xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision d6b92ffa990dc57d8a969a98e3ca4e25ba39cbb2)
1 /*-
2  * Copyright (c) 2011 Chelsio Communications, Inc.
3  * All rights reserved.
4  * Written by: Navdeep Parhar <np@FreeBSD.org>
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_ddb.h"
32 #include "opt_inet.h"
33 #include "opt_inet6.h"
34 #include "opt_rss.h"
35 
36 #include <sys/param.h>
37 #include <sys/conf.h>
38 #include <sys/priv.h>
39 #include <sys/kernel.h>
40 #include <sys/bus.h>
41 #include <sys/module.h>
42 #include <sys/malloc.h>
43 #include <sys/queue.h>
44 #include <sys/taskqueue.h>
45 #include <sys/pciio.h>
46 #include <dev/pci/pcireg.h>
47 #include <dev/pci/pcivar.h>
48 #include <dev/pci/pci_private.h>
49 #include <sys/firmware.h>
50 #include <sys/sbuf.h>
51 #include <sys/smp.h>
52 #include <sys/socket.h>
53 #include <sys/sockio.h>
54 #include <sys/sysctl.h>
55 #include <net/ethernet.h>
56 #include <net/if.h>
57 #include <net/if_types.h>
58 #include <net/if_dl.h>
59 #include <net/if_vlan_var.h>
60 #ifdef RSS
61 #include <net/rss_config.h>
62 #endif
63 #if defined(__i386__) || defined(__amd64__)
64 #include <vm/vm.h>
65 #include <vm/pmap.h>
66 #endif
67 #ifdef DDB
68 #include <ddb/ddb.h>
69 #include <ddb/db_lex.h>
70 #endif
71 
72 #include "common/common.h"
73 #include "common/t4_msg.h"
74 #include "common/t4_regs.h"
75 #include "common/t4_regs_values.h"
76 #include "t4_ioctl.h"
77 #include "t4_l2t.h"
78 #include "t4_mp_ring.h"
79 #include "t4_if.h"
80 
81 /* T4 bus driver interface */
82 static int t4_probe(device_t);
83 static int t4_attach(device_t);
84 static int t4_detach(device_t);
85 static int t4_ready(device_t);
86 static int t4_read_port_device(device_t, int, device_t *);
87 static device_method_t t4_methods[] = {
88 	DEVMETHOD(device_probe,		t4_probe),
89 	DEVMETHOD(device_attach,	t4_attach),
90 	DEVMETHOD(device_detach,	t4_detach),
91 
92 	DEVMETHOD(t4_is_main_ready,	t4_ready),
93 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
94 
95 	DEVMETHOD_END
96 };
97 static driver_t t4_driver = {
98 	"t4nex",
99 	t4_methods,
100 	sizeof(struct adapter)
101 };
102 
103 
104 /* T4 port (cxgbe) interface */
105 static int cxgbe_probe(device_t);
106 static int cxgbe_attach(device_t);
107 static int cxgbe_detach(device_t);
108 device_method_t cxgbe_methods[] = {
109 	DEVMETHOD(device_probe,		cxgbe_probe),
110 	DEVMETHOD(device_attach,	cxgbe_attach),
111 	DEVMETHOD(device_detach,	cxgbe_detach),
112 	{ 0, 0 }
113 };
114 static driver_t cxgbe_driver = {
115 	"cxgbe",
116 	cxgbe_methods,
117 	sizeof(struct port_info)
118 };
119 
120 /* T4 VI (vcxgbe) interface */
121 static int vcxgbe_probe(device_t);
122 static int vcxgbe_attach(device_t);
123 static int vcxgbe_detach(device_t);
124 static device_method_t vcxgbe_methods[] = {
125 	DEVMETHOD(device_probe,		vcxgbe_probe),
126 	DEVMETHOD(device_attach,	vcxgbe_attach),
127 	DEVMETHOD(device_detach,	vcxgbe_detach),
128 	{ 0, 0 }
129 };
130 static driver_t vcxgbe_driver = {
131 	"vcxgbe",
132 	vcxgbe_methods,
133 	sizeof(struct vi_info)
134 };
135 
136 static d_ioctl_t t4_ioctl;
137 
138 static struct cdevsw t4_cdevsw = {
139        .d_version = D_VERSION,
140        .d_ioctl = t4_ioctl,
141        .d_name = "t4nex",
142 };
143 
144 /* T5 bus driver interface */
145 static int t5_probe(device_t);
146 static device_method_t t5_methods[] = {
147 	DEVMETHOD(device_probe,		t5_probe),
148 	DEVMETHOD(device_attach,	t4_attach),
149 	DEVMETHOD(device_detach,	t4_detach),
150 
151 	DEVMETHOD(t4_is_main_ready,	t4_ready),
152 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
153 
154 	DEVMETHOD_END
155 };
156 static driver_t t5_driver = {
157 	"t5nex",
158 	t5_methods,
159 	sizeof(struct adapter)
160 };
161 
162 
163 /* T5 port (cxl) interface */
164 static driver_t cxl_driver = {
165 	"cxl",
166 	cxgbe_methods,
167 	sizeof(struct port_info)
168 };
169 
170 /* T5 VI (vcxl) interface */
171 static driver_t vcxl_driver = {
172 	"vcxl",
173 	vcxgbe_methods,
174 	sizeof(struct vi_info)
175 };
176 
177 /* T6 bus driver interface */
178 static int t6_probe(device_t);
179 static device_method_t t6_methods[] = {
180 	DEVMETHOD(device_probe,		t6_probe),
181 	DEVMETHOD(device_attach,	t4_attach),
182 	DEVMETHOD(device_detach,	t4_detach),
183 
184 	DEVMETHOD(t4_is_main_ready,	t4_ready),
185 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
186 
187 	DEVMETHOD_END
188 };
189 static driver_t t6_driver = {
190 	"t6nex",
191 	t6_methods,
192 	sizeof(struct adapter)
193 };
194 
195 
196 /* T6 port (cc) interface */
197 static driver_t cc_driver = {
198 	"cc",
199 	cxgbe_methods,
200 	sizeof(struct port_info)
201 };
202 
203 /* T6 VI (vcc) interface */
204 static driver_t vcc_driver = {
205 	"vcc",
206 	vcxgbe_methods,
207 	sizeof(struct vi_info)
208 };
209 
210 /* ifnet + media interface */
211 static void cxgbe_init(void *);
212 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
213 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
214 static void cxgbe_qflush(struct ifnet *);
215 static int cxgbe_media_change(struct ifnet *);
216 static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
217 
218 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
219 
220 /*
221  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
222  * then ADAPTER_LOCK, then t4_uld_list_lock.
223  */
224 static struct sx t4_list_lock;
225 SLIST_HEAD(, adapter) t4_list;
226 #ifdef TCP_OFFLOAD
227 static struct sx t4_uld_list_lock;
228 SLIST_HEAD(, uld_info) t4_uld_list;
229 #endif
230 
231 /*
232  * Tunables.  See tweak_tunables() too.
233  *
234  * Each tunable is set to a default value here if it's known at compile-time.
235  * Otherwise it is set to -n as an indication to tweak_tunables() that it should
236  * provide a reasonable default (upto n) when the driver is loaded.
237  *
238  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
239  * T5 are under hw.cxl.
240  */
241 
242 /*
243  * Number of queues for tx and rx, 10G and 1G, NIC and offload.
244  */
245 #define NTXQ_10G 16
246 int t4_ntxq10g = -NTXQ_10G;
247 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq10g);
248 
249 #define NRXQ_10G 8
250 int t4_nrxq10g = -NRXQ_10G;
251 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq10g);
252 
253 #define NTXQ_1G 4
254 int t4_ntxq1g = -NTXQ_1G;
255 TUNABLE_INT("hw.cxgbe.ntxq1g", &t4_ntxq1g);
256 
257 #define NRXQ_1G 2
258 int t4_nrxq1g = -NRXQ_1G;
259 TUNABLE_INT("hw.cxgbe.nrxq1g", &t4_nrxq1g);
260 
261 #define NTXQ_VI 1
262 static int t4_ntxq_vi = -NTXQ_VI;
263 TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
264 
265 #define NRXQ_VI 1
266 static int t4_nrxq_vi = -NRXQ_VI;
267 TUNABLE_INT("hw.cxgbe.nrxq_vi", &t4_nrxq_vi);
268 
269 static int t4_rsrv_noflowq = 0;
270 TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
271 
272 #ifdef TCP_OFFLOAD
273 #define NOFLDTXQ_10G 8
274 static int t4_nofldtxq10g = -NOFLDTXQ_10G;
275 TUNABLE_INT("hw.cxgbe.nofldtxq10g", &t4_nofldtxq10g);
276 
277 #define NOFLDRXQ_10G 2
278 static int t4_nofldrxq10g = -NOFLDRXQ_10G;
279 TUNABLE_INT("hw.cxgbe.nofldrxq10g", &t4_nofldrxq10g);
280 
281 #define NOFLDTXQ_1G 2
282 static int t4_nofldtxq1g = -NOFLDTXQ_1G;
283 TUNABLE_INT("hw.cxgbe.nofldtxq1g", &t4_nofldtxq1g);
284 
285 #define NOFLDRXQ_1G 1
286 static int t4_nofldrxq1g = -NOFLDRXQ_1G;
287 TUNABLE_INT("hw.cxgbe.nofldrxq1g", &t4_nofldrxq1g);
288 
289 #define NOFLDTXQ_VI 1
290 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
291 TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
292 
293 #define NOFLDRXQ_VI 1
294 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
295 TUNABLE_INT("hw.cxgbe.nofldrxq_vi", &t4_nofldrxq_vi);
296 #endif
297 
298 #ifdef DEV_NETMAP
299 #define NNMTXQ_VI 2
300 static int t4_nnmtxq_vi = -NNMTXQ_VI;
301 TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
302 
303 #define NNMRXQ_VI 2
304 static int t4_nnmrxq_vi = -NNMRXQ_VI;
305 TUNABLE_INT("hw.cxgbe.nnmrxq_vi", &t4_nnmrxq_vi);
306 #endif
307 
308 /*
309  * Holdoff parameters for 10G and 1G ports.
310  */
311 #define TMR_IDX_10G 1
312 int t4_tmr_idx_10g = TMR_IDX_10G;
313 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx_10g);
314 
315 #define PKTC_IDX_10G (-1)
316 int t4_pktc_idx_10g = PKTC_IDX_10G;
317 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx_10g);
318 
319 #define TMR_IDX_1G 1
320 int t4_tmr_idx_1g = TMR_IDX_1G;
321 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_1G", &t4_tmr_idx_1g);
322 
323 #define PKTC_IDX_1G (-1)
324 int t4_pktc_idx_1g = PKTC_IDX_1G;
325 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_1G", &t4_pktc_idx_1g);
326 
327 /*
328  * Size (# of entries) of each tx and rx queue.
329  */
330 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
331 TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
332 
333 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
334 TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
335 
336 /*
337  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
338  */
339 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
340 TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
341 
342 /*
343  * Configuration file.
344  */
345 #define DEFAULT_CF	"default"
346 #define FLASH_CF	"flash"
347 #define UWIRE_CF	"uwire"
348 #define FPGA_CF		"fpga"
349 static char t4_cfg_file[32] = DEFAULT_CF;
350 TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
351 
352 /*
353  * PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
354  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
355  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
356  *            mark or when signalled to do so, 0 to never emit PAUSE.
357  */
358 static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
359 TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
360 
361 /*
362  * Forward Error Correction settings (bit 0, 1, 2 = FEC_RS, FEC_BASER_RS,
363  * FEC_RESERVED respectively).
364  * -1 to run with the firmware default.
365  *  0 to disable FEC.
366  */
367 static int t4_fec = -1;
368 TUNABLE_INT("hw.cxgbe.fec", &t4_fec);
369 
370 /*
371  * Link autonegotiation.
372  * -1 to run with the firmware default.
373  *  0 to disable.
374  *  1 to enable.
375  */
376 static int t4_autoneg = -1;
377 TUNABLE_INT("hw.cxgbe.autoneg", &t4_autoneg);
378 
379 /*
380  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
381  * encouraged respectively).
382  */
383 static unsigned int t4_fw_install = 1;
384 TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
385 
386 /*
387  * ASIC features that will be used.  Disable the ones you don't want so that the
388  * chip resources aren't wasted on features that will not be used.
389  */
390 static int t4_nbmcaps_allowed = 0;
391 TUNABLE_INT("hw.cxgbe.nbmcaps_allowed", &t4_nbmcaps_allowed);
392 
393 static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
394 TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
395 
396 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
397     FW_CAPS_CONFIG_SWITCH_EGRESS;
398 TUNABLE_INT("hw.cxgbe.switchcaps_allowed", &t4_switchcaps_allowed);
399 
400 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC;
401 TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
402 
403 static int t4_toecaps_allowed = -1;
404 TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
405 
406 static int t4_rdmacaps_allowed = -1;
407 TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
408 
409 static int t4_cryptocaps_allowed = 0;
410 TUNABLE_INT("hw.cxgbe.cryptocaps_allowed", &t4_cryptocaps_allowed);
411 
412 static int t4_iscsicaps_allowed = -1;
413 TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
414 
415 static int t4_fcoecaps_allowed = 0;
416 TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
417 
418 static int t5_write_combine = 0;
419 TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
420 
421 static int t4_num_vis = 1;
422 TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
423 
424 /* Functions used by extra VIs to obtain unique MAC addresses for each VI. */
425 static int vi_mac_funcs[] = {
426 	FW_VI_FUNC_OFLD,
427 	FW_VI_FUNC_IWARP,
428 	FW_VI_FUNC_OPENISCSI,
429 	FW_VI_FUNC_OPENFCOE,
430 	FW_VI_FUNC_FOISCSI,
431 	FW_VI_FUNC_FOFCOE,
432 };
433 
434 struct intrs_and_queues {
435 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
436 	uint16_t nirq;		/* Total # of vectors */
437 	uint16_t intr_flags_10g;/* Interrupt flags for each 10G port */
438 	uint16_t intr_flags_1g;	/* Interrupt flags for each 1G port */
439 	uint16_t ntxq10g;	/* # of NIC txq's for each 10G port */
440 	uint16_t nrxq10g;	/* # of NIC rxq's for each 10G port */
441 	uint16_t ntxq1g;	/* # of NIC txq's for each 1G port */
442 	uint16_t nrxq1g;	/* # of NIC rxq's for each 1G port */
443 	uint16_t rsrv_noflowq;	/* Flag whether to reserve queue 0 */
444 	uint16_t nofldtxq10g;	/* # of TOE txq's for each 10G port */
445 	uint16_t nofldrxq10g;	/* # of TOE rxq's for each 10G port */
446 	uint16_t nofldtxq1g;	/* # of TOE txq's for each 1G port */
447 	uint16_t nofldrxq1g;	/* # of TOE rxq's for each 1G port */
448 
449 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
450 	uint16_t ntxq_vi;	/* # of NIC txq's */
451 	uint16_t nrxq_vi;	/* # of NIC rxq's */
452 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
453 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
454 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
455 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
456 };
457 
458 struct filter_entry {
459         uint32_t valid:1;	/* filter allocated and valid */
460         uint32_t locked:1;	/* filter is administratively locked */
461         uint32_t pending:1;	/* filter action is pending firmware reply */
462 	uint32_t smtidx:8;	/* Source MAC Table index for smac */
463 	struct l2t_entry *l2t;	/* Layer Two Table entry for dmac */
464 
465         struct t4_filter_specification fs;
466 };
467 
468 static void setup_memwin(struct adapter *);
469 static void position_memwin(struct adapter *, int, uint32_t);
470 static int rw_via_memwin(struct adapter *, int, uint32_t, uint32_t *, int, int);
471 static inline int read_via_memwin(struct adapter *, int, uint32_t, uint32_t *,
472     int);
473 static inline int write_via_memwin(struct adapter *, int, uint32_t,
474     const uint32_t *, int);
475 static int validate_mem_range(struct adapter *, uint32_t, int);
476 static int fwmtype_to_hwmtype(int);
477 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
478     uint32_t *);
479 static int fixup_devlog_params(struct adapter *);
480 static int cfg_itype_and_nqueues(struct adapter *, int, int, int,
481     struct intrs_and_queues *);
482 static int prep_firmware(struct adapter *);
483 static int partition_resources(struct adapter *, const struct firmware *,
484     const char *);
485 static int get_params__pre_init(struct adapter *);
486 static int get_params__post_init(struct adapter *);
487 static int set_params__post_init(struct adapter *);
488 static void t4_set_desc(struct adapter *);
489 static void build_medialist(struct port_info *, struct ifmedia *);
490 static void init_l1cfg(struct port_info *);
491 static int cxgbe_init_synchronized(struct vi_info *);
492 static int cxgbe_uninit_synchronized(struct vi_info *);
493 static void quiesce_txq(struct adapter *, struct sge_txq *);
494 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
495 static void quiesce_iq(struct adapter *, struct sge_iq *);
496 static void quiesce_fl(struct adapter *, struct sge_fl *);
497 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
498     driver_intr_t *, void *, char *);
499 static int t4_free_irq(struct adapter *, struct irq *);
500 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
501 static void vi_refresh_stats(struct adapter *, struct vi_info *);
502 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
503 static void cxgbe_tick(void *);
504 static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
505 static void cxgbe_sysctls(struct port_info *);
506 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
507 static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
508 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
509 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
510 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
511 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
512 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
513 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
514 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
515 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
516 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
517 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
518 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
519 #ifdef SBUF_DRAIN
520 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
521 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
522 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
523 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
524 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
525 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
526 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
527 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
528 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
529 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
530 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
531 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
532 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
533 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
534 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
535 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
536 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
537 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
538 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
539 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
540 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
541 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
542 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
543 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
544 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
545 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
546 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
547 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
548 static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
549 #endif
550 #ifdef TCP_OFFLOAD
551 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
552 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
553 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
554 #endif
555 static uint32_t fconf_iconf_to_mode(uint32_t, uint32_t);
556 static uint32_t mode_to_fconf(uint32_t);
557 static uint32_t mode_to_iconf(uint32_t);
558 static int check_fspec_against_fconf_iconf(struct adapter *,
559     struct t4_filter_specification *);
560 static int get_filter_mode(struct adapter *, uint32_t *);
561 static int set_filter_mode(struct adapter *, uint32_t);
562 static inline uint64_t get_filter_hits(struct adapter *, uint32_t);
563 static int get_filter(struct adapter *, struct t4_filter *);
564 static int set_filter(struct adapter *, struct t4_filter *);
565 static int del_filter(struct adapter *, struct t4_filter *);
566 static void clear_filter(struct filter_entry *);
567 static int set_filter_wr(struct adapter *, int);
568 static int del_filter_wr(struct adapter *, int);
569 static int set_tcb_rpl(struct sge_iq *, const struct rss_header *,
570     struct mbuf *);
571 static int get_sge_context(struct adapter *, struct t4_sge_context *);
572 static int load_fw(struct adapter *, struct t4_data *);
573 static int load_cfg(struct adapter *, struct t4_data *);
574 static int load_boot(struct adapter *, struct t4_bootrom *);
575 static int load_bootcfg(struct adapter *, struct t4_data *);
576 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
577 static int read_i2c(struct adapter *, struct t4_i2c_data *);
578 #ifdef TCP_OFFLOAD
579 static int toe_capability(struct vi_info *, int);
580 #endif
581 static int mod_event(module_t, int, void *);
582 static int notify_siblings(device_t, int);
583 
584 struct {
585 	uint16_t device;
586 	char *desc;
587 } t4_pciids[] = {
588 	{0xa000, "Chelsio Terminator 4 FPGA"},
589 	{0x4400, "Chelsio T440-dbg"},
590 	{0x4401, "Chelsio T420-CR"},
591 	{0x4402, "Chelsio T422-CR"},
592 	{0x4403, "Chelsio T440-CR"},
593 	{0x4404, "Chelsio T420-BCH"},
594 	{0x4405, "Chelsio T440-BCH"},
595 	{0x4406, "Chelsio T440-CH"},
596 	{0x4407, "Chelsio T420-SO"},
597 	{0x4408, "Chelsio T420-CX"},
598 	{0x4409, "Chelsio T420-BT"},
599 	{0x440a, "Chelsio T404-BT"},
600 	{0x440e, "Chelsio T440-LP-CR"},
601 }, t5_pciids[] = {
602 	{0xb000, "Chelsio Terminator 5 FPGA"},
603 	{0x5400, "Chelsio T580-dbg"},
604 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
605 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
606 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
607 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
608 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
609 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
610 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
611 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
612 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
613 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
614 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
615 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
616 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
617 #ifdef notyet
618 	{0x5404,  "Chelsio T520-BCH"},
619 	{0x5405,  "Chelsio T540-BCH"},
620 	{0x5406,  "Chelsio T540-CH"},
621 	{0x5408,  "Chelsio T520-CX"},
622 	{0x540b,  "Chelsio B520-SR"},
623 	{0x540c,  "Chelsio B504-BT"},
624 	{0x540f,  "Chelsio Amsterdam"},
625 	{0x5413,  "Chelsio T580-CHR"},
626 #endif
627 }, t6_pciids[] = {
628 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
629 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
630 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
631 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
632 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
633 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
634 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
635 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
636 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
637 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
638 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
639 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
640 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
641 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
642 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
643 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
644 
645 	/* Custom */
646 	{0x6480, "Chelsio T6225 80"},
647 	{0x6481, "Chelsio T62100 81"},
648 };
649 
650 #ifdef TCP_OFFLOAD
651 /*
652  * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
653  * exactly the same for both rxq and ofld_rxq.
654  */
655 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
656 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
657 #endif
658 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
659 
660 static int
661 t4_probe(device_t dev)
662 {
663 	int i;
664 	uint16_t v = pci_get_vendor(dev);
665 	uint16_t d = pci_get_device(dev);
666 	uint8_t f = pci_get_function(dev);
667 
668 	if (v != PCI_VENDOR_ID_CHELSIO)
669 		return (ENXIO);
670 
671 	/* Attach only to PF0 of the FPGA */
672 	if (d == 0xa000 && f != 0)
673 		return (ENXIO);
674 
675 	for (i = 0; i < nitems(t4_pciids); i++) {
676 		if (d == t4_pciids[i].device) {
677 			device_set_desc(dev, t4_pciids[i].desc);
678 			return (BUS_PROBE_DEFAULT);
679 		}
680 	}
681 
682 	return (ENXIO);
683 }
684 
685 static int
686 t5_probe(device_t dev)
687 {
688 	int i;
689 	uint16_t v = pci_get_vendor(dev);
690 	uint16_t d = pci_get_device(dev);
691 	uint8_t f = pci_get_function(dev);
692 
693 	if (v != PCI_VENDOR_ID_CHELSIO)
694 		return (ENXIO);
695 
696 	/* Attach only to PF0 of the FPGA */
697 	if (d == 0xb000 && f != 0)
698 		return (ENXIO);
699 
700 	for (i = 0; i < nitems(t5_pciids); i++) {
701 		if (d == t5_pciids[i].device) {
702 			device_set_desc(dev, t5_pciids[i].desc);
703 			return (BUS_PROBE_DEFAULT);
704 		}
705 	}
706 
707 	return (ENXIO);
708 }
709 
710 static int
711 t6_probe(device_t dev)
712 {
713 	int i;
714 	uint16_t v = pci_get_vendor(dev);
715 	uint16_t d = pci_get_device(dev);
716 
717 	if (v != PCI_VENDOR_ID_CHELSIO)
718 		return (ENXIO);
719 
720 	for (i = 0; i < nitems(t6_pciids); i++) {
721 		if (d == t6_pciids[i].device) {
722 			device_set_desc(dev, t6_pciids[i].desc);
723 			return (BUS_PROBE_DEFAULT);
724 		}
725 	}
726 
727 	return (ENXIO);
728 }
729 
730 static void
731 t5_attribute_workaround(device_t dev)
732 {
733 	device_t root_port;
734 	uint32_t v;
735 
736 	/*
737 	 * The T5 chips do not properly echo the No Snoop and Relaxed
738 	 * Ordering attributes when replying to a TLP from a Root
739 	 * Port.  As a workaround, find the parent Root Port and
740 	 * disable No Snoop and Relaxed Ordering.  Note that this
741 	 * affects all devices under this root port.
742 	 */
743 	root_port = pci_find_pcie_root_port(dev);
744 	if (root_port == NULL) {
745 		device_printf(dev, "Unable to find parent root port\n");
746 		return;
747 	}
748 
749 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
750 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
751 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
752 	    0)
753 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
754 		    device_get_nameunit(root_port));
755 }
756 
757 static const struct devnames devnames[] = {
758 	{
759 		.nexus_name = "t4nex",
760 		.ifnet_name = "cxgbe",
761 		.vi_ifnet_name = "vcxgbe",
762 		.pf03_drv_name = "t4iov",
763 		.vf_nexus_name = "t4vf",
764 		.vf_ifnet_name = "cxgbev"
765 	}, {
766 		.nexus_name = "t5nex",
767 		.ifnet_name = "cxl",
768 		.vi_ifnet_name = "vcxl",
769 		.pf03_drv_name = "t5iov",
770 		.vf_nexus_name = "t5vf",
771 		.vf_ifnet_name = "cxlv"
772 	}, {
773 		.nexus_name = "t6nex",
774 		.ifnet_name = "cc",
775 		.vi_ifnet_name = "vcc",
776 		.pf03_drv_name = "t6iov",
777 		.vf_nexus_name = "t6vf",
778 		.vf_ifnet_name = "ccv"
779 	}
780 };
781 
782 void
783 t4_init_devnames(struct adapter *sc)
784 {
785 	int id;
786 
787 	id = chip_id(sc);
788 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
789 		sc->names = &devnames[id - CHELSIO_T4];
790 	else {
791 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
792 		sc->names = NULL;
793 	}
794 }
795 
796 static int
797 t4_attach(device_t dev)
798 {
799 	struct adapter *sc;
800 	int rc = 0, i, j, n10g, n1g, rqidx, tqidx;
801 	struct make_dev_args mda;
802 	struct intrs_and_queues iaq;
803 	struct sge *s;
804 	uint8_t *buf;
805 #ifdef TCP_OFFLOAD
806 	int ofld_rqidx, ofld_tqidx;
807 #endif
808 #ifdef DEV_NETMAP
809 	int nm_rqidx, nm_tqidx;
810 #endif
811 	int num_vis;
812 
813 	sc = device_get_softc(dev);
814 	sc->dev = dev;
815 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
816 
817 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
818 		t5_attribute_workaround(dev);
819 	pci_enable_busmaster(dev);
820 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
821 		uint32_t v;
822 
823 		pci_set_max_read_req(dev, 4096);
824 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
825 		v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
826 		pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
827 
828 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
829 	}
830 
831 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
832 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
833 	sc->traceq = -1;
834 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
835 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
836 	    device_get_nameunit(dev));
837 
838 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
839 	    device_get_nameunit(dev));
840 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
841 	t4_add_adapter(sc);
842 
843 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
844 	TAILQ_INIT(&sc->sfl);
845 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
846 
847 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
848 
849 	rc = t4_map_bars_0_and_4(sc);
850 	if (rc != 0)
851 		goto done; /* error message displayed already */
852 
853 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
854 
855 	/* Prepare the adapter for operation. */
856 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
857 	rc = -t4_prep_adapter(sc, buf);
858 	free(buf, M_CXGBE);
859 	if (rc != 0) {
860 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
861 		goto done;
862 	}
863 
864 	/*
865 	 * This is the real PF# to which we're attaching.  Works from within PCI
866 	 * passthrough environments too, where pci_get_function() could return a
867 	 * different PF# depending on the passthrough configuration.  We need to
868 	 * use the real PF# in all our communication with the firmware.
869 	 */
870 	j = t4_read_reg(sc, A_PL_WHOAMI);
871 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
872 	sc->mbox = sc->pf;
873 
874 	t4_init_devnames(sc);
875 	if (sc->names == NULL) {
876 		rc = ENOTSUP;
877 		goto done; /* error message displayed already */
878 	}
879 
880 	/*
881 	 * Do this really early, with the memory windows set up even before the
882 	 * character device.  The userland tool's register i/o and mem read
883 	 * will work even in "recovery mode".
884 	 */
885 	setup_memwin(sc);
886 	if (t4_init_devlog_params(sc, 0) == 0)
887 		fixup_devlog_params(sc);
888 	make_dev_args_init(&mda);
889 	mda.mda_devsw = &t4_cdevsw;
890 	mda.mda_uid = UID_ROOT;
891 	mda.mda_gid = GID_WHEEL;
892 	mda.mda_mode = 0600;
893 	mda.mda_si_drv1 = sc;
894 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
895 	if (rc != 0)
896 		device_printf(dev, "failed to create nexus char device: %d.\n",
897 		    rc);
898 
899 	/* Go no further if recovery mode has been requested. */
900 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
901 		device_printf(dev, "recovery mode.\n");
902 		goto done;
903 	}
904 
905 #if defined(__i386__)
906 	if ((cpu_feature & CPUID_CX8) == 0) {
907 		device_printf(dev, "64 bit atomics not available.\n");
908 		rc = ENOTSUP;
909 		goto done;
910 	}
911 #endif
912 
913 	/* Prepare the firmware for operation */
914 	rc = prep_firmware(sc);
915 	if (rc != 0)
916 		goto done; /* error message displayed already */
917 
918 	rc = get_params__post_init(sc);
919 	if (rc != 0)
920 		goto done; /* error message displayed already */
921 
922 	rc = set_params__post_init(sc);
923 	if (rc != 0)
924 		goto done; /* error message displayed already */
925 
926 	rc = t4_map_bar_2(sc);
927 	if (rc != 0)
928 		goto done; /* error message displayed already */
929 
930 	rc = t4_create_dma_tag(sc);
931 	if (rc != 0)
932 		goto done; /* error message displayed already */
933 
934 	/*
935 	 * Number of VIs to create per-port.  The first VI is the "main" regular
936 	 * VI for the port.  The rest are additional virtual interfaces on the
937 	 * same physical port.  Note that the main VI does not have native
938 	 * netmap support but the extra VIs do.
939 	 *
940 	 * Limit the number of VIs per port to the number of available
941 	 * MAC addresses per port.
942 	 */
943 	if (t4_num_vis >= 1)
944 		num_vis = t4_num_vis;
945 	else
946 		num_vis = 1;
947 	if (num_vis > nitems(vi_mac_funcs)) {
948 		num_vis = nitems(vi_mac_funcs);
949 		device_printf(dev, "Number of VIs limited to %d\n", num_vis);
950 	}
951 
952 	/*
953 	 * First pass over all the ports - allocate VIs and initialize some
954 	 * basic parameters like mac address, port type, etc.  We also figure
955 	 * out whether a port is 10G or 1G and use that information when
956 	 * calculating how many interrupts to attempt to allocate.
957 	 */
958 	n10g = n1g = 0;
959 	for_each_port(sc, i) {
960 		struct port_info *pi;
961 
962 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
963 		sc->port[i] = pi;
964 
965 		/* These must be set before t4_port_init */
966 		pi->adapter = sc;
967 		pi->port_id = i;
968 		/*
969 		 * XXX: vi[0] is special so we can't delay this allocation until
970 		 * pi->nvi's final value is known.
971 		 */
972 		pi->vi = malloc(sizeof(struct vi_info) * num_vis, M_CXGBE,
973 		    M_ZERO | M_WAITOK);
974 
975 		/*
976 		 * Allocate the "main" VI and initialize parameters
977 		 * like mac addr.
978 		 */
979 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
980 		if (rc != 0) {
981 			device_printf(dev, "unable to initialize port %d: %d\n",
982 			    i, rc);
983 			free(pi->vi, M_CXGBE);
984 			free(pi, M_CXGBE);
985 			sc->port[i] = NULL;
986 			goto done;
987 		}
988 
989 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
990 		    device_get_nameunit(dev), i);
991 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
992 		sc->chan_map[pi->tx_chan] = i;
993 
994 		if (port_top_speed(pi) >= 10) {
995 			n10g++;
996 		} else {
997 			n1g++;
998 		}
999 
1000 		pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
1001 		if (pi->dev == NULL) {
1002 			device_printf(dev,
1003 			    "failed to add device for port %d.\n", i);
1004 			rc = ENXIO;
1005 			goto done;
1006 		}
1007 		pi->vi[0].dev = pi->dev;
1008 		device_set_softc(pi->dev, pi);
1009 	}
1010 
1011 	/*
1012 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1013 	 */
1014 	rc = cfg_itype_and_nqueues(sc, n10g, n1g, num_vis, &iaq);
1015 	if (rc != 0)
1016 		goto done; /* error message displayed already */
1017 	if (iaq.nrxq_vi + iaq.nofldrxq_vi + iaq.nnmrxq_vi == 0)
1018 		num_vis = 1;
1019 
1020 	sc->intr_type = iaq.intr_type;
1021 	sc->intr_count = iaq.nirq;
1022 
1023 	s = &sc->sge;
1024 	s->nrxq = n10g * iaq.nrxq10g + n1g * iaq.nrxq1g;
1025 	s->ntxq = n10g * iaq.ntxq10g + n1g * iaq.ntxq1g;
1026 	if (num_vis > 1) {
1027 		s->nrxq += (n10g + n1g) * (num_vis - 1) * iaq.nrxq_vi;
1028 		s->ntxq += (n10g + n1g) * (num_vis - 1) * iaq.ntxq_vi;
1029 	}
1030 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1031 	s->neq += sc->params.nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
1032 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1033 #ifdef TCP_OFFLOAD
1034 	if (is_offload(sc)) {
1035 		s->nofldrxq = n10g * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
1036 		s->nofldtxq = n10g * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
1037 		if (num_vis > 1) {
1038 			s->nofldrxq += (n10g + n1g) * (num_vis - 1) *
1039 			    iaq.nofldrxq_vi;
1040 			s->nofldtxq += (n10g + n1g) * (num_vis - 1) *
1041 			    iaq.nofldtxq_vi;
1042 		}
1043 		s->neq += s->nofldtxq + s->nofldrxq;
1044 		s->niq += s->nofldrxq;
1045 
1046 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1047 		    M_CXGBE, M_ZERO | M_WAITOK);
1048 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1049 		    M_CXGBE, M_ZERO | M_WAITOK);
1050 	}
1051 #endif
1052 #ifdef DEV_NETMAP
1053 	if (num_vis > 1) {
1054 		s->nnmrxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmrxq_vi;
1055 		s->nnmtxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmtxq_vi;
1056 	}
1057 	s->neq += s->nnmtxq + s->nnmrxq;
1058 	s->niq += s->nnmrxq;
1059 
1060 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1061 	    M_CXGBE, M_ZERO | M_WAITOK);
1062 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1063 	    M_CXGBE, M_ZERO | M_WAITOK);
1064 #endif
1065 
1066 	s->ctrlq = malloc(sc->params.nports * sizeof(struct sge_wrq), M_CXGBE,
1067 	    M_ZERO | M_WAITOK);
1068 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1069 	    M_ZERO | M_WAITOK);
1070 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1071 	    M_ZERO | M_WAITOK);
1072 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1073 	    M_ZERO | M_WAITOK);
1074 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1075 	    M_ZERO | M_WAITOK);
1076 
1077 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1078 	    M_ZERO | M_WAITOK);
1079 
1080 	t4_init_l2t(sc, M_WAITOK);
1081 	t4_init_tx_sched(sc);
1082 
1083 	/*
1084 	 * Second pass over the ports.  This time we know the number of rx and
1085 	 * tx queues that each port should get.
