t4_main.c revision 311261
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: stable/10/sys/dev/cxgbe/t4_main.c 311261 2017-01-04 06:58:35Z np $");
30
31#include "opt_ddb.h"
32#include "opt_inet.h"
33#include "opt_inet6.h"
34
35#include <sys/param.h>
36#include <sys/conf.h>
37#include <sys/priv.h>
38#include <sys/kernel.h>
39#include <sys/bus.h>
40#include <sys/systm.h>
41#include <sys/counter.h>
42#include <sys/module.h>
43#include <sys/malloc.h>
44#include <sys/queue.h>
45#include <sys/taskqueue.h>
46#include <sys/pciio.h>
47#include <dev/pci/pcireg.h>
48#include <dev/pci/pcivar.h>
49#include <dev/pci/pci_private.h>
50#include <sys/firmware.h>
51#include <sys/sbuf.h>
52#include <sys/smp.h>
53#include <sys/socket.h>
54#include <sys/sockio.h>
55#include <sys/sysctl.h>
56#include <net/ethernet.h>
57#include <net/if.h>
58#include <net/if_types.h>
59#include <net/if_dl.h>
60#include <net/if_vlan_var.h>
61#ifdef RSS
62#include <net/rss_config.h>
63#endif
64#if defined(__i386__) || defined(__amd64__)
65#include <vm/vm.h>
66#include <vm/pmap.h>
67#endif
68#ifdef DDB
69#include <ddb/ddb.h>
70#include <ddb/db_lex.h>
71#endif
72
73#include "common/common.h"
74#include "common/t4_msg.h"
75#include "common/t4_regs.h"
76#include "common/t4_regs_values.h"
77#include "t4_ioctl.h"
78#include "t4_l2t.h"
79#include "t4_mp_ring.h"
80
81/* T4 bus driver interface */
82static int t4_probe(device_t);
83static int t4_attach(device_t);
84static int t4_detach(device_t);
85static device_method_t t4_methods[] = {
86	DEVMETHOD(device_probe,		t4_probe),
87	DEVMETHOD(device_attach,	t4_attach),
88	DEVMETHOD(device_detach,	t4_detach),
89
90	DEVMETHOD_END
91};
92static driver_t t4_driver = {
93	"t4nex",
94	t4_methods,
95	sizeof(struct adapter)
96};
97
98
99/* T4 port (cxgbe) interface */
100static int cxgbe_probe(device_t);
101static int cxgbe_attach(device_t);
102static int cxgbe_detach(device_t);
103device_method_t cxgbe_methods[] = {
104	DEVMETHOD(device_probe,		cxgbe_probe),
105	DEVMETHOD(device_attach,	cxgbe_attach),
106	DEVMETHOD(device_detach,	cxgbe_detach),
107	{ 0, 0 }
108};
109static driver_t cxgbe_driver = {
110	"cxgbe",
111	cxgbe_methods,
112	sizeof(struct port_info)
113};
114
115/* T4 VI (vcxgbe) interface */
116static int vcxgbe_probe(device_t);
117static int vcxgbe_attach(device_t);
118static int vcxgbe_detach(device_t);
119static device_method_t vcxgbe_methods[] = {
120	DEVMETHOD(device_probe,		vcxgbe_probe),
121	DEVMETHOD(device_attach,	vcxgbe_attach),
122	DEVMETHOD(device_detach,	vcxgbe_detach),
123	{ 0, 0 }
124};
125static driver_t vcxgbe_driver = {
126	"vcxgbe",
127	vcxgbe_methods,
128	sizeof(struct vi_info)
129};
130
131static d_ioctl_t t4_ioctl;
132
133static struct cdevsw t4_cdevsw = {
134       .d_version = D_VERSION,
135       .d_ioctl = t4_ioctl,
136       .d_name = "t4nex",
137};
138
139/* T5 bus driver interface */
140static int t5_probe(device_t);
141static device_method_t t5_methods[] = {
142	DEVMETHOD(device_probe,		t5_probe),
143	DEVMETHOD(device_attach,	t4_attach),
144	DEVMETHOD(device_detach,	t4_detach),
145
146	DEVMETHOD_END
147};
148static driver_t t5_driver = {
149	"t5nex",
150	t5_methods,
151	sizeof(struct adapter)
152};
153
154
155/* T5 port (cxl) interface */
156static driver_t cxl_driver = {
157	"cxl",
158	cxgbe_methods,
159	sizeof(struct port_info)
160};
161
162/* T5 VI (vcxl) interface */
163static driver_t vcxl_driver = {
164	"vcxl",
165	vcxgbe_methods,
166	sizeof(struct vi_info)
167};
168
169/* T6 bus driver interface */
170static int t6_probe(device_t);
171static device_method_t t6_methods[] = {
172	DEVMETHOD(device_probe,		t6_probe),
173	DEVMETHOD(device_attach,	t4_attach),
174	DEVMETHOD(device_detach,	t4_detach),
175
176	DEVMETHOD_END
177};
178static driver_t t6_driver = {
179	"t6nex",
180	t6_methods,
181	sizeof(struct adapter)
182};
183
184
185/* T6 port (cc) interface */
186static driver_t cc_driver = {
187	"cc",
188	cxgbe_methods,
189	sizeof(struct port_info)
190};
191
192/* T6 VI (vcc) interface */
193static driver_t vcc_driver = {
194	"vcc",
195	vcxgbe_methods,
196	sizeof(struct vi_info)
197};
198
199/* ifnet + media interface */
200static void cxgbe_init(void *);
201static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
202static int cxgbe_transmit(struct ifnet *, struct mbuf *);
203static void cxgbe_qflush(struct ifnet *);
204static int cxgbe_media_change(struct ifnet *);
205static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
206
207MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
208
209/*
210 * Correct lock order when you need to acquire multiple locks is t4_list_lock,
211 * then ADAPTER_LOCK, then t4_uld_list_lock.
212 */
213static struct sx t4_list_lock;
214SLIST_HEAD(, adapter) t4_list;
215#ifdef TCP_OFFLOAD
216static struct sx t4_uld_list_lock;
217SLIST_HEAD(, uld_info) t4_uld_list;
218#endif
219
220/*
221 * Tunables.  See tweak_tunables() too.
222 *
223 * Each tunable is set to a default value here if it's known at compile-time.
224 * Otherwise it is set to -1 as an indication to tweak_tunables() that it should
225 * provide a reasonable default when the driver is loaded.
226 *
227 * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
228 * T5 are under hw.cxl.
229 */
230
231/*
232 * Number of queues for tx and rx, 10G and 1G, NIC and offload.
