "\004PO_INITIATOR\005PO_TARGET",
5602 };
5603 
5604 void
5605 t4_sysctls(struct adapter *sc)
5606 {
5607 	struct sysctl_ctx_list *ctx;
5608 	struct sysctl_oid *oid;
5609 	struct sysctl_oid_list *children, *c0;
5610 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5611 
5612 	ctx = device_get_sysctl_ctx(sc->dev);
5613 
5614 	/*
5615 	 * dev.t4nex.X.
5616 	 */
5617 	oid = device_get_sysctl_tree(sc->dev);
5618 	c0 = children = SYSCTL_CHILDREN(oid);
5619 
5620 	sc->sc_do_rxcopy = 1;
5621 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5622 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5623 
5624 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5625 	    sc->params.nports, "# of ports");
5626 
5627 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5628 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, (uintptr_t)&sc->doorbells,
5629 	    sysctl_bitfield_8b, "A", "available doorbells");
5630 
5631 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5632 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
5633 
5634 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5635 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5636 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5637 	    "interrupt holdoff timer values (us)");
5638 
5639 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5640 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5641 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5642 	    "interrupt holdoff packet counter values");
5643 
5644 	t4_sge_sysctls(sc, ctx, children);
5645 
5646 	sc->lro_timeout = 100;
5647 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5648 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5649 
5650 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5651 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
5652 
5653 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5654 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5655 
5656 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5657 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5658 
5659 	if (sc->flags & IS_VF)
5660 		return;
5661 
5662 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5663 	    NULL, chip_rev(sc), "chip hardware revision");
5664 
5665 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5666 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5667 
5668 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5669 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5670 
5671 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5672 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5673 
5674 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5675 	    CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5676 
5677 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5678 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5679 
5680 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5681 	    sc->er_version, 0, "expansion ROM version");
5682 
5683 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5684 	    sc->bs_version, 0, "bootstrap firmware version");
5685 
5686 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5687 	    NULL, sc->params.scfg_vers, "serial config version");
5688 
5689 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5690 	    NULL, sc->params.vpd_vers, "VPD version");
5691 
5692 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5693 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5694 
5695 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5696 	    sc->cfcsum, "config file checksum");
5697 
5698 #define SYSCTL_CAP(name, n, text) \
5699 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5700 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], (uintptr_t)&sc->name, \
5701 	    sysctl_bitfield_16b, "A", "available " text " capabilities")
5702 
5703 	SYSCTL_CAP(nbmcaps, 0, "NBM");
5704 	SYSCTL_CAP(linkcaps, 1, "link");
5705 	SYSCTL_CAP(switchcaps, 2, "switch");
5706 	SYSCTL_CAP(niccaps, 3, "NIC");
5707 	SYSCTL_CAP(toecaps, 4, "TCP offload");
5708 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
5709 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5710 	SYSCTL_CAP(cryptocaps, 7, "crypto");
5711 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
5712 #undef SYSCTL_CAP
5713 
5714 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5715 	    NULL, sc->tids.nftids, "number of filters");
5716 
5717 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5718 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5719 	    "chip temperature (in Celsius)");
5720 
5721 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "loadavg", CTLTYPE_STRING |
5722 	    CTLFLAG_RD, sc, 0, sysctl_loadavg, "A",
5723 	    "microprocessor load averages (debug firmwares only)");
5724 
5725 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
5726 	    &sc->params.core_vdd, 0, "core Vdd (in mV)");
5727 
5728 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "local_cpus",
5729 	    CTLTYPE_STRING | CTLFLAG_RD, sc, LOCAL_CPUS,
5730 	    sysctl_cpus, "A", "local CPUs");
5731 
5732 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "intr_cpus",
5733 	    CTLTYPE_STRING | CTLFLAG_RD, sc, INTR_CPUS,
5734 	    sysctl_cpus, "A", "preferred CPUs for interrupts");
5735 
5736 	/*
5737 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
5738 	 */
5739 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5740 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5741 	    "logs and miscellaneous information");
5742 	children = SYSCTL_CHILDREN(oid);
5743 
5744 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
5745 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5746 	    sysctl_cctrl, "A", "congestion control");
5747 
5748 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
5749 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5750 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
5751 
5752 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
5753 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
5754 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
5755 
5756 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
5757 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
5758 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
5759 
5760 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
5761 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
5762 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
5763 
5764 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
5765 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
5766 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
5767 
5768 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
5769 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
5770 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
5771 
5772 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
5773 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5774 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
5775 	    "A", "CIM logic analyzer");
5776 
5777 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
5778 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5779 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
5780 
5781 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
5782 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
5783 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
5784 
5785 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
5786 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
5787 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
5788 
5789 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
5790 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
5791 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
5792 
5793 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5794 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5795 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5796 
5797 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5798 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5799 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5800 
5801 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5802 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5803 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5804 
5805 	if (chip_id(sc) > CHELSIO_T4) {
5806 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5807 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5808 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5809 
5810 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5811 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5812 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5813 	}
5814 
5815 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5816 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5817 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5818 
5819 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5820 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5821 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
5822 
5823 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5824 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5825 	    sysctl_cpl_stats, "A", "CPL statistics");
5826 
5827 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5828 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5829 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5830 
5831 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5832 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5833 	    sysctl_devlog, "A", "firmware's device log");
5834 
5835 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5836 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5837 	    sysctl_fcoe_stats, "A", "FCoE statistics");
5838 
5839 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5840 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5841 	    sysctl_hw_sched, "A", "hardware scheduler ");
5842 
5843 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5844 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5845 	    sysctl_l2t, "A", "hardware L2 table");
5846 
5847 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "smt",
5848 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5849 	    sysctl_smt, "A", "hardware source MAC table");
5850 
5851 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5852 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5853 	    sysctl_lb_stats, "A", "loopback statistics");
5854 
5855 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5856 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5857 	    sysctl_meminfo, "A", "memory regions");
5858 
5859 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5860 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5861 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5862 	    "A", "MPS TCAM entries");
5863 
5864 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5865 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5866 	    sysctl_path_mtus, "A", "path MTUs");
5867 
5868 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5869 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5870 	    sysctl_pm_stats, "A", "PM statistics");
5871 
5872 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5873 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5874 	    sysctl_rdma_stats, "A", "RDMA statistics");
5875 
5876 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5877 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5878 	    sysctl_tcp_stats, "A", "TCP statistics");
5879 
5880 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5881 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5882 	    sysctl_tids, "A", "TID information");
5883 
5884 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5885 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5886 	    sysctl_tp_err_stats, "A", "TP error statistics");
5887 
5888 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5889 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5890 	    "TP logic analyzer event capture mask");
5891 
5892 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5893 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5894 	    sysctl_tp_la, "A", "TP logic analyzer");
5895 
5896 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5897 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5898 	    sysctl_tx_rate, "A", "Tx rate");
5899 
5900 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5901 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5902 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5903 
5904 	if (chip_id(sc) >= CHELSIO_T5) {
5905 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5906 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5907 		    sysctl_wcwr_stats, "A", "write combined work requests");
5908 	}
5909 
5910 #ifdef TCP_OFFLOAD
5911 	if (is_offload(sc)) {
5912 		int i;
5913 		char s[4];
5914 
5915 		/*
5916 		 * dev.t4nex.X.toe.
5917 		 */
5918 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5919 		    NULL, "TOE parameters");
5920 		children = SYSCTL_CHILDREN(oid);
5921 
5922 		sc->tt.cong_algorithm = -1;
5923 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
5924 		    CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
5925 		    "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
5926 		    "3 = highspeed)");
5927 
5928 		sc->tt.sndbuf = 256 * 1024;
5929 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5930 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
5931 
5932 		sc->tt.ddp = 0;
5933 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5934 		    &sc->tt.ddp, 0, "DDP allowed");
5935 
5936 		sc->tt.rx_coalesce = 1;
5937 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5938 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5939 
5940 		sc->tt.tls = 0;
5941 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tls", CTLFLAG_RW,
5942 		    &sc->tt.tls, 0, "Inline TLS allowed");
5943 
5944 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls_rx_ports",
5945 		    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tls_rx_ports,
5946 		    "I", "TCP ports that use inline TLS+TOE RX");
5947 
5948 		sc->tt.tx_align = 1;
5949 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5950 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5951 
5952 		sc->tt.tx_zcopy = 0;
5953 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
5954 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
5955 		    "Enable zero-copy aio_write(2)");
5956 
5957 		sc->tt.cop_managed_offloading = !!t4_cop_managed_offloading;
5958 		SYSCTL_ADD_INT(ctx, children, OID_AUTO,
5959 		    "cop_managed_offloading", CTLFLAG_RW,
5960 		    &sc->tt.cop_managed_offloading, 0,
5961 		    "COP (Connection Offload Policy) controls all TOE offload");
5962 
5963 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
5964 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
5965 		    "TP timer tick (us)");
5966 
5967 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
5968 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
5969 		    "TCP timestamp tick (us)");
5970 
5971 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5972 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5973 		    "DACK tick (us)");
5974 
5975 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5976 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5977 		    "IU", "DACK timer (us)");
5978 
5979 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5980 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5981 		    sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
5982 
5983 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5984 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5985 		    sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
5986 
5987 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5988 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5989 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
5990 
5991 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5992 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5993 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
5994 
5995 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5996 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5997 		    sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
5998 
5999 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
6000 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
6001 		    sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
6002 
6003 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
6004 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
6005 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
6006 
6007 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
6008 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
6009 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
6010 
6011 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
6012 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
6013 		    sysctl_tp_shift_cnt, "IU",
6014 		    "Number of SYN retransmissions before abort");
6015 
6016 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
6017 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
6018 		    sysctl_tp_shift_cnt, "IU",
6019 		    "Number of retransmissions before abort");
6020 
6021 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
6022 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
6023 		    sysctl_tp_shift_cnt, "IU",
6024 		    "Number of keepalive probes before abort");
6025 
6026 		oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
6027 		    CTLFLAG_RD, NULL, "TOE retransmit backoffs");
6028 		children = SYSCTL_CHILDREN(oid);
6029 		for (i = 0; i < 16; i++) {
6030 			snprintf(s, sizeof(s), "%u", i);
6031 			SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
6032 			    CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
6033 			    "IU", "TOE retransmit backoff");
6034 		}
6035 	}
6036 #endif
6037 }
6038 
6039 void
6040 vi_sysctls(struct vi_info *vi)
6041 {
6042 	struct sysctl_ctx_list *ctx;
6043 	struct sysctl_oid *oid;
6044 	struct sysctl_oid_list *children;
6045 
6046 	ctx = device_get_sysctl_ctx(vi->dev);
6047 
6048 	/*
6049 	 * dev.v?(cxgbe|cxl).X.
6050 	 */
6051 	oid = device_get_sysctl_tree(vi->dev);
6052 	children = SYSCTL_CHILDREN(oid);
6053 
6054 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
6055 	    vi->viid, "VI identifer");
6056 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
6057 	    &vi->nrxq, 0, "# of rx queues");
6058 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
6059 	    &vi->ntxq, 0, "# of tx queues");
6060 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
6061 	    &vi->first_rxq, 0, "index of first rx queue");
6062 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
6063 	    &vi->first_txq, 0, "index of first tx queue");
6064 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
6065 	    vi->rss_size, "size of RSS indirection table");
6066 
6067 	if (IS_MAIN_VI(vi)) {
6068 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
6069 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
6070 		    "Reserve queue 0 for non-flowid packets");
6071 	}
6072 
6073 #ifdef TCP_OFFLOAD
6074 	if (vi->nofldrxq != 0) {
6075 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
6076 		    &vi->nofldrxq, 0,
6077 		    "# of rx queues for offloaded TCP connections");
6078 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
6079 		    &vi->nofldtxq, 0,
6080 		    "# of tx queues for offloaded TCP connections");
6081 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
6082 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
6083 		    "index of first TOE rx queue");
6084 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
6085 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
6086 		    "index of first TOE tx queue");
6087 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
6088 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
6089 		    sysctl_holdoff_tmr_idx_ofld, "I",
6090 		    "holdoff timer index for TOE queues");
6091 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
6092 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
6093 		    sysctl_holdoff_pktc_idx_ofld, "I",
6094 		    "holdoff packet counter index for TOE queues");
6095 	}
6096 #endif
6097 #ifdef DEV_NETMAP
6098 	if (vi->nnmrxq != 0) {
6099 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
6100 		    &vi->nnmrxq, 0, "# of netmap rx queues");
6101 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
6102 		    &vi->nnmtxq, 0, "# of netmap tx queues");
6103 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
6104 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
6105 		    "index of first netmap rx queue");
6106 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
6107 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
6108 		    "index of first netmap tx queue");
6109 	}
6110 #endif
6111 
6112 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
6113 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
6114 	    "holdoff timer index");
6115 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
6116 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
6117 	    "holdoff packet counter index");
6118 
6119 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
6120 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
6121 	    "rx queue size");
6122 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
6123 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
6124 	    "tx queue size");
6125 }
6126 
6127 static void
6128 cxgbe_sysctls(struct port_info *pi)
6129 {
6130 	struct sysctl_ctx_list *ctx;
6131 	struct sysctl_oid *oid;
6132 	struct sysctl_oid_list *children, *children2;
6133 	struct adapter *sc = pi->adapter;
6134 	int i;
6135 	char name[16];
6136 	static char *tc_flags = {"\20\1USER\2SYNC\3ASYNC\4ERR"};
6137 
6138 	ctx = device_get_sysctl_ctx(pi->dev);
6139 
6140 	/*
6141 	 * dev.cxgbe.X.
6142 	 */
6143 	oid = device_get_sysctl_tree(pi->dev);
6144 	children = SYSCTL_CHILDREN(oid);
6145 
6146 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
6147 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
6148 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
6149 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
6150 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
6151 		    "PHY temperature (in Celsius)");
6152 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
6153 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
6154 		    "PHY firmware version");
6155 	}
6156 
6157 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
6158 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
6159 	    "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
6160 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
6161 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
6162 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
6163 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
6164 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
6165 	    "autonegotiation (-1 = not supported)");
6166 
6167 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
6168 	    port_top_speed(pi), "max speed (in Gbps)");
6169 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
6170 	    pi->mps_bg_map, "MPS buffer group map");
6171 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
6172 	    NULL, pi->rx_e_chan_map, "TP rx e-channel map");
6173 
6174 	if (sc->flags & IS_VF)
6175 		return;
6176 
6177 	/*
6178 	 * dev.(cxgbe|cxl).X.tc.
6179 	 */
6180 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
6181 	    "Tx scheduler traffic classes (cl_rl)");
6182 	children2 = SYSCTL_CHILDREN(oid);
6183 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "pktsize",
6184 	    CTLFLAG_RW, &pi->sched_params->pktsize, 0,
6185 	    "pktsize for per-flow cl-rl (0 means up to the driver )");
6186 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "burstsize",
6187 	    CTLFLAG_RW, &pi->sched_params->burstsize, 0,
6188 	    "burstsize for per-flow cl-rl (0 means up to the driver)");
6189 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
6190 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
6191 
6192 		snprintf(name, sizeof(name), "%d", i);
6193 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
6194 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
6195 		    "traffic class"));
6196 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "flags",
6197 		    CTLTYPE_STRING | CTLFLAG_RD, tc_flags, (uintptr_t)&tc->flags,
6198 		    sysctl_bitfield_8b, "A", "flags");
6199 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
6200 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
6201 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
6202 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
6203 		    sysctl_tc_params, "A", "traffic class parameters");
6204 	}
6205 
6206 	/*
6207 	 * dev.cxgbe.X.stats.
