xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision d93a896ef95946b0bf1219866fcb324b78543444)
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 "cudbg/cudbg.h"
77 #include "t4_ioctl.h"
78 #include "t4_l2t.h"
79 #include "t4_mp_ring.h"
80 #include "t4_if.h"
81 
82 /* T4 bus driver interface */
83 static int t4_probe(device_t);
84 static int t4_attach(device_t);
85 static int t4_detach(device_t);
86 static int t4_ready(device_t);
87 static int t4_read_port_device(device_t, int, device_t *);
88 static device_method_t t4_methods[] = {
89 	DEVMETHOD(device_probe,		t4_probe),
90 	DEVMETHOD(device_attach,	t4_attach),
91 	DEVMETHOD(device_detach,	t4_detach),
92 
93 	DEVMETHOD(t4_is_main_ready,	t4_ready),
94 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
95 
96 	DEVMETHOD_END
97 };
98 static driver_t t4_driver = {
99 	"t4nex",
100 	t4_methods,
101 	sizeof(struct adapter)
102 };
103 
104 
105 /* T4 port (cxgbe) interface */
106 static int cxgbe_probe(device_t);
107 static int cxgbe_attach(device_t);
108 static int cxgbe_detach(device_t);
109 device_method_t cxgbe_methods[] = {
110 	DEVMETHOD(device_probe,		cxgbe_probe),
111 	DEVMETHOD(device_attach,	cxgbe_attach),
112 	DEVMETHOD(device_detach,	cxgbe_detach),
113 	{ 0, 0 }
114 };
115 static driver_t cxgbe_driver = {
116 	"cxgbe",
117 	cxgbe_methods,
118 	sizeof(struct port_info)
119 };
120 
121 /* T4 VI (vcxgbe) interface */
122 static int vcxgbe_probe(device_t);
123 static int vcxgbe_attach(device_t);
124 static int vcxgbe_detach(device_t);
125 static device_method_t vcxgbe_methods[] = {
126 	DEVMETHOD(device_probe,		vcxgbe_probe),
127 	DEVMETHOD(device_attach,	vcxgbe_attach),
128 	DEVMETHOD(device_detach,	vcxgbe_detach),
129 	{ 0, 0 }
130 };
131 static driver_t vcxgbe_driver = {
132 	"vcxgbe",
133 	vcxgbe_methods,
134 	sizeof(struct vi_info)
135 };
136 
137 static d_ioctl_t t4_ioctl;
138 
139 static struct cdevsw t4_cdevsw = {
140        .d_version = D_VERSION,
141        .d_ioctl = t4_ioctl,
142        .d_name = "t4nex",
143 };
144 
145 /* T5 bus driver interface */
146 static int t5_probe(device_t);
147 static device_method_t t5_methods[] = {
148 	DEVMETHOD(device_probe,		t5_probe),
149 	DEVMETHOD(device_attach,	t4_attach),
150 	DEVMETHOD(device_detach,	t4_detach),
151 
152 	DEVMETHOD(t4_is_main_ready,	t4_ready),
153 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
154 
155 	DEVMETHOD_END
156 };
157 static driver_t t5_driver = {
158 	"t5nex",
159 	t5_methods,
160 	sizeof(struct adapter)
161 };
162 
163 
164 /* T5 port (cxl) interface */
165 static driver_t cxl_driver = {
166 	"cxl",
167 	cxgbe_methods,
168 	sizeof(struct port_info)
169 };
170 
171 /* T5 VI (vcxl) interface */
172 static driver_t vcxl_driver = {
173 	"vcxl",
174 	vcxgbe_methods,
175 	sizeof(struct vi_info)
176 };
177 
178 /* T6 bus driver interface */
179 static int t6_probe(device_t);
180 static device_method_t t6_methods[] = {
181 	DEVMETHOD(device_probe,		t6_probe),
182 	DEVMETHOD(device_attach,	t4_attach),
183 	DEVMETHOD(device_detach,	t4_detach),
184 
185 	DEVMETHOD(t4_is_main_ready,	t4_ready),
186 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
187 
188 	DEVMETHOD_END
189 };
190 static driver_t t6_driver = {
191 	"t6nex",
192 	t6_methods,
193 	sizeof(struct adapter)
194 };
195 
196 
197 /* T6 port (cc) interface */
198 static driver_t cc_driver = {
199 	"cc",
200 	cxgbe_methods,
201 	sizeof(struct port_info)
202 };
203 
204 /* T6 VI (vcc) interface */
205 static driver_t vcc_driver = {
206 	"vcc",
207 	vcxgbe_methods,
208 	sizeof(struct vi_info)
209 };
210 
211 /* ifnet + media interface */
212 static void cxgbe_init(void *);
213 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
214 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
215 static void cxgbe_qflush(struct ifnet *);
216 static int cxgbe_media_change(struct ifnet *);
217 static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
218 
219 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
220 
221 /*
222  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
223  * then ADAPTER_LOCK, then t4_uld_list_lock.
224  */
225 static struct sx t4_list_lock;
226 SLIST_HEAD(, adapter) t4_list;
227 #ifdef TCP_OFFLOAD
228 static struct sx t4_uld_list_lock;
229 SLIST_HEAD(, uld_info) t4_uld_list;
230 #endif
231 
232 /*
233  * Tunables.  See tweak_tunables() too.
234  *
235  * Each tunable is set to a default value here if it's known at compile-time.
236  * Otherwise it is set to -n as an indication to tweak_tunables() that it should
237  * provide a reasonable default (upto n) when the driver is loaded.
238  *
239  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
240  * T5 are under hw.cxl.
241  */
242 
243 /*
244  * Number of queues for tx and rx, 10G and 1G, NIC and offload.
245  */
246 #define NTXQ_10G 16
247 int t4_ntxq10g = -NTXQ_10G;
248 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq10g);
249 
250 #define NRXQ_10G 8
251 int t4_nrxq10g = -NRXQ_10G;
252 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq10g);
253 
254 #define NTXQ_1G 4
255 int t4_ntxq1g = -NTXQ_1G;
256 TUNABLE_INT("hw.cxgbe.ntxq1g", &t4_ntxq1g);
257 
258 #define NRXQ_1G 2
259 int t4_nrxq1g = -NRXQ_1G;
260 TUNABLE_INT("hw.cxgbe.nrxq1g", &t4_nrxq1g);
261 
262 #define NTXQ_VI 1
263 static int t4_ntxq_vi = -NTXQ_VI;
264 TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
265 
266 #define NRXQ_VI 1
267 static int t4_nrxq_vi = -NRXQ_VI;
268 TUNABLE_INT("hw.cxgbe.nrxq_vi", &t4_nrxq_vi);
269 
270 static int t4_rsrv_noflowq = 0;
271 TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
272 
273 #ifdef TCP_OFFLOAD
274 #define NOFLDTXQ_10G 8
275 static int t4_nofldtxq10g = -NOFLDTXQ_10G;
276 TUNABLE_INT("hw.cxgbe.nofldtxq10g", &t4_nofldtxq10g);
277 
278 #define NOFLDRXQ_10G 2
279 static int t4_nofldrxq10g = -NOFLDRXQ_10G;
280 TUNABLE_INT("hw.cxgbe.nofldrxq10g", &t4_nofldrxq10g);
281 
282 #define NOFLDTXQ_1G 2
283 static int t4_nofldtxq1g = -NOFLDTXQ_1G;
284 TUNABLE_INT("hw.cxgbe.nofldtxq1g", &t4_nofldtxq1g);
285 
286 #define NOFLDRXQ_1G 1
287 static int t4_nofldrxq1g = -NOFLDRXQ_1G;
288 TUNABLE_INT("hw.cxgbe.nofldrxq1g", &t4_nofldrxq1g);
289 
290 #define NOFLDTXQ_VI 1
291 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
292 TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
293 
294 #define NOFLDRXQ_VI 1
295 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
296 TUNABLE_INT("hw.cxgbe.nofldrxq_vi", &t4_nofldrxq_vi);
297 #endif
298 
299 #ifdef DEV_NETMAP
300 #define NNMTXQ_VI 2
301 static int t4_nnmtxq_vi = -NNMTXQ_VI;
302 TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
303 
304 #define NNMRXQ_VI 2
305 static int t4_nnmrxq_vi = -NNMRXQ_VI;
306 TUNABLE_INT("hw.cxgbe.nnmrxq_vi", &t4_nnmrxq_vi);
307 #endif
308 
309 /*
310  * Holdoff parameters for 10G and 1G ports.
311  */
312 #define TMR_IDX_10G 1
313 int t4_tmr_idx_10g = TMR_IDX_10G;
314 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx_10g);
315 
316 #define PKTC_IDX_10G (-1)
317 int t4_pktc_idx_10g = PKTC_IDX_10G;
318 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx_10g);
319 
320 #define TMR_IDX_1G 1
321 int t4_tmr_idx_1g = TMR_IDX_1G;
322 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_1G", &t4_tmr_idx_1g);
323 
324 #define PKTC_IDX_1G (-1)
325 int t4_pktc_idx_1g = PKTC_IDX_1G;
326 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_1G", &t4_pktc_idx_1g);
327 
328 /*
329  * Size (# of entries) of each tx and rx queue.
330  */
331 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
332 TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
333 
334 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
335 TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
336 
337 /*
338  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
339  */
340 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
341 TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
342 
343 /*
344  * Configuration file.
345  */
346 #define DEFAULT_CF	"default"
347 #define FLASH_CF	"flash"
348 #define UWIRE_CF	"uwire"
349 #define FPGA_CF		"fpga"
350 static char t4_cfg_file[32] = DEFAULT_CF;
351 TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
352 
353 /*
354  * PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
355  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
356  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
357  *            mark or when signalled to do so, 0 to never emit PAUSE.
358  */
359 static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
360 TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
361 
362 /*
363  * Forward Error Correction settings (bit 0, 1, 2 = FEC_RS, FEC_BASER_RS,
364  * FEC_RESERVED respectively).
365  * -1 to run with the firmware default.
366  *  0 to disable FEC.
367  */
368 static int t4_fec = -1;
369 TUNABLE_INT("hw.cxgbe.fec", &t4_fec);
370 
371 /*
372  * Link autonegotiation.
373  * -1 to run with the firmware default.
374  *  0 to disable.
375  *  1 to enable.
376  */
377 static int t4_autoneg = -1;
378 TUNABLE_INT("hw.cxgbe.autoneg", &t4_autoneg);
379 
380 /*
381  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
382  * encouraged respectively).
383  */
384 static unsigned int t4_fw_install = 1;
385 TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
386 
387 /*
388  * ASIC features that will be used.  Disable the ones you don't want so that the
389  * chip resources aren't wasted on features that will not be used.
390  */
391 static int t4_nbmcaps_allowed = 0;
392 TUNABLE_INT("hw.cxgbe.nbmcaps_allowed", &t4_nbmcaps_allowed);
393 
394 static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
395 TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
396 
397 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
398     FW_CAPS_CONFIG_SWITCH_EGRESS;
399 TUNABLE_INT("hw.cxgbe.switchcaps_allowed", &t4_switchcaps_allowed);
400 
401 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC;
402 TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
403 
404 static int t4_toecaps_allowed = -1;
405 TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
406 
407 static int t4_rdmacaps_allowed = -1;
408 TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
409 
410 static int t4_cryptocaps_allowed = 0;
411 TUNABLE_INT("hw.cxgbe.cryptocaps_allowed", &t4_cryptocaps_allowed);
412 
413 static int t4_iscsicaps_allowed = -1;
414 TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
415 
416 static int t4_fcoecaps_allowed = 0;
417 TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
418 
419 static int t5_write_combine = 0;
420 TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
421 
422 static int t4_num_vis = 1;
423 TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
424 
425 /* Functions used by extra VIs to obtain unique MAC addresses for each VI. */
426 static int vi_mac_funcs[] = {
427 	FW_VI_FUNC_OFLD,
428 	FW_VI_FUNC_IWARP,
429 	FW_VI_FUNC_OPENISCSI,
430 	FW_VI_FUNC_OPENFCOE,
431 	FW_VI_FUNC_FOISCSI,
432 	FW_VI_FUNC_FOFCOE,
433 };
434 
435 struct intrs_and_queues {
436 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
437 	uint16_t nirq;		/* Total # of vectors */
438 	uint16_t intr_flags_10g;/* Interrupt flags for each 10G port */
439 	uint16_t intr_flags_1g;	/* Interrupt flags for each 1G port */
440 	uint16_t ntxq10g;	/* # of NIC txq's for each 10G port */
441 	uint16_t nrxq10g;	/* # of NIC rxq's for each 10G port */
442 	uint16_t ntxq1g;	/* # of NIC txq's for each 1G port */
443 	uint16_t nrxq1g;	/* # of NIC rxq's for each 1G port */
444 	uint16_t rsrv_noflowq;	/* Flag whether to reserve queue 0 */
445 	uint16_t nofldtxq10g;	/* # of TOE txq's for each 10G port */
446 	uint16_t nofldrxq10g;	/* # of TOE rxq's for each 10G port */
447 	uint16_t nofldtxq1g;	/* # of TOE txq's for each 1G port */
448 	uint16_t nofldrxq1g;	/* # of TOE rxq's for each 1G port */
449 
450 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
451 	uint16_t ntxq_vi;	/* # of NIC txq's */
452 	uint16_t nrxq_vi;	/* # of NIC rxq's */
453 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
454 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
455 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
456 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
457 };
458 
459 struct filter_entry {
460         uint32_t valid:1;	/* filter allocated and valid */
461         uint32_t locked:1;	/* filter is administratively locked */
462         uint32_t pending:1;	/* filter action is pending firmware reply */
463 	uint32_t smtidx:8;	/* Source MAC Table index for smac */
464 	struct l2t_entry *l2t;	/* Layer Two Table entry for dmac */
465 
466         struct t4_filter_specification fs;
467 };
468 
469 static void setup_memwin(struct adapter *);
470 static void position_memwin(struct adapter *, int, uint32_t);
471 static int rw_via_memwin(struct adapter *, int, uint32_t, uint32_t *, int, int);
472 static inline int read_via_memwin(struct adapter *, int, uint32_t, uint32_t *,
473     int);
474 static inline int write_via_memwin(struct adapter *, int, uint32_t,
475     const uint32_t *, int);
476 static int validate_mem_range(struct adapter *, uint32_t, int);
477 static int fwmtype_to_hwmtype(int);
478 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
479     uint32_t *);
480 static int fixup_devlog_params(struct adapter *);
481 static int cfg_itype_and_nqueues(struct adapter *, int, int, int,
482     struct intrs_and_queues *);
483 static int prep_firmware(struct adapter *);
484 static int partition_resources(struct adapter *, const struct firmware *,
485     const char *);
486 static int get_params__pre_init(struct adapter *);
487 static int get_params__post_init(struct adapter *);
488 static int set_params__post_init(struct adapter *);
489 static void t4_set_desc(struct adapter *);
490 static void build_medialist(struct port_info *, struct ifmedia *);
491 static void init_l1cfg(struct port_info *);
492 static int cxgbe_init_synchronized(struct vi_info *);
493 static int cxgbe_uninit_synchronized(struct vi_info *);
494 static void quiesce_txq(struct adapter *, struct sge_txq *);
495 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
496 static void quiesce_iq(struct adapter *, struct sge_iq *);
497 static void quiesce_fl(struct adapter *, struct sge_fl *);
498 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
499     driver_intr_t *, void *, char *);
500 static int t4_free_irq(struct adapter *, struct irq *);
501 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
502 static void vi_refresh_stats(struct adapter *, struct vi_info *);
503 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
504 static void cxgbe_tick(void *);
505 static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
506 static void cxgbe_sysctls(struct port_info *);
507 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
508 static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
509 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
510 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
511 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
512 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
513 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
514 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
515 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
516 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
517 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
518 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
519 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
520 #ifdef SBUF_DRAIN
521 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
522 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
523 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
524 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
525 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
526 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
527 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
528 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
529 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
530 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
531 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
532 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
533 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
534 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
535 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
536 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
537 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
538 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
539 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
540 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
541 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
542 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
543 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
544 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
545 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
546 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
547 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
548 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
549 static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
550 #endif
551 #ifdef TCP_OFFLOAD
552 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
553 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
554 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
555 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
556 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
557 #endif
558 static uint32_t fconf_iconf_to_mode(uint32_t, uint32_t);
559 static uint32_t mode_to_fconf(uint32_t);
560 static uint32_t mode_to_iconf(uint32_t);
561 static int check_fspec_against_fconf_iconf(struct adapter *,
562     struct t4_filter_specification *);
563 static int get_filter_mode(struct adapter *, uint32_t *);
564 static int set_filter_mode(struct adapter *, uint32_t);
565 static inline uint64_t get_filter_hits(struct adapter *, uint32_t);
566 static int get_filter(struct adapter *, struct t4_filter *);
567 static int set_filter(struct adapter *, struct t4_filter *);
568 static int del_filter(struct adapter *, struct t4_filter *);
569 static void clear_filter(struct filter_entry *);
570 static int set_filter_wr(struct adapter *, int);
571 static int del_filter_wr(struct adapter *, int);
572 static int set_tcb_rpl(struct sge_iq *, const struct rss_header *,
573     struct mbuf *);
574 static int get_sge_context(struct adapter *, struct t4_sge_context *);
575 static int load_fw(struct adapter *, struct t4_data *);
576 static int load_cfg(struct adapter *, struct t4_data *);
577 static int load_boot(struct adapter *, struct t4_bootrom *);
578 static int load_bootcfg(struct adapter *, struct t4_data *);
579 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
580 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
581 static int read_i2c(struct adapter *, struct t4_i2c_data *);
582 #ifdef TCP_OFFLOAD
583 static int toe_capability(struct vi_info *, int);
584 #endif
585 static int mod_event(module_t, int, void *);
586 static int notify_siblings(device_t, int);
587 
588 struct {
589 	uint16_t device;
590 	char *desc;
591 } t4_pciids[] = {
592 	{0xa000, "Chelsio Terminator 4 FPGA"},
593 	{0x4400, "Chelsio T440-dbg"},
594 	{0x4401, "Chelsio T420-CR"},
595 	{0x4402, "Chelsio T422-CR"},
596 	{0x4403, "Chelsio T440-CR"},
597 	{0x4404, "Chelsio T420-BCH"},
598 	{0x4405, "Chelsio T440-BCH"},
599 	{0x4406, "Chelsio T440-CH"},
600 	{0x4407, "Chelsio T420-SO"},
601 	{0x4408, "Chelsio T420-CX"},
602 	{0x4409, "Chelsio T420-BT"},
603 	{0x440a, "Chelsio T404-BT"},
604 	{0x440e, "Chelsio T440-LP-CR"},
605 }, t5_pciids[] = {
606 	{0xb000, "Chelsio Terminator 5 FPGA"},
607 	{0x5400, "Chelsio T580-dbg"},
608 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
609 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
610 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
611 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
612 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
613 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
614 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
615 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
616 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
617 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
618 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
619 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
620 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
621 #ifdef notyet
622 	{0x5404,  "Chelsio T520-BCH"},
623 	{0x5405,  "Chelsio T540-BCH"},
624 	{0x5406,  "Chelsio T540-CH"},
625 	{0x5408,  "Chelsio T520-CX"},
626 	{0x540b,  "Chelsio B520-SR"},
627 	{0x540c,  "Chelsio B504-BT"},
628 	{0x540f,  "Chelsio Amsterdam"},
629 	{0x5413,  "Chelsio T580-CHR"},
630 #endif
631 }, t6_pciids[] = {
632 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
633 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
634 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
635 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
636 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
637 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
638 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
639 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
640 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
641 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
642 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
643 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
644 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
645 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
646 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
647 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
648 
649 	/* Custom */
650 	{0x6480, "Chelsio T6225 80"},
651 	{0x6481, "Chelsio T62100 81"},
652 };
653 
654 #ifdef TCP_OFFLOAD
655 /*
656  * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
657  * exactly the same for both rxq and ofld_rxq.
658  */
659 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
660 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
661 #endif
662 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
663 
664 static int
665 t4_probe(device_t dev)
666 {
667 	int i;
668 	uint16_t v = pci_get_vendor(dev);
669 	uint16_t d = pci_get_device(dev);
670 	uint8_t f = pci_get_function(dev);
671 
672 	if (v != PCI_VENDOR_ID_CHELSIO)
673 		return (ENXIO);
674 
675 	/* Attach only to PF0 of the FPGA */
676 	if (d == 0xa000 && f != 0)
677 		return (ENXIO);
678 
679 	for (i = 0; i < nitems(t4_pciids); i++) {
680 		if (d == t4_pciids[i].device) {
681 			device_set_desc(dev, t4_pciids[i].desc);
682 			return (BUS_PROBE_DEFAULT);
683 		}
684 	}
685 
686 	return (ENXIO);
687 }
688 
689 static int
690 t5_probe(device_t dev)
691 {
692 	int i;
693 	uint16_t v = pci_get_vendor(dev);
694 	uint16_t d = pci_get_device(dev);
695 	uint8_t f = pci_get_function(dev);
696 
697 	if (v != PCI_VENDOR_ID_CHELSIO)
698 		return (ENXIO);
699 
700 	/* Attach only to PF0 of the FPGA */
701 	if (d == 0xb000 && f != 0)
702 		return (ENXIO);
703 
704 	for (i = 0; i < nitems(t5_pciids); i++) {
705 		if (d == t5_pciids[i].device) {
706 			device_set_desc(dev, t5_pciids[i].desc);
707 			return (BUS_PROBE_DEFAULT);
708 		}
709 	}
710 
711 	return (ENXIO);
712 }
713 
714 static int
715 t6_probe(device_t dev)
716 {
717 	int i;
718 	uint16_t v = pci_get_vendor(dev);
719 	uint16_t d = pci_get_device(dev);
720 
721 	if (v != PCI_VENDOR_ID_CHELSIO)
722 		return (ENXIO);
723 
724 	for (i = 0; i < nitems(t6_pciids); i++) {
725 		if (d == t6_pciids[i].device) {
726 			device_set_desc(dev, t6_pciids[i].desc);
727 			return (BUS_PROBE_DEFAULT);
728 		}
729 	}
730 
731 	return (ENXIO);
732 }
733 
734 static void
735 t5_attribute_workaround(device_t dev)
736 {
737 	device_t root_port;
738 	uint32_t v;
739 
740 	/*
741 	 * The T5 chips do not properly echo the No Snoop and Relaxed
742 	 * Ordering attributes when replying to a TLP from a Root
743 	 * Port.  As a workaround, find the parent Root Port and
744 	 * disable No Snoop and Relaxed Ordering.  Note that this
745 	 * affects all devices under this root port.
746 	 */
747 	root_port = pci_find_pcie_root_port(dev);
748 	if (root_port == NULL) {
749 		device_printf(dev, "Unable to find parent root port\n");
750 		return;
751 	}
752 
753 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
754 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
755 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
756 	    0)
757 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
758 		    device_get_nameunit(root_port));
759 }
760 
761 static const struct devnames devnames[] = {
762 	{
763 		.nexus_name = "t4nex",
764 		.ifnet_name = "cxgbe",
765 		.vi_ifnet_name = "vcxgbe",
766 		.pf03_drv_name = "t4iov",
767 		.vf_nexus_name = "t4vf",
768 		.vf_ifnet_name = "cxgbev"
769 	}, {
770 		.nexus_name = "t5nex",
771 		.ifnet_name = "cxl",
772 		.vi_ifnet_name = "vcxl",
773 		.pf03_drv_name = "t5iov",
774 		.vf_nexus_name = "t5vf",
775 		.vf_ifnet_name = "cxlv"
776 	}, {
777 		.nexus_name = "t6nex",
778 		.ifnet_name = "cc",
779 		.vi_ifnet_name = "vcc",
780 		.pf03_drv_name = "t6iov",
781 		.vf_nexus_name = "t6vf",
782 		.vf_ifnet_name = "ccv"
783 	}
784 };
785 
786 void
787 t4_init_devnames(struct adapter *sc)
788 {
789 	int id;
790 
791 	id = chip_id(sc);
792 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
793 		sc->names = &devnames[id - CHELSIO_T4];
794 	else {
795 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
796 		sc->names = NULL;
797 	}
798 }
799 
800 static int
801 t4_attach(device_t dev)
802 {
803 	struct adapter *sc;
804 	int rc = 0, i, j, n10g, n1g, rqidx, tqidx;
805 	struct make_dev_args mda;
806 	struct intrs_and_queues iaq;
807 	struct sge *s;
808 	uint8_t *buf;
809 #ifdef TCP_OFFLOAD
810 	int ofld_rqidx, ofld_tqidx;
811 #endif
812 #ifdef DEV_NETMAP
813 	int nm_rqidx, nm_tqidx;
814 #endif
815 	int num_vis;
816 
817 	sc = device_get_softc(dev);
818 	sc->dev = dev;
819 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
820 
821 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
822 		t5_attribute_workaround(dev);
823 	pci_enable_busmaster(dev);
824 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
825 		uint32_t v;
826 
827 		pci_set_max_read_req(dev, 4096);
828 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
829 		v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
830 		pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
831 
832 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
833 	}
834 
835 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
836 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
837 	sc->traceq = -1;
838 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
839 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
840 	    device_get_nameunit(dev));
841 
842 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
843 	    device_get_nameunit(dev));
844 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
845 	t4_add_adapter(sc);
846 
847 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
848 	TAILQ_INIT(&sc->sfl);
849 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
850 
851 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
852 
853 	rc = t4_map_bars_0_and_4(sc);
854 	if (rc != 0)
855 		goto done; /* error message displayed already */
856 
857 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
858 
859 	/* Prepare the adapter for operation. */
860 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
861 	rc = -t4_prep_adapter(sc, buf);
862 	free(buf, M_CXGBE);
863 	if (rc != 0) {
864 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
865 		goto done;
866 	}
867 
868 	/*
869 	 * This is the real PF# to which we're attaching.  Works from within PCI
870 	 * passthrough environments too, where pci_get_function() could return a
871 	 * different PF# depending on the passthrough configuration.  We need to
872 	 * use the real PF# in all our communication with the firmware.
873 	 */
874 	j = t4_read_reg(sc, A_PL_WHOAMI);
875 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
876 	sc->mbox = sc->pf;
877 
878 	t4_init_devnames(sc);
879 	if (sc->names == NULL) {
880 		rc = ENOTSUP;
881 		goto done; /* error message displayed already */
882 	}
883 
884 	/*
885 	 * Do this really early, with the memory windows set up even before the
886 	 * character device.  The userland tool's register i/o and mem read
887 	 * will work even in "recovery mode".
888 	 */
889 	setup_memwin(sc);
890 	if (t4_init_devlog_params(sc, 0) == 0)
891 		fixup_devlog_params(sc);
892 	make_dev_args_init(&mda);
893 	mda.mda_devsw = &t4_cdevsw;
894 	mda.mda_uid = UID_ROOT;
895 	mda.mda_gid = GID_WHEEL;
896 	mda.mda_mode = 0600;
897 	mda.mda_si_drv1 = sc;
898 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
899 	if (rc != 0)
900 		device_printf(dev, "failed to create nexus char device: %d.\n",
901 		    rc);
902 
903 	/* Go no further if recovery mode has been requested. */
904 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
905 		device_printf(dev, "recovery mode.\n");
906 		goto done;
907 	}
908 
909 #if defined(__i386__)
910 	if ((cpu_feature & CPUID_CX8) == 0) {
911 		device_printf(dev, "64 bit atomics not available.\n");
912 		rc = ENOTSUP;
913 		goto done;
914 	}
915 #endif
916 
917 	/* Prepare the firmware for operation */
918 	rc = prep_firmware(sc);
919 	if (rc != 0)
920 		goto done; /* error message displayed already */
921 
922 	rc = get_params__post_init(sc);
923 	if (rc != 0)
924 		goto done; /* error message displayed already */
925 
926 	rc = set_params__post_init(sc);
927 	if (rc != 0)
928 		goto done; /* error message displayed already */
929 
930 	rc = t4_map_bar_2(sc);
931 	if (rc != 0)
932 		goto done; /* error message displayed already */
933 
934 	rc = t4_create_dma_tag(sc);
935 	if (rc != 0)
936 		goto done; /* error message displayed already */
937 
938 	/*
939 	 * Number of VIs to create per-port.  The first VI is the "main" regular
940 	 * VI for the port.  The rest are additional virtual interfaces on the
941 	 * same physical port.  Note that the main VI does not have native
942 	 * netmap support but the extra VIs do.
943 	 *
944 	 * Limit the number of VIs per port to the number of available
945 	 * MAC addresses per port.
946 	 */
947 	if (t4_num_vis >= 1)
948 		num_vis = t4_num_vis;
949 	else
950 		num_vis = 1;
951 	if (num_vis > nitems(vi_mac_funcs)) {
952 		num_vis = nitems(vi_mac_funcs);
953 		device_printf(dev, "Number of VIs limited to %d\n", num_vis);
954 	}
955 
956 	/*
957 	 * First pass over all the ports - allocate VIs and initialize some
958 	 * basic parameters like mac address, port type, etc.  We also figure
959 	 * out whether a port is 10G or 1G and use that information when
960 	 * calculating how many interrupts to attempt to allocate.
961 	 */
962 	n10g = n1g = 0;
963 	for_each_port(sc, i) {
964 		struct port_info *pi;
965 
966 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
967 		sc->port[i] = pi;
968 
969 		/* These must be set before t4_port_init */
970 		pi->adapter = sc;
971 		pi->port_id = i;
972 		/*
973 		 * XXX: vi[0] is special so we can't delay this allocation until
974 		 * pi->nvi's final value is known.
975 		 */
976 		pi->vi = malloc(sizeof(struct vi_info) * num_vis, M_CXGBE,
977 		    M_ZERO | M_WAITOK);
978 
979 		/*
980 		 * Allocate the "main" VI and initialize parameters
981 		 * like mac addr.
982 		 */
983 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
984 		if (rc != 0) {
985 			device_printf(dev, "unable to initialize port %d: %d\n",
986 			    i, rc);
987 			free(pi->vi, M_CXGBE);
988 			free(pi, M_CXGBE);
989 			sc->port[i] = NULL;
990 			goto done;
991 		}
992 
993 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
994 		    device_get_nameunit(dev), i);
995 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
996 		sc->chan_map[pi->tx_chan] = i;
997 
998 		if (port_top_speed(pi) >= 10) {
999 			n10g++;
1000 		} else {
1001 			n1g++;
1002 		}
1003 
1004 		pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
1005 		if (pi->dev == NULL) {
1006 			device_printf(dev,
1007 			    "failed to add device for port %d.\n", i);
1008 			rc = ENXIO;
1009 			goto done;
1010 		}
1011 		pi->vi[0].dev = pi->dev;
1012 		device_set_softc(pi->dev, pi);
1013 	}
1014 
1015 	/*
1016 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1017 	 */
1018 	rc = cfg_itype_and_nqueues(sc, n10g, n1g, num_vis, &iaq);
1019 	if (rc != 0)
1020 		goto done; /* error message displayed already */
1021 	if (iaq.nrxq_vi + iaq.nofldrxq_vi + iaq.nnmrxq_vi == 0)
1022 		num_vis = 1;
1023 
1024 	sc->intr_type = iaq.intr_type;
1025 	sc->intr_count = iaq.nirq;
1026 
1027 	s = &sc->sge;
1028 	s->nrxq = n10g * iaq.nrxq10g + n1g * iaq.nrxq1g;
1029 	s->ntxq = n10g * iaq.ntxq10g + n1g * iaq.ntxq1g;
1030 	if (num_vis > 1) {
1031 		s->nrxq += (n10g + n1g) * (num_vis - 1) * iaq.nrxq_vi;
1032 		s->ntxq += (n10g + n1g) * (num_vis - 1) * iaq.ntxq_vi;
1033 	}
1034 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1035 	s->neq += sc->params.nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
1036 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1037 #ifdef TCP_OFFLOAD
1038 	if (is_offload(sc)) {
1039 		s->nofldrxq = n10g * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
1040 		s->nofldtxq = n10g * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
1041 		if (num_vis > 1) {
1042 			s->nofldrxq += (n10g + n1g) * (num_vis - 1) *
1043 			    iaq.nofldrxq_vi;
1044 			s->nofldtxq += (n10g + n1g) * (num_vis - 1) *
1045 			    iaq.nofldtxq_vi;
1046 		}
1047 		s->neq += s->nofldtxq + s->nofldrxq;
1048 		s->niq += s->nofldrxq;
1049 
1050 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1051 		    M_CXGBE, M_ZERO | M_WAITOK);
1052 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1053 		    M_CXGBE, M_ZERO | M_WAITOK);
1054 	}
1055 #endif
1056 #ifdef DEV_NETMAP
1057 	if (num_vis > 1) {
1058 		s->nnmrxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmrxq_vi;
1059 		s->nnmtxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmtxq_vi;
1060 	}
1061 	s->neq += s->nnmtxq + s->nnmrxq;
1062 	s->niq += s->nnmrxq;
1063 
1064 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1065 	    M_CXGBE, M_ZERO | M_WAITOK);
1066 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1067 	    M_CXGBE, M_ZERO | M_WAITOK);
1068 #endif
1069 
1070 	s->ctrlq = malloc(sc->params.nports * sizeof(struct sge_wrq), M_CXGBE,
1071 	    M_ZERO | M_WAITOK);
1072 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1073 	    M_ZERO | M_WAITOK);
1074 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1075 	    M_ZERO | M_WAITOK);
1076 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1077 	    M_ZERO | M_WAITOK);
1078 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1079 	    M_ZERO | M_WAITOK);
1080 
1081 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1082 	    M_ZERO | M_WAITOK);
1083 
1084 	t4_init_l2t(sc, M_WAITOK);
1085 	t4_init_tx_sched(sc);
1086 
1087 	/*
1088 	 * Second pass over the ports.  This time we know the number of rx and
1089 	 * tx queues that each port should get.
