xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision 2e5b60079b7d8c3ca68f1390cd90f305e651f8d3)
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_inet.h"
32 #include "opt_inet6.h"
33 
34 #include <sys/param.h>
35 #include <sys/conf.h>
36 #include <sys/priv.h>
37 #include <sys/kernel.h>
38 #include <sys/bus.h>
39 #include <sys/module.h>
40 #include <sys/malloc.h>
41 #include <sys/queue.h>
42 #include <sys/taskqueue.h>
43 #include <sys/pciio.h>
44 #include <dev/pci/pcireg.h>
45 #include <dev/pci/pcivar.h>
46 #include <dev/pci/pci_private.h>
47 #include <sys/firmware.h>
48 #include <sys/sbuf.h>
49 #include <sys/smp.h>
50 #include <sys/socket.h>
51 #include <sys/sockio.h>
52 #include <sys/sysctl.h>
53 #include <net/ethernet.h>
54 #include <net/if.h>
55 #include <net/if_types.h>
56 #include <net/if_dl.h>
57 #include <net/if_vlan_var.h>
58 #if defined(__i386__) || defined(__amd64__)
59 #include <vm/vm.h>
60 #include <vm/pmap.h>
61 #endif
62 
63 #include "common/common.h"
64 #include "common/t4_msg.h"
65 #include "common/t4_regs.h"
66 #include "common/t4_regs_values.h"
67 #include "t4_ioctl.h"
68 #include "t4_l2t.h"
69 #include "t4_mp_ring.h"
70 
71 /* T4 bus driver interface */
72 static int t4_probe(device_t);
73 static int t4_attach(device_t);
74 static int t4_detach(device_t);
75 static device_method_t t4_methods[] = {
76 	DEVMETHOD(device_probe,		t4_probe),
77 	DEVMETHOD(device_attach,	t4_attach),
78 	DEVMETHOD(device_detach,	t4_detach),
79 
80 	DEVMETHOD_END
81 };
82 static driver_t t4_driver = {
83 	"t4nex",
84 	t4_methods,
85 	sizeof(struct adapter)
86 };
87 
88 
89 /* T4 port (cxgbe) interface */
90 static int cxgbe_probe(device_t);
91 static int cxgbe_attach(device_t);
92 static int cxgbe_detach(device_t);
93 static device_method_t cxgbe_methods[] = {
94 	DEVMETHOD(device_probe,		cxgbe_probe),
95 	DEVMETHOD(device_attach,	cxgbe_attach),
96 	DEVMETHOD(device_detach,	cxgbe_detach),
97 	{ 0, 0 }
98 };
99 static driver_t cxgbe_driver = {
100 	"cxgbe",
101 	cxgbe_methods,
102 	sizeof(struct port_info)
103 };
104 
105 static d_ioctl_t t4_ioctl;
106 static d_open_t t4_open;
107 static d_close_t t4_close;
108 
109 static struct cdevsw t4_cdevsw = {
110        .d_version = D_VERSION,
111        .d_flags = 0,
112        .d_open = t4_open,
113        .d_close = t4_close,
114        .d_ioctl = t4_ioctl,
115        .d_name = "t4nex",
116 };
117 
118 /* T5 bus driver interface */
119 static int t5_probe(device_t);
120 static device_method_t t5_methods[] = {
121 	DEVMETHOD(device_probe,		t5_probe),
122 	DEVMETHOD(device_attach,	t4_attach),
123 	DEVMETHOD(device_detach,	t4_detach),
124 
125 	DEVMETHOD_END
126 };
127 static driver_t t5_driver = {
128 	"t5nex",
129 	t5_methods,
130 	sizeof(struct adapter)
131 };
132 
133 
134 /* T5 port (cxl) interface */
135 static driver_t cxl_driver = {
136 	"cxl",
137 	cxgbe_methods,
138 	sizeof(struct port_info)
139 };
140 
141 static struct cdevsw t5_cdevsw = {
142        .d_version = D_VERSION,
143        .d_flags = 0,
144        .d_open = t4_open,
145        .d_close = t4_close,
146        .d_ioctl = t4_ioctl,
147        .d_name = "t5nex",
148 };
149 
150 /* ifnet + media interface */
151 static void cxgbe_init(void *);
152 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
153 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
154 static void cxgbe_qflush(struct ifnet *);
155 static uint64_t cxgbe_get_counter(struct ifnet *, ift_counter);
156 static int cxgbe_media_change(struct ifnet *);
157 static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
158 
159 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
160 
161 /*
162  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
163  * then ADAPTER_LOCK, then t4_uld_list_lock.
164  */
165 static struct sx t4_list_lock;
166 SLIST_HEAD(, adapter) t4_list;
167 #ifdef TCP_OFFLOAD
168 static struct sx t4_uld_list_lock;
169 SLIST_HEAD(, uld_info) t4_uld_list;
170 #endif
171 
172 /*
173  * Tunables.  See tweak_tunables() too.
174  *
175  * Each tunable is set to a default value here if it's known at compile-time.
176  * Otherwise it is set to -1 as an indication to tweak_tunables() that it should
177  * provide a reasonable default when the driver is loaded.
178  *
179  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
180  * T5 are under hw.cxl.
181  */
182 
183 /*
184  * Number of queues for tx and rx, 10G and 1G, NIC and offload.
185  */
186 #define NTXQ_10G 16
187 static int t4_ntxq10g = -1;
188 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq10g);
189 
190 #define NRXQ_10G 8
191 static int t4_nrxq10g = -1;
192 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq10g);
193 
194 #define NTXQ_1G 4
195 static int t4_ntxq1g = -1;
196 TUNABLE_INT("hw.cxgbe.ntxq1g", &t4_ntxq1g);
197 
198 #define NRXQ_1G 2
199 static int t4_nrxq1g = -1;
200 TUNABLE_INT("hw.cxgbe.nrxq1g", &t4_nrxq1g);
201 
202 static int t4_rsrv_noflowq = 0;
203 TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
204 
205 #ifdef TCP_OFFLOAD
206 #define NOFLDTXQ_10G 8
207 static int t4_nofldtxq10g = -1;
208 TUNABLE_INT("hw.cxgbe.nofldtxq10g", &t4_nofldtxq10g);
209 
210 #define NOFLDRXQ_10G 2
211 static int t4_nofldrxq10g = -1;
212 TUNABLE_INT("hw.cxgbe.nofldrxq10g", &t4_nofldrxq10g);
213 
214 #define NOFLDTXQ_1G 2
215 static int t4_nofldtxq1g = -1;
216 TUNABLE_INT("hw.cxgbe.nofldtxq1g", &t4_nofldtxq1g);
217 
218 #define NOFLDRXQ_1G 1
219 static int t4_nofldrxq1g = -1;
220 TUNABLE_INT("hw.cxgbe.nofldrxq1g", &t4_nofldrxq1g);
221 #endif
222 
223 #ifdef DEV_NETMAP
224 #define NNMTXQ_10G 2
225 static int t4_nnmtxq10g = -1;
226 TUNABLE_INT("hw.cxgbe.nnmtxq10g", &t4_nnmtxq10g);
227 
228 #define NNMRXQ_10G 2
229 static int t4_nnmrxq10g = -1;
230 TUNABLE_INT("hw.cxgbe.nnmrxq10g", &t4_nnmrxq10g);
231 
232 #define NNMTXQ_1G 1
233 static int t4_nnmtxq1g = -1;
234 TUNABLE_INT("hw.cxgbe.nnmtxq1g", &t4_nnmtxq1g);
235 
236 #define NNMRXQ_1G 1
237 static int t4_nnmrxq1g = -1;
238 TUNABLE_INT("hw.cxgbe.nnmrxq1g", &t4_nnmrxq1g);
239 #endif
240 
241 /*
242  * Holdoff parameters for 10G and 1G ports.
243  */
244 #define TMR_IDX_10G 1
245 static int t4_tmr_idx_10g = TMR_IDX_10G;
246 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx_10g);
247 
248 #define PKTC_IDX_10G (-1)
249 static int t4_pktc_idx_10g = PKTC_IDX_10G;
250 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx_10g);
251 
252 #define TMR_IDX_1G 1
253 static int t4_tmr_idx_1g = TMR_IDX_1G;
254 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_1G", &t4_tmr_idx_1g);
255 
256 #define PKTC_IDX_1G (-1)
257 static int t4_pktc_idx_1g = PKTC_IDX_1G;
258 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_1G", &t4_pktc_idx_1g);
259 
260 /*
261  * Size (# of entries) of each tx and rx queue.
262  */
263 static unsigned int t4_qsize_txq = TX_EQ_QSIZE;
264 TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
265 
266 static unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
267 TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
268 
269 /*
270  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
271  */
272 static int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
273 TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
274 
275 /*
276  * Configuration file.
277  */
278 #define DEFAULT_CF	"default"
279 #define FLASH_CF	"flash"
280 #define UWIRE_CF	"uwire"
281 #define FPGA_CF		"fpga"
282 static char t4_cfg_file[32] = DEFAULT_CF;
283 TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
284 
285 /*
286  * PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
287  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
288  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
289  *            mark or when signalled to do so, 0 to never emit PAUSE.
290  */
291 static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
292 TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
293 
294 /*
295  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
296  * encouraged respectively).
297  */
298 static unsigned int t4_fw_install = 1;
299 TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
300 
301 /*
302  * ASIC features that will be used.  Disable the ones you don't want so that the
303  * chip resources aren't wasted on features that will not be used.
304  */
305 static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
306 TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
307 
308 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC;
309 TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
310 
311 static int t4_toecaps_allowed = -1;
312 TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
313 
314 static int t4_rdmacaps_allowed = 0;
315 TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
316 
317 static int t4_iscsicaps_allowed = 0;
318 TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
319 
320 static int t4_fcoecaps_allowed = 0;
321 TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
322 
323 static int t5_write_combine = 0;
324 TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
325 
326 struct intrs_and_queues {
327 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
328 	uint16_t nirq;		/* Total # of vectors */
329 	uint16_t intr_flags_10g;/* Interrupt flags for each 10G port */
330 	uint16_t intr_flags_1g;	/* Interrupt flags for each 1G port */
331 	uint16_t ntxq10g;	/* # of NIC txq's for each 10G port */
332 	uint16_t nrxq10g;	/* # of NIC rxq's for each 10G port */
333 	uint16_t ntxq1g;	/* # of NIC txq's for each 1G port */
334 	uint16_t nrxq1g;	/* # of NIC rxq's for each 1G port */
335 	uint16_t rsrv_noflowq;	/* Flag whether to reserve queue 0 */
336 #ifdef TCP_OFFLOAD
337 	uint16_t nofldtxq10g;	/* # of TOE txq's for each 10G port */
338 	uint16_t nofldrxq10g;	/* # of TOE rxq's for each 10G port */
339 	uint16_t nofldtxq1g;	/* # of TOE txq's for each 1G port */
340 	uint16_t nofldrxq1g;	/* # of TOE rxq's for each 1G port */
341 #endif
342 #ifdef DEV_NETMAP
343 	uint16_t nnmtxq10g;	/* # of netmap txq's for each 10G port */
344 	uint16_t nnmrxq10g;	/* # of netmap rxq's for each 10G port */
345 	uint16_t nnmtxq1g;	/* # of netmap txq's for each 1G port */
346 	uint16_t nnmrxq1g;	/* # of netmap rxq's for each 1G port */
347 #endif
348 };
349 
350 struct filter_entry {
351         uint32_t valid:1;	/* filter allocated and valid */
352         uint32_t locked:1;	/* filter is administratively locked */
353         uint32_t pending:1;	/* filter action is pending firmware reply */
354 	uint32_t smtidx:8;	/* Source MAC Table index for smac */
355 	struct l2t_entry *l2t;	/* Layer Two Table entry for dmac */
356 
357         struct t4_filter_specification fs;
358 };
359 
360 static int map_bars_0_and_4(struct adapter *);
361 static int map_bar_2(struct adapter *);
362 static void setup_memwin(struct adapter *);
363 static int validate_mem_range(struct adapter *, uint32_t, int);
364 static int fwmtype_to_hwmtype(int);
365 static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
366     uint32_t *);
367 static void memwin_info(struct adapter *, int, uint32_t *, uint32_t *);
368 static uint32_t position_memwin(struct adapter *, int, uint32_t);
369 static int cfg_itype_and_nqueues(struct adapter *, int, int,
370     struct intrs_and_queues *);
371 static int prep_firmware(struct adapter *);
372 static int partition_resources(struct adapter *, const struct firmware *,
373     const char *);
374 static int get_params__pre_init(struct adapter *);
375 static int get_params__post_init(struct adapter *);
376 static int set_params__post_init(struct adapter *);
377 static void t4_set_desc(struct adapter *);
378 static void build_medialist(struct port_info *, struct ifmedia *);
379 static int cxgbe_init_synchronized(struct port_info *);
380 static int cxgbe_uninit_synchronized(struct port_info *);
381 static int setup_intr_handlers(struct adapter *);
382 static void quiesce_txq(struct adapter *, struct sge_txq *);
383 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
384 static void quiesce_iq(struct adapter *, struct sge_iq *);
385 static void quiesce_fl(struct adapter *, struct sge_fl *);
386 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
387     driver_intr_t *, void *, char *);
388 static int t4_free_irq(struct adapter *, struct irq *);
389 static void reg_block_dump(struct adapter *, uint8_t *, unsigned int,
390     unsigned int);
391 static void t4_get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
392 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
393 static void cxgbe_tick(void *);
394 static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
395 static int cpl_not_handled(struct sge_iq *, const struct rss_header *,
396     struct mbuf *);
397 static int an_not_handled(struct sge_iq *, const struct rsp_ctrl *);
398 static int fw_msg_not_handled(struct adapter *, const __be64 *);
399 static int t4_sysctls(struct adapter *);
400 static int cxgbe_sysctls(struct port_info *);
401 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
402 static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
403 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
404 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
405 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
406 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
407 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
408 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
409 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
410 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
411 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
412 #ifdef SBUF_DRAIN
413 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
414 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
415 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
416 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
417 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
418 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
419 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
420 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
421 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
422 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
423 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
424 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
425 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
426 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
427 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
428 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
429 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
430 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
431 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
432 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
433 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
434 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
435 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
436 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
437 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
438 #endif
439 static uint32_t fconf_to_mode(uint32_t);
440 static uint32_t mode_to_fconf(uint32_t);
441 static uint32_t fspec_to_fconf(struct t4_filter_specification *);
442 static int get_filter_mode(struct adapter *, uint32_t *);
443 static int set_filter_mode(struct adapter *, uint32_t);
444 static inline uint64_t get_filter_hits(struct adapter *, uint32_t);
445 static int get_filter(struct adapter *, struct t4_filter *);
446 static int set_filter(struct adapter *, struct t4_filter *);
447 static int del_filter(struct adapter *, struct t4_filter *);
448 static void clear_filter(struct filter_entry *);
449 static int set_filter_wr(struct adapter *, int);
450 static int del_filter_wr(struct adapter *, int);
451 static int get_sge_context(struct adapter *, struct t4_sge_context *);
452 static int load_fw(struct adapter *, struct t4_data *);
453 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
454 static int read_i2c(struct adapter *, struct t4_i2c_data *);
455 static int set_sched_class(struct adapter *, struct t4_sched_params *);
456 static int set_sched_queue(struct adapter *, struct t4_sched_queue *);
457 #ifdef TCP_OFFLOAD
458 static int toe_capability(struct port_info *, int);
459 #endif
460 static int mod_event(module_t, int, void *);
461 
462 struct {
463 	uint16_t device;
464 	char *desc;
465 } t4_pciids[] = {
466 	{0xa000, "Chelsio Terminator 4 FPGA"},
467 	{0x4400, "Chelsio T440-dbg"},
468 	{0x4401, "Chelsio T420-CR"},
469 	{0x4402, "Chelsio T422-CR"},
470 	{0x4403, "Chelsio T440-CR"},
471 	{0x4404, "Chelsio T420-BCH"},
472 	{0x4405, "Chelsio T440-BCH"},
473 	{0x4406, "Chelsio T440-CH"},
474 	{0x4407, "Chelsio T420-SO"},
475 	{0x4408, "Chelsio T420-CX"},
476 	{0x4409, "Chelsio T420-BT"},
477 	{0x440a, "Chelsio T404-BT"},
478 	{0x440e, "Chelsio T440-LP-CR"},
479 }, t5_pciids[] = {
480 	{0xb000, "Chelsio Terminator 5 FPGA"},
481 	{0x5400, "Chelsio T580-dbg"},
482 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
483 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
484 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
485 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
486 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
487 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
488 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
489 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
490 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
491 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
492 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
493 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
494 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
495 #ifdef notyet
496 	{0x5404,  "Chelsio T520-BCH"},
497 	{0x5405,  "Chelsio T540-BCH"},
498 	{0x5406,  "Chelsio T540-CH"},
499 	{0x5408,  "Chelsio T520-CX"},
500 	{0x540b,  "Chelsio B520-SR"},
501 	{0x540c,  "Chelsio B504-BT"},
502 	{0x540f,  "Chelsio Amsterdam"},
503 	{0x5413,  "Chelsio T580-CHR"},
504 #endif
505 };
506 
507 #ifdef TCP_OFFLOAD
508 /*
509  * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
510  * exactly the same for both rxq and ofld_rxq.
511  */
512 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
513 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
514 #endif
515 
516 /* No easy way to include t4_msg.h before adapter.h so we check this way */
517 CTASSERT(nitems(((struct adapter *)0)->cpl_handler) == NUM_CPL_CMDS);
518 CTASSERT(nitems(((struct adapter *)0)->fw_msg_handler) == NUM_FW6_TYPES);
519 
520 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
521 
522 static int
523 t4_probe(device_t dev)
524 {
525 	int i;
526 	uint16_t v = pci_get_vendor(dev);
527 	uint16_t d = pci_get_device(dev);
528 	uint8_t f = pci_get_function(dev);
529 
530 	if (v != PCI_VENDOR_ID_CHELSIO)
531 		return (ENXIO);
532 
533 	/* Attach only to PF0 of the FPGA */
534 	if (d == 0xa000 && f != 0)
535 		return (ENXIO);
536 
537 	for (i = 0; i < nitems(t4_pciids); i++) {
538 		if (d == t4_pciids[i].device) {
539 			device_set_desc(dev, t4_pciids[i].desc);
540 			return (BUS_PROBE_DEFAULT);
541 		}
542 	}
543 
544 	return (ENXIO);
545 }
546 
547 static int
548 t5_probe(device_t dev)
549 {
550 	int i;
551 	uint16_t v = pci_get_vendor(dev);
552 	uint16_t d = pci_get_device(dev);
553 	uint8_t f = pci_get_function(dev);
554 
555 	if (v != PCI_VENDOR_ID_CHELSIO)
556 		return (ENXIO);
557 
558 	/* Attach only to PF0 of the FPGA */
559 	if (d == 0xb000 && f != 0)
560 		return (ENXIO);
561 
562 	for (i = 0; i < nitems(t5_pciids); i++) {
563 		if (d == t5_pciids[i].device) {
564 			device_set_desc(dev, t5_pciids[i].desc);
565 			return (BUS_PROBE_DEFAULT);
566 		}
567 	}
568 
569 	return (ENXIO);
570 }
571 
572 static int
573 t4_attach(device_t dev)
574 {
575 	struct adapter *sc;
576 	int rc = 0, i, n10g, n1g, rqidx, tqidx;
577 	struct intrs_and_queues iaq;
578 	struct sge *s;
579 #ifdef TCP_OFFLOAD
580 	int ofld_rqidx, ofld_tqidx;
581 #endif
582 #ifdef DEV_NETMAP
583 	int nm_rqidx, nm_tqidx;
584 #endif
585 	const char *pcie_ts;
586 
587 	sc = device_get_softc(dev);
588 	sc->dev = dev;
589 
590 	pci_enable_busmaster(dev);
591 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
592 		uint32_t v;
593 
594 		pci_set_max_read_req(dev, 4096);
595 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
596 		v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
597 		pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
598 
599 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
600 	}
601 
602 	sc->traceq = -1;
603 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
604 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
605 	    device_get_nameunit(dev));
606 
607 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
608 	    device_get_nameunit(dev));
609 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
610 	sx_xlock(&t4_list_lock);
611 	SLIST_INSERT_HEAD(&t4_list, sc, link);
612 	sx_xunlock(&t4_list_lock);
613 
614 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
615 	TAILQ_INIT(&sc->sfl);
616 	callout_init(&sc->sfl_callout, CALLOUT_MPSAFE);
617 
618 	mtx_init(&sc->regwin_lock, "register and memory window", 0, MTX_DEF);
619 
620 	rc = map_bars_0_and_4(sc);
621 	if (rc != 0)
622 		goto done; /* error message displayed already */
623 
624 	/*
625 	 * This is the real PF# to which we're attaching.  Works from within PCI
626 	 * passthrough environments too, where pci_get_function() could return a
627 	 * different PF# depending on the passthrough configuration.  We need to
628 	 * use the real PF# in all our communication with the firmware.
629 	 */
630 	sc->pf = G_SOURCEPF(t4_read_reg(sc, A_PL_WHOAMI));
631 	sc->mbox = sc->pf;
632 
633 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
634 	sc->an_handler = an_not_handled;
635 	for (i = 0; i < nitems(sc->cpl_handler); i++)
636 		sc->cpl_handler[i] = cpl_not_handled;
637 	for (i = 0; i < nitems(sc->fw_msg_handler); i++)
638 		sc->fw_msg_handler[i] = fw_msg_not_handled;
639 	t4_register_cpl_handler(sc, CPL_SET_TCB_RPL, t4_filter_rpl);
640 	t4_register_cpl_handler(sc, CPL_TRACE_PKT, t4_trace_pkt);
641 	t4_register_cpl_handler(sc, CPL_TRACE_PKT_T5, t5_trace_pkt);
642 	t4_init_sge_cpl_handlers(sc);
643 
644 	/* Prepare the adapter for operation */
645 	rc = -t4_prep_adapter(sc);
646 	if (rc != 0) {
647 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
648 		goto done;
649 	}
650 
651 	/*
652 	 * Do this really early, with the memory windows set up even before the
653 	 * character device.  The userland tool's register i/o and mem read
654 	 * will work even in "recovery mode".
655 	 */
656 	setup_memwin(sc);
657 	sc->cdev = make_dev(is_t4(sc) ? &t4_cdevsw : &t5_cdevsw,
658 	    device_get_unit(dev), UID_ROOT, GID_WHEEL, 0600, "%s",
659 	    device_get_nameunit(dev));
660 	if (sc->cdev == NULL)
661 		device_printf(dev, "failed to create nexus char device.\n");
662 	else
663 		sc->cdev->si_drv1 = sc;
664 
665 	/* Go no further if recovery mode has been requested. */
666 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
667 		device_printf(dev, "recovery mode.\n");
668 		goto done;
669 	}
670 
671 #if defined(__i386__)
672 	if ((cpu_feature & CPUID_CX8) == 0) {
673 		device_printf(dev, "64 bit atomics not available.\n");
674 		rc = ENOTSUP;
675 		goto done;
676 	}
677 #endif
678 
679 	/* Prepare the firmware for operation */
680 	rc = prep_firmware(sc);
681 	if (rc != 0)
682 		goto done; /* error message displayed already */
683 
684 	rc = get_params__post_init(sc);
685 	if (rc != 0)
686 		goto done; /* error message displayed already */
687 
688 	rc = set_params__post_init(sc);
689 	if (rc != 0)
690 		goto done; /* error message displayed already */
691 
692 	rc = map_bar_2(sc);
693 	if (rc != 0)
694 		goto done; /* error message displayed already */
695 
696 	rc = t4_create_dma_tag(sc);
697 	if (rc != 0)
698 		goto done; /* error message displayed already */
699 
700 	/*
701 	 * First pass over all the ports - allocate VIs and initialize some
702 	 * basic parameters like mac address, port type, etc.  We also figure
703 	 * out whether a port is 10G or 1G and use that information when
704 	 * calculating how many interrupts to attempt to allocate.
705 	 */
706 	n10g = n1g = 0;
707 	for_each_port(sc, i) {
708 		struct port_info *pi;
709 
710 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
711 		sc->port[i] = pi;
712 
713 		/* These must be set before t4_port_init */
714 		pi->adapter = sc;
715 		pi->port_id = i;
716 
717 		/* Allocate the vi and initialize parameters like mac addr */
718 		rc = -t4_port_init(pi, sc->mbox, sc->pf, 0);
719 		if (rc != 0) {
720 			device_printf(dev, "unable to initialize port %d: %d\n",
721 			    i, rc);
722 			free(pi, M_CXGBE);
723 			sc->port[i] = NULL;
724 			goto done;
725 		}
726 
727 		pi->link_cfg.requested_fc &= ~(PAUSE_TX | PAUSE_RX);
728 		pi->link_cfg.requested_fc |= t4_pause_settings;
729 		pi->link_cfg.fc &= ~(PAUSE_TX | PAUSE_RX);
730 		pi->link_cfg.fc |= t4_pause_settings;
731 
732 		rc = -t4_link_start(sc, sc->mbox, pi->tx_chan, &pi->link_cfg);
733 		if (rc != 0) {
734 			device_printf(dev, "port %d l1cfg failed: %d\n", i, rc);
735 			free(pi, M_CXGBE);
736 			sc->port[i] = NULL;
737 			goto done;
738 		}
739 
740 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
741 		    device_get_nameunit(dev), i);
742 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
743 		sc->chan_map[pi->tx_chan] = i;
744 
745 		if (is_10G_port(pi) || is_40G_port(pi)) {
746 			n10g++;
747 			pi->tmr_idx = t4_tmr_idx_10g;
748 			pi->pktc_idx = t4_pktc_idx_10g;
749 		} else {
750 			n1g++;
751 			pi->tmr_idx = t4_tmr_idx_1g;
752 			pi->pktc_idx = t4_pktc_idx_1g;
753 		}
754 
755 		pi->xact_addr_filt = -1;
756 		pi->linkdnrc = -1;
757 
758 		pi->qsize_rxq = t4_qsize_rxq;
759 		pi->qsize_txq = t4_qsize_txq;
760 
761 		pi->dev = device_add_child(dev, is_t4(sc) ? "cxgbe" : "cxl", -1);
762 		if (pi->dev == NULL) {
763 			device_printf(dev,
764 			    "failed to add device for port %d.\n", i);
765 			rc = ENXIO;
766 			goto done;
767 		}
768 		device_set_softc(pi->dev, pi);
769 	}
770 
771 	/*
772 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
773 	 */
774 	rc = cfg_itype_and_nqueues(sc, n10g, n1g, &iaq);
775 	if (rc != 0)
776 		goto done; /* error message displayed already */
777 
778 	sc->intr_type = iaq.intr_type;
779 	sc->intr_count = iaq.nirq;
780 
781 	s = &sc->sge;
782 	s->nrxq = n10g * iaq.nrxq10g + n1g * iaq.nrxq1g;
783 	s->ntxq = n10g * iaq.ntxq10g + n1g * iaq.ntxq1g;
784 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
785 	s->neq += sc->params.nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
786 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
787 #ifdef TCP_OFFLOAD
788 	if (is_offload(sc)) {
789 		s->nofldrxq = n10g * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
790 		s->nofldtxq = n10g * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
791 		s->neq += s->nofldtxq + s->nofldrxq;
792 		s->niq += s->nofldrxq;
793 
794 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
795 		    M_CXGBE, M_ZERO | M_WAITOK);
796 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
797 		    M_CXGBE, M_ZERO | M_WAITOK);
798 	}
799 #endif
800 #ifdef DEV_NETMAP
801 	s->nnmrxq = n10g * iaq.nnmrxq10g + n1g * iaq.nnmrxq1g;
802 	s->nnmtxq = n10g * iaq.nnmtxq10g + n1g * iaq.nnmtxq1g;
803 	s->neq += s->nnmtxq + s->nnmrxq;
804 	s->niq += s->nnmrxq;
805 
806 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
807 	    M_CXGBE, M_ZERO | M_WAITOK);
808 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
809 	    M_CXGBE, M_ZERO | M_WAITOK);
810 #endif
811 
812 	s->ctrlq = malloc(sc->params.nports * sizeof(struct sge_wrq), M_CXGBE,
813 	    M_ZERO | M_WAITOK);
814 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
815 	    M_ZERO | M_WAITOK);
816 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
817 	    M_ZERO | M_WAITOK);
818 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
819 	    M_ZERO | M_WAITOK);
820 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
821 	    M_ZERO | M_WAITOK);
822 
823 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
824 	    M_ZERO | M_WAITOK);
825 
826 	t4_init_l2t(sc, M_WAITOK);
827 
828 	/*
829 	 * Second pass over the ports.  This time we know the number of rx and
830 	 * tx queues that each port should get.
