xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision cc349066556bcdeed0d6cc72aad340d0f383e35c)
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 -1 as an indication to tweak_tunables() that it should
236  * provide a reasonable default 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 = -1;
247 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq10g);
248 
249 #define NRXQ_10G 8
250 int t4_nrxq10g = -1;
251 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq10g);
252 
253 #define NTXQ_1G 4
254 int t4_ntxq1g = -1;
255 TUNABLE_INT("hw.cxgbe.ntxq1g", &t4_ntxq1g);
256 
257 #define NRXQ_1G 2
258 int t4_nrxq1g = -1;
259 TUNABLE_INT("hw.cxgbe.nrxq1g", &t4_nrxq1g);
260 
261 #define NTXQ_VI 1
262 static int t4_ntxq_vi = -1;
263 TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
264 
265 #define NRXQ_VI 1
266 static int t4_nrxq_vi = -1;
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 = -1;
275 TUNABLE_INT("hw.cxgbe.nofldtxq10g", &t4_nofldtxq10g);
276 
277 #define NOFLDRXQ_10G 2
278 static int t4_nofldrxq10g = -1;
279 TUNABLE_INT("hw.cxgbe.nofldrxq10g", &t4_nofldrxq10g);
280 
281 #define NOFLDTXQ_1G 2
282 static int t4_nofldtxq1g = -1;
283 TUNABLE_INT("hw.cxgbe.nofldtxq1g", &t4_nofldtxq1g);
284 
285 #define NOFLDRXQ_1G 1
286 static int t4_nofldrxq1g = -1;
287 TUNABLE_INT("hw.cxgbe.nofldrxq1g", &t4_nofldrxq1g);
288 
289 #define NOFLDTXQ_VI 1
290 static int t4_nofldtxq_vi = -1;
291 TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
292 
293 #define NOFLDRXQ_VI 1
294 static int t4_nofldrxq_vi = -1;
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 = -1;
301 TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
302 
303 #define NNMRXQ_VI 2
304 static int t4_nnmrxq_vi = -1;
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 int cxgbe_init_synchronized(struct vi_info *);
491 static int cxgbe_uninit_synchronized(struct vi_info *);
492 static void quiesce_txq(struct adapter *, struct sge_txq *);
493 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
494 static void quiesce_iq(struct adapter *, struct sge_iq *);
495 static void quiesce_fl(struct adapter *, struct sge_fl *);
496 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
497     driver_intr_t *, void *, char *);
498 static int t4_free_irq(struct adapter *, struct irq *);
499 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
500 static void vi_refresh_stats(struct adapter *, struct vi_info *);
501 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
502 static void cxgbe_tick(void *);
503 static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
504 static void cxgbe_sysctls(struct port_info *);
505 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
506 static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
507 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
508 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
509 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
510 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
511 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
512 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
513 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
514 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
515 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
516 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
517 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
518 #ifdef SBUF_DRAIN
519 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
520 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
521 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
522 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
523 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
524 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
525 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
526 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
527 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
528 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
529 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
530 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
531 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
532 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
533 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
534 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
535 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
536 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
537 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
538 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
539 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
540 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
541 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
542 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
543 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
544 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
545 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
546 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
547 static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
548 #endif
549 #ifdef TCP_OFFLOAD
550 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
551 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
552 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
553 #endif
554 static uint32_t fconf_iconf_to_mode(uint32_t, uint32_t);
555 static uint32_t mode_to_fconf(uint32_t);
556 static uint32_t mode_to_iconf(uint32_t);
557 static int check_fspec_against_fconf_iconf(struct adapter *,
558     struct t4_filter_specification *);
559 static int get_filter_mode(struct adapter *, uint32_t *);
560 static int set_filter_mode(struct adapter *, uint32_t);
561 static inline uint64_t get_filter_hits(struct adapter *, uint32_t);
562 static int get_filter(struct adapter *, struct t4_filter *);
563 static int set_filter(struct adapter *, struct t4_filter *);
564 static int del_filter(struct adapter *, struct t4_filter *);
565 static void clear_filter(struct filter_entry *);
566 static int set_filter_wr(struct adapter *, int);
567 static int del_filter_wr(struct adapter *, int);
568 static int set_tcb_rpl(struct sge_iq *, const struct rss_header *,
569     struct mbuf *);
570 static int get_sge_context(struct adapter *, struct t4_sge_context *);
571 static int load_fw(struct adapter *, struct t4_data *);
572 static int load_cfg(struct adapter *, struct t4_data *);
573 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
574 static int read_i2c(struct adapter *, struct t4_i2c_data *);
575 #ifdef TCP_OFFLOAD
576 static int toe_capability(struct vi_info *, int);
577 #endif
578 static int mod_event(module_t, int, void *);
579 static int notify_siblings(device_t, int);
580 
581 struct {
582 	uint16_t device;
583 	char *desc;
584 } t4_pciids[] = {
585 	{0xa000, "Chelsio Terminator 4 FPGA"},
586 	{0x4400, "Chelsio T440-dbg"},
587 	{0x4401, "Chelsio T420-CR"},
588 	{0x4402, "Chelsio T422-CR"},
589 	{0x4403, "Chelsio T440-CR"},
590 	{0x4404, "Chelsio T420-BCH"},
591 	{0x4405, "Chelsio T440-BCH"},
592 	{0x4406, "Chelsio T440-CH"},
593 	{0x4407, "Chelsio T420-SO"},
594 	{0x4408, "Chelsio T420-CX"},
595 	{0x4409, "Chelsio T420-BT"},
596 	{0x440a, "Chelsio T404-BT"},
597 	{0x440e, "Chelsio T440-LP-CR"},
598 }, t5_pciids[] = {
599 	{0xb000, "Chelsio Terminator 5 FPGA"},
600 	{0x5400, "Chelsio T580-dbg"},
601 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
602 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
603 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
604 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
605 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
606 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
607 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
608 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
609 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
610 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
611 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
612 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
613 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
614 #ifdef notyet
615 	{0x5404,  "Chelsio T520-BCH"},
616 	{0x5405,  "Chelsio T540-BCH"},
617 	{0x5406,  "Chelsio T540-CH"},
618 	{0x5408,  "Chelsio T520-CX"},
619 	{0x540b,  "Chelsio B520-SR"},
620 	{0x540c,  "Chelsio B504-BT"},
621 	{0x540f,  "Chelsio Amsterdam"},
622 	{0x5413,  "Chelsio T580-CHR"},
623 #endif
624 }, t6_pciids[] = {
625 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
626 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
627 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
628 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
629 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
630 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
631 	{0x6410, "Chelsio T62100-DBG"},		/* 2 x 40/50/100G, debug */
632 };
633 
634 #ifdef TCP_OFFLOAD
635 /*
636  * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
637  * exactly the same for both rxq and ofld_rxq.
638  */
639 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
640 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
641 #endif
642 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
643 
644 static int
645 t4_probe(device_t dev)
646 {
647 	int i;
648 	uint16_t v = pci_get_vendor(dev);
649 	uint16_t d = pci_get_device(dev);
650 	uint8_t f = pci_get_function(dev);
651 
652 	if (v != PCI_VENDOR_ID_CHELSIO)
653 		return (ENXIO);
654 
655 	/* Attach only to PF0 of the FPGA */
656 	if (d == 0xa000 && f != 0)
657 		return (ENXIO);
658 
659 	for (i = 0; i < nitems(t4_pciids); i++) {
660 		if (d == t4_pciids[i].device) {
661 			device_set_desc(dev, t4_pciids[i].desc);
662 			return (BUS_PROBE_DEFAULT);
663 		}
664 	}
665 
666 	return (ENXIO);
667 }
668 
669 static int
670 t5_probe(device_t dev)
671 {
672 	int i;
673 	uint16_t v = pci_get_vendor(dev);
674 	uint16_t d = pci_get_device(dev);
675 	uint8_t f = pci_get_function(dev);
676 
677 	if (v != PCI_VENDOR_ID_CHELSIO)
678 		return (ENXIO);
679 
680 	/* Attach only to PF0 of the FPGA */
681 	if (d == 0xb000 && f != 0)
682 		return (ENXIO);
683 
684 	for (i = 0; i < nitems(t5_pciids); i++) {
685 		if (d == t5_pciids[i].device) {
686 			device_set_desc(dev, t5_pciids[i].desc);
687 			return (BUS_PROBE_DEFAULT);
688 		}
689 	}
690 
691 	return (ENXIO);
692 }
693 
694 static int
695 t6_probe(device_t dev)
696 {
697 	int i;
698 	uint16_t v = pci_get_vendor(dev);
699 	uint16_t d = pci_get_device(dev);
700 
701 	if (v != PCI_VENDOR_ID_CHELSIO)
702 		return (ENXIO);
703 
704 	for (i = 0; i < nitems(t6_pciids); i++) {
705 		if (d == t6_pciids[i].device) {
706 			device_set_desc(dev, t6_pciids[i].desc);
707 			return (BUS_PROBE_DEFAULT);
708 		}
709 	}
710 
711 	return (ENXIO);
712 }
713 
714 static void
715 t5_attribute_workaround(device_t dev)
716 {
717 	device_t root_port;
718 	uint32_t v;
719 
720 	/*
721 	 * The T5 chips do not properly echo the No Snoop and Relaxed
722 	 * Ordering attributes when replying to a TLP from a Root
723 	 * Port.  As a workaround, find the parent Root Port and
724 	 * disable No Snoop and Relaxed Ordering.  Note that this
725 	 * affects all devices under this root port.
726 	 */
727 	root_port = pci_find_pcie_root_port(dev);
728 	if (root_port == NULL) {
729 		device_printf(dev, "Unable to find parent root port\n");
730 		return;
731 	}
732 
733 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
734 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
735 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
736 	    0)
737 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
738 		    device_get_nameunit(root_port));
739 }
740 
741 static const struct devnames devnames[] = {
742 	{
743 		.nexus_name = "t4nex",
744 		.ifnet_name = "cxgbe",
745 		.vi_ifnet_name = "vcxgbe",
746 		.pf03_drv_name = "t4iov",
747 		.vf_nexus_name = "t4vf",
748 		.vf_ifnet_name = "cxgbev"
749 	}, {
750 		.nexus_name = "t5nex",
751 		.ifnet_name = "cxl",
752 		.vi_ifnet_name = "vcxl",
753 		.pf03_drv_name = "t5iov",
754 		.vf_nexus_name = "t5vf",
755 		.vf_ifnet_name = "cxlv"
756 	}, {
757 		.nexus_name = "t6nex",
758 		.ifnet_name = "cc",
759 		.vi_ifnet_name = "vcc",
760 		.pf03_drv_name = "t6iov",
761 		.vf_nexus_name = "t6vf",
762 		.vf_ifnet_name = "ccv"
763 	}
764 };
765 
766 void
767 t4_init_devnames(struct adapter *sc)
768 {
769 	int id;
770 
771 	id = chip_id(sc);
772 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
773 		sc->names = &devnames[id - CHELSIO_T4];
774 	else {
775 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
776 		sc->names = NULL;
777 	}
778 }
779 
780 static int
781 t4_attach(device_t dev)
782 {
783 	struct adapter *sc;
784 	int rc = 0, i, j, n10g, n1g, rqidx, tqidx;
785 	struct make_dev_args mda;
786 	struct intrs_and_queues iaq;
787 	struct sge *s;
788 	uint8_t *buf;
789 #ifdef TCP_OFFLOAD
790 	int ofld_rqidx, ofld_tqidx;
791 #endif
792 #ifdef DEV_NETMAP
793 	int nm_rqidx, nm_tqidx;
794 #endif
795 	int num_vis;
796 
797 	sc = device_get_softc(dev);
798 	sc->dev = dev;
799 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
800 
801 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
802 		t5_attribute_workaround(dev);
803 	pci_enable_busmaster(dev);
804 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
805 		uint32_t v;
806 
807 		pci_set_max_read_req(dev, 4096);
808 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
809 		v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
810 		pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
811 
812 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
813 	}
814 
815 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
816 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
817 	sc->traceq = -1;
818 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
819 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
820 	    device_get_nameunit(dev));
821 
822 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
823 	    device_get_nameunit(dev));
824 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
825 	t4_add_adapter(sc);
826 
827 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
828 	TAILQ_INIT(&sc->sfl);
829 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
830 
831 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
832 
833 	rc = t4_map_bars_0_and_4(sc);
834 	if (rc != 0)
835 		goto done; /* error message displayed already */
836 
837 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
838 
839 	/* Prepare the adapter for operation. */
840 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
841 	rc = -t4_prep_adapter(sc, buf);
842 	free(buf, M_CXGBE);
843 	if (rc != 0) {
844 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
845 		goto done;
846 	}
847 
848 	/*
849 	 * This is the real PF# to which we're attaching.  Works from within PCI
850 	 * passthrough environments too, where pci_get_function() could return a
851 	 * different PF# depending on the passthrough configuration.  We need to
852 	 * use the real PF# in all our communication with the firmware.
853 	 */
854 	j = t4_read_reg(sc, A_PL_WHOAMI);
855 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
856 	sc->mbox = sc->pf;
857 
858 	t4_init_devnames(sc);
859 	if (sc->names == NULL) {
860 		rc = ENOTSUP;
861 		goto done; /* error message displayed already */
862 	}
863 
864 	/*
865 	 * Do this really early, with the memory windows set up even before the
866 	 * character device.  The userland tool's register i/o and mem read
867 	 * will work even in "recovery mode".
868 	 */
869 	setup_memwin(sc);
870 	if (t4_init_devlog_params(sc, 0) == 0)
871 		fixup_devlog_params(sc);
872 	make_dev_args_init(&mda);
873 	mda.mda_devsw = &t4_cdevsw;
874 	mda.mda_uid = UID_ROOT;
875 	mda.mda_gid = GID_WHEEL;
876 	mda.mda_mode = 0600;
877 	mda.mda_si_drv1 = sc;
878 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
879 	if (rc != 0)
880 		device_printf(dev, "failed to create nexus char device: %d.\n",
881 		    rc);
882 
883 	/* Go no further if recovery mode has been requested. */
884 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
885 		device_printf(dev, "recovery mode.\n");
886 		goto done;
887 	}
888 
889 #if defined(__i386__)
890 	if ((cpu_feature & CPUID_CX8) == 0) {
891 		device_printf(dev, "64 bit atomics not available.\n");
892 		rc = ENOTSUP;
893 		goto done;
894 	}
895 #endif
896 
897 	/* Prepare the firmware for operation */
898 	rc = prep_firmware(sc);
899 	if (rc != 0)
900 		goto done; /* error message displayed already */
901 
902 	rc = get_params__post_init(sc);
903 	if (rc != 0)
904 		goto done; /* error message displayed already */
905 
906 	rc = set_params__post_init(sc);
907 	if (rc != 0)
908 		goto done; /* error message displayed already */
909 
910 	rc = t4_map_bar_2(sc);
911 	if (rc != 0)
912 		goto done; /* error message displayed already */
913 
914 	rc = t4_create_dma_tag(sc);
915 	if (rc != 0)
916 		goto done; /* error message displayed already */
917 
918 	/*
919 	 * Number of VIs to create per-port.  The first VI is the "main" regular
920 	 * VI for the port.  The rest are additional virtual interfaces on the
921 	 * same physical port.  Note that the main VI does not have native
922 	 * netmap support but the extra VIs do.
923 	 *
924 	 * Limit the number of VIs per port to the number of available
925 	 * MAC addresses per port.
926 	 */
927 	if (t4_num_vis >= 1)
928 		num_vis = t4_num_vis;
929 	else
930 		num_vis = 1;
931 	if (num_vis > nitems(vi_mac_funcs)) {
932 		num_vis = nitems(vi_mac_funcs);
933 		device_printf(dev, "Number of VIs limited to %d\n", num_vis);
934 	}
935 
936 	/*
937 	 * First pass over all the ports - allocate VIs and initialize some
938 	 * basic parameters like mac address, port type, etc.  We also figure
939 	 * out whether a port is 10G or 1G and use that information when
940 	 * calculating how many interrupts to attempt to allocate.
941 	 */
942 	n10g = n1g = 0;
943 	for_each_port(sc, i) {
944 		struct port_info *pi;
945 		struct link_config *lc;
946 
947 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
948 		sc->port[i] = pi;
949 
950 		/* These must be set before t4_port_init */
951 		pi->adapter = sc;
952 		pi->port_id = i;
953 		/*
954 		 * XXX: vi[0] is special so we can't delay this allocation until
955 		 * pi->nvi's final value is known.
956 		 */
957 		pi->vi = malloc(sizeof(struct vi_info) * num_vis, M_CXGBE,
958 		    M_ZERO | M_WAITOK);
959 
960 		/*
961 		 * Allocate the "main" VI and initialize parameters
962 		 * like mac addr.
963 		 */
964 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
965 		if (rc != 0) {
966 			device_printf(dev, "unable to initialize port %d: %d\n",
967 			    i, rc);
968 			free(pi->vi, M_CXGBE);
969 			free(pi, M_CXGBE);
970 			sc->port[i] = NULL;
971 			goto done;
972 		}
973 
974 		lc = &pi->link_cfg;
975 		lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
976 		lc->requested_fc |= t4_pause_settings;
977 		if (t4_fec != -1) {
978 			lc->requested_fec = t4_fec &
979 			    G_FW_PORT_CAP_FEC(lc->supported);
980 		}
981 		if (lc->supported & FW_PORT_CAP_ANEG && t4_autoneg != -1) {
982 			lc->autoneg = t4_autoneg ? AUTONEG_ENABLE :
983 			    AUTONEG_DISABLE;
984 		}
985 
986 		rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
987 		if (rc != 0) {
988 			device_printf(dev, "port %d l1cfg failed: %d\n", i, rc);
989 			free(pi->vi, M_CXGBE);
990 			free(pi, M_CXGBE);
991 			sc->port[i] = NULL;
992 			goto done;
993 		}
994 
995 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
996 		    device_get_nameunit(dev), i);
997 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
998 		sc->chan_map[pi->tx_chan] = i;
999 
1000 		pi->tc = malloc(sizeof(struct tx_sched_class) *
1001 		    sc->chip_params->nsched_cls, M_CXGBE, M_ZERO | M_WAITOK);
1002 
1003 		if (port_top_speed(pi) >= 10) {
1004 			n10g++;
1005 		} else {
1006 			n1g++;
1007 		}
1008 
1009 		pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
1010 		if (pi->dev == NULL) {
1011 			device_printf(dev,
1012 			    "failed to add device for port %d.\n", i);
1013 			rc = ENXIO;
1014 			goto done;
1015 		}
1016 		pi->vi[0].dev = pi->dev;
1017 		device_set_softc(pi->dev, pi);
1018 	}
1019 
1020 	/*
1021 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1022 	 */
1023 	rc = cfg_itype_and_nqueues(sc, n10g, n1g, num_vis, &iaq);
1024 	if (rc != 0)
1025 		goto done; /* error message displayed already */
1026 	if (iaq.nrxq_vi + iaq.nofldrxq_vi + iaq.nnmrxq_vi == 0)
1027 		num_vis = 1;
1028 
1029 	sc->intr_type = iaq.intr_type;
1030 	sc->intr_count = iaq.nirq;
1031 
1032 	s = &sc->sge;
1033 	s->nrxq = n10g * iaq.nrxq10g + n1g * iaq.nrxq1g;
1034 	s->ntxq = n10g * iaq.ntxq10g + n1g * iaq.ntxq1g;
1035 	if (num_vis > 1) {
1036 		s->nrxq += (n10g + n1g) * (num_vis - 1) * iaq.nrxq_vi;
1037 		s->ntxq += (n10g + n1g) * (num_vis - 1) * iaq.ntxq_vi;
1038 	}
1039 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1040 	s->neq += sc->params.nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
1041 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1042 #ifdef TCP_OFFLOAD
1043 	if (is_offload(sc)) {
1044 		s->nofldrxq = n10g * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
1045 		s->nofldtxq = n10g * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
1046 		if (num_vis > 1) {
1047 			s->nofldrxq += (n10g + n1g) * (num_vis - 1) *
1048 			    iaq.nofldrxq_vi;
1049 			s->nofldtxq += (n10g + n1g) * (num_vis - 1) *
1050 			    iaq.nofldtxq_vi;
1051 		}
1052 		s->neq += s->nofldtxq + s->nofldrxq;
1053 		s->niq += s->nofldrxq;
1054 
1055 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1056 		    M_CXGBE, M_ZERO | M_WAITOK);
1057 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1058 		    M_CXGBE, M_ZERO | M_WAITOK);
1059 	}
1060 #endif
1061 #ifdef DEV_NETMAP
1062 	if (num_vis > 1) {
1063 		s->nnmrxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmrxq_vi;
1064 		s->nnmtxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmtxq_vi;
1065 	}
1066 	s->neq += s->nnmtxq + s->nnmrxq;
1067 	s->niq += s->nnmrxq;
1068 
1069 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1070 	    M_CXGBE, M_ZERO | M_WAITOK);
1071 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1072 	    M_CXGBE, M_ZERO | M_WAITOK);
1073 #endif
1074 
1075 	s->ctrlq = malloc(sc->params.nports * sizeof(struct sge_wrq), M_CXGBE,
1076 	    M_ZERO | M_WAITOK);
1077 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1078 	    M_ZERO | M_WAITOK);
1079 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1080 	    M_ZERO | M_WAITOK);
1081 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1082 	    M_ZERO | M_WAITOK);
1083 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1084 	    M_ZERO | M_WAITOK);
1085 
1086 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1087 	    M_ZERO | M_WAITOK);
1088 
1089 	t4_init_l2t(sc, M_WAITOK);
1090 
1091 	/*
1092 	 * Second pass over the ports.  This time we know the number of rx and
1093 	 * tx queues that each port should get.
1094 	 */
1095 	rqidx = tqidx = 0;
1096 #ifdef TCP_OFFLOAD
1097 	ofld_rqidx = ofld_tqidx = 0;
1098 #endif
1099 #ifdef DEV_NETMAP
1100 	nm_rqidx = nm_tqidx = 0;
1101 #endif
1102 	for_each_port(sc, i) {
1103 		struct port_info *pi = sc->port[i];
1104 		struct vi_info *vi;
1105 
1106 		if (pi == NULL)
1107 			continue;
1108 
1109 		pi->nvi = num_vis;
1110 		for_each_vi(pi, j, vi) {
1111 			vi->pi = pi;
1112 			vi->qsize_rxq = t4_qsize_rxq;
1113 			vi->qsize_txq = t4_qsize_txq;
1114 
1115 			vi->first_rxq = rqidx;
1116 			vi->first_txq = tqidx;
1117 			if (port_top_speed(pi) >= 10) {
1118 				vi->tmr_idx = t4_tmr_idx_10g;
1119 				vi->pktc_idx = t4_pktc_idx_10g;
1120 				vi->flags |= iaq.intr_flags_10g & INTR_RXQ;
1121 				vi->nrxq = j == 0 ? iaq.nrxq10g : iaq.nrxq_vi;
1122 				vi->ntxq = j == 0 ? iaq.ntxq10g : iaq.ntxq_vi;
1123 			} else {
1124 				vi->tmr_idx = t4_tmr_idx_1g;
1125 				vi->pktc_idx = t4_pktc_idx_1g;
1126 				vi->flags |= iaq.intr_flags_1g & INTR_RXQ;
1127 				vi->nrxq = j == 0 ? iaq.nrxq1g : iaq.nrxq_vi;
1128 				vi->ntxq = j == 0 ? iaq.ntxq1g : iaq.ntxq_vi;
1129 			}
1130 			rqidx += vi->nrxq;
1131 			tqidx += vi->ntxq;
1132 
1133 			if (j == 0 && vi->ntxq > 1)
1134 				vi->rsrv_noflowq = iaq.rsrv_noflowq ? 1 : 0;
1135 			else
1136 				vi->rsrv_noflowq = 0;
1137 
1138 #ifdef TCP_OFFLOAD
1139 			vi->first_ofld_rxq = ofld_rqidx;
1140 			vi->first_ofld_txq = ofld_tqidx;
1141 			if (port_top_speed(pi) >= 10) {
1142 				vi->flags |= iaq.intr_flags_10g & INTR_OFLD_RXQ;
1143 				vi->nofldrxq = j == 0 ? iaq.nofldrxq10g :
1144 				    iaq.nofldrxq_vi;
1145 				vi->nofldtxq = j == 0 ? iaq.nofldtxq10g :
1146 				    iaq.nofldtxq_vi;
1147 			} else {
1148 				vi->flags |= iaq.intr_flags_1g & INTR_OFLD_RXQ;
1149 				vi->nofldrxq = j == 0 ? iaq.nofldrxq1g :
1150 				    iaq.nofldrxq_vi;
1151 				vi->nofldtxq = j == 0 ? iaq.nofldtxq1g :
1152 				    iaq.nofldtxq_vi;
1153 			}
1154 			ofld_rqidx += vi->nofldrxq;
1155 			ofld_tqidx += vi->nofldtxq;
1156 #endif
1157 #ifdef DEV_NETMAP
1158 			if (j > 0) {
1159 				vi->first_nm_rxq = nm_rqidx;
1160 				vi->first_nm_txq = nm_tqidx;
1161 				vi->nnmrxq = iaq.nnmrxq_vi;
1162 				vi->nnmtxq = iaq.nnmtxq_vi;
1163 				nm_rqidx += vi->nnmrxq;
1164 				nm_tqidx += vi->nnmtxq;
1165 			}
1166 #endif
1167 		}
1168 	}
1169 
1170 	rc = t4_setup_intr_handlers(sc);
1171 	if (rc != 0) {
1172 		device_printf(dev,
1173 		    "failed to setup interrupt handlers: %d\n", rc);
1174 		goto done;
1175 	}
1176 
1177 	rc = bus_generic_attach(dev);
1178 	if (rc != 0) {
1179 		device_printf(dev,
1180 		    "failed to attach all child ports: %d\n", rc);
1181 		goto done;
1182 	}
1183 
1184 	device_printf(dev,
1185 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1186 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1187 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1188 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1189 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1190 
1191 	t4_set_desc(sc);
1192 
1193 	notify_siblings(dev, 0);
1194 
1195 done:
1196 	if (rc != 0 && sc->cdev) {
1197 		/* cdev was created and so cxgbetool works; recover that way. */
1198 		device_printf(dev,
1199 		    "error during attach, adapter is now in recovery mode.\n");
1200 		rc = 0;
1201 	}
1202 
1203 	if (rc != 0)
1204 		t4_detach_common(dev);
1205 	else
1206 		t4_sysctls(sc);
1207 
1208 	return (rc);
1209 }
1210 
1211 static int
1212 t4_ready(device_t dev)
1213 {
1214 	struct adapter *sc;
1215 
1216 	sc = device_get_softc(dev);
1217 	if (sc->flags & FW_OK)
1218 		return (0);
1219 	return (ENXIO);
1220 }
1221 
1222 static int
1223 t4_read_port_device(device_t dev, int port, device_t *child)
1224 {
1225 	struct adapter *sc;
1226 	struct port_info *pi;
1227 
1228 	sc = device_get_softc(dev);
1229 	if (port < 0 || port >= MAX_NPORTS)
1230 		return (EINVAL);
1231 	pi = sc->port[port];
1232 	if (pi == NULL || pi->dev == NULL)
1233 		return (ENXIO);
1234 	*child = pi->dev;
1235 	return (0);
1236 }
1237 
1238 static int
1239 notify_siblings(device_t dev, int detaching)
1240 {
1241 	device_t sibling;
1242 	int error, i;
1243 
1244 	error = 0;
1245 	for (i = 0; i < PCI_FUNCMAX; i++) {
1246 		if (i == pci_get_function(dev))
1247 			continue;
1248 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1249 		    pci_get_slot(dev), i);
1250 		if (sibling == NULL || !device_is_attached(sibling))
1251 			continue;
1252 		if (detaching)
1253 			error = T4_DETACH_CHILD(sibling);
1254 		else
1255 			(void)T4_ATTACH_CHILD(sibling);
1256 		if (error)
1257 			break;
1258 	}
1259 	return (error);
1260 }
1261 
1262 /*
1263  * Idempotent
1264  */
1265 static int
1266 t4_detach(device_t dev)
1267 {
1268 	struct adapter *sc;
1269 	int rc;
1270 
1271 	sc = device_get_softc(dev);
1272 
1273 	rc = notify_siblings(dev, 1);
1274 	if (rc) {
1275 		device_printf(dev,
1276 		    "failed to detach sibling devices: %d\n", rc);
1277 		return (rc);
1278 	}
1279 
1280 	return (t4_detach_common(dev));
1281 }
1282 
1283 int
1284 t4_detach_common(device_t dev)
1285 {
1286 	struct adapter *sc;
1287 	struct port_info *pi;
1288 	int i, rc;
1289 
1290 	sc = device_get_softc(dev);
1291 
1292 	if (sc->flags & FULL_INIT_DONE) {
1293 		if (!(sc->flags & IS_VF))
1294 			t4_intr_disable(sc);
1295 	}
1296 
1297 	if (sc->cdev) {
1298 		destroy_dev(sc->cdev);
1299 		sc->cdev = NULL;
1300 	}
1301 
1302 	if (device_is_attached(dev)) {
1303 		rc = bus_generic_detach(dev);
1304 		if (rc) {
1305 			device_printf(dev,
1306 			    "failed to detach child devices: %d\n", rc);
1307 			return (rc);
1308 		}
1309 	}
1310 
1311 	for (i = 0; i < sc->intr_count; i++)
1312 		t4_free_irq(sc, &sc->irq[i]);
1313 
1314 	for (i = 0; i < MAX_NPORTS; i++) {
1315 		pi = sc->port[i];
1316 		if (pi) {
1317 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1318 			if (pi->dev)
1319 				device_delete_child(dev, pi->dev);
1320 
1321 			mtx_destroy(&pi->pi_lock);
1322 			free(pi->vi, M_CXGBE);
1323 			free(pi->tc, M_CXGBE);
1324 			free(pi, M_CXGBE);
1325 		}
1326 	}
1327 
1328 	if (sc->flags & FULL_INIT_DONE)
1329 		adapter_full_uninit(sc);
1330 
1331 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1332 		t4_fw_bye(sc, sc->mbox);
1333 
1334 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1335 		pci_release_msi(dev);
1336 
1337 	if (sc->regs_res)
1338 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1339 		    sc->regs_res);
1340 
1341 	if (sc->udbs_res)
1342 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1343 		    sc->udbs_res);
1344 
1345 	if (sc->msix_res)
1346 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1347 		    sc->msix_res);
1348 
1349 	if (sc->l2t)
1350 		t4_free_l2t(sc->l2t);
1351 
1352 #ifdef TCP_OFFLOAD
1353 	free(sc->sge.ofld_rxq, M_CXGBE);
1354 	free(sc->sge.ofld_txq, M_CXGBE);
1355 #endif
1356 #ifdef DEV_NETMAP
1357 	free(sc->sge.nm_rxq, M_CXGBE);
1358 	free(sc->sge.nm_txq, M_CXGBE);
1359 #endif
1360 	free(sc->irq, M_CXGBE);
1361 	free(sc->sge.rxq, M_CXGBE);
1362 	free(sc->sge.txq, M_CXGBE);
1363 	free(sc->sge.ctrlq, M_CXGBE);
1364 	free(sc->sge.iqmap, M_CXGBE);
1365 	free(sc->sge.eqmap, M_CXGBE);
1366 	free(sc->tids.ftid_tab, M_CXGBE);
1367 	t4_destroy_dma_tag(sc);
1368 	if (mtx_initialized(&sc->sc_lock)) {
1369 		sx_xlock(&t4_list_lock);
1370 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1371 		sx_xunlock(&t4_list_lock);
1372 		mtx_destroy(&sc->sc_lock);
1373 	}
1374 
1375 	callout_drain(&sc->sfl_callout);
1376 	if (mtx_initialized(&sc->tids.ftid_lock))
1377 		mtx_destroy(&sc->tids.ftid_lock);
1378 	if (mtx_initialized(&sc->sfl_lock))
1379 		mtx_destroy(&sc->sfl_lock);
1380 	if (mtx_initialized(&sc->ifp_lock))
1381 		mtx_destroy(&sc->ifp_lock);
1382 	if (mtx_initialized(&sc->reg_lock))
1383 		mtx_destroy(&sc->reg_lock);
1384 
1385 	for (i = 0; i < NUM_MEMWIN; i++) {
1386 		struct memwin *mw = &sc->memwin[i];
1387 
1388 		if (rw_initialized(&mw->mw_lock))
1389 			rw_destroy(&mw->mw_lock);
1390 	}
1391 
1392 	bzero(sc, sizeof(*sc));
1393 
1394 	return (0);
1395 }
1396 
1397 static int
1398 cxgbe_probe(device_t dev)
1399 {
1400 	char buf[128];
1401 	struct port_info *pi = device_get_softc(dev);
1402 
1403 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1404 	device_set_desc_copy(dev, buf);
1405 
1406 	return (BUS_PROBE_DEFAULT);
1407 }
1408 
1409 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1410     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1411     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1412 #define T4_CAP_ENABLE (T4_CAP)
1413 
1414 static int
1415 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1416 {
1417 	struct ifnet *ifp;
1418 	struct sbuf *sb;
1419 
1420 	vi->xact_addr_filt = -1;
1421 	callout_init(&vi->tick, 1);
1422 
1423 	/* Allocate an ifnet and set it up */
1424 	ifp = if_alloc(IFT_ETHER);
1425 	if (ifp == NULL) {
1426 		device_printf(dev, "Cannot allocate ifnet\n");
1427 		return (ENOMEM);
1428 	}
1429 	vi->ifp = ifp;
1430 	ifp->if_softc = vi;
1431 
1432 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1433 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1434 
1435 	ifp->if_init = cxgbe_init;
1436 	ifp->if_ioctl = cxgbe_ioctl;
1437 	ifp->if_transmit = cxgbe_transmit;
1438 	ifp->if_qflush = cxgbe_qflush;
1439 	ifp->if_get_counter = cxgbe_get_counter;
1440 
1441 	ifp->if_capabilities = T4_CAP;
1442 #ifdef TCP_OFFLOAD
1443 	if (vi->nofldrxq != 0)
1444 		ifp->if_capabilities |= IFCAP_TOE;
1445 #endif
1446 #ifdef DEV_NETMAP
1447 	if (vi->nnmrxq != 0)
1448 		ifp->if_capabilities |= IFCAP_NETMAP;
1449 #endif
1450 	ifp->if_capenable = T4_CAP_ENABLE;
1451 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1452 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1453 
1454 	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1455 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1456 	ifp->if_hw_tsomaxsegsize = 65536;
1457 
1458 	/* Initialize ifmedia for this VI */
1459 	ifmedia_init(&vi->media, IFM_IMASK, cxgbe_media_change,
1460 	    cxgbe_media_status);
1461 	build_medialist(vi->pi, &vi->media);
1462 
1463 	vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1464 	    EVENTHANDLER_PRI_ANY);
1465 
1466 	ether_ifattach(ifp, vi->hw_addr);
1467 #ifdef DEV_NETMAP
1468 	if (ifp->if_capabilities & IFCAP_NETMAP)
1469 		cxgbe_nm_attach(vi);
1470 #endif
1471 	sb = sbuf_new_auto();
1472 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1473 #ifdef TCP_OFFLOAD
1474 	if (ifp->if_capabilities & IFCAP_TOE)
1475 		sbuf_printf(sb, "; %d txq, %d rxq (TOE)",
1476 		    vi->nofldtxq, vi->nofldrxq);
1477 #endif
1478 #ifdef DEV_NETMAP
1479 	if (ifp->if_capabilities & IFCAP_NETMAP)
1480 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1481 		    vi->nnmtxq, vi->nnmrxq);
1482 #endif
1483 	sbuf_finish(sb);
1484 	device_printf(dev, "%s\n", sbuf_data(sb));
1485 	sbuf_delete(sb);
1486 
1487 	vi_sysctls(vi);
1488 
1489 	return (0);
1490 }
1491 
1492 static int
1493 cxgbe_attach(device_t dev)
1494 {
1495 	struct port_info *pi = device_get_softc(dev);
1496 	struct adapter *sc = pi->adapter;
1497 	struct vi_info *vi;
1498 	int i, rc;
1499 
1500 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1501 
1502 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1503 	if (rc)
1504 		return (rc);
1505 
1506 	for_each_vi(pi, i, vi) {
1507 		if (i == 0)
1508 			continue;
1509 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1510 		if (vi->dev == NULL) {
1511 			device_printf(dev, "failed to add VI %d\n", i);
1512 			continue;
1513 		}
1514 		device_set_softc(vi->dev, vi);
1515 	}
1516 
1517 	cxgbe_sysctls(pi);
1518 
1519 	bus_generic_attach(dev);
1520 
1521 	return (0);
1522 }
1523 
1524 static void
1525 cxgbe_vi_detach(struct vi_info *vi)
1526 {
1527 	struct ifnet *ifp = vi->ifp;
1528 
1529 	ether_ifdetach(ifp);
1530 
1531 	if (vi->vlan_c)
1532 		EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1533 
1534 	/* Let detach proceed even if these fail. */
1535 #ifdef DEV_NETMAP
1536 	if (ifp->if_capabilities & IFCAP_NETMAP)
1537 		cxgbe_nm_detach(vi);
1538 #endif
1539 	cxgbe_uninit_synchronized(vi);
1540 	callout_drain(&vi->tick);
1541 	vi_full_uninit(vi);
1542 
1543 	ifmedia_removeall(&vi->media);
1544 	if_free(vi->ifp);
1545 	vi->ifp = NULL;
1546 }
1547 
1548 static int
1549 cxgbe_detach(device_t dev)
1550 {
1551 	struct port_info *pi = device_get_softc(dev);
1552 	struct adapter *sc = pi->adapter;
1553 	int rc;
1554 
1555 	/* Detach the extra VIs first. */
1556 	rc = bus_generic_detach(dev);
1557 	if (rc)
1558 		return (rc);
1559 	device_delete_children(dev);
1560 
1561 	doom_vi(sc, &pi->vi[0]);
1562 
1563 	if (pi->flags & HAS_TRACEQ) {
1564 		sc->traceq = -1;	/* cloner should not create ifnet */
1565 		t4_tracer_port_detach(sc);
1566 	}
1567 
1568 	cxgbe_vi_detach(&pi->vi[0]);
1569 	callout_drain(&pi->tick);
1570 
1571 	end_synchronized_op(sc, 0);
1572 
1573 	return (0);
1574 }
1575 
1576 static void
1577 cxgbe_init(void *arg)
1578 {
1579 	struct vi_info *vi = arg;
1580 	struct adapter *sc = vi->pi->adapter;
1581 
1582 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1583 		return;
1584 	cxgbe_init_synchronized(vi);
1585 	end_synchronized_op(sc, 0);
1586 }
1587 
1588 static int
1589 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1590 {
1591 	int rc = 0, mtu, flags, can_sleep;
1592 	struct vi_info *vi = ifp->if_softc;
1593 	struct adapter *sc = vi->pi->adapter;
1594 	struct ifreq *ifr = (struct ifreq *)data;
1595 	uint32_t mask;
1596 
1597 	switch (cmd) {
1598 	case SIOCSIFMTU:
1599 		mtu = ifr->ifr_mtu;
1600 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1601 			return (EINVAL);
1602 
1603 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1604 		if (rc)
1605 			return (rc);
1606 		ifp->if_mtu = mtu;
1607 		if (vi->flags & VI_INIT_DONE) {
1608 			t4_update_fl_bufsize(ifp);
1609 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1610 				rc = update_mac_settings(ifp, XGMAC_MTU);
1611 		}
1612 		end_synchronized_op(sc, 0);
1613 		break;
1614 
1615 	case SIOCSIFFLAGS:
1616 		can_sleep = 0;
1617 redo_sifflags:
1618 		rc = begin_synchronized_op(sc, vi,
1619 		    can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1620 		if (rc)
1621 			return (rc);
1622 
1623 		if (ifp->if_flags & IFF_UP) {
1624 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1625 				flags = vi->if_flags;
1626 				if ((ifp->if_flags ^ flags) &
1627 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1628 					if (can_sleep == 1) {
1629 						end_synchronized_op(sc, 0);
1630 						can_sleep = 0;
1631 						goto redo_sifflags;
1632 					}
1633 					rc = update_mac_settings(ifp,
1634 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1635 				}
1636 			} else {
1637 				if (can_sleep == 0) {
1638 					end_synchronized_op(sc, LOCK_HELD);
1639 					can_sleep = 1;
1640 					goto redo_sifflags;
1641 				}
1642 				rc = cxgbe_init_synchronized(vi);
1643 			}
1644 			vi->if_flags = ifp->if_flags;
1645 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1646 			if (can_sleep == 0) {
1647 				end_synchronized_op(sc, LOCK_HELD);
1648 				can_sleep = 1;
1649 				goto redo_sifflags;
1650 			}
1651 			rc = cxgbe_uninit_synchronized(vi);
1652 		}
1653 		end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1654 		break;
1655 
1656 	case SIOCADDMULTI:
1657 	case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1658 		rc = begin_synchronized_op(sc, vi, HOLD_LOCK, "t4multi");
1659 		if (rc)
1660 			return (rc);
1661 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1662 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1663 		end_synchronized_op(sc, LOCK_HELD);
1664 		break;
1665 
1666 	case SIOCSIFCAP:
1667 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1668 		if (rc)
1669 			return (rc);
1670 
1671 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1672 		if (mask & IFCAP_TXCSUM) {
1673 			ifp->if_capenable ^= IFCAP_TXCSUM;
1674 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1675 
1676 			if (IFCAP_TSO4 & ifp->if_capenable &&
1677 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1678 				ifp->if_capenable &= ~IFCAP_TSO4;
1679 				if_printf(ifp,
1680 				    "tso4 disabled due to -txcsum.\n");
1681 			}
1682 		}
1683 		if (mask & IFCAP_TXCSUM_IPV6) {
1684 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1685 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1686 
1687 			if (IFCAP_TSO6 & ifp->if_capenable &&
1688 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1689 				ifp->if_capenable &= ~IFCAP_TSO6;
1690 				if_printf(ifp,
1691 				    "tso6 disabled due to -txcsum6.\n");
1692 			}
1693 		}
1694 		if (mask & IFCAP_RXCSUM)
1695 			ifp->if_capenable ^= IFCAP_RXCSUM;
1696 		if (mask & IFCAP_RXCSUM_IPV6)
1697 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1698 
1699 		/*
1700 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1701 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1702 		 * sending a TSO request our way, so it's sufficient to toggle
1703 		 * IFCAP_TSOx only.
