xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision cbd30a72ca196976c1c700400ecd424baa1b9c16)
1 /*-
2  * Copyright (c) 2011 Chelsio Communications, Inc.
3  * All rights reserved.
4  * Written by: Navdeep Parhar <np@FreeBSD.org>
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_ddb.h"
32 #include "opt_inet.h"
33 #include "opt_inet6.h"
34 #include "opt_rss.h"
35 
36 #include <sys/param.h>
37 #include <sys/conf.h>
38 #include <sys/priv.h>
39 #include <sys/kernel.h>
40 #include <sys/bus.h>
41 #include <sys/module.h>
42 #include <sys/malloc.h>
43 #include <sys/queue.h>
44 #include <sys/taskqueue.h>
45 #include <sys/pciio.h>
46 #include <dev/pci/pcireg.h>
47 #include <dev/pci/pcivar.h>
48 #include <dev/pci/pci_private.h>
49 #include <sys/firmware.h>
50 #include <sys/sbuf.h>
51 #include <sys/smp.h>
52 #include <sys/socket.h>
53 #include <sys/sockio.h>
54 #include <sys/sysctl.h>
55 #include <net/ethernet.h>
56 #include <net/if.h>
57 #include <net/if_types.h>
58 #include <net/if_dl.h>
59 #include <net/if_vlan_var.h>
60 #ifdef RSS
61 #include <net/rss_config.h>
62 #endif
63 #if defined(__i386__) || defined(__amd64__)
64 #include <vm/vm.h>
65 #include <vm/pmap.h>
66 #endif
67 #ifdef DDB
68 #include <ddb/ddb.h>
69 #include <ddb/db_lex.h>
70 #endif
71 
72 #include "common/common.h"
73 #include "common/t4_msg.h"
74 #include "common/t4_regs.h"
75 #include "common/t4_regs_values.h"
76 #include "t4_ioctl.h"
77 #include "t4_l2t.h"
78 #include "t4_mp_ring.h"
79 #include "t4_if.h"
80 
81 /* T4 bus driver interface */
82 static int t4_probe(device_t);
83 static int t4_attach(device_t);
84 static int t4_detach(device_t);
85 static int t4_ready(device_t);
86 static int t4_read_port_device(device_t, int, device_t *);
87 static device_method_t t4_methods[] = {
88 	DEVMETHOD(device_probe,		t4_probe),
89 	DEVMETHOD(device_attach,	t4_attach),
90 	DEVMETHOD(device_detach,	t4_detach),
91 
92 	DEVMETHOD(t4_is_main_ready,	t4_ready),
93 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
94 
95 	DEVMETHOD_END
96 };
97 static driver_t t4_driver = {
98 	"t4nex",
99 	t4_methods,
100 	sizeof(struct adapter)
101 };
102 
103 
104 /* T4 port (cxgbe) interface */
105 static int cxgbe_probe(device_t);
106 static int cxgbe_attach(device_t);
107 static int cxgbe_detach(device_t);
108 device_method_t cxgbe_methods[] = {
109 	DEVMETHOD(device_probe,		cxgbe_probe),
110 	DEVMETHOD(device_attach,	cxgbe_attach),
111 	DEVMETHOD(device_detach,	cxgbe_detach),
112 	{ 0, 0 }
113 };
114 static driver_t cxgbe_driver = {
115 	"cxgbe",
116 	cxgbe_methods,
117 	sizeof(struct port_info)
118 };
119 
120 /* T4 VI (vcxgbe) interface */
121 static int vcxgbe_probe(device_t);
122 static int vcxgbe_attach(device_t);
123 static int vcxgbe_detach(device_t);
124 static device_method_t vcxgbe_methods[] = {
125 	DEVMETHOD(device_probe,		vcxgbe_probe),
126 	DEVMETHOD(device_attach,	vcxgbe_attach),
127 	DEVMETHOD(device_detach,	vcxgbe_detach),
128 	{ 0, 0 }
129 };
130 static driver_t vcxgbe_driver = {
131 	"vcxgbe",
132 	vcxgbe_methods,
133 	sizeof(struct vi_info)
134 };
135 
136 static d_ioctl_t t4_ioctl;
137 
138 static struct cdevsw t4_cdevsw = {
139        .d_version = D_VERSION,
140        .d_ioctl = t4_ioctl,
141        .d_name = "t4nex",
142 };
143 
144 /* T5 bus driver interface */
145 static int t5_probe(device_t);
146 static device_method_t t5_methods[] = {
147 	DEVMETHOD(device_probe,		t5_probe),
148 	DEVMETHOD(device_attach,	t4_attach),
149 	DEVMETHOD(device_detach,	t4_detach),
150 
151 	DEVMETHOD(t4_is_main_ready,	t4_ready),
152 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
153 
154 	DEVMETHOD_END
155 };
156 static driver_t t5_driver = {
157 	"t5nex",
158 	t5_methods,
159 	sizeof(struct adapter)
160 };
161 
162 
163 /* T5 port (cxl) interface */
164 static driver_t cxl_driver = {
165 	"cxl",
166 	cxgbe_methods,
167 	sizeof(struct port_info)
168 };
169 
170 /* T5 VI (vcxl) interface */
171 static driver_t vcxl_driver = {
172 	"vcxl",
173 	vcxgbe_methods,
174 	sizeof(struct vi_info)
175 };
176 
177 /* T6 bus driver interface */
178 static int t6_probe(device_t);
179 static device_method_t t6_methods[] = {
180 	DEVMETHOD(device_probe,		t6_probe),
181 	DEVMETHOD(device_attach,	t4_attach),
182 	DEVMETHOD(device_detach,	t4_detach),
183 
184 	DEVMETHOD(t4_is_main_ready,	t4_ready),
185 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
186 
187 	DEVMETHOD_END
188 };
189 static driver_t t6_driver = {
190 	"t6nex",
191 	t6_methods,
192 	sizeof(struct adapter)
193 };
194 
195 
196 /* T6 port (cc) interface */
197 static driver_t cc_driver = {
198 	"cc",
199 	cxgbe_methods,
200 	sizeof(struct port_info)
201 };
202 
203 /* T6 VI (vcc) interface */
204 static driver_t vcc_driver = {
205 	"vcc",
206 	vcxgbe_methods,
207 	sizeof(struct vi_info)
208 };
209 
210 /* ifnet + media interface */
211 static void cxgbe_init(void *);
212 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
213 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
214 static void cxgbe_qflush(struct ifnet *);
215 static int cxgbe_media_change(struct ifnet *);
216 static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
217 
218 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
219 
220 /*
221  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
222  * then ADAPTER_LOCK, then t4_uld_list_lock.
223  */
224 static struct sx t4_list_lock;
225 SLIST_HEAD(, adapter) t4_list;
226 #ifdef TCP_OFFLOAD
227 static struct sx t4_uld_list_lock;
228 SLIST_HEAD(, uld_info) t4_uld_list;
229 #endif
230 
231 /*
232  * Tunables.  See tweak_tunables() too.
233  *
234  * Each tunable is set to a default value here if it's known at compile-time.
235  * Otherwise it is set to -n as an indication to tweak_tunables() that it should
236  * provide a reasonable default (upto n) when the driver is loaded.
237  *
238  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
239  * T5 are under hw.cxl.
240  */
241 
242 /*
243  * Number of queues for tx and rx, 10G and 1G, NIC and offload.
244  */
245 #define NTXQ_10G 16
246 int t4_ntxq10g = -NTXQ_10G;
247 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq10g);
248 
249 #define NRXQ_10G 8
250 int t4_nrxq10g = -NRXQ_10G;
251 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq10g);
252 
253 #define NTXQ_1G 4
254 int t4_ntxq1g = -NTXQ_1G;
255 TUNABLE_INT("hw.cxgbe.ntxq1g", &t4_ntxq1g);
256 
257 #define NRXQ_1G 2
258 int t4_nrxq1g = -NRXQ_1G;
259 TUNABLE_INT("hw.cxgbe.nrxq1g", &t4_nrxq1g);
260 
261 #define NTXQ_VI 1
262 static int t4_ntxq_vi = -NTXQ_VI;
263 TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
264 
265 #define NRXQ_VI 1
266 static int t4_nrxq_vi = -NRXQ_VI;
267 TUNABLE_INT("hw.cxgbe.nrxq_vi", &t4_nrxq_vi);
268 
269 static int t4_rsrv_noflowq = 0;
270 TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
271 
272 #ifdef TCP_OFFLOAD
273 #define NOFLDTXQ_10G 8
274 static int t4_nofldtxq10g = -NOFLDTXQ_10G;
275 TUNABLE_INT("hw.cxgbe.nofldtxq10g", &t4_nofldtxq10g);
276 
277 #define NOFLDRXQ_10G 2
278 static int t4_nofldrxq10g = -NOFLDRXQ_10G;
279 TUNABLE_INT("hw.cxgbe.nofldrxq10g", &t4_nofldrxq10g);
280 
281 #define NOFLDTXQ_1G 2
282 static int t4_nofldtxq1g = -NOFLDTXQ_1G;
283 TUNABLE_INT("hw.cxgbe.nofldtxq1g", &t4_nofldtxq1g);
284 
285 #define NOFLDRXQ_1G 1
286 static int t4_nofldrxq1g = -NOFLDRXQ_1G;
287 TUNABLE_INT("hw.cxgbe.nofldrxq1g", &t4_nofldrxq1g);
288 
289 #define NOFLDTXQ_VI 1
290 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
291 TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
292 
293 #define NOFLDRXQ_VI 1
294 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
295 TUNABLE_INT("hw.cxgbe.nofldrxq_vi", &t4_nofldrxq_vi);
296 #endif
297 
298 #ifdef DEV_NETMAP
299 #define NNMTXQ_VI 2
300 static int t4_nnmtxq_vi = -NNMTXQ_VI;
301 TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
302 
303 #define NNMRXQ_VI 2
304 static int t4_nnmrxq_vi = -NNMRXQ_VI;
305 TUNABLE_INT("hw.cxgbe.nnmrxq_vi", &t4_nnmrxq_vi);
306 #endif
307 
308 /*
309  * Holdoff parameters for 10G and 1G ports.
310  */
311 #define TMR_IDX_10G 1
312 int t4_tmr_idx_10g = TMR_IDX_10G;
313 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx_10g);
314 
315 #define PKTC_IDX_10G (-1)
316 int t4_pktc_idx_10g = PKTC_IDX_10G;
317 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx_10g);
318 
319 #define TMR_IDX_1G 1
320 int t4_tmr_idx_1g = TMR_IDX_1G;
321 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_1G", &t4_tmr_idx_1g);
322 
323 #define PKTC_IDX_1G (-1)
324 int t4_pktc_idx_1g = PKTC_IDX_1G;
325 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_1G", &t4_pktc_idx_1g);
326 
327 /*
328  * Size (# of entries) of each tx and rx queue.
329  */
330 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
331 TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
332 
333 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
334 TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
335 
336 /*
337  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
338  */
339 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
340 TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
341 
342 /*
343  * Configuration file.
344  */
345 #define DEFAULT_CF	"default"
346 #define FLASH_CF	"flash"
347 #define UWIRE_CF	"uwire"
348 #define FPGA_CF		"fpga"
349 static char t4_cfg_file[32] = DEFAULT_CF;
350 TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
351 
352 /*
353  * PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
354  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
355  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
356  *            mark or when signalled to do so, 0 to never emit PAUSE.
357  */
358 static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
359 TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
360 
361 /*
362  * Forward Error Correction settings (bit 0, 1, 2 = FEC_RS, FEC_BASER_RS,
363  * FEC_RESERVED respectively).
364  * -1 to run with the firmware default.
365  *  0 to disable FEC.
366  */
367 static int t4_fec = -1;
368 TUNABLE_INT("hw.cxgbe.fec", &t4_fec);
369 
370 /*
371  * Link autonegotiation.
372  * -1 to run with the firmware default.
373  *  0 to disable.
374  *  1 to enable.
375  */
376 static int t4_autoneg = -1;
377 TUNABLE_INT("hw.cxgbe.autoneg", &t4_autoneg);
378 
379 /*
380  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
381  * encouraged respectively).
382  */
383 static unsigned int t4_fw_install = 1;
384 TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
385 
386 /*
387  * ASIC features that will be used.  Disable the ones you don't want so that the
388  * chip resources aren't wasted on features that will not be used.
389  */
390 static int t4_nbmcaps_allowed = 0;
391 TUNABLE_INT("hw.cxgbe.nbmcaps_allowed", &t4_nbmcaps_allowed);
392 
393 static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
394 TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
395 
396 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
397     FW_CAPS_CONFIG_SWITCH_EGRESS;
398 TUNABLE_INT("hw.cxgbe.switchcaps_allowed", &t4_switchcaps_allowed);
399 
400 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC;
401 TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
402 
403 static int t4_toecaps_allowed = -1;
404 TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
405 
406 static int t4_rdmacaps_allowed = -1;
407 TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
408 
409 static int t4_cryptocaps_allowed = 0;
410 TUNABLE_INT("hw.cxgbe.cryptocaps_allowed", &t4_cryptocaps_allowed);
411 
412 static int t4_iscsicaps_allowed = -1;
413 TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
414 
415 static int t4_fcoecaps_allowed = 0;
416 TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
417 
418 static int t5_write_combine = 0;
419 TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
420 
421 static int t4_num_vis = 1;
422 TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
423 
424 /* Functions used by extra VIs to obtain unique MAC addresses for each VI. */
425 static int vi_mac_funcs[] = {
426 	FW_VI_FUNC_OFLD,
427 	FW_VI_FUNC_IWARP,
428 	FW_VI_FUNC_OPENISCSI,
429 	FW_VI_FUNC_OPENFCOE,
430 	FW_VI_FUNC_FOISCSI,
431 	FW_VI_FUNC_FOFCOE,
432 };
433 
434 struct intrs_and_queues {
435 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
436 	uint16_t nirq;		/* Total # of vectors */
437 	uint16_t intr_flags_10g;/* Interrupt flags for each 10G port */
438 	uint16_t intr_flags_1g;	/* Interrupt flags for each 1G port */
439 	uint16_t ntxq10g;	/* # of NIC txq's for each 10G port */
440 	uint16_t nrxq10g;	/* # of NIC rxq's for each 10G port */
441 	uint16_t ntxq1g;	/* # of NIC txq's for each 1G port */
442 	uint16_t nrxq1g;	/* # of NIC rxq's for each 1G port */
443 	uint16_t rsrv_noflowq;	/* Flag whether to reserve queue 0 */
444 	uint16_t nofldtxq10g;	/* # of TOE txq's for each 10G port */
445 	uint16_t nofldrxq10g;	/* # of TOE rxq's for each 10G port */
446 	uint16_t nofldtxq1g;	/* # of TOE txq's for each 1G port */
447 	uint16_t nofldrxq1g;	/* # of TOE rxq's for each 1G port */
448 
449 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
450 	uint16_t ntxq_vi;	/* # of NIC txq's */
451 	uint16_t nrxq_vi;	/* # of NIC rxq's */
452 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
453 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
454 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
455 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
456 };
457 
458 struct filter_entry {
459         uint32_t valid:1;	/* filter allocated and valid */
460         uint32_t locked:1;	/* filter is administratively locked */
461         uint32_t pending:1;	/* filter action is pending firmware reply */
462 	uint32_t smtidx:8;	/* Source MAC Table index for smac */
463 	struct l2t_entry *l2t;	/* Layer Two Table entry for dmac */
464 
465         struct t4_filter_specification fs;
466 };
467 
468 static void setup_memwin(struct adapter *);
469 static void position_memwin(struct adapter *, int, uint32_t);
470 static int rw_via_memwin(struct adapter *, int, uint32_t, uint32_t *, int, int);
471 static inline int read_via_memwin(struct adapter *, int, uint32_t, uint32_t *,
472     int);
473 static inline int write_via_memwin(struct adapter *, int, uint32_t,
474     const uint32_t *, int);
475 static int validate_mem_range(struct adapter *, uint32_t, int);
476 static int fwmtype_to_hwmtype(int);
477 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
478     uint32_t *);
479 static int fixup_devlog_params(struct adapter *);
480 static int cfg_itype_and_nqueues(struct adapter *, int, int, int,
481     struct intrs_and_queues *);
482 static int prep_firmware(struct adapter *);
483 static int partition_resources(struct adapter *, const struct firmware *,
484     const char *);
485 static int get_params__pre_init(struct adapter *);
486 static int get_params__post_init(struct adapter *);
487 static int set_params__post_init(struct adapter *);
488 static void t4_set_desc(struct adapter *);
489 static void build_medialist(struct port_info *, struct ifmedia *);
490 static int cxgbe_init_synchronized(struct vi_info *);
491 static int cxgbe_uninit_synchronized(struct vi_info *);
492 static void quiesce_txq(struct adapter *, struct sge_txq *);
493 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
494 static void quiesce_iq(struct adapter *, struct sge_iq *);
495 static void quiesce_fl(struct adapter *, struct sge_fl *);
496 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
497     driver_intr_t *, void *, char *);
498 static int t4_free_irq(struct adapter *, struct irq *);
499 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
500 static void vi_refresh_stats(struct adapter *, struct vi_info *);
501 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
502 static void cxgbe_tick(void *);
503 static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
504 static void cxgbe_sysctls(struct port_info *);
505 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
506 static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
507 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
508 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
509 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
510 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
511 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
512 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
513 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
514 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
515 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
516 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
517 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
518 #ifdef SBUF_DRAIN
519 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
520 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
521 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
522 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
523 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
524 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
525 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
526 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
527 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
528 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
529 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
530 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
531 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
532 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
533 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
534 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
535 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
536 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
537 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
538 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
539 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
540 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
541 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
542 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
543 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
544 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
545 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
546 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
547 static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
548 #endif
549 #ifdef TCP_OFFLOAD
550 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
551 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
552 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
553 #endif
554 static uint32_t fconf_iconf_to_mode(uint32_t, uint32_t);
555 static uint32_t mode_to_fconf(uint32_t);
556 static uint32_t mode_to_iconf(uint32_t);
557 static int check_fspec_against_fconf_iconf(struct adapter *,
558     struct t4_filter_specification *);
559 static int get_filter_mode(struct adapter *, uint32_t *);
560 static int set_filter_mode(struct adapter *, uint32_t);
561 static inline uint64_t get_filter_hits(struct adapter *, uint32_t);
562 static int get_filter(struct adapter *, struct t4_filter *);
563 static int set_filter(struct adapter *, struct t4_filter *);
564 static int del_filter(struct adapter *, struct t4_filter *);
565 static void clear_filter(struct filter_entry *);
566 static int set_filter_wr(struct adapter *, int);
567 static int del_filter_wr(struct adapter *, int);
568 static int set_tcb_rpl(struct sge_iq *, const struct rss_header *,
569     struct mbuf *);
570 static int get_sge_context(struct adapter *, struct t4_sge_context *);
571 static int load_fw(struct adapter *, struct t4_data *);
572 static int load_cfg(struct adapter *, struct t4_data *);
573 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
574 static int read_i2c(struct adapter *, struct t4_i2c_data *);
575 #ifdef TCP_OFFLOAD
576 static int toe_capability(struct vi_info *, int);
577 #endif
578 static int mod_event(module_t, int, void *);
579 static int notify_siblings(device_t, int);
580 
581 struct {
582 	uint16_t device;
583 	char *desc;
584 } t4_pciids[] = {
585 	{0xa000, "Chelsio Terminator 4 FPGA"},
586 	{0x4400, "Chelsio T440-dbg"},
587 	{0x4401, "Chelsio T420-CR"},
588 	{0x4402, "Chelsio T422-CR"},
589 	{0x4403, "Chelsio T440-CR"},
590 	{0x4404, "Chelsio T420-BCH"},
591 	{0x4405, "Chelsio T440-BCH"},
592 	{0x4406, "Chelsio T440-CH"},
593 	{0x4407, "Chelsio T420-SO"},
594 	{0x4408, "Chelsio T420-CX"},
595 	{0x4409, "Chelsio T420-BT"},
596 	{0x440a, "Chelsio T404-BT"},
597 	{0x440e, "Chelsio T440-LP-CR"},
598 }, t5_pciids[] = {
599 	{0xb000, "Chelsio Terminator 5 FPGA"},
600 	{0x5400, "Chelsio T580-dbg"},
601 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
602 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
603 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
604 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
605 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
606 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
607 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
608 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
609 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
610 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
611 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
612 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
613 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
614 #ifdef notyet
615 	{0x5404,  "Chelsio T520-BCH"},
616 	{0x5405,  "Chelsio T540-BCH"},
617 	{0x5406,  "Chelsio T540-CH"},
618 	{0x5408,  "Chelsio T520-CX"},
619 	{0x540b,  "Chelsio B520-SR"},
620 	{0x540c,  "Chelsio B504-BT"},
621 	{0x540f,  "Chelsio Amsterdam"},
622 	{0x5413,  "Chelsio T580-CHR"},
623 #endif
624 }, t6_pciids[] = {
625 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
626 	{0x6400, "Chelsio T6225-DBG"},		/* 2 x 10/25G, debug */
627 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
628 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
629 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
630 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
631 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
632 	{0x6410, "Chelsio T62100-DBG"},		/* 2 x 40/50/100G, debug */
633 };
634 
635 #ifdef TCP_OFFLOAD
636 /*
637  * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
638  * exactly the same for both rxq and ofld_rxq.
639  */
640 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
641 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
642 #endif
643 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
644 
645 static int
646 t4_probe(device_t dev)
647 {
648 	int i;
649 	uint16_t v = pci_get_vendor(dev);
650 	uint16_t d = pci_get_device(dev);
651 	uint8_t f = pci_get_function(dev);
652 
653 	if (v != PCI_VENDOR_ID_CHELSIO)
654 		return (ENXIO);
655 
656 	/* Attach only to PF0 of the FPGA */
657 	if (d == 0xa000 && f != 0)
658 		return (ENXIO);
659 
660 	for (i = 0; i < nitems(t4_pciids); i++) {
661 		if (d == t4_pciids[i].device) {
662 			device_set_desc(dev, t4_pciids[i].desc);
663 			return (BUS_PROBE_DEFAULT);
664 		}
665 	}
666 
667 	return (ENXIO);
668 }
669 
670 static int
671 t5_probe(device_t dev)
672 {
673 	int i;
674 	uint16_t v = pci_get_vendor(dev);
675 	uint16_t d = pci_get_device(dev);
676 	uint8_t f = pci_get_function(dev);
677 
678 	if (v != PCI_VENDOR_ID_CHELSIO)
679 		return (ENXIO);
680 
681 	/* Attach only to PF0 of the FPGA */
682 	if (d == 0xb000 && f != 0)
683 		return (ENXIO);
684 
685 	for (i = 0; i < nitems(t5_pciids); i++) {
686 		if (d == t5_pciids[i].device) {
687 			device_set_desc(dev, t5_pciids[i].desc);
688 			return (BUS_PROBE_DEFAULT);
689 		}
690 	}
691 
692 	return (ENXIO);
693 }
694 
695 static int
696 t6_probe(device_t dev)
697 {
698 	int i;
699 	uint16_t v = pci_get_vendor(dev);
700 	uint16_t d = pci_get_device(dev);
701 
702 	if (v != PCI_VENDOR_ID_CHELSIO)
703 		return (ENXIO);
704 
705 	for (i = 0; i < nitems(t6_pciids); i++) {
706 		if (d == t6_pciids[i].device) {
707 			device_set_desc(dev, t6_pciids[i].desc);
708 			return (BUS_PROBE_DEFAULT);
709 		}
710 	}
711 
712 	return (ENXIO);
713 }
714 
715 static void
716 t5_attribute_workaround(device_t dev)
717 {
718 	device_t root_port;
719 	uint32_t v;
720 
721 	/*
722 	 * The T5 chips do not properly echo the No Snoop and Relaxed
723 	 * Ordering attributes when replying to a TLP from a Root
724 	 * Port.  As a workaround, find the parent Root Port and
725 	 * disable No Snoop and Relaxed Ordering.  Note that this
726 	 * affects all devices under this root port.
727 	 */
728 	root_port = pci_find_pcie_root_port(dev);
729 	if (root_port == NULL) {
730 		device_printf(dev, "Unable to find parent root port\n");
731 		return;
732 	}
733 
734 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
735 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
736 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
737 	    0)
738 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
739 		    device_get_nameunit(root_port));
740 }
741 
742 static const struct devnames devnames[] = {
743 	{
744 		.nexus_name = "t4nex",
745 		.ifnet_name = "cxgbe",
746 		.vi_ifnet_name = "vcxgbe",
747 		.pf03_drv_name = "t4iov",
748 		.vf_nexus_name = "t4vf",
749 		.vf_ifnet_name = "cxgbev"
750 	}, {
751 		.nexus_name = "t5nex",
752 		.ifnet_name = "cxl",
753 		.vi_ifnet_name = "vcxl",
754 		.pf03_drv_name = "t5iov",
755 		.vf_nexus_name = "t5vf",
756 		.vf_ifnet_name = "cxlv"
757 	}, {
758 		.nexus_name = "t6nex",
759 		.ifnet_name = "cc",
760 		.vi_ifnet_name = "vcc",
761 		.pf03_drv_name = "t6iov",
762 		.vf_nexus_name = "t6vf",
763 		.vf_ifnet_name = "ccv"
764 	}
765 };
766 
767 void
768 t4_init_devnames(struct adapter *sc)
769 {
770 	int id;
771 
772 	id = chip_id(sc);
773 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
774 		sc->names = &devnames[id - CHELSIO_T4];
775 	else {
776 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
777 		sc->names = NULL;
778 	}
779 }
780 
781 static int
782 t4_attach(device_t dev)
783 {
784 	struct adapter *sc;
785 	int rc = 0, i, j, n10g, n1g, rqidx, tqidx;
786 	struct make_dev_args mda;
787 	struct intrs_and_queues iaq;
788 	struct sge *s;
789 	uint8_t *buf;
790 #ifdef TCP_OFFLOAD
791 	int ofld_rqidx, ofld_tqidx;
792 #endif
793 #ifdef DEV_NETMAP
794 	int nm_rqidx, nm_tqidx;
795 #endif
796 	int num_vis;
797 
798 	sc = device_get_softc(dev);
799 	sc->dev = dev;
800 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
801 
802 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
803 		t5_attribute_workaround(dev);
804 	pci_enable_busmaster(dev);
805 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
806 		uint32_t v;
807 
808 		pci_set_max_read_req(dev, 4096);
809 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
810 		v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
811 		pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
812 
813 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
814 	}
815 
816 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
817 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
818 	sc->traceq = -1;
819 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
820 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
821 	    device_get_nameunit(dev));
822 
823 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
824 	    device_get_nameunit(dev));
825 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
826 	t4_add_adapter(sc);
827 
828 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
829 	TAILQ_INIT(&sc->sfl);
830 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
831 
832 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
833 
834 	rc = t4_map_bars_0_and_4(sc);
835 	if (rc != 0)
836 		goto done; /* error message displayed already */
837 
838 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
839 
840 	/* Prepare the adapter for operation. */
841 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
842 	rc = -t4_prep_adapter(sc, buf);
843 	free(buf, M_CXGBE);
844 	if (rc != 0) {
845 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
846 		goto done;
847 	}
848 
849 	/*
850 	 * This is the real PF# to which we're attaching.  Works from within PCI
851 	 * passthrough environments too, where pci_get_function() could return a
852 	 * different PF# depending on the passthrough configuration.  We need to
853 	 * use the real PF# in all our communication with the firmware.
854 	 */
855 	j = t4_read_reg(sc, A_PL_WHOAMI);
856 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
857 	sc->mbox = sc->pf;
858 
859 	t4_init_devnames(sc);
860 	if (sc->names == NULL) {
861 		rc = ENOTSUP;
862 		goto done; /* error message displayed already */
863 	}
864 
865 	/*
866 	 * Do this really early, with the memory windows set up even before the
867 	 * character device.  The userland tool's register i/o and mem read
868 	 * will work even in "recovery mode".
869 	 */
870 	setup_memwin(sc);
871 	if (t4_init_devlog_params(sc, 0) == 0)
872 		fixup_devlog_params(sc);
873 	make_dev_args_init(&mda);
874 	mda.mda_devsw = &t4_cdevsw;
875 	mda.mda_uid = UID_ROOT;
876 	mda.mda_gid = GID_WHEEL;
877 	mda.mda_mode = 0600;
878 	mda.mda_si_drv1 = sc;
879 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
880 	if (rc != 0)
881 		device_printf(dev, "failed to create nexus char device: %d.\n",
882 		    rc);
883 
884 	/* Go no further if recovery mode has been requested. */
885 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
886 		device_printf(dev, "recovery mode.\n");
887 		goto done;
888 	}
889 
890 #if defined(__i386__)
891 	if ((cpu_feature & CPUID_CX8) == 0) {
892 		device_printf(dev, "64 bit atomics not available.\n");
893 		rc = ENOTSUP;
894 		goto done;
895 	}
896 #endif
897 
898 	/* Prepare the firmware for operation */
899 	rc = prep_firmware(sc);
900 	if (rc != 0)
901 		goto done; /* error message displayed already */
902 
903 	rc = get_params__post_init(sc);
904 	if (rc != 0)
905 		goto done; /* error message displayed already */
906 
907 	rc = set_params__post_init(sc);
908 	if (rc != 0)
909 		goto done; /* error message displayed already */
910 
911 	rc = t4_map_bar_2(sc);
912 	if (rc != 0)
913 		goto done; /* error message displayed already */
914 
915 	rc = t4_create_dma_tag(sc);
916 	if (rc != 0)
917 		goto done; /* error message displayed already */
918 
919 	/*
920 	 * Number of VIs to create per-port.  The first VI is the "main" regular
921 	 * VI for the port.  The rest are additional virtual interfaces on the
922 	 * same physical port.  Note that the main VI does not have native
923 	 * netmap support but the extra VIs do.
924 	 *
925 	 * Limit the number of VIs per port to the number of available
926 	 * MAC addresses per port.
927 	 */
928 	if (t4_num_vis >= 1)
929 		num_vis = t4_num_vis;
930 	else
931 		num_vis = 1;
932 	if (num_vis > nitems(vi_mac_funcs)) {
933 		num_vis = nitems(vi_mac_funcs);
934 		device_printf(dev, "Number of VIs limited to %d\n", num_vis);
935 	}
936 
937 	/*
938 	 * First pass over all the ports - allocate VIs and initialize some
939 	 * basic parameters like mac address, port type, etc.  We also figure
940 	 * out whether a port is 10G or 1G and use that information when
941 	 * calculating how many interrupts to attempt to allocate.
942 	 */
943 	n10g = n1g = 0;
944 	for_each_port(sc, i) {
945 		struct port_info *pi;
946 		struct link_config *lc;
947 
948 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
949 		sc->port[i] = pi;
950 
951 		/* These must be set before t4_port_init */
952 		pi->adapter = sc;
953 		pi->port_id = i;
954 		/*
955 		 * XXX: vi[0] is special so we can't delay this allocation until
956 		 * pi->nvi's final value is known.
957 		 */
958 		pi->vi = malloc(sizeof(struct vi_info) * num_vis, M_CXGBE,
959 		    M_ZERO | M_WAITOK);
960 
961 		/*
962 		 * Allocate the "main" VI and initialize parameters
963 		 * like mac addr.
964 		 */
965 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
966 		if (rc != 0) {
967 			device_printf(dev, "unable to initialize port %d: %d\n",
968 			    i, rc);
969 			free(pi->vi, M_CXGBE);
970 			free(pi, M_CXGBE);
971 			sc->port[i] = NULL;
972 			goto done;
973 		}
974 
975 		lc = &pi->link_cfg;
976 		lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
977 		lc->requested_fc |= t4_pause_settings;
978 		if (t4_fec != -1) {
979 			lc->requested_fec = t4_fec &
980 			    G_FW_PORT_CAP_FEC(lc->supported);
981 		}
982 		if (lc->supported & FW_PORT_CAP_ANEG && t4_autoneg != -1) {
983 			lc->autoneg = t4_autoneg ? AUTONEG_ENABLE :
984 			    AUTONEG_DISABLE;
985 		}
986 
987 		rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
988 		if (rc != 0) {
989 			device_printf(dev, "port %d l1cfg failed: %d\n", i, rc);
990 			free(pi->vi, M_CXGBE);
991 			free(pi, M_CXGBE);
992 			sc->port[i] = NULL;
993 			goto done;
994 		}
995 
996 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
997 		    device_get_nameunit(dev), i);
998 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
999 		sc->chan_map[pi->tx_chan] = i;
1000 
1001 		pi->tc = malloc(sizeof(struct tx_sched_class) *
1002 		    sc->chip_params->nsched_cls, M_CXGBE, M_ZERO | M_WAITOK);
1003 
1004 		if (port_top_speed(pi) >= 10) {
1005 			n10g++;
1006 		} else {
1007 			n1g++;
1008 		}
1009 
1010 		pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
1011 		if (pi->dev == NULL) {
1012 			device_printf(dev,
1013 			    "failed to add device for port %d.\n", i);
1014 			rc = ENXIO;
1015 			goto done;
1016 		}
1017 		pi->vi[0].dev = pi->dev;
1018 		device_set_softc(pi->dev, pi);
1019 	}
1020 
1021 	/*
1022 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1023 	 */
1024 	rc = cfg_itype_and_nqueues(sc, n10g, n1g, num_vis, &iaq);
1025 	if (rc != 0)
1026 		goto done; /* error message displayed already */
1027 	if (iaq.nrxq_vi + iaq.nofldrxq_vi + iaq.nnmrxq_vi == 0)
1028 		num_vis = 1;
1029 
1030 	sc->intr_type = iaq.intr_type;
1031 	sc->intr_count = iaq.nirq;
1032 
1033 	s = &sc->sge;
1034 	s->nrxq = n10g * iaq.nrxq10g + n1g * iaq.nrxq1g;
1035 	s->ntxq = n10g * iaq.ntxq10g + n1g * iaq.ntxq1g;
1036 	if (num_vis > 1) {
1037 		s->nrxq += (n10g + n1g) * (num_vis - 1) * iaq.nrxq_vi;
1038 		s->ntxq += (n10g + n1g) * (num_vis - 1) * iaq.ntxq_vi;
1039 	}
1040 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1041 	s->neq += sc->params.nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
1042 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1043 #ifdef TCP_OFFLOAD
1044 	if (is_offload(sc)) {
1045 		s->nofldrxq = n10g * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
1046 		s->nofldtxq = n10g * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
1047 		if (num_vis > 1) {
1048 			s->nofldrxq += (n10g + n1g) * (num_vis - 1) *
1049 			    iaq.nofldrxq_vi;
1050 			s->nofldtxq += (n10g + n1g) * (num_vis - 1) *
1051 			    iaq.nofldtxq_vi;
1052 		}
1053 		s->neq += s->nofldtxq + s->nofldrxq;
1054 		s->niq += s->nofldrxq;
1055 
1056 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1057 		    M_CXGBE, M_ZERO | M_WAITOK);
1058 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1059 		    M_CXGBE, M_ZERO | M_WAITOK);
1060 	}
1061 #endif
1062 #ifdef DEV_NETMAP
1063 	if (num_vis > 1) {
1064 		s->nnmrxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmrxq_vi;
1065 		s->nnmtxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmtxq_vi;
1066 	}
1067 	s->neq += s->nnmtxq + s->nnmrxq;
1068 	s->niq += s->nnmrxq;
1069 
1070 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1071 	    M_CXGBE, M_ZERO | M_WAITOK);
1072 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1073 	    M_CXGBE, M_ZERO | M_WAITOK);
1074 #endif
1075 
1076 	s->ctrlq = malloc(sc->params.nports * sizeof(struct sge_wrq), M_CXGBE,
1077 	    M_ZERO | M_WAITOK);
1078 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1079 	    M_ZERO | M_WAITOK);
1080 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1081 	    M_ZERO | M_WAITOK);
1082 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1083 	    M_ZERO | M_WAITOK);
1084 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1085 	    M_ZERO | M_WAITOK);
1086 
1087 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1088 	    M_ZERO | M_WAITOK);
1089 
1090 	t4_init_l2t(sc, M_WAITOK);
1091 
1092 	/*
1093 	 * Second pass over the ports.  This time we know the number of rx and
1094 	 * tx queues that each port should get.