6208 	 */
6209 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
6210 	    NULL, "port statistics");
6211 	children = SYSCTL_CHILDREN(oid);
6212 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
6213 	    &pi->tx_parse_error, 0,
6214 	    "# of tx packets with invalid length or # of segments");
6215 
6216 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
6217 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
6218 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
6219 	    sysctl_handle_t4_reg64, "QU", desc)
6220 
6221 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
6222 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
6223 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
6224 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
6225 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
6226 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
6227 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
6228 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
6229 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
6230 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
6231 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
6232 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
6233 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
6234 	    "# of tx frames in this range",
6235 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
6236 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
6237 	    "# of tx frames in this range",
6238 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
6239 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
6240 	    "# of tx frames in this range",
6241 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
6242 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
6243 	    "# of tx frames in this range",
6244 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
6245 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
6246 	    "# of tx frames in this range",
6247 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
6248 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
6249 	    "# of tx frames in this range",
6250 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
6251 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
6252 	    "# of tx frames in this range",
6253 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
6254 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
6255 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
6256 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
6257 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
6258 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
6259 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
6260 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
6261 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
6262 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
6263 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
6264 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
6265 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
6266 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
6267 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
6268 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
6269 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
6270 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
6271 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
6272 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
6273 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
6274 
6275 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
6276 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
6277 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
6278 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
6279 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
6280 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
6281 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
6282 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
6283 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
6284 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
6285 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
6286 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
6287 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
6288 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
6289 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
6290 	    "# of frames received with bad FCS",
6291 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
6292 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
6293 	    "# of frames received with length error",
6294 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
6295 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
6296 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
6297 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
6298 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
6299 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
6300 	    "# of rx frames in this range",
6301 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
6302 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
6303 	    "# of rx frames in this range",
6304 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
6305 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
6306 	    "# of rx frames in this range",
6307 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
6308 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
6309 	    "# of rx frames in this range",
6310 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
6311 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
6312 	    "# of rx frames in this range",
6313 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
6314 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
6315 	    "# of rx frames in this range",
6316 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
6317 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
6318 	    "# of rx frames in this range",
6319 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
6320 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
6321 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
6322 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
6323 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
6324 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
6325 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
6326 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
6327 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
6328 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
6329 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
6330 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
6331 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
6332 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
6333 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
6334 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
6335 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
6336 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
6337 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
6338 
6339 #undef SYSCTL_ADD_T4_REG64
6340 
6341 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
6342 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
6343 	    &pi->stats.name, desc)
6344 
6345 	/* We get these from port_stats and they may be stale by up to 1s */
6346 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
6347 	    "# drops due to buffer-group 0 overflows");
6348 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
6349 	    "# drops due to buffer-group 1 overflows");
6350 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
6351 	    "# drops due to buffer-group 2 overflows");
6352 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
6353 	    "# drops due to buffer-group 3 overflows");
6354 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
6355 	    "# of buffer-group 0 truncated packets");
6356 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
6357 	    "# of buffer-group 1 truncated packets");
6358 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
6359 	    "# of buffer-group 2 truncated packets");
6360 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
6361 	    "# of buffer-group 3 truncated packets");
6362 
6363 #undef SYSCTL_ADD_T4_PORTSTAT
6364 
6365 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_records",
6366 	    CTLFLAG_RD, &pi->tx_tls_records,
6367 	    "# of TLS records transmitted");
6368 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_octets",
6369 	    CTLFLAG_RD, &pi->tx_tls_octets,
6370 	    "# of payload octets in transmitted TLS records");
6371 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_records",
6372 	    CTLFLAG_RD, &pi->rx_tls_records,
6373 	    "# of TLS records received");
6374 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_octets",
6375 	    CTLFLAG_RD, &pi->rx_tls_octets,
6376 	    "# of payload octets in received TLS records");
6377 }
6378 
6379 static int
6380 sysctl_int_array(SYSCTL_HANDLER_ARGS)
6381 {
6382 	int rc, *i, space = 0;
6383 	struct sbuf sb;
6384 
6385 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
6386 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
6387 		if (space)
6388 			sbuf_printf(&sb, " ");
6389 		sbuf_printf(&sb, "%d", *i);
6390 		space = 1;
6391 	}
6392 	rc = sbuf_finish(&sb);
6393 	sbuf_delete(&sb);
6394 	return (rc);
6395 }
6396 
6397 static int
6398 sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS)
6399 {
6400 	int rc;
6401 	struct sbuf *sb;
6402 
6403 	rc = sysctl_wire_old_buffer(req, 0);
6404 	if (rc != 0)
6405 		return(rc);
6406 
6407 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6408 	if (sb == NULL)
6409 		return (ENOMEM);
6410 
6411 	sbuf_printf(sb, "%b", *(uint8_t *)(uintptr_t)arg2, (char *)arg1);
6412 	rc = sbuf_finish(sb);
6413 	sbuf_delete(sb);
6414 
6415 	return (rc);
6416 }
6417 
6418 static int
6419 sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS)
6420 {
6421 	int rc;
6422 	struct sbuf *sb;
6423 
6424 	rc = sysctl_wire_old_buffer(req, 0);
6425 	if (rc != 0)
6426 		return(rc);
6427 
6428 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6429 	if (sb == NULL)
6430 		return (ENOMEM);
6431 
6432 	sbuf_printf(sb, "%b", *(uint16_t *)(uintptr_t)arg2, (char *)arg1);
6433 	rc = sbuf_finish(sb);
6434 	sbuf_delete(sb);
6435 
6436 	return (rc);
6437 }
6438 
6439 static int
6440 sysctl_btphy(SYSCTL_HANDLER_ARGS)
6441 {
6442 	struct port_info *pi = arg1;
6443 	int op = arg2;
6444 	struct adapter *sc = pi->adapter;
6445 	u_int v;
6446 	int rc;
6447 
6448 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
6449 	if (rc)
6450 		return (rc);
6451 	/* XXX: magic numbers */
6452 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
6453 	    &v);
6454 	end_synchronized_op(sc, 0);
6455 	if (rc)
6456 		return (rc);
6457 	if (op == 0)
6458 		v /= 256;
6459 
6460 	rc = sysctl_handle_int(oidp, &v, 0, req);
6461 	return (rc);
6462 }
6463 
6464 static int
6465 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
6466 {
6467 	struct vi_info *vi = arg1;
6468 	int rc, val;
6469 
6470 	val = vi->rsrv_noflowq;
6471 	rc = sysctl_handle_int(oidp, &val, 0, req);
6472 	if (rc != 0 || req->newptr == NULL)
6473 		return (rc);
6474 
6475 	if ((val >= 1) && (vi->ntxq > 1))
6476 		vi->rsrv_noflowq = 1;
6477 	else
6478 		vi->rsrv_noflowq = 0;
6479 
6480 	return (rc);
6481 }
6482 
6483 static int
6484 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
6485 {
6486 	struct vi_info *vi = arg1;
6487 	struct adapter *sc = vi->pi->adapter;
6488 	int idx, rc, i;
6489 	struct sge_rxq *rxq;
6490 	uint8_t v;
6491 
6492 	idx = vi->tmr_idx;
6493 
6494 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6495 	if (rc != 0 || req->newptr == NULL)
6496 		return (rc);
6497 
6498 	if (idx < 0 || idx >= SGE_NTIMERS)
6499 		return (EINVAL);
6500 
6501 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6502 	    "t4tmr");
6503 	if (rc)
6504 		return (rc);
6505 
6506 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
6507 	for_each_rxq(vi, i, rxq) {
6508 #ifdef atomic_store_rel_8
6509 		atomic_store_rel_8(&rxq->iq.intr_params, v);
6510 #else
6511 		rxq->iq.intr_params = v;
6512 #endif
6513 	}
6514 	vi->tmr_idx = idx;
6515 
6516 	end_synchronized_op(sc, LOCK_HELD);
6517 	return (0);
6518 }
6519 
6520 static int
6521 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
6522 {
6523 	struct vi_info *vi = arg1;
6524 	struct adapter *sc = vi->pi->adapter;
6525 	int idx, rc;
6526 
6527 	idx = vi->pktc_idx;
6528 
6529 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6530 	if (rc != 0 || req->newptr == NULL)
6531 		return (rc);
6532 
6533 	if (idx < -1 || idx >= SGE_NCOUNTERS)
6534 		return (EINVAL);
6535 
6536 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6537 	    "t4pktc");
6538 	if (rc)
6539 		return (rc);
6540 
6541 	if (vi->flags & VI_INIT_DONE)
6542 		rc = EBUSY; /* cannot be changed once the queues are created */
6543 	else
6544 		vi->pktc_idx = idx;
6545 
6546 	end_synchronized_op(sc, LOCK_HELD);
6547 	return (rc);
6548 }
6549 
6550 static int
6551 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
6552 {
6553 	struct vi_info *vi = arg1;
6554 	struct adapter *sc = vi->pi->adapter;
6555 	int qsize, rc;
6556 
6557 	qsize = vi->qsize_rxq;
6558 
6559 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6560 	if (rc != 0 || req->newptr == NULL)
6561 		return (rc);
6562 
6563 	if (qsize < 128 || (qsize & 7))
6564 		return (EINVAL);
6565 
6566 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6567 	    "t4rxqs");
6568 	if (rc)
6569 		return (rc);
6570 
6571 	if (vi->flags & VI_INIT_DONE)
6572 		rc = EBUSY; /* cannot be changed once the queues are created */
6573 	else
6574 		vi->qsize_rxq = qsize;
6575 
6576 	end_synchronized_op(sc, LOCK_HELD);
6577 	return (rc);
6578 }
6579 
6580 static int
6581 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
6582 {
6583 	struct vi_info *vi = arg1;
6584 	struct adapter *sc = vi->pi->adapter;
6585 	int qsize, rc;
6586 
6587 	qsize = vi->qsize_txq;
6588 
6589 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6590 	if (rc != 0 || req->newptr == NULL)
6591 		return (rc);
6592 
6593 	if (qsize < 128 || qsize > 65536)
6594 		return (EINVAL);
6595 
6596 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6597 	    "t4txqs");
6598 	if (rc)
6599 		return (rc);
6600 
6601 	if (vi->flags & VI_INIT_DONE)
6602 		rc = EBUSY; /* cannot be changed once the queues are created */
6603 	else
6604 		vi->qsize_txq = qsize;
6605 
6606 	end_synchronized_op(sc, LOCK_HELD);
6607 	return (rc);
6608 }
6609 
6610 static int
6611 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
6612 {
6613 	struct port_info *pi = arg1;
6614 	struct adapter *sc = pi->adapter;
6615 	struct link_config *lc = &pi->link_cfg;
6616 	int rc;
6617 
6618 	if (req->newptr == NULL) {
6619 		struct sbuf *sb;
6620 		static char *bits = "\20\1RX\2TX\3AUTO";
6621 
6622 		rc = sysctl_wire_old_buffer(req, 0);
6623 		if (rc != 0)
6624 			return(rc);
6625 
6626 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6627 		if (sb == NULL)
6628 			return (ENOMEM);
6629 
6630 		if (lc->link_ok) {
6631 			sbuf_printf(sb, "%b", (lc->fc & (PAUSE_TX | PAUSE_RX)) |
6632 			    (lc->requested_fc & PAUSE_AUTONEG), bits);
6633 		} else {
6634 			sbuf_printf(sb, "%b", lc->requested_fc & (PAUSE_TX |
6635 			    PAUSE_RX | PAUSE_AUTONEG), bits);
6636 		}
6637 		rc = sbuf_finish(sb);
6638 		sbuf_delete(sb);
6639 	} else {
6640 		char s[2];
6641 		int n;
6642 
6643 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX |
6644 		    PAUSE_AUTONEG));
6645 		s[1] = 0;
6646 
6647 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6648 		if (rc != 0)
6649 			return(rc);
6650 
6651 		if (s[1] != 0)
6652 			return (EINVAL);
6653 		if (s[0] < '0' || s[0] > '9')
6654 			return (EINVAL);	/* not a number */
6655 		n = s[0] - '0';
6656 		if (n & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG))
6657 			return (EINVAL);	/* some other bit is set too */
6658 
6659 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6660 		    "t4PAUSE");
6661 		if (rc)
6662 			return (rc);
6663 		PORT_LOCK(pi);
6664 		lc->requested_fc = n;
6665 		fixup_link_config(pi);
6666 		if (pi->up_vis > 0)
6667 			rc = apply_link_config(pi);
6668 		set_current_media(pi);
6669 		PORT_UNLOCK(pi);
6670 		end_synchronized_op(sc, 0);
6671 	}
6672 
6673 	return (rc);
6674 }
6675 
6676 static int
6677 sysctl_fec(SYSCTL_HANDLER_ARGS)
6678 {
6679 	struct port_info *pi = arg1;
6680 	struct adapter *sc = pi->adapter;
6681 	struct link_config *lc = &pi->link_cfg;
6682 	int rc;
6683 	int8_t old;
6684 
6685 	if (req->newptr == NULL) {
6686 		struct sbuf *sb;
6687 		static char *bits = "\20\1RS\2BASE-R\3RSVD1\4RSVD2\5RSVD3\6AUTO";
6688 
6689 		rc = sysctl_wire_old_buffer(req, 0);
6690 		if (rc != 0)
6691 			return(rc);
6692 
6693 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6694 		if (sb == NULL)
6695 			return (ENOMEM);
6696 
6697 		/*
6698 		 * Display the requested_fec when the link is down -- the actual
6699 		 * FEC makes sense only when the link is up.
6700 		 */
6701 		if (lc->link_ok) {
6702 			sbuf_printf(sb, "%b", (lc->fec & M_FW_PORT_CAP32_FEC) |
6703 			    (lc->requested_fec & FEC_AUTO), bits);
6704 		} else {
6705 			sbuf_printf(sb, "%b", lc->requested_fec, bits);
6706 		}
6707 		rc = sbuf_finish(sb);
6708 		sbuf_delete(sb);
6709 	} else {
6710 		char s[3];
6711 		int n;
6712 
6713 		snprintf(s, sizeof(s), "%d",
6714 		    lc->requested_fec == FEC_AUTO ? -1 :
6715 		    lc->requested_fec & M_FW_PORT_CAP32_FEC);
6716 
6717 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6718 		if (rc != 0)
6719 			return(rc);
6720 
6721 		n = strtol(&s[0], NULL, 0);
6722 		if (n < 0 || n & FEC_AUTO)
6723 			n = FEC_AUTO;
6724 		else {
6725 			if (n & ~M_FW_PORT_CAP32_FEC)
6726 				return (EINVAL);/* some other bit is set too */
6727 			if (!powerof2(n))
6728 				return (EINVAL);/* one bit can be set at most */
6729 		}
6730 
6731 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6732 		    "t4fec");
6733 		if (rc)
6734 			return (rc);
6735 		PORT_LOCK(pi);
6736 		old = lc->requested_fec;
6737 		if (n == FEC_AUTO)
6738 			lc->requested_fec = FEC_AUTO;
6739 		else if (n == 0)
6740 			lc->requested_fec = FEC_NONE;
6741 		else {
6742 			if ((lc->supported | V_FW_PORT_CAP32_FEC(n)) !=
6743 			    lc->supported) {
6744 				rc = ENOTSUP;
6745 				goto done;
6746 			}
6747 			lc->requested_fec = n;
6748 		}
6749 		fixup_link_config(pi);
6750 		if (pi->up_vis > 0) {
6751 			rc = apply_link_config(pi);
6752 			if (rc != 0) {
6753 				lc->requested_fec = old;
6754 				if (rc == FW_EPROTO)
6755 					rc = ENOTSUP;
6756 			}
6757 		}
6758 done:
6759 		PORT_UNLOCK(pi);
6760 		end_synchronized_op(sc, 0);
6761 	}
6762 
6763 	return (rc);
6764 }
6765 
6766 static int
6767 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
6768 {
6769 	struct port_info *pi = arg1;
6770 	struct adapter *sc = pi->adapter;
6771 	struct link_config *lc = &pi->link_cfg;
6772 	int rc, val;
6773 
6774 	if (lc->supported & FW_PORT_CAP32_ANEG)
6775 		val = lc->requested_aneg == AUTONEG_DISABLE ? 0 : 1;
6776 	else
6777 		val = -1;
6778 	rc = sysctl_handle_int(oidp, &val, 0, req);
6779 	if (rc != 0 || req->newptr == NULL)
6780 		return (rc);
6781 	if (val == 0)
6782 		val = AUTONEG_DISABLE;
6783 	else if (val == 1)
6784 		val = AUTONEG_ENABLE;