233 */
234#define NTXQ_10G 16
235int t4_ntxq10g = -1;
236TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq10g);
237
238#define NRXQ_10G 8
239int t4_nrxq10g = -1;
240TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq10g);
241
242#define NTXQ_1G 4
243int t4_ntxq1g = -1;
244TUNABLE_INT("hw.cxgbe.ntxq1g", &t4_ntxq1g);
245
246#define NRXQ_1G 2
247int t4_nrxq1g = -1;
248TUNABLE_INT("hw.cxgbe.nrxq1g", &t4_nrxq1g);
249
250#define NTXQ_VI 1
251static int t4_ntxq_vi = -1;
252TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
253
254#define NRXQ_VI 1
255static int t4_nrxq_vi = -1;
256TUNABLE_INT("hw.cxgbe.nrxq_vi", &t4_nrxq_vi);
257
258static int t4_rsrv_noflowq = 0;
259TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
260
261#ifdef TCP_OFFLOAD
262#define NOFLDTXQ_10G 8
263static int t4_nofldtxq10g = -1;
264TUNABLE_INT("hw.cxgbe.nofldtxq10g", &t4_nofldtxq10g);
265
266#define NOFLDRXQ_10G 2
267static int t4_nofldrxq10g = -1;
268TUNABLE_INT("hw.cxgbe.nofldrxq10g", &t4_nofldrxq10g);
269
270#define NOFLDTXQ_1G 2
271static int t4_nofldtxq1g = -1;
272TUNABLE_INT("hw.cxgbe.nofldtxq1g", &t4_nofldtxq1g);
273
274#define NOFLDRXQ_1G 1
275static int t4_nofldrxq1g = -1;
276TUNABLE_INT("hw.cxgbe.nofldrxq1g", &t4_nofldrxq1g);
277
278#define NOFLDTXQ_VI 1
279static int t4_nofldtxq_vi = -1;
280TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
281
282#define NOFLDRXQ_VI 1
283static int t4_nofldrxq_vi = -1;
284TUNABLE_INT("hw.cxgbe.nofldrxq_vi", &t4_nofldrxq_vi);
285#endif
286
287#ifdef DEV_NETMAP
288#define NNMTXQ_VI 2
289static int t4_nnmtxq_vi = -1;
290TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
291
292#define NNMRXQ_VI 2
293static int t4_nnmrxq_vi = -1;
294TUNABLE_INT("hw.cxgbe.nnmrxq_vi", &t4_nnmrxq_vi);
295#endif
296
297/*
298 * Holdoff parameters for 10G and 1G ports.
299 */
300#define TMR_IDX_10G 1
301int t4_tmr_idx_10g = TMR_IDX_10G;
302TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx_10g);
303
304#define PKTC_IDX_10G (-1)
305int t4_pktc_idx_10g = PKTC_IDX_10G;
306TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx_10g);
307
308#define TMR_IDX_1G 1
309int t4_tmr_idx_1g = TMR_IDX_1G;
310TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_1G", &t4_tmr_idx_1g);
311
312#define PKTC_IDX_1G (-1)
313int t4_pktc_idx_1g = PKTC_IDX_1G;
314TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_1G", &t4_pktc_idx_1g);
315
316/*
317 * Size (# of entries) of each tx and rx queue.
318 */
319unsigned int t4_qsize_txq = TX_EQ_QSIZE;
320TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
321
322unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
323TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
324
325/*
326 * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
327 */
328int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
329TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
330
331/*
332 * Configuration file.
333 */
334#define DEFAULT_CF	"default"
335#define FLASH_CF	"flash"
336#define UWIRE_CF	"uwire"
337#define FPGA_CF		"fpga"
338static char t4_cfg_file[32] = DEFAULT_CF;
339TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
340
341/*
342 * PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
343 * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
344 * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
345 *            mark or when signalled to do so, 0 to never emit PAUSE.
346 */
347static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
348TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
349
350/*
351 * Forward Error Correction settings (bit 0, 1, 2 = FEC_RS, FEC_BASER_RS,
352 * FEC_RESERVED respectively).
353 * -1 to run with the firmware default.
354 *  0 to disable FEC.
355 */
356static int t4_fec = -1;
357TUNABLE_INT("hw.cxgbe.fec", &t4_fec);
358
359/*
360 * Link autonegotiation.
361 * -1 to run with the firmware default.
362 *  0 to disable.
363 *  1 to enable.
364 */
365static int t4_autoneg = -1;
366TUNABLE_INT("hw.cxgbe.autoneg", &t4_autoneg);
367
368/*
369 * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
370 * encouraged respectively).
371 */
372static unsigned int t4_fw_install = 1;
373TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
374
375/*
376 * ASIC features that will be used.  Disable the ones you don't want so that the
377 * chip resources aren't wasted on features that will not be used.
378 */
379static int t4_nbmcaps_allowed = 0;
380TUNABLE_INT("hw.cxgbe.nbmcaps_allowed", &t4_nbmcaps_allowed);
381
382static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
383TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
384
385static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
386    FW_CAPS_CONFIG_SWITCH_EGRESS;
387TUNABLE_INT("hw.cxgbe.switchcaps_allowed", &t4_switchcaps_allowed);
388
389static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC;
390TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
391
392static int t4_toecaps_allowed = -1;
393TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
394
395static int t4_rdmacaps_allowed = -1;
396TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
397
398static int t4_cryptocaps_allowed = 0;
399TUNABLE_INT("hw.cxgbe.cryptocaps_allowed", &t4_cryptocaps_allowed);
400
401static int t4_iscsicaps_allowed = -1;
402TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
403
404static int t4_fcoecaps_allowed = 0;
405TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
406
407static int t5_write_combine = 0;
408TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
409
410static int t4_num_vis = 1;
411TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
412
413/* Functions used by extra VIs to obtain unique MAC addresses for each VI. */
414static int vi_mac_funcs[] = {
415	FW_VI_FUNC_OFLD,
416	FW_VI_FUNC_IWARP,
417	FW_VI_FUNC_OPENISCSI,
418	FW_VI_FUNC_OPENFCOE,
419	FW_VI_FUNC_FOISCSI,
420	FW_VI_FUNC_FOFCOE,
421};
422
423struct intrs_and_queues {
424	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
425	uint16_t nirq;		/* Total # of vectors */
426	uint16_t intr_flags_10g;/* Interrupt flags for each 10G port */
427	uint16_t intr_flags_1g;	/* Interrupt flags for each 1G port */
428	uint16_t ntxq10g;	/* # of NIC txq's for each 10G port */
429	uint16_t nrxq10g;	/* # of NIC rxq's for each 10G port */
430	uint16_t ntxq1g;	/* # of NIC txq's for each 1G port */
431	uint16_t nrxq1g;	/* # of NIC rxq's for each 1G port */
432	uint16_t rsrv_noflowq;	/* Flag whether to reserve queue 0 */
433	uint16_t nofldtxq10g;	/* # of TOE txq's for each 10G port */
434	uint16_t nofldrxq10g;	/* # of TOE rxq's for each 10G port */
435	uint16_t nofldtxq1g;	/* # of TOE txq's for each 1G port */
436	uint16_t nofldrxq1g;	/* # of TOE rxq's for each 1G port */
437
438	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
439	uint16_t ntxq_vi;	/* # of NIC txq's */
440	uint16_t nrxq_vi;	/* # of NIC rxq's */
441	uint16_t nofldtxq_vi;	/* # of TOE txq's */
442	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
443	uint16_t nnmtxq_vi;	/* # of netmap txq's */
444	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
445};
446
447struct filter_entry {
448        uint32_t valid:1;	/* filter allocated and valid */
449        uint32_t locked:1;	/* filter is administratively locked */
450        uint32_t pending:1;	/* filter action is pending firmware reply */
451	uint32_t smtidx:8;	/* Source MAC Table index for smac */
452	struct l2t_entry *l2t;	/* Layer Two Table entry for dmac */
453
454        struct t4_filter_specification fs;
455};
456
457static void setup_memwin(struct adapter *);
458static void position_memwin(struct adapter *, int, uint32_t);
459static int rw_via_memwin(struct adapter *, int, uint32_t, uint32_t *, int, int);
460static inline int read_via_memwin(struct adapter *, int, uint32_t, uint32_t *,
461    int);
462static inline int write_via_memwin(struct adapter *, int, uint32_t,
463    const uint32_t *, int);
464static int validate_mem_range(struct adapter *, uint32_t, int);
465static int fwmtype_to_hwmtype(int);