1704 		 */
1705 		if (mask & IFCAP_TSO4) {
1706 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1707 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1708 				if_printf(ifp, "enable txcsum first.\n");
1709 				rc = EAGAIN;
1710 				goto fail;
1711 			}
1712 			ifp->if_capenable ^= IFCAP_TSO4;
1713 		}
1714 		if (mask & IFCAP_TSO6) {
1715 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1716 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1717 				if_printf(ifp, "enable txcsum6 first.\n");
1718 				rc = EAGAIN;
1719 				goto fail;
1720 			}
1721 			ifp->if_capenable ^= IFCAP_TSO6;
1722 		}
1723 		if (mask & IFCAP_LRO) {
1724 #if defined(INET) || defined(INET6)
1725 			int i;
1726 			struct sge_rxq *rxq;
1727 
1728 			ifp->if_capenable ^= IFCAP_LRO;
1729 			for_each_rxq(vi, i, rxq) {
1730 				if (ifp->if_capenable & IFCAP_LRO)
1731 					rxq->iq.flags |= IQ_LRO_ENABLED;
1732 				else
1733 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1734 			}
1735 #endif
1736 		}
1737 #ifdef TCP_OFFLOAD
1738 		if (mask & IFCAP_TOE) {
1739 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1740 
1741 			rc = toe_capability(vi, enable);
1742 			if (rc != 0)
1743 				goto fail;
1744 
1745 			ifp->if_capenable ^= mask;
1746 		}
1747 #endif
1748 		if (mask & IFCAP_VLAN_HWTAGGING) {
1749 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1750 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1751 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1752 		}
1753 		if (mask & IFCAP_VLAN_MTU) {
1754 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1755 
1756 			/* Need to find out how to disable auto-mtu-inflation */
1757 		}
1758 		if (mask & IFCAP_VLAN_HWTSO)
1759 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1760 		if (mask & IFCAP_VLAN_HWCSUM)
1761 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1762 
1763 #ifdef VLAN_CAPABILITIES
1764 		VLAN_CAPABILITIES(ifp);
1765 #endif
1766 fail:
1767 		end_synchronized_op(sc, 0);
1768 		break;
1769 
1770 	case SIOCSIFMEDIA:
1771 	case SIOCGIFMEDIA:
1772 	case SIOCGIFXMEDIA:
1773 		ifmedia_ioctl(ifp, ifr, &vi->media, cmd);
1774 		break;
1775 
1776 	case SIOCGI2C: {
1777 		struct ifi2creq i2c;
1778 
1779 		rc = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
1780 		if (rc != 0)
1781 			break;
1782 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1783 			rc = EPERM;
1784 			break;
1785 		}
1786 		if (i2c.len > sizeof(i2c.data)) {
1787 			rc = EINVAL;
1788 			break;
1789 		}
1790 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1791 		if (rc)
1792 			return (rc);
1793 		rc = -t4_i2c_rd(sc, sc->mbox, vi->pi->port_id, i2c.dev_addr,
1794 		    i2c.offset, i2c.len, &i2c.data[0]);
1795 		end_synchronized_op(sc, 0);
1796 		if (rc == 0)
1797 			rc = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
1798 		break;
1799 	}
1800 
1801 	default:
1802 		rc = ether_ioctl(ifp, cmd, data);
1803 	}
1804 
1805 	return (rc);
1806 }
1807 
1808 static int
1809 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1810 {
1811 	struct vi_info *vi = ifp->if_softc;
1812 	struct port_info *pi = vi->pi;
1813 	struct adapter *sc = pi->adapter;
1814 	struct sge_txq *txq;
1815 	void *items[1];
1816 	int rc;
1817 
1818 	M_ASSERTPKTHDR(m);
1819 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1820 
1821 	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1822 		m_freem(m);
1823 		return (ENETDOWN);
1824 	}
1825 
1826 	rc = parse_pkt(sc, &m);
1827 	if (__predict_false(rc != 0)) {
1828 		MPASS(m == NULL);			/* was freed already */
1829 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1830 		return (rc);
1831 	}
1832 
1833 	/* Select a txq. */
1834 	txq = &sc->sge.txq[vi->first_txq];
1835 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1836 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1837 		    vi->rsrv_noflowq);
1838 
1839 	items[0] = m;
1840 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1841 	if (__predict_false(rc != 0))
1842 		m_freem(m);
1843 
1844 	return (rc);
1845 }
1846 
1847 static void
1848 cxgbe_qflush(struct ifnet *ifp)
1849 {
1850 	struct vi_info *vi = ifp->if_softc;
1851 	struct sge_txq *txq;
1852 	int i;
1853 
1854 	/* queues do not exist if !VI_INIT_DONE. */
1855 	if (vi->flags & VI_INIT_DONE) {
1856 		for_each_txq(vi, i, txq) {
1857 			TXQ_LOCK(txq);
1858 			txq->eq.flags &= ~EQ_ENABLED;
1859 			TXQ_UNLOCK(txq);
1860 			while (!mp_ring_is_idle(txq->r)) {
1861 				mp_ring_check_drainage(txq->r, 0);
1862 				pause("qflush", 1);
1863 			}
1864 		}
1865 	}
1866 	if_qflush(ifp);
1867 }
1868 
1869 static uint64_t
1870 vi_get_counter(struct ifnet *ifp, ift_counter c)
1871 {
1872 	struct vi_info *vi = ifp->if_softc;
1873 	struct fw_vi_stats_vf *s = &vi->stats;
1874 
1875 	vi_refresh_stats(vi->pi->adapter, vi);
1876 
1877 	switch (c) {
1878 	case IFCOUNTER_IPACKETS:
1879 		return (s->rx_bcast_frames + s->rx_mcast_frames +
1880 		    s->rx_ucast_frames);
1881 	case IFCOUNTER_IERRORS:
1882 		return (s->rx_err_frames);
1883 	case IFCOUNTER_OPACKETS:
1884 		return (s->tx_bcast_frames + s->tx_mcast_frames +
1885 		    s->tx_ucast_frames + s->tx_offload_frames);
1886 	case IFCOUNTER_OERRORS:
1887 		return (s->tx_drop_frames);
1888 	case IFCOUNTER_IBYTES:
1889 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
1890 		    s->rx_ucast_bytes);
1891 	case IFCOUNTER_OBYTES:
1892 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
1893 		    s->tx_ucast_bytes + s->tx_offload_bytes);
1894 	case IFCOUNTER_IMCASTS:
1895 		return (s->rx_mcast_frames);
1896 	case IFCOUNTER_OMCASTS:
1897 		return (s->tx_mcast_frames);
1898 	case IFCOUNTER_OQDROPS: {
1899 		uint64_t drops;
1900 
1901 		drops = 0;
1902 		if (vi->flags & VI_INIT_DONE) {
1903 			int i;
1904 			struct sge_txq *txq;
1905 
1906 			for_each_txq(vi, i, txq)
1907 				drops += counter_u64_fetch(txq->r->drops);
1908 		}
1909 
1910 		return (drops);
1911 
1912 	}
1913 
1914 	default:
1915 		return (if_get_counter_default(ifp, c));
1916 	}
1917 }
1918 
1919 uint64_t
1920 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
1921 {
1922 	struct vi_info *vi = ifp->if_softc;
1923 	struct port_info *pi = vi->pi;
1924 	struct adapter *sc = pi->adapter;
1925 	struct port_stats *s = &pi->stats;
1926 
1927 	if (pi->nvi > 1 || sc->flags & IS_VF)
1928 		return (vi_get_counter(ifp, c));
1929 
1930 	cxgbe_refresh_stats(sc, pi);
1931 
1932 	switch (c) {
1933 	case IFCOUNTER_IPACKETS:
1934 		return (s->rx_frames);
1935 
1936 	case IFCOUNTER_IERRORS:
1937 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
1938 		    s->rx_fcs_err + s->rx_len_err);
1939 
1940 	case IFCOUNTER_OPACKETS:
1941 		return (s->tx_frames);
1942 
1943 	case IFCOUNTER_OERRORS:
1944 		return (s->tx_error_frames);
1945 
1946 	case IFCOUNTER_IBYTES:
1947 		return (s->rx_octets);
1948 
1949 	case IFCOUNTER_OBYTES:
1950 		return (s->tx_octets);
1951 
1952 	case IFCOUNTER_IMCASTS:
1953 		return (s->rx_mcast_frames);
1954 
1955 	case IFCOUNTER_OMCASTS:
1956 		return (s->tx_mcast_frames);
1957 
1958 	case IFCOUNTER_IQDROPS:
1959 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
1960 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
1961 		    s->rx_trunc3 + pi->tnl_cong_drops);
1962 
1963 	case IFCOUNTER_OQDROPS: {
1964 		uint64_t drops;
1965 
1966 		drops = s->tx_drop;
1967 		if (vi->flags & VI_INIT_DONE) {
1968 			int i;
1969 			struct sge_txq *txq;
1970 
1971 			for_each_txq(vi, i, txq)
1972 				drops += counter_u64_fetch(txq->r->drops);
1973 		}
1974 
1975 		return (drops);
1976 
1977 	}
1978 
1979 	default:
1980 		return (if_get_counter_default(ifp, c));
1981 	}
1982 }
1983 
1984 static int
1985 cxgbe_media_change(struct ifnet *ifp)
1986 {
1987 	struct vi_info *vi = ifp->if_softc;
1988 
1989 	device_printf(vi->dev, "%s unimplemented.\n", __func__);
1990 
1991 	return (EOPNOTSUPP);
1992 }
1993 
1994 static void
1995 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1996 {
1997 	struct vi_info *vi = ifp->if_softc;
1998 	struct port_info *pi = vi->pi;
1999 	struct ifmedia_entry *cur;
2000 	int speed = pi->link_cfg.speed;
2001 
2002 	cur = vi->media.ifm_cur;
2003 
2004 	ifmr->ifm_status = IFM_AVALID;
2005 	if (!pi->link_cfg.link_ok)
2006 		return;
2007 
2008 	ifmr->ifm_status |= IFM_ACTIVE;
2009 
2010 	/* active and current will differ iff current media is autoselect. */
2011 	if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
2012 		return;
2013 
2014 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2015 	if (speed == 10000)
2016 		ifmr->ifm_active |= IFM_10G_T;
2017 	else if (speed == 1000)
2018 		ifmr->ifm_active |= IFM_1000_T;
2019 	else if (speed == 100)
2020 		ifmr->ifm_active |= IFM_100_TX;
2021 	else if (speed == 10)
2022 		ifmr->ifm_active |= IFM_10_T;
2023 	else
2024 		KASSERT(0, ("%s: link up but speed unknown (%u)", __func__,
2025 			    speed));
2026 }
2027 
2028 static int
2029 vcxgbe_probe(device_t dev)
2030 {
2031 	char buf[128];
2032 	struct vi_info *vi = device_get_softc(dev);
2033 
2034 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2035 	    vi - vi->pi->vi);
2036 	device_set_desc_copy(dev, buf);
2037 
2038 	return (BUS_PROBE_DEFAULT);
2039 }
2040 
2041 static int
2042 vcxgbe_attach(device_t dev)
2043 {
2044 	struct vi_info *vi;
2045 	struct port_info *pi;
2046 	struct adapter *sc;
2047 	int func, index, rc;
2048 	u32 param, val;
2049 
2050 	vi = device_get_softc(dev);
2051 	pi = vi->pi;
2052 	sc = pi->adapter;
2053 
2054 	index = vi - pi->vi;
2055 	KASSERT(index < nitems(vi_mac_funcs),
2056 	    ("%s: VI %s doesn't have a MAC func", __func__,
2057 	    device_get_nameunit(dev)));
2058 	func = vi_mac_funcs[index];
2059 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2060 	    vi->hw_addr, &vi->rss_size, func, 0);
2061 	if (rc < 0) {
2062 		device_printf(dev, "Failed to allocate virtual interface "
2063 		    "for port %d: %d\n", pi->port_id, -rc);
2064 		return (-rc);
2065 	}
2066 	vi->viid = rc;
2067 	if (chip_id(sc) <= CHELSIO_T5)
2068 		vi->smt_idx = (rc & 0x7f) << 1;
2069 	else
2070 		vi->smt_idx = (rc & 0x7f);
2071 
2072 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2073 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2074 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2075 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2076 	if (rc)
2077 		vi->rss_base = 0xffff;
2078 	else {
2079 		/* MPASS((val >> 16) == rss_size); */
2080 		vi->rss_base = val & 0xffff;
2081 	}
2082 
2083 	rc = cxgbe_vi_attach(dev, vi);
2084 	if (rc) {
2085 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2086 		return (rc);
2087 	}
2088 	return (0);
2089 }
2090 
2091 static int
2092 vcxgbe_detach(device_t dev)
2093 {
2094 	struct vi_info *vi;
2095 	struct adapter *sc;
2096 
2097 	vi = device_get_softc(dev);
2098 	sc = vi->pi->adapter;
2099 
2100 	doom_vi(sc, vi);
2101 
2102 	cxgbe_vi_detach(vi);
2103 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2104 
2105 	end_synchronized_op(sc, 0);
2106 
2107 	return (0);
2108 }
2109 
2110 void
2111 t4_fatal_err(struct adapter *sc)
2112 {
2113 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2114 	t4_intr_disable(sc);
2115 	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2116 	    device_get_nameunit(sc->dev));
2117 }
2118 
2119 void
2120 t4_add_adapter(struct adapter *sc)
2121 {
2122 	sx_xlock(&t4_list_lock);
2123 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2124 	sx_xunlock(&t4_list_lock);
2125 }
2126 
2127 int
2128 t4_map_bars_0_and_4(struct adapter *sc)
2129 {
2130 	sc->regs_rid = PCIR_BAR(0);
2131 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2132 	    &sc->regs_rid, RF_ACTIVE);
2133 	if (sc->regs_res == NULL) {
2134 		device_printf(sc->dev, "cannot map registers.\n");
2135 		return (ENXIO);
2136 	}
2137 	sc->bt = rman_get_bustag(sc->regs_res);
2138 	sc->bh = rman_get_bushandle(sc->regs_res);
2139 	sc->mmio_len = rman_get_size(sc->regs_res);
2140 	setbit(&sc->doorbells, DOORBELL_KDB);
2141 
2142 	sc->msix_rid = PCIR_BAR(4);
2143 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2144 	    &sc->msix_rid, RF_ACTIVE);
2145 	if (sc->msix_res == NULL) {
2146 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2147 		return (ENXIO);
2148 	}
2149 
2150 	return (0);
2151 }
2152 
2153 int
2154 t4_map_bar_2(struct adapter *sc)
2155 {
2156 
2157 	/*
2158 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2159 	 * to map it if RDMA is disabled.
2160 	 */
2161 	if (is_t4(sc) && sc->rdmacaps == 0)
2162 		return (0);
2163 
2164 	sc->udbs_rid = PCIR_BAR(2);
2165 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2166 	    &sc->udbs_rid, RF_ACTIVE);
2167 	if (sc->udbs_res == NULL) {
2168 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2169 		return (ENXIO);
2170 	}
2171 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2172 
2173 	if (chip_id(sc) >= CHELSIO_T5) {
2174 		setbit(&sc->doorbells, DOORBELL_UDB);
2175 #if defined(__i386__) || defined(__amd64__)
2176 		if (t5_write_combine) {
2177 			int rc, mode;
2178 
2179 			/*
2180 			 * Enable write combining on BAR2.  This is the
2181 			 * userspace doorbell BAR and is split into 128B
2182 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2183 			 * with an egress queue.  The first 64B has the doorbell
2184 			 * and the second 64B can be used to submit a tx work
2185 			 * request with an implicit doorbell.
2186 			 */
2187 
2188 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2189 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2190 			if (rc == 0) {
2191 				clrbit(&sc->doorbells, DOORBELL_UDB);
2192 				setbit(&sc->doorbells, DOORBELL_WCWR);
2193 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2194 			} else {
2195 				device_printf(sc->dev,
2196 				    "couldn't enable write combining: %d\n",
2197 				    rc);
2198 			}
2199 
2200 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2201 			t4_write_reg(sc, A_SGE_STAT_CFG,
2202 			    V_STATSOURCE_T5(7) | mode);
2203 		}
2204 #endif
2205 	}
2206 
2207 	return (0);
2208 }
2209 
2210 struct memwin_init {
2211 	uint32_t base;
2212 	uint32_t aperture;
2213 };
2214 
2215 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2216 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2217 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2218 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2219 };
2220 
2221 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2222 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2223 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2224 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2225 };
2226 
2227 static void
2228 setup_memwin(struct adapter *sc)
2229 {
2230 	const struct memwin_init *mw_init;
2231 	struct memwin *mw;
2232 	int i;
2233 	uint32_t bar0;
2234 
2235 	if (is_t4(sc)) {
2236 		/*
2237 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2238 		 * mechanism.  Works from within PCI passthrough environments
2239 		 * too, where rman_get_start() can return a different value.  We
2240 		 * need to program the T4 memory window decoders with the actual
2241 		 * addresses that will be coming across the PCIe link.
2242 		 */
2243 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2244 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2245 
2246 		mw_init = &t4_memwin[0];
2247 	} else {
2248 		/* T5+ use the relative offset inside the PCIe BAR */
2249 		bar0 = 0;
2250 
2251 		mw_init = &t5_memwin[0];
2252 	}
2253 
2254 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2255 		rw_init(&mw->mw_lock, "memory window access");
2256 		mw->mw_base = mw_init->base;
2257 		mw->mw_aperture = mw_init->aperture;
2258 		mw->mw_curpos = 0;
2259 		t4_write_reg(sc,
2260 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2261 		    (mw->mw_base + bar0) | V_BIR(0) |
2262 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2263 		rw_wlock(&mw->mw_lock);
2264 		position_memwin(sc, i, 0);
2265 		rw_wunlock(&mw->mw_lock);
2266 	}
2267 
2268 	/* flush */
2269 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2270 }
2271 
2272 /*
2273  * Positions the memory window at the given address in the card's address space.
2274  * There are some alignment requirements and the actual position may be at an
2275  * address prior to the requested address.  mw->mw_curpos always has the actual
2276  * position of the window.
2277  */
2278 static void
2279 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2280 {
2281 	struct memwin *mw;
2282 	uint32_t pf;
2283 	uint32_t reg;
2284 
2285 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2286 	mw = &sc->memwin[idx];
2287 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2288 
2289 	if (is_t4(sc)) {
2290 		pf = 0;
2291 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2292 	} else {
2293 		pf = V_PFNUM(sc->pf);
2294 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2295 	}
2296 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2297 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2298 	t4_read_reg(sc, reg);	/* flush */
2299 }
2300 
2301 static int
2302 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2303     int len, int rw)
2304 {
2305 	struct memwin *mw;
2306 	uint32_t mw_end, v;
2307 
2308 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2309 
2310 	/* Memory can only be accessed in naturally aligned 4 byte units */
2311 	if (addr & 3 || len & 3 || len <= 0)
2312 		return (EINVAL);
2313 
2314 	mw = &sc->memwin[idx];
2315 	while (len > 0) {
2316 		rw_rlock(&mw->mw_lock);
2317 		mw_end = mw->mw_curpos + mw->mw_aperture;
2318 		if (addr >= mw_end || addr < mw->mw_curpos) {
2319 			/* Will need to reposition the window */
2320 			if (!rw_try_upgrade(&mw->mw_lock)) {
2321 				rw_runlock(&mw->mw_lock);
2322 				rw_wlock(&mw->mw_lock);
2323 			}
2324 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2325 			position_memwin(sc, idx, addr);
2326 			rw_downgrade(&mw->mw_lock);
2327 			mw_end = mw->mw_curpos + mw->mw_aperture;
2328 		}
2329 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2330 		while (addr < mw_end && len > 0) {
2331 			if (rw == 0) {
2332 				v = t4_read_reg(sc, mw->mw_base + addr -
2333 				    mw->mw_curpos);
2334 				*val++ = le32toh(v);
2335 			} else {
2336 				v = *val++;
2337 				t4_write_reg(sc, mw->mw_base + addr -
2338 				    mw->mw_curpos, htole32(v));
2339 			}
2340 			addr += 4;
2341 			len -= 4;
2342 		}
2343 		rw_runlock(&mw->mw_lock);
2344 	}
2345 
2346 	return (0);
2347 }
2348 
2349 static inline int
2350 read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2351     int len)
2352 {
2353 
2354 	return (rw_via_memwin(sc, idx, addr, val, len, 0));
2355 }
2356 
2357 static inline int
2358 write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
2359     const uint32_t *val, int len)
2360 {
2361 
2362 	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
2363 }
2364 
2365 static int
2366 t4_range_cmp(const void *a, const void *b)
2367 {
2368 	return ((const struct t4_range *)a)->start -
2369 	       ((const struct t4_range *)b)->start;
2370 }
2371 
2372 /*
2373  * Verify that the memory range specified by the addr/len pair is valid within
2374  * the card's address space.
2375  */
2376 static int
2377 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2378 {
2379 	struct t4_range mem_ranges[4], *r, *next;
2380 	uint32_t em, addr_len;
2381 	int i, n, remaining;
2382 
2383 	/* Memory can only be accessed in naturally aligned 4 byte units */
2384 	if (addr & 3 || len & 3 || len <= 0)
2385 		return (EINVAL);
2386 
2387 	/* Enabled memories */
2388 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2389 
2390 	r = &mem_ranges[0];
2391 	n = 0;
2392 	bzero(r, sizeof(mem_ranges));
2393 	if (em & F_EDRAM0_ENABLE) {
2394 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2395 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2396 		if (r->size > 0) {
2397 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2398 			if (addr >= r->start &&
2399 			    addr + len <= r->start + r->size)
2400 				return (0);
2401 			r++;
2402 			n++;
2403 		}
2404 	}
2405 	if (em & F_EDRAM1_ENABLE) {
2406 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2407 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2408 		if (r->size > 0) {
2409 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2410 			if (addr >= r->start &&
2411 			    addr + len <= r->start + r->size)
2412 				return (0);
2413 			r++;
2414 			n++;
2415 		}
2416 	}
2417 	if (em & F_EXT_MEM_ENABLE) {
2418 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2419 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2420 		if (r->size > 0) {
2421 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2422 			if (addr >= r->start &&
2423 			    addr + len <= r->start + r->size)
2424 				return (0);
2425 			r++;
2426 			n++;
2427 		}
2428 	}
2429 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2430 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2431 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2432 		if (r->size > 0) {
2433 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2434 			if (addr >= r->start &&
2435 			    addr + len <= r->start + r->size)
2436 				return (0);
2437 			r++;
2438 			n++;
2439 		}
2440 	}
2441 	MPASS(n <= nitems(mem_ranges));
2442 
2443 	if (n > 1) {
2444 		/* Sort and merge the ranges. */
2445 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2446 
2447 		/* Start from index 0 and examine the next n - 1 entries. */
2448 		r = &mem_ranges[0];
2449 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2450 
2451 			MPASS(r->size > 0);	/* r is a valid entry. */
2452 			next = r + 1;
2453 			MPASS(next->size > 0);	/* and so is the next one. */
2454 
2455 			while (r->start + r->size >= next->start) {
2456 				/* Merge the next one into the current entry. */
2457 				r->size = max(r->start + r->size,
2458 				    next->start + next->size) - r->start;
2459 				n--;	/* One fewer entry in total. */
2460 				if (--remaining == 0)
2461 					goto done;	/* short circuit */
2462 				next++;
2463 			}
2464 			if (next != r + 1) {
2465 				/*
2466 				 * Some entries were merged into r and next
2467 				 * points to the first valid entry that couldn't
2468 				 * be merged.
2469 				 */
2470 				MPASS(next->size > 0);	/* must be valid */
2471 				memcpy(r + 1, next, remaining * sizeof(*r));
2472 #ifdef INVARIANTS
2473 				/*
2474 				 * This so that the foo->size assertion in the
2475 				 * next iteration of the loop do the right
2476 				 * thing for entries that were pulled up and are
2477 				 * no longer valid.
2478 				 */
2479 				MPASS(n < nitems(mem_ranges));
2480 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2481 				    sizeof(struct t4_range));
2482 #endif
2483 			}
2484 		}
2485 done:
2486 		/* Done merging the ranges. */
2487 		MPASS(n > 0);
2488 		r = &mem_ranges[0];
2489 		for (i = 0; i < n; i++, r++) {
2490 			if (addr >= r->start &&
2491 			    addr + len <= r->start + r->size)
2492 				return (0);
2493 		}
2494 	}
2495 
2496 	return (EFAULT);
2497 }
2498 
2499 static int
2500 fwmtype_to_hwmtype(int mtype)
2501 {
2502 
2503 	switch (mtype) {
2504 	case FW_MEMTYPE_EDC0:
2505 		return (MEM_EDC0);
2506 	case FW_MEMTYPE_EDC1:
2507 		return (MEM_EDC1);
2508 	case FW_MEMTYPE_EXTMEM:
2509 		return (MEM_MC0);
2510 	case FW_MEMTYPE_EXTMEM1:
2511 		return (MEM_MC1);
2512 	default:
2513 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2514 	}
2515 }
2516 
2517 /*
2518  * Verify that the memory range specified by the memtype/offset/len pair is
2519  * valid and lies entirely within the memtype specified.  The global address of
2520  * the start of the range is returned in addr.
2521  */
2522 static int
2523 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2524     uint32_t *addr)
2525 {
2526 	uint32_t em, addr_len, maddr;
2527 
2528 	/* Memory can only be accessed in naturally aligned 4 byte units */
2529 	if (off & 3 || len & 3 || len == 0)
2530 		return (EINVAL);
2531 
2532 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2533 	switch (fwmtype_to_hwmtype(mtype)) {
2534 	case MEM_EDC0:
2535 		if (!(em & F_EDRAM0_ENABLE))
2536 			return (EINVAL);
2537 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2538 		maddr = G_EDRAM0_BASE(addr_len) << 20;
2539 		break;
2540 	case MEM_EDC1:
2541 		if (!(em & F_EDRAM1_ENABLE))
2542 			return (EINVAL);
2543 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2544 		maddr = G_EDRAM1_BASE(addr_len) << 20;
2545 		break;
2546 	case MEM_MC:
2547 		if (!(em & F_EXT_MEM_ENABLE))
2548 			return (EINVAL);
2549 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2550 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2551 		break;
2552 	case MEM_MC1:
2553 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2554 			return (EINVAL);
2555 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2556 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2557 		break;
2558 	default:
2559 		return (EINVAL);
2560 	}
2561 
2562 	*addr = maddr + off;	/* global address */
2563 	return (validate_mem_range(sc, *addr, len));
2564 }
2565 
2566 static int
2567 fixup_devlog_params(struct adapter *sc)
2568 {
2569 	struct devlog_params *dparams = &sc->params.devlog;
2570 	int rc;
2571 
2572 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2573 	    dparams->size, &dparams->addr);
2574 
2575 	return (rc);
2576 }
2577 
2578 static int
2579 cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g, int num_vis,
2580     struct intrs_and_queues *iaq)
2581 {
2582 	int rc, itype, navail, nrxq10g, nrxq1g, n;
2583 	int nofldrxq10g = 0, nofldrxq1g = 0;
2584 
2585 	bzero(iaq, sizeof(*iaq));
2586 
2587 	iaq->ntxq10g = t4_ntxq10g;
2588 	iaq->ntxq1g = t4_ntxq1g;
2589 	iaq->ntxq_vi = t4_ntxq_vi;
2590 	iaq->nrxq10g = nrxq10g = t4_nrxq10g;
2591 	iaq->nrxq1g = nrxq1g = t4_nrxq1g;
2592 	iaq->nrxq_vi = t4_nrxq_vi;
2593 	iaq->rsrv_noflowq = t4_rsrv_noflowq;
2594 #ifdef TCP_OFFLOAD
2595 	if (is_offload(sc)) {
2596 		iaq->nofldtxq10g = t4_nofldtxq10g;
2597 		iaq->nofldtxq1g = t4_nofldtxq1g;
2598 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
2599 		iaq->nofldrxq10g = nofldrxq10g = t4_nofldrxq10g;
2600 		iaq->nofldrxq1g = nofldrxq1g = t4_nofldrxq1g;
2601 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
2602 	}
2603 #endif
2604 #ifdef DEV_NETMAP
2605 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
2606 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
2607 #endif
2608 
2609 	for (itype = INTR_MSIX; itype; itype >>= 1) {
2610 
2611 		if ((itype & t4_intr_types) == 0)
2612 			continue;	/* not allowed */
2613 
2614 		if (itype == INTR_MSIX)
2615 			navail = pci_msix_count(sc->dev);
2616 		else if (itype == INTR_MSI)
2617 			navail = pci_msi_count(sc->dev);
2618 		else
2619 			navail = 1;
2620 restart:
2621 		if (navail == 0)
2622 			continue;
2623 
2624 		iaq->intr_type = itype;
2625 		iaq->intr_flags_10g = 0;
2626 		iaq->intr_flags_1g = 0;
2627 
2628 		/*
2629 		 * Best option: an interrupt vector for errors, one for the
2630 		 * firmware event queue, and one for every rxq (NIC and TOE) of
2631 		 * every VI.  The VIs that support netmap use the same
2632 		 * interrupts for the NIC rx queues and the netmap rx queues
2633 		 * because only one set of queues is active at a time.
2634 		 */
2635 		iaq->nirq = T4_EXTRA_INTR;
2636 		iaq->nirq += n10g * (nrxq10g + nofldrxq10g);
2637 		iaq->nirq += n1g * (nrxq1g + nofldrxq1g);
2638 		iaq->nirq += (n10g + n1g) * (num_vis - 1) *
2639 		    max(iaq->nrxq_vi, iaq->nnmrxq_vi);	/* See comment above. */
2640 		iaq->nirq += (n10g + n1g) * (num_vis - 1) * iaq->nofldrxq_vi;
2641 		if (iaq->nirq <= navail &&
2642 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2643 			iaq->intr_flags_10g = INTR_ALL;
2644 			iaq->intr_flags_1g = INTR_ALL;
2645 			goto allocate;
2646 		}
2647 
2648 		/* Disable the VIs (and netmap) if there aren't enough intrs */
2649 		if (num_vis > 1) {
2650 			device_printf(sc->dev, "virtual interfaces disabled "
2651 			    "because num_vis=%u with current settings "
2652 			    "(nrxq10g=%u, nrxq1g=%u, nofldrxq10g=%u, "
2653 			    "nofldrxq1g=%u, nrxq_vi=%u nofldrxq_vi=%u, "
2654 			    "nnmrxq_vi=%u) would need %u interrupts but "
2655 			    "only %u are available.\n", num_vis, nrxq10g,
2656 			    nrxq1g, nofldrxq10g, nofldrxq1g, iaq->nrxq_vi,
2657 			    iaq->nofldrxq_vi, iaq->nnmrxq_vi, iaq->nirq,
2658 			    navail);
2659 			num_vis = 1;
2660 			iaq->ntxq_vi = iaq->nrxq_vi = 0;
2661 			iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
2662 			iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
2663 			goto restart;
2664 		}
2665 
2666 		/*
2667 		 * Second best option: a vector for errors, one for the firmware
2668 		 * event queue, and vectors for either all the NIC rx queues or
2669 		 * all the TOE rx queues.  The queues that don't get vectors
2670 		 * will forward their interrupts to those that do.