6785 	else
6786 		val = AUTONEG_AUTO;
6787 
6788 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6789 	    "t4aneg");
6790 	if (rc)
6791 		return (rc);
6792 	PORT_LOCK(pi);
6793 	if (val == AUTONEG_ENABLE && !(lc->supported & FW_PORT_CAP32_ANEG)) {
6794 		rc = ENOTSUP;
6795 		goto done;
6796 	}
6797 	lc->requested_aneg = val;
6798 	fixup_link_config(pi);
6799 	if (pi->up_vis > 0)
6800 		rc = apply_link_config(pi);
6801 	set_current_media(pi);
6802 done:
6803 	PORT_UNLOCK(pi);
6804 	end_synchronized_op(sc, 0);
6805 	return (rc);
6806 }
6807 
6808 static int
6809 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
6810 {
6811 	struct adapter *sc = arg1;
6812 	int reg = arg2;
6813 	uint64_t val;
6814 
6815 	val = t4_read_reg64(sc, reg);
6816 
6817 	return (sysctl_handle_64(oidp, &val, 0, req));
6818 }
6819 
6820 static int
6821 sysctl_temperature(SYSCTL_HANDLER_ARGS)
6822 {
6823 	struct adapter *sc = arg1;
6824 	int rc, t;
6825 	uint32_t param, val;
6826 
6827 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
6828 	if (rc)
6829 		return (rc);
6830 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6831 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
6832 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
6833 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6834 	end_synchronized_op(sc, 0);
6835 	if (rc)
6836 		return (rc);
6837 
6838 	/* unknown is returned as 0 but we display -1 in that case */
6839 	t = val == 0 ? -1 : val;
6840 
6841 	rc = sysctl_handle_int(oidp, &t, 0, req);
6842 	return (rc);
6843 }
6844 
6845 static int
6846 sysctl_loadavg(SYSCTL_HANDLER_ARGS)
6847 {
6848 	struct adapter *sc = arg1;
6849 	struct sbuf *sb;
6850 	int rc;
6851 	uint32_t param, val;
6852 
6853 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4lavg");
6854 	if (rc)
6855 		return (rc);
6856 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6857 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_LOAD);
6858 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6859 	end_synchronized_op(sc, 0);
6860 	if (rc)
6861 		return (rc);
6862 
6863 	rc = sysctl_wire_old_buffer(req, 0);
6864 	if (rc != 0)
6865 		return (rc);
6866 
6867 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6868 	if (sb == NULL)
6869 		return (ENOMEM);
6870 
6871 	if (val == 0xffffffff) {
6872 		/* Only debug and custom firmwares report load averages. */
6873 		sbuf_printf(sb, "not available");
6874 	} else {
6875 		sbuf_printf(sb, "%d %d %d", val & 0xff, (val >> 8) & 0xff,
6876 		    (val >> 16) & 0xff);
6877 	}
6878 	rc = sbuf_finish(sb);
6879 	sbuf_delete(sb);
6880 
6881 	return (rc);
6882 }
6883 
6884 static int
6885 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
6886 {
6887 	struct adapter *sc = arg1;
6888 	struct sbuf *sb;
6889 	int rc, i;
6890 	uint16_t incr[NMTUS][NCCTRL_WIN];
6891 	static const char *dec_fac[] = {
6892 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
6893 		"0.9375"
6894 	};
6895 
6896 	rc = sysctl_wire_old_buffer(req, 0);
6897 	if (rc != 0)
6898 		return (rc);
6899 
6900 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6901 	if (sb == NULL)
6902 		return (ENOMEM);
6903 
6904 	t4_read_cong_tbl(sc, incr);
6905 
6906 	for (i = 0; i < NCCTRL_WIN; ++i) {
6907 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
6908 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
6909 		    incr[5][i], incr[6][i], incr[7][i]);
6910 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
6911 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
6912 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
6913 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
6914 	}
6915 
6916 	rc = sbuf_finish(sb);
6917 	sbuf_delete(sb);
6918 
6919 	return (rc);
6920 }
6921 
6922 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
6923 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
6924 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
6925 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
6926 };
6927 
6928 static int
6929 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
6930 {
6931 	struct adapter *sc = arg1;
6932 	struct sbuf *sb;
6933 	int rc, i, n, qid = arg2;
6934 	uint32_t *buf, *p;
6935 	char *qtype;
6936 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
6937 
6938 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
6939 	    ("%s: bad qid %d\n", __func__, qid));
6940 
6941 	if (qid < CIM_NUM_IBQ) {
6942 		/* inbound queue */
6943 		qtype = "IBQ";
6944 		n = 4 * CIM_IBQ_SIZE;
6945 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6946 		rc = t4_read_cim_ibq(sc, qid, buf, n);
6947 	} else {
6948 		/* outbound queue */
6949 		qtype = "OBQ";
6950 		qid -= CIM_NUM_IBQ;
6951 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
6952 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6953 		rc = t4_read_cim_obq(sc, qid, buf, n);
6954 	}
6955 
6956 	if (rc < 0) {
6957 		rc = -rc;
6958 		goto done;
6959 	}
6960 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
6961 
6962 	rc = sysctl_wire_old_buffer(req, 0);
6963 	if (rc != 0)
6964 		goto done;
6965 
6966 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6967 	if (sb == NULL) {
6968 		rc = ENOMEM;
6969 		goto done;
6970 	}
6971 
6972 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
6973 	for (i = 0, p = buf; i < n; i += 16, p += 4)
6974 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
6975 		    p[2], p[3]);
6976 
6977 	rc = sbuf_finish(sb);
6978 	sbuf_delete(sb);
6979 done:
6980 	free(buf, M_CXGBE);
6981 	return (rc);
6982 }
6983 
6984 static int
6985 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
6986 {
6987 	struct adapter *sc = arg1;
6988 	u_int cfg;
6989 	struct sbuf *sb;
6990 	uint32_t *buf, *p;
6991 	int rc;
6992 
6993 	MPASS(chip_id(sc) <= CHELSIO_T5);
6994 
6995 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6996 	if (rc != 0)
6997 		return (rc);
6998 
6999 	rc = sysctl_wire_old_buffer(req, 0);
7000 	if (rc != 0)
7001 		return (rc);
7002 
7003 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7004 	if (sb == NULL)
7005 		return (ENOMEM);
7006 
7007 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
7008 	    M_ZERO | M_WAITOK);
7009 
7010 	rc = -t4_cim_read_la(sc, buf, NULL);
7011 	if (rc != 0)
7012 		goto done;
7013 
7014 	sbuf_printf(sb, "Status   Data      PC%s",
7015 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
7016 	    "     LS0Stat  LS0Addr             LS0Data");
7017 
7018 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
7019 		if (cfg & F_UPDBGLACAPTPCONLY) {
7020 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
7021 			    p[6], p[7]);
7022 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
7023 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
7024 			    p[4] & 0xff, p[5] >> 8);
7025 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
7026 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7027 			    p[1] & 0xf, p[2] >> 4);
7028 		} else {
7029 			sbuf_printf(sb,
7030 			    "\n  %02x   %x%07x %x%07x %08x %08x "
7031 			    "%08x%08x%08x%08x",
7032 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7033 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
7034 			    p[6], p[7]);
7035 		}
7036 	}
7037 
7038 	rc = sbuf_finish(sb);
7039 	sbuf_delete(sb);
7040 done:
7041 	free(buf, M_CXGBE);
7042 	return (rc);
7043 }
7044 
7045 static int
7046 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
7047 {
7048 	struct adapter *sc = arg1;
7049 	u_int cfg;
7050 	struct sbuf *sb;
7051 	uint32_t *buf, *p;
7052 	int rc;
7053 
7054 	MPASS(chip_id(sc) > CHELSIO_T5);
7055 
7056 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
7057 	if (rc != 0)
7058 		return (rc);
7059 
7060 	rc = sysctl_wire_old_buffer(req, 0);
7061 	if (rc != 0)
7062 		return (rc);
7063 
7064 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7065 	if (sb == NULL)
7066 		return (ENOMEM);
7067 
7068 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
7069 	    M_ZERO | M_WAITOK);
7070 
7071 	rc = -t4_cim_read_la(sc, buf, NULL);
7072 	if (rc != 0)
7073 		goto done;
7074 
7075 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
7076 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
7077 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
7078 
7079 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
7080 		if (cfg & F_UPDBGLACAPTPCONLY) {
7081 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
7082 			    p[3] & 0xff, p[2], p[1], p[0]);
7083 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
7084 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
7085 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
7086 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
7087 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
7088 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
7089 			    p[6] >> 16);
7090 		} else {
7091 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
7092 			    "%08x %08x %08x %08x %08x %08x",
7093 			    (p[9] >> 16) & 0xff,
7094 			    p[9] & 0xffff, p[8] >> 16,
7095 			    p[8] & 0xffff, p[7] >> 16,
7096 			    p[7] & 0xffff, p[6] >> 16,
7097 			    p[2], p[1], p[0], p[5], p[4], p[3]);
7098 		}
7099 	}
7100 
7101 	rc = sbuf_finish(sb);
7102 	sbuf_delete(sb);
7103 done:
7104 	free(buf, M_CXGBE);
7105 	return (rc);
7106 }
7107 
7108 static int
7109 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
7110 {
7111 	struct adapter *sc = arg1;
7112 	u_int i;
7113 	struct sbuf *sb;
7114 	uint32_t *buf, *p;
7115 	int rc;
7116 
7117 	rc = sysctl_wire_old_buffer(req, 0);
7118 	if (rc != 0)
7119 		return (rc);
7120 
7121 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7122 	if (sb == NULL)
7123 		return (ENOMEM);
7124 
7125 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
7126 	    M_ZERO | M_WAITOK);
7127 
7128 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
7129 	p = buf;
7130 
7131 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7132 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
7133 		    p[1], p[0]);
7134 	}
7135 
7136 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
7137 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7138 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
7139 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
7140 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
7141 		    (p[1] >> 2) | ((p[2] & 3) << 30),
7142 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
7143 		    p[0] & 1);
7144 	}
7145 
7146 	rc = sbuf_finish(sb);
7147 	sbuf_delete(sb);
7148 	free(buf, M_CXGBE);
7149 	return (rc);
7150 }
7151 
7152 static int
7153 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
7154 {
7155 	struct adapter *sc = arg1;
7156 	u_int i;
7157 	struct sbuf *sb;
7158 	uint32_t *buf, *p;
7159 	int rc;
7160 
7161 	rc = sysctl_wire_old_buffer(req, 0);
7162 	if (rc != 0)
7163 		return (rc);
7164 
7165 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7166 	if (sb == NULL)
7167 		return (ENOMEM);
7168 
7169 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
7170 	    M_ZERO | M_WAITOK);
7171 
7172 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
7173 	p = buf;
7174 
7175 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
7176 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7177 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
7178 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
7179 		    p[4], p[3], p[2], p[1], p[0]);
7180 	}
7181 
7182 	sbuf_printf(sb, "\n\nCntl ID               Data");
7183 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7184 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
7185 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
7186 	}
7187 
7188 	rc = sbuf_finish(sb);
7189 	sbuf_delete(sb);
7190 	free(buf, M_CXGBE);
7191 	return (rc);
7192 }
7193 
7194 static int
7195 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
7196 {
7197 	struct adapter *sc = arg1;
7198 	struct sbuf *sb;
7199 	int rc, i;
7200 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7201 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7202 	uint16_t thres[CIM_NUM_IBQ];
7203 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
7204 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
7205 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
7206 
7207 	cim_num_obq = sc->chip_params->cim_num_obq;
7208 	if (is_t4(sc)) {
7209 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
7210 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
7211 	} else {
7212 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
7213 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
7214 	}
7215 	nq = CIM_NUM_IBQ + cim_num_obq;
7216 
7217 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
7218 	if (rc == 0)
7219 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
7220 	if (rc != 0)
7221 		return (rc);
7222 
7223 	t4_read_cimq_cfg(sc, base, size, thres);
7224 
7225 	rc = sysctl_wire_old_buffer(req, 0);
7226 	if (rc != 0)
7227 		return (rc);
7228 
7229 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7230 	if (sb == NULL)
7231 		return (ENOMEM);
7232 
7233 	sbuf_printf(sb,
7234 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
7235 
7236 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
7237 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
7238 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
7239 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7240 		    G_QUEREMFLITS(p[2]) * 16);
7241 	for ( ; i < nq; i++, p += 4, wr += 2)
7242 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
7243 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
7244 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7245 		    G_QUEREMFLITS(p[2]) * 16);
7246 
7247 	rc = sbuf_finish(sb);
7248 	sbuf_delete(sb);
7249 
7250 	return (rc);
7251 }
7252 
7253 static int
7254 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
7255 {
7256 	struct adapter *sc = arg1;
7257 	struct sbuf *sb;
7258 	int rc;
7259 	struct tp_cpl_stats stats;
7260 
7261 	rc = sysctl_wire_old_buffer(req, 0);
7262 	if (rc != 0)
7263 		return (rc);
7264 
7265 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7266 	if (sb == NULL)
7267 		return (ENOMEM);
7268 
7269 	mtx_lock(&sc->reg_lock);
7270 	t4_tp_get_cpl_stats(sc, &stats, 0);
7271 	mtx_unlock(&sc->reg_lock);
7272 
7273 	if (sc->chip_params->nchan > 2) {
7274 		sbuf_printf(sb, "                 channel 0  channel 1"
7275 		    "  channel 2  channel 3");
7276 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
7277 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
7278 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
7279 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
7280 	} else {
7281 		sbuf_printf(sb, "                 channel 0  channel 1");
7282 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
7283 		    stats.req[0], stats.req[1]);
7284 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
7285 		    stats.rsp[0], stats.rsp[1]);
7286 	}
7287 
7288 	rc = sbuf_finish(sb);
7289 	sbuf_delete(sb);
7290 
7291 	return (rc);
7292 }
7293 
7294 static int
7295 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
7296 {
7297 	struct adapter *sc = arg1;
7298 	struct sbuf *sb;
7299 	int rc;
7300 	struct tp_usm_stats stats;
7301 
7302 	rc = sysctl_wire_old_buffer(req, 0);
7303 	if (rc != 0)
7304 		return(rc);
7305 
7306 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7307 	if (sb == NULL)
7308 		return (ENOMEM);
7309 
7310 	t4_get_usm_stats(sc, &stats, 1);
7311 
7312 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
7313 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
7314 	sbuf_printf(sb, "Drops:  %u", stats.drops);
7315 
7316 	rc = sbuf_finish(sb);
7317 	sbuf_delete(sb);
7318 
7319 	return (rc);
7320 }
7321 
7322 static const char * const devlog_level_strings[] = {
7323 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
7324 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
7325 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
7326 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
7327 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
7328 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
7329 };
7330 
7331 static const char * const devlog_facility_strings[] = {
7332 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
7333 	[FW_DEVLOG_FACILITY_CF]		= "CF",
7334 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
7335 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
7336 	[FW_DEVLOG_FACILITY_RES]	= "RES",
7337 	[FW_DEVLOG_FACILITY_HW]		= "HW",
7338 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
7339 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
7340 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
7341 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
7342 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
7343 	[FW_DEVLOG_FACILITY_VI]		= "VI",
7344 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
7345 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
7346 	[FW_DEVLOG_FACILITY_TM]		= "TM",
7347 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
7348 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
7349 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
7350 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
7351 	[FW_DEVLOG_FACILITY_RI]		= "RI",
7352 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
7353 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
7354 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
7355 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
7356 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
7357 };
7358 
7359 static int
7360 sysctl_devlog(SYSCTL_HANDLER_ARGS)
7361 {
7362 	struct adapter *sc = arg1;
7363 	struct devlog_params *dparams = &sc->params.devlog;
7364 	struct fw_devlog_e *buf, *e;
7365 	int i, j, rc, nentries, first = 0;
7366 	struct sbuf *sb;
7367 	uint64_t ftstamp = UINT64_MAX;
7368 
7369 	if (dparams->addr == 0)
7370 		return (ENXIO);
7371 
7372 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
7373 	if (buf == NULL)
7374 		return (ENOMEM);
7375 
7376 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
7377 	if (rc != 0)
7378 		goto done;
7379 
7380 	nentries = dparams->size / sizeof(struct fw_devlog_e);
7381 	for (i = 0; i < nentries; i++) {
7382 		e = &buf[i];
7383 
7384 		if (e->timestamp == 0)
7385 			break;	/* end */
7386 
7387 		e->timestamp = be64toh(e->timestamp);
7388 		e->seqno = be32toh(e->seqno);
7389 		for (j = 0; j < 8; j++)
7390 			e->params[j] = be32toh(e->params[j]);
7391 
7392 		if (e->timestamp < ftstamp) {
7393 			ftstamp = e->timestamp;
7394 			first = i;
7395 		}
7396 	}
7397 
7398 	if (buf[first].timestamp == 0)
7399 		goto done;	/* nothing in the log */
7400 
7401 	rc = sysctl_wire_old_buffer(req, 0);
7402 	if (rc != 0)
7403 		goto done;
7404 
7405 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7406 	if (sb == NULL) {
7407 		rc = ENOMEM;
7408 		goto done;
7409 	}
7410 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
7411 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
7412 
7413 	i = first;
7414 	do {
7415 		e = &buf[i];
7416 		if (e->timestamp == 0)
7417 			break;	/* end */
7418 
7419 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
7420 		    e->seqno, e->timestamp,
7421 		    (e->level < nitems(devlog_level_strings) ?
7422 			devlog_level_strings[e->level] : "UNKNOWN"),
7423 		    (e->facility < nitems(devlog_facility_strings) ?