1086 	 */
1087 	rqidx = tqidx = 0;
1088 #ifdef TCP_OFFLOAD
1089 	ofld_rqidx = ofld_tqidx = 0;
1090 #endif
1091 #ifdef DEV_NETMAP
1092 	nm_rqidx = nm_tqidx = 0;
1093 #endif
1094 	for_each_port(sc, i) {
1095 		struct port_info *pi = sc->port[i];
1096 		struct vi_info *vi;
1097 
1098 		if (pi == NULL)
1099 			continue;
1100 
1101 		pi->nvi = num_vis;
1102 		for_each_vi(pi, j, vi) {
1103 			vi->pi = pi;
1104 			vi->qsize_rxq = t4_qsize_rxq;
1105 			vi->qsize_txq = t4_qsize_txq;
1106 
1107 			vi->first_rxq = rqidx;
1108 			vi->first_txq = tqidx;
1109 			if (port_top_speed(pi) >= 10) {
1110 				vi->tmr_idx = t4_tmr_idx_10g;
1111 				vi->pktc_idx = t4_pktc_idx_10g;
1112 				vi->flags |= iaq.intr_flags_10g & INTR_RXQ;
1113 				vi->nrxq = j == 0 ? iaq.nrxq10g : iaq.nrxq_vi;
1114 				vi->ntxq = j == 0 ? iaq.ntxq10g : iaq.ntxq_vi;
1115 			} else {
1116 				vi->tmr_idx = t4_tmr_idx_1g;
1117 				vi->pktc_idx = t4_pktc_idx_1g;
1118 				vi->flags |= iaq.intr_flags_1g & INTR_RXQ;
1119 				vi->nrxq = j == 0 ? iaq.nrxq1g : iaq.nrxq_vi;
1120 				vi->ntxq = j == 0 ? iaq.ntxq1g : iaq.ntxq_vi;
1121 			}
1122 			rqidx += vi->nrxq;
1123 			tqidx += vi->ntxq;
1124 
1125 			if (j == 0 && vi->ntxq > 1)
1126 				vi->rsrv_noflowq = iaq.rsrv_noflowq ? 1 : 0;
1127 			else
1128 				vi->rsrv_noflowq = 0;
1129 
1130 #ifdef TCP_OFFLOAD
1131 			vi->first_ofld_rxq = ofld_rqidx;
1132 			vi->first_ofld_txq = ofld_tqidx;
1133 			if (port_top_speed(pi) >= 10) {
1134 				vi->flags |= iaq.intr_flags_10g & INTR_OFLD_RXQ;
1135 				vi->nofldrxq = j == 0 ? iaq.nofldrxq10g :
1136 				    iaq.nofldrxq_vi;
1137 				vi->nofldtxq = j == 0 ? iaq.nofldtxq10g :
1138 				    iaq.nofldtxq_vi;
1139 			} else {
1140 				vi->flags |= iaq.intr_flags_1g & INTR_OFLD_RXQ;
1141 				vi->nofldrxq = j == 0 ? iaq.nofldrxq1g :
1142 				    iaq.nofldrxq_vi;
1143 				vi->nofldtxq = j == 0 ? iaq.nofldtxq1g :
1144 				    iaq.nofldtxq_vi;
1145 			}
1146 			ofld_rqidx += vi->nofldrxq;
1147 			ofld_tqidx += vi->nofldtxq;
1148 #endif
1149 #ifdef DEV_NETMAP
1150 			if (j > 0) {
1151 				vi->first_nm_rxq = nm_rqidx;
1152 				vi->first_nm_txq = nm_tqidx;
1153 				vi->nnmrxq = iaq.nnmrxq_vi;
1154 				vi->nnmtxq = iaq.nnmtxq_vi;
1155 				nm_rqidx += vi->nnmrxq;
1156 				nm_tqidx += vi->nnmtxq;
1157 			}
1158 #endif
1159 		}
1160 	}
1161 
1162 	rc = t4_setup_intr_handlers(sc);
1163 	if (rc != 0) {
1164 		device_printf(dev,
1165 		    "failed to setup interrupt handlers: %d\n", rc);
1166 		goto done;
1167 	}
1168 
1169 	rc = bus_generic_probe(dev);
1170 	if (rc != 0) {
1171 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1172 		goto done;
1173 	}
1174 
1175 	rc = bus_generic_attach(dev);
1176 	if (rc != 0) {
1177 		device_printf(dev,
1178 		    "failed to attach all child ports: %d\n", rc);
1179 		goto done;
1180 	}
1181 
1182 	device_printf(dev,
1183 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1184 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1185 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1186 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1187 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1188 
1189 	t4_set_desc(sc);
1190 
1191 	notify_siblings(dev, 0);
1192 
1193 done:
1194 	if (rc != 0 && sc->cdev) {
1195 		/* cdev was created and so cxgbetool works; recover that way. */
1196 		device_printf(dev,
1197 		    "error during attach, adapter is now in recovery mode.\n");
1198 		rc = 0;
1199 	}
1200 
1201 	if (rc != 0)
1202 		t4_detach_common(dev);
1203 	else
1204 		t4_sysctls(sc);
1205 
1206 	return (rc);
1207 }
1208 
1209 static int
1210 t4_ready(device_t dev)
1211 {
1212 	struct adapter *sc;
1213 
1214 	sc = device_get_softc(dev);
1215 	if (sc->flags & FW_OK)
1216 		return (0);
1217 	return (ENXIO);
1218 }
1219 
1220 static int
1221 t4_read_port_device(device_t dev, int port, device_t *child)
1222 {
1223 	struct adapter *sc;
1224 	struct port_info *pi;
1225 
1226 	sc = device_get_softc(dev);
1227 	if (port < 0 || port >= MAX_NPORTS)
1228 		return (EINVAL);
1229 	pi = sc->port[port];
1230 	if (pi == NULL || pi->dev == NULL)
1231 		return (ENXIO);
1232 	*child = pi->dev;
1233 	return (0);
1234 }
1235 
1236 static int
1237 notify_siblings(device_t dev, int detaching)
1238 {
1239 	device_t sibling;
1240 	int error, i;
1241 
1242 	error = 0;
1243 	for (i = 0; i < PCI_FUNCMAX; i++) {
1244 		if (i == pci_get_function(dev))
1245 			continue;
1246 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1247 		    pci_get_slot(dev), i);
1248 		if (sibling == NULL || !device_is_attached(sibling))
1249 			continue;
1250 		if (detaching)
1251 			error = T4_DETACH_CHILD(sibling);
1252 		else
1253 			(void)T4_ATTACH_CHILD(sibling);
1254 		if (error)
1255 			break;
1256 	}
1257 	return (error);
1258 }
1259 
1260 /*
1261  * Idempotent
1262  */
1263 static int
1264 t4_detach(device_t dev)
1265 {
1266 	struct adapter *sc;
1267 	int rc;
1268 
1269 	sc = device_get_softc(dev);
1270 
1271 	rc = notify_siblings(dev, 1);
1272 	if (rc) {
1273 		device_printf(dev,
1274 		    "failed to detach sibling devices: %d\n", rc);
1275 		return (rc);
1276 	}
1277 
1278 	return (t4_detach_common(dev));
1279 }
1280 
1281 int
1282 t4_detach_common(device_t dev)
1283 {
1284 	struct adapter *sc;
1285 	struct port_info *pi;
1286 	int i, rc;
1287 
1288 	sc = device_get_softc(dev);
1289 
1290 	if (sc->flags & FULL_INIT_DONE) {
1291 		if (!(sc->flags & IS_VF))
1292 			t4_intr_disable(sc);
1293 	}
1294 
1295 	if (sc->cdev) {
1296 		destroy_dev(sc->cdev);
1297 		sc->cdev = NULL;
1298 	}
1299 
1300 	if (device_is_attached(dev)) {
1301 		rc = bus_generic_detach(dev);
1302 		if (rc) {
1303 			device_printf(dev,
1304 			    "failed to detach child devices: %d\n", rc);
1305 			return (rc);
1306 		}
1307 	}
1308 
1309 	for (i = 0; i < sc->intr_count; i++)
1310 		t4_free_irq(sc, &sc->irq[i]);
1311 
1312 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1313 		t4_free_tx_sched(sc);
1314 
1315 	for (i = 0; i < MAX_NPORTS; i++) {
1316 		pi = sc->port[i];
1317 		if (pi) {
1318 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1319 			if (pi->dev)
1320 				device_delete_child(dev, pi->dev);
1321 
1322 			mtx_destroy(&pi->pi_lock);
1323 			free(pi->vi, M_CXGBE);
1324 			free(pi, M_CXGBE);
1325 		}
1326 	}
1327 
1328 	device_delete_children(dev);
1329 
1330 	if (sc->flags & FULL_INIT_DONE)
1331 		adapter_full_uninit(sc);
1332 
1333 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1334 		t4_fw_bye(sc, sc->mbox);
1335 
1336 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1337 		pci_release_msi(dev);
1338 
1339 	if (sc->regs_res)
1340 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1341 		    sc->regs_res);
1342 
1343 	if (sc->udbs_res)
1344 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1345 		    sc->udbs_res);
1346 
1347 	if (sc->msix_res)
1348 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1349 		    sc->msix_res);
1350 
1351 	if (sc->l2t)
1352 		t4_free_l2t(sc->l2t);
1353 
1354 #ifdef TCP_OFFLOAD
1355 	free(sc->sge.ofld_rxq, M_CXGBE);
1356 	free(sc->sge.ofld_txq, M_CXGBE);
1357 #endif
1358 #ifdef DEV_NETMAP
1359 	free(sc->sge.nm_rxq, M_CXGBE);
1360 	free(sc->sge.nm_txq, M_CXGBE);
1361 #endif
1362 	free(sc->irq, M_CXGBE);
1363 	free(sc->sge.rxq, M_CXGBE);
1364 	free(sc->sge.txq, M_CXGBE);
1365 	free(sc->sge.ctrlq, M_CXGBE);
1366 	free(sc->sge.iqmap, M_CXGBE);
1367 	free(sc->sge.eqmap, M_CXGBE);
1368 	free(sc->tids.ftid_tab, M_CXGBE);
1369 	t4_destroy_dma_tag(sc);
1370 	if (mtx_initialized(&sc->sc_lock)) {
1371 		sx_xlock(&t4_list_lock);
1372 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1373 		sx_xunlock(&t4_list_lock);
1374 		mtx_destroy(&sc->sc_lock);
1375 	}
1376 
1377 	callout_drain(&sc->sfl_callout);
1378 	if (mtx_initialized(&sc->tids.ftid_lock))
1379 		mtx_destroy(&sc->tids.ftid_lock);
1380 	if (mtx_initialized(&sc->sfl_lock))
1381 		mtx_destroy(&sc->sfl_lock);
1382 	if (mtx_initialized(&sc->ifp_lock))
1383 		mtx_destroy(&sc->ifp_lock);
1384 	if (mtx_initialized(&sc->reg_lock))
1385 		mtx_destroy(&sc->reg_lock);
1386 
1387 	for (i = 0; i < NUM_MEMWIN; i++) {
1388 		struct memwin *mw = &sc->memwin[i];
1389 
1390 		if (rw_initialized(&mw->mw_lock))
1391 			rw_destroy(&mw->mw_lock);
1392 	}
1393 
1394 	bzero(sc, sizeof(*sc));
1395 
1396 	return (0);
1397 }
1398 
1399 static int
1400 cxgbe_probe(device_t dev)
1401 {
1402 	char buf[128];
1403 	struct port_info *pi = device_get_softc(dev);
1404 
1405 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1406 	device_set_desc_copy(dev, buf);
1407 
1408 	return (BUS_PROBE_DEFAULT);
1409 }
1410 
1411 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1412     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1413     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1414 #define T4_CAP_ENABLE (T4_CAP)
1415 
1416 static int
1417 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1418 {
1419 	struct ifnet *ifp;
1420 	struct sbuf *sb;
1421 
1422 	vi->xact_addr_filt = -1;
1423 	callout_init(&vi->tick, 1);
1424 
1425 	/* Allocate an ifnet and set it up */
1426 	ifp = if_alloc(IFT_ETHER);
1427 	if (ifp == NULL) {
1428 		device_printf(dev, "Cannot allocate ifnet\n");
1429 		return (ENOMEM);
1430 	}
1431 	vi->ifp = ifp;
1432 	ifp->if_softc = vi;
1433 
1434 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1435 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1436 
1437 	ifp->if_init = cxgbe_init;
1438 	ifp->if_ioctl = cxgbe_ioctl;
1439 	ifp->if_transmit = cxgbe_transmit;
1440 	ifp->if_qflush = cxgbe_qflush;
1441 	ifp->if_get_counter = cxgbe_get_counter;
1442 
1443 	ifp->if_capabilities = T4_CAP;
1444 #ifdef TCP_OFFLOAD
1445 	if (vi->nofldrxq != 0)
1446 		ifp->if_capabilities |= IFCAP_TOE;
1447 #endif
1448 #ifdef DEV_NETMAP
1449 	if (vi->nnmrxq != 0)
1450 		ifp->if_capabilities |= IFCAP_NETMAP;
1451 #endif
1452 	ifp->if_capenable = T4_CAP_ENABLE;
1453 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1454 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1455 
1456 	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1457 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1458 	ifp->if_hw_tsomaxsegsize = 65536;
1459 
1460 	/* Initialize ifmedia for this VI */
1461 	ifmedia_init(&vi->media, IFM_IMASK, cxgbe_media_change,
1462 	    cxgbe_media_status);
1463 
1464 	vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1465 	    EVENTHANDLER_PRI_ANY);
1466 
1467 	ether_ifattach(ifp, vi->hw_addr);
1468 #ifdef DEV_NETMAP
1469 	if (ifp->if_capabilities & IFCAP_NETMAP)
1470 		cxgbe_nm_attach(vi);
1471 #endif
1472 	sb = sbuf_new_auto();
1473 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1474 #ifdef TCP_OFFLOAD
1475 	if (ifp->if_capabilities & IFCAP_TOE)
1476 		sbuf_printf(sb, "; %d txq, %d rxq (TOE)",
1477 		    vi->nofldtxq, vi->nofldrxq);
1478 #endif
1479 #ifdef DEV_NETMAP
1480 	if (ifp->if_capabilities & IFCAP_NETMAP)
1481 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1482 		    vi->nnmtxq, vi->nnmrxq);
1483 #endif
1484 	sbuf_finish(sb);
1485 	device_printf(dev, "%s\n", sbuf_data(sb));
1486 	sbuf_delete(sb);
1487 
1488 	vi_sysctls(vi);
1489 
1490 	return (0);
1491 }
1492 
1493 static int
1494 cxgbe_attach(device_t dev)
1495 {
1496 	struct port_info *pi = device_get_softc(dev);
1497 	struct adapter *sc = pi->adapter;
1498 	struct vi_info *vi;
1499 	int i, rc;
1500 
1501 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1502 
1503 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1504 	if (rc)
1505 		return (rc);
1506 
1507 	for_each_vi(pi, i, vi) {
1508 		if (i == 0)
1509 			continue;
1510 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1511 		if (vi->dev == NULL) {
1512 			device_printf(dev, "failed to add VI %d\n", i);
1513 			continue;
1514 		}
1515 		device_set_softc(vi->dev, vi);
1516 	}
1517 
1518 	cxgbe_sysctls(pi);
1519 
1520 	bus_generic_attach(dev);
1521 
1522 	return (0);
1523 }
1524 
1525 static void
1526 cxgbe_vi_detach(struct vi_info *vi)
1527 {
1528 	struct ifnet *ifp = vi->ifp;
1529 
1530 	ether_ifdetach(ifp);
1531 
1532 	if (vi->vlan_c)
1533 		EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1534 
1535 	/* Let detach proceed even if these fail. */
1536 #ifdef DEV_NETMAP
1537 	if (ifp->if_capabilities & IFCAP_NETMAP)
1538 		cxgbe_nm_detach(vi);
1539 #endif
1540 	cxgbe_uninit_synchronized(vi);
1541 	callout_drain(&vi->tick);
1542 	vi_full_uninit(vi);
1543 
1544 	ifmedia_removeall(&vi->media);
1545 	if_free(vi->ifp);
1546 	vi->ifp = NULL;
1547 }
1548 
1549 static int
1550 cxgbe_detach(device_t dev)
1551 {
1552 	struct port_info *pi = device_get_softc(dev);
1553 	struct adapter *sc = pi->adapter;
1554 	int rc;
1555 
1556 	/* Detach the extra VIs first. */
1557 	rc = bus_generic_detach(dev);
1558 	if (rc)
1559 		return (rc);
1560 	device_delete_children(dev);
1561 
1562 	doom_vi(sc, &pi->vi[0]);
1563 
1564 	if (pi->flags & HAS_TRACEQ) {
1565 		sc->traceq = -1;	/* cloner should not create ifnet */
1566 		t4_tracer_port_detach(sc);
1567 	}
1568 
1569 	cxgbe_vi_detach(&pi->vi[0]);
1570 	callout_drain(&pi->tick);
1571 
1572 	end_synchronized_op(sc, 0);
1573 
1574 	return (0);
1575 }
1576 
1577 static void
1578 cxgbe_init(void *arg)
1579 {
1580 	struct vi_info *vi = arg;
1581 	struct adapter *sc = vi->pi->adapter;
1582 
1583 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1584 		return;
1585 	cxgbe_init_synchronized(vi);
1586 	end_synchronized_op(sc, 0);
1587 }
1588 
1589 static int
1590 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1591 {
1592 	int rc = 0, mtu, flags, can_sleep;
1593 	struct vi_info *vi = ifp->if_softc;
1594 	struct adapter *sc = vi->pi->adapter;
1595 	struct ifreq *ifr = (struct ifreq *)data;
1596 	uint32_t mask;
1597 
1598 	switch (cmd) {
1599 	case SIOCSIFMTU:
1600 		mtu = ifr->ifr_mtu;
1601 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1602 			return (EINVAL);
1603 
1604 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1605 		if (rc)
1606 			return (rc);
1607 		ifp->if_mtu = mtu;
1608 		if (vi->flags & VI_INIT_DONE) {
1609 			t4_update_fl_bufsize(ifp);
1610 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1611 				rc = update_mac_settings(ifp, XGMAC_MTU);
1612 		}
1613 		end_synchronized_op(sc, 0);
1614 		break;
1615 
1616 	case SIOCSIFFLAGS:
1617 		can_sleep = 0;
1618 redo_sifflags:
1619 		rc = begin_synchronized_op(sc, vi,
1620 		    can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1621 		if (rc)
1622 			return (rc);
1623 
1624 		if (ifp->if_flags & IFF_UP) {
1625 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1626 				flags = vi->if_flags;
1627 				if ((ifp->if_flags ^ flags) &
1628 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1629 					if (can_sleep == 1) {
1630 						end_synchronized_op(sc, 0);
1631 						can_sleep = 0;
1632 						goto redo_sifflags;
1633 					}
1634 					rc = update_mac_settings(ifp,
1635 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1636 				}
1637 			} else {
1638 				if (can_sleep == 0) {
1639 					end_synchronized_op(sc, LOCK_HELD);
1640 					can_sleep = 1;
1641 					goto redo_sifflags;
1642 				}
1643 				rc = cxgbe_init_synchronized(vi);
1644 			}
1645 			vi->if_flags = ifp->if_flags;
1646 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1647 			if (can_sleep == 0) {
1648 				end_synchronized_op(sc, LOCK_HELD);
1649 				can_sleep = 1;
1650 				goto redo_sifflags;
1651 			}
1652 			rc = cxgbe_uninit_synchronized(vi);
1653 		}
1654 		end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1655 		break;
1656 
1657 	case SIOCADDMULTI:
1658 	case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1659 		rc = begin_synchronized_op(sc, vi, HOLD_LOCK, "t4multi");
1660 		if (rc)
1661 			return (rc);
1662 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1663 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1664 		end_synchronized_op(sc, LOCK_HELD);
1665 		break;
1666 
1667 	case SIOCSIFCAP:
1668 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1669 		if (rc)
1670 			return (rc);
1671 
1672 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1673 		if (mask & IFCAP_TXCSUM) {
1674 			ifp->if_capenable ^= IFCAP_TXCSUM;
1675 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1676 
1677 			if (IFCAP_TSO4 & ifp->if_capenable &&
1678 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1679 				ifp->if_capenable &= ~IFCAP_TSO4;
1680 				if_printf(ifp,
1681 				    "tso4 disabled due to -txcsum.\n");
1682 			}
1683 		}
1684 		if (mask & IFCAP_TXCSUM_IPV6) {
1685 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1686 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1687 
1688 			if (IFCAP_TSO6 & ifp->if_capenable &&
1689 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1690 				ifp->if_capenable &= ~IFCAP_TSO6;
1691 				if_printf(ifp,
1692 				    "tso6 disabled due to -txcsum6.\n");
1693 			}
1694 		}
1695 		if (mask & IFCAP_RXCSUM)
1696 			ifp->if_capenable ^= IFCAP_RXCSUM;
1697 		if (mask & IFCAP_RXCSUM_IPV6)
1698 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1699 
1700 		/*
1701 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1702 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1703 		 * sending a TSO request our way, so it's sufficient to toggle
1704 		 * IFCAP_TSOx only.
1705 		 */
1706 		if (mask & IFCAP_TSO4) {
1707 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1708 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1709 				if_printf(ifp, "enable txcsum first.\n");
1710 				rc = EAGAIN;
1711 				goto fail;
1712 			}
1713 			ifp->if_capenable ^= IFCAP_TSO4;
1714 		}
1715 		if (mask & IFCAP_TSO6) {
1716 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1717 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1718 				if_printf(ifp, "enable txcsum6 first.\n");
1719 				rc = EAGAIN;
1720 				goto fail;
1721 			}
1722 			ifp->if_capenable ^= IFCAP_TSO6;
1723 		}
1724 		if (mask & IFCAP_LRO) {
1725 #if defined(INET) || defined(INET6)
1726 			int i;
1727 			struct sge_rxq *rxq;
1728 
1729 			ifp->if_capenable ^= IFCAP_LRO;
1730 			for_each_rxq(vi, i, rxq) {
1731 				if (ifp->if_capenable & IFCAP_LRO)
1732 					rxq->iq.flags |= IQ_LRO_ENABLED;
1733 				else
1734 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1735 			}
1736 #endif
1737 		}
1738 #ifdef TCP_OFFLOAD
1739 		if (mask & IFCAP_TOE) {
1740 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1741 
1742 			rc = toe_capability(vi, enable);
1743 			if (rc != 0)
1744 				goto fail;
1745 
1746 			ifp->if_capenable ^= mask;
1747 		}
1748 #endif
1749 		if (mask & IFCAP_VLAN_HWTAGGING) {
1750 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1751 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1752 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1753 		}
1754 		if (mask & IFCAP_VLAN_MTU) {
1755 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1756 
1757 			/* Need to find out how to disable auto-mtu-inflation */
1758 		}
1759 		if (mask & IFCAP_VLAN_HWTSO)
1760 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1761 		if (mask & IFCAP_VLAN_HWCSUM)
1762 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1763 
1764 #ifdef VLAN_CAPABILITIES
1765 		VLAN_CAPABILITIES(ifp);
1766 #endif
1767 fail:
1768 		end_synchronized_op(sc, 0);
1769 		break;
1770 
1771 	case SIOCSIFMEDIA:
1772 	case SIOCGIFMEDIA:
1773 	case SIOCGIFXMEDIA:
1774 		ifmedia_ioctl(ifp, ifr, &vi->media, cmd);
1775 		break;
1776 
1777 	case SIOCGI2C: {
1778 		struct ifi2creq i2c;
1779 
1780 		rc = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
1781 		if (rc != 0)
1782 			break;
1783 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1784 			rc = EPERM;
1785 			break;
1786 		}
1787 		if (i2c.len > sizeof(i2c.data)) {
1788 			rc = EINVAL;
1789 			break;
1790 		}
1791 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1792 		if (rc)
1793 			return (rc);
1794 		rc = -t4_i2c_rd(sc, sc->mbox, vi->pi->port_id, i2c.dev_addr,
1795 		    i2c.offset, i2c.len, &i2c.data[0]);
1796 		end_synchronized_op(sc, 0);
1797 		if (rc == 0)
1798 			rc = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
1799 		break;
1800 	}
1801 
1802 	default:
1803 		rc = ether_ioctl(ifp, cmd, data);
1804 	}
1805 
1806 	return (rc);
1807 }
1808 
1809 static int
1810 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1811 {
1812 	struct vi_info *vi = ifp->if_softc;
1813 	struct port_info *pi = vi->pi;
1814 	struct adapter *sc = pi->adapter;
1815 	struct sge_txq *txq;
1816 	void *items[1];
1817 	int rc;
1818 
1819 	M_ASSERTPKTHDR(m);
1820 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1821 
1822 	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1823 		m_freem(m);
1824 		return (ENETDOWN);
1825 	}
1826 
1827 	rc = parse_pkt(sc, &m);
1828 	if (__predict_false(rc != 0)) {
1829 		MPASS(m == NULL);			/* was freed already */
1830 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1831 		return (rc);
1832 	}
1833 
1834 	/* Select a txq. */
1835 	txq = &sc->sge.txq[vi->first_txq];
1836 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1837 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1838 		    vi->rsrv_noflowq);
1839 
1840 	items[0] = m;
1841 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1842 	if (__predict_false(rc != 0))
1843 		m_freem(m);
1844 
1845 	return (rc);
1846 }
1847 
1848 static void
1849 cxgbe_qflush(struct ifnet *ifp)
1850 {
1851 	struct vi_info *vi = ifp->if_softc;
1852 	struct sge_txq *txq;
1853 	int i;
1854 
1855 	/* queues do not exist if !VI_INIT_DONE. */
1856 	if (vi->flags & VI_INIT_DONE) {
1857 		for_each_txq(vi, i, txq) {
1858 			TXQ_LOCK(txq);
1859 			txq->eq.flags |= EQ_QFLUSH;
1860 			TXQ_UNLOCK(txq);
1861 			while (!mp_ring_is_idle(txq->r)) {
1862 				mp_ring_check_drainage(txq->r, 0);
1863 				pause("qflush", 1);
1864 			}
1865 			TXQ_LOCK(txq);
1866 			txq->eq.flags &= ~EQ_QFLUSH;
1867 			TXQ_UNLOCK(txq);
1868 		}
1869 	}
1870 	if_qflush(ifp);
1871 }
1872 
1873 static uint64_t
1874 vi_get_counter(struct ifnet *ifp, ift_counter c)
1875 {
1876 	struct vi_info *vi = ifp->if_softc;
1877 	struct fw_vi_stats_vf *s = &vi->stats;
1878 
1879 	vi_refresh_stats(vi->pi->adapter, vi);
1880 
1881 	switch (c) {
1882 	case IFCOUNTER_IPACKETS:
1883 		return (s->rx_bcast_frames + s->rx_mcast_frames +
1884 		    s->rx_ucast_frames);
1885 	case IFCOUNTER_IERRORS:
1886 		return (s->rx_err_frames);
1887 	case IFCOUNTER_OPACKETS:
1888 		return (s->tx_bcast_frames + s->tx_mcast_frames +
1889 		    s->tx_ucast_frames + s->tx_offload_frames);
1890 	case IFCOUNTER_OERRORS:
1891 		return (s->tx_drop_frames);
1892 	case IFCOUNTER_IBYTES:
1893 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
1894 		    s->rx_ucast_bytes);
1895 	case IFCOUNTER_OBYTES:
1896 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
1897 		    s->tx_ucast_bytes + s->tx_offload_bytes);
1898 	case IFCOUNTER_IMCASTS:
1899 		return (s->rx_mcast_frames);
1900 	case IFCOUNTER_OMCASTS:
1901 		return (s->tx_mcast_frames);
1902 	case IFCOUNTER_OQDROPS: {
1903 		uint64_t drops;
1904 
1905 		drops = 0;
1906 		if (vi->flags & VI_INIT_DONE) {
1907 			int i;
1908 			struct sge_txq *txq;
1909 
1910 			for_each_txq(vi, i, txq)
1911 				drops += counter_u64_fetch(txq->r->drops);
1912 		}
1913 
1914 		return (drops);
1915 
1916 	}
1917 
1918 	default:
1919 		return (if_get_counter_default(ifp, c));
1920 	}
1921 }
1922 
1923 uint64_t
1924 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
1925 {
1926 	struct vi_info *vi = ifp->if_softc;
1927 	struct port_info *pi = vi->pi;
1928 	struct adapter *sc = pi->adapter;
1929 	struct port_stats *s = &pi->stats;
1930 
1931 	if (pi->nvi > 1 || sc->flags & IS_VF)
1932 		return (vi_get_counter(ifp, c));
1933 
1934 	cxgbe_refresh_stats(sc, pi);
1935 
1936 	switch (c) {
1937 	case IFCOUNTER_IPACKETS:
1938 		return (s->rx_frames);
1939 
1940 	case IFCOUNTER_IERRORS:
1941 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
1942 		    s->rx_fcs_err + s->rx_len_err);
1943 
1944 	case IFCOUNTER_OPACKETS:
1945 		return (s->tx_frames);
1946 
1947 	case IFCOUNTER_OERRORS:
1948 		return (s->tx_error_frames);
1949 
1950 	case IFCOUNTER_IBYTES:
1951 		return (s->rx_octets);
1952 
1953 	case IFCOUNTER_OBYTES:
1954 		return (s->tx_octets);
1955 
1956 	case IFCOUNTER_IMCASTS:
1957 		return (s->rx_mcast_frames);
1958 
1959 	case IFCOUNTER_OMCASTS:
1960 		return (s->tx_mcast_frames);
1961 
1962 	case IFCOUNTER_IQDROPS:
1963 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
1964 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
1965 		    s->rx_trunc3 + pi->tnl_cong_drops);
1966 
1967 	case IFCOUNTER_OQDROPS: {
1968 		uint64_t drops;
1969 
1970 		drops = s->tx_drop;
1971 		if (vi->flags & VI_INIT_DONE) {
1972 			int i;
1973 			struct sge_txq *txq;
1974 
1975 			for_each_txq(vi, i, txq)
1976 				drops += counter_u64_fetch(txq->r->drops);
1977 		}
1978 
1979 		return (drops);
1980 
1981 	}
1982 
1983 	default:
1984 		return (if_get_counter_default(ifp, c));
1985 	}
1986 }
1987 
1988 static int
1989 cxgbe_media_change(struct ifnet *ifp)
1990 {
1991 	struct vi_info *vi = ifp->if_softc;
1992 
1993 	device_printf(vi->dev, "%s unimplemented.\n", __func__);
1994 
1995 	return (EOPNOTSUPP);
1996 }
1997 
1998 static void
1999 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2000 {
2001 	struct vi_info *vi = ifp->if_softc;
2002 	struct port_info *pi = vi->pi;
2003 	struct ifmedia_entry *cur;
2004 	struct link_config *lc = &pi->link_cfg;
2005 
2006 	/*
2007 	 * If all the interfaces are administratively down the firmware does not
2008 	 * report transceiver changes.  Refresh port info here so that ifconfig
2009 	 * displays accurate information at all times.
2010 	 */
2011 	if (begin_synchronized_op(pi->adapter, NULL, SLEEP_OK | INTR_OK,
2012 	    "t4med") == 0) {
2013 		PORT_LOCK(pi);
2014 		if (pi->up_vis == 0) {
2015 			t4_update_port_info(pi);
2016 			build_medialist(pi, &vi->media);
2017 		}
2018 		PORT_UNLOCK(pi);
2019 		end_synchronized_op(pi->adapter, 0);
2020 	}
2021 
2022 	cur = vi->media.ifm_cur;
2023 
2024 	ifmr->ifm_status = IFM_AVALID;
2025 	if (lc->link_ok == 0)
2026 		return;
2027 
2028 	ifmr->ifm_status |= IFM_ACTIVE;
2029 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2030 	if (lc->fc & PAUSE_RX)
2031 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2032 	if (lc->fc & PAUSE_TX)
2033 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2034 
2035 	/* active and current will differ iff current media is autoselect. */
2036 	if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
2037 		return;
2038 
2039 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2040 	if (lc->fc & PAUSE_RX)
2041 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2042 	if (lc->fc & PAUSE_TX)
2043 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2044 	switch (lc->speed) {
2045 	case 10000:
2046 		ifmr->ifm_active |= IFM_10G_T;
2047 		break;
2048 	case 1000:
2049 		ifmr->ifm_active |= IFM_1000_T;
2050 		break;
2051 	case 100:
2052 		ifmr->ifm_active |= IFM_100_TX;
2053 		break;
2054 	case 10:
2055 		ifmr->ifm_active |= IFM_10_T;
2056 		break;
2057 	default:
2058 		device_printf(vi->dev, "link up but speed unknown (%u)\n",
2059 		    lc->speed);
2060 	}
2061 }
2062 
2063 static int
2064 vcxgbe_probe(device_t dev)
2065 {
2066 	char buf[128];
2067 	struct vi_info *vi = device_get_softc(dev);
2068 
2069 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2070 	    vi - vi->pi->vi);
2071 	device_set_desc_copy(dev, buf);
2072 
2073 	return (BUS_PROBE_DEFAULT);
2074 }
2075 
2076 static int
2077 vcxgbe_attach(device_t dev)
2078 {
2079 	struct vi_info *vi;
2080 	struct port_info *pi;
2081 	struct adapter *sc;
2082 	int func, index, rc;
2083 	u32 param, val;
2084 
2085 	vi = device_get_softc(dev);
2086 	pi = vi->pi;
2087 	sc = pi->adapter;
2088 
2089 	index = vi - pi->vi;
2090 	KASSERT(index < nitems(vi_mac_funcs),
2091 	    ("%s: VI %s doesn't have a MAC func", __func__,
2092 	    device_get_nameunit(dev)));
2093 	func = vi_mac_funcs[index];
2094 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2095 	    vi->hw_addr, &vi->rss_size, func, 0);
2096 	if (rc < 0) {
2097 		device_printf(dev, "Failed to allocate virtual interface "
2098 		    "for port %d: %d\n", pi->port_id, -rc);
2099 		return (-rc);
2100 	}
2101 	vi->viid = rc;
2102 	if (chip_id(sc) <= CHELSIO_T5)
2103 		vi->smt_idx = (rc & 0x7f) << 1;
2104 	else
2105 		vi->smt_idx = (rc & 0x7f);
2106 
2107 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2108 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2109 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2110 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2111 	if (rc)
2112 		vi->rss_base = 0xffff;
2113 	else {
2114 		/* MPASS((val >> 16) == rss_size); */
2115 		vi->rss_base = val & 0xffff;
2116 	}
2117 
2118 	rc = cxgbe_vi_attach(dev, vi);
2119 	if (rc) {
2120 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2121 		return (rc);
2122 	}
2123 	return (0);
2124 }
2125 
2126 static int
2127 vcxgbe_detach(device_t dev)
2128 {
2129 	struct vi_info *vi;
2130 	struct adapter *sc;
2131 
2132 	vi = device_get_softc(dev);
2133 	sc = vi->pi->adapter;
2134 
2135 	doom_vi(sc, vi);
2136 
2137 	cxgbe_vi_detach(vi);
2138 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2139 
2140 	end_synchronized_op(sc, 0);
2141 
2142 	return (0);
2143 }
2144 
2145 void
2146 t4_fatal_err(struct adapter *sc)
2147 {
2148 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2149 	t4_intr_disable(sc);
2150 	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2151 	    device_get_nameunit(sc->dev));
2152 }
2153 
2154 void
2155 t4_add_adapter(struct adapter *sc)
2156 {
2157 	sx_xlock(&t4_list_lock);
2158 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2159 	sx_xunlock(&t4_list_lock);
2160 }
2161 
2162 int
2163 t4_map_bars_0_and_4(struct adapter *sc)
2164 {
2165 	sc->regs_rid = PCIR_BAR(0);
2166 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2167 	    &sc->regs_rid, RF_ACTIVE);
2168 	if (sc->regs_res == NULL) {
2169 		device_printf(sc->dev, "cannot map registers.\n");
2170 		return (ENXIO);
2171 	}
2172 	sc->bt = rman_get_bustag(sc->regs_res);
2173 	sc->bh = rman_get_bushandle(sc->regs_res);
2174 	sc->mmio_len = rman_get_size(sc->regs_res);
2175 	setbit(&sc->doorbells, DOORBELL_KDB);
2176 
2177 	sc->msix_rid = PCIR_BAR(4);
2178 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2179 	    &sc->msix_rid, RF_ACTIVE);
2180 	if (sc->msix_res == NULL) {
2181 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2182 		return (ENXIO);
2183 	}
2184 
2185 	return (0);
2186 }
2187 
2188 int
2189 t4_map_bar_2(struct adapter *sc)
2190 {
2191 
2192 	/*
2193 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2194 	 * to map it if RDMA is disabled.
2195 	 */
2196 	if (is_t4(sc) && sc->rdmacaps == 0)
2197 		return (0);
2198 
2199 	sc->udbs_rid = PCIR_BAR(2);
2200 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2201 	    &sc->udbs_rid, RF_ACTIVE);
2202 	if (sc->udbs_res == NULL) {
2203 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2204 		return (ENXIO);
2205 	}
2206 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2207 
2208 	if (chip_id(sc) >= CHELSIO_T5) {
2209 		setbit(&sc->doorbells, DOORBELL_UDB);
2210 #if defined(__i386__) || defined(__amd64__)
2211 		if (t5_write_combine) {
2212 			int rc, mode;
2213 
2214 			/*
2215 			 * Enable write combining on BAR2.  This is the
2216 			 * userspace doorbell BAR and is split into 128B
2217 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2218 			 * with an egress queue.  The first 64B has the doorbell
2219 			 * and the second 64B can be used to submit a tx work
2220 			 * request with an implicit doorbell.
2221 			 */
2222 
2223 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2224 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2225 			if (rc == 0) {
2226 				clrbit(&sc->doorbells, DOORBELL_UDB);
2227 				setbit(&sc->doorbells, DOORBELL_WCWR);
2228 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2229 			} else {
2230 				device_printf(sc->dev,
2231 				    "couldn't enable write combining: %d\n",
2232 				    rc);
2233 			}
2234 
2235 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2236 			t4_write_reg(sc, A_SGE_STAT_CFG,
2237 			    V_STATSOURCE_T5(7) | mode);
2238 		}
2239 #endif
2240 	}
2241 
2242 	return (0);
2243 }
2244 
2245 struct memwin_init {
2246 	uint32_t base;
2247 	uint32_t aperture;
2248 };
2249 
2250 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2251 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2252 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2253 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2254 };
2255 
2256 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2257 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2258 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2259 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2260 };
2261 
2262 static void
2263 setup_memwin(struct adapter *sc)
2264 {
2265 	const struct memwin_init *mw_init;
2266 	struct memwin *mw;
2267 	int i;
2268 	uint32_t bar0;
2269 
2270 	if (is_t4(sc)) {
2271 		/*
2272 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2273 		 * mechanism.  Works from within PCI passthrough environments
2274 		 * too, where rman_get_start() can return a different value.  We
2275 		 * need to program the T4 memory window decoders with the actual
2276 		 * addresses that will be coming across the PCIe link.
2277 		 */
2278 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2279 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2280 
2281 		mw_init = &t4_memwin[0];
2282 	} else {
2283 		/* T5+ use the relative offset inside the PCIe BAR */
2284 		bar0 = 0;
2285 
2286 		mw_init = &t5_memwin[0];
2287 	}
2288 
2289 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2290 		rw_init(&mw->mw_lock, "memory window access");
2291 		mw->mw_base = mw_init->base;
2292 		mw->mw_aperture = mw_init->aperture;
2293 		mw->mw_curpos = 0;
2294 		t4_write_reg(sc,
2295 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2296 		    (mw->mw_base + bar0) | V_BIR(0) |
2297 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2298 		rw_wlock(&mw->mw_lock);
2299 		position_memwin(sc, i, 0);
2300 		rw_wunlock(&mw->mw_lock);
2301 	}
2302 
2303 	/* flush */
2304 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2305 }
2306 
2307 /*
2308  * Positions the memory window at the given address in the card's address space.
2309  * There are some alignment requirements and the actual position may be at an
2310  * address prior to the requested address.  mw->mw_curpos always has the actual
2311  * position of the window.