466static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
467    uint32_t *);
468static int fixup_devlog_params(struct adapter *);
469static int cfg_itype_and_nqueues(struct adapter *, int, int, int,
470    struct intrs_and_queues *);
471static int prep_firmware(struct adapter *);
472static int partition_resources(struct adapter *, const struct firmware *,
473    const char *);
474static int get_params__pre_init(struct adapter *);
475static int get_params__post_init(struct adapter *);
476static int set_params__post_init(struct adapter *);
477static void t4_set_desc(struct adapter *);
478static void build_medialist(struct port_info *, struct ifmedia *);
479static int cxgbe_init_synchronized(struct vi_info *);
480static int cxgbe_uninit_synchronized(struct vi_info *);
481static void quiesce_txq(struct adapter *, struct sge_txq *);
482static void quiesce_wrq(struct adapter *, struct sge_wrq *);
483static void quiesce_iq(struct adapter *, struct sge_iq *);
484static void quiesce_fl(struct adapter *, struct sge_fl *);
485static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
486    driver_intr_t *, void *, char *);
487static int t4_free_irq(struct adapter *, struct irq *);
488static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
489static void vi_refresh_stats(struct adapter *, struct vi_info *);
490static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
491static void cxgbe_tick(void *);
492static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
493static void cxgbe_sysctls(struct port_info *);
494static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
495static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
496static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
497static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
498static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
499static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
500static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
501static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
502static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
503static int sysctl_fec(SYSCTL_HANDLER_ARGS);
504static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
505static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
506static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
507#ifdef SBUF_DRAIN
508static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
509static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
510static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
511static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
512static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
513static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
514static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
515static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
516static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
517static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
518static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
519static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
520static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
521static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
522static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
523static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
524static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
525static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
526static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
527static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
528static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
529static int sysctl_tids(SYSCTL_HANDLER_ARGS);
530static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
531static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
532static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
533static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
534static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
535static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
536static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
537#endif
538#ifdef TCP_OFFLOAD
539static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
540static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
541static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
542#endif
543static uint32_t fconf_iconf_to_mode(uint32_t, uint32_t);
544static uint32_t mode_to_fconf(uint32_t);
545static uint32_t mode_to_iconf(uint32_t);
546static int check_fspec_against_fconf_iconf(struct adapter *,
547    struct t4_filter_specification *);
548static int get_filter_mode(struct adapter *, uint32_t *);
549static int set_filter_mode(struct adapter *, uint32_t);
550static inline uint64_t get_filter_hits(struct adapter *, uint32_t);
551static int get_filter(struct adapter *, struct t4_filter *);
552static int set_filter(struct adapter *, struct t4_filter *);
553static int del_filter(struct adapter *, struct t4_filter *);
554static void clear_filter(struct filter_entry *);
555static int set_filter_wr(struct adapter *, int);
556static int del_filter_wr(struct adapter *, int);
557static int set_tcb_rpl(struct sge_iq *, const struct rss_header *,
558    struct mbuf *);
559static int get_sge_context(struct adapter *, struct t4_sge_context *);
560static int load_fw(struct adapter *, struct t4_data *);
561static int load_cfg(struct adapter *, struct t4_data *);
562static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
563static int read_i2c(struct adapter *, struct t4_i2c_data *);
564#ifdef TCP_OFFLOAD
565static int toe_capability(struct vi_info *, int);
566#endif
567static int mod_event(module_t, int, void *);
568
569struct {
570	uint16_t device;
571	char *desc;
572} t4_pciids[] = {
573	{0xa000, "Chelsio Terminator 4 FPGA"},
574	{0x4400, "Chelsio T440-dbg"},
575	{0x4401, "Chelsio T420-CR"},
576	{0x4402, "Chelsio T422-CR"},
577	{0x4403, "Chelsio T440-CR"},
578	{0x4404, "Chelsio T420-BCH"},
579	{0x4405, "Chelsio T440-BCH"},
580	{0x4406, "Chelsio T440-CH"},
581	{0x4407, "Chelsio T420-SO"},
582	{0x4408, "Chelsio T420-CX"},
583	{0x4409, "Chelsio T420-BT"},
584	{0x440a, "Chelsio T404-BT"},
585	{0x440e, "Chelsio T440-LP-CR"},
586}, t5_pciids[] = {
587	{0xb000, "Chelsio Terminator 5 FPGA"},
588	{0x5400, "Chelsio T580-dbg"},
589	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
590	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
591	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
592	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
593	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
594	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
595	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
596	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
597	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
598	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
599	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
600	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
601	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
602#ifdef notyet
603	{0x5404,  "Chelsio T520-BCH"},
604	{0x5405,  "Chelsio T540-BCH"},
605	{0x5406,  "Chelsio T540-CH"},
606	{0x5408,  "Chelsio T520-CX"},
607	{0x540b,  "Chelsio B520-SR"},
608	{0x540c,  "Chelsio B504-BT"},
609	{0x540f,  "Chelsio Amsterdam"},
610	{0x5413,  "Chelsio T580-CHR"},
611#endif
612}, t6_pciids[] = {
613	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
614	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
615	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
616	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
617	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
618	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
619	{0x6410, "Chelsio T62100-DBG"},		/* 2 x 40/50/100G, debug */
620};
621
622#ifdef TCP_OFFLOAD
623/*
624 * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
625 * exactly the same for both rxq and ofld_rxq.