1090 	 */
1091 	rqidx = tqidx = 0;
1092 #ifdef TCP_OFFLOAD
1093 	ofld_rqidx = ofld_tqidx = 0;
1094 #endif
1095 #ifdef DEV_NETMAP
1096 	nm_rqidx = nm_tqidx = 0;
1097 #endif
1098 	for_each_port(sc, i) {
1099 		struct port_info *pi = sc->port[i];
1100 		struct vi_info *vi;
1101 
1102 		if (pi == NULL)
1103 			continue;
1104 
1105 		pi->nvi = num_vis;
1106 		for_each_vi(pi, j, vi) {
1107 			vi->pi = pi;
1108 			vi->qsize_rxq = t4_qsize_rxq;
1109 			vi->qsize_txq = t4_qsize_txq;
1110 
1111 			vi->first_rxq = rqidx;
1112 			vi->first_txq = tqidx;
1113 			if (port_top_speed(pi) >= 10) {
1114 				vi->tmr_idx = t4_tmr_idx_10g;
1115 				vi->pktc_idx = t4_pktc_idx_10g;
1116 				vi->flags |= iaq.intr_flags_10g & INTR_RXQ;
1117 				vi->nrxq = j == 0 ? iaq.nrxq10g : iaq.nrxq_vi;
1118 				vi->ntxq = j == 0 ? iaq.ntxq10g : iaq.ntxq_vi;
1119 			} else {
1120 				vi->tmr_idx = t4_tmr_idx_1g;
1121 				vi->pktc_idx = t4_pktc_idx_1g;
1122 				vi->flags |= iaq.intr_flags_1g & INTR_RXQ;
1123 				vi->nrxq = j == 0 ? iaq.nrxq1g : iaq.nrxq_vi;
1124 				vi->ntxq = j == 0 ? iaq.ntxq1g : iaq.ntxq_vi;
1125 			}
1126 			rqidx += vi->nrxq;
1127 			tqidx += vi->ntxq;
1128 
1129 			if (j == 0 && vi->ntxq > 1)
1130 				vi->rsrv_noflowq = iaq.rsrv_noflowq ? 1 : 0;
1131 			else
1132 				vi->rsrv_noflowq = 0;
1133 
1134 #ifdef TCP_OFFLOAD
1135 			vi->first_ofld_rxq = ofld_rqidx;
1136 			vi->first_ofld_txq = ofld_tqidx;
1137 			if (port_top_speed(pi) >= 10) {
1138 				vi->flags |= iaq.intr_flags_10g & INTR_OFLD_RXQ;
1139 				vi->nofldrxq = j == 0 ? iaq.nofldrxq10g :
1140 				    iaq.nofldrxq_vi;
1141 				vi->nofldtxq = j == 0 ? iaq.nofldtxq10g :
1142 				    iaq.nofldtxq_vi;
1143 			} else {
1144 				vi->flags |= iaq.intr_flags_1g & INTR_OFLD_RXQ;
1145 				vi->nofldrxq = j == 0 ? iaq.nofldrxq1g :
1146 				    iaq.nofldrxq_vi;
1147 				vi->nofldtxq = j == 0 ? iaq.nofldtxq1g :
1148 				    iaq.nofldtxq_vi;
1149 			}
1150 			ofld_rqidx += vi->nofldrxq;
1151 			ofld_tqidx += vi->nofldtxq;
1152 #endif
1153 #ifdef DEV_NETMAP
1154 			if (j > 0) {
1155 				vi->first_nm_rxq = nm_rqidx;
1156 				vi->first_nm_txq = nm_tqidx;
1157 				vi->nnmrxq = iaq.nnmrxq_vi;
1158 				vi->nnmtxq = iaq.nnmtxq_vi;
1159 				nm_rqidx += vi->nnmrxq;
1160 				nm_tqidx += vi->nnmtxq;
1161 			}
1162 #endif
1163 		}
1164 	}
1165 
1166 	rc = t4_setup_intr_handlers(sc);
1167 	if (rc != 0) {
1168 		device_printf(dev,
1169 		    "failed to setup interrupt handlers: %d\n", rc);
1170 		goto done;
1171 	}
1172 
1173 	rc = bus_generic_probe(dev);
1174 	if (rc != 0) {
1175 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1176 		goto done;
1177 	}
1178 
1179 	rc = bus_generic_attach(dev);
1180 	if (rc != 0) {
1181 		device_printf(dev,
1182 		    "failed to attach all child ports: %d\n", rc);
1183 		goto done;
1184 	}
1185 
1186 	device_printf(dev,
1187 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1188 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1189 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1190 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1191 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1192 
1193 	t4_set_desc(sc);
1194 
1195 	notify_siblings(dev, 0);
1196 
1197 done:
1198 	if (rc != 0 && sc->cdev) {
1199 		/* cdev was created and so cxgbetool works; recover that way. */
1200 		device_printf(dev,
1201 		    "error during attach, adapter is now in recovery mode.\n");
1202 		rc = 0;
1203 	}
1204 
1205 	if (rc != 0)
1206 		t4_detach_common(dev);
1207 	else
1208 		t4_sysctls(sc);
1209 
1210 	return (rc);
1211 }
1212 
1213 static int
1214 t4_ready(device_t dev)
1215 {
1216 	struct adapter *sc;
1217 
1218 	sc = device_get_softc(dev);
1219 	if (sc->flags & FW_OK)
1220 		return (0);
1221 	return (ENXIO);
1222 }
1223 
1224 static int
1225 t4_read_port_device(device_t dev, int port, device_t *child)
1226 {
1227 	struct adapter *sc;
1228 	struct port_info *pi;
1229 
1230 	sc = device_get_softc(dev);
1231 	if (port < 0 || port >= MAX_NPORTS)
1232 		return (EINVAL);
1233 	pi = sc->port[port];
1234 	if (pi == NULL || pi->dev == NULL)
1235 		return (ENXIO);
1236 	*child = pi->dev;
1237 	return (0);
1238 }
1239 
1240 static int
1241 notify_siblings(device_t dev, int detaching)
1242 {
1243 	device_t sibling;
1244 	int error, i;
1245 
1246 	error = 0;
1247 	for (i = 0; i < PCI_FUNCMAX; i++) {
1248 		if (i == pci_get_function(dev))
1249 			continue;
1250 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1251 		    pci_get_slot(dev), i);
1252 		if (sibling == NULL || !device_is_attached(sibling))
1253 			continue;
1254 		if (detaching)
1255 			error = T4_DETACH_CHILD(sibling);
1256 		else
1257 			(void)T4_ATTACH_CHILD(sibling);
1258 		if (error)
1259 			break;
1260 	}
1261 	return (error);
1262 }
1263 
1264 /*
1265  * Idempotent
1266  */
1267 static int
1268 t4_detach(device_t dev)
1269 {
1270 	struct adapter *sc;
1271 	int rc;
1272 
1273 	sc = device_get_softc(dev);
1274 
1275 	rc = notify_siblings(dev, 1);
1276 	if (rc) {
1277 		device_printf(dev,
1278 		    "failed to detach sibling devices: %d\n", rc);
1279 		return (rc);
1280 	}
1281 
1282 	return (t4_detach_common(dev));
1283 }
1284 
1285 int
1286 t4_detach_common(device_t dev)
1287 {
1288 	struct adapter *sc;
1289 	struct port_info *pi;
1290 	int i, rc;
1291 
1292 	sc = device_get_softc(dev);
1293 
1294 	if (sc->flags & FULL_INIT_DONE) {
1295 		if (!(sc->flags & IS_VF))
1296 			t4_intr_disable(sc);
1297 	}
1298 
1299 	if (sc->cdev) {
1300 		destroy_dev(sc->cdev);
1301 		sc->cdev = NULL;
1302 	}
1303 
1304 	if (device_is_attached(dev)) {
1305 		rc = bus_generic_detach(dev);
1306 		if (rc) {
1307 			device_printf(dev,
1308 			    "failed to detach child devices: %d\n", rc);
1309 			return (rc);
1310 		}
1311 	}
1312 
1313 	for (i = 0; i < sc->intr_count; i++)
1314 		t4_free_irq(sc, &sc->irq[i]);
1315 
1316 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1317 		t4_free_tx_sched(sc);
1318 
1319 	for (i = 0; i < MAX_NPORTS; i++) {
1320 		pi = sc->port[i];
1321 		if (pi) {
1322 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1323 			if (pi->dev)
1324 				device_delete_child(dev, pi->dev);
1325 
1326 			mtx_destroy(&pi->pi_lock);
1327 			free(pi->vi, M_CXGBE);
1328 			free(pi, M_CXGBE);
1329 		}
1330 	}
1331 
1332 	device_delete_children(dev);
1333 
1334 	if (sc->flags & FULL_INIT_DONE)
1335 		adapter_full_uninit(sc);
1336 
1337 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1338 		t4_fw_bye(sc, sc->mbox);
1339 
1340 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1341 		pci_release_msi(dev);
1342 
1343 	if (sc->regs_res)
1344 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1345 		    sc->regs_res);
1346 
1347 	if (sc->udbs_res)
1348 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1349 		    sc->udbs_res);
1350 
1351 	if (sc->msix_res)
1352 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1353 		    sc->msix_res);
1354 
1355 	if (sc->l2t)
1356 		t4_free_l2t(sc->l2t);
1357 
1358 #ifdef TCP_OFFLOAD
1359 	free(sc->sge.ofld_rxq, M_CXGBE);
1360 	free(sc->sge.ofld_txq, M_CXGBE);
1361 #endif
1362 #ifdef DEV_NETMAP
1363 	free(sc->sge.nm_rxq, M_CXGBE);
1364 	free(sc->sge.nm_txq, M_CXGBE);
1365 #endif
1366 	free(sc->irq, M_CXGBE);
1367 	free(sc->sge.rxq, M_CXGBE);
1368 	free(sc->sge.txq, M_CXGBE);
1369 	free(sc->sge.ctrlq, M_CXGBE);
1370 	free(sc->sge.iqmap, M_CXGBE);
1371 	free(sc->sge.eqmap, M_CXGBE);
1372 	free(sc->tids.ftid_tab, M_CXGBE);
1373 	t4_destroy_dma_tag(sc);
1374 	if (mtx_initialized(&sc->sc_lock)) {
1375 		sx_xlock(&t4_list_lock);
1376 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1377 		sx_xunlock(&t4_list_lock);
1378 		mtx_destroy(&sc->sc_lock);
1379 	}
1380 
1381 	callout_drain(&sc->sfl_callout);
1382 	if (mtx_initialized(&sc->tids.ftid_lock))
1383 		mtx_destroy(&sc->tids.ftid_lock);
1384 	if (mtx_initialized(&sc->sfl_lock))
1385 		mtx_destroy(&sc->sfl_lock);
1386 	if (mtx_initialized(&sc->ifp_lock))
1387 		mtx_destroy(&sc->ifp_lock);
1388 	if (mtx_initialized(&sc->reg_lock))
1389 		mtx_destroy(&sc->reg_lock);
1390 
1391 	for (i = 0; i < NUM_MEMWIN; i++) {
1392 		struct memwin *mw = &sc->memwin[i];
1393 
1394 		if (rw_initialized(&mw->mw_lock))
1395 			rw_destroy(&mw->mw_lock);
1396 	}
1397 
1398 	bzero(sc, sizeof(*sc));
1399 
1400 	return (0);
1401 }
1402 
1403 static int
1404 cxgbe_probe(device_t dev)
1405 {
1406 	char buf[128];
1407 	struct port_info *pi = device_get_softc(dev);
1408 
1409 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1410 	device_set_desc_copy(dev, buf);
1411 
1412 	return (BUS_PROBE_DEFAULT);
1413 }
1414 
1415 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1416     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1417     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1418 #define T4_CAP_ENABLE (T4_CAP)
1419 
1420 static int
1421 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1422 {
1423 	struct ifnet *ifp;
1424 	struct sbuf *sb;
1425 
1426 	vi->xact_addr_filt = -1;
1427 	callout_init(&vi->tick, 1);
1428 
1429 	/* Allocate an ifnet and set it up */
1430 	ifp = if_alloc(IFT_ETHER);
1431 	if (ifp == NULL) {
1432 		device_printf(dev, "Cannot allocate ifnet\n");
1433 		return (ENOMEM);
1434 	}
1435 	vi->ifp = ifp;
1436 	ifp->if_softc = vi;
1437 
1438 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1439 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1440 
1441 	ifp->if_init = cxgbe_init;
1442 	ifp->if_ioctl = cxgbe_ioctl;
1443 	ifp->if_transmit = cxgbe_transmit;
1444 	ifp->if_qflush = cxgbe_qflush;
1445 	ifp->if_get_counter = cxgbe_get_counter;
1446 
1447 	ifp->if_capabilities = T4_CAP;
1448 #ifdef TCP_OFFLOAD
1449 	if (vi->nofldrxq != 0)
1450 		ifp->if_capabilities |= IFCAP_TOE;
1451 #endif
1452 #ifdef DEV_NETMAP
1453 	if (vi->nnmrxq != 0)
1454 		ifp->if_capabilities |= IFCAP_NETMAP;
1455 #endif
1456 	ifp->if_capenable = T4_CAP_ENABLE;
1457 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1458 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1459 
1460 	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1461 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1462 	ifp->if_hw_tsomaxsegsize = 65536;
1463 
1464 	/* Initialize ifmedia for this VI */
1465 	ifmedia_init(&vi->media, IFM_IMASK, cxgbe_media_change,
1466 	    cxgbe_media_status);
1467 
1468 	vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1469 	    EVENTHANDLER_PRI_ANY);
1470 
1471 	ether_ifattach(ifp, vi->hw_addr);
1472 #ifdef DEV_NETMAP
1473 	if (ifp->if_capabilities & IFCAP_NETMAP)
1474 		cxgbe_nm_attach(vi);
1475 #endif
1476 	sb = sbuf_new_auto();
1477 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1478 #ifdef TCP_OFFLOAD
1479 	if (ifp->if_capabilities & IFCAP_TOE)
1480 		sbuf_printf(sb, "; %d txq, %d rxq (TOE)",
1481 		    vi->nofldtxq, vi->nofldrxq);
1482 #endif
1483 #ifdef DEV_NETMAP
1484 	if (ifp->if_capabilities & IFCAP_NETMAP)
1485 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1486 		    vi->nnmtxq, vi->nnmrxq);
1487 #endif
1488 	sbuf_finish(sb);
1489 	device_printf(dev, "%s\n", sbuf_data(sb));
1490 	sbuf_delete(sb);
1491 
1492 	vi_sysctls(vi);
1493 
1494 	return (0);
1495 }
1496 
1497 static int
1498 cxgbe_attach(device_t dev)
1499 {
1500 	struct port_info *pi = device_get_softc(dev);
1501 	struct adapter *sc = pi->adapter;
1502 	struct vi_info *vi;
1503 	int i, rc;
1504 
1505 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1506 
1507 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1508 	if (rc)
1509 		return (rc);
1510 
1511 	for_each_vi(pi, i, vi) {
1512 		if (i == 0)
1513 			continue;
1514 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1515 		if (vi->dev == NULL) {
1516 			device_printf(dev, "failed to add VI %d\n", i);
1517 			continue;
1518 		}
1519 		device_set_softc(vi->dev, vi);
1520 	}
1521 
1522 	cxgbe_sysctls(pi);
1523 
1524 	bus_generic_attach(dev);
1525 
1526 	return (0);
1527 }
1528 
1529 static void
1530 cxgbe_vi_detach(struct vi_info *vi)
1531 {
1532 	struct ifnet *ifp = vi->ifp;
1533 
1534 	ether_ifdetach(ifp);
1535 
1536 	if (vi->vlan_c)
1537 		EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1538 
1539 	/* Let detach proceed even if these fail. */
1540 #ifdef DEV_NETMAP
1541 	if (ifp->if_capabilities & IFCAP_NETMAP)
1542 		cxgbe_nm_detach(vi);
1543 #endif
1544 	cxgbe_uninit_synchronized(vi);
1545 	callout_drain(&vi->tick);
1546 	vi_full_uninit(vi);
1547 
1548 	ifmedia_removeall(&vi->media);
1549 	if_free(vi->ifp);
1550 	vi->ifp = NULL;
1551 }
1552 
1553 static int
1554 cxgbe_detach(device_t dev)
1555 {
1556 	struct port_info *pi = device_get_softc(dev);
1557 	struct adapter *sc = pi->adapter;
1558 	int rc;
1559 
1560 	/* Detach the extra VIs first. */
1561 	rc = bus_generic_detach(dev);
1562 	if (rc)
1563 		return (rc);
1564 	device_delete_children(dev);
1565 
1566 	doom_vi(sc, &pi->vi[0]);
1567 
1568 	if (pi->flags & HAS_TRACEQ) {
1569 		sc->traceq = -1;	/* cloner should not create ifnet */
1570 		t4_tracer_port_detach(sc);
1571 	}
1572 
1573 	cxgbe_vi_detach(&pi->vi[0]);
1574 	callout_drain(&pi->tick);
1575 
1576 	end_synchronized_op(sc, 0);
1577 
1578 	return (0);
1579 }
1580 
1581 static void
1582 cxgbe_init(void *arg)
1583 {
1584 	struct vi_info *vi = arg;
1585 	struct adapter *sc = vi->pi->adapter;
1586 
1587 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1588 		return;
1589 	cxgbe_init_synchronized(vi);
1590 	end_synchronized_op(sc, 0);
1591 }
1592 
1593 static int
1594 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1595 {
1596 	int rc = 0, mtu, flags, can_sleep;
1597 	struct vi_info *vi = ifp->if_softc;
1598 	struct adapter *sc = vi->pi->adapter;
1599 	struct ifreq *ifr = (struct ifreq *)data;
1600 	uint32_t mask;
1601 
1602 	switch (cmd) {
1603 	case SIOCSIFMTU:
1604 		mtu = ifr->ifr_mtu;
1605 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1606 			return (EINVAL);
1607 
1608 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1609 		if (rc)
1610 			return (rc);
1611 		ifp->if_mtu = mtu;
1612 		if (vi->flags & VI_INIT_DONE) {
1613 			t4_update_fl_bufsize(ifp);
1614 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1615 				rc = update_mac_settings(ifp, XGMAC_MTU);
1616 		}
1617 		end_synchronized_op(sc, 0);
1618 		break;
1619 
1620 	case SIOCSIFFLAGS:
1621 		can_sleep = 0;
1622 redo_sifflags:
1623 		rc = begin_synchronized_op(sc, vi,
1624 		    can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1625 		if (rc)
1626 			return (rc);
1627 
1628 		if (ifp->if_flags & IFF_UP) {
1629 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1630 				flags = vi->if_flags;
1631 				if ((ifp->if_flags ^ flags) &
1632 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1633 					if (can_sleep == 1) {
1634 						end_synchronized_op(sc, 0);
1635 						can_sleep = 0;
1636 						goto redo_sifflags;
1637 					}
1638 					rc = update_mac_settings(ifp,
1639 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1640 				}
1641 			} else {
1642 				if (can_sleep == 0) {
1643 					end_synchronized_op(sc, LOCK_HELD);
1644 					can_sleep = 1;
1645 					goto redo_sifflags;
1646 				}
1647 				rc = cxgbe_init_synchronized(vi);
1648 			}
1649 			vi->if_flags = ifp->if_flags;
1650 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1651 			if (can_sleep == 0) {
1652 				end_synchronized_op(sc, LOCK_HELD);
1653 				can_sleep = 1;
1654 				goto redo_sifflags;
1655 			}
1656 			rc = cxgbe_uninit_synchronized(vi);
1657 		}
1658 		end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1659 		break;
1660 
1661 	case SIOCADDMULTI:
1662 	case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1663 		rc = begin_synchronized_op(sc, vi, HOLD_LOCK, "t4multi");
1664 		if (rc)
1665 			return (rc);
1666 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1667 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1668 		end_synchronized_op(sc, LOCK_HELD);
1669 		break;
1670 
1671 	case SIOCSIFCAP:
1672 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1673 		if (rc)
1674 			return (rc);
1675 
1676 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1677 		if (mask & IFCAP_TXCSUM) {
1678 			ifp->if_capenable ^= IFCAP_TXCSUM;
1679 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1680 
1681 			if (IFCAP_TSO4 & ifp->if_capenable &&
1682 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1683 				ifp->if_capenable &= ~IFCAP_TSO4;
1684 				if_printf(ifp,
1685 				    "tso4 disabled due to -txcsum.\n");
1686 			}
1687 		}
1688 		if (mask & IFCAP_TXCSUM_IPV6) {
1689 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1690 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1691 
1692 			if (IFCAP_TSO6 & ifp->if_capenable &&
1693 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1694 				ifp->if_capenable &= ~IFCAP_TSO6;
1695 				if_printf(ifp,
1696 				    "tso6 disabled due to -txcsum6.\n");
1697 			}
1698 		}
1699 		if (mask & IFCAP_RXCSUM)
1700 			ifp->if_capenable ^= IFCAP_RXCSUM;
1701 		if (mask & IFCAP_RXCSUM_IPV6)
1702 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1703 
1704 		/*
1705 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1706 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1707 		 * sending a TSO request our way, so it's sufficient to toggle
1708 		 * IFCAP_TSOx only.
1709 		 */
1710 		if (mask & IFCAP_TSO4) {
1711 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1712 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1713 				if_printf(ifp, "enable txcsum first.\n");
1714 				rc = EAGAIN;
1715 				goto fail;
1716 			}
1717 			ifp->if_capenable ^= IFCAP_TSO4;
1718 		}
1719 		if (mask & IFCAP_TSO6) {
1720 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1721 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1722 				if_printf(ifp, "enable txcsum6 first.\n");
1723 				rc = EAGAIN;
1724 				goto fail;
1725 			}
1726 			ifp->if_capenable ^= IFCAP_TSO6;
1727 		}
1728 		if (mask & IFCAP_LRO) {
1729 #if defined(INET) || defined(INET6)
1730 			int i;
1731 			struct sge_rxq *rxq;
1732 
1733 			ifp->if_capenable ^= IFCAP_LRO;
1734 			for_each_rxq(vi, i, rxq) {
1735 				if (ifp->if_capenable & IFCAP_LRO)
1736 					rxq->iq.flags |= IQ_LRO_ENABLED;
1737 				else
1738 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1739 			}
1740 #endif
1741 		}
1742 #ifdef TCP_OFFLOAD
1743 		if (mask & IFCAP_TOE) {
1744 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1745 
1746 			rc = toe_capability(vi, enable);
1747 			if (rc != 0)
1748 				goto fail;
1749 
1750 			ifp->if_capenable ^= mask;
1751 		}
1752 #endif
1753 		if (mask & IFCAP_VLAN_HWTAGGING) {
1754 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1755 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1756 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1757 		}
1758 		if (mask & IFCAP_VLAN_MTU) {
1759 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1760 
1761 			/* Need to find out how to disable auto-mtu-inflation */
1762 		}
1763 		if (mask & IFCAP_VLAN_HWTSO)
1764 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1765 		if (mask & IFCAP_VLAN_HWCSUM)
1766 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1767 
1768 #ifdef VLAN_CAPABILITIES
1769 		VLAN_CAPABILITIES(ifp);
1770 #endif
1771 fail:
1772 		end_synchronized_op(sc, 0);
1773 		break;
1774 
1775 	case SIOCSIFMEDIA:
1776 	case SIOCGIFMEDIA:
1777 	case SIOCGIFXMEDIA:
1778 		ifmedia_ioctl(ifp, ifr, &vi->media, cmd);
1779 		break;
1780 
1781 	case SIOCGI2C: {
1782 		struct ifi2creq i2c;
1783 
1784 		rc = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
1785 		if (rc != 0)
1786 			break;
1787 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1788 			rc = EPERM;
1789 			break;
1790 		}
1791 		if (i2c.len > sizeof(i2c.data)) {
1792 			rc = EINVAL;
1793 			break;
1794 		}
1795 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1796 		if (rc)
1797 			return (rc);
1798 		rc = -t4_i2c_rd(sc, sc->mbox, vi->pi->port_id, i2c.dev_addr,
1799 		    i2c.offset, i2c.len, &i2c.data[0]);
1800 		end_synchronized_op(sc, 0);
1801 		if (rc == 0)
1802 			rc = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
1803 		break;
1804 	}
1805 
1806 	default:
1807 		rc = ether_ioctl(ifp, cmd, data);
1808 	}
1809 
1810 	return (rc);
1811 }
1812 
1813 static int
1814 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1815 {
1816 	struct vi_info *vi = ifp->if_softc;
1817 	struct port_info *pi = vi->pi;
1818 	struct adapter *sc = pi->adapter;
1819 	struct sge_txq *txq;
1820 	void *items[1];
1821 	int rc;
1822 
1823 	M_ASSERTPKTHDR(m);
1824 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1825 
1826 	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1827 		m_freem(m);
1828 		return (ENETDOWN);
1829 	}
1830 
1831 	rc = parse_pkt(sc, &m);
1832 	if (__predict_false(rc != 0)) {
1833 		MPASS(m == NULL);			/* was freed already */
1834 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1835 		return (rc);
1836 	}
1837 
1838 	/* Select a txq. */
1839 	txq = &sc->sge.txq[vi->first_txq];
1840 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1841 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1842 		    vi->rsrv_noflowq);
1843 
1844 	items[0] = m;
1845 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1846 	if (__predict_false(rc != 0))
1847 		m_freem(m);
1848 
1849 	return (rc);
1850 }
1851 
1852 static void
1853 cxgbe_qflush(struct ifnet *ifp)
1854 {
1855 	struct vi_info *vi = ifp->if_softc;
1856 	struct sge_txq *txq;
1857 	int i;
1858 
1859 	/* queues do not exist if !VI_INIT_DONE. */
1860 	if (vi->flags & VI_INIT_DONE) {
1861 		for_each_txq(vi, i, txq) {
1862 			TXQ_LOCK(txq);
1863 			txq->eq.flags |= EQ_QFLUSH;
1864 			TXQ_UNLOCK(txq);
1865 			while (!mp_ring_is_idle(txq->r)) {
1866 				mp_ring_check_drainage(txq->r, 0);
1867 				pause("qflush", 1);
1868 			}
1869 			TXQ_LOCK(txq);
1870 			txq->eq.flags &= ~EQ_QFLUSH;
1871 			TXQ_UNLOCK(txq);
1872 		}
1873 	}
1874 	if_qflush(ifp);
1875 }
1876 
1877 static uint64_t
1878 vi_get_counter(struct ifnet *ifp, ift_counter c)
1879 {
1880 	struct vi_info *vi = ifp->if_softc;
1881 	struct fw_vi_stats_vf *s = &vi->stats;
1882 
1883 	vi_refresh_stats(vi->pi->adapter, vi);
1884 
1885 	switch (c) {
1886 	case IFCOUNTER_IPACKETS:
1887 		return (s->rx_bcast_frames + s->rx_mcast_frames +
1888 		    s->rx_ucast_frames);
1889 	case IFCOUNTER_IERRORS:
1890 		return (s->rx_err_frames);
1891 	case IFCOUNTER_OPACKETS:
1892 		return (s->tx_bcast_frames + s->tx_mcast_frames +
1893 		    s->tx_ucast_frames + s->tx_offload_frames);
1894 	case IFCOUNTER_OERRORS:
1895 		return (s->tx_drop_frames);
1896 	case IFCOUNTER_IBYTES:
1897 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
1898 		    s->rx_ucast_bytes);
1899 	case IFCOUNTER_OBYTES:
1900 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
1901 		    s->tx_ucast_bytes + s->tx_offload_bytes);
1902 	case IFCOUNTER_IMCASTS:
1903 		return (s->rx_mcast_frames);
1904 	case IFCOUNTER_OMCASTS:
1905 		return (s->tx_mcast_frames);
1906 	case IFCOUNTER_OQDROPS: {
1907 		uint64_t drops;
1908 
1909 		drops = 0;
1910 		if (vi->flags & VI_INIT_DONE) {
1911 			int i;
1912 			struct sge_txq *txq;
1913 
1914 			for_each_txq(vi, i, txq)
1915 				drops += counter_u64_fetch(txq->r->drops);
1916 		}
1917 
1918 		return (drops);
1919 
1920 	}
1921 
1922 	default:
1923 		return (if_get_counter_default(ifp, c));
1924 	}
1925 }
1926 
1927 uint64_t
1928 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
1929 {
1930 	struct vi_info *vi = ifp->if_softc;
1931 	struct port_info *pi = vi->pi;
1932 	struct adapter *sc = pi->adapter;
1933 	struct port_stats *s = &pi->stats;
1934 
1935 	if (pi->nvi > 1 || sc->flags & IS_VF)
1936 		return (vi_get_counter(ifp, c));
1937 
1938 	cxgbe_refresh_stats(sc, pi);
1939 
1940 	switch (c) {
1941 	case IFCOUNTER_IPACKETS:
1942 		return (s->rx_frames);
1943 
1944 	case IFCOUNTER_IERRORS:
1945 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
1946 		    s->rx_fcs_err + s->rx_len_err);
1947 
1948 	case IFCOUNTER_OPACKETS:
1949 		return (s->tx_frames);
1950 
1951 	case IFCOUNTER_OERRORS:
1952 		return (s->tx_error_frames);
1953 
1954 	case IFCOUNTER_IBYTES:
1955 		return (s->rx_octets);
1956 
1957 	case IFCOUNTER_OBYTES:
1958 		return (s->tx_octets);
1959 
1960 	case IFCOUNTER_IMCASTS:
1961 		return (s->rx_mcast_frames);
1962 
1963 	case IFCOUNTER_OMCASTS:
1964 		return (s->tx_mcast_frames);
1965 
1966 	case IFCOUNTER_IQDROPS:
1967 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
1968 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
1969 		    s->rx_trunc3 + pi->tnl_cong_drops);
1970 
1971 	case IFCOUNTER_OQDROPS: {
1972 		uint64_t drops;
1973 
1974 		drops = s->tx_drop;
1975 		if (vi->flags & VI_INIT_DONE) {
1976 			int i;
1977 			struct sge_txq *txq;
1978 
1979 			for_each_txq(vi, i, txq)
1980 				drops += counter_u64_fetch(txq->r->drops);
1981 		}
1982 
1983 		return (drops);
1984 
1985 	}
1986 
1987 	default:
1988 		return (if_get_counter_default(ifp, c));
1989 	}
1990 }
1991 
1992 static int
1993 cxgbe_media_change(struct ifnet *ifp)
1994 {
1995 	struct vi_info *vi = ifp->if_softc;
1996 
1997 	device_printf(vi->dev, "%s unimplemented.\n", __func__);
1998 
1999 	return (EOPNOTSUPP);
2000 }
2001 
2002 static void
2003 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2004 {
2005 	struct vi_info *vi = ifp->if_softc;
2006 	struct port_info *pi = vi->pi;
2007 	struct ifmedia_entry *cur;
2008 	struct link_config *lc = &pi->link_cfg;
2009 
2010 	/*
2011 	 * If all the interfaces are administratively down the firmware does not
2012 	 * report transceiver changes.  Refresh port info here so that ifconfig
2013 	 * displays accurate information at all times.
2014 	 */
2015 	if (begin_synchronized_op(pi->adapter, NULL, SLEEP_OK | INTR_OK,
2016 	    "t4med") == 0) {
2017 		PORT_LOCK(pi);
2018 		if (pi->up_vis == 0) {
2019 			t4_update_port_info(pi);
2020 			build_medialist(pi, &vi->media);
2021 		}
2022 		PORT_UNLOCK(pi);
2023 		end_synchronized_op(pi->adapter, 0);
2024 	}
2025 
2026 	cur = vi->media.ifm_cur;
2027 
2028 	ifmr->ifm_status = IFM_AVALID;
2029 	if (lc->link_ok == 0)
2030 		return;
2031 
2032 	ifmr->ifm_status |= IFM_ACTIVE;
2033 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2034 	if (lc->fc & PAUSE_RX)
2035 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2036 	if (lc->fc & PAUSE_TX)
2037 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2038 
2039 	/* active and current will differ iff current media is autoselect. */
2040 	if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
2041 		return;
2042 
2043 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2044 	if (lc->fc & PAUSE_RX)
2045 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2046 	if (lc->fc & PAUSE_TX)
2047 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2048 	switch (lc->speed) {
2049 	case 10000:
2050 		ifmr->ifm_active |= IFM_10G_T;
2051 		break;
2052 	case 1000:
2053 		ifmr->ifm_active |= IFM_1000_T;
2054 		break;
2055 	case 100:
2056 		ifmr->ifm_active |= IFM_100_TX;
2057 		break;
2058 	case 10:
2059 		ifmr->ifm_active |= IFM_10_T;
2060 		break;
2061 	default:
2062 		device_printf(vi->dev, "link up but speed unknown (%u)\n",
2063 		    lc->speed);
2064 	}
2065 }
2066 
2067 static int
2068 vcxgbe_probe(device_t dev)
2069 {
2070 	char buf[128];
2071 	struct vi_info *vi = device_get_softc(dev);
2072 
2073 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2074 	    vi - vi->pi->vi);
2075 	device_set_desc_copy(dev, buf);
2076 
2077 	return (BUS_PROBE_DEFAULT);
2078 }
2079 
2080 static int
2081 vcxgbe_attach(device_t dev)
2082 {
2083 	struct vi_info *vi;
2084 	struct port_info *pi;
2085 	struct adapter *sc;
2086 	int func, index, rc;
2087 	u32 param, val;
2088 
2089 	vi = device_get_softc(dev);
2090 	pi = vi->pi;
2091 	sc = pi->adapter;
2092 
2093 	index = vi - pi->vi;
2094 	KASSERT(index < nitems(vi_mac_funcs),
2095 	    ("%s: VI %s doesn't have a MAC func", __func__,
2096 	    device_get_nameunit(dev)));
2097 	func = vi_mac_funcs[index];
2098 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2099 	    vi->hw_addr, &vi->rss_size, func, 0);
2100 	if (rc < 0) {
2101 		device_printf(dev, "Failed to allocate virtual interface "
2102 		    "for port %d: %d\n", pi->port_id, -rc);
2103 		return (-rc);
2104 	}
2105 	vi->viid = rc;
2106 	if (chip_id(sc) <= CHELSIO_T5)
2107 		vi->smt_idx = (rc & 0x7f) << 1;
2108 	else
2109 		vi->smt_idx = (rc & 0x7f);
2110 
2111 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2112 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2113 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2114 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2115 	if (rc)
2116 		vi->rss_base = 0xffff;
2117 	else {
2118 		/* MPASS((val >> 16) == rss_size); */
2119 		vi->rss_base = val & 0xffff;
2120 	}
2121 
2122 	rc = cxgbe_vi_attach(dev, vi);
2123 	if (rc) {
2124 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2125 		return (rc);
2126 	}
2127 	return (0);
2128 }
2129 
2130 static int
2131 vcxgbe_detach(device_t dev)
2132 {
2133 	struct vi_info *vi;
2134 	struct adapter *sc;
2135 
2136 	vi = device_get_softc(dev);
2137 	sc = vi->pi->adapter;
2138 
2139 	doom_vi(sc, vi);
2140 
2141 	cxgbe_vi_detach(vi);
2142 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2143 
2144 	end_synchronized_op(sc, 0);
2145 
2146 	return (0);
2147 }
2148 
2149 void
2150 t4_fatal_err(struct adapter *sc)
2151 {
2152 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2153 	t4_intr_disable(sc);
2154 	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2155 	    device_get_nameunit(sc->dev));
2156 }
2157 
2158 void
2159 t4_add_adapter(struct adapter *sc)
2160 {
2161 	sx_xlock(&t4_list_lock);
2162 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2163 	sx_xunlock(&t4_list_lock);
2164 }
2165 
2166 int
2167 t4_map_bars_0_and_4(struct adapter *sc)
2168 {
2169 	sc->regs_rid = PCIR_BAR(0);
2170 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2171 	    &sc->regs_rid, RF_ACTIVE);
2172 	if (sc->regs_res == NULL) {
2173 		device_printf(sc->dev, "cannot map registers.\n");
2174 		return (ENXIO);
2175 	}
2176 	sc->bt = rman_get_bustag(sc->regs_res);
2177 	sc->bh = rman_get_bushandle(sc->regs_res);
2178 	sc->mmio_len = rman_get_size(sc->regs_res);
2179 	setbit(&sc->doorbells, DOORBELL_KDB);
2180 
2181 	sc->msix_rid = PCIR_BAR(4);
2182 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2183 	    &sc->msix_rid, RF_ACTIVE);
2184 	if (sc->msix_res == NULL) {
2185 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2186 		return (ENXIO);
2187 	}
2188 
2189 	return (0);
2190 }
2191 
2192 int
2193 t4_map_bar_2(struct adapter *sc)
2194 {
2195 
2196 	/*
2197 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2198 	 * to map it if RDMA is disabled.
2199 	 */
2200 	if (is_t4(sc) && sc->rdmacaps == 0)
2201 		return (0);
2202 
2203 	sc->udbs_rid = PCIR_BAR(2);
2204 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2205 	    &sc->udbs_rid, RF_ACTIVE);
2206 	if (sc->udbs_res == NULL) {
2207 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2208 		return (ENXIO);
2209 	}
2210 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2211 
2212 	if (chip_id(sc) >= CHELSIO_T5) {
2213 		setbit(&sc->doorbells, DOORBELL_UDB);
2214 #if defined(__i386__) || defined(__amd64__)
2215 		if (t5_write_combine) {
2216 			int rc, mode;
2217 
2218 			/*
2219 			 * Enable write combining on BAR2.  This is the
2220 			 * userspace doorbell BAR and is split into 128B
2221 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2222 			 * with an egress queue.  The first 64B has the doorbell
2223 			 * and the second 64B can be used to submit a tx work
2224 			 * request with an implicit doorbell.
2225 			 */
2226 
2227 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2228 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2229 			if (rc == 0) {
2230 				clrbit(&sc->doorbells, DOORBELL_UDB);
2231 				setbit(&sc->doorbells, DOORBELL_WCWR);
2232 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2233 			} else {
2234 				device_printf(sc->dev,
2235 				    "couldn't enable write combining: %d\n",
2236 				    rc);
2237 			}
2238 
2239 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2240 			t4_write_reg(sc, A_SGE_STAT_CFG,
2241 			    V_STATSOURCE_T5(7) | mode);
2242 		}
2243 #endif
2244 	}
2245 
2246 	return (0);
2247 }
2248 
2249 struct memwin_init {
2250 	uint32_t base;
2251 	uint32_t aperture;
2252 };
2253 
2254 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2255 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2256 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2257 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2258 };
2259 
2260 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2261 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2262 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2263 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2264 };
2265 
2266 static void
2267 setup_memwin(struct adapter *sc)
2268 {
2269 	const struct memwin_init *mw_init;
2270 	struct memwin *mw;
2271 	int i;
2272 	uint32_t bar0;
2273 
2274 	if (is_t4(sc)) {
2275 		/*
2276 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2277 		 * mechanism.  Works from within PCI passthrough environments
2278 		 * too, where rman_get_start() can return a different value.  We
2279 		 * need to program the T4 memory window decoders with the actual
2280 		 * addresses that will be coming across the PCIe link.
2281 		 */
2282 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2283 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2284 
2285 		mw_init = &t4_memwin[0];
2286 	} else {
2287 		/* T5+ use the relative offset inside the PCIe BAR */
2288 		bar0 = 0;
2289 
2290 		mw_init = &t5_memwin[0];
2291 	}
2292 
2293 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2294 		rw_init(&mw->mw_lock, "memory window access");
2295 		mw->mw_base = mw_init->base;
2296 		mw->mw_aperture = mw_init->aperture;
2297 		mw->mw_curpos = 0;
2298 		t4_write_reg(sc,
2299 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2300 		    (mw->mw_base + bar0) | V_BIR(0) |
2301 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2302 		rw_wlock(&mw->mw_lock);
2303 		position_memwin(sc, i, 0);
2304 		rw_wunlock(&mw->mw_lock);
2305 	}
2306 
2307 	/* flush */
2308 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2309 }
2310 
2311 /*
2312  * Positions the memory window at the given address in the card's address space.
2313  * There are some alignment requirements and the actual position may be at an
2314  * address prior to the requested address.  mw->mw_curpos always has the actual
2315  * position of the window.
2316  */
2317 static void
2318 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2319 {
2320 	struct memwin *mw;
2321 	uint32_t pf;
2322 	uint32_t reg;
2323 
2324 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2325 	mw = &sc->memwin[idx];
2326 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2327 
2328 	if (is_t4(sc)) {
2329 		pf = 0;
2330 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2331 	} else {
2332 		pf = V_PFNUM(sc->pf);
2333 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2334 	}
2335 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2336 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2337 	t4_read_reg(sc, reg);	/* flush */
2338 }
2339 
2340 static int
2341 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2342     int len, int rw)
2343 {
2344 	struct memwin *mw;
2345 	uint32_t mw_end, v;
2346 
2347 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2348 
2349 	/* Memory can only be accessed in naturally aligned 4 byte units */
2350 	if (addr & 3 || len & 3 || len <= 0)
2351 		return (EINVAL);
2352 
2353 	mw = &sc->memwin[idx];
2354 	while (len > 0) {
2355 		rw_rlock(&mw->mw_lock);
2356 		mw_end = mw->mw_curpos + mw->mw_aperture;
2357 		if (addr >= mw_end || addr < mw->mw_curpos) {
2358 			/* Will need to reposition the window */
2359 			if (!rw_try_upgrade(&mw->mw_lock)) {
2360 				rw_runlock(&mw->mw_lock);
2361 				rw_wlock(&mw->mw_lock);
2362 			}
2363 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2364 			position_memwin(sc, idx, addr);
2365 			rw_downgrade(&mw->mw_lock);
2366 			mw_end = mw->mw_curpos + mw->mw_aperture;
2367 		}
2368 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2369 		while (addr < mw_end && len > 0) {
2370 			if (rw == 0) {
2371 				v = t4_read_reg(sc, mw->mw_base + addr -
2372 				    mw->mw_curpos);
2373 				*val++ = le32toh(v);
2374 			} else {
2375 				v = *val++;
2376 				t4_write_reg(sc, mw->mw_base + addr -
2377 				    mw->mw_curpos, htole32(v));
2378 			}
2379 			addr += 4;
2380 			len -= 4;
2381 		}
2382 		rw_runlock(&mw->mw_lock);
2383 	}
2384 
2385 	return (0);
2386 }
2387 
2388 static inline int
2389 read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2390     int len)
2391 {
2392 
2393 	return (rw_via_memwin(sc, idx, addr, val, len, 0));
2394 }
2395 
2396 static inline int
2397 write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
2398     const uint32_t *val, int len)
2399 {
2400 
2401 	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
2402 }
2403 
2404 static int
2405 t4_range_cmp(const void *a, const void *b)
2406 {
2407 	return ((const struct t4_range *)a)->start -
2408 	       ((const struct t4_range *)b)->start;
2409 }
2410 
2411 /*
2412  * Verify that the memory range specified by the addr/len pair is valid within
2413  * the card's address space.