2312  */
2313 static void
2314 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2315 {
2316 	struct memwin *mw;
2317 	uint32_t pf;
2318 	uint32_t reg;
2319 
2320 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2321 	mw = &sc->memwin[idx];
2322 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2323 
2324 	if (is_t4(sc)) {
2325 		pf = 0;
2326 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2327 	} else {
2328 		pf = V_PFNUM(sc->pf);
2329 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2330 	}
2331 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2332 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2333 	t4_read_reg(sc, reg);	/* flush */
2334 }
2335 
2336 static int
2337 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2338     int len, int rw)
2339 {
2340 	struct memwin *mw;
2341 	uint32_t mw_end, v;
2342 
2343 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2344 
2345 	/* Memory can only be accessed in naturally aligned 4 byte units */
2346 	if (addr & 3 || len & 3 || len <= 0)
2347 		return (EINVAL);
2348 
2349 	mw = &sc->memwin[idx];
2350 	while (len > 0) {
2351 		rw_rlock(&mw->mw_lock);
2352 		mw_end = mw->mw_curpos + mw->mw_aperture;
2353 		if (addr >= mw_end || addr < mw->mw_curpos) {
2354 			/* Will need to reposition the window */
2355 			if (!rw_try_upgrade(&mw->mw_lock)) {
2356 				rw_runlock(&mw->mw_lock);
2357 				rw_wlock(&mw->mw_lock);
2358 			}
2359 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2360 			position_memwin(sc, idx, addr);
2361 			rw_downgrade(&mw->mw_lock);
2362 			mw_end = mw->mw_curpos + mw->mw_aperture;
2363 		}
2364 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2365 		while (addr < mw_end && len > 0) {
2366 			if (rw == 0) {
2367 				v = t4_read_reg(sc, mw->mw_base + addr -
2368 				    mw->mw_curpos);
2369 				*val++ = le32toh(v);
2370 			} else {
2371 				v = *val++;
2372 				t4_write_reg(sc, mw->mw_base + addr -
2373 				    mw->mw_curpos, htole32(v));
2374 			}
2375 			addr += 4;
2376 			len -= 4;
2377 		}
2378 		rw_runlock(&mw->mw_lock);
2379 	}
2380 
2381 	return (0);
2382 }
2383 
2384 static inline int
2385 read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2386     int len)
2387 {
2388 
2389 	return (rw_via_memwin(sc, idx, addr, val, len, 0));
2390 }
2391 
2392 static inline int
2393 write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
2394     const uint32_t *val, int len)
2395 {
2396 
2397 	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
2398 }
2399 
2400 static int
2401 t4_range_cmp(const void *a, const void *b)
2402 {
2403 	return ((const struct t4_range *)a)->start -
2404 	       ((const struct t4_range *)b)->start;
2405 }
2406 
2407 /*
2408  * Verify that the memory range specified by the addr/len pair is valid within
2409  * the card's address space.
2410  */
2411 static int
2412 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2413 {
2414 	struct t4_range mem_ranges[4], *r, *next;
2415 	uint32_t em, addr_len;
2416 	int i, n, remaining;
2417 
2418 	/* Memory can only be accessed in naturally aligned 4 byte units */
2419 	if (addr & 3 || len & 3 || len <= 0)
2420 		return (EINVAL);
2421 
2422 	/* Enabled memories */
2423 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2424 
2425 	r = &mem_ranges[0];
2426 	n = 0;
2427 	bzero(r, sizeof(mem_ranges));
2428 	if (em & F_EDRAM0_ENABLE) {
2429 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2430 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2431 		if (r->size > 0) {
2432 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2433 			if (addr >= r->start &&
2434 			    addr + len <= r->start + r->size)
2435 				return (0);
2436 			r++;
2437 			n++;
2438 		}
2439 	}
2440 	if (em & F_EDRAM1_ENABLE) {
2441 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2442 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2443 		if (r->size > 0) {
2444 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2445 			if (addr >= r->start &&
2446 			    addr + len <= r->start + r->size)
2447 				return (0);
2448 			r++;
2449 			n++;
2450 		}
2451 	}
2452 	if (em & F_EXT_MEM_ENABLE) {
2453 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2454 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2455 		if (r->size > 0) {
2456 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2457 			if (addr >= r->start &&
2458 			    addr + len <= r->start + r->size)
2459 				return (0);
2460 			r++;
2461 			n++;
2462 		}
2463 	}
2464 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2465 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2466 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2467 		if (r->size > 0) {
2468 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2469 			if (addr >= r->start &&
2470 			    addr + len <= r->start + r->size)
2471 				return (0);
2472 			r++;
2473 			n++;
2474 		}
2475 	}
2476 	MPASS(n <= nitems(mem_ranges));
2477 
2478 	if (n > 1) {
2479 		/* Sort and merge the ranges. */
2480 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2481 
2482 		/* Start from index 0 and examine the next n - 1 entries. */
2483 		r = &mem_ranges[0];
2484 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2485 
2486 			MPASS(r->size > 0);	/* r is a valid entry. */
2487 			next = r + 1;
2488 			MPASS(next->size > 0);	/* and so is the next one. */
2489 
2490 			while (r->start + r->size >= next->start) {
2491 				/* Merge the next one into the current entry. */
2492 				r->size = max(r->start + r->size,
2493 				    next->start + next->size) - r->start;
2494 				n--;	/* One fewer entry in total. */
2495 				if (--remaining == 0)
2496 					goto done;	/* short circuit */
2497 				next++;
2498 			}
2499 			if (next != r + 1) {
2500 				/*
2501 				 * Some entries were merged into r and next
2502 				 * points to the first valid entry that couldn't
2503 				 * be merged.
2504 				 */
2505 				MPASS(next->size > 0);	/* must be valid */
2506 				memcpy(r + 1, next, remaining * sizeof(*r));
2507 #ifdef INVARIANTS
2508 				/*
2509 				 * This so that the foo->size assertion in the
2510 				 * next iteration of the loop do the right
2511 				 * thing for entries that were pulled up and are
2512 				 * no longer valid.
2513 				 */
2514 				MPASS(n < nitems(mem_ranges));
2515 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2516 				    sizeof(struct t4_range));
2517 #endif
2518 			}
2519 		}
2520 done:
2521 		/* Done merging the ranges. */
2522 		MPASS(n > 0);
2523 		r = &mem_ranges[0];
2524 		for (i = 0; i < n; i++, r++) {
2525 			if (addr >= r->start &&
2526 			    addr + len <= r->start + r->size)
2527 				return (0);
2528 		}
2529 	}
2530 
2531 	return (EFAULT);
2532 }
2533 
2534 static int
2535 fwmtype_to_hwmtype(int mtype)
2536 {
2537 
2538 	switch (mtype) {
2539 	case FW_MEMTYPE_EDC0:
2540 		return (MEM_EDC0);
2541 	case FW_MEMTYPE_EDC1:
2542 		return (MEM_EDC1);
2543 	case FW_MEMTYPE_EXTMEM:
2544 		return (MEM_MC0);
2545 	case FW_MEMTYPE_EXTMEM1:
2546 		return (MEM_MC1);
2547 	default:
2548 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2549 	}
2550 }
2551 
2552 /*
2553  * Verify that the memory range specified by the memtype/offset/len pair is
2554  * valid and lies entirely within the memtype specified.  The global address of
2555  * the start of the range is returned in addr.
2556  */
2557 static int
2558 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2559     uint32_t *addr)
2560 {
2561 	uint32_t em, addr_len, maddr;
2562 
2563 	/* Memory can only be accessed in naturally aligned 4 byte units */
2564 	if (off & 3 || len & 3 || len == 0)
2565 		return (EINVAL);
2566 
2567 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2568 	switch (fwmtype_to_hwmtype(mtype)) {
2569 	case MEM_EDC0:
2570 		if (!(em & F_EDRAM0_ENABLE))
2571 			return (EINVAL);
2572 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2573 		maddr = G_EDRAM0_BASE(addr_len) << 20;
2574 		break;
2575 	case MEM_EDC1:
2576 		if (!(em & F_EDRAM1_ENABLE))
2577 			return (EINVAL);
2578 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2579 		maddr = G_EDRAM1_BASE(addr_len) << 20;
2580 		break;
2581 	case MEM_MC:
2582 		if (!(em & F_EXT_MEM_ENABLE))
2583 			return (EINVAL);
2584 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2585 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2586 		break;
2587 	case MEM_MC1:
2588 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2589 			return (EINVAL);
2590 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2591 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2592 		break;
2593 	default:
2594 		return (EINVAL);
2595 	}
2596 
2597 	*addr = maddr + off;	/* global address */
2598 	return (validate_mem_range(sc, *addr, len));
2599 }
2600 
2601 static int
2602 fixup_devlog_params(struct adapter *sc)
2603 {
2604 	struct devlog_params *dparams = &sc->params.devlog;
2605 	int rc;
2606 
2607 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2608 	    dparams->size, &dparams->addr);
2609 
2610 	return (rc);
2611 }
2612 
2613 static int
2614 cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g, int num_vis,
2615     struct intrs_and_queues *iaq)
2616 {
2617 	int rc, itype, navail, nrxq10g, nrxq1g, n;
2618 	int nofldrxq10g = 0, nofldrxq1g = 0;
2619 
2620 	bzero(iaq, sizeof(*iaq));
2621 
2622 	iaq->ntxq10g = t4_ntxq10g;
2623 	iaq->ntxq1g = t4_ntxq1g;
2624 	iaq->ntxq_vi = t4_ntxq_vi;
2625 	iaq->nrxq10g = nrxq10g = t4_nrxq10g;
2626 	iaq->nrxq1g = nrxq1g = t4_nrxq1g;
2627 	iaq->nrxq_vi = t4_nrxq_vi;
2628 	iaq->rsrv_noflowq = t4_rsrv_noflowq;
2629 #ifdef TCP_OFFLOAD
2630 	if (is_offload(sc)) {
2631 		iaq->nofldtxq10g = t4_nofldtxq10g;
2632 		iaq->nofldtxq1g = t4_nofldtxq1g;
2633 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
2634 		iaq->nofldrxq10g = nofldrxq10g = t4_nofldrxq10g;
2635 		iaq->nofldrxq1g = nofldrxq1g = t4_nofldrxq1g;
2636 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
2637 	}
2638 #endif
2639 #ifdef DEV_NETMAP
2640 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
2641 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
2642 #endif
2643 
2644 	for (itype = INTR_MSIX; itype; itype >>= 1) {
2645 
2646 		if ((itype & t4_intr_types) == 0)
2647 			continue;	/* not allowed */
2648 
2649 		if (itype == INTR_MSIX)
2650 			navail = pci_msix_count(sc->dev);
2651 		else if (itype == INTR_MSI)
2652 			navail = pci_msi_count(sc->dev);
2653 		else
2654 			navail = 1;
2655 restart:
2656 		if (navail == 0)
2657 			continue;
2658 
2659 		iaq->intr_type = itype;
2660 		iaq->intr_flags_10g = 0;
2661 		iaq->intr_flags_1g = 0;
2662 
2663 		/*
2664 		 * Best option: an interrupt vector for errors, one for the
2665 		 * firmware event queue, and one for every rxq (NIC and TOE) of
2666 		 * every VI.  The VIs that support netmap use the same
2667 		 * interrupts for the NIC rx queues and the netmap rx queues
2668 		 * because only one set of queues is active at a time.
2669 		 */
2670 		iaq->nirq = T4_EXTRA_INTR;
2671 		iaq->nirq += n10g * (nrxq10g + nofldrxq10g);
2672 		iaq->nirq += n1g * (nrxq1g + nofldrxq1g);
2673 		iaq->nirq += (n10g + n1g) * (num_vis - 1) *
2674 		    max(iaq->nrxq_vi, iaq->nnmrxq_vi);	/* See comment above. */
2675 		iaq->nirq += (n10g + n1g) * (num_vis - 1) * iaq->nofldrxq_vi;
2676 		if (iaq->nirq <= navail &&
2677 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2678 			iaq->intr_flags_10g = INTR_ALL;
2679 			iaq->intr_flags_1g = INTR_ALL;
2680 			goto allocate;
2681 		}
2682 
2683 		/* Disable the VIs (and netmap) if there aren't enough intrs */
2684 		if (num_vis > 1) {
2685 			device_printf(sc->dev, "virtual interfaces disabled "
2686 			    "because num_vis=%u with current settings "
2687 			    "(nrxq10g=%u, nrxq1g=%u, nofldrxq10g=%u, "
2688 			    "nofldrxq1g=%u, nrxq_vi=%u nofldrxq_vi=%u, "
2689 			    "nnmrxq_vi=%u) would need %u interrupts but "
2690 			    "only %u are available.\n", num_vis, nrxq10g,
2691 			    nrxq1g, nofldrxq10g, nofldrxq1g, iaq->nrxq_vi,
2692 			    iaq->nofldrxq_vi, iaq->nnmrxq_vi, iaq->nirq,
2693 			    navail);
2694 			num_vis = 1;
2695 			iaq->ntxq_vi = iaq->nrxq_vi = 0;
2696 			iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
2697 			iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
2698 			goto restart;
2699 		}
2700 
2701 		/*
2702 		 * Second best option: a vector for errors, one for the firmware
2703 		 * event queue, and vectors for either all the NIC rx queues or
2704 		 * all the TOE rx queues.  The queues that don't get vectors
2705 		 * will forward their interrupts to those that do.
2706 		 */
2707 		iaq->nirq = T4_EXTRA_INTR;
2708 		if (nrxq10g >= nofldrxq10g) {
2709 			iaq->intr_flags_10g = INTR_RXQ;
2710 			iaq->nirq += n10g * nrxq10g;
2711 		} else {
2712 			iaq->intr_flags_10g = INTR_OFLD_RXQ;
2713 			iaq->nirq += n10g * nofldrxq10g;
2714 		}
2715 		if (nrxq1g >= nofldrxq1g) {
2716 			iaq->intr_flags_1g = INTR_RXQ;
2717 			iaq->nirq += n1g * nrxq1g;
2718 		} else {
2719 			iaq->intr_flags_1g = INTR_OFLD_RXQ;
2720 			iaq->nirq += n1g * nofldrxq1g;
2721 		}
2722 		if (iaq->nirq <= navail &&
2723 		    (itype != INTR_MSI || powerof2(iaq->nirq)))
2724 			goto allocate;
2725 
2726 		/*
2727 		 * Next best option: an interrupt vector for errors, one for the
2728 		 * firmware event queue, and at least one per main-VI.  At this
2729 		 * point we know we'll have to downsize nrxq and/or nofldrxq to
2730 		 * fit what's available to us.
2731 		 */
2732 		iaq->nirq = T4_EXTRA_INTR;
2733 		iaq->nirq += n10g + n1g;
2734 		if (iaq->nirq <= navail) {
2735 			int leftover = navail - iaq->nirq;
2736 
2737 			if (n10g > 0) {
2738 				int target = max(nrxq10g, nofldrxq10g);
2739 
2740 				iaq->intr_flags_10g = nrxq10g >= nofldrxq10g ?
2741 				    INTR_RXQ : INTR_OFLD_RXQ;
2742 
2743 				n = 1;
2744 				while (n < target && leftover >= n10g) {
2745 					leftover -= n10g;
2746 					iaq->nirq += n10g;
2747 					n++;
2748 				}
2749 				iaq->nrxq10g = min(n, nrxq10g);
2750 #ifdef TCP_OFFLOAD
2751 				iaq->nofldrxq10g = min(n, nofldrxq10g);
2752 #endif
2753 			}
2754 
2755 			if (n1g > 0) {
2756 				int target = max(nrxq1g, nofldrxq1g);
2757 
2758 				iaq->intr_flags_1g = nrxq1g >= nofldrxq1g ?
2759 				    INTR_RXQ : INTR_OFLD_RXQ;
2760 
2761 				n = 1;
2762 				while (n < target && leftover >= n1g) {
2763 					leftover -= n1g;
2764 					iaq->nirq += n1g;
2765 					n++;
2766 				}
2767 				iaq->nrxq1g = min(n, nrxq1g);
2768 #ifdef TCP_OFFLOAD
2769 				iaq->nofldrxq1g = min(n, nofldrxq1g);
2770 #endif
2771 			}
2772 
2773 			if (itype != INTR_MSI || powerof2(iaq->nirq))
2774 				goto allocate;
2775 		}
2776 
2777 		/*
2778 		 * Least desirable option: one interrupt vector for everything.
2779 		 */
2780 		iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2781 		iaq->intr_flags_10g = iaq->intr_flags_1g = 0;
2782 #ifdef TCP_OFFLOAD
2783 		if (is_offload(sc))
2784 			iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2785 #endif
2786 allocate:
2787 		navail = iaq->nirq;
2788 		rc = 0;
2789 		if (itype == INTR_MSIX)
2790 			rc = pci_alloc_msix(sc->dev, &navail);
2791 		else if (itype == INTR_MSI)
2792 			rc = pci_alloc_msi(sc->dev, &navail);
2793 
2794 		if (rc == 0) {
2795 			if (navail == iaq->nirq)
2796 				return (0);
2797 
2798 			/*
2799 			 * Didn't get the number requested.  Use whatever number
2800 			 * the kernel is willing to allocate (it's in navail).
2801 			 */
2802 			device_printf(sc->dev, "fewer vectors than requested, "
2803 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2804 			    itype, iaq->nirq, navail);
2805 			pci_release_msi(sc->dev);
2806 			goto restart;
2807 		}
2808 
2809 		device_printf(sc->dev,
2810 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2811 		    itype, rc, iaq->nirq, navail);
2812 	}
2813 
2814 	device_printf(sc->dev,
2815 	    "failed to find a usable interrupt type.  "
2816 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2817 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2818 
2819 	return (ENXIO);
2820 }
2821 
2822 #define FW_VERSION(chip) ( \
2823     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2824     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2825     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2826     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2827 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2828 
2829 struct fw_info {
2830 	uint8_t chip;
2831 	char *kld_name;
2832 	char *fw_mod_name;
2833 	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
2834 } fw_info[] = {
2835 	{
2836 		.chip = CHELSIO_T4,
2837 		.kld_name = "t4fw_cfg",
2838 		.fw_mod_name = "t4fw",
2839 		.fw_hdr = {
2840 			.chip = FW_HDR_CHIP_T4,
2841 			.fw_ver = htobe32_const(FW_VERSION(T4)),
2842 			.intfver_nic = FW_INTFVER(T4, NIC),
2843 			.intfver_vnic = FW_INTFVER(T4, VNIC),
2844 			.intfver_ofld = FW_INTFVER(T4, OFLD),
2845 			.intfver_ri = FW_INTFVER(T4, RI),
2846 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2847 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
2848 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2849 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
2850 		},
2851 	}, {
2852 		.chip = CHELSIO_T5,
2853 		.kld_name = "t5fw_cfg",
2854 		.fw_mod_name = "t5fw",
2855 		.fw_hdr = {
2856 			.chip = FW_HDR_CHIP_T5,
2857 			.fw_ver = htobe32_const(FW_VERSION(T5)),
2858 			.intfver_nic = FW_INTFVER(T5, NIC),
2859 			.intfver_vnic = FW_INTFVER(T5, VNIC),
2860 			.intfver_ofld = FW_INTFVER(T5, OFLD),
2861 			.intfver_ri = FW_INTFVER(T5, RI),
2862 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
2863 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
2864 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
2865 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
2866 		},
2867 	}, {
2868 		.chip = CHELSIO_T6,
2869 		.kld_name = "t6fw_cfg",
2870 		.fw_mod_name = "t6fw",
2871 		.fw_hdr = {
2872 			.chip = FW_HDR_CHIP_T6,
2873 			.fw_ver = htobe32_const(FW_VERSION(T6)),
2874 			.intfver_nic = FW_INTFVER(T6, NIC),
2875 			.intfver_vnic = FW_INTFVER(T6, VNIC),
2876 			.intfver_ofld = FW_INTFVER(T6, OFLD),
2877 			.intfver_ri = FW_INTFVER(T6, RI),
2878 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
2879 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
2880 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
2881 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
2882 		},
2883 	}
2884 };
2885 
2886 static struct fw_info *
2887 find_fw_info(int chip)
2888 {
2889 	int i;
2890 
2891 	for (i = 0; i < nitems(fw_info); i++) {
2892 		if (fw_info[i].chip == chip)
2893 			return (&fw_info[i]);
2894 	}
2895 	return (NULL);
2896 }
2897 
2898 /*
2899  * Is the given firmware API compatible with the one the driver was compiled
2900  * with?
2901  */
2902 static int
2903 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
2904 {
2905 
2906 	/* short circuit if it's the exact same firmware version */
2907 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
2908 		return (1);
2909 
2910 	/*
2911 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
2912 	 * features that are supported in the driver.
2913 	 */
2914 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
2915 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
2916 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
2917 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
2918 		return (1);
2919 #undef SAME_INTF
2920 
2921 	return (0);
2922 }
2923 
2924 /*
2925  * The firmware in the KLD is usable, but should it be installed?  This routine
2926  * explains itself in detail if it indicates the KLD firmware should be
2927  * installed.
2928  */
2929 static int
2930 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
2931 {
2932 	const char *reason;
2933 
2934 	if (!card_fw_usable) {
2935 		reason = "incompatible or unusable";
2936 		goto install;
2937 	}
2938 
2939 	if (k > c) {
2940 		reason = "older than the version bundled with this driver";
2941 		goto install;
2942 	}
2943 
2944 	if (t4_fw_install == 2 && k != c) {
2945 		reason = "different than the version bundled with this driver";
2946 		goto install;
2947 	}
2948 
2949 	return (0);
2950 
2951 install:
2952 	if (t4_fw_install == 0) {
2953 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2954 		    "but the driver is prohibited from installing a different "
2955 		    "firmware on the card.\n",
2956 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2957 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
2958 
2959 		return (0);
2960 	}
2961 
2962 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2963 	    "installing firmware %u.%u.%u.%u on card.\n",
2964 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2965 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
2966 	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2967 	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2968 
2969 	return (1);
2970 }
2971 /*
2972  * Establish contact with the firmware and determine if we are the master driver
2973  * or not, and whether we are responsible for chip initialization.
2974  */
2975 static int
2976 prep_firmware(struct adapter *sc)
2977 {
2978 	const struct firmware *fw = NULL, *default_cfg;
2979 	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
2980 	enum dev_state state;
2981 	struct fw_info *fw_info;
2982 	struct fw_hdr *card_fw;		/* fw on the card */
2983 	const struct fw_hdr *kld_fw;	/* fw in the KLD */
2984 	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
2985 					   against */
2986 
2987 	/* Contact firmware. */
2988 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
2989 	if (rc < 0 || state == DEV_STATE_ERR) {
2990 		rc = -rc;
2991 		device_printf(sc->dev,
2992 		    "failed to connect to the firmware: %d, %d.\n", rc, state);
2993 		return (rc);
2994 	}
2995 	pf = rc;
2996 	if (pf == sc->mbox)
2997 		sc->flags |= MASTER_PF;
2998 	else if (state == DEV_STATE_UNINIT) {
2999 		/*
3000 		 * We didn't get to be the master so we definitely won't be
3001 		 * configuring the chip.  It's a bug if someone else hasn't
3002 		 * configured it already.
3003 		 */
3004 		device_printf(sc->dev, "couldn't be master(%d), "
3005 		    "device not already initialized either(%d).\n", rc, state);
3006 		return (EDOOFUS);
3007 	}
3008 
3009 	/* This is the firmware whose headers the driver was compiled against */
3010 	fw_info = find_fw_info(chip_id(sc));
3011 	if (fw_info == NULL) {
3012 		device_printf(sc->dev,
3013 		    "unable to look up firmware information for chip %d.\n",
3014 		    chip_id(sc));
3015 		return (EINVAL);
3016 	}
3017 	drv_fw = &fw_info->fw_hdr;
3018 
3019 	/*
3020 	 * The firmware KLD contains many modules.  The KLD name is also the
3021 	 * name of the module that contains the default config file.
3022 	 */
3023 	default_cfg = firmware_get(fw_info->kld_name);
3024 
3025 	/* Read the header of the firmware on the card */
3026 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3027 	rc = -t4_read_flash(sc, FLASH_FW_START,
3028 	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
3029 	if (rc == 0)
3030 		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
3031 	else {
3032 		device_printf(sc->dev,
3033 		    "Unable to read card's firmware header: %d\n", rc);
3034 		card_fw_usable = 0;
3035 	}
3036 
3037 	/* This is the firmware in the KLD */
3038 	fw = firmware_get(fw_info->fw_mod_name);
3039 	if (fw != NULL) {
3040 		kld_fw = (const void *)fw->data;
3041 		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3042 	} else {
3043 		kld_fw = NULL;
3044 		kld_fw_usable = 0;
3045 	}
3046 
3047 	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3048 	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3049 		/*
3050 		 * Common case: the firmware on the card is an exact match and
3051 		 * the KLD is an exact match too, or the KLD is
3052 		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
3053 		 * here -- use cxgbetool loadfw if you want to reinstall the
3054 		 * same firmware as the one on the card.
3055 		 */
3056 	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3057 	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3058 	    be32toh(card_fw->fw_ver))) {
3059 
3060 		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3061 		if (rc != 0) {
3062 			device_printf(sc->dev,
3063 			    "failed to install firmware: %d\n", rc);
3064 			goto done;
3065 		}
3066 
3067 		/* Installed successfully, update the cached header too. */
3068 		memcpy(card_fw, kld_fw, sizeof(*card_fw));
3069 		card_fw_usable = 1;
3070 		need_fw_reset = 0;	/* already reset as part of load_fw */
3071 	}
3072 
3073 	if (!card_fw_usable) {
3074 		uint32_t d, c, k;
3075 
3076 		d = ntohl(drv_fw->fw_ver);
3077 		c = ntohl(card_fw->fw_ver);
3078 		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3079 
3080 		device_printf(sc->dev, "Cannot find a usable firmware: "
3081 		    "fw_install %d, chip state %d, "
3082 		    "driver compiled with %d.%d.%d.%d, "
3083 		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3084 		    t4_fw_install, state,
3085 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3086 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3087 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3088 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3089 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3090 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3091 		rc = EINVAL;
3092 		goto done;
3093 	}
3094 
3095 	/* Reset device */
3096 	if (need_fw_reset &&
3097 	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3098 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3099 		if (rc != ETIMEDOUT && rc != EIO)
3100 			t4_fw_bye(sc, sc->mbox);
3101 		goto done;
3102 	}
3103 	sc->flags |= FW_OK;
3104 
3105 	rc = get_params__pre_init(sc);
3106 	if (rc != 0)
3107 		goto done; /* error message displayed already */
3108 
3109 	/* Partition adapter resources as specified in the config file. */
3110 	if (state == DEV_STATE_UNINIT) {
3111 
3112 		KASSERT(sc->flags & MASTER_PF,
3113 		    ("%s: trying to change chip settings when not master.",
3114 		    __func__));
3115 
3116 		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3117 		if (rc != 0)
3118 			goto done;	/* error message displayed already */
3119 
3120 		t4_tweak_chip_settings(sc);
3121 
3122 		/* get basic stuff going */
3123 		rc = -t4_fw_initialize(sc, sc->mbox);
3124 		if (rc != 0) {
3125 			device_printf(sc->dev, "fw init failed: %d.\n", rc);
3126 			goto done;
3127 		}
3128 	} else {
3129 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3130 		sc->cfcsum = 0;
3131 	}
3132 
3133 done:
3134 	free(card_fw, M_CXGBE);
3135 	if (fw != NULL)
3136 		firmware_put(fw, FIRMWARE_UNLOAD);
3137 	if (default_cfg != NULL)
3138 		firmware_put(default_cfg, FIRMWARE_UNLOAD);
3139 
3140 	return (rc);
3141 }
3142 
3143 #define FW_PARAM_DEV(param) \
3144 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3145 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3146 #define FW_PARAM_PFVF(param) \
3147 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3148 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3149 
3150 /*
3151  * Partition chip resources for use between various PFs, VFs, etc.
3152  */
3153 static int
3154 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3155     const char *name_prefix)
3156 {
3157 	const struct firmware *cfg = NULL;
3158 	int rc = 0;
3159 	struct fw_caps_config_cmd caps;
3160 	uint32_t mtype, moff, finicsum, cfcsum;
3161 
3162 	/*
3163 	 * Figure out what configuration file to use.  Pick the default config
3164 	 * file for the card if the user hasn't specified one explicitly.
3165 	 */
3166 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3167 	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3168 		/* Card specific overrides go here. */
3169 		if (pci_get_device(sc->dev) == 0x440a)
3170 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3171 		if (is_fpga(sc))
3172 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3173 	}
3174 
3175 	/*
3176 	 * We need to load another module if the profile is anything except
3177 	 * "default" or "flash".
3178 	 */
3179 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3180 	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3181 		char s[32];
3182 
3183 		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3184 		cfg = firmware_get(s);
3185 		if (cfg == NULL) {
3186 			if (default_cfg != NULL) {
3187 				device_printf(sc->dev,
3188 				    "unable to load module \"%s\" for "
3189 				    "configuration profile \"%s\", will use "
3190 				    "the default config file instead.\n",
3191 				    s, sc->cfg_file);
3192 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3193 				    "%s", DEFAULT_CF);
3194 			} else {
3195 				device_printf(sc->dev,
3196 				    "unable to load module \"%s\" for "
3197 				    "configuration profile \"%s\", will use "
3198 				    "the config file on the card's flash "
3199 				    "instead.\n", s, sc->cfg_file);
3200 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3201 				    "%s", FLASH_CF);
3202 			}
3203 		}
3204 	}
3205 
3206 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3207 	    default_cfg == NULL) {
3208 		device_printf(sc->dev,
3209 		    "default config file not available, will use the config "
3210 		    "file on the card's flash instead.\n");
3211 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3212 	}
3213 
3214 	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3215 		u_int cflen;
3216 		const uint32_t *cfdata;
3217 		uint32_t param, val, addr;
3218 
3219 		KASSERT(cfg != NULL || default_cfg != NULL,
3220 		    ("%s: no config to upload", __func__));
3221 
3222 		/*
3223 		 * Ask the firmware where it wants us to upload the config file.
3224 		 */
3225 		param = FW_PARAM_DEV(CF);
3226 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3227 		if (rc != 0) {
3228 			/* No support for config file?  Shouldn't happen. */
3229 			device_printf(sc->dev,
3230 			    "failed to query config file location: %d.\n", rc);
3231 			goto done;
3232 		}
3233 		mtype = G_FW_PARAMS_PARAM_Y(val);
3234 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3235 
3236 		/*
3237 		 * XXX: sheer laziness.  We deliberately added 4 bytes of
3238 		 * useless stuffing/comments at the end of the config file so
3239 		 * it's ok to simply throw away the last remaining bytes when
3240 		 * the config file is not an exact multiple of 4.  This also
3241 		 * helps with the validate_mt_off_len check.
3242 		 */
3243 		if (cfg != NULL) {
3244 			cflen = cfg->datasize & ~3;
3245 			cfdata = cfg->data;
3246 		} else {
3247 			cflen = default_cfg->datasize & ~3;
3248 			cfdata = default_cfg->data;
3249 		}
3250 
3251 		if (cflen > FLASH_CFG_MAX_SIZE) {
3252 			device_printf(sc->dev,
3253 			    "config file too long (%d, max allowed is %d).  "
3254 			    "Will try to use the config on the card, if any.\n",
3255 			    cflen, FLASH_CFG_MAX_SIZE);
3256 			goto use_config_on_flash;
3257 		}
3258 
3259 		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3260 		if (rc != 0) {
3261 			device_printf(sc->dev,
3262 			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3263 			    "Will try to use the config on the card, if any.\n",
3264 			    __func__, mtype, moff, cflen, rc);
3265 			goto use_config_on_flash;
3266 		}
3267 		write_via_memwin(sc, 2, addr, cfdata, cflen);
3268 	} else {
3269 use_config_on_flash:
3270 		mtype = FW_MEMTYPE_FLASH;
3271 		moff = t4_flash_cfg_addr(sc);
3272 	}
3273 
3274 	bzero(&caps, sizeof(caps));
3275 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3276 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3277 	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3278 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3279 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3280 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3281 	if (rc != 0) {
3282 		device_printf(sc->dev,
3283 		    "failed to pre-process config file: %d "
3284 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3285 		goto done;
3286 	}
3287 
3288 	finicsum = be32toh(caps.finicsum);
3289 	cfcsum = be32toh(caps.cfcsum);
3290 	if (finicsum != cfcsum) {
3291 		device_printf(sc->dev,
3292 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3293 		    finicsum, cfcsum);
3294 	}
3295 	sc->cfcsum = cfcsum;
3296 
3297 #define LIMIT_CAPS(x) do { \
3298 	caps.x &= htobe16(t4_##x##_allowed); \
3299 } while (0)
3300 
3301 	/*
3302 	 * Let the firmware know what features will (not) be used so it can tune
3303 	 * things accordingly.
3304 	 */
3305 	LIMIT_CAPS(nbmcaps);
3306 	LIMIT_CAPS(linkcaps);
3307 	LIMIT_CAPS(switchcaps);
3308 	LIMIT_CAPS(niccaps);
3309 	LIMIT_CAPS(toecaps);
3310 	LIMIT_CAPS(rdmacaps);
3311 	LIMIT_CAPS(cryptocaps);
3312 	LIMIT_CAPS(iscsicaps);
3313 	LIMIT_CAPS(fcoecaps);
3314 #undef LIMIT_CAPS
3315 
3316 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3317 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3318 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3319 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3320 	if (rc != 0) {
3321 		device_printf(sc->dev,
3322 		    "failed to process config file: %d.\n", rc);
3323 	}
3324 done:
3325 	if (cfg != NULL)
3326 		firmware_put(cfg, FIRMWARE_UNLOAD);
3327 	return (rc);
3328 }
3329 
3330 /*
3331  * Retrieve parameters that are needed (or nice to have) very early.
3332  */
3333 static int
3334 get_params__pre_init(struct adapter *sc)
3335 {
3336 	int rc;
3337 	uint32_t param[2], val[2];
3338 
3339 	t4_get_version_info(sc);
3340 
3341 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3342 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3343 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3344 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3345 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3346 
3347 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3348 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3349 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3350 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3351 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3352 
3353 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3354 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3355 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3356 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3357 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3358 
3359 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3360 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3361 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3362 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3363 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3364 
3365 	param[0] = FW_PARAM_DEV(PORTVEC);
3366 	param[1] = FW_PARAM_DEV(CCLK);
3367 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3368 	if (rc != 0) {
3369 		device_printf(sc->dev,
3370 		    "failed to query parameters (pre_init): %d.\n", rc);
3371 		return (rc);
3372 	}
3373 
3374 	sc->params.portvec = val[0];
3375 	sc->params.nports = bitcount32(val[0]);
3376 	sc->params.vpd.cclk = val[1];
3377 
3378 	/* Read device log parameters. */
3379 	rc = -t4_init_devlog_params(sc, 1);
3380 	if (rc == 0)
3381 		fixup_devlog_params(sc);
3382 	else {
3383 		device_printf(sc->dev,
3384 		    "failed to get devlog parameters: %d.\n", rc);
3385 		rc = 0;	/* devlog isn't critical for device operation */
3386 	}
3387 
3388 	return (rc);
3389 }
3390 
3391 /*
3392  * Retrieve various parameters that are of interest to the driver.  The device
3393  * has been initialized by the firmware at this point.
3394  */
3395 static int
3396 get_params__post_init(struct adapter *sc)
3397 {
3398 	int rc;
3399 	uint32_t param[7], val[7];
3400 	struct fw_caps_config_cmd caps;
3401 
3402 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3403 	param[1] = FW_PARAM_PFVF(EQ_START);
3404 	param[2] = FW_PARAM_PFVF(FILTER_START);
3405 	param[3] = FW_PARAM_PFVF(FILTER_END);
3406 	param[4] = FW_PARAM_PFVF(L2T_START);
3407 	param[5] = FW_PARAM_PFVF(L2T_END);
3408 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3409 	if (rc != 0) {
3410 		device_printf(sc->dev,
3411 		    "failed to query parameters (post_init): %d.\n", rc);
3412 		return (rc);
3413 	}
3414 
3415 	sc->sge.iq_start = val[0];
3416 	sc->sge.eq_start = val[1];
3417 	sc->tids.ftid_base = val[2];
3418 	sc->tids.nftids = val[3] - val[2] + 1;
3419 	sc->params.ftid_min = val[2];
3420 	sc->params.ftid_max = val[3];
3421 	sc->vres.l2t.start = val[4];
3422 	sc->vres.l2t.size = val[5] - val[4] + 1;
3423 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3424 	    ("%s: L2 table size (%u) larger than expected (%u)",
3425 	    __func__, sc->vres.l2t.size, L2T_SIZE));
3426 
3427 	/* get capabilites */
3428 	bzero(&caps, sizeof(caps));
3429 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3430 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3431 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3432 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3433 	if (rc != 0) {
3434 		device_printf(sc->dev,
3435 		    "failed to get card capabilities: %d.\n", rc);
3436 		return (rc);
3437 	}
3438 
3439 #define READ_CAPS(x) do { \
3440 	sc->x = htobe16(caps.x); \
3441 } while (0)
3442 	READ_CAPS(nbmcaps);
3443 	READ_CAPS(linkcaps);
3444 	READ_CAPS(switchcaps);
3445 	READ_CAPS(niccaps);
3446 	READ_CAPS(toecaps);
3447 	READ_CAPS(rdmacaps);
3448 	READ_CAPS(cryptocaps);
3449 	READ_CAPS(iscsicaps);
3450 	READ_CAPS(fcoecaps);
3451 
3452 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3453 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3454 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3455 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3456 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3457 		if (rc != 0) {
3458 			device_printf(sc->dev,
3459 			    "failed to query NIC parameters: %d.\n", rc);
3460 			return (rc);
3461 		}
3462 		sc->tids.etid_base = val[0];
3463 		sc->params.etid_min = val[0];
3464 		sc->tids.netids = val[1] - val[0] + 1;
3465 		sc->params.netids = sc->tids.netids;
3466 		sc->params.eo_wr_cred = val[2];
3467 		sc->params.ethoffload = 1;
3468 	}
3469 
3470 	if (sc->toecaps) {
3471 		/* query offload-related parameters */
3472 		param[0] = FW_PARAM_DEV(NTID);
3473 		param[1] = FW_PARAM_PFVF(SERVER_START);
3474 		param[2] = FW_PARAM_PFVF(SERVER_END);
3475 		param[3] = FW_PARAM_PFVF(TDDP_START);
3476 		param[4] = FW_PARAM_PFVF(TDDP_END);
3477 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3478 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3479 		if (rc != 0) {
3480 			device_printf(sc->dev,
3481 			    "failed to query TOE parameters: %d.\n", rc);
3482 			return (rc);
3483 		}
3484 		sc->tids.ntids = val[0];
3485 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3486 		sc->tids.stid_base = val[1];
3487 		sc->tids.nstids = val[2] - val[1] + 1;
3488 		sc->vres.ddp.start = val[3];
3489 		sc->vres.ddp.size = val[4] - val[3] + 1;
3490 		sc->params.ofldq_wr_cred = val[5];
3491 		sc->params.offload = 1;
3492 	}
3493 	if (sc->rdmacaps) {
3494 		param[0] = FW_PARAM_PFVF(STAG_START);
3495 		param[1] = FW_PARAM_PFVF(STAG_END);
3496 		param[2] = FW_PARAM_PFVF(RQ_START);
3497 		param[3] = FW_PARAM_PFVF(RQ_END);
3498 		param[4] = FW_PARAM_PFVF(PBL_START);
3499 		param[5] = FW_PARAM_PFVF(PBL_END);
3500 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3501 		if (rc != 0) {
3502 			device_printf(sc->dev,
3503 			    "failed to query RDMA parameters(1): %d.\n", rc);
3504 			return (rc);
3505 		}
3506 		sc->vres.stag.start = val[0];
3507 		sc->vres.stag.size = val[1] - val[0] + 1;
3508 		sc->vres.rq.start = val[2];
3509 		sc->vres.rq.size = val[3] - val[2] + 1;
3510 		sc->vres.pbl.start = val[4];
3511 		sc->vres.pbl.size = val[5] - val[4] + 1;
3512 
3513 		param[0] = FW_PARAM_PFVF(SQRQ_START);
3514 		param[1] = FW_PARAM_PFVF(SQRQ_END);
3515 		param[2] = FW_PARAM_PFVF(CQ_START);
3516 		param[3] = FW_PARAM_PFVF(CQ_END);
3517 		param[4] = FW_PARAM_PFVF(OCQ_START);
3518 		param[5] = FW_PARAM_PFVF(OCQ_END);
3519 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3520 		if (rc != 0) {
3521 			device_printf(sc->dev,
3522 			    "failed to query RDMA parameters(2): %d.\n", rc);
3523 			return (rc);
3524 		}
3525 		sc->vres.qp.start = val[0];
3526 		sc->vres.qp.size = val[1] - val[0] + 1;
3527 		sc->vres.cq.start = val[2];
3528 		sc->vres.cq.size = val[3] - val[2] + 1;
3529 		sc->vres.ocq.start = val[4];
3530 		sc->vres.ocq.size = val[5] - val[4] + 1;
3531 
3532 		param[0] = FW_PARAM_PFVF(SRQ_START);
3533 		param[1] = FW_PARAM_PFVF(SRQ_END);
3534 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
3535 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
3536 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
3537 		if (rc != 0) {
3538 			device_printf(sc->dev,
3539 			    "failed to query RDMA parameters(3): %d.\n", rc);
3540 			return (rc);
3541 		}
3542 		sc->vres.srq.start = val[0];
3543 		sc->vres.srq.size = val[1] - val[0] + 1;
3544 		sc->params.max_ordird_qp = val[2];
3545 		sc->params.max_ird_adapter = val[3];
3546 	}
3547 	if (sc->iscsicaps) {
3548 		param[0] = FW_PARAM_PFVF(ISCSI_START);
3549 		param[1] = FW_PARAM_PFVF(ISCSI_END);
3550 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3551 		if (rc != 0) {
3552 			device_printf(sc->dev,
3553 			    "failed to query iSCSI parameters: %d.\n", rc);
3554 			return (rc);
3555 		}
3556 		sc->vres.iscsi.start = val[0];
3557 		sc->vres.iscsi.size = val[1] - val[0] + 1;
3558 	}
3559 
3560 	t4_init_sge_params(sc);
3561 
3562 	/*
3563 	 * We've got the params we wanted to query via the firmware.  Now grab
3564 	 * some others directly from the chip.