831 	 */
832 	rqidx = tqidx = 0;
833 #ifdef TCP_OFFLOAD
834 	ofld_rqidx = ofld_tqidx = 0;
835 #endif
836 #ifdef DEV_NETMAP
837 	nm_rqidx = nm_tqidx = 0;
838 #endif
839 	for_each_port(sc, i) {
840 		struct port_info *pi = sc->port[i];
841 
842 		if (pi == NULL)
843 			continue;
844 
845 		pi->first_rxq = rqidx;
846 		pi->first_txq = tqidx;
847 		if (is_10G_port(pi) || is_40G_port(pi)) {
848 			pi->flags |= iaq.intr_flags_10g;
849 			pi->nrxq = iaq.nrxq10g;
850 			pi->ntxq = iaq.ntxq10g;
851 		} else {
852 			pi->flags |= iaq.intr_flags_1g;
853 			pi->nrxq = iaq.nrxq1g;
854 			pi->ntxq = iaq.ntxq1g;
855 		}
856 
857 		if (pi->ntxq > 1)
858 			pi->rsrv_noflowq = iaq.rsrv_noflowq ? 1 : 0;
859 		else
860 			pi->rsrv_noflowq = 0;
861 
862 		rqidx += pi->nrxq;
863 		tqidx += pi->ntxq;
864 #ifdef TCP_OFFLOAD
865 		if (is_offload(sc)) {
866 			pi->first_ofld_rxq = ofld_rqidx;
867 			pi->first_ofld_txq = ofld_tqidx;
868 			if (is_10G_port(pi) || is_40G_port(pi)) {
869 				pi->nofldrxq = iaq.nofldrxq10g;
870 				pi->nofldtxq = iaq.nofldtxq10g;
871 			} else {
872 				pi->nofldrxq = iaq.nofldrxq1g;
873 				pi->nofldtxq = iaq.nofldtxq1g;
874 			}
875 			ofld_rqidx += pi->nofldrxq;
876 			ofld_tqidx += pi->nofldtxq;
877 		}
878 #endif
879 #ifdef DEV_NETMAP
880 		pi->first_nm_rxq = nm_rqidx;
881 		pi->first_nm_txq = nm_tqidx;
882 		if (is_10G_port(pi) || is_40G_port(pi)) {
883 			pi->nnmrxq = iaq.nnmrxq10g;
884 			pi->nnmtxq = iaq.nnmtxq10g;
885 		} else {
886 			pi->nnmrxq = iaq.nnmrxq1g;
887 			pi->nnmtxq = iaq.nnmtxq1g;
888 		}
889 		nm_rqidx += pi->nnmrxq;
890 		nm_tqidx += pi->nnmtxq;
891 #endif
892 	}
893 
894 	rc = setup_intr_handlers(sc);
895 	if (rc != 0) {
896 		device_printf(dev,
897 		    "failed to setup interrupt handlers: %d\n", rc);
898 		goto done;
899 	}
900 
901 	rc = bus_generic_attach(dev);
902 	if (rc != 0) {
903 		device_printf(dev,
904 		    "failed to attach all child ports: %d\n", rc);
905 		goto done;
906 	}
907 
908 	switch (sc->params.pci.speed) {
909 		case 0x1:
910 			pcie_ts = "2.5";
911 			break;
912 		case 0x2:
913 			pcie_ts = "5.0";
914 			break;
915 		case 0x3:
916 			pcie_ts = "8.0";
917 			break;
918 		default:
919 			pcie_ts = "??";
920 			break;
921 	}
922 	device_printf(dev,
923 	    "PCIe x%d (%s GTS/s) (%d), %d ports, %d %s interrupt%s, %d eq, %d iq\n",
924 	    sc->params.pci.width, pcie_ts, sc->params.pci.speed,
925 	    sc->params.nports, sc->intr_count,
926 	    sc->intr_type == INTR_MSIX ? "MSI-X" :
927 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
928 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
929 
930 	t4_set_desc(sc);
931 
932 done:
933 	if (rc != 0 && sc->cdev) {
934 		/* cdev was created and so cxgbetool works; recover that way. */
935 		device_printf(dev,
936 		    "error during attach, adapter is now in recovery mode.\n");
937 		rc = 0;
938 	}
939 
940 	if (rc != 0)
941 		t4_detach(dev);
942 	else
943 		t4_sysctls(sc);
944 
945 	return (rc);
946 }
947 
948 /*
949  * Idempotent
950  */
951 static int
952 t4_detach(device_t dev)
953 {
954 	struct adapter *sc;
955 	struct port_info *pi;
956 	int i, rc;
957 
958 	sc = device_get_softc(dev);
959 
960 	if (sc->flags & FULL_INIT_DONE)
961 		t4_intr_disable(sc);
962 
963 	if (sc->cdev) {
964 		destroy_dev(sc->cdev);
965 		sc->cdev = NULL;
966 	}
967 
968 	rc = bus_generic_detach(dev);
969 	if (rc) {
970 		device_printf(dev,
971 		    "failed to detach child devices: %d\n", rc);
972 		return (rc);
973 	}
974 
975 	for (i = 0; i < sc->intr_count; i++)
976 		t4_free_irq(sc, &sc->irq[i]);
977 
978 	for (i = 0; i < MAX_NPORTS; i++) {
979 		pi = sc->port[i];
980 		if (pi) {
981 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->viid);
982 			if (pi->dev)
983 				device_delete_child(dev, pi->dev);
984 
985 			mtx_destroy(&pi->pi_lock);
986 			free(pi, M_CXGBE);
987 		}
988 	}
989 
990 	if (sc->flags & FULL_INIT_DONE)
991 		adapter_full_uninit(sc);
992 
993 	if (sc->flags & FW_OK)
994 		t4_fw_bye(sc, sc->mbox);
995 
996 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
997 		pci_release_msi(dev);
998 
999 	if (sc->regs_res)
1000 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1001 		    sc->regs_res);
1002 
1003 	if (sc->udbs_res)
1004 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1005 		    sc->udbs_res);
1006 
1007 	if (sc->msix_res)
1008 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1009 		    sc->msix_res);
1010 
1011 	if (sc->l2t)
1012 		t4_free_l2t(sc->l2t);
1013 
1014 #ifdef TCP_OFFLOAD
1015 	free(sc->sge.ofld_rxq, M_CXGBE);
1016 	free(sc->sge.ofld_txq, M_CXGBE);
1017 #endif
1018 #ifdef DEV_NETMAP
1019 	free(sc->sge.nm_rxq, M_CXGBE);
1020 	free(sc->sge.nm_txq, M_CXGBE);
1021 #endif
1022 	free(sc->irq, M_CXGBE);
1023 	free(sc->sge.rxq, M_CXGBE);
1024 	free(sc->sge.txq, M_CXGBE);
1025 	free(sc->sge.ctrlq, M_CXGBE);
1026 	free(sc->sge.iqmap, M_CXGBE);
1027 	free(sc->sge.eqmap, M_CXGBE);
1028 	free(sc->tids.ftid_tab, M_CXGBE);
1029 	t4_destroy_dma_tag(sc);
1030 	if (mtx_initialized(&sc->sc_lock)) {
1031 		sx_xlock(&t4_list_lock);
1032 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1033 		sx_xunlock(&t4_list_lock);
1034 		mtx_destroy(&sc->sc_lock);
1035 	}
1036 
1037 	if (mtx_initialized(&sc->tids.ftid_lock))
1038 		mtx_destroy(&sc->tids.ftid_lock);
1039 	if (mtx_initialized(&sc->sfl_lock))
1040 		mtx_destroy(&sc->sfl_lock);
1041 	if (mtx_initialized(&sc->ifp_lock))
1042 		mtx_destroy(&sc->ifp_lock);
1043 	if (mtx_initialized(&sc->regwin_lock))
1044 		mtx_destroy(&sc->regwin_lock);
1045 
1046 	bzero(sc, sizeof(*sc));
1047 
1048 	return (0);
1049 }
1050 
1051 static int
1052 cxgbe_probe(device_t dev)
1053 {
1054 	char buf[128];
1055 	struct port_info *pi = device_get_softc(dev);
1056 
1057 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1058 	device_set_desc_copy(dev, buf);
1059 
1060 	return (BUS_PROBE_DEFAULT);
1061 }
1062 
1063 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1064     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1065     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1066 #define T4_CAP_ENABLE (T4_CAP)
1067 
1068 static int
1069 cxgbe_attach(device_t dev)
1070 {
1071 	struct port_info *pi = device_get_softc(dev);
1072 	struct ifnet *ifp;
1073 	char *s;
1074 	int n, o;
1075 
1076 	/* Allocate an ifnet and set it up */
1077 	ifp = if_alloc(IFT_ETHER);
1078 	if (ifp == NULL) {
1079 		device_printf(dev, "Cannot allocate ifnet\n");
1080 		return (ENOMEM);
1081 	}
1082 	pi->ifp = ifp;
1083 	ifp->if_softc = pi;
1084 
1085 	callout_init(&pi->tick, CALLOUT_MPSAFE);
1086 
1087 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1088 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1089 
1090 	ifp->if_init = cxgbe_init;
1091 	ifp->if_ioctl = cxgbe_ioctl;
1092 	ifp->if_transmit = cxgbe_transmit;
1093 	ifp->if_qflush = cxgbe_qflush;
1094 	ifp->if_get_counter = cxgbe_get_counter;
1095 
1096 	ifp->if_capabilities = T4_CAP;
1097 #ifdef TCP_OFFLOAD
1098 	if (is_offload(pi->adapter))
1099 		ifp->if_capabilities |= IFCAP_TOE;
1100 #endif
1101 	ifp->if_capenable = T4_CAP_ENABLE;
1102 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1103 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1104 
1105 	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1106 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1107 	ifp->if_hw_tsomaxsegsize = 65536;
1108 
1109 	/* Initialize ifmedia for this port */
1110 	ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1111 	    cxgbe_media_status);
1112 	build_medialist(pi, &pi->media);
1113 
1114 	pi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1115 	    EVENTHANDLER_PRI_ANY);
1116 
1117 	ether_ifattach(ifp, pi->hw_addr);
1118 
1119 	n = 128;
1120 	s = malloc(n, M_CXGBE, M_WAITOK);
1121 	o = snprintf(s, n, "%d txq, %d rxq (NIC)", pi->ntxq, pi->nrxq);
1122 	MPASS(n > o);
1123 #ifdef TCP_OFFLOAD
1124 	if (is_offload(pi->adapter)) {
1125 		o += snprintf(s + o, n - o, "; %d txq, %d rxq (TOE)",
1126 		    pi->nofldtxq, pi->nofldrxq);
1127 		MPASS(n > o);
1128 	}
1129 #endif
1130 #ifdef DEV_NETMAP
1131 	o += snprintf(s + o, n - o, "; %d txq, %d rxq (netmap)", pi->nnmtxq,
1132 	    pi->nnmrxq);
1133 	MPASS(n > o);
1134 #endif
1135 	device_printf(dev, "%s\n", s);
1136 	free(s, M_CXGBE);
1137 
1138 #ifdef DEV_NETMAP
1139 	/* nm_media handled here to keep implementation private to this file */
1140 	ifmedia_init(&pi->nm_media, IFM_IMASK, cxgbe_media_change,
1141 	    cxgbe_media_status);
1142 	build_medialist(pi, &pi->nm_media);
1143 	create_netmap_ifnet(pi);	/* logs errors it something fails */
1144 #endif
1145 	cxgbe_sysctls(pi);
1146 
1147 	return (0);
1148 }
1149 
1150 static int
1151 cxgbe_detach(device_t dev)
1152 {
1153 	struct port_info *pi = device_get_softc(dev);
1154 	struct adapter *sc = pi->adapter;
1155 	struct ifnet *ifp = pi->ifp;
1156 
1157 	/* Tell if_ioctl and if_init that the port is going away */
1158 	ADAPTER_LOCK(sc);
1159 	SET_DOOMED(pi);
1160 	wakeup(&sc->flags);
1161 	while (IS_BUSY(sc))
1162 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
1163 	SET_BUSY(sc);
1164 #ifdef INVARIANTS
1165 	sc->last_op = "t4detach";
1166 	sc->last_op_thr = curthread;
1167 #endif
1168 	ADAPTER_UNLOCK(sc);
1169 
1170 	if (pi->flags & HAS_TRACEQ) {
1171 		sc->traceq = -1;	/* cloner should not create ifnet */
1172 		t4_tracer_port_detach(sc);
1173 	}
1174 
1175 	if (pi->vlan_c)
1176 		EVENTHANDLER_DEREGISTER(vlan_config, pi->vlan_c);
1177 
1178 	PORT_LOCK(pi);
1179 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1180 	callout_stop(&pi->tick);
1181 	PORT_UNLOCK(pi);
1182 	callout_drain(&pi->tick);
1183 
1184 	/* Let detach proceed even if these fail. */
1185 	cxgbe_uninit_synchronized(pi);
1186 	port_full_uninit(pi);
1187 
1188 	ifmedia_removeall(&pi->media);
1189 	ether_ifdetach(pi->ifp);
1190 	if_free(pi->ifp);
1191 
1192 #ifdef DEV_NETMAP
1193 	/* XXXNM: equivalent of cxgbe_uninit_synchronized to ifdown nm_ifp */
1194 	destroy_netmap_ifnet(pi);
1195 #endif
1196 
1197 	ADAPTER_LOCK(sc);
1198 	CLR_BUSY(sc);
1199 	wakeup(&sc->flags);
1200 	ADAPTER_UNLOCK(sc);
1201 
1202 	return (0);
1203 }
1204 
1205 static void
1206 cxgbe_init(void *arg)
1207 {
1208 	struct port_info *pi = arg;
1209 	struct adapter *sc = pi->adapter;
1210 
1211 	if (begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4init") != 0)
1212 		return;
1213 	cxgbe_init_synchronized(pi);
1214 	end_synchronized_op(sc, 0);
1215 }
1216 
1217 static int
1218 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1219 {
1220 	int rc = 0, mtu, flags, can_sleep;
1221 	struct port_info *pi = ifp->if_softc;
1222 	struct adapter *sc = pi->adapter;
1223 	struct ifreq *ifr = (struct ifreq *)data;
1224 	uint32_t mask;
1225 
1226 	switch (cmd) {
1227 	case SIOCSIFMTU:
1228 		mtu = ifr->ifr_mtu;
1229 		if ((mtu < ETHERMIN) || (mtu > ETHERMTU_JUMBO))
1230 			return (EINVAL);
1231 
1232 		rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4mtu");
1233 		if (rc)
1234 			return (rc);
1235 		ifp->if_mtu = mtu;
1236 		if (pi->flags & PORT_INIT_DONE) {
1237 			t4_update_fl_bufsize(ifp);
1238 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1239 				rc = update_mac_settings(ifp, XGMAC_MTU);
1240 		}
1241 		end_synchronized_op(sc, 0);
1242 		break;
1243 
1244 	case SIOCSIFFLAGS:
1245 		can_sleep = 0;
1246 redo_sifflags:
1247 		rc = begin_synchronized_op(sc, pi,
1248 		    can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1249 		if (rc)
1250 			return (rc);
1251 
1252 		if (ifp->if_flags & IFF_UP) {
1253 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1254 				flags = pi->if_flags;
1255 				if ((ifp->if_flags ^ flags) &
1256 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1257 					if (can_sleep == 1) {
1258 						end_synchronized_op(sc, 0);
1259 						can_sleep = 0;
1260 						goto redo_sifflags;
1261 					}
1262 					rc = update_mac_settings(ifp,
1263 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1264 				}
1265 			} else {
1266 				if (can_sleep == 0) {
1267 					end_synchronized_op(sc, LOCK_HELD);
1268 					can_sleep = 1;
1269 					goto redo_sifflags;
1270 				}
1271 				rc = cxgbe_init_synchronized(pi);
1272 			}
1273 			pi->if_flags = ifp->if_flags;
1274 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1275 			if (can_sleep == 0) {
1276 				end_synchronized_op(sc, LOCK_HELD);
1277 				can_sleep = 1;
1278 				goto redo_sifflags;
1279 			}
1280 			rc = cxgbe_uninit_synchronized(pi);
1281 		}
1282 		end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1283 		break;
1284 
1285 	case SIOCADDMULTI:
1286 	case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1287 		rc = begin_synchronized_op(sc, pi, HOLD_LOCK, "t4multi");
1288 		if (rc)
1289 			return (rc);
1290 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1291 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1292 		end_synchronized_op(sc, LOCK_HELD);
1293 		break;
1294 
1295 	case SIOCSIFCAP:
1296 		rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4cap");
1297 		if (rc)
1298 			return (rc);
1299 
1300 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1301 		if (mask & IFCAP_TXCSUM) {
1302 			ifp->if_capenable ^= IFCAP_TXCSUM;
1303 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1304 
1305 			if (IFCAP_TSO4 & ifp->if_capenable &&
1306 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1307 				ifp->if_capenable &= ~IFCAP_TSO4;
1308 				if_printf(ifp,
1309 				    "tso4 disabled due to -txcsum.\n");
1310 			}
1311 		}
1312 		if (mask & IFCAP_TXCSUM_IPV6) {
1313 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1314 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1315 
1316 			if (IFCAP_TSO6 & ifp->if_capenable &&
1317 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1318 				ifp->if_capenable &= ~IFCAP_TSO6;
1319 				if_printf(ifp,
1320 				    "tso6 disabled due to -txcsum6.\n");
1321 			}
1322 		}
1323 		if (mask & IFCAP_RXCSUM)
1324 			ifp->if_capenable ^= IFCAP_RXCSUM;
1325 		if (mask & IFCAP_RXCSUM_IPV6)
1326 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1327 
1328 		/*
1329 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1330 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1331 		 * sending a TSO request our way, so it's sufficient to toggle
1332 		 * IFCAP_TSOx only.
1333 		 */
1334 		if (mask & IFCAP_TSO4) {
1335 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1336 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1337 				if_printf(ifp, "enable txcsum first.\n");
1338 				rc = EAGAIN;
1339 				goto fail;
1340 			}
1341 			ifp->if_capenable ^= IFCAP_TSO4;
1342 		}
1343 		if (mask & IFCAP_TSO6) {
1344 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1345 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1346 				if_printf(ifp, "enable txcsum6 first.\n");
1347 				rc = EAGAIN;
1348 				goto fail;
1349 			}
1350 			ifp->if_capenable ^= IFCAP_TSO6;
1351 		}
1352 		if (mask & IFCAP_LRO) {
1353 #if defined(INET) || defined(INET6)
1354 			int i;
1355 			struct sge_rxq *rxq;
1356 
1357 			ifp->if_capenable ^= IFCAP_LRO;
1358 			for_each_rxq(pi, i, rxq) {
1359 				if (ifp->if_capenable & IFCAP_LRO)
1360 					rxq->iq.flags |= IQ_LRO_ENABLED;
1361 				else
1362 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1363 			}
1364 #endif
1365 		}
1366 #ifdef TCP_OFFLOAD
1367 		if (mask & IFCAP_TOE) {
1368 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1369 
1370 			rc = toe_capability(pi, enable);
1371 			if (rc != 0)
1372 				goto fail;
1373 
1374 			ifp->if_capenable ^= mask;
1375 		}
1376 #endif
1377 		if (mask & IFCAP_VLAN_HWTAGGING) {
1378 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1379 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1380 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1381 		}
1382 		if (mask & IFCAP_VLAN_MTU) {
1383 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1384 
1385 			/* Need to find out how to disable auto-mtu-inflation */
1386 		}
1387 		if (mask & IFCAP_VLAN_HWTSO)
1388 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1389 		if (mask & IFCAP_VLAN_HWCSUM)
1390 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1391 
1392 #ifdef VLAN_CAPABILITIES
1393 		VLAN_CAPABILITIES(ifp);
1394 #endif
1395 fail:
1396 		end_synchronized_op(sc, 0);
1397 		break;
1398 
1399 	case SIOCSIFMEDIA:
1400 	case SIOCGIFMEDIA:
1401 		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1402 		break;
1403 
1404 	case SIOCGI2C: {
1405 		struct ifi2creq i2c;
1406 
1407 		rc = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
1408 		if (rc != 0)
1409 			break;
1410 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1411 			rc = EPERM;
1412 			break;
1413 		}
1414 		if (i2c.len > sizeof(i2c.data)) {
1415 			rc = EINVAL;
1416 			break;
1417 		}
1418 		rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4i2c");
1419 		if (rc)
1420 			return (rc);
1421 		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
1422 		    i2c.offset, i2c.len, &i2c.data[0]);
1423 		end_synchronized_op(sc, 0);
1424 		if (rc == 0)
1425 			rc = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
1426 		break;
1427 	}
1428 
1429 	default:
1430 		rc = ether_ioctl(ifp, cmd, data);
1431 	}
1432 
1433 	return (rc);
1434 }
1435 
1436 static int
1437 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1438 {
1439 	struct port_info *pi = ifp->if_softc;
1440 	struct adapter *sc = pi->adapter;
1441 	struct sge_txq *txq;
1442 	void *items[1];
1443 	int rc;
1444 
1445 	M_ASSERTPKTHDR(m);
1446 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1447 
1448 	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1449 		m_freem(m);
1450 		return (ENETDOWN);
1451 	}
1452 
1453 	rc = parse_pkt(&m);
1454 	if (__predict_false(rc != 0)) {
1455 		MPASS(m == NULL);			/* was freed already */
1456 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1457 		return (rc);
1458 	}
1459 
1460 	/* Select a txq. */
1461 	txq = &sc->sge.txq[pi->first_txq];
1462 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1463 		txq += ((m->m_pkthdr.flowid % (pi->ntxq - pi->rsrv_noflowq)) +
1464 		    pi->rsrv_noflowq);
1465 
1466 	items[0] = m;
1467 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1468 	if (__predict_false(rc != 0))
1469 		m_freem(m);
1470 
1471 	return (rc);
1472 }
1473 
1474 static void
1475 cxgbe_qflush(struct ifnet *ifp)
1476 {
1477 	struct port_info *pi = ifp->if_softc;
1478 	struct sge_txq *txq;
1479 	int i;
1480 
1481 	/* queues do not exist if !PORT_INIT_DONE. */
1482 	if (pi->flags & PORT_INIT_DONE) {
1483 		for_each_txq(pi, i, txq) {
1484 			TXQ_LOCK(txq);
1485 			txq->eq.flags &= ~EQ_ENABLED;
1486 			TXQ_UNLOCK(txq);
1487 			while (!mp_ring_is_idle(txq->r)) {
1488 				mp_ring_check_drainage(txq->r, 0);
1489 				pause("qflush", 1);
1490 			}
1491 		}
1492 	}
1493 	if_qflush(ifp);
1494 }
1495 
1496 static uint64_t
1497 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
1498 {
1499 	struct port_info *pi = ifp->if_softc;
1500 	struct adapter *sc = pi->adapter;
1501 	struct port_stats *s = &pi->stats;
1502 
1503 	cxgbe_refresh_stats(sc, pi);
1504 
1505 	switch (c) {
1506 	case IFCOUNTER_IPACKETS:
1507 		return (s->rx_frames - s->rx_pause);
1508 
1509 	case IFCOUNTER_IERRORS:
1510 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
1511 		    s->rx_fcs_err + s->rx_len_err);
1512 
1513 	case IFCOUNTER_OPACKETS:
1514 		return (s->tx_frames - s->tx_pause);
1515 
1516 	case IFCOUNTER_OERRORS:
1517 		return (s->tx_error_frames);
1518 
1519 	case IFCOUNTER_IBYTES:
1520 		return (s->rx_octets - s->rx_pause * 64);
1521 
1522 	case IFCOUNTER_OBYTES:
1523 		return (s->tx_octets - s->tx_pause * 64);
1524 
1525 	case IFCOUNTER_IMCASTS:
1526 		return (s->rx_mcast_frames - s->rx_pause);
1527 
1528 	case IFCOUNTER_OMCASTS:
1529 		return (s->tx_mcast_frames - s->tx_pause);
1530 
1531 	case IFCOUNTER_IQDROPS:
1532 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
1533 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
1534 		    s->rx_trunc3 + pi->tnl_cong_drops);
1535 
1536 	case IFCOUNTER_OQDROPS: {
1537 		uint64_t drops;
1538 
1539 		drops = s->tx_drop;
1540 		if (pi->flags & PORT_INIT_DONE) {
1541 			int i;
1542 			struct sge_txq *txq;
1543 
1544 			for_each_txq(pi, i, txq)
1545 				drops += counter_u64_fetch(txq->r->drops);
1546 		}
1547 
1548 		return (drops);
1549 
1550 	}
1551 
1552 	default:
1553 		return (if_get_counter_default(ifp, c));
1554 	}
1555 }
1556 
1557 static int
1558 cxgbe_media_change(struct ifnet *ifp)
1559 {
1560 	struct port_info *pi = ifp->if_softc;
1561 
1562 	device_printf(pi->dev, "%s unimplemented.\n", __func__);
1563 
1564 	return (EOPNOTSUPP);
1565 }
1566 
1567 static void
1568 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1569 {
1570 	struct port_info *pi = ifp->if_softc;
1571 	struct ifmedia *media = NULL;
1572 	struct ifmedia_entry *cur;
1573 	int speed = pi->link_cfg.speed;
1574 #ifdef INVARIANTS
1575 	int data = (pi->port_type << 8) | pi->mod_type;
1576 #endif
1577 
1578 	if (ifp == pi->ifp)
1579 		media = &pi->media;
1580 #ifdef DEV_NETMAP
1581 	else if (ifp == pi->nm_ifp)
1582 		media = &pi->nm_media;
1583 #endif
1584 	MPASS(media != NULL);
1585 
1586 	cur = media->ifm_cur;
1587 	MPASS(cur->ifm_data == data);
1588 
1589 	ifmr->ifm_status = IFM_AVALID;
1590 	if (!pi->link_cfg.link_ok)
1591 		return;
1592 
1593 	ifmr->ifm_status |= IFM_ACTIVE;
1594 
1595 	/* active and current will differ iff current media is autoselect. */
1596 	if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
1597 		return;
1598 
1599 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
1600 	if (speed == SPEED_10000)
1601 		ifmr->ifm_active |= IFM_10G_T;
1602 	else if (speed == SPEED_1000)
1603 		ifmr->ifm_active |= IFM_1000_T;
1604 	else if (speed == SPEED_100)
1605 		ifmr->ifm_active |= IFM_100_TX;
1606 	else if (speed == SPEED_10)
1607 		ifmr->ifm_active |= IFM_10_T;
1608 	else
1609 		KASSERT(0, ("%s: link up but speed unknown (%u)", __func__,
1610 			    speed));
1611 }
1612 
1613 void
1614 t4_fatal_err(struct adapter *sc)
1615 {
1616 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
1617 	t4_intr_disable(sc);
1618 	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
1619 	    device_get_nameunit(sc->dev));
1620 }
1621 
1622 static int
1623 map_bars_0_and_4(struct adapter *sc)
1624 {
1625 	sc->regs_rid = PCIR_BAR(0);
1626 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1627 	    &sc->regs_rid, RF_ACTIVE);
1628 	if (sc->regs_res == NULL) {
1629 		device_printf(sc->dev, "cannot map registers.\n");
1630 		return (ENXIO);
1631 	}
1632 	sc->bt = rman_get_bustag(sc->regs_res);
1633 	sc->bh = rman_get_bushandle(sc->regs_res);
1634 	sc->mmio_len = rman_get_size(sc->regs_res);
1635 	setbit(&sc->doorbells, DOORBELL_KDB);
1636 
1637 	sc->msix_rid = PCIR_BAR(4);
1638 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1639 	    &sc->msix_rid, RF_ACTIVE);
1640 	if (sc->msix_res == NULL) {
1641 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
1642 		return (ENXIO);
1643 	}
1644 
1645 	return (0);
1646 }
1647 
1648 static int
1649 map_bar_2(struct adapter *sc)
1650 {
1651 
1652 	/*
1653 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
1654 	 * to map it if RDMA is disabled.
1655 	 */
1656 	if (is_t4(sc) && sc->rdmacaps == 0)
1657 		return (0);
1658 
1659 	sc->udbs_rid = PCIR_BAR(2);
1660 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1661 	    &sc->udbs_rid, RF_ACTIVE);
1662 	if (sc->udbs_res == NULL) {
1663 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
1664 		return (ENXIO);
1665 	}
1666 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
1667 
1668 	if (is_t5(sc)) {
1669 		setbit(&sc->doorbells, DOORBELL_UDB);
1670 #if defined(__i386__) || defined(__amd64__)
1671 		if (t5_write_combine) {
1672 			int rc;
1673 
1674 			/*
1675 			 * Enable write combining on BAR2.  This is the
1676 			 * userspace doorbell BAR and is split into 128B
1677 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
1678 			 * with an egress queue.  The first 64B has the doorbell
1679 			 * and the second 64B can be used to submit a tx work
1680 			 * request with an implicit doorbell.
1681 			 */
1682 
1683 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
1684 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
1685 			if (rc == 0) {
1686 				clrbit(&sc->doorbells, DOORBELL_UDB);
1687 				setbit(&sc->doorbells, DOORBELL_WCWR);
1688 				setbit(&sc->doorbells, DOORBELL_UDBWC);
1689 			} else {
1690 				device_printf(sc->dev,
1691 				    "couldn't enable write combining: %d\n",
1692 				    rc);
1693 			}
1694 
1695 			t4_write_reg(sc, A_SGE_STAT_CFG,
1696 			    V_STATSOURCE_T5(7) | V_STATMODE(0));
1697 		}
1698 #endif
1699 	}
1700 
1701 	return (0);
1702 }
1703 
1704 static const struct memwin t4_memwin[] = {
1705 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1706 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1707 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
1708 };
1709 
1710 static const struct memwin t5_memwin[] = {
1711 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1712 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1713 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
1714 };
1715 
1716 static void
1717 setup_memwin(struct adapter *sc)
1718 {
1719 	const struct memwin *mw;
1720 	int i, n;
1721 	uint32_t bar0;
1722 
1723 	if (is_t4(sc)) {
1724 		/*
1725 		 * Read low 32b of bar0 indirectly via the hardware backdoor
1726 		 * mechanism.  Works from within PCI passthrough environments
1727 		 * too, where rman_get_start() can return a different value.  We
1728 		 * need to program the T4 memory window decoders with the actual
1729 		 * addresses that will be coming across the PCIe link.
1730 		 */
1731 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
1732 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
1733 
1734 		mw = &t4_memwin[0];
1735 		n = nitems(t4_memwin);
1736 	} else {
1737 		/* T5 uses the relative offset inside the PCIe BAR */
1738 		bar0 = 0;
1739 
1740 		mw = &t5_memwin[0];
1741 		n = nitems(t5_memwin);
1742 	}
1743 
1744 	for (i = 0; i < n; i++, mw++) {
1745 		t4_write_reg(sc,
1746 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
1747 		    (mw->base + bar0) | V_BIR(0) |
1748 		    V_WINDOW(ilog2(mw->aperture) - 10));
1749 	}
1750 
1751 	/* flush */
1752 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
1753 }
1754 
1755 /*
1756  * Verify that the memory range specified by the addr/len pair is valid and lies
1757  * entirely within a single region (EDCx or MCx).
1758  */
1759 static int
1760 validate_mem_range(struct adapter *sc, uint32_t addr, int len)
1761 {
1762 	uint32_t em, addr_len, maddr, mlen;
1763 
1764 	/* Memory can only be accessed in naturally aligned 4 byte units */
1765 	if (addr & 3 || len & 3 || len == 0)
1766 		return (EINVAL);
1767 
1768 	/* Enabled memories */
1769 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
1770 	if (em & F_EDRAM0_ENABLE) {
1771 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
1772 		maddr = G_EDRAM0_BASE(addr_len) << 20;
1773 		mlen = G_EDRAM0_SIZE(addr_len) << 20;
1774 		if (mlen > 0 && addr >= maddr && addr < maddr + mlen &&
1775 		    addr + len <= maddr + mlen)
1776 			return (0);
1777 	}
1778 	if (em & F_EDRAM1_ENABLE) {
1779 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
1780 		maddr = G_EDRAM1_BASE(addr_len) << 20;
1781 		mlen = G_EDRAM1_SIZE(addr_len) << 20;
1782 		if (mlen > 0 && addr >= maddr && addr < maddr + mlen &&
1783 		    addr + len <= maddr + mlen)
1784 			return (0);
1785 	}
1786 	if (em & F_EXT_MEM_ENABLE) {
1787 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
1788 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
1789 		mlen = G_EXT_MEM_SIZE(addr_len) << 20;
1790 		if (mlen > 0 && addr >= maddr && addr < maddr + mlen &&
1791 		    addr + len <= maddr + mlen)
1792 			return (0);
1793 	}
1794 	if (!is_t4(sc) && em & F_EXT_MEM1_ENABLE) {
1795 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
1796 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
1797 		mlen = G_EXT_MEM1_SIZE(addr_len) << 20;
1798 		if (mlen > 0 && addr >= maddr && addr < maddr + mlen &&
1799 		    addr + len <= maddr + mlen)
1800 			return (0);
1801 	}
1802 
1803 	return (EFAULT);
1804 }
1805 
1806 static int
1807 fwmtype_to_hwmtype(int mtype)
1808 {
1809 
1810 	switch (mtype) {
1811 	case FW_MEMTYPE_EDC0:
1812 		return (MEM_EDC0);
1813 	case FW_MEMTYPE_EDC1:
1814 		return (MEM_EDC1);
1815 	case FW_MEMTYPE_EXTMEM:
1816 		return (MEM_MC0);
1817 	case FW_MEMTYPE_EXTMEM1:
1818 		return (MEM_MC1);
1819 	default:
1820 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
1821 	}
1822 }
1823 
1824 /*
1825  * Verify that the memory range specified by the memtype/offset/len pair is
1826  * valid and lies entirely within the memtype specified.  The global address of
1827  * the start of the range is returned in addr.
1828  */
1829 static int
1830 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
1831     uint32_t *addr)
1832 {
1833 	uint32_t em, addr_len, maddr, mlen;
1834 
1835 	/* Memory can only be accessed in naturally aligned 4 byte units */
1836 	if (off & 3 || len & 3 || len == 0)
1837 		return (EINVAL);
1838 
1839 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
1840 	switch (fwmtype_to_hwmtype(mtype)) {
1841 	case MEM_EDC0:
1842 		if (!(em & F_EDRAM0_ENABLE))
1843 			return (EINVAL);
1844 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
1845 		maddr = G_EDRAM0_BASE(addr_len) << 20;
1846 		mlen = G_EDRAM0_SIZE(addr_len) << 20;
1847 		break;
1848 	case MEM_EDC1:
1849 		if (!(em & F_EDRAM1_ENABLE))
1850 			return (EINVAL);
1851 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
1852 		maddr = G_EDRAM1_BASE(addr_len) << 20;
1853 		mlen = G_EDRAM1_SIZE(addr_len) << 20;
1854 		break;
1855 	case MEM_MC:
1856 		if (!(em & F_EXT_MEM_ENABLE))
1857 			return (EINVAL);
1858 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
1859 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
1860 		mlen = G_EXT_MEM_SIZE(addr_len) << 20;
1861 		break;
1862 	case MEM_MC1:
1863 		if (is_t4(sc) || !(em & F_EXT_MEM1_ENABLE))
1864 			return (EINVAL);
1865 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
1866 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
1867 		mlen = G_EXT_MEM1_SIZE(addr_len) << 20;
1868 		break;
1869 	default:
1870 		return (EINVAL);
1871 	}
1872 
1873 	if (mlen > 0 && off < mlen && off + len <= mlen) {
1874 		*addr = maddr + off;	/* global address */
1875 		return (0);
1876 	}
1877 
1878 	return (EFAULT);
1879 }
1880 
1881 static void
1882 memwin_info(struct adapter *sc, int win, uint32_t *base, uint32_t *aperture)
1883 {
1884 	const struct memwin *mw;
1885 
1886 	if (is_t4(sc)) {
1887 		KASSERT(win >= 0 && win < nitems(t4_memwin),
1888 		    ("%s: incorrect memwin# (%d)", __func__, win));
1889 		mw = &t4_memwin[win];
1890 	} else {
1891 		KASSERT(win >= 0 && win < nitems(t5_memwin),
1892 		    ("%s: incorrect memwin# (%d)", __func__, win));
1893 		mw = &t5_memwin[win];
1894 	}
1895 
1896 	if (base != NULL)
1897 		*base = mw->base;
1898 	if (aperture != NULL)
1899 		*aperture = mw->aperture;
1900 }
1901 
1902 /*
1903  * Positions the memory window such that it can be used to access the specified
1904  * address in the chip's address space.  The return value is the offset of addr
1905  * from the start of the window.
1906  */
1907 static uint32_t
1908 position_memwin(struct adapter *sc, int n, uint32_t addr)
1909 {
1910 	uint32_t start, pf;
1911 	uint32_t reg;
1912 
1913 	KASSERT(n >= 0 && n <= 3,
1914 	    ("%s: invalid window %d.", __func__, n));
1915 	KASSERT((addr & 3) == 0,
1916 	    ("%s: addr (0x%x) is not at a 4B boundary.", __func__, addr));
1917 
1918 	if (is_t4(sc)) {
1919 		pf = 0;
1920 		start = addr & ~0xf;	/* start must be 16B aligned */
1921 	} else {
1922 		pf = V_PFNUM(sc->pf);
1923 		start = addr & ~0x7f;	/* start must be 128B aligned */
1924 	}
1925 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, n);
1926 
1927 	t4_write_reg(sc, reg, start | pf);
1928 	t4_read_reg(sc, reg);
1929 
1930 	return (addr - start);
1931 }
1932 
1933 static int
1934 cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g,
1935     struct intrs_and_queues *iaq)
1936 {
1937 	int rc, itype, navail, nrxq10g, nrxq1g, n;
1938 	int nofldrxq10g = 0, nofldrxq1g = 0;
1939 	int nnmrxq10g = 0, nnmrxq1g = 0;
1940 
1941 	bzero(iaq, sizeof(*iaq));
1942 
1943 	iaq->ntxq10g = t4_ntxq10g;
1944 	iaq->ntxq1g = t4_ntxq1g;
1945 	iaq->nrxq10g = nrxq10g = t4_nrxq10g;
1946 	iaq->nrxq1g = nrxq1g = t4_nrxq1g;
1947 	iaq->rsrv_noflowq = t4_rsrv_noflowq;
1948 #ifdef TCP_OFFLOAD
1949 	if (is_offload(sc)) {
1950 		iaq->nofldtxq10g = t4_nofldtxq10g;
1951 		iaq->nofldtxq1g = t4_nofldtxq1g;
1952 		iaq->nofldrxq10g = nofldrxq10g = t4_nofldrxq10g;
1953 		iaq->nofldrxq1g = nofldrxq1g = t4_nofldrxq1g;
1954 	}
1955 #endif
1956 #ifdef DEV_NETMAP
1957 	iaq->nnmtxq10g = t4_nnmtxq10g;
1958 	iaq->nnmtxq1g = t4_nnmtxq1g;
1959 	iaq->nnmrxq10g = nnmrxq10g = t4_nnmrxq10g;
1960 	iaq->nnmrxq1g = nnmrxq1g = t4_nnmrxq1g;
1961 #endif
1962 
1963 	for (itype = INTR_MSIX; itype; itype >>= 1) {
1964 
1965 		if ((itype & t4_intr_types) == 0)
1966 			continue;	/* not allowed */
1967 
1968 		if (itype == INTR_MSIX)
1969 			navail = pci_msix_count(sc->dev);
1970 		else if (itype == INTR_MSI)
1971 			navail = pci_msi_count(sc->dev);
1972 		else
1973 			navail = 1;
1974 restart:
1975 		if (navail == 0)
1976 			continue;
1977 
1978 		iaq->intr_type = itype;
1979 		iaq->intr_flags_10g = 0;
1980 		iaq->intr_flags_1g = 0;
1981 
1982 		/*
1983 		 * Best option: an interrupt vector for errors, one for the
1984 		 * firmware event queue, and one for every rxq (NIC, TOE, and
1985 		 * netmap).
1986 		 */
1987 		iaq->nirq = T4_EXTRA_INTR;
1988 		iaq->nirq += n10g * (nrxq10g + nofldrxq10g + nnmrxq10g);
1989 		iaq->nirq += n1g * (nrxq1g + nofldrxq1g + nnmrxq1g);
1990 		if (iaq->nirq <= navail &&
1991 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
1992 			iaq->intr_flags_10g = INTR_ALL;
1993 			iaq->intr_flags_1g = INTR_ALL;
1994 			goto allocate;
1995 		}
1996 
1997 		/*
1998 		 * Second best option: a vector for errors, one for the firmware
1999 		 * event queue, and vectors for either all the NIC rx queues or
2000 		 * all the TOE rx queues.  The queues that don't get vectors
2001 		 * will forward their interrupts to those that do.
2002 		 *
2003 		 * Note: netmap rx queues cannot be created early and so they
2004 		 * can't be setup to receive forwarded interrupts for others.
2005 		 */
2006 		iaq->nirq = T4_EXTRA_INTR;
2007 		if (nrxq10g >= nofldrxq10g) {
2008 			iaq->intr_flags_10g = INTR_RXQ;
2009 			iaq->nirq += n10g * nrxq10g;
2010 #ifdef DEV_NETMAP
2011 			iaq->nnmrxq10g = min(nnmrxq10g, nrxq10g);
2012 #endif
2013 		} else {
2014 			iaq->intr_flags_10g = INTR_OFLD_RXQ;
2015 			iaq->nirq += n10g * nofldrxq10g;
2016 #ifdef DEV_NETMAP
2017 			iaq->nnmrxq10g = min(nnmrxq10g, nofldrxq10g);
2018 #endif
2019 		}
2020 		if (nrxq1g >= nofldrxq1g) {
2021 			iaq->intr_flags_1g = INTR_RXQ;
2022 			iaq->nirq += n1g * nrxq1g;
2023 #ifdef DEV_NETMAP
2024 			iaq->nnmrxq1g = min(nnmrxq1g, nrxq1g);
2025 #endif
2026 		} else {
2027 			iaq->intr_flags_1g = INTR_OFLD_RXQ;
2028 			iaq->nirq += n1g * nofldrxq1g;
2029 #ifdef DEV_NETMAP
2030 			iaq->nnmrxq1g = min(nnmrxq1g, nofldrxq1g);
2031 #endif
2032 		}
2033 		if (iaq->nirq <= navail &&
2034 		    (itype != INTR_MSI || powerof2(iaq->nirq)))
2035 			goto allocate;
2036 
2037 		/*
2038 		 * Next best option: an interrupt vector for errors, one for the
2039 		 * firmware event queue, and at least one per port.  At this
2040 		 * point we know we'll have to downsize nrxq and/or nofldrxq
2041 		 * and/or nnmrxq to fit what's available to us.
2042 		 */
2043 		iaq->nirq = T4_EXTRA_INTR;
2044 		iaq->nirq += n10g + n1g;
2045 		if (iaq->nirq <= navail) {
2046 			int leftover = navail - iaq->nirq;
2047 
2048 			if (n10g > 0) {
2049 				int target = max(nrxq10g, nofldrxq10g);
2050 
2051 				iaq->intr_flags_10g = nrxq10g >= nofldrxq10g ?
2052 				    INTR_RXQ : INTR_OFLD_RXQ;
2053 
2054 				n = 1;
2055 				while (n < target && leftover >= n10g) {
2056 					leftover -= n10g;
2057 					iaq->nirq += n10g;
2058 					n++;
2059 				}
2060 				iaq->nrxq10g = min(n, nrxq10g);
2061 #ifdef TCP_OFFLOAD
2062 				iaq->nofldrxq10g = min(n, nofldrxq10g);
2063 #endif
2064 #ifdef DEV_NETMAP
2065 				iaq->nnmrxq10g = min(n, nnmrxq10g);
2066 #endif
2067 			}
2068 
2069 			if (n1g > 0) {
2070 				int target = max(nrxq1g, nofldrxq1g);
2071 
2072 				iaq->intr_flags_1g = nrxq1g >= nofldrxq1g ?