2671 		 */
2672 		iaq->nirq = T4_EXTRA_INTR;
2673 		if (nrxq10g >= nofldrxq10g) {
2674 			iaq->intr_flags_10g = INTR_RXQ;
2675 			iaq->nirq += n10g * nrxq10g;
2676 		} else {
2677 			iaq->intr_flags_10g = INTR_OFLD_RXQ;
2678 			iaq->nirq += n10g * nofldrxq10g;
2679 		}
2680 		if (nrxq1g >= nofldrxq1g) {
2681 			iaq->intr_flags_1g = INTR_RXQ;
2682 			iaq->nirq += n1g * nrxq1g;
2683 		} else {
2684 			iaq->intr_flags_1g = INTR_OFLD_RXQ;
2685 			iaq->nirq += n1g * nofldrxq1g;
2686 		}
2687 		if (iaq->nirq <= navail &&
2688 		    (itype != INTR_MSI || powerof2(iaq->nirq)))
2689 			goto allocate;
2690 
2691 		/*
2692 		 * Next best option: an interrupt vector for errors, one for the
2693 		 * firmware event queue, and at least one per main-VI.  At this
2694 		 * point we know we'll have to downsize nrxq and/or nofldrxq to
2695 		 * fit what's available to us.
2696 		 */
2697 		iaq->nirq = T4_EXTRA_INTR;
2698 		iaq->nirq += n10g + n1g;
2699 		if (iaq->nirq <= navail) {
2700 			int leftover = navail - iaq->nirq;
2701 
2702 			if (n10g > 0) {
2703 				int target = max(nrxq10g, nofldrxq10g);
2704 
2705 				iaq->intr_flags_10g = nrxq10g >= nofldrxq10g ?
2706 				    INTR_RXQ : INTR_OFLD_RXQ;
2707 
2708 				n = 1;
2709 				while (n < target && leftover >= n10g) {
2710 					leftover -= n10g;
2711 					iaq->nirq += n10g;
2712 					n++;
2713 				}
2714 				iaq->nrxq10g = min(n, nrxq10g);
2715 #ifdef TCP_OFFLOAD
2716 				iaq->nofldrxq10g = min(n, nofldrxq10g);
2717 #endif
2718 			}
2719 
2720 			if (n1g > 0) {
2721 				int target = max(nrxq1g, nofldrxq1g);
2722 
2723 				iaq->intr_flags_1g = nrxq1g >= nofldrxq1g ?
2724 				    INTR_RXQ : INTR_OFLD_RXQ;
2725 
2726 				n = 1;
2727 				while (n < target && leftover >= n1g) {
2728 					leftover -= n1g;
2729 					iaq->nirq += n1g;
2730 					n++;
2731 				}
2732 				iaq->nrxq1g = min(n, nrxq1g);
2733 #ifdef TCP_OFFLOAD
2734 				iaq->nofldrxq1g = min(n, nofldrxq1g);
2735 #endif
2736 			}
2737 
2738 			if (itype != INTR_MSI || powerof2(iaq->nirq))
2739 				goto allocate;
2740 		}
2741 
2742 		/*
2743 		 * Least desirable option: one interrupt vector for everything.
2744 		 */
2745 		iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2746 		iaq->intr_flags_10g = iaq->intr_flags_1g = 0;
2747 #ifdef TCP_OFFLOAD
2748 		if (is_offload(sc))
2749 			iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2750 #endif
2751 allocate:
2752 		navail = iaq->nirq;
2753 		rc = 0;
2754 		if (itype == INTR_MSIX)
2755 			rc = pci_alloc_msix(sc->dev, &navail);
2756 		else if (itype == INTR_MSI)
2757 			rc = pci_alloc_msi(sc->dev, &navail);
2758 
2759 		if (rc == 0) {
2760 			if (navail == iaq->nirq)
2761 				return (0);
2762 
2763 			/*
2764 			 * Didn't get the number requested.  Use whatever number
2765 			 * the kernel is willing to allocate (it's in navail).
2766 			 */
2767 			device_printf(sc->dev, "fewer vectors than requested, "
2768 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2769 			    itype, iaq->nirq, navail);
2770 			pci_release_msi(sc->dev);
2771 			goto restart;
2772 		}
2773 
2774 		device_printf(sc->dev,
2775 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2776 		    itype, rc, iaq->nirq, navail);
2777 	}
2778 
2779 	device_printf(sc->dev,
2780 	    "failed to find a usable interrupt type.  "
2781 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2782 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2783 
2784 	return (ENXIO);
2785 }
2786 
2787 #define FW_VERSION(chip) ( \
2788     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2789     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2790     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2791     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2792 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2793 
2794 struct fw_info {
2795 	uint8_t chip;
2796 	char *kld_name;
2797 	char *fw_mod_name;
2798 	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
2799 } fw_info[] = {
2800 	{
2801 		.chip = CHELSIO_T4,
2802 		.kld_name = "t4fw_cfg",
2803 		.fw_mod_name = "t4fw",
2804 		.fw_hdr = {
2805 			.chip = FW_HDR_CHIP_T4,
2806 			.fw_ver = htobe32_const(FW_VERSION(T4)),
2807 			.intfver_nic = FW_INTFVER(T4, NIC),
2808 			.intfver_vnic = FW_INTFVER(T4, VNIC),
2809 			.intfver_ofld = FW_INTFVER(T4, OFLD),
2810 			.intfver_ri = FW_INTFVER(T4, RI),
2811 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2812 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
2813 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2814 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
2815 		},
2816 	}, {
2817 		.chip = CHELSIO_T5,
2818 		.kld_name = "t5fw_cfg",
2819 		.fw_mod_name = "t5fw",
2820 		.fw_hdr = {
2821 			.chip = FW_HDR_CHIP_T5,
2822 			.fw_ver = htobe32_const(FW_VERSION(T5)),
2823 			.intfver_nic = FW_INTFVER(T5, NIC),
2824 			.intfver_vnic = FW_INTFVER(T5, VNIC),
2825 			.intfver_ofld = FW_INTFVER(T5, OFLD),
2826 			.intfver_ri = FW_INTFVER(T5, RI),
2827 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
2828 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
2829 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
2830 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
2831 		},
2832 	}, {
2833 		.chip = CHELSIO_T6,
2834 		.kld_name = "t6fw_cfg",
2835 		.fw_mod_name = "t6fw",
2836 		.fw_hdr = {
2837 			.chip = FW_HDR_CHIP_T6,
2838 			.fw_ver = htobe32_const(FW_VERSION(T6)),
2839 			.intfver_nic = FW_INTFVER(T6, NIC),
2840 			.intfver_vnic = FW_INTFVER(T6, VNIC),
2841 			.intfver_ofld = FW_INTFVER(T6, OFLD),
2842 			.intfver_ri = FW_INTFVER(T6, RI),
2843 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
2844 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
2845 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
2846 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
2847 		},
2848 	}
2849 };
2850 
2851 static struct fw_info *
2852 find_fw_info(int chip)
2853 {
2854 	int i;
2855 
2856 	for (i = 0; i < nitems(fw_info); i++) {
2857 		if (fw_info[i].chip == chip)
2858 			return (&fw_info[i]);
2859 	}
2860 	return (NULL);
2861 }
2862 
2863 /*
2864  * Is the given firmware API compatible with the one the driver was compiled
2865  * with?
2866  */
2867 static int
2868 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
2869 {
2870 
2871 	/* short circuit if it's the exact same firmware version */
2872 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
2873 		return (1);
2874 
2875 	/*
2876 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
2877 	 * features that are supported in the driver.
2878 	 */
2879 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
2880 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
2881 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
2882 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
2883 		return (1);
2884 #undef SAME_INTF
2885 
2886 	return (0);
2887 }
2888 
2889 /*
2890  * The firmware in the KLD is usable, but should it be installed?  This routine
2891  * explains itself in detail if it indicates the KLD firmware should be
2892  * installed.
2893  */
2894 static int
2895 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
2896 {
2897 	const char *reason;
2898 
2899 	if (!card_fw_usable) {
2900 		reason = "incompatible or unusable";
2901 		goto install;
2902 	}
2903 
2904 	if (k > c) {
2905 		reason = "older than the version bundled with this driver";
2906 		goto install;
2907 	}
2908 
2909 	if (t4_fw_install == 2 && k != c) {
2910 		reason = "different than the version bundled with this driver";
2911 		goto install;
2912 	}
2913 
2914 	return (0);
2915 
2916 install:
2917 	if (t4_fw_install == 0) {
2918 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2919 		    "but the driver is prohibited from installing a different "
2920 		    "firmware on the card.\n",
2921 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2922 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
2923 
2924 		return (0);
2925 	}
2926 
2927 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2928 	    "installing firmware %u.%u.%u.%u on card.\n",
2929 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2930 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
2931 	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2932 	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2933 
2934 	return (1);
2935 }
2936 /*
2937  * Establish contact with the firmware and determine if we are the master driver
2938  * or not, and whether we are responsible for chip initialization.
2939  */
2940 static int
2941 prep_firmware(struct adapter *sc)
2942 {
2943 	const struct firmware *fw = NULL, *default_cfg;
2944 	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
2945 	enum dev_state state;
2946 	struct fw_info *fw_info;
2947 	struct fw_hdr *card_fw;		/* fw on the card */
2948 	const struct fw_hdr *kld_fw;	/* fw in the KLD */
2949 	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
2950 					   against */
2951 
2952 	/* Contact firmware. */
2953 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
2954 	if (rc < 0 || state == DEV_STATE_ERR) {
2955 		rc = -rc;
2956 		device_printf(sc->dev,
2957 		    "failed to connect to the firmware: %d, %d.\n", rc, state);
2958 		return (rc);
2959 	}
2960 	pf = rc;
2961 	if (pf == sc->mbox)
2962 		sc->flags |= MASTER_PF;
2963 	else if (state == DEV_STATE_UNINIT) {
2964 		/*
2965 		 * We didn't get to be the master so we definitely won't be
2966 		 * configuring the chip.  It's a bug if someone else hasn't
2967 		 * configured it already.
2968 		 */
2969 		device_printf(sc->dev, "couldn't be master(%d), "
2970 		    "device not already initialized either(%d).\n", rc, state);
2971 		return (EDOOFUS);
2972 	}
2973 
2974 	/* This is the firmware whose headers the driver was compiled against */
2975 	fw_info = find_fw_info(chip_id(sc));
2976 	if (fw_info == NULL) {
2977 		device_printf(sc->dev,
2978 		    "unable to look up firmware information for chip %d.\n",
2979 		    chip_id(sc));
2980 		return (EINVAL);
2981 	}
2982 	drv_fw = &fw_info->fw_hdr;
2983 
2984 	/*
2985 	 * The firmware KLD contains many modules.  The KLD name is also the
2986 	 * name of the module that contains the default config file.
2987 	 */
2988 	default_cfg = firmware_get(fw_info->kld_name);
2989 
2990 	/* Read the header of the firmware on the card */
2991 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
2992 	rc = -t4_read_flash(sc, FLASH_FW_START,
2993 	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
2994 	if (rc == 0)
2995 		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
2996 	else {
2997 		device_printf(sc->dev,
2998 		    "Unable to read card's firmware header: %d\n", rc);
2999 		card_fw_usable = 0;
3000 	}
3001 
3002 	/* This is the firmware in the KLD */
3003 	fw = firmware_get(fw_info->fw_mod_name);
3004 	if (fw != NULL) {
3005 		kld_fw = (const void *)fw->data;
3006 		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3007 	} else {
3008 		kld_fw = NULL;
3009 		kld_fw_usable = 0;
3010 	}
3011 
3012 	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3013 	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3014 		/*
3015 		 * Common case: the firmware on the card is an exact match and
3016 		 * the KLD is an exact match too, or the KLD is
3017 		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
3018 		 * here -- use cxgbetool loadfw if you want to reinstall the
3019 		 * same firmware as the one on the card.
3020 		 */
3021 	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3022 	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3023 	    be32toh(card_fw->fw_ver))) {
3024 
3025 		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3026 		if (rc != 0) {
3027 			device_printf(sc->dev,
3028 			    "failed to install firmware: %d\n", rc);
3029 			goto done;
3030 		}
3031 
3032 		/* Installed successfully, update the cached header too. */
3033 		memcpy(card_fw, kld_fw, sizeof(*card_fw));
3034 		card_fw_usable = 1;
3035 		need_fw_reset = 0;	/* already reset as part of load_fw */
3036 	}
3037 
3038 	if (!card_fw_usable) {
3039 		uint32_t d, c, k;
3040 
3041 		d = ntohl(drv_fw->fw_ver);
3042 		c = ntohl(card_fw->fw_ver);
3043 		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3044 
3045 		device_printf(sc->dev, "Cannot find a usable firmware: "
3046 		    "fw_install %d, chip state %d, "
3047 		    "driver compiled with %d.%d.%d.%d, "
3048 		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3049 		    t4_fw_install, state,
3050 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3051 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3052 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3053 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3054 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3055 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3056 		rc = EINVAL;
3057 		goto done;
3058 	}
3059 
3060 	/* Reset device */
3061 	if (need_fw_reset &&
3062 	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3063 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3064 		if (rc != ETIMEDOUT && rc != EIO)
3065 			t4_fw_bye(sc, sc->mbox);
3066 		goto done;
3067 	}
3068 	sc->flags |= FW_OK;
3069 
3070 	rc = get_params__pre_init(sc);
3071 	if (rc != 0)
3072 		goto done; /* error message displayed already */
3073 
3074 	/* Partition adapter resources as specified in the config file. */
3075 	if (state == DEV_STATE_UNINIT) {
3076 
3077 		KASSERT(sc->flags & MASTER_PF,
3078 		    ("%s: trying to change chip settings when not master.",
3079 		    __func__));
3080 
3081 		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3082 		if (rc != 0)
3083 			goto done;	/* error message displayed already */
3084 
3085 		t4_tweak_chip_settings(sc);
3086 
3087 		/* get basic stuff going */
3088 		rc = -t4_fw_initialize(sc, sc->mbox);
3089 		if (rc != 0) {
3090 			device_printf(sc->dev, "fw init failed: %d.\n", rc);
3091 			goto done;
3092 		}
3093 	} else {
3094 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3095 		sc->cfcsum = 0;
3096 	}
3097 
3098 done:
3099 	free(card_fw, M_CXGBE);
3100 	if (fw != NULL)
3101 		firmware_put(fw, FIRMWARE_UNLOAD);
3102 	if (default_cfg != NULL)
3103 		firmware_put(default_cfg, FIRMWARE_UNLOAD);
3104 
3105 	return (rc);
3106 }
3107 
3108 #define FW_PARAM_DEV(param) \
3109 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3110 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3111 #define FW_PARAM_PFVF(param) \
3112 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3113 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3114 
3115 /*
3116  * Partition chip resources for use between various PFs, VFs, etc.
3117  */
3118 static int
3119 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3120     const char *name_prefix)
3121 {
3122 	const struct firmware *cfg = NULL;
3123 	int rc = 0;
3124 	struct fw_caps_config_cmd caps;
3125 	uint32_t mtype, moff, finicsum, cfcsum;
3126 
3127 	/*
3128 	 * Figure out what configuration file to use.  Pick the default config
3129 	 * file for the card if the user hasn't specified one explicitly.
3130 	 */
3131 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3132 	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3133 		/* Card specific overrides go here. */
3134 		if (pci_get_device(sc->dev) == 0x440a)
3135 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3136 		if (is_fpga(sc))
3137 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3138 	}
3139 
3140 	/*
3141 	 * We need to load another module if the profile is anything except
3142 	 * "default" or "flash".
3143 	 */
3144 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3145 	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3146 		char s[32];
3147 
3148 		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3149 		cfg = firmware_get(s);
3150 		if (cfg == NULL) {
3151 			if (default_cfg != NULL) {
3152 				device_printf(sc->dev,
3153 				    "unable to load module \"%s\" for "
3154 				    "configuration profile \"%s\", will use "
3155 				    "the default config file instead.\n",
3156 				    s, sc->cfg_file);
3157 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3158 				    "%s", DEFAULT_CF);
3159 			} else {
3160 				device_printf(sc->dev,
3161 				    "unable to load module \"%s\" for "
3162 				    "configuration profile \"%s\", will use "
3163 				    "the config file on the card's flash "
3164 				    "instead.\n", s, sc->cfg_file);
3165 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3166 				    "%s", FLASH_CF);
3167 			}
3168 		}
3169 	}
3170 
3171 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3172 	    default_cfg == NULL) {
3173 		device_printf(sc->dev,
3174 		    "default config file not available, will use the config "
3175 		    "file on the card's flash instead.\n");
3176 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3177 	}
3178 
3179 	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3180 		u_int cflen;
3181 		const uint32_t *cfdata;
3182 		uint32_t param, val, addr;
3183 
3184 		KASSERT(cfg != NULL || default_cfg != NULL,
3185 		    ("%s: no config to upload", __func__));
3186 
3187 		/*
3188 		 * Ask the firmware where it wants us to upload the config file.
3189 		 */
3190 		param = FW_PARAM_DEV(CF);
3191 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3192 		if (rc != 0) {
3193 			/* No support for config file?  Shouldn't happen. */
3194 			device_printf(sc->dev,
3195 			    "failed to query config file location: %d.\n", rc);
3196 			goto done;
3197 		}
3198 		mtype = G_FW_PARAMS_PARAM_Y(val);
3199 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3200 
3201 		/*
3202 		 * XXX: sheer laziness.  We deliberately added 4 bytes of
3203 		 * useless stuffing/comments at the end of the config file so
3204 		 * it's ok to simply throw away the last remaining bytes when
3205 		 * the config file is not an exact multiple of 4.  This also
3206 		 * helps with the validate_mt_off_len check.
3207 		 */
3208 		if (cfg != NULL) {
3209 			cflen = cfg->datasize & ~3;
3210 			cfdata = cfg->data;
3211 		} else {
3212 			cflen = default_cfg->datasize & ~3;
3213 			cfdata = default_cfg->data;
3214 		}
3215 
3216 		if (cflen > FLASH_CFG_MAX_SIZE) {
3217 			device_printf(sc->dev,
3218 			    "config file too long (%d, max allowed is %d).  "
3219 			    "Will try to use the config on the card, if any.\n",
3220 			    cflen, FLASH_CFG_MAX_SIZE);
3221 			goto use_config_on_flash;
3222 		}
3223 
3224 		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3225 		if (rc != 0) {
3226 			device_printf(sc->dev,
3227 			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3228 			    "Will try to use the config on the card, if any.\n",
3229 			    __func__, mtype, moff, cflen, rc);
3230 			goto use_config_on_flash;
3231 		}
3232 		write_via_memwin(sc, 2, addr, cfdata, cflen);
3233 	} else {
3234 use_config_on_flash:
3235 		mtype = FW_MEMTYPE_FLASH;
3236 		moff = t4_flash_cfg_addr(sc);
3237 	}
3238 
3239 	bzero(&caps, sizeof(caps));
3240 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3241 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3242 	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3243 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3244 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3245 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3246 	if (rc != 0) {
3247 		device_printf(sc->dev,
3248 		    "failed to pre-process config file: %d "
3249 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3250 		goto done;
3251 	}
3252 
3253 	finicsum = be32toh(caps.finicsum);
3254 	cfcsum = be32toh(caps.cfcsum);
3255 	if (finicsum != cfcsum) {
3256 		device_printf(sc->dev,
3257 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3258 		    finicsum, cfcsum);
3259 	}
3260 	sc->cfcsum = cfcsum;
3261 
3262 #define LIMIT_CAPS(x) do { \
3263 	caps.x &= htobe16(t4_##x##_allowed); \
3264 } while (0)
3265 
3266 	/*
3267 	 * Let the firmware know what features will (not) be used so it can tune
3268 	 * things accordingly.
3269 	 */
3270 	LIMIT_CAPS(nbmcaps);
3271 	LIMIT_CAPS(linkcaps);
3272 	LIMIT_CAPS(switchcaps);
3273 	LIMIT_CAPS(niccaps);
3274 	LIMIT_CAPS(toecaps);
3275 	LIMIT_CAPS(rdmacaps);
3276 	LIMIT_CAPS(cryptocaps);
3277 	LIMIT_CAPS(iscsicaps);
3278 	LIMIT_CAPS(fcoecaps);
3279 #undef LIMIT_CAPS
3280 
3281 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3282 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3283 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3284 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3285 	if (rc != 0) {
3286 		device_printf(sc->dev,
3287 		    "failed to process config file: %d.\n", rc);
3288 	}
3289 done:
3290 	if (cfg != NULL)
3291 		firmware_put(cfg, FIRMWARE_UNLOAD);
3292 	return (rc);
3293 }
3294 
3295 /*
3296  * Retrieve parameters that are needed (or nice to have) very early.
3297  */
3298 static int
3299 get_params__pre_init(struct adapter *sc)
3300 {
3301 	int rc;
3302 	uint32_t param[2], val[2];
3303 
3304 	t4_get_version_info(sc);
3305 
3306 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3307 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3308 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3309 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3310 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3311 
3312 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3313 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3314 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3315 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3316 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3317 
3318 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3319 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3320 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3321 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3322 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3323 
3324 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3325 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3326 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3327 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3328 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3329 
3330 	param[0] = FW_PARAM_DEV(PORTVEC);
3331 	param[1] = FW_PARAM_DEV(CCLK);
3332 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3333 	if (rc != 0) {
3334 		device_printf(sc->dev,
3335 		    "failed to query parameters (pre_init): %d.\n", rc);
3336 		return (rc);
3337 	}
3338 
3339 	sc->params.portvec = val[0];
3340 	sc->params.nports = bitcount32(val[0]);
3341 	sc->params.vpd.cclk = val[1];
3342 
3343 	/* Read device log parameters. */
3344 	rc = -t4_init_devlog_params(sc, 1);
3345 	if (rc == 0)
3346 		fixup_devlog_params(sc);
3347 	else {
3348 		device_printf(sc->dev,
3349 		    "failed to get devlog parameters: %d.\n", rc);
3350 		rc = 0;	/* devlog isn't critical for device operation */
3351 	}
3352 
3353 	return (rc);
3354 }
3355 
3356 /*
3357  * Retrieve various parameters that are of interest to the driver.  The device
3358  * has been initialized by the firmware at this point.
3359  */
3360 static int
3361 get_params__post_init(struct adapter *sc)
3362 {
3363 	int rc;
3364 	uint32_t param[7], val[7];
3365 	struct fw_caps_config_cmd caps;
3366 
3367 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3368 	param[1] = FW_PARAM_PFVF(EQ_START);
3369 	param[2] = FW_PARAM_PFVF(FILTER_START);
3370 	param[3] = FW_PARAM_PFVF(FILTER_END);
3371 	param[4] = FW_PARAM_PFVF(L2T_START);
3372 	param[5] = FW_PARAM_PFVF(L2T_END);
3373 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3374 	if (rc != 0) {
3375 		device_printf(sc->dev,
3376 		    "failed to query parameters (post_init): %d.\n", rc);
3377 		return (rc);
3378 	}
3379 
3380 	sc->sge.iq_start = val[0];
3381 	sc->sge.eq_start = val[1];
3382 	sc->tids.ftid_base = val[2];
3383 	sc->tids.nftids = val[3] - val[2] + 1;
3384 	sc->params.ftid_min = val[2];
3385 	sc->params.ftid_max = val[3];
3386 	sc->vres.l2t.start = val[4];
3387 	sc->vres.l2t.size = val[5] - val[4] + 1;
3388 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3389 	    ("%s: L2 table size (%u) larger than expected (%u)",
3390 	    __func__, sc->vres.l2t.size, L2T_SIZE));
3391 
3392 	/* get capabilites */
3393 	bzero(&caps, sizeof(caps));
3394 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3395 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3396 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3397 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3398 	if (rc != 0) {
3399 		device_printf(sc->dev,
3400 		    "failed to get card capabilities: %d.\n", rc);
3401 		return (rc);
3402 	}
3403 
3404 #define READ_CAPS(x) do { \
3405 	sc->x = htobe16(caps.x); \
3406 } while (0)
3407 	READ_CAPS(nbmcaps);
3408 	READ_CAPS(linkcaps);
3409 	READ_CAPS(switchcaps);
3410 	READ_CAPS(niccaps);
3411 	READ_CAPS(toecaps);
3412 	READ_CAPS(rdmacaps);
3413 	READ_CAPS(cryptocaps);
3414 	READ_CAPS(iscsicaps);
3415 	READ_CAPS(fcoecaps);
3416 
3417 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3418 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3419 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3420 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3421 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3422 		if (rc != 0) {
3423 			device_printf(sc->dev,
3424 			    "failed to query NIC parameters: %d.\n", rc);
3425 			return (rc);
3426 		}
3427 		sc->tids.etid_base = val[0];
3428 		sc->params.etid_min = val[0];
3429 		sc->tids.netids = val[1] - val[0] + 1;
3430 		sc->params.netids = sc->tids.netids;
3431 		sc->params.eo_wr_cred = val[2];
3432 		sc->params.ethoffload = 1;
3433 	}
3434 
3435 	if (sc->toecaps) {
3436 		/* query offload-related parameters */
3437 		param[0] = FW_PARAM_DEV(NTID);
3438 		param[1] = FW_PARAM_PFVF(SERVER_START);
3439 		param[2] = FW_PARAM_PFVF(SERVER_END);
3440 		param[3] = FW_PARAM_PFVF(TDDP_START);
3441 		param[4] = FW_PARAM_PFVF(TDDP_END);
3442 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3443 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3444 		if (rc != 0) {
3445 			device_printf(sc->dev,
3446 			    "failed to query TOE parameters: %d.\n", rc);
3447 			return (rc);
3448 		}
3449 		sc->tids.ntids = val[0];
3450 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3451 		sc->tids.stid_base = val[1];
3452 		sc->tids.nstids = val[2] - val[1] + 1;
3453 		sc->vres.ddp.start = val[3];
3454 		sc->vres.ddp.size = val[4] - val[3] + 1;
3455 		sc->params.ofldq_wr_cred = val[5];
3456 		sc->params.offload = 1;
3457 	}
3458 	if (sc->rdmacaps) {
3459 		param[0] = FW_PARAM_PFVF(STAG_START);
3460 		param[1] = FW_PARAM_PFVF(STAG_END);
3461 		param[2] = FW_PARAM_PFVF(RQ_START);
3462 		param[3] = FW_PARAM_PFVF(RQ_END);
3463 		param[4] = FW_PARAM_PFVF(PBL_START);
3464 		param[5] = FW_PARAM_PFVF(PBL_END);
3465 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3466 		if (rc != 0) {
3467 			device_printf(sc->dev,
3468 			    "failed to query RDMA parameters(1): %d.\n", rc);
3469 			return (rc);
3470 		}
3471 		sc->vres.stag.start = val[0];
3472 		sc->vres.stag.size = val[1] - val[0] + 1;
3473 		sc->vres.rq.start = val[2];
3474 		sc->vres.rq.size = val[3] - val[2] + 1;
3475 		sc->vres.pbl.start = val[4];
3476 		sc->vres.pbl.size = val[5] - val[4] + 1;
3477 
3478 		param[0] = FW_PARAM_PFVF(SQRQ_START);
3479 		param[1] = FW_PARAM_PFVF(SQRQ_END);
3480 		param[2] = FW_PARAM_PFVF(CQ_START);
3481 		param[3] = FW_PARAM_PFVF(CQ_END);
3482 		param[4] = FW_PARAM_PFVF(OCQ_START);
3483 		param[5] = FW_PARAM_PFVF(OCQ_END);
3484 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3485 		if (rc != 0) {
3486 			device_printf(sc->dev,
3487 			    "failed to query RDMA parameters(2): %d.\n", rc);
3488 			return (rc);
3489 		}
3490 		sc->vres.qp.start = val[0];
3491 		sc->vres.qp.size = val[1] - val[0] + 1;
3492 		sc->vres.cq.start = val[2];
3493 		sc->vres.cq.size = val[3] - val[2] + 1;
3494 		sc->vres.ocq.start = val[4];
3495 		sc->vres.ocq.size = val[5] - val[4] + 1;
3496 	}
3497 	if (sc->iscsicaps) {
3498 		param[0] = FW_PARAM_PFVF(ISCSI_START);
3499 		param[1] = FW_PARAM_PFVF(ISCSI_END);
3500 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3501 		if (rc != 0) {
3502 			device_printf(sc->dev,
3503 			    "failed to query iSCSI parameters: %d.\n", rc);
3504 			return (rc);
3505 		}
3506 		sc->vres.iscsi.start = val[0];
3507 		sc->vres.iscsi.size = val[1] - val[0] + 1;
3508 	}
3509 
3510 	t4_init_sge_params(sc);
3511 
3512 	/*
3513 	 * We've got the params we wanted to query via the firmware.  Now grab
3514 	 * some others directly from the chip.
3515 	 */
3516 	rc = t4_read_chip_settings(sc);
3517 
3518 	return (rc);
3519 }
3520 
3521 static int
3522 set_params__post_init(struct adapter *sc)
3523 {
3524 	uint32_t param, val;
3525 
3526 	/* ask for encapsulated CPLs */
3527 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3528 	val = 1;
3529 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3530 
3531 	return (0);
3532 }
3533 
3534 #undef FW_PARAM_PFVF
3535 #undef FW_PARAM_DEV
3536 
3537 static void
3538 t4_set_desc(struct adapter *sc)
3539 {
3540 	char buf[128];
3541 	struct adapter_params *p = &sc->params;
3542 
3543 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
3544 
3545 	device_set_desc_copy(sc->dev, buf);
3546 }
3547 
3548 static void
3549 build_medialist(struct port_info *pi, struct ifmedia *media)
3550 {
3551 	int m;
3552 
3553 	PORT_LOCK(pi);
3554 
3555 	ifmedia_removeall(media);
3556 
3557 	m = IFM_ETHER | IFM_FDX;
3558 
3559 	switch(pi->port_type) {
3560 	case FW_PORT_TYPE_BT_XFI:
3561 	case FW_PORT_TYPE_BT_XAUI:
3562 		ifmedia_add(media, m | IFM_10G_T, 0, NULL);
3563 		/* fall through */
3564 
3565 	case FW_PORT_TYPE_BT_SGMII:
3566 		ifmedia_add(media, m | IFM_1000_T, 0, NULL);
3567 		ifmedia_add(media, m | IFM_100_TX, 0, NULL);
3568 		ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL);
3569 		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
3570 		break;
3571 
3572 	case FW_PORT_TYPE_CX4:
3573 		ifmedia_add(media, m | IFM_10G_CX4, 0, NULL);
3574 		ifmedia_set(media, m | IFM_10G_CX4);
3575 		break;
3576 
3577 	case FW_PORT_TYPE_QSFP_10G:
3578 	case FW_PORT_TYPE_SFP:
3579 	case FW_PORT_TYPE_FIBER_XFI:
3580 	case FW_PORT_TYPE_FIBER_XAUI:
3581 		switch (pi->mod_type) {
3582 
3583 		case FW_PORT_MOD_TYPE_LR:
3584 			ifmedia_add(media, m | IFM_10G_LR, 0, NULL);
3585 			ifmedia_set(media, m | IFM_10G_LR);
3586 			break;
3587 
3588 		case FW_PORT_MOD_TYPE_SR:
3589 			ifmedia_add(media, m | IFM_10G_SR, 0, NULL);
3590 			ifmedia_set(media, m | IFM_10G_SR);
3591 			break;
3592 
3593 		case FW_PORT_MOD_TYPE_LRM:
3594 			ifmedia_add(media, m | IFM_10G_LRM, 0, NULL);
3595 			ifmedia_set(media, m | IFM_10G_LRM);
3596 			break;
3597 
3598 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3599 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3600 			ifmedia_add(media, m | IFM_10G_TWINAX, 0, NULL);
3601 			ifmedia_set(media, m | IFM_10G_TWINAX);
3602 			break;
3603 
3604 		case FW_PORT_MOD_TYPE_NONE:
3605 			m &= ~IFM_FDX;
3606 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3607 			ifmedia_set(media, m | IFM_NONE);
3608 			break;
3609 
3610 		case FW_PORT_MOD_TYPE_NA:
3611 		case FW_PORT_MOD_TYPE_ER:
3612 		default:
3613 			device_printf(pi->dev,
3614 			    "unknown port_type (%d), mod_type (%d)\n",
3615 			    pi->port_type, pi->mod_type);
3616 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3617 			ifmedia_set(media, m | IFM_UNKNOWN);
3618 			break;
3619 		}
3620 		break;
3621 
3622 	case FW_PORT_TYPE_CR_QSFP:
3623 	case FW_PORT_TYPE_SFP28:
3624 	case FW_PORT_TYPE_KR_SFP28:
3625 		switch (pi->mod_type) {
3626 
3627 		case FW_PORT_MOD_TYPE_SR:
3628 			ifmedia_add(media, m | IFM_25G_SR, 0, NULL);
3629 			ifmedia_set(media, m | IFM_25G_SR);
3630 			break;
3631 
3632 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3633 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3634 			ifmedia_add(media, m | IFM_25G_CR, 0, NULL);
3635 			ifmedia_set(media, m | IFM_25G_CR);
3636 			break;
3637 
3638 		case FW_PORT_MOD_TYPE_NONE:
3639 			m &= ~IFM_FDX;
3640 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3641 			ifmedia_set(media, m | IFM_NONE);
3642 			break;
3643 
3644 		default:
3645 			device_printf(pi->dev,
3646 			    "unknown port_type (%d), mod_type (%d)\n",
3647 			    pi->port_type, pi->mod_type);
3648 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3649 			ifmedia_set(media, m | IFM_UNKNOWN);
3650 			break;
3651 		}
3652 		break;
3653 
3654 	case FW_PORT_TYPE_QSFP:
3655 		switch (pi->mod_type) {
3656 
3657 		case FW_PORT_MOD_TYPE_LR:
3658 			ifmedia_add(media, m | IFM_40G_LR4, 0, NULL);
3659 			ifmedia_set(media, m | IFM_40G_LR4);
3660 			break;
3661 
3662 		case FW_PORT_MOD_TYPE_SR:
3663 			ifmedia_add(media, m | IFM_40G_SR4, 0, NULL);
3664 			ifmedia_set(media, m | IFM_40G_SR4);
3665 			break;
3666 
3667 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3668 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3669 			ifmedia_add(media, m | IFM_40G_CR4, 0, NULL);
3670 			ifmedia_set(media, m | IFM_40G_CR4);
3671 			break;
3672 
3673 		case FW_PORT_MOD_TYPE_NONE:
3674 			m &= ~IFM_FDX;
3675 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3676 			ifmedia_set(media, m | IFM_NONE);
3677 			break;
3678 
3679 		default:
3680 			device_printf(pi->dev,
3681 			    "unknown port_type (%d), mod_type (%d)\n",
3682 			    pi->port_type, pi->mod_type);
3683 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3684 			ifmedia_set(media, m | IFM_UNKNOWN);
3685 			break;
3686 		}
3687 		break;
3688 
3689 	case FW_PORT_TYPE_KR4_100G:
3690 	case FW_PORT_TYPE_CR4_QSFP:
3691 		switch (pi->mod_type) {
3692 
3693 		case FW_PORT_MOD_TYPE_LR:
3694 			ifmedia_add(media, m | IFM_100G_LR4, 0, NULL);
3695 			ifmedia_set(media, m | IFM_100G_LR4);
3696 			break;
3697 
3698 		case FW_PORT_MOD_TYPE_SR:
3699 			ifmedia_add(media, m | IFM_100G_SR4, 0, NULL);
3700 			ifmedia_set(media, m | IFM_100G_SR4);
3701 			break;
3702 
3703 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3704 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3705 			ifmedia_add(media, m | IFM_100G_CR4, 0, NULL);
3706 			ifmedia_set(media, m | IFM_100G_CR4);
3707 			break;
3708 
3709 		case FW_PORT_MOD_TYPE_NONE:
3710 			m &= ~IFM_FDX;
3711 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3712 			ifmedia_set(media, m | IFM_NONE);
3713 			break;
3714 
3715 		default:
3716 			device_printf(pi->dev,
3717 			    "unknown port_type (%d), mod_type (%d)\n",
3718 			    pi->port_type, pi->mod_type);
3719 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3720 			ifmedia_set(media, m | IFM_UNKNOWN);
3721 			break;
3722 		}
3723 		break;
3724 
3725 	default:
3726 		device_printf(pi->dev,
3727 		    "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
3728 		    pi->mod_type);
3729 		ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3730 		ifmedia_set(media, m | IFM_UNKNOWN);
3731 		break;
3732 	}
3733 
3734 	PORT_UNLOCK(pi);
3735 }
3736 
3737 #define FW_MAC_EXACT_CHUNK	7
3738 
3739 /*
3740  * Program the port's XGMAC based on parameters in ifnet.  The caller also
3741  * indicates which parameters should be programmed (the rest are left alone).