3565 	 */
3566 	rc = t4_read_chip_settings(sc);
3567 
3568 	return (rc);
3569 }
3570 
3571 static int
3572 set_params__post_init(struct adapter *sc)
3573 {
3574 	uint32_t param, val;
3575 
3576 	/* ask for encapsulated CPLs */
3577 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3578 	val = 1;
3579 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3580 
3581 	return (0);
3582 }
3583 
3584 #undef FW_PARAM_PFVF
3585 #undef FW_PARAM_DEV
3586 
3587 static void
3588 t4_set_desc(struct adapter *sc)
3589 {
3590 	char buf[128];
3591 	struct adapter_params *p = &sc->params;
3592 
3593 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
3594 
3595 	device_set_desc_copy(sc->dev, buf);
3596 }
3597 
3598 static void
3599 build_medialist(struct port_info *pi, struct ifmedia *media)
3600 {
3601 	int m;
3602 
3603 	PORT_LOCK_ASSERT_OWNED(pi);
3604 
3605 	ifmedia_removeall(media);
3606 
3607 	/*
3608 	 * XXX: Would it be better to ifmedia_add all 4 combinations of pause
3609 	 * settings for every speed instead of just txpause|rxpause?  ifconfig
3610 	 * media display looks much better if autoselect is the only case where
3611 	 * ifm_current is different from ifm_active.  If the user picks anything
3612 	 * except txpause|rxpause the display is ugly.
3613 	 */
3614 	m = IFM_ETHER | IFM_FDX | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
3615 
3616 	switch(pi->port_type) {
3617 	case FW_PORT_TYPE_BT_XFI:
3618 	case FW_PORT_TYPE_BT_XAUI:
3619 		ifmedia_add(media, m | IFM_10G_T, 0, NULL);
3620 		/* fall through */
3621 
3622 	case FW_PORT_TYPE_BT_SGMII:
3623 		ifmedia_add(media, m | IFM_1000_T, 0, NULL);
3624 		ifmedia_add(media, m | IFM_100_TX, 0, NULL);
3625 		ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL);
3626 		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
3627 		break;
3628 
3629 	case FW_PORT_TYPE_CX4:
3630 		ifmedia_add(media, m | IFM_10G_CX4, 0, NULL);
3631 		ifmedia_set(media, m | IFM_10G_CX4);
3632 		break;
3633 
3634 	case FW_PORT_TYPE_QSFP_10G:
3635 	case FW_PORT_TYPE_SFP:
3636 	case FW_PORT_TYPE_FIBER_XFI:
3637 	case FW_PORT_TYPE_FIBER_XAUI:
3638 		switch (pi->mod_type) {
3639 
3640 		case FW_PORT_MOD_TYPE_LR:
3641 			ifmedia_add(media, m | IFM_10G_LR, 0, NULL);
3642 			ifmedia_set(media, m | IFM_10G_LR);
3643 			break;
3644 
3645 		case FW_PORT_MOD_TYPE_SR:
3646 			ifmedia_add(media, m | IFM_10G_SR, 0, NULL);
3647 			ifmedia_set(media, m | IFM_10G_SR);
3648 			break;
3649 
3650 		case FW_PORT_MOD_TYPE_LRM:
3651 			ifmedia_add(media, m | IFM_10G_LRM, 0, NULL);
3652 			ifmedia_set(media, m | IFM_10G_LRM);
3653 			break;
3654 
3655 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3656 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3657 			ifmedia_add(media, m | IFM_10G_TWINAX, 0, NULL);
3658 			ifmedia_set(media, m | IFM_10G_TWINAX);
3659 			break;
3660 
3661 		case FW_PORT_MOD_TYPE_NONE:
3662 			m &= ~IFM_FDX;
3663 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3664 			ifmedia_set(media, m | IFM_NONE);
3665 			break;
3666 
3667 		case FW_PORT_MOD_TYPE_NA:
3668 		case FW_PORT_MOD_TYPE_ER:
3669 		default:
3670 			device_printf(pi->dev,
3671 			    "unknown port_type (%d), mod_type (%d)\n",
3672 			    pi->port_type, pi->mod_type);
3673 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3674 			ifmedia_set(media, m | IFM_UNKNOWN);
3675 			break;
3676 		}
3677 		break;
3678 
3679 	case FW_PORT_TYPE_CR_QSFP:
3680 	case FW_PORT_TYPE_SFP28:
3681 	case FW_PORT_TYPE_KR_SFP28:
3682 		switch (pi->mod_type) {
3683 
3684 		case FW_PORT_MOD_TYPE_SR:
3685 			ifmedia_add(media, m | IFM_25G_SR, 0, NULL);
3686 			ifmedia_set(media, m | IFM_25G_SR);
3687 			break;
3688 
3689 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3690 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3691 			ifmedia_add(media, m | IFM_25G_CR, 0, NULL);
3692 			ifmedia_set(media, m | IFM_25G_CR);
3693 			break;
3694 
3695 		case FW_PORT_MOD_TYPE_NONE:
3696 			m &= ~IFM_FDX;
3697 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3698 			ifmedia_set(media, m | IFM_NONE);
3699 			break;
3700 
3701 		default:
3702 			device_printf(pi->dev,
3703 			    "unknown port_type (%d), mod_type (%d)\n",
3704 			    pi->port_type, pi->mod_type);
3705 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3706 			ifmedia_set(media, m | IFM_UNKNOWN);
3707 			break;
3708 		}
3709 		break;
3710 
3711 	case FW_PORT_TYPE_QSFP:
3712 		switch (pi->mod_type) {
3713 
3714 		case FW_PORT_MOD_TYPE_LR:
3715 			ifmedia_add(media, m | IFM_40G_LR4, 0, NULL);
3716 			ifmedia_set(media, m | IFM_40G_LR4);
3717 			break;
3718 
3719 		case FW_PORT_MOD_TYPE_SR:
3720 			ifmedia_add(media, m | IFM_40G_SR4, 0, NULL);
3721 			ifmedia_set(media, m | IFM_40G_SR4);
3722 			break;
3723 
3724 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3725 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3726 			ifmedia_add(media, m | IFM_40G_CR4, 0, NULL);
3727 			ifmedia_set(media, m | IFM_40G_CR4);
3728 			break;
3729 
3730 		case FW_PORT_MOD_TYPE_NONE:
3731 			m &= ~IFM_FDX;
3732 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3733 			ifmedia_set(media, m | IFM_NONE);
3734 			break;
3735 
3736 		default:
3737 			device_printf(pi->dev,
3738 			    "unknown port_type (%d), mod_type (%d)\n",
3739 			    pi->port_type, pi->mod_type);
3740 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3741 			ifmedia_set(media, m | IFM_UNKNOWN);
3742 			break;
3743 		}
3744 		break;
3745 
3746 	case FW_PORT_TYPE_KR4_100G:
3747 	case FW_PORT_TYPE_CR4_QSFP:
3748 		switch (pi->mod_type) {
3749 
3750 		case FW_PORT_MOD_TYPE_LR:
3751 			ifmedia_add(media, m | IFM_100G_LR4, 0, NULL);
3752 			ifmedia_set(media, m | IFM_100G_LR4);
3753 			break;
3754 
3755 		case FW_PORT_MOD_TYPE_SR:
3756 			ifmedia_add(media, m | IFM_100G_SR4, 0, NULL);
3757 			ifmedia_set(media, m | IFM_100G_SR4);
3758 			break;
3759 
3760 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3761 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3762 			ifmedia_add(media, m | IFM_100G_CR4, 0, NULL);
3763 			ifmedia_set(media, m | IFM_100G_CR4);
3764 			break;
3765 
3766 		case FW_PORT_MOD_TYPE_NONE:
3767 			m &= ~IFM_FDX;
3768 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3769 			ifmedia_set(media, m | IFM_NONE);
3770 			break;
3771 
3772 		default:
3773 			device_printf(pi->dev,
3774 			    "unknown port_type (%d), mod_type (%d)\n",
3775 			    pi->port_type, pi->mod_type);
3776 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3777 			ifmedia_set(media, m | IFM_UNKNOWN);
3778 			break;
3779 		}
3780 		break;
3781 
3782 	default:
3783 		device_printf(pi->dev,
3784 		    "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
3785 		    pi->mod_type);
3786 		ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3787 		ifmedia_set(media, m | IFM_UNKNOWN);
3788 		break;
3789 	}
3790 }
3791 
3792 /*
3793  * Update all the requested_* fields in the link config and then send a mailbox
3794  * command to apply the settings.
3795  */
3796 static void
3797 init_l1cfg(struct port_info *pi)
3798 {
3799 	struct adapter *sc = pi->adapter;
3800 	struct link_config *lc = &pi->link_cfg;
3801 	int rc;
3802 
3803 	ASSERT_SYNCHRONIZED_OP(sc);
3804 
3805 	if (t4_autoneg != 0 && lc->supported & FW_PORT_CAP_ANEG) {
3806 		lc->requested_aneg = AUTONEG_ENABLE;
3807 		lc->requested_speed = 0;
3808 	} else {
3809 		lc->requested_aneg = AUTONEG_DISABLE;
3810 		lc->requested_speed = port_top_speed(pi);	/* in Gbps */
3811 	}
3812 
3813 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX);
3814 
3815 	if (t4_fec != -1) {
3816 		lc->requested_fec = t4_fec & (FEC_RS | FEC_BASER_RS |
3817 		    FEC_RESERVED);
3818 	} else {
3819 		/* Use the suggested value provided by the firmware in acaps */
3820 		if (lc->advertising & FW_PORT_CAP_FEC_RS)
3821 			lc->requested_fec = FEC_RS;
3822 		else if (lc->advertising & FW_PORT_CAP_FEC_BASER_RS)
3823 			lc->requested_fec = FEC_BASER_RS;
3824 		else if (lc->advertising & FW_PORT_CAP_FEC_RESERVED)
3825 			lc->requested_fec = FEC_RESERVED;
3826 		else
3827 			lc->requested_fec = 0;
3828 	}
3829 
3830 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
3831 	if (rc != 0) {
3832 		device_printf(pi->dev, "l1cfg failed: %d\n", rc);
3833 	} else {
3834 		lc->fc = lc->requested_fc;
3835 		lc->fec = lc->requested_fec;
3836 	}
3837 }
3838 
3839 #define FW_MAC_EXACT_CHUNK	7
3840 
3841 /*
3842  * Program the port's XGMAC based on parameters in ifnet.  The caller also
3843  * indicates which parameters should be programmed (the rest are left alone).
3844  */
3845 int
3846 update_mac_settings(struct ifnet *ifp, int flags)
3847 {
3848 	int rc = 0;
3849 	struct vi_info *vi = ifp->if_softc;
3850 	struct port_info *pi = vi->pi;
3851 	struct adapter *sc = pi->adapter;
3852 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
3853 
3854 	ASSERT_SYNCHRONIZED_OP(sc);
3855 	KASSERT(flags, ("%s: not told what to update.", __func__));
3856 
3857 	if (flags & XGMAC_MTU)
3858 		mtu = ifp->if_mtu;
3859 
3860 	if (flags & XGMAC_PROMISC)
3861 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
3862 
3863 	if (flags & XGMAC_ALLMULTI)
3864 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
3865 
3866 	if (flags & XGMAC_VLANEX)
3867 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
3868 
3869 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
3870 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
3871 		    allmulti, 1, vlanex, false);
3872 		if (rc) {
3873 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
3874 			    rc);
3875 			return (rc);
3876 		}
3877 	}
3878 
3879 	if (flags & XGMAC_UCADDR) {
3880 		uint8_t ucaddr[ETHER_ADDR_LEN];
3881 
3882 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
3883 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
3884 		    ucaddr, true, true);
3885 		if (rc < 0) {
3886 			rc = -rc;
3887 			if_printf(ifp, "change_mac failed: %d\n", rc);
3888 			return (rc);
3889 		} else {
3890 			vi->xact_addr_filt = rc;
3891 			rc = 0;
3892 		}
3893 	}
3894 
3895 	if (flags & XGMAC_MCADDRS) {
3896 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
3897 		int del = 1;
3898 		uint64_t hash = 0;
3899 		struct ifmultiaddr *ifma;
3900 		int i = 0, j;
3901 
3902 		if_maddr_rlock(ifp);
3903 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3904 			if (ifma->ifma_addr->sa_family != AF_LINK)
3905 				continue;
3906 			mcaddr[i] =
3907 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
3908 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
3909 			i++;
3910 
3911 			if (i == FW_MAC_EXACT_CHUNK) {
3912 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
3913 				    del, i, mcaddr, NULL, &hash, 0);
3914 				if (rc < 0) {
3915 					rc = -rc;
3916 					for (j = 0; j < i; j++) {
3917 						if_printf(ifp,
3918 						    "failed to add mc address"
3919 						    " %02x:%02x:%02x:"
3920 						    "%02x:%02x:%02x rc=%d\n",
3921 						    mcaddr[j][0], mcaddr[j][1],
3922 						    mcaddr[j][2], mcaddr[j][3],
3923 						    mcaddr[j][4], mcaddr[j][5],
3924 						    rc);
3925 					}
3926 					goto mcfail;
3927 				}
3928 				del = 0;
3929 				i = 0;
3930 			}
3931 		}
3932 		if (i > 0) {
3933 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
3934 			    mcaddr, NULL, &hash, 0);
3935 			if (rc < 0) {
3936 				rc = -rc;
3937 				for (j = 0; j < i; j++) {
3938 					if_printf(ifp,
3939 					    "failed to add mc address"
3940 					    " %02x:%02x:%02x:"
3941 					    "%02x:%02x:%02x rc=%d\n",
3942 					    mcaddr[j][0], mcaddr[j][1],
3943 					    mcaddr[j][2], mcaddr[j][3],
3944 					    mcaddr[j][4], mcaddr[j][5],
3945 					    rc);
3946 				}
3947 				goto mcfail;
3948 			}
3949 		}
3950 
3951 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
3952 		if (rc != 0)
3953 			if_printf(ifp, "failed to set mc address hash: %d", rc);
3954 mcfail:
3955 		if_maddr_runlock(ifp);
3956 	}
3957 
3958 	return (rc);
3959 }
3960 
3961 /*
3962  * {begin|end}_synchronized_op must be called from the same thread.
3963  */
3964 int
3965 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
3966     char *wmesg)
3967 {
3968 	int rc, pri;
3969 
3970 #ifdef WITNESS
3971 	/* the caller thinks it's ok to sleep, but is it really? */
3972 	if (flags & SLEEP_OK)
3973 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
3974 		    "begin_synchronized_op");
3975 #endif
3976 
3977 	if (INTR_OK)
3978 		pri = PCATCH;
3979 	else
3980 		pri = 0;
3981 
3982 	ADAPTER_LOCK(sc);
3983 	for (;;) {
3984 
3985 		if (vi && IS_DOOMED(vi)) {
3986 			rc = ENXIO;
3987 			goto done;
3988 		}
3989 
3990 		if (!IS_BUSY(sc)) {
3991 			rc = 0;
3992 			break;
3993 		}
3994 
3995 		if (!(flags & SLEEP_OK)) {
3996 			rc = EBUSY;
3997 			goto done;
3998 		}
3999 
4000 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4001 			rc = EINTR;
4002 			goto done;
4003 		}
4004 	}
4005 
4006 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4007 	SET_BUSY(sc);
4008 #ifdef INVARIANTS
4009 	sc->last_op = wmesg;
4010 	sc->last_op_thr = curthread;
4011 	sc->last_op_flags = flags;
4012 #endif
4013 
4014 done:
4015 	if (!(flags & HOLD_LOCK) || rc)
4016 		ADAPTER_UNLOCK(sc);
4017 
4018 	return (rc);
4019 }
4020 
4021 /*
4022  * Tell if_ioctl and if_init that the VI is going away.  This is
4023  * special variant of begin_synchronized_op and must be paired with a
4024  * call to end_synchronized_op.
4025  */
4026 void
4027 doom_vi(struct adapter *sc, struct vi_info *vi)
4028 {
4029 
4030 	ADAPTER_LOCK(sc);
4031 	SET_DOOMED(vi);
4032 	wakeup(&sc->flags);
4033 	while (IS_BUSY(sc))
4034 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4035 	SET_BUSY(sc);
4036 #ifdef INVARIANTS
4037 	sc->last_op = "t4detach";
4038 	sc->last_op_thr = curthread;
4039 	sc->last_op_flags = 0;
4040 #endif
4041 	ADAPTER_UNLOCK(sc);
4042 }
4043 
4044 /*
4045  * {begin|end}_synchronized_op must be called from the same thread.
4046  */
4047 void
4048 end_synchronized_op(struct adapter *sc, int flags)
4049 {
4050 
4051 	if (flags & LOCK_HELD)
4052 		ADAPTER_LOCK_ASSERT_OWNED(sc);
4053 	else
4054 		ADAPTER_LOCK(sc);
4055 
4056 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
4057 	CLR_BUSY(sc);
4058 	wakeup(&sc->flags);
4059 	ADAPTER_UNLOCK(sc);
4060 }
4061 
4062 static int
4063 cxgbe_init_synchronized(struct vi_info *vi)
4064 {
4065 	struct port_info *pi = vi->pi;
4066 	struct adapter *sc = pi->adapter;
4067 	struct ifnet *ifp = vi->ifp;
4068 	int rc = 0, i;
4069 	struct sge_txq *txq;
4070 
4071 	ASSERT_SYNCHRONIZED_OP(sc);
4072 
4073 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
4074 		return (0);	/* already running */
4075 
4076 	if (!(sc->flags & FULL_INIT_DONE) &&
4077 	    ((rc = adapter_full_init(sc)) != 0))
4078 		return (rc);	/* error message displayed already */
4079 
4080 	if (!(vi->flags & VI_INIT_DONE) &&
4081 	    ((rc = vi_full_init(vi)) != 0))
4082 		return (rc); /* error message displayed already */
4083 
4084 	rc = update_mac_settings(ifp, XGMAC_ALL);
4085 	if (rc)
4086 		goto done;	/* error message displayed already */
4087 
4088 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
4089 	if (rc != 0) {
4090 		if_printf(ifp, "enable_vi failed: %d\n", rc);
4091 		goto done;
4092 	}
4093 
4094 	/*
4095 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
4096 	 * if this changes.
4097 	 */
4098 
4099 	for_each_txq(vi, i, txq) {
4100 		TXQ_LOCK(txq);
4101 		txq->eq.flags |= EQ_ENABLED;
4102 		TXQ_UNLOCK(txq);
4103 	}
4104 
4105 	/*
4106 	 * The first iq of the first port to come up is used for tracing.
4107 	 */
4108 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4109 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4110 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4111 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4112 		    V_QUEUENUMBER(sc->traceq));
4113 		pi->flags |= HAS_TRACEQ;
4114 	}
4115 
4116 	/* all ok */
4117 	PORT_LOCK(pi);
4118 	if (pi->up_vis++ == 0) {
4119 		t4_update_port_info(pi);
4120 		build_medialist(vi->pi, &vi->media);
4121 		init_l1cfg(pi);
4122 	}
4123 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
4124 
4125 	if (pi->nvi > 1 || sc->flags & IS_VF)
4126 		callout_reset(&vi->tick, hz, vi_tick, vi);
4127 	else
4128 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4129 	PORT_UNLOCK(pi);
4130 done:
4131 	if (rc != 0)
4132 		cxgbe_uninit_synchronized(vi);
4133 
4134 	return (rc);
4135 }
4136 
4137 /*
4138  * Idempotent.
4139  */
4140 static int
4141 cxgbe_uninit_synchronized(struct vi_info *vi)
4142 {
4143 	struct port_info *pi = vi->pi;
4144 	struct adapter *sc = pi->adapter;
4145 	struct ifnet *ifp = vi->ifp;
4146 	int rc, i;
4147 	struct sge_txq *txq;
4148 
4149 	ASSERT_SYNCHRONIZED_OP(sc);
4150 
4151 	if (!(vi->flags & VI_INIT_DONE)) {
4152 		KASSERT(!(ifp->if_drv_flags & IFF_DRV_RUNNING),
4153 		    ("uninited VI is running"));
4154 		return (0);
4155 	}
4156 
4157 	/*
4158 	 * Disable the VI so that all its data in either direction is discarded
4159 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4160 	 * tick) intact as the TP can deliver negative advice or data that it's
4161 	 * holding in its RAM (for an offloaded connection) even after the VI is
4162 	 * disabled.
4163 	 */
4164 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4165 	if (rc) {
4166 		if_printf(ifp, "disable_vi failed: %d\n", rc);
4167 		return (rc);
4168 	}
4169 
4170 	for_each_txq(vi, i, txq) {
4171 		TXQ_LOCK(txq);
4172 		txq->eq.flags &= ~EQ_ENABLED;
4173 		TXQ_UNLOCK(txq);
4174 	}
4175 
4176 	PORT_LOCK(pi);
4177 	if (pi->nvi > 1 || sc->flags & IS_VF)
4178 		callout_stop(&vi->tick);
4179 	else
4180 		callout_stop(&pi->tick);
4181 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4182 		PORT_UNLOCK(pi);
4183 		return (0);
4184 	}
4185 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4186 	pi->up_vis--;
4187 	if (pi->up_vis > 0) {
4188 		PORT_UNLOCK(pi);
4189 		return (0);
4190 	}
4191 	PORT_UNLOCK(pi);
4192 
4193 	pi->link_cfg.link_ok = 0;
4194 	pi->link_cfg.speed = 0;
4195 	pi->link_cfg.link_down_rc = 255;
4196 	t4_os_link_changed(pi, NULL);
4197 
4198 	return (0);
4199 }
4200 
4201 /*
4202  * It is ok for this function to fail midway and return right away.  t4_detach
4203  * will walk the entire sc->irq list and clean up whatever is valid.
4204  */
4205 int
4206 t4_setup_intr_handlers(struct adapter *sc)
4207 {
4208 	int rc, rid, p, q, v;
4209 	char s[8];
4210 	struct irq *irq;
4211 	struct port_info *pi;
4212 	struct vi_info *vi;
4213 	struct sge *sge = &sc->sge;
4214 	struct sge_rxq *rxq;
4215 #ifdef TCP_OFFLOAD
4216 	struct sge_ofld_rxq *ofld_rxq;
4217 #endif
4218 #ifdef DEV_NETMAP
4219 	struct sge_nm_rxq *nm_rxq;
4220 #endif
4221 #ifdef RSS
4222 	int nbuckets = rss_getnumbuckets();
4223 #endif
4224 
4225 	/*
4226 	 * Setup interrupts.
4227 	 */
4228 	irq = &sc->irq[0];
4229 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
4230 	if (sc->intr_count == 1)
4231 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4232 
4233 	/* Multiple interrupts. */
4234 	if (sc->flags & IS_VF)
4235 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4236 		    ("%s: too few intr.", __func__));
4237 	else
4238 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4239 		    ("%s: too few intr.", __func__));
4240 
4241 	/* The first one is always error intr on PFs */
4242 	if (!(sc->flags & IS_VF)) {
4243 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4244 		if (rc != 0)
4245 			return (rc);
4246 		irq++;
4247 		rid++;
4248 	}
4249 
4250 	/* The second one is always the firmware event queue (first on VFs) */
4251 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4252 	if (rc != 0)
4253 		return (rc);
4254 	irq++;
4255 	rid++;
4256 
4257 	for_each_port(sc, p) {
4258 		pi = sc->port[p];
4259 		for_each_vi(pi, v, vi) {
4260 			vi->first_intr = rid - 1;
4261 
4262 			if (vi->nnmrxq > 0) {
4263 				int n = max(vi->nrxq, vi->nnmrxq);
4264 
4265 				MPASS(vi->flags & INTR_RXQ);
4266 
4267 				rxq = &sge->rxq[vi->first_rxq];
4268 #ifdef DEV_NETMAP
4269 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4270 #endif
4271 				for (q = 0; q < n; q++) {
4272 					snprintf(s, sizeof(s), "%x%c%x", p,
4273 					    'a' + v, q);
4274 					if (q < vi->nrxq)
4275 						irq->rxq = rxq++;
4276 #ifdef DEV_NETMAP
4277 					if (q < vi->nnmrxq)
4278 						irq->nm_rxq = nm_rxq++;
4279 #endif
4280 					rc = t4_alloc_irq(sc, irq, rid,
4281 					    t4_vi_intr, irq, s);
4282 					if (rc != 0)
4283 						return (rc);
4284 					irq++;
4285 					rid++;
4286 					vi->nintr++;
4287 				}
4288 			} else if (vi->flags & INTR_RXQ) {
4289 				for_each_rxq(vi, q, rxq) {
4290 					snprintf(s, sizeof(s), "%x%c%x", p,
4291 					    'a' + v, q);
4292 					rc = t4_alloc_irq(sc, irq, rid,
4293 					    t4_intr, rxq, s);
4294 					if (rc != 0)
4295 						return (rc);
4296 #ifdef RSS
4297 					bus_bind_intr(sc->dev, irq->res,
4298 					    rss_getcpu(q % nbuckets));
4299 #endif
4300 					irq++;
4301 					rid++;
4302 					vi->nintr++;
4303 				}
4304 			}
4305 #ifdef TCP_OFFLOAD
4306 			if (vi->flags & INTR_OFLD_RXQ) {
4307 				for_each_ofld_rxq(vi, q, ofld_rxq) {
4308 					snprintf(s, sizeof(s), "%x%c%x", p,
4309 					    'A' + v, q);
4310 					rc = t4_alloc_irq(sc, irq, rid,
4311 					    t4_intr, ofld_rxq, s);
4312 					if (rc != 0)
4313 						return (rc);
4314 					irq++;
4315 					rid++;
4316 					vi->nintr++;
4317 				}
4318 			}
4319 #endif
4320 		}
4321 	}
4322 	MPASS(irq == &sc->irq[sc->intr_count]);
4323 
4324 	return (0);
4325 }
4326 
4327 int
4328 adapter_full_init(struct adapter *sc)
4329 {
4330 	int rc, i;
4331 #ifdef RSS
4332 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4333 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4334 #endif
4335 
4336 	ASSERT_SYNCHRONIZED_OP(sc);
4337 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4338 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
4339 	    ("%s: FULL_INIT_DONE already", __func__));
4340 
4341 	/*
4342 	 * queues that belong to the adapter (not any particular port).
4343 	 */
4344 	rc = t4_setup_adapter_queues(sc);
4345 	if (rc != 0)
4346 		goto done;
4347 
4348 	for (i = 0; i < nitems(sc->tq); i++) {
4349 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
4350 		    taskqueue_thread_enqueue, &sc->tq[i]);
4351 		if (sc->tq[i] == NULL) {
4352 			device_printf(sc->dev,
4353 			    "failed to allocate task queue %d\n", i);
4354 			rc = ENOMEM;
4355 			goto done;
4356 		}
4357 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
4358 		    device_get_nameunit(sc->dev), i);
4359 	}
4360 #ifdef RSS
4361 	MPASS(RSS_KEYSIZE == 40);
4362 	rss_getkey((void *)&raw_rss_key[0]);
4363 	for (i = 0; i < nitems(rss_key); i++) {
4364 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
4365 	}
4366 	t4_write_rss_key(sc, &rss_key[0], -1);
4367 #endif
4368 
4369 	if (!(sc->flags & IS_VF))
4370 		t4_intr_enable(sc);
4371 	sc->flags |= FULL_INIT_DONE;
4372 done:
4373 	if (rc != 0)
4374 		adapter_full_uninit(sc);
4375 
4376 	return (rc);
4377 }
4378 
4379 int
4380 adapter_full_uninit(struct adapter *sc)
4381 {
4382 	int i;
4383 
4384 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4385 
4386 	t4_teardown_adapter_queues(sc);
4387 
4388 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
4389 		taskqueue_free(sc->tq[i]);
4390 		sc->tq[i] = NULL;
4391 	}
4392 
4393 	sc->flags &= ~FULL_INIT_DONE;
4394 
4395 	return (0);
4396 }
4397 
4398 #ifdef RSS
4399 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
4400     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
4401     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
4402     RSS_HASHTYPE_RSS_UDP_IPV6)
4403 
4404 /* Translates kernel hash types to hardware. */
4405 static int
4406 hashconfig_to_hashen(int hashconfig)
4407 {
4408 	int hashen = 0;
4409 
4410 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
4411 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
4412 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
4413 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
4414 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
4415 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4416 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4417 	}
4418 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
4419 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4420 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4421 	}
4422 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
4423 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4424 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
4425 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4426 
4427 	return (hashen);
4428 }
4429 
4430 /* Translates hardware hash types to kernel. */
4431 static int
4432 hashen_to_hashconfig(int hashen)
4433 {
4434 	int hashconfig = 0;
4435 
4436 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
4437 		/*
4438 		 * If UDP hashing was enabled it must have been enabled for
4439 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
4440 		 * enabling any 4-tuple hash is nonsense configuration.
4441 		 */
4442 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4443 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
4444 
4445 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4446 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
4447 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4448 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
4449 	}
4450 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4451 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
4452 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4453 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
4454 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
4455 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
4456 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
4457 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
4458 
4459 	return (hashconfig);
4460 }
4461 #endif
4462 
4463 int
4464 vi_full_init(struct vi_info *vi)
4465 {
4466 	struct adapter *sc = vi->pi->adapter;
4467 	struct ifnet *ifp = vi->ifp;
4468 	uint16_t *rss;
4469 	struct sge_rxq *rxq;
4470 	int rc, i, j, hashen;
4471 #ifdef RSS
4472 	int nbuckets = rss_getnumbuckets();
4473 	int hashconfig = rss_gethashconfig();
4474 	int extra;
4475 #endif
4476 
4477 	ASSERT_SYNCHRONIZED_OP(sc);
4478 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
4479 	    ("%s: VI_INIT_DONE already", __func__));
4480 
4481 	sysctl_ctx_init(&vi->ctx);
4482 	vi->flags |= VI_SYSCTL_CTX;
4483 
4484 	/*
4485 	 * Allocate tx/rx/fl queues for this VI.
4486 	 */
4487 	rc = t4_setup_vi_queues(vi);
4488 	if (rc != 0)
4489 		goto done;	/* error message displayed already */
4490 
4491 	/*
4492 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
4493 	 */
4494 	if (vi->nrxq > vi->rss_size) {
4495 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
4496 		    "some queues will never receive traffic.\n", vi->nrxq,
4497 		    vi->rss_size);
4498 	} else if (vi->rss_size % vi->nrxq) {
4499 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
4500 		    "expect uneven traffic distribution.\n", vi->nrxq,
4501 		    vi->rss_size);
4502 	}
4503 #ifdef RSS
4504 	if (vi->nrxq != nbuckets) {
4505 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
4506 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
4507 	}
4508 #endif
4509 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
4510 	for (i = 0; i < vi->rss_size;) {
4511 #ifdef RSS
4512 		j = rss_get_indirection_to_bucket(i);
4513 		j %= vi->nrxq;
4514 		rxq = &sc->sge.rxq[vi->first_rxq + j];
4515 		rss[i++] = rxq->iq.abs_id;
4516 #else
4517 		for_each_rxq(vi, j, rxq) {
4518 			rss[i++] = rxq->iq.abs_id;
4519 			if (i == vi->rss_size)
4520 				break;
4521 		}
4522 #endif
4523 	}
4524 
4525 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
4526 	    vi->rss_size);
4527 	if (rc != 0) {
4528 		if_printf(ifp, "rss_config failed: %d\n", rc);
4529 		goto done;
4530 	}
4531 
4532 #ifdef RSS
4533 	hashen = hashconfig_to_hashen(hashconfig);
4534 
4535 	/*
4536 	 * We may have had to enable some hashes even though the global config
4537 	 * wants them disabled.  This is a potential problem that must be
4538 	 * reported to the user.
4539 	 */
4540 	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
4541 
4542 	/*
4543 	 * If we consider only the supported hash types, then the enabled hashes
4544 	 * are a superset of the requested hashes.  In other words, there cannot
4545 	 * be any supported hash that was requested but not enabled, but there
4546 	 * can be hashes that were not requested but had to be enabled.
4547 	 */
4548 	extra &= SUPPORTED_RSS_HASHTYPES;
4549 	MPASS((extra & hashconfig) == 0);
4550 
4551 	if (extra) {
4552 		if_printf(ifp,
4553 		    "global RSS config (0x%x) cannot be accommodated.\n",
4554 		    hashconfig);
4555 	}
4556 	if (extra & RSS_HASHTYPE_RSS_IPV4)
4557 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
4558 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
4559 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
4560 	if (extra & RSS_HASHTYPE_RSS_IPV6)
4561 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
4562 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
4563 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
4564 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
4565 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
4566 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
4567 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
4568 #else
4569 	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
4570 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
4571 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4572 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
4573 #endif
4574 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
4575 	if (rc != 0) {
4576 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
4577 		goto done;
4578 	}
4579 
4580 	vi->rss = rss;
4581 	vi->flags |= VI_INIT_DONE;
4582 done:
4583 	if (rc != 0)
4584 		vi_full_uninit(vi);
4585 
4586 	return (rc);
4587 }
4588 
4589 /*
4590  * Idempotent.