2073 				    INTR_RXQ : INTR_OFLD_RXQ;
2074 
2075 				n = 1;
2076 				while (n < target && leftover >= n1g) {
2077 					leftover -= n1g;
2078 					iaq->nirq += n1g;
2079 					n++;
2080 				}
2081 				iaq->nrxq1g = min(n, nrxq1g);
2082 #ifdef TCP_OFFLOAD
2083 				iaq->nofldrxq1g = min(n, nofldrxq1g);
2084 #endif
2085 #ifdef DEV_NETMAP
2086 				iaq->nnmrxq1g = min(n, nnmrxq1g);
2087 #endif
2088 			}
2089 
2090 			if (itype != INTR_MSI || powerof2(iaq->nirq))
2091 				goto allocate;
2092 		}
2093 
2094 		/*
2095 		 * Least desirable option: one interrupt vector for everything.
2096 		 */
2097 		iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2098 		iaq->intr_flags_10g = iaq->intr_flags_1g = 0;
2099 #ifdef TCP_OFFLOAD
2100 		if (is_offload(sc))
2101 			iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2102 #endif
2103 #ifdef DEV_NETMAP
2104 		iaq->nnmrxq10g = iaq->nnmrxq1g = 1;
2105 #endif
2106 
2107 allocate:
2108 		navail = iaq->nirq;
2109 		rc = 0;
2110 		if (itype == INTR_MSIX)
2111 			rc = pci_alloc_msix(sc->dev, &navail);
2112 		else if (itype == INTR_MSI)
2113 			rc = pci_alloc_msi(sc->dev, &navail);
2114 
2115 		if (rc == 0) {
2116 			if (navail == iaq->nirq)
2117 				return (0);
2118 
2119 			/*
2120 			 * Didn't get the number requested.  Use whatever number
2121 			 * the kernel is willing to allocate (it's in navail).
2122 			 */
2123 			device_printf(sc->dev, "fewer vectors than requested, "
2124 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2125 			    itype, iaq->nirq, navail);
2126 			pci_release_msi(sc->dev);
2127 			goto restart;
2128 		}
2129 
2130 		device_printf(sc->dev,
2131 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2132 		    itype, rc, iaq->nirq, navail);
2133 	}
2134 
2135 	device_printf(sc->dev,
2136 	    "failed to find a usable interrupt type.  "
2137 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2138 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2139 
2140 	return (ENXIO);
2141 }
2142 
2143 #define FW_VERSION(chip) ( \
2144     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2145     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2146     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2147     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2148 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2149 
2150 struct fw_info {
2151 	uint8_t chip;
2152 	char *kld_name;
2153 	char *fw_mod_name;
2154 	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
2155 } fw_info[] = {
2156 	{
2157 		.chip = CHELSIO_T4,
2158 		.kld_name = "t4fw_cfg",
2159 		.fw_mod_name = "t4fw",
2160 		.fw_hdr = {
2161 			.chip = FW_HDR_CHIP_T4,
2162 			.fw_ver = htobe32_const(FW_VERSION(T4)),
2163 			.intfver_nic = FW_INTFVER(T4, NIC),
2164 			.intfver_vnic = FW_INTFVER(T4, VNIC),
2165 			.intfver_ofld = FW_INTFVER(T4, OFLD),
2166 			.intfver_ri = FW_INTFVER(T4, RI),
2167 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2168 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
2169 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2170 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
2171 		},
2172 	}, {
2173 		.chip = CHELSIO_T5,
2174 		.kld_name = "t5fw_cfg",
2175 		.fw_mod_name = "t5fw",
2176 		.fw_hdr = {
2177 			.chip = FW_HDR_CHIP_T5,
2178 			.fw_ver = htobe32_const(FW_VERSION(T5)),
2179 			.intfver_nic = FW_INTFVER(T5, NIC),
2180 			.intfver_vnic = FW_INTFVER(T5, VNIC),
2181 			.intfver_ofld = FW_INTFVER(T5, OFLD),
2182 			.intfver_ri = FW_INTFVER(T5, RI),
2183 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
2184 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
2185 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
2186 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
2187 		},
2188 	}
2189 };
2190 
2191 static struct fw_info *
2192 find_fw_info(int chip)
2193 {
2194 	int i;
2195 
2196 	for (i = 0; i < nitems(fw_info); i++) {
2197 		if (fw_info[i].chip == chip)
2198 			return (&fw_info[i]);
2199 	}
2200 	return (NULL);
2201 }
2202 
2203 /*
2204  * Is the given firmware API compatible with the one the driver was compiled
2205  * with?
2206  */
2207 static int
2208 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
2209 {
2210 
2211 	/* short circuit if it's the exact same firmware version */
2212 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
2213 		return (1);
2214 
2215 	/*
2216 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
2217 	 * features that are supported in the driver.
2218 	 */
2219 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
2220 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
2221 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
2222 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
2223 		return (1);
2224 #undef SAME_INTF
2225 
2226 	return (0);
2227 }
2228 
2229 /*
2230  * The firmware in the KLD is usable, but should it be installed?  This routine
2231  * explains itself in detail if it indicates the KLD firmware should be
2232  * installed.
2233  */
2234 static int
2235 should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
2236 {
2237 	const char *reason;
2238 
2239 	if (!card_fw_usable) {
2240 		reason = "incompatible or unusable";
2241 		goto install;
2242 	}
2243 
2244 	if (k > c) {
2245 		reason = "older than the version bundled with this driver";
2246 		goto install;
2247 	}
2248 
2249 	if (t4_fw_install == 2 && k != c) {
2250 		reason = "different than the version bundled with this driver";
2251 		goto install;
2252 	}
2253 
2254 	return (0);
2255 
2256 install:
2257 	if (t4_fw_install == 0) {
2258 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2259 		    "but the driver is prohibited from installing a different "
2260 		    "firmware on the card.\n",
2261 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2262 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
2263 
2264 		return (0);
2265 	}
2266 
2267 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2268 	    "installing firmware %u.%u.%u.%u on card.\n",
2269 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2270 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
2271 	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2272 	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2273 
2274 	return (1);
2275 }
2276 /*
2277  * Establish contact with the firmware and determine if we are the master driver
2278  * or not, and whether we are responsible for chip initialization.
2279  */
2280 static int
2281 prep_firmware(struct adapter *sc)
2282 {
2283 	const struct firmware *fw = NULL, *default_cfg;
2284 	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
2285 	enum dev_state state;
2286 	struct fw_info *fw_info;
2287 	struct fw_hdr *card_fw;		/* fw on the card */
2288 	const struct fw_hdr *kld_fw;	/* fw in the KLD */
2289 	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
2290 					   against */
2291 
2292 	/* Contact firmware. */
2293 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
2294 	if (rc < 0 || state == DEV_STATE_ERR) {
2295 		rc = -rc;
2296 		device_printf(sc->dev,
2297 		    "failed to connect to the firmware: %d, %d.\n", rc, state);
2298 		return (rc);
2299 	}
2300 	pf = rc;
2301 	if (pf == sc->mbox)
2302 		sc->flags |= MASTER_PF;
2303 	else if (state == DEV_STATE_UNINIT) {
2304 		/*
2305 		 * We didn't get to be the master so we definitely won't be
2306 		 * configuring the chip.  It's a bug if someone else hasn't
2307 		 * configured it already.
2308 		 */
2309 		device_printf(sc->dev, "couldn't be master(%d), "
2310 		    "device not already initialized either(%d).\n", rc, state);
2311 		return (EDOOFUS);
2312 	}
2313 
2314 	/* This is the firmware whose headers the driver was compiled against */
2315 	fw_info = find_fw_info(chip_id(sc));
2316 	if (fw_info == NULL) {
2317 		device_printf(sc->dev,
2318 		    "unable to look up firmware information for chip %d.\n",
2319 		    chip_id(sc));
2320 		return (EINVAL);
2321 	}
2322 	drv_fw = &fw_info->fw_hdr;
2323 
2324 	/*
2325 	 * The firmware KLD contains many modules.  The KLD name is also the
2326 	 * name of the module that contains the default config file.
2327 	 */
2328 	default_cfg = firmware_get(fw_info->kld_name);
2329 
2330 	/* Read the header of the firmware on the card */
2331 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
2332 	rc = -t4_read_flash(sc, FLASH_FW_START,
2333 	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
2334 	if (rc == 0)
2335 		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
2336 	else {
2337 		device_printf(sc->dev,
2338 		    "Unable to read card's firmware header: %d\n", rc);
2339 		card_fw_usable = 0;
2340 	}
2341 
2342 	/* This is the firmware in the KLD */
2343 	fw = firmware_get(fw_info->fw_mod_name);
2344 	if (fw != NULL) {
2345 		kld_fw = (const void *)fw->data;
2346 		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
2347 	} else {
2348 		kld_fw = NULL;
2349 		kld_fw_usable = 0;
2350 	}
2351 
2352 	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
2353 	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
2354 		/*
2355 		 * Common case: the firmware on the card is an exact match and
2356 		 * the KLD is an exact match too, or the KLD is
2357 		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
2358 		 * here -- use cxgbetool loadfw if you want to reinstall the
2359 		 * same firmware as the one on the card.
2360 		 */
2361 	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
2362 	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
2363 	    be32toh(card_fw->fw_ver))) {
2364 
2365 		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
2366 		if (rc != 0) {
2367 			device_printf(sc->dev,
2368 			    "failed to install firmware: %d\n", rc);
2369 			goto done;
2370 		}
2371 
2372 		/* Installed successfully, update the cached header too. */
2373 		memcpy(card_fw, kld_fw, sizeof(*card_fw));
2374 		card_fw_usable = 1;
2375 		need_fw_reset = 0;	/* already reset as part of load_fw */
2376 	}
2377 
2378 	if (!card_fw_usable) {
2379 		uint32_t d, c, k;
2380 
2381 		d = ntohl(drv_fw->fw_ver);
2382 		c = ntohl(card_fw->fw_ver);
2383 		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
2384 
2385 		device_printf(sc->dev, "Cannot find a usable firmware: "
2386 		    "fw_install %d, chip state %d, "
2387 		    "driver compiled with %d.%d.%d.%d, "
2388 		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
2389 		    t4_fw_install, state,
2390 		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
2391 		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
2392 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2393 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
2394 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2395 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2396 		rc = EINVAL;
2397 		goto done;
2398 	}
2399 
2400 	/* We're using whatever's on the card and it's known to be good. */
2401 	sc->params.fw_vers = ntohl(card_fw->fw_ver);
2402 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
2403 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
2404 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
2405 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
2406 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
2407 	t4_get_tp_version(sc, &sc->params.tp_vers);
2408 
2409 	/* Reset device */
2410 	if (need_fw_reset &&
2411 	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
2412 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
2413 		if (rc != ETIMEDOUT && rc != EIO)
2414 			t4_fw_bye(sc, sc->mbox);
2415 		goto done;
2416 	}
2417 	sc->flags |= FW_OK;
2418 
2419 	rc = get_params__pre_init(sc);
2420 	if (rc != 0)
2421 		goto done; /* error message displayed already */
2422 
2423 	/* Partition adapter resources as specified in the config file. */
2424 	if (state == DEV_STATE_UNINIT) {
2425 
2426 		KASSERT(sc->flags & MASTER_PF,
2427 		    ("%s: trying to change chip settings when not master.",
2428 		    __func__));
2429 
2430 		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
2431 		if (rc != 0)
2432 			goto done;	/* error message displayed already */
2433 
2434 		t4_tweak_chip_settings(sc);
2435 
2436 		/* get basic stuff going */
2437 		rc = -t4_fw_initialize(sc, sc->mbox);
2438 		if (rc != 0) {
2439 			device_printf(sc->dev, "fw init failed: %d.\n", rc);
2440 			goto done;
2441 		}
2442 	} else {
2443 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
2444 		sc->cfcsum = 0;
2445 	}
2446 
2447 done:
2448 	free(card_fw, M_CXGBE);
2449 	if (fw != NULL)
2450 		firmware_put(fw, FIRMWARE_UNLOAD);
2451 	if (default_cfg != NULL)
2452 		firmware_put(default_cfg, FIRMWARE_UNLOAD);
2453 
2454 	return (rc);
2455 }
2456 
2457 #define FW_PARAM_DEV(param) \
2458 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
2459 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
2460 #define FW_PARAM_PFVF(param) \
2461 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
2462 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
2463 
2464 /*
2465  * Partition chip resources for use between various PFs, VFs, etc.
2466  */
2467 static int
2468 partition_resources(struct adapter *sc, const struct firmware *default_cfg,
2469     const char *name_prefix)
2470 {
2471 	const struct firmware *cfg = NULL;
2472 	int rc = 0;
2473 	struct fw_caps_config_cmd caps;
2474 	uint32_t mtype, moff, finicsum, cfcsum;
2475 
2476 	/*
2477 	 * Figure out what configuration file to use.  Pick the default config
2478 	 * file for the card if the user hasn't specified one explicitly.
2479 	 */
2480 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
2481 	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
2482 		/* Card specific overrides go here. */
2483 		if (pci_get_device(sc->dev) == 0x440a)
2484 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
2485 		if (is_fpga(sc))
2486 			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
2487 	}
2488 
2489 	/*
2490 	 * We need to load another module if the profile is anything except
2491 	 * "default" or "flash".
2492 	 */
2493 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
2494 	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
2495 		char s[32];
2496 
2497 		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
2498 		cfg = firmware_get(s);
2499 		if (cfg == NULL) {
2500 			if (default_cfg != NULL) {
2501 				device_printf(sc->dev,
2502 				    "unable to load module \"%s\" for "
2503 				    "configuration profile \"%s\", will use "
2504 				    "the default config file instead.\n",
2505 				    s, sc->cfg_file);
2506 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
2507 				    "%s", DEFAULT_CF);
2508 			} else {
2509 				device_printf(sc->dev,
2510 				    "unable to load module \"%s\" for "
2511 				    "configuration profile \"%s\", will use "
2512 				    "the config file on the card's flash "
2513 				    "instead.\n", s, sc->cfg_file);
2514 				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
2515 				    "%s", FLASH_CF);
2516 			}
2517 		}
2518 	}
2519 
2520 	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
2521 	    default_cfg == NULL) {
2522 		device_printf(sc->dev,
2523 		    "default config file not available, will use the config "
2524 		    "file on the card's flash instead.\n");
2525 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
2526 	}
2527 
2528 	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
2529 		u_int cflen, i, n;
2530 		const uint32_t *cfdata;
2531 		uint32_t param, val, addr, off, mw_base, mw_aperture;
2532 
2533 		KASSERT(cfg != NULL || default_cfg != NULL,
2534 		    ("%s: no config to upload", __func__));
2535 
2536 		/*
2537 		 * Ask the firmware where it wants us to upload the config file.
2538 		 */
2539 		param = FW_PARAM_DEV(CF);
2540 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2541 		if (rc != 0) {
2542 			/* No support for config file?  Shouldn't happen. */
2543 			device_printf(sc->dev,
2544 			    "failed to query config file location: %d.\n", rc);
2545 			goto done;
2546 		}
2547 		mtype = G_FW_PARAMS_PARAM_Y(val);
2548 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
2549 
2550 		/*
2551 		 * XXX: sheer laziness.  We deliberately added 4 bytes of
2552 		 * useless stuffing/comments at the end of the config file so
2553 		 * it's ok to simply throw away the last remaining bytes when
2554 		 * the config file is not an exact multiple of 4.  This also
2555 		 * helps with the validate_mt_off_len check.
2556 		 */
2557 		if (cfg != NULL) {
2558 			cflen = cfg->datasize & ~3;
2559 			cfdata = cfg->data;
2560 		} else {
2561 			cflen = default_cfg->datasize & ~3;
2562 			cfdata = default_cfg->data;
2563 		}
2564 
2565 		if (cflen > FLASH_CFG_MAX_SIZE) {
2566 			device_printf(sc->dev,
2567 			    "config file too long (%d, max allowed is %d).  "
2568 			    "Will try to use the config on the card, if any.\n",
2569 			    cflen, FLASH_CFG_MAX_SIZE);
2570 			goto use_config_on_flash;
2571 		}
2572 
2573 		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
2574 		if (rc != 0) {
2575 			device_printf(sc->dev,
2576 			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
2577 			    "Will try to use the config on the card, if any.\n",
2578 			    __func__, mtype, moff, cflen, rc);
2579 			goto use_config_on_flash;
2580 		}
2581 
2582 		memwin_info(sc, 2, &mw_base, &mw_aperture);
2583 		while (cflen) {
2584 			off = position_memwin(sc, 2, addr);
2585 			n = min(cflen, mw_aperture - off);
2586 			for (i = 0; i < n; i += 4)
2587 				t4_write_reg(sc, mw_base + off + i, *cfdata++);
2588 			cflen -= n;
2589 			addr += n;
2590 		}
2591 	} else {
2592 use_config_on_flash:
2593 		mtype = FW_MEMTYPE_FLASH;
2594 		moff = t4_flash_cfg_addr(sc);
2595 	}
2596 
2597 	bzero(&caps, sizeof(caps));
2598 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2599 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
2600 	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
2601 	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
2602 	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
2603 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
2604 	if (rc != 0) {
2605 		device_printf(sc->dev,
2606 		    "failed to pre-process config file: %d "
2607 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
2608 		goto done;
2609 	}
2610 
2611 	finicsum = be32toh(caps.finicsum);
2612 	cfcsum = be32toh(caps.cfcsum);
2613 	if (finicsum != cfcsum) {
2614 		device_printf(sc->dev,
2615 		    "WARNING: config file checksum mismatch: %08x %08x\n",
2616 		    finicsum, cfcsum);
2617 	}
2618 	sc->cfcsum = cfcsum;
2619 
2620 #define LIMIT_CAPS(x) do { \
2621 	caps.x &= htobe16(t4_##x##_allowed); \
2622 } while (0)
2623 
2624 	/*
2625 	 * Let the firmware know what features will (not) be used so it can tune
2626 	 * things accordingly.
2627 	 */
2628 	LIMIT_CAPS(linkcaps);
2629 	LIMIT_CAPS(niccaps);
2630 	LIMIT_CAPS(toecaps);
2631 	LIMIT_CAPS(rdmacaps);
2632 	LIMIT_CAPS(iscsicaps);
2633 	LIMIT_CAPS(fcoecaps);
2634 #undef LIMIT_CAPS
2635 
2636 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2637 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
2638 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
2639 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
2640 	if (rc != 0) {
2641 		device_printf(sc->dev,
2642 		    "failed to process config file: %d.\n", rc);
2643 	}
2644 done:
2645 	if (cfg != NULL)
2646 		firmware_put(cfg, FIRMWARE_UNLOAD);
2647 	return (rc);
2648 }
2649 
2650 /*
2651  * Retrieve parameters that are needed (or nice to have) very early.
2652  */
2653 static int
2654 get_params__pre_init(struct adapter *sc)
2655 {
2656 	int rc;
2657 	uint32_t param[2], val[2];
2658 	struct fw_devlog_cmd cmd;
2659 	struct devlog_params *dlog = &sc->params.devlog;
2660 
2661 	param[0] = FW_PARAM_DEV(PORTVEC);
2662 	param[1] = FW_PARAM_DEV(CCLK);
2663 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
2664 	if (rc != 0) {
2665 		device_printf(sc->dev,
2666 		    "failed to query parameters (pre_init): %d.\n", rc);
2667 		return (rc);
2668 	}
2669 
2670 	sc->params.portvec = val[0];
2671 	sc->params.nports = bitcount32(val[0]);
2672 	sc->params.vpd.cclk = val[1];
2673 
2674 	/* Read device log parameters. */
2675 	bzero(&cmd, sizeof(cmd));
2676 	cmd.op_to_write = htobe32(V_FW_CMD_OP(FW_DEVLOG_CMD) |
2677 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
2678 	cmd.retval_len16 = htobe32(FW_LEN16(cmd));
2679 	rc = -t4_wr_mbox(sc, sc->mbox, &cmd, sizeof(cmd), &cmd);
2680 	if (rc != 0) {
2681 		device_printf(sc->dev,
2682 		    "failed to get devlog parameters: %d.\n", rc);
2683 		bzero(dlog, sizeof (*dlog));
2684 		rc = 0;	/* devlog isn't critical for device operation */
2685 	} else {
2686 		val[0] = be32toh(cmd.memtype_devlog_memaddr16_devlog);
2687 		dlog->memtype = G_FW_DEVLOG_CMD_MEMTYPE_DEVLOG(val[0]);
2688 		dlog->start = G_FW_DEVLOG_CMD_MEMADDR16_DEVLOG(val[0]) << 4;
2689 		dlog->size = be32toh(cmd.memsize_devlog);
2690 	}
2691 
2692 	return (rc);
2693 }
2694 
2695 /*
2696  * Retrieve various parameters that are of interest to the driver.  The device
2697  * has been initialized by the firmware at this point.
2698  */
2699 static int
2700 get_params__post_init(struct adapter *sc)
2701 {
2702 	int rc;
2703 	uint32_t param[7], val[7];
2704 	struct fw_caps_config_cmd caps;
2705 
2706 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
2707 	param[1] = FW_PARAM_PFVF(EQ_START);
2708 	param[2] = FW_PARAM_PFVF(FILTER_START);
2709 	param[3] = FW_PARAM_PFVF(FILTER_END);
2710 	param[4] = FW_PARAM_PFVF(L2T_START);
2711 	param[5] = FW_PARAM_PFVF(L2T_END);
2712 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
2713 	if (rc != 0) {
2714 		device_printf(sc->dev,
2715 		    "failed to query parameters (post_init): %d.\n", rc);
2716 		return (rc);
2717 	}
2718 
2719 	sc->sge.iq_start = val[0];
2720 	sc->sge.eq_start = val[1];
2721 	sc->tids.ftid_base = val[2];
2722 	sc->tids.nftids = val[3] - val[2] + 1;
2723 	sc->params.ftid_min = val[2];
2724 	sc->params.ftid_max = val[3];
2725 	sc->vres.l2t.start = val[4];
2726 	sc->vres.l2t.size = val[5] - val[4] + 1;
2727 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
2728 	    ("%s: L2 table size (%u) larger than expected (%u)",
2729 	    __func__, sc->vres.l2t.size, L2T_SIZE));
2730 
2731 	/* get capabilites */
2732 	bzero(&caps, sizeof(caps));
2733 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2734 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
2735 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
2736 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
2737 	if (rc != 0) {
2738 		device_printf(sc->dev,
2739 		    "failed to get card capabilities: %d.\n", rc);
2740 		return (rc);
2741 	}
2742 
2743 #define READ_CAPS(x) do { \
2744 	sc->x = htobe16(caps.x); \
2745 } while (0)
2746 	READ_CAPS(linkcaps);
2747 	READ_CAPS(niccaps);
2748 	READ_CAPS(toecaps);
2749 	READ_CAPS(rdmacaps);
2750 	READ_CAPS(iscsicaps);
2751 	READ_CAPS(fcoecaps);
2752 
2753 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
2754 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
2755 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
2756 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
2757 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
2758 		if (rc != 0) {
2759 			device_printf(sc->dev,
2760 			    "failed to query NIC parameters: %d.\n", rc);
2761 			return (rc);
2762 		}
2763 		sc->tids.etid_base = val[0];
2764 		sc->params.etid_min = val[0];
2765 		sc->tids.netids = val[1] - val[0] + 1;
2766 		sc->params.netids = sc->tids.netids;
2767 		sc->params.eo_wr_cred = val[2];
2768 		sc->params.ethoffload = 1;
2769 	}
2770 
2771 	if (sc->toecaps) {
2772 		/* query offload-related parameters */
2773 		param[0] = FW_PARAM_DEV(NTID);
2774 		param[1] = FW_PARAM_PFVF(SERVER_START);
2775 		param[2] = FW_PARAM_PFVF(SERVER_END);
2776 		param[3] = FW_PARAM_PFVF(TDDP_START);
2777 		param[4] = FW_PARAM_PFVF(TDDP_END);
2778 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
2779 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
2780 		if (rc != 0) {
2781 			device_printf(sc->dev,
2782 			    "failed to query TOE parameters: %d.\n", rc);
2783 			return (rc);
2784 		}
2785 		sc->tids.ntids = val[0];
2786 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
2787 		sc->tids.stid_base = val[1];
2788 		sc->tids.nstids = val[2] - val[1] + 1;
2789 		sc->vres.ddp.start = val[3];
2790 		sc->vres.ddp.size = val[4] - val[3] + 1;
2791 		sc->params.ofldq_wr_cred = val[5];
2792 		sc->params.offload = 1;
2793 	}
2794 	if (sc->rdmacaps) {
2795 		param[0] = FW_PARAM_PFVF(STAG_START);
2796 		param[1] = FW_PARAM_PFVF(STAG_END);
2797 		param[2] = FW_PARAM_PFVF(RQ_START);
2798 		param[3] = FW_PARAM_PFVF(RQ_END);
2799 		param[4] = FW_PARAM_PFVF(PBL_START);
2800 		param[5] = FW_PARAM_PFVF(PBL_END);
2801 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
2802 		if (rc != 0) {
2803 			device_printf(sc->dev,
2804 			    "failed to query RDMA parameters(1): %d.\n", rc);
2805 			return (rc);
2806 		}
2807 		sc->vres.stag.start = val[0];
2808 		sc->vres.stag.size = val[1] - val[0] + 1;
2809 		sc->vres.rq.start = val[2];
2810 		sc->vres.rq.size = val[3] - val[2] + 1;
2811 		sc->vres.pbl.start = val[4];
2812 		sc->vres.pbl.size = val[5] - val[4] + 1;
2813 
2814 		param[0] = FW_PARAM_PFVF(SQRQ_START);
2815 		param[1] = FW_PARAM_PFVF(SQRQ_END);
2816 		param[2] = FW_PARAM_PFVF(CQ_START);
2817 		param[3] = FW_PARAM_PFVF(CQ_END);
2818 		param[4] = FW_PARAM_PFVF(OCQ_START);
2819 		param[5] = FW_PARAM_PFVF(OCQ_END);
2820 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
2821 		if (rc != 0) {
2822 			device_printf(sc->dev,
2823 			    "failed to query RDMA parameters(2): %d.\n", rc);
2824 			return (rc);
2825 		}
2826 		sc->vres.qp.start = val[0];
2827 		sc->vres.qp.size = val[1] - val[0] + 1;
2828 		sc->vres.cq.start = val[2];
2829 		sc->vres.cq.size = val[3] - val[2] + 1;
2830 		sc->vres.ocq.start = val[4];
2831 		sc->vres.ocq.size = val[5] - val[4] + 1;
2832 	}
2833 	if (sc->iscsicaps) {
2834 		param[0] = FW_PARAM_PFVF(ISCSI_START);
2835 		param[1] = FW_PARAM_PFVF(ISCSI_END);
2836 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
2837 		if (rc != 0) {
2838 			device_printf(sc->dev,
2839 			    "failed to query iSCSI parameters: %d.\n", rc);
2840 			return (rc);
2841 		}
2842 		sc->vres.iscsi.start = val[0];
2843 		sc->vres.iscsi.size = val[1] - val[0] + 1;
2844 	}
2845 
2846 	/*
2847 	 * We've got the params we wanted to query via the firmware.  Now grab
2848 	 * some others directly from the chip.
2849 	 */
2850 	rc = t4_read_chip_settings(sc);
2851 
2852 	return (rc);
2853 }
2854 
2855 static int
2856 set_params__post_init(struct adapter *sc)
2857 {
2858 	uint32_t param, val;
2859 
2860 	/* ask for encapsulated CPLs */
2861 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
2862 	val = 1;
2863 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2864 
2865 	return (0);
2866 }
2867 
2868 #undef FW_PARAM_PFVF
2869 #undef FW_PARAM_DEV
2870 
2871 static void
2872 t4_set_desc(struct adapter *sc)
2873 {
2874 	char buf[128];
2875 	struct adapter_params *p = &sc->params;
2876 
2877 	snprintf(buf, sizeof(buf), "Chelsio %s %sNIC (rev %d), S/N:%s, "
2878 	    "P/N:%s, E/C:%s", p->vpd.id, is_offload(sc) ? "R" : "",
2879 	    chip_rev(sc), p->vpd.sn, p->vpd.pn, p->vpd.ec);
2880 
2881 	device_set_desc_copy(sc->dev, buf);
2882 }
2883 
2884 static void
2885 build_medialist(struct port_info *pi, struct ifmedia *media)
2886 {
2887 	int data, m;
2888 
2889 	PORT_LOCK(pi);
2890 
2891 	ifmedia_removeall(media);
2892 
2893 	m = IFM_ETHER | IFM_FDX;
2894 	data = (pi->port_type << 8) | pi->mod_type;
2895 
2896 	switch(pi->port_type) {
2897 	case FW_PORT_TYPE_BT_XFI:
2898 		ifmedia_add(media, m | IFM_10G_T, data, NULL);
2899 		break;
2900 
2901 	case FW_PORT_TYPE_BT_XAUI:
2902 		ifmedia_add(media, m | IFM_10G_T, data, NULL);
2903 		/* fall through */
2904 
2905 	case FW_PORT_TYPE_BT_SGMII:
2906 		ifmedia_add(media, m | IFM_1000_T, data, NULL);
2907 		ifmedia_add(media, m | IFM_100_TX, data, NULL);
2908 		ifmedia_add(media, IFM_ETHER | IFM_AUTO, data, NULL);
2909 		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
2910 		break;
2911 
2912 	case FW_PORT_TYPE_CX4:
2913 		ifmedia_add(media, m | IFM_10G_CX4, data, NULL);
2914 		ifmedia_set(media, m | IFM_10G_CX4);
2915 		break;
2916 
2917 	case FW_PORT_TYPE_QSFP_10G:
2918 	case FW_PORT_TYPE_SFP:
2919 	case FW_PORT_TYPE_FIBER_XFI:
2920 	case FW_PORT_TYPE_FIBER_XAUI:
2921 		switch (pi->mod_type) {
2922 
2923 		case FW_PORT_MOD_TYPE_LR:
2924 			ifmedia_add(media, m | IFM_10G_LR, data, NULL);
2925 			ifmedia_set(media, m | IFM_10G_LR);
2926 			break;
2927 
2928 		case FW_PORT_MOD_TYPE_SR:
2929 			ifmedia_add(media, m | IFM_10G_SR, data, NULL);
2930 			ifmedia_set(media, m | IFM_10G_SR);
2931 			break;
2932 
2933 		case FW_PORT_MOD_TYPE_LRM:
2934 			ifmedia_add(media, m | IFM_10G_LRM, data, NULL);
2935 			ifmedia_set(media, m | IFM_10G_LRM);
2936 			break;
2937 
2938 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2939 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2940 			ifmedia_add(media, m | IFM_10G_TWINAX, data, NULL);
2941 			ifmedia_set(media, m | IFM_10G_TWINAX);
2942 			break;
2943 
2944 		case FW_PORT_MOD_TYPE_NONE:
2945 			m &= ~IFM_FDX;
2946 			ifmedia_add(media, m | IFM_NONE, data, NULL);
2947 			ifmedia_set(media, m | IFM_NONE);
2948 			break;
2949 
2950 		case FW_PORT_MOD_TYPE_NA:
2951 		case FW_PORT_MOD_TYPE_ER:
2952 		default:
2953 			device_printf(pi->dev,
2954 			    "unknown port_type (%d), mod_type (%d)\n",
2955 			    pi->port_type, pi->mod_type);
2956 			ifmedia_add(media, m | IFM_UNKNOWN, data, NULL);
2957 			ifmedia_set(media, m | IFM_UNKNOWN);
2958 			break;
2959 		}
2960 		break;
2961 
2962 	case FW_PORT_TYPE_QSFP:
2963 		switch (pi->mod_type) {
2964 
2965 		case FW_PORT_MOD_TYPE_LR:
2966 			ifmedia_add(media, m | IFM_40G_LR4, data, NULL);
2967 			ifmedia_set(media, m | IFM_40G_LR4);
2968 			break;
2969 
2970 		case FW_PORT_MOD_TYPE_SR:
2971 			ifmedia_add(media, m | IFM_40G_SR4, data, NULL);
2972 			ifmedia_set(media, m | IFM_40G_SR4);
2973 			break;
2974 
2975 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2976 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2977 			ifmedia_add(media, m | IFM_40G_CR4, data, NULL);
2978 			ifmedia_set(media, m | IFM_40G_CR4);
2979 			break;
2980 
2981 		case FW_PORT_MOD_TYPE_NONE:
2982 			m &= ~IFM_FDX;
2983 			ifmedia_add(media, m | IFM_NONE, data, NULL);
2984 			ifmedia_set(media, m | IFM_NONE);
2985 			break;
2986 
2987 		default:
2988 			device_printf(pi->dev,
2989 			    "unknown port_type (%d), mod_type (%d)\n",
2990 			    pi->port_type, pi->mod_type);
2991 			ifmedia_add(media, m | IFM_UNKNOWN, data, NULL);
2992 			ifmedia_set(media, m | IFM_UNKNOWN);
2993 			break;
2994 		}
2995 		break;
2996 
2997 	default:
2998 		device_printf(pi->dev,
2999 		    "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
3000 		    pi->mod_type);
3001 		ifmedia_add(media, m | IFM_UNKNOWN, data, NULL);
3002 		ifmedia_set(media, m | IFM_UNKNOWN);
3003 		break;
3004 	}
3005 
3006 	PORT_UNLOCK(pi);
3007 }
3008 
3009 #define FW_MAC_EXACT_CHUNK	7
3010 
3011 /*
3012  * Program the port's XGMAC based on parameters in ifnet.  The caller also
3013  * indicates which parameters should be programmed (the rest are left alone).