2414  */
2415 static int
2416 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2417 {
2418 	struct t4_range mem_ranges[4], *r, *next;
2419 	uint32_t em, addr_len;
2420 	int i, n, remaining;
2421 
2422 	/* Memory can only be accessed in naturally aligned 4 byte units */
2423 	if (addr & 3 || len & 3 || len <= 0)
2424 		return (EINVAL);
2425 
2426 	/* Enabled memories */
2427 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2428 
2429 	r = &mem_ranges[0];
2430 	n = 0;
2431 	bzero(r, sizeof(mem_ranges));
2432 	if (em & F_EDRAM0_ENABLE) {
2433 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2434 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2435 		if (r->size > 0) {
2436 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2437 			if (addr >= r->start &&
2438 			    addr + len <= r->start + r->size)
2439 				return (0);
2440 			r++;
2441 			n++;
2442 		}
2443 	}
2444 	if (em & F_EDRAM1_ENABLE) {
2445 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2446 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2447 		if (r->size > 0) {
2448 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2449 			if (addr >= r->start &&
2450 			    addr + len <= r->start + r->size)
2451 				return (0);
2452 			r++;
2453 			n++;
2454 		}
2455 	}
2456 	if (em & F_EXT_MEM_ENABLE) {
2457 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2458 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2459 		if (r->size > 0) {
2460 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2461 			if (addr >= r->start &&
2462 			    addr + len <= r->start + r->size)
2463 				return (0);
2464 			r++;
2465 			n++;
2466 		}
2467 	}
2468 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2469 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2470 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2471 		if (r->size > 0) {
2472 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2473 			if (addr >= r->start &&
2474 			    addr + len <= r->start + r->size)
2475 				return (0);
2476 			r++;
2477 			n++;
2478 		}
2479 	}
2480 	MPASS(n <= nitems(mem_ranges));
2481 
2482 	if (n > 1) {
2483 		/* Sort and merge the ranges. */
2484 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2485 
2486 		/* Start from index 0 and examine the next n - 1 entries. */
2487 		r = &mem_ranges[0];
2488 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2489 
2490 			MPASS(r->size > 0);	/* r is a valid entry. */
2491 			next = r + 1;
2492 			MPASS(next->size > 0);	/* and so is the next one. */
2493 
2494 			while (r->start + r->size >= next->start) {
2495 				/* Merge the next one into the current entry. */
2496 				r->size = max(r->start + r->size,
2497 				    next->start + next->size) - r->start;
2498 				n--;	/* One fewer entry in total. */
2499 				if (--remaining == 0)
2500 					goto done;	/* short circuit */
2501 				next++;
2502 			}
2503 			if (next != r + 1) {
2504 				/*
2505 				 * Some entries were merged into r and next
2506 				 * points to the first valid entry that couldn't
2507 				 * be merged.
2508 				 */
2509 				MPASS(next->size > 0);	/* must be valid */
2510 				memcpy(r + 1, next, remaining * sizeof(*r));
2511 #ifdef INVARIANTS
2512 				/*
2513 				 * This so that the foo->size assertion in the
2514 				 * next iteration of the loop do the right
2515 				 * thing for entries that were pulled up and are
2516 				 * no longer valid.
2517 				 */
2518 				MPASS(n < nitems(mem_ranges));
2519 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2520 				    sizeof(struct t4_range));
2521 #endif
2522 			}
2523 		}
2524 done:
2525 		/* Done merging the ranges. */
2526 		MPASS(n > 0);
2527 		r = &mem_ranges[0];
2528 		for (i = 0; i < n; i++, r++) {
2529 			if (addr >= r->start &&
2530 			    addr + len <= r->start + r->size)
2531 				return (0);
2532 		}
2533 	}
2534 
2535 	return (EFAULT);
2536 }
2537 
2538 static int
2539 fwmtype_to_hwmtype(int mtype)
2540 {
2541 
2542 	switch (mtype) {
2543 	case FW_MEMTYPE_EDC0:
2544 		return (MEM_EDC0);
2545 	case FW_MEMTYPE_EDC1:
2546 		return (MEM_EDC1);
2547 	case FW_MEMTYPE_EXTMEM:
2548 		return (MEM_MC0);
2549 	case FW_MEMTYPE_EXTMEM1:
2550 		return (MEM_MC1);
2551 	default:
2552 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2553 	}
2554 }
2555 
2556 /*
2557  * Verify that the memory range specified by the memtype/offset/len pair is
2558  * valid and lies entirely within the memtype specified.  The global address of
2559  * the start of the range is returned in addr.
2560  */
2561 static int
2562 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2563     uint32_t *addr)
2564 {
2565 	uint32_t em, addr_len, maddr;
2566 
2567 	/* Memory can only be accessed in naturally aligned 4 byte units */
2568 	if (off & 3 || len & 3 || len == 0)
2569 		return (EINVAL);
2570 
2571 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2572 	switch (fwmtype_to_hwmtype(mtype)) {
2573 	case MEM_EDC0:
2574 		if (!(em & F_EDRAM0_ENABLE))
2575 			return (EINVAL);
2576 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2577 		maddr = G_EDRAM0_BASE(addr_len) << 20;
2578 		break;
2579 	case MEM_EDC1:
2580 		if (!(em & F_EDRAM1_ENABLE))
2581 			return (EINVAL);
2582 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2583 		maddr = G_EDRAM1_BASE(addr_len) << 20;
2584 		break;
2585 	case MEM_MC:
2586 		if (!(em & F_EXT_MEM_ENABLE))
2587 			return (EINVAL);
2588 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2589 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2590 		break;
2591 	case MEM_MC1:
2592 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2593 			return (EINVAL);
2594 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2595 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2596 		break;
2597 	default:
2598 		return (EINVAL);
2599 	}
2600 
2601 	*addr = maddr + off;	/* global address */
2602 	return (validate_mem_range(sc, *addr, len));
2603 }
2604 
2605 static int
2606 fixup_devlog_params(struct adapter *sc)
2607 {
2608 	struct devlog_params *dparams = &sc->params.devlog;
2609 	int rc;
2610 
2611 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2612 	    dparams->size, &dparams->addr);
2613 
2614 	return (rc);
2615 }
2616 
2617 static int
2618 cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g, int num_vis,
2619     struct intrs_and_queues *iaq)
2620 {
2621 	int rc, itype, navail, nrxq10g, nrxq1g, n;
2622 	int nofldrxq10g = 0, nofldrxq1g = 0;
2623 
2624 	bzero(iaq, sizeof(*iaq));
2625 
2626 	iaq->ntxq10g = t4_ntxq10g;
2627 	iaq->ntxq1g = t4_ntxq1g;
2628 	iaq->ntxq_vi = t4_ntxq_vi;
2629 	iaq->nrxq10g = nrxq10g = t4_nrxq10g;
2630 	iaq->nrxq1g = nrxq1g = t4_nrxq1g;
2631 	iaq->nrxq_vi = t4_nrxq_vi;
2632 	iaq->rsrv_noflowq = t4_rsrv_noflowq;
2633 #ifdef TCP_OFFLOAD
2634 	if (is_offload(sc)) {
2635 		iaq->nofldtxq10g = t4_nofldtxq10g;
2636 		iaq->nofldtxq1g = t4_nofldtxq1g;
2637 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
2638 		iaq->nofldrxq10g = nofldrxq10g = t4_nofldrxq10g;
2639 		iaq->nofldrxq1g = nofldrxq1g = t4_nofldrxq1g;
2640 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
2641 	}
2642 #endif
2643 #ifdef DEV_NETMAP
2644 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
2645 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
2646 #endif
2647 
2648 	for (itype = INTR_MSIX; itype; itype >>= 1) {
2649 
2650 		if ((itype & t4_intr_types) == 0)
2651 			continue;	/* not allowed */
2652 
2653 		if (itype == INTR_MSIX)
2654 			navail = pci_msix_count(sc->dev);
2655 		else if (itype == INTR_MSI)
2656 			navail = pci_msi_count(sc->dev);
2657 		else
2658 			navail = 1;
2659 restart:
2660 		if (navail == 0)
2661 			continue;
2662 
2663 		iaq->intr_type = itype;
2664 		iaq->intr_flags_10g = 0;
2665 		iaq->intr_flags_1g = 0;
2666 
2667 		/*
2668 		 * Best option: an interrupt vector for errors, one for the
2669 		 * firmware event queue, and one for every rxq (NIC and TOE) of
2670 		 * every VI.  The VIs that support netmap use the same
2671 		 * interrupts for the NIC rx queues and the netmap rx queues
2672 		 * because only one set of queues is active at a time.
2673 		 */
2674 		iaq->nirq = T4_EXTRA_INTR;
2675 		iaq->nirq += n10g * (nrxq10g + nofldrxq10g);
2676 		iaq->nirq += n1g * (nrxq1g + nofldrxq1g);
2677 		iaq->nirq += (n10g + n1g) * (num_vis - 1) *
2678 		    max(iaq->nrxq_vi, iaq->nnmrxq_vi);	/* See comment above. */
2679 		iaq->nirq += (n10g + n1g) * (num_vis - 1) * iaq->nofldrxq_vi;
2680 		if (iaq->nirq <= navail &&
2681 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2682 			iaq->intr_flags_10g = INTR_ALL;
2683 			iaq->intr_flags_1g = INTR_ALL;
2684 			goto allocate;
2685 		}
2686 
2687 		/* Disable the VIs (and netmap) if there aren't enough intrs */
2688 		if (num_vis > 1) {
2689 			device_printf(sc->dev, "virtual interfaces disabled "
2690 			    "because num_vis=%u with current settings "
2691 			    "(nrxq10g=%u, nrxq1g=%u, nofldrxq10g=%u, "
2692 			    "nofldrxq1g=%u, nrxq_vi=%u nofldrxq_vi=%u, "
2693 			    "nnmrxq_vi=%u) would need %u interrupts but "
2694 			    "only %u are available.\n", num_vis, nrxq10g,
2695 			    nrxq1g, nofldrxq10g, nofldrxq1g, iaq->nrxq_vi,
2696 			    iaq->nofldrxq_vi, iaq->nnmrxq_vi, iaq->nirq,
2697 			    navail);
2698 			num_vis = 1;
2699 			iaq->ntxq_vi = iaq->nrxq_vi = 0;
2700 			iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
2701 			iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
2702 			goto restart;
2703 		}
2704 
2705 		/*
2706 		 * Second best option: a vector for errors, one for the firmware
2707 		 * event queue, and vectors for either all the NIC rx queues or
2708 		 * all the TOE rx queues.  The queues that don't get vectors
2709 		 * will forward their interrupts to those that do.
2710 		 */
2711 		iaq->nirq = T4_EXTRA_INTR;
2712 		if (nrxq10g >= nofldrxq10g) {
2713 			iaq->intr_flags_10g = INTR_RXQ;
2714 			iaq->nirq += n10g * nrxq10g;
2715 		} else {
2716 			iaq->intr_flags_10g = INTR_OFLD_RXQ;
2717 			iaq->nirq += n10g * nofldrxq10g;
2718 		}
2719 		if (nrxq1g >= nofldrxq1g) {
2720 			iaq->intr_flags_1g = INTR_RXQ;
2721 			iaq->nirq += n1g * nrxq1g;
2722 		} else {
2723 			iaq->intr_flags_1g = INTR_OFLD_RXQ;
2724 			iaq->nirq += n1g * nofldrxq1g;
2725 		}
2726 		if (iaq->nirq <= navail &&
2727 		    (itype != INTR_MSI || powerof2(iaq->nirq)))
2728 			goto allocate;
2729 
2730 		/*
2731 		 * Next best option: an interrupt vector for errors, one for the
2732 		 * firmware event queue, and at least one per main-VI.  At this
2733 		 * point we know we'll have to downsize nrxq and/or nofldrxq to
2734 		 * fit what's available to us.
2735 		 */
2736 		iaq->nirq = T4_EXTRA_INTR;
2737 		iaq->nirq += n10g + n1g;
2738 		if (iaq->nirq <= navail) {
2739 			int leftover = navail - iaq->nirq;
2740 
2741 			if (n10g > 0) {
2742 				int target = max(nrxq10g, nofldrxq10g);
2743 
2744 				iaq->intr_flags_10g = nrxq10g >= nofldrxq10g ?
2745 				    INTR_RXQ : INTR_OFLD_RXQ;
2746 
2747 				n = 1;
2748 				while (n < target && leftover >= n10g) {
2749 					leftover -= n10g;
2750 					iaq->nirq += n10g;
2751 					n++;
2752 				}
2753 				iaq->nrxq10g = min(n, nrxq10g);
2754 #ifdef TCP_OFFLOAD
2755 				iaq->nofldrxq10g = min(n, nofldrxq10g);
2756 #endif
2757 			}
2758 
2759 			if (n1g > 0) {
2760 				int target = max(nrxq1g, nofldrxq1g);
2761 
2762 				iaq->intr_flags_1g = nrxq1g >= nofldrxq1g ?
2763 				    INTR_RXQ : INTR_OFLD_RXQ;
2764 
2765 				n = 1;
2766 				while (n < target && leftover >= n1g) {
2767 					leftover -= n1g;
2768 					iaq->nirq += n1g;
2769 					n++;
2770 				}
2771 				iaq->nrxq1g = min(n, nrxq1g);
2772 #ifdef TCP_OFFLOAD
2773 				iaq->nofldrxq1g = min(n, nofldrxq1g);
2774 #endif
2775 			}
2776 
2777 			if (itype != INTR_MSI || powerof2(iaq->nirq))
2778 				goto allocate;
2779 		}
2780 
2781 		/*
2782 		 * Least desirable option: one interrupt vector for everything.
2783 		 */
2784 		iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2785 		iaq->intr_flags_10g = iaq->intr_flags_1g = 0;
2786 #ifdef TCP_OFFLOAD
2787 		if (is_offload(sc))
2788 			iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2789 #endif
2790 allocate:
2791 		navail = iaq->nirq;
2792 		rc = 0;
2793 		if (itype == INTR_MSIX)
2794 			rc = pci_alloc_msix(sc->dev, &navail);
2795 		else if (itype == INTR_MSI)
2796 			rc = pci_alloc_msi(sc->dev, &navail);
2797 
2798 		if (rc == 0) {
2799 			if (navail == iaq->nirq)
2800 				return (0);
2801 
2802 			/*
2803 			 * Didn't get the number requested.  Use whatever number
2804 			 * the kernel is willing to allocate (it's in navail).
2805 			 */
2806 			device_printf(sc->dev, "fewer vectors than requested, "
2807 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2808 			    itype, iaq->nirq, navail);
2809 			pci_release_msi(sc->dev);
2810 			goto restart;
2811 		}
2812 
2813 		device_printf(sc->dev,
2814 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2815 		    itype, rc, iaq->nirq, navail);
2816 	}
2817 
2818 	device_printf(sc->dev,
2819 	    "failed to find a usable interrupt type.  "
2820 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2821 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2822 
2823 	return (ENXIO);
2824 }
2825 
2826 #define FW_VERSION(chip) ( \
2827     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2828     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2829     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2830     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2831 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2832 
2833 struct fw_info {
2834 	uint8_t chip;
2835 	char *kld_name;
2836 	char *fw_mod_name;
2837 	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
2838 } fw_info[] = {
2839 	{
2840 		.chip = CHELSIO_T4,
2841 		.kld_name = "t4fw_cfg",
2842 		.fw_mod_name = "t4fw",
2843 		.fw_hdr = {
2844 			.chip = FW_HDR_CHIP_T4,
2845 			.fw_ver = htobe32_const(FW_VERSION(T4)),
2846 			.intfver_nic = FW_INTFVER(T4, NIC),
2847 			.intfver_vnic = FW_INTFVER(T4, VNIC),
2848 			.intfver_ofld = FW_INTFVER(T4, OFLD),
2849 			.intfver_ri = FW_INTFVER(T4, RI),
2850 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2851 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
2852 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2853 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
2854 		},
2855 	}, {
2856 		.chip = CHELSIO_T5,
2857 		.kld_name = "t5fw_cfg",
2858 		.fw_mod_name = "t5fw",
2859 		.fw_hdr = {
2860 			.chip = FW_HDR_CHIP_T5,
2861 			.fw_ver = htobe32_const(FW_VERSION(T5)),
2862 			.intfver_nic = FW_INTFVER(T5, NIC),
2863 			.intfver_vnic = FW_INTFVER(T5, VNIC),
2864 			.intfver_ofld = FW_INTFVER(T5, OFLD),
2865 			.intfver_ri = FW_INTFVER(T5, RI),
2866 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
2867 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
2868 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
2869 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
2870 		},
2871 	}, {
2872 		.chip = CHELSIO_T6,
2873 		.kld_name = "t6fw_cfg",
2874 		.fw_mod_name = "t6fw",
2875 		.fw_hdr = {
2876 			.chip = FW_HDR_CHIP_T6,
2877 			.fw_ver = htobe32_const(FW_VERSION(T6)),
2878 			.intfver_nic = FW_INTFVER(T6, NIC),
2879 			.intfver_vnic = FW_INTFVER(T6, VNIC),
2880 			.intfver_ofld = FW_INTFVER(T6, OFLD),
2881 			.intfver_ri = FW_INTFVER(T6, RI),
2882 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
2883 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
2884 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
2885 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
2886 		},
2887 	}
2888 };
2889 
2890 static struct fw_info *
2891 find_fw_info(int chip)
2892 {
2893 	int i;
2894 
2895 	for (i = 0; i < nitems(fw_info); i++) {
2896 		if (fw_info[i].chip == chip)
2897 			return (&fw_info[i]);
2898 	}
2899 	return (NULL);
2900 }
2901 
2902 /*
2903  * Is the given firmware API compatible with the one the driver was compiled
2904  * with?
2905  */
2906 static int
2907 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
2908 {
2909 
2910 	/* short circuit if it's the exact same firmware version */
2911 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
2912 		return (1);
2913 
2914 	/*
2915 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
2916 	 * features that are supported in the driver.
2917 	 */
2918 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
2919 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
2920 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
2921 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
2922 		return (1);
2923 #undef SAME_INTF
2924 
2925 	return (0);
2926 }
2927 
2928 /*
2929  * The firmware in the KLD is usable, but should it be installed?  This routine
2930  * explains itself in detail if it indicates the KLD firmware should be
2931  * installed.
2932  */
2933 static int
2934 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
2935 {
2936 	const char *reason;
2937 
2938 	if (!card_fw_usable) {
2939 		reason = "incompatible or unusable";
2940 		goto install;
2941 	}
2942 
2943 	if (k > c) {
2944 		reason = "older than the version bundled with this driver";
2945 		goto install;
2946 	}
2947 
2948 	if (t4_fw_install == 2 && k != c) {
2949 		reason = "different than the version bundled with this driver";
2950 		goto install;
2951 	}
2952 
2953 	return (0);
2954 
2955 install:
2956 	if (t4_fw_install == 0) {
2957 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2958 		    "but the driver is prohibited from installing a different "
2959 		    "firmware on the card.\n",
2960 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2961 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
2962 
2963 		return (0);
2964 	}
2965 
2966 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2967 	    "installing firmware %u.%u.%u.%u on card.\n",
2968 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2969 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
2970 	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2971 	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2972 
2973 	return (1);
2974 }
2975 
2976 /*
2977  * Establish contact with the firmware and determine if we are the master driver
2978  * or not, and whether we are responsible for chip initialization.
2979  */
2980 static int
2981 prep_firmware(struct adapter *sc)
2982 {
2983 	const struct firmware *fw = NULL, *default_cfg;
2984 	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
2985 	enum dev_state state;
2986 	struct fw_info *fw_info;
2987 	struct fw_hdr *card_fw;		/* fw on the card */
2988 	const struct fw_hdr *kld_fw;	/* fw in the KLD */
2989 	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
2990 					   against */
2991 
2992 	/* This is the firmware whose headers the driver was compiled against */
2993 	fw_info = find_fw_info(chip_id(sc));
2994 	if (fw_info == NULL) {
2995 		device_printf(sc->dev,
2996 		    "unable to look up firmware information for chip %d.\n",
2997 		    chip_id(sc));
2998 		return (EINVAL);
2999 	}
3000 	drv_fw = &fw_info->fw_hdr;
3001 
3002 	/*
3003 	 * The firmware KLD contains many modules.  The KLD name is also the
3004 	 * name of the module that contains the default config file.
3005 	 */
3006 	default_cfg = firmware_get(fw_info->kld_name);
3007 
3008 	/* This is the firmware in the KLD */
3009 	fw = firmware_get(fw_info->fw_mod_name);
3010 	if (fw != NULL) {
3011 		kld_fw = (const void *)fw->data;
3012 		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3013 	} else {
3014 		kld_fw = NULL;
3015 		kld_fw_usable = 0;
3016 	}
3017 
3018 	/* Read the header of the firmware on the card */
3019 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3020 	rc = -t4_read_flash(sc, FLASH_FW_START,
3021 	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
3022 	if (rc == 0) {
3023 		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
3024 		if (card_fw->fw_ver == be32toh(0xffffffff)) {
3025 			uint32_t d = be32toh(kld_fw->fw_ver);
3026 
3027 			if (!kld_fw_usable) {
3028 				device_printf(sc->dev,
3029 				    "no firmware on the card and no usable "
3030 				    "firmware bundled with the driver.\n");
3031 				rc = EIO;
3032 				goto done;
3033 			} else if (t4_fw_install == 0) {
3034 				device_printf(sc->dev,
3035 				    "no firmware on the card and the driver "
3036 				    "is prohibited from installing new "
3037 				    "firmware.\n");
3038 				rc = EIO;
3039 				goto done;
3040 			}
3041 
3042 			device_printf(sc->dev, "no firmware on the card, "
3043 			    "installing firmware %d.%d.%d.%d\n",
3044 			    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3045 			    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3046 			rc = t4_fw_forceinstall(sc, fw->data, fw->datasize);
3047 			if (rc < 0) {
3048 				rc = -rc;
3049 				device_printf(sc->dev,
3050 				    "firmware install failed: %d.\n", rc);
3051 				goto done;
3052 			}
3053 			memcpy(card_fw, kld_fw, sizeof(*card_fw));
3054 			card_fw_usable = 1;
3055 			need_fw_reset = 0;
3056 		}
3057 	} else {
3058 		device_printf(sc->dev,
3059 		    "Unable to read card's firmware header: %d\n", rc);
3060 		card_fw_usable = 0;
3061 	}
3062 
3063 	/* Contact firmware. */
3064 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3065 	if (rc < 0 || state == DEV_STATE_ERR) {
3066 		rc = -rc;
3067 		device_printf(sc->dev,
3068 		    "failed to connect to the firmware: %d, %d.\n", rc, state);
3069 		goto done;
3070 	}
3071 	pf = rc;
3072 	if (pf == sc->mbox)
3073 		sc->flags |= MASTER_PF;
3074 	else if (state == DEV_STATE_UNINIT) {
3075 		/*
3076 		 * We didn't get to be the master so we definitely won't be
3077 		 * configuring the chip.  It's a bug if someone else hasn't
3078 		 * configured it already.
3079 		 */
3080 		device_printf(sc->dev, "couldn't be master(%d), "
3081 		    "device not already initialized either(%d).\n", rc, state);
3082 		rc = EPROTO;
3083 		goto done;
3084 	}
3085 
3086 	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3087 	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3088 		/*
3089 		 * Common case: the firmware on the card is an exact match and
3090 		 * the KLD is an exact match too, or the KLD is
3091 		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
3092 		 * here -- use cxgbetool loadfw if you want to reinstall the
3093 		 * same firmware as the one on the card.
3094 		 */
3095 	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3096 	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3097 	    be32toh(card_fw->fw_ver))) {
3098 
3099 		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3100 		if (rc != 0) {
3101 			device_printf(sc->dev,
3102 			    "failed to install firmware: %d\n", rc);
3103 			goto done;
3104 		}
3105 
3106 		/* Installed successfully, update the cached header too. */
3107 		memcpy(card_fw, kld_fw, sizeof(*card_fw));
3108 		card_fw_usable = 1;
3109 		need_fw_reset = 0;	/* already reset as part of load_fw */
3110 	}
3111 
3112 	if (!card_fw_usable) {
3113 		uint32_t d, c, k;
3114 
3115 		d = ntohl(drv_fw->fw_ver);
3116 		c = ntohl(card_fw->fw_ver);
3117 		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3118 
3119 		device_printf(sc->dev, "Cannot find a usable firmware: "
3120 		    "fw_install %d, chip state %d, "
3121 		    "driver compiled with %d.%d.%d.%d, "
3122 		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3123 		    t4_fw_install, state,
3124 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3125 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3126 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3127 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3128 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3129 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3130 		rc = EINVAL;
3131 		goto done;
3132 	}
3133 
3134 	/* Reset device */
3135 	if (need_fw_reset &&
3136 	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3137 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3138 		if (rc != ETIMEDOUT && rc != EIO)
3139 			t4_fw_bye(sc, sc->mbox);
3140 		goto done;
3141 	}
3142 	sc->flags |= FW_OK;
3143 
3144 	rc = get_params__pre_init(sc);
3145 	if (rc != 0)
3146 		goto done; /* error message displayed already */
3147 
3148 	/* Partition adapter resources as specified in the config file. */
3149 	if (state == DEV_STATE_UNINIT) {
3150 
3151 		KASSERT(sc->flags & MASTER_PF,
3152 		    ("%s: trying to change chip settings when not master.",
3153 		    __func__));
3154 
3155 		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3156 		if (rc != 0)
3157 			goto done;	/* error message displayed already */
3158 
3159 		t4_tweak_chip_settings(sc);
3160 
3161 		/* get basic stuff going */
3162 		rc = -t4_fw_initialize(sc, sc->mbox);
3163 		if (rc != 0) {
3164 			device_printf(sc->dev, "fw init failed: %d.\n", rc);
3165 			goto done;
3166 		}
3167 	} else {
3168 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3169 		sc->cfcsum = 0;
3170 	}
3171 
3172 done:
3173 	free(card_fw, M_CXGBE);
3174 	if (fw != NULL)
3175 		firmware_put(fw, FIRMWARE_UNLOAD);
3176 	if (default_cfg != NULL)
3177 		firmware_put(default_cfg, FIRMWARE_UNLOAD);
3178 
3179 	return (rc);
3180 }
3181 
3182 #define FW_PARAM_DEV(param) \
3183 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3184 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3185 #define FW_PARAM_PFVF(param) \
3186 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3187 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3188 
3189 /*
3190  * Partition chip resources for use between various PFs, VFs, etc.
3191  */
3192 static int
3193 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3194     const char *name_prefix)
3195 {
3196 	const struct firmware *cfg = NULL;
3197 	int rc = 0;
3198 	struct fw_caps_config_cmd caps;
3199 	uint32_t mtype, moff, finicsum, cfcsum;
3200 
3201 	/*
3202 	 * Figure out what configuration file to use.  Pick the default config
3203 	 * file for the card if the user hasn't specified one explicitly.
3204 	 */
3205 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3206 	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3207 		/* Card specific overrides go here. */
3208 		if (pci_get_device(sc->dev) == 0x440a)
3209 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3210 		if (is_fpga(sc))
3211 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3212 	}
3213 
3214 	/*
3215 	 * We need to load another module if the profile is anything except
3216 	 * "default" or "flash".
3217 	 */
3218 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3219 	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3220 		char s[32];
3221 
3222 		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3223 		cfg = firmware_get(s);
3224 		if (cfg == NULL) {
3225 			if (default_cfg != NULL) {
3226 				device_printf(sc->dev,
3227 				    "unable to load module \"%s\" for "
3228 				    "configuration profile \"%s\", will use "
3229 				    "the default config file instead.\n",
3230 				    s, sc->cfg_file);
3231 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3232 				    "%s", DEFAULT_CF);
3233 			} else {
3234 				device_printf(sc->dev,
3235 				    "unable to load module \"%s\" for "
3236 				    "configuration profile \"%s\", will use "
3237 				    "the config file on the card's flash "
3238 				    "instead.\n", s, sc->cfg_file);
3239 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3240 				    "%s", FLASH_CF);
3241 			}
3242 		}
3243 	}
3244 
3245 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3246 	    default_cfg == NULL) {
3247 		device_printf(sc->dev,
3248 		    "default config file not available, will use the config "
3249 		    "file on the card's flash instead.\n");
3250 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3251 	}
3252 
3253 	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3254 		u_int cflen;
3255 		const uint32_t *cfdata;
3256 		uint32_t param, val, addr;
3257 
3258 		KASSERT(cfg != NULL || default_cfg != NULL,
3259 		    ("%s: no config to upload", __func__));
3260 
3261 		/*
3262 		 * Ask the firmware where it wants us to upload the config file.
3263 		 */
3264 		param = FW_PARAM_DEV(CF);
3265 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3266 		if (rc != 0) {
3267 			/* No support for config file?  Shouldn't happen. */
3268 			device_printf(sc->dev,
3269 			    "failed to query config file location: %d.\n", rc);
3270 			goto done;
3271 		}
3272 		mtype = G_FW_PARAMS_PARAM_Y(val);
3273 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3274 
3275 		/*
3276 		 * XXX: sheer laziness.  We deliberately added 4 bytes of
3277 		 * useless stuffing/comments at the end of the config file so
3278 		 * it's ok to simply throw away the last remaining bytes when
3279 		 * the config file is not an exact multiple of 4.  This also
3280 		 * helps with the validate_mt_off_len check.
3281 		 */
3282 		if (cfg != NULL) {
3283 			cflen = cfg->datasize & ~3;
3284 			cfdata = cfg->data;
3285 		} else {
3286 			cflen = default_cfg->datasize & ~3;
3287 			cfdata = default_cfg->data;
3288 		}
3289 
3290 		if (cflen > FLASH_CFG_MAX_SIZE) {
3291 			device_printf(sc->dev,
3292 			    "config file too long (%d, max allowed is %d).  "
3293 			    "Will try to use the config on the card, if any.\n",
3294 			    cflen, FLASH_CFG_MAX_SIZE);
3295 			goto use_config_on_flash;
3296 		}
3297 
3298 		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3299 		if (rc != 0) {
3300 			device_printf(sc->dev,
3301 			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3302 			    "Will try to use the config on the card, if any.\n",
3303 			    __func__, mtype, moff, cflen, rc);
3304 			goto use_config_on_flash;
3305 		}
3306 		write_via_memwin(sc, 2, addr, cfdata, cflen);
3307 	} else {
3308 use_config_on_flash:
3309 		mtype = FW_MEMTYPE_FLASH;
3310 		moff = t4_flash_cfg_addr(sc);
3311 	}
3312 
3313 	bzero(&caps, sizeof(caps));
3314 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3315 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3316 	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3317 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3318 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3319 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3320 	if (rc != 0) {
3321 		device_printf(sc->dev,
3322 		    "failed to pre-process config file: %d "
3323 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3324 		goto done;
3325 	}
3326 
3327 	finicsum = be32toh(caps.finicsum);
3328 	cfcsum = be32toh(caps.cfcsum);
3329 	if (finicsum != cfcsum) {
3330 		device_printf(sc->dev,
3331 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3332 		    finicsum, cfcsum);
3333 	}
3334 	sc->cfcsum = cfcsum;
3335 
3336 #define LIMIT_CAPS(x) do { \
3337 	caps.x &= htobe16(t4_##x##_allowed); \
3338 } while (0)
3339 
3340 	/*
3341 	 * Let the firmware know what features will (not) be used so it can tune
3342 	 * things accordingly.
3343 	 */
3344 	LIMIT_CAPS(nbmcaps);
3345 	LIMIT_CAPS(linkcaps);
3346 	LIMIT_CAPS(switchcaps);
3347 	LIMIT_CAPS(niccaps);
3348 	LIMIT_CAPS(toecaps);
3349 	LIMIT_CAPS(rdmacaps);
3350 	LIMIT_CAPS(cryptocaps);
3351 	LIMIT_CAPS(iscsicaps);
3352 	LIMIT_CAPS(fcoecaps);
3353 #undef LIMIT_CAPS
3354 
3355 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3356 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3357 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3358 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3359 	if (rc != 0) {
3360 		device_printf(sc->dev,
3361 		    "failed to process config file: %d.\n", rc);
3362 	}
3363 done:
3364 	if (cfg != NULL)
3365 		firmware_put(cfg, FIRMWARE_UNLOAD);
3366 	return (rc);
3367 }
3368 
3369 /*
3370  * Retrieve parameters that are needed (or nice to have) very early.
3371  */
3372 static int
3373 get_params__pre_init(struct adapter *sc)
3374 {
3375 	int rc;
3376 	uint32_t param[2], val[2];
3377 
3378 	t4_get_version_info(sc);
3379 
3380 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3381 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3382 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3383 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3384 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3385 
3386 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3387 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3388 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3389 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3390 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3391 
3392 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3393 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3394 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3395 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3396 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3397 
3398 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3399 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3400 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3401 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3402 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3403 
3404 	param[0] = FW_PARAM_DEV(PORTVEC);
3405 	param[1] = FW_PARAM_DEV(CCLK);
3406 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3407 	if (rc != 0) {
3408 		device_printf(sc->dev,
3409 		    "failed to query parameters (pre_init): %d.\n", rc);
3410 		return (rc);
3411 	}
3412 
3413 	sc->params.portvec = val[0];
3414 	sc->params.nports = bitcount32(val[0]);
3415 	sc->params.vpd.cclk = val[1];
3416 
3417 	/* Read device log parameters. */
3418 	rc = -t4_init_devlog_params(sc, 1);
3419 	if (rc == 0)
3420 		fixup_devlog_params(sc);
3421 	else {
3422 		device_printf(sc->dev,
3423 		    "failed to get devlog parameters: %d.\n", rc);
3424 		rc = 0;	/* devlog isn't critical for device operation */
3425 	}
3426 
3427 	return (rc);
3428 }
3429 
3430 /*
3431  * Retrieve various parameters that are of interest to the driver.  The device
3432  * has been initialized by the firmware at this point.
3433  */
3434 static int
3435 get_params__post_init(struct adapter *sc)
3436 {
3437 	int rc;
3438 	uint32_t param[7], val[7];
3439 	struct fw_caps_config_cmd caps;
3440 
3441 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3442 	param[1] = FW_PARAM_PFVF(EQ_START);
3443 	param[2] = FW_PARAM_PFVF(FILTER_START);
3444 	param[3] = FW_PARAM_PFVF(FILTER_END);
3445 	param[4] = FW_PARAM_PFVF(L2T_START);
3446 	param[5] = FW_PARAM_PFVF(L2T_END);
3447 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3448 	if (rc != 0) {
3449 		device_printf(sc->dev,
3450 		    "failed to query parameters (post_init): %d.\n", rc);
3451 		return (rc);
3452 	}
3453 
3454 	sc->sge.iq_start = val[0];
3455 	sc->sge.eq_start = val[1];
3456 	sc->tids.ftid_base = val[2];
3457 	sc->tids.nftids = val[3] - val[2] + 1;
3458 	sc->params.ftid_min = val[2];
3459 	sc->params.ftid_max = val[3];
3460 	sc->vres.l2t.start = val[4];
3461 	sc->vres.l2t.size = val[5] - val[4] + 1;
3462 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3463 	    ("%s: L2 table size (%u) larger than expected (%u)",
3464 	    __func__, sc->vres.l2t.size, L2T_SIZE));
3465 
3466 	/* get capabilites */
3467 	bzero(&caps, sizeof(caps));
3468 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3469 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3470 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3471 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3472 	if (rc != 0) {
3473 		device_printf(sc->dev,
3474 		    "failed to get card capabilities: %d.\n", rc);
3475 		return (rc);
3476 	}
3477 
3478 #define READ_CAPS(x) do { \
3479 	sc->x = htobe16(caps.x); \
3480 } while (0)
3481 	READ_CAPS(nbmcaps);
3482 	READ_CAPS(linkcaps);
3483 	READ_CAPS(switchcaps);
3484 	READ_CAPS(niccaps);
3485 	READ_CAPS(toecaps);
3486 	READ_CAPS(rdmacaps);
3487 	READ_CAPS(cryptocaps);
3488 	READ_CAPS(iscsicaps);
3489 	READ_CAPS(fcoecaps);
3490 
3491 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3492 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3493 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3494 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3495 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3496 		if (rc != 0) {
3497 			device_printf(sc->dev,
3498 			    "failed to query NIC parameters: %d.\n", rc);
3499 			return (rc);
3500 		}
3501 		sc->tids.etid_base = val[0];
3502 		sc->params.etid_min = val[0];
3503 		sc->tids.netids = val[1] - val[0] + 1;
3504 		sc->params.netids = sc->tids.netids;
3505 		sc->params.eo_wr_cred = val[2];
3506 		sc->params.ethoffload = 1;
3507 	}
3508 
3509 	if (sc->toecaps) {
3510 		/* query offload-related parameters */
3511 		param[0] = FW_PARAM_DEV(NTID);
3512 		param[1] = FW_PARAM_PFVF(SERVER_START);
3513 		param[2] = FW_PARAM_PFVF(SERVER_END);
3514 		param[3] = FW_PARAM_PFVF(TDDP_START);
3515 		param[4] = FW_PARAM_PFVF(TDDP_END);
3516 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3517 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3518 		if (rc != 0) {
3519 			device_printf(sc->dev,
3520 			    "failed to query TOE parameters: %d.\n", rc);
3521 			return (rc);
3522 		}
3523 		sc->tids.ntids = val[0];
3524 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3525 		sc->tids.stid_base = val[1];
3526 		sc->tids.nstids = val[2] - val[1] + 1;
3527 		sc->vres.ddp.start = val[3];
3528 		sc->vres.ddp.size = val[4] - val[3] + 1;
3529 		sc->params.ofldq_wr_cred = val[5];
3530 		sc->params.offload = 1;
3531 	}
3532 	if (sc->rdmacaps) {
3533 		param[0] = FW_PARAM_PFVF(STAG_START);
3534 		param[1] = FW_PARAM_PFVF(STAG_END);
3535 		param[2] = FW_PARAM_PFVF(RQ_START);
3536 		param[3] = FW_PARAM_PFVF(RQ_END);
3537 		param[4] = FW_PARAM_PFVF(PBL_START);
3538 		param[5] = FW_PARAM_PFVF(PBL_END);
3539 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3540 		if (rc != 0) {
3541 			device_printf(sc->dev,
3542 			    "failed to query RDMA parameters(1): %d.\n", rc);
3543 			return (rc);
3544 		}
3545 		sc->vres.stag.start = val[0];
3546 		sc->vres.stag.size = val[1] - val[0] + 1;
3547 		sc->vres.rq.start = val[2];
3548 		sc->vres.rq.size = val[3] - val[2] + 1;
3549 		sc->vres.pbl.start = val[4];
3550 		sc->vres.pbl.size = val[5] - val[4] + 1;
3551 
3552 		param[0] = FW_PARAM_PFVF(SQRQ_START);
3553 		param[1] = FW_PARAM_PFVF(SQRQ_END);
3554 		param[2] = FW_PARAM_PFVF(CQ_START);
3555 		param[3] = FW_PARAM_PFVF(CQ_END);
3556 		param[4] = FW_PARAM_PFVF(OCQ_START);
3557 		param[5] = FW_PARAM_PFVF(OCQ_END);
3558 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3559 		if (rc != 0) {
3560 			device_printf(sc->dev,
3561 			    "failed to query RDMA parameters(2): %d.\n", rc);
3562 			return (rc);
3563 		}
3564 		sc->vres.qp.start = val[0];
3565 		sc->vres.qp.size = val[1] - val[0] + 1;
3566 		sc->vres.cq.start = val[2];
3567 		sc->vres.cq.size = val[3] - val[2] + 1;
3568 		sc->vres.ocq.start = val[4];
3569 		sc->vres.ocq.size = val[5] - val[4] + 1;
3570 
3571 		param[0] = FW_PARAM_PFVF(SRQ_START);
3572 		param[1] = FW_PARAM_PFVF(SRQ_END);
3573 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
3574 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
3575 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
3576 		if (rc != 0) {
3577 			device_printf(sc->dev,
3578 			    "failed to query RDMA parameters(3): %d.\n", rc);
3579 			return (rc);
3580 		}
3581 		sc->vres.srq.start = val[0];
3582 		sc->vres.srq.size = val[1] - val[0] + 1;
3583 		sc->params.max_ordird_qp = val[2];
3584 		sc->params.max_ird_adapter = val[3];
3585 	}
3586 	if (sc->iscsicaps) {
3587 		param[0] = FW_PARAM_PFVF(ISCSI_START);
3588 		param[1] = FW_PARAM_PFVF(ISCSI_END);
3589 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3590 		if (rc != 0) {
3591 			device_printf(sc->dev,
3592 			    "failed to query iSCSI parameters: %d.\n", rc);
3593 			return (rc);
3594 		}
3595 		sc->vres.iscsi.start = val[0];
3596 		sc->vres.iscsi.size = val[1] - val[0] + 1;
3597 	}
3598 
3599 	t4_init_sge_params(sc);
3600 
3601 	/*
3602 	 * We've got the params we wanted to query via the firmware.  Now grab
3603 	 * some others directly from the chip.