1095 	 */
1096 	rqidx = tqidx = 0;
1097 #ifdef TCP_OFFLOAD
1098 	ofld_rqidx = ofld_tqidx = 0;
1099 #endif
1100 #ifdef DEV_NETMAP
1101 	nm_rqidx = nm_tqidx = 0;
1102 #endif
1103 	for_each_port(sc, i) {
1104 		struct port_info *pi = sc->port[i];
1105 		struct vi_info *vi;
1106 
1107 		if (pi == NULL)
1108 			continue;
1109 
1110 		pi->nvi = num_vis;
1111 		for_each_vi(pi, j, vi) {
1112 			vi->pi = pi;
1113 			vi->qsize_rxq = t4_qsize_rxq;
1114 			vi->qsize_txq = t4_qsize_txq;
1115 
1116 			vi->first_rxq = rqidx;
1117 			vi->first_txq = tqidx;
1118 			if (port_top_speed(pi) >= 10) {
1119 				vi->tmr_idx = t4_tmr_idx_10g;
1120 				vi->pktc_idx = t4_pktc_idx_10g;
1121 				vi->flags |= iaq.intr_flags_10g & INTR_RXQ;
1122 				vi->nrxq = j == 0 ? iaq.nrxq10g : iaq.nrxq_vi;
1123 				vi->ntxq = j == 0 ? iaq.ntxq10g : iaq.ntxq_vi;
1124 			} else {
1125 				vi->tmr_idx = t4_tmr_idx_1g;
1126 				vi->pktc_idx = t4_pktc_idx_1g;
1127 				vi->flags |= iaq.intr_flags_1g & INTR_RXQ;
1128 				vi->nrxq = j == 0 ? iaq.nrxq1g : iaq.nrxq_vi;
1129 				vi->ntxq = j == 0 ? iaq.ntxq1g : iaq.ntxq_vi;
1130 			}
1131 			rqidx += vi->nrxq;
1132 			tqidx += vi->ntxq;
1133 
1134 			if (j == 0 && vi->ntxq > 1)
1135 				vi->rsrv_noflowq = iaq.rsrv_noflowq ? 1 : 0;
1136 			else
1137 				vi->rsrv_noflowq = 0;
1138 
1139 #ifdef TCP_OFFLOAD
1140 			vi->first_ofld_rxq = ofld_rqidx;
1141 			vi->first_ofld_txq = ofld_tqidx;
1142 			if (port_top_speed(pi) >= 10) {
1143 				vi->flags |= iaq.intr_flags_10g & INTR_OFLD_RXQ;
1144 				vi->nofldrxq = j == 0 ? iaq.nofldrxq10g :
1145 				    iaq.nofldrxq_vi;
1146 				vi->nofldtxq = j == 0 ? iaq.nofldtxq10g :
1147 				    iaq.nofldtxq_vi;
1148 			} else {
1149 				vi->flags |= iaq.intr_flags_1g & INTR_OFLD_RXQ;
1150 				vi->nofldrxq = j == 0 ? iaq.nofldrxq1g :
1151 				    iaq.nofldrxq_vi;
1152 				vi->nofldtxq = j == 0 ? iaq.nofldtxq1g :
1153 				    iaq.nofldtxq_vi;
1154 			}
1155 			ofld_rqidx += vi->nofldrxq;
1156 			ofld_tqidx += vi->nofldtxq;
1157 #endif
1158 #ifdef DEV_NETMAP
1159 			if (j > 0) {
1160 				vi->first_nm_rxq = nm_rqidx;
1161 				vi->first_nm_txq = nm_tqidx;
1162 				vi->nnmrxq = iaq.nnmrxq_vi;
1163 				vi->nnmtxq = iaq.nnmtxq_vi;
1164 				nm_rqidx += vi->nnmrxq;
1165 				nm_tqidx += vi->nnmtxq;
1166 			}
1167 #endif
1168 		}
1169 	}
1170 
1171 	rc = t4_setup_intr_handlers(sc);
1172 	if (rc != 0) {
1173 		device_printf(dev,
1174 		    "failed to setup interrupt handlers: %d\n", rc);
1175 		goto done;
1176 	}
1177 
1178 	rc = bus_generic_attach(dev);
1179 	if (rc != 0) {
1180 		device_printf(dev,
1181 		    "failed to attach all child ports: %d\n", rc);
1182 		goto done;
1183 	}
1184 
1185 	device_printf(dev,
1186 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1187 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1188 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1189 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1190 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1191 
1192 	t4_set_desc(sc);
1193 
1194 	notify_siblings(dev, 0);
1195 
1196 done:
1197 	if (rc != 0 && sc->cdev) {
1198 		/* cdev was created and so cxgbetool works; recover that way. */
1199 		device_printf(dev,
1200 		    "error during attach, adapter is now in recovery mode.\n");
1201 		rc = 0;
1202 	}
1203 
1204 	if (rc != 0)
1205 		t4_detach_common(dev);
1206 	else
1207 		t4_sysctls(sc);
1208 
1209 	return (rc);
1210 }
1211 
1212 static int
1213 t4_ready(device_t dev)
1214 {
1215 	struct adapter *sc;
1216 
1217 	sc = device_get_softc(dev);
1218 	if (sc->flags & FW_OK)
1219 		return (0);
1220 	return (ENXIO);
1221 }
1222 
1223 static int
1224 t4_read_port_device(device_t dev, int port, device_t *child)
1225 {
1226 	struct adapter *sc;
1227 	struct port_info *pi;
1228 
1229 	sc = device_get_softc(dev);
1230 	if (port < 0 || port >= MAX_NPORTS)
1231 		return (EINVAL);
1232 	pi = sc->port[port];
1233 	if (pi == NULL || pi->dev == NULL)
1234 		return (ENXIO);
1235 	*child = pi->dev;
1236 	return (0);
1237 }
1238 
1239 static int
1240 notify_siblings(device_t dev, int detaching)
1241 {
1242 	device_t sibling;
1243 	int error, i;
1244 
1245 	error = 0;
1246 	for (i = 0; i < PCI_FUNCMAX; i++) {
1247 		if (i == pci_get_function(dev))
1248 			continue;
1249 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1250 		    pci_get_slot(dev), i);
1251 		if (sibling == NULL || !device_is_attached(sibling))
1252 			continue;
1253 		if (detaching)
1254 			error = T4_DETACH_CHILD(sibling);
1255 		else
1256 			(void)T4_ATTACH_CHILD(sibling);
1257 		if (error)
1258 			break;
1259 	}
1260 	return (error);
1261 }
1262 
1263 /*
1264  * Idempotent
1265  */
1266 static int
1267 t4_detach(device_t dev)
1268 {
1269 	struct adapter *sc;
1270 	int rc;
1271 
1272 	sc = device_get_softc(dev);
1273 
1274 	rc = notify_siblings(dev, 1);
1275 	if (rc) {
1276 		device_printf(dev,
1277 		    "failed to detach sibling devices: %d\n", rc);
1278 		return (rc);
1279 	}
1280 
1281 	return (t4_detach_common(dev));
1282 }
1283 
1284 int
1285 t4_detach_common(device_t dev)
1286 {
1287 	struct adapter *sc;
1288 	struct port_info *pi;
1289 	int i, rc;
1290 
1291 	sc = device_get_softc(dev);
1292 
1293 	if (sc->flags & FULL_INIT_DONE) {
1294 		if (!(sc->flags & IS_VF))
1295 			t4_intr_disable(sc);
1296 	}
1297 
1298 	if (sc->cdev) {
1299 		destroy_dev(sc->cdev);
1300 		sc->cdev = NULL;
1301 	}
1302 
1303 	if (device_is_attached(dev)) {
1304 		rc = bus_generic_detach(dev);
1305 		if (rc) {
1306 			device_printf(dev,
1307 			    "failed to detach child devices: %d\n", rc);
1308 			return (rc);
1309 		}
1310 	}
1311 
1312 	for (i = 0; i < sc->intr_count; i++)
1313 		t4_free_irq(sc, &sc->irq[i]);
1314 
1315 	for (i = 0; i < MAX_NPORTS; i++) {
1316 		pi = sc->port[i];
1317 		if (pi) {
1318 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1319 			if (pi->dev)
1320 				device_delete_child(dev, pi->dev);
1321 
1322 			mtx_destroy(&pi->pi_lock);
1323 			free(pi->vi, M_CXGBE);
1324 			free(pi->tc, M_CXGBE);
1325 			free(pi, M_CXGBE);
1326 		}
1327 	}
1328 
1329 	if (sc->flags & FULL_INIT_DONE)
1330 		adapter_full_uninit(sc);
1331 
1332 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1333 		t4_fw_bye(sc, sc->mbox);
1334 
1335 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1336 		pci_release_msi(dev);
1337 
1338 	if (sc->regs_res)
1339 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1340 		    sc->regs_res);
1341 
1342 	if (sc->udbs_res)
1343 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1344 		    sc->udbs_res);
1345 
1346 	if (sc->msix_res)
1347 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1348 		    sc->msix_res);
1349 
1350 	if (sc->l2t)
1351 		t4_free_l2t(sc->l2t);
1352 
1353 #ifdef TCP_OFFLOAD
1354 	free(sc->sge.ofld_rxq, M_CXGBE);
1355 	free(sc->sge.ofld_txq, M_CXGBE);
1356 #endif
1357 #ifdef DEV_NETMAP
1358 	free(sc->sge.nm_rxq, M_CXGBE);
1359 	free(sc->sge.nm_txq, M_CXGBE);
1360 #endif
1361 	free(sc->irq, M_CXGBE);
1362 	free(sc->sge.rxq, M_CXGBE);
1363 	free(sc->sge.txq, M_CXGBE);
1364 	free(sc->sge.ctrlq, M_CXGBE);
1365 	free(sc->sge.iqmap, M_CXGBE);
1366 	free(sc->sge.eqmap, M_CXGBE);
1367 	free(sc->tids.ftid_tab, M_CXGBE);
1368 	t4_destroy_dma_tag(sc);
1369 	if (mtx_initialized(&sc->sc_lock)) {
1370 		sx_xlock(&t4_list_lock);
1371 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1372 		sx_xunlock(&t4_list_lock);
1373 		mtx_destroy(&sc->sc_lock);
1374 	}
1375 
1376 	callout_drain(&sc->sfl_callout);
1377 	if (mtx_initialized(&sc->tids.ftid_lock))
1378 		mtx_destroy(&sc->tids.ftid_lock);
1379 	if (mtx_initialized(&sc->sfl_lock))
1380 		mtx_destroy(&sc->sfl_lock);
1381 	if (mtx_initialized(&sc->ifp_lock))
1382 		mtx_destroy(&sc->ifp_lock);
1383 	if (mtx_initialized(&sc->reg_lock))
1384 		mtx_destroy(&sc->reg_lock);
1385 
1386 	for (i = 0; i < NUM_MEMWIN; i++) {
1387 		struct memwin *mw = &sc->memwin[i];
1388 
1389 		if (rw_initialized(&mw->mw_lock))
1390 			rw_destroy(&mw->mw_lock);
1391 	}
1392 
1393 	bzero(sc, sizeof(*sc));
1394 
1395 	return (0);
1396 }
1397 
1398 static int
1399 cxgbe_probe(device_t dev)
1400 {
1401 	char buf[128];
1402 	struct port_info *pi = device_get_softc(dev);
1403 
1404 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1405 	device_set_desc_copy(dev, buf);
1406 
1407 	return (BUS_PROBE_DEFAULT);
1408 }
1409 
1410 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1411     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1412     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1413 #define T4_CAP_ENABLE (T4_CAP)
1414 
1415 static int
1416 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1417 {
1418 	struct ifnet *ifp;
1419 	struct sbuf *sb;
1420 
1421 	vi->xact_addr_filt = -1;
1422 	callout_init(&vi->tick, 1);
1423 
1424 	/* Allocate an ifnet and set it up */
1425 	ifp = if_alloc(IFT_ETHER);
1426 	if (ifp == NULL) {
1427 		device_printf(dev, "Cannot allocate ifnet\n");
1428 		return (ENOMEM);
1429 	}
1430 	vi->ifp = ifp;
1431 	ifp->if_softc = vi;
1432 
1433 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1434 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1435 
1436 	ifp->if_init = cxgbe_init;
1437 	ifp->if_ioctl = cxgbe_ioctl;
1438 	ifp->if_transmit = cxgbe_transmit;
1439 	ifp->if_qflush = cxgbe_qflush;
1440 	ifp->if_get_counter = cxgbe_get_counter;
1441 
1442 	ifp->if_capabilities = T4_CAP;
1443 #ifdef TCP_OFFLOAD
1444 	if (vi->nofldrxq != 0)
1445 		ifp->if_capabilities |= IFCAP_TOE;
1446 #endif
1447 #ifdef DEV_NETMAP
1448 	if (vi->nnmrxq != 0)
1449 		ifp->if_capabilities |= IFCAP_NETMAP;
1450 #endif
1451 	ifp->if_capenable = T4_CAP_ENABLE;
1452 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1453 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1454 
1455 	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1456 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1457 	ifp->if_hw_tsomaxsegsize = 65536;
1458 
1459 	/* Initialize ifmedia for this VI */
1460 	ifmedia_init(&vi->media, IFM_IMASK, cxgbe_media_change,
1461 	    cxgbe_media_status);
1462 	build_medialist(vi->pi, &vi->media);
1463 
1464 	vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1465 	    EVENTHANDLER_PRI_ANY);
1466 
1467 	ether_ifattach(ifp, vi->hw_addr);
1468 #ifdef DEV_NETMAP
1469 	if (ifp->if_capabilities & IFCAP_NETMAP)
1470 		cxgbe_nm_attach(vi);
1471 #endif
1472 	sb = sbuf_new_auto();
1473 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1474 #ifdef TCP_OFFLOAD
1475 	if (ifp->if_capabilities & IFCAP_TOE)
1476 		sbuf_printf(sb, "; %d txq, %d rxq (TOE)",
1477 		    vi->nofldtxq, vi->nofldrxq);
1478 #endif
1479 #ifdef DEV_NETMAP
1480 	if (ifp->if_capabilities & IFCAP_NETMAP)
1481 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1482 		    vi->nnmtxq, vi->nnmrxq);
1483 #endif
1484 	sbuf_finish(sb);
1485 	device_printf(dev, "%s\n", sbuf_data(sb));
1486 	sbuf_delete(sb);
1487 
1488 	vi_sysctls(vi);
1489 
1490 	return (0);
1491 }
1492 
1493 static int
1494 cxgbe_attach(device_t dev)
1495 {
1496 	struct port_info *pi = device_get_softc(dev);
1497 	struct adapter *sc = pi->adapter;
1498 	struct vi_info *vi;
1499 	int i, rc;
1500 
1501 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1502 
1503 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1504 	if (rc)
1505 		return (rc);
1506 
1507 	for_each_vi(pi, i, vi) {
1508 		if (i == 0)
1509 			continue;
1510 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1511 		if (vi->dev == NULL) {
1512 			device_printf(dev, "failed to add VI %d\n", i);
1513 			continue;
1514 		}
1515 		device_set_softc(vi->dev, vi);
1516 	}
1517 
1518 	cxgbe_sysctls(pi);
1519 
1520 	bus_generic_attach(dev);
1521 
1522 	return (0);
1523 }
1524 
1525 static void
1526 cxgbe_vi_detach(struct vi_info *vi)
1527 {
1528 	struct ifnet *ifp = vi->ifp;
1529 
1530 	ether_ifdetach(ifp);
1531 
1532 	if (vi->vlan_c)
1533 		EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1534 
1535 	/* Let detach proceed even if these fail. */
1536 #ifdef DEV_NETMAP
1537 	if (ifp->if_capabilities & IFCAP_NETMAP)
1538 		cxgbe_nm_detach(vi);
1539 #endif
1540 	cxgbe_uninit_synchronized(vi);
1541 	callout_drain(&vi->tick);
1542 	vi_full_uninit(vi);
1543 
1544 	ifmedia_removeall(&vi->media);
1545 	if_free(vi->ifp);
1546 	vi->ifp = NULL;
1547 }
1548 
1549 static int
1550 cxgbe_detach(device_t dev)
1551 {
1552 	struct port_info *pi = device_get_softc(dev);
1553 	struct adapter *sc = pi->adapter;
1554 	int rc;
1555 
1556 	/* Detach the extra VIs first. */
1557 	rc = bus_generic_detach(dev);
1558 	if (rc)
1559 		return (rc);
1560 	device_delete_children(dev);
1561 
1562 	doom_vi(sc, &pi->vi[0]);
1563 
1564 	if (pi->flags & HAS_TRACEQ) {
1565 		sc->traceq = -1;	/* cloner should not create ifnet */
1566 		t4_tracer_port_detach(sc);
1567 	}
1568 
1569 	cxgbe_vi_detach(&pi->vi[0]);
1570 	callout_drain(&pi->tick);
1571 
1572 	end_synchronized_op(sc, 0);
1573 
1574 	return (0);
1575 }
1576 
1577 static void
1578 cxgbe_init(void *arg)
1579 {
1580 	struct vi_info *vi = arg;
1581 	struct adapter *sc = vi->pi->adapter;
1582 
1583 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1584 		return;
1585 	cxgbe_init_synchronized(vi);
1586 	end_synchronized_op(sc, 0);
1587 }
1588 
1589 static int
1590 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1591 {
1592 	int rc = 0, mtu, flags, can_sleep;
1593 	struct vi_info *vi = ifp->if_softc;
1594 	struct adapter *sc = vi->pi->adapter;
1595 	struct ifreq *ifr = (struct ifreq *)data;
1596 	uint32_t mask;
1597 
1598 	switch (cmd) {
1599 	case SIOCSIFMTU:
1600 		mtu = ifr->ifr_mtu;
1601 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1602 			return (EINVAL);
1603 
1604 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1605 		if (rc)
1606 			return (rc);
1607 		ifp->if_mtu = mtu;
1608 		if (vi->flags & VI_INIT_DONE) {
1609 			t4_update_fl_bufsize(ifp);
1610 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1611 				rc = update_mac_settings(ifp, XGMAC_MTU);
1612 		}
1613 		end_synchronized_op(sc, 0);
1614 		break;
1615 
1616 	case SIOCSIFFLAGS:
1617 		can_sleep = 0;
1618 redo_sifflags:
1619 		rc = begin_synchronized_op(sc, vi,
1620 		    can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1621 		if (rc)
1622 			return (rc);
1623 
1624 		if (ifp->if_flags & IFF_UP) {
1625 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1626 				flags = vi->if_flags;
1627 				if ((ifp->if_flags ^ flags) &
1628 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1629 					if (can_sleep == 1) {
1630 						end_synchronized_op(sc, 0);
1631 						can_sleep = 0;
1632 						goto redo_sifflags;
1633 					}
1634 					rc = update_mac_settings(ifp,
1635 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1636 				}
1637 			} else {
1638 				if (can_sleep == 0) {
1639 					end_synchronized_op(sc, LOCK_HELD);
1640 					can_sleep = 1;
1641 					goto redo_sifflags;
1642 				}
1643 				rc = cxgbe_init_synchronized(vi);
1644 			}
1645 			vi->if_flags = ifp->if_flags;
1646 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1647 			if (can_sleep == 0) {
1648 				end_synchronized_op(sc, LOCK_HELD);
1649 				can_sleep = 1;
1650 				goto redo_sifflags;
1651 			}
1652 			rc = cxgbe_uninit_synchronized(vi);
1653 		}
1654 		end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1655 		break;
1656 
1657 	case SIOCADDMULTI:
1658 	case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1659 		rc = begin_synchronized_op(sc, vi, HOLD_LOCK, "t4multi");
1660 		if (rc)
1661 			return (rc);
1662 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1663 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1664 		end_synchronized_op(sc, LOCK_HELD);
1665 		break;
1666 
1667 	case SIOCSIFCAP:
1668 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1669 		if (rc)
1670 			return (rc);
1671 
1672 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1673 		if (mask & IFCAP_TXCSUM) {
1674 			ifp->if_capenable ^= IFCAP_TXCSUM;
1675 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1676 
1677 			if (IFCAP_TSO4 & ifp->if_capenable &&
1678 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1679 				ifp->if_capenable &= ~IFCAP_TSO4;
1680 				if_printf(ifp,
1681 				    "tso4 disabled due to -txcsum.\n");
1682 			}
1683 		}
1684 		if (mask & IFCAP_TXCSUM_IPV6) {
1685 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1686 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1687 
1688 			if (IFCAP_TSO6 & ifp->if_capenable &&
1689 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1690 				ifp->if_capenable &= ~IFCAP_TSO6;
1691 				if_printf(ifp,
1692 				    "tso6 disabled due to -txcsum6.\n");
1693 			}
1694 		}
1695 		if (mask & IFCAP_RXCSUM)
1696 			ifp->if_capenable ^= IFCAP_RXCSUM;
1697 		if (mask & IFCAP_RXCSUM_IPV6)
1698 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1699 
1700 		/*
1701 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1702 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1703 		 * sending a TSO request our way, so it's sufficient to toggle
1704 		 * IFCAP_TSOx only.
1705 		 */
1706 		if (mask & IFCAP_TSO4) {
1707 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1708 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1709 				if_printf(ifp, "enable txcsum first.\n");
1710 				rc = EAGAIN;
1711 				goto fail;
1712 			}
1713 			ifp->if_capenable ^= IFCAP_TSO4;
1714 		}
1715 		if (mask & IFCAP_TSO6) {
1716 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1717 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1718 				if_printf(ifp, "enable txcsum6 first.\n");
1719 				rc = EAGAIN;
1720 				goto fail;
1721 			}
1722 			ifp->if_capenable ^= IFCAP_TSO6;
1723 		}
1724 		if (mask & IFCAP_LRO) {
1725 #if defined(INET) || defined(INET6)
1726 			int i;
1727 			struct sge_rxq *rxq;
1728 
1729 			ifp->if_capenable ^= IFCAP_LRO;
1730 			for_each_rxq(vi, i, rxq) {
1731 				if (ifp->if_capenable & IFCAP_LRO)
1732 					rxq->iq.flags |= IQ_LRO_ENABLED;
1733 				else
1734 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1735 			}
1736 #endif
1737 		}
1738 #ifdef TCP_OFFLOAD
1739 		if (mask & IFCAP_TOE) {
1740 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1741 
1742 			rc = toe_capability(vi, enable);
1743 			if (rc != 0)
1744 				goto fail;
1745 
1746 			ifp->if_capenable ^= mask;
1747 		}
1748 #endif
1749 		if (mask & IFCAP_VLAN_HWTAGGING) {
1750 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1751 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1752 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1753 		}
1754 		if (mask & IFCAP_VLAN_MTU) {
1755 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1756 
1757 			/* Need to find out how to disable auto-mtu-inflation */
1758 		}
1759 		if (mask & IFCAP_VLAN_HWTSO)
1760 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1761 		if (mask & IFCAP_VLAN_HWCSUM)
1762 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1763 
1764 #ifdef VLAN_CAPABILITIES
1765 		VLAN_CAPABILITIES(ifp);
1766 #endif
1767 fail:
1768 		end_synchronized_op(sc, 0);
1769 		break;
1770 
1771 	case SIOCSIFMEDIA:
1772 	case SIOCGIFMEDIA:
1773 	case SIOCGIFXMEDIA:
1774 		ifmedia_ioctl(ifp, ifr, &vi->media, cmd);
1775 		break;
1776 
1777 	case SIOCGI2C: {
1778 		struct ifi2creq i2c;
1779 
1780 		rc = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
1781 		if (rc != 0)
1782 			break;
1783 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1784 			rc = EPERM;
1785 			break;
1786 		}
1787 		if (i2c.len > sizeof(i2c.data)) {
1788 			rc = EINVAL;
1789 			break;
1790 		}
1791 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1792 		if (rc)
1793 			return (rc);
1794 		rc = -t4_i2c_rd(sc, sc->mbox, vi->pi->port_id, i2c.dev_addr,
1795 		    i2c.offset, i2c.len, &i2c.data[0]);
1796 		end_synchronized_op(sc, 0);
1797 		if (rc == 0)
1798 			rc = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
1799 		break;
1800 	}
1801 
1802 	default:
1803 		rc = ether_ioctl(ifp, cmd, data);
1804 	}
1805 
1806 	return (rc);
1807 }
1808 
1809 static int
1810 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1811 {
1812 	struct vi_info *vi = ifp->if_softc;
1813 	struct port_info *pi = vi->pi;
1814 	struct adapter *sc = pi->adapter;
1815 	struct sge_txq *txq;
1816 	void *items[1];
1817 	int rc;
1818 
1819 	M_ASSERTPKTHDR(m);
1820 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1821 
1822 	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1823 		m_freem(m);
1824 		return (ENETDOWN);
1825 	}
1826 
1827 	rc = parse_pkt(sc, &m);
1828 	if (__predict_false(rc != 0)) {
1829 		MPASS(m == NULL);			/* was freed already */
1830 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1831 		return (rc);
1832 	}
1833 
1834 	/* Select a txq. */
1835 	txq = &sc->sge.txq[vi->first_txq];
1836 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1837 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1838 		    vi->rsrv_noflowq);
1839 
1840 	items[0] = m;
1841 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1842 	if (__predict_false(rc != 0))
1843 		m_freem(m);
1844 
1845 	return (rc);
1846 }
1847 
1848 static void
1849 cxgbe_qflush(struct ifnet *ifp)
1850 {
1851 	struct vi_info *vi = ifp->if_softc;
1852 	struct sge_txq *txq;
1853 	int i;
1854 
1855 	/* queues do not exist if !VI_INIT_DONE. */
1856 	if (vi->flags & VI_INIT_DONE) {
1857 		for_each_txq(vi, i, txq) {
1858 			TXQ_LOCK(txq);
1859 			txq->eq.flags &= ~EQ_ENABLED;
1860 			TXQ_UNLOCK(txq);
1861 			while (!mp_ring_is_idle(txq->r)) {
1862 				mp_ring_check_drainage(txq->r, 0);
1863 				pause("qflush", 1);
1864 			}
1865 		}
1866 	}
1867 	if_qflush(ifp);
1868 }
1869 
1870 static uint64_t
1871 vi_get_counter(struct ifnet *ifp, ift_counter c)
1872 {
1873 	struct vi_info *vi = ifp->if_softc;
1874 	struct fw_vi_stats_vf *s = &vi->stats;
1875 
1876 	vi_refresh_stats(vi->pi->adapter, vi);
1877 
1878 	switch (c) {
1879 	case IFCOUNTER_IPACKETS:
1880 		return (s->rx_bcast_frames + s->rx_mcast_frames +
1881 		    s->rx_ucast_frames);
1882 	case IFCOUNTER_IERRORS:
1883 		return (s->rx_err_frames);
1884 	case IFCOUNTER_OPACKETS:
1885 		return (s->tx_bcast_frames + s->tx_mcast_frames +
1886 		    s->tx_ucast_frames + s->tx_offload_frames);
1887 	case IFCOUNTER_OERRORS:
1888 		return (s->tx_drop_frames);
1889 	case IFCOUNTER_IBYTES:
1890 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
1891 		    s->rx_ucast_bytes);
1892 	case IFCOUNTER_OBYTES:
1893 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
1894 		    s->tx_ucast_bytes + s->tx_offload_bytes);
1895 	case IFCOUNTER_IMCASTS:
1896 		return (s->rx_mcast_frames);
1897 	case IFCOUNTER_OMCASTS:
1898 		return (s->tx_mcast_frames);
1899 	case IFCOUNTER_OQDROPS: {
1900 		uint64_t drops;
1901 
1902 		drops = 0;
1903 		if (vi->flags & VI_INIT_DONE) {
1904 			int i;
1905 			struct sge_txq *txq;
1906 
1907 			for_each_txq(vi, i, txq)
1908 				drops += counter_u64_fetch(txq->r->drops);
1909 		}
1910 
1911 		return (drops);
1912 
1913 	}
1914 
1915 	default:
1916 		return (if_get_counter_default(ifp, c));
1917 	}
1918 }
1919 
1920 uint64_t
1921 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
1922 {
1923 	struct vi_info *vi = ifp->if_softc;
1924 	struct port_info *pi = vi->pi;
1925 	struct adapter *sc = pi->adapter;
1926 	struct port_stats *s = &pi->stats;
1927 
1928 	if (pi->nvi > 1 || sc->flags & IS_VF)
1929 		return (vi_get_counter(ifp, c));
1930 
1931 	cxgbe_refresh_stats(sc, pi);
1932 
1933 	switch (c) {
1934 	case IFCOUNTER_IPACKETS:
1935 		return (s->rx_frames);
1936 
1937 	case IFCOUNTER_IERRORS:
1938 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
1939 		    s->rx_fcs_err + s->rx_len_err);
1940 
1941 	case IFCOUNTER_OPACKETS:
1942 		return (s->tx_frames);
1943 
1944 	case IFCOUNTER_OERRORS:
1945 		return (s->tx_error_frames);
1946 
1947 	case IFCOUNTER_IBYTES:
1948 		return (s->rx_octets);
1949 
1950 	case IFCOUNTER_OBYTES:
1951 		return (s->tx_octets);
1952 
1953 	case IFCOUNTER_IMCASTS:
1954 		return (s->rx_mcast_frames);
1955 
1956 	case IFCOUNTER_OMCASTS:
1957 		return (s->tx_mcast_frames);
1958 
1959 	case IFCOUNTER_IQDROPS:
1960 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
1961 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
1962 		    s->rx_trunc3 + pi->tnl_cong_drops);
1963 
1964 	case IFCOUNTER_OQDROPS: {
1965 		uint64_t drops;
1966 
1967 		drops = s->tx_drop;
1968 		if (vi->flags & VI_INIT_DONE) {
1969 			int i;
1970 			struct sge_txq *txq;
1971 
1972 			for_each_txq(vi, i, txq)
1973 				drops += counter_u64_fetch(txq->r->drops);
1974 		}
1975 
1976 		return (drops);
1977 
1978 	}
1979 
1980 	default:
1981 		return (if_get_counter_default(ifp, c));
1982 	}
1983 }
1984 
1985 static int
1986 cxgbe_media_change(struct ifnet *ifp)
1987 {
1988 	struct vi_info *vi = ifp->if_softc;
1989 
1990 	device_printf(vi->dev, "%s unimplemented.\n", __func__);
1991 
1992 	return (EOPNOTSUPP);
1993 }
1994 
1995 static void
1996 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1997 {
1998 	struct vi_info *vi = ifp->if_softc;
1999 	struct port_info *pi = vi->pi;
2000 	struct ifmedia_entry *cur;
2001 	int speed = pi->link_cfg.speed;
2002 
2003 	cur = vi->media.ifm_cur;
2004 
2005 	ifmr->ifm_status = IFM_AVALID;
2006 	if (!pi->link_cfg.link_ok)
2007 		return;
2008 
2009 	ifmr->ifm_status |= IFM_ACTIVE;
2010 
2011 	/* active and current will differ iff current media is autoselect. */
2012 	if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
2013 		return;
2014 
2015 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2016 	if (speed == 10000)
2017 		ifmr->ifm_active |= IFM_10G_T;
2018 	else if (speed == 1000)
2019 		ifmr->ifm_active |= IFM_1000_T;
2020 	else if (speed == 100)
2021 		ifmr->ifm_active |= IFM_100_TX;
2022 	else if (speed == 10)
2023 		ifmr->ifm_active |= IFM_10_T;
2024 	else
2025 		KASSERT(0, ("%s: link up but speed unknown (%u)", __func__,
2026 			    speed));
2027 }
2028 
2029 static int
2030 vcxgbe_probe(device_t dev)
2031 {
2032 	char buf[128];
2033 	struct vi_info *vi = device_get_softc(dev);
2034 
2035 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2036 	    vi - vi->pi->vi);
2037 	device_set_desc_copy(dev, buf);
2038 
2039 	return (BUS_PROBE_DEFAULT);
2040 }
2041 
2042 static int
2043 vcxgbe_attach(device_t dev)
2044 {
2045 	struct vi_info *vi;
2046 	struct port_info *pi;
2047 	struct adapter *sc;
2048 	int func, index, rc;
2049 	u32 param, val;
2050 
2051 	vi = device_get_softc(dev);
2052 	pi = vi->pi;
2053 	sc = pi->adapter;
2054 
2055 	index = vi - pi->vi;
2056 	KASSERT(index < nitems(vi_mac_funcs),
2057 	    ("%s: VI %s doesn't have a MAC func", __func__,
2058 	    device_get_nameunit(dev)));
2059 	func = vi_mac_funcs[index];
2060 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2061 	    vi->hw_addr, &vi->rss_size, func, 0);
2062 	if (rc < 0) {
2063 		device_printf(dev, "Failed to allocate virtual interface "
2064 		    "for port %d: %d\n", pi->port_id, -rc);
2065 		return (-rc);
2066 	}
2067 	vi->viid = rc;
2068 	if (chip_id(sc) <= CHELSIO_T5)
2069 		vi->smt_idx = (rc & 0x7f) << 1;
2070 	else
2071 		vi->smt_idx = (rc & 0x7f);
2072 
2073 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2074 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2075 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2076 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2077 	if (rc)
2078 		vi->rss_base = 0xffff;
2079 	else {
2080 		/* MPASS((val >> 16) == rss_size); */
2081 		vi->rss_base = val & 0xffff;
2082 	}
2083 
2084 	rc = cxgbe_vi_attach(dev, vi);
2085 	if (rc) {
2086 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2087 		return (rc);
2088 	}
2089 	return (0);
2090 }
2091 
2092 static int
2093 vcxgbe_detach(device_t dev)
2094 {
2095 	struct vi_info *vi;
2096 	struct adapter *sc;
2097 
2098 	vi = device_get_softc(dev);
2099 	sc = vi->pi->adapter;
2100 
2101 	doom_vi(sc, vi);
2102 
2103 	cxgbe_vi_detach(vi);
2104 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2105 
2106 	end_synchronized_op(sc, 0);
2107 
2108 	return (0);
2109 }
2110 
2111 void
2112 t4_fatal_err(struct adapter *sc)
2113 {
2114 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2115 	t4_intr_disable(sc);
2116 	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2117 	    device_get_nameunit(sc->dev));
2118 }
2119 
2120 void
2121 t4_add_adapter(struct adapter *sc)
2122 {
2123 	sx_xlock(&t4_list_lock);
2124 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2125 	sx_xunlock(&t4_list_lock);
2126 }
2127 
2128 int
2129 t4_map_bars_0_and_4(struct adapter *sc)
2130 {
2131 	sc->regs_rid = PCIR_BAR(0);
2132 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2133 	    &sc->regs_rid, RF_ACTIVE);
2134 	if (sc->regs_res == NULL) {
2135 		device_printf(sc->dev, "cannot map registers.\n");
2136 		return (ENXIO);
2137 	}
2138 	sc->bt = rman_get_bustag(sc->regs_res);
2139 	sc->bh = rman_get_bushandle(sc->regs_res);
2140 	sc->mmio_len = rman_get_size(sc->regs_res);
2141 	setbit(&sc->doorbells, DOORBELL_KDB);
2142 
2143 	sc->msix_rid = PCIR_BAR(4);
2144 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2145 	    &sc->msix_rid, RF_ACTIVE);
2146 	if (sc->msix_res == NULL) {
2147 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2148 		return (ENXIO);
2149 	}
2150 
2151 	return (0);
2152 }
2153 
2154 int
2155 t4_map_bar_2(struct adapter *sc)
2156 {
2157 
2158 	/*
2159 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2160 	 * to map it if RDMA is disabled.
2161 	 */
2162 	if (is_t4(sc) && sc->rdmacaps == 0)
2163 		return (0);
2164 
2165 	sc->udbs_rid = PCIR_BAR(2);
2166 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2167 	    &sc->udbs_rid, RF_ACTIVE);
2168 	if (sc->udbs_res == NULL) {
2169 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2170 		return (ENXIO);
2171 	}
2172 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2173 
2174 	if (chip_id(sc) >= CHELSIO_T5) {
2175 		setbit(&sc->doorbells, DOORBELL_UDB);
2176 #if defined(__i386__) || defined(__amd64__)
2177 		if (t5_write_combine) {
2178 			int rc, mode;
2179 
2180 			/*
2181 			 * Enable write combining on BAR2.  This is the
2182 			 * userspace doorbell BAR and is split into 128B
2183 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2184 			 * with an egress queue.  The first 64B has the doorbell
2185 			 * and the second 64B can be used to submit a tx work
2186 			 * request with an implicit doorbell.
2187 			 */
2188 
2189 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2190 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2191 			if (rc == 0) {
2192 				clrbit(&sc->doorbells, DOORBELL_UDB);
2193 				setbit(&sc->doorbells, DOORBELL_WCWR);
2194 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2195 			} else {
2196 				device_printf(sc->dev,
2197 				    "couldn't enable write combining: %d\n",
2198 				    rc);
2199 			}
2200 
2201 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2202 			t4_write_reg(sc, A_SGE_STAT_CFG,
2203 			    V_STATSOURCE_T5(7) | mode);
2204 		}
2205 #endif
2206 	}
2207 
2208 	return (0);
2209 }
2210 
2211 struct memwin_init {
2212 	uint32_t base;
2213 	uint32_t aperture;
2214 };
2215 
2216 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2217 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2218 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2219 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2220 };
2221 
2222 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2223 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2224 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2225 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2226 };
2227 
2228 static void
2229 setup_memwin(struct adapter *sc)
2230 {
2231 	const struct memwin_init *mw_init;
2232 	struct memwin *mw;
2233 	int i;
2234 	uint32_t bar0;
2235 
2236 	if (is_t4(sc)) {
2237 		/*
2238 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2239 		 * mechanism.  Works from within PCI passthrough environments
2240 		 * too, where rman_get_start() can return a different value.  We
2241 		 * need to program the T4 memory window decoders with the actual
2242 		 * addresses that will be coming across the PCIe link.
2243 		 */
2244 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2245 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2246 
2247 		mw_init = &t4_memwin[0];
2248 	} else {
2249 		/* T5+ use the relative offset inside the PCIe BAR */
2250 		bar0 = 0;
2251 
2252 		mw_init = &t5_memwin[0];
2253 	}
2254 
2255 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2256 		rw_init(&mw->mw_lock, "memory window access");
2257 		mw->mw_base = mw_init->base;
2258 		mw->mw_aperture = mw_init->aperture;
2259 		mw->mw_curpos = 0;
2260 		t4_write_reg(sc,
2261 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2262 		    (mw->mw_base + bar0) | V_BIR(0) |
2263 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2264 		rw_wlock(&mw->mw_lock);
2265 		position_memwin(sc, i, 0);
2266 		rw_wunlock(&mw->mw_lock);
2267 	}
2268 
2269 	/* flush */
2270 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2271 }
2272 
2273 /*
2274  * Positions the memory window at the given address in the card's address space.
2275  * There are some alignment requirements and the actual position may be at an
2276  * address prior to the requested address.  mw->mw_curpos always has the actual
2277  * position of the window.
2278  */
2279 static void
2280 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2281 {
2282 	struct memwin *mw;
2283 	uint32_t pf;
2284 	uint32_t reg;
2285 
2286 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2287 	mw = &sc->memwin[idx];
2288 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2289 
2290 	if (is_t4(sc)) {
2291 		pf = 0;
2292 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2293 	} else {
2294 		pf = V_PFNUM(sc->pf);
2295 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2296 	}
2297 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2298 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2299 	t4_read_reg(sc, reg);	/* flush */
2300 }
2301 
2302 static int
2303 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2304     int len, int rw)
2305 {
2306 	struct memwin *mw;
2307 	uint32_t mw_end, v;
2308 
2309 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2310 
2311 	/* Memory can only be accessed in naturally aligned 4 byte units */
2312 	if (addr & 3 || len & 3 || len <= 0)
2313 		return (EINVAL);
2314 
2315 	mw = &sc->memwin[idx];
2316 	while (len > 0) {
2317 		rw_rlock(&mw->mw_lock);
2318 		mw_end = mw->mw_curpos + mw->mw_aperture;
2319 		if (addr >= mw_end || addr < mw->mw_curpos) {
2320 			/* Will need to reposition the window */
2321 			if (!rw_try_upgrade(&mw->mw_lock)) {
2322 				rw_runlock(&mw->mw_lock);
2323 				rw_wlock(&mw->mw_lock);
2324 			}
2325 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2326 			position_memwin(sc, idx, addr);
2327 			rw_downgrade(&mw->mw_lock);
2328 			mw_end = mw->mw_curpos + mw->mw_aperture;
2329 		}
2330 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2331 		while (addr < mw_end && len > 0) {
2332 			if (rw == 0) {
2333 				v = t4_read_reg(sc, mw->mw_base + addr -
2334 				    mw->mw_curpos);
2335 				*val++ = le32toh(v);
2336 			} else {
2337 				v = *val++;
2338 				t4_write_reg(sc, mw->mw_base + addr -
2339 				    mw->mw_curpos, htole32(v));
2340 			}
2341 			addr += 4;
2342 			len -= 4;
2343 		}
2344 		rw_runlock(&mw->mw_lock);
2345 	}
2346 
2347 	return (0);
2348 }
2349 
2350 static inline int
2351 read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2352     int len)
2353 {
2354 
2355 	return (rw_via_memwin(sc, idx, addr, val, len, 0));
2356 }
2357 
2358 static inline int
2359 write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
2360     const uint32_t *val, int len)
2361 {
2362 
2363 	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
2364 }
2365 
2366 static int
2367 t4_range_cmp(const void *a, const void *b)
2368 {
2369 	return ((const struct t4_range *)a)->start -
2370 	       ((const struct t4_range *)b)->start;
2371 }
2372 
2373 /*
2374  * Verify that the memory range specified by the addr/len pair is valid within
2375  * the card's address space.