7424 			devlog_facility_strings[e->facility] : "UNKNOWN"));
7425 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
7426 		    e->params[2], e->params[3], e->params[4],
7427 		    e->params[5], e->params[6], e->params[7]);
7428 
7429 		if (++i == nentries)
7430 			i = 0;
7431 	} while (i != first);
7432 
7433 	rc = sbuf_finish(sb);
7434 	sbuf_delete(sb);
7435 done:
7436 	free(buf, M_CXGBE);
7437 	return (rc);
7438 }
7439 
7440 static int
7441 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
7442 {
7443 	struct adapter *sc = arg1;
7444 	struct sbuf *sb;
7445 	int rc;
7446 	struct tp_fcoe_stats stats[MAX_NCHAN];
7447 	int i, nchan = sc->chip_params->nchan;
7448 
7449 	rc = sysctl_wire_old_buffer(req, 0);
7450 	if (rc != 0)
7451 		return (rc);
7452 
7453 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7454 	if (sb == NULL)
7455 		return (ENOMEM);
7456 
7457 	for (i = 0; i < nchan; i++)
7458 		t4_get_fcoe_stats(sc, i, &stats[i], 1);
7459 
7460 	if (nchan > 2) {
7461 		sbuf_printf(sb, "                   channel 0        channel 1"
7462 		    "        channel 2        channel 3");
7463 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
7464 		    stats[0].octets_ddp, stats[1].octets_ddp,
7465 		    stats[2].octets_ddp, stats[3].octets_ddp);
7466 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
7467 		    stats[0].frames_ddp, stats[1].frames_ddp,
7468 		    stats[2].frames_ddp, stats[3].frames_ddp);
7469 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
7470 		    stats[0].frames_drop, stats[1].frames_drop,
7471 		    stats[2].frames_drop, stats[3].frames_drop);
7472 	} else {
7473 		sbuf_printf(sb, "                   channel 0        channel 1");
7474 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
7475 		    stats[0].octets_ddp, stats[1].octets_ddp);
7476 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
7477 		    stats[0].frames_ddp, stats[1].frames_ddp);
7478 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
7479 		    stats[0].frames_drop, stats[1].frames_drop);
7480 	}
7481 
7482 	rc = sbuf_finish(sb);
7483 	sbuf_delete(sb);
7484 
7485 	return (rc);
7486 }
7487 
7488 static int
7489 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
7490 {
7491 	struct adapter *sc = arg1;
7492 	struct sbuf *sb;
7493 	int rc, i;
7494 	unsigned int map, kbps, ipg, mode;
7495 	unsigned int pace_tab[NTX_SCHED];
7496 
7497 	rc = sysctl_wire_old_buffer(req, 0);
7498 	if (rc != 0)
7499 		return (rc);
7500 
7501 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7502 	if (sb == NULL)
7503 		return (ENOMEM);
7504 
7505 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
7506 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
7507 	t4_read_pace_tbl(sc, pace_tab);
7508 
7509 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
7510 	    "Class IPG (0.1 ns)   Flow IPG (us)");
7511 
7512 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
7513 		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
7514 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
7515 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
7516 		if (kbps)
7517 			sbuf_printf(sb, "%9u     ", kbps);
7518 		else
7519 			sbuf_printf(sb, " disabled     ");
7520 
7521 		if (ipg)
7522 			sbuf_printf(sb, "%13u        ", ipg);
7523 		else
7524 			sbuf_printf(sb, "     disabled        ");
7525 
7526 		if (pace_tab[i])
7527 			sbuf_printf(sb, "%10u", pace_tab[i]);
7528 		else
7529 			sbuf_printf(sb, "  disabled");
7530 	}
7531 
7532 	rc = sbuf_finish(sb);
7533 	sbuf_delete(sb);
7534 
7535 	return (rc);
7536 }
7537 
7538 static int
7539 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
7540 {
7541 	struct adapter *sc = arg1;
7542 	struct sbuf *sb;
7543 	int rc, i, j;
7544 	uint64_t *p0, *p1;
7545 	struct lb_port_stats s[2];
7546 	static const char *stat_name[] = {
7547 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
7548 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
7549 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
7550 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
7551 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
7552 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
7553 		"BG2FramesTrunc:", "BG3FramesTrunc:"
7554 	};
7555 
7556 	rc = sysctl_wire_old_buffer(req, 0);
7557 	if (rc != 0)
7558 		return (rc);
7559 
7560 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7561 	if (sb == NULL)
7562 		return (ENOMEM);
7563 
7564 	memset(s, 0, sizeof(s));
7565 
7566 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
7567 		t4_get_lb_stats(sc, i, &s[0]);
7568 		t4_get_lb_stats(sc, i + 1, &s[1]);
7569 
7570 		p0 = &s[0].octets;
7571 		p1 = &s[1].octets;
7572 		sbuf_printf(sb, "%s                       Loopback %u"
7573 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
7574 
7575 		for (j = 0; j < nitems(stat_name); j++)
7576 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
7577 				   *p0++, *p1++);
7578 	}
7579 
7580 	rc = sbuf_finish(sb);
7581 	sbuf_delete(sb);
7582 
7583 	return (rc);
7584 }
7585 
7586 static int
7587 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
7588 {
7589 	int rc = 0;
7590 	struct port_info *pi = arg1;
7591 	struct link_config *lc = &pi->link_cfg;
7592 	struct sbuf *sb;
7593 
7594 	rc = sysctl_wire_old_buffer(req, 0);
7595 	if (rc != 0)
7596 		return(rc);
7597 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
7598 	if (sb == NULL)
7599 		return (ENOMEM);
7600 
7601 	if (lc->link_ok || lc->link_down_rc == 255)
7602 		sbuf_printf(sb, "n/a");
7603 	else
7604 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
7605 
7606 	rc = sbuf_finish(sb);
7607 	sbuf_delete(sb);
7608 
7609 	return (rc);
7610 }
7611 
7612 struct mem_desc {
7613 	unsigned int base;
7614 	unsigned int limit;
7615 	unsigned int idx;
7616 };
7617 
7618 static int
7619 mem_desc_cmp(const void *a, const void *b)
7620 {
7621 	return ((const struct mem_desc *)a)->base -
7622 	       ((const struct mem_desc *)b)->base;
7623 }
7624 
7625 static void
7626 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
7627     unsigned int to)
7628 {
7629 	unsigned int size;
7630 
7631 	if (from == to)
7632 		return;
7633 
7634 	size = to - from + 1;
7635 	if (size == 0)
7636 		return;
7637 
7638 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
7639 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
7640 }
7641 
7642 static int
7643 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
7644 {
7645 	struct adapter *sc = arg1;
7646 	struct sbuf *sb;
7647 	int rc, i, n;
7648 	uint32_t lo, hi, used, alloc;
7649 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
7650 	static const char *region[] = {
7651 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
7652 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
7653 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
7654 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
7655 		"RQUDP region:", "PBL region:", "TXPBL region:",
7656 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
7657 		"On-chip queues:", "TLS keys:",
7658 	};
7659 	struct mem_desc avail[4];
7660 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
7661 	struct mem_desc *md = mem;
7662 
7663 	rc = sysctl_wire_old_buffer(req, 0);
7664 	if (rc != 0)
7665 		return (rc);
7666 
7667 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7668 	if (sb == NULL)
7669 		return (ENOMEM);
7670 
7671 	for (i = 0; i < nitems(mem); i++) {
7672 		mem[i].limit = 0;
7673 		mem[i].idx = i;
7674 	}
7675 
7676 	/* Find and sort the populated memory ranges */
7677 	i = 0;
7678 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
7679 	if (lo & F_EDRAM0_ENABLE) {
7680 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
7681 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
7682 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
7683 		avail[i].idx = 0;
7684 		i++;
7685 	}
7686 	if (lo & F_EDRAM1_ENABLE) {
7687 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
7688 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
7689 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
7690 		avail[i].idx = 1;
7691 		i++;
7692 	}
7693 	if (lo & F_EXT_MEM_ENABLE) {
7694 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
7695 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
7696 		avail[i].limit = avail[i].base +
7697 		    (G_EXT_MEM_SIZE(hi) << 20);
7698 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
7699 		i++;
7700 	}
7701 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
7702 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
7703 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
7704 		avail[i].limit = avail[i].base +
7705 		    (G_EXT_MEM1_SIZE(hi) << 20);
7706 		avail[i].idx = 4;
7707 		i++;
7708 	}
7709 	if (!i)                                    /* no memory available */
7710 		return 0;
7711 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
7712 
7713 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
7714 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
7715 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
7716 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
7717 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
7718 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
7719 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
7720 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
7721 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
7722 
7723 	/* the next few have explicit upper bounds */
7724 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
7725 	md->limit = md->base - 1 +
7726 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
7727 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
7728 	md++;
7729 
7730 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
7731 	md->limit = md->base - 1 +
7732 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
7733 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
7734 	md++;
7735 
7736 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7737 		if (chip_id(sc) <= CHELSIO_T5)
7738 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
7739 		else
7740 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
7741 		md->limit = 0;
7742 	} else {
7743 		md->base = 0;
7744 		md->idx = nitems(region);  /* hide it */
7745 	}
7746 	md++;
7747 
7748 #define ulp_region(reg) \
7749 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
7750 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
7751 
7752 	ulp_region(RX_ISCSI);
7753 	ulp_region(RX_TDDP);
7754 	ulp_region(TX_TPT);
7755 	ulp_region(RX_STAG);
7756 	ulp_region(RX_RQ);
7757 	ulp_region(RX_RQUDP);
7758 	ulp_region(RX_PBL);
7759 	ulp_region(TX_PBL);
7760 #undef ulp_region
7761 
7762 	md->base = 0;
7763 	md->idx = nitems(region);
7764 	if (!is_t4(sc)) {
7765 		uint32_t size = 0;
7766 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
7767 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
7768 
7769 		if (is_t5(sc)) {
7770 			if (sge_ctrl & F_VFIFO_ENABLE)
7771 				size = G_DBVFIFO_SIZE(fifo_size);
7772 		} else
7773 			size = G_T6_DBVFIFO_SIZE(fifo_size);
7774 
7775 		if (size) {
7776 			md->base = G_BASEADDR(t4_read_reg(sc,
7777 			    A_SGE_DBVFIFO_BADDR));
7778 			md->limit = md->base + (size << 2) - 1;
7779 		}
7780 	}
7781 	md++;
7782 
7783 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
7784 	md->limit = 0;
7785 	md++;
7786 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
7787 	md->limit = 0;
7788 	md++;
7789 
7790 	md->base = sc->vres.ocq.start;
7791 	if (sc->vres.ocq.size)
7792 		md->limit = md->base + sc->vres.ocq.size - 1;
7793 	else
7794 		md->idx = nitems(region);  /* hide it */
7795 	md++;
7796 
7797 	md->base = sc->vres.key.start;
7798 	if (sc->vres.key.size)
7799 		md->limit = md->base + sc->vres.key.size - 1;
7800 	else
7801 		md->idx = nitems(region);  /* hide it */
7802 	md++;
7803 
7804 	/* add any address-space holes, there can be up to 3 */
7805 	for (n = 0; n < i - 1; n++)
7806 		if (avail[n].limit < avail[n + 1].base)
7807 			(md++)->base = avail[n].limit;
7808 	if (avail[n].limit)
7809 		(md++)->base = avail[n].limit;
7810 
7811 	n = md - mem;
7812 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
7813 
7814 	for (lo = 0; lo < i; lo++)
7815 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
7816 				avail[lo].limit - 1);
7817 
7818 	sbuf_printf(sb, "\n");
7819 	for (i = 0; i < n; i++) {
7820 		if (mem[i].idx >= nitems(region))
7821 			continue;                        /* skip holes */
7822 		if (!mem[i].limit)
7823 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
7824 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
7825 				mem[i].limit);
7826 	}
7827 
7828 	sbuf_printf(sb, "\n");
7829 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
7830 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
7831 	mem_region_show(sb, "uP RAM:", lo, hi);
7832 
7833 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
7834 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
7835 	mem_region_show(sb, "uP Extmem2:", lo, hi);
7836 
7837 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
7838 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
7839 		   G_PMRXMAXPAGE(lo),
7840 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
7841 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
7842 
7843 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
7844 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
7845 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
7846 		   G_PMTXMAXPAGE(lo),
7847 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
7848 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
7849 	sbuf_printf(sb, "%u p-structs\n",
7850 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
7851 
7852 	for (i = 0; i < 4; i++) {
7853 		if (chip_id(sc) > CHELSIO_T5)
7854 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
7855 		else
7856 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
7857 		if (is_t5(sc)) {
7858 			used = G_T5_USED(lo);
7859 			alloc = G_T5_ALLOC(lo);
7860 		} else {
7861 			used = G_USED(lo);
7862 			alloc = G_ALLOC(lo);
7863 		}
7864 		/* For T6 these are MAC buffer groups */
7865 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
7866 		    i, used, alloc);
7867 	}
7868 	for (i = 0; i < sc->chip_params->nchan; i++) {
7869 		if (chip_id(sc) > CHELSIO_T5)
7870 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
7871 		else
7872 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
7873 		if (is_t5(sc)) {
7874 			used = G_T5_USED(lo);
7875 			alloc = G_T5_ALLOC(lo);
7876 		} else {
7877 			used = G_USED(lo);
7878 			alloc = G_ALLOC(lo);
7879 		}
7880 		/* For T6 these are MAC buffer groups */
7881 		sbuf_printf(sb,
7882 		    "\nLoopback %d using %u pages out of %u allocated",
7883 		    i, used, alloc);
7884 	}
7885 
7886 	rc = sbuf_finish(sb);
7887 	sbuf_delete(sb);
7888 
7889 	return (rc);
7890 }
7891 
7892 static inline void
7893 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
7894 {
7895 	*mask = x | y;
7896 	y = htobe64(y);
7897 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
7898 }
7899 
7900 static int
7901 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
7902 {
7903 	struct adapter *sc = arg1;
7904 	struct sbuf *sb;
7905 	int rc, i;
7906 
7907 	MPASS(chip_id(sc) <= CHELSIO_T5);
7908 
7909 	rc = sysctl_wire_old_buffer(req, 0);
7910 	if (rc != 0)
7911 		return (rc);
7912 
7913 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7914 	if (sb == NULL)
7915 		return (ENOMEM);
7916 
7917 	sbuf_printf(sb,
7918 	    "Idx  Ethernet address     Mask     Vld Ports PF"
7919 	    "  VF              Replication             P0 P1 P2 P3  ML");
7920 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7921 		uint64_t tcamx, tcamy, mask;
7922 		uint32_t cls_lo, cls_hi;
7923 		uint8_t addr[ETHER_ADDR_LEN];
7924 
7925 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
7926 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
7927 		if (tcamx & tcamy)
7928 			continue;
7929 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7930 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7931 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7932 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
7933 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
7934 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
7935 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
7936 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
7937 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
7938 
7939 		if (cls_lo & F_REPLICATE) {
7940 			struct fw_ldst_cmd ldst_cmd;
7941 
7942 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7943 			ldst_cmd.op_to_addrspace =
7944 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7945 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7946 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7947 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7948 			ldst_cmd.u.mps.rplc.fid_idx =
7949 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7950 				V_FW_LDST_CMD_IDX(i));
7951 
7952 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7953 			    "t4mps");
7954 			if (rc)
7955 				break;
7956 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7957 			    sizeof(ldst_cmd), &ldst_cmd);
7958 			end_synchronized_op(sc, 0);
7959 
7960 			if (rc != 0) {
7961 				sbuf_printf(sb, "%36d", rc);
7962 				rc = 0;
7963 			} else {
7964 				sbuf_printf(sb, " %08x %08x %08x %08x",
7965 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7966 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7967 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7968 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7969 			}
7970 		} else
7971 			sbuf_printf(sb, "%36s", "");
7972 
7973 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
7974 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
7975 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
7976 	}
7977 
7978 	if (rc)
7979 		(void) sbuf_finish(sb);
7980 	else
7981 		rc = sbuf_finish(sb);
7982 	sbuf_delete(sb);
7983 
7984 	return (rc);
7985 }
7986 
7987 static int
7988 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
7989 {
7990 	struct adapter *sc = arg1;
7991 	struct sbuf *sb;
7992 	int rc, i;
7993 
7994 	MPASS(chip_id(sc) > CHELSIO_T5);
7995 
7996 	rc = sysctl_wire_old_buffer(req, 0);
7997 	if (rc != 0)
7998 		return (rc);
7999 
8000 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8001 	if (sb == NULL)
8002 		return (ENOMEM);
8003 
8004 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
8005 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
8006 	    "                           Replication"
8007 	    "                                    P0 P1 P2 P3  ML\n");
8008 
8009 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
8010 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
8011 		uint16_t ivlan;
8012 		uint64_t tcamx, tcamy, val, mask;
8013 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
8014 		uint8_t addr[ETHER_ADDR_LEN];
8015 
8016 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
8017 		if (i < 256)
8018 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
8019 		else
8020 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
8021 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8022 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8023 		tcamy = G_DMACH(val) << 32;
8024 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8025 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8026 		lookup_type = G_DATALKPTYPE(data2);
8027 		port_num = G_DATAPORTNUM(data2);
8028 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8029 			/* Inner header VNI */
8030 			vniy = ((data2 & F_DATAVIDH2) << 23) |