3604 	 */
3605 	rc = t4_read_chip_settings(sc);
3606 
3607 	return (rc);
3608 }
3609 
3610 static int
3611 set_params__post_init(struct adapter *sc)
3612 {
3613 	uint32_t param, val;
3614 	int i, v, shift;
3615 	char s[32];
3616 
3617 	/* ask for encapsulated CPLs */
3618 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3619 	val = 1;
3620 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3621 
3622 	/*
3623 	 * Override the TOE timers with user provided tunables.  This is not the
3624 	 * recommended way to change the timers (the firmware config file is) so
3625 	 * these tunables are not documented.
3626 	 *
3627 	 * All the timer tunables are in milliseconds.
3628 	 */
3629 	if (TUNABLE_INT_FETCH("hw.cxgbe.toe.keepalive_idle", &v)) {
3630 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
3631 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE),
3632 		    V_KEEPALIVEIDLE(ms_to_tcp_ticks(sc, v)));
3633 	}
3634 	if (TUNABLE_INT_FETCH("hw.cxgbe.toe.keepalive_interval", &v)) {
3635 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
3636 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL),
3637 		    V_KEEPALIVEINTVL(ms_to_tcp_ticks(sc, v)));
3638 	}
3639 	if (TUNABLE_INT_FETCH("hw.cxgbe.toe.keepalive_count", &v)) {
3640 		v &= M_KEEPALIVEMAXR1;
3641 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
3642 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
3643 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
3644 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
3645 	}
3646 	if (TUNABLE_INT_FETCH("hw.cxgbe.toe.rexmt_min", &v)) {
3647 		t4_set_reg_field(sc, A_TP_RXT_MIN,
3648 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(ms_to_tcp_ticks(sc, v)));
3649 	}
3650 	if (TUNABLE_INT_FETCH("hw.cxgbe.toe.rexmt_max", &v)) {
3651 		t4_set_reg_field(sc, A_TP_RXT_MAX,
3652 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(ms_to_tcp_ticks(sc, v)));
3653 	}
3654 	if (TUNABLE_INT_FETCH("hw.cxgbe.toe.rexmt_count", &v)) {
3655 		v &= M_RXTSHIFTMAXR1;
3656 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
3657 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
3658 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
3659 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
3660 	}
3661 	for (i = 0; i < 16; i++) {
3662 		snprintf(s, sizeof(s), "hw.cxgbe.toe.rexmt_backoff.%d", i);
3663 		if (TUNABLE_INT_FETCH(s, &v)) {
3664 			v &= M_TIMERBACKOFFINDEX0;
3665 			shift = (i & 3) << 3;
3666 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
3667 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
3668 		}
3669 	}
3670 	return (0);
3671 }
3672 
3673 #undef FW_PARAM_PFVF
3674 #undef FW_PARAM_DEV
3675 
3676 static void
3677 t4_set_desc(struct adapter *sc)
3678 {
3679 	char buf[128];
3680 	struct adapter_params *p = &sc->params;
3681 
3682 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
3683 
3684 	device_set_desc_copy(sc->dev, buf);
3685 }
3686 
3687 static void
3688 build_medialist(struct port_info *pi, struct ifmedia *media)
3689 {
3690 	int m;
3691 
3692 	PORT_LOCK_ASSERT_OWNED(pi);
3693 
3694 	ifmedia_removeall(media);
3695 
3696 	/*
3697 	 * XXX: Would it be better to ifmedia_add all 4 combinations of pause
3698 	 * settings for every speed instead of just txpause|rxpause?  ifconfig
3699 	 * media display looks much better if autoselect is the only case where
3700 	 * ifm_current is different from ifm_active.  If the user picks anything
3701 	 * except txpause|rxpause the display is ugly.
3702 	 */
3703 	m = IFM_ETHER | IFM_FDX | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE;
3704 
3705 	switch(pi->port_type) {
3706 	case FW_PORT_TYPE_BT_XFI:
3707 	case FW_PORT_TYPE_BT_XAUI:
3708 		ifmedia_add(media, m | IFM_10G_T, 0, NULL);
3709 		/* fall through */
3710 
3711 	case FW_PORT_TYPE_BT_SGMII:
3712 		ifmedia_add(media, m | IFM_1000_T, 0, NULL);
3713 		ifmedia_add(media, m | IFM_100_TX, 0, NULL);
3714 		ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL);
3715 		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
3716 		break;
3717 
3718 	case FW_PORT_TYPE_CX4:
3719 		ifmedia_add(media, m | IFM_10G_CX4, 0, NULL);
3720 		ifmedia_set(media, m | IFM_10G_CX4);
3721 		break;
3722 
3723 	case FW_PORT_TYPE_QSFP_10G:
3724 	case FW_PORT_TYPE_SFP:
3725 	case FW_PORT_TYPE_FIBER_XFI:
3726 	case FW_PORT_TYPE_FIBER_XAUI:
3727 		switch (pi->mod_type) {
3728 
3729 		case FW_PORT_MOD_TYPE_LR:
3730 			ifmedia_add(media, m | IFM_10G_LR, 0, NULL);
3731 			ifmedia_set(media, m | IFM_10G_LR);
3732 			break;
3733 
3734 		case FW_PORT_MOD_TYPE_SR:
3735 			ifmedia_add(media, m | IFM_10G_SR, 0, NULL);
3736 			ifmedia_set(media, m | IFM_10G_SR);
3737 			break;
3738 
3739 		case FW_PORT_MOD_TYPE_LRM:
3740 			ifmedia_add(media, m | IFM_10G_LRM, 0, NULL);
3741 			ifmedia_set(media, m | IFM_10G_LRM);
3742 			break;
3743 
3744 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3745 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3746 			ifmedia_add(media, m | IFM_10G_TWINAX, 0, NULL);
3747 			ifmedia_set(media, m | IFM_10G_TWINAX);
3748 			break;
3749 
3750 		case FW_PORT_MOD_TYPE_NONE:
3751 			m &= ~IFM_FDX;
3752 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3753 			ifmedia_set(media, m | IFM_NONE);
3754 			break;
3755 
3756 		case FW_PORT_MOD_TYPE_NA:
3757 		case FW_PORT_MOD_TYPE_ER:
3758 		default:
3759 			device_printf(pi->dev,
3760 			    "unknown port_type (%d), mod_type (%d)\n",
3761 			    pi->port_type, pi->mod_type);
3762 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3763 			ifmedia_set(media, m | IFM_UNKNOWN);
3764 			break;
3765 		}
3766 		break;
3767 
3768 	case FW_PORT_TYPE_CR_QSFP:
3769 	case FW_PORT_TYPE_SFP28:
3770 	case FW_PORT_TYPE_KR_SFP28:
3771 		switch (pi->mod_type) {
3772 
3773 		case FW_PORT_MOD_TYPE_SR:
3774 			ifmedia_add(media, m | IFM_25G_SR, 0, NULL);
3775 			ifmedia_set(media, m | IFM_25G_SR);
3776 			break;
3777 
3778 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3779 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3780 			ifmedia_add(media, m | IFM_25G_CR, 0, NULL);
3781 			ifmedia_set(media, m | IFM_25G_CR);
3782 			break;
3783 
3784 		case FW_PORT_MOD_TYPE_NONE:
3785 			m &= ~IFM_FDX;
3786 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3787 			ifmedia_set(media, m | IFM_NONE);
3788 			break;
3789 
3790 		default:
3791 			device_printf(pi->dev,
3792 			    "unknown port_type (%d), mod_type (%d)\n",
3793 			    pi->port_type, pi->mod_type);
3794 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3795 			ifmedia_set(media, m | IFM_UNKNOWN);
3796 			break;
3797 		}
3798 		break;
3799 
3800 	case FW_PORT_TYPE_QSFP:
3801 		switch (pi->mod_type) {
3802 
3803 		case FW_PORT_MOD_TYPE_LR:
3804 			ifmedia_add(media, m | IFM_40G_LR4, 0, NULL);
3805 			ifmedia_set(media, m | IFM_40G_LR4);
3806 			break;
3807 
3808 		case FW_PORT_MOD_TYPE_SR:
3809 			ifmedia_add(media, m | IFM_40G_SR4, 0, NULL);
3810 			ifmedia_set(media, m | IFM_40G_SR4);
3811 			break;
3812 
3813 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3814 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3815 			ifmedia_add(media, m | IFM_40G_CR4, 0, NULL);
3816 			ifmedia_set(media, m | IFM_40G_CR4);
3817 			break;
3818 
3819 		case FW_PORT_MOD_TYPE_NONE:
3820 			m &= ~IFM_FDX;
3821 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3822 			ifmedia_set(media, m | IFM_NONE);
3823 			break;
3824 
3825 		default:
3826 			device_printf(pi->dev,
3827 			    "unknown port_type (%d), mod_type (%d)\n",
3828 			    pi->port_type, pi->mod_type);
3829 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3830 			ifmedia_set(media, m | IFM_UNKNOWN);
3831 			break;
3832 		}
3833 		break;
3834 
3835 	case FW_PORT_TYPE_KR4_100G:
3836 	case FW_PORT_TYPE_CR4_QSFP:
3837 		switch (pi->mod_type) {
3838 
3839 		case FW_PORT_MOD_TYPE_LR:
3840 			ifmedia_add(media, m | IFM_100G_LR4, 0, NULL);
3841 			ifmedia_set(media, m | IFM_100G_LR4);
3842 			break;
3843 
3844 		case FW_PORT_MOD_TYPE_SR:
3845 			ifmedia_add(media, m | IFM_100G_SR4, 0, NULL);
3846 			ifmedia_set(media, m | IFM_100G_SR4);
3847 			break;
3848 
3849 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3850 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3851 			ifmedia_add(media, m | IFM_100G_CR4, 0, NULL);
3852 			ifmedia_set(media, m | IFM_100G_CR4);
3853 			break;
3854 
3855 		case FW_PORT_MOD_TYPE_NONE:
3856 			m &= ~IFM_FDX;
3857 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3858 			ifmedia_set(media, m | IFM_NONE);
3859 			break;
3860 
3861 		default:
3862 			device_printf(pi->dev,
3863 			    "unknown port_type (%d), mod_type (%d)\n",
3864 			    pi->port_type, pi->mod_type);
3865 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3866 			ifmedia_set(media, m | IFM_UNKNOWN);
3867 			break;
3868 		}
3869 		break;
3870 
3871 	default:
3872 		device_printf(pi->dev,
3873 		    "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
3874 		    pi->mod_type);
3875 		ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3876 		ifmedia_set(media, m | IFM_UNKNOWN);
3877 		break;
3878 	}
3879 }
3880 
3881 /*
3882  * Update all the requested_* fields in the link config and then send a mailbox
3883  * command to apply the settings.
3884  */
3885 static void
3886 init_l1cfg(struct port_info *pi)
3887 {
3888 	struct adapter *sc = pi->adapter;
3889 	struct link_config *lc = &pi->link_cfg;
3890 	int rc;
3891 
3892 	ASSERT_SYNCHRONIZED_OP(sc);
3893 
3894 	if (t4_autoneg != 0 && lc->supported & FW_PORT_CAP_ANEG) {
3895 		lc->requested_aneg = AUTONEG_ENABLE;
3896 		lc->requested_speed = 0;
3897 	} else {
3898 		lc->requested_aneg = AUTONEG_DISABLE;
3899 		lc->requested_speed = port_top_speed(pi);	/* in Gbps */
3900 	}
3901 
3902 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX);
3903 
3904 	if (t4_fec != -1) {
3905 		lc->requested_fec = t4_fec & (FEC_RS | FEC_BASER_RS |
3906 		    FEC_RESERVED);
3907 	} else {
3908 		/* Use the suggested value provided by the firmware in acaps */
3909 		if (lc->advertising & FW_PORT_CAP_FEC_RS)
3910 			lc->requested_fec = FEC_RS;
3911 		else if (lc->advertising & FW_PORT_CAP_FEC_BASER_RS)
3912 			lc->requested_fec = FEC_BASER_RS;
3913 		else if (lc->advertising & FW_PORT_CAP_FEC_RESERVED)
3914 			lc->requested_fec = FEC_RESERVED;
3915 		else
3916 			lc->requested_fec = 0;
3917 	}
3918 
3919 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
3920 	if (rc != 0) {
3921 		device_printf(pi->dev, "l1cfg failed: %d\n", rc);
3922 	} else {
3923 		lc->fc = lc->requested_fc;
3924 		lc->fec = lc->requested_fec;
3925 	}
3926 }
3927 
3928 #define FW_MAC_EXACT_CHUNK	7
3929 
3930 /*
3931  * Program the port's XGMAC based on parameters in ifnet.  The caller also
3932  * indicates which parameters should be programmed (the rest are left alone).
3933  */
3934 int
3935 update_mac_settings(struct ifnet *ifp, int flags)
3936 {
3937 	int rc = 0;
3938 	struct vi_info *vi = ifp->if_softc;
3939 	struct port_info *pi = vi->pi;
3940 	struct adapter *sc = pi->adapter;
3941 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
3942 
3943 	ASSERT_SYNCHRONIZED_OP(sc);
3944 	KASSERT(flags, ("%s: not told what to update.", __func__));
3945 
3946 	if (flags & XGMAC_MTU)
3947 		mtu = ifp->if_mtu;
3948 
3949 	if (flags & XGMAC_PROMISC)
3950 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
3951 
3952 	if (flags & XGMAC_ALLMULTI)
3953 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
3954 
3955 	if (flags & XGMAC_VLANEX)
3956 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
3957 
3958 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
3959 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
3960 		    allmulti, 1, vlanex, false);
3961 		if (rc) {
3962 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
3963 			    rc);
3964 			return (rc);
3965 		}
3966 	}
3967 
3968 	if (flags & XGMAC_UCADDR) {
3969 		uint8_t ucaddr[ETHER_ADDR_LEN];
3970 
3971 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
3972 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
3973 		    ucaddr, true, true);
3974 		if (rc < 0) {
3975 			rc = -rc;
3976 			if_printf(ifp, "change_mac failed: %d\n", rc);
3977 			return (rc);
3978 		} else {
3979 			vi->xact_addr_filt = rc;
3980 			rc = 0;
3981 		}
3982 	}
3983 
3984 	if (flags & XGMAC_MCADDRS) {
3985 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
3986 		int del = 1;
3987 		uint64_t hash = 0;
3988 		struct ifmultiaddr *ifma;
3989 		int i = 0, j;
3990 
3991 		if_maddr_rlock(ifp);
3992 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3993 			if (ifma->ifma_addr->sa_family != AF_LINK)
3994 				continue;
3995 			mcaddr[i] =
3996 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
3997 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
3998 			i++;
3999 
4000 			if (i == FW_MAC_EXACT_CHUNK) {
4001 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4002 				    del, i, mcaddr, NULL, &hash, 0);
4003 				if (rc < 0) {
4004 					rc = -rc;
4005 					for (j = 0; j < i; j++) {
4006 						if_printf(ifp,
4007 						    "failed to add mc address"
4008 						    " %02x:%02x:%02x:"
4009 						    "%02x:%02x:%02x rc=%d\n",
4010 						    mcaddr[j][0], mcaddr[j][1],
4011 						    mcaddr[j][2], mcaddr[j][3],
4012 						    mcaddr[j][4], mcaddr[j][5],
4013 						    rc);
4014 					}
4015 					goto mcfail;
4016 				}
4017 				del = 0;
4018 				i = 0;
4019 			}
4020 		}
4021 		if (i > 0) {
4022 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4023 			    mcaddr, NULL, &hash, 0);
4024 			if (rc < 0) {
4025 				rc = -rc;
4026 				for (j = 0; j < i; j++) {
4027 					if_printf(ifp,
4028 					    "failed to add mc address"
4029 					    " %02x:%02x:%02x:"
4030 					    "%02x:%02x:%02x rc=%d\n",
4031 					    mcaddr[j][0], mcaddr[j][1],
4032 					    mcaddr[j][2], mcaddr[j][3],
4033 					    mcaddr[j][4], mcaddr[j][5],
4034 					    rc);
4035 				}
4036 				goto mcfail;
4037 			}
4038 		}
4039 
4040 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4041 		if (rc != 0)
4042 			if_printf(ifp, "failed to set mc address hash: %d", rc);
4043 mcfail:
4044 		if_maddr_runlock(ifp);
4045 	}
4046 
4047 	return (rc);
4048 }
4049 
4050 /*
4051  * {begin|end}_synchronized_op must be called from the same thread.
4052  */
4053 int
4054 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4055     char *wmesg)
4056 {
4057 	int rc, pri;
4058 
4059 #ifdef WITNESS
4060 	/* the caller thinks it's ok to sleep, but is it really? */
4061 	if (flags & SLEEP_OK)
4062 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4063 		    "begin_synchronized_op");
4064 #endif
4065 
4066 	if (INTR_OK)
4067 		pri = PCATCH;
4068 	else
4069 		pri = 0;
4070 
4071 	ADAPTER_LOCK(sc);
4072 	for (;;) {
4073 
4074 		if (vi && IS_DOOMED(vi)) {
4075 			rc = ENXIO;
4076 			goto done;
4077 		}
4078 
4079 		if (!IS_BUSY(sc)) {
4080 			rc = 0;
4081 			break;
4082 		}
4083 
4084 		if (!(flags & SLEEP_OK)) {
4085 			rc = EBUSY;
4086 			goto done;
4087 		}
4088 
4089 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4090 			rc = EINTR;
4091 			goto done;
4092 		}
4093 	}
4094 
4095 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4096 	SET_BUSY(sc);
4097 #ifdef INVARIANTS
4098 	sc->last_op = wmesg;
4099 	sc->last_op_thr = curthread;
4100 	sc->last_op_flags = flags;
4101 #endif
4102 
4103 done:
4104 	if (!(flags & HOLD_LOCK) || rc)
4105 		ADAPTER_UNLOCK(sc);
4106 
4107 	return (rc);
4108 }
4109 
4110 /*
4111  * Tell if_ioctl and if_init that the VI is going away.  This is
4112  * special variant of begin_synchronized_op and must be paired with a
4113  * call to end_synchronized_op.
4114  */
4115 void
4116 doom_vi(struct adapter *sc, struct vi_info *vi)
4117 {
4118 
4119 	ADAPTER_LOCK(sc);
4120 	SET_DOOMED(vi);
4121 	wakeup(&sc->flags);
4122 	while (IS_BUSY(sc))
4123 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4124 	SET_BUSY(sc);
4125 #ifdef INVARIANTS
4126 	sc->last_op = "t4detach";
4127 	sc->last_op_thr = curthread;
4128 	sc->last_op_flags = 0;
4129 #endif
4130 	ADAPTER_UNLOCK(sc);
4131 }
4132 
4133 /*
4134  * {begin|end}_synchronized_op must be called from the same thread.
4135  */
4136 void
4137 end_synchronized_op(struct adapter *sc, int flags)
4138 {
4139 
4140 	if (flags & LOCK_HELD)
4141 		ADAPTER_LOCK_ASSERT_OWNED(sc);
4142 	else
4143 		ADAPTER_LOCK(sc);
4144 
4145 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
4146 	CLR_BUSY(sc);
4147 	wakeup(&sc->flags);
4148 	ADAPTER_UNLOCK(sc);
4149 }
4150 
4151 static int
4152 cxgbe_init_synchronized(struct vi_info *vi)
4153 {
4154 	struct port_info *pi = vi->pi;
4155 	struct adapter *sc = pi->adapter;
4156 	struct ifnet *ifp = vi->ifp;
4157 	int rc = 0, i;
4158 	struct sge_txq *txq;
4159 
4160 	ASSERT_SYNCHRONIZED_OP(sc);
4161 
4162 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
4163 		return (0);	/* already running */
4164 
4165 	if (!(sc->flags & FULL_INIT_DONE) &&
4166 	    ((rc = adapter_full_init(sc)) != 0))
4167 		return (rc);	/* error message displayed already */
4168 
4169 	if (!(vi->flags & VI_INIT_DONE) &&
4170 	    ((rc = vi_full_init(vi)) != 0))
4171 		return (rc); /* error message displayed already */
4172 
4173 	rc = update_mac_settings(ifp, XGMAC_ALL);
4174 	if (rc)
4175 		goto done;	/* error message displayed already */
4176 
4177 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
4178 	if (rc != 0) {
4179 		if_printf(ifp, "enable_vi failed: %d\n", rc);
4180 		goto done;
4181 	}
4182 
4183 	/*
4184 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
4185 	 * if this changes.
4186 	 */
4187 
4188 	for_each_txq(vi, i, txq) {
4189 		TXQ_LOCK(txq);
4190 		txq->eq.flags |= EQ_ENABLED;
4191 		TXQ_UNLOCK(txq);
4192 	}
4193 
4194 	/*
4195 	 * The first iq of the first port to come up is used for tracing.
4196 	 */
4197 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4198 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4199 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4200 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4201 		    V_QUEUENUMBER(sc->traceq));
4202 		pi->flags |= HAS_TRACEQ;
4203 	}
4204 
4205 	/* all ok */
4206 	PORT_LOCK(pi);
4207 	if (pi->up_vis++ == 0) {
4208 		t4_update_port_info(pi);
4209 		build_medialist(vi->pi, &vi->media);
4210 		init_l1cfg(pi);
4211 	}
4212 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
4213 
4214 	if (pi->nvi > 1 || sc->flags & IS_VF)
4215 		callout_reset(&vi->tick, hz, vi_tick, vi);
4216 	else
4217 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4218 	PORT_UNLOCK(pi);
4219 done:
4220 	if (rc != 0)
4221 		cxgbe_uninit_synchronized(vi);
4222 
4223 	return (rc);
4224 }
4225 
4226 /*
4227  * Idempotent.
4228  */
4229 static int
4230 cxgbe_uninit_synchronized(struct vi_info *vi)
4231 {
4232 	struct port_info *pi = vi->pi;
4233 	struct adapter *sc = pi->adapter;
4234 	struct ifnet *ifp = vi->ifp;
4235 	int rc, i;
4236 	struct sge_txq *txq;
4237 
4238 	ASSERT_SYNCHRONIZED_OP(sc);
4239 
4240 	if (!(vi->flags & VI_INIT_DONE)) {
4241 		KASSERT(!(ifp->if_drv_flags & IFF_DRV_RUNNING),
4242 		    ("uninited VI is running"));
4243 		return (0);
4244 	}
4245 
4246 	/*
4247 	 * Disable the VI so that all its data in either direction is discarded
4248 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4249 	 * tick) intact as the TP can deliver negative advice or data that it's
4250 	 * holding in its RAM (for an offloaded connection) even after the VI is
4251 	 * disabled.
4252 	 */
4253 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4254 	if (rc) {
4255 		if_printf(ifp, "disable_vi failed: %d\n", rc);
4256 		return (rc);
4257 	}
4258 
4259 	for_each_txq(vi, i, txq) {
4260 		TXQ_LOCK(txq);
4261 		txq->eq.flags &= ~EQ_ENABLED;
4262 		TXQ_UNLOCK(txq);
4263 	}
4264 
4265 	PORT_LOCK(pi);
4266 	if (pi->nvi > 1 || sc->flags & IS_VF)
4267 		callout_stop(&vi->tick);
4268 	else
4269 		callout_stop(&pi->tick);
4270 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4271 		PORT_UNLOCK(pi);
4272 		return (0);
4273 	}
4274 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4275 	pi->up_vis--;
4276 	if (pi->up_vis > 0) {
4277 		PORT_UNLOCK(pi);
4278 		return (0);
4279 	}
4280 	PORT_UNLOCK(pi);
4281 
4282 	pi->link_cfg.link_ok = 0;
4283 	pi->link_cfg.speed = 0;
4284 	pi->link_cfg.link_down_rc = 255;
4285 	t4_os_link_changed(pi, NULL);
4286 
4287 	return (0);
4288 }
4289 
4290 /*
4291  * It is ok for this function to fail midway and return right away.  t4_detach
4292  * will walk the entire sc->irq list and clean up whatever is valid.
4293  */
4294 int
4295 t4_setup_intr_handlers(struct adapter *sc)
4296 {
4297 	int rc, rid, p, q, v;
4298 	char s[8];
4299 	struct irq *irq;
4300 	struct port_info *pi;
4301 	struct vi_info *vi;
4302 	struct sge *sge = &sc->sge;
4303 	struct sge_rxq *rxq;
4304 #ifdef TCP_OFFLOAD
4305 	struct sge_ofld_rxq *ofld_rxq;
4306 #endif
4307 #ifdef DEV_NETMAP
4308 	struct sge_nm_rxq *nm_rxq;
4309 #endif
4310 #ifdef RSS
4311 	int nbuckets = rss_getnumbuckets();
4312 #endif
4313 
4314 	/*
4315 	 * Setup interrupts.
4316 	 */
4317 	irq = &sc->irq[0];
4318 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
4319 	if (sc->intr_count == 1)
4320 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4321 
4322 	/* Multiple interrupts. */
4323 	if (sc->flags & IS_VF)
4324 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4325 		    ("%s: too few intr.", __func__));
4326 	else
4327 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4328 		    ("%s: too few intr.", __func__));
4329 
4330 	/* The first one is always error intr on PFs */
4331 	if (!(sc->flags & IS_VF)) {
4332 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4333 		if (rc != 0)
4334 			return (rc);
4335 		irq++;
4336 		rid++;
4337 	}
4338 
4339 	/* The second one is always the firmware event queue (first on VFs) */
4340 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4341 	if (rc != 0)
4342 		return (rc);
4343 	irq++;
4344 	rid++;
4345 
4346 	for_each_port(sc, p) {
4347 		pi = sc->port[p];
4348 		for_each_vi(pi, v, vi) {
4349 			vi->first_intr = rid - 1;
4350 
4351 			if (vi->nnmrxq > 0) {
4352 				int n = max(vi->nrxq, vi->nnmrxq);
4353 
4354 				MPASS(vi->flags & INTR_RXQ);
4355 
4356 				rxq = &sge->rxq[vi->first_rxq];
4357 #ifdef DEV_NETMAP
4358 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4359 #endif
4360 				for (q = 0; q < n; q++) {
4361 					snprintf(s, sizeof(s), "%x%c%x", p,
4362 					    'a' + v, q);
4363 					if (q < vi->nrxq)
4364 						irq->rxq = rxq++;
4365 #ifdef DEV_NETMAP
4366 					if (q < vi->nnmrxq)
4367 						irq->nm_rxq = nm_rxq++;
4368 #endif
4369 					rc = t4_alloc_irq(sc, irq, rid,
4370 					    t4_vi_intr, irq, s);
4371 					if (rc != 0)
4372 						return (rc);
4373 					irq++;
4374 					rid++;
4375 					vi->nintr++;
4376 				}
4377 			} else if (vi->flags & INTR_RXQ) {
4378 				for_each_rxq(vi, q, rxq) {
4379 					snprintf(s, sizeof(s), "%x%c%x", p,
4380 					    'a' + v, q);
4381 					rc = t4_alloc_irq(sc, irq, rid,
4382 					    t4_intr, rxq, s);
4383 					if (rc != 0)
4384 						return (rc);
4385 #ifdef RSS
4386 					bus_bind_intr(sc->dev, irq->res,
4387 					    rss_getcpu(q % nbuckets));
4388 #endif
4389 					irq++;
4390 					rid++;
4391 					vi->nintr++;
4392 				}
4393 			}
4394 #ifdef TCP_OFFLOAD
4395 			if (vi->flags & INTR_OFLD_RXQ) {
4396 				for_each_ofld_rxq(vi, q, ofld_rxq) {
4397 					snprintf(s, sizeof(s), "%x%c%x", p,
4398 					    'A' + v, q);
4399 					rc = t4_alloc_irq(sc, irq, rid,
4400 					    t4_intr, ofld_rxq, s);
4401 					if (rc != 0)
4402 						return (rc);
4403 					irq++;
4404 					rid++;
4405 					vi->nintr++;
4406 				}
4407 			}
4408 #endif
4409 		}
4410 	}
4411 	MPASS(irq == &sc->irq[sc->intr_count]);
4412 
4413 	return (0);
4414 }
4415 
4416 int
4417 adapter_full_init(struct adapter *sc)
4418 {
4419 	int rc, i;
4420 #ifdef RSS
4421 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4422 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4423 #endif
4424 
4425 	ASSERT_SYNCHRONIZED_OP(sc);
4426 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4427 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
4428 	    ("%s: FULL_INIT_DONE already", __func__));
4429 
4430 	/*
4431 	 * queues that belong to the adapter (not any particular port).
4432 	 */
4433 	rc = t4_setup_adapter_queues(sc);
4434 	if (rc != 0)
4435 		goto done;
4436 
4437 	for (i = 0; i < nitems(sc->tq); i++) {
4438 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
4439 		    taskqueue_thread_enqueue, &sc->tq[i]);
4440 		if (sc->tq[i] == NULL) {
4441 			device_printf(sc->dev,
4442 			    "failed to allocate task queue %d\n", i);
4443 			rc = ENOMEM;
4444 			goto done;
4445 		}
4446 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
4447 		    device_get_nameunit(sc->dev), i);
4448 	}
4449 #ifdef RSS
4450 	MPASS(RSS_KEYSIZE == 40);
4451 	rss_getkey((void *)&raw_rss_key[0]);
4452 	for (i = 0; i < nitems(rss_key); i++) {
4453 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
4454 	}
4455 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
4456 #endif
4457 
4458 	if (!(sc->flags & IS_VF))
4459 		t4_intr_enable(sc);
4460 	sc->flags |= FULL_INIT_DONE;
4461 done:
4462 	if (rc != 0)
4463 		adapter_full_uninit(sc);
4464 
4465 	return (rc);
4466 }
4467 
4468 int
4469 adapter_full_uninit(struct adapter *sc)
4470 {
4471 	int i;
4472 
4473 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4474 
4475 	t4_teardown_adapter_queues(sc);
4476 
4477 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
4478 		taskqueue_free(sc->tq[i]);
4479 		sc->tq[i] = NULL;
4480 	}
4481 
4482 	sc->flags &= ~FULL_INIT_DONE;
4483 
4484 	return (0);
4485 }
4486 
4487 #ifdef RSS
4488 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
4489     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
4490     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
4491     RSS_HASHTYPE_RSS_UDP_IPV6)
4492 
4493 /* Translates kernel hash types to hardware. */
4494 static int
4495 hashconfig_to_hashen(int hashconfig)
4496 {
4497 	int hashen = 0;
4498 
4499 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
4500 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
4501 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
4502 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
4503 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
4504 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4505 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4506 	}
4507 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
4508 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4509 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4510 	}
4511 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
4512 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4513 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
4514 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4515 
4516 	return (hashen);
4517 }
4518 
4519 /* Translates hardware hash types to kernel. */
4520 static int
4521 hashen_to_hashconfig(int hashen)
4522 {
4523 	int hashconfig = 0;
4524 
4525 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
4526 		/*
4527 		 * If UDP hashing was enabled it must have been enabled for
4528 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
4529 		 * enabling any 4-tuple hash is nonsense configuration.
4530 		 */
4531 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4532 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
4533 
4534 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4535 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
4536 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4537 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
4538 	}
4539 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4540 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
4541 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4542 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
4543 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
4544 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
4545 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
4546 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
4547 
4548 	return (hashconfig);
4549 }
4550 #endif
4551 
4552 int
4553 vi_full_init(struct vi_info *vi)
4554 {
4555 	struct adapter *sc = vi->pi->adapter;
4556 	struct ifnet *ifp = vi->ifp;
4557 	uint16_t *rss;
4558 	struct sge_rxq *rxq;
4559 	int rc, i, j, hashen;
4560 #ifdef RSS
4561 	int nbuckets = rss_getnumbuckets();
4562 	int hashconfig = rss_gethashconfig();
4563 	int extra;
4564 #endif
4565 
4566 	ASSERT_SYNCHRONIZED_OP(sc);
4567 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
4568 	    ("%s: VI_INIT_DONE already", __func__));
4569 
4570 	sysctl_ctx_init(&vi->ctx);
4571 	vi->flags |= VI_SYSCTL_CTX;
4572 
4573 	/*
4574 	 * Allocate tx/rx/fl queues for this VI.
4575 	 */
4576 	rc = t4_setup_vi_queues(vi);
4577 	if (rc != 0)
4578 		goto done;	/* error message displayed already */
4579 
4580 	/*
4581 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
4582 	 */
4583 	if (vi->nrxq > vi->rss_size) {
4584 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
4585 		    "some queues will never receive traffic.\n", vi->nrxq,
4586 		    vi->rss_size);
4587 	} else if (vi->rss_size % vi->nrxq) {
4588 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
4589 		    "expect uneven traffic distribution.\n", vi->nrxq,
4590 		    vi->rss_size);
4591 	}
4592 #ifdef RSS
4593 	if (vi->nrxq != nbuckets) {
4594 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
4595 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
4596 	}
4597 #endif
4598 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
4599 	for (i = 0; i < vi->rss_size;) {
4600 #ifdef RSS
4601 		j = rss_get_indirection_to_bucket(i);
4602 		j %= vi->nrxq;
4603 		rxq = &sc->sge.rxq[vi->first_rxq + j];
4604 		rss[i++] = rxq->iq.abs_id;
4605 #else
4606 		for_each_rxq(vi, j, rxq) {
4607 			rss[i++] = rxq->iq.abs_id;
4608 			if (i == vi->rss_size)
4609 				break;
4610 		}
4611 #endif
4612 	}
4613 
4614 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
4615 	    vi->rss_size);
4616 	if (rc != 0) {
4617 		if_printf(ifp, "rss_config failed: %d\n", rc);
4618 		goto done;
4619 	}
4620 
4621 #ifdef RSS
4622 	hashen = hashconfig_to_hashen(hashconfig);
4623 
4624 	/*
4625 	 * We may have had to enable some hashes even though the global config
4626 	 * wants them disabled.  This is a potential problem that must be
4627 	 * reported to the user.
4628 	 */
4629 	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
4630 
4631 	/*
4632 	 * If we consider only the supported hash types, then the enabled hashes
4633 	 * are a superset of the requested hashes.  In other words, there cannot
4634 	 * be any supported hash that was requested but not enabled, but there
4635 	 * can be hashes that were not requested but had to be enabled.
4636 	 */
4637 	extra &= SUPPORTED_RSS_HASHTYPES;
4638 	MPASS((extra & hashconfig) == 0);
4639 
4640 	if (extra) {
4641 		if_printf(ifp,
4642 		    "global RSS config (0x%x) cannot be accommodated.\n",
4643 		    hashconfig);
4644 	}
4645 	if (extra & RSS_HASHTYPE_RSS_IPV4)
4646 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
4647 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
4648 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
4649 	if (extra & RSS_HASHTYPE_RSS_IPV6)
4650 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
4651 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
4652 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
4653 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
4654 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
4655 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
4656 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
4657 #else
4658 	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
4659 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
4660 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4661 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
4662 #endif
4663 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
4664 	if (rc != 0) {
4665 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
4666 		goto done;
4667 	}
4668 
4669 	vi->rss = rss;
4670 	vi->flags |= VI_INIT_DONE;
4671 done:
4672 	if (rc != 0)
4673 		vi_full_uninit(vi);
4674 
4675 	return (rc);
4676 }
4677 
4678 /*
4679  * Idempotent.