3014  */
3015 int
3016 update_mac_settings(struct ifnet *ifp, int flags)
3017 {
3018 	int rc = 0;
3019 	struct port_info *pi = ifp->if_softc;
3020 	struct adapter *sc = pi->adapter;
3021 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
3022 	uint16_t viid = 0xffff;
3023 	int16_t *xact_addr_filt = NULL;
3024 
3025 	ASSERT_SYNCHRONIZED_OP(sc);
3026 	KASSERT(flags, ("%s: not told what to update.", __func__));
3027 
3028 	if (ifp == pi->ifp) {
3029 		viid = pi->viid;
3030 		xact_addr_filt = &pi->xact_addr_filt;
3031 	}
3032 #ifdef DEV_NETMAP
3033 	else if (ifp == pi->nm_ifp) {
3034 		viid = pi->nm_viid;
3035 		xact_addr_filt = &pi->nm_xact_addr_filt;
3036 	}
3037 #endif
3038 	if (flags & XGMAC_MTU)
3039 		mtu = ifp->if_mtu;
3040 
3041 	if (flags & XGMAC_PROMISC)
3042 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
3043 
3044 	if (flags & XGMAC_ALLMULTI)
3045 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
3046 
3047 	if (flags & XGMAC_VLANEX)
3048 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
3049 
3050 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
3051 		rc = -t4_set_rxmode(sc, sc->mbox, viid, mtu, promisc, allmulti,
3052 		    1, vlanex, false);
3053 		if (rc) {
3054 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
3055 			    rc);
3056 			return (rc);
3057 		}
3058 	}
3059 
3060 	if (flags & XGMAC_UCADDR) {
3061 		uint8_t ucaddr[ETHER_ADDR_LEN];
3062 
3063 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
3064 		rc = t4_change_mac(sc, sc->mbox, viid, *xact_addr_filt, ucaddr,
3065 		    true, true);
3066 		if (rc < 0) {
3067 			rc = -rc;
3068 			if_printf(ifp, "change_mac failed: %d\n", rc);
3069 			return (rc);
3070 		} else {
3071 			*xact_addr_filt = rc;
3072 			rc = 0;
3073 		}
3074 	}
3075 
3076 	if (flags & XGMAC_MCADDRS) {
3077 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
3078 		int del = 1;
3079 		uint64_t hash = 0;
3080 		struct ifmultiaddr *ifma;
3081 		int i = 0, j;
3082 
3083 		if_maddr_rlock(ifp);
3084 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3085 			if (ifma->ifma_addr->sa_family != AF_LINK)
3086 				continue;
3087 			mcaddr[i] =
3088 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
3089 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
3090 			i++;
3091 
3092 			if (i == FW_MAC_EXACT_CHUNK) {
3093 				rc = t4_alloc_mac_filt(sc, sc->mbox, viid, del,
3094 				    i, mcaddr, NULL, &hash, 0);
3095 				if (rc < 0) {
3096 					rc = -rc;
3097 					for (j = 0; j < i; j++) {
3098 						if_printf(ifp,
3099 						    "failed to add mc address"
3100 						    " %02x:%02x:%02x:"
3101 						    "%02x:%02x:%02x rc=%d\n",
3102 						    mcaddr[j][0], mcaddr[j][1],
3103 						    mcaddr[j][2], mcaddr[j][3],
3104 						    mcaddr[j][4], mcaddr[j][5],
3105 						    rc);
3106 					}
3107 					goto mcfail;
3108 				}
3109 				del = 0;
3110 				i = 0;
3111 			}
3112 		}
3113 		if (i > 0) {
3114 			rc = t4_alloc_mac_filt(sc, sc->mbox, viid, del, i,
3115 			    mcaddr, NULL, &hash, 0);
3116 			if (rc < 0) {
3117 				rc = -rc;
3118 				for (j = 0; j < i; j++) {
3119 					if_printf(ifp,
3120 					    "failed to add mc address"
3121 					    " %02x:%02x:%02x:"
3122 					    "%02x:%02x:%02x rc=%d\n",
3123 					    mcaddr[j][0], mcaddr[j][1],
3124 					    mcaddr[j][2], mcaddr[j][3],
3125 					    mcaddr[j][4], mcaddr[j][5],
3126 					    rc);
3127 				}
3128 				goto mcfail;
3129 			}
3130 		}
3131 
3132 		rc = -t4_set_addr_hash(sc, sc->mbox, viid, 0, hash, 0);
3133 		if (rc != 0)
3134 			if_printf(ifp, "failed to set mc address hash: %d", rc);
3135 mcfail:
3136 		if_maddr_runlock(ifp);
3137 	}
3138 
3139 	return (rc);
3140 }
3141 
3142 int
3143 begin_synchronized_op(struct adapter *sc, struct port_info *pi, int flags,
3144     char *wmesg)
3145 {
3146 	int rc, pri;
3147 
3148 #ifdef WITNESS
3149 	/* the caller thinks it's ok to sleep, but is it really? */
3150 	if (flags & SLEEP_OK)
3151 		pause("t4slptst", 1);
3152 #endif
3153 
3154 	if (INTR_OK)
3155 		pri = PCATCH;
3156 	else
3157 		pri = 0;
3158 
3159 	ADAPTER_LOCK(sc);
3160 	for (;;) {
3161 
3162 		if (pi && IS_DOOMED(pi)) {
3163 			rc = ENXIO;
3164 			goto done;
3165 		}
3166 
3167 		if (!IS_BUSY(sc)) {
3168 			rc = 0;
3169 			break;
3170 		}
3171 
3172 		if (!(flags & SLEEP_OK)) {
3173 			rc = EBUSY;
3174 			goto done;
3175 		}
3176 
3177 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
3178 			rc = EINTR;
3179 			goto done;
3180 		}
3181 	}
3182 
3183 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
3184 	SET_BUSY(sc);
3185 #ifdef INVARIANTS
3186 	sc->last_op = wmesg;
3187 	sc->last_op_thr = curthread;
3188 #endif
3189 
3190 done:
3191 	if (!(flags & HOLD_LOCK) || rc)
3192 		ADAPTER_UNLOCK(sc);
3193 
3194 	return (rc);
3195 }
3196 
3197 void
3198 end_synchronized_op(struct adapter *sc, int flags)
3199 {
3200 
3201 	if (flags & LOCK_HELD)
3202 		ADAPTER_LOCK_ASSERT_OWNED(sc);
3203 	else
3204 		ADAPTER_LOCK(sc);
3205 
3206 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
3207 	CLR_BUSY(sc);
3208 	wakeup(&sc->flags);
3209 	ADAPTER_UNLOCK(sc);
3210 }
3211 
3212 static int
3213 cxgbe_init_synchronized(struct port_info *pi)
3214 {
3215 	struct adapter *sc = pi->adapter;
3216 	struct ifnet *ifp = pi->ifp;
3217 	int rc = 0, i;
3218 	struct sge_txq *txq;
3219 
3220 	ASSERT_SYNCHRONIZED_OP(sc);
3221 
3222 	if (isset(&sc->open_device_map, pi->port_id)) {
3223 		KASSERT(ifp->if_drv_flags & IFF_DRV_RUNNING,
3224 		    ("mismatch between open_device_map and if_drv_flags"));
3225 		return (0);	/* already running */
3226 	}
3227 
3228 	if (!(sc->flags & FULL_INIT_DONE) &&
3229 	    ((rc = adapter_full_init(sc)) != 0))
3230 		return (rc);	/* error message displayed already */
3231 
3232 	if (!(pi->flags & PORT_INIT_DONE) &&
3233 	    ((rc = port_full_init(pi)) != 0))
3234 		return (rc); /* error message displayed already */
3235 
3236 	rc = update_mac_settings(ifp, XGMAC_ALL);
3237 	if (rc)
3238 		goto done;	/* error message displayed already */
3239 
3240 	rc = -t4_enable_vi(sc, sc->mbox, pi->viid, true, true);
3241 	if (rc != 0) {
3242 		if_printf(ifp, "enable_vi failed: %d\n", rc);
3243 		goto done;
3244 	}
3245 
3246 	/*
3247 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
3248 	 * if this changes.
3249 	 */
3250 
3251 	for_each_txq(pi, i, txq) {
3252 		TXQ_LOCK(txq);
3253 		txq->eq.flags |= EQ_ENABLED;
3254 		TXQ_UNLOCK(txq);
3255 	}
3256 
3257 	/*
3258 	 * The first iq of the first port to come up is used for tracing.
3259 	 */
3260 	if (sc->traceq < 0) {
3261 		sc->traceq = sc->sge.rxq[pi->first_rxq].iq.abs_id;
3262 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
3263 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
3264 		    V_QUEUENUMBER(sc->traceq));
3265 		pi->flags |= HAS_TRACEQ;
3266 	}
3267 
3268 	/* all ok */
3269 	setbit(&sc->open_device_map, pi->port_id);
3270 	PORT_LOCK(pi);
3271 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3272 	PORT_UNLOCK(pi);
3273 
3274 	callout_reset(&pi->tick, hz, cxgbe_tick, pi);
3275 done:
3276 	if (rc != 0)
3277 		cxgbe_uninit_synchronized(pi);
3278 
3279 	return (rc);
3280 }
3281 
3282 /*
3283  * Idempotent.
3284  */
3285 static int
3286 cxgbe_uninit_synchronized(struct port_info *pi)
3287 {
3288 	struct adapter *sc = pi->adapter;
3289 	struct ifnet *ifp = pi->ifp;
3290 	int rc, i;
3291 	struct sge_txq *txq;
3292 
3293 	ASSERT_SYNCHRONIZED_OP(sc);
3294 
3295 	/*
3296 	 * Disable the VI so that all its data in either direction is discarded
3297 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
3298 	 * tick) intact as the TP can deliver negative advice or data that it's
3299 	 * holding in its RAM (for an offloaded connection) even after the VI is
3300 	 * disabled.
3301 	 */
3302 	rc = -t4_enable_vi(sc, sc->mbox, pi->viid, false, false);
3303 	if (rc) {
3304 		if_printf(ifp, "disable_vi failed: %d\n", rc);
3305 		return (rc);
3306 	}
3307 
3308 	for_each_txq(pi, i, txq) {
3309 		TXQ_LOCK(txq);
3310 		txq->eq.flags &= ~EQ_ENABLED;
3311 		TXQ_UNLOCK(txq);
3312 	}
3313 
3314 	clrbit(&sc->open_device_map, pi->port_id);
3315 	PORT_LOCK(pi);
3316 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3317 	PORT_UNLOCK(pi);
3318 
3319 	pi->link_cfg.link_ok = 0;
3320 	pi->link_cfg.speed = 0;
3321 	pi->linkdnrc = -1;
3322 	t4_os_link_changed(sc, pi->port_id, 0, -1);
3323 
3324 	return (0);
3325 }
3326 
3327 /*
3328  * It is ok for this function to fail midway and return right away.  t4_detach
3329  * will walk the entire sc->irq list and clean up whatever is valid.
3330  */
3331 static int
3332 setup_intr_handlers(struct adapter *sc)
3333 {
3334 	int rc, rid, p, q;
3335 	char s[8];
3336 	struct irq *irq;
3337 	struct port_info *pi;
3338 	struct sge_rxq *rxq;
3339 #ifdef TCP_OFFLOAD
3340 	struct sge_ofld_rxq *ofld_rxq;
3341 #endif
3342 #ifdef DEV_NETMAP
3343 	struct sge_nm_rxq *nm_rxq;
3344 #endif
3345 
3346 	/*
3347 	 * Setup interrupts.
3348 	 */
3349 	irq = &sc->irq[0];
3350 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
3351 	if (sc->intr_count == 1)
3352 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
3353 
3354 	/* Multiple interrupts. */
3355 	KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
3356 	    ("%s: too few intr.", __func__));
3357 
3358 	/* The first one is always error intr */
3359 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
3360 	if (rc != 0)
3361 		return (rc);
3362 	irq++;
3363 	rid++;
3364 
3365 	/* The second one is always the firmware event queue */
3366 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sc->sge.fwq, "evt");
3367 	if (rc != 0)
3368 		return (rc);
3369 	irq++;
3370 	rid++;
3371 
3372 	for_each_port(sc, p) {
3373 		pi = sc->port[p];
3374 
3375 		if (pi->flags & INTR_RXQ) {
3376 			for_each_rxq(pi, q, rxq) {
3377 				snprintf(s, sizeof(s), "%d.%d", p, q);
3378 				rc = t4_alloc_irq(sc, irq, rid, t4_intr, rxq,
3379 				    s);
3380 				if (rc != 0)
3381 					return (rc);
3382 				irq++;
3383 				rid++;
3384 			}
3385 		}
3386 #ifdef TCP_OFFLOAD
3387 		if (pi->flags & INTR_OFLD_RXQ) {
3388 			for_each_ofld_rxq(pi, q, ofld_rxq) {
3389 				snprintf(s, sizeof(s), "%d,%d", p, q);
3390 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
3391 				    ofld_rxq, s);
3392 				if (rc != 0)
3393 					return (rc);
3394 				irq++;
3395 				rid++;
3396 			}
3397 		}
3398 #endif
3399 #ifdef DEV_NETMAP
3400 		if (pi->flags & INTR_NM_RXQ) {
3401 			for_each_nm_rxq(pi, q, nm_rxq) {
3402 				snprintf(s, sizeof(s), "%d-%d", p, q);
3403 				rc = t4_alloc_irq(sc, irq, rid, t4_nm_intr,
3404 				    nm_rxq, s);
3405 				if (rc != 0)
3406 					return (rc);
3407 				irq++;
3408 				rid++;
3409 			}
3410 		}
3411 #endif
3412 	}
3413 	MPASS(irq == &sc->irq[sc->intr_count]);
3414 
3415 	return (0);
3416 }
3417 
3418 int
3419 adapter_full_init(struct adapter *sc)
3420 {
3421 	int rc, i;
3422 
3423 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
3424 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
3425 	    ("%s: FULL_INIT_DONE already", __func__));
3426 
3427 	/*
3428 	 * queues that belong to the adapter (not any particular port).
3429 	 */
3430 	rc = t4_setup_adapter_queues(sc);
3431 	if (rc != 0)
3432 		goto done;
3433 
3434 	for (i = 0; i < nitems(sc->tq); i++) {
3435 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
3436 		    taskqueue_thread_enqueue, &sc->tq[i]);
3437 		if (sc->tq[i] == NULL) {
3438 			device_printf(sc->dev,
3439 			    "failed to allocate task queue %d\n", i);
3440 			rc = ENOMEM;
3441 			goto done;
3442 		}
3443 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
3444 		    device_get_nameunit(sc->dev), i);
3445 	}
3446 
3447 	t4_intr_enable(sc);
3448 	sc->flags |= FULL_INIT_DONE;
3449 done:
3450 	if (rc != 0)
3451 		adapter_full_uninit(sc);
3452 
3453 	return (rc);
3454 }
3455 
3456 int
3457 adapter_full_uninit(struct adapter *sc)
3458 {
3459 	int i;
3460 
3461 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
3462 
3463 	t4_teardown_adapter_queues(sc);
3464 
3465 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
3466 		taskqueue_free(sc->tq[i]);
3467 		sc->tq[i] = NULL;
3468 	}
3469 
3470 	sc->flags &= ~FULL_INIT_DONE;
3471 
3472 	return (0);
3473 }
3474 
3475 int
3476 port_full_init(struct port_info *pi)
3477 {
3478 	struct adapter *sc = pi->adapter;
3479 	struct ifnet *ifp = pi->ifp;
3480 	uint16_t *rss;
3481 	struct sge_rxq *rxq;
3482 	int rc, i, j;
3483 
3484 	ASSERT_SYNCHRONIZED_OP(sc);
3485 	KASSERT((pi->flags & PORT_INIT_DONE) == 0,
3486 	    ("%s: PORT_INIT_DONE already", __func__));
3487 
3488 	sysctl_ctx_init(&pi->ctx);
3489 	pi->flags |= PORT_SYSCTL_CTX;
3490 
3491 	/*
3492 	 * Allocate tx/rx/fl queues for this port.
3493 	 */
3494 	rc = t4_setup_port_queues(pi);
3495 	if (rc != 0)
3496 		goto done;	/* error message displayed already */
3497 
3498 	/*
3499 	 * Setup RSS for this port.  Save a copy of the RSS table for later use.
3500 	 */
3501 	rss = malloc(pi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
3502 	for (i = 0; i < pi->rss_size;) {
3503 		for_each_rxq(pi, j, rxq) {
3504 			rss[i++] = rxq->iq.abs_id;
3505 			if (i == pi->rss_size)
3506 				break;
3507 		}
3508 	}
3509 
3510 	rc = -t4_config_rss_range(sc, sc->mbox, pi->viid, 0, pi->rss_size, rss,
3511 	    pi->rss_size);
3512 	if (rc != 0) {
3513 		if_printf(ifp, "rss_config failed: %d\n", rc);
3514 		goto done;
3515 	}
3516 
3517 	pi->rss = rss;
3518 	pi->flags |= PORT_INIT_DONE;
3519 done:
3520 	if (rc != 0)
3521 		port_full_uninit(pi);
3522 
3523 	return (rc);
3524 }
3525 
3526 /*
3527  * Idempotent.
3528  */
3529 int
3530 port_full_uninit(struct port_info *pi)
3531 {
3532 	struct adapter *sc = pi->adapter;
3533 	int i;
3534 	struct sge_rxq *rxq;
3535 	struct sge_txq *txq;
3536 #ifdef TCP_OFFLOAD
3537 	struct sge_ofld_rxq *ofld_rxq;
3538 	struct sge_wrq *ofld_txq;
3539 #endif
3540 
3541 	if (pi->flags & PORT_INIT_DONE) {
3542 
3543 		/* Need to quiesce queues.  */
3544 
3545 		quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
3546 
3547 		for_each_txq(pi, i, txq) {
3548 			quiesce_txq(sc, txq);
3549 		}
3550 
3551 #ifdef TCP_OFFLOAD
3552 		for_each_ofld_txq(pi, i, ofld_txq) {
3553 			quiesce_wrq(sc, ofld_txq);
3554 		}
3555 #endif
3556 
3557 		for_each_rxq(pi, i, rxq) {
3558 			quiesce_iq(sc, &rxq->iq);
3559 			quiesce_fl(sc, &rxq->fl);
3560 		}
3561 
3562 #ifdef TCP_OFFLOAD
3563 		for_each_ofld_rxq(pi, i, ofld_rxq) {
3564 			quiesce_iq(sc, &ofld_rxq->iq);
3565 			quiesce_fl(sc, &ofld_rxq->fl);
3566 		}
3567 #endif
3568 		free(pi->rss, M_CXGBE);
3569 	}
3570 
3571 	t4_teardown_port_queues(pi);
3572 	pi->flags &= ~PORT_INIT_DONE;
3573 
3574 	return (0);
3575 }
3576 
3577 static void
3578 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
3579 {
3580 	struct sge_eq *eq = &txq->eq;
3581 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
3582 
3583 	(void) sc;	/* unused */
3584 
3585 #ifdef INVARIANTS
3586 	TXQ_LOCK(txq);
3587 	MPASS((eq->flags & EQ_ENABLED) == 0);
3588 	TXQ_UNLOCK(txq);
3589 #endif
3590 
3591 	/* Wait for the mp_ring to empty. */
3592 	while (!mp_ring_is_idle(txq->r)) {
3593 		mp_ring_check_drainage(txq->r, 0);
3594 		pause("rquiesce", 1);
3595 	}
3596 
3597 	/* Then wait for the hardware to finish. */
3598 	while (spg->cidx != htobe16(eq->pidx))
3599 		pause("equiesce", 1);
3600 
3601 	/* Finally, wait for the driver to reclaim all descriptors. */
3602 	while (eq->cidx != eq->pidx)
3603 		pause("dquiesce", 1);
3604 }
3605 
3606 static void
3607 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
3608 {
3609 
3610 	/* XXXTX */
3611 }
3612 
3613 static void
3614 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
3615 {
3616 	(void) sc;	/* unused */
3617 
3618 	/* Synchronize with the interrupt handler */
3619 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
3620 		pause("iqfree", 1);
3621 }
3622 
3623 static void
3624 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
3625 {
3626 	mtx_lock(&sc->sfl_lock);
3627 	FL_LOCK(fl);
3628 	fl->flags |= FL_DOOMED;
3629 	FL_UNLOCK(fl);
3630 	mtx_unlock(&sc->sfl_lock);
3631 
3632 	callout_drain(&sc->sfl_callout);
3633 	KASSERT((fl->flags & FL_STARVING) == 0,
3634 	    ("%s: still starving", __func__));
3635 }
3636 
3637 static int
3638 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
3639     driver_intr_t *handler, void *arg, char *name)
3640 {
3641 	int rc;
3642 
3643 	irq->rid = rid;
3644 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
3645 	    RF_SHAREABLE | RF_ACTIVE);
3646 	if (irq->res == NULL) {
3647 		device_printf(sc->dev,
3648 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
3649 		return (ENOMEM);
3650 	}
3651 
3652 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
3653 	    NULL, handler, arg, &irq->tag);
3654 	if (rc != 0) {
3655 		device_printf(sc->dev,
3656 		    "failed to setup interrupt for rid %d, name %s: %d\n",
3657 		    rid, name, rc);
3658 	} else if (name)
3659 		bus_describe_intr(sc->dev, irq->res, irq->tag, name);
3660 
3661 	return (rc);
3662 }
3663 
3664 static int
3665 t4_free_irq(struct adapter *sc, struct irq *irq)
3666 {
3667 	if (irq->tag)
3668 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
3669 	if (irq->res)
3670 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
3671 
3672 	bzero(irq, sizeof(*irq));
3673 
3674 	return (0);
3675 }
3676 
3677 static void
3678 reg_block_dump(struct adapter *sc, uint8_t *buf, unsigned int start,
3679     unsigned int end)
3680 {
3681 	uint32_t *p = (uint32_t *)(buf + start);
3682 
3683 	for ( ; start <= end; start += sizeof(uint32_t))
3684 		*p++ = t4_read_reg(sc, start);
3685 }
3686 
3687 static void
3688 t4_get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
3689 {
3690 	int i, n;
3691 	const unsigned int *reg_ranges;
3692 	static const unsigned int t4_reg_ranges[] = {
3693 		0x1008, 0x1108,
3694 		0x1180, 0x11b4,
3695 		0x11fc, 0x123c,
3696 		0x1300, 0x173c,
3697 		0x1800, 0x18fc,
3698 		0x3000, 0x30d8,
3699 		0x30e0, 0x5924,
3700 		0x5960, 0x59d4,
3701 		0x5a00, 0x5af8,
3702 		0x6000, 0x6098,
3703 		0x6100, 0x6150,
3704 		0x6200, 0x6208,
3705 		0x6240, 0x6248,
3706 		0x6280, 0x6338,
3707 		0x6370, 0x638c,
3708 		0x6400, 0x643c,
3709 		0x6500, 0x6524,
3710 		0x6a00, 0x6a38,
3711 		0x6a60, 0x6a78,
3712 		0x6b00, 0x6b84,
3713 		0x6bf0, 0x6c84,
3714 		0x6cf0, 0x6d84,
3715 		0x6df0, 0x6e84,
3716 		0x6ef0, 0x6f84,
3717 		0x6ff0, 0x7084,
3718 		0x70f0, 0x7184,
3719 		0x71f0, 0x7284,
3720 		0x72f0, 0x7384,
3721 		0x73f0, 0x7450,
3722 		0x7500, 0x7530,
3723 		0x7600, 0x761c,
3724 		0x7680, 0x76cc,
3725 		0x7700, 0x7798,
3726 		0x77c0, 0x77fc,
3727 		0x7900, 0x79fc,
3728 		0x7b00, 0x7c38,
3729 		0x7d00, 0x7efc,
3730 		0x8dc0, 0x8e1c,
3731 		0x8e30, 0x8e78,
3732 		0x8ea0, 0x8f6c,
3733 		0x8fc0, 0x9074,
3734 		0x90fc, 0x90fc,
3735 		0x9400, 0x9458,
3736 		0x9600, 0x96bc,
3737 		0x9800, 0x9808,
3738 		0x9820, 0x983c,
3739 		0x9850, 0x9864,
3740 		0x9c00, 0x9c6c,
3741 		0x9c80, 0x9cec,
3742 		0x9d00, 0x9d6c,
3743 		0x9d80, 0x9dec,
3744 		0x9e00, 0x9e6c,
3745 		0x9e80, 0x9eec,
3746 		0x9f00, 0x9f6c,
3747 		0x9f80, 0x9fec,
3748 		0xd004, 0xd03c,
3749 		0xdfc0, 0xdfe0,
3750 		0xe000, 0xea7c,
3751 		0xf000, 0x11110,
3752 		0x11118, 0x11190,
3753 		0x19040, 0x1906c,
3754 		0x19078, 0x19080,
3755 		0x1908c, 0x19124,
3756 		0x19150, 0x191b0,
3757 		0x191d0, 0x191e8,
3758 		0x19238, 0x1924c,
3759 		0x193f8, 0x19474,
3760 		0x19490, 0x194f8,
3761 		0x19800, 0x19f30,
3762 		0x1a000, 0x1a06c,
3763 		0x1a0b0, 0x1a120,
3764 		0x1a128, 0x1a138,
3765 		0x1a190, 0x1a1c4,
3766 		0x1a1fc, 0x1a1fc,
3767 		0x1e040, 0x1e04c,
3768 		0x1e284, 0x1e28c,
3769 		0x1e2c0, 0x1e2c0,
3770 		0x1e2e0, 0x1e2e0,
3771 		0x1e300, 0x1e384,
3772 		0x1e3c0, 0x1e3c8,
3773 		0x1e440, 0x1e44c,
3774 		0x1e684, 0x1e68c,
3775 		0x1e6c0, 0x1e6c0,
3776 		0x1e6e0, 0x1e6e0,
3777 		0x1e700, 0x1e784,
3778 		0x1e7c0, 0x1e7c8,
3779 		0x1e840, 0x1e84c,
3780 		0x1ea84, 0x1ea8c,
3781 		0x1eac0, 0x1eac0,
3782 		0x1eae0, 0x1eae0,
3783 		0x1eb00, 0x1eb84,
3784 		0x1ebc0, 0x1ebc8,
3785 		0x1ec40, 0x1ec4c,
3786 		0x1ee84, 0x1ee8c,
3787 		0x1eec0, 0x1eec0,
3788 		0x1eee0, 0x1eee0,
3789 		0x1ef00, 0x1ef84,
3790 		0x1efc0, 0x1efc8,
3791 		0x1f040, 0x1f04c,
3792 		0x1f284, 0x1f28c,
3793 		0x1f2c0, 0x1f2c0,
3794 		0x1f2e0, 0x1f2e0,
3795 		0x1f300, 0x1f384,
3796 		0x1f3c0, 0x1f3c8,
3797 		0x1f440, 0x1f44c,
3798 		0x1f684, 0x1f68c,
3799 		0x1f6c0, 0x1f6c0,
3800 		0x1f6e0, 0x1f6e0,
3801 		0x1f700, 0x1f784,
3802 		0x1f7c0, 0x1f7c8,
3803 		0x1f840, 0x1f84c,
3804 		0x1fa84, 0x1fa8c,
3805 		0x1fac0, 0x1fac0,
3806 		0x1fae0, 0x1fae0,
3807 		0x1fb00, 0x1fb84,
3808 		0x1fbc0, 0x1fbc8,
3809 		0x1fc40, 0x1fc4c,
3810 		0x1fe84, 0x1fe8c,
3811 		0x1fec0, 0x1fec0,
3812 		0x1fee0, 0x1fee0,
3813 		0x1ff00, 0x1ff84,
3814 		0x1ffc0, 0x1ffc8,
3815 		0x20000, 0x2002c,
3816 		0x20100, 0x2013c,
3817 		0x20190, 0x201c8,
3818 		0x20200, 0x20318,
3819 		0x20400, 0x20528,
3820 		0x20540, 0x20614,
3821 		0x21000, 0x21040,
3822 		0x2104c, 0x21060,
3823 		0x210c0, 0x210ec,
3824 		0x21200, 0x21268,
3825 		0x21270, 0x21284,
3826 		0x212fc, 0x21388,
3827 		0x21400, 0x21404,
3828 		0x21500, 0x21518,
3829 		0x2152c, 0x2153c,
3830 		0x21550, 0x21554,
3831 		0x21600, 0x21600,
3832 		0x21608, 0x21628,
3833 		0x21630, 0x2163c,
3834 		0x21700, 0x2171c,
3835 		0x21780, 0x2178c,
3836 		0x21800, 0x21c38,
3837 		0x21c80, 0x21d7c,
3838 		0x21e00, 0x21e04,
3839 		0x22000, 0x2202c,
3840 		0x22100, 0x2213c,
3841 		0x22190, 0x221c8,
3842 		0x22200, 0x22318,
3843 		0x22400, 0x22528,
3844 		0x22540, 0x22614,
3845 		0x23000, 0x23040,
3846 		0x2304c, 0x23060,
3847 		0x230c0, 0x230ec,
3848 		0x23200, 0x23268,
3849 		0x23270, 0x23284,
3850 		0x232fc, 0x23388,
3851 		0x23400, 0x23404,
3852 		0x23500, 0x23518,
3853 		0x2352c, 0x2353c,
3854 		0x23550, 0x23554,
3855 		0x23600, 0x23600,
3856 		0x23608, 0x23628,
3857 		0x23630, 0x2363c,
3858 		0x23700, 0x2371c,
3859 		0x23780, 0x2378c,
3860 		0x23800, 0x23c38,
3861 		0x23c80, 0x23d7c,
3862 		0x23e00, 0x23e04,
3863 		0x24000, 0x2402c,
3864 		0x24100, 0x2413c,
3865 		0x24190, 0x241c8,
3866 		0x24200, 0x24318,
3867 		0x24400, 0x24528,
3868 		0x24540, 0x24614,
3869 		0x25000, 0x25040,
3870 		0x2504c, 0x25060,
3871 		0x250c0, 0x250ec,
3872 		0x25200, 0x25268,
3873 		0x25270, 0x25284,
3874 		0x252fc, 0x25388,
3875 		0x25400, 0x25404,
3876 		0x25500, 0x25518,
3877 		0x2552c, 0x2553c,
3878 		0x25550, 0x25554,
3879 		0x25600, 0x25600,
3880 		0x25608, 0x25628,
3881 		0x25630, 0x2563c,
3882 		0x25700, 0x2571c,
3883 		0x25780, 0x2578c,
3884 		0x25800, 0x25c38,
3885 		0x25c80, 0x25d7c,
3886 		0x25e00, 0x25e04,
3887 		0x26000, 0x2602c,
3888 		0x26100, 0x2613c,
3889 		0x26190, 0x261c8,
3890 		0x26200, 0x26318,
3891 		0x26400, 0x26528,
3892 		0x26540, 0x26614,
3893 		0x27000, 0x27040,
3894 		0x2704c, 0x27060,
3895 		0x270c0, 0x270ec,
3896 		0x27200, 0x27268,
3897 		0x27270, 0x27284,
3898 		0x272fc, 0x27388,
3899 		0x27400, 0x27404,
3900 		0x27500, 0x27518,
3901 		0x2752c, 0x2753c,
3902 		0x27550, 0x27554,
3903 		0x27600, 0x27600,
3904 		0x27608, 0x27628,
3905 		0x27630, 0x2763c,
3906 		0x27700, 0x2771c,
3907 		0x27780, 0x2778c,
3908 		0x27800, 0x27c38,
3909 		0x27c80, 0x27d7c,
3910 		0x27e00, 0x27e04
3911 	};
3912 	static const unsigned int t5_reg_ranges[] = {
3913 		0x1008, 0x1148,
3914 		0x1180, 0x11b4,
3915 		0x11fc, 0x123c,
3916 		0x1280, 0x173c,
3917 		0x1800, 0x18fc,
3918 		0x3000, 0x3028,
3919 		0x3060, 0x30d8,
3920 		0x30e0, 0x30fc,
3921 		0x3140, 0x357c,
3922 		0x35a8, 0x35cc,
3923 		0x35ec, 0x35ec,
3924 		0x3600, 0x5624,
3925 		0x56cc, 0x575c,
3926 		0x580c, 0x5814,
3927 		0x5890, 0x58bc,
3928 		0x5940, 0x59dc,
3929 		0x59fc, 0x5a18,
3930 		0x5a60, 0x5a9c,
3931 		0x5b94, 0x5bfc,
3932 		0x6000, 0x6040,
3933 		0x6058, 0x614c,
3934 		0x7700, 0x7798,
3935 		0x77c0, 0x78fc,
3936 		0x7b00, 0x7c54,
3937 		0x7d00, 0x7efc,
3938 		0x8dc0, 0x8de0,
3939 		0x8df8, 0x8e84,
3940 		0x8ea0, 0x8f84,
3941 		0x8fc0, 0x90f8,
3942 		0x9400, 0x9470,
3943 		0x9600, 0x96f4,
3944 		0x9800, 0x9808,
3945 		0x9820, 0x983c,
3946 		0x9850, 0x9864,
3947 		0x9c00, 0x9c6c,
3948 		0x9c80, 0x9cec,
3949 		0x9d00, 0x9d6c,
3950 		0x9d80, 0x9dec,
3951 		0x9e00, 0x9e6c,
3952 		0x9e80, 0x9eec,
3953 		0x9f00, 0x9f6c,
3954 		0x9f80, 0xa020,
3955 		0xd004, 0xd03c,
3956 		0xdfc0, 0xdfe0,
3957 		0xe000, 0x11088,
3958 		0x1109c, 0x11110,
3959 		0x11118, 0x1117c,
3960 		0x11190, 0x11204,
3961 		0x19040, 0x1906c,
3962 		0x19078, 0x19080,
3963 		0x1908c, 0x19124,
3964 		0x19150, 0x191b0,
3965 		0x191d0, 0x191e8,
3966 		0x19238, 0x19290,
3967 		0x193f8, 0x19474,
3968 		0x19490, 0x194cc,
3969 		0x194f0, 0x194f8,
3970 		0x19c00, 0x19c60,
3971 		0x19c94, 0x19e10,
3972 		0x19e50, 0x19f34,
3973 		0x19f40, 0x19f50,
3974 		0x19f90, 0x19fe4,
3975 		0x1a000, 0x1a06c,
3976 		0x1a0b0, 0x1a120,
3977 		0x1a128, 0x1a138,
3978 		0x1a190, 0x1a1c4,
3979 		0x1a1fc, 0x1a1fc,
3980 		0x1e008, 0x1e00c,
3981 		0x1e040, 0x1e04c,
3982 		0x1e284, 0x1e290,
3983 		0x1e2c0, 0x1e2c0,
3984 		0x1e2e0, 0x1e2e0,
3985 		0x1e300, 0x1e384,
3986 		0x1e3c0, 0x1e3c8,
3987 		0x1e408, 0x1e40c,
3988 		0x1e440, 0x1e44c,
3989 		0x1e684, 0x1e690,
3990 		0x1e6c0, 0x1e6c0,
3991 		0x1e6e0, 0x1e6e0,
3992 		0x1e700, 0x1e784,
3993 		0x1e7c0, 0x1e7c8,
3994 		0x1e808, 0x1e80c,
3995 		0x1e840, 0x1e84c,
3996 		0x1ea84, 0x1ea90,
3997 		0x1eac0, 0x1eac0,
3998 		0x1eae0, 0x1eae0,
3999 		0x1eb00, 0x1eb84,
4000 		0x1ebc0, 0x1ebc8,
4001 		0x1ec08, 0x1ec0c,
4002 		0x1ec40, 0x1ec4c,
4003 		0x1ee84, 0x1ee90,
4004 		0x1eec0, 0x1eec0,
4005 		0x1eee0, 0x1eee0,
4006 		0x1ef00, 0x1ef84,
4007 		0x1efc0, 0x1efc8,
4008 		0x1f008, 0x1f00c,
4009 		0x1f040, 0x1f04c,
4010 		0x1f284, 0x1f290,
4011 		0x1f2c0, 0x1f2c0,
4012 		0x1f2e0, 0x1f2e0,
4013 		0x1f300, 0x1f384,
4014 		0x1f3c0, 0x1f3c8,
4015 		0x1f408, 0x1f40c,
4016 		0x1f440, 0x1f44c,
4017 		0x1f684, 0x1f690,
4018 		0x1f6c0, 0x1f6c0,
4019 		0x1f6e0, 0x1f6e0,
4020 		0x1f700, 0x1f784,
4021 		0x1f7c0, 0x1f7c8,
4022 		0x1f808, 0x1f80c,
4023 		0x1f840, 0x1f84c,
4024 		0x1fa84, 0x1fa90,
4025 		0x1fac0, 0x1fac0,
4026 		0x1fae0, 0x1fae0,
4027 		0x1fb00, 0x1fb84,
4028 		0x1fbc0, 0x1fbc8,
4029 		0x1fc08, 0x1fc0c,
4030 		0x1fc40, 0x1fc4c,
4031 		0x1fe84, 0x1fe90,
4032 		0x1fec0, 0x1fec0,
4033 		0x1fee0, 0x1fee0,
4034 		0x1ff00, 0x1ff84,
4035 		0x1ffc0, 0x1ffc8,
4036 		0x30000, 0x30030,
4037 		0x30100, 0x30144,
4038 		0x30190, 0x301d0,
4039 		0x30200, 0x30318,
4040 		0x30400, 0x3052c,
4041 		0x30540, 0x3061c,
4042 		0x30800, 0x30834,
4043 		0x308c0, 0x30908,
4044 		0x30910, 0x309ac,
4045 		0x30a00, 0x30a2c,
4046 		0x30a44, 0x30a50,
4047 		0x30a74, 0x30c24,
4048 		0x30d00, 0x30d00,
4049 		0x30d08, 0x30d14,
4050 		0x30d1c, 0x30d20,
4051 		0x30d3c, 0x30d50,
4052 		0x31200, 0x3120c,
4053 		0x31220, 0x31220,
4054 		0x31240, 0x31240,
4055 		0x31600, 0x3160c,
4056 		0x31a00, 0x31a1c,
4057 		0x31e00, 0x31e20,
4058 		0x31e38, 0x31e3c,
4059 		0x31e80, 0x31e80,
4060 		0x31e88, 0x31ea8,
4061 		0x31eb0, 0x31eb4,
4062 		0x31ec8, 0x31ed4,
4063 		0x31fb8, 0x32004,
4064 		0x32200, 0x32200,
4065 		0x32208, 0x32240,
4066 		0x32248, 0x32280,
4067 		0x32288, 0x322c0,
4068 		0x322c8, 0x322fc,
4069 		0x32600, 0x32630,
4070 		0x32a00, 0x32abc,
4071 		0x32b00, 0x32b70,
4072 		0x33000, 0x33048,
4073 		0x33060, 0x3309c,
4074 		0x330f0, 0x33148,
4075 		0x33160, 0x3319c,
4076 		0x331f0, 0x332e4,
4077 		0x332f8, 0x333e4,
4078 		0x333f8, 0x33448,
4079 		0x33460, 0x3349c,
4080 		0x334f0, 0x33548,
4081 		0x33560, 0x3359c,
4082 		0x335f0, 0x336e4,
4083 		0x336f8, 0x337e4,
4084 		0x337f8, 0x337fc,
4085 		0x33814, 0x33814,
4086 		0x3382c, 0x3382c,
4087 		0x33880, 0x3388c,
4088 		0x338e8, 0x338ec,