8031 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8032 			dip_hit = data2 & F_DATADIPHIT;
8033 			vlan_vld = 0;
8034 		} else {
8035 			vniy = 0;
8036 			dip_hit = 0;
8037 			vlan_vld = data2 & F_DATAVIDH2;
8038 			ivlan = G_VIDL(val);
8039 		}
8040 
8041 		ctl |= V_CTLXYBITSEL(1);
8042 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8043 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8044 		tcamx = G_DMACH(val) << 32;
8045 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8046 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8047 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8048 			/* Inner header VNI mask */
8049 			vnix = ((data2 & F_DATAVIDH2) << 23) |
8050 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8051 		} else
8052 			vnix = 0;
8053 
8054 		if (tcamx & tcamy)
8055 			continue;
8056 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
8057 
8058 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
8059 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
8060 
8061 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8062 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8063 			    "%012jx %06x %06x    -    -   %3c"
8064 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
8065 			    addr[1], addr[2], addr[3], addr[4], addr[5],
8066 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
8067 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8068 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8069 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8070 		} else {
8071 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8072 			    "%012jx    -       -   ", i, addr[0], addr[1],
8073 			    addr[2], addr[3], addr[4], addr[5],
8074 			    (uintmax_t)mask);
8075 
8076 			if (vlan_vld)
8077 				sbuf_printf(sb, "%4u   Y     ", ivlan);
8078 			else
8079 				sbuf_printf(sb, "  -    N     ");
8080 
8081 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
8082 			    lookup_type ? 'I' : 'O', port_num,
8083 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8084 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8085 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8086 		}
8087 
8088 
8089 		if (cls_lo & F_T6_REPLICATE) {
8090 			struct fw_ldst_cmd ldst_cmd;
8091 
8092 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
8093 			ldst_cmd.op_to_addrspace =
8094 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
8095 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
8096 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
8097 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
8098 			ldst_cmd.u.mps.rplc.fid_idx =
8099 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
8100 				V_FW_LDST_CMD_IDX(i));
8101 
8102 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8103 			    "t6mps");
8104 			if (rc)
8105 				break;
8106 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
8107 			    sizeof(ldst_cmd), &ldst_cmd);
8108 			end_synchronized_op(sc, 0);
8109 
8110 			if (rc != 0) {
8111 				sbuf_printf(sb, "%72d", rc);
8112 				rc = 0;
8113 			} else {
8114 				sbuf_printf(sb, " %08x %08x %08x %08x"
8115 				    " %08x %08x %08x %08x",
8116 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
8117 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
8118 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
8119 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
8120 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
8121 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
8122 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
8123 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
8124 			}
8125 		} else
8126 			sbuf_printf(sb, "%72s", "");
8127 
8128 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
8129 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
8130 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
8131 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
8132 	}
8133 
8134 	if (rc)
8135 		(void) sbuf_finish(sb);
8136 	else
8137 		rc = sbuf_finish(sb);
8138 	sbuf_delete(sb);
8139 
8140 	return (rc);
8141 }
8142 
8143 static int
8144 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
8145 {
8146 	struct adapter *sc = arg1;
8147 	struct sbuf *sb;
8148 	int rc;
8149 	uint16_t mtus[NMTUS];
8150 
8151 	rc = sysctl_wire_old_buffer(req, 0);
8152 	if (rc != 0)
8153 		return (rc);
8154 
8155 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8156 	if (sb == NULL)
8157 		return (ENOMEM);
8158 
8159 	t4_read_mtu_tbl(sc, mtus, NULL);
8160 
8161 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
8162 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
8163 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
8164 	    mtus[14], mtus[15]);
8165 
8166 	rc = sbuf_finish(sb);
8167 	sbuf_delete(sb);
8168 
8169 	return (rc);
8170 }
8171 
8172 static int
8173 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
8174 {
8175 	struct adapter *sc = arg1;
8176 	struct sbuf *sb;
8177 	int rc, i;
8178 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
8179 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
8180 	static const char *tx_stats[MAX_PM_NSTATS] = {
8181 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
8182 		"Tx FIFO wait", NULL, "Tx latency"
8183 	};
8184 	static const char *rx_stats[MAX_PM_NSTATS] = {
8185 		"Read:", "Write bypass:", "Write mem:", "Flush:",
8186 		"Rx FIFO wait", NULL, "Rx latency"
8187 	};
8188 
8189 	rc = sysctl_wire_old_buffer(req, 0);
8190 	if (rc != 0)
8191 		return (rc);
8192 
8193 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8194 	if (sb == NULL)
8195 		return (ENOMEM);
8196 
8197 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
8198 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
8199 
8200 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
8201 	for (i = 0; i < 4; i++) {
8202 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8203 		    tx_cyc[i]);
8204 	}
8205 
8206 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
8207 	for (i = 0; i < 4; i++) {
8208 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8209 		    rx_cyc[i]);
8210 	}
8211 
8212 	if (chip_id(sc) > CHELSIO_T5) {
8213 		sbuf_printf(sb,
8214 		    "\n              Total wait      Total occupancy");
8215 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8216 		    tx_cyc[i]);
8217 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8218 		    rx_cyc[i]);
8219 
8220 		i += 2;
8221 		MPASS(i < nitems(tx_stats));
8222 
8223 		sbuf_printf(sb,
8224 		    "\n                   Reads           Total wait");
8225 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8226 		    tx_cyc[i]);
8227 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8228 		    rx_cyc[i]);
8229 	}
8230 
8231 	rc = sbuf_finish(sb);
8232 	sbuf_delete(sb);
8233 
8234 	return (rc);
8235 }
8236 
8237 static int
8238 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
8239 {
8240 	struct adapter *sc = arg1;
8241 	struct sbuf *sb;
8242 	int rc;
8243 	struct tp_rdma_stats stats;
8244 
8245 	rc = sysctl_wire_old_buffer(req, 0);
8246 	if (rc != 0)
8247 		return (rc);
8248 
8249 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8250 	if (sb == NULL)
8251 		return (ENOMEM);
8252 
8253 	mtx_lock(&sc->reg_lock);
8254 	t4_tp_get_rdma_stats(sc, &stats, 0);
8255 	mtx_unlock(&sc->reg_lock);
8256 
8257 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
8258 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
8259 
8260 	rc = sbuf_finish(sb);
8261 	sbuf_delete(sb);
8262 
8263 	return (rc);
8264 }
8265 
8266 static int
8267 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
8268 {
8269 	struct adapter *sc = arg1;
8270 	struct sbuf *sb;
8271 	int rc;
8272 	struct tp_tcp_stats v4, v6;
8273 
8274 	rc = sysctl_wire_old_buffer(req, 0);
8275 	if (rc != 0)
8276 		return (rc);
8277 
8278 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8279 	if (sb == NULL)
8280 		return (ENOMEM);
8281 
8282 	mtx_lock(&sc->reg_lock);
8283 	t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
8284 	mtx_unlock(&sc->reg_lock);
8285 
8286 	sbuf_printf(sb,
8287 	    "                                IP                 IPv6\n");
8288 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
8289 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
8290 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
8291 	    v4.tcp_in_segs, v6.tcp_in_segs);
8292 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
8293 	    v4.tcp_out_segs, v6.tcp_out_segs);
8294 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
8295 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
8296 
8297 	rc = sbuf_finish(sb);
8298 	sbuf_delete(sb);
8299 
8300 	return (rc);
8301 }
8302 
8303 static int
8304 sysctl_tids(SYSCTL_HANDLER_ARGS)
8305 {
8306 	struct adapter *sc = arg1;
8307 	struct sbuf *sb;
8308 	int rc;
8309 	struct tid_info *t = &sc->tids;
8310 
8311 	rc = sysctl_wire_old_buffer(req, 0);
8312 	if (rc != 0)
8313 		return (rc);
8314 
8315 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8316 	if (sb == NULL)
8317 		return (ENOMEM);
8318 
8319 	if (t->natids) {
8320 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
8321 		    t->atids_in_use);
8322 	}
8323 
8324 	if (t->nhpftids) {
8325 		sbuf_printf(sb, "HPFTID range: %u-%u, in use: %u\n",
8326 		    t->hpftid_base, t->hpftid_end, t->hpftids_in_use);
8327 	}
8328 
8329 	if (t->ntids) {
8330 		sbuf_printf(sb, "TID range: ");
8331 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
8332 			uint32_t b, hb;
8333 
8334 			if (chip_id(sc) <= CHELSIO_T5) {
8335 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
8336 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
8337 			} else {
8338 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
8339 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
8340 			}
8341 
8342 			if (b)
8343 				sbuf_printf(sb, "%u-%u, ", t->tid_base, b - 1);
8344 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
8345 		} else
8346 			sbuf_printf(sb, "%u-%u", t->tid_base, t->ntids - 1);
8347 		sbuf_printf(sb, ", in use: %u\n",
8348 		    atomic_load_acq_int(&t->tids_in_use));
8349 	}
8350 
8351 	if (t->nstids) {
8352 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
8353 		    t->stid_base + t->nstids - 1, t->stids_in_use);
8354 	}
8355 
8356 	if (t->nftids) {
8357 		sbuf_printf(sb, "FTID range: %u-%u, in use: %u\n", t->ftid_base,
8358 		    t->ftid_end, t->ftids_in_use);
8359 	}
8360 
8361 	if (t->netids) {
8362 		sbuf_printf(sb, "ETID range: %u-%u, in use: %u\n", t->etid_base,
8363 		    t->etid_base + t->netids - 1, t->etids_in_use);
8364 	}
8365 
8366 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
8367 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
8368 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
8369 
8370 	rc = sbuf_finish(sb);
8371 	sbuf_delete(sb);
8372 
8373 	return (rc);
8374 }
8375 
8376 static int
8377 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
8378 {
8379 	struct adapter *sc = arg1;
8380 	struct sbuf *sb;
8381 	int rc;
8382 	struct tp_err_stats stats;
8383 
8384 	rc = sysctl_wire_old_buffer(req, 0);
8385 	if (rc != 0)
8386 		return (rc);
8387 
8388 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8389 	if (sb == NULL)
8390 		return (ENOMEM);
8391 
8392 	mtx_lock(&sc->reg_lock);
8393 	t4_tp_get_err_stats(sc, &stats, 0);
8394 	mtx_unlock(&sc->reg_lock);
8395 
8396 	if (sc->chip_params->nchan > 2) {
8397 		sbuf_printf(sb, "                 channel 0  channel 1"
8398 		    "  channel 2  channel 3\n");
8399 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
8400 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
8401 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
8402 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
8403 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
8404 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
8405 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
8406 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
8407 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
8408 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
8409 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
8410 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
8411 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
8412 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
8413 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
8414 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
8415 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
8416 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
8417 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
8418 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
8419 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
8420 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
8421 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
8422 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
8423 	} else {
8424 		sbuf_printf(sb, "                 channel 0  channel 1\n");
8425 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
8426 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
8427 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
8428 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
8429 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
8430 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
8431 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
8432 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
8433 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
8434 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
8435 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
8436 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
8437 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
8438 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
8439 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
8440 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
8441 	}
8442 
8443 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
8444 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
8445 
8446 	rc = sbuf_finish(sb);
8447 	sbuf_delete(sb);
8448 
8449 	return (rc);
8450 }
8451 
8452 static int
8453 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
8454 {
8455 	struct adapter *sc = arg1;
8456 	struct tp_params *tpp = &sc->params.tp;
8457 	u_int mask;
8458 	int rc;
8459 
8460 	mask = tpp->la_mask >> 16;
8461 	rc = sysctl_handle_int(oidp, &mask, 0, req);
8462 	if (rc != 0 || req->newptr == NULL)
8463 		return (rc);
8464 	if (mask > 0xffff)
8465 		return (EINVAL);
8466 	tpp->la_mask = mask << 16;
8467 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
8468 
8469 	return (0);
8470 }
8471 
8472 struct field_desc {
8473 	const char *name;
8474 	u_int start;
8475 	u_int width;
8476 };
8477 
8478 static void
8479 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
8480 {
8481 	char buf[32];
8482 	int line_size = 0;
8483 
8484 	while (f->name) {
8485 		uint64_t mask = (1ULL << f->width) - 1;
8486 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
8487 		    ((uintmax_t)v >> f->start) & mask);
8488 
8489 		if (line_size + len >= 79) {
8490 			line_size = 8;
8491 			sbuf_printf(sb, "\n        ");
8492 		}
8493 		sbuf_printf(sb, "%s ", buf);
8494 		line_size += len + 1;
8495 		f++;
8496 	}
8497 	sbuf_printf(sb, "\n");
8498 }
8499 
8500 static const struct field_desc tp_la0[] = {
8501 	{ "RcfOpCodeOut", 60, 4 },
8502 	{ "State", 56, 4 },
8503 	{ "WcfState", 52, 4 },
8504 	{ "RcfOpcSrcOut", 50, 2 },
8505 	{ "CRxError", 49, 1 },
8506 	{ "ERxError", 48, 1 },
8507 	{ "SanityFailed", 47, 1 },
8508 	{ "SpuriousMsg", 46, 1 },
8509 	{ "FlushInputMsg", 45, 1 },
8510 	{ "FlushInputCpl", 44, 1 },
8511 	{ "RssUpBit", 43, 1 },
8512 	{ "RssFilterHit", 42, 1 },
8513 	{ "Tid", 32, 10 },
8514 	{ "InitTcb", 31, 1 },
8515 	{ "LineNumber", 24, 7 },
8516 	{ "Emsg", 23, 1 },
8517 	{ "EdataOut", 22, 1 },
8518 	{ "Cmsg", 21, 1 },
8519 	{ "CdataOut", 20, 1 },
8520 	{ "EreadPdu", 19, 1 },
8521 	{ "CreadPdu", 18, 1 },
8522 	{ "TunnelPkt", 17, 1 },
8523 	{ "RcfPeerFin", 16, 1 },
8524 	{ "RcfReasonOut", 12, 4 },
8525 	{ "TxCchannel", 10, 2 },
8526 	{ "RcfTxChannel", 8, 2 },
8527 	{ "RxEchannel", 6, 2 },
8528 	{ "RcfRxChannel", 5, 1 },
8529 	{ "RcfDataOutSrdy", 4, 1 },
8530 	{ "RxDvld", 3, 1 },
8531 	{ "RxOoDvld", 2, 1 },
8532 	{ "RxCongestion", 1, 1 },
8533 	{ "TxCongestion", 0, 1 },
8534 	{ NULL }
8535 };
8536 
8537 static const struct field_desc tp_la1[] = {
8538 	{ "CplCmdIn", 56, 8 },
8539 	{ "CplCmdOut", 48, 8 },
8540 	{ "ESynOut", 47, 1 },
8541 	{ "EAckOut", 46, 1 },
8542 	{ "EFinOut", 45, 1 },
8543 	{ "ERstOut", 44, 1 },
8544 	{ "SynIn", 43, 1 },
8545 	{ "AckIn", 42, 1 },
8546 	{ "FinIn", 41, 1 },
8547 	{ "RstIn", 40, 1 },
8548 	{ "DataIn", 39, 1 },
8549 	{ "DataInVld", 38, 1 },
8550 	{ "PadIn", 37, 1 },
8551 	{ "RxBufEmpty", 36, 1 },
8552 	{ "RxDdp", 35, 1 },
8553 	{ "RxFbCongestion", 34, 1 },
8554 	{ "TxFbCongestion", 33, 1 },
8555 	{ "TxPktSumSrdy", 32, 1 },
8556 	{ "RcfUlpType", 28, 4 },
8557 	{ "Eread", 27, 1 },
8558 	{ "Ebypass", 26, 1 },
8559 	{ "Esave", 25, 1 },
8560 	{ "Static0", 24, 1 },
8561 	{ "Cread", 23, 1 },
8562 	{ "Cbypass", 22, 1 },
8563 	{ "Csave", 21, 1 },
8564 	{ "CPktOut", 20, 1 },
8565 	{ "RxPagePoolFull", 18, 2 },
8566 	{ "RxLpbkPkt", 17, 1 },
8567 	{ "TxLpbkPkt", 16, 1 },
8568 	{ "RxVfValid", 15, 1 },
8569 	{ "SynLearned", 14, 1 },
8570 	{ "SetDelEntry", 13, 1 },
8571 	{ "SetInvEntry", 12, 1 },
8572 	{ "CpcmdDvld", 11, 1 },
8573 	{ "CpcmdSave", 10, 1 },
8574 	{ "RxPstructsFull", 8, 2 },
8575 	{ "EpcmdDvld", 7, 1 },
8576 	{ "EpcmdFlush", 6, 1 },
8577 	{ "EpcmdTrimPrefix", 5, 1 },
8578 	{ "EpcmdTrimPostfix", 4, 1 },
8579 	{ "ERssIp4Pkt", 3, 1 },
8580 	{ "ERssIp6Pkt", 2, 1 },
8581 	{ "ERssTcpUdpPkt", 1, 1 },
8582 	{ "ERssFceFipPkt", 0, 1 },
8583 	{ NULL }
8584 };
8585 
8586 static const struct field_desc tp_la2[] = {
8587 	{ "CplCmdIn", 56, 8 },
8588 	{ "MpsVfVld", 55, 1 },
8589 	{ "MpsPf", 52, 3 },
8590 	{ "MpsVf", 44, 8 },
8591 	{ "SynIn", 43, 1 },
8592 	{ "AckIn", 42, 1 },
8593 	{ "FinIn", 41, 1 },
8594 	{ "RstIn", 40, 1 },
8595 	{ "DataIn", 39, 1 },
8596 	{ "DataInVld", 38, 1 },
8597 	{ "PadIn", 37, 1 },
8598 	{ "RxBufEmpty", 36, 1 },
8599 	{ "RxDdp", 35, 1 },
8600 	{ "RxFbCongestion", 34, 1 },
8601 	{ "TxFbCongestion", 33, 1 },
8602 	{ "TxPktSumSrdy", 32, 1 },
8603 	{ "RcfUlpType", 28, 4 },
8604 	{ "Eread", 27, 1 },
8605 	{ "Ebypass", 26, 1 },
8606 	{ "Esave", 25, 1 },
8607 	{ "Static0", 24, 1 },
8608 	{ "Cread", 23, 1 },
8609 	{ "Cbypass", 22, 1 },
8610 	{ "Csave", 21, 1 },
8611 	{ "CPktOut", 20, 1 },
8612 	{ "RxPagePoolFull", 18, 2 },
8613 	{ "RxLpbkPkt", 17, 1 },
8614 	{ "TxLpbkPkt", 16, 1 },
8615 	{ "RxVfValid", 15, 1 },
8616 	{ "SynLearned", 14, 1 },
8617 	{ "SetDelEntry", 13, 1 },
8618 	{ "SetInvEntry", 12, 1 },
8619 	{ "CpcmdDvld", 11, 1 },
8620 	{ "CpcmdSave", 10, 1 },
8621 	{ "RxPstructsFull", 8, 2 },
8622 	{ "EpcmdDvld", 7, 1 },
8623 	{ "EpcmdFlush", 6, 1 },
8624 	{ "EpcmdTrimPrefix", 5, 1 },
8625 	{ "EpcmdTrimPostfix", 4, 1 },
8626 	{ "ERssIp4Pkt", 3, 1 },
8627 	{ "ERssIp6Pkt", 2, 1 },
8628 	{ "ERssTcpUdpPkt", 1, 1 },
8629 	{ "ERssFceFipPkt", 0, 1 },
8630 	{ NULL }
8631 };
8632 
8633 static void
8634 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
8635 {
8636 
8637 	field_desc_show(sb, *p, tp_la0);
8638 }
8639 
8640 static void
8641 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
8642 {
8643 
8644 	if (idx)
8645 		sbuf_printf(sb, "\n");
8646 	field_desc_show(sb, p[0], tp_la0);
8647 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8648 		field_desc_show(sb, p[1], tp_la0);
8649 }
8650 
8651 static void