4680  */
4681 int
4682 vi_full_uninit(struct vi_info *vi)
4683 {
4684 	struct port_info *pi = vi->pi;
4685 	struct adapter *sc = pi->adapter;
4686 	int i;
4687 	struct sge_rxq *rxq;
4688 	struct sge_txq *txq;
4689 #ifdef TCP_OFFLOAD
4690 	struct sge_ofld_rxq *ofld_rxq;
4691 	struct sge_wrq *ofld_txq;
4692 #endif
4693 
4694 	if (vi->flags & VI_INIT_DONE) {
4695 
4696 		/* Need to quiesce queues.  */
4697 
4698 		/* XXX: Only for the first VI? */
4699 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
4700 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
4701 
4702 		for_each_txq(vi, i, txq) {
4703 			quiesce_txq(sc, txq);
4704 		}
4705 
4706 #ifdef TCP_OFFLOAD
4707 		for_each_ofld_txq(vi, i, ofld_txq) {
4708 			quiesce_wrq(sc, ofld_txq);
4709 		}
4710 #endif
4711 
4712 		for_each_rxq(vi, i, rxq) {
4713 			quiesce_iq(sc, &rxq->iq);
4714 			quiesce_fl(sc, &rxq->fl);
4715 		}
4716 
4717 #ifdef TCP_OFFLOAD
4718 		for_each_ofld_rxq(vi, i, ofld_rxq) {
4719 			quiesce_iq(sc, &ofld_rxq->iq);
4720 			quiesce_fl(sc, &ofld_rxq->fl);
4721 		}
4722 #endif
4723 		free(vi->rss, M_CXGBE);
4724 		free(vi->nm_rss, M_CXGBE);
4725 	}
4726 
4727 	t4_teardown_vi_queues(vi);
4728 	vi->flags &= ~VI_INIT_DONE;
4729 
4730 	return (0);
4731 }
4732 
4733 static void
4734 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
4735 {
4736 	struct sge_eq *eq = &txq->eq;
4737 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
4738 
4739 	(void) sc;	/* unused */
4740 
4741 #ifdef INVARIANTS
4742 	TXQ_LOCK(txq);
4743 	MPASS((eq->flags & EQ_ENABLED) == 0);
4744 	TXQ_UNLOCK(txq);
4745 #endif
4746 
4747 	/* Wait for the mp_ring to empty. */
4748 	while (!mp_ring_is_idle(txq->r)) {
4749 		mp_ring_check_drainage(txq->r, 0);
4750 		pause("rquiesce", 1);
4751 	}
4752 
4753 	/* Then wait for the hardware to finish. */
4754 	while (spg->cidx != htobe16(eq->pidx))
4755 		pause("equiesce", 1);
4756 
4757 	/* Finally, wait for the driver to reclaim all descriptors. */
4758 	while (eq->cidx != eq->pidx)
4759 		pause("dquiesce", 1);
4760 }
4761 
4762 static void
4763 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
4764 {
4765 
4766 	/* XXXTX */
4767 }
4768 
4769 static void
4770 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
4771 {
4772 	(void) sc;	/* unused */
4773 
4774 	/* Synchronize with the interrupt handler */
4775 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
4776 		pause("iqfree", 1);
4777 }
4778 
4779 static void
4780 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
4781 {
4782 	mtx_lock(&sc->sfl_lock);
4783 	FL_LOCK(fl);
4784 	fl->flags |= FL_DOOMED;
4785 	FL_UNLOCK(fl);
4786 	callout_stop(&sc->sfl_callout);
4787 	mtx_unlock(&sc->sfl_lock);
4788 
4789 	KASSERT((fl->flags & FL_STARVING) == 0,
4790 	    ("%s: still starving", __func__));
4791 }
4792 
4793 static int
4794 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
4795     driver_intr_t *handler, void *arg, char *name)
4796 {
4797 	int rc;
4798 
4799 	irq->rid = rid;
4800 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
4801 	    RF_SHAREABLE | RF_ACTIVE);
4802 	if (irq->res == NULL) {
4803 		device_printf(sc->dev,
4804 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
4805 		return (ENOMEM);
4806 	}
4807 
4808 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
4809 	    NULL, handler, arg, &irq->tag);
4810 	if (rc != 0) {
4811 		device_printf(sc->dev,
4812 		    "failed to setup interrupt for rid %d, name %s: %d\n",
4813 		    rid, name, rc);
4814 	} else if (name)
4815 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
4816 
4817 	return (rc);
4818 }
4819 
4820 static int
4821 t4_free_irq(struct adapter *sc, struct irq *irq)
4822 {
4823 	if (irq->tag)
4824 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
4825 	if (irq->res)
4826 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
4827 
4828 	bzero(irq, sizeof(*irq));
4829 
4830 	return (0);
4831 }
4832 
4833 static void
4834 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
4835 {
4836 
4837 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
4838 	t4_get_regs(sc, buf, regs->len);
4839 }
4840 
4841 #define	A_PL_INDIR_CMD	0x1f8
4842 
4843 #define	S_PL_AUTOINC	31
4844 #define	M_PL_AUTOINC	0x1U
4845 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
4846 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
4847 
4848 #define	S_PL_VFID	20
4849 #define	M_PL_VFID	0xffU
4850 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
4851 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
4852 
4853 #define	S_PL_ADDR	0
4854 #define	M_PL_ADDR	0xfffffU
4855 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
4856 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
4857 
4858 #define	A_PL_INDIR_DATA	0x1fc
4859 
4860 static uint64_t
4861 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
4862 {
4863 	u32 stats[2];
4864 
4865 	mtx_assert(&sc->reg_lock, MA_OWNED);
4866 	if (sc->flags & IS_VF) {
4867 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
4868 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
4869 	} else {
4870 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4871 		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4872 		    V_PL_ADDR(VF_MPS_REG(reg)));
4873 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
4874 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
4875 	}
4876 	return (((uint64_t)stats[1]) << 32 | stats[0]);
4877 }
4878 
4879 static void
4880 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
4881     struct fw_vi_stats_vf *stats)
4882 {
4883 
4884 #define GET_STAT(name) \
4885 	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
4886 
4887 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
4888 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
4889 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
4890 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
4891 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
4892 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
4893 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
4894 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
4895 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
4896 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
4897 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
4898 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
4899 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
4900 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
4901 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
4902 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
4903 
4904 #undef GET_STAT
4905 }
4906 
4907 static void
4908 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
4909 {
4910 	int reg;
4911 
4912 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4913 	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4914 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
4915 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
4916 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
4917 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
4918 }
4919 
4920 static void
4921 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
4922 {
4923 	struct timeval tv;
4924 	const struct timeval interval = {0, 250000};	/* 250ms */
4925 
4926 	if (!(vi->flags & VI_INIT_DONE))
4927 		return;
4928 
4929 	getmicrotime(&tv);
4930 	timevalsub(&tv, &interval);
4931 	if (timevalcmp(&tv, &vi->last_refreshed, <))
4932 		return;
4933 
4934 	mtx_lock(&sc->reg_lock);
4935 	t4_get_vi_stats(sc, vi->viid, &vi->stats);
4936 	getmicrotime(&vi->last_refreshed);
4937 	mtx_unlock(&sc->reg_lock);
4938 }
4939 
4940 static void
4941 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
4942 {
4943 	int i;
4944 	u_int v, tnl_cong_drops;
4945 	struct timeval tv;
4946 	const struct timeval interval = {0, 250000};	/* 250ms */
4947 
4948 	getmicrotime(&tv);
4949 	timevalsub(&tv, &interval);
4950 	if (timevalcmp(&tv, &pi->last_refreshed, <))
4951 		return;
4952 
4953 	tnl_cong_drops = 0;
4954 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
4955 	for (i = 0; i < sc->chip_params->nchan; i++) {
4956 		if (pi->rx_chan_map & (1 << i)) {
4957 			mtx_lock(&sc->reg_lock);
4958 			t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
4959 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
4960 			mtx_unlock(&sc->reg_lock);
4961 			tnl_cong_drops += v;
4962 		}
4963 	}
4964 	pi->tnl_cong_drops = tnl_cong_drops;
4965 	getmicrotime(&pi->last_refreshed);
4966 }
4967 
4968 static void
4969 cxgbe_tick(void *arg)
4970 {
4971 	struct port_info *pi = arg;
4972 	struct adapter *sc = pi->adapter;
4973 
4974 	PORT_LOCK_ASSERT_OWNED(pi);
4975 	cxgbe_refresh_stats(sc, pi);
4976 
4977 	callout_schedule(&pi->tick, hz);
4978 }
4979 
4980 void
4981 vi_tick(void *arg)
4982 {
4983 	struct vi_info *vi = arg;
4984 	struct adapter *sc = vi->pi->adapter;
4985 
4986 	vi_refresh_stats(sc, vi);
4987 
4988 	callout_schedule(&vi->tick, hz);
4989 }
4990 
4991 static void
4992 cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
4993 {
4994 	struct ifnet *vlan;
4995 
4996 	if (arg != ifp || ifp->if_type != IFT_ETHER)
4997 		return;
4998 
4999 	vlan = VLAN_DEVAT(ifp, vid);
5000 	VLAN_SETCOOKIE(vlan, ifp);
5001 }
5002 
5003 /*
5004  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5005  */
5006 static char *caps_decoder[] = {
5007 	"\20\001IPMI\002NCSI",				/* 0: NBM */
5008 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
5009 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
5010 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
5011 	    "\006HASHFILTER\007ETHOFLD",
5012 	"\20\001TOE",					/* 4: TOE */
5013 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
5014 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
5015 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5016 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5017 	    "\007T10DIF"
5018 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5019 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
5020 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
5021 		    "\004PO_INITIATOR\005PO_TARGET",
5022 };
5023 
5024 void
5025 t4_sysctls(struct adapter *sc)
5026 {
5027 	struct sysctl_ctx_list *ctx;
5028 	struct sysctl_oid *oid;
5029 	struct sysctl_oid_list *children, *c0;
5030 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5031 
5032 	ctx = device_get_sysctl_ctx(sc->dev);
5033 
5034 	/*
5035 	 * dev.t4nex.X.
5036 	 */
5037 	oid = device_get_sysctl_tree(sc->dev);
5038 	c0 = children = SYSCTL_CHILDREN(oid);
5039 
5040 	sc->sc_do_rxcopy = 1;
5041 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5042 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5043 
5044 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5045 	    sc->params.nports, "# of ports");
5046 
5047 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5048 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
5049 	    sysctl_bitfield, "A", "available doorbells");
5050 
5051 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5052 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
5053 
5054 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5055 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5056 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5057 	    "interrupt holdoff timer values (us)");
5058 
5059 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5060 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5061 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5062 	    "interrupt holdoff packet counter values");
5063 
5064 	t4_sge_sysctls(sc, ctx, children);
5065 
5066 	sc->lro_timeout = 100;
5067 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5068 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5069 
5070 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5071 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
5072 
5073 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5074 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5075 
5076 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5077 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5078 
5079 	if (sc->flags & IS_VF)
5080 		return;
5081 
5082 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5083 	    NULL, chip_rev(sc), "chip hardware revision");
5084 
5085 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5086 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5087 
5088 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5089 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5090 
5091 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5092 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5093 
5094 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5095 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5096 
5097 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5098 	    sc->er_version, 0, "expansion ROM version");
5099 
5100 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5101 	    sc->bs_version, 0, "bootstrap firmware version");
5102 
5103 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5104 	    NULL, sc->params.scfg_vers, "serial config version");
5105 
5106 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5107 	    NULL, sc->params.vpd_vers, "VPD version");
5108 
5109 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5110 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5111 
5112 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5113 	    sc->cfcsum, "config file checksum");
5114 
5115 #define SYSCTL_CAP(name, n, text) \
5116 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5117 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
5118 	    sysctl_bitfield, "A", "available " text " capabilities")
5119 
5120 	SYSCTL_CAP(nbmcaps, 0, "NBM");
5121 	SYSCTL_CAP(linkcaps, 1, "link");
5122 	SYSCTL_CAP(switchcaps, 2, "switch");
5123 	SYSCTL_CAP(niccaps, 3, "NIC");
5124 	SYSCTL_CAP(toecaps, 4, "TCP offload");
5125 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
5126 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5127 	SYSCTL_CAP(cryptocaps, 7, "crypto");
5128 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
5129 #undef SYSCTL_CAP
5130 
5131 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5132 	    NULL, sc->tids.nftids, "number of filters");
5133 
5134 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5135 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5136 	    "chip temperature (in Celsius)");
5137 
5138 #ifdef SBUF_DRAIN
5139 	/*
5140 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
5141 	 */
5142 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5143 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5144 	    "logs and miscellaneous information");
5145 	children = SYSCTL_CHILDREN(oid);
5146 
5147 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
5148 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5149 	    sysctl_cctrl, "A", "congestion control");
5150 
5151 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
5152 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5153 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
5154 
5155 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
5156 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
5157 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
5158 
5159 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
5160 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
5161 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
5162 
5163 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
5164 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
5165 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
5166 
5167 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
5168 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
5169 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
5170 
5171 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
5172 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
5173 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
5174 
5175 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
5176 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5177 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
5178 	    "A", "CIM logic analyzer");
5179 
5180 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
5181 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5182 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
5183 
5184 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
5185 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
5186 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
5187 
5188 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
5189 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
5190 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
5191 
5192 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
5193 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
5194 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
5195 
5196 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5197 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5198 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5199 
5200 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5201 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5202 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5203 
5204 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5205 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5206 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5207 
5208 	if (chip_id(sc) > CHELSIO_T4) {
5209 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5210 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5211 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5212 
5213 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5214 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5215 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5216 	}
5217 
5218 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5219 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5220 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5221 
5222 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5223 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5224 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
5225 
5226 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5227 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5228 	    sysctl_cpl_stats, "A", "CPL statistics");
5229 
5230 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5231 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5232 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5233 
5234 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5235 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5236 	    sysctl_devlog, "A", "firmware's device log");
5237 
5238 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5239 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5240 	    sysctl_fcoe_stats, "A", "FCoE statistics");
5241 
5242 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5243 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5244 	    sysctl_hw_sched, "A", "hardware scheduler ");
5245 
5246 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5247 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5248 	    sysctl_l2t, "A", "hardware L2 table");
5249 
5250 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5251 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5252 	    sysctl_lb_stats, "A", "loopback statistics");
5253 
5254 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5255 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5256 	    sysctl_meminfo, "A", "memory regions");
5257 
5258 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5259 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5260 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5261 	    "A", "MPS TCAM entries");
5262 
5263 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5264 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5265 	    sysctl_path_mtus, "A", "path MTUs");
5266 
5267 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5268 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5269 	    sysctl_pm_stats, "A", "PM statistics");
5270 
5271 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5272 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5273 	    sysctl_rdma_stats, "A", "RDMA statistics");
5274 
5275 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5276 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5277 	    sysctl_tcp_stats, "A", "TCP statistics");
5278 
5279 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5280 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5281 	    sysctl_tids, "A", "TID information");
5282 
5283 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5284 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5285 	    sysctl_tp_err_stats, "A", "TP error statistics");
5286 
5287 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5288 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5289 	    "TP logic analyzer event capture mask");
5290 
5291 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5292 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5293 	    sysctl_tp_la, "A", "TP logic analyzer");
5294 
5295 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5296 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5297 	    sysctl_tx_rate, "A", "Tx rate");
5298 
5299 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5300 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5301 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5302 
5303 	if (chip_id(sc) >= CHELSIO_T5) {
5304 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5305 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5306 		    sysctl_wcwr_stats, "A", "write combined work requests");
5307 	}
5308 #endif
5309 
5310 #ifdef TCP_OFFLOAD
5311 	if (is_offload(sc)) {
5312 		int i;
5313 		char s[4];
5314 
5315 		/*
5316 		 * dev.t4nex.X.toe.
5317 		 */
5318 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5319 		    NULL, "TOE parameters");
5320 		children = SYSCTL_CHILDREN(oid);
5321 
5322 		sc->tt.sndbuf = 256 * 1024;
5323 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5324 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
5325 
5326 		sc->tt.ddp = 0;
5327 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5328 		    &sc->tt.ddp, 0, "DDP allowed");
5329 
5330 		sc->tt.rx_coalesce = 1;
5331 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5332 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5333 
5334 		sc->tt.tx_align = 1;
5335 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5336 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5337 
5338 		sc->tt.tx_zcopy = 0;
5339 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
5340 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
5341 		    "Enable zero-copy aio_write(2)");
5342 
5343 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
5344 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
5345 		    "TP timer tick (us)");
5346 
5347 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
5348 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
5349 		    "TCP timestamp tick (us)");
5350 
5351 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5352 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5353 		    "DACK tick (us)");
5354 
5355 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5356 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5357 		    "IU", "DACK timer (us)");
5358 
5359 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5360 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5361 		    sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
5362 
5363 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5364 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5365 		    sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
5366 
5367 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5368 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5369 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
5370 
5371 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5372 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5373 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
5374 
5375 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5376 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5377 		    sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
5378 
5379 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
5380 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
5381 		    sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
5382 
5383 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
5384 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
5385 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
5386 
5387 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
5388 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
5389 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
5390 
5391 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
5392 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
5393 		    sysctl_tp_shift_cnt, "IU",
5394 		    "Number of SYN retransmissions before abort");
5395 
5396 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
5397 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
5398 		    sysctl_tp_shift_cnt, "IU",
5399 		    "Number of retransmissions before abort");
5400 
5401 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
5402 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
5403 		    sysctl_tp_shift_cnt, "IU",
5404 		    "Number of keepalive probes before abort");
5405 
5406 		oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
5407 		    CTLFLAG_RD, NULL, "TOE retransmit backoffs");
5408 		children = SYSCTL_CHILDREN(oid);
5409 		for (i = 0; i < 16; i++) {
5410 			snprintf(s, sizeof(s), "%u", i);
5411 			SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
5412 			    CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
5413 			    "IU", "TOE retransmit backoff");
5414 		}
5415 	}
5416 #endif
5417 }
5418 
5419 void
5420 vi_sysctls(struct vi_info *vi)
5421 {
5422 	struct sysctl_ctx_list *ctx;
5423 	struct sysctl_oid *oid;
5424 	struct sysctl_oid_list *children;
5425 
5426 	ctx = device_get_sysctl_ctx(vi->dev);
5427 
5428 	/*
5429 	 * dev.v?(cxgbe|cxl).X.
5430 	 */
5431 	oid = device_get_sysctl_tree(vi->dev);
5432 	children = SYSCTL_CHILDREN(oid);
5433 
5434 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
5435 	    vi->viid, "VI identifer");
5436 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
5437 	    &vi->nrxq, 0, "# of rx queues");
5438 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5439 	    &vi->ntxq, 0, "# of tx queues");
5440 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5441 	    &vi->first_rxq, 0, "index of first rx queue");
5442 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5443 	    &vi->first_txq, 0, "index of first tx queue");
5444 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
5445 	    vi->rss_size, "size of RSS indirection table");
5446 
5447 	if (IS_MAIN_VI(vi)) {
5448 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
5449 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
5450 		    "Reserve queue 0 for non-flowid packets");
5451 	}
5452 
5453 #ifdef TCP_OFFLOAD
5454 	if (vi->nofldrxq != 0) {
5455 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5456 		    &vi->nofldrxq, 0,
5457 		    "# of rx queues for offloaded TCP connections");
5458 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5459 		    &vi->nofldtxq, 0,
5460 		    "# of tx queues for offloaded TCP connections");
5461 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5462 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
5463 		    "index of first TOE rx queue");
5464 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5465 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
5466 		    "index of first TOE tx queue");
5467 	}
5468 #endif
5469 #ifdef DEV_NETMAP
5470 	if (vi->nnmrxq != 0) {
5471 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5472 		    &vi->nnmrxq, 0, "# of netmap rx queues");
5473 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5474 		    &vi->nnmtxq, 0, "# of netmap tx queues");
5475 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5476 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
5477 		    "index of first netmap rx queue");
5478 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
5479 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
5480 		    "index of first netmap tx queue");
5481 	}
5482 #endif
5483 
5484 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
5485 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
5486 	    "holdoff timer index");
5487 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
5488 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
5489 	    "holdoff packet counter index");
5490 
5491 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
5492 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
5493 	    "rx queue size");
5494 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
5495 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
5496 	    "tx queue size");
5497 }
5498 
5499 static void
5500 cxgbe_sysctls(struct port_info *pi)
5501 {
5502 	struct sysctl_ctx_list *ctx;
5503 	struct sysctl_oid *oid;
5504 	struct sysctl_oid_list *children, *children2;
5505 	struct adapter *sc = pi->adapter;
5506 	int i;
5507 	char name[16];
5508 
5509 	ctx = device_get_sysctl_ctx(pi->dev);
5510 
5511 	/*
5512 	 * dev.cxgbe.X.
5513 	 */
5514 	oid = device_get_sysctl_tree(pi->dev);
5515 	children = SYSCTL_CHILDREN(oid);
5516 
5517 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
5518 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
5519 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
5520 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
5521 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
5522 		    "PHY temperature (in Celsius)");
5523 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
5524 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
5525 		    "PHY firmware version");
5526 	}
5527 
5528 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
5529 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
5530 	    "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
5531 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
5532 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
5533 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
5534 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
5535 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
5536 	    "autonegotiation (-1 = not supported)");
5537 
5538 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
5539 	    port_top_speed(pi), "max speed (in Gbps)");
5540 
5541 	if (sc->flags & IS_VF)
5542 		return;
5543 
5544 	/*
5545 	 * dev.(cxgbe|cxl).X.tc.
5546 	 */
5547 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
5548 	    "Tx scheduler traffic classes (cl_rl)");
5549 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
5550 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
5551 
5552 		snprintf(name, sizeof(name), "%d", i);
5553 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
5554 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
5555 		    "traffic class"));
5556 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
5557 		    &tc->flags, 0, "flags");
5558 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
5559 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
5560 #ifdef SBUF_DRAIN
5561 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
5562 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
5563 		    sysctl_tc_params, "A", "traffic class parameters");
5564 #endif
5565 	}
5566 
5567 	/*
5568 	 * dev.cxgbe.X.stats.
5569 	 */
5570 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
5571 	    NULL, "port statistics");
5572 	children = SYSCTL_CHILDREN(oid);
5573 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
5574 	    &pi->tx_parse_error, 0,
5575 	    "# of tx packets with invalid length or # of segments");
5576 
5577 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
5578 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
5579 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
5580 	    sysctl_handle_t4_reg64, "QU", desc)
5581 
5582 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
5583 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
5584 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
5585 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
5586 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
5587 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
5588 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
5589 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
5590 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
5591 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
5592 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
5593 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
5594 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
5595 	    "# of tx frames in this range",
5596 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
5597 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
5598 	    "# of tx frames in this range",
5599 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
5600 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
5601 	    "# of tx frames in this range",
5602 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
5603 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
5604 	    "# of tx frames in this range",
5605 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
5606 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
5607 	    "# of tx frames in this range",
5608 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
5609 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
5610 	    "# of tx frames in this range",
5611 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
5612 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
5613 	    "# of tx frames in this range",
5614 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
5615 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
5616 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
5617 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
5618 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
5619 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
5620 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
5621 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5622 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5623 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5624 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5625 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5626 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5627 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5628 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5629 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5630 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5631 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5632 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5633 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5634 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5635 
5636 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5637 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5638 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5639 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5640 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5641 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5642 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5643 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5644 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5645 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5646 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5647 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5648 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5649 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5650 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5651 	    "# of frames received with bad FCS",
5652 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5653 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5654 	    "# of frames received with length error",
5655 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5656 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5657 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5658 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5659 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5660 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5661 	    "# of rx frames in this range",
5662 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5663 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5664 	    "# of rx frames in this range",
5665 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5666 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5667 	    "# of rx frames in this range",
5668 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5669 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5670 	    "# of rx frames in this range",
5671 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5672 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5673 	    "# of rx frames in this range",
5674 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5675 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5676 	    "# of rx frames in this range",
5677 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5678 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5679 	    "# of rx frames in this range",
5680 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5681 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5682 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5683 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5684 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5685 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5686 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5687 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5688 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5689 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5690 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5691 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5692 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5693 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5694 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5695 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5696 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5697 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5698 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5699 
5700 #undef SYSCTL_ADD_T4_REG64
5701 
5702 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5703 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5704 	    &pi->stats.name, desc)
5705 
5706 	/* We get these from port_stats and they may be stale by up to 1s */
5707 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5708 	    "# drops due to buffer-group 0 overflows");
5709 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5710 	    "# drops due to buffer-group 1 overflows");
5711 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5712 	    "# drops due to buffer-group 2 overflows");
5713 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5714 	    "# drops due to buffer-group 3 overflows");
5715 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5716 	    "# of buffer-group 0 truncated packets");
5717 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5718 	    "# of buffer-group 1 truncated packets");
5719 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5720 	    "# of buffer-group 2 truncated packets");
5721 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5722 	    "# of buffer-group 3 truncated packets");
5723 
5724 #undef SYSCTL_ADD_T4_PORTSTAT
5725 }
5726 
5727 static int
5728 sysctl_int_array(SYSCTL_HANDLER_ARGS)
5729 {
5730 	int rc, *i, space = 0;
5731 	struct sbuf sb;
5732 
5733 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
5734 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
5735 		if (space)
5736 			sbuf_printf(&sb, " ");
5737 		sbuf_printf(&sb, "%d", *i);
5738 		space = 1;
5739 	}
5740 	rc = sbuf_finish(&sb);
5741 	sbuf_delete(&sb);
5742 	return (rc);
5743 }
5744 
5745 static int
5746 sysctl_bitfield(SYSCTL_HANDLER_ARGS)
5747 {
5748 	int rc;
5749 	struct sbuf *sb;
5750 
5751 	rc = sysctl_wire_old_buffer(req, 0);
5752 	if (rc != 0)
5753 		return(rc);
5754 
5755 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5756 	if (sb == NULL)
5757 		return (ENOMEM);
5758 
5759 	sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
5760 	rc = sbuf_finish(sb);
5761 	sbuf_delete(sb);
5762 
5763 	return (rc);
5764 }
5765 
5766 static int
5767 sysctl_btphy(SYSCTL_HANDLER_ARGS)
5768 {
5769 	struct port_info *pi = arg1;
5770 	int op = arg2;
5771 	struct adapter *sc = pi->adapter;
5772 	u_int v;
5773 	int rc;
5774 
5775 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
5776 	if (rc)
5777 		return (rc);
5778 	/* XXX: magic numbers */
5779 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
5780 	    &v);
5781 	end_synchronized_op(sc, 0);
5782 	if (rc)
5783 		return (rc);
5784 	if (op == 0)
5785 		v /= 256;
5786 
5787 	rc = sysctl_handle_int(oidp, &v, 0, req);
5788 	return (rc);
5789 }
5790 
5791 static int
5792 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
5793 {
5794 	struct vi_info *vi = arg1;
5795 	int rc, val;
5796 
5797 	val = vi->rsrv_noflowq;
5798 	rc = sysctl_handle_int(oidp, &val, 0, req);
5799 	if (rc != 0 || req->newptr == NULL)
5800 		return (rc);
5801 
5802 	if ((val >= 1) && (vi->ntxq > 1))
5803 		vi->rsrv_noflowq = 1;
5804 	else
5805 		vi->rsrv_noflowq = 0;
5806 
5807 	return (rc);
5808 }
5809 
5810 static int
5811 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
5812 {
5813 	struct vi_info *vi = arg1;
5814 	struct adapter *sc = vi->pi->adapter;
5815 	int idx, rc, i;
5816 	struct sge_rxq *rxq;
5817 #ifdef TCP_OFFLOAD
5818 	struct sge_ofld_rxq *ofld_rxq;
5819 #endif
5820 	uint8_t v;
5821 
5822 	idx = vi->tmr_idx;
5823 
5824 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5825 	if (rc != 0 || req->newptr == NULL)
5826 		return (rc);
5827 
5828 	if (idx < 0 || idx >= SGE_NTIMERS)
5829 		return (EINVAL);
5830 
5831 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5832 	    "t4tmr");
5833 	if (rc)
5834 		return (rc);
5835 
5836 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
5837 	for_each_rxq(vi, i, rxq) {
5838 #ifdef atomic_store_rel_8
5839 		atomic_store_rel_8(&rxq->iq.intr_params, v);
5840 #else
5841 		rxq->iq.intr_params = v;
5842 #endif
5843 	}
5844 #ifdef TCP_OFFLOAD
5845 	for_each_ofld_rxq(vi, i, ofld_rxq) {
5846 #ifdef atomic_store_rel_8
5847 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
5848 #else
5849 		ofld_rxq->iq.intr_params = v;
5850 #endif
5851 	}
5852 #endif
5853 	vi->tmr_idx = idx;
5854 
5855 	end_synchronized_op(sc, LOCK_HELD);
5856 	return (0);
5857 }
5858 
5859 static int
5860 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
5861 {
5862 	struct vi_info *vi = arg1;
5863 	struct adapter *sc = vi->pi->adapter;
5864 	int idx, rc;
5865 
5866 	idx = vi->pktc_idx;
5867 
5868 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5869 	if (rc != 0 || req->newptr == NULL)
5870 		return (rc);
5871 
5872 	if (idx < -1 || idx >= SGE_NCOUNTERS)
5873 		return (EINVAL);
5874 
5875 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5876 	    "t4pktc");
5877 	if (rc)
5878 		return (rc);
5879 
5880 	if (vi->flags & VI_INIT_DONE)
5881 		rc = EBUSY; /* cannot be changed once the queues are created */
5882 	else
5883 		vi->pktc_idx = idx;
5884 
5885 	end_synchronized_op(sc, LOCK_HELD);
5886 	return (rc);
5887 }
5888 
5889 static int
5890 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
5891 {
5892 	struct vi_info *vi = arg1;
5893 	struct adapter *sc = vi->pi->adapter;
5894 	int qsize, rc;
5895 
5896 	qsize = vi->qsize_rxq;
5897 
5898 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5899 	if (rc != 0 || req->newptr == NULL)
5900 		return (rc);
5901 
5902 	if (qsize < 128 || (qsize & 7))
5903 		return (EINVAL);
5904 
5905 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5906 	    "t4rxqs");
5907 	if (rc)
5908 		return (rc);
5909 
5910 	if (vi->flags & VI_INIT_DONE)
5911 		rc = EBUSY; /* cannot be changed once the queues are created */
5912 	else
5913 		vi->qsize_rxq = qsize;
5914 
5915 	end_synchronized_op(sc, LOCK_HELD);
5916 	return (rc);
5917 }
5918 
5919 static int
5920 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
5921 {
5922 	struct vi_info *vi = arg1;
5923 	struct adapter *sc = vi->pi->adapter;
5924 	int qsize, rc;
5925 
5926 	qsize = vi->qsize_txq;
5927 
5928 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5929 	if (rc != 0 || req->newptr == NULL)
5930 		return (rc);
5931 
5932 	if (qsize < 128 || qsize > 65536)
5933 		return (EINVAL);
5934 
5935 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5936 	    "t4txqs");
5937 	if (rc)
5938 		return (rc);
5939 
5940 	if (vi->flags & VI_INIT_DONE)
5941 		rc = EBUSY; /* cannot be changed once the queues are created */
5942 	else
5943 		vi->qsize_txq = qsize;
5944 
5945 	end_synchronized_op(sc, LOCK_HELD);
5946 	return (rc);
5947 }
5948 
5949 static int
5950 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
5951 {
5952 	struct port_info *pi = arg1;
5953 	struct adapter *sc = pi->adapter;
5954 	struct link_config *lc = &pi->link_cfg;
5955 	int rc;
5956 
5957 	if (req->newptr == NULL) {
5958 		struct sbuf *sb;
5959 		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
5960 
5961 		rc = sysctl_wire_old_buffer(req, 0);
5962 		if (rc != 0)
5963 			return(rc);
5964 
5965 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5966 		if (sb == NULL)
5967 			return (ENOMEM);
5968 
5969 		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
5970 		rc = sbuf_finish(sb);
5971 		sbuf_delete(sb);
5972 	} else {
5973 		char s[2];
5974 		int n;
5975 
5976 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
5977 		s[1] = 0;
5978 
5979 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5980 		if (rc != 0)
5981 			return(rc);
5982 
5983 		if (s[1] != 0)
5984 			return (EINVAL);
5985 		if (s[0] < '0' || s[0] > '9')
5986 			return (EINVAL);	/* not a number */
5987 		n = s[0] - '0';
5988 		if (n & ~(PAUSE_TX | PAUSE_RX))
5989 			return (EINVAL);	/* some other bit is set too */
5990 
5991 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5992 		    "t4PAUSE");
5993 		if (rc)
5994 			return (rc);
5995 		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
5996 			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
5997 			lc->requested_fc |= n;
5998 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5999 			if (rc == 0) {
6000 				lc->fc = lc->requested_fc;
6001 			}
6002 		}
6003 		end_synchronized_op(sc, 0);
6004 	}
6005 
6006 	return (rc);
6007 }
6008 
6009 static int
6010 sysctl_fec(SYSCTL_HANDLER_ARGS)
6011 {
6012 	struct port_info *pi = arg1;
6013 	struct adapter *sc = pi->adapter;
6014 	struct link_config *lc = &pi->link_cfg;
6015 	int rc;
6016 
6017 	if (req->newptr == NULL) {
6018 		struct sbuf *sb;
6019 		static char *bits = "\20\1RS\2BASER_RS\3RESERVED";
6020 
6021 		rc = sysctl_wire_old_buffer(req, 0);
6022 		if (rc != 0)
6023 			return(rc);
6024 
6025 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6026 		if (sb == NULL)
6027 			return (ENOMEM);
6028 
6029 		sbuf_printf(sb, "%b", lc->fec & M_FW_PORT_CAP_FEC, bits);
6030 		rc = sbuf_finish(sb);
6031 		sbuf_delete(sb);
6032 	} else {
6033 		char s[2];
6034 		int n;
6035 
6036 		s[0] = '0' + (lc->requested_fec & M_FW_PORT_CAP_FEC);
6037 		s[1] = 0;
6038 
6039 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6040 		if (rc != 0)
6041 			return(rc);
6042 
6043 		if (s[1] != 0)
6044 			return (EINVAL);
6045 		if (s[0] < '0' || s[0] > '9')
6046 			return (EINVAL);	/* not a number */
6047 		n = s[0] - '0';
6048 		if (n & ~M_FW_PORT_CAP_FEC)
6049 			return (EINVAL);	/* some other bit is set too */
6050 
6051 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6052 		    "t4fec");
6053 		if (rc)
6054 			return (rc);
6055 		if ((lc->requested_fec & M_FW_PORT_CAP_FEC) != n) {
6056 			lc->requested_fec = n &
6057 			    G_FW_PORT_CAP_FEC(lc->supported);
6058 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6059 			if (rc == 0) {
6060 				lc->fec = lc->requested_fec;
6061 			}
6062 		}
6063 		end_synchronized_op(sc, 0);
6064 	}
6065 
6066 	return (rc);
6067 }
6068 
6069 static int
6070 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
6071 {
6072 	struct port_info *pi = arg1;
6073 	struct adapter *sc = pi->adapter;
6074 	struct link_config *lc = &pi->link_cfg;
6075 	int rc, val, old;
6076 
6077 	if (lc->supported & FW_PORT_CAP_ANEG)
6078 		val = lc->requested_aneg == AUTONEG_ENABLE ? 1 : 0;
6079 	else
6080 		val = -1;
6081 	rc = sysctl_handle_int(oidp, &val, 0, req);
6082 	if (rc != 0 || req->newptr == NULL)
6083 		return (rc);
6084 	if ((lc->supported & FW_PORT_CAP_ANEG) == 0)
6085 		return (ENOTSUP);
6086 
6087 	if (val == 0)
6088 		val = AUTONEG_DISABLE;
6089 	else if (val == 1)
6090 		val = AUTONEG_ENABLE;
6091 	else
6092 		return (EINVAL);
6093 	if (lc->requested_aneg == val)
6094 		return (0);	/* no change */
6095 
6096 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6097 	    "t4aneg");
6098 	if (rc)
6099 		return (rc);
6100 	old = lc->requested_aneg;
6101 	lc->requested_aneg = val;
6102 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
6103 	if (rc != 0)
6104 		lc->requested_aneg = old;
6105 	end_synchronized_op(sc, 0);
6106 	return (rc);
6107 }
6108 
6109 static int
6110 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
6111 {
6112 	struct adapter *sc = arg1;
6113 	int reg = arg2;
6114 	uint64_t val;
6115 
6116 	val = t4_read_reg64(sc, reg);
6117 
6118 	return (sysctl_handle_64(oidp, &val, 0, req));
6119 }
6120 
6121 static int
6122 sysctl_temperature(SYSCTL_HANDLER_ARGS)
6123 {
6124 	struct adapter *sc = arg1;
6125 	int rc, t;
6126 	uint32_t param, val;
6127 
6128 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
6129 	if (rc)
6130 		return (rc);
6131 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
6132 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
6133 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
6134 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6135 	end_synchronized_op(sc, 0);
6136 	if (rc)
6137 		return (rc);
6138 
6139 	/* unknown is returned as 0 but we display -1 in that case */
6140 	t = val == 0 ? -1 : val;
6141 
6142 	rc = sysctl_handle_int(oidp, &t, 0, req);
6143 	return (rc);
6144 }
6145 
6146 #ifdef SBUF_DRAIN
6147 static int
6148 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
6149 {
6150 	struct adapter *sc = arg1;
6151 	struct sbuf *sb;
6152 	int rc, i;
6153 	uint16_t incr[NMTUS][NCCTRL_WIN];
6154 	static const char *dec_fac[] = {
6155 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
6156 		"0.9375"
6157 	};
6158 
6159 	rc = sysctl_wire_old_buffer(req, 0);
6160 	if (rc != 0)
6161 		return (rc);
6162 
6163 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6164 	if (sb == NULL)
6165 		return (ENOMEM);
6166 
6167 	t4_read_cong_tbl(sc, incr);
6168 
6169 	for (i = 0; i < NCCTRL_WIN; ++i) {
6170 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
6171 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
6172 		    incr[5][i], incr[6][i], incr[7][i]);
6173 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
6174 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
6175 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
6176 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
6177 	}
6178 
6179 	rc = sbuf_finish(sb);
6180 	sbuf_delete(sb);
6181 
6182 	return (rc);
6183 }
6184 
6185 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
6186 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
6187 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
6188 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
6189 };
6190 
6191 static int
6192 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
6193 {
6194 	struct adapter *sc = arg1;
6195 	struct sbuf *sb;
6196 	int rc, i, n, qid = arg2;
6197 	uint32_t *buf, *p;
6198 	char *qtype;
6199 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
6200 
6201 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
6202 	    ("%s: bad qid %d\n", __func__, qid));
6203 
6204 	if (qid < CIM_NUM_IBQ) {
6205 		/* inbound queue */
6206 		qtype = "IBQ";
6207 		n = 4 * CIM_IBQ_SIZE;
6208 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6209 		rc = t4_read_cim_ibq(sc, qid, buf, n);
6210 	} else {
6211 		/* outbound queue */
6212 		qtype = "OBQ";
6213 		qid -= CIM_NUM_IBQ;
6214 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
6215 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6216 		rc = t4_read_cim_obq(sc, qid, buf, n);
6217 	}
6218 
6219 	if (rc < 0) {
6220 		rc = -rc;
6221 		goto done;
6222 	}
6223 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
6224 
6225 	rc = sysctl_wire_old_buffer(req, 0);
6226 	if (rc != 0)
6227 		goto done;
6228 
6229 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6230 	if (sb == NULL) {
6231 		rc = ENOMEM;
6232 		goto done;
6233 	}
6234 
6235 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
6236 	for (i = 0, p = buf; i < n; i += 16, p += 4)
6237 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
6238 		    p[2], p[3]);
6239 
6240 	rc = sbuf_finish(sb);
6241 	sbuf_delete(sb);
6242 done:
6243 	free(buf, M_CXGBE);
6244 	return (rc);
6245 }
6246 
6247 static int
6248 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
6249 {
6250 	struct adapter *sc = arg1;
6251 	u_int cfg;
6252 	struct sbuf *sb;
6253 	uint32_t *buf, *p;
6254 	int rc;
6255 
6256 	MPASS(chip_id(sc) <= CHELSIO_T5);
6257 
6258 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6259 	if (rc != 0)
6260 		return (rc);
6261 
6262 	rc = sysctl_wire_old_buffer(req, 0);
6263 	if (rc != 0)
6264 		return (rc);
6265 
6266 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6267 	if (sb == NULL)
6268 		return (ENOMEM);
6269 
6270 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6271 	    M_ZERO | M_WAITOK);
6272 
6273 	rc = -t4_cim_read_la(sc, buf, NULL);
6274 	if (rc != 0)
6275 		goto done;
6276 
6277 	sbuf_printf(sb, "Status   Data      PC%s",
6278 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6279 	    "     LS0Stat  LS0Addr             LS0Data");
6280 
6281 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
6282 		if (cfg & F_UPDBGLACAPTPCONLY) {
6283 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
6284 			    p[6], p[7]);
6285 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
6286 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
6287 			    p[4] & 0xff, p[5] >> 8);
6288 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
6289 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6290 			    p[1] & 0xf, p[2] >> 4);
6291 		} else {
6292 			sbuf_printf(sb,
6293 			    "\n  %02x   %x%07x %x%07x %08x %08x "
6294 			    "%08x%08x%08x%08x",
6295 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6296 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
6297 			    p[6], p[7]);
6298 		}
6299 	}
6300 
6301 	rc = sbuf_finish(sb);
6302 	sbuf_delete(sb);
6303 done:
6304 	free(buf, M_CXGBE);
6305 	return (rc);
6306 }
6307 
6308 static int
6309 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
6310 {
6311 	struct adapter *sc = arg1;
6312 	u_int cfg;
6313 	struct sbuf *sb;
6314 	uint32_t *buf, *p;
6315 	int rc;
6316 
6317 	MPASS(chip_id(sc) > CHELSIO_T5);
6318 
6319 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6320 	if (rc != 0)
6321 		return (rc);
6322 
6323 	rc = sysctl_wire_old_buffer(req, 0);
6324 	if (rc != 0)
6325 		return (rc);
6326 
6327 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6328 	if (sb == NULL)
6329 		return (ENOMEM);
6330 
6331 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6332 	    M_ZERO | M_WAITOK);
6333 
6334 	rc = -t4_cim_read_la(sc, buf, NULL);
6335 	if (rc != 0)
6336 		goto done;
6337 
6338 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
6339 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6340 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
6341 
6342 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
6343 		if (cfg & F_UPDBGLACAPTPCONLY) {
6344 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
6345 			    p[3] & 0xff, p[2], p[1], p[0]);
6346 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
6347 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
6348 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
6349 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
6350 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
6351 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
6352 			    p[6] >> 16);
6353 		} else {
6354 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
6355 			    "%08x %08x %08x %08x %08x %08x",
6356 			    (p[9] >> 16) & 0xff,
6357 			    p[9] & 0xffff, p[8] >> 16,
6358 			    p[8] & 0xffff, p[7] >> 16,
6359 			    p[7] & 0xffff, p[6] >> 16,
6360 			    p[2], p[1], p[0], p[5], p[4], p[3]);
6361 		}
6362 	}
6363 
6364 	rc = sbuf_finish(sb);
6365 	sbuf_delete(sb);
6366 done:
6367 	free(buf, M_CXGBE);
6368 	return (rc);
6369 }
6370 
6371 static int
6372 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
6373 {
6374 	struct adapter *sc = arg1;
6375 	u_int i;
6376 	struct sbuf *sb;
6377 	uint32_t *buf, *p;
6378 	int rc;
6379 
6380 	rc = sysctl_wire_old_buffer(req, 0);
6381 	if (rc != 0)
6382 		return (rc);
6383 
6384 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6385 	if (sb == NULL)
6386 		return (ENOMEM);
6387 
6388 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
6389 	    M_ZERO | M_WAITOK);
6390 
6391 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
6392 	p = buf;
6393 
6394 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6395 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
6396 		    p[1], p[0]);
6397 	}
6398 
6399 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
6400 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6401 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
6402 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
6403 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
6404 		    (p[1] >> 2) | ((p[2] & 3) << 30),
6405 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
6406 		    p[0] & 1);
6407 	}
6408 
6409 	rc = sbuf_finish(sb);
6410 	sbuf_delete(sb);
6411 	free(buf, M_CXGBE);
6412 	return (rc);
6413 }
6414 
6415 static int
6416 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
6417 {
6418 	struct adapter *sc = arg1;
6419 	u_int i;
6420 	struct sbuf *sb;
6421 	uint32_t *buf, *p;
6422 	int rc;
6423 
6424 	rc = sysctl_wire_old_buffer(req, 0);
6425 	if (rc != 0)
6426 		return (rc);
6427 
6428 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6429 	if (sb == NULL)
6430 		return (ENOMEM);
6431 
6432 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
6433 	    M_ZERO | M_WAITOK);
6434 
6435 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
6436 	p = buf;
6437 
6438 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
6439 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6440 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
6441 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
6442 		    p[4], p[3], p[2], p[1], p[0]);
6443 	}
6444 
6445 	sbuf_printf(sb, "\n\nCntl ID               Data");
6446 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6447 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
6448 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
6449 	}
6450 
6451 	rc = sbuf_finish(sb);
6452 	sbuf_delete(sb);
6453 	free(buf, M_CXGBE);
6454 	return (rc);
6455 }
6456 
6457 static int
6458 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
6459 {
6460 	struct adapter *sc = arg1;
6461 	struct sbuf *sb;
6462 	int rc, i;
6463 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6464 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6465 	uint16_t thres[CIM_NUM_IBQ];
6466 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
6467 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
6468 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
6469 
6470 	cim_num_obq = sc->chip_params->cim_num_obq;
6471 	if (is_t4(sc)) {
6472 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
6473 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
6474 	} else {
6475 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
6476 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
6477 	}
6478 	nq = CIM_NUM_IBQ + cim_num_obq;
6479 
6480 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
6481 	if (rc == 0)
6482 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
6483 	if (rc != 0)
6484 		return (rc);
6485 
6486 	t4_read_cimq_cfg(sc, base, size, thres);
6487 
6488 	rc = sysctl_wire_old_buffer(req, 0);
6489 	if (rc != 0)
6490 		return (rc);
6491 
6492 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6493 	if (sb == NULL)
6494 		return (ENOMEM);
6495 
6496 	sbuf_printf(sb,
6497 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
6498 
6499 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
6500 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
6501 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
6502 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6503 		    G_QUEREMFLITS(p[2]) * 16);
6504 	for ( ; i < nq; i++, p += 4, wr += 2)
6505 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
6506 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
6507 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6508 		    G_QUEREMFLITS(p[2]) * 16);
6509 
6510 	rc = sbuf_finish(sb);
6511 	sbuf_delete(sb);
6512 
6513 	return (rc);
6514 }
6515 
6516 static int
6517 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
6518 {
6519 	struct adapter *sc = arg1;
6520 	struct sbuf *sb;
6521 	int rc;
6522 	struct tp_cpl_stats stats;
6523 
6524 	rc = sysctl_wire_old_buffer(req, 0);
6525 	if (rc != 0)
6526 		return (rc);
6527 
6528 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6529 	if (sb == NULL)
6530 		return (ENOMEM);
6531 
6532 	mtx_lock(&sc->reg_lock);
6533 	t4_tp_get_cpl_stats(sc, &stats, 0);
6534 	mtx_unlock(&sc->reg_lock);
6535 
6536 	if (sc->chip_params->nchan > 2) {
6537 		sbuf_printf(sb, "                 channel 0  channel 1"
6538 		    "  channel 2  channel 3");
6539 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
6540 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
6541 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
6542 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
6543 	} else {
6544 		sbuf_printf(sb, "                 channel 0  channel 1");
6545 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
6546 		    stats.req[0], stats.req[1]);
6547 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
6548 		    stats.rsp[0], stats.rsp[1]);
6549 	}
6550 
6551 	rc = sbuf_finish(sb);
6552 	sbuf_delete(sb);
6553 
6554 	return (rc);
6555 }
6556 
6557 static int
6558 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
6559 {
6560 	struct adapter *sc = arg1;
6561 	struct sbuf *sb;
6562 	int rc;
6563 	struct tp_usm_stats stats;
6564 
6565 	rc = sysctl_wire_old_buffer(req, 0);
6566 	if (rc != 0)
6567 		return(rc);
6568 
6569 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6570 	if (sb == NULL)
6571 		return (ENOMEM);
6572 
6573 	t4_get_usm_stats(sc, &stats, 1);
6574 
6575 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
6576 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
6577 	sbuf_printf(sb, "Drops:  %u", stats.drops);
6578 
6579 	rc = sbuf_finish(sb);
6580 	sbuf_delete(sb);
6581 
6582 	return (rc);
6583 }
6584 
6585 static const char * const devlog_level_strings[] = {
6586 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
6587 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
6588 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
6589 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
6590 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
6591 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
6592 };
6593 
6594 static const char * const devlog_facility_strings[] = {
6595 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
6596 	[FW_DEVLOG_FACILITY_CF]		= "CF",
6597 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
6598 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
6599 	[FW_DEVLOG_FACILITY_RES]	= "RES",
6600 	[FW_DEVLOG_FACILITY_HW]		= "HW",
6601 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
6602 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
6603 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
6604 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
6605 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
6606 	[FW_DEVLOG_FACILITY_VI]		= "VI",
6607 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
6608 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
6609 	[FW_DEVLOG_FACILITY_TM]		= "TM",
6610 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
6611 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
6612 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
6613 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
6614 	[FW_DEVLOG_FACILITY_RI]		= "RI",
6615 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
6616 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
6617 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
6618 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
6619 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
6620 };
6621 
6622 static int
6623 sysctl_devlog(SYSCTL_HANDLER_ARGS)
6624 {
6625 	struct adapter *sc = arg1;
6626 	struct devlog_params *dparams = &sc->params.devlog;
6627 	struct fw_devlog_e *buf, *e;
6628 	int i, j, rc, nentries, first = 0;
6629 	struct sbuf *sb;
6630 	uint64_t ftstamp = UINT64_MAX;
6631 
6632 	if (dparams->addr == 0)
6633 		return (ENXIO);
6634 
6635 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
6636 	if (buf == NULL)
6637 		return (ENOMEM);
6638 
6639 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
6640 	if (rc != 0)
6641 		goto done;
6642 
6643 	nentries = dparams->size / sizeof(struct fw_devlog_e);
6644 	for (i = 0; i < nentries; i++) {
6645 		e = &buf[i];
6646 
6647 		if (e->timestamp == 0)
6648 			break;	/* end */
6649 
6650 		e->timestamp = be64toh(e->timestamp);
6651 		e->seqno = be32toh(e->seqno);
6652 		for (j = 0; j < 8; j++)
6653 			e->params[j] = be32toh(e->params[j]);
6654 
6655 		if (e->timestamp < ftstamp) {
6656 			ftstamp = e->timestamp;
6657 			first = i;
6658 		}
6659 	}
6660 
6661 	if (buf[first].timestamp == 0)
6662 		goto done;	/* nothing in the log */
6663 
6664 	rc = sysctl_wire_old_buffer(req, 0);
6665 	if (rc != 0)
6666 		goto done;
6667 
6668 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6669 	if (sb == NULL) {
6670 		rc = ENOMEM;
6671 		goto done;
6672 	}
6673 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
6674 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
6675 
6676 	i = first;
6677 	do {
6678 		e = &buf[i];
6679 		if (e->timestamp == 0)
6680 			break;	/* end */
6681 
6682 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
6683 		    e->seqno, e->timestamp,
6684 		    (e->level < nitems(devlog_level_strings) ?