3742  */
3743 int
3744 update_mac_settings(struct ifnet *ifp, int flags)
3745 {
3746 	int rc = 0;
3747 	struct vi_info *vi = ifp->if_softc;
3748 	struct port_info *pi = vi->pi;
3749 	struct adapter *sc = pi->adapter;
3750 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
3751 
3752 	ASSERT_SYNCHRONIZED_OP(sc);
3753 	KASSERT(flags, ("%s: not told what to update.", __func__));
3754 
3755 	if (flags & XGMAC_MTU)
3756 		mtu = ifp->if_mtu;
3757 
3758 	if (flags & XGMAC_PROMISC)
3759 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
3760 
3761 	if (flags & XGMAC_ALLMULTI)
3762 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
3763 
3764 	if (flags & XGMAC_VLANEX)
3765 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
3766 
3767 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
3768 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
3769 		    allmulti, 1, vlanex, false);
3770 		if (rc) {
3771 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
3772 			    rc);
3773 			return (rc);
3774 		}
3775 	}
3776 
3777 	if (flags & XGMAC_UCADDR) {
3778 		uint8_t ucaddr[ETHER_ADDR_LEN];
3779 
3780 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
3781 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
3782 		    ucaddr, true, true);
3783 		if (rc < 0) {
3784 			rc = -rc;
3785 			if_printf(ifp, "change_mac failed: %d\n", rc);
3786 			return (rc);
3787 		} else {
3788 			vi->xact_addr_filt = rc;
3789 			rc = 0;
3790 		}
3791 	}
3792 
3793 	if (flags & XGMAC_MCADDRS) {
3794 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
3795 		int del = 1;
3796 		uint64_t hash = 0;
3797 		struct ifmultiaddr *ifma;
3798 		int i = 0, j;
3799 
3800 		if_maddr_rlock(ifp);
3801 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3802 			if (ifma->ifma_addr->sa_family != AF_LINK)
3803 				continue;
3804 			mcaddr[i] =
3805 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
3806 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
3807 			i++;
3808 
3809 			if (i == FW_MAC_EXACT_CHUNK) {
3810 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
3811 				    del, i, mcaddr, NULL, &hash, 0);
3812 				if (rc < 0) {
3813 					rc = -rc;
3814 					for (j = 0; j < i; j++) {
3815 						if_printf(ifp,
3816 						    "failed to add mc address"
3817 						    " %02x:%02x:%02x:"
3818 						    "%02x:%02x:%02x rc=%d\n",
3819 						    mcaddr[j][0], mcaddr[j][1],
3820 						    mcaddr[j][2], mcaddr[j][3],
3821 						    mcaddr[j][4], mcaddr[j][5],
3822 						    rc);
3823 					}
3824 					goto mcfail;
3825 				}
3826 				del = 0;
3827 				i = 0;
3828 			}
3829 		}
3830 		if (i > 0) {
3831 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
3832 			    mcaddr, NULL, &hash, 0);
3833 			if (rc < 0) {
3834 				rc = -rc;
3835 				for (j = 0; j < i; j++) {
3836 					if_printf(ifp,
3837 					    "failed to add mc address"
3838 					    " %02x:%02x:%02x:"
3839 					    "%02x:%02x:%02x rc=%d\n",
3840 					    mcaddr[j][0], mcaddr[j][1],
3841 					    mcaddr[j][2], mcaddr[j][3],
3842 					    mcaddr[j][4], mcaddr[j][5],
3843 					    rc);
3844 				}
3845 				goto mcfail;
3846 			}
3847 		}
3848 
3849 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
3850 		if (rc != 0)
3851 			if_printf(ifp, "failed to set mc address hash: %d", rc);
3852 mcfail:
3853 		if_maddr_runlock(ifp);
3854 	}
3855 
3856 	return (rc);
3857 }
3858 
3859 /*
3860  * {begin|end}_synchronized_op must be called from the same thread.
3861  */
3862 int
3863 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
3864     char *wmesg)
3865 {
3866 	int rc, pri;
3867 
3868 #ifdef WITNESS
3869 	/* the caller thinks it's ok to sleep, but is it really? */
3870 	if (flags & SLEEP_OK)
3871 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
3872 		    "begin_synchronized_op");
3873 #endif
3874 
3875 	if (INTR_OK)
3876 		pri = PCATCH;
3877 	else
3878 		pri = 0;
3879 
3880 	ADAPTER_LOCK(sc);
3881 	for (;;) {
3882 
3883 		if (vi && IS_DOOMED(vi)) {
3884 			rc = ENXIO;
3885 			goto done;
3886 		}
3887 
3888 		if (!IS_BUSY(sc)) {
3889 			rc = 0;
3890 			break;
3891 		}
3892 
3893 		if (!(flags & SLEEP_OK)) {
3894 			rc = EBUSY;
3895 			goto done;
3896 		}
3897 
3898 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
3899 			rc = EINTR;
3900 			goto done;
3901 		}
3902 	}
3903 
3904 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
3905 	SET_BUSY(sc);
3906 #ifdef INVARIANTS
3907 	sc->last_op = wmesg;
3908 	sc->last_op_thr = curthread;
3909 	sc->last_op_flags = flags;
3910 #endif
3911 
3912 done:
3913 	if (!(flags & HOLD_LOCK) || rc)
3914 		ADAPTER_UNLOCK(sc);
3915 
3916 	return (rc);
3917 }
3918 
3919 /*
3920  * Tell if_ioctl and if_init that the VI is going away.  This is
3921  * special variant of begin_synchronized_op and must be paired with a
3922  * call to end_synchronized_op.
3923  */
3924 void
3925 doom_vi(struct adapter *sc, struct vi_info *vi)
3926 {
3927 
3928 	ADAPTER_LOCK(sc);
3929 	SET_DOOMED(vi);
3930 	wakeup(&sc->flags);
3931 	while (IS_BUSY(sc))
3932 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
3933 	SET_BUSY(sc);
3934 #ifdef INVARIANTS
3935 	sc->last_op = "t4detach";
3936 	sc->last_op_thr = curthread;
3937 	sc->last_op_flags = 0;
3938 #endif
3939 	ADAPTER_UNLOCK(sc);
3940 }
3941 
3942 /*
3943  * {begin|end}_synchronized_op must be called from the same thread.
3944  */
3945 void
3946 end_synchronized_op(struct adapter *sc, int flags)
3947 {
3948 
3949 	if (flags & LOCK_HELD)
3950 		ADAPTER_LOCK_ASSERT_OWNED(sc);
3951 	else
3952 		ADAPTER_LOCK(sc);
3953 
3954 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
3955 	CLR_BUSY(sc);
3956 	wakeup(&sc->flags);
3957 	ADAPTER_UNLOCK(sc);
3958 }
3959 
3960 static int
3961 cxgbe_init_synchronized(struct vi_info *vi)
3962 {
3963 	struct port_info *pi = vi->pi;
3964 	struct adapter *sc = pi->adapter;
3965 	struct ifnet *ifp = vi->ifp;
3966 	int rc = 0, i;
3967 	struct sge_txq *txq;
3968 
3969 	ASSERT_SYNCHRONIZED_OP(sc);
3970 
3971 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3972 		return (0);	/* already running */
3973 
3974 	if (!(sc->flags & FULL_INIT_DONE) &&
3975 	    ((rc = adapter_full_init(sc)) != 0))
3976 		return (rc);	/* error message displayed already */
3977 
3978 	if (!(vi->flags & VI_INIT_DONE) &&
3979 	    ((rc = vi_full_init(vi)) != 0))
3980 		return (rc); /* error message displayed already */
3981 
3982 	rc = update_mac_settings(ifp, XGMAC_ALL);
3983 	if (rc)
3984 		goto done;	/* error message displayed already */
3985 
3986 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
3987 	if (rc != 0) {
3988 		if_printf(ifp, "enable_vi failed: %d\n", rc);
3989 		goto done;
3990 	}
3991 
3992 	/*
3993 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
3994 	 * if this changes.
3995 	 */
3996 
3997 	for_each_txq(vi, i, txq) {
3998 		TXQ_LOCK(txq);
3999 		txq->eq.flags |= EQ_ENABLED;
4000 		TXQ_UNLOCK(txq);
4001 	}
4002 
4003 	/*
4004 	 * The first iq of the first port to come up is used for tracing.
4005 	 */
4006 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4007 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4008 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4009 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4010 		    V_QUEUENUMBER(sc->traceq));
4011 		pi->flags |= HAS_TRACEQ;
4012 	}
4013 
4014 	/* all ok */
4015 	PORT_LOCK(pi);
4016 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
4017 	pi->up_vis++;
4018 
4019 	if (pi->nvi > 1 || sc->flags & IS_VF)
4020 		callout_reset(&vi->tick, hz, vi_tick, vi);
4021 	else
4022 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4023 	PORT_UNLOCK(pi);
4024 done:
4025 	if (rc != 0)
4026 		cxgbe_uninit_synchronized(vi);
4027 
4028 	return (rc);
4029 }
4030 
4031 /*
4032  * Idempotent.
4033  */
4034 static int
4035 cxgbe_uninit_synchronized(struct vi_info *vi)
4036 {
4037 	struct port_info *pi = vi->pi;
4038 	struct adapter *sc = pi->adapter;
4039 	struct ifnet *ifp = vi->ifp;
4040 	int rc, i;
4041 	struct sge_txq *txq;
4042 
4043 	ASSERT_SYNCHRONIZED_OP(sc);
4044 
4045 	if (!(vi->flags & VI_INIT_DONE)) {
4046 		KASSERT(!(ifp->if_drv_flags & IFF_DRV_RUNNING),
4047 		    ("uninited VI is running"));
4048 		return (0);
4049 	}
4050 
4051 	/*
4052 	 * Disable the VI so that all its data in either direction is discarded
4053 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4054 	 * tick) intact as the TP can deliver negative advice or data that it's
4055 	 * holding in its RAM (for an offloaded connection) even after the VI is
4056 	 * disabled.
4057 	 */
4058 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4059 	if (rc) {
4060 		if_printf(ifp, "disable_vi failed: %d\n", rc);
4061 		return (rc);
4062 	}
4063 
4064 	for_each_txq(vi, i, txq) {
4065 		TXQ_LOCK(txq);
4066 		txq->eq.flags &= ~EQ_ENABLED;
4067 		TXQ_UNLOCK(txq);
4068 	}
4069 
4070 	PORT_LOCK(pi);
4071 	if (pi->nvi > 1 || sc->flags & IS_VF)
4072 		callout_stop(&vi->tick);
4073 	else
4074 		callout_stop(&pi->tick);
4075 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4076 		PORT_UNLOCK(pi);
4077 		return (0);
4078 	}
4079 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4080 	pi->up_vis--;
4081 	if (pi->up_vis > 0) {
4082 		PORT_UNLOCK(pi);
4083 		return (0);
4084 	}
4085 	PORT_UNLOCK(pi);
4086 
4087 	pi->link_cfg.link_ok = 0;
4088 	pi->link_cfg.speed = 0;
4089 	pi->link_cfg.link_down_rc = 255;
4090 	t4_os_link_changed(sc, pi->port_id, 0);
4091 
4092 	return (0);
4093 }
4094 
4095 /*
4096  * It is ok for this function to fail midway and return right away.  t4_detach
4097  * will walk the entire sc->irq list and clean up whatever is valid.
4098  */
4099 int
4100 t4_setup_intr_handlers(struct adapter *sc)
4101 {
4102 	int rc, rid, p, q, v;
4103 	char s[8];
4104 	struct irq *irq;
4105 	struct port_info *pi;
4106 	struct vi_info *vi;
4107 	struct sge *sge = &sc->sge;
4108 	struct sge_rxq *rxq;
4109 #ifdef TCP_OFFLOAD
4110 	struct sge_ofld_rxq *ofld_rxq;
4111 #endif
4112 #ifdef DEV_NETMAP
4113 	struct sge_nm_rxq *nm_rxq;
4114 #endif
4115 #ifdef RSS
4116 	int nbuckets = rss_getnumbuckets();
4117 #endif
4118 
4119 	/*
4120 	 * Setup interrupts.
4121 	 */
4122 	irq = &sc->irq[0];
4123 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
4124 	if (sc->intr_count == 1)
4125 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4126 
4127 	/* Multiple interrupts. */
4128 	if (sc->flags & IS_VF)
4129 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4130 		    ("%s: too few intr.", __func__));
4131 	else
4132 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4133 		    ("%s: too few intr.", __func__));
4134 
4135 	/* The first one is always error intr on PFs */
4136 	if (!(sc->flags & IS_VF)) {
4137 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4138 		if (rc != 0)
4139 			return (rc);
4140 		irq++;
4141 		rid++;
4142 	}
4143 
4144 	/* The second one is always the firmware event queue (first on VFs) */
4145 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4146 	if (rc != 0)
4147 		return (rc);
4148 	irq++;
4149 	rid++;
4150 
4151 	for_each_port(sc, p) {
4152 		pi = sc->port[p];
4153 		for_each_vi(pi, v, vi) {
4154 			vi->first_intr = rid - 1;
4155 
4156 			if (vi->nnmrxq > 0) {
4157 				int n = max(vi->nrxq, vi->nnmrxq);
4158 
4159 				MPASS(vi->flags & INTR_RXQ);
4160 
4161 				rxq = &sge->rxq[vi->first_rxq];
4162 #ifdef DEV_NETMAP
4163 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4164 #endif
4165 				for (q = 0; q < n; q++) {
4166 					snprintf(s, sizeof(s), "%x%c%x", p,
4167 					    'a' + v, q);
4168 					if (q < vi->nrxq)
4169 						irq->rxq = rxq++;
4170 #ifdef DEV_NETMAP
4171 					if (q < vi->nnmrxq)
4172 						irq->nm_rxq = nm_rxq++;
4173 #endif
4174 					rc = t4_alloc_irq(sc, irq, rid,
4175 					    t4_vi_intr, irq, s);
4176 					if (rc != 0)
4177 						return (rc);
4178 					irq++;
4179 					rid++;
4180 					vi->nintr++;
4181 				}
4182 			} else if (vi->flags & INTR_RXQ) {
4183 				for_each_rxq(vi, q, rxq) {
4184 					snprintf(s, sizeof(s), "%x%c%x", p,
4185 					    'a' + v, q);
4186 					rc = t4_alloc_irq(sc, irq, rid,
4187 					    t4_intr, rxq, s);
4188 					if (rc != 0)
4189 						return (rc);
4190 #ifdef RSS
4191 					bus_bind_intr(sc->dev, irq->res,
4192 					    rss_getcpu(q % nbuckets));
4193 #endif
4194 					irq++;
4195 					rid++;
4196 					vi->nintr++;
4197 				}
4198 			}
4199 #ifdef TCP_OFFLOAD
4200 			if (vi->flags & INTR_OFLD_RXQ) {
4201 				for_each_ofld_rxq(vi, q, ofld_rxq) {
4202 					snprintf(s, sizeof(s), "%x%c%x", p,
4203 					    'A' + v, q);
4204 					rc = t4_alloc_irq(sc, irq, rid,
4205 					    t4_intr, ofld_rxq, s);
4206 					if (rc != 0)
4207 						return (rc);
4208 					irq++;
4209 					rid++;
4210 					vi->nintr++;
4211 				}
4212 			}
4213 #endif
4214 		}
4215 	}
4216 	MPASS(irq == &sc->irq[sc->intr_count]);
4217 
4218 	return (0);
4219 }
4220 
4221 int
4222 adapter_full_init(struct adapter *sc)
4223 {
4224 	int rc, i;
4225 
4226 	ASSERT_SYNCHRONIZED_OP(sc);
4227 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4228 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
4229 	    ("%s: FULL_INIT_DONE already", __func__));
4230 
4231 	/*
4232 	 * queues that belong to the adapter (not any particular port).
4233 	 */
4234 	rc = t4_setup_adapter_queues(sc);
4235 	if (rc != 0)
4236 		goto done;
4237 
4238 	for (i = 0; i < nitems(sc->tq); i++) {
4239 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
4240 		    taskqueue_thread_enqueue, &sc->tq[i]);
4241 		if (sc->tq[i] == NULL) {
4242 			device_printf(sc->dev,
4243 			    "failed to allocate task queue %d\n", i);
4244 			rc = ENOMEM;
4245 			goto done;
4246 		}
4247 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
4248 		    device_get_nameunit(sc->dev), i);
4249 	}
4250 
4251 	if (!(sc->flags & IS_VF))
4252 		t4_intr_enable(sc);
4253 	sc->flags |= FULL_INIT_DONE;
4254 done:
4255 	if (rc != 0)
4256 		adapter_full_uninit(sc);
4257 
4258 	return (rc);
4259 }
4260 
4261 int
4262 adapter_full_uninit(struct adapter *sc)
4263 {
4264 	int i;
4265 
4266 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4267 
4268 	t4_teardown_adapter_queues(sc);
4269 
4270 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
4271 		taskqueue_free(sc->tq[i]);
4272 		sc->tq[i] = NULL;
4273 	}
4274 
4275 	sc->flags &= ~FULL_INIT_DONE;
4276 
4277 	return (0);
4278 }
4279 
4280 #ifdef RSS
4281 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
4282     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
4283     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
4284     RSS_HASHTYPE_RSS_UDP_IPV6)
4285 
4286 /* Translates kernel hash types to hardware. */
4287 static int
4288 hashconfig_to_hashen(int hashconfig)
4289 {
4290 	int hashen = 0;
4291 
4292 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
4293 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
4294 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
4295 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
4296 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
4297 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4298 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4299 	}
4300 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
4301 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4302 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4303 	}
4304 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
4305 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4306 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
4307 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4308 
4309 	return (hashen);
4310 }
4311 
4312 /* Translates hardware hash types to kernel. */
4313 static int
4314 hashen_to_hashconfig(int hashen)
4315 {
4316 	int hashconfig = 0;
4317 
4318 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
4319 		/*
4320 		 * If UDP hashing was enabled it must have been enabled for
4321 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
4322 		 * enabling any 4-tuple hash is nonsense configuration.
4323 		 */
4324 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4325 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
4326 
4327 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4328 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
4329 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4330 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
4331 	}
4332 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4333 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
4334 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4335 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
4336 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
4337 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
4338 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
4339 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
4340 
4341 	return (hashconfig);
4342 }
4343 #endif
4344 
4345 int
4346 vi_full_init(struct vi_info *vi)
4347 {
4348 	struct adapter *sc = vi->pi->adapter;
4349 	struct ifnet *ifp = vi->ifp;
4350 	uint16_t *rss;
4351 	struct sge_rxq *rxq;
4352 	int rc, i, j, hashen;
4353 #ifdef RSS
4354 	int nbuckets = rss_getnumbuckets();
4355 	int hashconfig = rss_gethashconfig();
4356 	int extra;
4357 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4358 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4359 #endif
4360 
4361 	ASSERT_SYNCHRONIZED_OP(sc);
4362 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
4363 	    ("%s: VI_INIT_DONE already", __func__));
4364 
4365 	sysctl_ctx_init(&vi->ctx);
4366 	vi->flags |= VI_SYSCTL_CTX;
4367 
4368 	/*
4369 	 * Allocate tx/rx/fl queues for this VI.
4370 	 */
4371 	rc = t4_setup_vi_queues(vi);
4372 	if (rc != 0)
4373 		goto done;	/* error message displayed already */
4374 
4375 	/*
4376 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
4377 	 */
4378 	if (vi->nrxq > vi->rss_size) {
4379 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
4380 		    "some queues will never receive traffic.\n", vi->nrxq,
4381 		    vi->rss_size);
4382 	} else if (vi->rss_size % vi->nrxq) {
4383 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
4384 		    "expect uneven traffic distribution.\n", vi->nrxq,
4385 		    vi->rss_size);
4386 	}
4387 #ifdef RSS
4388 	MPASS(RSS_KEYSIZE == 40);
4389 	if (vi->nrxq != nbuckets) {
4390 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
4391 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
4392 	}
4393 
4394 	rss_getkey((void *)&raw_rss_key[0]);
4395 	for (i = 0; i < nitems(rss_key); i++) {
4396 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
4397 	}
4398 	t4_write_rss_key(sc, &rss_key[0], -1);
4399 #endif
4400 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
4401 	for (i = 0; i < vi->rss_size;) {
4402 #ifdef RSS
4403 		j = rss_get_indirection_to_bucket(i);
4404 		j %= vi->nrxq;
4405 		rxq = &sc->sge.rxq[vi->first_rxq + j];
4406 		rss[i++] = rxq->iq.abs_id;
4407 #else
4408 		for_each_rxq(vi, j, rxq) {
4409 			rss[i++] = rxq->iq.abs_id;
4410 			if (i == vi->rss_size)
4411 				break;
4412 		}
4413 #endif
4414 	}
4415 
4416 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
4417 	    vi->rss_size);
4418 	if (rc != 0) {
4419 		if_printf(ifp, "rss_config failed: %d\n", rc);
4420 		goto done;
4421 	}
4422 
4423 #ifdef RSS
4424 	hashen = hashconfig_to_hashen(hashconfig);
4425 
4426 	/*
4427 	 * We may have had to enable some hashes even though the global config
4428 	 * wants them disabled.  This is a potential problem that must be
4429 	 * reported to the user.
4430 	 */
4431 	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
4432 
4433 	/*
4434 	 * If we consider only the supported hash types, then the enabled hashes
4435 	 * are a superset of the requested hashes.  In other words, there cannot
4436 	 * be any supported hash that was requested but not enabled, but there
4437 	 * can be hashes that were not requested but had to be enabled.
4438 	 */
4439 	extra &= SUPPORTED_RSS_HASHTYPES;
4440 	MPASS((extra & hashconfig) == 0);
4441 
4442 	if (extra) {
4443 		if_printf(ifp,
4444 		    "global RSS config (0x%x) cannot be accommodated.\n",
4445 		    hashconfig);
4446 	}
4447 	if (extra & RSS_HASHTYPE_RSS_IPV4)
4448 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
4449 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
4450 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
4451 	if (extra & RSS_HASHTYPE_RSS_IPV6)
4452 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
4453 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
4454 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
4455 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
4456 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
4457 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
4458 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
4459 #else
4460 	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
4461 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
4462 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4463 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
4464 #endif
4465 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
4466 	if (rc != 0) {
4467 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
4468 		goto done;
4469 	}
4470 
4471 	vi->rss = rss;
4472 	vi->flags |= VI_INIT_DONE;
4473 done:
4474 	if (rc != 0)
4475 		vi_full_uninit(vi);
4476 
4477 	return (rc);
4478 }
4479 
4480 /*
4481  * Idempotent.
4482  */
4483 int
4484 vi_full_uninit(struct vi_info *vi)
4485 {
4486 	struct port_info *pi = vi->pi;
4487 	struct adapter *sc = pi->adapter;
4488 	int i;
4489 	struct sge_rxq *rxq;
4490 	struct sge_txq *txq;
4491 #ifdef TCP_OFFLOAD
4492 	struct sge_ofld_rxq *ofld_rxq;
4493 	struct sge_wrq *ofld_txq;
4494 #endif
4495 
4496 	if (vi->flags & VI_INIT_DONE) {
4497 
4498 		/* Need to quiesce queues.  */
4499 
4500 		/* XXX: Only for the first VI? */
4501 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
4502 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
4503 
4504 		for_each_txq(vi, i, txq) {
4505 			quiesce_txq(sc, txq);
4506 		}
4507 
4508 #ifdef TCP_OFFLOAD
4509 		for_each_ofld_txq(vi, i, ofld_txq) {
4510 			quiesce_wrq(sc, ofld_txq);
4511 		}
4512 #endif
4513 
4514 		for_each_rxq(vi, i, rxq) {
4515 			quiesce_iq(sc, &rxq->iq);
4516 			quiesce_fl(sc, &rxq->fl);
4517 		}
4518 
4519 #ifdef TCP_OFFLOAD
4520 		for_each_ofld_rxq(vi, i, ofld_rxq) {
4521 			quiesce_iq(sc, &ofld_rxq->iq);
4522 			quiesce_fl(sc, &ofld_rxq->fl);
4523 		}
4524 #endif
4525 		free(vi->rss, M_CXGBE);
4526 		free(vi->nm_rss, M_CXGBE);
4527 	}
4528 
4529 	t4_teardown_vi_queues(vi);
4530 	vi->flags &= ~VI_INIT_DONE;
4531 
4532 	return (0);
4533 }
4534 
4535 static void
4536 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
4537 {
4538 	struct sge_eq *eq = &txq->eq;
4539 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
4540 
4541 	(void) sc;	/* unused */
4542 
4543 #ifdef INVARIANTS
4544 	TXQ_LOCK(txq);
4545 	MPASS((eq->flags & EQ_ENABLED) == 0);
4546 	TXQ_UNLOCK(txq);
4547 #endif
4548 
4549 	/* Wait for the mp_ring to empty. */
4550 	while (!mp_ring_is_idle(txq->r)) {
4551 		mp_ring_check_drainage(txq->r, 0);
4552 		pause("rquiesce", 1);
4553 	}
4554 
4555 	/* Then wait for the hardware to finish. */
4556 	while (spg->cidx != htobe16(eq->pidx))
4557 		pause("equiesce", 1);
4558 
4559 	/* Finally, wait for the driver to reclaim all descriptors. */
4560 	while (eq->cidx != eq->pidx)
4561 		pause("dquiesce", 1);
4562 }
4563 
4564 static void
4565 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
4566 {
4567 
4568 	/* XXXTX */
4569 }
4570 
4571 static void
4572 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
4573 {
4574 	(void) sc;	/* unused */
4575 
4576 	/* Synchronize with the interrupt handler */
4577 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
4578 		pause("iqfree", 1);
4579 }
4580 
4581 static void
4582 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
4583 {
4584 	mtx_lock(&sc->sfl_lock);
4585 	FL_LOCK(fl);
4586 	fl->flags |= FL_DOOMED;
4587 	FL_UNLOCK(fl);
4588 	callout_stop(&sc->sfl_callout);
4589 	mtx_unlock(&sc->sfl_lock);
4590 
4591 	KASSERT((fl->flags & FL_STARVING) == 0,
4592 	    ("%s: still starving", __func__));
4593 }
4594 
4595 static int
4596 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
4597     driver_intr_t *handler, void *arg, char *name)
4598 {
4599 	int rc;
4600 
4601 	irq->rid = rid;
4602 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
4603 	    RF_SHAREABLE | RF_ACTIVE);
4604 	if (irq->res == NULL) {
4605 		device_printf(sc->dev,
4606 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
4607 		return (ENOMEM);
4608 	}
4609 
4610 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
4611 	    NULL, handler, arg, &irq->tag);
4612 	if (rc != 0) {
4613 		device_printf(sc->dev,
4614 		    "failed to setup interrupt for rid %d, name %s: %d\n",
4615 		    rid, name, rc);
4616 	} else if (name)
4617 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
4618 
4619 	return (rc);
4620 }
4621 
4622 static int
4623 t4_free_irq(struct adapter *sc, struct irq *irq)
4624 {
4625 	if (irq->tag)
4626 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
4627 	if (irq->res)
4628 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
4629 
4630 	bzero(irq, sizeof(*irq));
4631 
4632 	return (0);
4633 }
4634 
4635 static void
4636 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
4637 {
4638 
4639 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
4640 	t4_get_regs(sc, buf, regs->len);
4641 }
4642 
4643 #define	A_PL_INDIR_CMD	0x1f8
4644 
4645 #define	S_PL_AUTOINC	31
4646 #define	M_PL_AUTOINC	0x1U
4647 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
4648 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
4649 
4650 #define	S_PL_VFID	20
4651 #define	M_PL_VFID	0xffU
4652 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
4653 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
4654 
4655 #define	S_PL_ADDR	0
4656 #define	M_PL_ADDR	0xfffffU
4657 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
4658 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
4659 
4660 #define	A_PL_INDIR_DATA	0x1fc
4661 
4662 static uint64_t
4663 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
4664 {
4665 	u32 stats[2];
4666 
4667 	mtx_assert(&sc->reg_lock, MA_OWNED);
4668 	if (sc->flags & IS_VF) {
4669 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
4670 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
4671 	} else {
4672 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4673 		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4674 		    V_PL_ADDR(VF_MPS_REG(reg)));
4675 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
4676 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
4677 	}
4678 	return (((uint64_t)stats[1]) << 32 | stats[0]);
4679 }
4680 
4681 static void
4682 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
4683     struct fw_vi_stats_vf *stats)
4684 {
4685 
4686 #define GET_STAT(name) \
4687 	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
4688 
4689 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
4690 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
4691 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
4692 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
4693 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
4694 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
4695 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
4696 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
4697 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
4698 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
4699 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
4700 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
4701 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
4702 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
4703 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
4704 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
4705 
4706 #undef GET_STAT
4707 }
4708 
4709 static void
4710 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
4711 {
4712 	int reg;
4713 
4714 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4715 	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4716 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
4717 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
4718 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
4719 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
4720 }
4721 
4722 static void
4723 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
4724 {
4725 	struct timeval tv;
4726 	const struct timeval interval = {0, 250000};	/* 250ms */
4727 
4728 	if (!(vi->flags & VI_INIT_DONE))
4729 		return;
4730 
4731 	getmicrotime(&tv);
4732 	timevalsub(&tv, &interval);
4733 	if (timevalcmp(&tv, &vi->last_refreshed, <))
4734 		return;
4735 
4736 	mtx_lock(&sc->reg_lock);
4737 	t4_get_vi_stats(sc, vi->viid, &vi->stats);
4738 	getmicrotime(&vi->last_refreshed);
4739 	mtx_unlock(&sc->reg_lock);
4740 }
4741 
4742 static void
4743 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
4744 {
4745 	int i;
4746 	u_int v, tnl_cong_drops;
4747 	struct timeval tv;
4748 	const struct timeval interval = {0, 250000};	/* 250ms */
4749 
4750 	getmicrotime(&tv);
4751 	timevalsub(&tv, &interval);
4752 	if (timevalcmp(&tv, &pi->last_refreshed, <))
4753 		return;
4754 
4755 	tnl_cong_drops = 0;
4756 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
4757 	for (i = 0; i < sc->chip_params->nchan; i++) {
4758 		if (pi->rx_chan_map & (1 << i)) {
4759 			mtx_lock(&sc->reg_lock);
4760 			t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
4761 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
4762 			mtx_unlock(&sc->reg_lock);
4763 			tnl_cong_drops += v;
4764 		}
4765 	}
4766 	pi->tnl_cong_drops = tnl_cong_drops;
4767 	getmicrotime(&pi->last_refreshed);
4768 }
4769 
4770 static void
4771 cxgbe_tick(void *arg)
4772 {
4773 	struct port_info *pi = arg;
4774 	struct adapter *sc = pi->adapter;
4775 
4776 	PORT_LOCK_ASSERT_OWNED(pi);
4777 	cxgbe_refresh_stats(sc, pi);
4778 
4779 	callout_schedule(&pi->tick, hz);
4780 }
4781 
4782 void
4783 vi_tick(void *arg)
4784 {
4785 	struct vi_info *vi = arg;
4786 	struct adapter *sc = vi->pi->adapter;
4787 
4788 	vi_refresh_stats(sc, vi);
4789 
4790 	callout_schedule(&vi->tick, hz);
4791 }
4792 
4793 static void
4794 cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
4795 {
4796 	struct ifnet *vlan;
4797 
4798 	if (arg != ifp || ifp->if_type != IFT_ETHER)
4799 		return;
4800 
4801 	vlan = VLAN_DEVAT(ifp, vid);
4802 	VLAN_SETCOOKIE(vlan, ifp);
4803 }
4804 
4805 /*
4806  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
4807  */
4808 static char *caps_decoder[] = {
4809 	"\20\001IPMI\002NCSI",				/* 0: NBM */
4810 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
4811 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
4812 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
4813 	    "\006HASHFILTER\007ETHOFLD",
4814 	"\20\001TOE",					/* 4: TOE */
4815 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
4816 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
4817 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
4818 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
4819 	    "\007T10DIF"
4820 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
4821 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
4822 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
4823 		    "\004PO_INITIATOR\005PO_TARGET",
4824 };
4825 
4826 void
4827 t4_sysctls(struct adapter *sc)
4828 {
4829 	struct sysctl_ctx_list *ctx;
4830 	struct sysctl_oid *oid;
4831 	struct sysctl_oid_list *children, *c0;
4832 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
4833 
4834 	ctx = device_get_sysctl_ctx(sc->dev);
4835 
4836 	/*
4837 	 * dev.t4nex.X.
4838 	 */
4839 	oid = device_get_sysctl_tree(sc->dev);
4840 	c0 = children = SYSCTL_CHILDREN(oid);
4841 
4842 	sc->sc_do_rxcopy = 1;
4843 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
4844 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
4845 
4846 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
4847 	    sc->params.nports, "# of ports");
4848 
4849 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
4850 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
4851 	    sysctl_bitfield, "A", "available doorbells");
4852 
4853 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
4854 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
4855 
4856 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
4857 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
4858 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
4859 	    "interrupt holdoff timer values (us)");
4860 
4861 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
4862 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
4863 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
4864 	    "interrupt holdoff packet counter values");
4865 
4866 	t4_sge_sysctls(sc, ctx, children);
4867 
4868 	sc->lro_timeout = 100;
4869 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
4870 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
4871 
4872 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
4873 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
4874 
4875 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
4876 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
4877 
4878 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
4879 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
4880 
4881 	if (sc->flags & IS_VF)
4882 		return;
4883 
4884 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
4885 	    NULL, chip_rev(sc), "chip hardware revision");
4886 
4887 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
4888 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
4889 
4890 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
4891 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
4892 
4893 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
4894 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
4895 
4896 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
4897 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
4898 
4899 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
4900 	    sc->er_version, 0, "expansion ROM version");
4901 
4902 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
4903 	    sc->bs_version, 0, "bootstrap firmware version");
4904 
4905 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
4906 	    NULL, sc->params.scfg_vers, "serial config version");
4907 
4908 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
4909 	    NULL, sc->params.vpd_vers, "VPD version");
4910 
4911 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
4912 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
4913 
4914 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
4915 	    sc->cfcsum, "config file checksum");
4916 
4917 #define SYSCTL_CAP(name, n, text) \
4918 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
4919 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
4920 	    sysctl_bitfield, "A", "available " text " capabilities")
4921 
4922 	SYSCTL_CAP(nbmcaps, 0, "NBM");
4923 	SYSCTL_CAP(linkcaps, 1, "link");
4924 	SYSCTL_CAP(switchcaps, 2, "switch");
4925 	SYSCTL_CAP(niccaps, 3, "NIC");
4926 	SYSCTL_CAP(toecaps, 4, "TCP offload");
4927 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
4928 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
4929 	SYSCTL_CAP(cryptocaps, 7, "crypto");
4930 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
4931 #undef SYSCTL_CAP
4932 
4933 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
4934 	    NULL, sc->tids.nftids, "number of filters");
4935 
4936 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
4937 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
4938 	    "chip temperature (in Celsius)");
4939 
4940 #ifdef SBUF_DRAIN
4941 	/*
4942 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
4943 	 */
4944 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
4945 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
4946 	    "logs and miscellaneous information");
4947 	children = SYSCTL_CHILDREN(oid);
4948 
4949 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
4950 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4951 	    sysctl_cctrl, "A", "congestion control");
4952 
4953 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
4954 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4955 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
4956 
4957 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
4958 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
4959 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
4960 
4961 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
4962 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
4963 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
4964 
4965 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
4966 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
4967 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
4968 
4969 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
4970 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
4971 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
4972 
4973 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
4974 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
4975 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
4976 
4977 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
4978 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4979 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
4980 	    "A", "CIM logic analyzer");
4981 
4982 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
4983 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4984 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
4985 
4986 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
4987 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
4988 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
4989 
4990 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
4991 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
4992 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
4993 
4994 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
4995 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
4996 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
4997 
4998 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
4999 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5000 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5001 
5002 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5003 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5004 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5005 
5006 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5007 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5008 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5009 
5010 	if (chip_id(sc) > CHELSIO_T4) {
5011 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5012 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5013 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5014 
5015 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5016 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5017 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5018 	}
5019 
5020 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5021 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5022 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5023 
5024 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5025 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5026 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
5027 
5028 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5029 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5030 	    sysctl_cpl_stats, "A", "CPL statistics");
5031 
5032 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5033 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5034 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5035 
5036 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5037 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5038 	    sysctl_devlog, "A", "firmware's device log");
5039 
5040 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5041 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5042 	    sysctl_fcoe_stats, "A", "FCoE statistics");
5043 
5044 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5045 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5046 	    sysctl_hw_sched, "A", "hardware scheduler ");
5047 
5048 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5049 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5050 	    sysctl_l2t, "A", "hardware L2 table");
5051 
5052 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5053 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5054 	    sysctl_lb_stats, "A", "loopback statistics");
5055 
5056 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5057 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5058 	    sysctl_meminfo, "A", "memory regions");
5059 
5060 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5061 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5062 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5063 	    "A", "MPS TCAM entries");
5064 
5065 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5066 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5067 	    sysctl_path_mtus, "A", "path MTUs");
5068 
5069 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5070 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5071 	    sysctl_pm_stats, "A", "PM statistics");
5072 
5073 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5074 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5075 	    sysctl_rdma_stats, "A", "RDMA statistics");
5076 
5077 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5078 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5079 	    sysctl_tcp_stats, "A", "TCP statistics");
5080 
5081 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5082 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5083 	    sysctl_tids, "A", "TID information");
5084 
5085 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5086 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5087 	    sysctl_tp_err_stats, "A", "TP error statistics");
5088 
5089 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5090 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5091 	    "TP logic analyzer event capture mask");
5092 
5093 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5094 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5095 	    sysctl_tp_la, "A", "TP logic analyzer");
5096 
5097 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5098 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5099 	    sysctl_tx_rate, "A", "Tx rate");
5100 
5101 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5102 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5103 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5104 
5105 	if (chip_id(sc) >= CHELSIO_T5) {
5106 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5107 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5108 		    sysctl_wcwr_stats, "A", "write combined work requests");
5109 	}
5110 #endif
5111 
5112 #ifdef TCP_OFFLOAD
5113 	if (is_offload(sc)) {
5114 		/*
5115 		 * dev.t4nex.X.toe.