626 */
627CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
628CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
629#endif
630CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
631
632static int
633t4_probe(device_t dev)
634{
635	int i;
636	uint16_t v = pci_get_vendor(dev);
637	uint16_t d = pci_get_device(dev);
638	uint8_t f = pci_get_function(dev);
639
640	if (v != PCI_VENDOR_ID_CHELSIO)
641		return (ENXIO);
642
643	/* Attach only to PF0 of the FPGA */
644	if (d == 0xa000 && f != 0)
645		return (ENXIO);
646
647	for (i = 0; i < nitems(t4_pciids); i++) {
648		if (d == t4_pciids[i].device) {
649			device_set_desc(dev, t4_pciids[i].desc);
650			return (BUS_PROBE_DEFAULT);
651		}
652	}
653
654	return (ENXIO);
655}
656
657static int
658t5_probe(device_t dev)
659{
660	int i;
661	uint16_t v = pci_get_vendor(dev);
662	uint16_t d = pci_get_device(dev);
663	uint8_t f = pci_get_function(dev);
664
665	if (v != PCI_VENDOR_ID_CHELSIO)
666		return (ENXIO);
667
668	/* Attach only to PF0 of the FPGA */
669	if (d == 0xb000 && f != 0)
670		return (ENXIO);
671
672	for (i = 0; i < nitems(t5_pciids); i++) {
673		if (d == t5_pciids[i].device) {
674			device_set_desc(dev, t5_pciids[i].desc);
675			return (BUS_PROBE_DEFAULT);
676		}
677	}
678
679	return (ENXIO);
680}
681
682static int
683t6_probe(device_t dev)
684{
685	int i;
686	uint16_t v = pci_get_vendor(dev);
687	uint16_t d = pci_get_device(dev);
688
689	if (v != PCI_VENDOR_ID_CHELSIO)
690		return (ENXIO);
691
692	for (i = 0; i < nitems(t6_pciids); i++) {
693		if (d == t6_pciids[i].device) {
694			device_set_desc(dev, t6_pciids[i].desc);
695			return (BUS_PROBE_DEFAULT);
696		}
697	}
698
699	return (ENXIO);
700}
701
702static void
703t5_attribute_workaround(device_t dev)
704{
705	device_t root_port;
706	uint32_t v;
707
708	/*
709	 * The T5 chips do not properly echo the No Snoop and Relaxed
710	 * Ordering attributes when replying to a TLP from a Root
711	 * Port.  As a workaround, find the parent Root Port and
712	 * disable No Snoop and Relaxed Ordering.  Note that this
713	 * affects all devices under this root port.
714	 */
715	root_port = pci_find_pcie_root_port(dev);
716	if (root_port == NULL) {
717		device_printf(dev, "Unable to find parent root port\n");
718		return;
719	}
720
721	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
722	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
723	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
724	    0)
725		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
726		    device_get_nameunit(root_port));
727}
728
729static const struct devnames devnames[] = {
730	{
731		.nexus_name = "t4nex",
732		.ifnet_name = "cxgbe",
733		.vi_ifnet_name = "vcxgbe",
734		.pf03_drv_name = "t4iov",
735		.vf_nexus_name = "t4vf",
736		.vf_ifnet_name = "cxgbev"
737	}, {
738		.nexus_name = "t5nex",
739		.ifnet_name = "cxl",
740		.vi_ifnet_name = "vcxl",
741		.pf03_drv_name = "t5iov",
742		.vf_nexus_name = "t5vf",
743		.vf_ifnet_name = "cxlv"
744	}, {
745		.nexus_name = "t6nex",
746		.ifnet_name = "cc",
747		.vi_ifnet_name = "vcc",
748		.pf03_drv_name = "t6iov",
749		.vf_nexus_name = "t6vf",
750		.vf_ifnet_name = "ccv"
751	}
752};
753
754void
755t4_init_devnames(struct adapter *sc)
756{
757	int id;
758
759	id = chip_id(sc);
760	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
761		sc->names = &devnames[id - CHELSIO_T4];
762	else {
763		device_printf(sc->dev, "chip id %d is not supported.\n", id);
764		sc->names = NULL;
765	}
766}
767
768static int
769t4_attach(device_t dev)
770{
771	struct adapter *sc;
772	int rc = 0, i, j, n10g, n1g, rqidx, tqidx;
773	struct make_dev_args mda;
774	struct intrs_and_queues iaq;
775	struct sge *s;
776	uint8_t *buf;
777#ifdef TCP_OFFLOAD
778	int ofld_rqidx, ofld_tqidx;
779#endif
780#ifdef DEV_NETMAP
781	int nm_rqidx, nm_tqidx;
782#endif
783	int num_vis;
784
785	sc = device_get_softc(dev);
786	sc->dev = dev;
787	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
788
789	if ((pci_get_device(dev) & 0xff00) == 0x5400)
790		t5_attribute_workaround(dev);
791	pci_enable_busmaster(dev);
792	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
793		uint32_t v;
794
795		pci_set_max_read_req(dev, 4096);
796		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
797		v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
798		pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
799
800		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
801	}
802
803	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
804	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
805	sc->traceq = -1;
806	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
807	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
808	    device_get_nameunit(dev));
809
810	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
811	    device_get_nameunit(dev));
812	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
813	t4_add_adapter(sc);
814
815	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
816	TAILQ_INIT(&sc->sfl);
817	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
818
819	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
820
821	rc = t4_map_bars_0_and_4(sc);
822	if (rc != 0)
823		goto done; /* error message displayed already */
824
825	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
826
827	/* Prepare the adapter for operation. */
828	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
829	rc = -t4_prep_adapter(sc, buf);
830	free(buf, M_CXGBE);
831	if (rc != 0) {
832		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
833		goto done;
834	}
835
836	/*
837	 * This is the real PF# to which we're attaching.  Works from within PCI
838	 * passthrough environments too, where pci_get_function() could return a
839	 * different PF# depending on the passthrough configuration.  We need to
840	 * use the real PF# in all our communication with the firmware.
841	 */
842	j = t4_read_reg(sc, A_PL_WHOAMI);
843	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
844	sc->mbox = sc->pf;
845
846	t4_init_devnames(sc);
847	if (sc->names == NULL) {
848		rc = ENOTSUP;
849		goto done; /* error message displayed already */
850	}
851
852	/*
853	 * Do this really early, with the memory windows set up even before the
854	 * character device.  The userland tool's register i/o and mem read
855	 * will work even in "recovery mode".
856	 */
857	setup_memwin(sc);
858	if (t4_init_devlog_params(sc, 0) == 0)
859		fixup_devlog_params(sc);
860	make_dev_args_init(&mda);
861	mda.mda_devsw = &t4_cdevsw;
862	mda.mda_uid = UID_ROOT;
863	mda.mda_gid = GID_WHEEL;
864	mda.mda_mode = 0600;
865	mda.mda_si_drv1 = sc;
866	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
867	if (rc != 0)
868		device_printf(dev, "failed to create nexus char device: %d.\n",
869		    rc);
870
871	/* Go no further if recovery mode has been requested. */
872	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
873		device_printf(dev, "recovery mode.\n");
874		goto done;
875	}
876
877#if defined(__i386__)
878	if ((cpu_feature & CPUID_CX8) == 0) {
879		device_printf(dev, "64 bit atomics not available.\n");
880		rc = ENOTSUP;
881		goto done;
882	}
883#endif
884
885	/* Prepare the firmware for operation */
886	rc = prep_firmware(sc);
887	if (rc != 0)
888		goto done; /* error message displayed already */
889
890	rc = get_params__post_init(sc);
891	if (rc != 0)
892		goto done; /* error message displayed already */
893
894	rc = set_params__post_init(sc);
895	if (rc != 0)
896		goto done; /* error message displayed already */
897
898	rc = t4_map_bar_2(sc);
899	if (rc != 0)
900		goto done; /* error message displayed already */
901
902	rc = t4_create_dma_tag(sc);
903	if (rc != 0)
904		goto done; /* error message displayed already */
905
906	/*
907	 * Number of VIs to create per-port.  The first VI is the "main" regular
908	 * VI for the port.  The rest are additional virtual interfaces on the
909	 * same physical port.  Note that the main VI does not have native
910	 * netmap support but the extra VIs do.