4591  */
4592 int
4593 vi_full_uninit(struct vi_info *vi)
4594 {
4595 	struct port_info *pi = vi->pi;
4596 	struct adapter *sc = pi->adapter;
4597 	int i;
4598 	struct sge_rxq *rxq;
4599 	struct sge_txq *txq;
4600 #ifdef TCP_OFFLOAD
4601 	struct sge_ofld_rxq *ofld_rxq;
4602 	struct sge_wrq *ofld_txq;
4603 #endif
4604 
4605 	if (vi->flags & VI_INIT_DONE) {
4606 
4607 		/* Need to quiesce queues.  */
4608 
4609 		/* XXX: Only for the first VI? */
4610 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
4611 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
4612 
4613 		for_each_txq(vi, i, txq) {
4614 			quiesce_txq(sc, txq);
4615 		}
4616 
4617 #ifdef TCP_OFFLOAD
4618 		for_each_ofld_txq(vi, i, ofld_txq) {
4619 			quiesce_wrq(sc, ofld_txq);
4620 		}
4621 #endif
4622 
4623 		for_each_rxq(vi, i, rxq) {
4624 			quiesce_iq(sc, &rxq->iq);
4625 			quiesce_fl(sc, &rxq->fl);
4626 		}
4627 
4628 #ifdef TCP_OFFLOAD
4629 		for_each_ofld_rxq(vi, i, ofld_rxq) {
4630 			quiesce_iq(sc, &ofld_rxq->iq);
4631 			quiesce_fl(sc, &ofld_rxq->fl);
4632 		}
4633 #endif
4634 		free(vi->rss, M_CXGBE);
4635 		free(vi->nm_rss, M_CXGBE);
4636 	}
4637 
4638 	t4_teardown_vi_queues(vi);
4639 	vi->flags &= ~VI_INIT_DONE;
4640 
4641 	return (0);
4642 }
4643 
4644 static void
4645 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
4646 {
4647 	struct sge_eq *eq = &txq->eq;
4648 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
4649 
4650 	(void) sc;	/* unused */
4651 
4652 #ifdef INVARIANTS
4653 	TXQ_LOCK(txq);
4654 	MPASS((eq->flags & EQ_ENABLED) == 0);
4655 	TXQ_UNLOCK(txq);
4656 #endif
4657 
4658 	/* Wait for the mp_ring to empty. */
4659 	while (!mp_ring_is_idle(txq->r)) {
4660 		mp_ring_check_drainage(txq->r, 0);
4661 		pause("rquiesce", 1);
4662 	}
4663 
4664 	/* Then wait for the hardware to finish. */
4665 	while (spg->cidx != htobe16(eq->pidx))
4666 		pause("equiesce", 1);
4667 
4668 	/* Finally, wait for the driver to reclaim all descriptors. */
4669 	while (eq->cidx != eq->pidx)
4670 		pause("dquiesce", 1);
4671 }
4672 
4673 static void
4674 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
4675 {
4676 
4677 	/* XXXTX */
4678 }
4679 
4680 static void
4681 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
4682 {
4683 	(void) sc;	/* unused */
4684 
4685 	/* Synchronize with the interrupt handler */
4686 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
4687 		pause("iqfree", 1);
4688 }
4689 
4690 static void
4691 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
4692 {
4693 	mtx_lock(&sc->sfl_lock);
4694 	FL_LOCK(fl);
4695 	fl->flags |= FL_DOOMED;
4696 	FL_UNLOCK(fl);
4697 	callout_stop(&sc->sfl_callout);
4698 	mtx_unlock(&sc->sfl_lock);
4699 
4700 	KASSERT((fl->flags & FL_STARVING) == 0,
4701 	    ("%s: still starving", __func__));
4702 }
4703 
4704 static int
4705 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
4706     driver_intr_t *handler, void *arg, char *name)
4707 {
4708 	int rc;
4709 
4710 	irq->rid = rid;
4711 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
4712 	    RF_SHAREABLE | RF_ACTIVE);
4713 	if (irq->res == NULL) {
4714 		device_printf(sc->dev,
4715 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
4716 		return (ENOMEM);
4717 	}
4718 
4719 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
4720 	    NULL, handler, arg, &irq->tag);
4721 	if (rc != 0) {
4722 		device_printf(sc->dev,
4723 		    "failed to setup interrupt for rid %d, name %s: %d\n",
4724 		    rid, name, rc);
4725 	} else if (name)
4726 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
4727 
4728 	return (rc);
4729 }
4730 
4731 static int
4732 t4_free_irq(struct adapter *sc, struct irq *irq)
4733 {
4734 	if (irq->tag)
4735 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
4736 	if (irq->res)
4737 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
4738 
4739 	bzero(irq, sizeof(*irq));
4740 
4741 	return (0);
4742 }
4743 
4744 static void
4745 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
4746 {
4747 
4748 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
4749 	t4_get_regs(sc, buf, regs->len);
4750 }
4751 
4752 #define	A_PL_INDIR_CMD	0x1f8
4753 
4754 #define	S_PL_AUTOINC	31
4755 #define	M_PL_AUTOINC	0x1U
4756 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
4757 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
4758 
4759 #define	S_PL_VFID	20
4760 #define	M_PL_VFID	0xffU
4761 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
4762 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
4763 
4764 #define	S_PL_ADDR	0
4765 #define	M_PL_ADDR	0xfffffU
4766 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
4767 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
4768 
4769 #define	A_PL_INDIR_DATA	0x1fc
4770 
4771 static uint64_t
4772 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
4773 {
4774 	u32 stats[2];
4775 
4776 	mtx_assert(&sc->reg_lock, MA_OWNED);
4777 	if (sc->flags & IS_VF) {
4778 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
4779 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
4780 	} else {
4781 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4782 		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4783 		    V_PL_ADDR(VF_MPS_REG(reg)));
4784 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
4785 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
4786 	}
4787 	return (((uint64_t)stats[1]) << 32 | stats[0]);
4788 }
4789 
4790 static void
4791 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
4792     struct fw_vi_stats_vf *stats)
4793 {
4794 
4795 #define GET_STAT(name) \
4796 	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
4797 
4798 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
4799 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
4800 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
4801 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
4802 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
4803 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
4804 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
4805 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
4806 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
4807 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
4808 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
4809 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
4810 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
4811 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
4812 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
4813 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
4814 
4815 #undef GET_STAT
4816 }
4817 
4818 static void
4819 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
4820 {
4821 	int reg;
4822 
4823 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4824 	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4825 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
4826 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
4827 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
4828 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
4829 }
4830 
4831 static void
4832 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
4833 {
4834 	struct timeval tv;
4835 	const struct timeval interval = {0, 250000};	/* 250ms */
4836 
4837 	if (!(vi->flags & VI_INIT_DONE))
4838 		return;
4839 
4840 	getmicrotime(&tv);
4841 	timevalsub(&tv, &interval);
4842 	if (timevalcmp(&tv, &vi->last_refreshed, <))
4843 		return;
4844 
4845 	mtx_lock(&sc->reg_lock);
4846 	t4_get_vi_stats(sc, vi->viid, &vi->stats);
4847 	getmicrotime(&vi->last_refreshed);
4848 	mtx_unlock(&sc->reg_lock);
4849 }
4850 
4851 static void
4852 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
4853 {
4854 	int i;
4855 	u_int v, tnl_cong_drops;
4856 	struct timeval tv;
4857 	const struct timeval interval = {0, 250000};	/* 250ms */
4858 
4859 	getmicrotime(&tv);
4860 	timevalsub(&tv, &interval);
4861 	if (timevalcmp(&tv, &pi->last_refreshed, <))
4862 		return;
4863 
4864 	tnl_cong_drops = 0;
4865 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
4866 	for (i = 0; i < sc->chip_params->nchan; i++) {
4867 		if (pi->rx_chan_map & (1 << i)) {
4868 			mtx_lock(&sc->reg_lock);
4869 			t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
4870 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
4871 			mtx_unlock(&sc->reg_lock);
4872 			tnl_cong_drops += v;
4873 		}
4874 	}
4875 	pi->tnl_cong_drops = tnl_cong_drops;
4876 	getmicrotime(&pi->last_refreshed);
4877 }
4878 
4879 static void
4880 cxgbe_tick(void *arg)
4881 {
4882 	struct port_info *pi = arg;
4883 	struct adapter *sc = pi->adapter;
4884 
4885 	PORT_LOCK_ASSERT_OWNED(pi);
4886 	cxgbe_refresh_stats(sc, pi);
4887 
4888 	callout_schedule(&pi->tick, hz);
4889 }
4890 
4891 void
4892 vi_tick(void *arg)
4893 {
4894 	struct vi_info *vi = arg;
4895 	struct adapter *sc = vi->pi->adapter;
4896 
4897 	vi_refresh_stats(sc, vi);
4898 
4899 	callout_schedule(&vi->tick, hz);
4900 }
4901 
4902 static void
4903 cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
4904 {
4905 	struct ifnet *vlan;
4906 
4907 	if (arg != ifp || ifp->if_type != IFT_ETHER)
4908 		return;
4909 
4910 	vlan = VLAN_DEVAT(ifp, vid);
4911 	VLAN_SETCOOKIE(vlan, ifp);
4912 }
4913 
4914 /*
4915  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
4916  */
4917 static char *caps_decoder[] = {
4918 	"\20\001IPMI\002NCSI",				/* 0: NBM */
4919 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
4920 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
4921 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
4922 	    "\006HASHFILTER\007ETHOFLD",
4923 	"\20\001TOE",					/* 4: TOE */
4924 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
4925 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
4926 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
4927 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
4928 	    "\007T10DIF"
4929 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
4930 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
4931 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
4932 		    "\004PO_INITIATOR\005PO_TARGET",
4933 };
4934 
4935 void
4936 t4_sysctls(struct adapter *sc)
4937 {
4938 	struct sysctl_ctx_list *ctx;
4939 	struct sysctl_oid *oid;
4940 	struct sysctl_oid_list *children, *c0;
4941 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
4942 
4943 	ctx = device_get_sysctl_ctx(sc->dev);
4944 
4945 	/*
4946 	 * dev.t4nex.X.
4947 	 */
4948 	oid = device_get_sysctl_tree(sc->dev);
4949 	c0 = children = SYSCTL_CHILDREN(oid);
4950 
4951 	sc->sc_do_rxcopy = 1;
4952 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
4953 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
4954 
4955 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
4956 	    sc->params.nports, "# of ports");
4957 
4958 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
4959 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
4960 	    sysctl_bitfield, "A", "available doorbells");
4961 
4962 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
4963 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
4964 
4965 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
4966 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
4967 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
4968 	    "interrupt holdoff timer values (us)");
4969 
4970 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
4971 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
4972 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
4973 	    "interrupt holdoff packet counter values");
4974 
4975 	t4_sge_sysctls(sc, ctx, children);
4976 
4977 	sc->lro_timeout = 100;
4978 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
4979 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
4980 
4981 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
4982 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
4983 
4984 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
4985 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
4986 
4987 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
4988 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
4989 
4990 	if (sc->flags & IS_VF)
4991 		return;
4992 
4993 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
4994 	    NULL, chip_rev(sc), "chip hardware revision");
4995 
4996 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
4997 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
4998 
4999 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5000 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5001 
5002 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5003 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5004 
5005 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5006 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5007 
5008 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5009 	    sc->er_version, 0, "expansion ROM version");
5010 
5011 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5012 	    sc->bs_version, 0, "bootstrap firmware version");
5013 
5014 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5015 	    NULL, sc->params.scfg_vers, "serial config version");
5016 
5017 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5018 	    NULL, sc->params.vpd_vers, "VPD version");
5019 
5020 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5021 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5022 
5023 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5024 	    sc->cfcsum, "config file checksum");
5025 
5026 #define SYSCTL_CAP(name, n, text) \
5027 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5028 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
5029 	    sysctl_bitfield, "A", "available " text " capabilities")
5030 
5031 	SYSCTL_CAP(nbmcaps, 0, "NBM");
5032 	SYSCTL_CAP(linkcaps, 1, "link");
5033 	SYSCTL_CAP(switchcaps, 2, "switch");
5034 	SYSCTL_CAP(niccaps, 3, "NIC");
5035 	SYSCTL_CAP(toecaps, 4, "TCP offload");
5036 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
5037 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5038 	SYSCTL_CAP(cryptocaps, 7, "crypto");
5039 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
5040 #undef SYSCTL_CAP
5041 
5042 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5043 	    NULL, sc->tids.nftids, "number of filters");
5044 
5045 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5046 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5047 	    "chip temperature (in Celsius)");
5048 
5049 #ifdef SBUF_DRAIN
5050 	/*
5051 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
5052 	 */
5053 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5054 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5055 	    "logs and miscellaneous information");
5056 	children = SYSCTL_CHILDREN(oid);
5057 
5058 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
5059 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5060 	    sysctl_cctrl, "A", "congestion control");
5061 
5062 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
5063 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5064 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
5065 
5066 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
5067 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
5068 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
5069 
5070 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
5071 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
5072 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
5073 
5074 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
5075 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
5076 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
5077 
5078 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
5079 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
5080 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
5081 
5082 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
5083 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
5084 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
5085 
5086 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
5087 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5088 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
5089 	    "A", "CIM logic analyzer");
5090 
5091 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
5092 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5093 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
5094 
5095 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
5096 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
5097 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
5098 
5099 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
5100 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
5101 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
5102 
5103 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
5104 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
5105 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
5106 
5107 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5108 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5109 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5110 
5111 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5112 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5113 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5114 
5115 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5116 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5117 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5118 
5119 	if (chip_id(sc) > CHELSIO_T4) {
5120 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5121 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5122 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5123 
5124 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5125 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5126 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5127 	}
5128 
5129 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5130 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5131 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5132 
5133 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5134 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5135 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
5136 
5137 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5138 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5139 	    sysctl_cpl_stats, "A", "CPL statistics");
5140 
5141 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5142 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5143 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5144 
5145 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5146 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5147 	    sysctl_devlog, "A", "firmware's device log");
5148 
5149 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5150 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5151 	    sysctl_fcoe_stats, "A", "FCoE statistics");
5152 
5153 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5154 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5155 	    sysctl_hw_sched, "A", "hardware scheduler ");
5156 
5157 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5158 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5159 	    sysctl_l2t, "A", "hardware L2 table");
5160 
5161 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5162 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5163 	    sysctl_lb_stats, "A", "loopback statistics");
5164 
5165 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5166 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5167 	    sysctl_meminfo, "A", "memory regions");
5168 
5169 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5170 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5171 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5172 	    "A", "MPS TCAM entries");
5173 
5174 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5175 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5176 	    sysctl_path_mtus, "A", "path MTUs");
5177 
5178 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5179 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5180 	    sysctl_pm_stats, "A", "PM statistics");
5181 
5182 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5183 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5184 	    sysctl_rdma_stats, "A", "RDMA statistics");
5185 
5186 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5187 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5188 	    sysctl_tcp_stats, "A", "TCP statistics");
5189 
5190 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5191 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5192 	    sysctl_tids, "A", "TID information");
5193 
5194 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5195 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5196 	    sysctl_tp_err_stats, "A", "TP error statistics");
5197 
5198 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5199 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5200 	    "TP logic analyzer event capture mask");
5201 
5202 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5203 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5204 	    sysctl_tp_la, "A", "TP logic analyzer");
5205 
5206 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5207 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5208 	    sysctl_tx_rate, "A", "Tx rate");
5209 
5210 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5211 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5212 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5213 
5214 	if (chip_id(sc) >= CHELSIO_T5) {
5215 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5216 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5217 		    sysctl_wcwr_stats, "A", "write combined work requests");
5218 	}
5219 #endif
5220 
5221 #ifdef TCP_OFFLOAD
5222 	if (is_offload(sc)) {
5223 		/*
5224 		 * dev.t4nex.X.toe.
5225 		 */
5226 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5227 		    NULL, "TOE parameters");
5228 		children = SYSCTL_CHILDREN(oid);
5229 
5230 		sc->tt.sndbuf = 256 * 1024;
5231 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5232 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
5233 
5234 		sc->tt.ddp = 0;
5235 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5236 		    &sc->tt.ddp, 0, "DDP allowed");
5237 
5238 		sc->tt.rx_coalesce = 1;
5239 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5240 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5241 
5242 		sc->tt.tx_align = 1;
5243 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5244 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5245 
5246 		sc->tt.tx_zcopy = 0;
5247 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
5248 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
5249 		    "Enable zero-copy aio_write(2)");
5250 
5251 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
5252 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
5253 		    "TP timer tick (us)");
5254 
5255 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
5256 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
5257 		    "TCP timestamp tick (us)");
5258 
5259 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5260 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5261 		    "DACK tick (us)");
5262 
5263 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5264 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5265 		    "IU", "DACK timer (us)");
5266 
5267 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5268 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5269 		    sysctl_tp_timer, "LU", "Retransmit min (us)");
5270 
5271 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5272 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5273 		    sysctl_tp_timer, "LU", "Retransmit max (us)");
5274 
5275 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5276 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5277 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
5278 
5279 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5280 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5281 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
5282 
5283 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5284 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5285 		    sysctl_tp_timer, "LU", "Keepidle idle timer (us)");
5286 
5287 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_intvl",
5288 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
5289 		    sysctl_tp_timer, "LU", "Keepidle interval (us)");
5290 
5291 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
5292 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
5293 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
5294 
5295 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
5296 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
5297 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
5298 	}
5299 #endif
5300 }
5301 
5302 void
5303 vi_sysctls(struct vi_info *vi)
5304 {
5305 	struct sysctl_ctx_list *ctx;
5306 	struct sysctl_oid *oid;
5307 	struct sysctl_oid_list *children;
5308 
5309 	ctx = device_get_sysctl_ctx(vi->dev);
5310 
5311 	/*
5312 	 * dev.v?(cxgbe|cxl).X.
5313 	 */
5314 	oid = device_get_sysctl_tree(vi->dev);
5315 	children = SYSCTL_CHILDREN(oid);
5316 
5317 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
5318 	    vi->viid, "VI identifer");
5319 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
5320 	    &vi->nrxq, 0, "# of rx queues");
5321 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5322 	    &vi->ntxq, 0, "# of tx queues");
5323 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5324 	    &vi->first_rxq, 0, "index of first rx queue");
5325 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5326 	    &vi->first_txq, 0, "index of first tx queue");
5327 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
5328 	    vi->rss_size, "size of RSS indirection table");
5329 
5330 	if (IS_MAIN_VI(vi)) {
5331 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
5332 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
5333 		    "Reserve queue 0 for non-flowid packets");
5334 	}
5335 
5336 #ifdef TCP_OFFLOAD
5337 	if (vi->nofldrxq != 0) {
5338 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5339 		    &vi->nofldrxq, 0,
5340 		    "# of rx queues for offloaded TCP connections");
5341 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5342 		    &vi->nofldtxq, 0,
5343 		    "# of tx queues for offloaded TCP connections");
5344 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5345 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
5346 		    "index of first TOE rx queue");
5347 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5348 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
5349 		    "index of first TOE tx queue");
5350 	}
5351 #endif
5352 #ifdef DEV_NETMAP
5353 	if (vi->nnmrxq != 0) {
5354 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5355 		    &vi->nnmrxq, 0, "# of netmap rx queues");
5356 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5357 		    &vi->nnmtxq, 0, "# of netmap tx queues");
5358 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5359 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
5360 		    "index of first netmap rx queue");
5361 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
5362 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
5363 		    "index of first netmap tx queue");
5364 	}
5365 #endif
5366 
5367 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
5368 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
5369 	    "holdoff timer index");
5370 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
5371 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
5372 	    "holdoff packet counter index");
5373 
5374 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
5375 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
5376 	    "rx queue size");
5377 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
5378 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
5379 	    "tx queue size");
5380 }
5381 
5382 static void
5383 cxgbe_sysctls(struct port_info *pi)
5384 {
5385 	struct sysctl_ctx_list *ctx;
5386 	struct sysctl_oid *oid;
5387 	struct sysctl_oid_list *children, *children2;
5388 	struct adapter *sc = pi->adapter;
5389 	int i;
5390 	char name[16];
5391 
5392 	ctx = device_get_sysctl_ctx(pi->dev);
5393 
5394 	/*
5395 	 * dev.cxgbe.X.
5396 	 */
5397 	oid = device_get_sysctl_tree(pi->dev);
5398 	children = SYSCTL_CHILDREN(oid);
5399 
5400 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
5401 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
5402 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
5403 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
5404 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
5405 		    "PHY temperature (in Celsius)");
5406 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
5407 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
5408 		    "PHY firmware version");
5409 	}
5410 
5411 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
5412 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
5413 	    "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
5414 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
5415 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
5416 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
5417 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
5418 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
5419 	    "autonegotiation (-1 = not supported)");
5420 
5421 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
5422 	    port_top_speed(pi), "max speed (in Gbps)");
5423 
5424 	if (sc->flags & IS_VF)
5425 		return;
5426 
5427 	/*
5428 	 * dev.(cxgbe|cxl).X.tc.
5429 	 */
5430 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
5431 	    "Tx scheduler traffic classes (cl_rl)");
5432 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
5433 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
5434 
5435 		snprintf(name, sizeof(name), "%d", i);
5436 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
5437 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
5438 		    "traffic class"));
5439 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
5440 		    &tc->flags, 0, "flags");
5441 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
5442 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
5443 #ifdef SBUF_DRAIN
5444 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
5445 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
5446 		    sysctl_tc_params, "A", "traffic class parameters");
5447 #endif
5448 	}
5449 
5450 	/*
5451 	 * dev.cxgbe.X.stats.
5452 	 */
5453 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
5454 	    NULL, "port statistics");
5455 	children = SYSCTL_CHILDREN(oid);
5456 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
5457 	    &pi->tx_parse_error, 0,
5458 	    "# of tx packets with invalid length or # of segments");
5459 
5460 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
5461 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
5462 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
5463 	    sysctl_handle_t4_reg64, "QU", desc)
5464 
5465 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
5466 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
5467 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
5468 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
5469 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
5470 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
5471 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
5472 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
5473 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
5474 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
5475 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
5476 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
5477 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
5478 	    "# of tx frames in this range",
5479 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
5480 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
5481 	    "# of tx frames in this range",
5482 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
5483 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
5484 	    "# of tx frames in this range",
5485 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
5486 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
5487 	    "# of tx frames in this range",
5488 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
5489 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
5490 	    "# of tx frames in this range",
5491 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
5492 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
5493 	    "# of tx frames in this range",
5494 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
5495 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
5496 	    "# of tx frames in this range",
5497 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
5498 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
5499 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
5500 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
5501 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
5502 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
5503 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
5504 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5505 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5506 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5507 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5508 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5509 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5510 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5511 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5512 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5513 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5514 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5515 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5516 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5517 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5518 
5519 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5520 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5521 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5522 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5523 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5524 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5525 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5526 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5527 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5528 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5529 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5530 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5531 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5532 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5533 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5534 	    "# of frames received with bad FCS",
5535 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5536 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5537 	    "# of frames received with length error",
5538 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5539 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5540 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5541 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5542 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5543 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5544 	    "# of rx frames in this range",
5545 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5546 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5547 	    "# of rx frames in this range",
5548 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5549 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5550 	    "# of rx frames in this range",
5551 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5552 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5553 	    "# of rx frames in this range",
5554 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5555 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5556 	    "# of rx frames in this range",
5557 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5558 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5559 	    "# of rx frames in this range",
5560 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5561 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5562 	    "# of rx frames in this range",
5563 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5564 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5565 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5566 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5567 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5568 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5569 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5570 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5571 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5572 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5573 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5574 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5575 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5576 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5577 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5578 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5579 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5580 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5581 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5582 
5583 #undef SYSCTL_ADD_T4_REG64
5584 
5585 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5586 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5587 	    &pi->stats.name, desc)
5588 
5589 	/* We get these from port_stats and they may be stale by up to 1s */
5590 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5591 	    "# drops due to buffer-group 0 overflows");
5592 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5593 	    "# drops due to buffer-group 1 overflows");
5594 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5595 	    "# drops due to buffer-group 2 overflows");
5596 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5597 	    "# drops due to buffer-group 3 overflows");
5598 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5599 	    "# of buffer-group 0 truncated packets");
5600 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5601 	    "# of buffer-group 1 truncated packets");
5602 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5603 	    "# of buffer-group 2 truncated packets");
5604 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5605 	    "# of buffer-group 3 truncated packets");
5606 
5607 #undef SYSCTL_ADD_T4_PORTSTAT
5608 }
5609 
5610 static int
5611 sysctl_int_array(SYSCTL_HANDLER_ARGS)
5612 {
5613 	int rc, *i, space = 0;
5614 	struct sbuf sb;
5615 
5616 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
5617 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
5618 		if (space)
5619 			sbuf_printf(&sb, " ");
5620 		sbuf_printf(&sb, "%d", *i);
5621 		space = 1;
5622 	}
5623 	rc = sbuf_finish(&sb);
5624 	sbuf_delete(&sb);
5625 	return (rc);
5626 }
5627 
5628 static int
5629 sysctl_bitfield(SYSCTL_HANDLER_ARGS)
5630 {
5631 	int rc;
5632 	struct sbuf *sb;
5633 
5634 	rc = sysctl_wire_old_buffer(req, 0);
5635 	if (rc != 0)
5636 		return(rc);
5637 
5638 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5639 	if (sb == NULL)
5640 		return (ENOMEM);
5641 
5642 	sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
5643 	rc = sbuf_finish(sb);
5644 	sbuf_delete(sb);
5645 
5646 	return (rc);
5647 }
5648 
5649 static int
5650 sysctl_btphy(SYSCTL_HANDLER_ARGS)
5651 {
5652 	struct port_info *pi = arg1;
5653 	int op = arg2;
5654 	struct adapter *sc = pi->adapter;
5655 	u_int v;
5656 	int rc;
5657 
5658 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
5659 	if (rc)
5660 		return (rc);
5661 	/* XXX: magic numbers */
5662 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
5663 	    &v);
5664 	end_synchronized_op(sc, 0);
5665 	if (rc)
5666 		return (rc);
5667 	if (op == 0)
5668 		v /= 256;
5669 
5670 	rc = sysctl_handle_int(oidp, &v, 0, req);
5671 	return (rc);
5672 }
5673 
5674 static int
5675 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
5676 {
5677 	struct vi_info *vi = arg1;
5678 	int rc, val;
5679 
5680 	val = vi->rsrv_noflowq;
5681 	rc = sysctl_handle_int(oidp, &val, 0, req);
5682 	if (rc != 0 || req->newptr == NULL)
5683 		return (rc);
5684 
5685 	if ((val >= 1) && (vi->ntxq > 1))
5686 		vi->rsrv_noflowq = 1;
5687 	else
5688 		vi->rsrv_noflowq = 0;
5689 
5690 	return (rc);
5691 }
5692 
5693 static int
5694 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
5695 {
5696 	struct vi_info *vi = arg1;
5697 	struct adapter *sc = vi->pi->adapter;
5698 	int idx, rc, i;
5699 	struct sge_rxq *rxq;
5700 #ifdef TCP_OFFLOAD
5701 	struct sge_ofld_rxq *ofld_rxq;
5702 #endif
5703 	uint8_t v;
5704 
5705 	idx = vi->tmr_idx;
5706 
5707 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5708 	if (rc != 0 || req->newptr == NULL)
5709 		return (rc);
5710 
5711 	if (idx < 0 || idx >= SGE_NTIMERS)
5712 		return (EINVAL);
5713 
5714 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5715 	    "t4tmr");
5716 	if (rc)
5717 		return (rc);
5718 
5719 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
5720 	for_each_rxq(vi, i, rxq) {
5721 #ifdef atomic_store_rel_8
5722 		atomic_store_rel_8(&rxq->iq.intr_params, v);
5723 #else
5724 		rxq->iq.intr_params = v;
5725 #endif
5726 	}
5727 #ifdef TCP_OFFLOAD
5728 	for_each_ofld_rxq(vi, i, ofld_rxq) {
5729 #ifdef atomic_store_rel_8
5730 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
5731 #else
5732 		ofld_rxq->iq.intr_params = v;
5733 #endif
5734 	}
5735 #endif
5736 	vi->tmr_idx = idx;
5737 
5738 	end_synchronized_op(sc, LOCK_HELD);
5739 	return (0);
5740 }
5741 
5742 static int
5743 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
5744 {
5745 	struct vi_info *vi = arg1;
5746 	struct adapter *sc = vi->pi->adapter;
5747 	int idx, rc;
5748 
5749 	idx = vi->pktc_idx;
5750 
5751 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5752 	if (rc != 0 || req->newptr == NULL)
5753 		return (rc);
5754 
5755 	if (idx < -1 || idx >= SGE_NCOUNTERS)
5756 		return (EINVAL);
5757 
5758 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5759 	    "t4pktc");
5760 	if (rc)
5761 		return (rc);
5762 
5763 	if (vi->flags & VI_INIT_DONE)
5764 		rc = EBUSY; /* cannot be changed once the queues are created */
5765 	else
5766 		vi->pktc_idx = idx;
5767 
5768 	end_synchronized_op(sc, LOCK_HELD);
5769 	return (rc);
5770 }
5771 
5772 static int
5773 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
5774 {
5775 	struct vi_info *vi = arg1;
5776 	struct adapter *sc = vi->pi->adapter;
5777 	int qsize, rc;
5778 
5779 	qsize = vi->qsize_rxq;
5780 
5781 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5782 	if (rc != 0 || req->newptr == NULL)
5783 		return (rc);
5784 
5785 	if (qsize < 128 || (qsize & 7))
5786 		return (EINVAL);
5787 
5788 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5789 	    "t4rxqs");
5790 	if (rc)
5791 		return (rc);
5792 
5793 	if (vi->flags & VI_INIT_DONE)
5794 		rc = EBUSY; /* cannot be changed once the queues are created */
5795 	else
5796 		vi->qsize_rxq = qsize;
5797 
5798 	end_synchronized_op(sc, LOCK_HELD);
5799 	return (rc);
5800 }
5801 
5802 static int
5803 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
5804 {
5805 	struct vi_info *vi = arg1;
5806 	struct adapter *sc = vi->pi->adapter;
5807 	int qsize, rc;
5808 
5809 	qsize = vi->qsize_txq;
5810 
5811 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5812 	if (rc != 0 || req->newptr == NULL)
5813 		return (rc);
5814 
5815 	if (qsize < 128 || qsize > 65536)
5816 		return (EINVAL);
5817 
5818 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5819 	    "t4txqs");
5820 	if (rc)
5821 		return (rc);
5822 
5823 	if (vi->flags & VI_INIT_DONE)
5824 		rc = EBUSY; /* cannot be changed once the queues are created */
5825 	else
5826 		vi->qsize_txq = qsize;
5827 
5828 	end_synchronized_op(sc, LOCK_HELD);
5829 	return (rc);
5830 }
5831 
5832 static int
5833 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
5834 {
5835 	struct port_info *pi = arg1;
5836 	struct adapter *sc = pi->adapter;
5837 	struct link_config *lc = &pi->link_cfg;
5838 	int rc;
5839 
5840 	if (req->newptr == NULL) {
5841 		struct sbuf *sb;
5842 		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
5843 
5844 		rc = sysctl_wire_old_buffer(req, 0);
5845 		if (rc != 0)
5846 			return(rc);
5847 
5848 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5849 		if (sb == NULL)
5850 			return (ENOMEM);
5851 
5852 		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
5853 		rc = sbuf_finish(sb);
5854 		sbuf_delete(sb);
5855 	} else {
5856 		char s[2];
5857 		int n;
5858 
5859 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
5860 		s[1] = 0;
5861 
5862 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5863 		if (rc != 0)
5864 			return(rc);
5865 
5866 		if (s[1] != 0)
5867 			return (EINVAL);
5868 		if (s[0] < '0' || s[0] > '9')
5869 			return (EINVAL);	/* not a number */
5870 		n = s[0] - '0';
5871 		if (n & ~(PAUSE_TX | PAUSE_RX))
5872 			return (EINVAL);	/* some other bit is set too */
5873 
5874 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5875 		    "t4PAUSE");
5876 		if (rc)
5877 			return (rc);
5878 		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
5879 			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
5880 			lc->requested_fc |= n;
5881 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5882 			if (rc == 0) {
5883 				lc->fc = lc->requested_fc;
5884 			}
5885 		}
5886 		end_synchronized_op(sc, 0);
5887 	}
5888 
5889 	return (rc);
5890 }
5891 
5892 static int
5893 sysctl_fec(SYSCTL_HANDLER_ARGS)
5894 {
5895 	struct port_info *pi = arg1;
5896 	struct adapter *sc = pi->adapter;
5897 	struct link_config *lc = &pi->link_cfg;
5898 	int rc;
5899 
5900 	if (req->newptr == NULL) {
5901 		struct sbuf *sb;
5902 		static char *bits = "\20\1RS\2BASER_RS\3RESERVED";
5903 
5904 		rc = sysctl_wire_old_buffer(req, 0);
5905 		if (rc != 0)
5906 			return(rc);
5907 
5908 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5909 		if (sb == NULL)
5910 			return (ENOMEM);
5911 
5912 		sbuf_printf(sb, "%b", lc->fec & M_FW_PORT_CAP_FEC, bits);
5913 		rc = sbuf_finish(sb);
5914 		sbuf_delete(sb);
5915 	} else {
5916 		char s[2];
5917 		int n;
5918 
5919 		s[0] = '0' + (lc->requested_fec & M_FW_PORT_CAP_FEC);
5920 		s[1] = 0;
5921 
5922 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5923 		if (rc != 0)
5924 			return(rc);
5925 
5926 		if (s[1] != 0)
5927 			return (EINVAL);
5928 		if (s[0] < '0' || s[0] > '9')
5929 			return (EINVAL);	/* not a number */
5930 		n = s[0] - '0';
5931 		if (n & ~M_FW_PORT_CAP_FEC)
5932 			return (EINVAL);	/* some other bit is set too */
5933 
5934 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5935 		    "t4fec");
5936 		if (rc)
5937 			return (rc);
5938 		if ((lc->requested_fec & M_FW_PORT_CAP_FEC) != n) {
5939 			lc->requested_fec = n &
5940 			    G_FW_PORT_CAP_FEC(lc->supported);
5941 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5942 			if (rc == 0) {
5943 				lc->fec = lc->requested_fec;
5944 			}
5945 		}
5946 		end_synchronized_op(sc, 0);
5947 	}
5948 
5949 	return (rc);
5950 }
5951 
5952 static int
5953 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
5954 {
5955 	struct port_info *pi = arg1;
5956 	struct adapter *sc = pi->adapter;
5957 	struct link_config *lc = &pi->link_cfg;
5958 	int rc, val, old;
5959 
5960 	if (lc->supported & FW_PORT_CAP_ANEG)
5961 		val = lc->requested_aneg == AUTONEG_ENABLE ? 1 : 0;
5962 	else
5963 		val = -1;
5964 	rc = sysctl_handle_int(oidp, &val, 0, req);
5965 	if (rc != 0 || req->newptr == NULL)
5966 		return (rc);
5967 	if ((lc->supported & FW_PORT_CAP_ANEG) == 0)
5968 		return (ENOTSUP);
5969 
5970 	if (val == 0)
5971 		val = AUTONEG_DISABLE;
5972 	else if (val == 1)
5973 		val = AUTONEG_ENABLE;
5974 	else
5975 		return (EINVAL);
5976 	if (lc->requested_aneg == val)
5977 		return (0);	/* no change */
5978 
5979 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5980 	    "t4aneg");
5981 	if (rc)
5982 		return (rc);
5983 	old = lc->requested_aneg;
5984 	lc->requested_aneg = val;
5985 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5986 	if (rc != 0)
5987 		lc->requested_aneg = old;
5988 	end_synchronized_op(sc, 0);
5989 	return (rc);
5990 }
5991 
5992 static int
5993 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
5994 {
5995 	struct adapter *sc = arg1;
5996 	int reg = arg2;
5997 	uint64_t val;
5998 
5999 	val = t4_read_reg64(sc, reg);
6000 
6001 	return (sysctl_handle_64(oidp, &val, 0, req));
6002 }
6003 
6004 static int
6005 sysctl_temperature(SYSCTL_HANDLER_ARGS)
6006 {
6007 	struct adapter *sc = arg1;
6008 	int rc, t;
6009 	uint32_t param, val;
6010 
6011 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
6012 	if (rc)
6013 		return (rc);
6014 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6015 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
6016 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
6017 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6018 	end_synchronized_op(sc, 0);
6019 	if (rc)
6020 		return (rc);
6021 
6022 	/* unknown is returned as 0 but we display -1 in that case */
6023 	t = val == 0 ? -1 : val;
6024 
6025 	rc = sysctl_handle_int(oidp, &t, 0, req);
6026 	return (rc);
6027 }
6028 
6029 #ifdef SBUF_DRAIN
6030 static int
6031 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
6032 {
6033 	struct adapter *sc = arg1;
6034 	struct sbuf *sb;
6035 	int rc, i;
6036 	uint16_t incr[NMTUS][NCCTRL_WIN];
6037 	static const char *dec_fac[] = {
6038 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
6039 		"0.9375"
6040 	};
6041 
6042 	rc = sysctl_wire_old_buffer(req, 0);
6043 	if (rc != 0)
6044 		return (rc);
6045 
6046 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6047 	if (sb == NULL)
6048 		return (ENOMEM);
6049 
6050 	t4_read_cong_tbl(sc, incr);
6051 
6052 	for (i = 0; i < NCCTRL_WIN; ++i) {
6053 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
6054 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
6055 		    incr[5][i], incr[6][i], incr[7][i]);
6056 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
6057 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
6058 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
6059 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
6060 	}
6061 
6062 	rc = sbuf_finish(sb);
6063 	sbuf_delete(sb);
6064 
6065 	return (rc);
6066 }
6067 
6068 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
6069 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
6070 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
6071 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
6072 };
6073 
6074 static int
6075 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
6076 {
6077 	struct adapter *sc = arg1;
6078 	struct sbuf *sb;
6079 	int rc, i, n, qid = arg2;
6080 	uint32_t *buf, *p;
6081 	char *qtype;
6082 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
6083 
6084 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
6085 	    ("%s: bad qid %d\n", __func__, qid));
6086 
6087 	if (qid < CIM_NUM_IBQ) {
6088 		/* inbound queue */
6089 		qtype = "IBQ";
6090 		n = 4 * CIM_IBQ_SIZE;
6091 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6092 		rc = t4_read_cim_ibq(sc, qid, buf, n);
6093 	} else {
6094 		/* outbound queue */
6095 		qtype = "OBQ";
6096 		qid -= CIM_NUM_IBQ;
6097 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
6098 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6099 		rc = t4_read_cim_obq(sc, qid, buf, n);
6100 	}
6101 
6102 	if (rc < 0) {
6103 		rc = -rc;
6104 		goto done;
6105 	}
6106 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
6107 
6108 	rc = sysctl_wire_old_buffer(req, 0);
6109 	if (rc != 0)
6110 		goto done;
6111 
6112 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6113 	if (sb == NULL) {
6114 		rc = ENOMEM;
6115 		goto done;
6116 	}
6117 
6118 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
6119 	for (i = 0, p = buf; i < n; i += 16, p += 4)
6120 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
6121 		    p[2], p[3]);
6122 
6123 	rc = sbuf_finish(sb);
6124 	sbuf_delete(sb);
6125 done:
6126 	free(buf, M_CXGBE);
6127 	return (rc);
6128 }
6129 
6130 static int
6131 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
6132 {
6133 	struct adapter *sc = arg1;
6134 	u_int cfg;
6135 	struct sbuf *sb;
6136 	uint32_t *buf, *p;
6137 	int rc;
6138 
6139 	MPASS(chip_id(sc) <= CHELSIO_T5);
6140 
6141 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6142 	if (rc != 0)
6143 		return (rc);
6144 
6145 	rc = sysctl_wire_old_buffer(req, 0);
6146 	if (rc != 0)
6147 		return (rc);
6148 
6149 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6150 	if (sb == NULL)
6151 		return (ENOMEM);
6152 
6153 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6154 	    M_ZERO | M_WAITOK);
6155 
6156 	rc = -t4_cim_read_la(sc, buf, NULL);
6157 	if (rc != 0)
6158 		goto done;
6159 
6160 	sbuf_printf(sb, "Status   Data      PC%s",
6161 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6162 	    "     LS0Stat  LS0Addr             LS0Data");
6163 
6164 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
6165 		if (cfg & F_UPDBGLACAPTPCONLY) {
6166 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
6167 			    p[6], p[7]);
6168 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
6169 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
6170 			    p[4] & 0xff, p[5] >> 8);
6171 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
6172 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6173 			    p[1] & 0xf, p[2] >> 4);
6174 		} else {
6175 			sbuf_printf(sb,
6176 			    "\n  %02x   %x%07x %x%07x %08x %08x "
6177 			    "%08x%08x%08x%08x",
6178 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6179 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
6180 			    p[6], p[7]);
6181 		}
6182 	}
6183 
6184 	rc = sbuf_finish(sb);
6185 	sbuf_delete(sb);
6186 done:
6187 	free(buf, M_CXGBE);
6188 	return (rc);
6189 }
6190 
6191 static int
6192 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
6193 {
6194 	struct adapter *sc = arg1;
6195 	u_int cfg;
6196 	struct sbuf *sb;
6197 	uint32_t *buf, *p;
6198 	int rc;
6199 
6200 	MPASS(chip_id(sc) > CHELSIO_T5);
6201 
6202 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6203 	if (rc != 0)
6204 		return (rc);
6205 
6206 	rc = sysctl_wire_old_buffer(req, 0);
6207 	if (rc != 0)
6208 		return (rc);
6209 
6210 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6211 	if (sb == NULL)
6212 		return (ENOMEM);
6213 
6214 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6215 	    M_ZERO | M_WAITOK);
6216 
6217 	rc = -t4_cim_read_la(sc, buf, NULL);
6218 	if (rc != 0)
6219 		goto done;
6220 
6221 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
6222 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6223 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
6224 
6225 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
6226 		if (cfg & F_UPDBGLACAPTPCONLY) {
6227 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
6228 			    p[3] & 0xff, p[2], p[1], p[0]);
6229 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
6230 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
6231 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
6232 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
6233 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
6234 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
6235 			    p[6] >> 16);
6236 		} else {
6237 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
6238 			    "%08x %08x %08x %08x %08x %08x",
6239 			    (p[9] >> 16) & 0xff,
6240 			    p[9] & 0xffff, p[8] >> 16,
6241 			    p[8] & 0xffff, p[7] >> 16,
6242 			    p[7] & 0xffff, p[6] >> 16,
6243 			    p[2], p[1], p[0], p[5], p[4], p[3]);
6244 		}
6245 	}
6246 
6247 	rc = sbuf_finish(sb);
6248 	sbuf_delete(sb);
6249 done:
6250 	free(buf, M_CXGBE);
6251 	return (rc);
6252 }
6253 
6254 static int
6255 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
6256 {
6257 	struct adapter *sc = arg1;
6258 	u_int i;
6259 	struct sbuf *sb;
6260 	uint32_t *buf, *p;
6261 	int rc;
6262 
6263 	rc = sysctl_wire_old_buffer(req, 0);
6264 	if (rc != 0)
6265 		return (rc);
6266 
6267 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6268 	if (sb == NULL)
6269 		return (ENOMEM);
6270 
6271 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
6272 	    M_ZERO | M_WAITOK);
6273 
6274 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
6275 	p = buf;
6276 
6277 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6278 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
6279 		    p[1], p[0]);
6280 	}
6281 
6282 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
6283 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6284 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
6285 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
6286 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
6287 		    (p[1] >> 2) | ((p[2] & 3) << 30),
6288 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
6289 		    p[0] & 1);
6290 	}
6291 
6292 	rc = sbuf_finish(sb);
6293 	sbuf_delete(sb);
6294 	free(buf, M_CXGBE);
6295 	return (rc);
6296 }
6297 
6298 static int
6299 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
6300 {
6301 	struct adapter *sc = arg1;
6302 	u_int i;
6303 	struct sbuf *sb;
6304 	uint32_t *buf, *p;
6305 	int rc;
6306 
6307 	rc = sysctl_wire_old_buffer(req, 0);
6308 	if (rc != 0)
6309 		return (rc);
6310 
6311 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6312 	if (sb == NULL)
6313 		return (ENOMEM);
6314 
6315 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
6316 	    M_ZERO | M_WAITOK);
6317 
6318 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
6319 	p = buf;
6320 
6321 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
6322 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6323 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
6324 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
6325 		    p[4], p[3], p[2], p[1], p[0]);
6326 	}
6327 
6328 	sbuf_printf(sb, "\n\nCntl ID               Data");
6329 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6330 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
6331 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
6332 	}
6333 
6334 	rc = sbuf_finish(sb);
6335 	sbuf_delete(sb);
6336 	free(buf, M_CXGBE);
6337 	return (rc);
6338 }
6339 
6340 static int
6341 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
6342 {
6343 	struct adapter *sc = arg1;
6344 	struct sbuf *sb;
6345 	int rc, i;
6346 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6347 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6348 	uint16_t thres[CIM_NUM_IBQ];
6349 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
6350 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
6351 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
6352 
6353 	cim_num_obq = sc->chip_params->cim_num_obq;
6354 	if (is_t4(sc)) {
6355 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
6356 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
6357 	} else {
6358 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
6359 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
6360 	}
6361 	nq = CIM_NUM_IBQ + cim_num_obq;
6362 
6363 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
6364 	if (rc == 0)
6365 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
6366 	if (rc != 0)
6367 		return (rc);
6368 
6369 	t4_read_cimq_cfg(sc, base, size, thres);
6370 
6371 	rc = sysctl_wire_old_buffer(req, 0);
6372 	if (rc != 0)
6373 		return (rc);
6374 
6375 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6376 	if (sb == NULL)
6377 		return (ENOMEM);
6378 
6379 	sbuf_printf(sb,
6380 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
6381 
6382 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
6383 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
6384 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
6385 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6386 		    G_QUEREMFLITS(p[2]) * 16);
6387 	for ( ; i < nq; i++, p += 4, wr += 2)
6388 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
6389 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
6390 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6391 		    G_QUEREMFLITS(p[2]) * 16);
6392 
6393 	rc = sbuf_finish(sb);
6394 	sbuf_delete(sb);
6395 
6396 	return (rc);
6397 }
6398 
6399 static int
6400 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
6401 {
6402 	struct adapter *sc = arg1;
6403 	struct sbuf *sb;
6404 	int rc;
6405 	struct tp_cpl_stats stats;
6406 
6407 	rc = sysctl_wire_old_buffer(req, 0);
6408 	if (rc != 0)
6409 		return (rc);
6410 
6411 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6412 	if (sb == NULL)
6413 		return (ENOMEM);
6414 
6415 	mtx_lock(&sc->reg_lock);
6416 	t4_tp_get_cpl_stats(sc, &stats);
6417 	mtx_unlock(&sc->reg_lock);
6418 
6419 	if (sc->chip_params->nchan > 2) {
6420 		sbuf_printf(sb, "                 channel 0  channel 1"
6421 		    "  channel 2  channel 3");
6422 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
6423 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
6424 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
6425 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
6426 	} else {
6427 		sbuf_printf(sb, "                 channel 0  channel 1");
6428 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
6429 		    stats.req[0], stats.req[1]);
6430 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
6431 		    stats.rsp[0], stats.rsp[1]);
6432 	}
6433 
6434 	rc = sbuf_finish(sb);
6435 	sbuf_delete(sb);
6436 
6437 	return (rc);
6438 }
6439 
6440 static int
6441 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
6442 {
6443 	struct adapter *sc = arg1;
6444 	struct sbuf *sb;
6445 	int rc;
6446 	struct tp_usm_stats stats;
6447 
6448 	rc = sysctl_wire_old_buffer(req, 0);
6449 	if (rc != 0)
6450 		return(rc);
6451 
6452 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6453 	if (sb == NULL)
6454 		return (ENOMEM);
6455 
6456 	t4_get_usm_stats(sc, &stats);
6457 
6458 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
6459 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
6460 	sbuf_printf(sb, "Drops:  %u", stats.drops);
6461 
6462 	rc = sbuf_finish(sb);
6463 	sbuf_delete(sb);
6464 
6465 	return (rc);
6466 }
6467 
6468 static const char * const devlog_level_strings[] = {
6469 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
6470 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
6471 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
6472 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
6473 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
6474 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
6475 };
6476 
6477 static const char * const devlog_facility_strings[] = {
6478 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
6479 	[FW_DEVLOG_FACILITY_CF]		= "CF",
6480 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
6481 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
6482 	[FW_DEVLOG_FACILITY_RES]	= "RES",
6483 	[FW_DEVLOG_FACILITY_HW]		= "HW",
6484 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
6485 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
6486 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
6487 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
6488 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
6489 	[FW_DEVLOG_FACILITY_VI]		= "VI",
6490 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
6491 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
6492 	[FW_DEVLOG_FACILITY_TM]		= "TM",
6493 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
6494 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
6495 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
6496 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
6497 	[FW_DEVLOG_FACILITY_RI]		= "RI",
6498 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
6499 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
6500 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
6501 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
6502 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
6503 };
6504 
6505 static int
6506 sysctl_devlog(SYSCTL_HANDLER_ARGS)
6507 {
6508 	struct adapter *sc = arg1;
6509 	struct devlog_params *dparams = &sc->params.devlog;
6510 	struct fw_devlog_e *buf, *e;
6511 	int i, j, rc, nentries, first = 0;
6512 	struct sbuf *sb;
6513 	uint64_t ftstamp = UINT64_MAX;
6514 
6515 	if (dparams->addr == 0)
6516 		return (ENXIO);
6517 
6518 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
6519 	if (buf == NULL)
6520 		return (ENOMEM);
6521 
6522 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
6523 	if (rc != 0)
6524 		goto done;
6525 
6526 	nentries = dparams->size / sizeof(struct fw_devlog_e);
6527 	for (i = 0; i < nentries; i++) {
6528 		e = &buf[i];
6529 
6530 		if (e->timestamp == 0)
6531 			break;	/* end */
6532 
6533 		e->timestamp = be64toh(e->timestamp);
6534 		e->seqno = be32toh(e->seqno);
6535 		for (j = 0; j < 8; j++)
6536 			e->params[j] = be32toh(e->params[j]);
6537 
6538 		if (e->timestamp < ftstamp) {
6539 			ftstamp = e->timestamp;
6540 			first = i;
6541 		}
6542 	}
6543 
6544 	if (buf[first].timestamp == 0)
6545 		goto done;	/* nothing in the log */
6546 
6547 	rc = sysctl_wire_old_buffer(req, 0);
6548 	if (rc != 0)
6549 		goto done;
6550 
6551 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6552 	if (sb == NULL) {
6553 		rc = ENOMEM;
6554 		goto done;
6555 	}
6556 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
6557 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
6558 
6559 	i = first;
6560 	do {
6561 		e = &buf[i];
6562 		if (e->timestamp == 0)
6563 			break;	/* end */
6564 
6565 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
6566 		    e->seqno, e->timestamp,
6567 		    (e->level < nitems(devlog_level_strings) ?