8652 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
8653 {
8654 
8655 	if (idx)
8656 		sbuf_printf(sb, "\n");
8657 	field_desc_show(sb, p[0], tp_la0);
8658 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8659 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
8660 }
8661 
8662 static int
8663 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
8664 {
8665 	struct adapter *sc = arg1;
8666 	struct sbuf *sb;
8667 	uint64_t *buf, *p;
8668 	int rc;
8669 	u_int i, inc;
8670 	void (*show_func)(struct sbuf *, uint64_t *, int);
8671 
8672 	rc = sysctl_wire_old_buffer(req, 0);
8673 	if (rc != 0)
8674 		return (rc);
8675 
8676 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8677 	if (sb == NULL)
8678 		return (ENOMEM);
8679 
8680 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
8681 
8682 	t4_tp_read_la(sc, buf, NULL);
8683 	p = buf;
8684 
8685 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
8686 	case 2:
8687 		inc = 2;
8688 		show_func = tp_la_show2;
8689 		break;
8690 	case 3:
8691 		inc = 2;
8692 		show_func = tp_la_show3;
8693 		break;
8694 	default:
8695 		inc = 1;
8696 		show_func = tp_la_show;
8697 	}
8698 
8699 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
8700 		(*show_func)(sb, p, i);
8701 
8702 	rc = sbuf_finish(sb);
8703 	sbuf_delete(sb);
8704 	free(buf, M_CXGBE);
8705 	return (rc);
8706 }
8707 
8708 static int
8709 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
8710 {
8711 	struct adapter *sc = arg1;
8712 	struct sbuf *sb;
8713 	int rc;
8714 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
8715 
8716 	rc = sysctl_wire_old_buffer(req, 0);
8717 	if (rc != 0)
8718 		return (rc);
8719 
8720 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8721 	if (sb == NULL)
8722 		return (ENOMEM);
8723 
8724 	t4_get_chan_txrate(sc, nrate, orate);
8725 
8726 	if (sc->chip_params->nchan > 2) {
8727 		sbuf_printf(sb, "              channel 0   channel 1"
8728 		    "   channel 2   channel 3\n");
8729 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
8730 		    nrate[0], nrate[1], nrate[2], nrate[3]);
8731 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
8732 		    orate[0], orate[1], orate[2], orate[3]);
8733 	} else {
8734 		sbuf_printf(sb, "              channel 0   channel 1\n");
8735 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
8736 		    nrate[0], nrate[1]);
8737 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
8738 		    orate[0], orate[1]);
8739 	}
8740 
8741 	rc = sbuf_finish(sb);
8742 	sbuf_delete(sb);
8743 
8744 	return (rc);
8745 }
8746 
8747 static int
8748 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
8749 {
8750 	struct adapter *sc = arg1;
8751 	struct sbuf *sb;
8752 	uint32_t *buf, *p;
8753 	int rc, i;
8754 
8755 	rc = sysctl_wire_old_buffer(req, 0);
8756 	if (rc != 0)
8757 		return (rc);
8758 
8759 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8760 	if (sb == NULL)
8761 		return (ENOMEM);
8762 
8763 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
8764 	    M_ZERO | M_WAITOK);
8765 
8766 	t4_ulprx_read_la(sc, buf);
8767 	p = buf;
8768 
8769 	sbuf_printf(sb, "      Pcmd        Type   Message"
8770 	    "                Data");
8771 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
8772 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
8773 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
8774 	}
8775 
8776 	rc = sbuf_finish(sb);
8777 	sbuf_delete(sb);
8778 	free(buf, M_CXGBE);
8779 	return (rc);
8780 }
8781 
8782 static int
8783 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
8784 {
8785 	struct adapter *sc = arg1;
8786 	struct sbuf *sb;
8787 	int rc, v;
8788 
8789 	MPASS(chip_id(sc) >= CHELSIO_T5);
8790 
8791 	rc = sysctl_wire_old_buffer(req, 0);
8792 	if (rc != 0)
8793 		return (rc);
8794 
8795 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8796 	if (sb == NULL)
8797 		return (ENOMEM);
8798 
8799 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
8800 	if (G_STATSOURCE_T5(v) == 7) {
8801 		int mode;
8802 
8803 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
8804 		if (mode == 0) {
8805 			sbuf_printf(sb, "total %d, incomplete %d",
8806 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8807 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8808 		} else if (mode == 1) {
8809 			sbuf_printf(sb, "total %d, data overflow %d",
8810 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8811 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8812 		} else {
8813 			sbuf_printf(sb, "unknown mode %d", mode);
8814 		}
8815 	}
8816 	rc = sbuf_finish(sb);
8817 	sbuf_delete(sb);
8818 
8819 	return (rc);
8820 }
8821 
8822 static int
8823 sysctl_cpus(SYSCTL_HANDLER_ARGS)
8824 {
8825 	struct adapter *sc = arg1;
8826 	enum cpu_sets op = arg2;
8827 	cpuset_t cpuset;
8828 	struct sbuf *sb;
8829 	int i, rc;
8830 
8831 	MPASS(op == LOCAL_CPUS || op == INTR_CPUS);
8832 
8833 	CPU_ZERO(&cpuset);
8834 	rc = bus_get_cpus(sc->dev, op, sizeof(cpuset), &cpuset);
8835 	if (rc != 0)
8836 		return (rc);
8837 
8838 	rc = sysctl_wire_old_buffer(req, 0);
8839 	if (rc != 0)
8840 		return (rc);
8841 
8842 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8843 	if (sb == NULL)
8844 		return (ENOMEM);
8845 
8846 	CPU_FOREACH(i)
8847 		sbuf_printf(sb, "%d ", i);
8848 	rc = sbuf_finish(sb);
8849 	sbuf_delete(sb);
8850 
8851 	return (rc);
8852 }
8853 
8854 #ifdef TCP_OFFLOAD
8855 static int
8856 sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS)
8857 {
8858 	struct adapter *sc = arg1;
8859 	int *old_ports, *new_ports;
8860 	int i, new_count, rc;
8861 
8862 	if (req->newptr == NULL && req->oldptr == NULL)
8863 		return (SYSCTL_OUT(req, NULL, imax(sc->tt.num_tls_rx_ports, 1) *
8864 		    sizeof(sc->tt.tls_rx_ports[0])));
8865 
8866 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tlsrx");
8867 	if (rc)
8868 		return (rc);
8869 
8870 	if (sc->tt.num_tls_rx_ports == 0) {
8871 		i = -1;
8872 		rc = SYSCTL_OUT(req, &i, sizeof(i));
8873 	} else
8874 		rc = SYSCTL_OUT(req, sc->tt.tls_rx_ports,
8875 		    sc->tt.num_tls_rx_ports * sizeof(sc->tt.tls_rx_ports[0]));
8876 	if (rc == 0 && req->newptr != NULL) {
8877 		new_count = req->newlen / sizeof(new_ports[0]);
8878 		new_ports = malloc(new_count * sizeof(new_ports[0]), M_CXGBE,
8879 		    M_WAITOK);
8880 		rc = SYSCTL_IN(req, new_ports, new_count *
8881 		    sizeof(new_ports[0]));
8882 		if (rc)
8883 			goto err;
8884 
8885 		/* Allow setting to a single '-1' to clear the list. */
8886 		if (new_count == 1 && new_ports[0] == -1) {
8887 			ADAPTER_LOCK(sc);
8888 			old_ports = sc->tt.tls_rx_ports;
8889 			sc->tt.tls_rx_ports = NULL;
8890 			sc->tt.num_tls_rx_ports = 0;
8891 			ADAPTER_UNLOCK(sc);
8892 			free(old_ports, M_CXGBE);
8893 		} else {
8894 			for (i = 0; i < new_count; i++) {
8895 				if (new_ports[i] < 1 ||
8896 				    new_ports[i] > IPPORT_MAX) {
8897 					rc = EINVAL;
8898 					goto err;
8899 				}
8900 			}
8901 
8902 			ADAPTER_LOCK(sc);
8903 			old_ports = sc->tt.tls_rx_ports;
8904 			sc->tt.tls_rx_ports = new_ports;
8905 			sc->tt.num_tls_rx_ports = new_count;
8906 			ADAPTER_UNLOCK(sc);
8907 			free(old_ports, M_CXGBE);
8908 			new_ports = NULL;
8909 		}
8910 	err:
8911 		free(new_ports, M_CXGBE);
8912 	}
8913 	end_synchronized_op(sc, 0);
8914 	return (rc);
8915 }
8916 
8917 static void
8918 unit_conv(char *buf, size_t len, u_int val, u_int factor)
8919 {
8920 	u_int rem = val % factor;
8921 
8922 	if (rem == 0)
8923 		snprintf(buf, len, "%u", val / factor);
8924 	else {
8925 		while (rem % 10 == 0)
8926 			rem /= 10;
8927 		snprintf(buf, len, "%u.%u", val / factor, rem);
8928 	}
8929 }
8930 
8931 static int
8932 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
8933 {
8934 	struct adapter *sc = arg1;
8935 	char buf[16];
8936 	u_int res, re;
8937 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8938 
8939 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8940 	switch (arg2) {
8941 	case 0:
8942 		/* timer_tick */
8943 		re = G_TIMERRESOLUTION(res);
8944 		break;
8945 	case 1:
8946 		/* TCP timestamp tick */
8947 		re = G_TIMESTAMPRESOLUTION(res);
8948 		break;
8949 	case 2:
8950 		/* DACK tick */
8951 		re = G_DELAYEDACKRESOLUTION(res);
8952 		break;
8953 	default:
8954 		return (EDOOFUS);
8955 	}
8956 
8957 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
8958 
8959 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
8960 }
8961 
8962 static int
8963 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
8964 {
8965 	struct adapter *sc = arg1;
8966 	u_int res, dack_re, v;
8967 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8968 
8969 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8970 	dack_re = G_DELAYEDACKRESOLUTION(res);
8971 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
8972 
8973 	return (sysctl_handle_int(oidp, &v, 0, req));
8974 }
8975 
8976 static int
8977 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
8978 {
8979 	struct adapter *sc = arg1;
8980 	int reg = arg2;
8981 	u_int tre;
8982 	u_long tp_tick_us, v;
8983 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8984 
8985 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
8986 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
8987 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
8988 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
8989 
8990 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
8991 	tp_tick_us = (cclk_ps << tre) / 1000000;
8992 
8993 	if (reg == A_TP_INIT_SRTT)
8994 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
8995 	else
8996 		v = tp_tick_us * t4_read_reg(sc, reg);
8997 
8998 	return (sysctl_handle_long(oidp, &v, 0, req));
8999 }
9000 
9001 /*
9002  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
9003  * passed to this function.
9004  */
9005 static int
9006 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
9007 {
9008 	struct adapter *sc = arg1;
9009 	int idx = arg2;
9010 	u_int v;
9011 
9012 	MPASS(idx >= 0 && idx <= 24);
9013 
9014 	v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
9015 
9016 	return (sysctl_handle_int(oidp, &v, 0, req));
9017 }
9018 
9019 static int
9020 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
9021 {
9022 	struct adapter *sc = arg1;
9023 	int idx = arg2;
9024 	u_int shift, v, r;
9025 
9026 	MPASS(idx >= 0 && idx < 16);
9027 
9028 	r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
9029 	shift = (idx & 3) << 3;
9030 	v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
9031 
9032 	return (sysctl_handle_int(oidp, &v, 0, req));
9033 }
9034 
9035 static int
9036 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
9037 {
9038 	struct vi_info *vi = arg1;
9039 	struct adapter *sc = vi->pi->adapter;
9040 	int idx, rc, i;
9041 	struct sge_ofld_rxq *ofld_rxq;
9042 	uint8_t v;
9043 
9044 	idx = vi->ofld_tmr_idx;
9045 
9046 	rc = sysctl_handle_int(oidp, &idx, 0, req);
9047 	if (rc != 0 || req->newptr == NULL)
9048 		return (rc);
9049 
9050 	if (idx < 0 || idx >= SGE_NTIMERS)
9051 		return (EINVAL);
9052 
9053 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9054 	    "t4otmr");
9055 	if (rc)
9056 		return (rc);
9057 
9058 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
9059 	for_each_ofld_rxq(vi, i, ofld_rxq) {
9060 #ifdef atomic_store_rel_8
9061 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
9062 #else
9063 		ofld_rxq->iq.intr_params = v;
9064 #endif
9065 	}
9066 	vi->ofld_tmr_idx = idx;
9067 
9068 	end_synchronized_op(sc, LOCK_HELD);
9069 	return (0);
9070 }
9071 
9072 static int
9073 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
9074 {
9075 	struct vi_info *vi = arg1;
9076 	struct adapter *sc = vi->pi->adapter;
9077 	int idx, rc;
9078 
9079 	idx = vi->ofld_pktc_idx;
9080 
9081 	rc = sysctl_handle_int(oidp, &idx, 0, req);
9082 	if (rc != 0 || req->newptr == NULL)
9083 		return (rc);
9084 
9085 	if (idx < -1 || idx >= SGE_NCOUNTERS)
9086 		return (EINVAL);
9087 
9088 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9089 	    "t4opktc");
9090 	if (rc)
9091 		return (rc);
9092 
9093 	if (vi->flags & VI_INIT_DONE)
9094 		rc = EBUSY; /* cannot be changed once the queues are created */
9095 	else
9096 		vi->ofld_pktc_idx = idx;
9097 
9098 	end_synchronized_op(sc, LOCK_HELD);
9099 	return (rc);
9100 }
9101 #endif
9102 
9103 static int
9104 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
9105 {
9106 	int rc;
9107 
9108 	if (cntxt->cid > M_CTXTQID)
9109 		return (EINVAL);
9110 
9111 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
9112 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
9113 		return (EINVAL);
9114 
9115 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
9116 	if (rc)
9117 		return (rc);
9118 
9119 	if (sc->flags & FW_OK) {
9120 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
9121 		    &cntxt->data[0]);
9122 		if (rc == 0)
9123 			goto done;
9124 	}
9125 
9126 	/*
9127 	 * Read via firmware failed or wasn't even attempted.  Read directly via
9128 	 * the backdoor.
9129 	 */
9130 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
9131 done:
9132 	end_synchronized_op(sc, 0);
9133 	return (rc);
9134 }
9135 
9136 static int
9137 load_fw(struct adapter *sc, struct t4_data *fw)
9138 {
9139 	int rc;
9140 	uint8_t *fw_data;
9141 
9142 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
9143 	if (rc)
9144 		return (rc);
9145 
9146 	/*
9147 	 * The firmware, with the sole exception of the memory parity error
9148 	 * handler, runs from memory and not flash.  It is almost always safe to
9149 	 * install a new firmware on a running system.  Just set bit 1 in
9150 	 * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
9151 	 */
9152 	if (sc->flags & FULL_INIT_DONE &&
9153 	    (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
9154 		rc = EBUSY;
9155 		goto done;
9156 	}
9157 
9158 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
9159 	if (fw_data == NULL) {
9160 		rc = ENOMEM;
9161 		goto done;
9162 	}
9163 
9164 	rc = copyin(fw->data, fw_data, fw->len);
9165 	if (rc == 0)
9166 		rc = -t4_load_fw(sc, fw_data, fw->len);
9167 
9168 	free(fw_data, M_CXGBE);
9169 done:
9170 	end_synchronized_op(sc, 0);
9171 	return (rc);
9172 }
9173 
9174 static int
9175 load_cfg(struct adapter *sc, struct t4_data *cfg)
9176 {
9177 	int rc;
9178 	uint8_t *cfg_data = NULL;
9179 
9180 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9181 	if (rc)
9182 		return (rc);
9183 
9184 	if (cfg->len == 0) {
9185 		/* clear */
9186 		rc = -t4_load_cfg(sc, NULL, 0);
9187 		goto done;
9188 	}
9189 
9190 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
9191 	if (cfg_data == NULL) {
9192 		rc = ENOMEM;
9193 		goto done;
9194 	}
9195 
9196 	rc = copyin(cfg->data, cfg_data, cfg->len);
9197 	if (rc == 0)
9198 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
9199 
9200 	free(cfg_data, M_CXGBE);
9201 done:
9202 	end_synchronized_op(sc, 0);
9203 	return (rc);
9204 }
9205 
9206 static int
9207 load_boot(struct adapter *sc, struct t4_bootrom *br)
9208 {
9209 	int rc;
9210 	uint8_t *br_data = NULL;
9211 	u_int offset;
9212 
9213 	if (br->len > 1024 * 1024)
9214 		return (EFBIG);
9215 
9216 	if (br->pf_offset == 0) {
9217 		/* pfidx */
9218 		if (br->pfidx_addr > 7)
9219 			return (EINVAL);
9220 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
9221 		    A_PCIE_PF_EXPROM_OFST)));
9222 	} else if (br->pf_offset == 1) {
9223 		/* offset */
9224 		offset = G_OFFSET(br->pfidx_addr);
9225 	} else {
9226 		return (EINVAL);
9227 	}
9228 
9229 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
9230 	if (rc)
9231 		return (rc);
9232 
9233 	if (br->len == 0) {
9234 		/* clear */
9235 		rc = -t4_load_boot(sc, NULL, offset, 0);
9236 		goto done;
9237 	}
9238 
9239 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
9240 	if (br_data == NULL) {
9241 		rc = ENOMEM;
9242 		goto done;
9243 	}
9244 
9245 	rc = copyin(br->data, br_data, br->len);
9246 	if (rc == 0)
9247 		rc = -t4_load_boot(sc, br_data, offset, br->len);
9248 
9249 	free(br_data, M_CXGBE);
9250 done:
9251 	end_synchronized_op(sc, 0);
9252 	return (rc);
9253 }
9254 
9255 static int
9256 load_bootcfg(struct adapter *sc, struct t4_data *bc)
9257 {
9258 	int rc;
9259 	uint8_t *bc_data = NULL;
9260 
9261 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9262 	if (rc)
9263 		return (rc);
9264 
9265 	if (bc->len == 0) {
9266 		/* clear */
9267 		rc = -t4_load_bootcfg(sc, NULL, 0);
9268 		goto done;
9269 	}
9270 
9271 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
9272 	if (bc_data == NULL) {
9273 		rc = ENOMEM;
9274 		goto done;
9275 	}
9276 
9277 	rc = copyin(bc->data, bc_data, bc->len);
9278 	if (rc == 0)
9279 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
9280 
9281 	free(bc_data, M_CXGBE);
9282 done:
9283 	end_synchronized_op(sc, 0);
9284 	return (rc);
9285 }
9286 
9287 static int
9288 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
9289 {
9290 	int rc;
9291 	struct cudbg_init *cudbg;
9292 	void *handle, *buf;
9293 
9294 	/* buf is large, don't block if no memory is available */
9295 	buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
9296 	if (buf == NULL)
9297 		return (ENOMEM);
9298 
9299 	handle = cudbg_alloc_handle();
9300 	if (handle == NULL) {
9301 		rc = ENOMEM;
9302 		goto done;
9303 	}
9304 
9305 	cudbg = cudbg_get_init(handle);
9306 	cudbg->adap = sc;
9307 	cudbg->print = (cudbg_print_cb)printf;
9308 
9309 #ifndef notyet
9310 	device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
9311 	    __func__, dump->wr_flash, dump->len, dump->data);
9312 #endif
9313 
9314 	if (dump->wr_flash)
9315 		cudbg->use_flash = 1;
9316 	MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
9317 	memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
9318 
9319 	rc = cudbg_collect(handle, buf, &dump->len);
9320 	if (rc != 0)
9321 		goto done;
9322 
9323 	rc = copyout(buf, dump->data, dump->len);
9324 done:
9325 	cudbg_free_handle(handle);
9326 	free(buf, M_CXGBE);
9327 	return (rc);
9328 }
9329 
9330 static void
9331 free_offload_policy(struct t4_offload_policy *op)
9332 {
9333 	struct offload_rule *r;
9334 	int i;
9335 
9336 	if (op == NULL)
9337 		return;
9338 
9339 	r = &op->rule[0];
9340 	for (i = 0; i < op->nrules; i++, r++) {
9341 		free(r->bpf_prog.bf_insns, M_CXGBE);
9342 	}
9343 	free(op->rule, M_CXGBE);
9344 	free(op, M_CXGBE);
9345 }
9346 
9347 static int
9348 set_offload_policy(struct adapter *sc, struct t4_offload_policy *uop)
9349 {
9350 	int i, rc, len;
9351 	struct t4_offload_policy *op, *old;
9352 	struct bpf_program *bf;
9353 	const struct offload_settings *s;
9354 	struct offload_rule *r;
9355 	void *u;
9356 
9357 	if (!is_offload(sc))
9358 		return (ENODEV);
9359 
9360 	if (uop->nrules == 0) {
9361 		/* Delete installed policies. */
9362 		op = NULL;
9363 		goto set_policy;
9364 	} if (uop->nrules > 256) { /* arbitrary */
9365 		return (E2BIG);
9366 	}
9367 
9368 	/* Copy userspace offload policy to kernel */
9369 	op = malloc(sizeof(*op), M_CXGBE, M_ZERO | M_WAITOK);
9370 	op->nrules = uop->nrules;
9371 	len = op->nrules * sizeof(struct offload_rule);
9372 	op->rule = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9373 	rc = copyin(uop->rule, op->rule, len);
9374 	if (rc) {
9375 		free(op->rule, M_CXGBE);
9376 		free(op, M_CXGBE);
9377 		return (rc);
9378 	}
9379 
9380 	r = &op->rule[0];
9381 	for (i = 0; i < op->nrules; i++, r++) {
9382 
9383 		/* Validate open_type */
9384 		if (r->open_type != OPEN_TYPE_LISTEN &&
9385 		    r->open_type != OPEN_TYPE_ACTIVE &&
9386 		    r->open_type != OPEN_TYPE_PASSIVE &&
9387 		    r->open_type != OPEN_TYPE_DONTCARE) {
9388 error:
9389 			/*
9390 			 * Rules 0 to i have malloc'd filters that need to be
9391 			 * freed.  Rules i+1 to nrules have userspace pointers
9392 			 * and should be left alone.