911	 *
912	 * Limit the number of VIs per port to the number of available
913	 * MAC addresses per port.
914	 */
915	if (t4_num_vis >= 1)
916		num_vis = t4_num_vis;
917	else
918		num_vis = 1;
919	if (num_vis > nitems(vi_mac_funcs)) {
920		num_vis = nitems(vi_mac_funcs);
921		device_printf(dev, "Number of VIs limited to %d\n", num_vis);
922	}
923
924	/*
925	 * First pass over all the ports - allocate VIs and initialize some
926	 * basic parameters like mac address, port type, etc.  We also figure
927	 * out whether a port is 10G or 1G and use that information when
928	 * calculating how many interrupts to attempt to allocate.
929	 */
930	n10g = n1g = 0;
931	for_each_port(sc, i) {
932		struct port_info *pi;
933		struct link_config *lc;
934
935		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
936		sc->port[i] = pi;
937
938		/* These must be set before t4_port_init */
939		pi->adapter = sc;
940		pi->port_id = i;
941		/*
942		 * XXX: vi[0] is special so we can't delay this allocation until
943		 * pi->nvi's final value is known.
944		 */
945		pi->vi = malloc(sizeof(struct vi_info) * num_vis, M_CXGBE,
946		    M_ZERO | M_WAITOK);
947
948		/*
949		 * Allocate the "main" VI and initialize parameters
950		 * like mac addr.
951		 */
952		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
953		if (rc != 0) {
954			device_printf(dev, "unable to initialize port %d: %d\n",
955			    i, rc);
956			free(pi->vi, M_CXGBE);
957			free(pi, M_CXGBE);
958			sc->port[i] = NULL;
959			goto done;
960		}
961
962		lc = &pi->link_cfg;
963		lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
964		lc->requested_fc |= t4_pause_settings;
965		if (t4_fec != -1) {
966			lc->requested_fec = t4_fec &
967			    G_FW_PORT_CAP_FEC(lc->supported);
968		}
969		if (lc->supported & FW_PORT_CAP_ANEG && t4_autoneg != -1) {
970			lc->autoneg = t4_autoneg ? AUTONEG_ENABLE :
971			    AUTONEG_DISABLE;
972		}
973
974		rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
975		if (rc != 0) {
976			device_printf(dev, "port %d l1cfg failed: %d\n", i, rc);
977			free(pi->vi, M_CXGBE);
978			free(pi, M_CXGBE);
979			sc->port[i] = NULL;
980			goto done;
981		}
982
983		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
984		    device_get_nameunit(dev), i);
985		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
986		sc->chan_map[pi->tx_chan] = i;
987
988		pi->tc = malloc(sizeof(struct tx_sched_class) *
989		    sc->chip_params->nsched_cls, M_CXGBE, M_ZERO | M_WAITOK);
990
991		if (port_top_speed(pi) >= 10) {
992			n10g++;
993		} else {
994			n1g++;
995		}
996
997		pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
998		if (pi->dev == NULL) {
999			device_printf(dev,
1000			    "failed to add device for port %d.\n", i);
1001			rc = ENXIO;
1002			goto done;
1003		}
1004		pi->vi[0].dev = pi->dev;
1005		device_set_softc(pi->dev, pi);
1006	}
1007
1008	/*
1009	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1010	 */
1011	rc = cfg_itype_and_nqueues(sc, n10g, n1g, num_vis, &iaq);
1012	if (rc != 0)
1013		goto done; /* error message displayed already */
1014	if (iaq.nrxq_vi + iaq.nofldrxq_vi + iaq.nnmrxq_vi == 0)
1015		num_vis = 1;
1016
1017	sc->intr_type = iaq.intr_type;
1018	sc->intr_count = iaq.nirq;
1019
1020	s = &sc->sge;
1021	s->nrxq = n10g * iaq.nrxq10g + n1g * iaq.nrxq1g;
1022	s->ntxq = n10g * iaq.ntxq10g + n1g * iaq.ntxq1g;
1023	if (num_vis > 1) {
1024		s->nrxq += (n10g + n1g) * (num_vis - 1) * iaq.nrxq_vi;
1025		s->ntxq += (n10g + n1g) * (num_vis - 1) * iaq.ntxq_vi;
1026	}
1027	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1028	s->neq += sc->params.nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
1029	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1030#ifdef TCP_OFFLOAD
1031	if (is_offload(sc)) {
1032		s->nofldrxq = n10g * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
1033		s->nofldtxq = n10g * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
1034		if (num_vis > 1) {
1035			s->nofldrxq += (n10g + n1g) * (num_vis - 1) *
1036			    iaq.nofldrxq_vi;
1037			s->nofldtxq += (n10g + n1g) * (num_vis - 1) *
1038			    iaq.nofldtxq_vi;
1039		}
1040		s->neq += s->nofldtxq + s->nofldrxq;
1041		s->niq += s->nofldrxq;
1042
1043		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1044		    M_CXGBE, M_ZERO | M_WAITOK);
1045		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1046		    M_CXGBE, M_ZERO | M_WAITOK);
1047	}
1048#endif
1049#ifdef DEV_NETMAP
1050	if (num_vis > 1) {
1051		s->nnmrxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmrxq_vi;
1052		s->nnmtxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmtxq_vi;
1053	}
1054	s->neq += s->nnmtxq + s->nnmrxq;
1055	s->niq += s->nnmrxq;
1056
1057	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1058	    M_CXGBE, M_ZERO | M_WAITOK);
1059	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1060	    M_CXGBE, M_ZERO | M_WAITOK);
1061#endif
1062
1063	s->ctrlq = malloc(sc->params.nports * sizeof(struct sge_wrq), M_CXGBE,
1064	    M_ZERO | M_WAITOK);
1065	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1066	    M_ZERO | M_WAITOK);
1067	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1068	    M_ZERO | M_WAITOK);
1069	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1070	    M_ZERO | M_WAITOK);
1071	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1072	    M_ZERO | M_WAITOK);
1073
1074	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1075	    M_ZERO | M_WAITOK);
1076
1077	t4_init_l2t(sc, M_WAITOK);
1078
1079	/*
1080	 * Second pass over the ports.  This time we know the number of rx and
1081	 * tx queues that each port should get.