5116 		 */
5117 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5118 		    NULL, "TOE parameters");
5119 		children = SYSCTL_CHILDREN(oid);
5120 
5121 		sc->tt.sndbuf = 256 * 1024;
5122 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5123 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
5124 
5125 		sc->tt.ddp = 0;
5126 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5127 		    &sc->tt.ddp, 0, "DDP allowed");
5128 
5129 		sc->tt.rx_coalesce = 1;
5130 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5131 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5132 
5133 		sc->tt.tx_align = 1;
5134 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5135 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5136 
5137 		sc->tt.tx_zcopy = 0;
5138 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
5139 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
5140 		    "Enable zero-copy aio_write(2)");
5141 
5142 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
5143 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
5144 		    "TP timer tick (us)");
5145 
5146 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
5147 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
5148 		    "TCP timestamp tick (us)");
5149 
5150 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5151 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5152 		    "DACK tick (us)");
5153 
5154 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5155 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5156 		    "IU", "DACK timer (us)");
5157 
5158 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5159 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5160 		    sysctl_tp_timer, "LU", "Retransmit min (us)");
5161 
5162 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5163 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5164 		    sysctl_tp_timer, "LU", "Retransmit max (us)");
5165 
5166 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5167 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5168 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
5169 
5170 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5171 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5172 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
5173 
5174 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5175 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5176 		    sysctl_tp_timer, "LU", "Keepidle idle timer (us)");
5177 
5178 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_intvl",
5179 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
5180 		    sysctl_tp_timer, "LU", "Keepidle interval (us)");
5181 
5182 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
5183 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
5184 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
5185 
5186 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
5187 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
5188 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
5189 	}
5190 #endif
5191 }
5192 
5193 void
5194 vi_sysctls(struct vi_info *vi)
5195 {
5196 	struct sysctl_ctx_list *ctx;
5197 	struct sysctl_oid *oid;
5198 	struct sysctl_oid_list *children;
5199 
5200 	ctx = device_get_sysctl_ctx(vi->dev);
5201 
5202 	/*
5203 	 * dev.v?(cxgbe|cxl).X.
5204 	 */
5205 	oid = device_get_sysctl_tree(vi->dev);
5206 	children = SYSCTL_CHILDREN(oid);
5207 
5208 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
5209 	    vi->viid, "VI identifer");
5210 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
5211 	    &vi->nrxq, 0, "# of rx queues");
5212 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5213 	    &vi->ntxq, 0, "# of tx queues");
5214 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5215 	    &vi->first_rxq, 0, "index of first rx queue");
5216 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5217 	    &vi->first_txq, 0, "index of first tx queue");
5218 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
5219 	    vi->rss_size, "size of RSS indirection table");
5220 
5221 	if (IS_MAIN_VI(vi)) {
5222 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
5223 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
5224 		    "Reserve queue 0 for non-flowid packets");
5225 	}
5226 
5227 #ifdef TCP_OFFLOAD
5228 	if (vi->nofldrxq != 0) {
5229 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5230 		    &vi->nofldrxq, 0,
5231 		    "# of rx queues for offloaded TCP connections");
5232 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5233 		    &vi->nofldtxq, 0,
5234 		    "# of tx queues for offloaded TCP connections");
5235 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5236 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
5237 		    "index of first TOE rx queue");
5238 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5239 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
5240 		    "index of first TOE tx queue");
5241 	}
5242 #endif
5243 #ifdef DEV_NETMAP
5244 	if (vi->nnmrxq != 0) {
5245 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5246 		    &vi->nnmrxq, 0, "# of netmap rx queues");
5247 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5248 		    &vi->nnmtxq, 0, "# of netmap tx queues");
5249 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5250 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
5251 		    "index of first netmap rx queue");
5252 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
5253 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
5254 		    "index of first netmap tx queue");
5255 	}
5256 #endif
5257 
5258 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
5259 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
5260 	    "holdoff timer index");
5261 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
5262 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
5263 	    "holdoff packet counter index");
5264 
5265 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
5266 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
5267 	    "rx queue size");
5268 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
5269 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
5270 	    "tx queue size");
5271 }
5272 
5273 static void
5274 cxgbe_sysctls(struct port_info *pi)
5275 {
5276 	struct sysctl_ctx_list *ctx;
5277 	struct sysctl_oid *oid;
5278 	struct sysctl_oid_list *children, *children2;
5279 	struct adapter *sc = pi->adapter;
5280 	int i;
5281 	char name[16];
5282 
5283 	ctx = device_get_sysctl_ctx(pi->dev);
5284 
5285 	/*
5286 	 * dev.cxgbe.X.
5287 	 */
5288 	oid = device_get_sysctl_tree(pi->dev);
5289 	children = SYSCTL_CHILDREN(oid);
5290 
5291 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
5292 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
5293 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
5294 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
5295 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
5296 		    "PHY temperature (in Celsius)");
5297 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
5298 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
5299 		    "PHY firmware version");
5300 	}
5301 
5302 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
5303 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
5304 	    "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
5305 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
5306 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
5307 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
5308 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
5309 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
5310 	    "autonegotiation (-1 = not supported)");
5311 
5312 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
5313 	    port_top_speed(pi), "max speed (in Gbps)");
5314 
5315 	if (sc->flags & IS_VF)
5316 		return;
5317 
5318 	/*
5319 	 * dev.(cxgbe|cxl).X.tc.
5320 	 */
5321 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
5322 	    "Tx scheduler traffic classes");
5323 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
5324 		struct tx_sched_class *tc = &pi->tc[i];
5325 
5326 		snprintf(name, sizeof(name), "%d", i);
5327 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
5328 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
5329 		    "traffic class"));
5330 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
5331 		    &tc->flags, 0, "flags");
5332 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
5333 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
5334 #ifdef SBUF_DRAIN
5335 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
5336 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
5337 		    sysctl_tc_params, "A", "traffic class parameters");
5338 #endif
5339 	}
5340 
5341 	/*
5342 	 * dev.cxgbe.X.stats.
5343 	 */
5344 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
5345 	    NULL, "port statistics");
5346 	children = SYSCTL_CHILDREN(oid);
5347 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
5348 	    &pi->tx_parse_error, 0,
5349 	    "# of tx packets with invalid length or # of segments");
5350 
5351 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
5352 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
5353 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
5354 	    sysctl_handle_t4_reg64, "QU", desc)
5355 
5356 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
5357 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
5358 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
5359 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
5360 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
5361 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
5362 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
5363 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
5364 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
5365 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
5366 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
5367 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
5368 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
5369 	    "# of tx frames in this range",
5370 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
5371 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
5372 	    "# of tx frames in this range",
5373 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
5374 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
5375 	    "# of tx frames in this range",
5376 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
5377 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
5378 	    "# of tx frames in this range",
5379 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
5380 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
5381 	    "# of tx frames in this range",
5382 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
5383 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
5384 	    "# of tx frames in this range",
5385 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
5386 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
5387 	    "# of tx frames in this range",
5388 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
5389 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
5390 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
5391 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
5392 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
5393 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
5394 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
5395 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5396 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5397 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5398 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5399 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5400 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5401 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5402 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5403 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5404 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5405 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5406 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5407 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5408 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5409 
5410 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5411 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5412 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5413 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5414 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5415 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5416 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5417 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5418 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5419 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5420 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5421 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5422 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5423 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5424 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5425 	    "# of frames received with bad FCS",
5426 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5427 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5428 	    "# of frames received with length error",
5429 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5430 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5431 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5432 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5433 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5434 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5435 	    "# of rx frames in this range",
5436 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5437 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5438 	    "# of rx frames in this range",
5439 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5440 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5441 	    "# of rx frames in this range",
5442 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5443 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5444 	    "# of rx frames in this range",
5445 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5446 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5447 	    "# of rx frames in this range",
5448 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5449 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5450 	    "# of rx frames in this range",
5451 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5452 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5453 	    "# of rx frames in this range",
5454 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5455 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5456 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5457 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5458 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5459 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5460 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5461 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5462 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5463 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5464 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5465 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5466 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5467 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5468 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5469 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5470 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5471 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5472 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5473 
5474 #undef SYSCTL_ADD_T4_REG64
5475 
5476 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5477 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5478 	    &pi->stats.name, desc)
5479 
5480 	/* We get these from port_stats and they may be stale by up to 1s */
5481 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5482 	    "# drops due to buffer-group 0 overflows");
5483 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5484 	    "# drops due to buffer-group 1 overflows");
5485 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5486 	    "# drops due to buffer-group 2 overflows");
5487 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5488 	    "# drops due to buffer-group 3 overflows");
5489 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5490 	    "# of buffer-group 0 truncated packets");
5491 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5492 	    "# of buffer-group 1 truncated packets");
5493 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5494 	    "# of buffer-group 2 truncated packets");
5495 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5496 	    "# of buffer-group 3 truncated packets");
5497 
5498 #undef SYSCTL_ADD_T4_PORTSTAT
5499 }
5500 
5501 static int
5502 sysctl_int_array(SYSCTL_HANDLER_ARGS)
5503 {
5504 	int rc, *i, space = 0;
5505 	struct sbuf sb;
5506 
5507 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
5508 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
5509 		if (space)
5510 			sbuf_printf(&sb, " ");
5511 		sbuf_printf(&sb, "%d", *i);
5512 		space = 1;
5513 	}
5514 	rc = sbuf_finish(&sb);
5515 	sbuf_delete(&sb);
5516 	return (rc);
5517 }
5518 
5519 static int
5520 sysctl_bitfield(SYSCTL_HANDLER_ARGS)
5521 {
5522 	int rc;
5523 	struct sbuf *sb;
5524 
5525 	rc = sysctl_wire_old_buffer(req, 0);
5526 	if (rc != 0)
5527 		return(rc);
5528 
5529 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5530 	if (sb == NULL)
5531 		return (ENOMEM);
5532 
5533 	sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
5534 	rc = sbuf_finish(sb);
5535 	sbuf_delete(sb);
5536 
5537 	return (rc);
5538 }
5539 
5540 static int
5541 sysctl_btphy(SYSCTL_HANDLER_ARGS)
5542 {
5543 	struct port_info *pi = arg1;
5544 	int op = arg2;
5545 	struct adapter *sc = pi->adapter;
5546 	u_int v;
5547 	int rc;
5548 
5549 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
5550 	if (rc)
5551 		return (rc);
5552 	/* XXX: magic numbers */
5553 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
5554 	    &v);
5555 	end_synchronized_op(sc, 0);
5556 	if (rc)
5557 		return (rc);
5558 	if (op == 0)
5559 		v /= 256;
5560 
5561 	rc = sysctl_handle_int(oidp, &v, 0, req);
5562 	return (rc);
5563 }
5564 
5565 static int
5566 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
5567 {
5568 	struct vi_info *vi = arg1;
5569 	int rc, val;
5570 
5571 	val = vi->rsrv_noflowq;
5572 	rc = sysctl_handle_int(oidp, &val, 0, req);
5573 	if (rc != 0 || req->newptr == NULL)
5574 		return (rc);
5575 
5576 	if ((val >= 1) && (vi->ntxq > 1))
5577 		vi->rsrv_noflowq = 1;
5578 	else
5579 		vi->rsrv_noflowq = 0;
5580 
5581 	return (rc);
5582 }
5583 
5584 static int
5585 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
5586 {
5587 	struct vi_info *vi = arg1;
5588 	struct adapter *sc = vi->pi->adapter;
5589 	int idx, rc, i;
5590 	struct sge_rxq *rxq;
5591 #ifdef TCP_OFFLOAD
5592 	struct sge_ofld_rxq *ofld_rxq;
5593 #endif
5594 	uint8_t v;
5595 
5596 	idx = vi->tmr_idx;
5597 
5598 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5599 	if (rc != 0 || req->newptr == NULL)
5600 		return (rc);
5601 
5602 	if (idx < 0 || idx >= SGE_NTIMERS)
5603 		return (EINVAL);
5604 
5605 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5606 	    "t4tmr");
5607 	if (rc)
5608 		return (rc);
5609 
5610 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
5611 	for_each_rxq(vi, i, rxq) {
5612 #ifdef atomic_store_rel_8
5613 		atomic_store_rel_8(&rxq->iq.intr_params, v);
5614 #else
5615 		rxq->iq.intr_params = v;
5616 #endif
5617 	}
5618 #ifdef TCP_OFFLOAD
5619 	for_each_ofld_rxq(vi, i, ofld_rxq) {
5620 #ifdef atomic_store_rel_8
5621 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
5622 #else
5623 		ofld_rxq->iq.intr_params = v;
5624 #endif
5625 	}
5626 #endif
5627 	vi->tmr_idx = idx;
5628 
5629 	end_synchronized_op(sc, LOCK_HELD);
5630 	return (0);
5631 }
5632 
5633 static int
5634 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
5635 {
5636 	struct vi_info *vi = arg1;
5637 	struct adapter *sc = vi->pi->adapter;
5638 	int idx, rc;
5639 
5640 	idx = vi->pktc_idx;
5641 
5642 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5643 	if (rc != 0 || req->newptr == NULL)
5644 		return (rc);
5645 
5646 	if (idx < -1 || idx >= SGE_NCOUNTERS)
5647 		return (EINVAL);
5648 
5649 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5650 	    "t4pktc");
5651 	if (rc)
5652 		return (rc);
5653 
5654 	if (vi->flags & VI_INIT_DONE)
5655 		rc = EBUSY; /* cannot be changed once the queues are created */
5656 	else
5657 		vi->pktc_idx = idx;
5658 
5659 	end_synchronized_op(sc, LOCK_HELD);
5660 	return (rc);
5661 }
5662 
5663 static int
5664 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
5665 {
5666 	struct vi_info *vi = arg1;
5667 	struct adapter *sc = vi->pi->adapter;
5668 	int qsize, rc;
5669 
5670 	qsize = vi->qsize_rxq;
5671 
5672 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5673 	if (rc != 0 || req->newptr == NULL)
5674 		return (rc);
5675 
5676 	if (qsize < 128 || (qsize & 7))
5677 		return (EINVAL);
5678 
5679 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5680 	    "t4rxqs");
5681 	if (rc)
5682 		return (rc);
5683 
5684 	if (vi->flags & VI_INIT_DONE)
5685 		rc = EBUSY; /* cannot be changed once the queues are created */
5686 	else
5687 		vi->qsize_rxq = qsize;
5688 
5689 	end_synchronized_op(sc, LOCK_HELD);
5690 	return (rc);
5691 }
5692 
5693 static int
5694 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
5695 {
5696 	struct vi_info *vi = arg1;
5697 	struct adapter *sc = vi->pi->adapter;
5698 	int qsize, rc;
5699 
5700 	qsize = vi->qsize_txq;
5701 
5702 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5703 	if (rc != 0 || req->newptr == NULL)
5704 		return (rc);
5705 
5706 	if (qsize < 128 || qsize > 65536)
5707 		return (EINVAL);
5708 
5709 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5710 	    "t4txqs");
5711 	if (rc)
5712 		return (rc);
5713 
5714 	if (vi->flags & VI_INIT_DONE)
5715 		rc = EBUSY; /* cannot be changed once the queues are created */
5716 	else
5717 		vi->qsize_txq = qsize;
5718 
5719 	end_synchronized_op(sc, LOCK_HELD);
5720 	return (rc);
5721 }
5722 
5723 static int
5724 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
5725 {
5726 	struct port_info *pi = arg1;
5727 	struct adapter *sc = pi->adapter;
5728 	struct link_config *lc = &pi->link_cfg;
5729 	int rc;
5730 
5731 	if (req->newptr == NULL) {
5732 		struct sbuf *sb;
5733 		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
5734 
5735 		rc = sysctl_wire_old_buffer(req, 0);
5736 		if (rc != 0)
5737 			return(rc);
5738 
5739 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5740 		if (sb == NULL)
5741 			return (ENOMEM);
5742 
5743 		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
5744 		rc = sbuf_finish(sb);
5745 		sbuf_delete(sb);
5746 	} else {
5747 		char s[2];
5748 		int n;
5749 
5750 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
5751 		s[1] = 0;
5752 
5753 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5754 		if (rc != 0)
5755 			return(rc);
5756 
5757 		if (s[1] != 0)
5758 			return (EINVAL);
5759 		if (s[0] < '0' || s[0] > '9')
5760 			return (EINVAL);	/* not a number */
5761 		n = s[0] - '0';
5762 		if (n & ~(PAUSE_TX | PAUSE_RX))
5763 			return (EINVAL);	/* some other bit is set too */
5764 
5765 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5766 		    "t4PAUSE");
5767 		if (rc)
5768 			return (rc);
5769 		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
5770 			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
5771 			lc->requested_fc |= n;
5772 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5773 		}
5774 		end_synchronized_op(sc, 0);
5775 	}
5776 
5777 	return (rc);
5778 }
5779 
5780 static int
5781 sysctl_fec(SYSCTL_HANDLER_ARGS)
5782 {
5783 	struct port_info *pi = arg1;
5784 	struct adapter *sc = pi->adapter;
5785 	struct link_config *lc = &pi->link_cfg;
5786 	int rc;
5787 
5788 	if (req->newptr == NULL) {
5789 		struct sbuf *sb;
5790 		static char *bits = "\20\1RS\2BASER_RS\3RESERVED";
5791 
5792 		rc = sysctl_wire_old_buffer(req, 0);
5793 		if (rc != 0)
5794 			return(rc);
5795 
5796 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5797 		if (sb == NULL)
5798 			return (ENOMEM);
5799 
5800 		sbuf_printf(sb, "%b", lc->fec & M_FW_PORT_CAP_FEC, bits);
5801 		rc = sbuf_finish(sb);
5802 		sbuf_delete(sb);
5803 	} else {
5804 		char s[2];
5805 		int n;
5806 
5807 		s[0] = '0' + (lc->requested_fec & M_FW_PORT_CAP_FEC);
5808 		s[1] = 0;
5809 
5810 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5811 		if (rc != 0)
5812 			return(rc);
5813 
5814 		if (s[1] != 0)
5815 			return (EINVAL);
5816 		if (s[0] < '0' || s[0] > '9')
5817 			return (EINVAL);	/* not a number */
5818 		n = s[0] - '0';
5819 		if (n & ~M_FW_PORT_CAP_FEC)
5820 			return (EINVAL);	/* some other bit is set too */
5821 
5822 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5823 		    "t4fec");
5824 		if (rc)
5825 			return (rc);
5826 		if ((lc->requested_fec & M_FW_PORT_CAP_FEC) != n) {
5827 			lc->requested_fec = n &
5828 			    G_FW_PORT_CAP_FEC(lc->supported);
5829 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5830 		}
5831 		end_synchronized_op(sc, 0);
5832 	}
5833 
5834 	return (rc);
5835 }
5836 
5837 static int
5838 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
5839 {
5840 	struct port_info *pi = arg1;
5841 	struct adapter *sc = pi->adapter;
5842 	struct link_config *lc = &pi->link_cfg;
5843 	int rc, val, old;
5844 
5845 	if (lc->supported & FW_PORT_CAP_ANEG)
5846 		val = lc->autoneg == AUTONEG_ENABLE ? 1 : 0;
5847 	else
5848 		val = -1;
5849 	rc = sysctl_handle_int(oidp, &val, 0, req);
5850 	if (rc != 0 || req->newptr == NULL)
5851 		return (rc);
5852 	if ((lc->supported & FW_PORT_CAP_ANEG) == 0)
5853 		return (ENOTSUP);
5854 
5855 	val = val ? AUTONEG_ENABLE : AUTONEG_DISABLE;
5856 	if (lc->autoneg == val)
5857 		return (0);	/* no change */
5858 
5859 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5860 	    "t4aneg");
5861 	if (rc)
5862 		return (rc);
5863 	old = lc->autoneg;
5864 	lc->autoneg = val;
5865 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5866 	if (rc != 0)
5867 		lc->autoneg = old;
5868 	return (rc);
5869 }
5870 
5871 static int
5872 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
5873 {
5874 	struct adapter *sc = arg1;
5875 	int reg = arg2;
5876 	uint64_t val;
5877 
5878 	val = t4_read_reg64(sc, reg);
5879 
5880 	return (sysctl_handle_64(oidp, &val, 0, req));
5881 }
5882 
5883 static int
5884 sysctl_temperature(SYSCTL_HANDLER_ARGS)
5885 {
5886 	struct adapter *sc = arg1;
5887 	int rc, t;
5888 	uint32_t param, val;
5889 
5890 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
5891 	if (rc)
5892 		return (rc);
5893 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5894 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
5895 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
5896 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
5897 	end_synchronized_op(sc, 0);
5898 	if (rc)
5899 		return (rc);
5900 
5901 	/* unknown is returned as 0 but we display -1 in that case */
5902 	t = val == 0 ? -1 : val;
5903 
5904 	rc = sysctl_handle_int(oidp, &t, 0, req);
5905 	return (rc);
5906 }
5907 
5908 #ifdef SBUF_DRAIN
5909 static int
5910 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
5911 {
5912 	struct adapter *sc = arg1;
5913 	struct sbuf *sb;
5914 	int rc, i;
5915 	uint16_t incr[NMTUS][NCCTRL_WIN];
5916 	static const char *dec_fac[] = {
5917 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
5918 		"0.9375"
5919 	};
5920 
5921 	rc = sysctl_wire_old_buffer(req, 0);
5922 	if (rc != 0)
5923 		return (rc);
5924 
5925 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5926 	if (sb == NULL)
5927 		return (ENOMEM);
5928 
5929 	t4_read_cong_tbl(sc, incr);
5930 
5931 	for (i = 0; i < NCCTRL_WIN; ++i) {
5932 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
5933 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
5934 		    incr[5][i], incr[6][i], incr[7][i]);
5935 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
5936 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
5937 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
5938 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
5939 	}
5940 
5941 	rc = sbuf_finish(sb);
5942 	sbuf_delete(sb);
5943 
5944 	return (rc);
5945 }
5946 
5947 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
5948 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
5949 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
5950 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
5951 };
5952 
5953 static int
5954 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
5955 {
5956 	struct adapter *sc = arg1;
5957 	struct sbuf *sb;
5958 	int rc, i, n, qid = arg2;
5959 	uint32_t *buf, *p;
5960 	char *qtype;
5961 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
5962 
5963 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
5964 	    ("%s: bad qid %d\n", __func__, qid));
5965 
5966 	if (qid < CIM_NUM_IBQ) {
5967 		/* inbound queue */
5968 		qtype = "IBQ";
5969 		n = 4 * CIM_IBQ_SIZE;
5970 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5971 		rc = t4_read_cim_ibq(sc, qid, buf, n);
5972 	} else {
5973 		/* outbound queue */
5974 		qtype = "OBQ";
5975 		qid -= CIM_NUM_IBQ;
5976 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
5977 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5978 		rc = t4_read_cim_obq(sc, qid, buf, n);
5979 	}
5980 
5981 	if (rc < 0) {
5982 		rc = -rc;
5983 		goto done;
5984 	}
5985 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
5986 
5987 	rc = sysctl_wire_old_buffer(req, 0);
5988 	if (rc != 0)
5989 		goto done;
5990 
5991 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
5992 	if (sb == NULL) {
5993 		rc = ENOMEM;
5994 		goto done;
5995 	}
5996 
5997 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
5998 	for (i = 0, p = buf; i < n; i += 16, p += 4)
5999 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
6000 		    p[2], p[3]);
6001 
6002 	rc = sbuf_finish(sb);
6003 	sbuf_delete(sb);
6004 done:
6005 	free(buf, M_CXGBE);
6006 	return (rc);
6007 }
6008 
6009 static int
6010 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
6011 {
6012 	struct adapter *sc = arg1;
6013 	u_int cfg;
6014 	struct sbuf *sb;
6015 	uint32_t *buf, *p;
6016 	int rc;
6017 
6018 	MPASS(chip_id(sc) <= CHELSIO_T5);
6019 
6020 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6021 	if (rc != 0)
6022 		return (rc);
6023 
6024 	rc = sysctl_wire_old_buffer(req, 0);
6025 	if (rc != 0)
6026 		return (rc);
6027 
6028 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6029 	if (sb == NULL)
6030 		return (ENOMEM);
6031 
6032 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6033 	    M_ZERO | M_WAITOK);
6034 
6035 	rc = -t4_cim_read_la(sc, buf, NULL);
6036 	if (rc != 0)
6037 		goto done;
6038 
6039 	sbuf_printf(sb, "Status   Data      PC%s",
6040 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6041 	    "     LS0Stat  LS0Addr             LS0Data");
6042 
6043 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
6044 		if (cfg & F_UPDBGLACAPTPCONLY) {
6045 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
6046 			    p[6], p[7]);
6047 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
6048 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
6049 			    p[4] & 0xff, p[5] >> 8);
6050 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
6051 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6052 			    p[1] & 0xf, p[2] >> 4);
6053 		} else {
6054 			sbuf_printf(sb,
6055 			    "\n  %02x   %x%07x %x%07x %08x %08x "
6056 			    "%08x%08x%08x%08x",
6057 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6058 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
6059 			    p[6], p[7]);
6060 		}
6061 	}
6062 
6063 	rc = sbuf_finish(sb);
6064 	sbuf_delete(sb);
6065 done:
6066 	free(buf, M_CXGBE);
6067 	return (rc);
6068 }
6069 
6070 static int
6071 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
6072 {
6073 	struct adapter *sc = arg1;
6074 	u_int cfg;
6075 	struct sbuf *sb;
6076 	uint32_t *buf, *p;
6077 	int rc;
6078 
6079 	MPASS(chip_id(sc) > CHELSIO_T5);
6080 
6081 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6082 	if (rc != 0)
6083 		return (rc);
6084 
6085 	rc = sysctl_wire_old_buffer(req, 0);
6086 	if (rc != 0)
6087 		return (rc);
6088 
6089 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6090 	if (sb == NULL)
6091 		return (ENOMEM);
6092 
6093 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6094 	    M_ZERO | M_WAITOK);
6095 
6096 	rc = -t4_cim_read_la(sc, buf, NULL);
6097 	if (rc != 0)
6098 		goto done;
6099 
6100 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
6101 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6102 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
6103 
6104 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
6105 		if (cfg & F_UPDBGLACAPTPCONLY) {
6106 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
6107 			    p[3] & 0xff, p[2], p[1], p[0]);
6108 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
6109 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
6110 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
6111 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
6112 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
6113 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
6114 			    p[6] >> 16);
6115 		} else {
6116 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
6117 			    "%08x %08x %08x %08x %08x %08x",
6118 			    (p[9] >> 16) & 0xff,
6119 			    p[9] & 0xffff, p[8] >> 16,
6120 			    p[8] & 0xffff, p[7] >> 16,
6121 			    p[7] & 0xffff, p[6] >> 16,
6122 			    p[2], p[1], p[0], p[5], p[4], p[3]);
6123 		}
6124 	}
6125 
6126 	rc = sbuf_finish(sb);
6127 	sbuf_delete(sb);
6128 done:
6129 	free(buf, M_CXGBE);
6130 	return (rc);
6131 }
6132 
6133 static int
6134 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
6135 {
6136 	struct adapter *sc = arg1;
6137 	u_int i;
6138 	struct sbuf *sb;
6139 	uint32_t *buf, *p;
6140 	int rc;
6141 
6142 	rc = sysctl_wire_old_buffer(req, 0);
6143 	if (rc != 0)
6144 		return (rc);
6145 
6146 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6147 	if (sb == NULL)
6148 		return (ENOMEM);
6149 
6150 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
6151 	    M_ZERO | M_WAITOK);
6152 
6153 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
6154 	p = buf;
6155 
6156 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6157 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
6158 		    p[1], p[0]);
6159 	}
6160 
6161 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
6162 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6163 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
6164 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
6165 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
6166 		    (p[1] >> 2) | ((p[2] & 3) << 30),
6167 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
6168 		    p[0] & 1);
6169 	}
6170 
6171 	rc = sbuf_finish(sb);
6172 	sbuf_delete(sb);
6173 	free(buf, M_CXGBE);
6174 	return (rc);
6175 }
6176 
6177 static int
6178 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
6179 {
6180 	struct adapter *sc = arg1;
6181 	u_int i;
6182 	struct sbuf *sb;
6183 	uint32_t *buf, *p;
6184 	int rc;
6185 
6186 	rc = sysctl_wire_old_buffer(req, 0);
6187 	if (rc != 0)
6188 		return (rc);
6189 
6190 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6191 	if (sb == NULL)
6192 		return (ENOMEM);
6193 
6194 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
6195 	    M_ZERO | M_WAITOK);
6196 
6197 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
6198 	p = buf;
6199 
6200 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
6201 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6202 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
6203 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
6204 		    p[4], p[3], p[2], p[1], p[0]);
6205 	}
6206 
6207 	sbuf_printf(sb, "\n\nCntl ID               Data");
6208 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6209 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
6210 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
6211 	}
6212 
6213 	rc = sbuf_finish(sb);
6214 	sbuf_delete(sb);
6215 	free(buf, M_CXGBE);
6216 	return (rc);
6217 }
6218 
6219 static int
6220 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
6221 {
6222 	struct adapter *sc = arg1;
6223 	struct sbuf *sb;
6224 	int rc, i;
6225 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6226 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6227 	uint16_t thres[CIM_NUM_IBQ];
6228 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
6229 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
6230 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
6231 
6232 	cim_num_obq = sc->chip_params->cim_num_obq;
6233 	if (is_t4(sc)) {
6234 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
6235 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
6236 	} else {
6237 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
6238 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
6239 	}
6240 	nq = CIM_NUM_IBQ + cim_num_obq;
6241 
6242 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
6243 	if (rc == 0)
6244 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
6245 	if (rc != 0)
6246 		return (rc);
6247 
6248 	t4_read_cimq_cfg(sc, base, size, thres);
6249 
6250 	rc = sysctl_wire_old_buffer(req, 0);
6251 	if (rc != 0)
6252 		return (rc);
6253 
6254 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6255 	if (sb == NULL)
6256 		return (ENOMEM);
6257 
6258 	sbuf_printf(sb,
6259 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
6260 
6261 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
6262 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
6263 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
6264 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6265 		    G_QUEREMFLITS(p[2]) * 16);
6266 	for ( ; i < nq; i++, p += 4, wr += 2)
6267 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
6268 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
6269 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6270 		    G_QUEREMFLITS(p[2]) * 16);
6271 
6272 	rc = sbuf_finish(sb);
6273 	sbuf_delete(sb);
6274 
6275 	return (rc);
6276 }
6277 
6278 static int
6279 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
6280 {
6281 	struct adapter *sc = arg1;
6282 	struct sbuf *sb;
6283 	int rc;
6284 	struct tp_cpl_stats stats;
6285 
6286 	rc = sysctl_wire_old_buffer(req, 0);
6287 	if (rc != 0)
6288 		return (rc);
6289 
6290 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6291 	if (sb == NULL)
6292 		return (ENOMEM);
6293 
6294 	mtx_lock(&sc->reg_lock);
6295 	t4_tp_get_cpl_stats(sc, &stats);
6296 	mtx_unlock(&sc->reg_lock);
6297 
6298 	if (sc->chip_params->nchan > 2) {
6299 		sbuf_printf(sb, "                 channel 0  channel 1"
6300 		    "  channel 2  channel 3");
6301 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
6302 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
6303 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
6304 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
6305 	} else {
6306 		sbuf_printf(sb, "                 channel 0  channel 1");
6307 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
6308 		    stats.req[0], stats.req[1]);
6309 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
6310 		    stats.rsp[0], stats.rsp[1]);
6311 	}
6312 
6313 	rc = sbuf_finish(sb);
6314 	sbuf_delete(sb);
6315 
6316 	return (rc);
6317 }
6318 
6319 static int
6320 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
6321 {
6322 	struct adapter *sc = arg1;
6323 	struct sbuf *sb;
6324 	int rc;
6325 	struct tp_usm_stats stats;
6326 
6327 	rc = sysctl_wire_old_buffer(req, 0);
6328 	if (rc != 0)
6329 		return(rc);
6330 
6331 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6332 	if (sb == NULL)
6333 		return (ENOMEM);
6334 
6335 	t4_get_usm_stats(sc, &stats);
6336 
6337 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
6338 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
6339 	sbuf_printf(sb, "Drops:  %u", stats.drops);
6340 
6341 	rc = sbuf_finish(sb);
6342 	sbuf_delete(sb);
6343 
6344 	return (rc);
6345 }
6346 
6347 static const char * const devlog_level_strings[] = {
6348 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
6349 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
6350 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
6351 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
6352 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
6353 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
6354 };
6355 
6356 static const char * const devlog_facility_strings[] = {
6357 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
6358 	[FW_DEVLOG_FACILITY_CF]		= "CF",
6359 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
6360 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
6361 	[FW_DEVLOG_FACILITY_RES]	= "RES",
6362 	[FW_DEVLOG_FACILITY_HW]		= "HW",
6363 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
6364 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
6365 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
6366 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
6367 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
6368 	[FW_DEVLOG_FACILITY_VI]		= "VI",
6369 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
6370 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
6371 	[FW_DEVLOG_FACILITY_TM]		= "TM",
6372 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
6373 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
6374 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
6375 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
6376 	[FW_DEVLOG_FACILITY_RI]		= "RI",
6377 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
6378 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
6379 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
6380 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
6381 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
6382 };
6383 
6384 static int
6385 sysctl_devlog(SYSCTL_HANDLER_ARGS)
6386 {
6387 	struct adapter *sc = arg1;
6388 	struct devlog_params *dparams = &sc->params.devlog;
6389 	struct fw_devlog_e *buf, *e;
6390 	int i, j, rc, nentries, first = 0;
6391 	struct sbuf *sb;
6392 	uint64_t ftstamp = UINT64_MAX;
6393 
6394 	if (dparams->addr == 0)
6395 		return (ENXIO);
6396 
6397 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
6398 	if (buf == NULL)
6399 		return (ENOMEM);
6400 
6401 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
6402 	if (rc != 0)
6403 		goto done;
6404 
6405 	nentries = dparams->size / sizeof(struct fw_devlog_e);
6406 	for (i = 0; i < nentries; i++) {
6407 		e = &buf[i];
6408 
6409 		if (e->timestamp == 0)
6410 			break;	/* end */
6411 
6412 		e->timestamp = be64toh(e->timestamp);
6413 		e->seqno = be32toh(e->seqno);
6414 		for (j = 0; j < 8; j++)
6415 			e->params[j] = be32toh(e->params[j]);
6416 
6417 		if (e->timestamp < ftstamp) {
6418 			ftstamp = e->timestamp;
6419 			first = i;
6420 		}
6421 	}
6422 
6423 	if (buf[first].timestamp == 0)
6424 		goto done;	/* nothing in the log */
6425 
6426 	rc = sysctl_wire_old_buffer(req, 0);
6427 	if (rc != 0)
6428 		goto done;
6429 
6430 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6431 	if (sb == NULL) {
6432 		rc = ENOMEM;
6433 		goto done;
6434 	}
6435 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
6436 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
6437 
6438 	i = first;
6439 	do {
6440 		e = &buf[i];
6441 		if (e->timestamp == 0)
6442 			break;	/* end */
6443 
6444 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
6445 		    e->seqno, e->timestamp,
6446 		    (e->level < nitems(devlog_level_strings) ?
6447 			devlog_level_strings[e->level] : "UNKNOWN"),
6448 		    (e->facility < nitems(devlog_facility_strings) ?