4089 		0x33900, 0x33948,
4090 		0x33960, 0x3399c,
4091 		0x339f0, 0x33ae4,
4092 		0x33af8, 0x33b10,
4093 		0x33b28, 0x33b28,
4094 		0x33b3c, 0x33b50,
4095 		0x33bf0, 0x33c10,
4096 		0x33c28, 0x33c28,
4097 		0x33c3c, 0x33c50,
4098 		0x33cf0, 0x33cfc,
4099 		0x34000, 0x34030,
4100 		0x34100, 0x34144,
4101 		0x34190, 0x341d0,
4102 		0x34200, 0x34318,
4103 		0x34400, 0x3452c,
4104 		0x34540, 0x3461c,
4105 		0x34800, 0x34834,
4106 		0x348c0, 0x34908,
4107 		0x34910, 0x349ac,
4108 		0x34a00, 0x34a2c,
4109 		0x34a44, 0x34a50,
4110 		0x34a74, 0x34c24,
4111 		0x34d00, 0x34d00,
4112 		0x34d08, 0x34d14,
4113 		0x34d1c, 0x34d20,
4114 		0x34d3c, 0x34d50,
4115 		0x35200, 0x3520c,
4116 		0x35220, 0x35220,
4117 		0x35240, 0x35240,
4118 		0x35600, 0x3560c,
4119 		0x35a00, 0x35a1c,
4120 		0x35e00, 0x35e20,
4121 		0x35e38, 0x35e3c,
4122 		0x35e80, 0x35e80,
4123 		0x35e88, 0x35ea8,
4124 		0x35eb0, 0x35eb4,
4125 		0x35ec8, 0x35ed4,
4126 		0x35fb8, 0x36004,
4127 		0x36200, 0x36200,
4128 		0x36208, 0x36240,
4129 		0x36248, 0x36280,
4130 		0x36288, 0x362c0,
4131 		0x362c8, 0x362fc,
4132 		0x36600, 0x36630,
4133 		0x36a00, 0x36abc,
4134 		0x36b00, 0x36b70,
4135 		0x37000, 0x37048,
4136 		0x37060, 0x3709c,
4137 		0x370f0, 0x37148,
4138 		0x37160, 0x3719c,
4139 		0x371f0, 0x372e4,
4140 		0x372f8, 0x373e4,
4141 		0x373f8, 0x37448,
4142 		0x37460, 0x3749c,
4143 		0x374f0, 0x37548,
4144 		0x37560, 0x3759c,
4145 		0x375f0, 0x376e4,
4146 		0x376f8, 0x377e4,
4147 		0x377f8, 0x377fc,
4148 		0x37814, 0x37814,
4149 		0x3782c, 0x3782c,
4150 		0x37880, 0x3788c,
4151 		0x378e8, 0x378ec,
4152 		0x37900, 0x37948,
4153 		0x37960, 0x3799c,
4154 		0x379f0, 0x37ae4,
4155 		0x37af8, 0x37b10,
4156 		0x37b28, 0x37b28,
4157 		0x37b3c, 0x37b50,
4158 		0x37bf0, 0x37c10,
4159 		0x37c28, 0x37c28,
4160 		0x37c3c, 0x37c50,
4161 		0x37cf0, 0x37cfc,
4162 		0x38000, 0x38030,
4163 		0x38100, 0x38144,
4164 		0x38190, 0x381d0,
4165 		0x38200, 0x38318,
4166 		0x38400, 0x3852c,
4167 		0x38540, 0x3861c,
4168 		0x38800, 0x38834,
4169 		0x388c0, 0x38908,
4170 		0x38910, 0x389ac,
4171 		0x38a00, 0x38a2c,
4172 		0x38a44, 0x38a50,
4173 		0x38a74, 0x38c24,
4174 		0x38d00, 0x38d00,
4175 		0x38d08, 0x38d14,
4176 		0x38d1c, 0x38d20,
4177 		0x38d3c, 0x38d50,
4178 		0x39200, 0x3920c,
4179 		0x39220, 0x39220,
4180 		0x39240, 0x39240,
4181 		0x39600, 0x3960c,
4182 		0x39a00, 0x39a1c,
4183 		0x39e00, 0x39e20,
4184 		0x39e38, 0x39e3c,
4185 		0x39e80, 0x39e80,
4186 		0x39e88, 0x39ea8,
4187 		0x39eb0, 0x39eb4,
4188 		0x39ec8, 0x39ed4,
4189 		0x39fb8, 0x3a004,
4190 		0x3a200, 0x3a200,
4191 		0x3a208, 0x3a240,
4192 		0x3a248, 0x3a280,
4193 		0x3a288, 0x3a2c0,
4194 		0x3a2c8, 0x3a2fc,
4195 		0x3a600, 0x3a630,
4196 		0x3aa00, 0x3aabc,
4197 		0x3ab00, 0x3ab70,
4198 		0x3b000, 0x3b048,
4199 		0x3b060, 0x3b09c,
4200 		0x3b0f0, 0x3b148,
4201 		0x3b160, 0x3b19c,
4202 		0x3b1f0, 0x3b2e4,
4203 		0x3b2f8, 0x3b3e4,
4204 		0x3b3f8, 0x3b448,
4205 		0x3b460, 0x3b49c,
4206 		0x3b4f0, 0x3b548,
4207 		0x3b560, 0x3b59c,
4208 		0x3b5f0, 0x3b6e4,
4209 		0x3b6f8, 0x3b7e4,
4210 		0x3b7f8, 0x3b7fc,
4211 		0x3b814, 0x3b814,
4212 		0x3b82c, 0x3b82c,
4213 		0x3b880, 0x3b88c,
4214 		0x3b8e8, 0x3b8ec,
4215 		0x3b900, 0x3b948,
4216 		0x3b960, 0x3b99c,
4217 		0x3b9f0, 0x3bae4,
4218 		0x3baf8, 0x3bb10,
4219 		0x3bb28, 0x3bb28,
4220 		0x3bb3c, 0x3bb50,
4221 		0x3bbf0, 0x3bc10,
4222 		0x3bc28, 0x3bc28,
4223 		0x3bc3c, 0x3bc50,
4224 		0x3bcf0, 0x3bcfc,
4225 		0x3c000, 0x3c030,
4226 		0x3c100, 0x3c144,
4227 		0x3c190, 0x3c1d0,
4228 		0x3c200, 0x3c318,
4229 		0x3c400, 0x3c52c,
4230 		0x3c540, 0x3c61c,
4231 		0x3c800, 0x3c834,
4232 		0x3c8c0, 0x3c908,
4233 		0x3c910, 0x3c9ac,
4234 		0x3ca00, 0x3ca2c,
4235 		0x3ca44, 0x3ca50,
4236 		0x3ca74, 0x3cc24,
4237 		0x3cd00, 0x3cd00,
4238 		0x3cd08, 0x3cd14,
4239 		0x3cd1c, 0x3cd20,
4240 		0x3cd3c, 0x3cd50,
4241 		0x3d200, 0x3d20c,
4242 		0x3d220, 0x3d220,
4243 		0x3d240, 0x3d240,
4244 		0x3d600, 0x3d60c,
4245 		0x3da00, 0x3da1c,
4246 		0x3de00, 0x3de20,
4247 		0x3de38, 0x3de3c,
4248 		0x3de80, 0x3de80,
4249 		0x3de88, 0x3dea8,
4250 		0x3deb0, 0x3deb4,
4251 		0x3dec8, 0x3ded4,
4252 		0x3dfb8, 0x3e004,
4253 		0x3e200, 0x3e200,
4254 		0x3e208, 0x3e240,
4255 		0x3e248, 0x3e280,
4256 		0x3e288, 0x3e2c0,
4257 		0x3e2c8, 0x3e2fc,
4258 		0x3e600, 0x3e630,
4259 		0x3ea00, 0x3eabc,
4260 		0x3eb00, 0x3eb70,
4261 		0x3f000, 0x3f048,
4262 		0x3f060, 0x3f09c,
4263 		0x3f0f0, 0x3f148,
4264 		0x3f160, 0x3f19c,
4265 		0x3f1f0, 0x3f2e4,
4266 		0x3f2f8, 0x3f3e4,
4267 		0x3f3f8, 0x3f448,
4268 		0x3f460, 0x3f49c,
4269 		0x3f4f0, 0x3f548,
4270 		0x3f560, 0x3f59c,
4271 		0x3f5f0, 0x3f6e4,
4272 		0x3f6f8, 0x3f7e4,
4273 		0x3f7f8, 0x3f7fc,
4274 		0x3f814, 0x3f814,
4275 		0x3f82c, 0x3f82c,
4276 		0x3f880, 0x3f88c,
4277 		0x3f8e8, 0x3f8ec,
4278 		0x3f900, 0x3f948,
4279 		0x3f960, 0x3f99c,
4280 		0x3f9f0, 0x3fae4,
4281 		0x3faf8, 0x3fb10,
4282 		0x3fb28, 0x3fb28,
4283 		0x3fb3c, 0x3fb50,
4284 		0x3fbf0, 0x3fc10,
4285 		0x3fc28, 0x3fc28,
4286 		0x3fc3c, 0x3fc50,
4287 		0x3fcf0, 0x3fcfc,
4288 		0x40000, 0x4000c,
4289 		0x40040, 0x40068,
4290 		0x4007c, 0x40144,
4291 		0x40180, 0x4018c,
4292 		0x40200, 0x40298,
4293 		0x402ac, 0x4033c,
4294 		0x403f8, 0x403fc,
4295 		0x41304, 0x413c4,
4296 		0x41400, 0x4141c,
4297 		0x41480, 0x414d0,
4298 		0x44000, 0x44078,
4299 		0x440c0, 0x44278,
4300 		0x442c0, 0x44478,
4301 		0x444c0, 0x44678,
4302 		0x446c0, 0x44878,
4303 		0x448c0, 0x449fc,
4304 		0x45000, 0x45068,
4305 		0x45080, 0x45084,
4306 		0x450a0, 0x450b0,
4307 		0x45200, 0x45268,
4308 		0x45280, 0x45284,
4309 		0x452a0, 0x452b0,
4310 		0x460c0, 0x460e4,
4311 		0x47000, 0x4708c,
4312 		0x47200, 0x47250,
4313 		0x47400, 0x47420,
4314 		0x47600, 0x47618,
4315 		0x47800, 0x47814,
4316 		0x48000, 0x4800c,
4317 		0x48040, 0x48068,
4318 		0x4807c, 0x48144,
4319 		0x48180, 0x4818c,
4320 		0x48200, 0x48298,
4321 		0x482ac, 0x4833c,
4322 		0x483f8, 0x483fc,
4323 		0x49304, 0x493c4,
4324 		0x49400, 0x4941c,
4325 		0x49480, 0x494d0,
4326 		0x4c000, 0x4c078,
4327 		0x4c0c0, 0x4c278,
4328 		0x4c2c0, 0x4c478,
4329 		0x4c4c0, 0x4c678,
4330 		0x4c6c0, 0x4c878,
4331 		0x4c8c0, 0x4c9fc,
4332 		0x4d000, 0x4d068,
4333 		0x4d080, 0x4d084,
4334 		0x4d0a0, 0x4d0b0,
4335 		0x4d200, 0x4d268,
4336 		0x4d280, 0x4d284,
4337 		0x4d2a0, 0x4d2b0,
4338 		0x4e0c0, 0x4e0e4,
4339 		0x4f000, 0x4f08c,
4340 		0x4f200, 0x4f250,
4341 		0x4f400, 0x4f420,
4342 		0x4f600, 0x4f618,
4343 		0x4f800, 0x4f814,
4344 		0x50000, 0x500cc,
4345 		0x50400, 0x50400,
4346 		0x50800, 0x508cc,
4347 		0x50c00, 0x50c00,
4348 		0x51000, 0x5101c,
4349 		0x51300, 0x51308,
4350 	};
4351 
4352 	if (is_t4(sc)) {
4353 		reg_ranges = &t4_reg_ranges[0];
4354 		n = nitems(t4_reg_ranges);
4355 	} else {
4356 		reg_ranges = &t5_reg_ranges[0];
4357 		n = nitems(t5_reg_ranges);
4358 	}
4359 
4360 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
4361 	for (i = 0; i < n; i += 2)
4362 		reg_block_dump(sc, buf, reg_ranges[i], reg_ranges[i + 1]);
4363 }
4364 
4365 static void
4366 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
4367 {
4368 	int i;
4369 	u_int v, tnl_cong_drops;
4370 	struct timeval tv;
4371 	const struct timeval interval = {0, 250000};	/* 250ms */
4372 
4373 	getmicrotime(&tv);
4374 	timevalsub(&tv, &interval);
4375 	if (timevalcmp(&tv, &pi->last_refreshed, <))
4376 		return;
4377 
4378 	tnl_cong_drops = 0;
4379 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
4380 	for (i = 0; i < NCHAN; i++) {
4381 		if (pi->rx_chan_map & (1 << i)) {
4382 			mtx_lock(&sc->regwin_lock);
4383 			t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
4384 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
4385 			mtx_unlock(&sc->regwin_lock);
4386 			tnl_cong_drops += v;
4387 		}
4388 	}
4389 	pi->tnl_cong_drops = tnl_cong_drops;
4390 	getmicrotime(&pi->last_refreshed);
4391 }
4392 
4393 static void
4394 cxgbe_tick(void *arg)
4395 {
4396 	struct port_info *pi = arg;
4397 	struct adapter *sc = pi->adapter;
4398 	struct ifnet *ifp = pi->ifp;
4399 
4400 	PORT_LOCK(pi);
4401 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4402 		PORT_UNLOCK(pi);
4403 		return;	/* without scheduling another callout */
4404 	}
4405 
4406 	cxgbe_refresh_stats(sc, pi);
4407 
4408 	callout_schedule(&pi->tick, hz);
4409 	PORT_UNLOCK(pi);
4410 }
4411 
4412 static void
4413 cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
4414 {
4415 	struct ifnet *vlan;
4416 
4417 	if (arg != ifp || ifp->if_type != IFT_ETHER)
4418 		return;
4419 
4420 	vlan = VLAN_DEVAT(ifp, vid);
4421 	VLAN_SETCOOKIE(vlan, ifp);
4422 }
4423 
4424 static int
4425 cpl_not_handled(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
4426 {
4427 
4428 #ifdef INVARIANTS
4429 	panic("%s: opcode 0x%02x on iq %p with payload %p",
4430 	    __func__, rss->opcode, iq, m);
4431 #else
4432 	log(LOG_ERR, "%s: opcode 0x%02x on iq %p with payload %p\n",
4433 	    __func__, rss->opcode, iq, m);
4434 	m_freem(m);
4435 #endif
4436 	return (EDOOFUS);
4437 }
4438 
4439 int
4440 t4_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h)
4441 {
4442 	uintptr_t *loc, new;
4443 
4444 	if (opcode >= nitems(sc->cpl_handler))
4445 		return (EINVAL);
4446 
4447 	new = h ? (uintptr_t)h : (uintptr_t)cpl_not_handled;
4448 	loc = (uintptr_t *) &sc->cpl_handler[opcode];
4449 	atomic_store_rel_ptr(loc, new);
4450 
4451 	return (0);
4452 }
4453 
4454 static int
4455 an_not_handled(struct sge_iq *iq, const struct rsp_ctrl *ctrl)
4456 {
4457 
4458 #ifdef INVARIANTS
4459 	panic("%s: async notification on iq %p (ctrl %p)", __func__, iq, ctrl);
4460 #else
4461 	log(LOG_ERR, "%s: async notification on iq %p (ctrl %p)\n",
4462 	    __func__, iq, ctrl);
4463 #endif
4464 	return (EDOOFUS);
4465 }
4466 
4467 int
4468 t4_register_an_handler(struct adapter *sc, an_handler_t h)
4469 {
4470 	uintptr_t *loc, new;
4471 
4472 	new = h ? (uintptr_t)h : (uintptr_t)an_not_handled;
4473 	loc = (uintptr_t *) &sc->an_handler;
4474 	atomic_store_rel_ptr(loc, new);
4475 
4476 	return (0);
4477 }
4478 
4479 static int
4480 fw_msg_not_handled(struct adapter *sc, const __be64 *rpl)
4481 {
4482 	const struct cpl_fw6_msg *cpl =
4483 	    __containerof(rpl, struct cpl_fw6_msg, data[0]);
4484 
4485 #ifdef INVARIANTS
4486 	panic("%s: fw_msg type %d", __func__, cpl->type);
4487 #else
4488 	log(LOG_ERR, "%s: fw_msg type %d\n", __func__, cpl->type);
4489 #endif
4490 	return (EDOOFUS);
4491 }
4492 
4493 int
4494 t4_register_fw_msg_handler(struct adapter *sc, int type, fw_msg_handler_t h)
4495 {
4496 	uintptr_t *loc, new;
4497 
4498 	if (type >= nitems(sc->fw_msg_handler))
4499 		return (EINVAL);
4500 
4501 	/*
4502 	 * These are dispatched by the handler for FW{4|6}_CPL_MSG using the CPL
4503 	 * handler dispatch table.  Reject any attempt to install a handler for
4504 	 * this subtype.
4505 	 */
4506 	if (type == FW_TYPE_RSSCPL || type == FW6_TYPE_RSSCPL)
4507 		return (EINVAL);
4508 
4509 	new = h ? (uintptr_t)h : (uintptr_t)fw_msg_not_handled;
4510 	loc = (uintptr_t *) &sc->fw_msg_handler[type];
4511 	atomic_store_rel_ptr(loc, new);
4512 
4513 	return (0);
4514 }
4515 
4516 static int
4517 t4_sysctls(struct adapter *sc)
4518 {
4519 	struct sysctl_ctx_list *ctx;
4520 	struct sysctl_oid *oid;
4521 	struct sysctl_oid_list *children, *c0;
4522 	static char *caps[] = {
4523 		"\20\1PPP\2QFC\3DCBX",			/* caps[0] linkcaps */
4524 		"\20\1NIC\2VM\3IDS\4UM\5UM_ISGL"	/* caps[1] niccaps */
4525 		    "\6HASHFILTER\7ETHOFLD",
4526 		"\20\1TOE",				/* caps[2] toecaps */
4527 		"\20\1RDDP\2RDMAC",			/* caps[3] rdmacaps */
4528 		"\20\1INITIATOR_PDU\2TARGET_PDU"	/* caps[4] iscsicaps */
4529 		    "\3INITIATOR_CNXOFLD\4TARGET_CNXOFLD"
4530 		    "\5INITIATOR_SSNOFLD\6TARGET_SSNOFLD",
4531 		"\20\1INITIATOR\2TARGET\3CTRL_OFLD"	/* caps[5] fcoecaps */
4532 		    "\4PO_INITIAOR\5PO_TARGET"
4533 	};
4534 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
4535 
4536 	ctx = device_get_sysctl_ctx(sc->dev);
4537 
4538 	/*
4539 	 * dev.t4nex.X.
4540 	 */
4541 	oid = device_get_sysctl_tree(sc->dev);
4542 	c0 = children = SYSCTL_CHILDREN(oid);
4543 
4544 	sc->sc_do_rxcopy = 1;
4545 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
4546 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
4547 
4548 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
4549 	    sc->params.nports, "# of ports");
4550 
4551 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
4552 	    NULL, chip_rev(sc), "chip hardware revision");
4553 
4554 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
4555 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
4556 
4557 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
4558 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
4559 
4560 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
4561 	    sc->cfcsum, "config file checksum");
4562 
4563 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
4564 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
4565 	    sysctl_bitfield, "A", "available doorbells");
4566 
4567 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkcaps",
4568 	    CTLTYPE_STRING | CTLFLAG_RD, caps[0], sc->linkcaps,
4569 	    sysctl_bitfield, "A", "available link capabilities");
4570 
4571 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "niccaps",
4572 	    CTLTYPE_STRING | CTLFLAG_RD, caps[1], sc->niccaps,
4573 	    sysctl_bitfield, "A", "available NIC capabilities");
4574 
4575 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "toecaps",
4576 	    CTLTYPE_STRING | CTLFLAG_RD, caps[2], sc->toecaps,
4577 	    sysctl_bitfield, "A", "available TCP offload capabilities");
4578 
4579 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdmacaps",
4580 	    CTLTYPE_STRING | CTLFLAG_RD, caps[3], sc->rdmacaps,
4581 	    sysctl_bitfield, "A", "available RDMA capabilities");
4582 
4583 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "iscsicaps",
4584 	    CTLTYPE_STRING | CTLFLAG_RD, caps[4], sc->iscsicaps,
4585 	    sysctl_bitfield, "A", "available iSCSI capabilities");
4586 
4587 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoecaps",
4588 	    CTLTYPE_STRING | CTLFLAG_RD, caps[5], sc->fcoecaps,
4589 	    sysctl_bitfield, "A", "available FCoE capabilities");
4590 
4591 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
4592 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
4593 
4594 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
4595 	    CTLTYPE_STRING | CTLFLAG_RD, sc->sge.timer_val,
4596 	    sizeof(sc->sge.timer_val), sysctl_int_array, "A",
4597 	    "interrupt holdoff timer values (us)");
4598 
4599 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
4600 	    CTLTYPE_STRING | CTLFLAG_RD, sc->sge.counter_val,
4601 	    sizeof(sc->sge.counter_val), sysctl_int_array, "A",
4602 	    "interrupt holdoff packet counter values");
4603 
4604 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
4605 	    NULL, sc->tids.nftids, "number of filters");
4606 
4607 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
4608 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
4609 	    "chip temperature (in Celsius)");
4610 
4611 	t4_sge_sysctls(sc, ctx, children);
4612 
4613 	sc->lro_timeout = 100;
4614 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
4615 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
4616 
4617 #ifdef SBUF_DRAIN
4618 	/*
4619 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
4620 	 */
4621 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
4622 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
4623 	    "logs and miscellaneous information");
4624 	children = SYSCTL_CHILDREN(oid);
4625 
4626 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
4627 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4628 	    sysctl_cctrl, "A", "congestion control");
4629 
4630 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
4631 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4632 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
4633 
4634 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
4635 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
4636 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
4637 
4638 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
4639 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
4640 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
4641 
4642 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
4643 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
4644 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
4645 
4646 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
4647 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
4648 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
4649 
4650 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
4651 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
4652 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
4653 
4654 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
4655 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4656 	    sysctl_cim_la, "A", "CIM logic analyzer");
4657 
4658 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
4659 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4660 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
4661 
4662 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
4663 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
4664 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
4665 
4666 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
4667 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
4668 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
4669 
4670 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
4671 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
4672 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
4673 
4674 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
4675 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
4676 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
4677 
4678 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
4679 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
4680 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
4681 
4682 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
4683 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
4684 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
4685 
4686 	if (is_t5(sc)) {
4687 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
4688 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
4689 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
4690 
4691 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
4692 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
4693 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
4694 	}
4695 
4696 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
4697 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4698 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
4699 
4700 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
4701 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4702 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
4703 
4704 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
4705 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4706 	    sysctl_cpl_stats, "A", "CPL statistics");
4707 
4708 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
4709 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4710 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
4711 
4712 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
4713 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4714 	    sysctl_devlog, "A", "firmware's device log");
4715 
4716 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
4717 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4718 	    sysctl_fcoe_stats, "A", "FCoE statistics");
4719 
4720 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
4721 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4722 	    sysctl_hw_sched, "A", "hardware scheduler ");
4723 
4724 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
4725 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4726 	    sysctl_l2t, "A", "hardware L2 table");
4727 
4728 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
4729 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4730 	    sysctl_lb_stats, "A", "loopback statistics");
4731 
4732 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
4733 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4734 	    sysctl_meminfo, "A", "memory regions");
4735 
4736 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
4737 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4738 	    sysctl_mps_tcam, "A", "MPS TCAM entries");
4739 
4740 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
4741 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4742 	    sysctl_path_mtus, "A", "path MTUs");
4743 
4744 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
4745 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4746 	    sysctl_pm_stats, "A", "PM statistics");
4747 
4748 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
4749 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4750 	    sysctl_rdma_stats, "A", "RDMA statistics");
4751 
4752 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
4753 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4754 	    sysctl_tcp_stats, "A", "TCP statistics");
4755 
4756 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
4757 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4758 	    sysctl_tids, "A", "TID information");
4759 
4760 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
4761 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4762 	    sysctl_tp_err_stats, "A", "TP error statistics");
4763 
4764 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
4765 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4766 	    sysctl_tp_la, "A", "TP logic analyzer");
4767 
4768 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
4769 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4770 	    sysctl_tx_rate, "A", "Tx rate");
4771 
4772 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
4773 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4774 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
4775 
4776 	if (is_t5(sc)) {
4777 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
4778 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4779 		    sysctl_wcwr_stats, "A", "write combined work requests");
4780 	}
4781 #endif
4782 
4783 #ifdef TCP_OFFLOAD
4784 	if (is_offload(sc)) {
4785 		/*
4786 		 * dev.t4nex.X.toe.
4787 		 */
4788 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
4789 		    NULL, "TOE parameters");
4790 		children = SYSCTL_CHILDREN(oid);
4791 
4792 		sc->tt.sndbuf = 256 * 1024;
4793 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
4794 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
4795 
4796 		sc->tt.ddp = 0;
4797 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
4798 		    &sc->tt.ddp, 0, "DDP allowed");
4799 
4800 		sc->tt.indsz = G_INDICATESIZE(t4_read_reg(sc, A_TP_PARA_REG5));
4801 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "indsz", CTLFLAG_RW,
4802 		    &sc->tt.indsz, 0, "DDP max indicate size allowed");
4803 
4804 		sc->tt.ddp_thres =
4805 		    G_RXCOALESCESIZE(t4_read_reg(sc, A_TP_PARA_REG2));
4806 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp_thres", CTLFLAG_RW,
4807 		    &sc->tt.ddp_thres, 0, "DDP threshold");
4808 
4809 		sc->tt.rx_coalesce = 1;
4810 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
4811 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
4812 
4813 		sc->tt.tx_align = 1;
4814 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
4815 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
4816 	}
4817 #endif
4818 
4819 
4820 	return (0);
4821 }
4822 
4823 static int
4824 cxgbe_sysctls(struct port_info *pi)
4825 {
4826 	struct sysctl_ctx_list *ctx;
4827 	struct sysctl_oid *oid;
4828 	struct sysctl_oid_list *children;
4829 	struct adapter *sc = pi->adapter;
4830 
4831 	ctx = device_get_sysctl_ctx(pi->dev);
4832 
4833 	/*
4834 	 * dev.cxgbe.X.
4835 	 */
4836 	oid = device_get_sysctl_tree(pi->dev);
4837 	children = SYSCTL_CHILDREN(oid);
4838 
4839 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
4840 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
4841 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
4842 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
4843 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
4844 		    "PHY temperature (in Celsius)");
4845 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
4846 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
4847 		    "PHY firmware version");
4848 	}
4849 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
4850 	    &pi->nrxq, 0, "# of rx queues");
4851 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
4852 	    &pi->ntxq, 0, "# of tx queues");
4853 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
4854 	    &pi->first_rxq, 0, "index of first rx queue");
4855 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
4856 	    &pi->first_txq, 0, "index of first tx queue");
4857 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq", CTLTYPE_INT |
4858 	    CTLFLAG_RW, pi, 0, sysctl_noflowq, "IU",
4859 	    "Reserve queue 0 for non-flowid packets");
4860 
4861 #ifdef TCP_OFFLOAD
4862 	if (is_offload(sc)) {
4863 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
4864 		    &pi->nofldrxq, 0,
4865 		    "# of rx queues for offloaded TCP connections");
4866 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
4867 		    &pi->nofldtxq, 0,
4868 		    "# of tx queues for offloaded TCP connections");
4869 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
4870 		    CTLFLAG_RD, &pi->first_ofld_rxq, 0,
4871 		    "index of first TOE rx queue");
4872 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
4873 		    CTLFLAG_RD, &pi->first_ofld_txq, 0,
4874 		    "index of first TOE tx queue");
4875 	}
4876 #endif
4877 #ifdef DEV_NETMAP
4878 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
4879 	    &pi->nnmrxq, 0, "# of rx queues for netmap");
4880 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
4881 	    &pi->nnmtxq, 0, "# of tx queues for netmap");
4882 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
4883 	    CTLFLAG_RD, &pi->first_nm_rxq, 0,
4884 	    "index of first netmap rx queue");
4885 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
4886 	    CTLFLAG_RD, &pi->first_nm_txq, 0,
4887 	    "index of first netmap tx queue");
4888 #endif
4889 
4890 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
4891 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_holdoff_tmr_idx, "I",
4892 	    "holdoff timer index");
4893 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
4894 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_holdoff_pktc_idx, "I",
4895 	    "holdoff packet counter index");
4896 
4897 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
4898 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_qsize_rxq, "I",
4899 	    "rx queue size");
4900 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
4901 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_qsize_txq, "I",
4902 	    "tx queue size");
4903 
4904 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
4905 	    CTLTYPE_STRING | CTLFLAG_RW, pi, PAUSE_TX, sysctl_pause_settings,
4906 	    "A", "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
4907 
4908 	/*
4909 	 * dev.cxgbe.X.stats.