1082	 */
1083	rqidx = tqidx = 0;
1084#ifdef TCP_OFFLOAD
1085	ofld_rqidx = ofld_tqidx = 0;
1086#endif
1087#ifdef DEV_NETMAP
1088	nm_rqidx = nm_tqidx = 0;
1089#endif
1090	for_each_port(sc, i) {
1091		struct port_info *pi = sc->port[i];
1092		struct vi_info *vi;
1093
1094		if (pi == NULL)
1095			continue;
1096
1097		pi->nvi = num_vis;
1098		for_each_vi(pi, j, vi) {
1099			vi->pi = pi;
1100			vi->qsize_rxq = t4_qsize_rxq;
1101			vi->qsize_txq = t4_qsize_txq;
1102
1103			vi->first_rxq = rqidx;
1104			vi->first_txq = tqidx;
1105			if (port_top_speed(pi) >= 10) {
1106				vi->tmr_idx = t4_tmr_idx_10g;
1107				vi->pktc_idx = t4_pktc_idx_10g;
1108				vi->flags |= iaq.intr_flags_10g & INTR_RXQ;
1109				vi->nrxq = j == 0 ? iaq.nrxq10g : iaq.nrxq_vi;
1110				vi->ntxq = j == 0 ? iaq.ntxq10g : iaq.ntxq_vi;
1111			} else {
1112				vi->tmr_idx = t4_tmr_idx_1g;
1113				vi->pktc_idx = t4_pktc_idx_1g;
1114				vi->flags |= iaq.intr_flags_1g & INTR_RXQ;
1115				vi->nrxq = j == 0 ? iaq.nrxq1g : iaq.nrxq_vi;
1116				vi->ntxq = j == 0 ? iaq.ntxq1g : iaq.ntxq_vi;
1117			}
1118			rqidx += vi->nrxq;
1119			tqidx += vi->ntxq;
1120
1121			if (j == 0 && vi->ntxq > 1)
1122				vi->rsrv_noflowq = iaq.rsrv_noflowq ? 1 : 0;
1123			else
1124				vi->rsrv_noflowq = 0;
1125
1126#ifdef TCP_OFFLOAD
1127			vi->first_ofld_rxq = ofld_rqidx;
1128			vi->first_ofld_txq = ofld_tqidx;
1129			if (port_top_speed(pi) >= 10) {
1130				vi->flags |= iaq.intr_flags_10g & INTR_OFLD_RXQ;
1131				vi->nofldrxq = j == 0 ? iaq.nofldrxq10g :
1132				    iaq.nofldrxq_vi;
1133				vi->nofldtxq = j == 0 ? iaq.nofldtxq10g :
1134				    iaq.nofldtxq_vi;
1135			} else {
1136				vi->flags |= iaq.intr_flags_1g & INTR_OFLD_RXQ;
1137				vi->nofldrxq = j == 0 ? iaq.nofldrxq1g :
1138				    iaq.nofldrxq_vi;
1139				vi->nofldtxq = j == 0 ? iaq.nofldtxq1g :
1140				    iaq.nofldtxq_vi;
1141			}
1142			ofld_rqidx += vi->nofldrxq;
1143			ofld_tqidx += vi->nofldtxq;
1144#endif
1145#ifdef DEV_NETMAP
1146			if (j > 0) {
1147				vi->first_nm_rxq = nm_rqidx;
1148				vi->first_nm_txq = nm_tqidx;
1149				vi->nnmrxq = iaq.nnmrxq_vi;
1150				vi->nnmtxq = iaq.nnmtxq_vi;
1151				nm_rqidx += vi->nnmrxq;
1152				nm_tqidx += vi->nnmtxq;
1153			}
1154#endif
1155		}
1156	}
1157
1158	rc = t4_setup_intr_handlers(sc);
1159	if (rc != 0) {
1160		device_printf(dev,
1161		    "failed to setup interrupt handlers: %d\n", rc);
1162		goto done;
1163	}
1164
1165	rc = bus_generic_attach(dev);
1166	if (rc != 0) {
1167		device_printf(dev,
1168		    "failed to attach all child ports: %d\n", rc);
1169		goto done;
1170	}
1171
1172	device_printf(dev,
1173	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1174	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1175	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1176	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1177	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1178
1179	t4_set_desc(sc);
1180
1181done:
1182	if (rc != 0 && sc->cdev) {
1183		/* cdev was created and so cxgbetool works; recover that way. */
1184		device_printf(dev,
1185		    "error during attach, adapter is now in recovery mode.\n");
1186		rc = 0;
1187	}
1188
1189	if (rc != 0)
1190		t4_detach_common(dev);
1191	else
1192		t4_sysctls(sc);
1193
1194	return (rc);
1195}
1196
1197/*
1198 * Idempotent
1199 */
1200static int
1201t4_detach(device_t dev)
1202{
1203	struct adapter *sc;
1204
1205	sc = device_get_softc(dev);
1206
1207	return (t4_detach_common(dev));
1208}
1209
1210int
1211t4_detach_common(device_t dev)
1212{
1213	struct adapter *sc;
1214	struct port_info *pi;
1215	int i, rc;
1216
1217	sc = device_get_softc(dev);
1218
1219	if (sc->flags & FULL_INIT_DONE) {
1220		if (!(sc->flags & IS_VF))
1221			t4_intr_disable(sc);
1222	}
1223
1224	if (sc->cdev) {
1225		destroy_dev(sc->cdev);
1226		sc->cdev = NULL;
1227	}
1228
1229	if (device_is_attached(dev)) {
1230		rc = bus_generic_detach(dev);
1231		if (rc) {
1232			device_printf(dev,
1233			    "failed to detach child devices: %d\n", rc);
1234			return (rc);
1235		}
1236	}
1237
1238	for (i = 0; i < sc->intr_count; i++)
1239		t4_free_irq(sc, &sc->irq[i]);
1240
1241	for (i = 0; i < MAX_NPORTS; i++) {
1242		pi = sc->port[i];
1243		if (pi) {
1244			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1245			if (pi->dev)
1246				device_delete_child(dev, pi->dev);
1247
1248			mtx_destroy(&pi->pi_lock);
1249			free(pi->vi, M_CXGBE);
1250			free(pi->tc, M_CXGBE);
1251			free(pi, M_CXGBE);
1252		}
1253	}
1254
1255	if (sc->flags & FULL_INIT_DONE)
1256		adapter_full_uninit(sc);
1257
1258	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1259		t4_fw_bye(sc, sc->mbox);
1260
1261	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1262		pci_release_msi(dev);
1263
1264	if (sc->regs_res)
1265		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1266		    sc->regs_res);
1267
1268	if (sc->udbs_res)
1269		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1270		    sc->udbs_res);
1271
1272	if (sc->msix_res)
1273		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1274		    sc->msix_res);
1275
1276	if (sc->l2t)
1277		t4_free_l2t(sc->l2t);
1278
1279#ifdef TCP_OFFLOAD
1280	free(sc->sge.ofld_rxq, M_CXGBE);
1281	free(sc->sge.ofld_txq, M_CXGBE);
1282#endif
1283#ifdef DEV_NETMAP
1284	free(sc->sge.nm_rxq, M_CXGBE);
1285	free(sc->sge.nm_txq, M_CXGBE);
1286#endif
1287	free(sc->irq, M_CXGBE);
1288	free(sc->sge.rxq, M_CXGBE);
1289	free(sc->sge.txq, M_CXGBE);
1290	free(sc->sge.ctrlq, M_CXGBE);
1291	free(sc->sge.iqmap, M_CXGBE);
1292	free(sc->sge.eqmap, M_CXGBE);
1293	free(sc->tids.ftid_tab, M_CXGBE);
1294	t4_destroy_dma_tag(sc);
1295	if (mtx_initialized(&sc->sc_lock)) {
1296		sx_xlock(&t4_list_lock);
1297		SLIST_REMOVE(&t4_list, sc, adapter, link);
1298		sx_xunlock(&t4_list_lock);
1299		mtx_destroy(&sc->sc_lock);
1300	}
1301
1302	callout_drain(&sc->sfl_callout);
1303	if (mtx_initialized(&sc->tids.ftid_lock))
1304		mtx_destroy(&sc->tids.ftid_lock);
1305	if (mtx_initialized(&sc->sfl_lock))
1306		mtx_destroy(&sc->sfl_lock);
1307	if (mtx_initialized(&sc->ifp_lock))
1308		mtx_destroy(&sc->ifp_lock);
1309	if (mtx_initialized(&sc->reg_lock))