2376  */
2377 static int
2378 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2379 {
2380 	struct t4_range mem_ranges[4], *r, *next;
2381 	uint32_t em, addr_len;
2382 	int i, n, remaining;
2383 
2384 	/* Memory can only be accessed in naturally aligned 4 byte units */
2385 	if (addr & 3 || len & 3 || len <= 0)
2386 		return (EINVAL);
2387 
2388 	/* Enabled memories */
2389 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2390 
2391 	r = &mem_ranges[0];
2392 	n = 0;
2393 	bzero(r, sizeof(mem_ranges));
2394 	if (em & F_EDRAM0_ENABLE) {
2395 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2396 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2397 		if (r->size > 0) {
2398 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2399 			if (addr >= r->start &&
2400 			    addr + len <= r->start + r->size)
2401 				return (0);
2402 			r++;
2403 			n++;
2404 		}
2405 	}
2406 	if (em & F_EDRAM1_ENABLE) {
2407 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2408 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2409 		if (r->size > 0) {
2410 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2411 			if (addr >= r->start &&
2412 			    addr + len <= r->start + r->size)
2413 				return (0);
2414 			r++;
2415 			n++;
2416 		}
2417 	}
2418 	if (em & F_EXT_MEM_ENABLE) {
2419 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2420 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2421 		if (r->size > 0) {
2422 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2423 			if (addr >= r->start &&
2424 			    addr + len <= r->start + r->size)
2425 				return (0);
2426 			r++;
2427 			n++;
2428 		}
2429 	}
2430 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2431 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2432 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2433 		if (r->size > 0) {
2434 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2435 			if (addr >= r->start &&
2436 			    addr + len <= r->start + r->size)
2437 				return (0);
2438 			r++;
2439 			n++;
2440 		}
2441 	}
2442 	MPASS(n <= nitems(mem_ranges));
2443 
2444 	if (n > 1) {
2445 		/* Sort and merge the ranges. */
2446 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2447 
2448 		/* Start from index 0 and examine the next n - 1 entries. */
2449 		r = &mem_ranges[0];
2450 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2451 
2452 			MPASS(r->size > 0);	/* r is a valid entry. */
2453 			next = r + 1;
2454 			MPASS(next->size > 0);	/* and so is the next one. */
2455 
2456 			while (r->start + r->size >= next->start) {
2457 				/* Merge the next one into the current entry. */
2458 				r->size = max(r->start + r->size,
2459 				    next->start + next->size) - r->start;
2460 				n--;	/* One fewer entry in total. */
2461 				if (--remaining == 0)
2462 					goto done;	/* short circuit */
2463 				next++;
2464 			}
2465 			if (next != r + 1) {
2466 				/*
2467 				 * Some entries were merged into r and next
2468 				 * points to the first valid entry that couldn't
2469 				 * be merged.
2470 				 */
2471 				MPASS(next->size > 0);	/* must be valid */
2472 				memcpy(r + 1, next, remaining * sizeof(*r));
2473 #ifdef INVARIANTS
2474 				/*
2475 				 * This so that the foo->size assertion in the
2476 				 * next iteration of the loop do the right
2477 				 * thing for entries that were pulled up and are
2478 				 * no longer valid.
2479 				 */
2480 				MPASS(n < nitems(mem_ranges));
2481 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2482 				    sizeof(struct t4_range));
2483 #endif
2484 			}
2485 		}
2486 done:
2487 		/* Done merging the ranges. */
2488 		MPASS(n > 0);
2489 		r = &mem_ranges[0];
2490 		for (i = 0; i < n; i++, r++) {
2491 			if (addr >= r->start &&
2492 			    addr + len <= r->start + r->size)
2493 				return (0);
2494 		}
2495 	}
2496 
2497 	return (EFAULT);
2498 }
2499 
2500 static int
2501 fwmtype_to_hwmtype(int mtype)
2502 {
2503 
2504 	switch (mtype) {
2505 	case FW_MEMTYPE_EDC0:
2506 		return (MEM_EDC0);
2507 	case FW_MEMTYPE_EDC1:
2508 		return (MEM_EDC1);
2509 	case FW_MEMTYPE_EXTMEM:
2510 		return (MEM_MC0);
2511 	case FW_MEMTYPE_EXTMEM1:
2512 		return (MEM_MC1);
2513 	default:
2514 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2515 	}
2516 }
2517 
2518 /*
2519  * Verify that the memory range specified by the memtype/offset/len pair is
2520  * valid and lies entirely within the memtype specified.  The global address of
2521  * the start of the range is returned in addr.
2522  */
2523 static int
2524 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2525     uint32_t *addr)
2526 {
2527 	uint32_t em, addr_len, maddr;
2528 
2529 	/* Memory can only be accessed in naturally aligned 4 byte units */
2530 	if (off & 3 || len & 3 || len == 0)
2531 		return (EINVAL);
2532 
2533 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2534 	switch (fwmtype_to_hwmtype(mtype)) {
2535 	case MEM_EDC0:
2536 		if (!(em & F_EDRAM0_ENABLE))
2537 			return (EINVAL);
2538 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2539 		maddr = G_EDRAM0_BASE(addr_len) << 20;
2540 		break;
2541 	case MEM_EDC1:
2542 		if (!(em & F_EDRAM1_ENABLE))
2543 			return (EINVAL);
2544 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2545 		maddr = G_EDRAM1_BASE(addr_len) << 20;
2546 		break;
2547 	case MEM_MC:
2548 		if (!(em & F_EXT_MEM_ENABLE))
2549 			return (EINVAL);
2550 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2551 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2552 		break;
2553 	case MEM_MC1:
2554 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2555 			return (EINVAL);
2556 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2557 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2558 		break;
2559 	default:
2560 		return (EINVAL);
2561 	}
2562 
2563 	*addr = maddr + off;	/* global address */
2564 	return (validate_mem_range(sc, *addr, len));
2565 }
2566 
2567 static int
2568 fixup_devlog_params(struct adapter *sc)
2569 {
2570 	struct devlog_params *dparams = &sc->params.devlog;
2571 	int rc;
2572 
2573 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2574 	    dparams->size, &dparams->addr);
2575 
2576 	return (rc);
2577 }
2578 
2579 static int
2580 cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g, int num_vis,
2581     struct intrs_and_queues *iaq)
2582 {
2583 	int rc, itype, navail, nrxq10g, nrxq1g, n;
2584 	int nofldrxq10g = 0, nofldrxq1g = 0;
2585 
2586 	bzero(iaq, sizeof(*iaq));
2587 
2588 	iaq->ntxq10g = t4_ntxq10g;
2589 	iaq->ntxq1g = t4_ntxq1g;
2590 	iaq->ntxq_vi = t4_ntxq_vi;
2591 	iaq->nrxq10g = nrxq10g = t4_nrxq10g;
2592 	iaq->nrxq1g = nrxq1g = t4_nrxq1g;
2593 	iaq->nrxq_vi = t4_nrxq_vi;
2594 	iaq->rsrv_noflowq = t4_rsrv_noflowq;
2595 #ifdef TCP_OFFLOAD
2596 	if (is_offload(sc)) {
2597 		iaq->nofldtxq10g = t4_nofldtxq10g;
2598 		iaq->nofldtxq1g = t4_nofldtxq1g;
2599 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
2600 		iaq->nofldrxq10g = nofldrxq10g = t4_nofldrxq10g;
2601 		iaq->nofldrxq1g = nofldrxq1g = t4_nofldrxq1g;
2602 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
2603 	}
2604 #endif
2605 #ifdef DEV_NETMAP
2606 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
2607 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
2608 #endif
2609 
2610 	for (itype = INTR_MSIX; itype; itype >>= 1) {
2611 
2612 		if ((itype & t4_intr_types) == 0)
2613 			continue;	/* not allowed */
2614 
2615 		if (itype == INTR_MSIX)
2616 			navail = pci_msix_count(sc->dev);
2617 		else if (itype == INTR_MSI)
2618 			navail = pci_msi_count(sc->dev);
2619 		else
2620 			navail = 1;
2621 restart:
2622 		if (navail == 0)
2623 			continue;
2624 
2625 		iaq->intr_type = itype;
2626 		iaq->intr_flags_10g = 0;
2627 		iaq->intr_flags_1g = 0;
2628 
2629 		/*
2630 		 * Best option: an interrupt vector for errors, one for the
2631 		 * firmware event queue, and one for every rxq (NIC and TOE) of
2632 		 * every VI.  The VIs that support netmap use the same
2633 		 * interrupts for the NIC rx queues and the netmap rx queues
2634 		 * because only one set of queues is active at a time.
2635 		 */
2636 		iaq->nirq = T4_EXTRA_INTR;
2637 		iaq->nirq += n10g * (nrxq10g + nofldrxq10g);
2638 		iaq->nirq += n1g * (nrxq1g + nofldrxq1g);
2639 		iaq->nirq += (n10g + n1g) * (num_vis - 1) *
2640 		    max(iaq->nrxq_vi, iaq->nnmrxq_vi);	/* See comment above. */
2641 		iaq->nirq += (n10g + n1g) * (num_vis - 1) * iaq->nofldrxq_vi;
2642 		if (iaq->nirq <= navail &&
2643 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2644 			iaq->intr_flags_10g = INTR_ALL;
2645 			iaq->intr_flags_1g = INTR_ALL;
2646 			goto allocate;
2647 		}
2648 
2649 		/* Disable the VIs (and netmap) if there aren't enough intrs */
2650 		if (num_vis > 1) {
2651 			device_printf(sc->dev, "virtual interfaces disabled "
2652 			    "because num_vis=%u with current settings "
2653 			    "(nrxq10g=%u, nrxq1g=%u, nofldrxq10g=%u, "
2654 			    "nofldrxq1g=%u, nrxq_vi=%u nofldrxq_vi=%u, "
2655 			    "nnmrxq_vi=%u) would need %u interrupts but "
2656 			    "only %u are available.\n", num_vis, nrxq10g,
2657 			    nrxq1g, nofldrxq10g, nofldrxq1g, iaq->nrxq_vi,
2658 			    iaq->nofldrxq_vi, iaq->nnmrxq_vi, iaq->nirq,
2659 			    navail);
2660 			num_vis = 1;
2661 			iaq->ntxq_vi = iaq->nrxq_vi = 0;
2662 			iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
2663 			iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
2664 			goto restart;
2665 		}
2666 
2667 		/*
2668 		 * Second best option: a vector for errors, one for the firmware
2669 		 * event queue, and vectors for either all the NIC rx queues or
2670 		 * all the TOE rx queues.  The queues that don't get vectors
2671 		 * will forward their interrupts to those that do.
2672 		 */
2673 		iaq->nirq = T4_EXTRA_INTR;
2674 		if (nrxq10g >= nofldrxq10g) {
2675 			iaq->intr_flags_10g = INTR_RXQ;
2676 			iaq->nirq += n10g * nrxq10g;
2677 		} else {
2678 			iaq->intr_flags_10g = INTR_OFLD_RXQ;
2679 			iaq->nirq += n10g * nofldrxq10g;
2680 		}
2681 		if (nrxq1g >= nofldrxq1g) {
2682 			iaq->intr_flags_1g = INTR_RXQ;
2683 			iaq->nirq += n1g * nrxq1g;
2684 		} else {
2685 			iaq->intr_flags_1g = INTR_OFLD_RXQ;
2686 			iaq->nirq += n1g * nofldrxq1g;
2687 		}
2688 		if (iaq->nirq <= navail &&
2689 		    (itype != INTR_MSI || powerof2(iaq->nirq)))
2690 			goto allocate;
2691 
2692 		/*
2693 		 * Next best option: an interrupt vector for errors, one for the
2694 		 * firmware event queue, and at least one per main-VI.  At this
2695 		 * point we know we'll have to downsize nrxq and/or nofldrxq to
2696 		 * fit what's available to us.
2697 		 */
2698 		iaq->nirq = T4_EXTRA_INTR;
2699 		iaq->nirq += n10g + n1g;
2700 		if (iaq->nirq <= navail) {
2701 			int leftover = navail - iaq->nirq;
2702 
2703 			if (n10g > 0) {
2704 				int target = max(nrxq10g, nofldrxq10g);
2705 
2706 				iaq->intr_flags_10g = nrxq10g >= nofldrxq10g ?
2707 				    INTR_RXQ : INTR_OFLD_RXQ;
2708 
2709 				n = 1;
2710 				while (n < target && leftover >= n10g) {
2711 					leftover -= n10g;
2712 					iaq->nirq += n10g;
2713 					n++;
2714 				}
2715 				iaq->nrxq10g = min(n, nrxq10g);
2716 #ifdef TCP_OFFLOAD
2717 				iaq->nofldrxq10g = min(n, nofldrxq10g);
2718 #endif
2719 			}
2720 
2721 			if (n1g > 0) {
2722 				int target = max(nrxq1g, nofldrxq1g);
2723 
2724 				iaq->intr_flags_1g = nrxq1g >= nofldrxq1g ?
2725 				    INTR_RXQ : INTR_OFLD_RXQ;
2726 
2727 				n = 1;
2728 				while (n < target && leftover >= n1g) {
2729 					leftover -= n1g;
2730 					iaq->nirq += n1g;
2731 					n++;
2732 				}
2733 				iaq->nrxq1g = min(n, nrxq1g);
2734 #ifdef TCP_OFFLOAD
2735 				iaq->nofldrxq1g = min(n, nofldrxq1g);
2736 #endif
2737 			}
2738 
2739 			if (itype != INTR_MSI || powerof2(iaq->nirq))
2740 				goto allocate;
2741 		}
2742 
2743 		/*
2744 		 * Least desirable option: one interrupt vector for everything.
2745 		 */
2746 		iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2747 		iaq->intr_flags_10g = iaq->intr_flags_1g = 0;
2748 #ifdef TCP_OFFLOAD
2749 		if (is_offload(sc))
2750 			iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2751 #endif
2752 allocate:
2753 		navail = iaq->nirq;
2754 		rc = 0;
2755 		if (itype == INTR_MSIX)
2756 			rc = pci_alloc_msix(sc->dev, &navail);
2757 		else if (itype == INTR_MSI)
2758 			rc = pci_alloc_msi(sc->dev, &navail);
2759 
2760 		if (rc == 0) {
2761 			if (navail == iaq->nirq)
2762 				return (0);
2763 
2764 			/*
2765 			 * Didn't get the number requested.  Use whatever number
2766 			 * the kernel is willing to allocate (it's in navail).
2767 			 */
2768 			device_printf(sc->dev, "fewer vectors than requested, "
2769 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2770 			    itype, iaq->nirq, navail);
2771 			pci_release_msi(sc->dev);
2772 			goto restart;
2773 		}
2774 
2775 		device_printf(sc->dev,
2776 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2777 		    itype, rc, iaq->nirq, navail);
2778 	}
2779 
2780 	device_printf(sc->dev,
2781 	    "failed to find a usable interrupt type.  "
2782 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2783 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2784 
2785 	return (ENXIO);
2786 }
2787 
2788 #define FW_VERSION(chip) ( \
2789     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2790     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2791     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2792     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2793 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2794 
2795 struct fw_info {
2796 	uint8_t chip;
2797 	char *kld_name;
2798 	char *fw_mod_name;
2799 	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
2800 } fw_info[] = {
2801 	{
2802 		.chip = CHELSIO_T4,
2803 		.kld_name = "t4fw_cfg",
2804 		.fw_mod_name = "t4fw",
2805 		.fw_hdr = {
2806 			.chip = FW_HDR_CHIP_T4,
2807 			.fw_ver = htobe32_const(FW_VERSION(T4)),
2808 			.intfver_nic = FW_INTFVER(T4, NIC),
2809 			.intfver_vnic = FW_INTFVER(T4, VNIC),
2810 			.intfver_ofld = FW_INTFVER(T4, OFLD),
2811 			.intfver_ri = FW_INTFVER(T4, RI),
2812 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2813 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
2814 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2815 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
2816 		},
2817 	}, {
2818 		.chip = CHELSIO_T5,
2819 		.kld_name = "t5fw_cfg",
2820 		.fw_mod_name = "t5fw",
2821 		.fw_hdr = {
2822 			.chip = FW_HDR_CHIP_T5,
2823 			.fw_ver = htobe32_const(FW_VERSION(T5)),
2824 			.intfver_nic = FW_INTFVER(T5, NIC),
2825 			.intfver_vnic = FW_INTFVER(T5, VNIC),
2826 			.intfver_ofld = FW_INTFVER(T5, OFLD),
2827 			.intfver_ri = FW_INTFVER(T5, RI),
2828 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
2829 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
2830 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
2831 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
2832 		},
2833 	}, {
2834 		.chip = CHELSIO_T6,
2835 		.kld_name = "t6fw_cfg",
2836 		.fw_mod_name = "t6fw",
2837 		.fw_hdr = {
2838 			.chip = FW_HDR_CHIP_T6,
2839 			.fw_ver = htobe32_const(FW_VERSION(T6)),
2840 			.intfver_nic = FW_INTFVER(T6, NIC),
2841 			.intfver_vnic = FW_INTFVER(T6, VNIC),
2842 			.intfver_ofld = FW_INTFVER(T6, OFLD),
2843 			.intfver_ri = FW_INTFVER(T6, RI),
2844 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
2845 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
2846 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
2847 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
2848 		},
2849 	}
2850 };
2851 
2852 static struct fw_info *
2853 find_fw_info(int chip)
2854 {
2855 	int i;
2856 
2857 	for (i = 0; i < nitems(fw_info); i++) {
2858 		if (fw_info[i].chip == chip)
2859 			return (&fw_info[i]);
2860 	}
2861 	return (NULL);
2862 }
2863 
2864 /*
2865  * Is the given firmware API compatible with the one the driver was compiled
2866  * with?
2867  */
2868 static int
2869 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
2870 {
2871 
2872 	/* short circuit if it's the exact same firmware version */
2873 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
2874 		return (1);
2875 
2876 	/*
2877 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
2878 	 * features that are supported in the driver.
2879 	 */
2880 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
2881 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
2882 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
2883 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
2884 		return (1);
2885 #undef SAME_INTF
2886 
2887 	return (0);
2888 }
2889 
2890 /*
2891  * The firmware in the KLD is usable, but should it be installed?  This routine
2892  * explains itself in detail if it indicates the KLD firmware should be
2893  * installed.
2894  */
2895 static int
2896 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
2897 {
2898 	const char *reason;
2899 
2900 	if (!card_fw_usable) {
2901 		reason = "incompatible or unusable";
2902 		goto install;
2903 	}
2904 
2905 	if (k > c) {
2906 		reason = "older than the version bundled with this driver";
2907 		goto install;
2908 	}
2909 
2910 	if (t4_fw_install == 2 && k != c) {
2911 		reason = "different than the version bundled with this driver";
2912 		goto install;
2913 	}
2914 
2915 	return (0);
2916 
2917 install:
2918 	if (t4_fw_install == 0) {
2919 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2920 		    "but the driver is prohibited from installing a different "
2921 		    "firmware on the card.\n",
2922 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2923 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
2924 
2925 		return (0);
2926 	}
2927 
2928 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2929 	    "installing firmware %u.%u.%u.%u on card.\n",
2930 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2931 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
2932 	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2933 	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2934 
2935 	return (1);
2936 }
2937 /*
2938  * Establish contact with the firmware and determine if we are the master driver
2939  * or not, and whether we are responsible for chip initialization.
2940  */
2941 static int
2942 prep_firmware(struct adapter *sc)
2943 {
2944 	const struct firmware *fw = NULL, *default_cfg;
2945 	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
2946 	enum dev_state state;
2947 	struct fw_info *fw_info;
2948 	struct fw_hdr *card_fw;		/* fw on the card */
2949 	const struct fw_hdr *kld_fw;	/* fw in the KLD */
2950 	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
2951 					   against */
2952 
2953 	/* Contact firmware. */
2954 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
2955 	if (rc < 0 || state == DEV_STATE_ERR) {
2956 		rc = -rc;
2957 		device_printf(sc->dev,
2958 		    "failed to connect to the firmware: %d, %d.\n", rc, state);
2959 		return (rc);
2960 	}
2961 	pf = rc;
2962 	if (pf == sc->mbox)
2963 		sc->flags |= MASTER_PF;
2964 	else if (state == DEV_STATE_UNINIT) {
2965 		/*
2966 		 * We didn't get to be the master so we definitely won't be
2967 		 * configuring the chip.  It's a bug if someone else hasn't
2968 		 * configured it already.
2969 		 */
2970 		device_printf(sc->dev, "couldn't be master(%d), "
2971 		    "device not already initialized either(%d).\n", rc, state);
2972 		return (EDOOFUS);
2973 	}
2974 
2975 	/* This is the firmware whose headers the driver was compiled against */
2976 	fw_info = find_fw_info(chip_id(sc));
2977 	if (fw_info == NULL) {
2978 		device_printf(sc->dev,
2979 		    "unable to look up firmware information for chip %d.\n",
2980 		    chip_id(sc));
2981 		return (EINVAL);
2982 	}
2983 	drv_fw = &fw_info->fw_hdr;
2984 
2985 	/*
2986 	 * The firmware KLD contains many modules.  The KLD name is also the
2987 	 * name of the module that contains the default config file.
2988 	 */
2989 	default_cfg = firmware_get(fw_info->kld_name);
2990 
2991 	/* Read the header of the firmware on the card */
2992 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
2993 	rc = -t4_read_flash(sc, FLASH_FW_START,
2994 	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
2995 	if (rc == 0)
2996 		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
2997 	else {
2998 		device_printf(sc->dev,
2999 		    "Unable to read card's firmware header: %d\n", rc);
3000 		card_fw_usable = 0;
3001 	}
3002 
3003 	/* This is the firmware in the KLD */
3004 	fw = firmware_get(fw_info->fw_mod_name);
3005 	if (fw != NULL) {
3006 		kld_fw = (const void *)fw->data;
3007 		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3008 	} else {
3009 		kld_fw = NULL;
3010 		kld_fw_usable = 0;
3011 	}
3012 
3013 	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3014 	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3015 		/*
3016 		 * Common case: the firmware on the card is an exact match and
3017 		 * the KLD is an exact match too, or the KLD is
3018 		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
3019 		 * here -- use cxgbetool loadfw if you want to reinstall the
3020 		 * same firmware as the one on the card.
3021 		 */
3022 	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3023 	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3024 	    be32toh(card_fw->fw_ver))) {
3025 
3026 		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3027 		if (rc != 0) {
3028 			device_printf(sc->dev,
3029 			    "failed to install firmware: %d\n", rc);
3030 			goto done;
3031 		}
3032 
3033 		/* Installed successfully, update the cached header too. */
3034 		memcpy(card_fw, kld_fw, sizeof(*card_fw));
3035 		card_fw_usable = 1;
3036 		need_fw_reset = 0;	/* already reset as part of load_fw */
3037 	}
3038 
3039 	if (!card_fw_usable) {
3040 		uint32_t d, c, k;
3041 
3042 		d = ntohl(drv_fw->fw_ver);
3043 		c = ntohl(card_fw->fw_ver);
3044 		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3045 
3046 		device_printf(sc->dev, "Cannot find a usable firmware: "
3047 		    "fw_install %d, chip state %d, "
3048 		    "driver compiled with %d.%d.%d.%d, "
3049 		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3050 		    t4_fw_install, state,
3051 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3052 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3053 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3054 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3055 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3056 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3057 		rc = EINVAL;
3058 		goto done;
3059 	}
3060 
3061 	/* Reset device */
3062 	if (need_fw_reset &&
3063 	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3064 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3065 		if (rc != ETIMEDOUT && rc != EIO)
3066 			t4_fw_bye(sc, sc->mbox);
3067 		goto done;
3068 	}
3069 	sc->flags |= FW_OK;
3070 
3071 	rc = get_params__pre_init(sc);
3072 	if (rc != 0)
3073 		goto done; /* error message displayed already */
3074 
3075 	/* Partition adapter resources as specified in the config file. */
3076 	if (state == DEV_STATE_UNINIT) {
3077 
3078 		KASSERT(sc->flags & MASTER_PF,
3079 		    ("%s: trying to change chip settings when not master.",
3080 		    __func__));
3081 
3082 		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3083 		if (rc != 0)
3084 			goto done;	/* error message displayed already */
3085 
3086 		t4_tweak_chip_settings(sc);
3087 
3088 		/* get basic stuff going */
3089 		rc = -t4_fw_initialize(sc, sc->mbox);
3090 		if (rc != 0) {
3091 			device_printf(sc->dev, "fw init failed: %d.\n", rc);
3092 			goto done;
3093 		}
3094 	} else {
3095 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3096 		sc->cfcsum = 0;
3097 	}
3098 
3099 done:
3100 	free(card_fw, M_CXGBE);
3101 	if (fw != NULL)
3102 		firmware_put(fw, FIRMWARE_UNLOAD);
3103 	if (default_cfg != NULL)
3104 		firmware_put(default_cfg, FIRMWARE_UNLOAD);
3105 
3106 	return (rc);
3107 }
3108 
3109 #define FW_PARAM_DEV(param) \
3110 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3111 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3112 #define FW_PARAM_PFVF(param) \
3113 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3114 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3115 
3116 /*
3117  * Partition chip resources for use between various PFs, VFs, etc.
3118  */
3119 static int
3120 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3121     const char *name_prefix)
3122 {
3123 	const struct firmware *cfg = NULL;
3124 	int rc = 0;
3125 	struct fw_caps_config_cmd caps;
3126 	uint32_t mtype, moff, finicsum, cfcsum;
3127 
3128 	/*
3129 	 * Figure out what configuration file to use.  Pick the default config
3130 	 * file for the card if the user hasn't specified one explicitly.
3131 	 */
3132 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3133 	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3134 		/* Card specific overrides go here. */
3135 		if (pci_get_device(sc->dev) == 0x440a)
3136 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3137 		if (is_fpga(sc))
3138 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3139 	}
3140 
3141 	/*
3142 	 * We need to load another module if the profile is anything except
3143 	 * "default" or "flash".
3144 	 */
3145 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3146 	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3147 		char s[32];
3148 
3149 		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3150 		cfg = firmware_get(s);
3151 		if (cfg == NULL) {
3152 			if (default_cfg != NULL) {
3153 				device_printf(sc->dev,
3154 				    "unable to load module \"%s\" for "
3155 				    "configuration profile \"%s\", will use "
3156 				    "the default config file instead.\n",
3157 				    s, sc->cfg_file);
3158 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3159 				    "%s", DEFAULT_CF);
3160 			} else {
3161 				device_printf(sc->dev,
3162 				    "unable to load module \"%s\" for "
3163 				    "configuration profile \"%s\", will use "
3164 				    "the config file on the card's flash "
3165 				    "instead.\n", s, sc->cfg_file);
3166 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3167 				    "%s", FLASH_CF);
3168 			}
3169 		}
3170 	}
3171 
3172 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3173 	    default_cfg == NULL) {
3174 		device_printf(sc->dev,
3175 		    "default config file not available, will use the config "
3176 		    "file on the card's flash instead.\n");
3177 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3178 	}
3179 
3180 	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3181 		u_int cflen;
3182 		const uint32_t *cfdata;
3183 		uint32_t param, val, addr;
3184 
3185 		KASSERT(cfg != NULL || default_cfg != NULL,
3186 		    ("%s: no config to upload", __func__));
3187 
3188 		/*
3189 		 * Ask the firmware where it wants us to upload the config file.
3190 		 */
3191 		param = FW_PARAM_DEV(CF);
3192 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3193 		if (rc != 0) {
3194 			/* No support for config file?  Shouldn't happen. */
3195 			device_printf(sc->dev,
3196 			    "failed to query config file location: %d.\n", rc);
3197 			goto done;
3198 		}
3199 		mtype = G_FW_PARAMS_PARAM_Y(val);
3200 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3201 
3202 		/*
3203 		 * XXX: sheer laziness.  We deliberately added 4 bytes of
3204 		 * useless stuffing/comments at the end of the config file so
3205 		 * it's ok to simply throw away the last remaining bytes when
3206 		 * the config file is not an exact multiple of 4.  This also
3207 		 * helps with the validate_mt_off_len check.
3208 		 */
3209 		if (cfg != NULL) {
3210 			cflen = cfg->datasize & ~3;
3211 			cfdata = cfg->data;
3212 		} else {
3213 			cflen = default_cfg->datasize & ~3;
3214 			cfdata = default_cfg->data;
3215 		}
3216 
3217 		if (cflen > FLASH_CFG_MAX_SIZE) {
3218 			device_printf(sc->dev,
3219 			    "config file too long (%d, max allowed is %d).  "
3220 			    "Will try to use the config on the card, if any.\n",
3221 			    cflen, FLASH_CFG_MAX_SIZE);
3222 			goto use_config_on_flash;
3223 		}
3224 
3225 		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3226 		if (rc != 0) {
3227 			device_printf(sc->dev,
3228 			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3229 			    "Will try to use the config on the card, if any.\n",
3230 			    __func__, mtype, moff, cflen, rc);
3231 			goto use_config_on_flash;
3232 		}
3233 		write_via_memwin(sc, 2, addr, cfdata, cflen);
3234 	} else {
3235 use_config_on_flash:
3236 		mtype = FW_MEMTYPE_FLASH;
3237 		moff = t4_flash_cfg_addr(sc);
3238 	}
3239 
3240 	bzero(&caps, sizeof(caps));
3241 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3242 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3243 	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3244 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3245 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3246 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3247 	if (rc != 0) {
3248 		device_printf(sc->dev,
3249 		    "failed to pre-process config file: %d "
3250 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3251 		goto done;
3252 	}
3253 
3254 	finicsum = be32toh(caps.finicsum);
3255 	cfcsum = be32toh(caps.cfcsum);
3256 	if (finicsum != cfcsum) {
3257 		device_printf(sc->dev,
3258 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3259 		    finicsum, cfcsum);
3260 	}
3261 	sc->cfcsum = cfcsum;
3262 
3263 #define LIMIT_CAPS(x) do { \
3264 	caps.x &= htobe16(t4_##x##_allowed); \
3265 } while (0)
3266 
3267 	/*
3268 	 * Let the firmware know what features will (not) be used so it can tune
3269 	 * things accordingly.
3270 	 */
3271 	LIMIT_CAPS(nbmcaps);
3272 	LIMIT_CAPS(linkcaps);
3273 	LIMIT_CAPS(switchcaps);
3274 	LIMIT_CAPS(niccaps);
3275 	LIMIT_CAPS(toecaps);
3276 	LIMIT_CAPS(rdmacaps);
3277 	LIMIT_CAPS(cryptocaps);
3278 	LIMIT_CAPS(iscsicaps);
3279 	LIMIT_CAPS(fcoecaps);
3280 #undef LIMIT_CAPS
3281 
3282 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3283 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3284 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3285 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3286 	if (rc != 0) {
3287 		device_printf(sc->dev,
3288 		    "failed to process config file: %d.\n", rc);
3289 	}
3290 done:
3291 	if (cfg != NULL)
3292 		firmware_put(cfg, FIRMWARE_UNLOAD);
3293 	return (rc);
3294 }
3295 
3296 /*
3297  * Retrieve parameters that are needed (or nice to have) very early.
3298  */
3299 static int
3300 get_params__pre_init(struct adapter *sc)
3301 {
3302 	int rc;
3303 	uint32_t param[2], val[2];
3304 
3305 	t4_get_version_info(sc);
3306 
3307 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3308 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3309 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3310 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3311 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3312 
3313 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3314 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3315 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3316 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3317 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3318 
3319 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3320 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3321 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3322 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3323 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3324 
3325 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3326 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3327 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3328 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3329 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3330 
3331 	param[0] = FW_PARAM_DEV(PORTVEC);
3332 	param[1] = FW_PARAM_DEV(CCLK);
3333 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3334 	if (rc != 0) {
3335 		device_printf(sc->dev,
3336 		    "failed to query parameters (pre_init): %d.\n", rc);
3337 		return (rc);
3338 	}
3339 
3340 	sc->params.portvec = val[0];
3341 	sc->params.nports = bitcount32(val[0]);
3342 	sc->params.vpd.cclk = val[1];
3343 
3344 	/* Read device log parameters. */
3345 	rc = -t4_init_devlog_params(sc, 1);
3346 	if (rc == 0)
3347 		fixup_devlog_params(sc);
3348 	else {
3349 		device_printf(sc->dev,
3350 		    "failed to get devlog parameters: %d.\n", rc);
3351 		rc = 0;	/* devlog isn't critical for device operation */
3352 	}
3353 
3354 	return (rc);
3355 }
3356 
3357 /*
3358  * Retrieve various parameters that are of interest to the driver.  The device
3359  * has been initialized by the firmware at this point.
3360  */
3361 static int
3362 get_params__post_init(struct adapter *sc)
3363 {
3364 	int rc;
3365 	uint32_t param[7], val[7];
3366 	struct fw_caps_config_cmd caps;
3367 
3368 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3369 	param[1] = FW_PARAM_PFVF(EQ_START);
3370 	param[2] = FW_PARAM_PFVF(FILTER_START);
3371 	param[3] = FW_PARAM_PFVF(FILTER_END);
3372 	param[4] = FW_PARAM_PFVF(L2T_START);
3373 	param[5] = FW_PARAM_PFVF(L2T_END);
3374 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3375 	if (rc != 0) {
3376 		device_printf(sc->dev,
3377 		    "failed to query parameters (post_init): %d.\n", rc);
3378 		return (rc);
3379 	}
3380 
3381 	sc->sge.iq_start = val[0];
3382 	sc->sge.eq_start = val[1];
3383 	sc->tids.ftid_base = val[2];
3384 	sc->tids.nftids = val[3] - val[2] + 1;
3385 	sc->params.ftid_min = val[2];
3386 	sc->params.ftid_max = val[3];
3387 	sc->vres.l2t.start = val[4];
3388 	sc->vres.l2t.size = val[5] - val[4] + 1;
3389 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3390 	    ("%s: L2 table size (%u) larger than expected (%u)",
3391 	    __func__, sc->vres.l2t.size, L2T_SIZE));
3392 
3393 	/* get capabilites */
3394 	bzero(&caps, sizeof(caps));
3395 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3396 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3397 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3398 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3399 	if (rc != 0) {
3400 		device_printf(sc->dev,
3401 		    "failed to get card capabilities: %d.\n", rc);
3402 		return (rc);
3403 	}
3404 
3405 #define READ_CAPS(x) do { \
3406 	sc->x = htobe16(caps.x); \
3407 } while (0)
3408 	READ_CAPS(nbmcaps);
3409 	READ_CAPS(linkcaps);
3410 	READ_CAPS(switchcaps);
3411 	READ_CAPS(niccaps);
3412 	READ_CAPS(toecaps);
3413 	READ_CAPS(rdmacaps);
3414 	READ_CAPS(cryptocaps);
3415 	READ_CAPS(iscsicaps);
3416 	READ_CAPS(fcoecaps);
3417 
3418 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3419 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3420 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3421 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3422 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3423 		if (rc != 0) {
3424 			device_printf(sc->dev,
3425 			    "failed to query NIC parameters: %d.\n", rc);
3426 			return (rc);
3427 		}
3428 		sc->tids.etid_base = val[0];
3429 		sc->params.etid_min = val[0];
3430 		sc->tids.netids = val[1] - val[0] + 1;
3431 		sc->params.netids = sc->tids.netids;
3432 		sc->params.eo_wr_cred = val[2];
3433 		sc->params.ethoffload = 1;
3434 	}
3435 
3436 	if (sc->toecaps) {
3437 		/* query offload-related parameters */
3438 		param[0] = FW_PARAM_DEV(NTID);
3439 		param[1] = FW_PARAM_PFVF(SERVER_START);
3440 		param[2] = FW_PARAM_PFVF(SERVER_END);
3441 		param[3] = FW_PARAM_PFVF(TDDP_START);
3442 		param[4] = FW_PARAM_PFVF(TDDP_END);
3443 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3444 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3445 		if (rc != 0) {
3446 			device_printf(sc->dev,
3447 			    "failed to query TOE parameters: %d.\n", rc);
3448 			return (rc);
3449 		}
3450 		sc->tids.ntids = val[0];
3451 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3452 		sc->tids.stid_base = val[1];
3453 		sc->tids.nstids = val[2] - val[1] + 1;
3454 		sc->vres.ddp.start = val[3];
3455 		sc->vres.ddp.size = val[4] - val[3] + 1;
3456 		sc->params.ofldq_wr_cred = val[5];
3457 		sc->params.offload = 1;
3458 	}
3459 	if (sc->rdmacaps) {
3460 		param[0] = FW_PARAM_PFVF(STAG_START);
3461 		param[1] = FW_PARAM_PFVF(STAG_END);
3462 		param[2] = FW_PARAM_PFVF(RQ_START);
3463 		param[3] = FW_PARAM_PFVF(RQ_END);
3464 		param[4] = FW_PARAM_PFVF(PBL_START);
3465 		param[5] = FW_PARAM_PFVF(PBL_END);
3466 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3467 		if (rc != 0) {
3468 			device_printf(sc->dev,
3469 			    "failed to query RDMA parameters(1): %d.\n", rc);
3470 			return (rc);
3471 		}
3472 		sc->vres.stag.start = val[0];
3473 		sc->vres.stag.size = val[1] - val[0] + 1;
3474 		sc->vres.rq.start = val[2];
3475 		sc->vres.rq.size = val[3] - val[2] + 1;
3476 		sc->vres.pbl.start = val[4];
3477 		sc->vres.pbl.size = val[5] - val[4] + 1;
3478 
3479 		param[0] = FW_PARAM_PFVF(SQRQ_START);
3480 		param[1] = FW_PARAM_PFVF(SQRQ_END);
3481 		param[2] = FW_PARAM_PFVF(CQ_START);
3482 		param[3] = FW_PARAM_PFVF(CQ_END);
3483 		param[4] = FW_PARAM_PFVF(OCQ_START);
3484 		param[5] = FW_PARAM_PFVF(OCQ_END);
3485 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3486 		if (rc != 0) {
3487 			device_printf(sc->dev,
3488 			    "failed to query RDMA parameters(2): %d.\n", rc);
3489 			return (rc);
3490 		}
3491 		sc->vres.qp.start = val[0];
3492 		sc->vres.qp.size = val[1] - val[0] + 1;
3493 		sc->vres.cq.start = val[2];
3494 		sc->vres.cq.size = val[3] - val[2] + 1;
3495 		sc->vres.ocq.start = val[4];
3496 		sc->vres.ocq.size = val[5] - val[4] + 1;
3497 	}
3498 	if (sc->iscsicaps) {
3499 		param[0] = FW_PARAM_PFVF(ISCSI_START);
3500 		param[1] = FW_PARAM_PFVF(ISCSI_END);
3501 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3502 		if (rc != 0) {
3503 			device_printf(sc->dev,
3504 			    "failed to query iSCSI parameters: %d.\n", rc);
3505 			return (rc);
3506 		}
3507 		sc->vres.iscsi.start = val[0];
3508 		sc->vres.iscsi.size = val[1] - val[0] + 1;
3509 	}
3510 
3511 	t4_init_sge_params(sc);
3512 
3513 	/*
3514 	 * We've got the params we wanted to query via the firmware.  Now grab
3515 	 * some others directly from the chip.