4910 	 */
4911 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
4912 	    NULL, "port statistics");
4913 	children = SYSCTL_CHILDREN(oid);
4914 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
4915 	    &pi->tx_parse_error, 0,
4916 	    "# of tx packets with invalid length or # of segments");
4917 
4918 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
4919 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
4920 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
4921 	    sysctl_handle_t4_reg64, "QU", desc)
4922 
4923 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
4924 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
4925 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
4926 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
4927 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
4928 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
4929 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
4930 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
4931 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
4932 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
4933 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
4934 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
4935 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
4936 	    "# of tx frames in this range",
4937 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
4938 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
4939 	    "# of tx frames in this range",
4940 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
4941 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
4942 	    "# of tx frames in this range",
4943 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
4944 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
4945 	    "# of tx frames in this range",
4946 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
4947 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
4948 	    "# of tx frames in this range",
4949 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
4950 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
4951 	    "# of tx frames in this range",
4952 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
4953 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
4954 	    "# of tx frames in this range",
4955 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
4956 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
4957 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
4958 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
4959 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
4960 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
4961 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
4962 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
4963 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
4964 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
4965 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
4966 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
4967 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
4968 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
4969 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
4970 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
4971 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
4972 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
4973 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
4974 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
4975 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
4976 
4977 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
4978 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
4979 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
4980 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
4981 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
4982 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
4983 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
4984 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
4985 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
4986 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
4987 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
4988 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
4989 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
4990 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
4991 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
4992 	    "# of frames received with bad FCS",
4993 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
4994 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
4995 	    "# of frames received with length error",
4996 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
4997 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
4998 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
4999 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5000 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5001 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5002 	    "# of rx frames in this range",
5003 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5004 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5005 	    "# of rx frames in this range",
5006 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5007 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5008 	    "# of rx frames in this range",
5009 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5010 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5011 	    "# of rx frames in this range",
5012 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5013 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5014 	    "# of rx frames in this range",
5015 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5016 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5017 	    "# of rx frames in this range",
5018 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5019 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5020 	    "# of rx frames in this range",
5021 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5022 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5023 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5024 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5025 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5026 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5027 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5028 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5029 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5030 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5031 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5032 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5033 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5034 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5035 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5036 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5037 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5038 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5039 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5040 
5041 #undef SYSCTL_ADD_T4_REG64
5042 
5043 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5044 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5045 	    &pi->stats.name, desc)
5046 
5047 	/* We get these from port_stats and they may be stale by upto 1s */
5048 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5049 	    "# drops due to buffer-group 0 overflows");
5050 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5051 	    "# drops due to buffer-group 1 overflows");
5052 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5053 	    "# drops due to buffer-group 2 overflows");
5054 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5055 	    "# drops due to buffer-group 3 overflows");
5056 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5057 	    "# of buffer-group 0 truncated packets");
5058 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5059 	    "# of buffer-group 1 truncated packets");
5060 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5061 	    "# of buffer-group 2 truncated packets");
5062 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5063 	    "# of buffer-group 3 truncated packets");
5064 
5065 #undef SYSCTL_ADD_T4_PORTSTAT
5066 
5067 	return (0);
5068 }
5069 
5070 static int
5071 sysctl_int_array(SYSCTL_HANDLER_ARGS)
5072 {
5073 	int rc, *i;
5074 	struct sbuf sb;
5075 
5076 	sbuf_new(&sb, NULL, 32, SBUF_AUTOEXTEND);
5077 	for (i = arg1; arg2; arg2 -= sizeof(int), i++)
5078 		sbuf_printf(&sb, "%d ", *i);
5079 	sbuf_trim(&sb);
5080 	sbuf_finish(&sb);
5081 	rc = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
5082 	sbuf_delete(&sb);
5083 	return (rc);
5084 }
5085 
5086 static int
5087 sysctl_bitfield(SYSCTL_HANDLER_ARGS)
5088 {
5089 	int rc;
5090 	struct sbuf *sb;
5091 
5092 	rc = sysctl_wire_old_buffer(req, 0);
5093 	if (rc != 0)
5094 		return(rc);
5095 
5096 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5097 	if (sb == NULL)
5098 		return (ENOMEM);
5099 
5100 	sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
5101 	rc = sbuf_finish(sb);
5102 	sbuf_delete(sb);
5103 
5104 	return (rc);
5105 }
5106 
5107 static int
5108 sysctl_btphy(SYSCTL_HANDLER_ARGS)
5109 {
5110 	struct port_info *pi = arg1;
5111 	int op = arg2;
5112 	struct adapter *sc = pi->adapter;
5113 	u_int v;
5114 	int rc;
5115 
5116 	rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4btt");
5117 	if (rc)
5118 		return (rc);
5119 	/* XXX: magic numbers */
5120 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
5121 	    &v);
5122 	end_synchronized_op(sc, 0);
5123 	if (rc)
5124 		return (rc);
5125 	if (op == 0)
5126 		v /= 256;
5127 
5128 	rc = sysctl_handle_int(oidp, &v, 0, req);
5129 	return (rc);
5130 }
5131 
5132 static int
5133 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
5134 {
5135 	struct port_info *pi = arg1;
5136 	int rc, val;
5137 
5138 	val = pi->rsrv_noflowq;
5139 	rc = sysctl_handle_int(oidp, &val, 0, req);
5140 	if (rc != 0 || req->newptr == NULL)
5141 		return (rc);
5142 
5143 	if ((val >= 1) && (pi->ntxq > 1))
5144 		pi->rsrv_noflowq = 1;
5145 	else
5146 		pi->rsrv_noflowq = 0;
5147 
5148 	return (rc);
5149 }
5150 
5151 static int
5152 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
5153 {
5154 	struct port_info *pi = arg1;
5155 	struct adapter *sc = pi->adapter;
5156 	int idx, rc, i;
5157 	struct sge_rxq *rxq;
5158 #ifdef TCP_OFFLOAD
5159 	struct sge_ofld_rxq *ofld_rxq;
5160 #endif
5161 	uint8_t v;
5162 
5163 	idx = pi->tmr_idx;
5164 
5165 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5166 	if (rc != 0 || req->newptr == NULL)
5167 		return (rc);
5168 
5169 	if (idx < 0 || idx >= SGE_NTIMERS)
5170 		return (EINVAL);
5171 
5172 	rc = begin_synchronized_op(sc, pi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5173 	    "t4tmr");
5174 	if (rc)
5175 		return (rc);
5176 
5177 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(pi->pktc_idx != -1);
5178 	for_each_rxq(pi, i, rxq) {
5179 #ifdef atomic_store_rel_8
5180 		atomic_store_rel_8(&rxq->iq.intr_params, v);
5181 #else
5182 		rxq->iq.intr_params = v;
5183 #endif
5184 	}
5185 #ifdef TCP_OFFLOAD
5186 	for_each_ofld_rxq(pi, i, ofld_rxq) {
5187 #ifdef atomic_store_rel_8
5188 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
5189 #else
5190 		ofld_rxq->iq.intr_params = v;
5191 #endif
5192 	}
5193 #endif
5194 	pi->tmr_idx = idx;
5195 
5196 	end_synchronized_op(sc, LOCK_HELD);
5197 	return (0);
5198 }
5199 
5200 static int
5201 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
5202 {
5203 	struct port_info *pi = arg1;
5204 	struct adapter *sc = pi->adapter;
5205 	int idx, rc;
5206 
5207 	idx = pi->pktc_idx;
5208 
5209 	rc = sysctl_handle_int(oidp, &idx, 0, req);
5210 	if (rc != 0 || req->newptr == NULL)
5211 		return (rc);
5212 
5213 	if (idx < -1 || idx >= SGE_NCOUNTERS)
5214 		return (EINVAL);
5215 
5216 	rc = begin_synchronized_op(sc, pi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5217 	    "t4pktc");
5218 	if (rc)
5219 		return (rc);
5220 
5221 	if (pi->flags & PORT_INIT_DONE)
5222 		rc = EBUSY; /* cannot be changed once the queues are created */
5223 	else
5224 		pi->pktc_idx = idx;
5225 
5226 	end_synchronized_op(sc, LOCK_HELD);
5227 	return (rc);
5228 }
5229 
5230 static int
5231 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
5232 {
5233 	struct port_info *pi = arg1;
5234 	struct adapter *sc = pi->adapter;
5235 	int qsize, rc;
5236 
5237 	qsize = pi->qsize_rxq;
5238 
5239 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5240 	if (rc != 0 || req->newptr == NULL)
5241 		return (rc);
5242 
5243 	if (qsize < 128 || (qsize & 7))
5244 		return (EINVAL);
5245 
5246 	rc = begin_synchronized_op(sc, pi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5247 	    "t4rxqs");
5248 	if (rc)
5249 		return (rc);
5250 
5251 	if (pi->flags & PORT_INIT_DONE)
5252 		rc = EBUSY; /* cannot be changed once the queues are created */
5253 	else
5254 		pi->qsize_rxq = qsize;
5255 
5256 	end_synchronized_op(sc, LOCK_HELD);
5257 	return (rc);
5258 }
5259 
5260 static int
5261 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
5262 {
5263 	struct port_info *pi = arg1;
5264 	struct adapter *sc = pi->adapter;
5265 	int qsize, rc;
5266 
5267 	qsize = pi->qsize_txq;
5268 
5269 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5270 	if (rc != 0 || req->newptr == NULL)
5271 		return (rc);
5272 
5273 	if (qsize < 128 || qsize > 65536)
5274 		return (EINVAL);
5275 
5276 	rc = begin_synchronized_op(sc, pi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5277 	    "t4txqs");
5278 	if (rc)
5279 		return (rc);
5280 
5281 	if (pi->flags & PORT_INIT_DONE)
5282 		rc = EBUSY; /* cannot be changed once the queues are created */
5283 	else
5284 		pi->qsize_txq = qsize;
5285 
5286 	end_synchronized_op(sc, LOCK_HELD);
5287 	return (rc);
5288 }
5289 
5290 static int
5291 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
5292 {
5293 	struct port_info *pi = arg1;
5294 	struct adapter *sc = pi->adapter;
5295 	struct link_config *lc = &pi->link_cfg;
5296 	int rc;
5297 
5298 	if (req->newptr == NULL) {
5299 		struct sbuf *sb;
5300 		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
5301 
5302 		rc = sysctl_wire_old_buffer(req, 0);
5303 		if (rc != 0)
5304 			return(rc);
5305 
5306 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5307 		if (sb == NULL)
5308 			return (ENOMEM);
5309 
5310 		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
5311 		rc = sbuf_finish(sb);
5312 		sbuf_delete(sb);
5313 	} else {
5314 		char s[2];
5315 		int n;
5316 
5317 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
5318 		s[1] = 0;
5319 
5320 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5321 		if (rc != 0)
5322 			return(rc);
5323 
5324 		if (s[1] != 0)
5325 			return (EINVAL);
5326 		if (s[0] < '0' || s[0] > '9')
5327 			return (EINVAL);	/* not a number */
5328 		n = s[0] - '0';
5329 		if (n & ~(PAUSE_TX | PAUSE_RX))
5330 			return (EINVAL);	/* some other bit is set too */
5331 
5332 		rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4PAUSE");
5333 		if (rc)
5334 			return (rc);
5335 		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
5336 			int link_ok = lc->link_ok;
5337 
5338 			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
5339 			lc->requested_fc |= n;
5340 			rc = -t4_link_start(sc, sc->mbox, pi->tx_chan, lc);
5341 			lc->link_ok = link_ok;	/* restore */
5342 		}
5343 		end_synchronized_op(sc, 0);
5344 	}
5345 
5346 	return (rc);
5347 }
5348 
5349 static int
5350 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
5351 {
5352 	struct adapter *sc = arg1;
5353 	int reg = arg2;
5354 	uint64_t val;
5355 
5356 	val = t4_read_reg64(sc, reg);
5357 
5358 	return (sysctl_handle_64(oidp, &val, 0, req));
5359 }
5360 
5361 static int
5362 sysctl_temperature(SYSCTL_HANDLER_ARGS)
5363 {
5364 	struct adapter *sc = arg1;
5365 	int rc, t;
5366 	uint32_t param, val;
5367 
5368 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
5369 	if (rc)
5370 		return (rc);
5371 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5372 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
5373 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
5374 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
5375 	end_synchronized_op(sc, 0);
5376 	if (rc)
5377 		return (rc);
5378 
5379 	/* unknown is returned as 0 but we display -1 in that case */
5380 	t = val == 0 ? -1 : val;
5381 
5382 	rc = sysctl_handle_int(oidp, &t, 0, req);
5383 	return (rc);
5384 }
5385 
5386 #ifdef SBUF_DRAIN
5387 static int
5388 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
5389 {
5390 	struct adapter *sc = arg1;
5391 	struct sbuf *sb;
5392 	int rc, i;
5393 	uint16_t incr[NMTUS][NCCTRL_WIN];
5394 	static const char *dec_fac[] = {
5395 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
5396 		"0.9375"
5397 	};
5398 
5399 	rc = sysctl_wire_old_buffer(req, 0);
5400 	if (rc != 0)
5401 		return (rc);
5402 
5403 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5404 	if (sb == NULL)
5405 		return (ENOMEM);
5406 
5407 	t4_read_cong_tbl(sc, incr);
5408 
5409 	for (i = 0; i < NCCTRL_WIN; ++i) {
5410 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
5411 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
5412 		    incr[5][i], incr[6][i], incr[7][i]);
5413 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
5414 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
5415 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
5416 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
5417 	}
5418 
5419 	rc = sbuf_finish(sb);
5420 	sbuf_delete(sb);
5421 
5422 	return (rc);
5423 }
5424 
5425 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
5426 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
5427 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
5428 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
5429 };
5430 
5431 static int
5432 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
5433 {
5434 	struct adapter *sc = arg1;
5435 	struct sbuf *sb;
5436 	int rc, i, n, qid = arg2;
5437 	uint32_t *buf, *p;
5438 	char *qtype;
5439 	u_int cim_num_obq = is_t4(sc) ? CIM_NUM_OBQ : CIM_NUM_OBQ_T5;
5440 
5441 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
5442 	    ("%s: bad qid %d\n", __func__, qid));
5443 
5444 	if (qid < CIM_NUM_IBQ) {
5445 		/* inbound queue */
5446 		qtype = "IBQ";
5447 		n = 4 * CIM_IBQ_SIZE;
5448 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5449 		rc = t4_read_cim_ibq(sc, qid, buf, n);
5450 	} else {
5451 		/* outbound queue */
5452 		qtype = "OBQ";
5453 		qid -= CIM_NUM_IBQ;
5454 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
5455 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5456 		rc = t4_read_cim_obq(sc, qid, buf, n);
5457 	}
5458 
5459 	if (rc < 0) {
5460 		rc = -rc;
5461 		goto done;
5462 	}
5463 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
5464 
5465 	rc = sysctl_wire_old_buffer(req, 0);
5466 	if (rc != 0)
5467 		goto done;
5468 
5469 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
5470 	if (sb == NULL) {
5471 		rc = ENOMEM;
5472 		goto done;
5473 	}
5474 
5475 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
5476 	for (i = 0, p = buf; i < n; i += 16, p += 4)
5477 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
5478 		    p[2], p[3]);
5479 
5480 	rc = sbuf_finish(sb);
5481 	sbuf_delete(sb);
5482 done:
5483 	free(buf, M_CXGBE);
5484 	return (rc);
5485 }
5486 
5487 static int
5488 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
5489 {
5490 	struct adapter *sc = arg1;
5491 	u_int cfg;
5492 	struct sbuf *sb;
5493 	uint32_t *buf, *p;
5494 	int rc;
5495 
5496 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
5497 	if (rc != 0)
5498 		return (rc);
5499 
5500 	rc = sysctl_wire_old_buffer(req, 0);
5501 	if (rc != 0)
5502 		return (rc);
5503 
5504 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5505 	if (sb == NULL)
5506 		return (ENOMEM);
5507 
5508 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
5509 	    M_ZERO | M_WAITOK);
5510 
5511 	rc = -t4_cim_read_la(sc, buf, NULL);
5512 	if (rc != 0)
5513 		goto done;
5514 
5515 	sbuf_printf(sb, "Status   Data      PC%s",
5516 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
5517 	    "     LS0Stat  LS0Addr             LS0Data");
5518 
5519 	KASSERT((sc->params.cim_la_size & 7) == 0,
5520 	    ("%s: p will walk off the end of buf", __func__));
5521 
5522 	for (p = buf; p < &buf[sc->params.cim_la_size]; p += 8) {
5523 		if (cfg & F_UPDBGLACAPTPCONLY) {
5524 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
5525 			    p[6], p[7]);
5526 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
5527 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
5528 			    p[4] & 0xff, p[5] >> 8);
5529 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
5530 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
5531 			    p[1] & 0xf, p[2] >> 4);
5532 		} else {
5533 			sbuf_printf(sb,
5534 			    "\n  %02x   %x%07x %x%07x %08x %08x "
5535 			    "%08x%08x%08x%08x",
5536 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
5537 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
5538 			    p[6], p[7]);
5539 		}
5540 	}
5541 
5542 	rc = sbuf_finish(sb);
5543 	sbuf_delete(sb);
5544 done:
5545 	free(buf, M_CXGBE);
5546 	return (rc);
5547 }
5548 
5549 static int
5550 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
5551 {
5552 	struct adapter *sc = arg1;
5553 	u_int i;
5554 	struct sbuf *sb;
5555 	uint32_t *buf, *p;
5556 	int rc;
5557 
5558 	rc = sysctl_wire_old_buffer(req, 0);
5559 	if (rc != 0)
5560 		return (rc);
5561 
5562 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5563 	if (sb == NULL)
5564 		return (ENOMEM);
5565 
5566 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
5567 	    M_ZERO | M_WAITOK);
5568 
5569 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
5570 	p = buf;
5571 
5572 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
5573 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
5574 		    p[1], p[0]);
5575 	}
5576 
5577 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
5578 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
5579 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
5580 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
5581 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
5582 		    (p[1] >> 2) | ((p[2] & 3) << 30),
5583 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
5584 		    p[0] & 1);
5585 	}
5586 
5587 	rc = sbuf_finish(sb);
5588 	sbuf_delete(sb);
5589 	free(buf, M_CXGBE);
5590 	return (rc);
5591 }
5592 
5593 static int
5594 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
5595 {
5596 	struct adapter *sc = arg1;
5597 	u_int i;
5598 	struct sbuf *sb;
5599 	uint32_t *buf, *p;
5600 	int rc;
5601 
5602 	rc = sysctl_wire_old_buffer(req, 0);
5603 	if (rc != 0)
5604 		return (rc);
5605 
5606 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5607 	if (sb == NULL)
5608 		return (ENOMEM);
5609 
5610 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
5611 	    M_ZERO | M_WAITOK);
5612 
5613 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
5614 	p = buf;
5615 
5616 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
5617 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 6) {
5618 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
5619 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
5620 		    p[4], p[3], p[2], p[1], p[0]);
5621 	}
5622 
5623 	sbuf_printf(sb, "\n\nCntl ID               Data");
5624 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 6) {
5625 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
5626 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
5627 	}
5628 
5629 	rc = sbuf_finish(sb);
5630 	sbuf_delete(sb);
5631 	free(buf, M_CXGBE);
5632 	return (rc);
5633 }
5634 
5635 static int
5636 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
5637 {
5638 	struct adapter *sc = arg1;
5639 	struct sbuf *sb;
5640 	int rc, i;
5641 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
5642 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
5643 	uint16_t thres[CIM_NUM_IBQ];
5644 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
5645 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
5646 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
5647 
5648 	if (is_t4(sc)) {
5649 		cim_num_obq = CIM_NUM_OBQ;
5650 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
5651 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
5652 	} else {
5653 		cim_num_obq = CIM_NUM_OBQ_T5;
5654 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
5655 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
5656 	}
5657 	nq = CIM_NUM_IBQ + cim_num_obq;
5658 
5659 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
5660 	if (rc == 0)
5661 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
5662 	if (rc != 0)
5663 		return (rc);
5664 
5665 	t4_read_cimq_cfg(sc, base, size, thres);
5666 
5667 	rc = sysctl_wire_old_buffer(req, 0);
5668 	if (rc != 0)
5669 		return (rc);
5670 
5671 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
5672 	if (sb == NULL)
5673 		return (ENOMEM);
5674 
5675 	sbuf_printf(sb, "Queue  Base  Size Thres RdPtr WrPtr  SOP  EOP Avail");
5676 
5677 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
5678 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
5679 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
5680 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
5681 		    G_QUEREMFLITS(p[2]) * 16);
5682 	for ( ; i < nq; i++, p += 4, wr += 2)
5683 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
5684 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
5685 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
5686 		    G_QUEREMFLITS(p[2]) * 16);
5687 
5688 	rc = sbuf_finish(sb);
5689 	sbuf_delete(sb);
5690 
5691 	return (rc);
5692 }
5693 
5694 static int
5695 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
5696 {
5697 	struct adapter *sc = arg1;
5698 	struct sbuf *sb;
5699 	int rc;
5700 	struct tp_cpl_stats stats;
5701 
5702 	rc = sysctl_wire_old_buffer(req, 0);
5703 	if (rc != 0)
5704 		return (rc);
5705 
5706 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
5707 	if (sb == NULL)
5708 		return (ENOMEM);
5709 
5710 	t4_tp_get_cpl_stats(sc, &stats);
5711 
5712 	sbuf_printf(sb, "                 channel 0  channel 1  channel 2  "
5713 	    "channel 3\n");
5714 	sbuf_printf(sb, "CPL requests:   %10u %10u %10u %10u\n",
5715 		   stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
5716 	sbuf_printf(sb, "CPL responses:  %10u %10u %10u %10u",
5717 		   stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
5718 
5719 	rc = sbuf_finish(sb);
5720 	sbuf_delete(sb);
5721 
5722 	return (rc);
5723 }
5724 
5725 static int
5726 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
5727 {
5728 	struct adapter *sc = arg1;
5729 	struct sbuf *sb;
5730 	int rc;
5731 	struct tp_usm_stats stats;
5732 
5733 	rc = sysctl_wire_old_buffer(req, 0);
5734 	if (rc != 0)
5735 		return(rc);
5736 
5737 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
5738 	if (sb == NULL)
5739 		return (ENOMEM);
5740 
5741 	t4_get_usm_stats(sc, &stats);
5742 
5743 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
5744 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
5745 	sbuf_printf(sb, "Drops:  %u", stats.drops);
5746 
5747 	rc = sbuf_finish(sb);
5748 	sbuf_delete(sb);
5749 
5750 	return (rc);
5751 }
5752 
5753 const char *devlog_level_strings[] = {
5754 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
5755 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
5756 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
5757 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
5758 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
5759 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
5760 };
5761 
5762 const char *devlog_facility_strings[] = {
5763 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
5764 	[FW_DEVLOG_FACILITY_CF]		= "CF",
5765 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
5766 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
5767 	[FW_DEVLOG_FACILITY_RES]	= "RES",
5768 	[FW_DEVLOG_FACILITY_HW]		= "HW",
5769 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
5770 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
5771 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
5772 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
5773 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
5774 	[FW_DEVLOG_FACILITY_VI]		= "VI",
5775 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
5776 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
5777 	[FW_DEVLOG_FACILITY_TM]		= "TM",
5778 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
5779 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
5780 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
5781 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
5782 	[FW_DEVLOG_FACILITY_RI]		= "RI",
5783 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
5784 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
5785 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
5786 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE"
5787 };
5788 
5789 static int
5790 sysctl_devlog(SYSCTL_HANDLER_ARGS)
5791 {
5792 	struct adapter *sc = arg1;
5793 	struct devlog_params *dparams = &sc->params.devlog;
5794 	struct fw_devlog_e *buf, *e;
5795 	int i, j, rc, nentries, first = 0, m;
5796 	struct sbuf *sb;
5797 	uint64_t ftstamp = UINT64_MAX;
5798 
5799 	if (dparams->start == 0) {
5800 		dparams->memtype = FW_MEMTYPE_EDC0;
5801 		dparams->start = 0x84000;
5802 		dparams->size = 32768;
5803 	}
5804 
5805 	nentries = dparams->size / sizeof(struct fw_devlog_e);
5806 
5807 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
5808 	if (buf == NULL)
5809 		return (ENOMEM);
5810 
5811 	m = fwmtype_to_hwmtype(dparams->memtype);
5812 	rc = -t4_mem_read(sc, m, dparams->start, dparams->size, (void *)buf);
5813 	if (rc != 0)
5814 		goto done;
5815 
5816 	for (i = 0; i < nentries; i++) {
5817 		e = &buf[i];
5818 
5819 		if (e->timestamp == 0)
5820 			break;	/* end */
5821 
5822 		e->timestamp = be64toh(e->timestamp);
5823 		e->seqno = be32toh(e->seqno);
5824 		for (j = 0; j < 8; j++)
5825 			e->params[j] = be32toh(e->params[j]);
5826 
5827 		if (e->timestamp < ftstamp) {
5828 			ftstamp = e->timestamp;
5829 			first = i;
5830 		}
5831 	}
5832 
5833 	if (buf[first].timestamp == 0)
5834 		goto done;	/* nothing in the log */
5835 
5836 	rc = sysctl_wire_old_buffer(req, 0);
5837 	if (rc != 0)
5838 		goto done;
5839 
5840 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5841 	if (sb == NULL) {
5842 		rc = ENOMEM;
5843 		goto done;
5844 	}
5845 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
5846 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
5847 
5848 	i = first;
5849 	do {
5850 		e = &buf[i];
5851 		if (e->timestamp == 0)
5852 			break;	/* end */
5853 
5854 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
5855 		    e->seqno, e->timestamp,
5856 		    (e->level < nitems(devlog_level_strings) ?
5857 			devlog_level_strings[e->level] : "UNKNOWN"),
5858 		    (e->facility < nitems(devlog_facility_strings) ?
5859 			devlog_facility_strings[e->facility] : "UNKNOWN"));
5860 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
5861 		    e->params[2], e->params[3], e->params[4],
5862 		    e->params[5], e->params[6], e->params[7]);
5863 
5864 		if (++i == nentries)
5865 			i = 0;
5866 	} while (i != first);
5867 
5868 	rc = sbuf_finish(sb);
5869 	sbuf_delete(sb);
5870 done:
5871 	free(buf, M_CXGBE);
5872 	return (rc);
5873 }
5874 
5875 static int
5876 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
5877 {
5878 	struct adapter *sc = arg1;
5879 	struct sbuf *sb;
5880 	int rc;
5881 	struct tp_fcoe_stats stats[4];
5882 
5883 	rc = sysctl_wire_old_buffer(req, 0);
5884 	if (rc != 0)
5885 		return (rc);
5886 
5887 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
5888 	if (sb == NULL)
5889 		return (ENOMEM);
5890 
5891 	t4_get_fcoe_stats(sc, 0, &stats[0]);
5892 	t4_get_fcoe_stats(sc, 1, &stats[1]);
5893 	t4_get_fcoe_stats(sc, 2, &stats[2]);
5894 	t4_get_fcoe_stats(sc, 3, &stats[3]);
5895 
5896 	sbuf_printf(sb, "                   channel 0        channel 1        "
5897 	    "channel 2        channel 3\n");
5898 	sbuf_printf(sb, "octetsDDP:  %16ju %16ju %16ju %16ju\n",
5899 	    stats[0].octetsDDP, stats[1].octetsDDP, stats[2].octetsDDP,
5900 	    stats[3].octetsDDP);
5901 	sbuf_printf(sb, "framesDDP:  %16u %16u %16u %16u\n", stats[0].framesDDP,
5902 	    stats[1].framesDDP, stats[2].framesDDP, stats[3].framesDDP);
5903 	sbuf_printf(sb, "framesDrop: %16u %16u %16u %16u",
5904 	    stats[0].framesDrop, stats[1].framesDrop, stats[2].framesDrop,
5905 	    stats[3].framesDrop);
5906 
5907 	rc = sbuf_finish(sb);
5908 	sbuf_delete(sb);
5909 
5910 	return (rc);
5911 }
5912 
5913 static int
5914 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
5915 {
5916 	struct adapter *sc = arg1;
5917 	struct sbuf *sb;
5918 	int rc, i;
5919 	unsigned int map, kbps, ipg, mode;
5920 	unsigned int pace_tab[NTX_SCHED];
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, 256, req);
5927 	if (sb == NULL)
5928 		return (ENOMEM);
5929 
5930 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
5931 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
5932 	t4_read_pace_tbl(sc, pace_tab);
5933 
5934 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
5935 	    "Class IPG (0.1 ns)   Flow IPG (us)");
5936 
5937 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
5938 		t4_get_tx_sched(sc, i, &kbps, &ipg);
5939 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
5940 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
5941 		if (kbps)
5942 			sbuf_printf(sb, "%9u     ", kbps);
5943 		else
5944 			sbuf_printf(sb, " disabled     ");
5945 
5946 		if (ipg)
5947 			sbuf_printf(sb, "%13u        ", ipg);
5948 		else
5949 			sbuf_printf(sb, "     disabled        ");
5950 
5951 		if (pace_tab[i])
5952 			sbuf_printf(sb, "%10u", pace_tab[i]);
5953 		else
5954 			sbuf_printf(sb, "  disabled");
5955 	}
5956 
5957 	rc = sbuf_finish(sb);
5958 	sbuf_delete(sb);
5959 
5960 	return (rc);
5961 }
5962 
5963 static int
5964 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
5965 {
5966 	struct adapter *sc = arg1;
5967 	struct sbuf *sb;
5968 	int rc, i, j;
5969 	uint64_t *p0, *p1;
5970 	struct lb_port_stats s[2];
5971 	static const char *stat_name[] = {
5972 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
5973 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
5974 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
5975 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
5976 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
5977 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
5978 		"BG2FramesTrunc:", "BG3FramesTrunc:"
5979 	};
5980 
5981 	rc = sysctl_wire_old_buffer(req, 0);
5982 	if (rc != 0)
5983 		return (rc);
5984 
5985 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5986 	if (sb == NULL)
5987 		return (ENOMEM);
5988 
5989 	memset(s, 0, sizeof(s));
5990 
5991 	for (i = 0; i < 4; i += 2) {
5992 		t4_get_lb_stats(sc, i, &s[0]);
5993 		t4_get_lb_stats(sc, i + 1, &s[1]);
5994 
5995 		p0 = &s[0].octets;
5996 		p1 = &s[1].octets;
5997 		sbuf_printf(sb, "%s                       Loopback %u"
5998 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
5999 
6000 		for (j = 0; j < nitems(stat_name); j++)
6001 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
6002 				   *p0++, *p1++);
6003 	}
6004 
6005 	rc = sbuf_finish(sb);
6006 	sbuf_delete(sb);
6007 
6008 	return (rc);
6009 }
6010 
6011 static int
6012 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
6013 {
6014 	int rc = 0;
6015 	struct port_info *pi = arg1;
6016 	struct sbuf *sb;
6017 	static const char *linkdnreasons[] = {
6018 		"non-specific", "remote fault", "autoneg failed", "reserved3",
6019 		"PHY overheated", "unknown", "rx los", "reserved7"
6020 	};
6021 
6022 	rc = sysctl_wire_old_buffer(req, 0);
6023 	if (rc != 0)
6024 		return(rc);
6025 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
6026 	if (sb == NULL)
6027 		return (ENOMEM);
6028 
6029 	if (pi->linkdnrc < 0)
6030 		sbuf_printf(sb, "n/a");
6031 	else if (pi->linkdnrc < nitems(linkdnreasons))
6032 		sbuf_printf(sb, "%s", linkdnreasons[pi->linkdnrc]);
6033 	else
6034 		sbuf_printf(sb, "%d", pi->linkdnrc);
6035 
6036 	rc = sbuf_finish(sb);
6037 	sbuf_delete(sb);
6038 
6039 	return (rc);
6040 }
6041 
6042 struct mem_desc {
6043 	unsigned int base;
6044 	unsigned int limit;
6045 	unsigned int idx;
6046 };
6047 
6048 static int
6049 mem_desc_cmp(const void *a, const void *b)
6050 {
6051 	return ((const struct mem_desc *)a)->base -
6052 	       ((const struct mem_desc *)b)->base;
6053 }
6054 
6055 static void
6056 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
6057     unsigned int to)
6058 {
6059 	unsigned int size;
6060 
6061 	size = to - from + 1;
6062 	if (size == 0)
6063 		return;
6064 
6065 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
6066 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
6067 }
6068 
6069 static int
6070 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
6071 {
6072 	struct adapter *sc = arg1;
6073 	struct sbuf *sb;
6074 	int rc, i, n;
6075 	uint32_t lo, hi, used, alloc;
6076 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
6077 	static const char *region[] = {
6078 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
6079 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
6080 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
6081 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
6082 		"RQUDP region:", "PBL region:", "TXPBL region:",
6083 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
6084 		"On-chip queues:"
6085 	};
6086 	struct mem_desc avail[4];
6087 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
6088 	struct mem_desc *md = mem;
6089 
6090 	rc = sysctl_wire_old_buffer(req, 0);
6091 	if (rc != 0)
6092 		return (rc);
6093 
6094 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6095 	if (sb == NULL)
6096 		return (ENOMEM);
6097 
6098 	for (i = 0; i < nitems(mem); i++) {
6099 		mem[i].limit = 0;
6100 		mem[i].idx = i;
6101 	}
6102 
6103 	/* Find and sort the populated memory ranges */
6104 	i = 0;
6105 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
6106 	if (lo & F_EDRAM0_ENABLE) {
6107 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
6108 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
6109 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
6110 		avail[i].idx = 0;
6111 		i++;
6112 	}
6113 	if (lo & F_EDRAM1_ENABLE) {
6114 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
6115 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
6116 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
6117 		avail[i].idx = 1;
6118 		i++;
6119 	}
6120 	if (lo & F_EXT_MEM_ENABLE) {
6121 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
6122 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
6123 		avail[i].limit = avail[i].base +
6124 		    (G_EXT_MEM_SIZE(hi) << 20);
6125 		avail[i].idx = is_t4(sc) ? 2 : 3;	/* Call it MC for T4 */
6126 		i++;
6127 	}
6128 	if (!is_t4(sc) && lo & F_EXT_MEM1_ENABLE) {
6129 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
6130 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
6131 		avail[i].limit = avail[i].base +
6132 		    (G_EXT_MEM1_SIZE(hi) << 20);
6133 		avail[i].idx = 4;
6134 		i++;
6135 	}
6136 	if (!i)                                    /* no memory available */
6137 		return 0;
6138 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
6139 
6140 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
6141 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
6142 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
6143 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
6144 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
6145 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
6146 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
6147 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
6148 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
6149 
6150 	/* the next few have explicit upper bounds */
6151 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
6152 	md->limit = md->base - 1 +
6153 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
6154 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
6155 	md++;
6156 
6157 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
6158 	md->limit = md->base - 1 +
6159 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
6160 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
6161 	md++;
6162 
6163 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