1310		mtx_destroy(&sc->reg_lock);
1311
1312	for (i = 0; i < NUM_MEMWIN; i++) {
1313		struct memwin *mw = &sc->memwin[i];
1314
1315		if (rw_initialized(&mw->mw_lock))
1316			rw_destroy(&mw->mw_lock);
1317	}
1318
1319	bzero(sc, sizeof(*sc));
1320
1321	return (0);
1322}
1323
1324static int
1325cxgbe_probe(device_t dev)
1326{
1327	char buf[128];
1328	struct port_info *pi = device_get_softc(dev);
1329
1330	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1331	device_set_desc_copy(dev, buf);
1332
1333	return (BUS_PROBE_DEFAULT);
1334}
1335
1336#define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1337    IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1338    IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1339#define T4_CAP_ENABLE (T4_CAP)
1340
1341static int
1342cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1343{
1344	struct ifnet *ifp;
1345	struct sbuf *sb;
1346
1347	vi->xact_addr_filt = -1;
1348	callout_init(&vi->tick, 1);
1349
1350	/* Allocate an ifnet and set it up */
1351	ifp = if_alloc(IFT_ETHER);
1352	if (ifp == NULL) {
1353		device_printf(dev, "Cannot allocate ifnet\n");
1354		return (ENOMEM);
1355	}
1356	vi->ifp = ifp;
1357	ifp->if_softc = vi;
1358
1359	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1360	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1361
1362	ifp->if_init = cxgbe_init;
1363	ifp->if_ioctl = cxgbe_ioctl;
1364	ifp->if_transmit = cxgbe_transmit;
1365	ifp->if_qflush = cxgbe_qflush;
1366
1367	ifp->if_capabilities = T4_CAP;
1368#ifdef TCP_OFFLOAD
1369	if (vi->nofldrxq != 0)
1370		ifp->if_capabilities |= IFCAP_TOE;
1371#endif
1372	ifp->if_capenable = T4_CAP_ENABLE;
1373	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1374	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1375
1376	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1377	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1378	ifp->if_hw_tsomaxsegsize = 65536;
1379
1380	/* Initialize ifmedia for this VI */
1381	ifmedia_init(&vi->media, IFM_IMASK, cxgbe_media_change,
1382	    cxgbe_media_status);
1383	build_medialist(vi->pi, &vi->media);
1384
1385	vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1386	    EVENTHANDLER_PRI_ANY);
1387
1388	ether_ifattach(ifp, vi->hw_addr);
1389#ifdef DEV_NETMAP
1390	if (vi->nnmrxq != 0)
1391		cxgbe_nm_attach(vi);
1392#endif
1393	sb = sbuf_new_auto();
1394	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1395#ifdef TCP_OFFLOAD
1396	if (ifp->if_capabilities & IFCAP_TOE)
1397		sbuf_printf(sb, "; %d txq, %d rxq (TOE)",
1398		    vi->nofldtxq, vi->nofldrxq);
1399#endif
1400#ifdef DEV_NETMAP
1401	if (ifp->if_capabilities & IFCAP_NETMAP)
1402		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1403		    vi->nnmtxq, vi->nnmrxq);
1404#endif
1405	sbuf_finish(sb);
1406	device_printf(dev, "%s\n", sbuf_data(sb));
1407	sbuf_delete(sb);
1408
1409	vi_sysctls(vi);
1410
1411	return (0);
1412}
1413
1414static int
1415cxgbe_attach(device_t dev)
1416{
1417	struct port_info *pi = device_get_softc(dev);
1418	struct adapter *sc = pi->adapter;
1419	struct vi_info *vi;
1420	int i, rc;
1421
1422	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1423
1424	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1425	if (rc)
1426		return (rc);
1427
1428	for_each_vi(pi, i, vi) {
1429		if (i == 0)
1430			continue;
1431		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1432		if (vi->dev == NULL) {
1433			device_printf(dev, "failed to add VI %d\n", i);
1434			continue;
1435		}
1436		device_set_softc(vi->dev, vi);
1437	}
1438
1439	cxgbe_sysctls(pi);
1440
1441	bus_generic_attach(dev);
1442
1443	return (0);
1444}
1445
1446static void
1447cxgbe_vi_detach(struct vi_info *vi)
1448{
1449	struct ifnet *ifp = vi->ifp;
1450
1451	ether_ifdetach(ifp);
1452
1453	if (vi->vlan_c)
1454		EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1455
1456	/* Let detach proceed even if these fail. */
1457#ifdef DEV_NETMAP
1458	if (ifp->if_capabilities & IFCAP_NETMAP)
1459		cxgbe_nm_detach(vi);
1460#endif
1461	cxgbe_uninit_synchronized(vi);
1462	callout_drain(&vi->tick);
1463	vi_full_uninit(vi);
1464
1465	ifmedia_removeall(&vi->media);
1466	if_free(vi->ifp);
1467	vi->ifp = NULL;
1468}
1469
1470static int
1471cxgbe_detach(device_t dev)
1472{
1473	struct port_info *pi = device_get_softc(dev);
1474	struct adapter *sc = pi->adapter;
1475	int rc;
1476
1477	/* Detach the extra VIs first. */
1478	rc = bus_generic_detach(dev);
1479	if (rc)
1480		return (rc);
1481	device_delete_children(dev);
1482
1483	doom_vi(sc, &pi->vi[0]);
1484
1485	if (pi->flags & HAS_TRACEQ) {
1486		sc->traceq = -1;	/* cloner should not create ifnet */
1487		t4_tracer_port_detach(sc);
1488	}
1489
1490	cxgbe_vi_detach(&pi->vi[0]);
1491	callout_drain(&pi->tick);
1492
1493	end_synchronized_op(sc, 0);
1494
1495	return (0);
1496}
1497
1498static void
1499cxgbe_init(void *arg)
1500{
1501	struct vi_info *vi = arg;
1502	struct adapter *sc = vi->pi->adapter;
1503
1504	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1505		return;
1506	cxgbe_init_synchronized(vi);
1507	end_synchronized_op(sc, 0);
1508}
1509
1510static int
1511cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1512{
1513	int rc = 0, mtu, flags, can_sleep;
1514	struct vi_info *vi = ifp->if_softc;
1515	struct adapter *sc = vi->pi->adapter;
1516	struct ifreq *ifr = (struct ifreq *)data;
1517	uint32_t mask;
1518
1519	switch (cmd) {
1520	case SIOCSIFMTU:
1521		mtu = ifr->ifr_mtu;
1522		if (mtu < ETHERMIN || mtu > MAX_MTU)
1523			return (EINVAL);
1524
1525		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1526		if (rc)
1527			return (rc);
1528		ifp->if_mtu = mtu;
1529		if (vi->flags & VI_INIT_DONE) {
1530			t4_update_fl_bufsize(ifp);
1531			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1532				rc = update_mac_settings(ifp, XGMAC_MTU);
1533		}
1534		end_synchronized_op(sc, 0);
1535		break;
1536
1537	case SIOCSIFFLAGS:
1538		can_sleep = 0;
1539redo_sifflags:
1540		rc = begin_synchronized_op(sc, vi,
1541		    can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1542		if (rc)
1543			return (rc);
1544
1545		if (ifp->if_flags & IFF_UP) {