6568 			devlog_level_strings[e->level] : "UNKNOWN"),
6569 		    (e->facility < nitems(devlog_facility_strings) ?
6570 			devlog_facility_strings[e->facility] : "UNKNOWN"));
6571 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
6572 		    e->params[2], e->params[3], e->params[4],
6573 		    e->params[5], e->params[6], e->params[7]);
6574 
6575 		if (++i == nentries)
6576 			i = 0;
6577 	} while (i != first);
6578 
6579 	rc = sbuf_finish(sb);
6580 	sbuf_delete(sb);
6581 done:
6582 	free(buf, M_CXGBE);
6583 	return (rc);
6584 }
6585 
6586 static int
6587 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
6588 {
6589 	struct adapter *sc = arg1;
6590 	struct sbuf *sb;
6591 	int rc;
6592 	struct tp_fcoe_stats stats[MAX_NCHAN];
6593 	int i, nchan = sc->chip_params->nchan;
6594 
6595 	rc = sysctl_wire_old_buffer(req, 0);
6596 	if (rc != 0)
6597 		return (rc);
6598 
6599 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6600 	if (sb == NULL)
6601 		return (ENOMEM);
6602 
6603 	for (i = 0; i < nchan; i++)
6604 		t4_get_fcoe_stats(sc, i, &stats[i]);
6605 
6606 	if (nchan > 2) {
6607 		sbuf_printf(sb, "                   channel 0        channel 1"
6608 		    "        channel 2        channel 3");
6609 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
6610 		    stats[0].octets_ddp, stats[1].octets_ddp,
6611 		    stats[2].octets_ddp, stats[3].octets_ddp);
6612 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
6613 		    stats[0].frames_ddp, stats[1].frames_ddp,
6614 		    stats[2].frames_ddp, stats[3].frames_ddp);
6615 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
6616 		    stats[0].frames_drop, stats[1].frames_drop,
6617 		    stats[2].frames_drop, stats[3].frames_drop);
6618 	} else {
6619 		sbuf_printf(sb, "                   channel 0        channel 1");
6620 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
6621 		    stats[0].octets_ddp, stats[1].octets_ddp);
6622 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
6623 		    stats[0].frames_ddp, stats[1].frames_ddp);
6624 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
6625 		    stats[0].frames_drop, stats[1].frames_drop);
6626 	}
6627 
6628 	rc = sbuf_finish(sb);
6629 	sbuf_delete(sb);
6630 
6631 	return (rc);
6632 }
6633 
6634 static int
6635 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
6636 {
6637 	struct adapter *sc = arg1;
6638 	struct sbuf *sb;
6639 	int rc, i;
6640 	unsigned int map, kbps, ipg, mode;
6641 	unsigned int pace_tab[NTX_SCHED];
6642 
6643 	rc = sysctl_wire_old_buffer(req, 0);
6644 	if (rc != 0)
6645 		return (rc);
6646 
6647 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6648 	if (sb == NULL)
6649 		return (ENOMEM);
6650 
6651 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
6652 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
6653 	t4_read_pace_tbl(sc, pace_tab);
6654 
6655 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
6656 	    "Class IPG (0.1 ns)   Flow IPG (us)");
6657 
6658 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
6659 		t4_get_tx_sched(sc, i, &kbps, &ipg);
6660 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
6661 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
6662 		if (kbps)
6663 			sbuf_printf(sb, "%9u     ", kbps);
6664 		else
6665 			sbuf_printf(sb, " disabled     ");
6666 
6667 		if (ipg)
6668 			sbuf_printf(sb, "%13u        ", ipg);
6669 		else
6670 			sbuf_printf(sb, "     disabled        ");
6671 
6672 		if (pace_tab[i])
6673 			sbuf_printf(sb, "%10u", pace_tab[i]);
6674 		else
6675 			sbuf_printf(sb, "  disabled");
6676 	}
6677 
6678 	rc = sbuf_finish(sb);
6679 	sbuf_delete(sb);
6680 
6681 	return (rc);
6682 }
6683 
6684 static int
6685 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
6686 {
6687 	struct adapter *sc = arg1;
6688 	struct sbuf *sb;
6689 	int rc, i, j;
6690 	uint64_t *p0, *p1;
6691 	struct lb_port_stats s[2];
6692 	static const char *stat_name[] = {
6693 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
6694 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
6695 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
6696 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
6697 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
6698 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
6699 		"BG2FramesTrunc:", "BG3FramesTrunc:"
6700 	};
6701 
6702 	rc = sysctl_wire_old_buffer(req, 0);
6703 	if (rc != 0)
6704 		return (rc);
6705 
6706 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6707 	if (sb == NULL)
6708 		return (ENOMEM);
6709 
6710 	memset(s, 0, sizeof(s));
6711 
6712 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
6713 		t4_get_lb_stats(sc, i, &s[0]);
6714 		t4_get_lb_stats(sc, i + 1, &s[1]);
6715 
6716 		p0 = &s[0].octets;
6717 		p1 = &s[1].octets;
6718 		sbuf_printf(sb, "%s                       Loopback %u"
6719 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
6720 
6721 		for (j = 0; j < nitems(stat_name); j++)
6722 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
6723 				   *p0++, *p1++);
6724 	}
6725 
6726 	rc = sbuf_finish(sb);
6727 	sbuf_delete(sb);
6728 
6729 	return (rc);
6730 }
6731 
6732 static int
6733 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
6734 {
6735 	int rc = 0;
6736 	struct port_info *pi = arg1;
6737 	struct link_config *lc = &pi->link_cfg;
6738 	struct sbuf *sb;
6739 
6740 	rc = sysctl_wire_old_buffer(req, 0);
6741 	if (rc != 0)
6742 		return(rc);
6743 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
6744 	if (sb == NULL)
6745 		return (ENOMEM);
6746 
6747 	if (lc->link_ok || lc->link_down_rc == 255)
6748 		sbuf_printf(sb, "n/a");
6749 	else
6750 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
6751 
6752 	rc = sbuf_finish(sb);
6753 	sbuf_delete(sb);
6754 
6755 	return (rc);
6756 }
6757 
6758 struct mem_desc {
6759 	unsigned int base;
6760 	unsigned int limit;
6761 	unsigned int idx;
6762 };
6763 
6764 static int
6765 mem_desc_cmp(const void *a, const void *b)
6766 {
6767 	return ((const struct mem_desc *)a)->base -
6768 	       ((const struct mem_desc *)b)->base;
6769 }
6770 
6771 static void
6772 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
6773     unsigned int to)
6774 {
6775 	unsigned int size;
6776 
6777 	if (from == to)
6778 		return;
6779 
6780 	size = to - from + 1;
6781 	if (size == 0)
6782 		return;
6783 
6784 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
6785 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
6786 }
6787 
6788 static int
6789 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
6790 {
6791 	struct adapter *sc = arg1;
6792 	struct sbuf *sb;
6793 	int rc, i, n;
6794 	uint32_t lo, hi, used, alloc;
6795 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
6796 	static const char *region[] = {
6797 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
6798 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
6799 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
6800 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
6801 		"RQUDP region:", "PBL region:", "TXPBL region:",
6802 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
6803 		"On-chip queues:"
6804 	};
6805 	struct mem_desc avail[4];
6806 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
6807 	struct mem_desc *md = mem;
6808 
6809 	rc = sysctl_wire_old_buffer(req, 0);
6810 	if (rc != 0)
6811 		return (rc);
6812 
6813 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6814 	if (sb == NULL)
6815 		return (ENOMEM);
6816 
6817 	for (i = 0; i < nitems(mem); i++) {
6818 		mem[i].limit = 0;
6819 		mem[i].idx = i;
6820 	}
6821 
6822 	/* Find and sort the populated memory ranges */
6823 	i = 0;
6824 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
6825 	if (lo & F_EDRAM0_ENABLE) {
6826 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
6827 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
6828 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
6829 		avail[i].idx = 0;
6830 		i++;
6831 	}
6832 	if (lo & F_EDRAM1_ENABLE) {
6833 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
6834 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
6835 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
6836 		avail[i].idx = 1;
6837 		i++;
6838 	}
6839 	if (lo & F_EXT_MEM_ENABLE) {
6840 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
6841 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
6842 		avail[i].limit = avail[i].base +
6843 		    (G_EXT_MEM_SIZE(hi) << 20);
6844 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
6845 		i++;
6846 	}
6847 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
6848 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
6849 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
6850 		avail[i].limit = avail[i].base +
6851 		    (G_EXT_MEM1_SIZE(hi) << 20);
6852 		avail[i].idx = 4;
6853 		i++;
6854 	}
6855 	if (!i)                                    /* no memory available */
6856 		return 0;
6857 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
6858 
6859 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
6860 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
6861 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
6862 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
6863 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
6864 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
6865 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
6866 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
6867 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
6868 
6869 	/* the next few have explicit upper bounds */
6870 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
6871 	md->limit = md->base - 1 +
6872 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
6873 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
6874 	md++;
6875 
6876 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
6877 	md->limit = md->base - 1 +
6878 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
6879 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
6880 	md++;
6881 
6882 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
6883 		if (chip_id(sc) <= CHELSIO_T5)
6884 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
6885 		else
6886 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
6887 		md->limit = 0;
6888 	} else {
6889 		md->base = 0;
6890 		md->idx = nitems(region);  /* hide it */
6891 	}
6892 	md++;
6893 
6894 #define ulp_region(reg) \
6895 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
6896 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
6897 
6898 	ulp_region(RX_ISCSI);
6899 	ulp_region(RX_TDDP);
6900 	ulp_region(TX_TPT);
6901 	ulp_region(RX_STAG);
6902 	ulp_region(RX_RQ);
6903 	ulp_region(RX_RQUDP);
6904 	ulp_region(RX_PBL);
6905 	ulp_region(TX_PBL);
6906 #undef ulp_region
6907 
6908 	md->base = 0;
6909 	md->idx = nitems(region);
6910 	if (!is_t4(sc)) {
6911 		uint32_t size = 0;
6912 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
6913 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
6914 
6915 		if (is_t5(sc)) {
6916 			if (sge_ctrl & F_VFIFO_ENABLE)
6917 				size = G_DBVFIFO_SIZE(fifo_size);
6918 		} else
6919 			size = G_T6_DBVFIFO_SIZE(fifo_size);
6920 
6921 		if (size) {
6922 			md->base = G_BASEADDR(t4_read_reg(sc,
6923 			    A_SGE_DBVFIFO_BADDR));
6924 			md->limit = md->base + (size << 2) - 1;
6925 		}
6926 	}
6927 	md++;
6928 
6929 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
6930 	md->limit = 0;
6931 	md++;
6932 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
6933 	md->limit = 0;
6934 	md++;
6935 
6936 	md->base = sc->vres.ocq.start;
6937 	if (sc->vres.ocq.size)
6938 		md->limit = md->base + sc->vres.ocq.size - 1;
6939 	else
6940 		md->idx = nitems(region);  /* hide it */
6941 	md++;
6942 
6943 	/* add any address-space holes, there can be up to 3 */
6944 	for (n = 0; n < i - 1; n++)
6945 		if (avail[n].limit < avail[n + 1].base)
6946 			(md++)->base = avail[n].limit;
6947 	if (avail[n].limit)
6948 		(md++)->base = avail[n].limit;
6949 
6950 	n = md - mem;
6951 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
6952 
6953 	for (lo = 0; lo < i; lo++)
6954 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
6955 				avail[lo].limit - 1);
6956 
6957 	sbuf_printf(sb, "\n");
6958 	for (i = 0; i < n; i++) {
6959 		if (mem[i].idx >= nitems(region))
6960 			continue;                        /* skip holes */
6961 		if (!mem[i].limit)
6962 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
6963 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
6964 				mem[i].limit);
6965 	}
6966 
6967 	sbuf_printf(sb, "\n");
6968 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
6969 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
6970 	mem_region_show(sb, "uP RAM:", lo, hi);
6971 
6972 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
6973 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
6974 	mem_region_show(sb, "uP Extmem2:", lo, hi);
6975 
6976 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
6977 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
6978 		   G_PMRXMAXPAGE(lo),
6979 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
6980 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
6981 
6982 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
6983 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
6984 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
6985 		   G_PMTXMAXPAGE(lo),
6986 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
6987 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
6988 	sbuf_printf(sb, "%u p-structs\n",
6989 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
6990 
6991 	for (i = 0; i < 4; i++) {
6992 		if (chip_id(sc) > CHELSIO_T5)
6993 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
6994 		else
6995 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
6996 		if (is_t5(sc)) {
6997 			used = G_T5_USED(lo);
6998 			alloc = G_T5_ALLOC(lo);
6999 		} else {
7000 			used = G_USED(lo);
7001 			alloc = G_ALLOC(lo);
7002 		}
7003 		/* For T6 these are MAC buffer groups */
7004 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
7005 		    i, used, alloc);
7006 	}
7007 	for (i = 0; i < sc->chip_params->nchan; i++) {
7008 		if (chip_id(sc) > CHELSIO_T5)
7009 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
7010 		else
7011 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
7012 		if (is_t5(sc)) {
7013 			used = G_T5_USED(lo);
7014 			alloc = G_T5_ALLOC(lo);
7015 		} else {
7016 			used = G_USED(lo);
7017 			alloc = G_ALLOC(lo);
7018 		}
7019 		/* For T6 these are MAC buffer groups */
7020 		sbuf_printf(sb,
7021 		    "\nLoopback %d using %u pages out of %u allocated",
7022 		    i, used, alloc);
7023 	}
7024 
7025 	rc = sbuf_finish(sb);
7026 	sbuf_delete(sb);
7027 
7028 	return (rc);
7029 }
7030 
7031 static inline void
7032 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
7033 {
7034 	*mask = x | y;
7035 	y = htobe64(y);
7036 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
7037 }
7038 
7039 static int
7040 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
7041 {
7042 	struct adapter *sc = arg1;
7043 	struct sbuf *sb;
7044 	int rc, i;
7045 
7046 	MPASS(chip_id(sc) <= CHELSIO_T5);
7047 
7048 	rc = sysctl_wire_old_buffer(req, 0);
7049 	if (rc != 0)
7050 		return (rc);
7051 
7052 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7053 	if (sb == NULL)
7054 		return (ENOMEM);
7055 
7056 	sbuf_printf(sb,
7057 	    "Idx  Ethernet address     Mask     Vld Ports PF"
7058 	    "  VF              Replication             P0 P1 P2 P3  ML");
7059 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7060 		uint64_t tcamx, tcamy, mask;
7061 		uint32_t cls_lo, cls_hi;
7062 		uint8_t addr[ETHER_ADDR_LEN];
7063 
7064 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
7065 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
7066 		if (tcamx & tcamy)
7067 			continue;
7068 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7069 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7070 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7071 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
7072 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
7073 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
7074 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
7075 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
7076 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
7077 
7078 		if (cls_lo & F_REPLICATE) {
7079 			struct fw_ldst_cmd ldst_cmd;
7080 
7081 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7082 			ldst_cmd.op_to_addrspace =
7083 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7084 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7085 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7086 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7087 			ldst_cmd.u.mps.rplc.fid_idx =
7088 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7089 				V_FW_LDST_CMD_IDX(i));
7090 
7091 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7092 			    "t4mps");
7093 			if (rc)
7094 				break;
7095 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7096 			    sizeof(ldst_cmd), &ldst_cmd);
7097 			end_synchronized_op(sc, 0);
7098 
7099 			if (rc != 0) {
7100 				sbuf_printf(sb, "%36d", rc);
7101 				rc = 0;
7102 			} else {
7103 				sbuf_printf(sb, " %08x %08x %08x %08x",
7104 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7105 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7106 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7107 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7108 			}
7109 		} else
7110 			sbuf_printf(sb, "%36s", "");
7111 
7112 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
7113 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
7114 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
7115 	}
7116 
7117 	if (rc)
7118 		(void) sbuf_finish(sb);
7119 	else
7120 		rc = sbuf_finish(sb);
7121 	sbuf_delete(sb);
7122 
7123 	return (rc);
7124 }
7125 
7126 static int
7127 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
7128 {
7129 	struct adapter *sc = arg1;
7130 	struct sbuf *sb;
7131 	int rc, i;
7132 
7133 	MPASS(chip_id(sc) > CHELSIO_T5);
7134 
7135 	rc = sysctl_wire_old_buffer(req, 0);
7136 	if (rc != 0)
7137 		return (rc);
7138 
7139 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7140 	if (sb == NULL)
7141 		return (ENOMEM);
7142 
7143 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
7144 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
7145 	    "                           Replication"
7146 	    "                                    P0 P1 P2 P3  ML\n");
7147 
7148 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7149 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
7150 		uint16_t ivlan;
7151 		uint64_t tcamx, tcamy, val, mask;
7152 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
7153 		uint8_t addr[ETHER_ADDR_LEN];
7154 
7155 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
7156 		if (i < 256)
7157 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
7158 		else
7159 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
7160 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7161 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7162 		tcamy = G_DMACH(val) << 32;
7163 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7164 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7165 		lookup_type = G_DATALKPTYPE(data2);
7166 		port_num = G_DATAPORTNUM(data2);
7167 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7168 			/* Inner header VNI */
7169 			vniy = ((data2 & F_DATAVIDH2) << 23) |
7170 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7171 			dip_hit = data2 & F_DATADIPHIT;
7172 			vlan_vld = 0;
7173 		} else {
7174 			vniy = 0;
7175 			dip_hit = 0;
7176 			vlan_vld = data2 & F_DATAVIDH2;
7177 			ivlan = G_VIDL(val);
7178 		}
7179 
7180 		ctl |= V_CTLXYBITSEL(1);
7181 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7182 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7183 		tcamx = G_DMACH(val) << 32;
7184 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7185 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7186 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7187 			/* Inner header VNI mask */
7188 			vnix = ((data2 & F_DATAVIDH2) << 23) |
7189 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7190 		} else
7191 			vnix = 0;
7192 
7193 		if (tcamx & tcamy)
7194 			continue;
7195 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7196 
7197 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7198 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7199 
7200 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7201 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7202 			    "%012jx %06x %06x    -    -   %3c"
7203 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
7204 			    addr[1], addr[2], addr[3], addr[4], addr[5],
7205 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
7206 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7207 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7208 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7209 		} else {
7210 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7211 			    "%012jx    -       -   ", i, addr[0], addr[1],
7212 			    addr[2], addr[3], addr[4], addr[5],
7213 			    (uintmax_t)mask);
7214 
7215 			if (vlan_vld)
7216 				sbuf_printf(sb, "%4u   Y     ", ivlan);
7217 			else
7218 				sbuf_printf(sb, "  -    N     ");
7219 
7220 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
7221 			    lookup_type ? 'I' : 'O', port_num,
7222 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7223 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7224 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7225 		}
7226 
7227 
7228 		if (cls_lo & F_T6_REPLICATE) {
7229 			struct fw_ldst_cmd ldst_cmd;
7230 
7231 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7232 			ldst_cmd.op_to_addrspace =
7233 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7234 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7235 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7236 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7237 			ldst_cmd.u.mps.rplc.fid_idx =
7238 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7239 				V_FW_LDST_CMD_IDX(i));
7240 
7241 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7242 			    "t6mps");
7243 			if (rc)
7244 				break;
7245 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7246 			    sizeof(ldst_cmd), &ldst_cmd);
7247 			end_synchronized_op(sc, 0);
7248 
7249 			if (rc != 0) {
7250 				sbuf_printf(sb, "%72d", rc);
7251 				rc = 0;
7252 			} else {
7253 				sbuf_printf(sb, " %08x %08x %08x %08x"
7254 				    " %08x %08x %08x %08x",
7255 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
7256 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
7257 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
7258 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
7259 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7260 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7261 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7262 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7263 			}
7264 		} else
7265 			sbuf_printf(sb, "%72s", "");
7266 
7267 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
7268 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
7269 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
7270 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
7271 	}
7272 
7273 	if (rc)
7274 		(void) sbuf_finish(sb);
7275 	else
7276 		rc = sbuf_finish(sb);
7277 	sbuf_delete(sb);
7278 
7279 	return (rc);
7280 }
7281 
7282 static int
7283 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
7284 {
7285 	struct adapter *sc = arg1;
7286 	struct sbuf *sb;
7287 	int rc;
7288 	uint16_t mtus[NMTUS];
7289 
7290 	rc = sysctl_wire_old_buffer(req, 0);
7291 	if (rc != 0)
7292 		return (rc);
7293 
7294 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7295 	if (sb == NULL)
7296 		return (ENOMEM);
7297 
7298 	t4_read_mtu_tbl(sc, mtus, NULL);
7299 
7300 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
7301 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
7302 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
7303 	    mtus[14], mtus[15]);
7304 
7305 	rc = sbuf_finish(sb);
7306 	sbuf_delete(sb);
7307 
7308 	return (rc);
7309 }
7310 
7311 static int
7312 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
7313 {
7314 	struct adapter *sc = arg1;
7315 	struct sbuf *sb;
7316 	int rc, i;
7317 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
7318 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
7319 	static const char *tx_stats[MAX_PM_NSTATS] = {
7320 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
7321 		"Tx FIFO wait", NULL, "Tx latency"
7322 	};
7323 	static const char *rx_stats[MAX_PM_NSTATS] = {
7324 		"Read:", "Write bypass:", "Write mem:", "Flush:",
7325 		"Rx FIFO wait", NULL, "Rx latency"
7326 	};
7327 
7328 	rc = sysctl_wire_old_buffer(req, 0);
7329 	if (rc != 0)
7330 		return (rc);
7331 
7332 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7333 	if (sb == NULL)
7334 		return (ENOMEM);
7335 
7336 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
7337 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
7338 
7339 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
7340 	for (i = 0; i < 4; i++) {
7341 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7342 		    tx_cyc[i]);
7343 	}
7344 
7345 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
7346 	for (i = 0; i < 4; i++) {
7347 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7348 		    rx_cyc[i]);
7349 	}
7350 
7351 	if (chip_id(sc) > CHELSIO_T5) {
7352 		sbuf_printf(sb,
7353 		    "\n              Total wait      Total occupancy");
7354 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7355 		    tx_cyc[i]);
7356 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7357 		    rx_cyc[i]);
7358 
7359 		i += 2;
7360 		MPASS(i < nitems(tx_stats));
7361 
7362 		sbuf_printf(sb,
7363 		    "\n                   Reads           Total wait");
7364 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7365 		    tx_cyc[i]);
7366 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7367 		    rx_cyc[i]);
7368 	}
7369 
7370 	rc = sbuf_finish(sb);
7371 	sbuf_delete(sb);
7372 
7373 	return (rc);
7374 }
7375 
7376 static int
7377 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
7378 {
7379 	struct adapter *sc = arg1;
7380 	struct sbuf *sb;
7381 	int rc;
7382 	struct tp_rdma_stats stats;
7383 
7384 	rc = sysctl_wire_old_buffer(req, 0);
7385 	if (rc != 0)
7386 		return (rc);
7387 
7388 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7389 	if (sb == NULL)
7390 		return (ENOMEM);
7391 
7392 	mtx_lock(&sc->reg_lock);
7393 	t4_tp_get_rdma_stats(sc, &stats);
7394 	mtx_unlock(&sc->reg_lock);
7395 
7396 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
7397 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
7398 
7399 	rc = sbuf_finish(sb);
7400 	sbuf_delete(sb);
7401 
7402 	return (rc);
7403 }
7404 
7405 static int
7406 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
7407 {
7408 	struct adapter *sc = arg1;
7409 	struct sbuf *sb;
7410 	int rc;
7411 	struct tp_tcp_stats v4, v6;
7412 
7413 	rc = sysctl_wire_old_buffer(req, 0);
7414 	if (rc != 0)
7415 		return (rc);
7416 
7417 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7418 	if (sb == NULL)
7419 		return (ENOMEM);
7420 
7421 	mtx_lock(&sc->reg_lock);
7422 	t4_tp_get_tcp_stats(sc, &v4, &v6);
7423 	mtx_unlock(&sc->reg_lock);
7424 
7425 	sbuf_printf(sb,
7426 	    "                                IP                 IPv6\n");
7427 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
7428 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
7429 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
7430 	    v4.tcp_in_segs, v6.tcp_in_segs);
7431 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
7432 	    v4.tcp_out_segs, v6.tcp_out_segs);
7433 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
7434 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
7435 
7436 	rc = sbuf_finish(sb);
7437 	sbuf_delete(sb);
7438 
7439 	return (rc);
7440 }
7441 
7442 static int
7443 sysctl_tids(SYSCTL_HANDLER_ARGS)
7444 {
7445 	struct adapter *sc = arg1;
7446 	struct sbuf *sb;
7447 	int rc;
7448 	struct tid_info *t = &sc->tids;
7449 
7450 	rc = sysctl_wire_old_buffer(req, 0);
7451 	if (rc != 0)
7452 		return (rc);
7453 
7454 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7455 	if (sb == NULL)
7456 		return (ENOMEM);
7457 
7458 	if (t->natids) {
7459 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
7460 		    t->atids_in_use);
7461 	}
7462 
7463 	if (t->ntids) {
7464 		sbuf_printf(sb, "TID range: ");
7465 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7466 			uint32_t b, hb;
7467 
7468 			if (chip_id(sc) <= CHELSIO_T5) {
7469 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
7470 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
7471 			} else {
7472 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
7473 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
7474 			}
7475 
7476 			if (b)
7477 				sbuf_printf(sb, "0-%u, ", b - 1);
7478 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
7479 		} else
7480 			sbuf_printf(sb, "0-%u", t->ntids - 1);
7481 		sbuf_printf(sb, ", in use: %u\n",
7482 		    atomic_load_acq_int(&t->tids_in_use));
7483 	}
7484 
7485 	if (t->nstids) {
7486 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
7487 		    t->stid_base + t->nstids - 1, t->stids_in_use);
7488 	}
7489 
7490 	if (t->nftids) {
7491 		sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
7492 		    t->ftid_base + t->nftids - 1);
7493 	}
7494 
7495 	if (t->netids) {
7496 		sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
7497 		    t->etid_base + t->netids - 1);
7498 	}
7499 
7500 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
7501 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
7502 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
7503 
7504 	rc = sbuf_finish(sb);
7505 	sbuf_delete(sb);
7506 
7507 	return (rc);
7508 }
7509 
7510 static int
7511 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
7512 {
7513 	struct adapter *sc = arg1;
7514 	struct sbuf *sb;
7515 	int rc;
7516 	struct tp_err_stats stats;
7517 
7518 	rc = sysctl_wire_old_buffer(req, 0);
7519 	if (rc != 0)
7520 		return (rc);
7521 
7522 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7523 	if (sb == NULL)
7524 		return (ENOMEM);
7525 
7526 	mtx_lock(&sc->reg_lock);
7527 	t4_tp_get_err_stats(sc, &stats);
7528 	mtx_unlock(&sc->reg_lock);
7529 
7530 	if (sc->chip_params->nchan > 2) {
7531 		sbuf_printf(sb, "                 channel 0  channel 1"
7532 		    "  channel 2  channel 3\n");
7533 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
7534 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
7535 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
7536 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
7537 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
7538 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
7539 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
7540 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
7541 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
7542 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
7543 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
7544 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
7545 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
7546 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
7547 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
7548 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
7549 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
7550 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
7551 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
7552 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
7553 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
7554 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
7555 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
7556 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
7557 	} else {
7558 		sbuf_printf(sb, "                 channel 0  channel 1\n");
7559 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
7560 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
7561 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
7562 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
7563 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
7564 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
7565 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
7566 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
7567 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
7568 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
7569 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
7570 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
7571 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