6685 			devlog_level_strings[e->level] : "UNKNOWN"),
6686 		    (e->facility < nitems(devlog_facility_strings) ?
6687 			devlog_facility_strings[e->facility] : "UNKNOWN"));
6688 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
6689 		    e->params[2], e->params[3], e->params[4],
6690 		    e->params[5], e->params[6], e->params[7]);
6691 
6692 		if (++i == nentries)
6693 			i = 0;
6694 	} while (i != first);
6695 
6696 	rc = sbuf_finish(sb);
6697 	sbuf_delete(sb);
6698 done:
6699 	free(buf, M_CXGBE);
6700 	return (rc);
6701 }
6702 
6703 static int
6704 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
6705 {
6706 	struct adapter *sc = arg1;
6707 	struct sbuf *sb;
6708 	int rc;
6709 	struct tp_fcoe_stats stats[MAX_NCHAN];
6710 	int i, nchan = sc->chip_params->nchan;
6711 
6712 	rc = sysctl_wire_old_buffer(req, 0);
6713 	if (rc != 0)
6714 		return (rc);
6715 
6716 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6717 	if (sb == NULL)
6718 		return (ENOMEM);
6719 
6720 	for (i = 0; i < nchan; i++)
6721 		t4_get_fcoe_stats(sc, i, &stats[i], 1);
6722 
6723 	if (nchan > 2) {
6724 		sbuf_printf(sb, "                   channel 0        channel 1"
6725 		    "        channel 2        channel 3");
6726 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
6727 		    stats[0].octets_ddp, stats[1].octets_ddp,
6728 		    stats[2].octets_ddp, stats[3].octets_ddp);
6729 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
6730 		    stats[0].frames_ddp, stats[1].frames_ddp,
6731 		    stats[2].frames_ddp, stats[3].frames_ddp);
6732 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
6733 		    stats[0].frames_drop, stats[1].frames_drop,
6734 		    stats[2].frames_drop, stats[3].frames_drop);
6735 	} else {
6736 		sbuf_printf(sb, "                   channel 0        channel 1");
6737 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
6738 		    stats[0].octets_ddp, stats[1].octets_ddp);
6739 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
6740 		    stats[0].frames_ddp, stats[1].frames_ddp);
6741 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
6742 		    stats[0].frames_drop, stats[1].frames_drop);
6743 	}
6744 
6745 	rc = sbuf_finish(sb);
6746 	sbuf_delete(sb);
6747 
6748 	return (rc);
6749 }
6750 
6751 static int
6752 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
6753 {
6754 	struct adapter *sc = arg1;
6755 	struct sbuf *sb;
6756 	int rc, i;
6757 	unsigned int map, kbps, ipg, mode;
6758 	unsigned int pace_tab[NTX_SCHED];
6759 
6760 	rc = sysctl_wire_old_buffer(req, 0);
6761 	if (rc != 0)
6762 		return (rc);
6763 
6764 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6765 	if (sb == NULL)
6766 		return (ENOMEM);
6767 
6768 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
6769 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
6770 	t4_read_pace_tbl(sc, pace_tab);
6771 
6772 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
6773 	    "Class IPG (0.1 ns)   Flow IPG (us)");
6774 
6775 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
6776 		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
6777 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
6778 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
6779 		if (kbps)
6780 			sbuf_printf(sb, "%9u     ", kbps);
6781 		else
6782 			sbuf_printf(sb, " disabled     ");
6783 
6784 		if (ipg)
6785 			sbuf_printf(sb, "%13u        ", ipg);
6786 		else
6787 			sbuf_printf(sb, "     disabled        ");
6788 
6789 		if (pace_tab[i])
6790 			sbuf_printf(sb, "%10u", pace_tab[i]);
6791 		else
6792 			sbuf_printf(sb, "  disabled");
6793 	}
6794 
6795 	rc = sbuf_finish(sb);
6796 	sbuf_delete(sb);
6797 
6798 	return (rc);
6799 }
6800 
6801 static int
6802 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
6803 {
6804 	struct adapter *sc = arg1;
6805 	struct sbuf *sb;
6806 	int rc, i, j;
6807 	uint64_t *p0, *p1;
6808 	struct lb_port_stats s[2];
6809 	static const char *stat_name[] = {
6810 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
6811 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
6812 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
6813 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
6814 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
6815 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
6816 		"BG2FramesTrunc:", "BG3FramesTrunc:"
6817 	};
6818 
6819 	rc = sysctl_wire_old_buffer(req, 0);
6820 	if (rc != 0)
6821 		return (rc);
6822 
6823 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6824 	if (sb == NULL)
6825 		return (ENOMEM);
6826 
6827 	memset(s, 0, sizeof(s));
6828 
6829 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
6830 		t4_get_lb_stats(sc, i, &s[0]);
6831 		t4_get_lb_stats(sc, i + 1, &s[1]);
6832 
6833 		p0 = &s[0].octets;
6834 		p1 = &s[1].octets;
6835 		sbuf_printf(sb, "%s                       Loopback %u"
6836 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
6837 
6838 		for (j = 0; j < nitems(stat_name); j++)
6839 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
6840 				   *p0++, *p1++);
6841 	}
6842 
6843 	rc = sbuf_finish(sb);
6844 	sbuf_delete(sb);
6845 
6846 	return (rc);
6847 }
6848 
6849 static int
6850 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
6851 {
6852 	int rc = 0;
6853 	struct port_info *pi = arg1;
6854 	struct link_config *lc = &pi->link_cfg;
6855 	struct sbuf *sb;
6856 
6857 	rc = sysctl_wire_old_buffer(req, 0);
6858 	if (rc != 0)
6859 		return(rc);
6860 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
6861 	if (sb == NULL)
6862 		return (ENOMEM);
6863 
6864 	if (lc->link_ok || lc->link_down_rc == 255)
6865 		sbuf_printf(sb, "n/a");
6866 	else
6867 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
6868 
6869 	rc = sbuf_finish(sb);
6870 	sbuf_delete(sb);
6871 
6872 	return (rc);
6873 }
6874 
6875 struct mem_desc {
6876 	unsigned int base;
6877 	unsigned int limit;
6878 	unsigned int idx;
6879 };
6880 
6881 static int
6882 mem_desc_cmp(const void *a, const void *b)
6883 {
6884 	return ((const struct mem_desc *)a)->base -
6885 	       ((const struct mem_desc *)b)->base;
6886 }
6887 
6888 static void
6889 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
6890     unsigned int to)
6891 {
6892 	unsigned int size;
6893 
6894 	if (from == to)
6895 		return;
6896 
6897 	size = to - from + 1;
6898 	if (size == 0)
6899 		return;
6900 
6901 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
6902 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
6903 }
6904 
6905 static int
6906 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
6907 {
6908 	struct adapter *sc = arg1;
6909 	struct sbuf *sb;
6910 	int rc, i, n;
6911 	uint32_t lo, hi, used, alloc;
6912 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
6913 	static const char *region[] = {
6914 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
6915 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
6916 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
6917 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
6918 		"RQUDP region:", "PBL region:", "TXPBL region:",
6919 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
6920 		"On-chip queues:"
6921 	};
6922 	struct mem_desc avail[4];
6923 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
6924 	struct mem_desc *md = mem;
6925 
6926 	rc = sysctl_wire_old_buffer(req, 0);
6927 	if (rc != 0)
6928 		return (rc);
6929 
6930 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6931 	if (sb == NULL)
6932 		return (ENOMEM);
6933 
6934 	for (i = 0; i < nitems(mem); i++) {
6935 		mem[i].limit = 0;
6936 		mem[i].idx = i;
6937 	}
6938 
6939 	/* Find and sort the populated memory ranges */
6940 	i = 0;
6941 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
6942 	if (lo & F_EDRAM0_ENABLE) {
6943 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
6944 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
6945 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
6946 		avail[i].idx = 0;
6947 		i++;
6948 	}
6949 	if (lo & F_EDRAM1_ENABLE) {
6950 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
6951 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
6952 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
6953 		avail[i].idx = 1;
6954 		i++;
6955 	}
6956 	if (lo & F_EXT_MEM_ENABLE) {
6957 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
6958 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
6959 		avail[i].limit = avail[i].base +
6960 		    (G_EXT_MEM_SIZE(hi) << 20);
6961 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
6962 		i++;
6963 	}
6964 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
6965 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
6966 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
6967 		avail[i].limit = avail[i].base +
6968 		    (G_EXT_MEM1_SIZE(hi) << 20);
6969 		avail[i].idx = 4;
6970 		i++;
6971 	}
6972 	if (!i)                                    /* no memory available */
6973 		return 0;
6974 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
6975 
6976 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
6977 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
6978 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
6979 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
6980 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
6981 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
6982 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
6983 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
6984 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
6985 
6986 	/* the next few have explicit upper bounds */
6987 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
6988 	md->limit = md->base - 1 +
6989 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
6990 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
6991 	md++;
6992 
6993 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
6994 	md->limit = md->base - 1 +
6995 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
6996 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
6997 	md++;
6998 
6999 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7000 		if (chip_id(sc) <= CHELSIO_T5)
7001 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
7002 		else
7003 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
7004 		md->limit = 0;
7005 	} else {
7006 		md->base = 0;
7007 		md->idx = nitems(region);  /* hide it */
7008 	}
7009 	md++;
7010 
7011 #define ulp_region(reg) \
7012 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
7013 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
7014 
7015 	ulp_region(RX_ISCSI);
7016 	ulp_region(RX_TDDP);
7017 	ulp_region(TX_TPT);
7018 	ulp_region(RX_STAG);
7019 	ulp_region(RX_RQ);
7020 	ulp_region(RX_RQUDP);
7021 	ulp_region(RX_PBL);
7022 	ulp_region(TX_PBL);
7023 #undef ulp_region
7024 
7025 	md->base = 0;
7026 	md->idx = nitems(region);
7027 	if (!is_t4(sc)) {
7028 		uint32_t size = 0;
7029 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
7030 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
7031 
7032 		if (is_t5(sc)) {
7033 			if (sge_ctrl & F_VFIFO_ENABLE)
7034 				size = G_DBVFIFO_SIZE(fifo_size);
7035 		} else
7036 			size = G_T6_DBVFIFO_SIZE(fifo_size);
7037 
7038 		if (size) {
7039 			md->base = G_BASEADDR(t4_read_reg(sc,
7040 			    A_SGE_DBVFIFO_BADDR));
7041 			md->limit = md->base + (size << 2) - 1;
7042 		}
7043 	}
7044 	md++;
7045 
7046 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
7047 	md->limit = 0;
7048 	md++;
7049 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
7050 	md->limit = 0;
7051 	md++;
7052 
7053 	md->base = sc->vres.ocq.start;
7054 	if (sc->vres.ocq.size)
7055 		md->limit = md->base + sc->vres.ocq.size - 1;
7056 	else
7057 		md->idx = nitems(region);  /* hide it */
7058 	md++;
7059 
7060 	/* add any address-space holes, there can be up to 3 */
7061 	for (n = 0; n < i - 1; n++)
7062 		if (avail[n].limit < avail[n + 1].base)
7063 			(md++)->base = avail[n].limit;
7064 	if (avail[n].limit)
7065 		(md++)->base = avail[n].limit;
7066 
7067 	n = md - mem;
7068 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
7069 
7070 	for (lo = 0; lo < i; lo++)
7071 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
7072 				avail[lo].limit - 1);
7073 
7074 	sbuf_printf(sb, "\n");
7075 	for (i = 0; i < n; i++) {
7076 		if (mem[i].idx >= nitems(region))
7077 			continue;                        /* skip holes */
7078 		if (!mem[i].limit)
7079 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
7080 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
7081 				mem[i].limit);
7082 	}
7083 
7084 	sbuf_printf(sb, "\n");
7085 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
7086 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
7087 	mem_region_show(sb, "uP RAM:", lo, hi);
7088 
7089 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
7090 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
7091 	mem_region_show(sb, "uP Extmem2:", lo, hi);
7092 
7093 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
7094 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
7095 		   G_PMRXMAXPAGE(lo),
7096 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
7097 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
7098 
7099 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
7100 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
7101 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
7102 		   G_PMTXMAXPAGE(lo),
7103 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
7104 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
7105 	sbuf_printf(sb, "%u p-structs\n",
7106 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
7107 
7108 	for (i = 0; i < 4; i++) {
7109 		if (chip_id(sc) > CHELSIO_T5)
7110 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
7111 		else
7112 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
7113 		if (is_t5(sc)) {
7114 			used = G_T5_USED(lo);
7115 			alloc = G_T5_ALLOC(lo);
7116 		} else {
7117 			used = G_USED(lo);
7118 			alloc = G_ALLOC(lo);
7119 		}
7120 		/* For T6 these are MAC buffer groups */
7121 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
7122 		    i, used, alloc);
7123 	}
7124 	for (i = 0; i < sc->chip_params->nchan; i++) {
7125 		if (chip_id(sc) > CHELSIO_T5)
7126 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
7127 		else
7128 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
7129 		if (is_t5(sc)) {
7130 			used = G_T5_USED(lo);
7131 			alloc = G_T5_ALLOC(lo);
7132 		} else {
7133 			used = G_USED(lo);
7134 			alloc = G_ALLOC(lo);
7135 		}
7136 		/* For T6 these are MAC buffer groups */
7137 		sbuf_printf(sb,
7138 		    "\nLoopback %d using %u pages out of %u allocated",
7139 		    i, used, alloc);
7140 	}
7141 
7142 	rc = sbuf_finish(sb);
7143 	sbuf_delete(sb);
7144 
7145 	return (rc);
7146 }
7147 
7148 static inline void
7149 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
7150 {
7151 	*mask = x | y;
7152 	y = htobe64(y);
7153 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
7154 }
7155 
7156 static int
7157 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
7158 {
7159 	struct adapter *sc = arg1;
7160 	struct sbuf *sb;
7161 	int rc, i;
7162 
7163 	MPASS(chip_id(sc) <= CHELSIO_T5);
7164 
7165 	rc = sysctl_wire_old_buffer(req, 0);
7166 	if (rc != 0)
7167 		return (rc);
7168 
7169 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7170 	if (sb == NULL)
7171 		return (ENOMEM);
7172 
7173 	sbuf_printf(sb,
7174 	    "Idx  Ethernet address     Mask     Vld Ports PF"
7175 	    "  VF              Replication             P0 P1 P2 P3  ML");
7176 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7177 		uint64_t tcamx, tcamy, mask;
7178 		uint32_t cls_lo, cls_hi;
7179 		uint8_t addr[ETHER_ADDR_LEN];
7180 
7181 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
7182 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
7183 		if (tcamx & tcamy)
7184 			continue;
7185 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7186 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7187 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7188 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
7189 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
7190 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
7191 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
7192 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
7193 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
7194 
7195 		if (cls_lo & F_REPLICATE) {
7196 			struct fw_ldst_cmd ldst_cmd;
7197 
7198 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7199 			ldst_cmd.op_to_addrspace =
7200 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7201 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7202 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7203 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7204 			ldst_cmd.u.mps.rplc.fid_idx =
7205 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7206 				V_FW_LDST_CMD_IDX(i));
7207 
7208 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7209 			    "t4mps");
7210 			if (rc)
7211 				break;
7212 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7213 			    sizeof(ldst_cmd), &ldst_cmd);
7214 			end_synchronized_op(sc, 0);
7215 
7216 			if (rc != 0) {
7217 				sbuf_printf(sb, "%36d", rc);
7218 				rc = 0;
7219 			} else {
7220 				sbuf_printf(sb, " %08x %08x %08x %08x",
7221 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7222 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7223 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7224 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7225 			}
7226 		} else
7227 			sbuf_printf(sb, "%36s", "");
7228 
7229 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
7230 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
7231 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
7232 	}
7233 
7234 	if (rc)
7235 		(void) sbuf_finish(sb);
7236 	else
7237 		rc = sbuf_finish(sb);
7238 	sbuf_delete(sb);
7239 
7240 	return (rc);
7241 }
7242 
7243 static int
7244 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
7245 {
7246 	struct adapter *sc = arg1;
7247 	struct sbuf *sb;
7248 	int rc, i;
7249 
7250 	MPASS(chip_id(sc) > CHELSIO_T5);
7251 
7252 	rc = sysctl_wire_old_buffer(req, 0);
7253 	if (rc != 0)
7254 		return (rc);
7255 
7256 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7257 	if (sb == NULL)
7258 		return (ENOMEM);
7259 
7260 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
7261 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
7262 	    "                           Replication"
7263 	    "                                    P0 P1 P2 P3  ML\n");
7264 
7265 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7266 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
7267 		uint16_t ivlan;
7268 		uint64_t tcamx, tcamy, val, mask;
7269 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
7270 		uint8_t addr[ETHER_ADDR_LEN];
7271 
7272 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
7273 		if (i < 256)
7274 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
7275 		else
7276 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
7277 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7278 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7279 		tcamy = G_DMACH(val) << 32;
7280 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7281 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7282 		lookup_type = G_DATALKPTYPE(data2);
7283 		port_num = G_DATAPORTNUM(data2);
7284 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7285 			/* Inner header VNI */
7286 			vniy = ((data2 & F_DATAVIDH2) << 23) |
7287 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7288 			dip_hit = data2 & F_DATADIPHIT;
7289 			vlan_vld = 0;
7290 		} else {
7291 			vniy = 0;
7292 			dip_hit = 0;
7293 			vlan_vld = data2 & F_DATAVIDH2;
7294 			ivlan = G_VIDL(val);
7295 		}
7296 
7297 		ctl |= V_CTLXYBITSEL(1);
7298 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7299 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7300 		tcamx = G_DMACH(val) << 32;
7301 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7302 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7303 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7304 			/* Inner header VNI mask */
7305 			vnix = ((data2 & F_DATAVIDH2) << 23) |
7306 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7307 		} else
7308 			vnix = 0;
7309 
7310 		if (tcamx & tcamy)
7311 			continue;
7312 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7313 
7314 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7315 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7316 
7317 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7318 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7319 			    "%012jx %06x %06x    -    -   %3c"
7320 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
7321 			    addr[1], addr[2], addr[3], addr[4], addr[5],
7322 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
7323 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7324 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7325 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7326 		} else {
7327 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7328 			    "%012jx    -       -   ", i, addr[0], addr[1],
7329 			    addr[2], addr[3], addr[4], addr[5],
7330 			    (uintmax_t)mask);
7331 
7332 			if (vlan_vld)
7333 				sbuf_printf(sb, "%4u   Y     ", ivlan);
7334 			else
7335 				sbuf_printf(sb, "  -    N     ");
7336 
7337 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
7338 			    lookup_type ? 'I' : 'O', port_num,
7339 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7340 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7341 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7342 		}
7343 
7344 
7345 		if (cls_lo & F_T6_REPLICATE) {
7346 			struct fw_ldst_cmd ldst_cmd;
7347 
7348 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7349 			ldst_cmd.op_to_addrspace =
7350 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7351 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7352 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7353 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7354 			ldst_cmd.u.mps.rplc.fid_idx =
7355 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7356 				V_FW_LDST_CMD_IDX(i));
7357 
7358 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7359 			    "t6mps");
7360 			if (rc)
7361 				break;
7362 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7363 			    sizeof(ldst_cmd), &ldst_cmd);
7364 			end_synchronized_op(sc, 0);
7365 
7366 			if (rc != 0) {
7367 				sbuf_printf(sb, "%72d", rc);
7368 				rc = 0;
7369 			} else {
7370 				sbuf_printf(sb, " %08x %08x %08x %08x"
7371 				    " %08x %08x %08x %08x",
7372 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
7373 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
7374 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
7375 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
7376 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7377 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7378 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7379 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7380 			}
7381 		} else
7382 			sbuf_printf(sb, "%72s", "");
7383 
7384 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
7385 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
7386 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
7387 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
7388 	}
7389 
7390 	if (rc)
7391 		(void) sbuf_finish(sb);
7392 	else
7393 		rc = sbuf_finish(sb);
7394 	sbuf_delete(sb);
7395 
7396 	return (rc);
7397 }
7398 
7399 static int
7400 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
7401 {
7402 	struct adapter *sc = arg1;
7403 	struct sbuf *sb;
7404 	int rc;
7405 	uint16_t mtus[NMTUS];
7406 
7407 	rc = sysctl_wire_old_buffer(req, 0);
7408 	if (rc != 0)
7409 		return (rc);
7410 
7411 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7412 	if (sb == NULL)
7413 		return (ENOMEM);
7414 
7415 	t4_read_mtu_tbl(sc, mtus, NULL);
7416 
7417 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
7418 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
7419 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
7420 	    mtus[14], mtus[15]);
7421 
7422 	rc = sbuf_finish(sb);
7423 	sbuf_delete(sb);
7424 
7425 	return (rc);
7426 }
7427 
7428 static int
7429 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
7430 {
7431 	struct adapter *sc = arg1;
7432 	struct sbuf *sb;
7433 	int rc, i;
7434 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
7435 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
7436 	static const char *tx_stats[MAX_PM_NSTATS] = {
7437 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
7438 		"Tx FIFO wait", NULL, "Tx latency"
7439 	};
7440 	static const char *rx_stats[MAX_PM_NSTATS] = {
7441 		"Read:", "Write bypass:", "Write mem:", "Flush:",
7442 		"Rx FIFO wait", NULL, "Rx latency"
7443 	};
7444 
7445 	rc = sysctl_wire_old_buffer(req, 0);
7446 	if (rc != 0)
7447 		return (rc);
7448 
7449 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7450 	if (sb == NULL)
7451 		return (ENOMEM);
7452 
7453 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
7454 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
7455 
7456 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
7457 	for (i = 0; i < 4; i++) {
7458 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7459 		    tx_cyc[i]);
7460 	}
7461 
7462 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
7463 	for (i = 0; i < 4; i++) {
7464 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7465 		    rx_cyc[i]);
7466 	}
7467 
7468 	if (chip_id(sc) > CHELSIO_T5) {
7469 		sbuf_printf(sb,
7470 		    "\n              Total wait      Total occupancy");
7471 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7472 		    tx_cyc[i]);
7473 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7474 		    rx_cyc[i]);
7475 
7476 		i += 2;
7477 		MPASS(i < nitems(tx_stats));
7478 
7479 		sbuf_printf(sb,
7480 		    "\n                   Reads           Total wait");
7481 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7482 		    tx_cyc[i]);
7483 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7484 		    rx_cyc[i]);
7485 	}
7486 
7487 	rc = sbuf_finish(sb);
7488 	sbuf_delete(sb);
7489 
7490 	return (rc);
7491 }
7492 
7493 static int
7494 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
7495 {
7496 	struct adapter *sc = arg1;
7497 	struct sbuf *sb;
7498 	int rc;
7499 	struct tp_rdma_stats stats;
7500 
7501 	rc = sysctl_wire_old_buffer(req, 0);
7502 	if (rc != 0)
7503 		return (rc);
7504 
7505 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7506 	if (sb == NULL)
7507 		return (ENOMEM);
7508 
7509 	mtx_lock(&sc->reg_lock);
7510 	t4_tp_get_rdma_stats(sc, &stats, 0);
7511 	mtx_unlock(&sc->reg_lock);
7512 
7513 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
7514 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
7515 
7516 	rc = sbuf_finish(sb);
7517 	sbuf_delete(sb);
7518 
7519 	return (rc);
7520 }
7521 
7522 static int
7523 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
7524 {
7525 	struct adapter *sc = arg1;
7526 	struct sbuf *sb;
7527 	int rc;
7528 	struct tp_tcp_stats v4, v6;
7529 
7530 	rc = sysctl_wire_old_buffer(req, 0);
7531 	if (rc != 0)
7532 		return (rc);
7533 
7534 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7535 	if (sb == NULL)
7536 		return (ENOMEM);
7537 
7538 	mtx_lock(&sc->reg_lock);
7539 	t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
7540 	mtx_unlock(&sc->reg_lock);
7541 
7542 	sbuf_printf(sb,
7543 	    "                                IP                 IPv6\n");
7544 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
7545 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
7546 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
7547 	    v4.tcp_in_segs, v6.tcp_in_segs);
7548 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
7549 	    v4.tcp_out_segs, v6.tcp_out_segs);
7550 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
7551 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
7552 
7553 	rc = sbuf_finish(sb);
7554 	sbuf_delete(sb);
7555 
7556 	return (rc);
7557 }
7558 
7559 static int
7560 sysctl_tids(SYSCTL_HANDLER_ARGS)
7561 {
7562 	struct adapter *sc = arg1;
7563 	struct sbuf *sb;
7564 	int rc;
7565 	struct tid_info *t = &sc->tids;
7566 
7567 	rc = sysctl_wire_old_buffer(req, 0);
7568 	if (rc != 0)
7569 		return (rc);
7570 
7571 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7572 	if (sb == NULL)
7573 		return (ENOMEM);
7574 
7575 	if (t->natids) {
7576 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
7577 		    t->atids_in_use);
7578 	}
7579 
7580 	if (t->ntids) {
7581 		sbuf_printf(sb, "TID range: ");
7582 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7583 			uint32_t b, hb;
7584 
7585 			if (chip_id(sc) <= CHELSIO_T5) {
7586 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
7587 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
7588 			} else {
7589 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
7590 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
7591 			}
7592 
7593 			if (b)
7594 				sbuf_printf(sb, "0-%u, ", b - 1);
7595 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
7596 		} else
7597 			sbuf_printf(sb, "0-%u", t->ntids - 1);
7598 		sbuf_printf(sb, ", in use: %u\n",
7599 		    atomic_load_acq_int(&t->tids_in_use));
7600 	}
7601 
7602 	if (t->nstids) {
7603 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
7604 		    t->stid_base + t->nstids - 1, t->stids_in_use);
7605 	}
7606 
7607 	if (t->nftids) {
7608 		sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
7609 		    t->ftid_base + t->nftids - 1);
7610 	}
7611 
7612 	if (t->netids) {
7613 		sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
7614 		    t->etid_base + t->netids - 1);
7615 	}
7616 
7617 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
7618 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
7619 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
7620 
7621 	rc = sbuf_finish(sb);
7622 	sbuf_delete(sb);
7623 
7624 	return (rc);
7625 }
7626 
7627 static int
7628 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
7629 {
7630 	struct adapter *sc = arg1;
7631 	struct sbuf *sb;
7632 	int rc;
7633 	struct tp_err_stats stats;
7634 
7635 	rc = sysctl_wire_old_buffer(req, 0);
7636 	if (rc != 0)
7637 		return (rc);
7638 
7639 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7640 	if (sb == NULL)
7641 		return (ENOMEM);
7642 
7643 	mtx_lock(&sc->reg_lock);
7644 	t4_tp_get_err_stats(sc, &stats, 0);
7645 	mtx_unlock(&sc->reg_lock);
7646 
7647 	if (sc->chip_params->nchan > 2) {
7648 		sbuf_printf(sb, "                 channel 0  channel 1"
7649 		    "  channel 2  channel 3\n");
7650 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
7651 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
7652 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
7653 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
7654 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
7655 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
7656 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
7657 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
7658 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
7659 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
7660 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
7661 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
7662 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
7663 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
7664 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
7665 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
7666 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
7667 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
7668 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
7669 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
7670 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
7671 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
7672 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
7673 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
7674 	} else {
7675 		sbuf_printf(sb, "                 channel 0  channel 1\n");
7676 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
7677 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
7678 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
7679 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
7680 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
7681 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
7682 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
7683 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
7684 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
7685 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
7686 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
7687 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
7688 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
7689 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
7690 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
7691 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
7692 	}
7693 
7694 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
7695 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
7696 
7697 	rc = sbuf_finish(sb);
7698 	sbuf_delete(sb);
7699 
7700 	return (rc);
7701 }
7702 
7703 static int