3516 	 */
3517 	rc = t4_read_chip_settings(sc);
3518 
3519 	return (rc);
3520 }
3521 
3522 static int
3523 set_params__post_init(struct adapter *sc)
3524 {
3525 	uint32_t param, val;
3526 
3527 	/* ask for encapsulated CPLs */
3528 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3529 	val = 1;
3530 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3531 
3532 	return (0);
3533 }
3534 
3535 #undef FW_PARAM_PFVF
3536 #undef FW_PARAM_DEV
3537 
3538 static void
3539 t4_set_desc(struct adapter *sc)
3540 {
3541 	char buf[128];
3542 	struct adapter_params *p = &sc->params;
3543 
3544 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
3545 
3546 	device_set_desc_copy(sc->dev, buf);
3547 }
3548 
3549 static void
3550 build_medialist(struct port_info *pi, struct ifmedia *media)
3551 {
3552 	int m;
3553 
3554 	PORT_LOCK(pi);
3555 
3556 	ifmedia_removeall(media);
3557 
3558 	m = IFM_ETHER | IFM_FDX;
3559 
3560 	switch(pi->port_type) {
3561 	case FW_PORT_TYPE_BT_XFI:
3562 	case FW_PORT_TYPE_BT_XAUI:
3563 		ifmedia_add(media, m | IFM_10G_T, 0, NULL);
3564 		/* fall through */
3565 
3566 	case FW_PORT_TYPE_BT_SGMII:
3567 		ifmedia_add(media, m | IFM_1000_T, 0, NULL);
3568 		ifmedia_add(media, m | IFM_100_TX, 0, NULL);
3569 		ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL);
3570 		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
3571 		break;
3572 
3573 	case FW_PORT_TYPE_CX4:
3574 		ifmedia_add(media, m | IFM_10G_CX4, 0, NULL);
3575 		ifmedia_set(media, m | IFM_10G_CX4);
3576 		break;
3577 
3578 	case FW_PORT_TYPE_QSFP_10G:
3579 	case FW_PORT_TYPE_SFP:
3580 	case FW_PORT_TYPE_FIBER_XFI:
3581 	case FW_PORT_TYPE_FIBER_XAUI:
3582 		switch (pi->mod_type) {
3583 
3584 		case FW_PORT_MOD_TYPE_LR:
3585 			ifmedia_add(media, m | IFM_10G_LR, 0, NULL);
3586 			ifmedia_set(media, m | IFM_10G_LR);
3587 			break;
3588 
3589 		case FW_PORT_MOD_TYPE_SR:
3590 			ifmedia_add(media, m | IFM_10G_SR, 0, NULL);
3591 			ifmedia_set(media, m | IFM_10G_SR);
3592 			break;
3593 
3594 		case FW_PORT_MOD_TYPE_LRM:
3595 			ifmedia_add(media, m | IFM_10G_LRM, 0, NULL);
3596 			ifmedia_set(media, m | IFM_10G_LRM);
3597 			break;
3598 
3599 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3600 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3601 			ifmedia_add(media, m | IFM_10G_TWINAX, 0, NULL);
3602 			ifmedia_set(media, m | IFM_10G_TWINAX);
3603 			break;
3604 
3605 		case FW_PORT_MOD_TYPE_NONE:
3606 			m &= ~IFM_FDX;
3607 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3608 			ifmedia_set(media, m | IFM_NONE);
3609 			break;
3610 
3611 		case FW_PORT_MOD_TYPE_NA:
3612 		case FW_PORT_MOD_TYPE_ER:
3613 		default:
3614 			device_printf(pi->dev,
3615 			    "unknown port_type (%d), mod_type (%d)\n",
3616 			    pi->port_type, pi->mod_type);
3617 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3618 			ifmedia_set(media, m | IFM_UNKNOWN);
3619 			break;
3620 		}
3621 		break;
3622 
3623 	case FW_PORT_TYPE_CR_QSFP:
3624 	case FW_PORT_TYPE_SFP28:
3625 	case FW_PORT_TYPE_KR_SFP28:
3626 		switch (pi->mod_type) {
3627 
3628 		case FW_PORT_MOD_TYPE_SR:
3629 			ifmedia_add(media, m | IFM_25G_SR, 0, NULL);
3630 			ifmedia_set(media, m | IFM_25G_SR);
3631 			break;
3632 
3633 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3634 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3635 			ifmedia_add(media, m | IFM_25G_CR, 0, NULL);
3636 			ifmedia_set(media, m | IFM_25G_CR);
3637 			break;
3638 
3639 		case FW_PORT_MOD_TYPE_NONE:
3640 			m &= ~IFM_FDX;
3641 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3642 			ifmedia_set(media, m | IFM_NONE);
3643 			break;
3644 
3645 		default:
3646 			device_printf(pi->dev,
3647 			    "unknown port_type (%d), mod_type (%d)\n",
3648 			    pi->port_type, pi->mod_type);
3649 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3650 			ifmedia_set(media, m | IFM_UNKNOWN);
3651 			break;
3652 		}
3653 		break;
3654 
3655 	case FW_PORT_TYPE_QSFP:
3656 		switch (pi->mod_type) {
3657 
3658 		case FW_PORT_MOD_TYPE_LR:
3659 			ifmedia_add(media, m | IFM_40G_LR4, 0, NULL);
3660 			ifmedia_set(media, m | IFM_40G_LR4);
3661 			break;
3662 
3663 		case FW_PORT_MOD_TYPE_SR:
3664 			ifmedia_add(media, m | IFM_40G_SR4, 0, NULL);
3665 			ifmedia_set(media, m | IFM_40G_SR4);
3666 			break;
3667 
3668 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3669 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3670 			ifmedia_add(media, m | IFM_40G_CR4, 0, NULL);
3671 			ifmedia_set(media, m | IFM_40G_CR4);
3672 			break;
3673 
3674 		case FW_PORT_MOD_TYPE_NONE:
3675 			m &= ~IFM_FDX;
3676 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3677 			ifmedia_set(media, m | IFM_NONE);
3678 			break;
3679 
3680 		default:
3681 			device_printf(pi->dev,
3682 			    "unknown port_type (%d), mod_type (%d)\n",
3683 			    pi->port_type, pi->mod_type);
3684 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3685 			ifmedia_set(media, m | IFM_UNKNOWN);
3686 			break;
3687 		}
3688 		break;
3689 
3690 	case FW_PORT_TYPE_KR4_100G:
3691 	case FW_PORT_TYPE_CR4_QSFP:
3692 		switch (pi->mod_type) {
3693 
3694 		case FW_PORT_MOD_TYPE_LR:
3695 			ifmedia_add(media, m | IFM_100G_LR4, 0, NULL);
3696 			ifmedia_set(media, m | IFM_100G_LR4);
3697 			break;
3698 
3699 		case FW_PORT_MOD_TYPE_SR:
3700 			ifmedia_add(media, m | IFM_100G_SR4, 0, NULL);
3701 			ifmedia_set(media, m | IFM_100G_SR4);
3702 			break;
3703 
3704 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3705 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3706 			ifmedia_add(media, m | IFM_100G_CR4, 0, NULL);
3707 			ifmedia_set(media, m | IFM_100G_CR4);
3708 			break;
3709 
3710 		case FW_PORT_MOD_TYPE_NONE:
3711 			m &= ~IFM_FDX;
3712 			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3713 			ifmedia_set(media, m | IFM_NONE);
3714 			break;
3715 
3716 		default:
3717 			device_printf(pi->dev,
3718 			    "unknown port_type (%d), mod_type (%d)\n",
3719 			    pi->port_type, pi->mod_type);
3720 			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3721 			ifmedia_set(media, m | IFM_UNKNOWN);
3722 			break;
3723 		}
3724 		break;
3725 
3726 	default:
3727 		device_printf(pi->dev,
3728 		    "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
3729 		    pi->mod_type);
3730 		ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3731 		ifmedia_set(media, m | IFM_UNKNOWN);
3732 		break;
3733 	}
3734 
3735 	PORT_UNLOCK(pi);
3736 }
3737 
3738 #define FW_MAC_EXACT_CHUNK	7
3739 
3740 /*
3741  * Program the port's XGMAC based on parameters in ifnet.  The caller also
3742  * indicates which parameters should be programmed (the rest are left alone).
3743  */
3744 int
3745 update_mac_settings(struct ifnet *ifp, int flags)
3746 {
3747 	int rc = 0;
3748 	struct vi_info *vi = ifp->if_softc;
3749 	struct port_info *pi = vi->pi;
3750 	struct adapter *sc = pi->adapter;
3751 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
3752 
3753 	ASSERT_SYNCHRONIZED_OP(sc);
3754 	KASSERT(flags, ("%s: not told what to update.", __func__));
3755 
3756 	if (flags & XGMAC_MTU)
3757 		mtu = ifp->if_mtu;
3758 
3759 	if (flags & XGMAC_PROMISC)
3760 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
3761 
3762 	if (flags & XGMAC_ALLMULTI)
3763 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
3764 
3765 	if (flags & XGMAC_VLANEX)
3766 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
3767 
3768 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
3769 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
3770 		    allmulti, 1, vlanex, false);
3771 		if (rc) {
3772 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
3773 			    rc);
3774 			return (rc);
3775 		}
3776 	}
3777 
3778 	if (flags & XGMAC_UCADDR) {
3779 		uint8_t ucaddr[ETHER_ADDR_LEN];
3780 
3781 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
3782 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
3783 		    ucaddr, true, true);
3784 		if (rc < 0) {
3785 			rc = -rc;
3786 			if_printf(ifp, "change_mac failed: %d\n", rc);
3787 			return (rc);
3788 		} else {
3789 			vi->xact_addr_filt = rc;
3790 			rc = 0;
3791 		}
3792 	}
3793 
3794 	if (flags & XGMAC_MCADDRS) {
3795 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
3796 		int del = 1;
3797 		uint64_t hash = 0;
3798 		struct ifmultiaddr *ifma;
3799 		int i = 0, j;
3800 
3801 		if_maddr_rlock(ifp);
3802 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3803 			if (ifma->ifma_addr->sa_family != AF_LINK)
3804 				continue;
3805 			mcaddr[i] =
3806 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
3807 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
3808 			i++;
3809 
3810 			if (i == FW_MAC_EXACT_CHUNK) {
3811 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
3812 				    del, i, mcaddr, NULL, &hash, 0);
3813 				if (rc < 0) {
3814 					rc = -rc;
3815 					for (j = 0; j < i; j++) {
3816 						if_printf(ifp,
3817 						    "failed to add mc address"
3818 						    " %02x:%02x:%02x:"
3819 						    "%02x:%02x:%02x rc=%d\n",
3820 						    mcaddr[j][0], mcaddr[j][1],
3821 						    mcaddr[j][2], mcaddr[j][3],
3822 						    mcaddr[j][4], mcaddr[j][5],
3823 						    rc);
3824 					}
3825 					goto mcfail;
3826 				}
3827 				del = 0;
3828 				i = 0;
3829 			}
3830 		}
3831 		if (i > 0) {
3832 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
3833 			    mcaddr, NULL, &hash, 0);
3834 			if (rc < 0) {
3835 				rc = -rc;
3836 				for (j = 0; j < i; j++) {
3837 					if_printf(ifp,
3838 					    "failed to add mc address"
3839 					    " %02x:%02x:%02x:"
3840 					    "%02x:%02x:%02x rc=%d\n",
3841 					    mcaddr[j][0], mcaddr[j][1],
3842 					    mcaddr[j][2], mcaddr[j][3],
3843 					    mcaddr[j][4], mcaddr[j][5],
3844 					    rc);
3845 				}
3846 				goto mcfail;
3847 			}
3848 		}
3849 
3850 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
3851 		if (rc != 0)
3852 			if_printf(ifp, "failed to set mc address hash: %d", rc);
3853 mcfail:
3854 		if_maddr_runlock(ifp);
3855 	}
3856 
3857 	return (rc);
3858 }
3859 
3860 /*
3861  * {begin|end}_synchronized_op must be called from the same thread.
3862  */
3863 int
3864 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
3865     char *wmesg)
3866 {
3867 	int rc, pri;
3868 
3869 #ifdef WITNESS
3870 	/* the caller thinks it's ok to sleep, but is it really? */
3871 	if (flags & SLEEP_OK)
3872 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
3873 		    "begin_synchronized_op");
3874 #endif
3875 
3876 	if (INTR_OK)
3877 		pri = PCATCH;
3878 	else
3879 		pri = 0;
3880 
3881 	ADAPTER_LOCK(sc);
3882 	for (;;) {
3883 
3884 		if (vi && IS_DOOMED(vi)) {
3885 			rc = ENXIO;
3886 			goto done;
3887 		}
3888 
3889 		if (!IS_BUSY(sc)) {
3890 			rc = 0;
3891 			break;
3892 		}
3893 
3894 		if (!(flags & SLEEP_OK)) {
3895 			rc = EBUSY;
3896 			goto done;
3897 		}
3898 
3899 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
3900 			rc = EINTR;
3901 			goto done;
3902 		}
3903 	}
3904 
3905 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
3906 	SET_BUSY(sc);
3907 #ifdef INVARIANTS
3908 	sc->last_op = wmesg;
3909 	sc->last_op_thr = curthread;
3910 	sc->last_op_flags = flags;
3911 #endif
3912 
3913 done:
3914 	if (!(flags & HOLD_LOCK) || rc)
3915 		ADAPTER_UNLOCK(sc);
3916 
3917 	return (rc);
3918 }
3919 
3920 /*
3921  * Tell if_ioctl and if_init that the VI is going away.  This is
3922  * special variant of begin_synchronized_op and must be paired with a
3923  * call to end_synchronized_op.
3924  */
3925 void
3926 doom_vi(struct adapter *sc, struct vi_info *vi)
3927 {
3928 
3929 	ADAPTER_LOCK(sc);
3930 	SET_DOOMED(vi);
3931 	wakeup(&sc->flags);
3932 	while (IS_BUSY(sc))
3933 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
3934 	SET_BUSY(sc);
3935 #ifdef INVARIANTS
3936 	sc->last_op = "t4detach";
3937 	sc->last_op_thr = curthread;
3938 	sc->last_op_flags = 0;
3939 #endif
3940 	ADAPTER_UNLOCK(sc);
3941 }
3942 
3943 /*
3944  * {begin|end}_synchronized_op must be called from the same thread.
3945  */
3946 void
3947 end_synchronized_op(struct adapter *sc, int flags)
3948 {
3949 
3950 	if (flags & LOCK_HELD)
3951 		ADAPTER_LOCK_ASSERT_OWNED(sc);
3952 	else
3953 		ADAPTER_LOCK(sc);
3954 
3955 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
3956 	CLR_BUSY(sc);
3957 	wakeup(&sc->flags);
3958 	ADAPTER_UNLOCK(sc);
3959 }
3960 
3961 static int
3962 cxgbe_init_synchronized(struct vi_info *vi)
3963 {
3964 	struct port_info *pi = vi->pi;
3965 	struct adapter *sc = pi->adapter;
3966 	struct ifnet *ifp = vi->ifp;
3967 	int rc = 0, i;
3968 	struct sge_txq *txq;
3969 
3970 	ASSERT_SYNCHRONIZED_OP(sc);
3971 
3972 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3973 		return (0);	/* already running */
3974 
3975 	if (!(sc->flags & FULL_INIT_DONE) &&
3976 	    ((rc = adapter_full_init(sc)) != 0))
3977 		return (rc);	/* error message displayed already */
3978 
3979 	if (!(vi->flags & VI_INIT_DONE) &&
3980 	    ((rc = vi_full_init(vi)) != 0))
3981 		return (rc); /* error message displayed already */
3982 
3983 	rc = update_mac_settings(ifp, XGMAC_ALL);
3984 	if (rc)
3985 		goto done;	/* error message displayed already */
3986 
3987 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
3988 	if (rc != 0) {
3989 		if_printf(ifp, "enable_vi failed: %d\n", rc);
3990 		goto done;
3991 	}
3992 
3993 	/*
3994 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
3995 	 * if this changes.
3996 	 */
3997 
3998 	for_each_txq(vi, i, txq) {
3999 		TXQ_LOCK(txq);
4000 		txq->eq.flags |= EQ_ENABLED;
4001 		TXQ_UNLOCK(txq);
4002 	}
4003 
4004 	/*
4005 	 * The first iq of the first port to come up is used for tracing.
4006 	 */
4007 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4008 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4009 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4010 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4011 		    V_QUEUENUMBER(sc->traceq));
4012 		pi->flags |= HAS_TRACEQ;
4013 	}
4014 
4015 	/* all ok */
4016 	PORT_LOCK(pi);
4017 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
4018 	pi->up_vis++;
4019 
4020 	if (pi->nvi > 1 || sc->flags & IS_VF)
4021 		callout_reset(&vi->tick, hz, vi_tick, vi);
4022 	else
4023 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4024 	PORT_UNLOCK(pi);
4025 done:
4026 	if (rc != 0)
4027 		cxgbe_uninit_synchronized(vi);
4028 
4029 	return (rc);
4030 }
4031 
4032 /*
4033  * Idempotent.
4034  */
4035 static int
4036 cxgbe_uninit_synchronized(struct vi_info *vi)
4037 {
4038 	struct port_info *pi = vi->pi;
4039 	struct adapter *sc = pi->adapter;
4040 	struct ifnet *ifp = vi->ifp;
4041 	int rc, i;
4042 	struct sge_txq *txq;
4043 
4044 	ASSERT_SYNCHRONIZED_OP(sc);
4045 
4046 	if (!(vi->flags & VI_INIT_DONE)) {
4047 		KASSERT(!(ifp->if_drv_flags & IFF_DRV_RUNNING),
4048 		    ("uninited VI is running"));
4049 		return (0);
4050 	}
4051 
4052 	/*
4053 	 * Disable the VI so that all its data in either direction is discarded
4054 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4055 	 * tick) intact as the TP can deliver negative advice or data that it's
4056 	 * holding in its RAM (for an offloaded connection) even after the VI is
4057 	 * disabled.
4058 	 */
4059 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4060 	if (rc) {
4061 		if_printf(ifp, "disable_vi failed: %d\n", rc);
4062 		return (rc);
4063 	}
4064 
4065 	for_each_txq(vi, i, txq) {
4066 		TXQ_LOCK(txq);
4067 		txq->eq.flags &= ~EQ_ENABLED;
4068 		TXQ_UNLOCK(txq);
4069 	}
4070 
4071 	PORT_LOCK(pi);
4072 	if (pi->nvi > 1 || sc->flags & IS_VF)
4073 		callout_stop(&vi->tick);
4074 	else
4075 		callout_stop(&pi->tick);
4076 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4077 		PORT_UNLOCK(pi);
4078 		return (0);
4079 	}
4080 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4081 	pi->up_vis--;
4082 	if (pi->up_vis > 0) {
4083 		PORT_UNLOCK(pi);
4084 		return (0);
4085 	}
4086 	PORT_UNLOCK(pi);
4087 
4088 	pi->link_cfg.link_ok = 0;
4089 	pi->link_cfg.speed = 0;
4090 	pi->link_cfg.link_down_rc = 255;
4091 	t4_os_link_changed(sc, pi->port_id, 0);
4092 
4093 	return (0);
4094 }
4095 
4096 /*
4097  * It is ok for this function to fail midway and return right away.  t4_detach
4098  * will walk the entire sc->irq list and clean up whatever is valid.
4099  */
4100 int
4101 t4_setup_intr_handlers(struct adapter *sc)
4102 {
4103 	int rc, rid, p, q, v;
4104 	char s[8];
4105 	struct irq *irq;
4106 	struct port_info *pi;
4107 	struct vi_info *vi;
4108 	struct sge *sge = &sc->sge;
4109 	struct sge_rxq *rxq;
4110 #ifdef TCP_OFFLOAD
4111 	struct sge_ofld_rxq *ofld_rxq;
4112 #endif
4113 #ifdef DEV_NETMAP
4114 	struct sge_nm_rxq *nm_rxq;
4115 #endif
4116 #ifdef RSS
4117 	int nbuckets = rss_getnumbuckets();
4118 #endif
4119 
4120 	/*
4121 	 * Setup interrupts.
4122 	 */
4123 	irq = &sc->irq[0];
4124 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
4125 	if (sc->intr_count == 1)
4126 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4127 
4128 	/* Multiple interrupts. */
4129 	if (sc->flags & IS_VF)
4130 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4131 		    ("%s: too few intr.", __func__));
4132 	else
4133 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4134 		    ("%s: too few intr.", __func__));
4135 
4136 	/* The first one is always error intr on PFs */
4137 	if (!(sc->flags & IS_VF)) {
4138 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4139 		if (rc != 0)
4140 			return (rc);
4141 		irq++;
4142 		rid++;
4143 	}
4144 
4145 	/* The second one is always the firmware event queue (first on VFs) */
4146 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4147 	if (rc != 0)
4148 		return (rc);
4149 	irq++;
4150 	rid++;
4151 
4152 	for_each_port(sc, p) {
4153 		pi = sc->port[p];
4154 		for_each_vi(pi, v, vi) {
4155 			vi->first_intr = rid - 1;
4156 
4157 			if (vi->nnmrxq > 0) {
4158 				int n = max(vi->nrxq, vi->nnmrxq);
4159 
4160 				MPASS(vi->flags & INTR_RXQ);
4161 
4162 				rxq = &sge->rxq[vi->first_rxq];
4163 #ifdef DEV_NETMAP
4164 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4165 #endif
4166 				for (q = 0; q < n; q++) {
4167 					snprintf(s, sizeof(s), "%x%c%x", p,
4168 					    'a' + v, q);
4169 					if (q < vi->nrxq)
4170 						irq->rxq = rxq++;
4171 #ifdef DEV_NETMAP
4172 					if (q < vi->nnmrxq)
4173 						irq->nm_rxq = nm_rxq++;
4174 #endif
4175 					rc = t4_alloc_irq(sc, irq, rid,
4176 					    t4_vi_intr, irq, s);
4177 					if (rc != 0)
4178 						return (rc);
4179 					irq++;
4180 					rid++;
4181 					vi->nintr++;
4182 				}
4183 			} else if (vi->flags & INTR_RXQ) {
4184 				for_each_rxq(vi, q, rxq) {
4185 					snprintf(s, sizeof(s), "%x%c%x", p,
4186 					    'a' + v, q);
4187 					rc = t4_alloc_irq(sc, irq, rid,
4188 					    t4_intr, rxq, s);
4189 					if (rc != 0)
4190 						return (rc);
4191 #ifdef RSS
4192 					bus_bind_intr(sc->dev, irq->res,
4193 					    rss_getcpu(q % nbuckets));
4194 #endif
4195 					irq++;
4196 					rid++;
4197 					vi->nintr++;
4198 				}
4199 			}
4200 #ifdef TCP_OFFLOAD
4201 			if (vi->flags & INTR_OFLD_RXQ) {
4202 				for_each_ofld_rxq(vi, q, ofld_rxq) {
4203 					snprintf(s, sizeof(s), "%x%c%x", p,
4204 					    'A' + v, q);
4205 					rc = t4_alloc_irq(sc, irq, rid,
4206 					    t4_intr, ofld_rxq, s);
4207 					if (rc != 0)
4208 						return (rc);
4209 					irq++;
4210 					rid++;
4211 					vi->nintr++;
4212 				}
4213 			}
4214 #endif
4215 		}
4216 	}
4217 	MPASS(irq == &sc->irq[sc->intr_count]);
4218 
4219 	return (0);
4220 }
4221 
4222 int
4223 adapter_full_init(struct adapter *sc)
4224 {
4225 	int rc, i;
4226 
4227 	ASSERT_SYNCHRONIZED_OP(sc);
4228 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4229 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
4230 	    ("%s: FULL_INIT_DONE already", __func__));
4231 
4232 	/*
4233 	 * queues that belong to the adapter (not any particular port).
4234 	 */
4235 	rc = t4_setup_adapter_queues(sc);
4236 	if (rc != 0)
4237 		goto done;
4238 
4239 	for (i = 0; i < nitems(sc->tq); i++) {
4240 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
4241 		    taskqueue_thread_enqueue, &sc->tq[i]);
4242 		if (sc->tq[i] == NULL) {
4243 			device_printf(sc->dev,
4244 			    "failed to allocate task queue %d\n", i);
4245 			rc = ENOMEM;
4246 			goto done;
4247 		}
4248 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
4249 		    device_get_nameunit(sc->dev), i);
4250 	}
4251 
4252 	if (!(sc->flags & IS_VF))
4253 		t4_intr_enable(sc);
4254 	sc->flags |= FULL_INIT_DONE;
4255 done:
4256 	if (rc != 0)
4257 		adapter_full_uninit(sc);
4258 
4259 	return (rc);
4260 }
4261 
4262 int
4263 adapter_full_uninit(struct adapter *sc)
4264 {
4265 	int i;
4266 
4267 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4268 
4269 	t4_teardown_adapter_queues(sc);
4270 
4271 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
4272 		taskqueue_free(sc->tq[i]);
4273 		sc->tq[i] = NULL;
4274 	}
4275 
4276 	sc->flags &= ~FULL_INIT_DONE;
4277 
4278 	return (0);
4279 }
4280 
4281 #ifdef RSS
4282 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
4283     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
4284     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
4285     RSS_HASHTYPE_RSS_UDP_IPV6)
4286 
4287 /* Translates kernel hash types to hardware. */
4288 static int
4289 hashconfig_to_hashen(int hashconfig)
4290 {
4291 	int hashen = 0;
4292 
4293 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
4294 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
4295 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
4296 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
4297 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
4298 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4299 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4300 	}
4301 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
4302 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4303 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4304 	}
4305 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
4306 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4307 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
4308 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4309 
4310 	return (hashen);
4311 }
4312 
4313 /* Translates hardware hash types to kernel. */
4314 static int
4315 hashen_to_hashconfig(int hashen)
4316 {
4317 	int hashconfig = 0;
4318 
4319 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
4320 		/*
4321 		 * If UDP hashing was enabled it must have been enabled for
4322 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
4323 		 * enabling any 4-tuple hash is nonsense configuration.
4324 		 */
4325 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4326 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
4327 
4328 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4329 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
4330 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4331 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
4332 	}
4333 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4334 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
4335 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4336 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
4337 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
4338 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
4339 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
4340 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
4341 
4342 	return (hashconfig);
4343 }
4344 #endif
4345 
4346 int
4347 vi_full_init(struct vi_info *vi)
4348 {
4349 	struct adapter *sc = vi->pi->adapter;
4350 	struct ifnet *ifp = vi->ifp;
4351 	uint16_t *rss;
4352 	struct sge_rxq *rxq;
4353 	int rc, i, j, hashen;
4354 #ifdef RSS
4355 	int nbuckets = rss_getnumbuckets();
4356 	int hashconfig = rss_gethashconfig();
4357 	int extra;
4358 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4359 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4360 #endif
4361 
4362 	ASSERT_SYNCHRONIZED_OP(sc);
4363 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
4364 	    ("%s: VI_INIT_DONE already", __func__));
4365 
4366 	sysctl_ctx_init(&vi->ctx);
4367 	vi->flags |= VI_SYSCTL_CTX;
4368 
4369 	/*
4370 	 * Allocate tx/rx/fl queues for this VI.
4371 	 */
4372 	rc = t4_setup_vi_queues(vi);
4373 	if (rc != 0)
4374 		goto done;	/* error message displayed already */
4375 
4376 	/*
4377 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
4378 	 */
4379 	if (vi->nrxq > vi->rss_size) {
4380 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
4381 		    "some queues will never receive traffic.\n", vi->nrxq,
4382 		    vi->rss_size);
4383 	} else if (vi->rss_size % vi->nrxq) {
4384 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
4385 		    "expect uneven traffic distribution.\n", vi->nrxq,
4386 		    vi->rss_size);
4387 	}
4388 #ifdef RSS
4389 	MPASS(RSS_KEYSIZE == 40);
4390 	if (vi->nrxq != nbuckets) {
4391 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
4392 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
4393 	}
4394 
4395 	rss_getkey((void *)&raw_rss_key[0]);
4396 	for (i = 0; i < nitems(rss_key); i++) {
4397 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
4398 	}
4399 	t4_write_rss_key(sc, &rss_key[0], -1);
4400 #endif
4401 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
4402 	for (i = 0; i < vi->rss_size;) {
4403 #ifdef RSS
4404 		j = rss_get_indirection_to_bucket(i);
4405 		j %= vi->nrxq;
4406 		rxq = &sc->sge.rxq[vi->first_rxq + j];
4407 		rss[i++] = rxq->iq.abs_id;
4408 #else
4409 		for_each_rxq(vi, j, rxq) {
4410 			rss[i++] = rxq->iq.abs_id;
4411 			if (i == vi->rss_size)
4412 				break;
4413 		}
4414 #endif
4415 	}
4416 
4417 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
4418 	    vi->rss_size);
4419 	if (rc != 0) {
4420 		if_printf(ifp, "rss_config failed: %d\n", rc);
4421 		goto done;
4422 	}
4423 
4424 #ifdef RSS
4425 	hashen = hashconfig_to_hashen(hashconfig);
4426 
4427 	/*
4428 	 * We may have had to enable some hashes even though the global config
4429 	 * wants them disabled.  This is a potential problem that must be
4430 	 * reported to the user.
4431 	 */
4432 	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
4433 
4434 	/*
4435 	 * If we consider only the supported hash types, then the enabled hashes
4436 	 * are a superset of the requested hashes.  In other words, there cannot
4437 	 * be any supported hash that was requested but not enabled, but there
4438 	 * can be hashes that were not requested but had to be enabled.
4439 	 */
4440 	extra &= SUPPORTED_RSS_HASHTYPES;
4441 	MPASS((extra & hashconfig) == 0);
4442 
4443 	if (extra) {
4444 		if_printf(ifp,
4445 		    "global RSS config (0x%x) cannot be accommodated.\n",
4446 		    hashconfig);
4447 	}
4448 	if (extra & RSS_HASHTYPE_RSS_IPV4)
4449 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
4450 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
4451 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
4452 	if (extra & RSS_HASHTYPE_RSS_IPV6)
4453 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
4454 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
4455 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
4456 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
4457 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
4458 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
4459 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
4460 #else
4461 	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
4462 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
4463 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4464 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
4465 #endif
4466 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
4467 	if (rc != 0) {
4468 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
4469 		goto done;
4470 	}
4471 
4472 	vi->rss = rss;
4473 	vi->flags |= VI_INIT_DONE;
4474 done:
4475 	if (rc != 0)
4476 		vi_full_uninit(vi);
4477 
4478 	return (rc);
4479 }
4480 
4481 /*
4482  * Idempotent.
4483  */
4484 int
4485 vi_full_uninit(struct vi_info *vi)
4486 {
4487 	struct port_info *pi = vi->pi;
4488 	struct adapter *sc = pi->adapter;
4489 	int i;
4490 	struct sge_rxq *rxq;
4491 	struct sge_txq *txq;
4492 #ifdef TCP_OFFLOAD
4493 	struct sge_ofld_rxq *ofld_rxq;
4494 	struct sge_wrq *ofld_txq;
4495 #endif
4496 
4497 	if (vi->flags & VI_INIT_DONE) {
4498 
4499 		/* Need to quiesce queues.  */
4500 
4501 		/* XXX: Only for the first VI? */
4502 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
4503 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
4504 
4505 		for_each_txq(vi, i, txq) {
4506 			quiesce_txq(sc, txq);
4507 		}
4508 
4509 #ifdef TCP_OFFLOAD
4510 		for_each_ofld_txq(vi, i, ofld_txq) {
4511 			quiesce_wrq(sc, ofld_txq);
4512 		}
4513 #endif
4514 
4515 		for_each_rxq(vi, i, rxq) {
4516 			quiesce_iq(sc, &rxq->iq);
4517 			quiesce_fl(sc, &rxq->fl);
4518 		}
4519 
4520 #ifdef TCP_OFFLOAD
4521 		for_each_ofld_rxq(vi, i, ofld_rxq) {
4522 			quiesce_iq(sc, &ofld_rxq->iq);
4523 			quiesce_fl(sc, &ofld_rxq->fl);
4524 		}
4525 #endif
4526 		free(vi->rss, M_CXGBE);
4527 		free(vi->nm_rss, M_CXGBE);
4528 	}
4529 
4530 	t4_teardown_vi_queues(vi);
4531 	vi->flags &= ~VI_INIT_DONE;
4532 
4533 	return (0);
4534 }
4535 
4536 static void
4537 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
4538 {
4539 	struct sge_eq *eq = &txq->eq;
4540 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
4541 
4542 	(void) sc;	/* unused */
4543 
4544 #ifdef INVARIANTS
4545 	TXQ_LOCK(txq);
4546 	MPASS((eq->flags & EQ_ENABLED) == 0);
4547 	TXQ_UNLOCK(txq);
4548 #endif
4549 
4550 	/* Wait for the mp_ring to empty. */
4551 	while (!mp_ring_is_idle(txq->r)) {
4552 		mp_ring_check_drainage(txq->r, 0);
4553 		pause("rquiesce", 1);
4554 	}
4555 
4556 	/* Then wait for the hardware to finish. */
4557 	while (spg->cidx != htobe16(eq->pidx))
4558 		pause("equiesce", 1);
4559 
4560 	/* Finally, wait for the driver to reclaim all descriptors. */
4561 	while (eq->cidx != eq->pidx)
4562 		pause("dquiesce", 1);
4563 }
4564 
4565 static void
4566 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
4567 {
4568 
4569 	/* XXXTX */
4570 }
4571 
4572 static void
4573 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
4574 {
4575 	(void) sc;	/* unused */
4576 
4577 	/* Synchronize with the interrupt handler */
4578 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
4579 		pause("iqfree", 1);
4580 }
4581 
4582 static void
4583 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
4584 {
4585 	mtx_lock(&sc->sfl_lock);
4586 	FL_LOCK(fl);
4587 	fl->flags |= FL_DOOMED;
4588 	FL_UNLOCK(fl);
4589 	callout_stop(&sc->sfl_callout);
4590 	mtx_unlock(&sc->sfl_lock);
4591 
4592 	KASSERT((fl->flags & FL_STARVING) == 0,
4593 	    ("%s: still starving", __func__));
4594 }
4595 
4596 static int
4597 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
4598     driver_intr_t *handler, void *arg, char *name)
4599 {
4600 	int rc;
4601 
4602 	irq->rid = rid;
4603 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
4604 	    RF_SHAREABLE | RF_ACTIVE);
4605 	if (irq->res == NULL) {
4606 		device_printf(sc->dev,
4607 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
4608 		return (ENOMEM);
4609 	}
4610 
4611 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
4612 	    NULL, handler, arg, &irq->tag);
4613 	if (rc != 0) {
4614 		device_printf(sc->dev,
4615 		    "failed to setup interrupt for rid %d, name %s: %d\n",
4616 		    rid, name, rc);
4617 	} else if (name)
4618 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
4619 
4620 	return (rc);
4621 }
4622 
4623 static int
4624 t4_free_irq(struct adapter *sc, struct irq *irq)
4625 {
4626 	if (irq->tag)
4627 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
4628 	if (irq->res)
4629 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
4630 
4631 	bzero(irq, sizeof(*irq));
4632 
4633 	return (0);
4634 }
4635 
4636 static void
4637 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
4638 {
4639 
4640 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
4641 	t4_get_regs(sc, buf, regs->len);
4642 }
4643 
4644 #define	A_PL_INDIR_CMD	0x1f8
4645 
4646 #define	S_PL_AUTOINC	31
4647 #define	M_PL_AUTOINC	0x1U
4648 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
4649 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
4650 
4651 #define	S_PL_VFID	20
4652 #define	M_PL_VFID	0xffU
4653 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
4654 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
4655 
4656 #define	S_PL_ADDR	0
4657 #define	M_PL_ADDR	0xfffffU
4658 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
4659 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
4660 
4661 #define	A_PL_INDIR_DATA	0x1fc
4662 
4663 static uint64_t
4664 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
4665 {
4666 	u32 stats[2];
4667 
4668 	mtx_assert(&sc->reg_lock, MA_OWNED);
4669 	if (sc->flags & IS_VF) {
4670 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
4671 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
4672 	} else {
4673 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4674 		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4675 		    V_PL_ADDR(VF_MPS_REG(reg)));
4676 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
4677 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
4678 	}
4679 	return (((uint64_t)stats[1]) << 32 | stats[0]);
4680 }
4681 
4682 static void
4683 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
4684     struct fw_vi_stats_vf *stats)
4685 {
4686 
4687 #define GET_STAT(name) \
4688 	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
4689 
4690 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
4691 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
4692 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
4693 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
4694 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
4695 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
4696 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
4697 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
4698 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
4699 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
4700 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
4701 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
4702 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
4703 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
4704 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
4705 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
4706 
4707 #undef GET_STAT
4708 }
4709 
4710 static void
4711 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
4712 {
4713 	int reg;
4714 
4715 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4716 	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4717 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
4718 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
4719 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
4720 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
4721 }
4722 
4723 static void
4724 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
4725 {
4726 	struct timeval tv;
4727 	const struct timeval interval = {0, 250000};	/* 250ms */
4728 
4729 	if (!(vi->flags & VI_INIT_DONE))
4730 		return;
4731 
4732 	getmicrotime(&tv);
4733 	timevalsub(&tv, &interval);
4734 	if (timevalcmp(&tv, &vi->last_refreshed, <))
4735 		return;
4736 
4737 	mtx_lock(&sc->reg_lock);
4738 	t4_get_vi_stats(sc, vi->viid, &vi->stats);
4739 	getmicrotime(&vi->last_refreshed);
4740 	mtx_unlock(&sc->reg_lock);
4741 }
4742 
4743 static void
4744 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
4745 {
4746 	int i;
4747 	u_int v, tnl_cong_drops;
4748 	struct timeval tv;
4749 	const struct timeval interval = {0, 250000};	/* 250ms */
4750 
4751 	getmicrotime(&tv);
4752 	timevalsub(&tv, &interval);
4753 	if (timevalcmp(&tv, &pi->last_refreshed, <))
4754 		return;
4755 
4756 	tnl_cong_drops = 0;
4757 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
4758 	for (i = 0; i < sc->chip_params->nchan; i++) {
4759 		if (pi->rx_chan_map & (1 << i)) {
4760 			mtx_lock(&sc->reg_lock);
4761 			t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
4762 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
4763 			mtx_unlock(&sc->reg_lock);
4764 			tnl_cong_drops += v;
4765 		}
4766 	}
4767 	pi->tnl_cong_drops = tnl_cong_drops;
4768 	getmicrotime(&pi->last_refreshed);
4769 }
4770 
4771 static void
4772 cxgbe_tick(void *arg)
4773 {
4774 	struct port_info *pi = arg;
4775 	struct adapter *sc = pi->adapter;
4776 
4777 	PORT_LOCK_ASSERT_OWNED(pi);
4778 	cxgbe_refresh_stats(sc, pi);
4779 
4780 	callout_schedule(&pi->tick, hz);
4781 }
4782 
4783 void
4784 vi_tick(void *arg)
4785 {
4786 	struct vi_info *vi = arg;
4787 	struct adapter *sc = vi->pi->adapter;
4788 
4789 	vi_refresh_stats(sc, vi);
4790 
4791 	callout_schedule(&vi->tick, hz);
4792 }
4793 
4794 static void
4795 cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
4796 {
4797 	struct ifnet *vlan;
4798 
4799 	if (arg != ifp || ifp->if_type != IFT_ETHER)
4800 		return;
4801 
4802 	vlan = VLAN_DEVAT(ifp, vid);
4803 	VLAN_SETCOOKIE(vlan, ifp);
4804 }
4805 
4806 /*
4807  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
4808  */
4809 static char *caps_decoder[] = {
4810 	"\20\001IPMI\002NCSI",				/* 0: NBM */
4811 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
4812 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
4813 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
4814 	    "\006HASHFILTER\007ETHOFLD",
4815 	"\20\001TOE",					/* 4: TOE */
4816 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
4817 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
4818 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
4819 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
4820 	    "\007T10DIF"
4821 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
4822 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
4823 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
4824 		    "\004PO_INITIATOR\005PO_TARGET",
4825 };
4826 
4827 void
4828 t4_sysctls(struct adapter *sc)
4829 {
4830 	struct sysctl_ctx_list *ctx;
4831 	struct sysctl_oid *oid;
4832 	struct sysctl_oid_list *children, *c0;
4833 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
4834 
4835 	ctx = device_get_sysctl_ctx(sc->dev);
4836 
4837 	/*
4838 	 * dev.t4nex.X.