6449 			devlog_facility_strings[e->facility] : "UNKNOWN"));
6450 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
6451 		    e->params[2], e->params[3], e->params[4],
6452 		    e->params[5], e->params[6], e->params[7]);
6453 
6454 		if (++i == nentries)
6455 			i = 0;
6456 	} while (i != first);
6457 
6458 	rc = sbuf_finish(sb);
6459 	sbuf_delete(sb);
6460 done:
6461 	free(buf, M_CXGBE);
6462 	return (rc);
6463 }
6464 
6465 static int
6466 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
6467 {
6468 	struct adapter *sc = arg1;
6469 	struct sbuf *sb;
6470 	int rc;
6471 	struct tp_fcoe_stats stats[MAX_NCHAN];
6472 	int i, nchan = sc->chip_params->nchan;
6473 
6474 	rc = sysctl_wire_old_buffer(req, 0);
6475 	if (rc != 0)
6476 		return (rc);
6477 
6478 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6479 	if (sb == NULL)
6480 		return (ENOMEM);
6481 
6482 	for (i = 0; i < nchan; i++)
6483 		t4_get_fcoe_stats(sc, i, &stats[i]);
6484 
6485 	if (nchan > 2) {
6486 		sbuf_printf(sb, "                   channel 0        channel 1"
6487 		    "        channel 2        channel 3");
6488 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
6489 		    stats[0].octets_ddp, stats[1].octets_ddp,
6490 		    stats[2].octets_ddp, stats[3].octets_ddp);
6491 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
6492 		    stats[0].frames_ddp, stats[1].frames_ddp,
6493 		    stats[2].frames_ddp, stats[3].frames_ddp);
6494 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
6495 		    stats[0].frames_drop, stats[1].frames_drop,
6496 		    stats[2].frames_drop, stats[3].frames_drop);
6497 	} else {
6498 		sbuf_printf(sb, "                   channel 0        channel 1");
6499 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
6500 		    stats[0].octets_ddp, stats[1].octets_ddp);
6501 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
6502 		    stats[0].frames_ddp, stats[1].frames_ddp);
6503 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
6504 		    stats[0].frames_drop, stats[1].frames_drop);
6505 	}
6506 
6507 	rc = sbuf_finish(sb);
6508 	sbuf_delete(sb);
6509 
6510 	return (rc);
6511 }
6512 
6513 static int
6514 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
6515 {
6516 	struct adapter *sc = arg1;
6517 	struct sbuf *sb;
6518 	int rc, i;
6519 	unsigned int map, kbps, ipg, mode;
6520 	unsigned int pace_tab[NTX_SCHED];
6521 
6522 	rc = sysctl_wire_old_buffer(req, 0);
6523 	if (rc != 0)
6524 		return (rc);
6525 
6526 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6527 	if (sb == NULL)
6528 		return (ENOMEM);
6529 
6530 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
6531 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
6532 	t4_read_pace_tbl(sc, pace_tab);
6533 
6534 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
6535 	    "Class IPG (0.1 ns)   Flow IPG (us)");
6536 
6537 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
6538 		t4_get_tx_sched(sc, i, &kbps, &ipg);
6539 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
6540 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
6541 		if (kbps)
6542 			sbuf_printf(sb, "%9u     ", kbps);
6543 		else
6544 			sbuf_printf(sb, " disabled     ");
6545 
6546 		if (ipg)
6547 			sbuf_printf(sb, "%13u        ", ipg);
6548 		else
6549 			sbuf_printf(sb, "     disabled        ");
6550 
6551 		if (pace_tab[i])
6552 			sbuf_printf(sb, "%10u", pace_tab[i]);
6553 		else
6554 			sbuf_printf(sb, "  disabled");
6555 	}
6556 
6557 	rc = sbuf_finish(sb);
6558 	sbuf_delete(sb);
6559 
6560 	return (rc);
6561 }
6562 
6563 static int
6564 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
6565 {
6566 	struct adapter *sc = arg1;
6567 	struct sbuf *sb;
6568 	int rc, i, j;
6569 	uint64_t *p0, *p1;
6570 	struct lb_port_stats s[2];
6571 	static const char *stat_name[] = {
6572 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
6573 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
6574 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
6575 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
6576 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
6577 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
6578 		"BG2FramesTrunc:", "BG3FramesTrunc:"
6579 	};
6580 
6581 	rc = sysctl_wire_old_buffer(req, 0);
6582 	if (rc != 0)
6583 		return (rc);
6584 
6585 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6586 	if (sb == NULL)
6587 		return (ENOMEM);
6588 
6589 	memset(s, 0, sizeof(s));
6590 
6591 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
6592 		t4_get_lb_stats(sc, i, &s[0]);
6593 		t4_get_lb_stats(sc, i + 1, &s[1]);
6594 
6595 		p0 = &s[0].octets;
6596 		p1 = &s[1].octets;
6597 		sbuf_printf(sb, "%s                       Loopback %u"
6598 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
6599 
6600 		for (j = 0; j < nitems(stat_name); j++)
6601 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
6602 				   *p0++, *p1++);
6603 	}
6604 
6605 	rc = sbuf_finish(sb);
6606 	sbuf_delete(sb);
6607 
6608 	return (rc);
6609 }
6610 
6611 static int
6612 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
6613 {
6614 	int rc = 0;
6615 	struct port_info *pi = arg1;
6616 	struct link_config *lc = &pi->link_cfg;
6617 	struct sbuf *sb;
6618 
6619 	rc = sysctl_wire_old_buffer(req, 0);
6620 	if (rc != 0)
6621 		return(rc);
6622 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
6623 	if (sb == NULL)
6624 		return (ENOMEM);
6625 
6626 	if (lc->link_ok || lc->link_down_rc == 255)
6627 		sbuf_printf(sb, "n/a");
6628 	else
6629 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
6630 
6631 	rc = sbuf_finish(sb);
6632 	sbuf_delete(sb);
6633 
6634 	return (rc);
6635 }
6636 
6637 struct mem_desc {
6638 	unsigned int base;
6639 	unsigned int limit;
6640 	unsigned int idx;
6641 };
6642 
6643 static int
6644 mem_desc_cmp(const void *a, const void *b)
6645 {
6646 	return ((const struct mem_desc *)a)->base -
6647 	       ((const struct mem_desc *)b)->base;
6648 }
6649 
6650 static void
6651 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
6652     unsigned int to)
6653 {
6654 	unsigned int size;
6655 
6656 	if (from == to)
6657 		return;
6658 
6659 	size = to - from + 1;
6660 	if (size == 0)
6661 		return;
6662 
6663 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
6664 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
6665 }
6666 
6667 static int
6668 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
6669 {
6670 	struct adapter *sc = arg1;
6671 	struct sbuf *sb;
6672 	int rc, i, n;
6673 	uint32_t lo, hi, used, alloc;
6674 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
6675 	static const char *region[] = {
6676 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
6677 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
6678 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
6679 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
6680 		"RQUDP region:", "PBL region:", "TXPBL region:",
6681 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
6682 		"On-chip queues:"
6683 	};
6684 	struct mem_desc avail[4];
6685 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
6686 	struct mem_desc *md = mem;
6687 
6688 	rc = sysctl_wire_old_buffer(req, 0);
6689 	if (rc != 0)
6690 		return (rc);
6691 
6692 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6693 	if (sb == NULL)
6694 		return (ENOMEM);
6695 
6696 	for (i = 0; i < nitems(mem); i++) {
6697 		mem[i].limit = 0;
6698 		mem[i].idx = i;
6699 	}
6700 
6701 	/* Find and sort the populated memory ranges */
6702 	i = 0;
6703 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
6704 	if (lo & F_EDRAM0_ENABLE) {
6705 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
6706 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
6707 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
6708 		avail[i].idx = 0;
6709 		i++;
6710 	}
6711 	if (lo & F_EDRAM1_ENABLE) {
6712 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
6713 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
6714 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
6715 		avail[i].idx = 1;
6716 		i++;
6717 	}
6718 	if (lo & F_EXT_MEM_ENABLE) {
6719 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
6720 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
6721 		avail[i].limit = avail[i].base +
6722 		    (G_EXT_MEM_SIZE(hi) << 20);
6723 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
6724 		i++;
6725 	}
6726 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
6727 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
6728 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
6729 		avail[i].limit = avail[i].base +
6730 		    (G_EXT_MEM1_SIZE(hi) << 20);
6731 		avail[i].idx = 4;
6732 		i++;
6733 	}
6734 	if (!i)                                    /* no memory available */
6735 		return 0;
6736 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
6737 
6738 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
6739 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
6740 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
6741 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
6742 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
6743 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
6744 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
6745 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
6746 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
6747 
6748 	/* the next few have explicit upper bounds */
6749 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
6750 	md->limit = md->base - 1 +
6751 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
6752 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
6753 	md++;
6754 
6755 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
6756 	md->limit = md->base - 1 +
6757 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
6758 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
6759 	md++;
6760 
6761 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
6762 		if (chip_id(sc) <= CHELSIO_T5)
6763 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
6764 		else
6765 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
6766 		md->limit = 0;
6767 	} else {
6768 		md->base = 0;
6769 		md->idx = nitems(region);  /* hide it */
6770 	}
6771 	md++;
6772 
6773 #define ulp_region(reg) \
6774 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
6775 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
6776 
6777 	ulp_region(RX_ISCSI);
6778 	ulp_region(RX_TDDP);
6779 	ulp_region(TX_TPT);
6780 	ulp_region(RX_STAG);
6781 	ulp_region(RX_RQ);
6782 	ulp_region(RX_RQUDP);
6783 	ulp_region(RX_PBL);
6784 	ulp_region(TX_PBL);
6785 #undef ulp_region
6786 
6787 	md->base = 0;
6788 	md->idx = nitems(region);
6789 	if (!is_t4(sc)) {
6790 		uint32_t size = 0;
6791 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
6792 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
6793 
6794 		if (is_t5(sc)) {
6795 			if (sge_ctrl & F_VFIFO_ENABLE)
6796 				size = G_DBVFIFO_SIZE(fifo_size);
6797 		} else
6798 			size = G_T6_DBVFIFO_SIZE(fifo_size);
6799 
6800 		if (size) {
6801 			md->base = G_BASEADDR(t4_read_reg(sc,
6802 			    A_SGE_DBVFIFO_BADDR));
6803 			md->limit = md->base + (size << 2) - 1;
6804 		}
6805 	}
6806 	md++;
6807 
6808 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
6809 	md->limit = 0;
6810 	md++;
6811 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
6812 	md->limit = 0;
6813 	md++;
6814 
6815 	md->base = sc->vres.ocq.start;
6816 	if (sc->vres.ocq.size)
6817 		md->limit = md->base + sc->vres.ocq.size - 1;
6818 	else
6819 		md->idx = nitems(region);  /* hide it */
6820 	md++;
6821 
6822 	/* add any address-space holes, there can be up to 3 */
6823 	for (n = 0; n < i - 1; n++)
6824 		if (avail[n].limit < avail[n + 1].base)
6825 			(md++)->base = avail[n].limit;
6826 	if (avail[n].limit)
6827 		(md++)->base = avail[n].limit;
6828 
6829 	n = md - mem;
6830 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
6831 
6832 	for (lo = 0; lo < i; lo++)
6833 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
6834 				avail[lo].limit - 1);
6835 
6836 	sbuf_printf(sb, "\n");
6837 	for (i = 0; i < n; i++) {
6838 		if (mem[i].idx >= nitems(region))
6839 			continue;                        /* skip holes */
6840 		if (!mem[i].limit)
6841 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
6842 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
6843 				mem[i].limit);
6844 	}
6845 
6846 	sbuf_printf(sb, "\n");
6847 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
6848 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
6849 	mem_region_show(sb, "uP RAM:", lo, hi);
6850 
6851 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
6852 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
6853 	mem_region_show(sb, "uP Extmem2:", lo, hi);
6854 
6855 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
6856 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
6857 		   G_PMRXMAXPAGE(lo),
6858 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
6859 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
6860 
6861 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
6862 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
6863 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
6864 		   G_PMTXMAXPAGE(lo),
6865 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
6866 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
6867 	sbuf_printf(sb, "%u p-structs\n",
6868 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
6869 
6870 	for (i = 0; i < 4; i++) {
6871 		if (chip_id(sc) > CHELSIO_T5)
6872 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
6873 		else
6874 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
6875 		if (is_t5(sc)) {
6876 			used = G_T5_USED(lo);
6877 			alloc = G_T5_ALLOC(lo);
6878 		} else {
6879 			used = G_USED(lo);
6880 			alloc = G_ALLOC(lo);
6881 		}
6882 		/* For T6 these are MAC buffer groups */
6883 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
6884 		    i, used, alloc);
6885 	}
6886 	for (i = 0; i < sc->chip_params->nchan; i++) {
6887 		if (chip_id(sc) > CHELSIO_T5)
6888 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
6889 		else
6890 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
6891 		if (is_t5(sc)) {
6892 			used = G_T5_USED(lo);
6893 			alloc = G_T5_ALLOC(lo);
6894 		} else {
6895 			used = G_USED(lo);
6896 			alloc = G_ALLOC(lo);
6897 		}
6898 		/* For T6 these are MAC buffer groups */
6899 		sbuf_printf(sb,
6900 		    "\nLoopback %d using %u pages out of %u allocated",
6901 		    i, used, alloc);
6902 	}
6903 
6904 	rc = sbuf_finish(sb);
6905 	sbuf_delete(sb);
6906 
6907 	return (rc);
6908 }
6909 
6910 static inline void
6911 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
6912 {
6913 	*mask = x | y;
6914 	y = htobe64(y);
6915 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
6916 }
6917 
6918 static int
6919 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
6920 {
6921 	struct adapter *sc = arg1;
6922 	struct sbuf *sb;
6923 	int rc, i;
6924 
6925 	MPASS(chip_id(sc) <= CHELSIO_T5);
6926 
6927 	rc = sysctl_wire_old_buffer(req, 0);
6928 	if (rc != 0)
6929 		return (rc);
6930 
6931 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6932 	if (sb == NULL)
6933 		return (ENOMEM);
6934 
6935 	sbuf_printf(sb,
6936 	    "Idx  Ethernet address     Mask     Vld Ports PF"
6937 	    "  VF              Replication             P0 P1 P2 P3  ML");
6938 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
6939 		uint64_t tcamx, tcamy, mask;
6940 		uint32_t cls_lo, cls_hi;
6941 		uint8_t addr[ETHER_ADDR_LEN];
6942 
6943 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
6944 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
6945 		if (tcamx & tcamy)
6946 			continue;
6947 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
6948 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
6949 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
6950 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
6951 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
6952 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
6953 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
6954 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
6955 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
6956 
6957 		if (cls_lo & F_REPLICATE) {
6958 			struct fw_ldst_cmd ldst_cmd;
6959 
6960 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
6961 			ldst_cmd.op_to_addrspace =
6962 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
6963 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
6964 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
6965 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
6966 			ldst_cmd.u.mps.rplc.fid_idx =
6967 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
6968 				V_FW_LDST_CMD_IDX(i));
6969 
6970 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
6971 			    "t4mps");
6972 			if (rc)
6973 				break;
6974 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
6975 			    sizeof(ldst_cmd), &ldst_cmd);
6976 			end_synchronized_op(sc, 0);
6977 
6978 			if (rc != 0) {
6979 				sbuf_printf(sb, "%36d", rc);
6980 				rc = 0;
6981 			} else {
6982 				sbuf_printf(sb, " %08x %08x %08x %08x",
6983 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
6984 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
6985 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
6986 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
6987 			}
6988 		} else
6989 			sbuf_printf(sb, "%36s", "");
6990 
6991 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
6992 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
6993 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
6994 	}
6995 
6996 	if (rc)
6997 		(void) sbuf_finish(sb);
6998 	else
6999 		rc = sbuf_finish(sb);
7000 	sbuf_delete(sb);
7001 
7002 	return (rc);
7003 }
7004 
7005 static int
7006 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
7007 {
7008 	struct adapter *sc = arg1;
7009 	struct sbuf *sb;
7010 	int rc, i;
7011 
7012 	MPASS(chip_id(sc) > CHELSIO_T5);
7013 
7014 	rc = sysctl_wire_old_buffer(req, 0);
7015 	if (rc != 0)
7016 		return (rc);
7017 
7018 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7019 	if (sb == NULL)
7020 		return (ENOMEM);
7021 
7022 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
7023 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
7024 	    "                           Replication"
7025 	    "                                    P0 P1 P2 P3  ML\n");
7026 
7027 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7028 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
7029 		uint16_t ivlan;
7030 		uint64_t tcamx, tcamy, val, mask;
7031 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
7032 		uint8_t addr[ETHER_ADDR_LEN];
7033 
7034 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
7035 		if (i < 256)
7036 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
7037 		else
7038 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
7039 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7040 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7041 		tcamy = G_DMACH(val) << 32;
7042 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7043 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7044 		lookup_type = G_DATALKPTYPE(data2);
7045 		port_num = G_DATAPORTNUM(data2);
7046 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7047 			/* Inner header VNI */
7048 			vniy = ((data2 & F_DATAVIDH2) << 23) |
7049 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7050 			dip_hit = data2 & F_DATADIPHIT;
7051 			vlan_vld = 0;
7052 		} else {
7053 			vniy = 0;
7054 			dip_hit = 0;
7055 			vlan_vld = data2 & F_DATAVIDH2;
7056 			ivlan = G_VIDL(val);
7057 		}
7058 
7059 		ctl |= V_CTLXYBITSEL(1);
7060 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7061 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7062 		tcamx = G_DMACH(val) << 32;
7063 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7064 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7065 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7066 			/* Inner header VNI mask */
7067 			vnix = ((data2 & F_DATAVIDH2) << 23) |
7068 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7069 		} else
7070 			vnix = 0;
7071 
7072 		if (tcamx & tcamy)
7073 			continue;
7074 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7075 
7076 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7077 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7078 
7079 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7080 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7081 			    "%012jx %06x %06x    -    -   %3c"
7082 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
7083 			    addr[1], addr[2], addr[3], addr[4], addr[5],
7084 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
7085 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7086 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7087 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7088 		} else {
7089 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7090 			    "%012jx    -       -   ", i, addr[0], addr[1],
7091 			    addr[2], addr[3], addr[4], addr[5],
7092 			    (uintmax_t)mask);
7093 
7094 			if (vlan_vld)
7095 				sbuf_printf(sb, "%4u   Y     ", ivlan);
7096 			else
7097 				sbuf_printf(sb, "  -    N     ");
7098 
7099 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
7100 			    lookup_type ? 'I' : 'O', port_num,
7101 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7102 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7103 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7104 		}
7105 
7106 
7107 		if (cls_lo & F_T6_REPLICATE) {
7108 			struct fw_ldst_cmd ldst_cmd;
7109 
7110 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7111 			ldst_cmd.op_to_addrspace =
7112 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7113 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7114 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7115 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7116 			ldst_cmd.u.mps.rplc.fid_idx =
7117 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7118 				V_FW_LDST_CMD_IDX(i));
7119 
7120 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7121 			    "t6mps");
7122 			if (rc)
7123 				break;
7124 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7125 			    sizeof(ldst_cmd), &ldst_cmd);
7126 			end_synchronized_op(sc, 0);
7127 
7128 			if (rc != 0) {
7129 				sbuf_printf(sb, "%72d", rc);
7130 				rc = 0;
7131 			} else {
7132 				sbuf_printf(sb, " %08x %08x %08x %08x"
7133 				    " %08x %08x %08x %08x",
7134 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
7135 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
7136 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
7137 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
7138 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7139 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7140 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7141 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7142 			}
7143 		} else
7144 			sbuf_printf(sb, "%72s", "");
7145 
7146 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
7147 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
7148 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
7149 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
7150 	}
7151 
7152 	if (rc)
7153 		(void) sbuf_finish(sb);
7154 	else
7155 		rc = sbuf_finish(sb);
7156 	sbuf_delete(sb);
7157 
7158 	return (rc);
7159 }
7160 
7161 static int
7162 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
7163 {
7164 	struct adapter *sc = arg1;
7165 	struct sbuf *sb;
7166 	int rc;
7167 	uint16_t mtus[NMTUS];
7168 
7169 	rc = sysctl_wire_old_buffer(req, 0);
7170 	if (rc != 0)
7171 		return (rc);
7172 
7173 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7174 	if (sb == NULL)
7175 		return (ENOMEM);
7176 
7177 	t4_read_mtu_tbl(sc, mtus, NULL);
7178 
7179 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
7180 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
7181 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
7182 	    mtus[14], mtus[15]);
7183 
7184 	rc = sbuf_finish(sb);
7185 	sbuf_delete(sb);
7186 
7187 	return (rc);
7188 }
7189 
7190 static int
7191 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
7192 {
7193 	struct adapter *sc = arg1;
7194 	struct sbuf *sb;
7195 	int rc, i;
7196 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
7197 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
7198 	static const char *tx_stats[MAX_PM_NSTATS] = {
7199 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
7200 		"Tx FIFO wait", NULL, "Tx latency"
7201 	};
7202 	static const char *rx_stats[MAX_PM_NSTATS] = {
7203 		"Read:", "Write bypass:", "Write mem:", "Flush:",
7204 		"Rx FIFO wait", NULL, "Rx latency"
7205 	};
7206 
7207 	rc = sysctl_wire_old_buffer(req, 0);
7208 	if (rc != 0)
7209 		return (rc);
7210 
7211 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7212 	if (sb == NULL)
7213 		return (ENOMEM);
7214 
7215 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
7216 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
7217 
7218 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
7219 	for (i = 0; i < 4; i++) {
7220 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7221 		    tx_cyc[i]);
7222 	}
7223 
7224 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
7225 	for (i = 0; i < 4; i++) {
7226 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7227 		    rx_cyc[i]);
7228 	}
7229 
7230 	if (chip_id(sc) > CHELSIO_T5) {
7231 		sbuf_printf(sb,
7232 		    "\n              Total wait      Total occupancy");
7233 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7234 		    tx_cyc[i]);
7235 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7236 		    rx_cyc[i]);
7237 
7238 		i += 2;
7239 		MPASS(i < nitems(tx_stats));
7240 
7241 		sbuf_printf(sb,
7242 		    "\n                   Reads           Total wait");
7243 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7244 		    tx_cyc[i]);
7245 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7246 		    rx_cyc[i]);
7247 	}
7248 
7249 	rc = sbuf_finish(sb);
7250 	sbuf_delete(sb);
7251 
7252 	return (rc);
7253 }
7254 
7255 static int
7256 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
7257 {
7258 	struct adapter *sc = arg1;
7259 	struct sbuf *sb;
7260 	int rc;
7261 	struct tp_rdma_stats stats;
7262 
7263 	rc = sysctl_wire_old_buffer(req, 0);
7264 	if (rc != 0)
7265 		return (rc);
7266 
7267 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7268 	if (sb == NULL)
7269 		return (ENOMEM);
7270 
7271 	mtx_lock(&sc->reg_lock);
7272 	t4_tp_get_rdma_stats(sc, &stats);
7273 	mtx_unlock(&sc->reg_lock);
7274 
7275 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
7276 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
7277 
7278 	rc = sbuf_finish(sb);
7279 	sbuf_delete(sb);
7280 
7281 	return (rc);
7282 }
7283 
7284 static int
7285 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
7286 {
7287 	struct adapter *sc = arg1;
7288 	struct sbuf *sb;
7289 	int rc;
7290 	struct tp_tcp_stats v4, v6;
7291 
7292 	rc = sysctl_wire_old_buffer(req, 0);
7293 	if (rc != 0)
7294 		return (rc);
7295 
7296 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7297 	if (sb == NULL)
7298 		return (ENOMEM);
7299 
7300 	mtx_lock(&sc->reg_lock);
7301 	t4_tp_get_tcp_stats(sc, &v4, &v6);
7302 	mtx_unlock(&sc->reg_lock);
7303 
7304 	sbuf_printf(sb,
7305 	    "                                IP                 IPv6\n");
7306 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
7307 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
7308 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
7309 	    v4.tcp_in_segs, v6.tcp_in_segs);
7310 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
7311 	    v4.tcp_out_segs, v6.tcp_out_segs);
7312 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
7313 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
7314 
7315 	rc = sbuf_finish(sb);
7316 	sbuf_delete(sb);
7317 
7318 	return (rc);
7319 }
7320 
7321 static int
7322 sysctl_tids(SYSCTL_HANDLER_ARGS)
7323 {
7324 	struct adapter *sc = arg1;
7325 	struct sbuf *sb;
7326 	int rc;
7327 	struct tid_info *t = &sc->tids;
7328 
7329 	rc = sysctl_wire_old_buffer(req, 0);
7330 	if (rc != 0)
7331 		return (rc);
7332 
7333 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7334 	if (sb == NULL)
7335 		return (ENOMEM);
7336 
7337 	if (t->natids) {
7338 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
7339 		    t->atids_in_use);
7340 	}
7341 
7342 	if (t->ntids) {
7343 		sbuf_printf(sb, "TID range: ");
7344 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7345 			uint32_t b, hb;
7346 
7347 			if (chip_id(sc) <= CHELSIO_T5) {
7348 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
7349 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
7350 			} else {
7351 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
7352 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
7353 			}
7354 
7355 			if (b)
7356 				sbuf_printf(sb, "0-%u, ", b - 1);
7357 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
7358 		} else
7359 			sbuf_printf(sb, "0-%u", t->ntids - 1);
7360 		sbuf_printf(sb, ", in use: %u\n",
7361 		    atomic_load_acq_int(&t->tids_in_use));
7362 	}
7363 
7364 	if (t->nstids) {
7365 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
7366 		    t->stid_base + t->nstids - 1, t->stids_in_use);
7367 	}
7368 
7369 	if (t->nftids) {
7370 		sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
7371 		    t->ftid_base + t->nftids - 1);
7372 	}
7373 
7374 	if (t->netids) {
7375 		sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
7376 		    t->etid_base + t->netids - 1);
7377 	}
7378 
7379 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
7380 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
7381 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
7382 
7383 	rc = sbuf_finish(sb);
7384 	sbuf_delete(sb);
7385 
7386 	return (rc);
7387 }
7388 
7389 static int
7390 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
7391 {
7392 	struct adapter *sc = arg1;
7393 	struct sbuf *sb;
7394 	int rc;
7395 	struct tp_err_stats stats;
7396 
7397 	rc = sysctl_wire_old_buffer(req, 0);
7398 	if (rc != 0)
7399 		return (rc);
7400 
7401 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7402 	if (sb == NULL)
7403 		return (ENOMEM);
7404 
7405 	mtx_lock(&sc->reg_lock);
7406 	t4_tp_get_err_stats(sc, &stats);
7407 	mtx_unlock(&sc->reg_lock);
7408 
7409 	if (sc->chip_params->nchan > 2) {
7410 		sbuf_printf(sb, "                 channel 0  channel 1"
7411 		    "  channel 2  channel 3\n");
7412 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
7413 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
7414 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
7415 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
7416 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
7417 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
7418 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
7419 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
7420 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
7421 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
7422 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
7423 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
7424 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
7425 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
7426 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
7427 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
7428 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
7429 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
7430 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
7431 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
7432 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
7433 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
7434 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
7435 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
7436 	} else {
7437 		sbuf_printf(sb, "                 channel 0  channel 1\n");
7438 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
7439 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
7440 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
7441 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
7442 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
7443 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
7444 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
7445 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
7446 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
7447 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
7448 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
7449 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
7450 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
7451 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
7452 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
7453 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
7454 	}
7455 
7456 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
7457 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
7458 
7459 	rc = sbuf_finish(sb);
7460 	sbuf_delete(sb);
7461 
7462 	return (rc);
7463 }
7464 
7465 static int
7466 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
7467 {
7468 	struct adapter *sc = arg1;
7469 	struct tp_params *tpp = &sc->params.tp;
7470 	u_int mask;
7471 	int rc;
7472 
7473 	mask = tpp->la_mask >> 16;
7474 	rc = sysctl_handle_int(oidp, &mask, 0, req);
7475 	if (rc != 0 || req->newptr == NULL)
7476 		return (rc);
7477 	if (mask > 0xffff)
7478 		return (EINVAL);
7479 	tpp->la_mask = mask << 16;
7480 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
7481 
7482 	return (0);
7483 }
7484 
7485 struct field_desc {
7486 	const char *name;
7487 	u_int start;
7488 	u_int width;
7489 };
7490 
7491 static void
7492 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
7493 {
7494 	char buf[32];
7495 	int line_size = 0;
7496 
7497 	while (f->name) {
7498 		uint64_t mask = (1ULL << f->width) - 1;
7499 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
7500 		    ((uintmax_t)v >> f->start) & mask);
7501 
7502 		if (line_size + len >= 79) {
7503 			line_size = 8;
7504 			sbuf_printf(sb, "\n        ");
7505 		}
7506 		sbuf_printf(sb, "%s ", buf);
7507 		line_size += len + 1;
7508 		f++;
7509 	}
7510 	sbuf_printf(sb, "\n");
7511 }
7512 
7513 static const struct field_desc tp_la0[] = {
7514 	{ "RcfOpCodeOut", 60, 4 },
7515 	{ "State", 56, 4 },
7516 	{ "WcfState", 52, 4 },
7517 	{ "RcfOpcSrcOut", 50, 2 },
7518 	{ "CRxError", 49, 1 },
7519 	{ "ERxError", 48, 1 },
7520 	{ "SanityFailed", 47, 1 },
7521 	{ "SpuriousMsg", 46, 1 },
7522 	{ "FlushInputMsg", 45, 1 },
7523 	{ "FlushInputCpl", 44, 1 },
7524 	{ "RssUpBit", 43, 1 },
7525 	{ "RssFilterHit", 42, 1 },
7526 	{ "Tid", 32, 10 },
7527 	{ "InitTcb", 31, 1 },
7528 	{ "LineNumber", 24, 7 },
7529 	{ "Emsg", 23, 1 },
7530 	{ "EdataOut", 22, 1 },
7531 	{ "Cmsg", 21, 1 },
7532 	{ "CdataOut", 20, 1 },
7533 	{ "EreadPdu", 19, 1 },
7534 	{ "CreadPdu", 18, 1 },
7535 	{ "TunnelPkt", 17, 1 },
7536 	{ "RcfPeerFin", 16, 1 },
7537 	{ "RcfReasonOut", 12, 4 },
7538 	{ "TxCchannel", 10, 2 },
7539 	{ "RcfTxChannel", 8, 2 },
7540 	{ "RxEchannel", 6, 2 },
7541 	{ "RcfRxChannel", 5, 1 },
7542 	{ "RcfDataOutSrdy", 4, 1 },
7543 	{ "RxDvld", 3, 1 },
7544 	{ "RxOoDvld", 2, 1 },
7545 	{ "RxCongestion", 1, 1 },
7546 	{ "TxCongestion", 0, 1 },
7547 	{ NULL }
7548 };
7549 
7550 static const struct field_desc tp_la1[] = {
7551 	{ "CplCmdIn", 56, 8 },
7552 	{ "CplCmdOut", 48, 8 },
7553 	{ "ESynOut", 47, 1 },
7554 	{ "EAckOut", 46, 1 },
7555 	{ "EFinOut", 45, 1 },
7556 	{ "ERstOut", 44, 1 },
7557 	{ "SynIn", 43, 1 },
7558 	{ "AckIn", 42, 1 },
7559 	{ "FinIn", 41, 1 },
7560 	{ "RstIn", 40, 1 },
7561 	{ "DataIn", 39, 1 },
7562 	{ "DataInVld", 38, 1 },
7563 	{ "PadIn", 37, 1 },