6164 		hi = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
6165 		md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
6166 		md->limit = (sc->tids.ntids - hi) * 16 + md->base - 1;
6167 	} else {
6168 		md->base = 0;
6169 		md->idx = nitems(region);  /* hide it */
6170 	}
6171 	md++;
6172 
6173 #define ulp_region(reg) \
6174 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
6175 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
6176 
6177 	ulp_region(RX_ISCSI);
6178 	ulp_region(RX_TDDP);
6179 	ulp_region(TX_TPT);
6180 	ulp_region(RX_STAG);
6181 	ulp_region(RX_RQ);
6182 	ulp_region(RX_RQUDP);
6183 	ulp_region(RX_PBL);
6184 	ulp_region(TX_PBL);
6185 #undef ulp_region
6186 
6187 	md->base = 0;
6188 	md->idx = nitems(region);
6189 	if (!is_t4(sc) && t4_read_reg(sc, A_SGE_CONTROL2) & F_VFIFO_ENABLE) {
6190 		md->base = G_BASEADDR(t4_read_reg(sc, A_SGE_DBVFIFO_BADDR));
6191 		md->limit = md->base + (G_DBVFIFO_SIZE((t4_read_reg(sc,
6192 		    A_SGE_DBVFIFO_SIZE))) << 2) - 1;
6193 	}
6194 	md++;
6195 
6196 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
6197 	md->limit = md->base + sc->tids.ntids - 1;
6198 	md++;
6199 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
6200 	md->limit = md->base + sc->tids.ntids - 1;
6201 	md++;
6202 
6203 	md->base = sc->vres.ocq.start;
6204 	if (sc->vres.ocq.size)
6205 		md->limit = md->base + sc->vres.ocq.size - 1;
6206 	else
6207 		md->idx = nitems(region);  /* hide it */
6208 	md++;
6209 
6210 	/* add any address-space holes, there can be up to 3 */
6211 	for (n = 0; n < i - 1; n++)
6212 		if (avail[n].limit < avail[n + 1].base)
6213 			(md++)->base = avail[n].limit;
6214 	if (avail[n].limit)
6215 		(md++)->base = avail[n].limit;
6216 
6217 	n = md - mem;
6218 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
6219 
6220 	for (lo = 0; lo < i; lo++)
6221 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
6222 				avail[lo].limit - 1);
6223 
6224 	sbuf_printf(sb, "\n");
6225 	for (i = 0; i < n; i++) {
6226 		if (mem[i].idx >= nitems(region))
6227 			continue;                        /* skip holes */
6228 		if (!mem[i].limit)
6229 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
6230 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
6231 				mem[i].limit);
6232 	}
6233 
6234 	sbuf_printf(sb, "\n");
6235 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
6236 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
6237 	mem_region_show(sb, "uP RAM:", lo, hi);
6238 
6239 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
6240 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
6241 	mem_region_show(sb, "uP Extmem2:", lo, hi);
6242 
6243 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
6244 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
6245 		   G_PMRXMAXPAGE(lo),
6246 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
6247 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
6248 
6249 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
6250 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
6251 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
6252 		   G_PMTXMAXPAGE(lo),
6253 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
6254 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
6255 	sbuf_printf(sb, "%u p-structs\n",
6256 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
6257 
6258 	for (i = 0; i < 4; i++) {
6259 		lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
6260 		if (is_t4(sc)) {
6261 			used = G_USED(lo);
6262 			alloc = G_ALLOC(lo);
6263 		} else {
6264 			used = G_T5_USED(lo);
6265 			alloc = G_T5_ALLOC(lo);
6266 		}
6267 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
6268 			   i, used, alloc);
6269 	}
6270 	for (i = 0; i < 4; i++) {
6271 		lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
6272 		if (is_t4(sc)) {
6273 			used = G_USED(lo);
6274 			alloc = G_ALLOC(lo);
6275 		} else {
6276 			used = G_T5_USED(lo);
6277 			alloc = G_T5_ALLOC(lo);
6278 		}
6279 		sbuf_printf(sb,
6280 			   "\nLoopback %d using %u pages out of %u allocated",
6281 			   i, used, alloc);
6282 	}
6283 
6284 	rc = sbuf_finish(sb);
6285 	sbuf_delete(sb);
6286 
6287 	return (rc);
6288 }
6289 
6290 static inline void
6291 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
6292 {
6293 	*mask = x | y;
6294 	y = htobe64(y);
6295 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
6296 }
6297 
6298 static int
6299 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
6300 {
6301 	struct adapter *sc = arg1;
6302 	struct sbuf *sb;
6303 	int rc, i, n;
6304 
6305 	rc = sysctl_wire_old_buffer(req, 0);
6306 	if (rc != 0)
6307 		return (rc);
6308 
6309 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6310 	if (sb == NULL)
6311 		return (ENOMEM);
6312 
6313 	sbuf_printf(sb,
6314 	    "Idx  Ethernet address     Mask     Vld Ports PF"
6315 	    "  VF              Replication             P0 P1 P2 P3  ML");
6316 	n = is_t4(sc) ? NUM_MPS_CLS_SRAM_L_INSTANCES :
6317 	    NUM_MPS_T5_CLS_SRAM_L_INSTANCES;
6318 	for (i = 0; i < n; i++) {
6319 		uint64_t tcamx, tcamy, mask;
6320 		uint32_t cls_lo, cls_hi;
6321 		uint8_t addr[ETHER_ADDR_LEN];
6322 
6323 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
6324 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
6325 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
6326 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
6327 
6328 		if (tcamx & tcamy)
6329 			continue;
6330 
6331 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
6332 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
6333 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
6334 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
6335 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
6336 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
6337 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
6338 
6339 		if (cls_lo & F_REPLICATE) {
6340 			struct fw_ldst_cmd ldst_cmd;
6341 
6342 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
6343 			ldst_cmd.op_to_addrspace =
6344 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
6345 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
6346 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
6347 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
6348 			ldst_cmd.u.mps.fid_ctl =
6349 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
6350 				V_FW_LDST_CMD_CTL(i));
6351 
6352 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
6353 			    "t4mps");
6354 			if (rc)
6355 				break;
6356 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
6357 			    sizeof(ldst_cmd), &ldst_cmd);
6358 			end_synchronized_op(sc, 0);
6359 
6360 			if (rc != 0) {
6361 				sbuf_printf(sb,
6362 				    " ------------ error %3u ------------", rc);
6363 				rc = 0;
6364 			} else {
6365 				sbuf_printf(sb, " %08x %08x %08x %08x",
6366 				    be32toh(ldst_cmd.u.mps.rplc127_96),
6367 				    be32toh(ldst_cmd.u.mps.rplc95_64),
6368 				    be32toh(ldst_cmd.u.mps.rplc63_32),
6369 				    be32toh(ldst_cmd.u.mps.rplc31_0));
6370 			}
6371 		} else
6372 			sbuf_printf(sb, "%36s", "");
6373 
6374 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
6375 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
6376 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
6377 	}
6378 
6379 	if (rc)
6380 		(void) sbuf_finish(sb);
6381 	else
6382 		rc = sbuf_finish(sb);
6383 	sbuf_delete(sb);
6384 
6385 	return (rc);
6386 }
6387 
6388 static int
6389 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
6390 {
6391 	struct adapter *sc = arg1;
6392 	struct sbuf *sb;
6393 	int rc;
6394 	uint16_t mtus[NMTUS];
6395 
6396 	rc = sysctl_wire_old_buffer(req, 0);
6397 	if (rc != 0)
6398 		return (rc);
6399 
6400 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6401 	if (sb == NULL)
6402 		return (ENOMEM);
6403 
6404 	t4_read_mtu_tbl(sc, mtus, NULL);
6405 
6406 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
6407 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
6408 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
6409 	    mtus[14], mtus[15]);
6410 
6411 	rc = sbuf_finish(sb);
6412 	sbuf_delete(sb);
6413 
6414 	return (rc);
6415 }
6416 
6417 static int
6418 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
6419 {
6420 	struct adapter *sc = arg1;
6421 	struct sbuf *sb;
6422 	int rc, i;
6423 	uint32_t cnt[PM_NSTATS];
6424 	uint64_t cyc[PM_NSTATS];
6425 	static const char *rx_stats[] = {
6426 		"Read:", "Write bypass:", "Write mem:", "Flush:"
6427 	};
6428 	static const char *tx_stats[] = {
6429 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:"
6430 	};
6431 
6432 	rc = sysctl_wire_old_buffer(req, 0);
6433 	if (rc != 0)
6434 		return (rc);
6435 
6436 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6437 	if (sb == NULL)
6438 		return (ENOMEM);
6439 
6440 	t4_pmtx_get_stats(sc, cnt, cyc);
6441 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
6442 	for (i = 0; i < ARRAY_SIZE(tx_stats); i++)
6443 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], cnt[i],
6444 		    cyc[i]);
6445 
6446 	t4_pmrx_get_stats(sc, cnt, cyc);
6447 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
6448 	for (i = 0; i < ARRAY_SIZE(rx_stats); i++)
6449 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], cnt[i],
6450 		    cyc[i]);
6451 
6452 	rc = sbuf_finish(sb);
6453 	sbuf_delete(sb);
6454 
6455 	return (rc);
6456 }
6457 
6458 static int
6459 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
6460 {
6461 	struct adapter *sc = arg1;
6462 	struct sbuf *sb;
6463 	int rc;
6464 	struct tp_rdma_stats stats;
6465 
6466 	rc = sysctl_wire_old_buffer(req, 0);
6467 	if (rc != 0)
6468 		return (rc);
6469 
6470 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6471 	if (sb == NULL)
6472 		return (ENOMEM);
6473 
6474 	t4_tp_get_rdma_stats(sc, &stats);
6475 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
6476 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
6477 
6478 	rc = sbuf_finish(sb);
6479 	sbuf_delete(sb);
6480 
6481 	return (rc);
6482 }
6483 
6484 static int
6485 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
6486 {
6487 	struct adapter *sc = arg1;
6488 	struct sbuf *sb;
6489 	int rc;
6490 	struct tp_tcp_stats v4, v6;
6491 
6492 	rc = sysctl_wire_old_buffer(req, 0);
6493 	if (rc != 0)
6494 		return (rc);
6495 
6496 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6497 	if (sb == NULL)
6498 		return (ENOMEM);
6499 
6500 	t4_tp_get_tcp_stats(sc, &v4, &v6);
6501 	sbuf_printf(sb,
6502 	    "                                IP                 IPv6\n");
6503 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
6504 	    v4.tcpOutRsts, v6.tcpOutRsts);
6505 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
6506 	    v4.tcpInSegs, v6.tcpInSegs);
6507 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
6508 	    v4.tcpOutSegs, v6.tcpOutSegs);
6509 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
6510 	    v4.tcpRetransSegs, v6.tcpRetransSegs);
6511 
6512 	rc = sbuf_finish(sb);
6513 	sbuf_delete(sb);
6514 
6515 	return (rc);
6516 }
6517 
6518 static int
6519 sysctl_tids(SYSCTL_HANDLER_ARGS)
6520 {
6521 	struct adapter *sc = arg1;
6522 	struct sbuf *sb;
6523 	int rc;
6524 	struct tid_info *t = &sc->tids;
6525 
6526 	rc = sysctl_wire_old_buffer(req, 0);
6527 	if (rc != 0)
6528 		return (rc);
6529 
6530 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6531 	if (sb == NULL)
6532 		return (ENOMEM);
6533 
6534 	if (t->natids) {
6535 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
6536 		    t->atids_in_use);
6537 	}
6538 
6539 	if (t->ntids) {
6540 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
6541 			uint32_t b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
6542 
6543 			if (b) {
6544 				sbuf_printf(sb, "TID range: 0-%u, %u-%u", b - 1,
6545 				    t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4,
6546 				    t->ntids - 1);
6547 			} else {
6548 				sbuf_printf(sb, "TID range: %u-%u",
6549 				    t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4,
6550 				    t->ntids - 1);
6551 			}
6552 		} else
6553 			sbuf_printf(sb, "TID range: 0-%u", t->ntids - 1);
6554 		sbuf_printf(sb, ", in use: %u\n",
6555 		    atomic_load_acq_int(&t->tids_in_use));
6556 	}
6557 
6558 	if (t->nstids) {
6559 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
6560 		    t->stid_base + t->nstids - 1, t->stids_in_use);
6561 	}
6562 
6563 	if (t->nftids) {
6564 		sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
6565 		    t->ftid_base + t->nftids - 1);
6566 	}
6567 
6568 	if (t->netids) {
6569 		sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
6570 		    t->etid_base + t->netids - 1);
6571 	}
6572 
6573 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
6574 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
6575 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
6576 
6577 	rc = sbuf_finish(sb);
6578 	sbuf_delete(sb);
6579 
6580 	return (rc);
6581 }
6582 
6583 static int
6584 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
6585 {
6586 	struct adapter *sc = arg1;
6587 	struct sbuf *sb;
6588 	int rc;
6589 	struct tp_err_stats stats;
6590 
6591 	rc = sysctl_wire_old_buffer(req, 0);
6592 	if (rc != 0)
6593 		return (rc);
6594 
6595 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6596 	if (sb == NULL)
6597 		return (ENOMEM);
6598 
6599 	t4_tp_get_err_stats(sc, &stats);
6600 
6601 	sbuf_printf(sb, "                 channel 0  channel 1  channel 2  "
6602 		      "channel 3\n");
6603 	sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
6604 	    stats.macInErrs[0], stats.macInErrs[1], stats.macInErrs[2],
6605 	    stats.macInErrs[3]);
6606 	sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
6607 	    stats.hdrInErrs[0], stats.hdrInErrs[1], stats.hdrInErrs[2],
6608 	    stats.hdrInErrs[3]);
6609 	sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
6610 	    stats.tcpInErrs[0], stats.tcpInErrs[1], stats.tcpInErrs[2],
6611 	    stats.tcpInErrs[3]);
6612 	sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
6613 	    stats.tcp6InErrs[0], stats.tcp6InErrs[1], stats.tcp6InErrs[2],
6614 	    stats.tcp6InErrs[3]);
6615 	sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
6616 	    stats.tnlCongDrops[0], stats.tnlCongDrops[1], stats.tnlCongDrops[2],
6617 	    stats.tnlCongDrops[3]);
6618 	sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
6619 	    stats.tnlTxDrops[0], stats.tnlTxDrops[1], stats.tnlTxDrops[2],
6620 	    stats.tnlTxDrops[3]);
6621 	sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
6622 	    stats.ofldVlanDrops[0], stats.ofldVlanDrops[1],
6623 	    stats.ofldVlanDrops[2], stats.ofldVlanDrops[3]);
6624 	sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
6625 	    stats.ofldChanDrops[0], stats.ofldChanDrops[1],
6626 	    stats.ofldChanDrops[2], stats.ofldChanDrops[3]);
6627 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
6628 	    stats.ofldNoNeigh, stats.ofldCongDefer);
6629 
6630 	rc = sbuf_finish(sb);
6631 	sbuf_delete(sb);
6632 
6633 	return (rc);
6634 }
6635 
6636 struct field_desc {
6637 	const char *name;
6638 	u_int start;
6639 	u_int width;
6640 };
6641 
6642 static void
6643 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
6644 {
6645 	char buf[32];
6646 	int line_size = 0;
6647 
6648 	while (f->name) {
6649 		uint64_t mask = (1ULL << f->width) - 1;
6650 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
6651 		    ((uintmax_t)v >> f->start) & mask);
6652 
6653 		if (line_size + len >= 79) {
6654 			line_size = 8;
6655 			sbuf_printf(sb, "\n        ");
6656 		}
6657 		sbuf_printf(sb, "%s ", buf);
6658 		line_size += len + 1;
6659 		f++;
6660 	}
6661 	sbuf_printf(sb, "\n");
6662 }
6663 
6664 static struct field_desc tp_la0[] = {
6665 	{ "RcfOpCodeOut", 60, 4 },
6666 	{ "State", 56, 4 },
6667 	{ "WcfState", 52, 4 },
6668 	{ "RcfOpcSrcOut", 50, 2 },
6669 	{ "CRxError", 49, 1 },
6670 	{ "ERxError", 48, 1 },
6671 	{ "SanityFailed", 47, 1 },
6672 	{ "SpuriousMsg", 46, 1 },
6673 	{ "FlushInputMsg", 45, 1 },
6674 	{ "FlushInputCpl", 44, 1 },
6675 	{ "RssUpBit", 43, 1 },
6676 	{ "RssFilterHit", 42, 1 },
6677 	{ "Tid", 32, 10 },
6678 	{ "InitTcb", 31, 1 },
6679 	{ "LineNumber", 24, 7 },
6680 	{ "Emsg", 23, 1 },
6681 	{ "EdataOut", 22, 1 },
6682 	{ "Cmsg", 21, 1 },
6683 	{ "CdataOut", 20, 1 },
6684 	{ "EreadPdu", 19, 1 },
6685 	{ "CreadPdu", 18, 1 },
6686 	{ "TunnelPkt", 17, 1 },
6687 	{ "RcfPeerFin", 16, 1 },
6688 	{ "RcfReasonOut", 12, 4 },
6689 	{ "TxCchannel", 10, 2 },
6690 	{ "RcfTxChannel", 8, 2 },
6691 	{ "RxEchannel", 6, 2 },
6692 	{ "RcfRxChannel", 5, 1 },
6693 	{ "RcfDataOutSrdy", 4, 1 },
6694 	{ "RxDvld", 3, 1 },
6695 	{ "RxOoDvld", 2, 1 },
6696 	{ "RxCongestion", 1, 1 },
6697 	{ "TxCongestion", 0, 1 },
6698 	{ NULL }
6699 };
6700 
6701 static struct field_desc tp_la1[] = {
6702 	{ "CplCmdIn", 56, 8 },
6703 	{ "CplCmdOut", 48, 8 },
6704 	{ "ESynOut", 47, 1 },
6705 	{ "EAckOut", 46, 1 },
6706 	{ "EFinOut", 45, 1 },
6707 	{ "ERstOut", 44, 1 },
6708 	{ "SynIn", 43, 1 },
6709 	{ "AckIn", 42, 1 },
6710 	{ "FinIn", 41, 1 },
6711 	{ "RstIn", 40, 1 },
6712 	{ "DataIn", 39, 1 },
6713 	{ "DataInVld", 38, 1 },
6714 	{ "PadIn", 37, 1 },
6715 	{ "RxBufEmpty", 36, 1 },
6716 	{ "RxDdp", 35, 1 },
6717 	{ "RxFbCongestion", 34, 1 },
6718 	{ "TxFbCongestion", 33, 1 },
6719 	{ "TxPktSumSrdy", 32, 1 },
6720 	{ "RcfUlpType", 28, 4 },
6721 	{ "Eread", 27, 1 },
6722 	{ "Ebypass", 26, 1 },
6723 	{ "Esave", 25, 1 },
6724 	{ "Static0", 24, 1 },
6725 	{ "Cread", 23, 1 },
6726 	{ "Cbypass", 22, 1 },
6727 	{ "Csave", 21, 1 },
6728 	{ "CPktOut", 20, 1 },
6729 	{ "RxPagePoolFull", 18, 2 },
6730 	{ "RxLpbkPkt", 17, 1 },
6731 	{ "TxLpbkPkt", 16, 1 },
6732 	{ "RxVfValid", 15, 1 },
6733 	{ "SynLearned", 14, 1 },
6734 	{ "SetDelEntry", 13, 1 },
6735 	{ "SetInvEntry", 12, 1 },
6736 	{ "CpcmdDvld", 11, 1 },
6737 	{ "CpcmdSave", 10, 1 },
6738 	{ "RxPstructsFull", 8, 2 },
6739 	{ "EpcmdDvld", 7, 1 },
6740 	{ "EpcmdFlush", 6, 1 },
6741 	{ "EpcmdTrimPrefix", 5, 1 },
6742 	{ "EpcmdTrimPostfix", 4, 1 },
6743 	{ "ERssIp4Pkt", 3, 1 },
6744 	{ "ERssIp6Pkt", 2, 1 },
6745 	{ "ERssTcpUdpPkt", 1, 1 },
6746 	{ "ERssFceFipPkt", 0, 1 },
6747 	{ NULL }
6748 };
6749 
6750 static struct field_desc tp_la2[] = {
6751 	{ "CplCmdIn", 56, 8 },
6752 	{ "MpsVfVld", 55, 1 },
6753 	{ "MpsPf", 52, 3 },
6754 	{ "MpsVf", 44, 8 },
6755 	{ "SynIn", 43, 1 },
6756 	{ "AckIn", 42, 1 },
6757 	{ "FinIn", 41, 1 },
6758 	{ "RstIn", 40, 1 },
6759 	{ "DataIn", 39, 1 },
6760 	{ "DataInVld", 38, 1 },
6761 	{ "PadIn", 37, 1 },
6762 	{ "RxBufEmpty", 36, 1 },
6763 	{ "RxDdp", 35, 1 },
6764 	{ "RxFbCongestion", 34, 1 },
6765 	{ "TxFbCongestion", 33, 1 },
6766 	{ "TxPktSumSrdy", 32, 1 },
6767 	{ "RcfUlpType", 28, 4 },
6768 	{ "Eread", 27, 1 },
6769 	{ "Ebypass", 26, 1 },
6770 	{ "Esave", 25, 1 },
6771 	{ "Static0", 24, 1 },
6772 	{ "Cread", 23, 1 },
6773 	{ "Cbypass", 22, 1 },
6774 	{ "Csave", 21, 1 },
6775 	{ "CPktOut", 20, 1 },
6776 	{ "RxPagePoolFull", 18, 2 },
6777 	{ "RxLpbkPkt", 17, 1 },
6778 	{ "TxLpbkPkt", 16, 1 },
6779 	{ "RxVfValid", 15, 1 },
6780 	{ "SynLearned", 14, 1 },
6781 	{ "SetDelEntry", 13, 1 },
6782 	{ "SetInvEntry", 12, 1 },
6783 	{ "CpcmdDvld", 11, 1 },
6784 	{ "CpcmdSave", 10, 1 },
6785 	{ "RxPstructsFull", 8, 2 },
6786 	{ "EpcmdDvld", 7, 1 },
6787 	{ "EpcmdFlush", 6, 1 },
6788 	{ "EpcmdTrimPrefix", 5, 1 },
6789 	{ "EpcmdTrimPostfix", 4, 1 },
6790 	{ "ERssIp4Pkt", 3, 1 },
6791 	{ "ERssIp6Pkt", 2, 1 },
6792 	{ "ERssTcpUdpPkt", 1, 1 },
6793 	{ "ERssFceFipPkt", 0, 1 },
6794 	{ NULL }
6795 };
6796 
6797 static void
6798 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
6799 {
6800 
6801 	field_desc_show(sb, *p, tp_la0);
6802 }
6803 
6804 static void
6805 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
6806 {
6807 
6808 	if (idx)
6809 		sbuf_printf(sb, "\n");
6810 	field_desc_show(sb, p[0], tp_la0);
6811 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
6812 		field_desc_show(sb, p[1], tp_la0);
6813 }
6814 
6815 static void
6816 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
6817 {
6818 
6819 	if (idx)
6820 		sbuf_printf(sb, "\n");
6821 	field_desc_show(sb, p[0], tp_la0);
6822 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
6823 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
6824 }
6825 
6826 static int
6827 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
6828 {
6829 	struct adapter *sc = arg1;
6830 	struct sbuf *sb;
6831 	uint64_t *buf, *p;
6832 	int rc;
6833 	u_int i, inc;
6834 	void (*show_func)(struct sbuf *, uint64_t *, int);
6835 
6836 	rc = sysctl_wire_old_buffer(req, 0);
6837 	if (rc != 0)
6838 		return (rc);
6839 
6840 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6841 	if (sb == NULL)
6842 		return (ENOMEM);
6843 
6844 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
6845 
6846 	t4_tp_read_la(sc, buf, NULL);
6847 	p = buf;
6848 
6849 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
6850 	case 2:
6851 		inc = 2;
6852 		show_func = tp_la_show2;
6853 		break;
6854 	case 3:
6855 		inc = 2;
6856 		show_func = tp_la_show3;
6857 		break;
6858 	default:
6859 		inc = 1;
6860 		show_func = tp_la_show;
6861 	}
6862 
6863 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
6864 		(*show_func)(sb, p, i);
6865 
6866 	rc = sbuf_finish(sb);
6867 	sbuf_delete(sb);
6868 	free(buf, M_CXGBE);
6869 	return (rc);
6870 }
6871 
6872 static int
6873 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
6874 {
6875 	struct adapter *sc = arg1;
6876 	struct sbuf *sb;
6877 	int rc;
6878 	u64 nrate[NCHAN], orate[NCHAN];
6879 
6880 	rc = sysctl_wire_old_buffer(req, 0);
6881 	if (rc != 0)
6882 		return (rc);
6883 
6884 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6885 	if (sb == NULL)
6886 		return (ENOMEM);
6887 
6888 	t4_get_chan_txrate(sc, nrate, orate);
6889 	sbuf_printf(sb, "              channel 0   channel 1   channel 2   "
6890 		 "channel 3\n");
6891 	sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
6892 	    nrate[0], nrate[1], nrate[2], nrate[3]);
6893 	sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
6894 	    orate[0], orate[1], orate[2], orate[3]);
6895 
6896 	rc = sbuf_finish(sb);
6897 	sbuf_delete(sb);
6898 
6899 	return (rc);
6900 }
6901 
6902 static int
6903 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
6904 {
6905 	struct adapter *sc = arg1;
6906 	struct sbuf *sb;
6907 	uint32_t *buf, *p;
6908 	int rc, i;
6909 
6910 	rc = sysctl_wire_old_buffer(req, 0);
6911 	if (rc != 0)
6912 		return (rc);
6913 
6914 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6915 	if (sb == NULL)
6916 		return (ENOMEM);
6917 
6918 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
6919 	    M_ZERO | M_WAITOK);
6920 
6921 	t4_ulprx_read_la(sc, buf);
6922 	p = buf;
6923 
6924 	sbuf_printf(sb, "      Pcmd        Type   Message"
6925 	    "                Data");
6926 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
6927 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
6928 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
6929 	}
6930 
6931 	rc = sbuf_finish(sb);
6932 	sbuf_delete(sb);
6933 	free(buf, M_CXGBE);
6934 	return (rc);
6935 }
6936 
6937 static int
6938 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
6939 {
6940 	struct adapter *sc = arg1;
6941 	struct sbuf *sb;
6942 	int rc, v;
6943 
6944 	rc = sysctl_wire_old_buffer(req, 0);
6945 	if (rc != 0)
6946 		return (rc);
6947 
6948 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6949 	if (sb == NULL)
6950 		return (ENOMEM);
6951 
6952 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
6953 	if (G_STATSOURCE_T5(v) == 7) {
6954 		if (G_STATMODE(v) == 0) {
6955 			sbuf_printf(sb, "total %d, incomplete %d",
6956 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
6957 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
6958 		} else if (G_STATMODE(v) == 1) {
6959 			sbuf_printf(sb, "total %d, data overflow %d",
6960 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
6961 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
6962 		}
6963 	}
6964 	rc = sbuf_finish(sb);
6965 	sbuf_delete(sb);
6966 
6967 	return (rc);
6968 }
6969 #endif
6970 
6971 static uint32_t
6972 fconf_to_mode(uint32_t fconf)
6973 {
6974 	uint32_t mode;
6975 
6976 	mode = T4_FILTER_IPv4 | T4_FILTER_IPv6 | T4_FILTER_IP_SADDR |
6977 	    T4_FILTER_IP_DADDR | T4_FILTER_IP_SPORT | T4_FILTER_IP_DPORT;
6978 
6979 	if (fconf & F_FRAGMENTATION)
6980 		mode |= T4_FILTER_IP_FRAGMENT;
6981 
6982 	if (fconf & F_MPSHITTYPE)
6983 		mode |= T4_FILTER_MPS_HIT_TYPE;
6984 
6985 	if (fconf & F_MACMATCH)
6986 		mode |= T4_FILTER_MAC_IDX;
6987 
6988 	if (fconf & F_ETHERTYPE)
6989 		mode |= T4_FILTER_ETH_TYPE;
6990 
6991 	if (fconf & F_PROTOCOL)
6992 		mode |= T4_FILTER_IP_PROTO;
6993 
6994 	if (fconf & F_TOS)
6995 		mode |= T4_FILTER_IP_TOS;
6996 
6997 	if (fconf & F_VLAN)
6998 		mode |= T4_FILTER_VLAN;
6999 
7000 	if (fconf & F_VNIC_ID)
7001 		mode |= T4_FILTER_VNIC;
7002 
7003 	if (fconf & F_PORT)
7004 		mode |= T4_FILTER_PORT;
7005 
7006 	if (fconf & F_FCOE)
7007 		mode |= T4_FILTER_FCoE;
7008 
7009 	return (mode);
7010 }
7011 
7012 static uint32_t
7013 mode_to_fconf(uint32_t mode)
7014 {
7015 	uint32_t fconf = 0;
7016 
7017 	if (mode & T4_FILTER_IP_FRAGMENT)
7018 		fconf |= F_FRAGMENTATION;
7019 
7020 	if (mode & T4_FILTER_MPS_HIT_TYPE)
7021 		fconf |= F_MPSHITTYPE;
7022 
7023 	if (mode & T4_FILTER_MAC_IDX)
7024 		fconf |= F_MACMATCH;
7025 
7026 	if (mode & T4_FILTER_ETH_TYPE)
7027 		fconf |= F_ETHERTYPE;
7028 
7029 	if (mode & T4_FILTER_IP_PROTO)
7030 		fconf |= F_PROTOCOL;
7031 
7032 	if (mode & T4_FILTER_IP_TOS)
7033 		fconf |= F_TOS;
7034 
7035 	if (mode & T4_FILTER_VLAN)
7036 		fconf |= F_VLAN;
7037 
7038 	if (mode & T4_FILTER_VNIC)
7039 		fconf |= F_VNIC_ID;
7040 
7041 	if (mode & T4_FILTER_PORT)
7042 		fconf |= F_PORT;
7043 
7044 	if (mode & T4_FILTER_FCoE)
7045 		fconf |= F_FCOE;
7046 
7047 	return (fconf);
7048 }
7049 
7050 static uint32_t
7051 fspec_to_fconf(struct t4_filter_specification *fs)
7052 {
7053 	uint32_t fconf = 0;
7054 
7055 	if (fs->val.frag || fs->mask.frag)
7056 		fconf |= F_FRAGMENTATION;
7057 
7058 	if (fs->val.matchtype || fs->mask.matchtype)
7059 		fconf |= F_MPSHITTYPE;
7060 
7061 	if (fs->val.macidx || fs->mask.macidx)
7062 		fconf |= F_MACMATCH;
7063 
7064 	if (fs->val.ethtype || fs->mask.ethtype)
7065 		fconf |= F_ETHERTYPE;
7066 
7067 	if (fs->val.proto || fs->mask.proto)
7068 		fconf |= F_PROTOCOL;
7069 
7070 	if (fs->val.tos || fs->mask.tos)
7071 		fconf |= F_TOS;
7072 
7073 	if (fs->val.vlan_vld || fs->mask.vlan_vld)
7074 		fconf |= F_VLAN;
7075 
7076 	if (fs->val.vnic_vld || fs->mask.vnic_vld)
7077 		fconf |= F_VNIC_ID;
7078 
7079 	if (fs->val.iport || fs->mask.iport)
7080 		fconf |= F_PORT;
7081 
7082 	if (fs->val.fcoe || fs->mask.fcoe)
7083 		fconf |= F_FCOE;
7084 
7085 	return (fconf);
7086 }
7087 
7088 static int
7089 get_filter_mode(struct adapter *sc, uint32_t *mode)
7090 {
7091 	int rc;
7092 	uint32_t fconf;
7093 
7094 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7095 	    "t4getfm");
7096 	if (rc)
7097 		return (rc);
7098 
7099 	t4_read_indirect(sc, A_TP_PIO_ADDR, A_TP_PIO_DATA, &fconf, 1,
7100 	    A_TP_VLAN_PRI_MAP);
7101 
7102 	if (sc->params.tp.vlan_pri_map != fconf) {
7103 		log(LOG_WARNING, "%s: cached filter mode out of sync %x %x.\n",
7104 		    device_get_nameunit(sc->dev), sc->params.tp.vlan_pri_map,
7105 		    fconf);
7106 		sc->params.tp.vlan_pri_map = fconf;
7107 	}
7108 
7109 	*mode = fconf_to_mode(sc->params.tp.vlan_pri_map);
7110 
7111 	end_synchronized_op(sc, LOCK_HELD);
7112 	return (0);
7113 }
7114 
7115 static int
7116 set_filter_mode(struct adapter *sc, uint32_t mode)
7117 {
7118 	uint32_t fconf;
7119 	int rc;
7120 
7121 	fconf = mode_to_fconf(mode);
7122 
7123 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7124 	    "t4setfm");
7125 	if (rc)
7126 		return (rc);
7127 
7128 	if (sc->tids.ftids_in_use > 0) {
7129 		rc = EBUSY;
7130 		goto done;
7131 	}
7132 
7133 #ifdef TCP_OFFLOAD
7134 	if (sc->offload_map) {
7135 		rc = EBUSY;
7136 		goto done;
7137 	}
7138 #endif
7139 
7140 #ifdef notyet
7141 	rc = -t4_set_filter_mode(sc, fconf);
7142 	if (rc == 0)
7143 		sc->filter_mode = fconf;
7144 #else
7145 	rc = ENOTSUP;
7146 #endif
7147 
7148 done:
7149 	end_synchronized_op(sc, LOCK_HELD);
7150 	return (rc);
7151 }
7152 
7153 static inline uint64_t
7154 get_filter_hits(struct adapter *sc, uint32_t fid)
7155 {
7156 	uint32_t mw_base, off, tcb_base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
7157 	uint64_t hits;
7158 
7159 	memwin_info(sc, 0, &mw_base, NULL);
7160 	off = position_memwin(sc, 0,
7161 	    tcb_base + (fid + sc->tids.ftid_base) * TCB_SIZE);
7162 	if (is_t4(sc)) {
7163 		hits = t4_read_reg64(sc, mw_base + off + 16);
7164 		hits = be64toh(hits);
7165 	} else {
7166 		hits = t4_read_reg(sc, mw_base + off + 24);
7167 		hits = be32toh(hits);
7168 	}
7169 
7170 	return (hits);
7171 }
7172 
7173 static int
7174 get_filter(struct adapter *sc, struct t4_filter *t)
7175 {
7176 	int i, rc, nfilters = sc->tids.nftids;
7177 	struct filter_entry *f;
7178 
7179 	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7180 	    "t4getf");
7181 	if (rc)
7182 		return (rc);
7183 
7184 	if (sc->tids.ftids_in_use == 0 || sc->tids.ftid_tab == NULL ||
7185 	    t->idx >= nfilters) {
7186 		t->idx = 0xffffffff;
7187 		goto done;
7188 	}
7189 
7190 	f = &sc->tids.ftid_tab[t->idx];
7191 	for (i = t->idx; i < nfilters; i++, f++) {
7192 		if (f->valid) {
7193 			t->idx = i;
7194 			t->l2tidx = f->l2t ? f->l2t->idx : 0;
7195 			t->smtidx = f->smtidx;
7196 			if (f->fs.hitcnts)
7197 				t->hits = get_filter_hits(sc, t->idx);
7198 			else
7199 				t->hits = UINT64_MAX;
7200 			t->fs = f->fs;
7201 
7202 			goto done;
7203 		}
7204 	}
7205 
7206 	t->idx = 0xffffffff;
7207 done:
7208 	end_synchronized_op(sc, LOCK_HELD);
7209 	return (0);
7210 }
7211 
7212 static int
7213 set_filter(struct adapter *sc, struct t4_filter *t)
7214 {
7215 	unsigned int nfilters, nports;
7216 	struct filter_entry *f;
7217 	int i, rc;
7218 
7219 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setf");
7220 	if (rc)
7221 		return (rc);
7222 
7223 	nfilters = sc->tids.nftids;
7224 	nports = sc->params.nports;
7225 
7226 	if (nfilters == 0) {
7227 		rc = ENOTSUP;
7228 		goto done;
7229 	}
7230 
7231 	if (!(sc->flags & FULL_INIT_DONE)) {
7232 		rc = EAGAIN;
7233 		goto done;
7234 	}
7235 
7236 	if (t->idx >= nfilters) {
7237 		rc = EINVAL;
7238 		goto done;
7239 	}
7240 
7241 	/* Validate against the global filter mode */
7242 	if ((sc->params.tp.vlan_pri_map | fspec_to_fconf(&t->fs)) !=
7243 	    sc->params.tp.vlan_pri_map) {
7244 		rc = E2BIG;
7245 		goto done;
7246 	}
7247 
7248 	if (t->fs.action == FILTER_SWITCH && t->fs.eport >= nports) {
7249 		rc = EINVAL;
7250 		goto done;
7251 	}
7252 
7253 	if (t->fs.val.iport >= nports) {
7254 		rc = EINVAL;
7255 		goto done;
7256 	}
7257 
7258 	/* Can't specify an iq if not steering to it */
7259 	if (!t->fs.dirsteer && t->fs.iq) {
7260 		rc = EINVAL;
7261 		goto done;
7262 	}
7263 
7264 	/* IPv6 filter idx must be 4 aligned */
7265 	if (t->fs.type == 1 &&
7266 	    ((t->idx & 0x3) || t->idx + 4 >= nfilters)) {
7267 		rc = EINVAL;
7268 		goto done;
7269 	}
7270 
7271 	if (sc->tids.ftid_tab == NULL) {
7272 		KASSERT(sc->tids.ftids_in_use == 0,
7273 		    ("%s: no memory allocated but filters_in_use > 0",
7274 		    __func__));
7275 
7276 		sc->tids.ftid_tab = malloc(sizeof (struct filter_entry) *
7277 		    nfilters, M_CXGBE, M_NOWAIT | M_ZERO);
7278 		if (sc->tids.ftid_tab == NULL) {
7279 			rc = ENOMEM;
7280 			goto done;
7281 		}
7282 		mtx_init(&sc->tids.ftid_lock, "T4 filters", 0, MTX_DEF);
7283 	}
7284 
7285 	for (i = 0; i < 4; i++) {
7286 		f = &sc->tids.ftid_tab[t->idx + i];
7287 
7288 		if (f->pending || f->valid) {
7289 			rc = EBUSY;
7290 			goto done;
7291 		}
7292 		if (f->locked) {
7293 			rc = EPERM;
7294 			goto done;
7295 		}
7296 
7297 		if (t->fs.type == 0)
7298 			break;
7299 	}
7300 
7301 	f = &sc->tids.ftid_tab[t->idx];
7302 	f->fs = t->fs;
7303 
7304 	rc = set_filter_wr(sc, t->idx);
7305 done:
7306 	end_synchronized_op(sc, 0);
7307 
7308 	if (rc == 0) {
7309 		mtx_lock(&sc->tids.ftid_lock);
7310 		for (;;) {
7311 			if (f->pending == 0) {
7312 				rc = f->valid ? 0 : EIO;
7313 				break;
7314 			}
7315 
7316 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
7317 			    PCATCH, "t4setfw", 0)) {
7318 				rc = EINPROGRESS;
7319 				break;
7320 			}
7321 		}
7322 		mtx_unlock(&sc->tids.ftid_lock);
7323 	}
7324 	return (rc);
7325 }
7326 
7327 static int
7328 del_filter(struct adapter *sc, struct t4_filter *t)
7329 {
7330 	unsigned int nfilters;
7331 	struct filter_entry *f;
7332 	int rc;
7333 
7334 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4delf");
7335 	if (rc)
7336 		return (rc);
7337 
7338 	nfilters = sc->tids.nftids;
7339 
7340 	if (nfilters == 0) {
7341 		rc = ENOTSUP;
7342 		goto done;
7343 	}
7344 
7345 	if (sc->tids.ftid_tab == NULL || sc->tids.ftids_in_use == 0 ||
7346 	    t->idx >= nfilters) {
7347 		rc = EINVAL;
7348 		goto done;
7349 	}
7350 
7351 	if (!(sc->flags & FULL_INIT_DONE)) {
7352 		rc = EAGAIN;
7353 		goto done;
7354 	}
7355 
7356 	f = &sc->tids.ftid_tab[t->idx];
7357 
7358 	if (f->pending) {
7359 		rc = EBUSY;
7360 		goto done;
7361 	}
7362 	if (f->locked) {
7363 		rc = EPERM;
7364 		goto done;
7365 	}
7366 
7367 	if (f->valid) {
7368 		t->fs = f->fs;	/* extra info for the caller */
7369 		rc = del_filter_wr(sc, t->idx);
7370 	}
7371 
7372 done:
7373 	end_synchronized_op(sc, 0);
7374 
7375 	if (rc == 0) {
7376 		mtx_lock(&sc->tids.ftid_lock);
7377 		for (;;) {
7378 			if (f->pending == 0) {
7379 				rc = f->valid ? EIO : 0;
7380 				break;
7381 			}
7382 
7383 			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
7384 			    PCATCH, "t4delfw", 0)) {
7385 				rc = EINPROGRESS;
7386 				break;
7387 			}
7388 		}
7389 		mtx_unlock(&sc->tids.ftid_lock);
7390 	}
7391 
7392 	return (rc);
7393 }
7394 
7395 static void
7396 clear_filter(struct filter_entry *f)
7397 {
7398 	if (f->l2t)
7399 		t4_l2t_release(f->l2t);
7400 
7401 	bzero(f, sizeof (*f));
7402 }
7403 
7404 static int
7405 set_filter_wr(struct adapter *sc, int fidx)
7406 {
7407 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
7408 	struct fw_filter_wr *fwr;
7409 	unsigned int ftid;
7410 	struct wrq_cookie cookie;
7411 
7412 	ASSERT_SYNCHRONIZED_OP(sc);
7413 
7414 	if (f->fs.newdmac || f->fs.newvlan) {
7415 		/* This filter needs an L2T entry; allocate one. */
7416 		f->l2t = t4_l2t_alloc_switching(sc->l2t);
7417 		if (f->l2t == NULL)
7418 			return (EAGAIN);
7419 		if (t4_l2t_set_switching(sc, f->l2t, f->fs.vlan, f->fs.eport,
7420 		    f->fs.dmac)) {
7421 			t4_l2t_release(f->l2t);
7422 			f->l2t = NULL;
7423 			return (ENOMEM);
7424 		}
7425 	}
7426 
7427 	ftid = sc->tids.ftid_base + fidx;
7428 
7429 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
7430 	if (fwr == NULL)
7431 		return (ENOMEM);
7432 	bzero(fwr, sizeof(*fwr));
7433 
7434 	fwr->op_pkd = htobe32(V_FW_WR_OP(FW_FILTER_WR));
7435 	fwr->len16_pkd = htobe32(FW_LEN16(*fwr));
7436 	fwr->tid_to_iq =
7437 	    htobe32(V_FW_FILTER_WR_TID(ftid) |
7438 		V_FW_FILTER_WR_RQTYPE(f->fs.type) |
7439 		V_FW_FILTER_WR_NOREPLY(0) |
7440 		V_FW_FILTER_WR_IQ(f->fs.iq));
7441 	fwr->del_filter_to_l2tix =
7442 	    htobe32(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
7443 		V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
7444 		V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
7445 		V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
7446 		V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
7447 		V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
7448 		V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
7449 		V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
7450 		V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
7451 		    f->fs.newvlan == VLAN_REWRITE) |
7452 		V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
7453 		    f->fs.newvlan == VLAN_REWRITE) |
7454 		V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
7455 		V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
7456 		V_FW_FILTER_WR_PRIO(f->fs.prio) |
7457 		V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
7458 	fwr->ethtype = htobe16(f->fs.val.ethtype);
7459 	fwr->ethtypem = htobe16(f->fs.mask.ethtype);
7460 	fwr->frag_to_ovlan_vldm =
7461 	    (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
7462 		V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
7463 		V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.vlan_vld) |
7464 		V_FW_FILTER_WR_OVLAN_VLD(f->fs.val.vnic_vld) |
7465 		V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.vlan_vld) |
7466 		V_FW_FILTER_WR_OVLAN_VLDM(f->fs.mask.vnic_vld));
7467 	fwr->smac_sel = 0;
7468 	fwr->rx_chan_rx_rpl_iq = htobe16(V_FW_FILTER_WR_RX_CHAN(0) |
7469 	    V_FW_FILTER_WR_RX_RPL_IQ(sc->sge.fwq.abs_id));
7470 	fwr->maci_to_matchtypem =
7471 	    htobe32(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
7472 		V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
7473 		V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
7474 		V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
7475 		V_FW_FILTER_WR_PORT(f->fs.val.iport) |
7476 		V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
7477 		V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
7478 		V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
7479 	fwr->ptcl = f->fs.val.proto;
7480 	fwr->ptclm = f->fs.mask.proto;
7481 	fwr->ttyp = f->fs.val.tos;
7482 	fwr->ttypm = f->fs.mask.tos;
7483 	fwr->ivlan = htobe16(f->fs.val.vlan);
7484 	fwr->ivlanm = htobe16(f->fs.mask.vlan);
7485 	fwr->ovlan = htobe16(f->fs.val.vnic);
7486 	fwr->ovlanm = htobe16(f->fs.mask.vnic);
7487 	bcopy(f->fs.val.dip, fwr->lip, sizeof (fwr->lip));
7488 	bcopy(f->fs.mask.dip, fwr->lipm, sizeof (fwr->lipm));
7489 	bcopy(f->fs.val.sip, fwr->fip, sizeof (fwr->fip));
7490 	bcopy(f->fs.mask.sip, fwr->fipm, sizeof (fwr->fipm));
7491 	fwr->lp = htobe16(f->fs.val.dport);
7492 	fwr->lpm = htobe16(f->fs.mask.dport);
7493 	fwr->fp = htobe16(f->fs.val.sport);
7494 	fwr->fpm = htobe16(f->fs.mask.sport);
7495 	if (f->fs.newsmac)
7496 		bcopy(f->fs.smac, fwr->sma, sizeof (fwr->sma));
7497 
7498 	f->pending = 1;
7499 	sc->tids.ftids_in_use++;
7500 
7501 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
7502 	return (0);
7503 }
7504 
7505 static int
7506 del_filter_wr(struct adapter *sc, int fidx)
7507 {
7508 	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
7509 	struct fw_filter_wr *fwr;
7510 	unsigned int ftid;
7511 	struct wrq_cookie cookie;
7512 
7513 	ftid = sc->tids.ftid_base + fidx;
7514 
7515 	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
7516 	if (fwr == NULL)
7517 		return (ENOMEM);
7518 	bzero(fwr, sizeof (*fwr));
7519 
7520 	t4_mk_filtdelwr(ftid, fwr, sc->sge.fwq.abs_id);
7521 
7522 	f->pending = 1;
7523 	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
7524 	return (0);
7525 }
7526 
7527 int
7528 t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
7529 {
7530 	struct adapter *sc = iq->adapter;
7531 	const struct cpl_set_tcb_rpl *rpl = (const void *)(rss + 1);
7532 	unsigned int idx = GET_TID(rpl);
7533 	unsigned int rc;
7534 	struct filter_entry *f;
7535 
7536 	KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
7537 	    rss->opcode));
7538 
7539 	if (is_ftid(sc, idx)) {
7540 
7541 		idx -= sc->tids.ftid_base;
7542 		f = &sc->tids.ftid_tab[idx];
7543 		rc = G_COOKIE(rpl->cookie);
7544 
7545 		mtx_lock(&sc->tids.ftid_lock);
7546 		if (rc == FW_FILTER_WR_FLT_ADDED) {
7547 			KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
7548 			    __func__, idx));
7549 			f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
7550 			f->pending = 0;  /* asynchronous setup completed */
7551 			f->valid = 1;
7552 		} else {
7553 			if (rc != FW_FILTER_WR_FLT_DELETED) {
7554 				/* Add or delete failed, display an error */
7555 				log(LOG_ERR,
7556 				    "filter %u setup failed with error %u\n",
7557 				    idx, rc);
7558 			}
7559 
7560 			clear_filter(f);
7561 			sc->tids.ftids_in_use--;
7562 		}
7563 		wakeup(&sc->tids.ftid_tab);
7564 		mtx_unlock(&sc->tids.ftid_lock);
7565 	}
7566 
7567 	return (0);
7568 }
7569 
7570 static int
7571 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
7572 {
7573 	int rc;
7574 
7575 	if (cntxt->cid > M_CTXTQID)
7576 		return (EINVAL);
7577 
7578 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
7579 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
7580 		return (EINVAL);
7581 
7582 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
7583 	if (rc)
7584 		return (rc);
7585 
7586 	if (sc->flags & FW_OK) {
7587 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
7588 		    &cntxt->data[0]);
7589 		if (rc == 0)
7590 			goto done;
7591 	}
7592 
7593 	/*
7594 	 * Read via firmware failed or wasn't even attempted.  Read directly via
7595 	 * the backdoor.