7572 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
7573 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
7574 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
7575 	}
7576 
7577 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
7578 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
7579 
7580 	rc = sbuf_finish(sb);
7581 	sbuf_delete(sb);
7582 
7583 	return (rc);
7584 }
7585 
7586 static int
7587 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
7588 {
7589 	struct adapter *sc = arg1;
7590 	struct tp_params *tpp = &sc->params.tp;
7591 	u_int mask;
7592 	int rc;
7593 
7594 	mask = tpp->la_mask >> 16;
7595 	rc = sysctl_handle_int(oidp, &mask, 0, req);
7596 	if (rc != 0 || req->newptr == NULL)
7597 		return (rc);
7598 	if (mask > 0xffff)
7599 		return (EINVAL);
7600 	tpp->la_mask = mask << 16;
7601 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
7602 
7603 	return (0);
7604 }
7605 
7606 struct field_desc {
7607 	const char *name;
7608 	u_int start;
7609 	u_int width;
7610 };
7611 
7612 static void
7613 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
7614 {
7615 	char buf[32];
7616 	int line_size = 0;
7617 
7618 	while (f->name) {
7619 		uint64_t mask = (1ULL << f->width) - 1;
7620 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
7621 		    ((uintmax_t)v >> f->start) & mask);
7622 
7623 		if (line_size + len >= 79) {
7624 			line_size = 8;
7625 			sbuf_printf(sb, "\n        ");
7626 		}
7627 		sbuf_printf(sb, "%s ", buf);
7628 		line_size += len + 1;
7629 		f++;
7630 	}
7631 	sbuf_printf(sb, "\n");
7632 }
7633 
7634 static const struct field_desc tp_la0[] = {
7635 	{ "RcfOpCodeOut", 60, 4 },
7636 	{ "State", 56, 4 },
7637 	{ "WcfState", 52, 4 },
7638 	{ "RcfOpcSrcOut", 50, 2 },
7639 	{ "CRxError", 49, 1 },
7640 	{ "ERxError", 48, 1 },
7641 	{ "SanityFailed", 47, 1 },
7642 	{ "SpuriousMsg", 46, 1 },
7643 	{ "FlushInputMsg", 45, 1 },
7644 	{ "FlushInputCpl", 44, 1 },
7645 	{ "RssUpBit", 43, 1 },
7646 	{ "RssFilterHit", 42, 1 },
7647 	{ "Tid", 32, 10 },
7648 	{ "InitTcb", 31, 1 },
7649 	{ "LineNumber", 24, 7 },
7650 	{ "Emsg", 23, 1 },
7651 	{ "EdataOut", 22, 1 },
7652 	{ "Cmsg", 21, 1 },
7653 	{ "CdataOut", 20, 1 },
7654 	{ "EreadPdu", 19, 1 },
7655 	{ "CreadPdu", 18, 1 },
7656 	{ "TunnelPkt", 17, 1 },
7657 	{ "RcfPeerFin", 16, 1 },
7658 	{ "RcfReasonOut", 12, 4 },
7659 	{ "TxCchannel", 10, 2 },
7660 	{ "RcfTxChannel", 8, 2 },
7661 	{ "RxEchannel", 6, 2 },
7662 	{ "RcfRxChannel", 5, 1 },
7663 	{ "RcfDataOutSrdy", 4, 1 },
7664 	{ "RxDvld", 3, 1 },
7665 	{ "RxOoDvld", 2, 1 },
7666 	{ "RxCongestion", 1, 1 },
7667 	{ "TxCongestion", 0, 1 },
7668 	{ NULL }
7669 };
7670 
7671 static const struct field_desc tp_la1[] = {
7672 	{ "CplCmdIn", 56, 8 },
7673 	{ "CplCmdOut", 48, 8 },
7674 	{ "ESynOut", 47, 1 },
7675 	{ "EAckOut", 46, 1 },
7676 	{ "EFinOut", 45, 1 },
7677 	{ "ERstOut", 44, 1 },
7678 	{ "SynIn", 43, 1 },
7679 	{ "AckIn", 42, 1 },
7680 	{ "FinIn", 41, 1 },
7681 	{ "RstIn", 40, 1 },
7682 	{ "DataIn", 39, 1 },
7683 	{ "DataInVld", 38, 1 },
7684 	{ "PadIn", 37, 1 },
7685 	{ "RxBufEmpty", 36, 1 },
7686 	{ "RxDdp", 35, 1 },
7687 	{ "RxFbCongestion", 34, 1 },
7688 	{ "TxFbCongestion", 33, 1 },
7689 	{ "TxPktSumSrdy", 32, 1 },
7690 	{ "RcfUlpType", 28, 4 },
7691 	{ "Eread", 27, 1 },
7692 	{ "Ebypass", 26, 1 },
7693 	{ "Esave", 25, 1 },
7694 	{ "Static0", 24, 1 },
7695 	{ "Cread", 23, 1 },
7696 	{ "Cbypass", 22, 1 },
7697 	{ "Csave", 21, 1 },
7698 	{ "CPktOut", 20, 1 },
7699 	{ "RxPagePoolFull", 18, 2 },
7700 	{ "RxLpbkPkt", 17, 1 },
7701 	{ "TxLpbkPkt", 16, 1 },
7702 	{ "RxVfValid", 15, 1 },
7703 	{ "SynLearned", 14, 1 },
7704 	{ "SetDelEntry", 13, 1 },
7705 	{ "SetInvEntry", 12, 1 },
7706 	{ "CpcmdDvld", 11, 1 },
7707 	{ "CpcmdSave", 10, 1 },
7708 	{ "RxPstructsFull", 8, 2 },
7709 	{ "EpcmdDvld", 7, 1 },
7710 	{ "EpcmdFlush", 6, 1 },
7711 	{ "EpcmdTrimPrefix", 5, 1 },
7712 	{ "EpcmdTrimPostfix", 4, 1 },
7713 	{ "ERssIp4Pkt", 3, 1 },
7714 	{ "ERssIp6Pkt", 2, 1 },
7715 	{ "ERssTcpUdpPkt", 1, 1 },
7716 	{ "ERssFceFipPkt", 0, 1 },
7717 	{ NULL }
7718 };
7719 
7720 static const struct field_desc tp_la2[] = {
7721 	{ "CplCmdIn", 56, 8 },
7722 	{ "MpsVfVld", 55, 1 },
7723 	{ "MpsPf", 52, 3 },
7724 	{ "MpsVf", 44, 8 },
7725 	{ "SynIn", 43, 1 },
7726 	{ "AckIn", 42, 1 },
7727 	{ "FinIn", 41, 1 },
7728 	{ "RstIn", 40, 1 },
7729 	{ "DataIn", 39, 1 },
7730 	{ "DataInVld", 38, 1 },
7731 	{ "PadIn", 37, 1 },
7732 	{ "RxBufEmpty", 36, 1 },
7733 	{ "RxDdp", 35, 1 },
7734 	{ "RxFbCongestion", 34, 1 },
7735 	{ "TxFbCongestion", 33, 1 },
7736 	{ "TxPktSumSrdy", 32, 1 },
7737 	{ "RcfUlpType", 28, 4 },
7738 	{ "Eread", 27, 1 },
7739 	{ "Ebypass", 26, 1 },
7740 	{ "Esave", 25, 1 },
7741 	{ "Static0", 24, 1 },
7742 	{ "Cread", 23, 1 },
7743 	{ "Cbypass", 22, 1 },
7744 	{ "Csave", 21, 1 },
7745 	{ "CPktOut", 20, 1 },
7746 	{ "RxPagePoolFull", 18, 2 },
7747 	{ "RxLpbkPkt", 17, 1 },
7748 	{ "TxLpbkPkt", 16, 1 },
7749 	{ "RxVfValid", 15, 1 },
7750 	{ "SynLearned", 14, 1 },
7751 	{ "SetDelEntry", 13, 1 },
7752 	{ "SetInvEntry", 12, 1 },
7753 	{ "CpcmdDvld", 11, 1 },
7754 	{ "CpcmdSave", 10, 1 },
7755 	{ "RxPstructsFull", 8, 2 },
7756 	{ "EpcmdDvld", 7, 1 },
7757 	{ "EpcmdFlush", 6, 1 },
7758 	{ "EpcmdTrimPrefix", 5, 1 },
7759 	{ "EpcmdTrimPostfix", 4, 1 },
7760 	{ "ERssIp4Pkt", 3, 1 },
7761 	{ "ERssIp6Pkt", 2, 1 },
7762 	{ "ERssTcpUdpPkt", 1, 1 },
7763 	{ "ERssFceFipPkt", 0, 1 },
7764 	{ NULL }
7765 };
7766 
7767 static void
7768 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
7769 {
7770 
7771 	field_desc_show(sb, *p, tp_la0);
7772 }
7773 
7774 static void
7775 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
7776 {
7777 
7778 	if (idx)
7779 		sbuf_printf(sb, "\n");
7780 	field_desc_show(sb, p[0], tp_la0);
7781 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7782 		field_desc_show(sb, p[1], tp_la0);
7783 }
7784 
7785 static void
7786 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
7787 {
7788 
7789 	if (idx)
7790 		sbuf_printf(sb, "\n");
7791 	field_desc_show(sb, p[0], tp_la0);
7792 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7793 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
7794 }
7795 
7796 static int
7797 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
7798 {
7799 	struct adapter *sc = arg1;
7800 	struct sbuf *sb;
7801 	uint64_t *buf, *p;
7802 	int rc;
7803 	u_int i, inc;
7804 	void (*show_func)(struct sbuf *, uint64_t *, int);
7805 
7806 	rc = sysctl_wire_old_buffer(req, 0);
7807 	if (rc != 0)
7808 		return (rc);
7809 
7810 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7811 	if (sb == NULL)
7812 		return (ENOMEM);
7813 
7814 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
7815 
7816 	t4_tp_read_la(sc, buf, NULL);
7817 	p = buf;
7818 
7819 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
7820 	case 2:
7821 		inc = 2;
7822 		show_func = tp_la_show2;
7823 		break;
7824 	case 3:
7825 		inc = 2;
7826 		show_func = tp_la_show3;
7827 		break;
7828 	default:
7829 		inc = 1;
7830 		show_func = tp_la_show;
7831 	}
7832 
7833 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
7834 		(*show_func)(sb, p, i);
7835 
7836 	rc = sbuf_finish(sb);
7837 	sbuf_delete(sb);
7838 	free(buf, M_CXGBE);
7839 	return (rc);
7840 }
7841 
7842 static int
7843 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
7844 {
7845 	struct adapter *sc = arg1;
7846 	struct sbuf *sb;
7847 	int rc;
7848 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
7849 
7850 	rc = sysctl_wire_old_buffer(req, 0);
7851 	if (rc != 0)
7852 		return (rc);
7853 
7854 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7855 	if (sb == NULL)
7856 		return (ENOMEM);
7857 
7858 	t4_get_chan_txrate(sc, nrate, orate);
7859 
7860 	if (sc->chip_params->nchan > 2) {
7861 		sbuf_printf(sb, "              channel 0   channel 1"
7862 		    "   channel 2   channel 3\n");
7863 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
7864 		    nrate[0], nrate[1], nrate[2], nrate[3]);
7865 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
7866 		    orate[0], orate[1], orate[2], orate[3]);
7867 	} else {
7868 		sbuf_printf(sb, "              channel 0   channel 1\n");
7869 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
7870 		    nrate[0], nrate[1]);
7871 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
7872 		    orate[0], orate[1]);
7873 	}
7874 
7875 	rc = sbuf_finish(sb);
7876 	sbuf_delete(sb);
7877 
7878 	return (rc);
7879 }
7880 
7881 static int
7882 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
7883 {
7884 	struct adapter *sc = arg1;
7885 	struct sbuf *sb;
7886 	uint32_t *buf, *p;
7887 	int rc, i;
7888 
7889 	rc = sysctl_wire_old_buffer(req, 0);
7890 	if (rc != 0)
7891 		return (rc);
7892 
7893 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7894 	if (sb == NULL)
7895 		return (ENOMEM);
7896 
7897 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
7898 	    M_ZERO | M_WAITOK);
7899 
7900 	t4_ulprx_read_la(sc, buf);
7901 	p = buf;
7902 
7903 	sbuf_printf(sb, "      Pcmd        Type   Message"
7904 	    "                Data");
7905 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
7906 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
7907 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
7908 	}
7909 
7910 	rc = sbuf_finish(sb);
7911 	sbuf_delete(sb);
7912 	free(buf, M_CXGBE);
7913 	return (rc);
7914 }
7915 
7916 static int
7917 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
7918 {
7919 	struct adapter *sc = arg1;
7920 	struct sbuf *sb;
7921 	int rc, v;
7922 
7923 	MPASS(chip_id(sc) >= CHELSIO_T5);
7924 
7925 	rc = sysctl_wire_old_buffer(req, 0);
7926 	if (rc != 0)
7927 		return (rc);
7928 
7929 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7930 	if (sb == NULL)
7931 		return (ENOMEM);
7932 
7933 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
7934 	if (G_STATSOURCE_T5(v) == 7) {
7935 		int mode;
7936 
7937 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
7938 		if (mode == 0) {
7939 			sbuf_printf(sb, "total %d, incomplete %d",
7940 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7941 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7942 		} else if (mode == 1) {
7943 			sbuf_printf(sb, "total %d, data overflow %d",
7944 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7945 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7946 		} else {
7947 			sbuf_printf(sb, "unknown mode %d", mode);
7948 		}
7949 	}
7950 	rc = sbuf_finish(sb);
7951 	sbuf_delete(sb);
7952 
7953 	return (rc);
7954 }
7955 
7956 static int
7957 sysctl_tc_params(SYSCTL_HANDLER_ARGS)
7958 {
7959 	struct adapter *sc = arg1;
7960 	struct tx_cl_rl_params tc;
7961 	struct sbuf *sb;
7962 	int i, rc, port_id, mbps, gbps;
7963 
7964 	rc = sysctl_wire_old_buffer(req, 0);
7965 	if (rc != 0)
7966 		return (rc);
7967 
7968 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7969 	if (sb == NULL)
7970 		return (ENOMEM);
7971 
7972 	port_id = arg2 >> 16;
7973 	MPASS(port_id < sc->params.nports);
7974 	MPASS(sc->port[port_id] != NULL);
7975 	i = arg2 & 0xffff;
7976 	MPASS(i < sc->chip_params->nsched_cls);
7977 
7978 	mtx_lock(&sc->tc_lock);
7979 	tc = sc->port[port_id]->sched_params->cl_rl[i];
7980 	mtx_unlock(&sc->tc_lock);
7981 
7982 	if (tc.flags & TX_CLRL_ERROR) {
7983 		sbuf_printf(sb, "error");
7984 		goto done;
7985 	}
7986 
7987 	if (tc.ratemode == SCHED_CLASS_RATEMODE_REL) {
7988 		/* XXX: top speed or actual link speed? */
7989 		gbps = port_top_speed(sc->port[port_id]);
7990 		sbuf_printf(sb, " %u%% of %uGbps", tc.maxrate, gbps);
7991 	} else if (tc.ratemode == SCHED_CLASS_RATEMODE_ABS) {
7992 		switch (tc.rateunit) {
7993 		case SCHED_CLASS_RATEUNIT_BITS:
7994 			mbps = tc.maxrate / 1000;
7995 			gbps = tc.maxrate / 1000000;
7996 			if (tc.maxrate == gbps * 1000000)
7997 				sbuf_printf(sb, " %uGbps", gbps);
7998 			else if (tc.maxrate == mbps * 1000)
7999 				sbuf_printf(sb, " %uMbps", mbps);
8000 			else
8001 				sbuf_printf(sb, " %uKbps", tc.maxrate);
8002 			break;
8003 		case SCHED_CLASS_RATEUNIT_PKTS:
8004 			sbuf_printf(sb, " %upps", tc.maxrate);
8005 			break;
8006 		default:
8007 			rc = ENXIO;
8008 			goto done;
8009 		}
8010 	}
8011 
8012 	switch (tc.mode) {
8013 	case SCHED_CLASS_MODE_CLASS:
8014 		sbuf_printf(sb, " aggregate");
8015 		break;
8016 	case SCHED_CLASS_MODE_FLOW:
8017 		sbuf_printf(sb, " per-flow");
8018 		break;
8019 	default:
8020 		rc = ENXIO;
8021 		goto done;
8022 	}
8023 
8024 done:
8025 	if (rc == 0)
8026 		rc = sbuf_finish(sb);
8027 	sbuf_delete(sb);
8028 
8029 	return (rc);
8030 }
8031 #endif
8032 
8033 #ifdef TCP_OFFLOAD
8034 static void
8035 unit_conv(char *buf, size_t len, u_int val, u_int factor)
8036 {
8037 	u_int rem = val % factor;
8038 
8039 	if (rem == 0)
8040 		snprintf(buf, len, "%u", val / factor);
8041 	else {
8042 		while (rem % 10 == 0)
8043 			rem /= 10;
8044 		snprintf(buf, len, "%u.%u", val / factor, rem);
8045 	}
8046 }
8047 
8048 static int
8049 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
8050 {
8051 	struct adapter *sc = arg1;
8052 	char buf[16];
8053 	u_int res, re;
8054 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8055 
8056 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8057 	switch (arg2) {
8058 	case 0:
8059 		/* timer_tick */
8060 		re = G_TIMERRESOLUTION(res);
8061 		break;
8062 	case 1:
8063 		/* TCP timestamp tick */
8064 		re = G_TIMESTAMPRESOLUTION(res);
8065 		break;
8066 	case 2:
8067 		/* DACK tick */
8068 		re = G_DELAYEDACKRESOLUTION(res);
8069 		break;
8070 	default:
8071 		return (EDOOFUS);
8072 	}
8073 
8074 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
8075 
8076 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
8077 }
8078 
8079 static int
8080 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
8081 {
8082 	struct adapter *sc = arg1;
8083 	u_int res, dack_re, v;
8084 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8085 
8086 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8087 	dack_re = G_DELAYEDACKRESOLUTION(res);
8088 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
8089 
8090 	return (sysctl_handle_int(oidp, &v, 0, req));
8091 }
8092 
8093 static int
8094 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
8095 {
8096 	struct adapter *sc = arg1;
8097 	int reg = arg2;
8098 	u_int tre;
8099 	u_long tp_tick_us, v;
8100 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8101 
8102 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
8103 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
8104 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
8105 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
8106 
8107 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
8108 	tp_tick_us = (cclk_ps << tre) / 1000000;
8109 
8110 	if (reg == A_TP_INIT_SRTT)
8111 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
8112 	else
8113 		v = tp_tick_us * t4_read_reg(sc, reg);
8114 
8115 	return (sysctl_handle_long(oidp, &v, 0, req));
8116 }
8117 #endif
8118 
8119 static uint32_t
8120 fconf_iconf_to_mode(uint32_t fconf, uint32_t iconf)
8121 {
8122 	uint32_t mode;
8123 
8124 	mode = T4_FILTER_IPv4 | T4_FILTER_IPv6 | T4_FILTER_IP_SADDR |
8125 	    T4_FILTER_IP_DADDR | T4_FILTER_IP_SPORT | T4_FILTER_IP_DPORT;
8126 
8127 	if (fconf & F_FRAGMENTATION)
8128 		mode |= T4_FILTER_IP_FRAGMENT;
8129 
8130 	if (fconf & F_MPSHITTYPE)
8131 		mode |= T4_FILTER_MPS_HIT_TYPE;
8132 
8133 	if (fconf & F_MACMATCH)
8134 		mode |= T4_FILTER_MAC_IDX;
8135 
8136 	if (fconf & F_ETHERTYPE)
8137 		mode |= T4_FILTER_ETH_TYPE;
8138 
8139 	if (fconf & F_PROTOCOL)
8140 		mode |= T4_FILTER_IP_PROTO;
8141 
8142 	if (fconf & F_TOS)
8143 		mode |= T4_FILTER_IP_TOS;
8144 
8145 	if (fconf & F_VLAN)
8146 		mode |= T4_FILTER_VLAN;
8147 
8148 	if (fconf & F_VNIC_ID) {
8149 		mode |= T4_FILTER_VNIC;
8150 		if (iconf & F_VNIC)
8151 			mode |= T4_FILTER_IC_VNIC;
8152 	}
8153 
8154 	if (fconf & F_PORT)
8155 		mode |= T4_FILTER_PORT;
8156 
8157 	if (fconf & F_FCOE)
8158 		mode |= T4_FILTER_FCoE;
8159 
8160 	return (mode);
8161 }
8162 
8163 static uint32_t
8164 mode_to_fconf(uint32_t mode)
8165 {
8166 	uint32_t fconf = 0;
8167 
8168 	if (mode & T4_FILTER_IP_FRAGMENT)
8169 		fconf |= F_FRAGMENTATION;
8170 
8171 	if (mode & T4_FILTER_MPS_HIT_TYPE)
8172 		fconf |= F_MPSHITTYPE;
8173 
8174 	if (mode & T4_FILTER_MAC_IDX)
8175 		fconf |= F_MACMATCH;
8176 
8177 	if (mode & T4_FILTER_ETH_TYPE)
8178 		fconf |= F_ETHERTYPE;
8179 
8180 	if (mode & T4_FILTER_IP_PROTO)
8181 		fconf |= F_PROTOCOL;
8182 
8183 	if (mode & T4_FILTER_IP_TOS)
8184 		fconf |= F_TOS;
8185 
8186 	if (mode & T4_FILTER_VLAN)
8187 		fconf |= F_VLAN;
8188 
8189 	if (mode & T4_FILTER_VNIC)
8190 		fconf |= F_VNIC_ID;
8191 
8192 	if (mode & T4_FILTER_PORT)
8193 		fconf |= F_PORT;
8194 
8195 	if (mode & T4_FILTER_FCoE)
8196 		fconf |= F_FCOE;
8197 
8198 	return (fconf);
8199 }
8200 
8201 static uint32_t
8202 mode_to_iconf(uint32_t mode)
8203 {
8204 
8205 	if (mode & T4_FILTER_IC_VNIC)
8206 		return (F_VNIC);
8207 	return (0);
8208 }
8209 
8210 static int check_fspec_against_fconf_iconf(struct adapter *sc,
8211     struct t4_filter_specification *fs)
8212 {
8213 	struct tp_params *tpp = &sc->params.tp;
8214 	uint32_t fconf = 0;
8215 
8216 	if (fs->val.frag || fs->mask.frag)
8217 		fconf |= F_FRAGMENTATION;
8218 
8219 	if (fs->val.matchtype || fs->mask.matchtype)
8220 		fconf |= F_MPSHITTYPE;
8221 
8222 	if (fs->val.macidx || fs->mask.macidx)
8223 		fconf |= F_MACMATCH;
8224 
8225 	if (fs->val.ethtype || fs->mask.ethtype)
8226 		fconf |= F_ETHERTYPE;
8227 
8228 	if (fs->val.proto || fs->mask.proto)
8229 		fconf |= F_PROTOCOL;
8230 
8231 	if (fs->val.tos || fs->mask.tos)
8232 		fconf |= F_TOS;
8233 
8234 	if (fs->val.vlan_vld || fs->mask.vlan_vld)
8235 		fconf |= F_VLAN;
8236 
8237 	if (fs->val.ovlan_vld || fs->mask.ovlan_vld) {
8238 		fconf |= F_VNIC_ID;
8239 		if (tpp->ingress_config & F_VNIC)
8240 			return (EINVAL);
8241 	}
8242 
8243 	if (fs->val.pfvf_vld || fs->mask.pfvf_vld) {
8244 		fconf |= F_VNIC_ID;
8245 		if ((tpp->ingress_config & F_VNIC) == 0)
8246 			return (EINVAL);
8247 	}
8248 
8249 	if (fs->val.iport || fs->mask.iport)
8250 		fconf |= F_PORT;
8251 
8252 	if (fs->val.fcoe || fs->mask.fcoe)
8253 		fconf |= F_FCOE;
8254 
8255 	if ((tpp->vlan_pri_map | fconf) != tpp->vlan_pri_map)
8256 		return (E2BIG);
8257 
8258 	return (0);
8259 }
8260 
8261 static int
8262 get_filter_mode(struct adapter *sc, uint32_t *mode)
8263 {
8264 	struct tp_params *tpp = &sc->params.tp;
8265 
8266 	/*
8267 	 * We trust the cached values of the relevant TP registers.  This means
8268 	 * things work reliably only if writes to those registers are always via
8269 	 * t4_set_filter_mode.
8270 	 */
8271 	*mode = fconf_iconf_to_mode(tpp->vlan_pri_map, tpp->ingress_config);
8272 
8273 	return (0);
8274 }
8275 
8276 static int
8277 set_filter_mode(struct adapter *sc, uint32_t mode)
8278 {
8279 	struct tp_params *tpp = &sc->params.tp;
8280 	uint32_t fconf, iconf;
8281 	int rc;
8282 
8283 	iconf = mode_to_iconf(mode);
8284 	if ((iconf ^ tpp->ingress_config) & F_VNIC) {
8285 		/*
8286 		 * For now we just complain if A_TP_INGRESS_CONFIG is not
8287 		 * already set to the correct value for the requested filter
8288 		 * mode.  It's not clear if it's safe to write to this register
8289 		 * on the fly.  (And we trust the cached value of the register).
8290 		 */
8291 		return (EBUSY);
8292 	}
8293 
8294 	fconf = mode_to_fconf(mode);
8295 
8296 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8297 	    "t4setfm");
8298 	if (rc)
8299 		return (rc);
8300 
8301 	if (sc->tids.ftids_in_use > 0) {
8302 		rc = EBUSY;
8303 		goto done;
8304 	}
8305 
8306 #ifdef TCP_OFFLOAD
8307 	if (uld_active(sc, ULD_TOM)) {
8308 		rc = EBUSY;
8309 		goto done;
8310 	}
8311 #endif
8312 
8313 	rc = -t4_set_filter_mode(sc, fconf);
8314 done:
8315 	end_synchronized_op(sc, LOCK_HELD);
8316 	return (rc);
8317 }
8318 
8319 static inline uint64_t
8320 get_filter_hits(struct adapter *sc, uint32_t fid)
8321 {
8322 	uint32_t tcb_addr;
8323 
8324 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE) +
8325 	    (fid + sc->tids.ftid_base) * TCB_SIZE;
8326 
8327 	if (is_t4(sc)) {
8328 		uint64_t hits;
8329 
8330 		read_via_memwin(sc, 0, tcb_addr + 16, (uint32_t *)&hits, 8);
8331 		return (be64toh(hits));
8332 	} else {
8333 		uint32_t hits;
8334 
8335 		read_via_memwin(sc, 0, tcb_addr + 24, &hits, 4);
8336 		return (be32toh(hits));
8337 	}
8338 }
8339 
8340 static int
8341 get_filter(struct adapter *sc, struct t4_filter *t)
8342 {
8343 	int i, rc, nfilters = sc->tids.nftids;
8344 	struct filter_entry *f;
8345 
8346 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8347 	    "t4getf");
8348 	if (rc)
8349 		return (rc);
8350 
8351 	if (sc->tids.ftids_in_use == 0 || sc->tids.ftid_tab == NULL ||
8352 	    t->idx >= nfilters) {
8353 		t->idx = 0xffffffff;
8354 		goto done;
8355 	}
8356 
8357 	f = &sc->tids.ftid_tab[t->idx];
8358 	for (i = t->idx; i < nfilters; i++, f++) {
8359 		if (f->valid) {
8360 			t->idx = i;
8361 			t->l2tidx = f->l2t ? f->l2t->idx : 0;
8362 			t->smtidx = f->smtidx;
8363 			if (f->fs.hitcnts)
8364 				t->hits = get_filter_hits(sc, t->idx);
8365 			else
8366 				t->hits = UINT64_MAX;
8367 			t->fs = f->fs;
8368 
8369 			goto done;
8370 		}
8371 	}
8372 
8373 	t->idx = 0xffffffff;
8374 done:
8375 	end_synchronized_op(sc, LOCK_HELD);
8376 	return (0);
8377 }
8378 
8379 static int
8380 set_filter(struct adapter *sc, struct t4_filter *t)
8381 {
8382 	unsigned int nfilters, nports;
8383 	struct filter_entry *f;
8384 	int i, rc;
8385 
8386 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setf");
8387 	if (rc)
8388 		return (rc);
8389 
8390 	nfilters = sc->tids.nftids;
8391 	nports = sc->params.nports;
8392 
8393 	if (nfilters == 0) {
8394 		rc = ENOTSUP;
8395 		goto done;
8396 	}
8397 
8398 	if (t->idx >= nfilters) {
8399 		rc = EINVAL;
8400 		goto done;
8401 	}
8402 
8403 	/* Validate against the global filter mode and ingress config */
8404 	rc = check_fspec_against_fconf_iconf(sc, &t->fs);
8405 	if (rc != 0)
8406 		goto done;
8407 
8408 	if (t->fs.action == FILTER_SWITCH && t->fs.eport >= nports) {
8409 		rc = EINVAL;
8410 		goto done;
8411 	}
8412 
8413 	if (t->fs.val.iport >= nports) {
8414 		rc = EINVAL;
8415 		goto done;
8416 	}
8417 
8418 	/* Can't specify an iq if not steering to it */
8419 	if (!t->fs.dirsteer && t->fs.iq) {
8420 		rc = EINVAL;
8421 		goto done;
8422 	}
8423 
8424 	/* IPv6 filter idx must be 4 aligned */
8425 	if (t->fs.type == 1 &&
8426 	    ((t->idx & 0x3) || t->idx + 4 >= nfilters)) {
8427 		rc = EINVAL;
8428 		goto done;
8429 	}
8430 
8431 	if (!(sc->flags & FULL_INIT_DONE) &&
8432 	    ((rc = adapter_full_init(sc)) != 0))
8433 		goto done;
8434 
8435 	if (sc->tids.ftid_tab == NULL) {
8436 		KASSERT(sc->tids.ftids_in_use == 0,
8437 		    ("%s: no memory allocated but filters_in_use > 0",
8438 		    __func__));
8439 
8440 		sc->tids.ftid_tab = malloc(sizeof (struct filter_entry) *
8441 		    nfilters, M_CXGBE, M_NOWAIT | M_ZERO);
8442 		if (sc->tids.ftid_tab == NULL) {
8443 			rc = ENOMEM;
8444 			goto done;
8445 		}
8446 		mtx_init(&sc->tids.ftid_lock, "T4 filters", 0, MTX_DEF);
8447 	}
8448 
8449 	for (i = 0; i < 4; i++) {
8450 		f = &sc->tids.ftid_tab[t->idx + i];
8451 
8452 		if (f->pending || f->valid) {
8453 			rc = EBUSY;
8454 			goto done;
8455 		}
8456 		if (f->locked) {
8457 			rc = EPERM;
8458 			goto done;
8459 		}
8460 
8461 		if (t->fs.type == 0)
8462 			break;
8463 	}
8464 
8465 	f = &sc->tids.ftid_tab[t->idx];
8466 	f->fs = t->fs;
8467 
8468 	rc = set_filter_wr(sc, t->idx);
8469 done:
8470 	end_synchronized_op(sc, 0);
8471 
8472 	if (rc == 0) {
8473 		mtx_lock(&sc->tids.ftid_lock);
8474 		for (;;) {
8475 			if (f->pending == 0) {
8476 				rc = f->valid ? 0 : EIO;
8477 				break;
8478 			}
8479 
8480 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8481 			    PCATCH, "t4setfw", 0)) {
8482 				rc = EINPROGRESS;
8483 				break;
8484 			}
8485 		}
8486 		mtx_unlock(&sc->tids.ftid_lock);
8487 	}
8488 	return (rc);
8489 }
8490 
8491 static int
8492 del_filter(struct adapter *sc, struct t4_filter *t)
8493 {
8494 	unsigned int nfilters;
8495 	struct filter_entry *f;
8496 	int rc;
8497 
8498 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4delf");
8499 	if (rc)
8500 		return (rc);
8501 
8502 	nfilters = sc->tids.nftids;
8503 
8504 	if (nfilters == 0) {
8505 		rc = ENOTSUP;
8506 		goto done;
8507 	}
8508 
8509 	if (sc->tids.ftid_tab == NULL || sc->tids.ftids_in_use == 0 ||
8510 	    t->idx >= nfilters) {
8511 		rc = EINVAL;
8512 		goto done;
8513 	}
8514 
8515 	if (!(sc->flags & FULL_INIT_DONE)) {
8516 		rc = EAGAIN;
8517 		goto done;
8518 	}
8519 
8520 	f = &sc->tids.ftid_tab[t->idx];
8521 
8522 	if (f->pending) {
8523 		rc = EBUSY;
8524 		goto done;
8525 	}
8526 	if (f->locked) {
8527 		rc = EPERM;
8528 		goto done;
8529 	}
8530 
8531 	if (f->valid) {
8532 		t->fs = f->fs;	/* extra info for the caller */
8533 		rc = del_filter_wr(sc, t->idx);
8534 	}
8535 
8536 done:
8537 	end_synchronized_op(sc, 0);
8538 
8539 	if (rc == 0) {
8540 		mtx_lock(&sc->tids.ftid_lock);
8541 		for (;;) {
8542 			if (f->pending == 0) {
8543 				rc = f->valid ? EIO : 0;
8544 				break;
8545 			}
8546 
8547 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8548 			    PCATCH, "t4delfw", 0)) {
8549 				rc = EINPROGRESS;
8550 				break;
8551 			}
8552 		}
8553 		mtx_unlock(&sc->tids.ftid_lock);
8554 	}
8555 
8556 	return (rc);
8557 }
8558 
8559 static void
8560 clear_filter(struct filter_entry *f)
8561 {
8562 	if (f->l2t)
8563 		t4_l2t_release(f->l2t);
8564 
8565 	bzero(f, sizeof (*f));
8566 }
8567 
8568 static int
8569 set_filter_wr(struct adapter *sc, int fidx)
8570 {
8571 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8572 	struct fw_filter_wr *fwr;
8573 	unsigned int ftid, vnic_vld, vnic_vld_mask;
8574 	struct wrq_cookie cookie;
8575 
8576 	ASSERT_SYNCHRONIZED_OP(sc);
8577 
8578 	if (f->fs.newdmac || f->fs.newvlan) {
8579 		/* This filter needs an L2T entry; allocate one. */
8580 		f->l2t = t4_l2t_alloc_switching(sc->l2t);
8581 		if (f->l2t == NULL)
8582 			return (EAGAIN);
8583 		if (t4_l2t_set_switching(sc, f->l2t, f->fs.vlan, f->fs.eport,
8584 		    f->fs.dmac)) {
8585 			t4_l2t_release(f->l2t);
8586 			f->l2t = NULL;
8587 			return (ENOMEM);
8588 		}
8589 	}
8590 
8591 	/* Already validated against fconf, iconf */
8592 	MPASS((f->fs.val.pfvf_vld & f->fs.val.ovlan_vld) == 0);
8593 	MPASS((f->fs.mask.pfvf_vld & f->fs.mask.ovlan_vld) == 0);
8594 	if (f->fs.val.pfvf_vld || f->fs.val.ovlan_vld)
8595 		vnic_vld = 1;
8596 	else
8597 		vnic_vld = 0;
8598 	if (f->fs.mask.pfvf_vld || f->fs.mask.ovlan_vld)
8599 		vnic_vld_mask = 1;
8600 	else
8601 		vnic_vld_mask = 0;
8602 
8603 	ftid = sc->tids.ftid_base + fidx;
8604 
8605 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8606 	if (fwr == NULL)
8607 		return (ENOMEM);
8608 	bzero(fwr, sizeof(*fwr));
8609 
8610 	fwr->op_pkd = htobe32(V_FW_WR_OP(FW_FILTER_WR));
8611 	fwr->len16_pkd = htobe32(FW_LEN16(*fwr));
8612 	fwr->tid_to_iq =
8613 	    htobe32(V_FW_FILTER_WR_TID(ftid) |
8614 		V_FW_FILTER_WR_RQTYPE(f->fs.type) |
8615 		V_FW_FILTER_WR_NOREPLY(0) |
8616 		V_FW_FILTER_WR_IQ(f->fs.iq));
8617 	fwr->del_filter_to_l2tix =
8618 	    htobe32(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
8619 		V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
8620 		V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
8621 		V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
8622 		V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
8623 		V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
8624 		V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
8625 		V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
8626 		V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
8627 		    f->fs.newvlan == VLAN_REWRITE) |
8628 		V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
8629 		    f->fs.newvlan == VLAN_REWRITE) |
8630 		V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
8631 		V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
8632 		V_FW_FILTER_WR_PRIO(f->fs.prio) |
8633 		V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
8634 	fwr->ethtype = htobe16(f->fs.val.ethtype);
8635 	fwr->ethtypem = htobe16(f->fs.mask.ethtype);
8636 	fwr->frag_to_ovlan_vldm =
8637 	    (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
8638 		V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
8639 		V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.vlan_vld) |
8640 		V_FW_FILTER_WR_OVLAN_VLD(vnic_vld) |
8641 		V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.vlan_vld) |
8642 		V_FW_FILTER_WR_OVLAN_VLDM(vnic_vld_mask));
8643 	fwr->smac_sel = 0;
8644 	fwr->rx_chan_rx_rpl_iq = htobe16(V_FW_FILTER_WR_RX_CHAN(0) |
8645 	    V_FW_FILTER_WR_RX_RPL_IQ(sc->sge.fwq.abs_id));
8646 	fwr->maci_to_matchtypem =
8647 	    htobe32(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
8648 		V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
8649 		V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
8650 		V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
8651 		V_FW_FILTER_WR_PORT(f->fs.val.iport) |
8652 		V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
8653 		V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
8654 		V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
8655 	fwr->ptcl = f->fs.val.proto;
8656 	fwr->ptclm = f->fs.mask.proto;
8657 	fwr->ttyp = f->fs.val.tos;
8658 	fwr->ttypm = f->fs.mask.tos;
8659 	fwr->ivlan = htobe16(f->fs.val.vlan);
8660 	fwr->ivlanm = htobe16(f->fs.mask.vlan);
8661 	fwr->ovlan = htobe16(f->fs.val.vnic);
8662 	fwr->ovlanm = htobe16(f->fs.mask.vnic);
8663 	bcopy(f->fs.val.dip, fwr->lip, sizeof (fwr->lip));
8664 	bcopy(f->fs.mask.dip, fwr->lipm, sizeof (fwr->lipm));
8665 	bcopy(f->fs.val.sip, fwr->fip, sizeof (fwr->fip));
8666 	bcopy(f->fs.mask.sip, fwr->fipm, sizeof (fwr->fipm));
8667 	fwr->lp = htobe16(f->fs.val.dport);
8668 	fwr->lpm = htobe16(f->fs.mask.dport);
8669 	fwr->fp = htobe16(f->fs.val.sport);
8670 	fwr->fpm = htobe16(f->fs.mask.sport);
8671 	if (f->fs.newsmac)
8672 		bcopy(f->fs.smac, fwr->sma, sizeof (fwr->sma));
8673 
8674 	f->pending = 1;
8675 	sc->tids.ftids_in_use++;
8676 
8677 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8678 	return (0);
8679 }
8680 
8681 static int
8682 del_filter_wr(struct adapter *sc, int fidx)
8683 {
8684 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8685 	struct fw_filter_wr *fwr;
8686 	unsigned int ftid;
8687 	struct wrq_cookie cookie;
8688 
8689 	ftid = sc->tids.ftid_base + fidx;
8690 
8691 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8692 	if (fwr == NULL)
8693 		return (ENOMEM);
8694 	bzero(fwr, sizeof (*fwr));
8695 
8696 	t4_mk_filtdelwr(ftid, fwr, sc->sge.fwq.abs_id);
8697 
8698 	f->pending = 1;
8699 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8700 	return (0);
8701 }
8702 
8703 int
8704 t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8705 {
8706 	struct adapter *sc = iq->adapter;
8707 	const struct cpl_set_tcb_rpl *rpl = (const void *)(rss + 1);
8708 	unsigned int idx = GET_TID(rpl);
8709 	unsigned int rc;
8710 	struct filter_entry *f;
8711 
8712 	KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
8713 	    rss->opcode));
8714 	MPASS(iq == &sc->sge.fwq);
8715 	MPASS(is_ftid(sc, idx));
8716 
8717 	idx -= sc->tids.ftid_base;
8718 	f = &sc->tids.ftid_tab[idx];
8719 	rc = G_COOKIE(rpl->cookie);
8720 
8721 	mtx_lock(&sc->tids.ftid_lock);
8722 	if (rc == FW_FILTER_WR_FLT_ADDED) {
8723 		KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
8724 		    __func__, idx));
8725 		f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
8726 		f->pending = 0;  /* asynchronous setup completed */
8727 		f->valid = 1;
8728 	} else {
8729 		if (rc != FW_FILTER_WR_FLT_DELETED) {
8730 			/* Add or delete failed, display an error */
8731 			log(LOG_ERR,
8732 			    "filter %u setup failed with error %u\n",
8733 			    idx, rc);
8734 		}
8735 
8736 		clear_filter(f);
8737 		sc->tids.ftids_in_use--;
8738 	}
8739 	wakeup(&sc->tids.ftid_tab);
8740 	mtx_unlock(&sc->tids.ftid_lock);
8741 
8742 	return (0);
8743 }
8744 
8745 static int
8746 set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8747 {
8748 
8749 	MPASS(iq->set_tcb_rpl != NULL);
8750 	return (iq->set_tcb_rpl(iq, rss, m));
8751 }
8752 
8753 static int
8754 l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8755 {
8756 
8757 	MPASS(iq->l2t_write_rpl != NULL);
8758 	return (iq->l2t_write_rpl(iq, rss, m));
8759 }
8760 
8761 static int
8762 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
8763 {
8764 	int rc;
8765 
8766 	if (cntxt->cid > M_CTXTQID)
8767 		return (EINVAL);
8768 
8769 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
8770 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
8771 		return (EINVAL);
8772 
8773 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
8774 	if (rc)
8775 		return (rc);
8776 
8777 	if (sc->flags & FW_OK) {
8778 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
8779 		    &cntxt->data[0]);
8780 		if (rc == 0)
8781 			goto done;
8782 	}
8783 
8784 	/*
8785 	 * Read via firmware failed or wasn't even attempted.  Read directly via
8786 	 * the backdoor.