4839 	 */
4840 	oid = device_get_sysctl_tree(sc->dev);
4841 	c0 = children = SYSCTL_CHILDREN(oid);
4842 
4843 	sc->sc_do_rxcopy = 1;
4844 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
4845 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
4846 
4847 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
4848 	    sc->params.nports, "# of ports");
4849 
4850 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
4851 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
4852 	    sysctl_bitfield, "A", "available doorbells");
4853 
4854 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
4855 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
4856 
4857 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
4858 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
4859 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
4860 	    "interrupt holdoff timer values (us)");
4861 
4862 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
4863 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
4864 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
4865 	    "interrupt holdoff packet counter values");
4866 
4867 	t4_sge_sysctls(sc, ctx, children);
4868 
4869 	sc->lro_timeout = 100;
4870 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
4871 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
4872 
4873 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
4874 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
4875 
4876 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
4877 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
4878 
4879 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
4880 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
4881 
4882 	if (sc->flags & IS_VF)
4883 		return;
4884 
4885 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
4886 	    NULL, chip_rev(sc), "chip hardware revision");
4887 
4888 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
4889 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
4890 
4891 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
4892 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
4893 
4894 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
4895 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
4896 
4897 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
4898 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
4899 
4900 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
4901 	    sc->er_version, 0, "expansion ROM version");
4902 
4903 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
4904 	    sc->bs_version, 0, "bootstrap firmware version");
4905 
4906 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
4907 	    NULL, sc->params.scfg_vers, "serial config version");
4908 
4909 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
4910 	    NULL, sc->params.vpd_vers, "VPD version");
4911 
4912 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
4913 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
4914 
4915 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
4916 	    sc->cfcsum, "config file checksum");
4917 
4918 #define SYSCTL_CAP(name, n, text) \
4919 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
4920 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
4921 	    sysctl_bitfield, "A", "available " text " capabilities")
4922 
4923 	SYSCTL_CAP(nbmcaps, 0, "NBM");
4924 	SYSCTL_CAP(linkcaps, 1, "link");
4925 	SYSCTL_CAP(switchcaps, 2, "switch");
4926 	SYSCTL_CAP(niccaps, 3, "NIC");
4927 	SYSCTL_CAP(toecaps, 4, "TCP offload");
4928 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
4929 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
4930 	SYSCTL_CAP(cryptocaps, 7, "crypto");
4931 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
4932 #undef SYSCTL_CAP
4933 
4934 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
4935 	    NULL, sc->tids.nftids, "number of filters");
4936 
4937 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
4938 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
4939 	    "chip temperature (in Celsius)");
4940 
4941 #ifdef SBUF_DRAIN
4942 	/*
4943 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
4944 	 */
4945 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
4946 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
4947 	    "logs and miscellaneous information");
4948 	children = SYSCTL_CHILDREN(oid);
4949 
4950 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
4951 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4952 	    sysctl_cctrl, "A", "congestion control");
4953 
4954 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
4955 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4956 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
4957 
4958 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
4959 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
4960 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
4961 
4962 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
4963 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
4964 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
4965 
4966 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
4967 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
4968 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
4969 
4970 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
4971 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
4972 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
4973 
4974 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
4975 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
4976 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
4977 
4978 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
4979 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4980 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
4981 	    "A", "CIM logic analyzer");
4982 
4983 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
4984 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4985 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
4986 
4987 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
4988 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
4989 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
4990 
4991 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
4992 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
4993 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
4994 
4995 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
4996 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
4997 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
4998 
4999 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5000 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5001 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5002 
5003 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5004 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5005 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5006 
5007 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5008 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5009 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5010 
5011 	if (chip_id(sc) > CHELSIO_T4) {
5012 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5013 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5014 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5015 
5016 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5017 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5018 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5019 	}
5020 
5021 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5022 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5023 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5024 
5025 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5026 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5027 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
5028 
5029 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5030 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5031 	    sysctl_cpl_stats, "A", "CPL statistics");
5032 
5033 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5034 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5035 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5036 
5037 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5038 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5039 	    sysctl_devlog, "A", "firmware's device log");
5040 
5041 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5042 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5043 	    sysctl_fcoe_stats, "A", "FCoE statistics");
5044 
5045 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5046 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5047 	    sysctl_hw_sched, "A", "hardware scheduler ");
5048 
5049 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5050 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5051 	    sysctl_l2t, "A", "hardware L2 table");
5052 
5053 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5054 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5055 	    sysctl_lb_stats, "A", "loopback statistics");
5056 
5057 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5058 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5059 	    sysctl_meminfo, "A", "memory regions");
5060 
5061 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5062 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5063 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5064 	    "A", "MPS TCAM entries");
5065 
5066 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5067 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5068 	    sysctl_path_mtus, "A", "path MTUs");
5069 
5070 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5071 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5072 	    sysctl_pm_stats, "A", "PM statistics");
5073 
5074 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5075 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5076 	    sysctl_rdma_stats, "A", "RDMA statistics");
5077 
5078 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5079 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5080 	    sysctl_tcp_stats, "A", "TCP statistics");
5081 
5082 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5083 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5084 	    sysctl_tids, "A", "TID information");
5085 
5086 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5087 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5088 	    sysctl_tp_err_stats, "A", "TP error statistics");
5089 
5090 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5091 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5092 	    "TP logic analyzer event capture mask");
5093 
5094 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5095 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5096 	    sysctl_tp_la, "A", "TP logic analyzer");
5097 
5098 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5099 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5100 	    sysctl_tx_rate, "A", "Tx rate");
5101 
5102 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5103 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5104 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5105 
5106 	if (chip_id(sc) >= CHELSIO_T5) {
5107 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5108 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5109 		    sysctl_wcwr_stats, "A", "write combined work requests");
5110 	}
5111 #endif
5112 
5113 #ifdef TCP_OFFLOAD
5114 	if (is_offload(sc)) {
5115 		/*
5116 		 * dev.t4nex.X.toe.
5117 		 */
5118 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5119 		    NULL, "TOE parameters");
5120 		children = SYSCTL_CHILDREN(oid);
5121 
5122 		sc->tt.sndbuf = 256 * 1024;
5123 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5124 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
5125 
5126 		sc->tt.ddp = 0;
5127 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5128 		    &sc->tt.ddp, 0, "DDP allowed");
5129 
5130 		sc->tt.rx_coalesce = 1;
5131 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5132 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5133 
5134 		sc->tt.tx_align = 1;
5135 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5136 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5137 
5138 		sc->tt.tx_zcopy = 0;
5139 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
5140 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
5141 		    "Enable zero-copy aio_write(2)");
5142 
5143 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
5144 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
5145 		    "TP timer tick (us)");
5146 
5147 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
5148 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
5149 		    "TCP timestamp tick (us)");
5150 
5151 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5152 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5153 		    "DACK tick (us)");
5154 
5155 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5156 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5157 		    "IU", "DACK timer (us)");
5158 
5159 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5160 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5161 		    sysctl_tp_timer, "LU", "Retransmit min (us)");
5162 
5163 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5164 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5165 		    sysctl_tp_timer, "LU", "Retransmit max (us)");
5166 
5167 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5168 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5169 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
5170 
5171 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5172 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5173 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
5174 
5175 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5176 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5177 		    sysctl_tp_timer, "LU", "Keepidle idle timer (us)");
5178 
5179 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_intvl",
5180 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
5181 		    sysctl_tp_timer, "LU", "Keepidle interval (us)");
5182 
5183 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
5184 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
5185 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
5186 
5187 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
5188 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
5189 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
5190 	}
5191 #endif
5192 }
5193 
5194 void
5195 vi_sysctls(struct vi_info *vi)
5196 {
5197 	struct sysctl_ctx_list *ctx;
5198 	struct sysctl_oid *oid;
5199 	struct sysctl_oid_list *children;
5200 
5201 	ctx = device_get_sysctl_ctx(vi->dev);
5202 
5203 	/*
5204 	 * dev.v?(cxgbe|cxl).X.
5205 	 */
5206 	oid = device_get_sysctl_tree(vi->dev);
5207 	children = SYSCTL_CHILDREN(oid);
5208 
5209 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
5210 	    vi->viid, "VI identifer");
5211 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
5212 	    &vi->nrxq, 0, "# of rx queues");
5213 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5214 	    &vi->ntxq, 0, "# of tx queues");
5215 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5216 	    &vi->first_rxq, 0, "index of first rx queue");
5217 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5218 	    &vi->first_txq, 0, "index of first tx queue");
5219 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
5220 	    vi->rss_size, "size of RSS indirection table");
5221 
5222 	if (IS_MAIN_VI(vi)) {
5223 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
5224 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
5225 		    "Reserve queue 0 for non-flowid packets");
5226 	}
5227 
5228 #ifdef TCP_OFFLOAD
5229 	if (vi->nofldrxq != 0) {
5230 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5231 		    &vi->nofldrxq, 0,
5232 		    "# of rx queues for offloaded TCP connections");
5233 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5234 		    &vi->nofldtxq, 0,
5235 		    "# of tx queues for offloaded TCP connections");
5236 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5237 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
5238 		    "index of first TOE rx queue");
5239 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5240 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
5241 		    "index of first TOE tx queue");
5242 	}
5243 #endif
5244 #ifdef DEV_NETMAP
5245 	if (vi->nnmrxq != 0) {
5246 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5247 		    &vi->nnmrxq, 0, "# of netmap rx queues");
5248 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5249 		    &vi->nnmtxq, 0, "# of netmap tx queues");
5250 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5251 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
5252 		    "index of first netmap rx queue");
5253 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
5254 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
5255 		    "index of first netmap tx queue");
5256 	}
5257 #endif
5258 
5259 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
5260 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
5261 	    "holdoff timer index");
5262 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
5263 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
5264 	    "holdoff packet counter index");
5265 
5266 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
5267 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
5268 	    "rx queue size");
5269 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
5270 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
5271 	    "tx queue size");
5272 }
5273 
5274 static void
5275 cxgbe_sysctls(struct port_info *pi)
5276 {
5277 	struct sysctl_ctx_list *ctx;
5278 	struct sysctl_oid *oid;
5279 	struct sysctl_oid_list *children, *children2;
5280 	struct adapter *sc = pi->adapter;
5281 	int i;
5282 	char name[16];
5283 
5284 	ctx = device_get_sysctl_ctx(pi->dev);
5285 
5286 	/*
5287 	 * dev.cxgbe.X.
5288 	 */
5289 	oid = device_get_sysctl_tree(pi->dev);
5290 	children = SYSCTL_CHILDREN(oid);
5291 
5292 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
5293 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
5294 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
5295 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
5296 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
5297 		    "PHY temperature (in Celsius)");
5298 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
5299 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
5300 		    "PHY firmware version");
5301 	}
5302 
5303 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
5304 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
5305 	    "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
5306 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
5307 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
5308 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
5309 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
5310 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
5311 	    "autonegotiation (-1 = not supported)");
5312 
5313 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
5314 	    port_top_speed(pi), "max speed (in Gbps)");
5315 
5316 	if (sc->flags & IS_VF)
5317 		return;
5318 
5319 	/*
5320 	 * dev.(cxgbe|cxl).X.tc.
5321 	 */
5322 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
5323 	    "Tx scheduler traffic classes");
5324 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
5325 		struct tx_sched_class *tc = &pi->tc[i];
5326 
5327 		snprintf(name, sizeof(name), "%d", i);
5328 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
5329 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
5330 		    "traffic class"));
5331 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
5332 		    &tc->flags, 0, "flags");
5333 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
5334 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
5335 #ifdef SBUF_DRAIN
5336 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
5337 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
5338 		    sysctl_tc_params, "A", "traffic class parameters");
5339 #endif
5340 	}
5341 
5342 	/*
5343 	 * dev.cxgbe.X.stats.
5344 	 */
5345 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
5346 	    NULL, "port statistics");
5347 	children = SYSCTL_CHILDREN(oid);
5348 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
5349 	    &pi->tx_parse_error, 0,
5350 	    "# of tx packets with invalid length or # of segments");
5351 
5352 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
5353 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
5354 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
5355 	    sysctl_handle_t4_reg64, "QU", desc)
5356 
5357 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
5358 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
5359 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
5360 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
5361 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
5362 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
5363 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
5364 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
5365 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
5366 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
5367 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
5368 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
5369 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
5370 	    "# of tx frames in this range",
5371 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
5372 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
5373 	    "# of tx frames in this range",
5374 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
5375 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
5376 	    "# of tx frames in this range",
5377 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
5378 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
5379 	    "# of tx frames in this range",
5380 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
5381 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
5382 	    "# of tx frames in this range",
5383 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
5384 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
5385 	    "# of tx frames in this range",
5386 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
5387 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
5388 	    "# of tx frames in this range",
5389 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
5390 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
5391 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
5392 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
5393 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
5394 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
5395 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
5396 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5397 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5398 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5399 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5400 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5401 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5402 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5403 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5404 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5405 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5406 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5407 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5408 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5409 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5410 
5411 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5412 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5413 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5414 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5415 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5416 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5417 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5418 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5419 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5420 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5421 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5422 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5423 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5424 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5425 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5426 	    "# of frames received with bad FCS",
5427 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5428 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5429 	    "# of frames received with length error",
5430 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5431 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5432 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5433 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5434 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5435 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5436 	    "# of rx frames in this range",
5437 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5438 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5439 	    "# of rx frames in this range",
5440 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5441 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5442 	    "# of rx frames in this range",
5443 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5444 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5445 	    "# of rx frames in this range",
5446 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5447 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5448 	    "# of rx frames in this range",
5449 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5450 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5451 	    "# of rx frames in this range",
5452 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5453 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5454 	    "# of rx frames in this range",
5455 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5456 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5457 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5458 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5459 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5460 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5461 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5462 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5463 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5464 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5465 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5466 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5467 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5468 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5469 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5470 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5471 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5472 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5473 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5474 
5475 #undef SYSCTL_ADD_T4_REG64
5476 
5477 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5478 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5479 	    &pi->stats.name, desc)
5480 
5481 	/* We get these from port_stats and they may be stale by up to 1s */
5482 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5483 	    "# drops due to buffer-group 0 overflows");
5484 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5485 	    "# drops due to buffer-group 1 overflows");
5486 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5487 	    "# drops due to buffer-group 2 overflows");
5488 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5489 	    "# drops due to buffer-group 3 overflows");
5490 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5491 	    "# of buffer-group 0 truncated packets");
5492 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5493 	    "# of buffer-group 1 truncated packets");
5494 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5495 	    "# of buffer-group 2 truncated packets");
5496 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5497 	    "# of buffer-group 3 truncated packets");
5498 
5499 #undef SYSCTL_ADD_T4_PORTSTAT
5500 }
5501 
5502 static int
5503 sysctl_int_array(SYSCTL_HANDLER_ARGS)
5504 {
5505 	int rc, *i, space = 0;
5506 	struct sbuf sb;
5507 
5508 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
5509 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
5510 		if (space)
5511 			sbuf_printf(&sb, " ");
5512 		sbuf_printf(&sb, "%d", *i);
5513 		space = 1;
5514 	}
5515 	rc = sbuf_finish(&sb);
5516 	sbuf_delete(&sb);
5517 	return (rc);
5518 }
5519 
5520 static int
5521 sysctl_bitfield(SYSCTL_HANDLER_ARGS)
5522 {
5523 	int rc;
5524 	struct sbuf *sb;
5525 
5526 	rc = sysctl_wire_old_buffer(req, 0);
5527 	if (rc != 0)
5528 		return(rc);
5529 
5530 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5531 	if (sb == NULL)
5532 		return (ENOMEM);
5533 
5534 	sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
5535 	rc = sbuf_finish(sb);
5536 	sbuf_delete(sb);
5537 
5538 	return (rc);
5539 }
5540 
5541 static int
5542 sysctl_btphy(SYSCTL_HANDLER_ARGS)
5543 {
5544 	struct port_info *pi = arg1;
5545 	int op = arg2;
5546 	struct adapter *sc = pi->adapter;
5547 	u_int v;
5548 	int rc;
5549 
5550 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
5551 	if (rc)
5552 		return (rc);
5553 	/* XXX: magic numbers */
5554 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
5555 	    &v);
5556 	end_synchronized_op(sc, 0);
5557 	if (rc)
5558 		return (rc);
5559 	if (op == 0)
5560 		v /= 256;
5561 
5562 	rc = sysctl_handle_int(oidp, &v, 0, req);
5563 	return (rc);
5564 }
5565 
5566 static int
5567 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
5568 {
5569 	struct vi_info *vi = arg1;
5570 	int rc, val;
5571 
5572 	val = vi->rsrv_noflowq;
5573 	rc = sysctl_handle_int(oidp, &val, 0, req);
5574 	if (rc != 0 || req->newptr == NULL)
5575 		return (rc);
5576 
5577 	if ((val >= 1) && (vi->ntxq > 1))
5578 		vi->rsrv_noflowq = 1;
5579 	else
5580 		vi->rsrv_noflowq = 0;
5581 
5582 	return (rc);
5583 }
5584 
5585 static int
5586 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
5587 {
5588 	struct vi_info *vi = arg1;
5589 	struct adapter *sc = vi->pi->adapter;
5590 	int idx, rc, i;
5591 	struct sge_rxq *rxq;
5592 #ifdef TCP_OFFLOAD
5593 	struct sge_ofld_rxq *ofld_rxq;
5594 #endif
5595 	uint8_t v;
5596 
5597 	idx = vi->tmr_idx;
5598 
5599 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5600 	if (rc != 0 || req->newptr == NULL)
5601 		return (rc);
5602 
5603 	if (idx < 0 || idx >= SGE_NTIMERS)
5604 		return (EINVAL);
5605 
5606 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5607 	    "t4tmr");
5608 	if (rc)
5609 		return (rc);
5610 
5611 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
5612 	for_each_rxq(vi, i, rxq) {
5613 #ifdef atomic_store_rel_8
5614 		atomic_store_rel_8(&rxq->iq.intr_params, v);
5615 #else
5616 		rxq->iq.intr_params = v;
5617 #endif
5618 	}
5619 #ifdef TCP_OFFLOAD
5620 	for_each_ofld_rxq(vi, i, ofld_rxq) {
5621 #ifdef atomic_store_rel_8
5622 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
5623 #else
5624 		ofld_rxq->iq.intr_params = v;
5625 #endif
5626 	}
5627 #endif
5628 	vi->tmr_idx = idx;
5629 
5630 	end_synchronized_op(sc, LOCK_HELD);
5631 	return (0);
5632 }
5633 
5634 static int
5635 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
5636 {
5637 	struct vi_info *vi = arg1;
5638 	struct adapter *sc = vi->pi->adapter;
5639 	int idx, rc;
5640 
5641 	idx = vi->pktc_idx;
5642 
5643 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5644 	if (rc != 0 || req->newptr == NULL)
5645 		return (rc);
5646 
5647 	if (idx < -1 || idx >= SGE_NCOUNTERS)
5648 		return (EINVAL);
5649 
5650 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5651 	    "t4pktc");
5652 	if (rc)
5653 		return (rc);
5654 
5655 	if (vi->flags & VI_INIT_DONE)
5656 		rc = EBUSY; /* cannot be changed once the queues are created */
5657 	else
5658 		vi->pktc_idx = idx;
5659 
5660 	end_synchronized_op(sc, LOCK_HELD);
5661 	return (rc);
5662 }
5663 
5664 static int
5665 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
5666 {
5667 	struct vi_info *vi = arg1;
5668 	struct adapter *sc = vi->pi->adapter;
5669 	int qsize, rc;
5670 
5671 	qsize = vi->qsize_rxq;
5672 
5673 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5674 	if (rc != 0 || req->newptr == NULL)
5675 		return (rc);
5676 
5677 	if (qsize < 128 || (qsize & 7))
5678 		return (EINVAL);
5679 
5680 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5681 	    "t4rxqs");
5682 	if (rc)
5683 		return (rc);
5684 
5685 	if (vi->flags & VI_INIT_DONE)
5686 		rc = EBUSY; /* cannot be changed once the queues are created */
5687 	else
5688 		vi->qsize_rxq = qsize;
5689 
5690 	end_synchronized_op(sc, LOCK_HELD);
5691 	return (rc);
5692 }
5693 
5694 static int
5695 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
5696 {
5697 	struct vi_info *vi = arg1;
5698 	struct adapter *sc = vi->pi->adapter;
5699 	int qsize, rc;
5700 
5701 	qsize = vi->qsize_txq;
5702 
5703 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5704 	if (rc != 0 || req->newptr == NULL)
5705 		return (rc);
5706 
5707 	if (qsize < 128 || qsize > 65536)
5708 		return (EINVAL);
5709 
5710 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5711 	    "t4txqs");
5712 	if (rc)
5713 		return (rc);
5714 
5715 	if (vi->flags & VI_INIT_DONE)
5716 		rc = EBUSY; /* cannot be changed once the queues are created */
5717 	else
5718 		vi->qsize_txq = qsize;
5719 
5720 	end_synchronized_op(sc, LOCK_HELD);
5721 	return (rc);
5722 }
5723 
5724 static int
5725 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
5726 {
5727 	struct port_info *pi = arg1;
5728 	struct adapter *sc = pi->adapter;
5729 	struct link_config *lc = &pi->link_cfg;
5730 	int rc;
5731 
5732 	if (req->newptr == NULL) {
5733 		struct sbuf *sb;
5734 		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
5735 
5736 		rc = sysctl_wire_old_buffer(req, 0);
5737 		if (rc != 0)
5738 			return(rc);
5739 
5740 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5741 		if (sb == NULL)
5742 			return (ENOMEM);
5743 
5744 		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
5745 		rc = sbuf_finish(sb);
5746 		sbuf_delete(sb);
5747 	} else {
5748 		char s[2];
5749 		int n;
5750 
5751 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
5752 		s[1] = 0;
5753 
5754 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5755 		if (rc != 0)
5756 			return(rc);
5757 
5758 		if (s[1] != 0)
5759 			return (EINVAL);
5760 		if (s[0] < '0' || s[0] > '9')
5761 			return (EINVAL);	/* not a number */
5762 		n = s[0] - '0';
5763 		if (n & ~(PAUSE_TX | PAUSE_RX))
5764 			return (EINVAL);	/* some other bit is set too */
5765 
5766 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5767 		    "t4PAUSE");
5768 		if (rc)
5769 			return (rc);
5770 		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
5771 			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
5772 			lc->requested_fc |= n;
5773 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5774 		}
5775 		end_synchronized_op(sc, 0);
5776 	}
5777 
5778 	return (rc);
5779 }
5780 
5781 static int
5782 sysctl_fec(SYSCTL_HANDLER_ARGS)
5783 {
5784 	struct port_info *pi = arg1;
5785 	struct adapter *sc = pi->adapter;
5786 	struct link_config *lc = &pi->link_cfg;
5787 	int rc;
5788 
5789 	if (req->newptr == NULL) {
5790 		struct sbuf *sb;
5791 		static char *bits = "\20\1RS\2BASER_RS\3RESERVED";
5792 
5793 		rc = sysctl_wire_old_buffer(req, 0);
5794 		if (rc != 0)
5795 			return(rc);
5796 
5797 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5798 		if (sb == NULL)
5799 			return (ENOMEM);
5800 
5801 		sbuf_printf(sb, "%b", lc->fec & M_FW_PORT_CAP_FEC, bits);
5802 		rc = sbuf_finish(sb);
5803 		sbuf_delete(sb);
5804 	} else {
5805 		char s[2];
5806 		int n;
5807 
5808 		s[0] = '0' + (lc->requested_fec & M_FW_PORT_CAP_FEC);
5809 		s[1] = 0;
5810 
5811 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5812 		if (rc != 0)
5813 			return(rc);
5814 
5815 		if (s[1] != 0)
5816 			return (EINVAL);
5817 		if (s[0] < '0' || s[0] > '9')
5818 			return (EINVAL);	/* not a number */
5819 		n = s[0] - '0';
5820 		if (n & ~M_FW_PORT_CAP_FEC)
5821 			return (EINVAL);	/* some other bit is set too */
5822 
5823 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5824 		    "t4fec");
5825 		if (rc)
5826 			return (rc);
5827 		if ((lc->requested_fec & M_FW_PORT_CAP_FEC) != n) {
5828 			lc->requested_fec = n &
5829 			    G_FW_PORT_CAP_FEC(lc->supported);
5830 			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5831 		}
5832 		end_synchronized_op(sc, 0);
5833 	}
5834 
5835 	return (rc);
5836 }
5837 
5838 static int
5839 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
5840 {
5841 	struct port_info *pi = arg1;
5842 	struct adapter *sc = pi->adapter;
5843 	struct link_config *lc = &pi->link_cfg;
5844 	int rc, val, old;
5845 
5846 	if (lc->supported & FW_PORT_CAP_ANEG)
5847 		val = lc->autoneg == AUTONEG_ENABLE ? 1 : 0;
5848 	else
5849 		val = -1;
5850 	rc = sysctl_handle_int(oidp, &val, 0, req);
5851 	if (rc != 0 || req->newptr == NULL)
5852 		return (rc);
5853 	if ((lc->supported & FW_PORT_CAP_ANEG) == 0)
5854 		return (ENOTSUP);
5855 
5856 	val = val ? AUTONEG_ENABLE : AUTONEG_DISABLE;
5857 	if (lc->autoneg == val)
5858 		return (0);	/* no change */
5859 
5860 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5861 	    "t4aneg");
5862 	if (rc)
5863 		return (rc);
5864 	old = lc->autoneg;
5865 	lc->autoneg = val;
5866 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5867 	if (rc != 0)
5868 		lc->autoneg = old;
5869 	return (rc);
5870 }
5871 
5872 static int
5873 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
5874 {
5875 	struct adapter *sc = arg1;
5876 	int reg = arg2;
5877 	uint64_t val;
5878 
5879 	val = t4_read_reg64(sc, reg);
5880 
5881 	return (sysctl_handle_64(oidp, &val, 0, req));
5882 }
5883 
5884 static int
5885 sysctl_temperature(SYSCTL_HANDLER_ARGS)
5886 {
5887 	struct adapter *sc = arg1;
5888 	int rc, t;
5889 	uint32_t param, val;
5890 
5891 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
5892 	if (rc)
5893 		return (rc);
5894 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5895 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
5896 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
5897 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
5898 	end_synchronized_op(sc, 0);
5899 	if (rc)
5900 		return (rc);
5901 
5902 	/* unknown is returned as 0 but we display -1 in that case */
5903 	t = val == 0 ? -1 : val;
5904 
5905 	rc = sysctl_handle_int(oidp, &t, 0, req);
5906 	return (rc);
5907 }
5908 
5909 #ifdef SBUF_DRAIN
5910 static int
5911 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
5912 {
5913 	struct adapter *sc = arg1;
5914 	struct sbuf *sb;
5915 	int rc, i;
5916 	uint16_t incr[NMTUS][NCCTRL_WIN];
5917 	static const char *dec_fac[] = {
5918 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
5919 		"0.9375"
5920 	};
5921 
5922 	rc = sysctl_wire_old_buffer(req, 0);
5923 	if (rc != 0)
5924 		return (rc);
5925 
5926 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5927 	if (sb == NULL)
5928 		return (ENOMEM);
5929 
5930 	t4_read_cong_tbl(sc, incr);
5931 
5932 	for (i = 0; i < NCCTRL_WIN; ++i) {
5933 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
5934 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
5935 		    incr[5][i], incr[6][i], incr[7][i]);
5936 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
5937 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
5938 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
5939 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
5940 	}
5941 
5942 	rc = sbuf_finish(sb);
5943 	sbuf_delete(sb);
5944 
5945 	return (rc);
5946 }
5947 
5948 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
5949 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
5950 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
5951 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
5952 };
5953 
5954 static int
5955 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
5956 {
5957 	struct adapter *sc = arg1;
5958 	struct sbuf *sb;
5959 	int rc, i, n, qid = arg2;
5960 	uint32_t *buf, *p;
5961 	char *qtype;
5962 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
5963 
5964 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
5965 	    ("%s: bad qid %d\n", __func__, qid));
5966 
5967 	if (qid < CIM_NUM_IBQ) {
5968 		/* inbound queue */
5969 		qtype = "IBQ";
5970 		n = 4 * CIM_IBQ_SIZE;
5971 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5972 		rc = t4_read_cim_ibq(sc, qid, buf, n);
5973 	} else {
5974 		/* outbound queue */
5975 		qtype = "OBQ";
5976 		qid -= CIM_NUM_IBQ;
5977 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
5978 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5979 		rc = t4_read_cim_obq(sc, qid, buf, n);
5980 	}
5981 
5982 	if (rc < 0) {
5983 		rc = -rc;
5984 		goto done;
5985 	}
5986 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
5987 
5988 	rc = sysctl_wire_old_buffer(req, 0);
5989 	if (rc != 0)
5990 		goto done;
5991 
5992 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
5993 	if (sb == NULL) {
5994 		rc = ENOMEM;
5995 		goto done;
5996 	}
5997 
5998 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
5999 	for (i = 0, p = buf; i < n; i += 16, p += 4)
6000 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
6001 		    p[2], p[3]);
6002 
6003 	rc = sbuf_finish(sb);
6004 	sbuf_delete(sb);
6005 done:
6006 	free(buf, M_CXGBE);
6007 	return (rc);
6008 }
6009 
6010 static int
6011 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
6012 {
6013 	struct adapter *sc = arg1;
6014 	u_int cfg;
6015 	struct sbuf *sb;
6016 	uint32_t *buf, *p;
6017 	int rc;
6018 
6019 	MPASS(chip_id(sc) <= CHELSIO_T5);
6020 
6021 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6022 	if (rc != 0)
6023 		return (rc);
6024 
6025 	rc = sysctl_wire_old_buffer(req, 0);
6026 	if (rc != 0)
6027 		return (rc);
6028 
6029 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6030 	if (sb == NULL)
6031 		return (ENOMEM);
6032 
6033 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6034 	    M_ZERO | M_WAITOK);
6035 
6036 	rc = -t4_cim_read_la(sc, buf, NULL);
6037 	if (rc != 0)
6038 		goto done;
6039 
6040 	sbuf_printf(sb, "Status   Data      PC%s",
6041 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6042 	    "     LS0Stat  LS0Addr             LS0Data");
6043 
6044 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
6045 		if (cfg & F_UPDBGLACAPTPCONLY) {
6046 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
6047 			    p[6], p[7]);
6048 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
6049 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
6050 			    p[4] & 0xff, p[5] >> 8);
6051 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
6052 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6053 			    p[1] & 0xf, p[2] >> 4);
6054 		} else {
6055 			sbuf_printf(sb,
6056 			    "\n  %02x   %x%07x %x%07x %08x %08x "
6057 			    "%08x%08x%08x%08x",
6058 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6059 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
6060 			    p[6], p[7]);
6061 		}
6062 	}
6063 
6064 	rc = sbuf_finish(sb);
6065 	sbuf_delete(sb);
6066 done:
6067 	free(buf, M_CXGBE);
6068 	return (rc);
6069 }
6070 
6071 static int
6072 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
6073 {
6074 	struct adapter *sc = arg1;
6075 	u_int cfg;
6076 	struct sbuf *sb;
6077 	uint32_t *buf, *p;
6078 	int rc;
6079 
6080 	MPASS(chip_id(sc) > CHELSIO_T5);
6081 
6082 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6083 	if (rc != 0)
6084 		return (rc);
6085 
6086 	rc = sysctl_wire_old_buffer(req, 0);
6087 	if (rc != 0)
6088 		return (rc);
6089 
6090 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6091 	if (sb == NULL)
6092 		return (ENOMEM);
6093 
6094 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6095 	    M_ZERO | M_WAITOK);
6096 
6097 	rc = -t4_cim_read_la(sc, buf, NULL);
6098 	if (rc != 0)
6099 		goto done;
6100 
6101 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
6102 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6103 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
6104 
6105 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
6106 		if (cfg & F_UPDBGLACAPTPCONLY) {
6107 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
6108 			    p[3] & 0xff, p[2], p[1], p[0]);
6109 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
6110 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
6111 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
6112 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
6113 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
6114 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
6115 			    p[6] >> 16);
6116 		} else {
6117 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
6118 			    "%08x %08x %08x %08x %08x %08x",
6119 			    (p[9] >> 16) & 0xff,
6120 			    p[9] & 0xffff, p[8] >> 16,
6121 			    p[8] & 0xffff, p[7] >> 16,
6122 			    p[7] & 0xffff, p[6] >> 16,
6123 			    p[2], p[1], p[0], p[5], p[4], p[3]);
6124 		}
6125 	}
6126 
6127 	rc = sbuf_finish(sb);
6128 	sbuf_delete(sb);
6129 done:
6130 	free(buf, M_CXGBE);
6131 	return (rc);
6132 }
6133 
6134 static int
6135 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
6136 {
6137 	struct adapter *sc = arg1;
6138 	u_int i;
6139 	struct sbuf *sb;
6140 	uint32_t *buf, *p;
6141 	int rc;
6142 
6143 	rc = sysctl_wire_old_buffer(req, 0);
6144 	if (rc != 0)
6145 		return (rc);
6146 
6147 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6148 	if (sb == NULL)
6149 		return (ENOMEM);
6150 
6151 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
6152 	    M_ZERO | M_WAITOK);
6153 
6154 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
6155 	p = buf;
6156 
6157 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6158 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
6159 		    p[1], p[0]);
6160 	}
6161 
6162 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
6163 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6164 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
6165 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
6166 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
6167 		    (p[1] >> 2) | ((p[2] & 3) << 30),
6168 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
6169 		    p[0] & 1);
6170 	}
6171 
6172 	rc = sbuf_finish(sb);
6173 	sbuf_delete(sb);
6174 	free(buf, M_CXGBE);
6175 	return (rc);
6176 }
6177 
6178 static int
6179 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
6180 {
6181 	struct adapter *sc = arg1;
6182 	u_int i;
6183 	struct sbuf *sb;
6184 	uint32_t *buf, *p;
6185 	int rc;
6186 
6187 	rc = sysctl_wire_old_buffer(req, 0);
6188 	if (rc != 0)
6189 		return (rc);
6190 
6191 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6192 	if (sb == NULL)
6193 		return (ENOMEM);
6194 
6195 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
6196 	    M_ZERO | M_WAITOK);
6197 
6198 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
6199 	p = buf;
6200 
6201 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
6202 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6203 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
6204 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
6205 		    p[4], p[3], p[2], p[1], p[0]);
6206 	}
6207 
6208 	sbuf_printf(sb, "\n\nCntl ID               Data");
6209 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6210 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
6211 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
6212 	}
6213 
6214 	rc = sbuf_finish(sb);
6215 	sbuf_delete(sb);
6216 	free(buf, M_CXGBE);
6217 	return (rc);
6218 }
6219 
6220 static int
6221 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
6222 {
6223 	struct adapter *sc = arg1;
6224 	struct sbuf *sb;
6225 	int rc, i;
6226 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6227 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6228 	uint16_t thres[CIM_NUM_IBQ];
6229 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
6230 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
6231 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
6232 
6233 	cim_num_obq = sc->chip_params->cim_num_obq;
6234 	if (is_t4(sc)) {
6235 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
6236 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
6237 	} else {
6238 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
6239 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
6240 	}
6241 	nq = CIM_NUM_IBQ + cim_num_obq;
6242 
6243 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
6244 	if (rc == 0)
6245 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
6246 	if (rc != 0)
6247 		return (rc);
6248 
6249 	t4_read_cimq_cfg(sc, base, size, thres);
6250 
6251 	rc = sysctl_wire_old_buffer(req, 0);
6252 	if (rc != 0)
6253 		return (rc);
6254 
6255 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6256 	if (sb == NULL)
6257 		return (ENOMEM);
6258 
6259 	sbuf_printf(sb,
6260 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
6261 
6262 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
6263 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
6264 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
6265 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6266 		    G_QUEREMFLITS(p[2]) * 16);
6267 	for ( ; i < nq; i++, p += 4, wr += 2)
6268 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
6269 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
6270 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6271 		    G_QUEREMFLITS(p[2]) * 16);
6272 
6273 	rc = sbuf_finish(sb);
6274 	sbuf_delete(sb);
6275 
6276 	return (rc);
6277 }
6278 
6279 static int
6280 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
6281 {
6282 	struct adapter *sc = arg1;
6283 	struct sbuf *sb;
6284 	int rc;
6285 	struct tp_cpl_stats stats;
6286 
6287 	rc = sysctl_wire_old_buffer(req, 0);
6288 	if (rc != 0)
6289 		return (rc);
6290 
6291 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6292 	if (sb == NULL)
6293 		return (ENOMEM);
6294 
6295 	mtx_lock(&sc->reg_lock);
6296 	t4_tp_get_cpl_stats(sc, &stats);
6297 	mtx_unlock(&sc->reg_lock);
6298 
6299 	if (sc->chip_params->nchan > 2) {
6300 		sbuf_printf(sb, "                 channel 0  channel 1"
6301 		    "  channel 2  channel 3");
6302 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
6303 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
6304 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
6305 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
6306 	} else {
6307 		sbuf_printf(sb, "                 channel 0  channel 1");
6308 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
6309 		    stats.req[0], stats.req[1]);
6310 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
6311 		    stats.rsp[0], stats.rsp[1]);
6312 	}
6313 
6314 	rc = sbuf_finish(sb);
6315 	sbuf_delete(sb);
6316 
6317 	return (rc);
6318 }
6319 
6320 static int
6321 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
6322 {
6323 	struct adapter *sc = arg1;
6324 	struct sbuf *sb;
6325 	int rc;
6326 	struct tp_usm_stats stats;
6327 
6328 	rc = sysctl_wire_old_buffer(req, 0);
6329 	if (rc != 0)
6330 		return(rc);
6331 
6332 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6333 	if (sb == NULL)
6334 		return (ENOMEM);
6335 
6336 	t4_get_usm_stats(sc, &stats);
6337 
6338 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
6339 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
6340 	sbuf_printf(sb, "Drops:  %u", stats.drops);
6341 
6342 	rc = sbuf_finish(sb);
6343 	sbuf_delete(sb);
6344 
6345 	return (rc);
6346 }
6347 
6348 static const char * const devlog_level_strings[] = {
6349 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
6350 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
6351 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
6352 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
6353 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
6354 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
6355 };
6356 
6357 static const char * const devlog_facility_strings[] = {
6358 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
6359 	[FW_DEVLOG_FACILITY_CF]		= "CF",
6360 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
6361 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
6362 	[FW_DEVLOG_FACILITY_RES]	= "RES",
6363 	[FW_DEVLOG_FACILITY_HW]		= "HW",
6364 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
6365 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
6366 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
6367 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
6368 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
6369 	[FW_DEVLOG_FACILITY_VI]		= "VI",
6370 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
6371 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
6372 	[FW_DEVLOG_FACILITY_TM]		= "TM",
6373 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
6374 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
6375 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
6376 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
6377 	[FW_DEVLOG_FACILITY_RI]		= "RI",
6378 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
6379 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
6380 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
6381 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
6382 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
6383 };
6384 
6385 static int
6386 sysctl_devlog(SYSCTL_HANDLER_ARGS)
6387 {
6388 	struct adapter *sc = arg1;
6389 	struct devlog_params *dparams = &sc->params.devlog;
6390 	struct fw_devlog_e *buf, *e;
6391 	int i, j, rc, nentries, first = 0;
6392 	struct sbuf *sb;
6393 	uint64_t ftstamp = UINT64_MAX;
6394 
6395 	if (dparams->addr == 0)
6396 		return (ENXIO);
6397 
6398 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
6399 	if (buf == NULL)
6400 		return (ENOMEM);
6401 
6402 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
6403 	if (rc != 0)
6404 		goto done;
6405 
6406 	nentries = dparams->size / sizeof(struct fw_devlog_e);
6407 	for (i = 0; i < nentries; i++) {
6408 		e = &buf[i];
6409 
6410 		if (e->timestamp == 0)
6411 			break;	/* end */
6412 
6413 		e->timestamp = be64toh(e->timestamp);
6414 		e->seqno = be32toh(e->seqno);
6415 		for (j = 0; j < 8; j++)
6416 			e->params[j] = be32toh(e->params[j]);
6417 
6418 		if (e->timestamp < ftstamp) {
6419 			ftstamp = e->timestamp;
6420 			first = i;
6421 		}
6422 	}
6423 
6424 	if (buf[first].timestamp == 0)
6425 		goto done;	/* nothing in the log */
6426 
6427 	rc = sysctl_wire_old_buffer(req, 0);
6428 	if (rc != 0)
6429 		goto done;
6430 
6431 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6432 	if (sb == NULL) {
6433 		rc = ENOMEM;
6434 		goto done;
6435 	}
6436 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
6437 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
6438 
6439 	i = first;
6440 	do {
6441 		e = &buf[i];
6442 		if (e->timestamp == 0)
6443 			break;	/* end */
6444 
6445 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
6446 		    e->seqno, e->timestamp,
6447 		    (e->level < nitems(devlog_level_strings) ?