7704 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
7705 {
7706 	struct adapter *sc = arg1;
7707 	struct tp_params *tpp = &sc->params.tp;
7708 	u_int mask;
7709 	int rc;
7710 
7711 	mask = tpp->la_mask >> 16;
7712 	rc = sysctl_handle_int(oidp, &mask, 0, req);
7713 	if (rc != 0 || req->newptr == NULL)
7714 		return (rc);
7715 	if (mask > 0xffff)
7716 		return (EINVAL);
7717 	tpp->la_mask = mask << 16;
7718 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
7719 
7720 	return (0);
7721 }
7722 
7723 struct field_desc {
7724 	const char *name;
7725 	u_int start;
7726 	u_int width;
7727 };
7728 
7729 static void
7730 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
7731 {
7732 	char buf[32];
7733 	int line_size = 0;
7734 
7735 	while (f->name) {
7736 		uint64_t mask = (1ULL << f->width) - 1;
7737 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
7738 		    ((uintmax_t)v >> f->start) & mask);
7739 
7740 		if (line_size + len >= 79) {
7741 			line_size = 8;
7742 			sbuf_printf(sb, "\n        ");
7743 		}
7744 		sbuf_printf(sb, "%s ", buf);
7745 		line_size += len + 1;
7746 		f++;
7747 	}
7748 	sbuf_printf(sb, "\n");
7749 }
7750 
7751 static const struct field_desc tp_la0[] = {
7752 	{ "RcfOpCodeOut", 60, 4 },
7753 	{ "State", 56, 4 },
7754 	{ "WcfState", 52, 4 },
7755 	{ "RcfOpcSrcOut", 50, 2 },
7756 	{ "CRxError", 49, 1 },
7757 	{ "ERxError", 48, 1 },
7758 	{ "SanityFailed", 47, 1 },
7759 	{ "SpuriousMsg", 46, 1 },
7760 	{ "FlushInputMsg", 45, 1 },
7761 	{ "FlushInputCpl", 44, 1 },
7762 	{ "RssUpBit", 43, 1 },
7763 	{ "RssFilterHit", 42, 1 },
7764 	{ "Tid", 32, 10 },
7765 	{ "InitTcb", 31, 1 },
7766 	{ "LineNumber", 24, 7 },
7767 	{ "Emsg", 23, 1 },
7768 	{ "EdataOut", 22, 1 },
7769 	{ "Cmsg", 21, 1 },
7770 	{ "CdataOut", 20, 1 },
7771 	{ "EreadPdu", 19, 1 },
7772 	{ "CreadPdu", 18, 1 },
7773 	{ "TunnelPkt", 17, 1 },
7774 	{ "RcfPeerFin", 16, 1 },
7775 	{ "RcfReasonOut", 12, 4 },
7776 	{ "TxCchannel", 10, 2 },
7777 	{ "RcfTxChannel", 8, 2 },
7778 	{ "RxEchannel", 6, 2 },
7779 	{ "RcfRxChannel", 5, 1 },
7780 	{ "RcfDataOutSrdy", 4, 1 },
7781 	{ "RxDvld", 3, 1 },
7782 	{ "RxOoDvld", 2, 1 },
7783 	{ "RxCongestion", 1, 1 },
7784 	{ "TxCongestion", 0, 1 },
7785 	{ NULL }
7786 };
7787 
7788 static const struct field_desc tp_la1[] = {
7789 	{ "CplCmdIn", 56, 8 },
7790 	{ "CplCmdOut", 48, 8 },
7791 	{ "ESynOut", 47, 1 },
7792 	{ "EAckOut", 46, 1 },
7793 	{ "EFinOut", 45, 1 },
7794 	{ "ERstOut", 44, 1 },
7795 	{ "SynIn", 43, 1 },
7796 	{ "AckIn", 42, 1 },
7797 	{ "FinIn", 41, 1 },
7798 	{ "RstIn", 40, 1 },
7799 	{ "DataIn", 39, 1 },
7800 	{ "DataInVld", 38, 1 },
7801 	{ "PadIn", 37, 1 },
7802 	{ "RxBufEmpty", 36, 1 },
7803 	{ "RxDdp", 35, 1 },
7804 	{ "RxFbCongestion", 34, 1 },
7805 	{ "TxFbCongestion", 33, 1 },
7806 	{ "TxPktSumSrdy", 32, 1 },
7807 	{ "RcfUlpType", 28, 4 },
7808 	{ "Eread", 27, 1 },
7809 	{ "Ebypass", 26, 1 },
7810 	{ "Esave", 25, 1 },
7811 	{ "Static0", 24, 1 },
7812 	{ "Cread", 23, 1 },
7813 	{ "Cbypass", 22, 1 },
7814 	{ "Csave", 21, 1 },
7815 	{ "CPktOut", 20, 1 },
7816 	{ "RxPagePoolFull", 18, 2 },
7817 	{ "RxLpbkPkt", 17, 1 },
7818 	{ "TxLpbkPkt", 16, 1 },
7819 	{ "RxVfValid", 15, 1 },
7820 	{ "SynLearned", 14, 1 },
7821 	{ "SetDelEntry", 13, 1 },
7822 	{ "SetInvEntry", 12, 1 },
7823 	{ "CpcmdDvld", 11, 1 },
7824 	{ "CpcmdSave", 10, 1 },
7825 	{ "RxPstructsFull", 8, 2 },
7826 	{ "EpcmdDvld", 7, 1 },
7827 	{ "EpcmdFlush", 6, 1 },
7828 	{ "EpcmdTrimPrefix", 5, 1 },
7829 	{ "EpcmdTrimPostfix", 4, 1 },
7830 	{ "ERssIp4Pkt", 3, 1 },
7831 	{ "ERssIp6Pkt", 2, 1 },
7832 	{ "ERssTcpUdpPkt", 1, 1 },
7833 	{ "ERssFceFipPkt", 0, 1 },
7834 	{ NULL }
7835 };
7836 
7837 static const struct field_desc tp_la2[] = {
7838 	{ "CplCmdIn", 56, 8 },
7839 	{ "MpsVfVld", 55, 1 },
7840 	{ "MpsPf", 52, 3 },
7841 	{ "MpsVf", 44, 8 },
7842 	{ "SynIn", 43, 1 },
7843 	{ "AckIn", 42, 1 },
7844 	{ "FinIn", 41, 1 },
7845 	{ "RstIn", 40, 1 },
7846 	{ "DataIn", 39, 1 },
7847 	{ "DataInVld", 38, 1 },
7848 	{ "PadIn", 37, 1 },
7849 	{ "RxBufEmpty", 36, 1 },
7850 	{ "RxDdp", 35, 1 },
7851 	{ "RxFbCongestion", 34, 1 },
7852 	{ "TxFbCongestion", 33, 1 },
7853 	{ "TxPktSumSrdy", 32, 1 },
7854 	{ "RcfUlpType", 28, 4 },
7855 	{ "Eread", 27, 1 },
7856 	{ "Ebypass", 26, 1 },
7857 	{ "Esave", 25, 1 },
7858 	{ "Static0", 24, 1 },
7859 	{ "Cread", 23, 1 },
7860 	{ "Cbypass", 22, 1 },
7861 	{ "Csave", 21, 1 },
7862 	{ "CPktOut", 20, 1 },
7863 	{ "RxPagePoolFull", 18, 2 },
7864 	{ "RxLpbkPkt", 17, 1 },
7865 	{ "TxLpbkPkt", 16, 1 },
7866 	{ "RxVfValid", 15, 1 },
7867 	{ "SynLearned", 14, 1 },
7868 	{ "SetDelEntry", 13, 1 },
7869 	{ "SetInvEntry", 12, 1 },
7870 	{ "CpcmdDvld", 11, 1 },
7871 	{ "CpcmdSave", 10, 1 },
7872 	{ "RxPstructsFull", 8, 2 },
7873 	{ "EpcmdDvld", 7, 1 },
7874 	{ "EpcmdFlush", 6, 1 },
7875 	{ "EpcmdTrimPrefix", 5, 1 },
7876 	{ "EpcmdTrimPostfix", 4, 1 },
7877 	{ "ERssIp4Pkt", 3, 1 },
7878 	{ "ERssIp6Pkt", 2, 1 },
7879 	{ "ERssTcpUdpPkt", 1, 1 },
7880 	{ "ERssFceFipPkt", 0, 1 },
7881 	{ NULL }
7882 };
7883 
7884 static void
7885 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
7886 {
7887 
7888 	field_desc_show(sb, *p, tp_la0);
7889 }
7890 
7891 static void
7892 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
7893 {
7894 
7895 	if (idx)
7896 		sbuf_printf(sb, "\n");
7897 	field_desc_show(sb, p[0], tp_la0);
7898 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7899 		field_desc_show(sb, p[1], tp_la0);
7900 }
7901 
7902 static void
7903 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
7904 {
7905 
7906 	if (idx)
7907 		sbuf_printf(sb, "\n");
7908 	field_desc_show(sb, p[0], tp_la0);
7909 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7910 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
7911 }
7912 
7913 static int
7914 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
7915 {
7916 	struct adapter *sc = arg1;
7917 	struct sbuf *sb;
7918 	uint64_t *buf, *p;
7919 	int rc;
7920 	u_int i, inc;
7921 	void (*show_func)(struct sbuf *, uint64_t *, int);
7922 
7923 	rc = sysctl_wire_old_buffer(req, 0);
7924 	if (rc != 0)
7925 		return (rc);
7926 
7927 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7928 	if (sb == NULL)
7929 		return (ENOMEM);
7930 
7931 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
7932 
7933 	t4_tp_read_la(sc, buf, NULL);
7934 	p = buf;
7935 
7936 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
7937 	case 2:
7938 		inc = 2;
7939 		show_func = tp_la_show2;
7940 		break;
7941 	case 3:
7942 		inc = 2;
7943 		show_func = tp_la_show3;
7944 		break;
7945 	default:
7946 		inc = 1;
7947 		show_func = tp_la_show;
7948 	}
7949 
7950 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
7951 		(*show_func)(sb, p, i);
7952 
7953 	rc = sbuf_finish(sb);
7954 	sbuf_delete(sb);
7955 	free(buf, M_CXGBE);
7956 	return (rc);
7957 }
7958 
7959 static int
7960 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
7961 {
7962 	struct adapter *sc = arg1;
7963 	struct sbuf *sb;
7964 	int rc;
7965 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
7966 
7967 	rc = sysctl_wire_old_buffer(req, 0);
7968 	if (rc != 0)
7969 		return (rc);
7970 
7971 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7972 	if (sb == NULL)
7973 		return (ENOMEM);
7974 
7975 	t4_get_chan_txrate(sc, nrate, orate);
7976 
7977 	if (sc->chip_params->nchan > 2) {
7978 		sbuf_printf(sb, "              channel 0   channel 1"
7979 		    "   channel 2   channel 3\n");
7980 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
7981 		    nrate[0], nrate[1], nrate[2], nrate[3]);
7982 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
7983 		    orate[0], orate[1], orate[2], orate[3]);
7984 	} else {
7985 		sbuf_printf(sb, "              channel 0   channel 1\n");
7986 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
7987 		    nrate[0], nrate[1]);
7988 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
7989 		    orate[0], orate[1]);
7990 	}
7991 
7992 	rc = sbuf_finish(sb);
7993 	sbuf_delete(sb);
7994 
7995 	return (rc);
7996 }
7997 
7998 static int
7999 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
8000 {
8001 	struct adapter *sc = arg1;
8002 	struct sbuf *sb;
8003 	uint32_t *buf, *p;
8004 	int rc, i;
8005 
8006 	rc = sysctl_wire_old_buffer(req, 0);
8007 	if (rc != 0)
8008 		return (rc);
8009 
8010 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8011 	if (sb == NULL)
8012 		return (ENOMEM);
8013 
8014 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
8015 	    M_ZERO | M_WAITOK);
8016 
8017 	t4_ulprx_read_la(sc, buf);
8018 	p = buf;
8019 
8020 	sbuf_printf(sb, "      Pcmd        Type   Message"
8021 	    "                Data");
8022 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
8023 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
8024 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
8025 	}
8026 
8027 	rc = sbuf_finish(sb);
8028 	sbuf_delete(sb);
8029 	free(buf, M_CXGBE);
8030 	return (rc);
8031 }
8032 
8033 static int
8034 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
8035 {
8036 	struct adapter *sc = arg1;
8037 	struct sbuf *sb;
8038 	int rc, v;
8039 
8040 	MPASS(chip_id(sc) >= CHELSIO_T5);
8041 
8042 	rc = sysctl_wire_old_buffer(req, 0);
8043 	if (rc != 0)
8044 		return (rc);
8045 
8046 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8047 	if (sb == NULL)
8048 		return (ENOMEM);
8049 
8050 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
8051 	if (G_STATSOURCE_T5(v) == 7) {
8052 		int mode;
8053 
8054 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
8055 		if (mode == 0) {
8056 			sbuf_printf(sb, "total %d, incomplete %d",
8057 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8058 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8059 		} else if (mode == 1) {
8060 			sbuf_printf(sb, "total %d, data overflow %d",
8061 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8062 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8063 		} else {
8064 			sbuf_printf(sb, "unknown mode %d", mode);
8065 		}
8066 	}
8067 	rc = sbuf_finish(sb);
8068 	sbuf_delete(sb);
8069 
8070 	return (rc);
8071 }
8072 
8073 static int
8074 sysctl_tc_params(SYSCTL_HANDLER_ARGS)
8075 {
8076 	struct adapter *sc = arg1;
8077 	struct tx_cl_rl_params tc;
8078 	struct sbuf *sb;
8079 	int i, rc, port_id, mbps, gbps;
8080 
8081 	rc = sysctl_wire_old_buffer(req, 0);
8082 	if (rc != 0)
8083 		return (rc);
8084 
8085 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8086 	if (sb == NULL)
8087 		return (ENOMEM);
8088 
8089 	port_id = arg2 >> 16;
8090 	MPASS(port_id < sc->params.nports);
8091 	MPASS(sc->port[port_id] != NULL);
8092 	i = arg2 & 0xffff;
8093 	MPASS(i < sc->chip_params->nsched_cls);
8094 
8095 	mtx_lock(&sc->tc_lock);
8096 	tc = sc->port[port_id]->sched_params->cl_rl[i];
8097 	mtx_unlock(&sc->tc_lock);
8098 
8099 	if (tc.flags & TX_CLRL_ERROR) {
8100 		sbuf_printf(sb, "error");
8101 		goto done;
8102 	}
8103 
8104 	if (tc.ratemode == SCHED_CLASS_RATEMODE_REL) {
8105 		/* XXX: top speed or actual link speed? */
8106 		gbps = port_top_speed(sc->port[port_id]);
8107 		sbuf_printf(sb, " %u%% of %uGbps", tc.maxrate, gbps);
8108 	} else if (tc.ratemode == SCHED_CLASS_RATEMODE_ABS) {
8109 		switch (tc.rateunit) {
8110 		case SCHED_CLASS_RATEUNIT_BITS:
8111 			mbps = tc.maxrate / 1000;
8112 			gbps = tc.maxrate / 1000000;
8113 			if (tc.maxrate == gbps * 1000000)
8114 				sbuf_printf(sb, " %uGbps", gbps);
8115 			else if (tc.maxrate == mbps * 1000)
8116 				sbuf_printf(sb, " %uMbps", mbps);
8117 			else
8118 				sbuf_printf(sb, " %uKbps", tc.maxrate);
8119 			break;
8120 		case SCHED_CLASS_RATEUNIT_PKTS:
8121 			sbuf_printf(sb, " %upps", tc.maxrate);
8122 			break;
8123 		default:
8124 			rc = ENXIO;
8125 			goto done;
8126 		}
8127 	}
8128 
8129 	switch (tc.mode) {
8130 	case SCHED_CLASS_MODE_CLASS:
8131 		sbuf_printf(sb, " aggregate");
8132 		break;
8133 	case SCHED_CLASS_MODE_FLOW:
8134 		sbuf_printf(sb, " per-flow");
8135 		break;
8136 	default:
8137 		rc = ENXIO;
8138 		goto done;
8139 	}
8140 
8141 done:
8142 	if (rc == 0)
8143 		rc = sbuf_finish(sb);
8144 	sbuf_delete(sb);
8145 
8146 	return (rc);
8147 }
8148 #endif
8149 
8150 #ifdef TCP_OFFLOAD
8151 static void
8152 unit_conv(char *buf, size_t len, u_int val, u_int factor)
8153 {
8154 	u_int rem = val % factor;
8155 
8156 	if (rem == 0)
8157 		snprintf(buf, len, "%u", val / factor);
8158 	else {
8159 		while (rem % 10 == 0)
8160 			rem /= 10;
8161 		snprintf(buf, len, "%u.%u", val / factor, rem);
8162 	}
8163 }
8164 
8165 static int
8166 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
8167 {
8168 	struct adapter *sc = arg1;
8169 	char buf[16];
8170 	u_int res, re;
8171 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8172 
8173 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8174 	switch (arg2) {
8175 	case 0:
8176 		/* timer_tick */
8177 		re = G_TIMERRESOLUTION(res);
8178 		break;
8179 	case 1:
8180 		/* TCP timestamp tick */
8181 		re = G_TIMESTAMPRESOLUTION(res);
8182 		break;
8183 	case 2:
8184 		/* DACK tick */
8185 		re = G_DELAYEDACKRESOLUTION(res);
8186 		break;
8187 	default:
8188 		return (EDOOFUS);
8189 	}
8190 
8191 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
8192 
8193 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
8194 }
8195 
8196 static int
8197 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
8198 {
8199 	struct adapter *sc = arg1;
8200 	u_int res, dack_re, v;
8201 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8202 
8203 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8204 	dack_re = G_DELAYEDACKRESOLUTION(res);
8205 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
8206 
8207 	return (sysctl_handle_int(oidp, &v, 0, req));
8208 }
8209 
8210 static int
8211 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
8212 {
8213 	struct adapter *sc = arg1;
8214 	int reg = arg2;
8215 	u_int tre;
8216 	u_long tp_tick_us, v;
8217 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8218 
8219 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
8220 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
8221 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
8222 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
8223 
8224 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
8225 	tp_tick_us = (cclk_ps << tre) / 1000000;
8226 
8227 	if (reg == A_TP_INIT_SRTT)
8228 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
8229 	else
8230 		v = tp_tick_us * t4_read_reg(sc, reg);
8231 
8232 	return (sysctl_handle_long(oidp, &v, 0, req));
8233 }
8234 
8235 /*
8236  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
8237  * passed to this function.
8238  */
8239 static int
8240 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
8241 {
8242 	struct adapter *sc = arg1;
8243 	int idx = arg2;
8244 	u_int v;
8245 
8246 	MPASS(idx >= 0 && idx <= 24);
8247 
8248 	v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
8249 
8250 	return (sysctl_handle_int(oidp, &v, 0, req));
8251 }
8252 
8253 static int
8254 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
8255 {
8256 	struct adapter *sc = arg1;
8257 	int idx = arg2;
8258 	u_int shift, v, r;
8259 
8260 	MPASS(idx >= 0 && idx < 16);
8261 
8262 	r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
8263 	shift = (idx & 3) << 3;
8264 	v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
8265 
8266 	return (sysctl_handle_int(oidp, &v, 0, req));
8267 }
8268 #endif
8269 
8270 static uint32_t
8271 fconf_iconf_to_mode(uint32_t fconf, uint32_t iconf)
8272 {
8273 	uint32_t mode;
8274 
8275 	mode = T4_FILTER_IPv4 | T4_FILTER_IPv6 | T4_FILTER_IP_SADDR |
8276 	    T4_FILTER_IP_DADDR | T4_FILTER_IP_SPORT | T4_FILTER_IP_DPORT;
8277 
8278 	if (fconf & F_FRAGMENTATION)
8279 		mode |= T4_FILTER_IP_FRAGMENT;
8280 
8281 	if (fconf & F_MPSHITTYPE)
8282 		mode |= T4_FILTER_MPS_HIT_TYPE;
8283 
8284 	if (fconf & F_MACMATCH)
8285 		mode |= T4_FILTER_MAC_IDX;
8286 
8287 	if (fconf & F_ETHERTYPE)
8288 		mode |= T4_FILTER_ETH_TYPE;
8289 
8290 	if (fconf & F_PROTOCOL)
8291 		mode |= T4_FILTER_IP_PROTO;
8292 
8293 	if (fconf & F_TOS)
8294 		mode |= T4_FILTER_IP_TOS;
8295 
8296 	if (fconf & F_VLAN)
8297 		mode |= T4_FILTER_VLAN;
8298 
8299 	if (fconf & F_VNIC_ID) {
8300 		mode |= T4_FILTER_VNIC;
8301 		if (iconf & F_VNIC)
8302 			mode |= T4_FILTER_IC_VNIC;
8303 	}
8304 
8305 	if (fconf & F_PORT)
8306 		mode |= T4_FILTER_PORT;
8307 
8308 	if (fconf & F_FCOE)
8309 		mode |= T4_FILTER_FCoE;
8310 
8311 	return (mode);
8312 }
8313 
8314 static uint32_t
8315 mode_to_fconf(uint32_t mode)
8316 {
8317 	uint32_t fconf = 0;
8318 
8319 	if (mode & T4_FILTER_IP_FRAGMENT)
8320 		fconf |= F_FRAGMENTATION;
8321 
8322 	if (mode & T4_FILTER_MPS_HIT_TYPE)
8323 		fconf |= F_MPSHITTYPE;
8324 
8325 	if (mode & T4_FILTER_MAC_IDX)
8326 		fconf |= F_MACMATCH;
8327 
8328 	if (mode & T4_FILTER_ETH_TYPE)
8329 		fconf |= F_ETHERTYPE;
8330 
8331 	if (mode & T4_FILTER_IP_PROTO)
8332 		fconf |= F_PROTOCOL;
8333 
8334 	if (mode & T4_FILTER_IP_TOS)
8335 		fconf |= F_TOS;
8336 
8337 	if (mode & T4_FILTER_VLAN)
8338 		fconf |= F_VLAN;
8339 
8340 	if (mode & T4_FILTER_VNIC)
8341 		fconf |= F_VNIC_ID;
8342 
8343 	if (mode & T4_FILTER_PORT)
8344 		fconf |= F_PORT;
8345 
8346 	if (mode & T4_FILTER_FCoE)
8347 		fconf |= F_FCOE;
8348 
8349 	return (fconf);
8350 }
8351 
8352 static uint32_t
8353 mode_to_iconf(uint32_t mode)
8354 {
8355 
8356 	if (mode & T4_FILTER_IC_VNIC)
8357 		return (F_VNIC);
8358 	return (0);
8359 }
8360 
8361 static int check_fspec_against_fconf_iconf(struct adapter *sc,
8362     struct t4_filter_specification *fs)
8363 {
8364 	struct tp_params *tpp = &sc->params.tp;
8365 	uint32_t fconf = 0;
8366 
8367 	if (fs->val.frag || fs->mask.frag)
8368 		fconf |= F_FRAGMENTATION;
8369 
8370 	if (fs->val.matchtype || fs->mask.matchtype)
8371 		fconf |= F_MPSHITTYPE;
8372 
8373 	if (fs->val.macidx || fs->mask.macidx)
8374 		fconf |= F_MACMATCH;
8375 
8376 	if (fs->val.ethtype || fs->mask.ethtype)
8377 		fconf |= F_ETHERTYPE;
8378 
8379 	if (fs->val.proto || fs->mask.proto)
8380 		fconf |= F_PROTOCOL;
8381 
8382 	if (fs->val.tos || fs->mask.tos)
8383 		fconf |= F_TOS;
8384 
8385 	if (fs->val.vlan_vld || fs->mask.vlan_vld)
8386 		fconf |= F_VLAN;
8387 
8388 	if (fs->val.ovlan_vld || fs->mask.ovlan_vld) {
8389 		fconf |= F_VNIC_ID;
8390 		if (tpp->ingress_config & F_VNIC)
8391 			return (EINVAL);
8392 	}
8393 
8394 	if (fs->val.pfvf_vld || fs->mask.pfvf_vld) {
8395 		fconf |= F_VNIC_ID;
8396 		if ((tpp->ingress_config & F_VNIC) == 0)
8397 			return (EINVAL);
8398 	}
8399 
8400 	if (fs->val.iport || fs->mask.iport)
8401 		fconf |= F_PORT;
8402 
8403 	if (fs->val.fcoe || fs->mask.fcoe)
8404 		fconf |= F_FCOE;
8405 
8406 	if ((tpp->vlan_pri_map | fconf) != tpp->vlan_pri_map)
8407 		return (E2BIG);
8408 
8409 	return (0);
8410 }
8411 
8412 static int
8413 get_filter_mode(struct adapter *sc, uint32_t *mode)
8414 {
8415 	struct tp_params *tpp = &sc->params.tp;
8416 
8417 	/*
8418 	 * We trust the cached values of the relevant TP registers.  This means
8419 	 * things work reliably only if writes to those registers are always via
8420 	 * t4_set_filter_mode.
8421 	 */
8422 	*mode = fconf_iconf_to_mode(tpp->vlan_pri_map, tpp->ingress_config);
8423 
8424 	return (0);
8425 }
8426 
8427 static int
8428 set_filter_mode(struct adapter *sc, uint32_t mode)
8429 {
8430 	struct tp_params *tpp = &sc->params.tp;
8431 	uint32_t fconf, iconf;
8432 	int rc;
8433 
8434 	iconf = mode_to_iconf(mode);
8435 	if ((iconf ^ tpp->ingress_config) & F_VNIC) {
8436 		/*
8437 		 * For now we just complain if A_TP_INGRESS_CONFIG is not
8438 		 * already set to the correct value for the requested filter
8439 		 * mode.  It's not clear if it's safe to write to this register
8440 		 * on the fly.  (And we trust the cached value of the register).
8441 		 */
8442 		return (EBUSY);
8443 	}
8444 
8445 	fconf = mode_to_fconf(mode);
8446 
8447 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8448 	    "t4setfm");
8449 	if (rc)
8450 		return (rc);
8451 
8452 	if (sc->tids.ftids_in_use > 0) {
8453 		rc = EBUSY;
8454 		goto done;
8455 	}
8456 
8457 #ifdef TCP_OFFLOAD
8458 	if (uld_active(sc, ULD_TOM)) {
8459 		rc = EBUSY;
8460 		goto done;
8461 	}
8462 #endif
8463 
8464 	rc = -t4_set_filter_mode(sc, fconf, true);
8465 done:
8466 	end_synchronized_op(sc, LOCK_HELD);
8467 	return (rc);
8468 }
8469 
8470 static inline uint64_t
8471 get_filter_hits(struct adapter *sc, uint32_t fid)
8472 {
8473 	uint32_t tcb_addr;
8474 
8475 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE) +
8476 	    (fid + sc->tids.ftid_base) * TCB_SIZE;
8477 
8478 	if (is_t4(sc)) {
8479 		uint64_t hits;
8480 
8481 		read_via_memwin(sc, 0, tcb_addr + 16, (uint32_t *)&hits, 8);
8482 		return (be64toh(hits));
8483 	} else {
8484 		uint32_t hits;
8485 
8486 		read_via_memwin(sc, 0, tcb_addr + 24, &hits, 4);
8487 		return (be32toh(hits));
8488 	}
8489 }
8490 
8491 static int
8492 get_filter(struct adapter *sc, struct t4_filter *t)
8493 {
8494 	int i, rc, nfilters = sc->tids.nftids;
8495 	struct filter_entry *f;
8496 
8497 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8498 	    "t4getf");
8499 	if (rc)
8500 		return (rc);
8501 
8502 	if (sc->tids.ftids_in_use == 0 || sc->tids.ftid_tab == NULL ||
8503 	    t->idx >= nfilters) {
8504 		t->idx = 0xffffffff;
8505 		goto done;
8506 	}
8507 
8508 	f = &sc->tids.ftid_tab[t->idx];
8509 	for (i = t->idx; i < nfilters; i++, f++) {
8510 		if (f->valid) {
8511 			t->idx = i;
8512 			t->l2tidx = f->l2t ? f->l2t->idx : 0;
8513 			t->smtidx = f->smtidx;
8514 			if (f->fs.hitcnts)
8515 				t->hits = get_filter_hits(sc, t->idx);
8516 			else
8517 				t->hits = UINT64_MAX;
8518 			t->fs = f->fs;
8519 
8520 			goto done;
8521 		}
8522 	}
8523 
8524 	t->idx = 0xffffffff;
8525 done:
8526 	end_synchronized_op(sc, LOCK_HELD);
8527 	return (0);
8528 }
8529 
8530 static int
8531 set_filter(struct adapter *sc, struct t4_filter *t)
8532 {
8533 	unsigned int nfilters, nports;
8534 	struct filter_entry *f;
8535 	int i, rc;
8536 
8537 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setf");
8538 	if (rc)
8539 		return (rc);
8540 
8541 	nfilters = sc->tids.nftids;
8542 	nports = sc->params.nports;
8543 
8544 	if (nfilters == 0) {
8545 		rc = ENOTSUP;
8546 		goto done;
8547 	}
8548 
8549 	if (t->idx >= nfilters) {
8550 		rc = EINVAL;
8551 		goto done;
8552 	}
8553 
8554 	/* Validate against the global filter mode and ingress config */
8555 	rc = check_fspec_against_fconf_iconf(sc, &t->fs);
8556 	if (rc != 0)
8557 		goto done;
8558 
8559 	if (t->fs.action == FILTER_SWITCH && t->fs.eport >= nports) {
8560 		rc = EINVAL;
8561 		goto done;
8562 	}
8563 
8564 	if (t->fs.val.iport >= nports) {
8565 		rc = EINVAL;
8566 		goto done;
8567 	}
8568 
8569 	/* Can't specify an iq if not steering to it */
8570 	if (!t->fs.dirsteer && t->fs.iq) {
8571 		rc = EINVAL;
8572 		goto done;
8573 	}
8574 
8575 	/* IPv6 filter idx must be 4 aligned */
8576 	if (t->fs.type == 1 &&
8577 	    ((t->idx & 0x3) || t->idx + 4 >= nfilters)) {
8578 		rc = EINVAL;
8579 		goto done;
8580 	}
8581 
8582 	if (!(sc->flags & FULL_INIT_DONE) &&
8583 	    ((rc = adapter_full_init(sc)) != 0))
8584 		goto done;
8585 
8586 	if (sc->tids.ftid_tab == NULL) {
8587 		KASSERT(sc->tids.ftids_in_use == 0,
8588 		    ("%s: no memory allocated but filters_in_use > 0",
8589 		    __func__));
8590 
8591 		sc->tids.ftid_tab = malloc(sizeof (struct filter_entry) *
8592 		    nfilters, M_CXGBE, M_NOWAIT | M_ZERO);
8593 		if (sc->tids.ftid_tab == NULL) {
8594 			rc = ENOMEM;
8595 			goto done;
8596 		}
8597 		mtx_init(&sc->tids.ftid_lock, "T4 filters", 0, MTX_DEF);
8598 	}
8599 
8600 	for (i = 0; i < 4; i++) {
8601 		f = &sc->tids.ftid_tab[t->idx + i];
8602 
8603 		if (f->pending || f->valid) {
8604 			rc = EBUSY;
8605 			goto done;
8606 		}
8607 		if (f->locked) {
8608 			rc = EPERM;
8609 			goto done;
8610 		}
8611 
8612 		if (t->fs.type == 0)
8613 			break;
8614 	}
8615 
8616 	f = &sc->tids.ftid_tab[t->idx];
8617 	f->fs = t->fs;
8618 
8619 	rc = set_filter_wr(sc, t->idx);
8620 done:
8621 	end_synchronized_op(sc, 0);
8622 
8623 	if (rc == 0) {
8624 		mtx_lock(&sc->tids.ftid_lock);
8625 		for (;;) {
8626 			if (f->pending == 0) {
8627 				rc = f->valid ? 0 : EIO;
8628 				break;
8629 			}
8630 
8631 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8632 			    PCATCH, "t4setfw", 0)) {
8633 				rc = EINPROGRESS;
8634 				break;
8635 			}
8636 		}
8637 		mtx_unlock(&sc->tids.ftid_lock);
8638 	}
8639 	return (rc);
8640 }
8641 
8642 static int
8643 del_filter(struct adapter *sc, struct t4_filter *t)
8644 {
8645 	unsigned int nfilters;
8646 	struct filter_entry *f;
8647 	int rc;
8648 
8649 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4delf");
8650 	if (rc)
8651 		return (rc);
8652 
8653 	nfilters = sc->tids.nftids;
8654 
8655 	if (nfilters == 0) {
8656 		rc = ENOTSUP;
8657 		goto done;
8658 	}
8659 
8660 	if (sc->tids.ftid_tab == NULL || sc->tids.ftids_in_use == 0 ||
8661 	    t->idx >= nfilters) {
8662 		rc = EINVAL;
8663 		goto done;
8664 	}
8665 
8666 	if (!(sc->flags & FULL_INIT_DONE)) {
8667 		rc = EAGAIN;
8668 		goto done;
8669 	}
8670 
8671 	f = &sc->tids.ftid_tab[t->idx];
8672 
8673 	if (f->pending) {
8674 		rc = EBUSY;
8675 		goto done;
8676 	}
8677 	if (f->locked) {
8678 		rc = EPERM;
8679 		goto done;
8680 	}
8681 
8682 	if (f->valid) {
8683 		t->fs = f->fs;	/* extra info for the caller */
8684 		rc = del_filter_wr(sc, t->idx);
8685 	}
8686 
8687 done:
8688 	end_synchronized_op(sc, 0);
8689 
8690 	if (rc == 0) {
8691 		mtx_lock(&sc->tids.ftid_lock);
8692 		for (;;) {
8693 			if (f->pending == 0) {
8694 				rc = f->valid ? EIO : 0;
8695 				break;
8696 			}
8697 
8698 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8699 			    PCATCH, "t4delfw", 0)) {
8700 				rc = EINPROGRESS;
8701 				break;
8702 			}
8703 		}
8704 		mtx_unlock(&sc->tids.ftid_lock);
8705 	}
8706 
8707 	return (rc);
8708 }
8709 
8710 static void
8711 clear_filter(struct filter_entry *f)
8712 {
8713 	if (f->l2t)
8714 		t4_l2t_release(f->l2t);
8715 
8716 	bzero(f, sizeof (*f));
8717 }
8718 
8719 static int
8720 set_filter_wr(struct adapter *sc, int fidx)
8721 {
8722 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8723 	struct fw_filter_wr *fwr;
8724 	unsigned int ftid, vnic_vld, vnic_vld_mask;
8725 	struct wrq_cookie cookie;
8726 
8727 	ASSERT_SYNCHRONIZED_OP(sc);
8728 
8729 	if (f->fs.newdmac || f->fs.newvlan) {
8730 		/* This filter needs an L2T entry; allocate one. */
8731 		f->l2t = t4_l2t_alloc_switching(sc->l2t);
8732 		if (f->l2t == NULL)
8733 			return (EAGAIN);
8734 		if (t4_l2t_set_switching(sc, f->l2t, f->fs.vlan, f->fs.eport,
8735 		    f->fs.dmac)) {
8736 			t4_l2t_release(f->l2t);
8737 			f->l2t = NULL;
8738 			return (ENOMEM);
8739 		}
8740 	}
8741 
8742 	/* Already validated against fconf, iconf */
8743 	MPASS((f->fs.val.pfvf_vld & f->fs.val.ovlan_vld) == 0);
8744 	MPASS((f->fs.mask.pfvf_vld & f->fs.mask.ovlan_vld) == 0);
8745 	if (f->fs.val.pfvf_vld || f->fs.val.ovlan_vld)
8746 		vnic_vld = 1;
8747 	else
8748 		vnic_vld = 0;
8749 	if (f->fs.mask.pfvf_vld || f->fs.mask.ovlan_vld)
8750 		vnic_vld_mask = 1;
8751 	else
8752 		vnic_vld_mask = 0;
8753 
8754 	ftid = sc->tids.ftid_base + fidx;
8755 
8756 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8757 	if (fwr == NULL)
8758 		return (ENOMEM);
8759 	bzero(fwr, sizeof(*fwr));
8760 
8761 	fwr->op_pkd = htobe32(V_FW_WR_OP(FW_FILTER_WR));
8762 	fwr->len16_pkd = htobe32(FW_LEN16(*fwr));
8763 	fwr->tid_to_iq =
8764 	    htobe32(V_FW_FILTER_WR_TID(ftid) |
8765 		V_FW_FILTER_WR_RQTYPE(f->fs.type) |
8766 		V_FW_FILTER_WR_NOREPLY(0) |
8767 		V_FW_FILTER_WR_IQ(f->fs.iq));
8768 	fwr->del_filter_to_l2tix =
8769 	    htobe32(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
8770 		V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
8771 		V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
8772 		V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
8773 		V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
8774 		V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
8775 		V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
8776 		V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
8777 		V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
8778 		    f->fs.newvlan == VLAN_REWRITE) |
8779 		V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
8780 		    f->fs.newvlan == VLAN_REWRITE) |
8781 		V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
8782 		V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
8783 		V_FW_FILTER_WR_PRIO(f->fs.prio) |
8784 		V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
8785 	fwr->ethtype = htobe16(f->fs.val.ethtype);
8786 	fwr->ethtypem = htobe16(f->fs.mask.ethtype);
8787 	fwr->frag_to_ovlan_vldm =
8788 	    (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
8789 		V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
8790 		V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.vlan_vld) |
8791 		V_FW_FILTER_WR_OVLAN_VLD(vnic_vld) |
8792 		V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.vlan_vld) |
8793 		V_FW_FILTER_WR_OVLAN_VLDM(vnic_vld_mask));
8794 	fwr->smac_sel = 0;
8795 	fwr->rx_chan_rx_rpl_iq = htobe16(V_FW_FILTER_WR_RX_CHAN(0) |
8796 	    V_FW_FILTER_WR_RX_RPL_IQ(sc->sge.fwq.abs_id));
8797 	fwr->maci_to_matchtypem =
8798 	    htobe32(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
8799 		V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
8800 		V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
8801 		V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
8802 		V_FW_FILTER_WR_PORT(f->fs.val.iport) |
8803 		V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
8804 		V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
8805 		V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
8806 	fwr->ptcl = f->fs.val.proto;
8807 	fwr->ptclm = f->fs.mask.proto;
8808 	fwr->ttyp = f->fs.val.tos;
8809 	fwr->ttypm = f->fs.mask.tos;
8810 	fwr->ivlan = htobe16(f->fs.val.vlan);
8811 	fwr->ivlanm = htobe16(f->fs.mask.vlan);
8812 	fwr->ovlan = htobe16(f->fs.val.vnic);
8813 	fwr->ovlanm = htobe16(f->fs.mask.vnic);
8814 	bcopy(f->fs.val.dip, fwr->lip, sizeof (fwr->lip));
8815 	bcopy(f->fs.mask.dip, fwr->lipm, sizeof (fwr->lipm));
8816 	bcopy(f->fs.val.sip, fwr->fip, sizeof (fwr->fip));
8817 	bcopy(f->fs.mask.sip, fwr->fipm, sizeof (fwr->fipm));
8818 	fwr->lp = htobe16(f->fs.val.dport);
8819 	fwr->lpm = htobe16(f->fs.mask.dport);
8820 	fwr->fp = htobe16(f->fs.val.sport);
8821 	fwr->fpm = htobe16(f->fs.mask.sport);
8822 	if (f->fs.newsmac)
8823 		bcopy(f->fs.smac, fwr->sma, sizeof (fwr->sma));
8824 
8825 	f->pending = 1;
8826 	sc->tids.ftids_in_use++;
8827 
8828 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8829 	return (0);
8830 }
8831 
8832 static int
8833 del_filter_wr(struct adapter *sc, int fidx)
8834 {
8835 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8836 	struct fw_filter_wr *fwr;
8837 	unsigned int ftid;
8838 	struct wrq_cookie cookie;
8839 
8840 	ftid = sc->tids.ftid_base + fidx;
8841 
8842 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8843 	if (fwr == NULL)
8844 		return (ENOMEM);
8845 	bzero(fwr, sizeof (*fwr));
8846 
8847 	t4_mk_filtdelwr(ftid, fwr, sc->sge.fwq.abs_id);
8848 
8849 	f->pending = 1;
8850 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8851 	return (0);
8852 }
8853 
8854 int
8855 t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8856 {
8857 	struct adapter *sc = iq->adapter;
8858 	const struct cpl_set_tcb_rpl *rpl = (const void *)(rss + 1);
8859 	unsigned int idx = GET_TID(rpl);
8860 	unsigned int rc;
8861 	struct filter_entry *f;
8862 
8863 	KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
8864 	    rss->opcode));
8865 	MPASS(iq == &sc->sge.fwq);
8866 	MPASS(is_ftid(sc, idx));
8867 
8868 	idx -= sc->tids.ftid_base;
8869 	f = &sc->tids.ftid_tab[idx];
8870 	rc = G_COOKIE(rpl->cookie);
8871 
8872 	mtx_lock(&sc->tids.ftid_lock);
8873 	if (rc == FW_FILTER_WR_FLT_ADDED) {
8874 		KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
8875 		    __func__, idx));
8876 		f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
8877 		f->pending = 0;  /* asynchronous setup completed */
8878 		f->valid = 1;
8879 	} else {
8880 		if (rc != FW_FILTER_WR_FLT_DELETED) {
8881 			/* Add or delete failed, display an error */
8882 			log(LOG_ERR,
8883 			    "filter %u setup failed with error %u\n",
8884 			    idx, rc);
8885 		}
8886 
8887 		clear_filter(f);
8888 		sc->tids.ftids_in_use--;
8889 	}
8890 	wakeup(&sc->tids.ftid_tab);
8891 	mtx_unlock(&sc->tids.ftid_lock);
8892 
8893 	return (0);
8894 }
8895 
8896 static int
8897 set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8898 {
8899 
8900 	MPASS(iq->set_tcb_rpl != NULL);
8901 	return (iq->set_tcb_rpl(iq, rss, m));
8902 }
8903 
8904 static int
8905 l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8906 {
8907 
8908 	MPASS(iq->l2t_write_rpl != NULL);
8909 	return (iq->l2t_write_rpl(iq, rss, m));
8910 }
8911 
8912 static int
8913 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
8914 {
8915 	int rc;
8916 
8917 	if (cntxt->cid > M_CTXTQID)
8918 		return (EINVAL);
8919 
8920 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
8921 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
8922 		return (EINVAL);
8923 
8924 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
8925 	if (rc)
8926 		return (rc);
8927 
8928 	if (sc->flags & FW_OK) {
8929 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
8930 		    &cntxt->data[0]);
8931 		if (rc == 0)
8932 			goto done;
8933 	}
8934 
8935 	/*
8936 	 * Read via firmware failed or wasn't even attempted.  Read directly via
8937 	 * the backdoor.