9393 			 */
9394 			op->nrules = i;
9395 			free_offload_policy(op);
9396 			return (rc);
9397 		}
9398 
9399 		/* Validate settings */
9400 		s = &r->settings;
9401 		if ((s->offload != 0 && s->offload != 1) ||
9402 		    s->cong_algo < -1 || s->cong_algo > CONG_ALG_HIGHSPEED ||
9403 		    s->sched_class < -1 ||
9404 		    s->sched_class >= sc->chip_params->nsched_cls) {
9405 			rc = EINVAL;
9406 			goto error;
9407 		}
9408 
9409 		bf = &r->bpf_prog;
9410 		u = bf->bf_insns;	/* userspace ptr */
9411 		bf->bf_insns = NULL;
9412 		if (bf->bf_len == 0) {
9413 			/* legal, matches everything */
9414 			continue;
9415 		}
9416 		len = bf->bf_len * sizeof(*bf->bf_insns);
9417 		bf->bf_insns = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9418 		rc = copyin(u, bf->bf_insns, len);
9419 		if (rc != 0)
9420 			goto error;
9421 
9422 		if (!bpf_validate(bf->bf_insns, bf->bf_len)) {
9423 			rc = EINVAL;
9424 			goto error;
9425 		}
9426 	}
9427 set_policy:
9428 	rw_wlock(&sc->policy_lock);
9429 	old = sc->policy;
9430 	sc->policy = op;
9431 	rw_wunlock(&sc->policy_lock);
9432 	free_offload_policy(old);
9433 
9434 	return (0);
9435 }
9436 
9437 #define MAX_READ_BUF_SIZE (128 * 1024)
9438 static int
9439 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
9440 {
9441 	uint32_t addr, remaining, n;
9442 	uint32_t *buf;
9443 	int rc;
9444 	uint8_t *dst;
9445 
9446 	rc = validate_mem_range(sc, mr->addr, mr->len);
9447 	if (rc != 0)
9448 		return (rc);
9449 
9450 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
9451 	addr = mr->addr;
9452 	remaining = mr->len;
9453 	dst = (void *)mr->data;
9454 
9455 	while (remaining) {
9456 		n = min(remaining, MAX_READ_BUF_SIZE);
9457 		read_via_memwin(sc, 2, addr, buf, n);
9458 
9459 		rc = copyout(buf, dst, n);
9460 		if (rc != 0)
9461 			break;
9462 
9463 		dst += n;
9464 		remaining -= n;
9465 		addr += n;
9466 	}
9467 
9468 	free(buf, M_CXGBE);
9469 	return (rc);
9470 }
9471 #undef MAX_READ_BUF_SIZE
9472 
9473 static int
9474 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
9475 {
9476 	int rc;
9477 
9478 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
9479 		return (EINVAL);
9480 
9481 	if (i2cd->len > sizeof(i2cd->data))
9482 		return (EFBIG);
9483 
9484 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9485 	if (rc)
9486 		return (rc);
9487 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9488 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
9489 	end_synchronized_op(sc, 0);
9490 
9491 	return (rc);
9492 }
9493 
9494 int
9495 t4_os_find_pci_capability(struct adapter *sc, int cap)
9496 {
9497 	int i;
9498 
9499 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9500 }
9501 
9502 int
9503 t4_os_pci_save_state(struct adapter *sc)
9504 {
9505 	device_t dev;
9506 	struct pci_devinfo *dinfo;
9507 
9508 	dev = sc->dev;
9509 	dinfo = device_get_ivars(dev);
9510 
9511 	pci_cfg_save(dev, dinfo, 0);
9512 	return (0);
9513 }
9514 
9515 int
9516 t4_os_pci_restore_state(struct adapter *sc)
9517 {
9518 	device_t dev;
9519 	struct pci_devinfo *dinfo;
9520 
9521 	dev = sc->dev;
9522 	dinfo = device_get_ivars(dev);
9523 
9524 	pci_cfg_restore(dev, dinfo);
9525 	return (0);
9526 }
9527 
9528 void
9529 t4_os_portmod_changed(struct port_info *pi)
9530 {
9531 	struct adapter *sc = pi->adapter;
9532 	struct vi_info *vi;
9533 	struct ifnet *ifp;
9534 	static const char *mod_str[] = {
9535 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9536 	};
9537 
9538 	KASSERT((pi->flags & FIXED_IFMEDIA) == 0,
9539 	    ("%s: port_type %u", __func__, pi->port_type));
9540 
9541 	vi = &pi->vi[0];
9542 	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9543 		PORT_LOCK(pi);
9544 		build_medialist(pi);
9545 		if (pi->mod_type != FW_PORT_MOD_TYPE_NONE) {
9546 			fixup_link_config(pi);
9547 			apply_link_config(pi);
9548 		}
9549 		PORT_UNLOCK(pi);
9550 		end_synchronized_op(sc, LOCK_HELD);
9551 	}
9552 
9553 	ifp = vi->ifp;
9554 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9555 		if_printf(ifp, "transceiver unplugged.\n");
9556 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9557 		if_printf(ifp, "unknown transceiver inserted.\n");
9558 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9559 		if_printf(ifp, "unsupported transceiver inserted.\n");
9560 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9561 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9562 		    port_top_speed(pi), mod_str[pi->mod_type]);
9563 	} else {
9564 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9565 		    pi->mod_type);
9566 	}
9567 }
9568 
9569 void
9570 t4_os_link_changed(struct port_info *pi)
9571 {
9572 	struct vi_info *vi;
9573 	struct ifnet *ifp;
9574 	struct link_config *lc;
9575 	int v;
9576 
9577 	PORT_LOCK_ASSERT_OWNED(pi);
9578 
9579 	for_each_vi(pi, v, vi) {
9580 		ifp = vi->ifp;
9581 		if (ifp == NULL)
9582 			continue;
9583 
9584 		lc = &pi->link_cfg;
9585 		if (lc->link_ok) {
9586 			ifp->if_baudrate = IF_Mbps(lc->speed);
9587 			if_link_state_change(ifp, LINK_STATE_UP);
9588 		} else {
9589 			if_link_state_change(ifp, LINK_STATE_DOWN);
9590 		}
9591 	}
9592 }
9593 
9594 void
9595 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9596 {
9597 	struct adapter *sc;
9598 
9599 	sx_slock(&t4_list_lock);
9600 	SLIST_FOREACH(sc, &t4_list, link) {
9601 		/*
9602 		 * func should not make any assumptions about what state sc is
9603 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9604 		 */
9605 		func(sc, arg);
9606 	}
9607 	sx_sunlock(&t4_list_lock);
9608 }
9609 
9610 static int
9611 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9612     struct thread *td)
9613 {
9614 	int rc;
9615 	struct adapter *sc = dev->si_drv1;
9616 
9617 	rc = priv_check(td, PRIV_DRIVER);
9618 	if (rc != 0)
9619 		return (rc);
9620 
9621 	switch (cmd) {
9622 	case CHELSIO_T4_GETREG: {
9623 		struct t4_reg *edata = (struct t4_reg *)data;
9624 
9625 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9626 			return (EFAULT);
9627 
9628 		if (edata->size == 4)
9629 			edata->val = t4_read_reg(sc, edata->addr);
9630 		else if (edata->size == 8)
9631 			edata->val = t4_read_reg64(sc, edata->addr);
9632 		else
9633 			return (EINVAL);
9634 
9635 		break;
9636 	}
9637 	case CHELSIO_T4_SETREG: {
9638 		struct t4_reg *edata = (struct t4_reg *)data;
9639 
9640 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9641 			return (EFAULT);
9642 
9643 		if (edata->size == 4) {
9644 			if (edata->val & 0xffffffff00000000)
9645 				return (EINVAL);
9646 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9647 		} else if (edata->size == 8)
9648 			t4_write_reg64(sc, edata->addr, edata->val);
9649 		else
9650 			return (EINVAL);
9651 		break;
9652 	}
9653 	case CHELSIO_T4_REGDUMP: {
9654 		struct t4_regdump *regs = (struct t4_regdump *)data;
9655 		int reglen = t4_get_regs_len(sc);
9656 		uint8_t *buf;
9657 
9658 		if (regs->len < reglen) {
9659 			regs->len = reglen; /* hint to the caller */
9660 			return (ENOBUFS);
9661 		}
9662 
9663 		regs->len = reglen;
9664 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9665 		get_regs(sc, regs, buf);
9666 		rc = copyout(buf, regs->data, reglen);
9667 		free(buf, M_CXGBE);
9668 		break;
9669 	}
9670 	case CHELSIO_T4_GET_FILTER_MODE:
9671 		rc = get_filter_mode(sc, (uint32_t *)data);
9672 		break;
9673 	case CHELSIO_T4_SET_FILTER_MODE:
9674 		rc = set_filter_mode(sc, *(uint32_t *)data);
9675 		break;
9676 	case CHELSIO_T4_GET_FILTER:
9677 		rc = get_filter(sc, (struct t4_filter *)data);
9678 		break;
9679 	case CHELSIO_T4_SET_FILTER:
9680 		rc = set_filter(sc, (struct t4_filter *)data);
9681 		break;
9682 	case CHELSIO_T4_DEL_FILTER:
9683 		rc = del_filter(sc, (struct t4_filter *)data);
9684 		break;
9685 	case CHELSIO_T4_GET_SGE_CONTEXT:
9686 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9687 		break;
9688 	case CHELSIO_T4_LOAD_FW:
9689 		rc = load_fw(sc, (struct t4_data *)data);
9690 		break;
9691 	case CHELSIO_T4_GET_MEM:
9692 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9693 		break;
9694 	case CHELSIO_T4_GET_I2C:
9695 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9696 		break;
9697 	case CHELSIO_T4_CLEAR_STATS: {
9698 		int i, v, bg_map;
9699 		u_int port_id = *(uint32_t *)data;
9700 		struct port_info *pi;
9701 		struct vi_info *vi;
9702 
9703 		if (port_id >= sc->params.nports)
9704 			return (EINVAL);
9705 		pi = sc->port[port_id];
9706 		if (pi == NULL)
9707 			return (EIO);
9708 
9709 		/* MAC stats */
9710 		t4_clr_port_stats(sc, pi->tx_chan);
9711 		pi->tx_parse_error = 0;
9712 		pi->tnl_cong_drops = 0;
9713 		mtx_lock(&sc->reg_lock);
9714 		for_each_vi(pi, v, vi) {
9715 			if (vi->flags & VI_INIT_DONE)
9716 				t4_clr_vi_stats(sc, vi->viid);
9717 		}
9718 		bg_map = pi->mps_bg_map;
9719 		v = 0;	/* reuse */
9720 		while (bg_map) {
9721 			i = ffs(bg_map) - 1;
9722 			t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
9723 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
9724 			bg_map &= ~(1 << i);
9725 		}
9726 		mtx_unlock(&sc->reg_lock);
9727 
9728 		/*
9729 		 * Since this command accepts a port, clear stats for
9730 		 * all VIs on this port.
9731 		 */
9732 		for_each_vi(pi, v, vi) {
9733 			if (vi->flags & VI_INIT_DONE) {
9734 				struct sge_rxq *rxq;
9735 				struct sge_txq *txq;
9736 				struct sge_wrq *wrq;
9737 
9738 				for_each_rxq(vi, i, rxq) {
9739 #if defined(INET) || defined(INET6)
9740 					rxq->lro.lro_queued = 0;
9741 					rxq->lro.lro_flushed = 0;
9742 #endif
9743 					rxq->rxcsum = 0;
9744 					rxq->vlan_extraction = 0;
9745 				}
9746 
9747 				for_each_txq(vi, i, txq) {
9748 					txq->txcsum = 0;
9749 					txq->tso_wrs = 0;
9750 					txq->vlan_insertion = 0;
9751 					txq->imm_wrs = 0;
9752 					txq->sgl_wrs = 0;
9753 					txq->txpkt_wrs = 0;
9754 					txq->txpkts0_wrs = 0;
9755 					txq->txpkts1_wrs = 0;
9756 					txq->txpkts0_pkts = 0;
9757 					txq->txpkts1_pkts = 0;
9758 					mp_ring_reset_stats(txq->r);
9759 				}
9760 
9761 #ifdef TCP_OFFLOAD
9762 				/* nothing to clear for each ofld_rxq */
9763 
9764 				for_each_ofld_txq(vi, i, wrq) {
9765 					wrq->tx_wrs_direct = 0;
9766 					wrq->tx_wrs_copied = 0;
9767 				}
9768 #endif
9769 
9770 				if (IS_MAIN_VI(vi)) {
9771 					wrq = &sc->sge.ctrlq[pi->port_id];
9772 					wrq->tx_wrs_direct = 0;
9773 					wrq->tx_wrs_copied = 0;
9774 				}
9775 			}
9776 		}
9777 		break;
9778 	}
9779 	case CHELSIO_T4_SCHED_CLASS:
9780 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9781 		break;
9782 	case CHELSIO_T4_SCHED_QUEUE:
9783 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9784 		break;
9785 	case CHELSIO_T4_GET_TRACER:
9786 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9787 		break;
9788 	case CHELSIO_T4_SET_TRACER:
9789 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9790 		break;
9791 	case CHELSIO_T4_LOAD_CFG:
9792 		rc = load_cfg(sc, (struct t4_data *)data);
9793 		break;
9794 	case CHELSIO_T4_LOAD_BOOT:
9795 		rc = load_boot(sc, (struct t4_bootrom *)data);
9796 		break;
9797 	case CHELSIO_T4_LOAD_BOOTCFG:
9798 		rc = load_bootcfg(sc, (struct t4_data *)data);
9799 		break;
9800 	case CHELSIO_T4_CUDBG_DUMP:
9801 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
9802 		break;
9803 	case CHELSIO_T4_SET_OFLD_POLICY:
9804 		rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
9805 		break;
9806 	default:
9807 		rc = ENOTTY;
9808 	}
9809 
9810 	return (rc);
9811 }
9812 
9813 void
9814 t4_db_full(struct adapter *sc)
9815 {
9816 
9817 	CXGBE_UNIMPLEMENTED(__func__);
9818 }
9819 
9820 void
9821 t4_db_dropped(struct adapter *sc)
9822 {
9823 
9824 	CXGBE_UNIMPLEMENTED(__func__);
9825 }
9826 
9827 #ifdef TCP_OFFLOAD
9828 static int
9829 toe_capability(struct vi_info *vi, int enable)
9830 {
9831 	int rc;
9832 	struct port_info *pi = vi->pi;
9833 	struct adapter *sc = pi->adapter;
9834 
9835 	ASSERT_SYNCHRONIZED_OP(sc);
9836 
9837 	if (!is_offload(sc))
9838 		return (ENODEV);
9839 
9840 	if (enable) {
9841 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9842 			/* TOE is already enabled. */
9843 			return (0);
9844 		}
9845 
9846 		/*
9847 		 * We need the port's queues around so that we're able to send
9848 		 * and receive CPLs to/from the TOE even if the ifnet for this
9849 		 * port has never been UP'd administratively.
9850 		 */
9851 		if (!(vi->flags & VI_INIT_DONE)) {
9852 			rc = vi_full_init(vi);
9853 			if (rc)
9854 				return (rc);
9855 		}
9856 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9857 			rc = vi_full_init(&pi->vi[0]);
9858 			if (rc)
9859 				return (rc);
9860 		}
9861 
9862 		if (isset(&sc->offload_map, pi->port_id)) {
9863 			/* TOE is enabled on another VI of this port. */
9864 			pi->uld_vis++;
9865 			return (0);
9866 		}
9867 
9868 		if (!uld_active(sc, ULD_TOM)) {
9869 			rc = t4_activate_uld(sc, ULD_TOM);
9870 			if (rc == EAGAIN) {
9871 				log(LOG_WARNING,
9872 				    "You must kldload t4_tom.ko before trying "
9873 				    "to enable TOE on a cxgbe interface.\n");
9874 			}
9875 			if (rc != 0)
9876 				return (rc);
9877 			KASSERT(sc->tom_softc != NULL,
9878 			    ("%s: TOM activated but softc NULL", __func__));
9879 			KASSERT(uld_active(sc, ULD_TOM),
9880 			    ("%s: TOM activated but flag not set", __func__));
9881 		}
9882 
9883 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
9884 		if (!uld_active(sc, ULD_IWARP))
9885 			(void) t4_activate_uld(sc, ULD_IWARP);
9886 		if (!uld_active(sc, ULD_ISCSI))
9887 			(void) t4_activate_uld(sc, ULD_ISCSI);
9888 
9889 		pi->uld_vis++;
9890 		setbit(&sc->offload_map, pi->port_id);
9891 	} else {
9892 		pi->uld_vis--;
9893 
9894 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9895 			return (0);
9896 
9897 		KASSERT(uld_active(sc, ULD_TOM),
9898 		    ("%s: TOM never initialized?", __func__));
9899 		clrbit(&sc->offload_map, pi->port_id);
9900 	}
9901 
9902 	return (0);
9903 }
9904 
9905 /*
9906  * Add an upper layer driver to the global list.