6448 			devlog_level_strings[e->level] : "UNKNOWN"),
6449 		    (e->facility < nitems(devlog_facility_strings) ?
6450 			devlog_facility_strings[e->facility] : "UNKNOWN"));
6451 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
6452 		    e->params[2], e->params[3], e->params[4],
6453 		    e->params[5], e->params[6], e->params[7]);
6454 
6455 		if (++i == nentries)
6456 			i = 0;
6457 	} while (i != first);
6458 
6459 	rc = sbuf_finish(sb);
6460 	sbuf_delete(sb);
6461 done:
6462 	free(buf, M_CXGBE);
6463 	return (rc);
6464 }
6465 
6466 static int
6467 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
6468 {
6469 	struct adapter *sc = arg1;
6470 	struct sbuf *sb;
6471 	int rc;
6472 	struct tp_fcoe_stats stats[MAX_NCHAN];
6473 	int i, nchan = sc->chip_params->nchan;
6474 
6475 	rc = sysctl_wire_old_buffer(req, 0);
6476 	if (rc != 0)
6477 		return (rc);
6478 
6479 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6480 	if (sb == NULL)
6481 		return (ENOMEM);
6482 
6483 	for (i = 0; i < nchan; i++)
6484 		t4_get_fcoe_stats(sc, i, &stats[i]);
6485 
6486 	if (nchan > 2) {
6487 		sbuf_printf(sb, "                   channel 0        channel 1"
6488 		    "        channel 2        channel 3");
6489 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
6490 		    stats[0].octets_ddp, stats[1].octets_ddp,
6491 		    stats[2].octets_ddp, stats[3].octets_ddp);
6492 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
6493 		    stats[0].frames_ddp, stats[1].frames_ddp,
6494 		    stats[2].frames_ddp, stats[3].frames_ddp);
6495 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
6496 		    stats[0].frames_drop, stats[1].frames_drop,
6497 		    stats[2].frames_drop, stats[3].frames_drop);
6498 	} else {
6499 		sbuf_printf(sb, "                   channel 0        channel 1");
6500 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
6501 		    stats[0].octets_ddp, stats[1].octets_ddp);
6502 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
6503 		    stats[0].frames_ddp, stats[1].frames_ddp);
6504 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
6505 		    stats[0].frames_drop, stats[1].frames_drop);
6506 	}
6507 
6508 	rc = sbuf_finish(sb);
6509 	sbuf_delete(sb);
6510 
6511 	return (rc);
6512 }
6513 
6514 static int
6515 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
6516 {
6517 	struct adapter *sc = arg1;
6518 	struct sbuf *sb;
6519 	int rc, i;
6520 	unsigned int map, kbps, ipg, mode;
6521 	unsigned int pace_tab[NTX_SCHED];
6522 
6523 	rc = sysctl_wire_old_buffer(req, 0);
6524 	if (rc != 0)
6525 		return (rc);
6526 
6527 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6528 	if (sb == NULL)
6529 		return (ENOMEM);
6530 
6531 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
6532 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
6533 	t4_read_pace_tbl(sc, pace_tab);
6534 
6535 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
6536 	    "Class IPG (0.1 ns)   Flow IPG (us)");
6537 
6538 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
6539 		t4_get_tx_sched(sc, i, &kbps, &ipg);
6540 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
6541 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
6542 		if (kbps)
6543 			sbuf_printf(sb, "%9u     ", kbps);
6544 		else
6545 			sbuf_printf(sb, " disabled     ");
6546 
6547 		if (ipg)
6548 			sbuf_printf(sb, "%13u        ", ipg);
6549 		else
6550 			sbuf_printf(sb, "     disabled        ");
6551 
6552 		if (pace_tab[i])
6553 			sbuf_printf(sb, "%10u", pace_tab[i]);
6554 		else
6555 			sbuf_printf(sb, "  disabled");
6556 	}
6557 
6558 	rc = sbuf_finish(sb);
6559 	sbuf_delete(sb);
6560 
6561 	return (rc);
6562 }
6563 
6564 static int
6565 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
6566 {
6567 	struct adapter *sc = arg1;
6568 	struct sbuf *sb;
6569 	int rc, i, j;
6570 	uint64_t *p0, *p1;
6571 	struct lb_port_stats s[2];
6572 	static const char *stat_name[] = {
6573 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
6574 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
6575 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
6576 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
6577 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
6578 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
6579 		"BG2FramesTrunc:", "BG3FramesTrunc:"
6580 	};
6581 
6582 	rc = sysctl_wire_old_buffer(req, 0);
6583 	if (rc != 0)
6584 		return (rc);
6585 
6586 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6587 	if (sb == NULL)
6588 		return (ENOMEM);
6589 
6590 	memset(s, 0, sizeof(s));
6591 
6592 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
6593 		t4_get_lb_stats(sc, i, &s[0]);
6594 		t4_get_lb_stats(sc, i + 1, &s[1]);
6595 
6596 		p0 = &s[0].octets;
6597 		p1 = &s[1].octets;
6598 		sbuf_printf(sb, "%s                       Loopback %u"
6599 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
6600 
6601 		for (j = 0; j < nitems(stat_name); j++)
6602 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
6603 				   *p0++, *p1++);
6604 	}
6605 
6606 	rc = sbuf_finish(sb);
6607 	sbuf_delete(sb);
6608 
6609 	return (rc);
6610 }
6611 
6612 static int
6613 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
6614 {
6615 	int rc = 0;
6616 	struct port_info *pi = arg1;
6617 	struct link_config *lc = &pi->link_cfg;
6618 	struct sbuf *sb;
6619 
6620 	rc = sysctl_wire_old_buffer(req, 0);
6621 	if (rc != 0)
6622 		return(rc);
6623 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
6624 	if (sb == NULL)
6625 		return (ENOMEM);
6626 
6627 	if (lc->link_ok || lc->link_down_rc == 255)
6628 		sbuf_printf(sb, "n/a");
6629 	else
6630 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
6631 
6632 	rc = sbuf_finish(sb);
6633 	sbuf_delete(sb);
6634 
6635 	return (rc);
6636 }
6637 
6638 struct mem_desc {
6639 	unsigned int base;
6640 	unsigned int limit;
6641 	unsigned int idx;
6642 };
6643 
6644 static int
6645 mem_desc_cmp(const void *a, const void *b)
6646 {
6647 	return ((const struct mem_desc *)a)->base -
6648 	       ((const struct mem_desc *)b)->base;
6649 }
6650 
6651 static void
6652 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
6653     unsigned int to)
6654 {
6655 	unsigned int size;
6656 
6657 	if (from == to)
6658 		return;
6659 
6660 	size = to - from + 1;
6661 	if (size == 0)
6662 		return;
6663 
6664 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
6665 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
6666 }
6667 
6668 static int
6669 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
6670 {
6671 	struct adapter *sc = arg1;
6672 	struct sbuf *sb;
6673 	int rc, i, n;
6674 	uint32_t lo, hi, used, alloc;
6675 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
6676 	static const char *region[] = {
6677 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
6678 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
6679 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
6680 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
6681 		"RQUDP region:", "PBL region:", "TXPBL region:",
6682 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
6683 		"On-chip queues:"
6684 	};
6685 	struct mem_desc avail[4];
6686 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
6687 	struct mem_desc *md = mem;
6688 
6689 	rc = sysctl_wire_old_buffer(req, 0);
6690 	if (rc != 0)
6691 		return (rc);
6692 
6693 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6694 	if (sb == NULL)
6695 		return (ENOMEM);
6696 
6697 	for (i = 0; i < nitems(mem); i++) {
6698 		mem[i].limit = 0;
6699 		mem[i].idx = i;
6700 	}
6701 
6702 	/* Find and sort the populated memory ranges */
6703 	i = 0;
6704 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
6705 	if (lo & F_EDRAM0_ENABLE) {
6706 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
6707 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
6708 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
6709 		avail[i].idx = 0;
6710 		i++;
6711 	}
6712 	if (lo & F_EDRAM1_ENABLE) {
6713 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
6714 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
6715 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
6716 		avail[i].idx = 1;
6717 		i++;
6718 	}
6719 	if (lo & F_EXT_MEM_ENABLE) {
6720 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
6721 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
6722 		avail[i].limit = avail[i].base +
6723 		    (G_EXT_MEM_SIZE(hi) << 20);
6724 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
6725 		i++;
6726 	}
6727 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
6728 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
6729 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
6730 		avail[i].limit = avail[i].base +
6731 		    (G_EXT_MEM1_SIZE(hi) << 20);
6732 		avail[i].idx = 4;
6733 		i++;
6734 	}
6735 	if (!i)                                    /* no memory available */
6736 		return 0;
6737 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
6738 
6739 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
6740 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
6741 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
6742 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
6743 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
6744 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
6745 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
6746 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
6747 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
6748 
6749 	/* the next few have explicit upper bounds */
6750 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
6751 	md->limit = md->base - 1 +
6752 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
6753 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
6754 	md++;
6755 
6756 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
6757 	md->limit = md->base - 1 +
6758 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
6759 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
6760 	md++;
6761 
6762 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
6763 		if (chip_id(sc) <= CHELSIO_T5)
6764 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
6765 		else
6766 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
6767 		md->limit = 0;
6768 	} else {
6769 		md->base = 0;
6770 		md->idx = nitems(region);  /* hide it */
6771 	}
6772 	md++;
6773 
6774 #define ulp_region(reg) \
6775 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
6776 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
6777 
6778 	ulp_region(RX_ISCSI);
6779 	ulp_region(RX_TDDP);
6780 	ulp_region(TX_TPT);
6781 	ulp_region(RX_STAG);
6782 	ulp_region(RX_RQ);
6783 	ulp_region(RX_RQUDP);
6784 	ulp_region(RX_PBL);
6785 	ulp_region(TX_PBL);
6786 #undef ulp_region
6787 
6788 	md->base = 0;
6789 	md->idx = nitems(region);
6790 	if (!is_t4(sc)) {
6791 		uint32_t size = 0;
6792 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
6793 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
6794 
6795 		if (is_t5(sc)) {
6796 			if (sge_ctrl & F_VFIFO_ENABLE)
6797 				size = G_DBVFIFO_SIZE(fifo_size);
6798 		} else
6799 			size = G_T6_DBVFIFO_SIZE(fifo_size);
6800 
6801 		if (size) {
6802 			md->base = G_BASEADDR(t4_read_reg(sc,
6803 			    A_SGE_DBVFIFO_BADDR));
6804 			md->limit = md->base + (size << 2) - 1;
6805 		}
6806 	}
6807 	md++;
6808 
6809 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
6810 	md->limit = 0;
6811 	md++;
6812 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
6813 	md->limit = 0;
6814 	md++;
6815 
6816 	md->base = sc->vres.ocq.start;
6817 	if (sc->vres.ocq.size)
6818 		md->limit = md->base + sc->vres.ocq.size - 1;
6819 	else
6820 		md->idx = nitems(region);  /* hide it */
6821 	md++;
6822 
6823 	/* add any address-space holes, there can be up to 3 */
6824 	for (n = 0; n < i - 1; n++)
6825 		if (avail[n].limit < avail[n + 1].base)
6826 			(md++)->base = avail[n].limit;
6827 	if (avail[n].limit)
6828 		(md++)->base = avail[n].limit;
6829 
6830 	n = md - mem;
6831 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
6832 
6833 	for (lo = 0; lo < i; lo++)
6834 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
6835 				avail[lo].limit - 1);
6836 
6837 	sbuf_printf(sb, "\n");
6838 	for (i = 0; i < n; i++) {
6839 		if (mem[i].idx >= nitems(region))
6840 			continue;                        /* skip holes */
6841 		if (!mem[i].limit)
6842 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
6843 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
6844 				mem[i].limit);
6845 	}
6846 
6847 	sbuf_printf(sb, "\n");
6848 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
6849 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
6850 	mem_region_show(sb, "uP RAM:", lo, hi);
6851 
6852 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
6853 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
6854 	mem_region_show(sb, "uP Extmem2:", lo, hi);
6855 
6856 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
6857 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
6858 		   G_PMRXMAXPAGE(lo),
6859 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
6860 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
6861 
6862 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
6863 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
6864 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
6865 		   G_PMTXMAXPAGE(lo),
6866 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
6867 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
6868 	sbuf_printf(sb, "%u p-structs\n",
6869 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
6870 
6871 	for (i = 0; i < 4; i++) {
6872 		if (chip_id(sc) > CHELSIO_T5)
6873 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
6874 		else
6875 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
6876 		if (is_t5(sc)) {
6877 			used = G_T5_USED(lo);
6878 			alloc = G_T5_ALLOC(lo);
6879 		} else {
6880 			used = G_USED(lo);
6881 			alloc = G_ALLOC(lo);
6882 		}
6883 		/* For T6 these are MAC buffer groups */
6884 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
6885 		    i, used, alloc);
6886 	}
6887 	for (i = 0; i < sc->chip_params->nchan; i++) {
6888 		if (chip_id(sc) > CHELSIO_T5)
6889 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
6890 		else
6891 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
6892 		if (is_t5(sc)) {
6893 			used = G_T5_USED(lo);
6894 			alloc = G_T5_ALLOC(lo);
6895 		} else {
6896 			used = G_USED(lo);
6897 			alloc = G_ALLOC(lo);
6898 		}
6899 		/* For T6 these are MAC buffer groups */
6900 		sbuf_printf(sb,
6901 		    "\nLoopback %d using %u pages out of %u allocated",
6902 		    i, used, alloc);
6903 	}
6904 
6905 	rc = sbuf_finish(sb);
6906 	sbuf_delete(sb);
6907 
6908 	return (rc);
6909 }
6910 
6911 static inline void
6912 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
6913 {
6914 	*mask = x | y;
6915 	y = htobe64(y);
6916 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
6917 }
6918 
6919 static int
6920 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
6921 {
6922 	struct adapter *sc = arg1;
6923 	struct sbuf *sb;
6924 	int rc, i;
6925 
6926 	MPASS(chip_id(sc) <= CHELSIO_T5);
6927 
6928 	rc = sysctl_wire_old_buffer(req, 0);
6929 	if (rc != 0)
6930 		return (rc);
6931 
6932 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6933 	if (sb == NULL)
6934 		return (ENOMEM);
6935 
6936 	sbuf_printf(sb,
6937 	    "Idx  Ethernet address     Mask     Vld Ports PF"
6938 	    "  VF              Replication             P0 P1 P2 P3  ML");
6939 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
6940 		uint64_t tcamx, tcamy, mask;
6941 		uint32_t cls_lo, cls_hi;
6942 		uint8_t addr[ETHER_ADDR_LEN];
6943 
6944 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
6945 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
6946 		if (tcamx & tcamy)
6947 			continue;
6948 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
6949 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
6950 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
6951 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
6952 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
6953 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
6954 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
6955 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
6956 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
6957 
6958 		if (cls_lo & F_REPLICATE) {
6959 			struct fw_ldst_cmd ldst_cmd;
6960 
6961 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
6962 			ldst_cmd.op_to_addrspace =
6963 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
6964 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
6965 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
6966 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
6967 			ldst_cmd.u.mps.rplc.fid_idx =
6968 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
6969 				V_FW_LDST_CMD_IDX(i));
6970 
6971 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
6972 			    "t4mps");
6973 			if (rc)
6974 				break;
6975 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
6976 			    sizeof(ldst_cmd), &ldst_cmd);
6977 			end_synchronized_op(sc, 0);
6978 
6979 			if (rc != 0) {
6980 				sbuf_printf(sb, "%36d", rc);
6981 				rc = 0;
6982 			} else {
6983 				sbuf_printf(sb, " %08x %08x %08x %08x",
6984 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
6985 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
6986 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
6987 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
6988 			}
6989 		} else
6990 			sbuf_printf(sb, "%36s", "");
6991 
6992 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
6993 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
6994 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
6995 	}
6996 
6997 	if (rc)
6998 		(void) sbuf_finish(sb);
6999 	else
7000 		rc = sbuf_finish(sb);
7001 	sbuf_delete(sb);
7002 
7003 	return (rc);
7004 }
7005 
7006 static int
7007 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
7008 {
7009 	struct adapter *sc = arg1;
7010 	struct sbuf *sb;
7011 	int rc, i;
7012 
7013 	MPASS(chip_id(sc) > CHELSIO_T5);
7014 
7015 	rc = sysctl_wire_old_buffer(req, 0);
7016 	if (rc != 0)
7017 		return (rc);
7018 
7019 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7020 	if (sb == NULL)
7021 		return (ENOMEM);
7022 
7023 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
7024 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
7025 	    "                           Replication"
7026 	    "                                    P0 P1 P2 P3  ML\n");
7027 
7028 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7029 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
7030 		uint16_t ivlan;
7031 		uint64_t tcamx, tcamy, val, mask;
7032 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
7033 		uint8_t addr[ETHER_ADDR_LEN];
7034 
7035 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
7036 		if (i < 256)
7037 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
7038 		else
7039 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
7040 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7041 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7042 		tcamy = G_DMACH(val) << 32;
7043 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7044 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7045 		lookup_type = G_DATALKPTYPE(data2);
7046 		port_num = G_DATAPORTNUM(data2);
7047 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7048 			/* Inner header VNI */
7049 			vniy = ((data2 & F_DATAVIDH2) << 23) |
7050 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7051 			dip_hit = data2 & F_DATADIPHIT;
7052 			vlan_vld = 0;
7053 		} else {
7054 			vniy = 0;
7055 			dip_hit = 0;
7056 			vlan_vld = data2 & F_DATAVIDH2;
7057 			ivlan = G_VIDL(val);
7058 		}
7059 
7060 		ctl |= V_CTLXYBITSEL(1);
7061 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7062 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7063 		tcamx = G_DMACH(val) << 32;
7064 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7065 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7066 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7067 			/* Inner header VNI mask */
7068 			vnix = ((data2 & F_DATAVIDH2) << 23) |
7069 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7070 		} else
7071 			vnix = 0;
7072 
7073 		if (tcamx & tcamy)
7074 			continue;
7075 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7076 
7077 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7078 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7079 
7080 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7081 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7082 			    "%012jx %06x %06x    -    -   %3c"
7083 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
7084 			    addr[1], addr[2], addr[3], addr[4], addr[5],
7085 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
7086 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7087 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7088 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7089 		} else {
7090 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7091 			    "%012jx    -       -   ", i, addr[0], addr[1],
7092 			    addr[2], addr[3], addr[4], addr[5],
7093 			    (uintmax_t)mask);
7094 
7095 			if (vlan_vld)
7096 				sbuf_printf(sb, "%4u   Y     ", ivlan);
7097 			else
7098 				sbuf_printf(sb, "  -    N     ");
7099 
7100 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
7101 			    lookup_type ? 'I' : 'O', port_num,
7102 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7103 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7104 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7105 		}
7106 
7107 
7108 		if (cls_lo & F_T6_REPLICATE) {
7109 			struct fw_ldst_cmd ldst_cmd;
7110 
7111 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7112 			ldst_cmd.op_to_addrspace =
7113 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7114 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7115 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7116 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7117 			ldst_cmd.u.mps.rplc.fid_idx =
7118 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7119 				V_FW_LDST_CMD_IDX(i));
7120 
7121 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7122 			    "t6mps");
7123 			if (rc)
7124 				break;
7125 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7126 			    sizeof(ldst_cmd), &ldst_cmd);
7127 			end_synchronized_op(sc, 0);
7128 
7129 			if (rc != 0) {
7130 				sbuf_printf(sb, "%72d", rc);
7131 				rc = 0;
7132 			} else {
7133 				sbuf_printf(sb, " %08x %08x %08x %08x"
7134 				    " %08x %08x %08x %08x",
7135 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
7136 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
7137 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
7138 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
7139 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7140 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7141 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7142 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7143 			}
7144 		} else
7145 			sbuf_printf(sb, "%72s", "");
7146 
7147 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
7148 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
7149 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
7150 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
7151 	}
7152 
7153 	if (rc)
7154 		(void) sbuf_finish(sb);
7155 	else
7156 		rc = sbuf_finish(sb);
7157 	sbuf_delete(sb);
7158 
7159 	return (rc);
7160 }
7161 
7162 static int
7163 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
7164 {
7165 	struct adapter *sc = arg1;
7166 	struct sbuf *sb;
7167 	int rc;
7168 	uint16_t mtus[NMTUS];
7169 
7170 	rc = sysctl_wire_old_buffer(req, 0);
7171 	if (rc != 0)
7172 		return (rc);
7173 
7174 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7175 	if (sb == NULL)
7176 		return (ENOMEM);
7177 
7178 	t4_read_mtu_tbl(sc, mtus, NULL);
7179 
7180 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
7181 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
7182 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
7183 	    mtus[14], mtus[15]);
7184 
7185 	rc = sbuf_finish(sb);
7186 	sbuf_delete(sb);
7187 
7188 	return (rc);
7189 }
7190 
7191 static int
7192 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
7193 {
7194 	struct adapter *sc = arg1;
7195 	struct sbuf *sb;
7196 	int rc, i;
7197 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
7198 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
7199 	static const char *tx_stats[MAX_PM_NSTATS] = {
7200 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
7201 		"Tx FIFO wait", NULL, "Tx latency"
7202 	};
7203 	static const char *rx_stats[MAX_PM_NSTATS] = {
7204 		"Read:", "Write bypass:", "Write mem:", "Flush:",
7205 		"Rx FIFO wait", NULL, "Rx latency"
7206 	};
7207 
7208 	rc = sysctl_wire_old_buffer(req, 0);
7209 	if (rc != 0)
7210 		return (rc);
7211 
7212 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7213 	if (sb == NULL)
7214 		return (ENOMEM);
7215 
7216 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
7217 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
7218 
7219 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
7220 	for (i = 0; i < 4; i++) {
7221 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7222 		    tx_cyc[i]);
7223 	}
7224 
7225 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
7226 	for (i = 0; i < 4; i++) {
7227 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7228 		    rx_cyc[i]);
7229 	}
7230 
7231 	if (chip_id(sc) > CHELSIO_T5) {
7232 		sbuf_printf(sb,
7233 		    "\n              Total wait      Total occupancy");
7234 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7235 		    tx_cyc[i]);
7236 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7237 		    rx_cyc[i]);
7238 
7239 		i += 2;
7240 		MPASS(i < nitems(tx_stats));
7241 
7242 		sbuf_printf(sb,
7243 		    "\n                   Reads           Total wait");
7244 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7245 		    tx_cyc[i]);
7246 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7247 		    rx_cyc[i]);
7248 	}
7249 
7250 	rc = sbuf_finish(sb);
7251 	sbuf_delete(sb);
7252 
7253 	return (rc);
7254 }
7255 
7256 static int
7257 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
7258 {
7259 	struct adapter *sc = arg1;
7260 	struct sbuf *sb;
7261 	int rc;
7262 	struct tp_rdma_stats stats;
7263 
7264 	rc = sysctl_wire_old_buffer(req, 0);
7265 	if (rc != 0)
7266 		return (rc);
7267 
7268 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7269 	if (sb == NULL)
7270 		return (ENOMEM);
7271 
7272 	mtx_lock(&sc->reg_lock);
7273 	t4_tp_get_rdma_stats(sc, &stats);
7274 	mtx_unlock(&sc->reg_lock);
7275 
7276 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
7277 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
7278 
7279 	rc = sbuf_finish(sb);
7280 	sbuf_delete(sb);
7281 
7282 	return (rc);
7283 }
7284 
7285 static int
7286 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
7287 {
7288 	struct adapter *sc = arg1;
7289 	struct sbuf *sb;
7290 	int rc;
7291 	struct tp_tcp_stats v4, v6;
7292 
7293 	rc = sysctl_wire_old_buffer(req, 0);
7294 	if (rc != 0)
7295 		return (rc);
7296 
7297 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7298 	if (sb == NULL)
7299 		return (ENOMEM);
7300 
7301 	mtx_lock(&sc->reg_lock);
7302 	t4_tp_get_tcp_stats(sc, &v4, &v6);
7303 	mtx_unlock(&sc->reg_lock);
7304 
7305 	sbuf_printf(sb,
7306 	    "                                IP                 IPv6\n");
7307 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
7308 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
7309 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
7310 	    v4.tcp_in_segs, v6.tcp_in_segs);
7311 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
7312 	    v4.tcp_out_segs, v6.tcp_out_segs);
7313 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
7314 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
7315 
7316 	rc = sbuf_finish(sb);
7317 	sbuf_delete(sb);
7318 
7319 	return (rc);
7320 }
7321 
7322 static int
7323 sysctl_tids(SYSCTL_HANDLER_ARGS)
7324 {
7325 	struct adapter *sc = arg1;
7326 	struct sbuf *sb;
7327 	int rc;
7328 	struct tid_info *t = &sc->tids;
7329 
7330 	rc = sysctl_wire_old_buffer(req, 0);
7331 	if (rc != 0)
7332 		return (rc);
7333 
7334 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7335 	if (sb == NULL)
7336 		return (ENOMEM);
7337 
7338 	if (t->natids) {
7339 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
7340 		    t->atids_in_use);
7341 	}
7342 
7343 	if (t->ntids) {
7344 		sbuf_printf(sb, "TID range: ");
7345 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7346 			uint32_t b, hb;
7347 
7348 			if (chip_id(sc) <= CHELSIO_T5) {
7349 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
7350 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
7351 			} else {
7352 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
7353 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
7354 			}
7355 
7356 			if (b)
7357 				sbuf_printf(sb, "0-%u, ", b - 1);
7358 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
7359 		} else
7360 			sbuf_printf(sb, "0-%u", t->ntids - 1);
7361 		sbuf_printf(sb, ", in use: %u\n",
7362 		    atomic_load_acq_int(&t->tids_in_use));
7363 	}
7364 
7365 	if (t->nstids) {
7366 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
7367 		    t->stid_base + t->nstids - 1, t->stids_in_use);
7368 	}
7369 
7370 	if (t->nftids) {
7371 		sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
7372 		    t->ftid_base + t->nftids - 1);
7373 	}
7374 
7375 	if (t->netids) {
7376 		sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
7377 		    t->etid_base + t->netids - 1);
7378 	}
7379 
7380 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
7381 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
7382 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
7383 
7384 	rc = sbuf_finish(sb);
7385 	sbuf_delete(sb);
7386 
7387 	return (rc);
7388 }
7389 
7390 static int
7391 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
7392 {
7393 	struct adapter *sc = arg1;
7394 	struct sbuf *sb;
7395 	int rc;
7396 	struct tp_err_stats stats;
7397 
7398 	rc = sysctl_wire_old_buffer(req, 0);
7399 	if (rc != 0)
7400 		return (rc);
7401 
7402 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7403 	if (sb == NULL)
7404 		return (ENOMEM);
7405 
7406 	mtx_lock(&sc->reg_lock);
7407 	t4_tp_get_err_stats(sc, &stats);
7408 	mtx_unlock(&sc->reg_lock);
7409 
7410 	if (sc->chip_params->nchan > 2) {
7411 		sbuf_printf(sb, "                 channel 0  channel 1"
7412 		    "  channel 2  channel 3\n");
7413 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
7414 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
7415 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
7416 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
7417 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
7418 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
7419 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
7420 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
7421 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
7422 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
7423 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
7424 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
7425 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
7426 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
7427 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
7428 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
7429 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
7430 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
7431 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
7432 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
7433 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
7434 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
7435 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
7436 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
7437 	} else {
7438 		sbuf_printf(sb, "                 channel 0  channel 1\n");
7439 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
7440 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
7441 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
7442 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
7443 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
7444 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
7445 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
7446 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
7447 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
7448 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
7449 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
7450 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
7451 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
7452 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
7453 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
7454 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
7455 	}
7456 
7457 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