8938 	 */
8939 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
8940 done:
8941 	end_synchronized_op(sc, 0);
8942 	return (rc);
8943 }
8944 
8945 static int
8946 load_fw(struct adapter *sc, struct t4_data *fw)
8947 {
8948 	int rc;
8949 	uint8_t *fw_data;
8950 
8951 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
8952 	if (rc)
8953 		return (rc);
8954 
8955 	if (sc->flags & FULL_INIT_DONE) {
8956 		rc = EBUSY;
8957 		goto done;
8958 	}
8959 
8960 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
8961 	if (fw_data == NULL) {
8962 		rc = ENOMEM;
8963 		goto done;
8964 	}
8965 
8966 	rc = copyin(fw->data, fw_data, fw->len);
8967 	if (rc == 0)
8968 		rc = -t4_load_fw(sc, fw_data, fw->len);
8969 
8970 	free(fw_data, M_CXGBE);
8971 done:
8972 	end_synchronized_op(sc, 0);
8973 	return (rc);
8974 }
8975 
8976 static int
8977 load_cfg(struct adapter *sc, struct t4_data *cfg)
8978 {
8979 	int rc;
8980 	uint8_t *cfg_data = NULL;
8981 
8982 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
8983 	if (rc)
8984 		return (rc);
8985 
8986 	if (cfg->len == 0) {
8987 		/* clear */
8988 		rc = -t4_load_cfg(sc, NULL, 0);
8989 		goto done;
8990 	}
8991 
8992 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
8993 	if (cfg_data == NULL) {
8994 		rc = ENOMEM;
8995 		goto done;
8996 	}
8997 
8998 	rc = copyin(cfg->data, cfg_data, cfg->len);
8999 	if (rc == 0)
9000 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
9001 
9002 	free(cfg_data, M_CXGBE);
9003 done:
9004 	end_synchronized_op(sc, 0);
9005 	return (rc);
9006 }
9007 
9008 static int
9009 load_boot(struct adapter *sc, struct t4_bootrom *br)
9010 {
9011 	int rc;
9012 	uint8_t *br_data = NULL;
9013 	u_int offset;
9014 
9015 	if (br->len > 1024 * 1024)
9016 		return (EFBIG);
9017 
9018 	if (br->pf_offset == 0) {
9019 		/* pfidx */
9020 		if (br->pfidx_addr > 7)
9021 			return (EINVAL);
9022 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
9023 		    A_PCIE_PF_EXPROM_OFST)));
9024 	} else if (br->pf_offset == 1) {
9025 		/* offset */
9026 		offset = G_OFFSET(br->pfidx_addr);
9027 	} else {
9028 		return (EINVAL);
9029 	}
9030 
9031 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
9032 	if (rc)
9033 		return (rc);
9034 
9035 	if (br->len == 0) {
9036 		/* clear */
9037 		rc = -t4_load_boot(sc, NULL, offset, 0);
9038 		goto done;
9039 	}
9040 
9041 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
9042 	if (br_data == NULL) {
9043 		rc = ENOMEM;
9044 		goto done;
9045 	}
9046 
9047 	rc = copyin(br->data, br_data, br->len);
9048 	if (rc == 0)
9049 		rc = -t4_load_boot(sc, br_data, offset, br->len);
9050 
9051 	free(br_data, M_CXGBE);
9052 done:
9053 	end_synchronized_op(sc, 0);
9054 	return (rc);
9055 }
9056 
9057 static int
9058 load_bootcfg(struct adapter *sc, struct t4_data *bc)
9059 {
9060 	int rc;
9061 	uint8_t *bc_data = NULL;
9062 
9063 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9064 	if (rc)
9065 		return (rc);
9066 
9067 	if (bc->len == 0) {
9068 		/* clear */
9069 		rc = -t4_load_bootcfg(sc, NULL, 0);
9070 		goto done;
9071 	}
9072 
9073 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
9074 	if (bc_data == NULL) {
9075 		rc = ENOMEM;
9076 		goto done;
9077 	}
9078 
9079 	rc = copyin(bc->data, bc_data, bc->len);
9080 	if (rc == 0)
9081 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
9082 
9083 	free(bc_data, M_CXGBE);
9084 done:
9085 	end_synchronized_op(sc, 0);
9086 	return (rc);
9087 }
9088 
9089 static int
9090 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
9091 {
9092 	int rc;
9093 	struct cudbg_init *cudbg;
9094 	void *handle, *buf;
9095 
9096 	/* buf is large, don't block if no memory is available */
9097 	buf = malloc(dump->len, M_CXGBE, M_NOWAIT);
9098 	if (buf == NULL)
9099 		return (ENOMEM);
9100 
9101 	handle = cudbg_alloc_handle();
9102 	if (handle == NULL) {
9103 		rc = ENOMEM;
9104 		goto done;
9105 	}
9106 
9107 	cudbg = cudbg_get_init(handle);
9108 	cudbg->adap = sc;
9109 	cudbg->print = (cudbg_print_cb)printf;
9110 
9111 #ifndef notyet
9112 	device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
9113 	    __func__, dump->wr_flash, dump->len, dump->data);
9114 #endif
9115 
9116 	if (dump->wr_flash)
9117 		cudbg->use_flash = 1;
9118 	MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
9119 	memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
9120 
9121 	rc = cudbg_collect(handle, buf, &dump->len);
9122 	if (rc != 0)
9123 		goto done;
9124 
9125 	rc = copyout(buf, dump->data, dump->len);
9126 done:
9127 	cudbg_free_handle(handle);
9128 	free(buf, M_CXGBE);
9129 	return (rc);
9130 }
9131 
9132 #define MAX_READ_BUF_SIZE (128 * 1024)
9133 static int
9134 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
9135 {
9136 	uint32_t addr, remaining, n;
9137 	uint32_t *buf;
9138 	int rc;
9139 	uint8_t *dst;
9140 
9141 	rc = validate_mem_range(sc, mr->addr, mr->len);
9142 	if (rc != 0)
9143 		return (rc);
9144 
9145 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
9146 	addr = mr->addr;
9147 	remaining = mr->len;
9148 	dst = (void *)mr->data;
9149 
9150 	while (remaining) {
9151 		n = min(remaining, MAX_READ_BUF_SIZE);
9152 		read_via_memwin(sc, 2, addr, buf, n);
9153 
9154 		rc = copyout(buf, dst, n);
9155 		if (rc != 0)
9156 			break;
9157 
9158 		dst += n;
9159 		remaining -= n;
9160 		addr += n;
9161 	}
9162 
9163 	free(buf, M_CXGBE);
9164 	return (rc);
9165 }
9166 #undef MAX_READ_BUF_SIZE
9167 
9168 static int
9169 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
9170 {
9171 	int rc;
9172 
9173 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
9174 		return (EINVAL);
9175 
9176 	if (i2cd->len > sizeof(i2cd->data))
9177 		return (EFBIG);
9178 
9179 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9180 	if (rc)
9181 		return (rc);
9182 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9183 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
9184 	end_synchronized_op(sc, 0);
9185 
9186 	return (rc);
9187 }
9188 
9189 int
9190 t4_os_find_pci_capability(struct adapter *sc, int cap)
9191 {
9192 	int i;
9193 
9194 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9195 }
9196 
9197 int
9198 t4_os_pci_save_state(struct adapter *sc)
9199 {
9200 	device_t dev;
9201 	struct pci_devinfo *dinfo;
9202 
9203 	dev = sc->dev;
9204 	dinfo = device_get_ivars(dev);
9205 
9206 	pci_cfg_save(dev, dinfo, 0);
9207 	return (0);
9208 }
9209 
9210 int
9211 t4_os_pci_restore_state(struct adapter *sc)
9212 {
9213 	device_t dev;
9214 	struct pci_devinfo *dinfo;
9215 
9216 	dev = sc->dev;
9217 	dinfo = device_get_ivars(dev);
9218 
9219 	pci_cfg_restore(dev, dinfo);
9220 	return (0);
9221 }
9222 
9223 void
9224 t4_os_portmod_changed(struct port_info *pi, int old_ptype, int old_mtype,
9225     struct link_config *old_lc)
9226 {
9227 	struct vi_info *vi;
9228 	struct ifnet *ifp;
9229 	int v;
9230 	static const char *mod_str[] = {
9231 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9232 	};
9233 
9234 	PORT_LOCK(pi);
9235 	for_each_vi(pi, v, vi) {
9236 		build_medialist(pi, &vi->media);
9237 	}
9238 	PORT_UNLOCK(pi);
9239 	vi = &pi->vi[0];
9240 	if (begin_synchronized_op(pi->adapter, vi, HOLD_LOCK, "t4mod") == 0) {
9241 		init_l1cfg(pi);
9242 		end_synchronized_op(pi->adapter, LOCK_HELD);
9243 	}
9244 
9245 	ifp = vi->ifp;
9246 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9247 		if_printf(ifp, "transceiver unplugged.\n");
9248 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9249 		if_printf(ifp, "unknown transceiver inserted.\n");
9250 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9251 		if_printf(ifp, "unsupported transceiver inserted.\n");
9252 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9253 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9254 		    port_top_speed(pi), mod_str[pi->mod_type]);
9255 	} else {
9256 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9257 		    pi->mod_type);
9258 	}
9259 }
9260 
9261 void
9262 t4_os_link_changed(struct port_info *pi, struct link_config *old_lc)
9263 {
9264 	struct vi_info *vi;
9265 	struct ifnet *ifp;
9266 	struct link_config *lc;
9267 	int v;
9268 
9269 	for_each_vi(pi, v, vi) {
9270 		ifp = vi->ifp;
9271 		if (ifp == NULL)
9272 			continue;
9273 
9274 		lc = &pi->link_cfg;
9275 		if (lc->link_ok) {
9276 			ifp->if_baudrate = IF_Mbps(lc->speed);
9277 			if_link_state_change(ifp, LINK_STATE_UP);
9278 		} else {
9279 			if_link_state_change(ifp, LINK_STATE_DOWN);
9280 		}
9281 	}
9282 }
9283 
9284 void
9285 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9286 {
9287 	struct adapter *sc;
9288 
9289 	sx_slock(&t4_list_lock);
9290 	SLIST_FOREACH(sc, &t4_list, link) {
9291 		/*
9292 		 * func should not make any assumptions about what state sc is
9293 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9294 		 */
9295 		func(sc, arg);
9296 	}
9297 	sx_sunlock(&t4_list_lock);
9298 }
9299 
9300 static int
9301 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9302     struct thread *td)
9303 {
9304 	int rc;
9305 	struct adapter *sc = dev->si_drv1;
9306 
9307 	rc = priv_check(td, PRIV_DRIVER);
9308 	if (rc != 0)
9309 		return (rc);
9310 
9311 	switch (cmd) {
9312 	case CHELSIO_T4_GETREG: {
9313 		struct t4_reg *edata = (struct t4_reg *)data;
9314 
9315 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9316 			return (EFAULT);
9317 
9318 		if (edata->size == 4)
9319 			edata->val = t4_read_reg(sc, edata->addr);
9320 		else if (edata->size == 8)
9321 			edata->val = t4_read_reg64(sc, edata->addr);
9322 		else
9323 			return (EINVAL);
9324 
9325 		break;
9326 	}
9327 	case CHELSIO_T4_SETREG: {
9328 		struct t4_reg *edata = (struct t4_reg *)data;
9329 
9330 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9331 			return (EFAULT);
9332 
9333 		if (edata->size == 4) {
9334 			if (edata->val & 0xffffffff00000000)
9335 				return (EINVAL);
9336 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9337 		} else if (edata->size == 8)
9338 			t4_write_reg64(sc, edata->addr, edata->val);
9339 		else
9340 			return (EINVAL);
9341 		break;
9342 	}
9343 	case CHELSIO_T4_REGDUMP: {
9344 		struct t4_regdump *regs = (struct t4_regdump *)data;
9345 		int reglen = t4_get_regs_len(sc);
9346 		uint8_t *buf;
9347 
9348 		if (regs->len < reglen) {
9349 			regs->len = reglen; /* hint to the caller */
9350 			return (ENOBUFS);
9351 		}
9352 
9353 		regs->len = reglen;
9354 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9355 		get_regs(sc, regs, buf);
9356 		rc = copyout(buf, regs->data, reglen);
9357 		free(buf, M_CXGBE);
9358 		break;
9359 	}
9360 	case CHELSIO_T4_GET_FILTER_MODE:
9361 		rc = get_filter_mode(sc, (uint32_t *)data);
9362 		break;
9363 	case CHELSIO_T4_SET_FILTER_MODE:
9364 		rc = set_filter_mode(sc, *(uint32_t *)data);
9365 		break;
9366 	case CHELSIO_T4_GET_FILTER:
9367 		rc = get_filter(sc, (struct t4_filter *)data);
9368 		break;
9369 	case CHELSIO_T4_SET_FILTER:
9370 		rc = set_filter(sc, (struct t4_filter *)data);
9371 		break;
9372 	case CHELSIO_T4_DEL_FILTER:
9373 		rc = del_filter(sc, (struct t4_filter *)data);
9374 		break;
9375 	case CHELSIO_T4_GET_SGE_CONTEXT:
9376 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9377 		break;
9378 	case CHELSIO_T4_LOAD_FW:
9379 		rc = load_fw(sc, (struct t4_data *)data);
9380 		break;
9381 	case CHELSIO_T4_GET_MEM:
9382 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9383 		break;
9384 	case CHELSIO_T4_GET_I2C:
9385 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9386 		break;
9387 	case CHELSIO_T4_CLEAR_STATS: {
9388 		int i, v;
9389 		u_int port_id = *(uint32_t *)data;
9390 		struct port_info *pi;
9391 		struct vi_info *vi;
9392 
9393 		if (port_id >= sc->params.nports)
9394 			return (EINVAL);
9395 		pi = sc->port[port_id];
9396 		if (pi == NULL)
9397 			return (EIO);
9398 
9399 		/* MAC stats */
9400 		t4_clr_port_stats(sc, pi->tx_chan);
9401 		pi->tx_parse_error = 0;
9402 		mtx_lock(&sc->reg_lock);
9403 		for_each_vi(pi, v, vi) {
9404 			if (vi->flags & VI_INIT_DONE)
9405 				t4_clr_vi_stats(sc, vi->viid);
9406 		}
9407 		mtx_unlock(&sc->reg_lock);
9408 
9409 		/*
9410 		 * Since this command accepts a port, clear stats for
9411 		 * all VIs on this port.
9412 		 */
9413 		for_each_vi(pi, v, vi) {
9414 			if (vi->flags & VI_INIT_DONE) {
9415 				struct sge_rxq *rxq;
9416 				struct sge_txq *txq;
9417 				struct sge_wrq *wrq;
9418 
9419 				for_each_rxq(vi, i, rxq) {
9420 #if defined(INET) || defined(INET6)
9421 					rxq->lro.lro_queued = 0;
9422 					rxq->lro.lro_flushed = 0;
9423 #endif
9424 					rxq->rxcsum = 0;
9425 					rxq->vlan_extraction = 0;
9426 				}
9427 
9428 				for_each_txq(vi, i, txq) {
9429 					txq->txcsum = 0;
9430 					txq->tso_wrs = 0;
9431 					txq->vlan_insertion = 0;
9432 					txq->imm_wrs = 0;
9433 					txq->sgl_wrs = 0;
9434 					txq->txpkt_wrs = 0;
9435 					txq->txpkts0_wrs = 0;
9436 					txq->txpkts1_wrs = 0;
9437 					txq->txpkts0_pkts = 0;
9438 					txq->txpkts1_pkts = 0;
9439 					mp_ring_reset_stats(txq->r);
9440 				}
9441 
9442 #ifdef TCP_OFFLOAD
9443 				/* nothing to clear for each ofld_rxq */
9444 
9445 				for_each_ofld_txq(vi, i, wrq) {
9446 					wrq->tx_wrs_direct = 0;
9447 					wrq->tx_wrs_copied = 0;
9448 				}
9449 #endif
9450 
9451 				if (IS_MAIN_VI(vi)) {
9452 					wrq = &sc->sge.ctrlq[pi->port_id];
9453 					wrq->tx_wrs_direct = 0;
9454 					wrq->tx_wrs_copied = 0;
9455 				}
9456 			}
9457 		}
9458 		break;
9459 	}
9460 	case CHELSIO_T4_SCHED_CLASS:
9461 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9462 		break;
9463 	case CHELSIO_T4_SCHED_QUEUE:
9464 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9465 		break;
9466 	case CHELSIO_T4_GET_TRACER:
9467 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9468 		break;
9469 	case CHELSIO_T4_SET_TRACER:
9470 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9471 		break;
9472 	case CHELSIO_T4_LOAD_CFG:
9473 		rc = load_cfg(sc, (struct t4_data *)data);
9474 		break;
9475 	case CHELSIO_T4_LOAD_BOOT:
9476 		rc = load_boot(sc, (struct t4_bootrom *)data);
9477 		break;
9478 	case CHELSIO_T4_LOAD_BOOTCFG:
9479 		rc = load_bootcfg(sc, (struct t4_data *)data);
9480 		break;
9481 	case CHELSIO_T4_CUDBG_DUMP:
9482 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
9483 		break;
9484 	default:
9485 		rc = ENOTTY;
9486 	}
9487 
9488 	return (rc);
9489 }
9490 
9491 void
9492 t4_db_full(struct adapter *sc)
9493 {
9494 
9495 	CXGBE_UNIMPLEMENTED(__func__);
9496 }
9497 
9498 void
9499 t4_db_dropped(struct adapter *sc)
9500 {
9501 
9502 	CXGBE_UNIMPLEMENTED(__func__);
9503 }
9504 
9505 #ifdef TCP_OFFLOAD
9506 static int
9507 toe_capability(struct vi_info *vi, int enable)
9508 {
9509 	int rc;
9510 	struct port_info *pi = vi->pi;
9511 	struct adapter *sc = pi->adapter;
9512 
9513 	ASSERT_SYNCHRONIZED_OP(sc);
9514 
9515 	if (!is_offload(sc))
9516 		return (ENODEV);
9517 
9518 	if (enable) {
9519 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9520 			/* TOE is already enabled. */
9521 			return (0);
9522 		}
9523 
9524 		/*
9525 		 * We need the port's queues around so that we're able to send
9526 		 * and receive CPLs to/from the TOE even if the ifnet for this
9527 		 * port has never been UP'd administratively.
9528 		 */
9529 		if (!(vi->flags & VI_INIT_DONE)) {
9530 			rc = vi_full_init(vi);
9531 			if (rc)
9532 				return (rc);
9533 		}
9534 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9535 			rc = vi_full_init(&pi->vi[0]);
9536 			if (rc)
9537 				return (rc);
9538 		}
9539 
9540 		if (isset(&sc->offload_map, pi->port_id)) {
9541 			/* TOE is enabled on another VI of this port. */
9542 			pi->uld_vis++;
9543 			return (0);
9544 		}
9545 
9546 		if (!uld_active(sc, ULD_TOM)) {
9547 			rc = t4_activate_uld(sc, ULD_TOM);
9548 			if (rc == EAGAIN) {
9549 				log(LOG_WARNING,
9550 				    "You must kldload t4_tom.ko before trying "
9551 				    "to enable TOE on a cxgbe interface.\n");
9552 			}
9553 			if (rc != 0)
9554 				return (rc);
9555 			KASSERT(sc->tom_softc != NULL,
9556 			    ("%s: TOM activated but softc NULL", __func__));
9557 			KASSERT(uld_active(sc, ULD_TOM),
9558 			    ("%s: TOM activated but flag not set", __func__));
9559 		}
9560 
9561 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
9562 		if (!uld_active(sc, ULD_IWARP))
9563 			(void) t4_activate_uld(sc, ULD_IWARP);
9564 		if (!uld_active(sc, ULD_ISCSI))
9565 			(void) t4_activate_uld(sc, ULD_ISCSI);
9566 
9567 		pi->uld_vis++;
9568 		setbit(&sc->offload_map, pi->port_id);
9569 	} else {
9570 		pi->uld_vis--;
9571 
9572 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9573 			return (0);
9574 
9575 		KASSERT(uld_active(sc, ULD_TOM),
9576 		    ("%s: TOM never initialized?", __func__));
9577 		clrbit(&sc->offload_map, pi->port_id);
9578 	}
9579 
9580 	return (0);
9581 }
9582 
9583 /*
9584  * Add an upper layer driver to the global list.
9585  */
9586 int
9587 t4_register_uld(struct uld_info *ui)
9588 {
9589 	int rc = 0;
9590 	struct uld_info *u;
9591 
9592 	sx_xlock(&t4_uld_list_lock);
9593 	SLIST_FOREACH(u, &t4_uld_list, link) {
9594 	    if (u->uld_id == ui->uld_id) {
9595 		    rc = EEXIST;
9596 		    goto done;
9597 	    }
9598 	}
9599 
9600 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9601 	ui->refcount = 0;
9602 done:
9603 	sx_xunlock(&t4_uld_list_lock);
9604 	return (rc);
9605 }
9606 
9607 int
9608 t4_unregister_uld(struct uld_info *ui)
9609 {
9610 	int rc = EINVAL;
9611 	struct uld_info *u;
9612 
9613 	sx_xlock(&t4_uld_list_lock);
9614 
9615 	SLIST_FOREACH(u, &t4_uld_list, link) {
9616 	    if (u == ui) {
9617 		    if (ui->refcount > 0) {
9618 			    rc = EBUSY;
9619 			    goto done;
9620 		    }
9621 
9622 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9623 		    rc = 0;
9624 		    goto done;
9625 	    }
9626 	}
9627 done:
9628 	sx_xunlock(&t4_uld_list_lock);
9629 	return (rc);
9630 }
9631 
9632 int
9633 t4_activate_uld(struct adapter *sc, int id)
9634 {
9635 	int rc;
9636 	struct uld_info *ui;
9637 
9638 	ASSERT_SYNCHRONIZED_OP(sc);
9639 
9640 	if (id < 0 || id > ULD_MAX)
9641 		return (EINVAL);
9642 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
9643 
9644 	sx_slock(&t4_uld_list_lock);
9645 
9646 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9647 		if (ui->uld_id == id) {
9648 			if (!(sc->flags & FULL_INIT_DONE)) {
9649 				rc = adapter_full_init(sc);
9650 				if (rc != 0)
9651 					break;
9652 			}
9653 
9654 			rc = ui->activate(sc);
9655 			if (rc == 0) {
9656 				setbit(&sc->active_ulds, id);
9657 				ui->refcount++;
9658 			}
9659 			break;
9660 		}
9661 	}
9662 
9663 	sx_sunlock(&t4_uld_list_lock);
9664 
9665 	return (rc);
9666 }
9667 
9668 int
9669 t4_deactivate_uld(struct adapter *sc, int id)
9670 {
9671 	int rc;
9672 	struct uld_info *ui;
9673 
9674 	ASSERT_SYNCHRONIZED_OP(sc);
9675 
9676 	if (id < 0 || id > ULD_MAX)
9677 		return (EINVAL);
9678 	rc = ENXIO;
9679 
9680 	sx_slock(&t4_uld_list_lock);
9681 
9682 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9683 		if (ui->uld_id == id) {
9684 			rc = ui->deactivate(sc);
9685 			if (rc == 0) {
9686 				clrbit(&sc->active_ulds, id);
9687 				ui->refcount--;
9688 			}
9689 			break;
9690 		}
9691 	}
9692 
9693 	sx_sunlock(&t4_uld_list_lock);
9694 
9695 	return (rc);
9696 }
9697 
9698 int
9699 uld_active(struct adapter *sc, int uld_id)
9700 {
9701 
9702 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9703 
9704 	return (isset(&sc->active_ulds, uld_id));
9705 }
9706 #endif
9707 
9708 /*
9709  * t  = ptr to tunable.
9710  * nc = number of CPUs.
9711  * c  = compiled in default for that tunable.
9712  */
9713 static void
9714 calculate_nqueues(int *t, int nc, const int c)
9715 {
9716 	int nq;
9717 
9718 	if (*t > 0)
9719 		return;
9720 	nq = *t < 0 ? -*t : c;
9721 	*t = min(nc, nq);
9722 }
9723 
9724 /*
9725  * Come up with reasonable defaults for some of the tunables, provided they're
9726  * not set by the user (in which case we'll use the values as is).
9727  */
9728 static void
9729 tweak_tunables(void)
9730 {
9731 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
9732 
9733 	if (t4_ntxq10g < 1) {
9734 #ifdef RSS
9735 		t4_ntxq10g = rss_getnumbuckets();
9736 #else
9737 		calculate_nqueues(&t4_ntxq10g, nc, NTXQ_10G);
9738 #endif
9739 	}
9740 
9741 	if (t4_ntxq1g < 1) {
9742 #ifdef RSS
9743 		/* XXX: way too many for 1GbE? */
9744 		t4_ntxq1g = rss_getnumbuckets();
9745 #else
9746 		calculate_nqueues(&t4_ntxq1g, nc, NTXQ_1G);
9747 #endif
9748 	}
9749 
9750 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
9751 
9752 	if (t4_nrxq10g < 1) {
9753 #ifdef RSS
9754 		t4_nrxq10g = rss_getnumbuckets();
9755 #else
9756 		calculate_nqueues(&t4_nrxq10g, nc, NRXQ_10G);
9757 #endif
9758 	}
9759 
9760 	if (t4_nrxq1g < 1) {
9761 #ifdef RSS
9762 		/* XXX: way too many for 1GbE? */
9763 		t4_nrxq1g = rss_getnumbuckets();
9764 #else
9765 		calculate_nqueues(&t4_nrxq1g, nc, NRXQ_1G);
9766 #endif
9767 	}
9768 
9769 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
9770 
9771 #ifdef TCP_OFFLOAD
9772 	calculate_nqueues(&t4_nofldtxq10g, nc, NOFLDTXQ_10G);
9773 	calculate_nqueues(&t4_nofldtxq1g, nc, NOFLDTXQ_1G);
9774 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
9775 	calculate_nqueues(&t4_nofldrxq10g, nc, NOFLDRXQ_10G);
9776 	calculate_nqueues(&t4_nofldrxq1g, nc, NOFLDRXQ_1G);
9777 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
9778 
9779 	if (t4_toecaps_allowed == -1)
9780 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9781 
9782 	if (t4_rdmacaps_allowed == -1) {
9783 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9784 		    FW_CAPS_CONFIG_RDMA_RDMAC;
9785 	}
9786 
9787 	if (t4_iscsicaps_allowed == -1) {
9788 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9789 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9790 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
9791 	}
9792 #else
9793 	if (t4_toecaps_allowed == -1)
9794 		t4_toecaps_allowed = 0;
9795 
9796 	if (t4_rdmacaps_allowed == -1)
9797 		t4_rdmacaps_allowed = 0;
9798 
9799 	if (t4_iscsicaps_allowed == -1)
9800 		t4_iscsicaps_allowed = 0;
9801 #endif
9802 
9803 #ifdef DEV_NETMAP
9804 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
9805 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
9806 #endif
9807 
9808 	if (t4_tmr_idx_10g < 0 || t4_tmr_idx_10g >= SGE_NTIMERS)
9809 		t4_tmr_idx_10g = TMR_IDX_10G;
9810 
9811 	if (t4_pktc_idx_10g < -1 || t4_pktc_idx_10g >= SGE_NCOUNTERS)
9812 		t4_pktc_idx_10g = PKTC_IDX_10G;
9813 
9814 	if (t4_tmr_idx_1g < 0 || t4_tmr_idx_1g >= SGE_NTIMERS)
9815 		t4_tmr_idx_1g = TMR_IDX_1G;
9816 
9817 	if (t4_pktc_idx_1g < -1 || t4_pktc_idx_1g >= SGE_NCOUNTERS)
9818 		t4_pktc_idx_1g = PKTC_IDX_1G;
9819 
9820 	if (t4_qsize_txq < 128)
9821 		t4_qsize_txq = 128;
9822 
9823 	if (t4_qsize_rxq < 128)
9824 		t4_qsize_rxq = 128;
9825 	while (t4_qsize_rxq & 7)
9826 		t4_qsize_rxq++;
9827 
9828 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9829 }
9830 
9831 #ifdef DDB
9832 static void
9833 t4_dump_tcb(struct adapter *sc, int tid)
9834 {
9835 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
9836 
9837 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
9838 	save = t4_read_reg(sc, reg);
9839 	base = sc->memwin[2].mw_base;
9840 
9841 	/* Dump TCB for the tid */
9842 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
9843 	tcb_addr += tid * TCB_SIZE;
9844 
9845 	if (is_t4(sc)) {
9846 		pf = 0;
9847 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
9848 	} else {
9849 		pf = V_PFNUM(sc->pf);
9850 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
9851 	}
9852 	t4_write_reg(sc, reg, win_pos | pf);
9853 	t4_read_reg(sc, reg);
9854 
9855 	off = tcb_addr - win_pos;
9856 	for (i = 0; i < 4; i++) {
9857 		uint32_t buf[8];
9858 		for (j = 0; j < 8; j++, off += 4)
9859 			buf[j] = htonl(t4_read_reg(sc, base + off));
9860 
9861 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
9862 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
9863 		    buf[7]);
9864 	}
9865 
9866 	t4_write_reg(sc, reg, save);
9867 	t4_read_reg(sc, reg);
9868 }
9869 
9870 static void
9871 t4_dump_devlog(struct adapter *sc)
9872 {
9873 	struct devlog_params *dparams = &sc->params.devlog;
9874 	struct fw_devlog_e e;
9875 	int i, first, j, m, nentries, rc;
9876 	uint64_t ftstamp = UINT64_MAX;
9877 
9878 	if (dparams->start == 0) {
9879 		db_printf("devlog params not valid\n");
9880 		return;
9881 	}
9882 
9883 	nentries = dparams->size / sizeof(struct fw_devlog_e);
9884 	m = fwmtype_to_hwmtype(dparams->memtype);
9885 
9886 	/* Find the first entry. */
9887 	first = -1;
9888 	for (i = 0; i < nentries && !db_pager_quit; i++) {
9889 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9890 		    sizeof(e), (void *)&e);
9891 		if (rc != 0)
9892 			break;
9893 
9894 		if (e.timestamp == 0)
9895 			break;
9896 
9897 		e.timestamp = be64toh(e.timestamp);
9898 		if (e.timestamp < ftstamp) {
9899 			ftstamp = e.timestamp;
9900 			first = i;
9901 		}
9902 	}
9903 
9904 	if (first == -1)
9905 		return;
9906 
9907 	i = first;
9908 	do {
9909 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9910 		    sizeof(e), (void *)&e);
9911 		if (rc != 0)
9912 			return;
9913 
9914 		if (e.timestamp == 0)
9915 			return;
9916 
9917 		e.timestamp = be64toh(e.timestamp);
9918 		e.seqno = be32toh(e.seqno);
9919 		for (j = 0; j < 8; j++)
9920 			e.params[j] = be32toh(e.params[j]);
9921 
9922 		db_printf("%10d  %15ju  %8s  %8s  ",
9923 		    e.seqno, e.timestamp,
9924 		    (e.level < nitems(devlog_level_strings) ?
9925 			devlog_level_strings[e.level] : "UNKNOWN"),
9926 		    (e.facility < nitems(devlog_facility_strings) ?
9927 			devlog_facility_strings[e.facility] : "UNKNOWN"));
9928 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
9929 		    e.params[3], e.params[4], e.params[5], e.params[6],
9930 		    e.params[7]);
9931 
9932 		if (++i == nentries)
9933 			i = 0;
9934 	} while (i != first && !db_pager_quit);
9935 }
9936 
9937 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
9938 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
9939 
9940 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
9941 {
9942 	device_t dev;
9943 	int t;
9944 	bool valid;
9945 
9946 	valid = false;
9947 	t = db_read_token();
9948 	if (t == tIDENT) {
9949 		dev = device_lookup_by_name(db_tok_string);
9950 		valid = true;
9951 	}
9952 	db_skip_to_eol();
9953 	if (!valid) {
9954 		db_printf("usage: show t4 devlog <nexus>\n");
9955 		return;
9956 	}
9957 
9958 	if (dev == NULL) {
9959 		db_printf("device not found\n");
9960 		return;
9961 	}
9962 
9963 	t4_dump_devlog(device_get_softc(dev));
9964 }
9965 
9966 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
9967 {
9968 	device_t dev;
9969 	int radix, tid, t;
9970 	bool valid;
9971 
9972 	valid = false;
9973 	radix = db_radix;
9974 	db_radix = 10;
9975 	t = db_read_token();
9976 	if (t == tIDENT) {
9977 		dev = device_lookup_by_name(db_tok_string);
9978 		t = db_read_token();
9979 		if (t == tNUMBER) {
9980 			tid = db_tok_number;
9981 			valid = true;
9982 		}
9983 	}
9984 	db_radix = radix;
9985 	db_skip_to_eol();
9986 	if (!valid) {
9987 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
9988 		return;
9989 	}
9990 
9991 	if (dev == NULL) {
9992 		db_printf("device not found\n");
9993 		return;
9994 	}
9995 	if (tid < 0) {
9996 		db_printf("invalid tid\n");
9997 		return;
9998 	}
9999 
10000 	t4_dump_tcb(device_get_softc(dev), tid);
10001 }
10002 #endif
10003 
10004 static struct sx mlu;	/* mod load unload */
10005 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
10006 
10007 static int
10008 mod_event(module_t mod, int cmd, void *arg)
10009 {
10010 	int rc = 0;
10011 	static int loaded = 0;
10012 
10013 	switch (cmd) {
10014 	case MOD_LOAD:
10015 		sx_xlock(&mlu);
10016 		if (loaded++ == 0) {
10017 			t4_sge_modload();
10018 			t4_register_cpl_handler(CPL_SET_TCB_RPL, set_tcb_rpl);
10019 			t4_register_cpl_handler(CPL_L2T_WRITE_RPL, l2t_write_rpl);
10020 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
10021 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
10022 			sx_init(&t4_list_lock, "T4/T5 adapters");
10023 			SLIST_INIT(&t4_list);
10024 #ifdef TCP_OFFLOAD
10025 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
10026 			SLIST_INIT(&t4_uld_list);
10027 #endif
10028 			t4_tracer_modload();
10029 			tweak_tunables();
10030 		}
10031 		sx_xunlock(&mlu);
10032 		break;
10033 
10034 	case MOD_UNLOAD:
10035 		sx_xlock(&mlu);
10036 		if (--loaded == 0) {
10037 			int tries;
10038 
10039 			sx_slock(&t4_list_lock);
10040 			if (!SLIST_EMPTY(&t4_list)) {
10041 				rc = EBUSY;
10042 				sx_sunlock(&t4_list_lock);
10043 				goto done_unload;
10044 			}
10045 #ifdef TCP_OFFLOAD
10046 			sx_slock(&t4_uld_list_lock);
10047 			if (!SLIST_EMPTY(&t4_uld_list)) {
10048 				rc = EBUSY;
10049 				sx_sunlock(&t4_uld_list_lock);
10050 				sx_sunlock(&t4_list_lock);
10051 				goto done_unload;
10052 			}
10053 #endif
10054 			tries = 0;
10055 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
10056 				uprintf("%ju clusters with custom free routine "
10057 				    "still is use.\n", t4_sge_extfree_refs());
10058 				pause("t4unload", 2 * hz);
10059 			}
10060 #ifdef TCP_OFFLOAD
10061 			sx_sunlock(&t4_uld_list_lock);
10062 #endif
10063 			sx_sunlock(&t4_list_lock);
10064 
10065 			if (t4_sge_extfree_refs() == 0) {
10066 				t4_tracer_modunload();
10067 #ifdef TCP_OFFLOAD
10068 				sx_destroy(&t4_uld_list_lock);
10069 #endif
10070 				sx_destroy(&t4_list_lock);
10071 				t4_sge_modunload();
10072 				loaded = 0;
10073 			} else {
10074 				rc = EBUSY;
10075 				loaded++;	/* undo earlier decrement */
10076 			}
10077 		}
10078 done_unload:
10079 		sx_xunlock(&mlu);
10080 		break;
10081 	}
10082 
10083 	return (rc);
10084 }
10085 
10086 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10087 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10088 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10089 
10090 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10091 MODULE_VERSION(t4nex, 1);
10092 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10093 #ifdef DEV_NETMAP
10094 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10095 #endif /* DEV_NETMAP */
10096 
10097 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10098 MODULE_VERSION(t5nex, 1);
10099 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10100 #ifdef DEV_NETMAP
10101 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10102 #endif /* DEV_NETMAP */
10103 
10104 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10105 MODULE_VERSION(t6nex, 1);
10106 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10107 #ifdef DEV_NETMAP
10108 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10109 #endif /* DEV_NETMAP */
10110 
10111 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10112 MODULE_VERSION(cxgbe, 1);
10113 
10114 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10115 MODULE_VERSION(cxl, 1);
10116 
10117 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10118 MODULE_VERSION(cc, 1);
10119 
10120 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10121 MODULE_VERSION(vcxgbe, 1);
10122 
10123 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10124 MODULE_VERSION(vcxl, 1);
10125 
10126 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10127 MODULE_VERSION(vcc, 1);
10128