7596 	 */
7597 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
7598 done:
7599 	end_synchronized_op(sc, 0);
7600 	return (rc);
7601 }
7602 
7603 static int
7604 load_fw(struct adapter *sc, struct t4_data *fw)
7605 {
7606 	int rc;
7607 	uint8_t *fw_data;
7608 
7609 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
7610 	if (rc)
7611 		return (rc);
7612 
7613 	if (sc->flags & FULL_INIT_DONE) {
7614 		rc = EBUSY;
7615 		goto done;
7616 	}
7617 
7618 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
7619 	if (fw_data == NULL) {
7620 		rc = ENOMEM;
7621 		goto done;
7622 	}
7623 
7624 	rc = copyin(fw->data, fw_data, fw->len);
7625 	if (rc == 0)
7626 		rc = -t4_load_fw(sc, fw_data, fw->len);
7627 
7628 	free(fw_data, M_CXGBE);
7629 done:
7630 	end_synchronized_op(sc, 0);
7631 	return (rc);
7632 }
7633 
7634 static int
7635 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
7636 {
7637 	uint32_t addr, off, remaining, i, n;
7638 	uint32_t *buf, *b;
7639 	uint32_t mw_base, mw_aperture;
7640 	int rc;
7641 	uint8_t *dst;
7642 
7643 	rc = validate_mem_range(sc, mr->addr, mr->len);
7644 	if (rc != 0)
7645 		return (rc);
7646 
7647 	memwin_info(sc, win, &mw_base, &mw_aperture);
7648 	buf = b = malloc(min(mr->len, mw_aperture), M_CXGBE, M_WAITOK);
7649 	addr = mr->addr;
7650 	remaining = mr->len;
7651 	dst = (void *)mr->data;
7652 
7653 	while (remaining) {
7654 		off = position_memwin(sc, win, addr);
7655 
7656 		/* number of bytes that we'll copy in the inner loop */
7657 		n = min(remaining, mw_aperture - off);
7658 		for (i = 0; i < n; i += 4)
7659 			*b++ = t4_read_reg(sc, mw_base + off + i);
7660 
7661 		rc = copyout(buf, dst, n);
7662 		if (rc != 0)
7663 			break;
7664 
7665 		b = buf;
7666 		dst += n;
7667 		remaining -= n;
7668 		addr += n;
7669 	}
7670 
7671 	free(buf, M_CXGBE);
7672 	return (rc);
7673 }
7674 
7675 static int
7676 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
7677 {
7678 	int rc;
7679 
7680 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
7681 		return (EINVAL);
7682 
7683 	if (i2cd->len > sizeof(i2cd->data))
7684 		return (EFBIG);
7685 
7686 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
7687 	if (rc)
7688 		return (rc);
7689 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
7690 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
7691 	end_synchronized_op(sc, 0);
7692 
7693 	return (rc);
7694 }
7695 
7696 static int
7697 in_range(int val, int lo, int hi)
7698 {
7699 
7700 	return (val < 0 || (val <= hi && val >= lo));
7701 }
7702 
7703 static int
7704 set_sched_class(struct adapter *sc, struct t4_sched_params *p)
7705 {
7706 	int fw_subcmd, fw_type, rc;
7707 
7708 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setsc");
7709 	if (rc)
7710 		return (rc);
7711 
7712 	if (!(sc->flags & FULL_INIT_DONE)) {
7713 		rc = EAGAIN;
7714 		goto done;
7715 	}
7716 
7717 	/*
7718 	 * Translate the cxgbetool parameters into T4 firmware parameters.  (The
7719 	 * sub-command and type are in common locations.)
7720 	 */
7721 	if (p->subcmd == SCHED_CLASS_SUBCMD_CONFIG)
7722 		fw_subcmd = FW_SCHED_SC_CONFIG;
7723 	else if (p->subcmd == SCHED_CLASS_SUBCMD_PARAMS)
7724 		fw_subcmd = FW_SCHED_SC_PARAMS;
7725 	else {
7726 		rc = EINVAL;
7727 		goto done;
7728 	}
7729 	if (p->type == SCHED_CLASS_TYPE_PACKET)
7730 		fw_type = FW_SCHED_TYPE_PKTSCHED;
7731 	else {
7732 		rc = EINVAL;
7733 		goto done;
7734 	}
7735 
7736 	if (fw_subcmd == FW_SCHED_SC_CONFIG) {
7737 		/* Vet our parameters ..*/
7738 		if (p->u.config.minmax < 0) {
7739 			rc = EINVAL;
7740 			goto done;
7741 		}
7742 
7743 		/* And pass the request to the firmware ...*/
7744 		rc = -t4_sched_config(sc, fw_type, p->u.config.minmax, 1);
7745 		goto done;
7746 	}
7747 
7748 	if (fw_subcmd == FW_SCHED_SC_PARAMS) {
7749 		int fw_level;
7750 		int fw_mode;
7751 		int fw_rateunit;
7752 		int fw_ratemode;
7753 
7754 		if (p->u.params.level == SCHED_CLASS_LEVEL_CL_RL)
7755 			fw_level = FW_SCHED_PARAMS_LEVEL_CL_RL;
7756 		else if (p->u.params.level == SCHED_CLASS_LEVEL_CL_WRR)
7757 			fw_level = FW_SCHED_PARAMS_LEVEL_CL_WRR;
7758 		else if (p->u.params.level == SCHED_CLASS_LEVEL_CH_RL)
7759 			fw_level = FW_SCHED_PARAMS_LEVEL_CH_RL;
7760 		else {
7761 			rc = EINVAL;
7762 			goto done;
7763 		}
7764 
7765 		if (p->u.params.mode == SCHED_CLASS_MODE_CLASS)
7766 			fw_mode = FW_SCHED_PARAMS_MODE_CLASS;
7767 		else if (p->u.params.mode == SCHED_CLASS_MODE_FLOW)
7768 			fw_mode = FW_SCHED_PARAMS_MODE_FLOW;
7769 		else {
7770 			rc = EINVAL;
7771 			goto done;
7772 		}
7773 
7774 		if (p->u.params.rateunit == SCHED_CLASS_RATEUNIT_BITS)
7775 			fw_rateunit = FW_SCHED_PARAMS_UNIT_BITRATE;
7776 		else if (p->u.params.rateunit == SCHED_CLASS_RATEUNIT_PKTS)
7777 			fw_rateunit = FW_SCHED_PARAMS_UNIT_PKTRATE;
7778 		else {
7779 			rc = EINVAL;
7780 			goto done;
7781 		}
7782 
7783 		if (p->u.params.ratemode == SCHED_CLASS_RATEMODE_REL)
7784 			fw_ratemode = FW_SCHED_PARAMS_RATE_REL;
7785 		else if (p->u.params.ratemode == SCHED_CLASS_RATEMODE_ABS)
7786 			fw_ratemode = FW_SCHED_PARAMS_RATE_ABS;
7787 		else {
7788 			rc = EINVAL;
7789 			goto done;
7790 		}
7791 
7792 		/* Vet our parameters ... */
7793 		if (!in_range(p->u.params.channel, 0, 3) ||
7794 		    !in_range(p->u.params.cl, 0, is_t4(sc) ? 15 : 16) ||
7795 		    !in_range(p->u.params.minrate, 0, 10000000) ||
7796 		    !in_range(p->u.params.maxrate, 0, 10000000) ||
7797 		    !in_range(p->u.params.weight, 0, 100)) {
7798 			rc = ERANGE;
7799 			goto done;
7800 		}
7801 
7802 		/*
7803 		 * Translate any unset parameters into the firmware's
7804 		 * nomenclature and/or fail the call if the parameters
7805 		 * are required ...
7806 		 */
7807 		if (p->u.params.rateunit < 0 || p->u.params.ratemode < 0 ||
7808 		    p->u.params.channel < 0 || p->u.params.cl < 0) {
7809 			rc = EINVAL;
7810 			goto done;
7811 		}
7812 		if (p->u.params.minrate < 0)
7813 			p->u.params.minrate = 0;
7814 		if (p->u.params.maxrate < 0) {
7815 			if (p->u.params.level == SCHED_CLASS_LEVEL_CL_RL ||
7816 			    p->u.params.level == SCHED_CLASS_LEVEL_CH_RL) {
7817 				rc = EINVAL;
7818 				goto done;
7819 			} else
7820 				p->u.params.maxrate = 0;
7821 		}
7822 		if (p->u.params.weight < 0) {
7823 			if (p->u.params.level == SCHED_CLASS_LEVEL_CL_WRR) {
7824 				rc = EINVAL;
7825 				goto done;
7826 			} else
7827 				p->u.params.weight = 0;
7828 		}
7829 		if (p->u.params.pktsize < 0) {
7830 			if (p->u.params.level == SCHED_CLASS_LEVEL_CL_RL ||
7831 			    p->u.params.level == SCHED_CLASS_LEVEL_CH_RL) {
7832 				rc = EINVAL;
7833 				goto done;
7834 			} else
7835 				p->u.params.pktsize = 0;
7836 		}
7837 
7838 		/* See what the firmware thinks of the request ... */
7839 		rc = -t4_sched_params(sc, fw_type, fw_level, fw_mode,
7840 		    fw_rateunit, fw_ratemode, p->u.params.channel,
7841 		    p->u.params.cl, p->u.params.minrate, p->u.params.maxrate,
7842 		    p->u.params.weight, p->u.params.pktsize, 1);
7843 		goto done;
7844 	}
7845 
7846 	rc = EINVAL;
7847 done:
7848 	end_synchronized_op(sc, 0);
7849 	return (rc);
7850 }
7851 
7852 static int
7853 set_sched_queue(struct adapter *sc, struct t4_sched_queue *p)
7854 {
7855 	struct port_info *pi = NULL;
7856 	struct sge_txq *txq;
7857 	uint32_t fw_mnem, fw_queue, fw_class;
7858 	int i, rc;
7859 
7860 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setsq");
7861 	if (rc)
7862 		return (rc);
7863 
7864 	if (!(sc->flags & FULL_INIT_DONE)) {
7865 		rc = EAGAIN;
7866 		goto done;
7867 	}
7868 
7869 	if (p->port >= sc->params.nports) {
7870 		rc = EINVAL;
7871 		goto done;
7872 	}
7873 
7874 	pi = sc->port[p->port];
7875 	if (!in_range(p->queue, 0, pi->ntxq - 1) || !in_range(p->cl, 0, 7)) {
7876 		rc = EINVAL;
7877 		goto done;
7878 	}
7879 
7880 	/*
7881 	 * Create a template for the FW_PARAMS_CMD mnemonic and value (TX
7882 	 * Scheduling Class in this case).
7883 	 */
7884 	fw_mnem = (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
7885 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_EQ_SCHEDCLASS_ETH));
7886 	fw_class = p->cl < 0 ? 0xffffffff : p->cl;
7887 
7888 	/*
7889 	 * If op.queue is non-negative, then we're only changing the scheduling
7890 	 * on a single specified TX queue.
7891 	 */
7892 	if (p->queue >= 0) {
7893 		txq = &sc->sge.txq[pi->first_txq + p->queue];
7894 		fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
7895 		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
7896 		    &fw_class);
7897 		goto done;
7898 	}
7899 
7900 	/*
7901 	 * Change the scheduling on all the TX queues for the
7902 	 * interface.
7903 	 */
7904 	for_each_txq(pi, i, txq) {
7905 		fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
7906 		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
7907 		    &fw_class);
7908 		if (rc)
7909 			goto done;
7910 	}
7911 
7912 	rc = 0;
7913 done:
7914 	end_synchronized_op(sc, 0);
7915 	return (rc);
7916 }
7917 
7918 int
7919 t4_os_find_pci_capability(struct adapter *sc, int cap)
7920 {
7921 	int i;
7922 
7923 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
7924 }
7925 
7926 int
7927 t4_os_pci_save_state(struct adapter *sc)
7928 {
7929 	device_t dev;
7930 	struct pci_devinfo *dinfo;
7931 
7932 	dev = sc->dev;
7933 	dinfo = device_get_ivars(dev);
7934 
7935 	pci_cfg_save(dev, dinfo, 0);
7936 	return (0);
7937 }
7938 
7939 int
7940 t4_os_pci_restore_state(struct adapter *sc)
7941 {
7942 	device_t dev;
7943 	struct pci_devinfo *dinfo;
7944 
7945 	dev = sc->dev;
7946 	dinfo = device_get_ivars(dev);
7947 
7948 	pci_cfg_restore(dev, dinfo);
7949 	return (0);
7950 }
7951 
7952 void
7953 t4_os_portmod_changed(const struct adapter *sc, int idx)
7954 {
7955 	struct port_info *pi = sc->port[idx];
7956 	static const char *mod_str[] = {
7957 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
7958 	};
7959 
7960 	build_medialist(pi, &pi->media);
7961 #ifdef DEV_NETMAP
7962 	build_medialist(pi, &pi->nm_media);
7963 #endif
7964 
7965 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
7966 		if_printf(pi->ifp, "transceiver unplugged.\n");
7967 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
7968 		if_printf(pi->ifp, "unknown transceiver inserted.\n");
7969 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
7970 		if_printf(pi->ifp, "unsupported transceiver inserted.\n");
7971 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
7972 		if_printf(pi->ifp, "%s transceiver inserted.\n",
7973 		    mod_str[pi->mod_type]);
7974 	} else {
7975 		if_printf(pi->ifp, "transceiver (type %d) inserted.\n",
7976 		    pi->mod_type);
7977 	}
7978 }
7979 
7980 void
7981 t4_os_link_changed(struct adapter *sc, int idx, int link_stat, int reason)
7982 {
7983 	struct port_info *pi = sc->port[idx];
7984 	struct ifnet *ifp = pi->ifp;
7985 
7986 	if (link_stat) {
7987 		pi->linkdnrc = -1;
7988 		ifp->if_baudrate = IF_Mbps(pi->link_cfg.speed);
7989 		if_link_state_change(ifp, LINK_STATE_UP);
7990 	} else {
7991 		if (reason >= 0)
7992 			pi->linkdnrc = reason;
7993 		if_link_state_change(ifp, LINK_STATE_DOWN);
7994 	}
7995 }
7996 
7997 void
7998 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
7999 {
8000 	struct adapter *sc;
8001 
8002 	sx_slock(&t4_list_lock);
8003 	SLIST_FOREACH(sc, &t4_list, link) {
8004 		/*
8005 		 * func should not make any assumptions about what state sc is
8006 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
8007 		 */
8008 		func(sc, arg);
8009 	}
8010 	sx_sunlock(&t4_list_lock);
8011 }
8012 
8013 static int
8014 t4_open(struct cdev *dev, int flags, int type, struct thread *td)
8015 {
8016        return (0);
8017 }
8018 
8019 static int
8020 t4_close(struct cdev *dev, int flags, int type, struct thread *td)
8021 {
8022        return (0);
8023 }
8024 
8025 static int
8026 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
8027     struct thread *td)
8028 {
8029 	int rc;
8030 	struct adapter *sc = dev->si_drv1;
8031 
8032 	rc = priv_check(td, PRIV_DRIVER);
8033 	if (rc != 0)
8034 		return (rc);
8035 
8036 	switch (cmd) {
8037 	case CHELSIO_T4_GETREG: {
8038 		struct t4_reg *edata = (struct t4_reg *)data;
8039 
8040 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
8041 			return (EFAULT);
8042 
8043 		if (edata->size == 4)
8044 			edata->val = t4_read_reg(sc, edata->addr);
8045 		else if (edata->size == 8)
8046 			edata->val = t4_read_reg64(sc, edata->addr);
8047 		else
8048 			return (EINVAL);
8049 
8050 		break;
8051 	}
8052 	case CHELSIO_T4_SETREG: {
8053 		struct t4_reg *edata = (struct t4_reg *)data;
8054 
8055 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
8056 			return (EFAULT);
8057 
8058 		if (edata->size == 4) {
8059 			if (edata->val & 0xffffffff00000000)
8060 				return (EINVAL);
8061 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
8062 		} else if (edata->size == 8)
8063 			t4_write_reg64(sc, edata->addr, edata->val);
8064 		else
8065 			return (EINVAL);
8066 		break;
8067 	}
8068 	case CHELSIO_T4_REGDUMP: {
8069 		struct t4_regdump *regs = (struct t4_regdump *)data;
8070 		int reglen = is_t4(sc) ? T4_REGDUMP_SIZE : T5_REGDUMP_SIZE;
8071 		uint8_t *buf;
8072 
8073 		if (regs->len < reglen) {
8074 			regs->len = reglen; /* hint to the caller */
8075 			return (ENOBUFS);
8076 		}
8077 
8078 		regs->len = reglen;
8079 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
8080 		t4_get_regs(sc, regs, buf);
8081 		rc = copyout(buf, regs->data, reglen);
8082 		free(buf, M_CXGBE);
8083 		break;
8084 	}
8085 	case CHELSIO_T4_GET_FILTER_MODE:
8086 		rc = get_filter_mode(sc, (uint32_t *)data);
8087 		break;
8088 	case CHELSIO_T4_SET_FILTER_MODE:
8089 		rc = set_filter_mode(sc, *(uint32_t *)data);
8090 		break;
8091 	case CHELSIO_T4_GET_FILTER:
8092 		rc = get_filter(sc, (struct t4_filter *)data);
8093 		break;
8094 	case CHELSIO_T4_SET_FILTER:
8095 		rc = set_filter(sc, (struct t4_filter *)data);
8096 		break;
8097 	case CHELSIO_T4_DEL_FILTER:
8098 		rc = del_filter(sc, (struct t4_filter *)data);
8099 		break;
8100 	case CHELSIO_T4_GET_SGE_CONTEXT:
8101 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
8102 		break;
8103 	case CHELSIO_T4_LOAD_FW:
8104 		rc = load_fw(sc, (struct t4_data *)data);
8105 		break;
8106 	case CHELSIO_T4_GET_MEM:
8107 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
8108 		break;
8109 	case CHELSIO_T4_GET_I2C:
8110 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
8111 		break;
8112 	case CHELSIO_T4_CLEAR_STATS: {
8113 		int i;
8114 		u_int port_id = *(uint32_t *)data;
8115 		struct port_info *pi;
8116 
8117 		if (port_id >= sc->params.nports)
8118 			return (EINVAL);
8119 		pi = sc->port[port_id];
8120 
8121 		/* MAC stats */
8122 		t4_clr_port_stats(sc, pi->tx_chan);
8123 		pi->tx_parse_error = 0;
8124 
8125 		if (pi->flags & PORT_INIT_DONE) {
8126 			struct sge_rxq *rxq;
8127 			struct sge_txq *txq;
8128 			struct sge_wrq *wrq;
8129 
8130 			for_each_rxq(pi, i, rxq) {
8131 #if defined(INET) || defined(INET6)
8132 				rxq->lro.lro_queued = 0;
8133 				rxq->lro.lro_flushed = 0;
8134 #endif
8135 				rxq->rxcsum = 0;
8136 				rxq->vlan_extraction = 0;
8137 			}
8138 
8139 			for_each_txq(pi, i, txq) {
8140 				txq->txcsum = 0;
8141 				txq->tso_wrs = 0;
8142 				txq->vlan_insertion = 0;
8143 				txq->imm_wrs = 0;
8144 				txq->sgl_wrs = 0;
8145 				txq->txpkt_wrs = 0;
8146 				txq->txpkts0_wrs = 0;
8147 				txq->txpkts1_wrs = 0;
8148 				txq->txpkts0_pkts = 0;
8149 				txq->txpkts1_pkts = 0;
8150 				mp_ring_reset_stats(txq->r);
8151 			}
8152 
8153 #ifdef TCP_OFFLOAD
8154 			/* nothing to clear for each ofld_rxq */
8155 
8156 			for_each_ofld_txq(pi, i, wrq) {
8157 				wrq->tx_wrs_direct = 0;
8158 				wrq->tx_wrs_copied = 0;
8159 			}
8160 #endif
8161 			wrq = &sc->sge.ctrlq[pi->port_id];
8162 			wrq->tx_wrs_direct = 0;
8163 			wrq->tx_wrs_copied = 0;
8164 		}
8165 		break;
8166 	}
8167 	case CHELSIO_T4_SCHED_CLASS:
8168 		rc = set_sched_class(sc, (struct t4_sched_params *)data);
8169 		break;
8170 	case CHELSIO_T4_SCHED_QUEUE:
8171 		rc = set_sched_queue(sc, (struct t4_sched_queue *)data);
8172 		break;
8173 	case CHELSIO_T4_GET_TRACER:
8174 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
8175 		break;
8176 	case CHELSIO_T4_SET_TRACER:
8177 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
8178 		break;
8179 	default:
8180 		rc = EINVAL;
8181 	}
8182 
8183 	return (rc);
8184 }
8185 
8186 #ifdef TCP_OFFLOAD
8187 void
8188 t4_iscsi_init(struct ifnet *ifp, unsigned int tag_mask,
8189     const unsigned int *pgsz_order)
8190 {
8191 	struct port_info *pi = ifp->if_softc;
8192 	struct adapter *sc = pi->adapter;
8193 
8194 	t4_write_reg(sc, A_ULP_RX_ISCSI_TAGMASK, tag_mask);
8195 	t4_write_reg(sc, A_ULP_RX_ISCSI_PSZ, V_HPZ0(pgsz_order[0]) |
8196 		V_HPZ1(pgsz_order[1]) | V_HPZ2(pgsz_order[2]) |
8197 		V_HPZ3(pgsz_order[3]));
8198 }
8199 
8200 static int
8201 toe_capability(struct port_info *pi, int enable)
8202 {
8203 	int rc;
8204 	struct adapter *sc = pi->adapter;
8205 
8206 	ASSERT_SYNCHRONIZED_OP(sc);
8207 
8208 	if (!is_offload(sc))
8209 		return (ENODEV);
8210 
8211 	if (enable) {
8212 		if (!(sc->flags & FULL_INIT_DONE)) {
8213 			rc = cxgbe_init_synchronized(pi);
8214 			if (rc)
8215 				return (rc);
8216 		}
8217 
8218 		if (isset(&sc->offload_map, pi->port_id))
8219 			return (0);
8220 
8221 		if (!(sc->flags & TOM_INIT_DONE)) {
8222 			rc = t4_activate_uld(sc, ULD_TOM);
8223 			if (rc == EAGAIN) {
8224 				log(LOG_WARNING,
8225 				    "You must kldload t4_tom.ko before trying "
8226 				    "to enable TOE on a cxgbe interface.\n");
8227 			}
8228 			if (rc != 0)
8229 				return (rc);
8230 			KASSERT(sc->tom_softc != NULL,
8231 			    ("%s: TOM activated but softc NULL", __func__));
8232 			KASSERT(sc->flags & TOM_INIT_DONE,
8233 			    ("%s: TOM activated but flag not set", __func__));
8234 		}
8235 
8236 		setbit(&sc->offload_map, pi->port_id);
8237 	} else {
8238 		if (!isset(&sc->offload_map, pi->port_id))
8239 			return (0);
8240 
8241 		KASSERT(sc->flags & TOM_INIT_DONE,
8242 		    ("%s: TOM never initialized?", __func__));
8243 		clrbit(&sc->offload_map, pi->port_id);
8244 	}
8245 
8246 	return (0);
8247 }
8248 
8249 /*
8250  * Add an upper layer driver to the global list.
8251  */
8252 int
8253 t4_register_uld(struct uld_info *ui)
8254 {
8255 	int rc = 0;
8256 	struct uld_info *u;
8257 
8258 	sx_xlock(&t4_uld_list_lock);
8259 	SLIST_FOREACH(u, &t4_uld_list, link) {
8260 	    if (u->uld_id == ui->uld_id) {
8261 		    rc = EEXIST;
8262 		    goto done;
8263 	    }
8264 	}
8265 
8266 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
8267 	ui->refcount = 0;
8268 done:
8269 	sx_xunlock(&t4_uld_list_lock);
8270 	return (rc);
8271 }
8272 
8273 int
8274 t4_unregister_uld(struct uld_info *ui)
8275 {
8276 	int rc = EINVAL;
8277 	struct uld_info *u;
8278 
8279 	sx_xlock(&t4_uld_list_lock);
8280 
8281 	SLIST_FOREACH(u, &t4_uld_list, link) {
8282 	    if (u == ui) {
8283 		    if (ui->refcount > 0) {
8284 			    rc = EBUSY;
8285 			    goto done;
8286 		    }
8287 
8288 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
8289 		    rc = 0;
8290 		    goto done;
8291 	    }
8292 	}
8293 done:
8294 	sx_xunlock(&t4_uld_list_lock);
8295 	return (rc);
8296 }
8297 
8298 int
8299 t4_activate_uld(struct adapter *sc, int id)
8300 {
8301 	int rc = EAGAIN;
8302 	struct uld_info *ui;
8303 
8304 	ASSERT_SYNCHRONIZED_OP(sc);
8305 
8306 	sx_slock(&t4_uld_list_lock);
8307 
8308 	SLIST_FOREACH(ui, &t4_uld_list, link) {
8309 		if (ui->uld_id == id) {
8310 			if (!(sc->flags & FULL_INIT_DONE)) {
8311 				rc = adapter_full_init(sc);
8312 				if (rc != 0)
8313 					goto done;
8314 			}
8315 
8316 			rc = ui->activate(sc);
8317 			if (rc == 0)
8318 				ui->refcount++;
8319 			goto done;
8320 		}
8321 	}
8322 done:
8323 	sx_sunlock(&t4_uld_list_lock);
8324 
8325 	return (rc);
8326 }
8327 
8328 int
8329 t4_deactivate_uld(struct adapter *sc, int id)
8330 {
8331 	int rc = EINVAL;
8332 	struct uld_info *ui;
8333 
8334 	ASSERT_SYNCHRONIZED_OP(sc);
8335 
8336 	sx_slock(&t4_uld_list_lock);
8337 
8338 	SLIST_FOREACH(ui, &t4_uld_list, link) {
8339 		if (ui->uld_id == id) {
8340 			rc = ui->deactivate(sc);
8341 			if (rc == 0)
8342 				ui->refcount--;
8343 			goto done;
8344 		}
8345 	}
8346 done:
8347 	sx_sunlock(&t4_uld_list_lock);
8348 
8349 	return (rc);
8350 }
8351 #endif
8352 
8353 /*
8354  * Come up with reasonable defaults for some of the tunables, provided they're
8355  * not set by the user (in which case we'll use the values as is).
8356  */
8357 static void
8358 tweak_tunables(void)
8359 {
8360 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
8361 
8362 	if (t4_ntxq10g < 1)
8363 		t4_ntxq10g = min(nc, NTXQ_10G);
8364 
8365 	if (t4_ntxq1g < 1)
8366 		t4_ntxq1g = min(nc, NTXQ_1G);
8367 
8368 	if (t4_nrxq10g < 1)
8369 		t4_nrxq10g = min(nc, NRXQ_10G);
8370 
8371 	if (t4_nrxq1g < 1)
8372 		t4_nrxq1g = min(nc, NRXQ_1G);
8373 
8374 #ifdef TCP_OFFLOAD
8375 	if (t4_nofldtxq10g < 1)
8376 		t4_nofldtxq10g = min(nc, NOFLDTXQ_10G);
8377 
8378 	if (t4_nofldtxq1g < 1)
8379 		t4_nofldtxq1g = min(nc, NOFLDTXQ_1G);
8380 
8381 	if (t4_nofldrxq10g < 1)
8382 		t4_nofldrxq10g = min(nc, NOFLDRXQ_10G);
8383 
8384 	if (t4_nofldrxq1g < 1)
8385 		t4_nofldrxq1g = min(nc, NOFLDRXQ_1G);
8386 
8387 	if (t4_toecaps_allowed == -1)
8388 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
8389 #else
8390 	if (t4_toecaps_allowed == -1)
8391 		t4_toecaps_allowed = 0;
8392 #endif
8393 
8394 #ifdef DEV_NETMAP
8395 	if (t4_nnmtxq10g < 1)
8396 		t4_nnmtxq10g = min(nc, NNMTXQ_10G);
8397 
8398 	if (t4_nnmtxq1g < 1)
8399 		t4_nnmtxq1g = min(nc, NNMTXQ_1G);
8400 
8401 	if (t4_nnmrxq10g < 1)
8402 		t4_nnmrxq10g = min(nc, NNMRXQ_10G);
8403 
8404 	if (t4_nnmrxq1g < 1)
8405 		t4_nnmrxq1g = min(nc, NNMRXQ_1G);
8406 #endif
8407 
8408 	if (t4_tmr_idx_10g < 0 || t4_tmr_idx_10g >= SGE_NTIMERS)
8409 		t4_tmr_idx_10g = TMR_IDX_10G;
8410 
8411 	if (t4_pktc_idx_10g < -1 || t4_pktc_idx_10g >= SGE_NCOUNTERS)
8412 		t4_pktc_idx_10g = PKTC_IDX_10G;
8413 
8414 	if (t4_tmr_idx_1g < 0 || t4_tmr_idx_1g >= SGE_NTIMERS)
8415 		t4_tmr_idx_1g = TMR_IDX_1G;
8416 
8417 	if (t4_pktc_idx_1g < -1 || t4_pktc_idx_1g >= SGE_NCOUNTERS)
8418 		t4_pktc_idx_1g = PKTC_IDX_1G;
8419 
8420 	if (t4_qsize_txq < 128)
8421 		t4_qsize_txq = 128;
8422 
8423 	if (t4_qsize_rxq < 128)
8424 		t4_qsize_rxq = 128;
8425 	while (t4_qsize_rxq & 7)
8426 		t4_qsize_rxq++;
8427 
8428 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
8429 }
8430 
8431 static struct sx mlu;	/* mod load unload */
8432 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
8433 
8434 static int
8435 mod_event(module_t mod, int cmd, void *arg)
8436 {
8437 	int rc = 0;
8438 	static int loaded = 0;
8439 
8440 	switch (cmd) {
8441 	case MOD_LOAD:
8442 		sx_xlock(&mlu);
8443 		if (loaded++ == 0) {
8444 			t4_sge_modload();
8445 			sx_init(&t4_list_lock, "T4/T5 adapters");
8446 			SLIST_INIT(&t4_list);
8447 #ifdef TCP_OFFLOAD
8448 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
8449 			SLIST_INIT(&t4_uld_list);
8450 #endif
8451 			t4_tracer_modload();
8452 			tweak_tunables();
8453 		}
8454 		sx_xunlock(&mlu);
8455 		break;
8456 
8457 	case MOD_UNLOAD:
8458 		sx_xlock(&mlu);
8459 		if (--loaded == 0) {
8460 			int tries;
8461 
8462 			sx_slock(&t4_list_lock);
8463 			if (!SLIST_EMPTY(&t4_list)) {
8464 				rc = EBUSY;
8465 				sx_sunlock(&t4_list_lock);
8466 				goto done_unload;
8467 			}
8468 #ifdef TCP_OFFLOAD
8469 			sx_slock(&t4_uld_list_lock);
8470 			if (!SLIST_EMPTY(&t4_uld_list)) {
8471 				rc = EBUSY;
8472 				sx_sunlock(&t4_uld_list_lock);
8473 				sx_sunlock(&t4_list_lock);
8474 				goto done_unload;
8475 			}
8476 #endif
8477 			tries = 0;
8478 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
8479 				uprintf("%ju clusters with custom free routine "
8480 				    "still is use.\n", t4_sge_extfree_refs());
8481 				pause("t4unload", 2 * hz);
8482 			}
8483 #ifdef TCP_OFFLOAD
8484 			sx_sunlock(&t4_uld_list_lock);
8485 #endif
8486 			sx_sunlock(&t4_list_lock);
8487 
8488 			if (t4_sge_extfree_refs() == 0) {
8489 				t4_tracer_modunload();
8490 #ifdef TCP_OFFLOAD
8491 				sx_destroy(&t4_uld_list_lock);
8492 #endif
8493 				sx_destroy(&t4_list_lock);
8494 				t4_sge_modunload();
8495 				loaded = 0;
8496 			} else {
8497 				rc = EBUSY;
8498 				loaded++;	/* undo earlier decrement */
8499 			}
8500 		}
8501 done_unload:
8502 		sx_xunlock(&mlu);
8503 		break;
8504 	}
8505 
8506 	return (rc);
8507 }
8508 
8509 static devclass_t t4_devclass, t5_devclass;
8510 static devclass_t cxgbe_devclass, cxl_devclass;
8511 
8512 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
8513 MODULE_VERSION(t4nex, 1);
8514 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
8515 
8516 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
8517 MODULE_VERSION(t5nex, 1);
8518 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
8519 
8520 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
8521 MODULE_VERSION(cxgbe, 1);
8522 
8523 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
8524 MODULE_VERSION(cxl, 1);
8525