7458 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
7459 
7460 	rc = sbuf_finish(sb);
7461 	sbuf_delete(sb);
7462 
7463 	return (rc);
7464 }
7465 
7466 static int
7467 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
7468 {
7469 	struct adapter *sc = arg1;
7470 	struct tp_params *tpp = &sc->params.tp;
7471 	u_int mask;
7472 	int rc;
7473 
7474 	mask = tpp->la_mask >> 16;
7475 	rc = sysctl_handle_int(oidp, &mask, 0, req);
7476 	if (rc != 0 || req->newptr == NULL)
7477 		return (rc);
7478 	if (mask > 0xffff)
7479 		return (EINVAL);
7480 	tpp->la_mask = mask << 16;
7481 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
7482 
7483 	return (0);
7484 }
7485 
7486 struct field_desc {
7487 	const char *name;
7488 	u_int start;
7489 	u_int width;
7490 };
7491 
7492 static void
7493 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
7494 {
7495 	char buf[32];
7496 	int line_size = 0;
7497 
7498 	while (f->name) {
7499 		uint64_t mask = (1ULL << f->width) - 1;
7500 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
7501 		    ((uintmax_t)v >> f->start) & mask);
7502 
7503 		if (line_size + len >= 79) {
7504 			line_size = 8;
7505 			sbuf_printf(sb, "\n        ");
7506 		}
7507 		sbuf_printf(sb, "%s ", buf);
7508 		line_size += len + 1;
7509 		f++;
7510 	}
7511 	sbuf_printf(sb, "\n");
7512 }
7513 
7514 static const struct field_desc tp_la0[] = {
7515 	{ "RcfOpCodeOut", 60, 4 },
7516 	{ "State", 56, 4 },
7517 	{ "WcfState", 52, 4 },
7518 	{ "RcfOpcSrcOut", 50, 2 },
7519 	{ "CRxError", 49, 1 },
7520 	{ "ERxError", 48, 1 },
7521 	{ "SanityFailed", 47, 1 },
7522 	{ "SpuriousMsg", 46, 1 },
7523 	{ "FlushInputMsg", 45, 1 },
7524 	{ "FlushInputCpl", 44, 1 },
7525 	{ "RssUpBit", 43, 1 },
7526 	{ "RssFilterHit", 42, 1 },
7527 	{ "Tid", 32, 10 },
7528 	{ "InitTcb", 31, 1 },
7529 	{ "LineNumber", 24, 7 },
7530 	{ "Emsg", 23, 1 },
7531 	{ "EdataOut", 22, 1 },
7532 	{ "Cmsg", 21, 1 },
7533 	{ "CdataOut", 20, 1 },
7534 	{ "EreadPdu", 19, 1 },
7535 	{ "CreadPdu", 18, 1 },
7536 	{ "TunnelPkt", 17, 1 },
7537 	{ "RcfPeerFin", 16, 1 },
7538 	{ "RcfReasonOut", 12, 4 },
7539 	{ "TxCchannel", 10, 2 },
7540 	{ "RcfTxChannel", 8, 2 },
7541 	{ "RxEchannel", 6, 2 },
7542 	{ "RcfRxChannel", 5, 1 },
7543 	{ "RcfDataOutSrdy", 4, 1 },
7544 	{ "RxDvld", 3, 1 },
7545 	{ "RxOoDvld", 2, 1 },
7546 	{ "RxCongestion", 1, 1 },
7547 	{ "TxCongestion", 0, 1 },
7548 	{ NULL }
7549 };
7550 
7551 static const struct field_desc tp_la1[] = {
7552 	{ "CplCmdIn", 56, 8 },
7553 	{ "CplCmdOut", 48, 8 },
7554 	{ "ESynOut", 47, 1 },
7555 	{ "EAckOut", 46, 1 },
7556 	{ "EFinOut", 45, 1 },
7557 	{ "ERstOut", 44, 1 },
7558 	{ "SynIn", 43, 1 },
7559 	{ "AckIn", 42, 1 },
7560 	{ "FinIn", 41, 1 },
7561 	{ "RstIn", 40, 1 },
7562 	{ "DataIn", 39, 1 },
7563 	{ "DataInVld", 38, 1 },
7564 	{ "PadIn", 37, 1 },
7565 	{ "RxBufEmpty", 36, 1 },
7566 	{ "RxDdp", 35, 1 },
7567 	{ "RxFbCongestion", 34, 1 },
7568 	{ "TxFbCongestion", 33, 1 },
7569 	{ "TxPktSumSrdy", 32, 1 },
7570 	{ "RcfUlpType", 28, 4 },
7571 	{ "Eread", 27, 1 },
7572 	{ "Ebypass", 26, 1 },
7573 	{ "Esave", 25, 1 },
7574 	{ "Static0", 24, 1 },
7575 	{ "Cread", 23, 1 },
7576 	{ "Cbypass", 22, 1 },
7577 	{ "Csave", 21, 1 },
7578 	{ "CPktOut", 20, 1 },
7579 	{ "RxPagePoolFull", 18, 2 },
7580 	{ "RxLpbkPkt", 17, 1 },
7581 	{ "TxLpbkPkt", 16, 1 },
7582 	{ "RxVfValid", 15, 1 },
7583 	{ "SynLearned", 14, 1 },
7584 	{ "SetDelEntry", 13, 1 },
7585 	{ "SetInvEntry", 12, 1 },
7586 	{ "CpcmdDvld", 11, 1 },
7587 	{ "CpcmdSave", 10, 1 },
7588 	{ "RxPstructsFull", 8, 2 },
7589 	{ "EpcmdDvld", 7, 1 },
7590 	{ "EpcmdFlush", 6, 1 },
7591 	{ "EpcmdTrimPrefix", 5, 1 },
7592 	{ "EpcmdTrimPostfix", 4, 1 },
7593 	{ "ERssIp4Pkt", 3, 1 },
7594 	{ "ERssIp6Pkt", 2, 1 },
7595 	{ "ERssTcpUdpPkt", 1, 1 },
7596 	{ "ERssFceFipPkt", 0, 1 },
7597 	{ NULL }
7598 };
7599 
7600 static const struct field_desc tp_la2[] = {
7601 	{ "CplCmdIn", 56, 8 },
7602 	{ "MpsVfVld", 55, 1 },
7603 	{ "MpsPf", 52, 3 },
7604 	{ "MpsVf", 44, 8 },
7605 	{ "SynIn", 43, 1 },
7606 	{ "AckIn", 42, 1 },
7607 	{ "FinIn", 41, 1 },
7608 	{ "RstIn", 40, 1 },
7609 	{ "DataIn", 39, 1 },
7610 	{ "DataInVld", 38, 1 },
7611 	{ "PadIn", 37, 1 },
7612 	{ "RxBufEmpty", 36, 1 },
7613 	{ "RxDdp", 35, 1 },
7614 	{ "RxFbCongestion", 34, 1 },
7615 	{ "TxFbCongestion", 33, 1 },
7616 	{ "TxPktSumSrdy", 32, 1 },
7617 	{ "RcfUlpType", 28, 4 },
7618 	{ "Eread", 27, 1 },
7619 	{ "Ebypass", 26, 1 },
7620 	{ "Esave", 25, 1 },
7621 	{ "Static0", 24, 1 },
7622 	{ "Cread", 23, 1 },
7623 	{ "Cbypass", 22, 1 },
7624 	{ "Csave", 21, 1 },
7625 	{ "CPktOut", 20, 1 },
7626 	{ "RxPagePoolFull", 18, 2 },
7627 	{ "RxLpbkPkt", 17, 1 },
7628 	{ "TxLpbkPkt", 16, 1 },
7629 	{ "RxVfValid", 15, 1 },
7630 	{ "SynLearned", 14, 1 },
7631 	{ "SetDelEntry", 13, 1 },
7632 	{ "SetInvEntry", 12, 1 },
7633 	{ "CpcmdDvld", 11, 1 },
7634 	{ "CpcmdSave", 10, 1 },
7635 	{ "RxPstructsFull", 8, 2 },
7636 	{ "EpcmdDvld", 7, 1 },
7637 	{ "EpcmdFlush", 6, 1 },
7638 	{ "EpcmdTrimPrefix", 5, 1 },
7639 	{ "EpcmdTrimPostfix", 4, 1 },
7640 	{ "ERssIp4Pkt", 3, 1 },
7641 	{ "ERssIp6Pkt", 2, 1 },
7642 	{ "ERssTcpUdpPkt", 1, 1 },
7643 	{ "ERssFceFipPkt", 0, 1 },
7644 	{ NULL }
7645 };
7646 
7647 static void
7648 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
7649 {
7650 
7651 	field_desc_show(sb, *p, tp_la0);
7652 }
7653 
7654 static void
7655 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
7656 {
7657 
7658 	if (idx)
7659 		sbuf_printf(sb, "\n");
7660 	field_desc_show(sb, p[0], tp_la0);
7661 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7662 		field_desc_show(sb, p[1], tp_la0);
7663 }
7664 
7665 static void
7666 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
7667 {
7668 
7669 	if (idx)
7670 		sbuf_printf(sb, "\n");
7671 	field_desc_show(sb, p[0], tp_la0);
7672 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7673 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
7674 }
7675 
7676 static int
7677 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
7678 {
7679 	struct adapter *sc = arg1;
7680 	struct sbuf *sb;
7681 	uint64_t *buf, *p;
7682 	int rc;
7683 	u_int i, inc;
7684 	void (*show_func)(struct sbuf *, uint64_t *, int);
7685 
7686 	rc = sysctl_wire_old_buffer(req, 0);
7687 	if (rc != 0)
7688 		return (rc);
7689 
7690 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7691 	if (sb == NULL)
7692 		return (ENOMEM);
7693 
7694 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
7695 
7696 	t4_tp_read_la(sc, buf, NULL);
7697 	p = buf;
7698 
7699 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
7700 	case 2:
7701 		inc = 2;
7702 		show_func = tp_la_show2;
7703 		break;
7704 	case 3:
7705 		inc = 2;
7706 		show_func = tp_la_show3;
7707 		break;
7708 	default:
7709 		inc = 1;
7710 		show_func = tp_la_show;
7711 	}
7712 
7713 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
7714 		(*show_func)(sb, p, i);
7715 
7716 	rc = sbuf_finish(sb);
7717 	sbuf_delete(sb);
7718 	free(buf, M_CXGBE);
7719 	return (rc);
7720 }
7721 
7722 static int
7723 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
7724 {
7725 	struct adapter *sc = arg1;
7726 	struct sbuf *sb;
7727 	int rc;
7728 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
7729 
7730 	rc = sysctl_wire_old_buffer(req, 0);
7731 	if (rc != 0)
7732 		return (rc);
7733 
7734 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7735 	if (sb == NULL)
7736 		return (ENOMEM);
7737 
7738 	t4_get_chan_txrate(sc, nrate, orate);
7739 
7740 	if (sc->chip_params->nchan > 2) {
7741 		sbuf_printf(sb, "              channel 0   channel 1"
7742 		    "   channel 2   channel 3\n");
7743 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
7744 		    nrate[0], nrate[1], nrate[2], nrate[3]);
7745 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
7746 		    orate[0], orate[1], orate[2], orate[3]);
7747 	} else {
7748 		sbuf_printf(sb, "              channel 0   channel 1\n");
7749 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
7750 		    nrate[0], nrate[1]);
7751 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
7752 		    orate[0], orate[1]);
7753 	}
7754 
7755 	rc = sbuf_finish(sb);
7756 	sbuf_delete(sb);
7757 
7758 	return (rc);
7759 }
7760 
7761 static int
7762 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
7763 {
7764 	struct adapter *sc = arg1;
7765 	struct sbuf *sb;
7766 	uint32_t *buf, *p;
7767 	int rc, i;
7768 
7769 	rc = sysctl_wire_old_buffer(req, 0);
7770 	if (rc != 0)
7771 		return (rc);
7772 
7773 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7774 	if (sb == NULL)
7775 		return (ENOMEM);
7776 
7777 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
7778 	    M_ZERO | M_WAITOK);
7779 
7780 	t4_ulprx_read_la(sc, buf);
7781 	p = buf;
7782 
7783 	sbuf_printf(sb, "      Pcmd        Type   Message"
7784 	    "                Data");
7785 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
7786 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
7787 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
7788 	}
7789 
7790 	rc = sbuf_finish(sb);
7791 	sbuf_delete(sb);
7792 	free(buf, M_CXGBE);
7793 	return (rc);
7794 }
7795 
7796 static int
7797 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
7798 {
7799 	struct adapter *sc = arg1;
7800 	struct sbuf *sb;
7801 	int rc, v;
7802 
7803 	MPASS(chip_id(sc) >= CHELSIO_T5);
7804 
7805 	rc = sysctl_wire_old_buffer(req, 0);
7806 	if (rc != 0)
7807 		return (rc);
7808 
7809 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7810 	if (sb == NULL)
7811 		return (ENOMEM);
7812 
7813 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
7814 	if (G_STATSOURCE_T5(v) == 7) {
7815 		int mode;
7816 
7817 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
7818 		if (mode == 0) {
7819 			sbuf_printf(sb, "total %d, incomplete %d",
7820 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7821 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7822 		} else if (mode == 1) {
7823 			sbuf_printf(sb, "total %d, data overflow %d",
7824 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7825 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7826 		} else {
7827 			sbuf_printf(sb, "unknown mode %d", mode);
7828 		}
7829 	}
7830 	rc = sbuf_finish(sb);
7831 	sbuf_delete(sb);
7832 
7833 	return (rc);
7834 }
7835 
7836 static int
7837 sysctl_tc_params(SYSCTL_HANDLER_ARGS)
7838 {
7839 	struct adapter *sc = arg1;
7840 	struct tx_sched_class *tc;
7841 	struct t4_sched_class_params p;
7842 	struct sbuf *sb;
7843 	int i, rc, port_id, flags, mbps, gbps;
7844 
7845 	rc = sysctl_wire_old_buffer(req, 0);
7846 	if (rc != 0)
7847 		return (rc);
7848 
7849 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7850 	if (sb == NULL)
7851 		return (ENOMEM);
7852 
7853 	port_id = arg2 >> 16;
7854 	MPASS(port_id < sc->params.nports);
7855 	MPASS(sc->port[port_id] != NULL);
7856 	i = arg2 & 0xffff;
7857 	MPASS(i < sc->chip_params->nsched_cls);
7858 	tc = &sc->port[port_id]->tc[i];
7859 
7860 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7861 	    "t4tc_p");
7862 	if (rc)
7863 		goto done;
7864 	flags = tc->flags;
7865 	p = tc->params;
7866 	end_synchronized_op(sc, LOCK_HELD);
7867 
7868 	if ((flags & TX_SC_OK) == 0) {
7869 		sbuf_printf(sb, "none");
7870 		goto done;
7871 	}
7872 
7873 	if (p.level == SCHED_CLASS_LEVEL_CL_WRR) {
7874 		sbuf_printf(sb, "cl-wrr weight %u", p.weight);
7875 		goto done;
7876 	} else if (p.level == SCHED_CLASS_LEVEL_CL_RL)
7877 		sbuf_printf(sb, "cl-rl");
7878 	else if (p.level == SCHED_CLASS_LEVEL_CH_RL)
7879 		sbuf_printf(sb, "ch-rl");
7880 	else {
7881 		rc = ENXIO;
7882 		goto done;
7883 	}
7884 
7885 	if (p.ratemode == SCHED_CLASS_RATEMODE_REL) {
7886 		/* XXX: top speed or actual link speed? */
7887 		gbps = port_top_speed(sc->port[port_id]);
7888 		sbuf_printf(sb, " %u%% of %uGbps", p.maxrate, gbps);
7889 	}
7890 	else if (p.ratemode == SCHED_CLASS_RATEMODE_ABS) {
7891 		switch (p.rateunit) {
7892 		case SCHED_CLASS_RATEUNIT_BITS:
7893 			mbps = p.maxrate / 1000;
7894 			gbps = p.maxrate / 1000000;
7895 			if (p.maxrate == gbps * 1000000)
7896 				sbuf_printf(sb, " %uGbps", gbps);
7897 			else if (p.maxrate == mbps * 1000)
7898 				sbuf_printf(sb, " %uMbps", mbps);
7899 			else
7900 				sbuf_printf(sb, " %uKbps", p.maxrate);
7901 			break;
7902 		case SCHED_CLASS_RATEUNIT_PKTS:
7903 			sbuf_printf(sb, " %upps", p.maxrate);
7904 			break;
7905 		default:
7906 			rc = ENXIO;
7907 			goto done;
7908 		}
7909 	}
7910 
7911 	switch (p.mode) {
7912 	case SCHED_CLASS_MODE_CLASS:
7913 		sbuf_printf(sb, " aggregate");
7914 		break;
7915 	case SCHED_CLASS_MODE_FLOW:
7916 		sbuf_printf(sb, " per-flow");
7917 		break;
7918 	default:
7919 		rc = ENXIO;
7920 		goto done;
7921 	}
7922 
7923 done:
7924 	if (rc == 0)
7925 		rc = sbuf_finish(sb);
7926 	sbuf_delete(sb);
7927 
7928 	return (rc);
7929 }
7930 #endif
7931 
7932 #ifdef TCP_OFFLOAD
7933 static void
7934 unit_conv(char *buf, size_t len, u_int val, u_int factor)
7935 {
7936 	u_int rem = val % factor;
7937 
7938 	if (rem == 0)
7939 		snprintf(buf, len, "%u", val / factor);
7940 	else {
7941 		while (rem % 10 == 0)
7942 			rem /= 10;
7943 		snprintf(buf, len, "%u.%u", val / factor, rem);
7944 	}
7945 }
7946 
7947 static int
7948 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
7949 {
7950 	struct adapter *sc = arg1;
7951 	char buf[16];
7952 	u_int res, re;
7953 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7954 
7955 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
7956 	switch (arg2) {
7957 	case 0:
7958 		/* timer_tick */
7959 		re = G_TIMERRESOLUTION(res);
7960 		break;
7961 	case 1:
7962 		/* TCP timestamp tick */
7963 		re = G_TIMESTAMPRESOLUTION(res);
7964 		break;
7965 	case 2:
7966 		/* DACK tick */
7967 		re = G_DELAYEDACKRESOLUTION(res);
7968 		break;
7969 	default:
7970 		return (EDOOFUS);
7971 	}
7972 
7973 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
7974 
7975 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
7976 }
7977 
7978 static int
7979 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
7980 {
7981 	struct adapter *sc = arg1;
7982 	u_int res, dack_re, v;
7983 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7984 
7985 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
7986 	dack_re = G_DELAYEDACKRESOLUTION(res);
7987 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
7988 
7989 	return (sysctl_handle_int(oidp, &v, 0, req));
7990 }
7991 
7992 static int
7993 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
7994 {
7995 	struct adapter *sc = arg1;
7996 	int reg = arg2;
7997 	u_int tre;
7998 	u_long tp_tick_us, v;
7999 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8000 
8001 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
8002 	    reg == A_TP_PERS_MIN || reg == A_TP_PERS_MAX ||
8003 	    reg == A_TP_KEEP_IDLE || A_TP_KEEP_INTVL || reg == A_TP_INIT_SRTT ||
8004 	    reg == A_TP_FINWAIT2_TIMER);
8005 
8006 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
8007 	tp_tick_us = (cclk_ps << tre) / 1000000;
8008 
8009 	if (reg == A_TP_INIT_SRTT)
8010 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
8011 	else
8012 		v = tp_tick_us * t4_read_reg(sc, reg);
8013 
8014 	return (sysctl_handle_long(oidp, &v, 0, req));
8015 }
8016 #endif
8017 
8018 static uint32_t
8019 fconf_iconf_to_mode(uint32_t fconf, uint32_t iconf)
8020 {
8021 	uint32_t mode;
8022 
8023 	mode = T4_FILTER_IPv4 | T4_FILTER_IPv6 | T4_FILTER_IP_SADDR |
8024 	    T4_FILTER_IP_DADDR | T4_FILTER_IP_SPORT | T4_FILTER_IP_DPORT;
8025 
8026 	if (fconf & F_FRAGMENTATION)
8027 		mode |= T4_FILTER_IP_FRAGMENT;
8028 
8029 	if (fconf & F_MPSHITTYPE)
8030 		mode |= T4_FILTER_MPS_HIT_TYPE;
8031 
8032 	if (fconf & F_MACMATCH)
8033 		mode |= T4_FILTER_MAC_IDX;
8034 
8035 	if (fconf & F_ETHERTYPE)
8036 		mode |= T4_FILTER_ETH_TYPE;
8037 
8038 	if (fconf & F_PROTOCOL)
8039 		mode |= T4_FILTER_IP_PROTO;
8040 
8041 	if (fconf & F_TOS)
8042 		mode |= T4_FILTER_IP_TOS;
8043 
8044 	if (fconf & F_VLAN)
8045 		mode |= T4_FILTER_VLAN;
8046 
8047 	if (fconf & F_VNIC_ID) {
8048 		mode |= T4_FILTER_VNIC;
8049 		if (iconf & F_VNIC)
8050 			mode |= T4_FILTER_IC_VNIC;
8051 	}
8052 
8053 	if (fconf & F_PORT)
8054 		mode |= T4_FILTER_PORT;
8055 
8056 	if (fconf & F_FCOE)
8057 		mode |= T4_FILTER_FCoE;
8058 
8059 	return (mode);
8060 }
8061 
8062 static uint32_t
8063 mode_to_fconf(uint32_t mode)
8064 {
8065 	uint32_t fconf = 0;
8066 
8067 	if (mode & T4_FILTER_IP_FRAGMENT)
8068 		fconf |= F_FRAGMENTATION;
8069 
8070 	if (mode & T4_FILTER_MPS_HIT_TYPE)
8071 		fconf |= F_MPSHITTYPE;
8072 
8073 	if (mode & T4_FILTER_MAC_IDX)
8074 		fconf |= F_MACMATCH;
8075 
8076 	if (mode & T4_FILTER_ETH_TYPE)
8077 		fconf |= F_ETHERTYPE;
8078 
8079 	if (mode & T4_FILTER_IP_PROTO)
8080 		fconf |= F_PROTOCOL;
8081 
8082 	if (mode & T4_FILTER_IP_TOS)
8083 		fconf |= F_TOS;
8084 
8085 	if (mode & T4_FILTER_VLAN)
8086 		fconf |= F_VLAN;
8087 
8088 	if (mode & T4_FILTER_VNIC)
8089 		fconf |= F_VNIC_ID;
8090 
8091 	if (mode & T4_FILTER_PORT)
8092 		fconf |= F_PORT;
8093 
8094 	if (mode & T4_FILTER_FCoE)
8095 		fconf |= F_FCOE;
8096 
8097 	return (fconf);
8098 }
8099 
8100 static uint32_t
8101 mode_to_iconf(uint32_t mode)
8102 {
8103 
8104 	if (mode & T4_FILTER_IC_VNIC)
8105 		return (F_VNIC);
8106 	return (0);
8107 }
8108 
8109 static int check_fspec_against_fconf_iconf(struct adapter *sc,
8110     struct t4_filter_specification *fs)
8111 {
8112 	struct tp_params *tpp = &sc->params.tp;
8113 	uint32_t fconf = 0;
8114 
8115 	if (fs->val.frag || fs->mask.frag)
8116 		fconf |= F_FRAGMENTATION;
8117 
8118 	if (fs->val.matchtype || fs->mask.matchtype)
8119 		fconf |= F_MPSHITTYPE;
8120 
8121 	if (fs->val.macidx || fs->mask.macidx)
8122 		fconf |= F_MACMATCH;
8123 
8124 	if (fs->val.ethtype || fs->mask.ethtype)
8125 		fconf |= F_ETHERTYPE;
8126 
8127 	if (fs->val.proto || fs->mask.proto)
8128 		fconf |= F_PROTOCOL;
8129 
8130 	if (fs->val.tos || fs->mask.tos)
8131 		fconf |= F_TOS;
8132 
8133 	if (fs->val.vlan_vld || fs->mask.vlan_vld)
8134 		fconf |= F_VLAN;
8135 
8136 	if (fs->val.ovlan_vld || fs->mask.ovlan_vld) {
8137 		fconf |= F_VNIC_ID;
8138 		if (tpp->ingress_config & F_VNIC)
8139 			return (EINVAL);
8140 	}
8141 
8142 	if (fs->val.pfvf_vld || fs->mask.pfvf_vld) {
8143 		fconf |= F_VNIC_ID;
8144 		if ((tpp->ingress_config & F_VNIC) == 0)
8145 			return (EINVAL);
8146 	}
8147 
8148 	if (fs->val.iport || fs->mask.iport)
8149 		fconf |= F_PORT;
8150 
8151 	if (fs->val.fcoe || fs->mask.fcoe)
8152 		fconf |= F_FCOE;
8153 
8154 	if ((tpp->vlan_pri_map | fconf) != tpp->vlan_pri_map)
8155 		return (E2BIG);
8156 
8157 	return (0);
8158 }
8159 
8160 static int
8161 get_filter_mode(struct adapter *sc, uint32_t *mode)
8162 {
8163 	struct tp_params *tpp = &sc->params.tp;
8164 
8165 	/*
8166 	 * We trust the cached values of the relevant TP registers.  This means
8167 	 * things work reliably only if writes to those registers are always via
8168 	 * t4_set_filter_mode.
8169 	 */
8170 	*mode = fconf_iconf_to_mode(tpp->vlan_pri_map, tpp->ingress_config);
8171 
8172 	return (0);
8173 }
8174 
8175 static int
8176 set_filter_mode(struct adapter *sc, uint32_t mode)
8177 {
8178 	struct tp_params *tpp = &sc->params.tp;
8179 	uint32_t fconf, iconf;
8180 	int rc;
8181 
8182 	iconf = mode_to_iconf(mode);
8183 	if ((iconf ^ tpp->ingress_config) & F_VNIC) {
8184 		/*
8185 		 * For now we just complain if A_TP_INGRESS_CONFIG is not
8186 		 * already set to the correct value for the requested filter
8187 		 * mode.  It's not clear if it's safe to write to this register
8188 		 * on the fly.  (And we trust the cached value of the register).
8189 		 */
8190 		return (EBUSY);
8191 	}
8192 
8193 	fconf = mode_to_fconf(mode);
8194 
8195 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8196 	    "t4setfm");
8197 	if (rc)
8198 		return (rc);
8199 
8200 	if (sc->tids.ftids_in_use > 0) {
8201 		rc = EBUSY;
8202 		goto done;
8203 	}
8204 
8205 #ifdef TCP_OFFLOAD
8206 	if (uld_active(sc, ULD_TOM)) {
8207 		rc = EBUSY;
8208 		goto done;
8209 	}
8210 #endif
8211 
8212 	rc = -t4_set_filter_mode(sc, fconf);
8213 done:
8214 	end_synchronized_op(sc, LOCK_HELD);
8215 	return (rc);
8216 }
8217 
8218 static inline uint64_t
8219 get_filter_hits(struct adapter *sc, uint32_t fid)
8220 {
8221 	uint32_t tcb_addr;
8222 
8223 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE) +
8224 	    (fid + sc->tids.ftid_base) * TCB_SIZE;
8225 
8226 	if (is_t4(sc)) {
8227 		uint64_t hits;
8228 
8229 		read_via_memwin(sc, 0, tcb_addr + 16, (uint32_t *)&hits, 8);
8230 		return (be64toh(hits));
8231 	} else {
8232 		uint32_t hits;
8233 
8234 		read_via_memwin(sc, 0, tcb_addr + 24, &hits, 4);
8235 		return (be32toh(hits));
8236 	}
8237 }
8238 
8239 static int
8240 get_filter(struct adapter *sc, struct t4_filter *t)
8241 {
8242 	int i, rc, nfilters = sc->tids.nftids;
8243 	struct filter_entry *f;
8244 
8245 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8246 	    "t4getf");
8247 	if (rc)
8248 		return (rc);
8249 
8250 	if (sc->tids.ftids_in_use == 0 || sc->tids.ftid_tab == NULL ||
8251 	    t->idx >= nfilters) {
8252 		t->idx = 0xffffffff;
8253 		goto done;
8254 	}
8255 
8256 	f = &sc->tids.ftid_tab[t->idx];
8257 	for (i = t->idx; i < nfilters; i++, f++) {
8258 		if (f->valid) {
8259 			t->idx = i;
8260 			t->l2tidx = f->l2t ? f->l2t->idx : 0;
8261 			t->smtidx = f->smtidx;
8262 			if (f->fs.hitcnts)
8263 				t->hits = get_filter_hits(sc, t->idx);
8264 			else
8265 				t->hits = UINT64_MAX;
8266 			t->fs = f->fs;
8267 
8268 			goto done;
8269 		}
8270 	}
8271 
8272 	t->idx = 0xffffffff;
8273 done:
8274 	end_synchronized_op(sc, LOCK_HELD);
8275 	return (0);
8276 }
8277 
8278 static int
8279 set_filter(struct adapter *sc, struct t4_filter *t)
8280 {
8281 	unsigned int nfilters, nports;
8282 	struct filter_entry *f;
8283 	int i, rc;
8284 
8285 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setf");
8286 	if (rc)
8287 		return (rc);
8288 
8289 	nfilters = sc->tids.nftids;
8290 	nports = sc->params.nports;
8291 
8292 	if (nfilters == 0) {
8293 		rc = ENOTSUP;
8294 		goto done;
8295 	}
8296 
8297 	if (t->idx >= nfilters) {
8298 		rc = EINVAL;
8299 		goto done;
8300 	}
8301 
8302 	/* Validate against the global filter mode and ingress config */
8303 	rc = check_fspec_against_fconf_iconf(sc, &t->fs);
8304 	if (rc != 0)
8305 		goto done;
8306 
8307 	if (t->fs.action == FILTER_SWITCH && t->fs.eport >= nports) {
8308 		rc = EINVAL;
8309 		goto done;
8310 	}
8311 
8312 	if (t->fs.val.iport >= nports) {
8313 		rc = EINVAL;
8314 		goto done;
8315 	}
8316 
8317 	/* Can't specify an iq if not steering to it */
8318 	if (!t->fs.dirsteer && t->fs.iq) {
8319 		rc = EINVAL;
8320 		goto done;
8321 	}
8322 
8323 	/* IPv6 filter idx must be 4 aligned */
8324 	if (t->fs.type == 1 &&
8325 	    ((t->idx & 0x3) || t->idx + 4 >= nfilters)) {
8326 		rc = EINVAL;
8327 		goto done;
8328 	}
8329 
8330 	if (!(sc->flags & FULL_INIT_DONE) &&
8331 	    ((rc = adapter_full_init(sc)) != 0))
8332 		goto done;
8333 
8334 	if (sc->tids.ftid_tab == NULL) {
8335 		KASSERT(sc->tids.ftids_in_use == 0,
8336 		    ("%s: no memory allocated but filters_in_use > 0",
8337 		    __func__));
8338 
8339 		sc->tids.ftid_tab = malloc(sizeof (struct filter_entry) *
8340 		    nfilters, M_CXGBE, M_NOWAIT | M_ZERO);
8341 		if (sc->tids.ftid_tab == NULL) {
8342 			rc = ENOMEM;
8343 			goto done;
8344 		}
8345 		mtx_init(&sc->tids.ftid_lock, "T4 filters", 0, MTX_DEF);
8346 	}
8347 
8348 	for (i = 0; i < 4; i++) {
8349 		f = &sc->tids.ftid_tab[t->idx + i];
8350 
8351 		if (f->pending || f->valid) {
8352 			rc = EBUSY;
8353 			goto done;
8354 		}
8355 		if (f->locked) {
8356 			rc = EPERM;
8357 			goto done;
8358 		}
8359 
8360 		if (t->fs.type == 0)
8361 			break;
8362 	}
8363 
8364 	f = &sc->tids.ftid_tab[t->idx];
8365 	f->fs = t->fs;
8366 
8367 	rc = set_filter_wr(sc, t->idx);
8368 done:
8369 	end_synchronized_op(sc, 0);
8370 
8371 	if (rc == 0) {
8372 		mtx_lock(&sc->tids.ftid_lock);
8373 		for (;;) {
8374 			if (f->pending == 0) {
8375 				rc = f->valid ? 0 : EIO;
8376 				break;
8377 			}
8378 
8379 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8380 			    PCATCH, "t4setfw", 0)) {
8381 				rc = EINPROGRESS;
8382 				break;
8383 			}
8384 		}
8385 		mtx_unlock(&sc->tids.ftid_lock);
8386 	}
8387 	return (rc);
8388 }
8389 
8390 static int
8391 del_filter(struct adapter *sc, struct t4_filter *t)
8392 {
8393 	unsigned int nfilters;
8394 	struct filter_entry *f;
8395 	int rc;
8396 
8397 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4delf");
8398 	if (rc)
8399 		return (rc);
8400 
8401 	nfilters = sc->tids.nftids;
8402 
8403 	if (nfilters == 0) {
8404 		rc = ENOTSUP;
8405 		goto done;
8406 	}
8407 
8408 	if (sc->tids.ftid_tab == NULL || sc->tids.ftids_in_use == 0 ||
8409 	    t->idx >= nfilters) {
8410 		rc = EINVAL;
8411 		goto done;
8412 	}
8413 
8414 	if (!(sc->flags & FULL_INIT_DONE)) {
8415 		rc = EAGAIN;
8416 		goto done;
8417 	}
8418 
8419 	f = &sc->tids.ftid_tab[t->idx];
8420 
8421 	if (f->pending) {
8422 		rc = EBUSY;
8423 		goto done;
8424 	}
8425 	if (f->locked) {
8426 		rc = EPERM;
8427 		goto done;
8428 	}
8429 
8430 	if (f->valid) {
8431 		t->fs = f->fs;	/* extra info for the caller */
8432 		rc = del_filter_wr(sc, t->idx);
8433 	}
8434 
8435 done:
8436 	end_synchronized_op(sc, 0);
8437 
8438 	if (rc == 0) {
8439 		mtx_lock(&sc->tids.ftid_lock);
8440 		for (;;) {
8441 			if (f->pending == 0) {
8442 				rc = f->valid ? EIO : 0;
8443 				break;
8444 			}
8445 
8446 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8447 			    PCATCH, "t4delfw", 0)) {
8448 				rc = EINPROGRESS;
8449 				break;
8450 			}
8451 		}
8452 		mtx_unlock(&sc->tids.ftid_lock);
8453 	}
8454 
8455 	return (rc);
8456 }
8457 
8458 static void
8459 clear_filter(struct filter_entry *f)
8460 {
8461 	if (f->l2t)
8462 		t4_l2t_release(f->l2t);
8463 
8464 	bzero(f, sizeof (*f));
8465 }
8466 
8467 static int
8468 set_filter_wr(struct adapter *sc, int fidx)
8469 {
8470 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8471 	struct fw_filter_wr *fwr;
8472 	unsigned int ftid, vnic_vld, vnic_vld_mask;
8473 	struct wrq_cookie cookie;
8474 
8475 	ASSERT_SYNCHRONIZED_OP(sc);
8476 
8477 	if (f->fs.newdmac || f->fs.newvlan) {
8478 		/* This filter needs an L2T entry; allocate one. */
8479 		f->l2t = t4_l2t_alloc_switching(sc->l2t);
8480 		if (f->l2t == NULL)
8481 			return (EAGAIN);
8482 		if (t4_l2t_set_switching(sc, f->l2t, f->fs.vlan, f->fs.eport,
8483 		    f->fs.dmac)) {
8484 			t4_l2t_release(f->l2t);
8485 			f->l2t = NULL;
8486 			return (ENOMEM);
8487 		}
8488 	}
8489 
8490 	/* Already validated against fconf, iconf */
8491 	MPASS((f->fs.val.pfvf_vld & f->fs.val.ovlan_vld) == 0);
8492 	MPASS((f->fs.mask.pfvf_vld & f->fs.mask.ovlan_vld) == 0);
8493 	if (f->fs.val.pfvf_vld || f->fs.val.ovlan_vld)
8494 		vnic_vld = 1;
8495 	else
8496 		vnic_vld = 0;
8497 	if (f->fs.mask.pfvf_vld || f->fs.mask.ovlan_vld)
8498 		vnic_vld_mask = 1;
8499 	else
8500 		vnic_vld_mask = 0;
8501 
8502 	ftid = sc->tids.ftid_base + fidx;
8503 
8504 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8505 	if (fwr == NULL)
8506 		return (ENOMEM);
8507 	bzero(fwr, sizeof(*fwr));
8508 
8509 	fwr->op_pkd = htobe32(V_FW_WR_OP(FW_FILTER_WR));
8510 	fwr->len16_pkd = htobe32(FW_LEN16(*fwr));
8511 	fwr->tid_to_iq =
8512 	    htobe32(V_FW_FILTER_WR_TID(ftid) |
8513 		V_FW_FILTER_WR_RQTYPE(f->fs.type) |
8514 		V_FW_FILTER_WR_NOREPLY(0) |
8515 		V_FW_FILTER_WR_IQ(f->fs.iq));
8516 	fwr->del_filter_to_l2tix =
8517 	    htobe32(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
8518 		V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
8519 		V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
8520 		V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
8521 		V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
8522 		V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
8523 		V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
8524 		V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
8525 		V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
8526 		    f->fs.newvlan == VLAN_REWRITE) |
8527 		V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
8528 		    f->fs.newvlan == VLAN_REWRITE) |
8529 		V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
8530 		V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
8531 		V_FW_FILTER_WR_PRIO(f->fs.prio) |
8532 		V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
8533 	fwr->ethtype = htobe16(f->fs.val.ethtype);
8534 	fwr->ethtypem = htobe16(f->fs.mask.ethtype);
8535 	fwr->frag_to_ovlan_vldm =
8536 	    (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
8537 		V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
8538 		V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.vlan_vld) |
8539 		V_FW_FILTER_WR_OVLAN_VLD(vnic_vld) |
8540 		V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.vlan_vld) |
8541 		V_FW_FILTER_WR_OVLAN_VLDM(vnic_vld_mask));
8542 	fwr->smac_sel = 0;
8543 	fwr->rx_chan_rx_rpl_iq = htobe16(V_FW_FILTER_WR_RX_CHAN(0) |
8544 	    V_FW_FILTER_WR_RX_RPL_IQ(sc->sge.fwq.abs_id));
8545 	fwr->maci_to_matchtypem =
8546 	    htobe32(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
8547 		V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
8548 		V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
8549 		V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
8550 		V_FW_FILTER_WR_PORT(f->fs.val.iport) |
8551 		V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
8552 		V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
8553 		V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
8554 	fwr->ptcl = f->fs.val.proto;
8555 	fwr->ptclm = f->fs.mask.proto;
8556 	fwr->ttyp = f->fs.val.tos;
8557 	fwr->ttypm = f->fs.mask.tos;
8558 	fwr->ivlan = htobe16(f->fs.val.vlan);
8559 	fwr->ivlanm = htobe16(f->fs.mask.vlan);
8560 	fwr->ovlan = htobe16(f->fs.val.vnic);
8561 	fwr->ovlanm = htobe16(f->fs.mask.vnic);
8562 	bcopy(f->fs.val.dip, fwr->lip, sizeof (fwr->lip));
8563 	bcopy(f->fs.mask.dip, fwr->lipm, sizeof (fwr->lipm));
8564 	bcopy(f->fs.val.sip, fwr->fip, sizeof (fwr->fip));
8565 	bcopy(f->fs.mask.sip, fwr->fipm, sizeof (fwr->fipm));
8566 	fwr->lp = htobe16(f->fs.val.dport);
8567 	fwr->lpm = htobe16(f->fs.mask.dport);
8568 	fwr->fp = htobe16(f->fs.val.sport);
8569 	fwr->fpm = htobe16(f->fs.mask.sport);
8570 	if (f->fs.newsmac)
8571 		bcopy(f->fs.smac, fwr->sma, sizeof (fwr->sma));
8572 
8573 	f->pending = 1;
8574 	sc->tids.ftids_in_use++;
8575 
8576 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8577 	return (0);
8578 }
8579 
8580 static int
8581 del_filter_wr(struct adapter *sc, int fidx)
8582 {
8583 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8584 	struct fw_filter_wr *fwr;
8585 	unsigned int ftid;
8586 	struct wrq_cookie cookie;
8587 
8588 	ftid = sc->tids.ftid_base + fidx;
8589 
8590 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8591 	if (fwr == NULL)
8592 		return (ENOMEM);
8593 	bzero(fwr, sizeof (*fwr));
8594 
8595 	t4_mk_filtdelwr(ftid, fwr, sc->sge.fwq.abs_id);
8596 
8597 	f->pending = 1;
8598 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8599 	return (0);
8600 }
8601 
8602 int
8603 t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8604 {
8605 	struct adapter *sc = iq->adapter;
8606 	const struct cpl_set_tcb_rpl *rpl = (const void *)(rss + 1);
8607 	unsigned int idx = GET_TID(rpl);
8608 	unsigned int rc;
8609 	struct filter_entry *f;
8610 
8611 	KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
8612 	    rss->opcode));
8613 	MPASS(iq == &sc->sge.fwq);
8614 	MPASS(is_ftid(sc, idx));
8615 
8616 	idx -= sc->tids.ftid_base;
8617 	f = &sc->tids.ftid_tab[idx];
8618 	rc = G_COOKIE(rpl->cookie);
8619 
8620 	mtx_lock(&sc->tids.ftid_lock);
8621 	if (rc == FW_FILTER_WR_FLT_ADDED) {
8622 		KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
8623 		    __func__, idx));
8624 		f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
8625 		f->pending = 0;  /* asynchronous setup completed */
8626 		f->valid = 1;
8627 	} else {
8628 		if (rc != FW_FILTER_WR_FLT_DELETED) {
8629 			/* Add or delete failed, display an error */
8630 			log(LOG_ERR,
8631 			    "filter %u setup failed with error %u\n",
8632 			    idx, rc);
8633 		}
8634 
8635 		clear_filter(f);
8636 		sc->tids.ftids_in_use--;
8637 	}
8638 	wakeup(&sc->tids.ftid_tab);
8639 	mtx_unlock(&sc->tids.ftid_lock);
8640 
8641 	return (0);
8642 }
8643 
8644 static int
8645 set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8646 {
8647 
8648 	MPASS(iq->set_tcb_rpl != NULL);
8649 	return (iq->set_tcb_rpl(iq, rss, m));
8650 }
8651 
8652 static int
8653 l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8654 {
8655 
8656 	MPASS(iq->l2t_write_rpl != NULL);
8657 	return (iq->l2t_write_rpl(iq, rss, m));
8658 }
8659 
8660 static int
8661 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
8662 {
8663 	int rc;
8664 
8665 	if (cntxt->cid > M_CTXTQID)
8666 		return (EINVAL);
8667 
8668 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
8669 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
8670 		return (EINVAL);
8671 
8672 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
8673 	if (rc)
8674 		return (rc);
8675 
8676 	if (sc->flags & FW_OK) {
8677 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
8678 		    &cntxt->data[0]);
8679 		if (rc == 0)
8680 			goto done;
8681 	}
8682 
8683 	/*
8684 	 * Read via firmware failed or wasn't even attempted.  Read directly via
8685 	 * the backdoor.