7564 	{ "RxBufEmpty", 36, 1 },
7565 	{ "RxDdp", 35, 1 },
7566 	{ "RxFbCongestion", 34, 1 },
7567 	{ "TxFbCongestion", 33, 1 },
7568 	{ "TxPktSumSrdy", 32, 1 },
7569 	{ "RcfUlpType", 28, 4 },
7570 	{ "Eread", 27, 1 },
7571 	{ "Ebypass", 26, 1 },
7572 	{ "Esave", 25, 1 },
7573 	{ "Static0", 24, 1 },
7574 	{ "Cread", 23, 1 },
7575 	{ "Cbypass", 22, 1 },
7576 	{ "Csave", 21, 1 },
7577 	{ "CPktOut", 20, 1 },
7578 	{ "RxPagePoolFull", 18, 2 },
7579 	{ "RxLpbkPkt", 17, 1 },
7580 	{ "TxLpbkPkt", 16, 1 },
7581 	{ "RxVfValid", 15, 1 },
7582 	{ "SynLearned", 14, 1 },
7583 	{ "SetDelEntry", 13, 1 },
7584 	{ "SetInvEntry", 12, 1 },
7585 	{ "CpcmdDvld", 11, 1 },
7586 	{ "CpcmdSave", 10, 1 },
7587 	{ "RxPstructsFull", 8, 2 },
7588 	{ "EpcmdDvld", 7, 1 },
7589 	{ "EpcmdFlush", 6, 1 },
7590 	{ "EpcmdTrimPrefix", 5, 1 },
7591 	{ "EpcmdTrimPostfix", 4, 1 },
7592 	{ "ERssIp4Pkt", 3, 1 },
7593 	{ "ERssIp6Pkt", 2, 1 },
7594 	{ "ERssTcpUdpPkt", 1, 1 },
7595 	{ "ERssFceFipPkt", 0, 1 },
7596 	{ NULL }
7597 };
7598 
7599 static const struct field_desc tp_la2[] = {
7600 	{ "CplCmdIn", 56, 8 },
7601 	{ "MpsVfVld", 55, 1 },
7602 	{ "MpsPf", 52, 3 },
7603 	{ "MpsVf", 44, 8 },
7604 	{ "SynIn", 43, 1 },
7605 	{ "AckIn", 42, 1 },
7606 	{ "FinIn", 41, 1 },
7607 	{ "RstIn", 40, 1 },
7608 	{ "DataIn", 39, 1 },
7609 	{ "DataInVld", 38, 1 },
7610 	{ "PadIn", 37, 1 },
7611 	{ "RxBufEmpty", 36, 1 },
7612 	{ "RxDdp", 35, 1 },
7613 	{ "RxFbCongestion", 34, 1 },
7614 	{ "TxFbCongestion", 33, 1 },
7615 	{ "TxPktSumSrdy", 32, 1 },
7616 	{ "RcfUlpType", 28, 4 },
7617 	{ "Eread", 27, 1 },
7618 	{ "Ebypass", 26, 1 },
7619 	{ "Esave", 25, 1 },
7620 	{ "Static0", 24, 1 },
7621 	{ "Cread", 23, 1 },
7622 	{ "Cbypass", 22, 1 },
7623 	{ "Csave", 21, 1 },
7624 	{ "CPktOut", 20, 1 },
7625 	{ "RxPagePoolFull", 18, 2 },
7626 	{ "RxLpbkPkt", 17, 1 },
7627 	{ "TxLpbkPkt", 16, 1 },
7628 	{ "RxVfValid", 15, 1 },
7629 	{ "SynLearned", 14, 1 },
7630 	{ "SetDelEntry", 13, 1 },
7631 	{ "SetInvEntry", 12, 1 },
7632 	{ "CpcmdDvld", 11, 1 },
7633 	{ "CpcmdSave", 10, 1 },
7634 	{ "RxPstructsFull", 8, 2 },
7635 	{ "EpcmdDvld", 7, 1 },
7636 	{ "EpcmdFlush", 6, 1 },
7637 	{ "EpcmdTrimPrefix", 5, 1 },
7638 	{ "EpcmdTrimPostfix", 4, 1 },
7639 	{ "ERssIp4Pkt", 3, 1 },
7640 	{ "ERssIp6Pkt", 2, 1 },
7641 	{ "ERssTcpUdpPkt", 1, 1 },
7642 	{ "ERssFceFipPkt", 0, 1 },
7643 	{ NULL }
7644 };
7645 
7646 static void
7647 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
7648 {
7649 
7650 	field_desc_show(sb, *p, tp_la0);
7651 }
7652 
7653 static void
7654 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
7655 {
7656 
7657 	if (idx)
7658 		sbuf_printf(sb, "\n");
7659 	field_desc_show(sb, p[0], tp_la0);
7660 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7661 		field_desc_show(sb, p[1], tp_la0);
7662 }
7663 
7664 static void
7665 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
7666 {
7667 
7668 	if (idx)
7669 		sbuf_printf(sb, "\n");
7670 	field_desc_show(sb, p[0], tp_la0);
7671 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7672 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
7673 }
7674 
7675 static int
7676 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
7677 {
7678 	struct adapter *sc = arg1;
7679 	struct sbuf *sb;
7680 	uint64_t *buf, *p;
7681 	int rc;
7682 	u_int i, inc;
7683 	void (*show_func)(struct sbuf *, uint64_t *, int);
7684 
7685 	rc = sysctl_wire_old_buffer(req, 0);
7686 	if (rc != 0)
7687 		return (rc);
7688 
7689 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7690 	if (sb == NULL)
7691 		return (ENOMEM);
7692 
7693 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
7694 
7695 	t4_tp_read_la(sc, buf, NULL);
7696 	p = buf;
7697 
7698 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
7699 	case 2:
7700 		inc = 2;
7701 		show_func = tp_la_show2;
7702 		break;
7703 	case 3:
7704 		inc = 2;
7705 		show_func = tp_la_show3;
7706 		break;
7707 	default:
7708 		inc = 1;
7709 		show_func = tp_la_show;
7710 	}
7711 
7712 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
7713 		(*show_func)(sb, p, i);
7714 
7715 	rc = sbuf_finish(sb);
7716 	sbuf_delete(sb);
7717 	free(buf, M_CXGBE);
7718 	return (rc);
7719 }
7720 
7721 static int
7722 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
7723 {
7724 	struct adapter *sc = arg1;
7725 	struct sbuf *sb;
7726 	int rc;
7727 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
7728 
7729 	rc = sysctl_wire_old_buffer(req, 0);
7730 	if (rc != 0)
7731 		return (rc);
7732 
7733 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7734 	if (sb == NULL)
7735 		return (ENOMEM);
7736 
7737 	t4_get_chan_txrate(sc, nrate, orate);
7738 
7739 	if (sc->chip_params->nchan > 2) {
7740 		sbuf_printf(sb, "              channel 0   channel 1"
7741 		    "   channel 2   channel 3\n");
7742 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
7743 		    nrate[0], nrate[1], nrate[2], nrate[3]);
7744 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
7745 		    orate[0], orate[1], orate[2], orate[3]);
7746 	} else {
7747 		sbuf_printf(sb, "              channel 0   channel 1\n");
7748 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
7749 		    nrate[0], nrate[1]);
7750 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
7751 		    orate[0], orate[1]);
7752 	}
7753 
7754 	rc = sbuf_finish(sb);
7755 	sbuf_delete(sb);
7756 
7757 	return (rc);
7758 }
7759 
7760 static int
7761 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
7762 {
7763 	struct adapter *sc = arg1;
7764 	struct sbuf *sb;
7765 	uint32_t *buf, *p;
7766 	int rc, i;
7767 
7768 	rc = sysctl_wire_old_buffer(req, 0);
7769 	if (rc != 0)
7770 		return (rc);
7771 
7772 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7773 	if (sb == NULL)
7774 		return (ENOMEM);
7775 
7776 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
7777 	    M_ZERO | M_WAITOK);
7778 
7779 	t4_ulprx_read_la(sc, buf);
7780 	p = buf;
7781 
7782 	sbuf_printf(sb, "      Pcmd        Type   Message"
7783 	    "                Data");
7784 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
7785 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
7786 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
7787 	}
7788 
7789 	rc = sbuf_finish(sb);
7790 	sbuf_delete(sb);
7791 	free(buf, M_CXGBE);
7792 	return (rc);
7793 }
7794 
7795 static int
7796 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
7797 {
7798 	struct adapter *sc = arg1;
7799 	struct sbuf *sb;
7800 	int rc, v;
7801 
7802 	MPASS(chip_id(sc) >= CHELSIO_T5);
7803 
7804 	rc = sysctl_wire_old_buffer(req, 0);
7805 	if (rc != 0)
7806 		return (rc);
7807 
7808 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7809 	if (sb == NULL)
7810 		return (ENOMEM);
7811 
7812 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
7813 	if (G_STATSOURCE_T5(v) == 7) {
7814 		int mode;
7815 
7816 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
7817 		if (mode == 0) {
7818 			sbuf_printf(sb, "total %d, incomplete %d",
7819 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7820 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7821 		} else if (mode == 1) {
7822 			sbuf_printf(sb, "total %d, data overflow %d",
7823 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7824 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7825 		} else {
7826 			sbuf_printf(sb, "unknown mode %d", mode);
7827 		}
7828 	}
7829 	rc = sbuf_finish(sb);
7830 	sbuf_delete(sb);
7831 
7832 	return (rc);
7833 }
7834 
7835 static int
7836 sysctl_tc_params(SYSCTL_HANDLER_ARGS)
7837 {
7838 	struct adapter *sc = arg1;
7839 	struct tx_sched_class *tc;
7840 	struct t4_sched_class_params p;
7841 	struct sbuf *sb;
7842 	int i, rc, port_id, flags, mbps, gbps;
7843 
7844 	rc = sysctl_wire_old_buffer(req, 0);
7845 	if (rc != 0)
7846 		return (rc);
7847 
7848 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7849 	if (sb == NULL)
7850 		return (ENOMEM);
7851 
7852 	port_id = arg2 >> 16;
7853 	MPASS(port_id < sc->params.nports);
7854 	MPASS(sc->port[port_id] != NULL);
7855 	i = arg2 & 0xffff;
7856 	MPASS(i < sc->chip_params->nsched_cls);
7857 	tc = &sc->port[port_id]->tc[i];
7858 
7859 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7860 	    "t4tc_p");
7861 	if (rc)
7862 		goto done;
7863 	flags = tc->flags;
7864 	p = tc->params;
7865 	end_synchronized_op(sc, LOCK_HELD);
7866 
7867 	if ((flags & TX_SC_OK) == 0) {
7868 		sbuf_printf(sb, "none");
7869 		goto done;
7870 	}
7871 
7872 	if (p.level == SCHED_CLASS_LEVEL_CL_WRR) {
7873 		sbuf_printf(sb, "cl-wrr weight %u", p.weight);
7874 		goto done;
7875 	} else if (p.level == SCHED_CLASS_LEVEL_CL_RL)
7876 		sbuf_printf(sb, "cl-rl");
7877 	else if (p.level == SCHED_CLASS_LEVEL_CH_RL)
7878 		sbuf_printf(sb, "ch-rl");
7879 	else {
7880 		rc = ENXIO;
7881 		goto done;
7882 	}
7883 
7884 	if (p.ratemode == SCHED_CLASS_RATEMODE_REL) {
7885 		/* XXX: top speed or actual link speed? */
7886 		gbps = port_top_speed(sc->port[port_id]);
7887 		sbuf_printf(sb, " %u%% of %uGbps", p.maxrate, gbps);
7888 	}
7889 	else if (p.ratemode == SCHED_CLASS_RATEMODE_ABS) {
7890 		switch (p.rateunit) {
7891 		case SCHED_CLASS_RATEUNIT_BITS:
7892 			mbps = p.maxrate / 1000;
7893 			gbps = p.maxrate / 1000000;
7894 			if (p.maxrate == gbps * 1000000)
7895 				sbuf_printf(sb, " %uGbps", gbps);
7896 			else if (p.maxrate == mbps * 1000)
7897 				sbuf_printf(sb, " %uMbps", mbps);
7898 			else
7899 				sbuf_printf(sb, " %uKbps", p.maxrate);
7900 			break;
7901 		case SCHED_CLASS_RATEUNIT_PKTS:
7902 			sbuf_printf(sb, " %upps", p.maxrate);
7903 			break;
7904 		default:
7905 			rc = ENXIO;
7906 			goto done;
7907 		}
7908 	}
7909 
7910 	switch (p.mode) {
7911 	case SCHED_CLASS_MODE_CLASS:
7912 		sbuf_printf(sb, " aggregate");
7913 		break;
7914 	case SCHED_CLASS_MODE_FLOW:
7915 		sbuf_printf(sb, " per-flow");
7916 		break;
7917 	default:
7918 		rc = ENXIO;
7919 		goto done;
7920 	}
7921 
7922 done:
7923 	if (rc == 0)
7924 		rc = sbuf_finish(sb);
7925 	sbuf_delete(sb);
7926 
7927 	return (rc);
7928 }
7929 #endif
7930 
7931 #ifdef TCP_OFFLOAD
7932 static void
7933 unit_conv(char *buf, size_t len, u_int val, u_int factor)
7934 {
7935 	u_int rem = val % factor;
7936 
7937 	if (rem == 0)
7938 		snprintf(buf, len, "%u", val / factor);
7939 	else {
7940 		while (rem % 10 == 0)
7941 			rem /= 10;
7942 		snprintf(buf, len, "%u.%u", val / factor, rem);
7943 	}
7944 }
7945 
7946 static int
7947 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
7948 {
7949 	struct adapter *sc = arg1;
7950 	char buf[16];
7951 	u_int res, re;
7952 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7953 
7954 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
7955 	switch (arg2) {
7956 	case 0:
7957 		/* timer_tick */
7958 		re = G_TIMERRESOLUTION(res);
7959 		break;
7960 	case 1:
7961 		/* TCP timestamp tick */
7962 		re = G_TIMESTAMPRESOLUTION(res);
7963 		break;
7964 	case 2:
7965 		/* DACK tick */
7966 		re = G_DELAYEDACKRESOLUTION(res);
7967 		break;
7968 	default:
7969 		return (EDOOFUS);
7970 	}
7971 
7972 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
7973 
7974 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
7975 }
7976 
7977 static int
7978 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
7979 {
7980 	struct adapter *sc = arg1;
7981 	u_int res, dack_re, v;
7982 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7983 
7984 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
7985 	dack_re = G_DELAYEDACKRESOLUTION(res);
7986 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
7987 
7988 	return (sysctl_handle_int(oidp, &v, 0, req));
7989 }
7990 
7991 static int
7992 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
7993 {
7994 	struct adapter *sc = arg1;
7995 	int reg = arg2;
7996 	u_int tre;
7997 	u_long tp_tick_us, v;
7998 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7999 
8000 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
8001 	    reg == A_TP_PERS_MIN || reg == A_TP_PERS_MAX ||
8002 	    reg == A_TP_KEEP_IDLE || A_TP_KEEP_INTVL || reg == A_TP_INIT_SRTT ||
8003 	    reg == A_TP_FINWAIT2_TIMER);
8004 
8005 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
8006 	tp_tick_us = (cclk_ps << tre) / 1000000;
8007 
8008 	if (reg == A_TP_INIT_SRTT)
8009 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
8010 	else
8011 		v = tp_tick_us * t4_read_reg(sc, reg);
8012 
8013 	return (sysctl_handle_long(oidp, &v, 0, req));
8014 }
8015 #endif
8016 
8017 static uint32_t
8018 fconf_iconf_to_mode(uint32_t fconf, uint32_t iconf)
8019 {
8020 	uint32_t mode;
8021 
8022 	mode = T4_FILTER_IPv4 | T4_FILTER_IPv6 | T4_FILTER_IP_SADDR |
8023 	    T4_FILTER_IP_DADDR | T4_FILTER_IP_SPORT | T4_FILTER_IP_DPORT;
8024 
8025 	if (fconf & F_FRAGMENTATION)
8026 		mode |= T4_FILTER_IP_FRAGMENT;
8027 
8028 	if (fconf & F_MPSHITTYPE)
8029 		mode |= T4_FILTER_MPS_HIT_TYPE;
8030 
8031 	if (fconf & F_MACMATCH)
8032 		mode |= T4_FILTER_MAC_IDX;
8033 
8034 	if (fconf & F_ETHERTYPE)
8035 		mode |= T4_FILTER_ETH_TYPE;
8036 
8037 	if (fconf & F_PROTOCOL)
8038 		mode |= T4_FILTER_IP_PROTO;
8039 
8040 	if (fconf & F_TOS)
8041 		mode |= T4_FILTER_IP_TOS;
8042 
8043 	if (fconf & F_VLAN)
8044 		mode |= T4_FILTER_VLAN;
8045 
8046 	if (fconf & F_VNIC_ID) {
8047 		mode |= T4_FILTER_VNIC;
8048 		if (iconf & F_VNIC)
8049 			mode |= T4_FILTER_IC_VNIC;
8050 	}
8051 
8052 	if (fconf & F_PORT)
8053 		mode |= T4_FILTER_PORT;
8054 
8055 	if (fconf & F_FCOE)
8056 		mode |= T4_FILTER_FCoE;
8057 
8058 	return (mode);
8059 }
8060 
8061 static uint32_t
8062 mode_to_fconf(uint32_t mode)
8063 {
8064 	uint32_t fconf = 0;
8065 
8066 	if (mode & T4_FILTER_IP_FRAGMENT)
8067 		fconf |= F_FRAGMENTATION;
8068 
8069 	if (mode & T4_FILTER_MPS_HIT_TYPE)
8070 		fconf |= F_MPSHITTYPE;
8071 
8072 	if (mode & T4_FILTER_MAC_IDX)
8073 		fconf |= F_MACMATCH;
8074 
8075 	if (mode & T4_FILTER_ETH_TYPE)
8076 		fconf |= F_ETHERTYPE;
8077 
8078 	if (mode & T4_FILTER_IP_PROTO)
8079 		fconf |= F_PROTOCOL;
8080 
8081 	if (mode & T4_FILTER_IP_TOS)
8082 		fconf |= F_TOS;
8083 
8084 	if (mode & T4_FILTER_VLAN)
8085 		fconf |= F_VLAN;
8086 
8087 	if (mode & T4_FILTER_VNIC)
8088 		fconf |= F_VNIC_ID;
8089 
8090 	if (mode & T4_FILTER_PORT)
8091 		fconf |= F_PORT;
8092 
8093 	if (mode & T4_FILTER_FCoE)
8094 		fconf |= F_FCOE;
8095 
8096 	return (fconf);
8097 }
8098 
8099 static uint32_t
8100 mode_to_iconf(uint32_t mode)
8101 {
8102 
8103 	if (mode & T4_FILTER_IC_VNIC)
8104 		return (F_VNIC);
8105 	return (0);
8106 }
8107 
8108 static int check_fspec_against_fconf_iconf(struct adapter *sc,
8109     struct t4_filter_specification *fs)
8110 {
8111 	struct tp_params *tpp = &sc->params.tp;
8112 	uint32_t fconf = 0;
8113 
8114 	if (fs->val.frag || fs->mask.frag)
8115 		fconf |= F_FRAGMENTATION;
8116 
8117 	if (fs->val.matchtype || fs->mask.matchtype)
8118 		fconf |= F_MPSHITTYPE;
8119 
8120 	if (fs->val.macidx || fs->mask.macidx)
8121 		fconf |= F_MACMATCH;
8122 
8123 	if (fs->val.ethtype || fs->mask.ethtype)
8124 		fconf |= F_ETHERTYPE;
8125 
8126 	if (fs->val.proto || fs->mask.proto)
8127 		fconf |= F_PROTOCOL;
8128 
8129 	if (fs->val.tos || fs->mask.tos)
8130 		fconf |= F_TOS;
8131 
8132 	if (fs->val.vlan_vld || fs->mask.vlan_vld)
8133 		fconf |= F_VLAN;
8134 
8135 	if (fs->val.ovlan_vld || fs->mask.ovlan_vld) {
8136 		fconf |= F_VNIC_ID;
8137 		if (tpp->ingress_config & F_VNIC)
8138 			return (EINVAL);
8139 	}
8140 
8141 	if (fs->val.pfvf_vld || fs->mask.pfvf_vld) {
8142 		fconf |= F_VNIC_ID;
8143 		if ((tpp->ingress_config & F_VNIC) == 0)
8144 			return (EINVAL);
8145 	}
8146 
8147 	if (fs->val.iport || fs->mask.iport)
8148 		fconf |= F_PORT;
8149 
8150 	if (fs->val.fcoe || fs->mask.fcoe)
8151 		fconf |= F_FCOE;
8152 
8153 	if ((tpp->vlan_pri_map | fconf) != tpp->vlan_pri_map)
8154 		return (E2BIG);
8155 
8156 	return (0);
8157 }
8158 
8159 static int
8160 get_filter_mode(struct adapter *sc, uint32_t *mode)
8161 {
8162 	struct tp_params *tpp = &sc->params.tp;
8163 
8164 	/*
8165 	 * We trust the cached values of the relevant TP registers.  This means
8166 	 * things work reliably only if writes to those registers are always via
8167 	 * t4_set_filter_mode.
8168 	 */
8169 	*mode = fconf_iconf_to_mode(tpp->vlan_pri_map, tpp->ingress_config);
8170 
8171 	return (0);
8172 }
8173 
8174 static int
8175 set_filter_mode(struct adapter *sc, uint32_t mode)
8176 {
8177 	struct tp_params *tpp = &sc->params.tp;
8178 	uint32_t fconf, iconf;
8179 	int rc;
8180 
8181 	iconf = mode_to_iconf(mode);
8182 	if ((iconf ^ tpp->ingress_config) & F_VNIC) {
8183 		/*
8184 		 * For now we just complain if A_TP_INGRESS_CONFIG is not
8185 		 * already set to the correct value for the requested filter
8186 		 * mode.  It's not clear if it's safe to write to this register
8187 		 * on the fly.  (And we trust the cached value of the register).
8188 		 */
8189 		return (EBUSY);
8190 	}
8191 
8192 	fconf = mode_to_fconf(mode);
8193 
8194 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8195 	    "t4setfm");
8196 	if (rc)
8197 		return (rc);
8198 
8199 	if (sc->tids.ftids_in_use > 0) {
8200 		rc = EBUSY;
8201 		goto done;
8202 	}
8203 
8204 #ifdef TCP_OFFLOAD
8205 	if (uld_active(sc, ULD_TOM)) {
8206 		rc = EBUSY;
8207 		goto done;
8208 	}
8209 #endif
8210 
8211 	rc = -t4_set_filter_mode(sc, fconf);
8212 done:
8213 	end_synchronized_op(sc, LOCK_HELD);
8214 	return (rc);
8215 }
8216 
8217 static inline uint64_t
8218 get_filter_hits(struct adapter *sc, uint32_t fid)
8219 {
8220 	uint32_t tcb_addr;
8221 
8222 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE) +
8223 	    (fid + sc->tids.ftid_base) * TCB_SIZE;
8224 
8225 	if (is_t4(sc)) {
8226 		uint64_t hits;
8227 
8228 		read_via_memwin(sc, 0, tcb_addr + 16, (uint32_t *)&hits, 8);
8229 		return (be64toh(hits));
8230 	} else {
8231 		uint32_t hits;
8232 
8233 		read_via_memwin(sc, 0, tcb_addr + 24, &hits, 4);
8234 		return (be32toh(hits));
8235 	}
8236 }
8237 
8238 static int
8239 get_filter(struct adapter *sc, struct t4_filter *t)
8240 {
8241 	int i, rc, nfilters = sc->tids.nftids;
8242 	struct filter_entry *f;
8243 
8244 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8245 	    "t4getf");
8246 	if (rc)
8247 		return (rc);
8248 
8249 	if (sc->tids.ftids_in_use == 0 || sc->tids.ftid_tab == NULL ||
8250 	    t->idx >= nfilters) {
8251 		t->idx = 0xffffffff;
8252 		goto done;
8253 	}
8254 
8255 	f = &sc->tids.ftid_tab[t->idx];
8256 	for (i = t->idx; i < nfilters; i++, f++) {
8257 		if (f->valid) {
8258 			t->idx = i;
8259 			t->l2tidx = f->l2t ? f->l2t->idx : 0;
8260 			t->smtidx = f->smtidx;
8261 			if (f->fs.hitcnts)
8262 				t->hits = get_filter_hits(sc, t->idx);
8263 			else
8264 				t->hits = UINT64_MAX;
8265 			t->fs = f->fs;
8266 
8267 			goto done;
8268 		}
8269 	}
8270 
8271 	t->idx = 0xffffffff;
8272 done:
8273 	end_synchronized_op(sc, LOCK_HELD);
8274 	return (0);
8275 }
8276 
8277 static int
8278 set_filter(struct adapter *sc, struct t4_filter *t)
8279 {
8280 	unsigned int nfilters, nports;
8281 	struct filter_entry *f;
8282 	int i, rc;
8283 
8284 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setf");
8285 	if (rc)
8286 		return (rc);
8287 
8288 	nfilters = sc->tids.nftids;
8289 	nports = sc->params.nports;
8290 
8291 	if (nfilters == 0) {
8292 		rc = ENOTSUP;
8293 		goto done;
8294 	}
8295 
8296 	if (t->idx >= nfilters) {
8297 		rc = EINVAL;
8298 		goto done;
8299 	}
8300 
8301 	/* Validate against the global filter mode and ingress config */
8302 	rc = check_fspec_against_fconf_iconf(sc, &t->fs);
8303 	if (rc != 0)
8304 		goto done;
8305 
8306 	if (t->fs.action == FILTER_SWITCH && t->fs.eport >= nports) {
8307 		rc = EINVAL;
8308 		goto done;
8309 	}
8310 
8311 	if (t->fs.val.iport >= nports) {
8312 		rc = EINVAL;
8313 		goto done;
8314 	}
8315 
8316 	/* Can't specify an iq if not steering to it */
8317 	if (!t->fs.dirsteer && t->fs.iq) {
8318 		rc = EINVAL;
8319 		goto done;
8320 	}
8321 
8322 	/* IPv6 filter idx must be 4 aligned */
8323 	if (t->fs.type == 1 &&
8324 	    ((t->idx & 0x3) || t->idx + 4 >= nfilters)) {
8325 		rc = EINVAL;
8326 		goto done;
8327 	}
8328 
8329 	if (!(sc->flags & FULL_INIT_DONE) &&
8330 	    ((rc = adapter_full_init(sc)) != 0))
8331 		goto done;
8332 
8333 	if (sc->tids.ftid_tab == NULL) {
8334 		KASSERT(sc->tids.ftids_in_use == 0,
8335 		    ("%s: no memory allocated but filters_in_use > 0",
8336 		    __func__));
8337 
8338 		sc->tids.ftid_tab = malloc(sizeof (struct filter_entry) *
8339 		    nfilters, M_CXGBE, M_NOWAIT | M_ZERO);
8340 		if (sc->tids.ftid_tab == NULL) {
8341 			rc = ENOMEM;
8342 			goto done;
8343 		}
8344 		mtx_init(&sc->tids.ftid_lock, "T4 filters", 0, MTX_DEF);
8345 	}
8346 
8347 	for (i = 0; i < 4; i++) {
8348 		f = &sc->tids.ftid_tab[t->idx + i];
8349 
8350 		if (f->pending || f->valid) {
8351 			rc = EBUSY;
8352 			goto done;
8353 		}
8354 		if (f->locked) {
8355 			rc = EPERM;
8356 			goto done;
8357 		}
8358 
8359 		if (t->fs.type == 0)
8360 			break;
8361 	}
8362 
8363 	f = &sc->tids.ftid_tab[t->idx];
8364 	f->fs = t->fs;
8365 
8366 	rc = set_filter_wr(sc, t->idx);
8367 done:
8368 	end_synchronized_op(sc, 0);
8369 
8370 	if (rc == 0) {
8371 		mtx_lock(&sc->tids.ftid_lock);
8372 		for (;;) {
8373 			if (f->pending == 0) {
8374 				rc = f->valid ? 0 : EIO;
8375 				break;
8376 			}
8377 
8378 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8379 			    PCATCH, "t4setfw", 0)) {
8380 				rc = EINPROGRESS;
8381 				break;
8382 			}
8383 		}
8384 		mtx_unlock(&sc->tids.ftid_lock);
8385 	}
8386 	return (rc);
8387 }
8388 
8389 static int
8390 del_filter(struct adapter *sc, struct t4_filter *t)
8391 {
8392 	unsigned int nfilters;
8393 	struct filter_entry *f;
8394 	int rc;
8395 
8396 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4delf");
8397 	if (rc)
8398 		return (rc);
8399 
8400 	nfilters = sc->tids.nftids;
8401 
8402 	if (nfilters == 0) {
8403 		rc = ENOTSUP;
8404 		goto done;
8405 	}
8406 
8407 	if (sc->tids.ftid_tab == NULL || sc->tids.ftids_in_use == 0 ||
8408 	    t->idx >= nfilters) {
8409 		rc = EINVAL;
8410 		goto done;
8411 	}
8412 
8413 	if (!(sc->flags & FULL_INIT_DONE)) {
8414 		rc = EAGAIN;
8415 		goto done;
8416 	}
8417 
8418 	f = &sc->tids.ftid_tab[t->idx];
8419 
8420 	if (f->pending) {
8421 		rc = EBUSY;
8422 		goto done;
8423 	}
8424 	if (f->locked) {
8425 		rc = EPERM;
8426 		goto done;
8427 	}
8428 
8429 	if (f->valid) {
8430 		t->fs = f->fs;	/* extra info for the caller */
8431 		rc = del_filter_wr(sc, t->idx);
8432 	}
8433 
8434 done:
8435 	end_synchronized_op(sc, 0);
8436 
8437 	if (rc == 0) {
8438 		mtx_lock(&sc->tids.ftid_lock);
8439 		for (;;) {
8440 			if (f->pending == 0) {
8441 				rc = f->valid ? EIO : 0;
8442 				break;
8443 			}
8444 
8445 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8446 			    PCATCH, "t4delfw", 0)) {
8447 				rc = EINPROGRESS;
8448 				break;
8449 			}
8450 		}
8451 		mtx_unlock(&sc->tids.ftid_lock);
8452 	}
8453 
8454 	return (rc);
8455 }
8456 
8457 static void
8458 clear_filter(struct filter_entry *f)
8459 {
8460 	if (f->l2t)
8461 		t4_l2t_release(f->l2t);
8462 
8463 	bzero(f, sizeof (*f));
8464 }
8465 
8466 static int
8467 set_filter_wr(struct adapter *sc, int fidx)
8468 {
8469 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8470 	struct fw_filter_wr *fwr;
8471 	unsigned int ftid, vnic_vld, vnic_vld_mask;
8472 	struct wrq_cookie cookie;
8473 
8474 	ASSERT_SYNCHRONIZED_OP(sc);
8475 
8476 	if (f->fs.newdmac || f->fs.newvlan) {
8477 		/* This filter needs an L2T entry; allocate one. */
8478 		f->l2t = t4_l2t_alloc_switching(sc->l2t);
8479 		if (f->l2t == NULL)
8480 			return (EAGAIN);
8481 		if (t4_l2t_set_switching(sc, f->l2t, f->fs.vlan, f->fs.eport,
8482 		    f->fs.dmac)) {
8483 			t4_l2t_release(f->l2t);
8484 			f->l2t = NULL;
8485 			return (ENOMEM);
8486 		}
8487 	}
8488 
8489 	/* Already validated against fconf, iconf */
8490 	MPASS((f->fs.val.pfvf_vld & f->fs.val.ovlan_vld) == 0);
8491 	MPASS((f->fs.mask.pfvf_vld & f->fs.mask.ovlan_vld) == 0);
8492 	if (f->fs.val.pfvf_vld || f->fs.val.ovlan_vld)
8493 		vnic_vld = 1;
8494 	else
8495 		vnic_vld = 0;
8496 	if (f->fs.mask.pfvf_vld || f->fs.mask.ovlan_vld)
8497 		vnic_vld_mask = 1;
8498 	else
8499 		vnic_vld_mask = 0;
8500 
8501 	ftid = sc->tids.ftid_base + fidx;
8502 
8503 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8504 	if (fwr == NULL)
8505 		return (ENOMEM);
8506 	bzero(fwr, sizeof(*fwr));
8507 
8508 	fwr->op_pkd = htobe32(V_FW_WR_OP(FW_FILTER_WR));
8509 	fwr->len16_pkd = htobe32(FW_LEN16(*fwr));
8510 	fwr->tid_to_iq =
8511 	    htobe32(V_FW_FILTER_WR_TID(ftid) |
8512 		V_FW_FILTER_WR_RQTYPE(f->fs.type) |
8513 		V_FW_FILTER_WR_NOREPLY(0) |
8514 		V_FW_FILTER_WR_IQ(f->fs.iq));
8515 	fwr->del_filter_to_l2tix =
8516 	    htobe32(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
8517 		V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
8518 		V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
8519 		V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
8520 		V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
8521 		V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
8522 		V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
8523 		V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
8524 		V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
8525 		    f->fs.newvlan == VLAN_REWRITE) |
8526 		V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
8527 		    f->fs.newvlan == VLAN_REWRITE) |
8528 		V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
8529 		V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
8530 		V_FW_FILTER_WR_PRIO(f->fs.prio) |
8531 		V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
8532 	fwr->ethtype = htobe16(f->fs.val.ethtype);
8533 	fwr->ethtypem = htobe16(f->fs.mask.ethtype);
8534 	fwr->frag_to_ovlan_vldm =
8535 	    (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
8536 		V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
8537 		V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.vlan_vld) |
8538 		V_FW_FILTER_WR_OVLAN_VLD(vnic_vld) |
8539 		V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.vlan_vld) |
8540 		V_FW_FILTER_WR_OVLAN_VLDM(vnic_vld_mask));
8541 	fwr->smac_sel = 0;
8542 	fwr->rx_chan_rx_rpl_iq = htobe16(V_FW_FILTER_WR_RX_CHAN(0) |
8543 	    V_FW_FILTER_WR_RX_RPL_IQ(sc->sge.fwq.abs_id));
8544 	fwr->maci_to_matchtypem =
8545 	    htobe32(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
8546 		V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
8547 		V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
8548 		V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
8549 		V_FW_FILTER_WR_PORT(f->fs.val.iport) |
8550 		V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
8551 		V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
8552 		V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
8553 	fwr->ptcl = f->fs.val.proto;
8554 	fwr->ptclm = f->fs.mask.proto;
8555 	fwr->ttyp = f->fs.val.tos;
8556 	fwr->ttypm = f->fs.mask.tos;
8557 	fwr->ivlan = htobe16(f->fs.val.vlan);
8558 	fwr->ivlanm = htobe16(f->fs.mask.vlan);
8559 	fwr->ovlan = htobe16(f->fs.val.vnic);
8560 	fwr->ovlanm = htobe16(f->fs.mask.vnic);
8561 	bcopy(f->fs.val.dip, fwr->lip, sizeof (fwr->lip));
8562 	bcopy(f->fs.mask.dip, fwr->lipm, sizeof (fwr->lipm));
8563 	bcopy(f->fs.val.sip, fwr->fip, sizeof (fwr->fip));
8564 	bcopy(f->fs.mask.sip, fwr->fipm, sizeof (fwr->fipm));
8565 	fwr->lp = htobe16(f->fs.val.dport);
8566 	fwr->lpm = htobe16(f->fs.mask.dport);
8567 	fwr->fp = htobe16(f->fs.val.sport);
8568 	fwr->fpm = htobe16(f->fs.mask.sport);
8569 	if (f->fs.newsmac)
8570 		bcopy(f->fs.smac, fwr->sma, sizeof (fwr->sma));
8571 
8572 	f->pending = 1;
8573 	sc->tids.ftids_in_use++;
8574 
8575 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8576 	return (0);
8577 }
8578 
8579 static int
8580 del_filter_wr(struct adapter *sc, int fidx)
8581 {
8582 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8583 	struct fw_filter_wr *fwr;
8584 	unsigned int ftid;
8585 	struct wrq_cookie cookie;
8586 
8587 	ftid = sc->tids.ftid_base + fidx;
8588 
8589 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8590 	if (fwr == NULL)
8591 		return (ENOMEM);
8592 	bzero(fwr, sizeof (*fwr));
8593 
8594 	t4_mk_filtdelwr(ftid, fwr, sc->sge.fwq.abs_id);
8595 
8596 	f->pending = 1;
8597 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8598 	return (0);
8599 }
8600 
8601 int
8602 t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8603 {
8604 	struct adapter *sc = iq->adapter;
8605 	const struct cpl_set_tcb_rpl *rpl = (const void *)(rss + 1);
8606 	unsigned int idx = GET_TID(rpl);
8607 	unsigned int rc;
8608 	struct filter_entry *f;
8609 
8610 	KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
8611 	    rss->opcode));
8612 	MPASS(iq == &sc->sge.fwq);
8613 	MPASS(is_ftid(sc, idx));
8614 
8615 	idx -= sc->tids.ftid_base;
8616 	f = &sc->tids.ftid_tab[idx];
8617 	rc = G_COOKIE(rpl->cookie);
8618 
8619 	mtx_lock(&sc->tids.ftid_lock);
8620 	if (rc == FW_FILTER_WR_FLT_ADDED) {
8621 		KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
8622 		    __func__, idx));
8623 		f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
8624 		f->pending = 0;  /* asynchronous setup completed */
8625 		f->valid = 1;
8626 	} else {
8627 		if (rc != FW_FILTER_WR_FLT_DELETED) {
8628 			/* Add or delete failed, display an error */
8629 			log(LOG_ERR,
8630 			    "filter %u setup failed with error %u\n",
8631 			    idx, rc);
8632 		}
8633 
8634 		clear_filter(f);
8635 		sc->tids.ftids_in_use--;
8636 	}
8637 	wakeup(&sc->tids.ftid_tab);
8638 	mtx_unlock(&sc->tids.ftid_lock);
8639 
8640 	return (0);
8641 }
8642 
8643 static int
8644 set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8645 {
8646 
8647 	MPASS(iq->set_tcb_rpl != NULL);
8648 	return (iq->set_tcb_rpl(iq, rss, m));
8649 }
8650 
8651 static int
8652 l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8653 {
8654 
8655 	MPASS(iq->l2t_write_rpl != NULL);
8656 	return (iq->l2t_write_rpl(iq, rss, m));
8657 }
8658 
8659 static int
8660 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
8661 {
8662 	int rc;
8663 
8664 	if (cntxt->cid > M_CTXTQID)
8665 		return (EINVAL);
8666 
8667 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
8668 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
8669 		return (EINVAL);
8670 
8671 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
8672 	if (rc)
8673 		return (rc);
8674 
8675 	if (sc->flags & FW_OK) {
8676 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
8677 		    &cntxt->data[0]);
8678 		if (rc == 0)
8679 			goto done;
8680 	}
8681 
8682 	/*
8683 	 * Read via firmware failed or wasn't even attempted.  Read directly via
8684 	 * the backdoor.
8685 	 */
8686 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
8687 done:
8688 	end_synchronized_op(sc, 0);
8689 	return (rc);
8690 }
8691 
8692 static int
8693 load_fw(struct adapter *sc, struct t4_data *fw)
8694 {
8695 	int rc;
8696 	uint8_t *fw_data;
8697 
8698 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
8699 	if (rc)
8700 		return (rc);
8701 
8702 	if (sc->flags & FULL_INIT_DONE) {
8703 		rc = EBUSY;
8704 		goto done;
8705 	}
8706 
8707 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
8708 	if (fw_data == NULL) {
8709 		rc = ENOMEM;
8710 		goto done;
8711 	}
8712 
8713 	rc = copyin(fw->data, fw_data, fw->len);
8714 	if (rc == 0)
8715 		rc = -t4_load_fw(sc, fw_data, fw->len);
8716 
8717 	free(fw_data, M_CXGBE);
8718 done:
8719 	end_synchronized_op(sc, 0);
8720 	return (rc);
8721 }
8722 
8723 static int
8724 load_cfg(struct adapter *sc, struct t4_data *cfg)
8725 {
8726 	int rc;
8727 	uint8_t *cfg_data = NULL;
8728 
8729 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
8730 	if (rc)
8731 		return (rc);
8732 
8733 	if (cfg->len == 0) {
8734 		/* clear */
8735 		rc = -t4_load_cfg(sc, NULL, 0);
8736 		goto done;
8737 	}
8738 
8739 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
8740 	if (cfg_data == NULL) {
8741 		rc = ENOMEM;
8742 		goto done;
8743 	}
8744 
8745 	rc = copyin(cfg->data, cfg_data, cfg->len);
8746 	if (rc == 0)
8747 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
8748 
8749 	free(cfg_data, M_CXGBE);
8750 done:
8751 	end_synchronized_op(sc, 0);
8752 	return (rc);
8753 }
8754 
8755 #define MAX_READ_BUF_SIZE (128 * 1024)
8756 static int
8757 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
8758 {
8759 	uint32_t addr, remaining, n;
8760 	uint32_t *buf;
8761 	int rc;
8762 	uint8_t *dst;
8763 
8764 	rc = validate_mem_range(sc, mr->addr, mr->len);
8765 	if (rc != 0)
8766 		return (rc);
8767 
8768 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
8769 	addr = mr->addr;
8770 	remaining = mr->len;
8771 	dst = (void *)mr->data;
8772 
8773 	while (remaining) {
8774 		n = min(remaining, MAX_READ_BUF_SIZE);
8775 		read_via_memwin(sc, 2, addr, buf, n);
8776 
8777 		rc = copyout(buf, dst, n);
8778 		if (rc != 0)
8779 			break;
8780 
8781 		dst += n;
8782 		remaining -= n;
8783 		addr += n;
8784 	}
8785 
8786 	free(buf, M_CXGBE);
8787 	return (rc);
8788 }
8789 #undef MAX_READ_BUF_SIZE
8790 
8791 static int
8792 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
8793 {
8794 	int rc;
8795 
8796 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
8797 		return (EINVAL);
8798 
8799 	if (i2cd->len > sizeof(i2cd->data))
8800 		return (EFBIG);
8801 
8802 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
8803 	if (rc)
8804 		return (rc);
8805 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
8806 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
8807 	end_synchronized_op(sc, 0);
8808 
8809 	return (rc);
8810 }
8811 
8812 static int
8813 in_range(int val, int lo, int hi)
8814 {
8815 
8816 	return (val < 0 || (val <= hi && val >= lo));
8817 }
8818 
8819 static int
8820 set_sched_class_config(struct adapter *sc, int minmax)
8821 {
8822 	int rc;
8823 
8824 	if (minmax < 0)
8825 		return (EINVAL);
8826 
8827 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4sscc");
8828 	if (rc)
8829 		return (rc);
8830 	rc = -t4_sched_config(sc, FW_SCHED_TYPE_PKTSCHED, minmax, 1);
8831 	end_synchronized_op(sc, 0);
8832 
8833 	return (rc);
8834 }
8835 
8836 static int
8837 set_sched_class_params(struct adapter *sc, struct t4_sched_class_params *p,
8838     int sleep_ok)
8839 {
8840 	int rc, top_speed, fw_level, fw_mode, fw_rateunit, fw_ratemode;
8841 	struct port_info *pi;
8842 	struct tx_sched_class *tc;
8843 
8844 	if (p->level == SCHED_CLASS_LEVEL_CL_RL)
8845 		fw_level = FW_SCHED_PARAMS_LEVEL_CL_RL;
8846 	else if (p->level == SCHED_CLASS_LEVEL_CL_WRR)
8847 		fw_level = FW_SCHED_PARAMS_LEVEL_CL_WRR;
8848 	else if (p->level == SCHED_CLASS_LEVEL_CH_RL)
8849 		fw_level = FW_SCHED_PARAMS_LEVEL_CH_RL;
8850 	else
8851 		return (EINVAL);
8852 
8853 	if (p->mode == SCHED_CLASS_MODE_CLASS)
8854 		fw_mode = FW_SCHED_PARAMS_MODE_CLASS;
8855 	else if (p->mode == SCHED_CLASS_MODE_FLOW)
8856 		fw_mode = FW_SCHED_PARAMS_MODE_FLOW;
8857 	else
8858 		return (EINVAL);
8859 
8860 	if (p->rateunit == SCHED_CLASS_RATEUNIT_BITS)
8861 		fw_rateunit = FW_SCHED_PARAMS_UNIT_BITRATE;
8862 	else if (p->rateunit == SCHED_CLASS_RATEUNIT_PKTS)
8863 		fw_rateunit = FW_SCHED_PARAMS_UNIT_PKTRATE;
8864 	else
8865 		return (EINVAL);
8866 
8867 	if (p->ratemode == SCHED_CLASS_RATEMODE_REL)
8868 		fw_ratemode = FW_SCHED_PARAMS_RATE_REL;
8869 	else if (p->ratemode == SCHED_CLASS_RATEMODE_ABS)
8870 		fw_ratemode = FW_SCHED_PARAMS_RATE_ABS;
8871 	else
8872 		return (EINVAL);
8873 
8874 	/* Vet our parameters ... */
8875 	if (!in_range(p->channel, 0, sc->chip_params->nchan - 1))
8876 		return (ERANGE);
8877 
8878 	pi = sc->port[sc->chan_map[p->channel]];
8879 	if (pi == NULL)
8880 		return (ENXIO);
8881 	MPASS(pi->tx_chan == p->channel);
8882 	top_speed = port_top_speed(pi) * 1000000; /* Gbps -> Kbps */
8883 
8884 	if (!in_range(p->cl, 0, sc->chip_params->nsched_cls) ||
8885 	    !in_range(p->minrate, 0, top_speed) ||
8886 	    !in_range(p->maxrate, 0, top_speed) ||
8887 	    !in_range(p->weight, 0, 100))
8888 		return (ERANGE);
8889 
8890 	/*
8891 	 * Translate any unset parameters into the firmware's
8892 	 * nomenclature and/or fail the call if the parameters
8893 	 * are required ...