8787 	 */
8788 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
8789 done:
8790 	end_synchronized_op(sc, 0);
8791 	return (rc);
8792 }
8793 
8794 static int
8795 load_fw(struct adapter *sc, struct t4_data *fw)
8796 {
8797 	int rc;
8798 	uint8_t *fw_data;
8799 
8800 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
8801 	if (rc)
8802 		return (rc);
8803 
8804 	if (sc->flags & FULL_INIT_DONE) {
8805 		rc = EBUSY;
8806 		goto done;
8807 	}
8808 
8809 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
8810 	if (fw_data == NULL) {
8811 		rc = ENOMEM;
8812 		goto done;
8813 	}
8814 
8815 	rc = copyin(fw->data, fw_data, fw->len);
8816 	if (rc == 0)
8817 		rc = -t4_load_fw(sc, fw_data, fw->len);
8818 
8819 	free(fw_data, M_CXGBE);
8820 done:
8821 	end_synchronized_op(sc, 0);
8822 	return (rc);
8823 }
8824 
8825 static int
8826 load_cfg(struct adapter *sc, struct t4_data *cfg)
8827 {
8828 	int rc;
8829 	uint8_t *cfg_data = NULL;
8830 
8831 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
8832 	if (rc)
8833 		return (rc);
8834 
8835 	if (cfg->len == 0) {
8836 		/* clear */
8837 		rc = -t4_load_cfg(sc, NULL, 0);
8838 		goto done;
8839 	}
8840 
8841 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
8842 	if (cfg_data == NULL) {
8843 		rc = ENOMEM;
8844 		goto done;
8845 	}
8846 
8847 	rc = copyin(cfg->data, cfg_data, cfg->len);
8848 	if (rc == 0)
8849 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
8850 
8851 	free(cfg_data, M_CXGBE);
8852 done:
8853 	end_synchronized_op(sc, 0);
8854 	return (rc);
8855 }
8856 
8857 static int
8858 load_boot(struct adapter *sc, struct t4_bootrom *br)
8859 {
8860 	int rc;
8861 	uint8_t *br_data = NULL;
8862 	u_int offset;
8863 
8864 	if (br->len > 1024 * 1024)
8865 		return (EFBIG);
8866 
8867 	if (br->pf_offset == 0) {
8868 		/* pfidx */
8869 		if (br->pfidx_addr > 7)
8870 			return (EINVAL);
8871 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
8872 		    A_PCIE_PF_EXPROM_OFST)));
8873 	} else if (br->pf_offset == 1) {
8874 		/* offset */
8875 		offset = G_OFFSET(br->pfidx_addr);
8876 	} else {
8877 		return (EINVAL);
8878 	}
8879 
8880 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
8881 	if (rc)
8882 		return (rc);
8883 
8884 	if (br->len == 0) {
8885 		/* clear */
8886 		rc = -t4_load_boot(sc, NULL, offset, 0);
8887 		goto done;
8888 	}
8889 
8890 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
8891 	if (br_data == NULL) {
8892 		rc = ENOMEM;
8893 		goto done;
8894 	}
8895 
8896 	rc = copyin(br->data, br_data, br->len);
8897 	if (rc == 0)
8898 		rc = -t4_load_boot(sc, br_data, offset, br->len);
8899 
8900 	free(br_data, M_CXGBE);
8901 done:
8902 	end_synchronized_op(sc, 0);
8903 	return (rc);
8904 }
8905 
8906 static int
8907 load_bootcfg(struct adapter *sc, struct t4_data *bc)
8908 {
8909 	int rc;
8910 	uint8_t *bc_data = NULL;
8911 
8912 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
8913 	if (rc)
8914 		return (rc);
8915 
8916 	if (bc->len == 0) {
8917 		/* clear */
8918 		rc = -t4_load_bootcfg(sc, NULL, 0);
8919 		goto done;
8920 	}
8921 
8922 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
8923 	if (bc_data == NULL) {
8924 		rc = ENOMEM;
8925 		goto done;
8926 	}
8927 
8928 	rc = copyin(bc->data, bc_data, bc->len);
8929 	if (rc == 0)
8930 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
8931 
8932 	free(bc_data, M_CXGBE);
8933 done:
8934 	end_synchronized_op(sc, 0);
8935 	return (rc);
8936 }
8937 
8938 #define MAX_READ_BUF_SIZE (128 * 1024)
8939 static int
8940 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
8941 {
8942 	uint32_t addr, remaining, n;
8943 	uint32_t *buf;
8944 	int rc;
8945 	uint8_t *dst;
8946 
8947 	rc = validate_mem_range(sc, mr->addr, mr->len);
8948 	if (rc != 0)
8949 		return (rc);
8950 
8951 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
8952 	addr = mr->addr;
8953 	remaining = mr->len;
8954 	dst = (void *)mr->data;
8955 
8956 	while (remaining) {
8957 		n = min(remaining, MAX_READ_BUF_SIZE);
8958 		read_via_memwin(sc, 2, addr, buf, n);
8959 
8960 		rc = copyout(buf, dst, n);
8961 		if (rc != 0)
8962 			break;
8963 
8964 		dst += n;
8965 		remaining -= n;
8966 		addr += n;
8967 	}
8968 
8969 	free(buf, M_CXGBE);
8970 	return (rc);
8971 }
8972 #undef MAX_READ_BUF_SIZE
8973 
8974 static int
8975 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
8976 {
8977 	int rc;
8978 
8979 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
8980 		return (EINVAL);
8981 
8982 	if (i2cd->len > sizeof(i2cd->data))
8983 		return (EFBIG);
8984 
8985 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
8986 	if (rc)
8987 		return (rc);
8988 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
8989 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
8990 	end_synchronized_op(sc, 0);
8991 
8992 	return (rc);
8993 }
8994 
8995 int
8996 t4_os_find_pci_capability(struct adapter *sc, int cap)
8997 {
8998 	int i;
8999 
9000 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9001 }
9002 
9003 int
9004 t4_os_pci_save_state(struct adapter *sc)
9005 {
9006 	device_t dev;
9007 	struct pci_devinfo *dinfo;
9008 
9009 	dev = sc->dev;
9010 	dinfo = device_get_ivars(dev);
9011 
9012 	pci_cfg_save(dev, dinfo, 0);
9013 	return (0);
9014 }
9015 
9016 int
9017 t4_os_pci_restore_state(struct adapter *sc)
9018 {
9019 	device_t dev;
9020 	struct pci_devinfo *dinfo;
9021 
9022 	dev = sc->dev;
9023 	dinfo = device_get_ivars(dev);
9024 
9025 	pci_cfg_restore(dev, dinfo);
9026 	return (0);
9027 }
9028 
9029 void
9030 t4_os_portmod_changed(struct port_info *pi, int old_ptype, int old_mtype,
9031     struct link_config *old_lc)
9032 {
9033 	struct vi_info *vi;
9034 	struct ifnet *ifp;
9035 	int v;
9036 	static const char *mod_str[] = {
9037 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9038 	};
9039 
9040 	PORT_LOCK(pi);
9041 	for_each_vi(pi, v, vi) {
9042 		build_medialist(pi, &vi->media);
9043 	}
9044 	PORT_UNLOCK(pi);
9045 	if (begin_synchronized_op(pi->adapter, vi, HOLD_LOCK, "t4mod") == 0) {
9046 		init_l1cfg(pi);
9047 		end_synchronized_op(pi->adapter, LOCK_HELD);
9048 	}
9049 
9050 	ifp = pi->vi[0].ifp;
9051 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9052 		if_printf(ifp, "transceiver unplugged.\n");
9053 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9054 		if_printf(ifp, "unknown transceiver inserted.\n");
9055 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9056 		if_printf(ifp, "unsupported transceiver inserted.\n");
9057 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9058 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9059 		    port_top_speed(pi), mod_str[pi->mod_type]);
9060 	} else {
9061 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9062 		    pi->mod_type);
9063 	}
9064 }
9065 
9066 void
9067 t4_os_link_changed(struct port_info *pi, struct link_config *old_lc)
9068 {
9069 	struct vi_info *vi;
9070 	struct ifnet *ifp;
9071 	struct link_config *lc;
9072 	int v;
9073 
9074 	for_each_vi(pi, v, vi) {
9075 		ifp = vi->ifp;
9076 		if (ifp == NULL)
9077 			continue;
9078 
9079 		lc = &pi->link_cfg;
9080 		if (lc->link_ok) {
9081 			ifp->if_baudrate = IF_Mbps(lc->speed);
9082 			if_link_state_change(ifp, LINK_STATE_UP);
9083 		} else {
9084 			if_link_state_change(ifp, LINK_STATE_DOWN);
9085 		}
9086 	}
9087 }
9088 
9089 void
9090 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9091 {
9092 	struct adapter *sc;
9093 
9094 	sx_slock(&t4_list_lock);
9095 	SLIST_FOREACH(sc, &t4_list, link) {
9096 		/*
9097 		 * func should not make any assumptions about what state sc is
9098 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9099 		 */
9100 		func(sc, arg);
9101 	}
9102 	sx_sunlock(&t4_list_lock);
9103 }
9104 
9105 static int
9106 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9107     struct thread *td)
9108 {
9109 	int rc;
9110 	struct adapter *sc = dev->si_drv1;
9111 
9112 	rc = priv_check(td, PRIV_DRIVER);
9113 	if (rc != 0)
9114 		return (rc);
9115 
9116 	switch (cmd) {
9117 	case CHELSIO_T4_GETREG: {
9118 		struct t4_reg *edata = (struct t4_reg *)data;
9119 
9120 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9121 			return (EFAULT);
9122 
9123 		if (edata->size == 4)
9124 			edata->val = t4_read_reg(sc, edata->addr);
9125 		else if (edata->size == 8)
9126 			edata->val = t4_read_reg64(sc, edata->addr);
9127 		else
9128 			return (EINVAL);
9129 
9130 		break;
9131 	}
9132 	case CHELSIO_T4_SETREG: {
9133 		struct t4_reg *edata = (struct t4_reg *)data;
9134 
9135 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9136 			return (EFAULT);
9137 
9138 		if (edata->size == 4) {
9139 			if (edata->val & 0xffffffff00000000)
9140 				return (EINVAL);
9141 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9142 		} else if (edata->size == 8)
9143 			t4_write_reg64(sc, edata->addr, edata->val);
9144 		else
9145 			return (EINVAL);
9146 		break;
9147 	}
9148 	case CHELSIO_T4_REGDUMP: {
9149 		struct t4_regdump *regs = (struct t4_regdump *)data;
9150 		int reglen = t4_get_regs_len(sc);
9151 		uint8_t *buf;
9152 
9153 		if (regs->len < reglen) {
9154 			regs->len = reglen; /* hint to the caller */
9155 			return (ENOBUFS);
9156 		}
9157 
9158 		regs->len = reglen;
9159 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9160 		get_regs(sc, regs, buf);
9161 		rc = copyout(buf, regs->data, reglen);
9162 		free(buf, M_CXGBE);
9163 		break;
9164 	}
9165 	case CHELSIO_T4_GET_FILTER_MODE:
9166 		rc = get_filter_mode(sc, (uint32_t *)data);
9167 		break;
9168 	case CHELSIO_T4_SET_FILTER_MODE:
9169 		rc = set_filter_mode(sc, *(uint32_t *)data);
9170 		break;
9171 	case CHELSIO_T4_GET_FILTER:
9172 		rc = get_filter(sc, (struct t4_filter *)data);
9173 		break;
9174 	case CHELSIO_T4_SET_FILTER:
9175 		rc = set_filter(sc, (struct t4_filter *)data);
9176 		break;
9177 	case CHELSIO_T4_DEL_FILTER:
9178 		rc = del_filter(sc, (struct t4_filter *)data);
9179 		break;
9180 	case CHELSIO_T4_GET_SGE_CONTEXT:
9181 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9182 		break;
9183 	case CHELSIO_T4_LOAD_FW:
9184 		rc = load_fw(sc, (struct t4_data *)data);
9185 		break;
9186 	case CHELSIO_T4_GET_MEM:
9187 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9188 		break;
9189 	case CHELSIO_T4_GET_I2C:
9190 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9191 		break;
9192 	case CHELSIO_T4_CLEAR_STATS: {
9193 		int i, v;
9194 		u_int port_id = *(uint32_t *)data;
9195 		struct port_info *pi;
9196 		struct vi_info *vi;
9197 
9198 		if (port_id >= sc->params.nports)
9199 			return (EINVAL);
9200 		pi = sc->port[port_id];
9201 		if (pi == NULL)
9202 			return (EIO);
9203 
9204 		/* MAC stats */
9205 		t4_clr_port_stats(sc, pi->tx_chan);
9206 		pi->tx_parse_error = 0;
9207 		mtx_lock(&sc->reg_lock);
9208 		for_each_vi(pi, v, vi) {
9209 			if (vi->flags & VI_INIT_DONE)
9210 				t4_clr_vi_stats(sc, vi->viid);
9211 		}
9212 		mtx_unlock(&sc->reg_lock);
9213 
9214 		/*
9215 		 * Since this command accepts a port, clear stats for
9216 		 * all VIs on this port.
9217 		 */
9218 		for_each_vi(pi, v, vi) {
9219 			if (vi->flags & VI_INIT_DONE) {
9220 				struct sge_rxq *rxq;
9221 				struct sge_txq *txq;
9222 				struct sge_wrq *wrq;
9223 
9224 				for_each_rxq(vi, i, rxq) {
9225 #if defined(INET) || defined(INET6)
9226 					rxq->lro.lro_queued = 0;
9227 					rxq->lro.lro_flushed = 0;
9228 #endif
9229 					rxq->rxcsum = 0;
9230 					rxq->vlan_extraction = 0;
9231 				}
9232 
9233 				for_each_txq(vi, i, txq) {
9234 					txq->txcsum = 0;
9235 					txq->tso_wrs = 0;
9236 					txq->vlan_insertion = 0;
9237 					txq->imm_wrs = 0;
9238 					txq->sgl_wrs = 0;
9239 					txq->txpkt_wrs = 0;
9240 					txq->txpkts0_wrs = 0;
9241 					txq->txpkts1_wrs = 0;
9242 					txq->txpkts0_pkts = 0;
9243 					txq->txpkts1_pkts = 0;
9244 					mp_ring_reset_stats(txq->r);
9245 				}
9246 
9247 #ifdef TCP_OFFLOAD
9248 				/* nothing to clear for each ofld_rxq */
9249 
9250 				for_each_ofld_txq(vi, i, wrq) {
9251 					wrq->tx_wrs_direct = 0;
9252 					wrq->tx_wrs_copied = 0;
9253 				}
9254 #endif
9255 
9256 				if (IS_MAIN_VI(vi)) {
9257 					wrq = &sc->sge.ctrlq[pi->port_id];
9258 					wrq->tx_wrs_direct = 0;
9259 					wrq->tx_wrs_copied = 0;
9260 				}
9261 			}
9262 		}
9263 		break;
9264 	}
9265 	case CHELSIO_T4_SCHED_CLASS:
9266 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9267 		break;
9268 	case CHELSIO_T4_SCHED_QUEUE:
9269 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9270 		break;
9271 	case CHELSIO_T4_GET_TRACER:
9272 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9273 		break;
9274 	case CHELSIO_T4_SET_TRACER:
9275 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9276 		break;
9277 	case CHELSIO_T4_LOAD_CFG:
9278 		rc = load_cfg(sc, (struct t4_data *)data);
9279 		break;
9280 	case CHELSIO_T4_LOAD_BOOT:
9281 		rc = load_boot(sc, (struct t4_bootrom *)data);
9282 		break;
9283 	case CHELSIO_T4_LOAD_BOOTCFG:
9284 		rc = load_bootcfg(sc, (struct t4_data *)data);
9285 		break;
9286 	default:
9287 		rc = ENOTTY;
9288 	}
9289 
9290 	return (rc);
9291 }
9292 
9293 void
9294 t4_db_full(struct adapter *sc)
9295 {
9296 
9297 	CXGBE_UNIMPLEMENTED(__func__);
9298 }
9299 
9300 void
9301 t4_db_dropped(struct adapter *sc)
9302 {
9303 
9304 	CXGBE_UNIMPLEMENTED(__func__);
9305 }
9306 
9307 #ifdef TCP_OFFLOAD
9308 static int
9309 toe_capability(struct vi_info *vi, int enable)
9310 {
9311 	int rc;
9312 	struct port_info *pi = vi->pi;
9313 	struct adapter *sc = pi->adapter;
9314 
9315 	ASSERT_SYNCHRONIZED_OP(sc);
9316 
9317 	if (!is_offload(sc))
9318 		return (ENODEV);
9319 
9320 	if (enable) {
9321 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9322 			/* TOE is already enabled. */
9323 			return (0);
9324 		}
9325 
9326 		/*
9327 		 * We need the port's queues around so that we're able to send
9328 		 * and receive CPLs to/from the TOE even if the ifnet for this
9329 		 * port has never been UP'd administratively.
9330 		 */
9331 		if (!(vi->flags & VI_INIT_DONE)) {
9332 			rc = vi_full_init(vi);
9333 			if (rc)
9334 				return (rc);
9335 		}
9336 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9337 			rc = vi_full_init(&pi->vi[0]);
9338 			if (rc)
9339 				return (rc);
9340 		}
9341 
9342 		if (isset(&sc->offload_map, pi->port_id)) {
9343 			/* TOE is enabled on another VI of this port. */
9344 			pi->uld_vis++;
9345 			return (0);
9346 		}
9347 
9348 		if (!uld_active(sc, ULD_TOM)) {
9349 			rc = t4_activate_uld(sc, ULD_TOM);
9350 			if (rc == EAGAIN) {
9351 				log(LOG_WARNING,
9352 				    "You must kldload t4_tom.ko before trying "
9353 				    "to enable TOE on a cxgbe interface.\n");
9354 			}
9355 			if (rc != 0)
9356 				return (rc);
9357 			KASSERT(sc->tom_softc != NULL,
9358 			    ("%s: TOM activated but softc NULL", __func__));
9359 			KASSERT(uld_active(sc, ULD_TOM),
9360 			    ("%s: TOM activated but flag not set", __func__));
9361 		}
9362 
9363 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
9364 		if (!uld_active(sc, ULD_IWARP))
9365 			(void) t4_activate_uld(sc, ULD_IWARP);
9366 		if (!uld_active(sc, ULD_ISCSI))
9367 			(void) t4_activate_uld(sc, ULD_ISCSI);
9368 
9369 		pi->uld_vis++;
9370 		setbit(&sc->offload_map, pi->port_id);
9371 	} else {
9372 		pi->uld_vis--;
9373 
9374 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9375 			return (0);
9376 
9377 		KASSERT(uld_active(sc, ULD_TOM),
9378 		    ("%s: TOM never initialized?", __func__));
9379 		clrbit(&sc->offload_map, pi->port_id);
9380 	}
9381 
9382 	return (0);
9383 }
9384 
9385 /*
9386  * Add an upper layer driver to the global list.
9387  */
9388 int
9389 t4_register_uld(struct uld_info *ui)
9390 {
9391 	int rc = 0;
9392 	struct uld_info *u;
9393 
9394 	sx_xlock(&t4_uld_list_lock);
9395 	SLIST_FOREACH(u, &t4_uld_list, link) {
9396 	    if (u->uld_id == ui->uld_id) {
9397 		    rc = EEXIST;
9398 		    goto done;
9399 	    }
9400 	}
9401 
9402 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9403 	ui->refcount = 0;
9404 done:
9405 	sx_xunlock(&t4_uld_list_lock);
9406 	return (rc);
9407 }
9408 
9409 int
9410 t4_unregister_uld(struct uld_info *ui)
9411 {
9412 	int rc = EINVAL;
9413 	struct uld_info *u;
9414 
9415 	sx_xlock(&t4_uld_list_lock);
9416 
9417 	SLIST_FOREACH(u, &t4_uld_list, link) {
9418 	    if (u == ui) {
9419 		    if (ui->refcount > 0) {
9420 			    rc = EBUSY;
9421 			    goto done;
9422 		    }
9423 
9424 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9425 		    rc = 0;
9426 		    goto done;
9427 	    }
9428 	}
9429 done:
9430 	sx_xunlock(&t4_uld_list_lock);
9431 	return (rc);
9432 }
9433 
9434 int
9435 t4_activate_uld(struct adapter *sc, int id)
9436 {
9437 	int rc;
9438 	struct uld_info *ui;
9439 
9440 	ASSERT_SYNCHRONIZED_OP(sc);
9441 
9442 	if (id < 0 || id > ULD_MAX)
9443 		return (EINVAL);
9444 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
9445 
9446 	sx_slock(&t4_uld_list_lock);
9447 
9448 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9449 		if (ui->uld_id == id) {
9450 			if (!(sc->flags & FULL_INIT_DONE)) {
9451 				rc = adapter_full_init(sc);
9452 				if (rc != 0)
9453 					break;
9454 			}
9455 
9456 			rc = ui->activate(sc);
9457 			if (rc == 0) {
9458 				setbit(&sc->active_ulds, id);
9459 				ui->refcount++;
9460 			}
9461 			break;
9462 		}
9463 	}
9464 
9465 	sx_sunlock(&t4_uld_list_lock);
9466 
9467 	return (rc);
9468 }
9469 
9470 int
9471 t4_deactivate_uld(struct adapter *sc, int id)
9472 {
9473 	int rc;
9474 	struct uld_info *ui;
9475 
9476 	ASSERT_SYNCHRONIZED_OP(sc);
9477 
9478 	if (id < 0 || id > ULD_MAX)
9479 		return (EINVAL);
9480 	rc = ENXIO;
9481 
9482 	sx_slock(&t4_uld_list_lock);
9483 
9484 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9485 		if (ui->uld_id == id) {
9486 			rc = ui->deactivate(sc);
9487 			if (rc == 0) {
9488 				clrbit(&sc->active_ulds, id);
9489 				ui->refcount--;
9490 			}
9491 			break;
9492 		}
9493 	}
9494 
9495 	sx_sunlock(&t4_uld_list_lock);
9496 
9497 	return (rc);
9498 }
9499 
9500 int
9501 uld_active(struct adapter *sc, int uld_id)
9502 {
9503 
9504 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9505 
9506 	return (isset(&sc->active_ulds, uld_id));
9507 }
9508 #endif
9509 
9510 /*
9511  * t  = ptr to tunable.
9512  * nc = number of CPUs.
9513  * c  = compiled in default for that tunable.
9514  */
9515 static void
9516 calculate_nqueues(int *t, int nc, const int c)
9517 {
9518 	int nq;
9519 
9520 	if (*t > 0)
9521 		return;
9522 	nq = *t < 0 ? -*t : c;
9523 	*t = min(nc, nq);
9524 }
9525 
9526 /*
9527  * Come up with reasonable defaults for some of the tunables, provided they're
9528  * not set by the user (in which case we'll use the values as is).
9529  */
9530 static void
9531 tweak_tunables(void)
9532 {
9533 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
9534 
9535 	if (t4_ntxq10g < 1) {
9536 #ifdef RSS
9537 		t4_ntxq10g = rss_getnumbuckets();
9538 #else
9539 		calculate_nqueues(&t4_ntxq10g, nc, NTXQ_10G);
9540 #endif
9541 	}
9542 
9543 	if (t4_ntxq1g < 1) {
9544 #ifdef RSS
9545 		/* XXX: way too many for 1GbE? */
9546 		t4_ntxq1g = rss_getnumbuckets();
9547 #else
9548 		calculate_nqueues(&t4_ntxq1g, nc, NTXQ_1G);
9549 #endif
9550 	}
9551 
9552 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
9553 
9554 	if (t4_nrxq10g < 1) {
9555 #ifdef RSS
9556 		t4_nrxq10g = rss_getnumbuckets();
9557 #else
9558 		calculate_nqueues(&t4_nrxq10g, nc, NRXQ_10G);
9559 #endif
9560 	}
9561 
9562 	if (t4_nrxq1g < 1) {
9563 #ifdef RSS
9564 		/* XXX: way too many for 1GbE? */
9565 		t4_nrxq1g = rss_getnumbuckets();
9566 #else
9567 		calculate_nqueues(&t4_nrxq1g, nc, NRXQ_1G);
9568 #endif
9569 	}
9570 
9571 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
9572 
9573 #ifdef TCP_OFFLOAD
9574 	calculate_nqueues(&t4_nofldtxq10g, nc, NOFLDTXQ_10G);
9575 	calculate_nqueues(&t4_nofldtxq1g, nc, NOFLDTXQ_1G);
9576 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
9577 	calculate_nqueues(&t4_nofldrxq10g, nc, NOFLDRXQ_10G);
9578 	calculate_nqueues(&t4_nofldrxq1g, nc, NOFLDRXQ_1G);
9579 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
9580 
9581 	if (t4_toecaps_allowed == -1)
9582 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9583 
9584 	if (t4_rdmacaps_allowed == -1) {
9585 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9586 		    FW_CAPS_CONFIG_RDMA_RDMAC;
9587 	}
9588 
9589 	if (t4_iscsicaps_allowed == -1) {
9590 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9591 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9592 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
9593 	}
9594 #else
9595 	if (t4_toecaps_allowed == -1)
9596 		t4_toecaps_allowed = 0;
9597 
9598 	if (t4_rdmacaps_allowed == -1)
9599 		t4_rdmacaps_allowed = 0;
9600 
9601 	if (t4_iscsicaps_allowed == -1)
9602 		t4_iscsicaps_allowed = 0;
9603 #endif
9604 
9605 #ifdef DEV_NETMAP
9606 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
9607 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
9608 #endif
9609 
9610 	if (t4_tmr_idx_10g < 0 || t4_tmr_idx_10g >= SGE_NTIMERS)
9611 		t4_tmr_idx_10g = TMR_IDX_10G;
9612 
9613 	if (t4_pktc_idx_10g < -1 || t4_pktc_idx_10g >= SGE_NCOUNTERS)
9614 		t4_pktc_idx_10g = PKTC_IDX_10G;
9615 
9616 	if (t4_tmr_idx_1g < 0 || t4_tmr_idx_1g >= SGE_NTIMERS)
9617 		t4_tmr_idx_1g = TMR_IDX_1G;
9618 
9619 	if (t4_pktc_idx_1g < -1 || t4_pktc_idx_1g >= SGE_NCOUNTERS)
9620 		t4_pktc_idx_1g = PKTC_IDX_1G;
9621 
9622 	if (t4_qsize_txq < 128)
9623 		t4_qsize_txq = 128;
9624 
9625 	if (t4_qsize_rxq < 128)
9626 		t4_qsize_rxq = 128;
9627 	while (t4_qsize_rxq & 7)
9628 		t4_qsize_rxq++;
9629 
9630 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9631 }
9632 
9633 #ifdef DDB
9634 static void
9635 t4_dump_tcb(struct adapter *sc, int tid)
9636 {
9637 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
9638 
9639 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
9640 	save = t4_read_reg(sc, reg);
9641 	base = sc->memwin[2].mw_base;
9642 
9643 	/* Dump TCB for the tid */
9644 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
9645 	tcb_addr += tid * TCB_SIZE;
9646 
9647 	if (is_t4(sc)) {
9648 		pf = 0;
9649 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
9650 	} else {
9651 		pf = V_PFNUM(sc->pf);
9652 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
9653 	}
9654 	t4_write_reg(sc, reg, win_pos | pf);
9655 	t4_read_reg(sc, reg);
9656 
9657 	off = tcb_addr - win_pos;
9658 	for (i = 0; i < 4; i++) {
9659 		uint32_t buf[8];
9660 		for (j = 0; j < 8; j++, off += 4)
9661 			buf[j] = htonl(t4_read_reg(sc, base + off));
9662 
9663 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
9664 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
9665 		    buf[7]);
9666 	}
9667 
9668 	t4_write_reg(sc, reg, save);
9669 	t4_read_reg(sc, reg);
9670 }
9671 
9672 static void
9673 t4_dump_devlog(struct adapter *sc)
9674 {
9675 	struct devlog_params *dparams = &sc->params.devlog;
9676 	struct fw_devlog_e e;
9677 	int i, first, j, m, nentries, rc;
9678 	uint64_t ftstamp = UINT64_MAX;
9679 
9680 	if (dparams->start == 0) {
9681 		db_printf("devlog params not valid\n");
9682 		return;
9683 	}
9684 
9685 	nentries = dparams->size / sizeof(struct fw_devlog_e);
9686 	m = fwmtype_to_hwmtype(dparams->memtype);
9687 
9688 	/* Find the first entry. */
9689 	first = -1;
9690 	for (i = 0; i < nentries && !db_pager_quit; i++) {
9691 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9692 		    sizeof(e), (void *)&e);
9693 		if (rc != 0)
9694 			break;
9695 
9696 		if (e.timestamp == 0)
9697 			break;
9698 
9699 		e.timestamp = be64toh(e.timestamp);
9700 		if (e.timestamp < ftstamp) {
9701 			ftstamp = e.timestamp;
9702 			first = i;
9703 		}
9704 	}
9705 
9706 	if (first == -1)
9707 		return;
9708 
9709 	i = first;
9710 	do {
9711 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9712 		    sizeof(e), (void *)&e);
9713 		if (rc != 0)
9714 			return;
9715 
9716 		if (e.timestamp == 0)
9717 			return;
9718 
9719 		e.timestamp = be64toh(e.timestamp);
9720 		e.seqno = be32toh(e.seqno);
9721 		for (j = 0; j < 8; j++)
9722 			e.params[j] = be32toh(e.params[j]);
9723 
9724 		db_printf("%10d  %15ju  %8s  %8s  ",
9725 		    e.seqno, e.timestamp,
9726 		    (e.level < nitems(devlog_level_strings) ?
9727 			devlog_level_strings[e.level] : "UNKNOWN"),
9728 		    (e.facility < nitems(devlog_facility_strings) ?
9729 			devlog_facility_strings[e.facility] : "UNKNOWN"));
9730 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
9731 		    e.params[3], e.params[4], e.params[5], e.params[6],
9732 		    e.params[7]);
9733 
9734 		if (++i == nentries)
9735 			i = 0;
9736 	} while (i != first && !db_pager_quit);
9737 }
9738 
9739 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
9740 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
9741 
9742 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
9743 {
9744 	device_t dev;
9745 	int t;
9746 	bool valid;
9747 
9748 	valid = false;
9749 	t = db_read_token();
9750 	if (t == tIDENT) {
9751 		dev = device_lookup_by_name(db_tok_string);
9752 		valid = true;
9753 	}
9754 	db_skip_to_eol();
9755 	if (!valid) {
9756 		db_printf("usage: show t4 devlog <nexus>\n");
9757 		return;
9758 	}
9759 
9760 	if (dev == NULL) {
9761 		db_printf("device not found\n");
9762 		return;
9763 	}
9764 
9765 	t4_dump_devlog(device_get_softc(dev));
9766 }
9767 
9768 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
9769 {
9770 	device_t dev;
9771 	int radix, tid, t;
9772 	bool valid;
9773 
9774 	valid = false;
9775 	radix = db_radix;
9776 	db_radix = 10;
9777 	t = db_read_token();
9778 	if (t == tIDENT) {
9779 		dev = device_lookup_by_name(db_tok_string);
9780 		t = db_read_token();
9781 		if (t == tNUMBER) {
9782 			tid = db_tok_number;
9783 			valid = true;
9784 		}
9785 	}
9786 	db_radix = radix;
9787 	db_skip_to_eol();
9788 	if (!valid) {
9789 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
9790 		return;
9791 	}
9792 
9793 	if (dev == NULL) {
9794 		db_printf("device not found\n");
9795 		return;
9796 	}
9797 	if (tid < 0) {
9798 		db_printf("invalid tid\n");
9799 		return;
9800 	}
9801 
9802 	t4_dump_tcb(device_get_softc(dev), tid);
9803 }
9804 #endif
9805 
9806 static struct sx mlu;	/* mod load unload */
9807 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
9808 
9809 static int
9810 mod_event(module_t mod, int cmd, void *arg)
9811 {
9812 	int rc = 0;
9813 	static int loaded = 0;
9814 
9815 	switch (cmd) {
9816 	case MOD_LOAD:
9817 		sx_xlock(&mlu);
9818 		if (loaded++ == 0) {
9819 			t4_sge_modload();
9820 			t4_register_cpl_handler(CPL_SET_TCB_RPL, set_tcb_rpl);
9821 			t4_register_cpl_handler(CPL_L2T_WRITE_RPL, l2t_write_rpl);
9822 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
9823 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
9824 			sx_init(&t4_list_lock, "T4/T5 adapters");
9825 			SLIST_INIT(&t4_list);
9826 #ifdef TCP_OFFLOAD
9827 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
9828 			SLIST_INIT(&t4_uld_list);
9829 #endif
9830 			t4_tracer_modload();
9831 			tweak_tunables();
9832 		}
9833 		sx_xunlock(&mlu);
9834 		break;
9835 
9836 	case MOD_UNLOAD:
9837 		sx_xlock(&mlu);
9838 		if (--loaded == 0) {
9839 			int tries;
9840 
9841 			sx_slock(&t4_list_lock);
9842 			if (!SLIST_EMPTY(&t4_list)) {
9843 				rc = EBUSY;
9844 				sx_sunlock(&t4_list_lock);
9845 				goto done_unload;
9846 			}
9847 #ifdef TCP_OFFLOAD
9848 			sx_slock(&t4_uld_list_lock);
9849 			if (!SLIST_EMPTY(&t4_uld_list)) {
9850 				rc = EBUSY;
9851 				sx_sunlock(&t4_uld_list_lock);
9852 				sx_sunlock(&t4_list_lock);
9853 				goto done_unload;
9854 			}
9855 #endif
9856 			tries = 0;
9857 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
9858 				uprintf("%ju clusters with custom free routine "
9859 				    "still is use.\n", t4_sge_extfree_refs());
9860 				pause("t4unload", 2 * hz);
9861 			}
9862 #ifdef TCP_OFFLOAD
9863 			sx_sunlock(&t4_uld_list_lock);
9864 #endif
9865 			sx_sunlock(&t4_list_lock);
9866 
9867 			if (t4_sge_extfree_refs() == 0) {
9868 				t4_tracer_modunload();
9869 #ifdef TCP_OFFLOAD
9870 				sx_destroy(&t4_uld_list_lock);
9871 #endif
9872 				sx_destroy(&t4_list_lock);
9873 				t4_sge_modunload();
9874 				loaded = 0;
9875 			} else {
9876 				rc = EBUSY;
9877 				loaded++;	/* undo earlier decrement */
9878 			}
9879 		}
9880 done_unload:
9881 		sx_xunlock(&mlu);
9882 		break;
9883 	}
9884 
9885 	return (rc);
9886 }
9887 
9888 static devclass_t t4_devclass, t5_devclass, t6_devclass;
9889 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
9890 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
9891 
9892 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
9893 MODULE_VERSION(t4nex, 1);
9894 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
9895 #ifdef DEV_NETMAP
9896 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
9897 #endif /* DEV_NETMAP */
9898 
9899 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
9900 MODULE_VERSION(t5nex, 1);
9901 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
9902 #ifdef DEV_NETMAP
9903 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
9904 #endif /* DEV_NETMAP */
9905 
9906 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
9907 MODULE_VERSION(t6nex, 1);
9908 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
9909 #ifdef DEV_NETMAP
9910 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
9911 #endif /* DEV_NETMAP */
9912 
9913 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
9914 MODULE_VERSION(cxgbe, 1);
9915 
9916 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
9917 MODULE_VERSION(cxl, 1);
9918 
9919 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
9920 MODULE_VERSION(cc, 1);
9921 
9922 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
9923 MODULE_VERSION(vcxgbe, 1);
9924 
9925 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
9926 MODULE_VERSION(vcxl, 1);
9927 
9928 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
9929 MODULE_VERSION(vcc, 1);
9930