8686 	 */
8687 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
8688 done:
8689 	end_synchronized_op(sc, 0);
8690 	return (rc);
8691 }
8692 
8693 static int
8694 load_fw(struct adapter *sc, struct t4_data *fw)
8695 {
8696 	int rc;
8697 	uint8_t *fw_data;
8698 
8699 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
8700 	if (rc)
8701 		return (rc);
8702 
8703 	if (sc->flags & FULL_INIT_DONE) {
8704 		rc = EBUSY;
8705 		goto done;
8706 	}
8707 
8708 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
8709 	if (fw_data == NULL) {
8710 		rc = ENOMEM;
8711 		goto done;
8712 	}
8713 
8714 	rc = copyin(fw->data, fw_data, fw->len);
8715 	if (rc == 0)
8716 		rc = -t4_load_fw(sc, fw_data, fw->len);
8717 
8718 	free(fw_data, M_CXGBE);
8719 done:
8720 	end_synchronized_op(sc, 0);
8721 	return (rc);
8722 }
8723 
8724 static int
8725 load_cfg(struct adapter *sc, struct t4_data *cfg)
8726 {
8727 	int rc;
8728 	uint8_t *cfg_data = NULL;
8729 
8730 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
8731 	if (rc)
8732 		return (rc);
8733 
8734 	if (cfg->len == 0) {
8735 		/* clear */
8736 		rc = -t4_load_cfg(sc, NULL, 0);
8737 		goto done;
8738 	}
8739 
8740 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
8741 	if (cfg_data == NULL) {
8742 		rc = ENOMEM;
8743 		goto done;
8744 	}
8745 
8746 	rc = copyin(cfg->data, cfg_data, cfg->len);
8747 	if (rc == 0)
8748 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
8749 
8750 	free(cfg_data, M_CXGBE);
8751 done:
8752 	end_synchronized_op(sc, 0);
8753 	return (rc);
8754 }
8755 
8756 #define MAX_READ_BUF_SIZE (128 * 1024)
8757 static int
8758 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
8759 {
8760 	uint32_t addr, remaining, n;
8761 	uint32_t *buf;
8762 	int rc;
8763 	uint8_t *dst;
8764 
8765 	rc = validate_mem_range(sc, mr->addr, mr->len);
8766 	if (rc != 0)
8767 		return (rc);
8768 
8769 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
8770 	addr = mr->addr;
8771 	remaining = mr->len;
8772 	dst = (void *)mr->data;
8773 
8774 	while (remaining) {
8775 		n = min(remaining, MAX_READ_BUF_SIZE);
8776 		read_via_memwin(sc, 2, addr, buf, n);
8777 
8778 		rc = copyout(buf, dst, n);
8779 		if (rc != 0)
8780 			break;
8781 
8782 		dst += n;
8783 		remaining -= n;
8784 		addr += n;
8785 	}
8786 
8787 	free(buf, M_CXGBE);
8788 	return (rc);
8789 }
8790 #undef MAX_READ_BUF_SIZE
8791 
8792 static int
8793 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
8794 {
8795 	int rc;
8796 
8797 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
8798 		return (EINVAL);
8799 
8800 	if (i2cd->len > sizeof(i2cd->data))
8801 		return (EFBIG);
8802 
8803 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
8804 	if (rc)
8805 		return (rc);
8806 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
8807 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
8808 	end_synchronized_op(sc, 0);
8809 
8810 	return (rc);
8811 }
8812 
8813 static int
8814 in_range(int val, int lo, int hi)
8815 {
8816 
8817 	return (val < 0 || (val <= hi && val >= lo));
8818 }
8819 
8820 static int
8821 set_sched_class_config(struct adapter *sc, int minmax)
8822 {
8823 	int rc;
8824 
8825 	if (minmax < 0)
8826 		return (EINVAL);
8827 
8828 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4sscc");
8829 	if (rc)
8830 		return (rc);
8831 	rc = -t4_sched_config(sc, FW_SCHED_TYPE_PKTSCHED, minmax, 1);
8832 	end_synchronized_op(sc, 0);
8833 
8834 	return (rc);
8835 }
8836 
8837 static int
8838 set_sched_class_params(struct adapter *sc, struct t4_sched_class_params *p,
8839     int sleep_ok)
8840 {
8841 	int rc, top_speed, fw_level, fw_mode, fw_rateunit, fw_ratemode;
8842 	struct port_info *pi;
8843 	struct tx_sched_class *tc;
8844 
8845 	if (p->level == SCHED_CLASS_LEVEL_CL_RL)
8846 		fw_level = FW_SCHED_PARAMS_LEVEL_CL_RL;
8847 	else if (p->level == SCHED_CLASS_LEVEL_CL_WRR)
8848 		fw_level = FW_SCHED_PARAMS_LEVEL_CL_WRR;
8849 	else if (p->level == SCHED_CLASS_LEVEL_CH_RL)
8850 		fw_level = FW_SCHED_PARAMS_LEVEL_CH_RL;
8851 	else
8852 		return (EINVAL);
8853 
8854 	if (p->mode == SCHED_CLASS_MODE_CLASS)
8855 		fw_mode = FW_SCHED_PARAMS_MODE_CLASS;
8856 	else if (p->mode == SCHED_CLASS_MODE_FLOW)
8857 		fw_mode = FW_SCHED_PARAMS_MODE_FLOW;
8858 	else
8859 		return (EINVAL);
8860 
8861 	if (p->rateunit == SCHED_CLASS_RATEUNIT_BITS)
8862 		fw_rateunit = FW_SCHED_PARAMS_UNIT_BITRATE;
8863 	else if (p->rateunit == SCHED_CLASS_RATEUNIT_PKTS)
8864 		fw_rateunit = FW_SCHED_PARAMS_UNIT_PKTRATE;
8865 	else
8866 		return (EINVAL);
8867 
8868 	if (p->ratemode == SCHED_CLASS_RATEMODE_REL)
8869 		fw_ratemode = FW_SCHED_PARAMS_RATE_REL;
8870 	else if (p->ratemode == SCHED_CLASS_RATEMODE_ABS)
8871 		fw_ratemode = FW_SCHED_PARAMS_RATE_ABS;
8872 	else
8873 		return (EINVAL);
8874 
8875 	/* Vet our parameters ... */
8876 	if (!in_range(p->channel, 0, sc->chip_params->nchan - 1))
8877 		return (ERANGE);
8878 
8879 	pi = sc->port[sc->chan_map[p->channel]];
8880 	if (pi == NULL)
8881 		return (ENXIO);
8882 	MPASS(pi->tx_chan == p->channel);
8883 	top_speed = port_top_speed(pi) * 1000000; /* Gbps -> Kbps */
8884 
8885 	if (!in_range(p->cl, 0, sc->chip_params->nsched_cls) ||
8886 	    !in_range(p->minrate, 0, top_speed) ||
8887 	    !in_range(p->maxrate, 0, top_speed) ||
8888 	    !in_range(p->weight, 0, 100))
8889 		return (ERANGE);
8890 
8891 	/*
8892 	 * Translate any unset parameters into the firmware's
8893 	 * nomenclature and/or fail the call if the parameters
8894 	 * are required ...
8895 	 */
8896 	if (p->rateunit < 0 || p->ratemode < 0 || p->channel < 0 || p->cl < 0)
8897 		return (EINVAL);
8898 
8899 	if (p->minrate < 0)
8900 		p->minrate = 0;
8901 	if (p->maxrate < 0) {
8902 		if (p->level == SCHED_CLASS_LEVEL_CL_RL ||
8903 		    p->level == SCHED_CLASS_LEVEL_CH_RL)
8904 			return (EINVAL);
8905 		else
8906 			p->maxrate = 0;
8907 	}
8908 	if (p->weight < 0) {
8909 		if (p->level == SCHED_CLASS_LEVEL_CL_WRR)
8910 			return (EINVAL);
8911 		else
8912 			p->weight = 0;
8913 	}
8914 	if (p->pktsize < 0) {
8915 		if (p->level == SCHED_CLASS_LEVEL_CL_RL ||
8916 		    p->level == SCHED_CLASS_LEVEL_CH_RL)
8917 			return (EINVAL);
8918 		else
8919 			p->pktsize = 0;
8920 	}
8921 
8922 	rc = begin_synchronized_op(sc, NULL,
8923 	    sleep_ok ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4sscp");
8924 	if (rc)
8925 		return (rc);
8926 	tc = &pi->tc[p->cl];
8927 	tc->params = *p;
8928 	rc = -t4_sched_params(sc, FW_SCHED_TYPE_PKTSCHED, fw_level, fw_mode,
8929 	    fw_rateunit, fw_ratemode, p->channel, p->cl, p->minrate, p->maxrate,
8930 	    p->weight, p->pktsize, sleep_ok);
8931 	if (rc == 0)
8932 		tc->flags |= TX_SC_OK;
8933 	else {
8934 		/*
8935 		 * Unknown state at this point, see tc->params for what was
8936 		 * attempted.
8937 		 */
8938 		tc->flags &= ~TX_SC_OK;
8939 	}
8940 	end_synchronized_op(sc, sleep_ok ? 0 : LOCK_HELD);
8941 
8942 	return (rc);
8943 }
8944 
8945 int
8946 t4_set_sched_class(struct adapter *sc, struct t4_sched_params *p)
8947 {
8948 
8949 	if (p->type != SCHED_CLASS_TYPE_PACKET)
8950 		return (EINVAL);
8951 
8952 	if (p->subcmd == SCHED_CLASS_SUBCMD_CONFIG)
8953 		return (set_sched_class_config(sc, p->u.config.minmax));
8954 
8955 	if (p->subcmd == SCHED_CLASS_SUBCMD_PARAMS)
8956 		return (set_sched_class_params(sc, &p->u.params, 1));
8957 
8958 	return (EINVAL);
8959 }
8960 
8961 int
8962 t4_set_sched_queue(struct adapter *sc, struct t4_sched_queue *p)
8963 {
8964 	struct port_info *pi = NULL;
8965 	struct vi_info *vi;
8966 	struct sge_txq *txq;
8967 	uint32_t fw_mnem, fw_queue, fw_class;
8968 	int i, rc;
8969 
8970 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setsq");
8971 	if (rc)
8972 		return (rc);
8973 
8974 	if (p->port >= sc->params.nports) {
8975 		rc = EINVAL;
8976 		goto done;
8977 	}
8978 
8979 	/* XXX: Only supported for the main VI. */
8980 	pi = sc->port[p->port];
8981 	vi = &pi->vi[0];
8982 	if (!(vi->flags & VI_INIT_DONE)) {
8983 		/* tx queues not set up yet */
8984 		rc = EAGAIN;
8985 		goto done;
8986 	}
8987 
8988 	if (!in_range(p->queue, 0, vi->ntxq - 1) ||
8989 	    !in_range(p->cl, 0, sc->chip_params->nsched_cls - 1)) {
8990 		rc = EINVAL;
8991 		goto done;
8992 	}
8993 
8994 	/*
8995 	 * Create a template for the FW_PARAMS_CMD mnemonic and value (TX
8996 	 * Scheduling Class in this case).
8997 	 */
8998 	fw_mnem = (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
8999 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_EQ_SCHEDCLASS_ETH));
9000 	fw_class = p->cl < 0 ? 0xffffffff : p->cl;
9001 
9002 	/*
9003 	 * If op.queue is non-negative, then we're only changing the scheduling
9004 	 * on a single specified TX queue.
9005 	 */
9006 	if (p->queue >= 0) {
9007 		txq = &sc->sge.txq[vi->first_txq + p->queue];
9008 		fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
9009 		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
9010 		    &fw_class);
9011 		goto done;
9012 	}
9013 
9014 	/*
9015 	 * Change the scheduling on all the TX queues for the
9016 	 * interface.
9017 	 */
9018 	for_each_txq(vi, i, txq) {
9019 		fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
9020 		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
9021 		    &fw_class);
9022 		if (rc)
9023 			goto done;
9024 	}
9025 
9026 	rc = 0;
9027 done:
9028 	end_synchronized_op(sc, 0);
9029 	return (rc);
9030 }
9031 
9032 int
9033 t4_os_find_pci_capability(struct adapter *sc, int cap)
9034 {
9035 	int i;
9036 
9037 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9038 }
9039 
9040 int
9041 t4_os_pci_save_state(struct adapter *sc)
9042 {
9043 	device_t dev;
9044 	struct pci_devinfo *dinfo;
9045 
9046 	dev = sc->dev;
9047 	dinfo = device_get_ivars(dev);
9048 
9049 	pci_cfg_save(dev, dinfo, 0);
9050 	return (0);
9051 }
9052 
9053 int
9054 t4_os_pci_restore_state(struct adapter *sc)
9055 {
9056 	device_t dev;
9057 	struct pci_devinfo *dinfo;
9058 
9059 	dev = sc->dev;
9060 	dinfo = device_get_ivars(dev);
9061 
9062 	pci_cfg_restore(dev, dinfo);
9063 	return (0);
9064 }
9065 
9066 void
9067 t4_os_portmod_changed(const struct adapter *sc, int idx)
9068 {
9069 	struct port_info *pi = sc->port[idx];
9070 	struct vi_info *vi;
9071 	struct ifnet *ifp;
9072 	int v;
9073 	static const char *mod_str[] = {
9074 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9075 	};
9076 
9077 	for_each_vi(pi, v, vi) {
9078 		build_medialist(pi, &vi->media);
9079 	}
9080 
9081 	ifp = pi->vi[0].ifp;
9082 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9083 		if_printf(ifp, "transceiver unplugged.\n");
9084 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9085 		if_printf(ifp, "unknown transceiver inserted.\n");
9086 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9087 		if_printf(ifp, "unsupported transceiver inserted.\n");
9088 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9089 		if_printf(ifp, "%s transceiver inserted.\n",
9090 		    mod_str[pi->mod_type]);
9091 	} else {
9092 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9093 		    pi->mod_type);
9094 	}
9095 }
9096 
9097 void
9098 t4_os_link_changed(struct adapter *sc, int idx, int link_stat)
9099 {
9100 	struct port_info *pi = sc->port[idx];
9101 	struct vi_info *vi;
9102 	struct ifnet *ifp;
9103 	int v;
9104 
9105 	for_each_vi(pi, v, vi) {
9106 		ifp = vi->ifp;
9107 		if (ifp == NULL)
9108 			continue;
9109 
9110 		if (link_stat) {
9111 			ifp->if_baudrate = IF_Mbps(pi->link_cfg.speed);
9112 			if_link_state_change(ifp, LINK_STATE_UP);
9113 		} else {
9114 			if_link_state_change(ifp, LINK_STATE_DOWN);
9115 		}
9116 	}
9117 }
9118 
9119 void
9120 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9121 {
9122 	struct adapter *sc;
9123 
9124 	sx_slock(&t4_list_lock);
9125 	SLIST_FOREACH(sc, &t4_list, link) {
9126 		/*
9127 		 * func should not make any assumptions about what state sc is
9128 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9129 		 */
9130 		func(sc, arg);
9131 	}
9132 	sx_sunlock(&t4_list_lock);
9133 }
9134 
9135 static int
9136 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9137     struct thread *td)
9138 {
9139 	int rc;
9140 	struct adapter *sc = dev->si_drv1;
9141 
9142 	rc = priv_check(td, PRIV_DRIVER);
9143 	if (rc != 0)
9144 		return (rc);
9145 
9146 	switch (cmd) {
9147 	case CHELSIO_T4_GETREG: {
9148 		struct t4_reg *edata = (struct t4_reg *)data;
9149 
9150 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9151 			return (EFAULT);
9152 
9153 		if (edata->size == 4)
9154 			edata->val = t4_read_reg(sc, edata->addr);
9155 		else if (edata->size == 8)
9156 			edata->val = t4_read_reg64(sc, edata->addr);
9157 		else
9158 			return (EINVAL);
9159 
9160 		break;
9161 	}
9162 	case CHELSIO_T4_SETREG: {
9163 		struct t4_reg *edata = (struct t4_reg *)data;
9164 
9165 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9166 			return (EFAULT);
9167 
9168 		if (edata->size == 4) {
9169 			if (edata->val & 0xffffffff00000000)
9170 				return (EINVAL);
9171 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9172 		} else if (edata->size == 8)
9173 			t4_write_reg64(sc, edata->addr, edata->val);
9174 		else
9175 			return (EINVAL);
9176 		break;
9177 	}
9178 	case CHELSIO_T4_REGDUMP: {
9179 		struct t4_regdump *regs = (struct t4_regdump *)data;
9180 		int reglen = t4_get_regs_len(sc);
9181 		uint8_t *buf;
9182 
9183 		if (regs->len < reglen) {
9184 			regs->len = reglen; /* hint to the caller */
9185 			return (ENOBUFS);
9186 		}
9187 
9188 		regs->len = reglen;
9189 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9190 		get_regs(sc, regs, buf);
9191 		rc = copyout(buf, regs->data, reglen);
9192 		free(buf, M_CXGBE);
9193 		break;
9194 	}
9195 	case CHELSIO_T4_GET_FILTER_MODE:
9196 		rc = get_filter_mode(sc, (uint32_t *)data);
9197 		break;
9198 	case CHELSIO_T4_SET_FILTER_MODE:
9199 		rc = set_filter_mode(sc, *(uint32_t *)data);
9200 		break;
9201 	case CHELSIO_T4_GET_FILTER:
9202 		rc = get_filter(sc, (struct t4_filter *)data);
9203 		break;
9204 	case CHELSIO_T4_SET_FILTER:
9205 		rc = set_filter(sc, (struct t4_filter *)data);
9206 		break;
9207 	case CHELSIO_T4_DEL_FILTER:
9208 		rc = del_filter(sc, (struct t4_filter *)data);
9209 		break;
9210 	case CHELSIO_T4_GET_SGE_CONTEXT:
9211 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9212 		break;
9213 	case CHELSIO_T4_LOAD_FW:
9214 		rc = load_fw(sc, (struct t4_data *)data);
9215 		break;
9216 	case CHELSIO_T4_GET_MEM:
9217 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9218 		break;
9219 	case CHELSIO_T4_GET_I2C:
9220 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9221 		break;
9222 	case CHELSIO_T4_CLEAR_STATS: {
9223 		int i, v;
9224 		u_int port_id = *(uint32_t *)data;
9225 		struct port_info *pi;
9226 		struct vi_info *vi;
9227 
9228 		if (port_id >= sc->params.nports)
9229 			return (EINVAL);
9230 		pi = sc->port[port_id];
9231 		if (pi == NULL)
9232 			return (EIO);
9233 
9234 		/* MAC stats */
9235 		t4_clr_port_stats(sc, pi->tx_chan);
9236 		pi->tx_parse_error = 0;
9237 		mtx_lock(&sc->reg_lock);
9238 		for_each_vi(pi, v, vi) {
9239 			if (vi->flags & VI_INIT_DONE)
9240 				t4_clr_vi_stats(sc, vi->viid);
9241 		}
9242 		mtx_unlock(&sc->reg_lock);
9243 
9244 		/*
9245 		 * Since this command accepts a port, clear stats for
9246 		 * all VIs on this port.
9247 		 */
9248 		for_each_vi(pi, v, vi) {
9249 			if (vi->flags & VI_INIT_DONE) {
9250 				struct sge_rxq *rxq;
9251 				struct sge_txq *txq;
9252 				struct sge_wrq *wrq;
9253 
9254 				for_each_rxq(vi, i, rxq) {
9255 #if defined(INET) || defined(INET6)
9256 					rxq->lro.lro_queued = 0;
9257 					rxq->lro.lro_flushed = 0;
9258 #endif
9259 					rxq->rxcsum = 0;
9260 					rxq->vlan_extraction = 0;
9261 				}
9262 
9263 				for_each_txq(vi, i, txq) {
9264 					txq->txcsum = 0;
9265 					txq->tso_wrs = 0;
9266 					txq->vlan_insertion = 0;
9267 					txq->imm_wrs = 0;
9268 					txq->sgl_wrs = 0;
9269 					txq->txpkt_wrs = 0;
9270 					txq->txpkts0_wrs = 0;
9271 					txq->txpkts1_wrs = 0;
9272 					txq->txpkts0_pkts = 0;
9273 					txq->txpkts1_pkts = 0;
9274 					mp_ring_reset_stats(txq->r);
9275 				}
9276 
9277 #ifdef TCP_OFFLOAD
9278 				/* nothing to clear for each ofld_rxq */
9279 
9280 				for_each_ofld_txq(vi, i, wrq) {
9281 					wrq->tx_wrs_direct = 0;
9282 					wrq->tx_wrs_copied = 0;
9283 				}
9284 #endif
9285 
9286 				if (IS_MAIN_VI(vi)) {
9287 					wrq = &sc->sge.ctrlq[pi->port_id];
9288 					wrq->tx_wrs_direct = 0;
9289 					wrq->tx_wrs_copied = 0;
9290 				}
9291 			}
9292 		}
9293 		break;
9294 	}
9295 	case CHELSIO_T4_SCHED_CLASS:
9296 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9297 		break;
9298 	case CHELSIO_T4_SCHED_QUEUE:
9299 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9300 		break;
9301 	case CHELSIO_T4_GET_TRACER:
9302 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9303 		break;
9304 	case CHELSIO_T4_SET_TRACER:
9305 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9306 		break;
9307 	case CHELSIO_T4_LOAD_CFG:
9308 		rc = load_cfg(sc, (struct t4_data *)data);
9309 		break;
9310 	default:
9311 		rc = ENOTTY;
9312 	}
9313 
9314 	return (rc);
9315 }
9316 
9317 void
9318 t4_db_full(struct adapter *sc)
9319 {
9320 
9321 	CXGBE_UNIMPLEMENTED(__func__);
9322 }
9323 
9324 void
9325 t4_db_dropped(struct adapter *sc)
9326 {
9327 
9328 	CXGBE_UNIMPLEMENTED(__func__);
9329 }
9330 
9331 #ifdef TCP_OFFLOAD
9332 static int
9333 toe_capability(struct vi_info *vi, int enable)
9334 {
9335 	int rc;
9336 	struct port_info *pi = vi->pi;
9337 	struct adapter *sc = pi->adapter;
9338 
9339 	ASSERT_SYNCHRONIZED_OP(sc);
9340 
9341 	if (!is_offload(sc))
9342 		return (ENODEV);
9343 
9344 	if (enable) {
9345 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9346 			/* TOE is already enabled. */
9347 			return (0);
9348 		}
9349 
9350 		/*
9351 		 * We need the port's queues around so that we're able to send
9352 		 * and receive CPLs to/from the TOE even if the ifnet for this
9353 		 * port has never been UP'd administratively.
9354 		 */
9355 		if (!(vi->flags & VI_INIT_DONE)) {
9356 			rc = vi_full_init(vi);
9357 			if (rc)
9358 				return (rc);
9359 		}
9360 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9361 			rc = vi_full_init(&pi->vi[0]);
9362 			if (rc)
9363 				return (rc);
9364 		}
9365 
9366 		if (isset(&sc->offload_map, pi->port_id)) {
9367 			/* TOE is enabled on another VI of this port. */
9368 			pi->uld_vis++;
9369 			return (0);
9370 		}
9371 
9372 		if (!uld_active(sc, ULD_TOM)) {
9373 			rc = t4_activate_uld(sc, ULD_TOM);
9374 			if (rc == EAGAIN) {
9375 				log(LOG_WARNING,
9376 				    "You must kldload t4_tom.ko before trying "
9377 				    "to enable TOE on a cxgbe interface.\n");
9378 			}
9379 			if (rc != 0)
9380 				return (rc);
9381 			KASSERT(sc->tom_softc != NULL,
9382 			    ("%s: TOM activated but softc NULL", __func__));
9383 			KASSERT(uld_active(sc, ULD_TOM),
9384 			    ("%s: TOM activated but flag not set", __func__));
9385 		}
9386 
9387 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
9388 		if (!uld_active(sc, ULD_IWARP))
9389 			(void) t4_activate_uld(sc, ULD_IWARP);
9390 		if (!uld_active(sc, ULD_ISCSI))
9391 			(void) t4_activate_uld(sc, ULD_ISCSI);
9392 
9393 		pi->uld_vis++;
9394 		setbit(&sc->offload_map, pi->port_id);
9395 	} else {
9396 		pi->uld_vis--;
9397 
9398 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9399 			return (0);
9400 
9401 		KASSERT(uld_active(sc, ULD_TOM),
9402 		    ("%s: TOM never initialized?", __func__));
9403 		clrbit(&sc->offload_map, pi->port_id);
9404 	}
9405 
9406 	return (0);
9407 }
9408 
9409 /*
9410  * Add an upper layer driver to the global list.
9411  */
9412 int
9413 t4_register_uld(struct uld_info *ui)
9414 {
9415 	int rc = 0;
9416 	struct uld_info *u;
9417 
9418 	sx_xlock(&t4_uld_list_lock);
9419 	SLIST_FOREACH(u, &t4_uld_list, link) {
9420 	    if (u->uld_id == ui->uld_id) {
9421 		    rc = EEXIST;
9422 		    goto done;
9423 	    }
9424 	}
9425 
9426 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9427 	ui->refcount = 0;
9428 done:
9429 	sx_xunlock(&t4_uld_list_lock);
9430 	return (rc);
9431 }
9432 
9433 int
9434 t4_unregister_uld(struct uld_info *ui)
9435 {
9436 	int rc = EINVAL;
9437 	struct uld_info *u;
9438 
9439 	sx_xlock(&t4_uld_list_lock);
9440 
9441 	SLIST_FOREACH(u, &t4_uld_list, link) {
9442 	    if (u == ui) {
9443 		    if (ui->refcount > 0) {
9444 			    rc = EBUSY;
9445 			    goto done;
9446 		    }
9447 
9448 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9449 		    rc = 0;
9450 		    goto done;
9451 	    }
9452 	}
9453 done:
9454 	sx_xunlock(&t4_uld_list_lock);
9455 	return (rc);
9456 }
9457 
9458 int
9459 t4_activate_uld(struct adapter *sc, int id)
9460 {
9461 	int rc;
9462 	struct uld_info *ui;
9463 
9464 	ASSERT_SYNCHRONIZED_OP(sc);
9465 
9466 	if (id < 0 || id > ULD_MAX)
9467 		return (EINVAL);
9468 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
9469 
9470 	sx_slock(&t4_uld_list_lock);
9471 
9472 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9473 		if (ui->uld_id == id) {
9474 			if (!(sc->flags & FULL_INIT_DONE)) {
9475 				rc = adapter_full_init(sc);
9476 				if (rc != 0)
9477 					break;
9478 			}
9479 
9480 			rc = ui->activate(sc);
9481 			if (rc == 0) {
9482 				setbit(&sc->active_ulds, id);
9483 				ui->refcount++;
9484 			}
9485 			break;
9486 		}
9487 	}
9488 
9489 	sx_sunlock(&t4_uld_list_lock);
9490 
9491 	return (rc);
9492 }
9493 
9494 int
9495 t4_deactivate_uld(struct adapter *sc, int id)
9496 {
9497 	int rc;
9498 	struct uld_info *ui;
9499 
9500 	ASSERT_SYNCHRONIZED_OP(sc);
9501 
9502 	if (id < 0 || id > ULD_MAX)
9503 		return (EINVAL);
9504 	rc = ENXIO;
9505 
9506 	sx_slock(&t4_uld_list_lock);
9507 
9508 	SLIST_FOREACH(ui, &t4_uld_list, link) {
9509 		if (ui->uld_id == id) {
9510 			rc = ui->deactivate(sc);
9511 			if (rc == 0) {
9512 				clrbit(&sc->active_ulds, id);
9513 				ui->refcount--;
9514 			}
9515 			break;
9516 		}
9517 	}
9518 
9519 	sx_sunlock(&t4_uld_list_lock);
9520 
9521 	return (rc);
9522 }
9523 
9524 int
9525 uld_active(struct adapter *sc, int uld_id)
9526 {
9527 
9528 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9529 
9530 	return (isset(&sc->active_ulds, uld_id));
9531 }
9532 #endif
9533 
9534 /*
9535  * t  = ptr to tunable.
9536  * nc = number of CPUs.
9537  * c  = compiled in default for that tunable.
9538  */
9539 static void
9540 calculate_nqueues(int *t, int nc, const int c)
9541 {
9542 	int nq;
9543 
9544 	if (*t > 0)
9545 		return;
9546 	nq = *t < 0 ? -*t : c;
9547 	*t = min(nc, nq);
9548 }
9549 
9550 /*
9551  * Come up with reasonable defaults for some of the tunables, provided they're
9552  * not set by the user (in which case we'll use the values as is).
9553  */
9554 static void
9555 tweak_tunables(void)
9556 {
9557 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
9558 
9559 	if (t4_ntxq10g < 1) {
9560 #ifdef RSS
9561 		t4_ntxq10g = rss_getnumbuckets();
9562 #else
9563 		calculate_nqueues(&t4_ntxq10g, nc, NTXQ_10G);
9564 #endif
9565 	}
9566 
9567 	if (t4_ntxq1g < 1) {
9568 #ifdef RSS
9569 		/* XXX: way too many for 1GbE? */
9570 		t4_ntxq1g = rss_getnumbuckets();
9571 #else
9572 		calculate_nqueues(&t4_ntxq1g, nc, NTXQ_1G);
9573 #endif
9574 	}
9575 
9576 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
9577 
9578 	if (t4_nrxq10g < 1) {
9579 #ifdef RSS
9580 		t4_nrxq10g = rss_getnumbuckets();
9581 #else
9582 		calculate_nqueues(&t4_nrxq10g, nc, NRXQ_10G);
9583 #endif
9584 	}
9585 
9586 	if (t4_nrxq1g < 1) {
9587 #ifdef RSS
9588 		/* XXX: way too many for 1GbE? */
9589 		t4_nrxq1g = rss_getnumbuckets();
9590 #else
9591 		calculate_nqueues(&t4_nrxq1g, nc, NRXQ_1G);
9592 #endif
9593 	}
9594 
9595 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
9596 
9597 #ifdef TCP_OFFLOAD
9598 	calculate_nqueues(&t4_nofldtxq10g, nc, NOFLDTXQ_10G);
9599 	calculate_nqueues(&t4_nofldtxq1g, nc, NOFLDTXQ_1G);
9600 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
9601 	calculate_nqueues(&t4_nofldrxq10g, nc, NOFLDRXQ_10G);
9602 	calculate_nqueues(&t4_nofldrxq1g, nc, NOFLDRXQ_1G);
9603 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
9604 
9605 	if (t4_toecaps_allowed == -1)
9606 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9607 
9608 	if (t4_rdmacaps_allowed == -1) {
9609 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9610 		    FW_CAPS_CONFIG_RDMA_RDMAC;
9611 	}
9612 
9613 	if (t4_iscsicaps_allowed == -1) {
9614 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9615 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9616 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
9617 	}
9618 #else
9619 	if (t4_toecaps_allowed == -1)
9620 		t4_toecaps_allowed = 0;
9621 
9622 	if (t4_rdmacaps_allowed == -1)
9623 		t4_rdmacaps_allowed = 0;
9624 
9625 	if (t4_iscsicaps_allowed == -1)
9626 		t4_iscsicaps_allowed = 0;
9627 #endif
9628 
9629 #ifdef DEV_NETMAP
9630 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
9631 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
9632 #endif
9633 
9634 	if (t4_tmr_idx_10g < 0 || t4_tmr_idx_10g >= SGE_NTIMERS)
9635 		t4_tmr_idx_10g = TMR_IDX_10G;
9636 
9637 	if (t4_pktc_idx_10g < -1 || t4_pktc_idx_10g >= SGE_NCOUNTERS)
9638 		t4_pktc_idx_10g = PKTC_IDX_10G;
9639 
9640 	if (t4_tmr_idx_1g < 0 || t4_tmr_idx_1g >= SGE_NTIMERS)
9641 		t4_tmr_idx_1g = TMR_IDX_1G;
9642 
9643 	if (t4_pktc_idx_1g < -1 || t4_pktc_idx_1g >= SGE_NCOUNTERS)
9644 		t4_pktc_idx_1g = PKTC_IDX_1G;
9645 
9646 	if (t4_qsize_txq < 128)
9647 		t4_qsize_txq = 128;
9648 
9649 	if (t4_qsize_rxq < 128)
9650 		t4_qsize_rxq = 128;
9651 	while (t4_qsize_rxq & 7)
9652 		t4_qsize_rxq++;
9653 
9654 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9655 }
9656 
9657 #ifdef DDB
9658 static void
9659 t4_dump_tcb(struct adapter *sc, int tid)
9660 {
9661 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
9662 
9663 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
9664 	save = t4_read_reg(sc, reg);
9665 	base = sc->memwin[2].mw_base;
9666 
9667 	/* Dump TCB for the tid */
9668 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
9669 	tcb_addr += tid * TCB_SIZE;
9670 
9671 	if (is_t4(sc)) {
9672 		pf = 0;
9673 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
9674 	} else {
9675 		pf = V_PFNUM(sc->pf);
9676 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
9677 	}
9678 	t4_write_reg(sc, reg, win_pos | pf);
9679 	t4_read_reg(sc, reg);
9680 
9681 	off = tcb_addr - win_pos;
9682 	for (i = 0; i < 4; i++) {
9683 		uint32_t buf[8];
9684 		for (j = 0; j < 8; j++, off += 4)
9685 			buf[j] = htonl(t4_read_reg(sc, base + off));
9686 
9687 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
9688 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
9689 		    buf[7]);
9690 	}
9691 
9692 	t4_write_reg(sc, reg, save);
9693 	t4_read_reg(sc, reg);
9694 }
9695 
9696 static void
9697 t4_dump_devlog(struct adapter *sc)
9698 {
9699 	struct devlog_params *dparams = &sc->params.devlog;
9700 	struct fw_devlog_e e;
9701 	int i, first, j, m, nentries, rc;
9702 	uint64_t ftstamp = UINT64_MAX;
9703 
9704 	if (dparams->start == 0) {
9705 		db_printf("devlog params not valid\n");
9706 		return;
9707 	}
9708 
9709 	nentries = dparams->size / sizeof(struct fw_devlog_e);
9710 	m = fwmtype_to_hwmtype(dparams->memtype);
9711 
9712 	/* Find the first entry. */
9713 	first = -1;
9714 	for (i = 0; i < nentries && !db_pager_quit; i++) {
9715 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9716 		    sizeof(e), (void *)&e);
9717 		if (rc != 0)
9718 			break;
9719 
9720 		if (e.timestamp == 0)
9721 			break;
9722 
9723 		e.timestamp = be64toh(e.timestamp);
9724 		if (e.timestamp < ftstamp) {
9725 			ftstamp = e.timestamp;
9726 			first = i;
9727 		}
9728 	}
9729 
9730 	if (first == -1)
9731 		return;
9732 
9733 	i = first;
9734 	do {
9735 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9736 		    sizeof(e), (void *)&e);
9737 		if (rc != 0)
9738 			return;
9739 
9740 		if (e.timestamp == 0)
9741 			return;
9742 
9743 		e.timestamp = be64toh(e.timestamp);
9744 		e.seqno = be32toh(e.seqno);
9745 		for (j = 0; j < 8; j++)
9746 			e.params[j] = be32toh(e.params[j]);
9747 
9748 		db_printf("%10d  %15ju  %8s  %8s  ",
9749 		    e.seqno, e.timestamp,
9750 		    (e.level < nitems(devlog_level_strings) ?
9751 			devlog_level_strings[e.level] : "UNKNOWN"),
9752 		    (e.facility < nitems(devlog_facility_strings) ?
9753 			devlog_facility_strings[e.facility] : "UNKNOWN"));
9754 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
9755 		    e.params[3], e.params[4], e.params[5], e.params[6],
9756 		    e.params[7]);
9757 
9758 		if (++i == nentries)
9759 			i = 0;
9760 	} while (i != first && !db_pager_quit);
9761 }
9762 
9763 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
9764 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
9765 
9766 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
9767 {
9768 	device_t dev;
9769 	int t;
9770 	bool valid;
9771 
9772 	valid = false;
9773 	t = db_read_token();
9774 	if (t == tIDENT) {
9775 		dev = device_lookup_by_name(db_tok_string);
9776 		valid = true;
9777 	}
9778 	db_skip_to_eol();
9779 	if (!valid) {
9780 		db_printf("usage: show t4 devlog <nexus>\n");
9781 		return;
9782 	}
9783 
9784 	if (dev == NULL) {
9785 		db_printf("device not found\n");
9786 		return;
9787 	}
9788 
9789 	t4_dump_devlog(device_get_softc(dev));
9790 }
9791 
9792 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
9793 {
9794 	device_t dev;
9795 	int radix, tid, t;
9796 	bool valid;
9797 
9798 	valid = false;
9799 	radix = db_radix;
9800 	db_radix = 10;
9801 	t = db_read_token();
9802 	if (t == tIDENT) {
9803 		dev = device_lookup_by_name(db_tok_string);
9804 		t = db_read_token();
9805 		if (t == tNUMBER) {
9806 			tid = db_tok_number;
9807 			valid = true;
9808 		}
9809 	}
9810 	db_radix = radix;
9811 	db_skip_to_eol();
9812 	if (!valid) {
9813 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
9814 		return;
9815 	}
9816 
9817 	if (dev == NULL) {
9818 		db_printf("device not found\n");
9819 		return;
9820 	}
9821 	if (tid < 0) {
9822 		db_printf("invalid tid\n");
9823 		return;
9824 	}
9825 
9826 	t4_dump_tcb(device_get_softc(dev), tid);
9827 }
9828 #endif
9829 
9830 static struct sx mlu;	/* mod load unload */
9831 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
9832 
9833 static int
9834 mod_event(module_t mod, int cmd, void *arg)
9835 {
9836 	int rc = 0;
9837 	static int loaded = 0;
9838 
9839 	switch (cmd) {
9840 	case MOD_LOAD:
9841 		sx_xlock(&mlu);
9842 		if (loaded++ == 0) {
9843 			t4_sge_modload();
9844 			t4_register_cpl_handler(CPL_SET_TCB_RPL, set_tcb_rpl);
9845 			t4_register_cpl_handler(CPL_L2T_WRITE_RPL, l2t_write_rpl);
9846 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
9847 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
9848 			sx_init(&t4_list_lock, "T4/T5 adapters");
9849 			SLIST_INIT(&t4_list);
9850 #ifdef TCP_OFFLOAD
9851 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
9852 			SLIST_INIT(&t4_uld_list);
9853 #endif
9854 			t4_tracer_modload();
9855 			tweak_tunables();
9856 		}
9857 		sx_xunlock(&mlu);
9858 		break;
9859 
9860 	case MOD_UNLOAD:
9861 		sx_xlock(&mlu);
9862 		if (--loaded == 0) {
9863 			int tries;
9864 
9865 			sx_slock(&t4_list_lock);
9866 			if (!SLIST_EMPTY(&t4_list)) {
9867 				rc = EBUSY;
9868 				sx_sunlock(&t4_list_lock);
9869 				goto done_unload;
9870 			}
9871 #ifdef TCP_OFFLOAD
9872 			sx_slock(&t4_uld_list_lock);
9873 			if (!SLIST_EMPTY(&t4_uld_list)) {
9874 				rc = EBUSY;
9875 				sx_sunlock(&t4_uld_list_lock);
9876 				sx_sunlock(&t4_list_lock);
9877 				goto done_unload;
9878 			}
9879 #endif
9880 			tries = 0;
9881 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
9882 				uprintf("%ju clusters with custom free routine "
9883 				    "still is use.\n", t4_sge_extfree_refs());
9884 				pause("t4unload", 2 * hz);
9885 			}
9886 #ifdef TCP_OFFLOAD
9887 			sx_sunlock(&t4_uld_list_lock);
9888 #endif
9889 			sx_sunlock(&t4_list_lock);
9890 
9891 			if (t4_sge_extfree_refs() == 0) {
9892 				t4_tracer_modunload();
9893 #ifdef TCP_OFFLOAD
9894 				sx_destroy(&t4_uld_list_lock);
9895 #endif
9896 				sx_destroy(&t4_list_lock);
9897 				t4_sge_modunload();
9898 				loaded = 0;
9899 			} else {
9900 				rc = EBUSY;
9901 				loaded++;	/* undo earlier decrement */
9902 			}
9903 		}
9904 done_unload:
9905 		sx_xunlock(&mlu);
9906 		break;
9907 	}
9908 
9909 	return (rc);
9910 }
9911 
9912 static devclass_t t4_devclass, t5_devclass, t6_devclass;
9913 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
9914 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
9915 
9916 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
9917 MODULE_VERSION(t4nex, 1);
9918 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
9919 #ifdef DEV_NETMAP
9920 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
9921 #endif /* DEV_NETMAP */
9922 
9923 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
9924 MODULE_VERSION(t5nex, 1);
9925 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
9926 #ifdef DEV_NETMAP
9927 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
9928 #endif /* DEV_NETMAP */
9929 
9930 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
9931 MODULE_VERSION(t6nex, 1);
9932 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
9933 #ifdef DEV_NETMAP
9934 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
9935 #endif /* DEV_NETMAP */
9936 
9937 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
9938 MODULE_VERSION(cxgbe, 1);
9939 
9940 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
9941 MODULE_VERSION(cxl, 1);
9942 
9943 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
9944 MODULE_VERSION(cc, 1);
9945 
9946 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
9947 MODULE_VERSION(vcxgbe, 1);
9948 
9949 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
9950 MODULE_VERSION(vcxl, 1);
9951 
9952 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
9953 MODULE_VERSION(vcc, 1);
9954