9907  */
9908 int
9909 t4_register_uld(struct uld_info *ui)
9910 {
9911 	int rc = 0;
9912 	struct uld_info *u;
9913 
9914 	sx_xlock(&t4_uld_list_lock);
9915 	SLIST_FOREACH(u, &t4_uld_list, link) {
9916 	    if (u->uld_id == ui->uld_id) {
9917 		    rc = EEXIST;
9918 		    goto done;
9919 	    }
9920 	}
9921 
9922 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9923 	ui->refcount = 0;
9924 done:
9925 	sx_xunlock(&t4_uld_list_lock);
9926 	return (rc);
9927 }
9928 
9929 int
9930 t4_unregister_uld(struct uld_info *ui)
9931 {
9932 	int rc = EINVAL;
9933 	struct uld_info *u;
9934 
9935 	sx_xlock(&t4_uld_list_lock);
9936 
9937 	SLIST_FOREACH(u, &t4_uld_list, link) {
9938 	    if (u == ui) {
9939 		    if (ui->refcount > 0) {
9940 			    rc = EBUSY;
9941 			    goto done;
9942 		    }
9943 
9944 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9945 		    rc = 0;
9946 		    goto done;
9947 	    }
9948 	}
9949 done:
9950 	sx_xunlock(&t4_uld_list_lock);
9951 	return (rc);
9952 }
9953 
9954 int
9955 t4_activate_uld(struct adapter *sc, int id)
9956 {
9957 	int rc;
9958 	struct uld_info *ui;
9959 
9960 	ASSERT_SYNCHRONIZED_OP(sc);
9961 
9962 	if (id < 0 || id > ULD_MAX)
9963 		return (EINVAL);
9964 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
9965 
9966 	sx_slock(&t4_uld_list_lock);
9967 
9968 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9969 		if (ui->uld_id == id) {
9970 			if (!(sc->flags & FULL_INIT_DONE)) {
9971 				rc = adapter_full_init(sc);
9972 				if (rc != 0)
9973 					break;
9974 			}
9975 
9976 			rc = ui->activate(sc);
9977 			if (rc == 0) {
9978 				setbit(&sc->active_ulds, id);
9979 				ui->refcount++;
9980 			}
9981 			break;
9982 		}
9983 	}
9984 
9985 	sx_sunlock(&t4_uld_list_lock);
9986 
9987 	return (rc);
9988 }
9989 
9990 int
9991 t4_deactivate_uld(struct adapter *sc, int id)
9992 {
9993 	int rc;
9994 	struct uld_info *ui;
9995 
9996 	ASSERT_SYNCHRONIZED_OP(sc);
9997 
9998 	if (id < 0 || id > ULD_MAX)
9999 		return (EINVAL);
10000 	rc = ENXIO;
10001 
10002 	sx_slock(&t4_uld_list_lock);
10003 
10004 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10005 		if (ui->uld_id == id) {
10006 			rc = ui->deactivate(sc);
10007 			if (rc == 0) {
10008 				clrbit(&sc->active_ulds, id);
10009 				ui->refcount--;
10010 			}
10011 			break;
10012 		}
10013 	}
10014 
10015 	sx_sunlock(&t4_uld_list_lock);
10016 
10017 	return (rc);
10018 }
10019 
10020 int
10021 uld_active(struct adapter *sc, int uld_id)
10022 {
10023 
10024 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
10025 
10026 	return (isset(&sc->active_ulds, uld_id));
10027 }
10028 #endif
10029 
10030 /*
10031  * t  = ptr to tunable.
10032  * nc = number of CPUs.
10033  * c  = compiled in default for that tunable.
10034  */
10035 static void
10036 calculate_nqueues(int *t, int nc, const int c)
10037 {
10038 	int nq;
10039 
10040 	if (*t > 0)
10041 		return;
10042 	nq = *t < 0 ? -*t : c;
10043 	*t = min(nc, nq);
10044 }
10045 
10046 /*
10047  * Come up with reasonable defaults for some of the tunables, provided they're
10048  * not set by the user (in which case we'll use the values as is).
10049  */
10050 static void
10051 tweak_tunables(void)
10052 {
10053 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
10054 
10055 	if (t4_ntxq < 1) {
10056 #ifdef RSS
10057 		t4_ntxq = rss_getnumbuckets();
10058 #else
10059 		calculate_nqueues(&t4_ntxq, nc, NTXQ);
10060 #endif
10061 	}
10062 
10063 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
10064 
10065 	if (t4_nrxq < 1) {
10066 #ifdef RSS
10067 		t4_nrxq = rss_getnumbuckets();
10068 #else
10069 		calculate_nqueues(&t4_nrxq, nc, NRXQ);
10070 #endif
10071 	}
10072 
10073 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
10074 
10075 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
10076 	calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
10077 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
10078 #endif
10079 #ifdef TCP_OFFLOAD
10080 	calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
10081 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
10082 
10083 	if (t4_toecaps_allowed == -1)
10084 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
10085 
10086 	if (t4_rdmacaps_allowed == -1) {
10087 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
10088 		    FW_CAPS_CONFIG_RDMA_RDMAC;
10089 	}
10090 
10091 	if (t4_iscsicaps_allowed == -1) {
10092 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
10093 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
10094 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
10095 	}
10096 
10097 	if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
10098 		t4_tmr_idx_ofld = TMR_IDX_OFLD;
10099 
10100 	if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
10101 		t4_pktc_idx_ofld = PKTC_IDX_OFLD;
10102 #else
10103 	if (t4_toecaps_allowed == -1)
10104 		t4_toecaps_allowed = 0;
10105 
10106 	if (t4_rdmacaps_allowed == -1)
10107 		t4_rdmacaps_allowed = 0;
10108 
10109 	if (t4_iscsicaps_allowed == -1)
10110 		t4_iscsicaps_allowed = 0;
10111 #endif
10112 
10113 #ifdef DEV_NETMAP
10114 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
10115 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
10116 #endif
10117 
10118 	if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
10119 		t4_tmr_idx = TMR_IDX;
10120 
10121 	if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
10122 		t4_pktc_idx = PKTC_IDX;
10123 
10124 	if (t4_qsize_txq < 128)
10125 		t4_qsize_txq = 128;
10126 
10127 	if (t4_qsize_rxq < 128)
10128 		t4_qsize_rxq = 128;
10129 	while (t4_qsize_rxq & 7)
10130 		t4_qsize_rxq++;
10131 
10132 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
10133 
10134 	/*
10135 	 * Number of VIs to create per-port.  The first VI is the "main" regular
10136 	 * VI for the port.  The rest are additional virtual interfaces on the
10137 	 * same physical port.  Note that the main VI does not have native
10138 	 * netmap support but the extra VIs do.
10139 	 *
10140 	 * Limit the number of VIs per port to the number of available
10141 	 * MAC addresses per port.
10142 	 */
10143 	if (t4_num_vis < 1)
10144 		t4_num_vis = 1;
10145 	if (t4_num_vis > nitems(vi_mac_funcs)) {
10146 		t4_num_vis = nitems(vi_mac_funcs);
10147 		printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
10148 	}
10149 
10150 	if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
10151 		pcie_relaxed_ordering = 1;
10152 #if defined(__i386__) || defined(__amd64__)
10153 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
10154 			pcie_relaxed_ordering = 0;
10155 #endif
10156 	}
10157 }
10158 
10159 #ifdef DDB
10160 static void
10161 t4_dump_tcb(struct adapter *sc, int tid)
10162 {
10163 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
10164 
10165 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
10166 	save = t4_read_reg(sc, reg);
10167 	base = sc->memwin[2].mw_base;
10168 
10169 	/* Dump TCB for the tid */
10170 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
10171 	tcb_addr += tid * TCB_SIZE;
10172 
10173 	if (is_t4(sc)) {
10174 		pf = 0;
10175 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
10176 	} else {
10177 		pf = V_PFNUM(sc->pf);
10178 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
10179 	}
10180 	t4_write_reg(sc, reg, win_pos | pf);
10181 	t4_read_reg(sc, reg);
10182 
10183 	off = tcb_addr - win_pos;
10184 	for (i = 0; i < 4; i++) {
10185 		uint32_t buf[8];
10186 		for (j = 0; j < 8; j++, off += 4)
10187 			buf[j] = htonl(t4_read_reg(sc, base + off));
10188 
10189 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
10190 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
10191 		    buf[7]);
10192 	}
10193 
10194 	t4_write_reg(sc, reg, save);
10195 	t4_read_reg(sc, reg);
10196 }
10197 
10198 static void
10199 t4_dump_devlog(struct adapter *sc)
10200 {
10201 	struct devlog_params *dparams = &sc->params.devlog;
10202 	struct fw_devlog_e e;
10203 	int i, first, j, m, nentries, rc;
10204 	uint64_t ftstamp = UINT64_MAX;
10205 
10206 	if (dparams->start == 0) {
10207 		db_printf("devlog params not valid\n");
10208 		return;
10209 	}
10210 
10211 	nentries = dparams->size / sizeof(struct fw_devlog_e);
10212 	m = fwmtype_to_hwmtype(dparams->memtype);
10213 
10214 	/* Find the first entry. */
10215 	first = -1;
10216 	for (i = 0; i < nentries && !db_pager_quit; i++) {
10217 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10218 		    sizeof(e), (void *)&e);
10219 		if (rc != 0)
10220 			break;
10221 
10222 		if (e.timestamp == 0)
10223 			break;
10224 
10225 		e.timestamp = be64toh(e.timestamp);
10226 		if (e.timestamp < ftstamp) {
10227 			ftstamp = e.timestamp;
10228 			first = i;
10229 		}
10230 	}
10231 
10232 	if (first == -1)
10233 		return;
10234 
10235 	i = first;
10236 	do {
10237 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10238 		    sizeof(e), (void *)&e);
10239 		if (rc != 0)
10240 			return;
10241 
10242 		if (e.timestamp == 0)
10243 			return;
10244 
10245 		e.timestamp = be64toh(e.timestamp);
10246 		e.seqno = be32toh(e.seqno);
10247 		for (j = 0; j < 8; j++)
10248 			e.params[j] = be32toh(e.params[j]);
10249 
10250 		db_printf("%10d  %15ju  %8s  %8s  ",
10251 		    e.seqno, e.timestamp,
10252 		    (e.level < nitems(devlog_level_strings) ?
10253 			devlog_level_strings[e.level] : "UNKNOWN"),
10254 		    (e.facility < nitems(devlog_facility_strings) ?
10255 			devlog_facility_strings[e.facility] : "UNKNOWN"));
10256 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
10257 		    e.params[3], e.params[4], e.params[5], e.params[6],
10258 		    e.params[7]);
10259 
10260 		if (++i == nentries)
10261 			i = 0;
10262 	} while (i != first && !db_pager_quit);
10263 }
10264 
10265 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
10266 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
10267 
10268 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
10269 {
10270 	device_t dev;
10271 	int t;
10272 	bool valid;
10273 
10274 	valid = false;
10275 	t = db_read_token();
10276 	if (t == tIDENT) {
10277 		dev = device_lookup_by_name(db_tok_string);
10278 		valid = true;
10279 	}
10280 	db_skip_to_eol();
10281 	if (!valid) {
10282 		db_printf("usage: show t4 devlog <nexus>\n");
10283 		return;
10284 	}
10285 
10286 	if (dev == NULL) {
10287 		db_printf("device not found\n");
10288 		return;
10289 	}
10290 
10291 	t4_dump_devlog(device_get_softc(dev));
10292 }
10293 
10294 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
10295 {
10296 	device_t dev;
10297 	int radix, tid, t;
10298 	bool valid;
10299 
10300 	valid = false;
10301 	radix = db_radix;
10302 	db_radix = 10;
10303 	t = db_read_token();
10304 	if (t == tIDENT) {
10305 		dev = device_lookup_by_name(db_tok_string);
10306 		t = db_read_token();
10307 		if (t == tNUMBER) {
10308 			tid = db_tok_number;
10309 			valid = true;
10310 		}
10311 	}
10312 	db_radix = radix;
10313 	db_skip_to_eol();
10314 	if (!valid) {
10315 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
10316 		return;
10317 	}
10318 
10319 	if (dev == NULL) {
10320 		db_printf("device not found\n");
10321 		return;
10322 	}
10323 	if (tid < 0) {
10324 		db_printf("invalid tid\n");
10325 		return;
10326 	}
10327 
10328 	t4_dump_tcb(device_get_softc(dev), tid);
10329 }
10330 #endif
10331 
10332 /*
10333  * Borrowed from cesa_prep_aes_key().
10334  *
10335  * NB: The crypto engine wants the words in the decryption key in reverse
10336  * order.
10337  */
10338 void
10339 t4_aes_getdeckey(void *dec_key, const void *enc_key, unsigned int kbits)
10340 {
10341 	uint32_t ek[4 * (RIJNDAEL_MAXNR + 1)];
10342 	uint32_t *dkey;
10343 	int i;
10344 
10345 	rijndaelKeySetupEnc(ek, enc_key, kbits);
10346 	dkey = dec_key;
10347 	dkey += (kbits / 8) / 4;
10348 
10349 	switch (kbits) {
10350 	case 128:
10351 		for (i = 0; i < 4; i++)
10352 			*--dkey = htobe32(ek[4 * 10 + i]);
10353 		break;
10354 	case 192:
10355 		for (i = 0; i < 2; i++)
10356 			*--dkey = htobe32(ek[4 * 11 + 2 + i]);
10357 		for (i = 0; i < 4; i++)
10358 			*--dkey = htobe32(ek[4 * 12 + i]);
10359 		break;
10360 	case 256:
10361 		for (i = 0; i < 4; i++)
10362 			*--dkey = htobe32(ek[4 * 13 + i]);
10363 		for (i = 0; i < 4; i++)
10364 			*--dkey = htobe32(ek[4 * 14 + i]);
10365 		break;
10366 	}
10367 	MPASS(dkey == dec_key);
10368 }
10369 
10370 static struct sx mlu;	/* mod load unload */
10371 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
10372 
10373 static int
10374 mod_event(module_t mod, int cmd, void *arg)
10375 {
10376 	int rc = 0;
10377 	static int loaded = 0;
10378 
10379 	switch (cmd) {
10380 	case MOD_LOAD:
10381 		sx_xlock(&mlu);
10382 		if (loaded++ == 0) {
10383 			t4_sge_modload();
10384 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10385 			    t4_filter_rpl, CPL_COOKIE_FILTER);
10386 			t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
10387 			    do_l2t_write_rpl, CPL_COOKIE_FILTER);
10388 			t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL,
10389 			    t4_hashfilter_ao_rpl, CPL_COOKIE_HASHFILTER);
10390 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10391 			    t4_hashfilter_tcb_rpl, CPL_COOKIE_HASHFILTER);
10392 			t4_register_shared_cpl_handler(CPL_ABORT_RPL_RSS,
10393 			    t4_del_hashfilter_rpl, CPL_COOKIE_HASHFILTER);
10394 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
10395 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
10396 			t4_register_cpl_handler(CPL_SMT_WRITE_RPL,
10397 			    do_smt_write_rpl);
10398 			sx_init(&t4_list_lock, "T4/T5 adapters");
10399 			SLIST_INIT(&t4_list);
10400 #ifdef TCP_OFFLOAD
10401 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
10402 			SLIST_INIT(&t4_uld_list);
10403 #endif
10404 			t4_tracer_modload();
10405 			tweak_tunables();
10406 		}
10407 		sx_xunlock(&mlu);
10408 		break;
10409 
10410 	case MOD_UNLOAD:
10411 		sx_xlock(&mlu);
10412 		if (--loaded == 0) {
10413 			int tries;
10414 
10415 			sx_slock(&t4_list_lock);
10416 			if (!SLIST_EMPTY(&t4_list)) {
10417 				rc = EBUSY;
10418 				sx_sunlock(&t4_list_lock);
10419 				goto done_unload;
10420 			}
10421 #ifdef TCP_OFFLOAD
10422 			sx_slock(&t4_uld_list_lock);
10423 			if (!SLIST_EMPTY(&t4_uld_list)) {
10424 				rc = EBUSY;
10425 				sx_sunlock(&t4_uld_list_lock);
10426 				sx_sunlock(&t4_list_lock);
10427 				goto done_unload;
10428 			}
10429 #endif
10430 			tries = 0;
10431 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
10432 				uprintf("%ju clusters with custom free routine "
10433 				    "still is use.\n", t4_sge_extfree_refs());
10434 				pause("t4unload", 2 * hz);
10435 			}
10436 #ifdef TCP_OFFLOAD
10437 			sx_sunlock(&t4_uld_list_lock);
10438 #endif
10439 			sx_sunlock(&t4_list_lock);
10440 
10441 			if (t4_sge_extfree_refs() == 0) {
10442 				t4_tracer_modunload();
10443 #ifdef TCP_OFFLOAD
10444 				sx_destroy(&t4_uld_list_lock);
10445 #endif
10446 				sx_destroy(&t4_list_lock);
10447 				t4_sge_modunload();
10448 				loaded = 0;
10449 			} else {
10450 				rc = EBUSY;
10451 				loaded++;	/* undo earlier decrement */
10452 			}
10453 		}
10454 done_unload:
10455 		sx_xunlock(&mlu);
10456 		break;
10457 	}
10458 
10459 	return (rc);
10460 }
10461 
10462 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10463 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10464 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10465 
10466 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10467 MODULE_VERSION(t4nex, 1);
10468 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10469 #ifdef DEV_NETMAP
10470 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10471 #endif /* DEV_NETMAP */
10472 
10473 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10474 MODULE_VERSION(t5nex, 1);
10475 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10476 #ifdef DEV_NETMAP
10477 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10478 #endif /* DEV_NETMAP */
10479 
10480 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10481 MODULE_VERSION(t6nex, 1);
10482 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10483 #ifdef DEV_NETMAP
10484 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10485 #endif /* DEV_NETMAP */
10486 
10487 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10488 MODULE_VERSION(cxgbe, 1);
10489 
10490 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10491 MODULE_VERSION(cxl, 1);
10492 
10493 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10494 MODULE_VERSION(cc, 1);
10495 
10496 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10497 MODULE_VERSION(vcxgbe, 1);
10498 
10499 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10500 MODULE_VERSION(vcxl, 1);
10501 
10502 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10503 MODULE_VERSION(vcc, 1);
10504