8894 	 */
8895 	if (p->rateunit < 0 || p->ratemode < 0 || p->channel < 0 || p->cl < 0)
8896 		return (EINVAL);
8897 
8898 	if (p->minrate < 0)
8899 		p->minrate = 0;
8900 	if (p->maxrate < 0) {
8901 		if (p->level == SCHED_CLASS_LEVEL_CL_RL ||
8902 		    p->level == SCHED_CLASS_LEVEL_CH_RL)
8903 			return (EINVAL);
8904 		else
8905 			p->maxrate = 0;
8906 	}
8907 	if (p->weight < 0) {
8908 		if (p->level == SCHED_CLASS_LEVEL_CL_WRR)
8909 			return (EINVAL);
8910 		else
8911 			p->weight = 0;
8912 	}
8913 	if (p->pktsize < 0) {
8914 		if (p->level == SCHED_CLASS_LEVEL_CL_RL ||
8915 		    p->level == SCHED_CLASS_LEVEL_CH_RL)
8916 			return (EINVAL);
8917 		else
8918 			p->pktsize = 0;
8919 	}
8920 
8921 	rc = begin_synchronized_op(sc, NULL,
8922 	    sleep_ok ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4sscp");
8923 	if (rc)
8924 		return (rc);
8925 	tc = &pi->tc[p->cl];
8926 	tc->params = *p;
8927 	rc = -t4_sched_params(sc, FW_SCHED_TYPE_PKTSCHED, fw_level, fw_mode,
8928 	    fw_rateunit, fw_ratemode, p->channel, p->cl, p->minrate, p->maxrate,
8929 	    p->weight, p->pktsize, sleep_ok);
8930 	if (rc == 0)
8931 		tc->flags |= TX_SC_OK;
8932 	else {
8933 		/*
8934 		 * Unknown state at this point, see tc->params for what was
8935 		 * attempted.
8936 		 */
8937 		tc->flags &= ~TX_SC_OK;
8938 	}
8939 	end_synchronized_op(sc, sleep_ok ? 0 : LOCK_HELD);
8940 
8941 	return (rc);
8942 }
8943 
8944 int
8945 t4_set_sched_class(struct adapter *sc, struct t4_sched_params *p)
8946 {
8947 
8948 	if (p->type != SCHED_CLASS_TYPE_PACKET)
8949 		return (EINVAL);
8950 
8951 	if (p->subcmd == SCHED_CLASS_SUBCMD_CONFIG)
8952 		return (set_sched_class_config(sc, p->u.config.minmax));
8953 
8954 	if (p->subcmd == SCHED_CLASS_SUBCMD_PARAMS)
8955 		return (set_sched_class_params(sc, &p->u.params, 1));
8956 
8957 	return (EINVAL);
8958 }
8959 
8960 int
8961 t4_set_sched_queue(struct adapter *sc, struct t4_sched_queue *p)
8962 {
8963 	struct port_info *pi = NULL;
8964 	struct vi_info *vi;
8965 	struct sge_txq *txq;
8966 	uint32_t fw_mnem, fw_queue, fw_class;
8967 	int i, rc;
8968 
8969 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setsq");
8970 	if (rc)
8971 		return (rc);
8972 
8973 	if (p->port >= sc->params.nports) {
8974 		rc = EINVAL;
8975 		goto done;
8976 	}
8977 
8978 	/* XXX: Only supported for the main VI. */
8979 	pi = sc->port[p->port];
8980 	vi = &pi->vi[0];
8981 	if (!(vi->flags & VI_INIT_DONE)) {
8982 		/* tx queues not set up yet */
8983 		rc = EAGAIN;
8984 		goto done;
8985 	}
8986 
8987 	if (!in_range(p->queue, 0, vi->ntxq - 1) ||
8988 	    !in_range(p->cl, 0, sc->chip_params->nsched_cls - 1)) {
8989 		rc = EINVAL;
8990 		goto done;
8991 	}
8992 
8993 	/*
8994 	 * Create a template for the FW_PARAMS_CMD mnemonic and value (TX
8995 	 * Scheduling Class in this case).
8996 	 */
8997 	fw_mnem = (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
8998 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_EQ_SCHEDCLASS_ETH));
8999 	fw_class = p->cl < 0 ? 0xffffffff : p->cl;
9000 
9001 	/*
9002 	 * If op.queue is non-negative, then we're only changing the scheduling
9003 	 * on a single specified TX queue.
9004 	 */
9005 	if (p->queue >= 0) {
9006 		txq = &sc->sge.txq[vi->first_txq + p->queue];
9007 		fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
9008 		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
9009 		    &fw_class);
9010 		goto done;
9011 	}
9012 
9013 	/*
9014 	 * Change the scheduling on all the TX queues for the
9015 	 * interface.
9016 	 */
9017 	for_each_txq(vi, i, txq) {
9018 		fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
9019 		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
9020 		    &fw_class);
9021 		if (rc)
9022 			goto done;
9023 	}
9024 
9025 	rc = 0;
9026 done:
9027 	end_synchronized_op(sc, 0);
9028 	return (rc);
9029 }
9030 
9031 int
9032 t4_os_find_pci_capability(struct adapter *sc, int cap)
9033 {
9034 	int i;
9035 
9036 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9037 }
9038 
9039 int
9040 t4_os_pci_save_state(struct adapter *sc)
9041 {
9042 	device_t dev;
9043 	struct pci_devinfo *dinfo;
9044 
9045 	dev = sc->dev;
9046 	dinfo = device_get_ivars(dev);
9047 
9048 	pci_cfg_save(dev, dinfo, 0);
9049 	return (0);
9050 }
9051 
9052 int
9053 t4_os_pci_restore_state(struct adapter *sc)
9054 {
9055 	device_t dev;
9056 	struct pci_devinfo *dinfo;
9057 
9058 	dev = sc->dev;
9059 	dinfo = device_get_ivars(dev);
9060 
9061 	pci_cfg_restore(dev, dinfo);
9062 	return (0);
9063 }
9064 
9065 void
9066 t4_os_portmod_changed(const struct adapter *sc, int idx)
9067 {
9068 	struct port_info *pi = sc->port[idx];
9069 	struct vi_info *vi;
9070 	struct ifnet *ifp;
9071 	int v;
9072 	static const char *mod_str[] = {
9073 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9074 	};
9075 
9076 	for_each_vi(pi, v, vi) {
9077 		build_medialist(pi, &vi->media);
9078 	}
9079 
9080 	ifp = pi->vi[0].ifp;
9081 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9082 		if_printf(ifp, "transceiver unplugged.\n");
9083 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9084 		if_printf(ifp, "unknown transceiver inserted.\n");
9085 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9086 		if_printf(ifp, "unsupported transceiver inserted.\n");
9087 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9088 		if_printf(ifp, "%s transceiver inserted.\n",
9089 		    mod_str[pi->mod_type]);
9090 	} else {
9091 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9092 		    pi->mod_type);
9093 	}
9094 }
9095 
9096 void
9097 t4_os_link_changed(struct adapter *sc, int idx, int link_stat)
9098 {
9099 	struct port_info *pi = sc->port[idx];
9100 	struct vi_info *vi;
9101 	struct ifnet *ifp;
9102 	int v;
9103 
9104 	for_each_vi(pi, v, vi) {
9105 		ifp = vi->ifp;
9106 		if (ifp == NULL)
9107 			continue;
9108 
9109 		if (link_stat) {
9110 			ifp->if_baudrate = IF_Mbps(pi->link_cfg.speed);
9111 			if_link_state_change(ifp, LINK_STATE_UP);
9112 		} else {
9113 			if_link_state_change(ifp, LINK_STATE_DOWN);
9114 		}
9115 	}
9116 }
9117 
9118 void
9119 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9120 {
9121 	struct adapter *sc;
9122 
9123 	sx_slock(&t4_list_lock);
9124 	SLIST_FOREACH(sc, &t4_list, link) {
9125 		/*
9126 		 * func should not make any assumptions about what state sc is
9127 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9128 		 */
9129 		func(sc, arg);
9130 	}
9131 	sx_sunlock(&t4_list_lock);
9132 }
9133 
9134 static int
9135 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9136     struct thread *td)
9137 {
9138 	int rc;
9139 	struct adapter *sc = dev->si_drv1;
9140 
9141 	rc = priv_check(td, PRIV_DRIVER);
9142 	if (rc != 0)
9143 		return (rc);
9144 
9145 	switch (cmd) {
9146 	case CHELSIO_T4_GETREG: {
9147 		struct t4_reg *edata = (struct t4_reg *)data;
9148 
9149 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9150 			return (EFAULT);
9151 
9152 		if (edata->size == 4)
9153 			edata->val = t4_read_reg(sc, edata->addr);
9154 		else if (edata->size == 8)
9155 			edata->val = t4_read_reg64(sc, edata->addr);
9156 		else
9157 			return (EINVAL);
9158 
9159 		break;
9160 	}
9161 	case CHELSIO_T4_SETREG: {
9162 		struct t4_reg *edata = (struct t4_reg *)data;
9163 
9164 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9165 			return (EFAULT);
9166 
9167 		if (edata->size == 4) {
9168 			if (edata->val & 0xffffffff00000000)
9169 				return (EINVAL);
9170 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9171 		} else if (edata->size == 8)
9172 			t4_write_reg64(sc, edata->addr, edata->val);
9173 		else
9174 			return (EINVAL);
9175 		break;
9176 	}
9177 	case CHELSIO_T4_REGDUMP: {
9178 		struct t4_regdump *regs = (struct t4_regdump *)data;
9179 		int reglen = t4_get_regs_len(sc);
9180 		uint8_t *buf;
9181 
9182 		if (regs->len < reglen) {
9183 			regs->len = reglen; /* hint to the caller */
9184 			return (ENOBUFS);
9185 		}
9186 
9187 		regs->len = reglen;
9188 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9189 		get_regs(sc, regs, buf);
9190 		rc = copyout(buf, regs->data, reglen);
9191 		free(buf, M_CXGBE);
9192 		break;
9193 	}
9194 	case CHELSIO_T4_GET_FILTER_MODE:
9195 		rc = get_filter_mode(sc, (uint32_t *)data);
9196 		break;
9197 	case CHELSIO_T4_SET_FILTER_MODE:
9198 		rc = set_filter_mode(sc, *(uint32_t *)data);
9199 		break;
9200 	case CHELSIO_T4_GET_FILTER:
9201 		rc = get_filter(sc, (struct t4_filter *)data);
9202 		break;
9203 	case CHELSIO_T4_SET_FILTER:
9204 		rc = set_filter(sc, (struct t4_filter *)data);
9205 		break;
9206 	case CHELSIO_T4_DEL_FILTER:
9207 		rc = del_filter(sc, (struct t4_filter *)data);
9208 		break;
9209 	case CHELSIO_T4_GET_SGE_CONTEXT:
9210 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9211 		break;
9212 	case CHELSIO_T4_LOAD_FW:
9213 		rc = load_fw(sc, (struct t4_data *)data);
9214 		break;
9215 	case CHELSIO_T4_GET_MEM:
9216 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9217 		break;
9218 	case CHELSIO_T4_GET_I2C:
9219 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9220 		break;
9221 	case CHELSIO_T4_CLEAR_STATS: {
9222 		int i, v;
9223 		u_int port_id = *(uint32_t *)data;
9224 		struct port_info *pi;
9225 		struct vi_info *vi;
9226 
9227 		if (port_id >= sc->params.nports)
9228 			return (EINVAL);
9229 		pi = sc->port[port_id];
9230 		if (pi == NULL)
9231 			return (EIO);
9232 
9233 		/* MAC stats */
9234 		t4_clr_port_stats(sc, pi->tx_chan);
9235 		pi->tx_parse_error = 0;
9236 		mtx_lock(&sc->reg_lock);
9237 		for_each_vi(pi, v, vi) {
9238 			if (vi->flags & VI_INIT_DONE)
9239 				t4_clr_vi_stats(sc, vi->viid);
9240 		}
9241 		mtx_unlock(&sc->reg_lock);
9242 
9243 		/*
9244 		 * Since this command accepts a port, clear stats for
9245 		 * all VIs on this port.
9246 		 */
9247 		for_each_vi(pi, v, vi) {
9248 			if (vi->flags & VI_INIT_DONE) {
9249 				struct sge_rxq *rxq;
9250 				struct sge_txq *txq;
9251 				struct sge_wrq *wrq;
9252 
9253 				for_each_rxq(vi, i, rxq) {
9254 #if defined(INET) || defined(INET6)
9255 					rxq->lro.lro_queued = 0;
9256 					rxq->lro.lro_flushed = 0;
9257 #endif
9258 					rxq->rxcsum = 0;
9259 					rxq->vlan_extraction = 0;
9260 				}
9261 
9262 				for_each_txq(vi, i, txq) {
9263 					txq->txcsum = 0;
9264 					txq->tso_wrs = 0;
9265 					txq->vlan_insertion = 0;
9266 					txq->imm_wrs = 0;
9267 					txq->sgl_wrs = 0;
9268 					txq->txpkt_wrs = 0;
9269 					txq->txpkts0_wrs = 0;
9270 					txq->txpkts1_wrs = 0;
9271 					txq->txpkts0_pkts = 0;
9272 					txq->txpkts1_pkts = 0;
9273 					mp_ring_reset_stats(txq->r);
9274 				}
9275 
9276 #ifdef TCP_OFFLOAD
9277 				/* nothing to clear for each ofld_rxq */
9278 
9279 				for_each_ofld_txq(vi, i, wrq) {
9280 					wrq->tx_wrs_direct = 0;
9281 					wrq->tx_wrs_copied = 0;
9282 				}
9283 #endif
9284 
9285 				if (IS_MAIN_VI(vi)) {
9286 					wrq = &sc->sge.ctrlq[pi->port_id];
9287 					wrq->tx_wrs_direct = 0;
9288 					wrq->tx_wrs_copied = 0;
9289 				}
9290 			}
9291 		}
9292 		break;
9293 	}
9294 	case CHELSIO_T4_SCHED_CLASS:
9295 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9296 		break;
9297 	case CHELSIO_T4_SCHED_QUEUE:
9298 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9299 		break;
9300 	case CHELSIO_T4_GET_TRACER:
9301 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9302 		break;
9303 	case CHELSIO_T4_SET_TRACER:
9304 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9305 		break;
9306 	case CHELSIO_T4_LOAD_CFG:
9307 		rc = load_cfg(sc, (struct t4_data *)data);
9308 		break;
9309 	default:
9310 		rc = ENOTTY;
9311 	}
9312 
9313 	return (rc);
9314 }
9315 
9316 void
9317 t4_db_full(struct adapter *sc)
9318 {
9319 
9320 	CXGBE_UNIMPLEMENTED(__func__);
9321 }
9322 
9323 void
9324 t4_db_dropped(struct adapter *sc)
9325 {
9326 
9327 	CXGBE_UNIMPLEMENTED(__func__);
9328 }
9329 
9330 #ifdef TCP_OFFLOAD
9331 static int
9332 toe_capability(struct vi_info *vi, int enable)
9333 {
9334 	int rc;
9335 	struct port_info *pi = vi->pi;
9336 	struct adapter *sc = pi->adapter;
9337 
9338 	ASSERT_SYNCHRONIZED_OP(sc);
9339 
9340 	if (!is_offload(sc))
9341 		return (ENODEV);
9342 
9343 	if (enable) {
9344 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9345 			/* TOE is already enabled. */
9346 			return (0);
9347 		}
9348 
9349 		/*
9350 		 * We need the port's queues around so that we're able to send
9351 		 * and receive CPLs to/from the TOE even if the ifnet for this
9352 		 * port has never been UP'd administratively.
9353 		 */
9354 		if (!(vi->flags & VI_INIT_DONE)) {
9355 			rc = vi_full_init(vi);
9356 			if (rc)
9357 				return (rc);
9358 		}
9359 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9360 			rc = vi_full_init(&pi->vi[0]);
9361 			if (rc)
9362 				return (rc);
9363 		}
9364 
9365 		if (isset(&sc->offload_map, pi->port_id)) {
9366 			/* TOE is enabled on another VI of this port. */
9367 			pi->uld_vis++;
9368 			return (0);
9369 		}
9370 
9371 		if (!uld_active(sc, ULD_TOM)) {
9372 			rc = t4_activate_uld(sc, ULD_TOM);
9373 			if (rc == EAGAIN) {
9374 				log(LOG_WARNING,
9375 				    "You must kldload t4_tom.ko before trying "
9376 				    "to enable TOE on a cxgbe interface.\n");
9377 			}
9378 			if (rc != 0)
9379 				return (rc);
9380 			KASSERT(sc->tom_softc != NULL,
9381 			    ("%s: TOM activated but softc NULL", __func__));
9382 			KASSERT(uld_active(sc, ULD_TOM),
9383 			    ("%s: TOM activated but flag not set", __func__));
9384 		}
9385 
9386 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
9387 		if (!uld_active(sc, ULD_IWARP))
9388 			(void) t4_activate_uld(sc, ULD_IWARP);
9389 		if (!uld_active(sc, ULD_ISCSI))
9390 			(void) t4_activate_uld(sc, ULD_ISCSI);
9391 
9392 		pi->uld_vis++;
9393 		setbit(&sc->offload_map, pi->port_id);
9394 	} else {
9395 		pi->uld_vis--;
9396 
9397 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9398 			return (0);
9399 
9400 		KASSERT(uld_active(sc, ULD_TOM),
9401 		    ("%s: TOM never initialized?", __func__));
9402 		clrbit(&sc->offload_map, pi->port_id);
9403 	}
9404 
9405 	return (0);
9406 }
9407 
9408 /*
9409  * Add an upper layer driver to the global list.
9410  */
9411 int
9412 t4_register_uld(struct uld_info *ui)
9413 {
9414 	int rc = 0;
9415 	struct uld_info *u;
9416 
9417 	sx_xlock(&t4_uld_list_lock);
9418 	SLIST_FOREACH(u, &t4_uld_list, link) {
9419 	    if (u->uld_id == ui->uld_id) {
9420 		    rc = EEXIST;
9421 		    goto done;
9422 	    }
9423 	}
9424 
9425 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9426 	ui->refcount = 0;
9427 done:
9428 	sx_xunlock(&t4_uld_list_lock);
9429 	return (rc);
9430 }
9431 
9432 int
9433 t4_unregister_uld(struct uld_info *ui)
9434 {
9435 	int rc = EINVAL;
9436 	struct uld_info *u;
9437 
9438 	sx_xlock(&t4_uld_list_lock);
9439 
9440 	SLIST_FOREACH(u, &t4_uld_list, link) {
9441 	    if (u == ui) {
9442 		    if (ui->refcount > 0) {
9443 			    rc = EBUSY;
9444 			    goto done;
9445 		    }
9446 
9447 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9448 		    rc = 0;
9449 		    goto done;
9450 	    }
9451 	}
9452 done:
9453 	sx_xunlock(&t4_uld_list_lock);
9454 	return (rc);
9455 }
9456 
9457 int
9458 t4_activate_uld(struct adapter *sc, int id)
9459 {
9460 	int rc;
9461 	struct uld_info *ui;
9462 
9463 	ASSERT_SYNCHRONIZED_OP(sc);
9464 
9465 	if (id < 0 || id > ULD_MAX)
9466 		return (EINVAL);
9467 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
9468 
9469 	sx_slock(&t4_uld_list_lock);
9470 
9471 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9472 		if (ui->uld_id == id) {
9473 			if (!(sc->flags & FULL_INIT_DONE)) {
9474 				rc = adapter_full_init(sc);
9475 				if (rc != 0)
9476 					break;
9477 			}
9478 
9479 			rc = ui->activate(sc);
9480 			if (rc == 0) {
9481 				setbit(&sc->active_ulds, id);
9482 				ui->refcount++;
9483 			}
9484 			break;
9485 		}
9486 	}
9487 
9488 	sx_sunlock(&t4_uld_list_lock);
9489 
9490 	return (rc);
9491 }
9492 
9493 int
9494 t4_deactivate_uld(struct adapter *sc, int id)
9495 {
9496 	int rc;
9497 	struct uld_info *ui;
9498 
9499 	ASSERT_SYNCHRONIZED_OP(sc);
9500 
9501 	if (id < 0 || id > ULD_MAX)
9502 		return (EINVAL);
9503 	rc = ENXIO;
9504 
9505 	sx_slock(&t4_uld_list_lock);
9506 
9507 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9508 		if (ui->uld_id == id) {
9509 			rc = ui->deactivate(sc);
9510 			if (rc == 0) {
9511 				clrbit(&sc->active_ulds, id);
9512 				ui->refcount--;
9513 			}
9514 			break;
9515 		}
9516 	}
9517 
9518 	sx_sunlock(&t4_uld_list_lock);
9519 
9520 	return (rc);
9521 }
9522 
9523 int
9524 uld_active(struct adapter *sc, int uld_id)
9525 {
9526 
9527 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9528 
9529 	return (isset(&sc->active_ulds, uld_id));
9530 }
9531 #endif
9532 
9533 /*
9534  * Come up with reasonable defaults for some of the tunables, provided they're
9535  * not set by the user (in which case we'll use the values as is).
9536  */
9537 static void
9538 tweak_tunables(void)
9539 {
9540 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
9541 
9542 	if (t4_ntxq10g < 1) {
9543 #ifdef RSS
9544 		t4_ntxq10g = rss_getnumbuckets();
9545 #else
9546 		t4_ntxq10g = min(nc, NTXQ_10G);
9547 #endif
9548 	}
9549 
9550 	if (t4_ntxq1g < 1) {
9551 #ifdef RSS
9552 		/* XXX: way too many for 1GbE? */
9553 		t4_ntxq1g = rss_getnumbuckets();
9554 #else
9555 		t4_ntxq1g = min(nc, NTXQ_1G);
9556 #endif
9557 	}
9558 
9559 	if (t4_ntxq_vi < 1)
9560 		t4_ntxq_vi = min(nc, NTXQ_VI);
9561 
9562 	if (t4_nrxq10g < 1) {
9563 #ifdef RSS
9564 		t4_nrxq10g = rss_getnumbuckets();
9565 #else
9566 		t4_nrxq10g = min(nc, NRXQ_10G);
9567 #endif
9568 	}
9569 
9570 	if (t4_nrxq1g < 1) {
9571 #ifdef RSS
9572 		/* XXX: way too many for 1GbE? */
9573 		t4_nrxq1g = rss_getnumbuckets();
9574 #else
9575 		t4_nrxq1g = min(nc, NRXQ_1G);
9576 #endif
9577 	}
9578 
9579 	if (t4_nrxq_vi < 1)
9580 		t4_nrxq_vi = min(nc, NRXQ_VI);
9581 
9582 #ifdef TCP_OFFLOAD
9583 	if (t4_nofldtxq10g < 1)
9584 		t4_nofldtxq10g = min(nc, NOFLDTXQ_10G);
9585 
9586 	if (t4_nofldtxq1g < 1)
9587 		t4_nofldtxq1g = min(nc, NOFLDTXQ_1G);
9588 
9589 	if (t4_nofldtxq_vi < 1)
9590 		t4_nofldtxq_vi = min(nc, NOFLDTXQ_VI);
9591 
9592 	if (t4_nofldrxq10g < 1)
9593 		t4_nofldrxq10g = min(nc, NOFLDRXQ_10G);
9594 
9595 	if (t4_nofldrxq1g < 1)
9596 		t4_nofldrxq1g = min(nc, NOFLDRXQ_1G);
9597 
9598 	if (t4_nofldrxq_vi < 1)
9599 		t4_nofldrxq_vi = min(nc, NOFLDRXQ_VI);
9600 
9601 	if (t4_toecaps_allowed == -1)
9602 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9603 
9604 	if (t4_rdmacaps_allowed == -1) {
9605 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9606 		    FW_CAPS_CONFIG_RDMA_RDMAC;
9607 	}
9608 
9609 	if (t4_iscsicaps_allowed == -1) {
9610 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9611 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9612 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
9613 	}
9614 #else
9615 	if (t4_toecaps_allowed == -1)
9616 		t4_toecaps_allowed = 0;
9617 
9618 	if (t4_rdmacaps_allowed == -1)
9619 		t4_rdmacaps_allowed = 0;
9620 
9621 	if (t4_iscsicaps_allowed == -1)
9622 		t4_iscsicaps_allowed = 0;
9623 #endif
9624 
9625 #ifdef DEV_NETMAP
9626 	if (t4_nnmtxq_vi < 1)
9627 		t4_nnmtxq_vi = min(nc, NNMTXQ_VI);
9628 
9629 	if (t4_nnmrxq_vi < 1)
9630 		t4_nnmrxq_vi = min(nc, NNMRXQ_VI);
9631 #endif
9632 
9633 	if (t4_tmr_idx_10g < 0 || t4_tmr_idx_10g >= SGE_NTIMERS)
9634 		t4_tmr_idx_10g = TMR_IDX_10G;
9635 
9636 	if (t4_pktc_idx_10g < -1 || t4_pktc_idx_10g >= SGE_NCOUNTERS)
9637 		t4_pktc_idx_10g = PKTC_IDX_10G;
9638 
9639 	if (t4_tmr_idx_1g < 0 || t4_tmr_idx_1g >= SGE_NTIMERS)
9640 		t4_tmr_idx_1g = TMR_IDX_1G;
9641 
9642 	if (t4_pktc_idx_1g < -1 || t4_pktc_idx_1g >= SGE_NCOUNTERS)
9643 		t4_pktc_idx_1g = PKTC_IDX_1G;
9644 
9645 	if (t4_qsize_txq < 128)
9646 		t4_qsize_txq = 128;
9647 
9648 	if (t4_qsize_rxq < 128)
9649 		t4_qsize_rxq = 128;
9650 	while (t4_qsize_rxq & 7)
9651 		t4_qsize_rxq++;
9652 
9653 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9654 }
9655 
9656 #ifdef DDB
9657 static void
9658 t4_dump_tcb(struct adapter *sc, int tid)
9659 {
9660 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
9661 
9662 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
9663 	save = t4_read_reg(sc, reg);
9664 	base = sc->memwin[2].mw_base;
9665 
9666 	/* Dump TCB for the tid */
9667 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
9668 	tcb_addr += tid * TCB_SIZE;
9669 
9670 	if (is_t4(sc)) {
9671 		pf = 0;
9672 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
9673 	} else {
9674 		pf = V_PFNUM(sc->pf);
9675 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
9676 	}
9677 	t4_write_reg(sc, reg, win_pos | pf);
9678 	t4_read_reg(sc, reg);
9679 
9680 	off = tcb_addr - win_pos;
9681 	for (i = 0; i < 4; i++) {
9682 		uint32_t buf[8];
9683 		for (j = 0; j < 8; j++, off += 4)
9684 			buf[j] = htonl(t4_read_reg(sc, base + off));
9685 
9686 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
9687 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
9688 		    buf[7]);
9689 	}
9690 
9691 	t4_write_reg(sc, reg, save);
9692 	t4_read_reg(sc, reg);
9693 }
9694 
9695 static void
9696 t4_dump_devlog(struct adapter *sc)
9697 {
9698 	struct devlog_params *dparams = &sc->params.devlog;
9699 	struct fw_devlog_e e;
9700 	int i, first, j, m, nentries, rc;
9701 	uint64_t ftstamp = UINT64_MAX;
9702 
9703 	if (dparams->start == 0) {
9704 		db_printf("devlog params not valid\n");
9705 		return;
9706 	}
9707 
9708 	nentries = dparams->size / sizeof(struct fw_devlog_e);
9709 	m = fwmtype_to_hwmtype(dparams->memtype);
9710 
9711 	/* Find the first entry. */
9712 	first = -1;
9713 	for (i = 0; i < nentries && !db_pager_quit; i++) {
9714 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9715 		    sizeof(e), (void *)&e);
9716 		if (rc != 0)
9717 			break;
9718 
9719 		if (e.timestamp == 0)
9720 			break;
9721 
9722 		e.timestamp = be64toh(e.timestamp);
9723 		if (e.timestamp < ftstamp) {
9724 			ftstamp = e.timestamp;
9725 			first = i;
9726 		}
9727 	}
9728 
9729 	if (first == -1)
9730 		return;
9731 
9732 	i = first;
9733 	do {
9734 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9735 		    sizeof(e), (void *)&e);
9736 		if (rc != 0)
9737 			return;
9738 
9739 		if (e.timestamp == 0)
9740 			return;
9741 
9742 		e.timestamp = be64toh(e.timestamp);
9743 		e.seqno = be32toh(e.seqno);
9744 		for (j = 0; j < 8; j++)
9745 			e.params[j] = be32toh(e.params[j]);
9746 
9747 		db_printf("%10d  %15ju  %8s  %8s  ",
9748 		    e.seqno, e.timestamp,
9749 		    (e.level < nitems(devlog_level_strings) ?
9750 			devlog_level_strings[e.level] : "UNKNOWN"),
9751 		    (e.facility < nitems(devlog_facility_strings) ?
9752 			devlog_facility_strings[e.facility] : "UNKNOWN"));
9753 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
9754 		    e.params[3], e.params[4], e.params[5], e.params[6],
9755 		    e.params[7]);
9756 
9757 		if (++i == nentries)
9758 			i = 0;
9759 	} while (i != first && !db_pager_quit);
9760 }
9761 
9762 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
9763 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
9764 
9765 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
9766 {
9767 	device_t dev;
9768 	int t;
9769 	bool valid;
9770 
9771 	valid = false;
9772 	t = db_read_token();
9773 	if (t == tIDENT) {
9774 		dev = device_lookup_by_name(db_tok_string);
9775 		valid = true;
9776 	}
9777 	db_skip_to_eol();
9778 	if (!valid) {
9779 		db_printf("usage: show t4 devlog <nexus>\n");
9780 		return;
9781 	}
9782 
9783 	if (dev == NULL) {
9784 		db_printf("device not found\n");
9785 		return;
9786 	}
9787 
9788 	t4_dump_devlog(device_get_softc(dev));
9789 }
9790 
9791 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
9792 {
9793 	device_t dev;
9794 	int radix, tid, t;
9795 	bool valid;
9796 
9797 	valid = false;
9798 	radix = db_radix;
9799 	db_radix = 10;
9800 	t = db_read_token();
9801 	if (t == tIDENT) {
9802 		dev = device_lookup_by_name(db_tok_string);
9803 		t = db_read_token();
9804 		if (t == tNUMBER) {
9805 			tid = db_tok_number;
9806 			valid = true;
9807 		}
9808 	}
9809 	db_radix = radix;
9810 	db_skip_to_eol();
9811 	if (!valid) {
9812 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
9813 		return;
9814 	}
9815 
9816 	if (dev == NULL) {
9817 		db_printf("device not found\n");
9818 		return;
9819 	}
9820 	if (tid < 0) {
9821 		db_printf("invalid tid\n");
9822 		return;
9823 	}
9824 
9825 	t4_dump_tcb(device_get_softc(dev), tid);
9826 }
9827 #endif
9828 
9829 static struct sx mlu;	/* mod load unload */
9830 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
9831 
9832 static int
9833 mod_event(module_t mod, int cmd, void *arg)
9834 {
9835 	int rc = 0;
9836 	static int loaded = 0;
9837 
9838 	switch (cmd) {
9839 	case MOD_LOAD:
9840 		sx_xlock(&mlu);
9841 		if (loaded++ == 0) {
9842 			t4_sge_modload();
9843 			t4_register_cpl_handler(CPL_SET_TCB_RPL, set_tcb_rpl);
9844 			t4_register_cpl_handler(CPL_L2T_WRITE_RPL, l2t_write_rpl);
9845 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
9846 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
9847 			sx_init(&t4_list_lock, "T4/T5 adapters");
9848 			SLIST_INIT(&t4_list);
9849 #ifdef TCP_OFFLOAD
9850 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
9851 			SLIST_INIT(&t4_uld_list);
9852 #endif
9853 			t4_tracer_modload();
9854 			tweak_tunables();
9855 		}
9856 		sx_xunlock(&mlu);
9857 		break;
9858 
9859 	case MOD_UNLOAD:
9860 		sx_xlock(&mlu);
9861 		if (--loaded == 0) {
9862 			int tries;
9863 
9864 			sx_slock(&t4_list_lock);
9865 			if (!SLIST_EMPTY(&t4_list)) {
9866 				rc = EBUSY;
9867 				sx_sunlock(&t4_list_lock);
9868 				goto done_unload;
9869 			}
9870 #ifdef TCP_OFFLOAD
9871 			sx_slock(&t4_uld_list_lock);
9872 			if (!SLIST_EMPTY(&t4_uld_list)) {
9873 				rc = EBUSY;
9874 				sx_sunlock(&t4_uld_list_lock);
9875 				sx_sunlock(&t4_list_lock);
9876 				goto done_unload;
9877 			}
9878 #endif
9879 			tries = 0;
9880 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
9881 				uprintf("%ju clusters with custom free routine "
9882 				    "still is use.\n", t4_sge_extfree_refs());
9883 				pause("t4unload", 2 * hz);
9884 			}
9885 #ifdef TCP_OFFLOAD
9886 			sx_sunlock(&t4_uld_list_lock);
9887 #endif
9888 			sx_sunlock(&t4_list_lock);
9889 
9890 			if (t4_sge_extfree_refs() == 0) {
9891 				t4_tracer_modunload();
9892 #ifdef TCP_OFFLOAD
9893 				sx_destroy(&t4_uld_list_lock);
9894 #endif
9895 				sx_destroy(&t4_list_lock);
9896 				t4_sge_modunload();
9897 				loaded = 0;
9898 			} else {
9899 				rc = EBUSY;
9900 				loaded++;	/* undo earlier decrement */
9901 			}
9902 		}
9903 done_unload:
9904 		sx_xunlock(&mlu);
9905 		break;
9906 	}
9907 
9908 	return (rc);
9909 }
9910 
9911 static devclass_t t4_devclass, t5_devclass, t6_devclass;
9912 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
9913 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
9914 
9915 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
9916 MODULE_VERSION(t4nex, 1);
9917 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
9918 #ifdef DEV_NETMAP
9919 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
9920 #endif /* DEV_NETMAP */
9921 
9922 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
9923 MODULE_VERSION(t5nex, 1);
9924 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
9925 #ifdef DEV_NETMAP
9926 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
9927 #endif /* DEV_NETMAP */
9928 
9929 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
9930 MODULE_VERSION(t6nex, 1);
9931 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
9932 #ifdef DEV_NETMAP
9933 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
9934 #endif /* DEV_NETMAP */
9935 
9936 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
9937 MODULE_VERSION(cxgbe, 1);
9938 
9939 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
9940 MODULE_VERSION(cxl, 1);
9941 
9942 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
9943 MODULE_VERSION(cc, 1);
9944 
9945 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
9946 MODULE_VERSION(vcxgbe, 1);
9947 
9948 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
9949 MODULE_VERSION(vcxl, 1);
9950 
9951 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
9952 MODULE_VERSION(vcc, 1);
9953