1546			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1547				flags = vi->if_flags;
1548				if ((ifp->if_flags ^ flags) &
1549				    (IFF_PROMISC | IFF_ALLMULTI)) {
1550					if (can_sleep == 1) {
1551						end_synchronized_op(sc, 0);
1552						can_sleep = 0;
1553						goto redo_sifflags;
1554					}
1555					rc = update_mac_settings(ifp,
1556					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1557				}
1558			} else {
1559				if (can_sleep == 0) {
1560					end_synchronized_op(sc, LOCK_HELD);
1561					can_sleep = 1;
1562					goto redo_sifflags;
1563				}
1564				rc = cxgbe_init_synchronized(vi);
1565			}
1566			vi->if_flags = ifp->if_flags;
1567		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1568			if (can_sleep == 0) {
1569				end_synchronized_op(sc, LOCK_HELD);
1570				can_sleep = 1;
1571				goto redo_sifflags;
1572			}
1573			rc = cxgbe_uninit_synchronized(vi);
1574		}
1575		end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1576		break;
1577
1578	case SIOCADDMULTI:
1579	case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1580		rc = begin_synchronized_op(sc, vi, HOLD_LOCK, "t4multi");
1581		if (rc)
1582			return (rc);
1583		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1584			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1585		end_synchronized_op(sc, LOCK_HELD);
1586		break;
1587
1588	case SIOCSIFCAP:
1589		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1590		if (rc)
1591			return (rc);
1592
1593		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1594		if (mask & IFCAP_TXCSUM) {
1595			ifp->if_capenable ^= IFCAP_TXCSUM;
1596			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1597
1598			if (IFCAP_TSO4 & ifp->if_capenable &&
1599			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1600				ifp->if_capenable &= ~IFCAP_TSO4;
1601				if_printf(ifp,
1602				    "tso4 disabled due to -txcsum.\n");
1603			}
1604		}
1605		if (mask & IFCAP_TXCSUM_IPV6) {
1606			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1607			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1608
1609			if (IFCAP_TSO6 & ifp->if_capenable &&
1610			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1611				ifp->if_capenable &= ~IFCAP_TSO6;
1612				if_printf(ifp,
1613				    "tso6 disabled due to -txcsum6.\n");
1614			}
1615		}
1616		if (mask & IFCAP_RXCSUM)
1617			ifp->if_capenable ^= IFCAP_RXCSUM;
1618		if (mask & IFCAP_RXCSUM_IPV6)
1619			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1620
1621		/*
1622		 * Note that we leave CSUM_TSO alone (it is always set).  The
1623		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1624		 * sending a TSO request our way, so it's sufficient to toggle
1625		 * IFCAP_TSOx only.
1626		 */
1627		if (mask & IFCAP_TSO4) {
1628			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1629			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1630				if_printf(ifp, "enable txcsum first.\n");
1631				rc = EAGAIN;
1632				goto fail;
1633			}
1634			ifp->if_capenable ^= IFCAP_TSO4;
1635		}
1636		if (mask & IFCAP_TSO6) {
1637			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1638			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1639				if_printf(ifp, "enable txcsum6 first.\n");
1640				rc = EAGAIN;
1641				goto fail;
1642			}
1643			ifp->if_capenable ^= IFCAP_TSO6;
1644		}
1645		if (mask & IFCAP_LRO) {
1646#if defined(INET) || defined(INET6)
1647			int i;
1648			struct sge_rxq *rxq;
1649
1650			ifp->if_capenable ^= IFCAP_LRO;
1651			for_each_rxq(vi, i, rxq) {
1652				if (ifp->if_capenable & IFCAP_LRO)
1653					rxq->iq.flags |= IQ_LRO_ENABLED;
1654				else
1655					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1656			}
1657#endif
1658		}
1659#ifdef TCP_OFFLOAD
1660		if (mask & IFCAP_TOE) {
1661			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1662
1663			rc = toe_capability(vi, enable);
1664			if (rc != 0)
1665				goto fail;
1666
1667			ifp->if_capenable ^= mask;
1668		}
1669#endif
1670		if (mask & IFCAP_VLAN_HWTAGGING) {
1671			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1672			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1673				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1674		}
1675		if (mask & IFCAP_VLAN_MTU) {
1676			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1677
1678			/* Need to find out how to disable auto-mtu-inflation */
1679		}
1680		if (mask & IFCAP_VLAN_HWTSO)
1681			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1682		if (mask & IFCAP_VLAN_HWCSUM)
1683			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1684
1685#ifdef VLAN_CAPABILITIES
1686		VLAN_CAPABILITIES(ifp);
1687#endif
1688fail:
1689		end_synchronized_op(sc, 0);
1690		break;
1691
1692	case SIOCSIFMEDIA:
1693	case SIOCGIFMEDIA:
1694	case SIOCGIFXMEDIA:
1695		ifmedia_ioctl(ifp, ifr, &vi->media, cmd);
1696		break;
1697
1698	case SIOCGI2C: {
1699		struct ifi2creq i2c;
1700
1701		rc = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
1702		if (rc != 0)
1703			break;
1704		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1705			rc = EPERM;
1706			break;
1707		}
1708		if (i2c.len > sizeof(i2c.data)) {
1709			rc = EINVAL;
1710			break;
1711		}
1712		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1713		if (rc)
1714			return (rc);
1715		rc = -t4_i2c_rd(sc, sc->mbox, vi->pi->port_id, i2c.dev_addr,
1716		    i2c.offset, i2c.len, &i2c.data[0]);
1717		end_synchronized_op(sc, 0);
1718		if (rc == 0)
1719			rc = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
1720		break;
1721	}
1722
1723	default:
1724		rc = ether_ioctl(ifp, cmd, data);
1725	}
1726
1727	return (rc);
1728}
1729
1730static int
1731cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1732{
1733	struct vi_info *vi = ifp->if_softc;
1734	struct port_info *pi = vi->pi;
1735	struct adapter *sc = pi->adapter;
1736	struct sge_txq *txq;
1737	void *items[1];
1738	int rc;
1739
1740	M_ASSERTPKTHDR(m);
1741	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1742
1743	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1744		m_freem(m);
1745		return (ENETDOWN);
1746	}
1747
1748	rc = parse_pkt(sc, &m);
1749	if (__predict_false(rc != 0)) {
1750		MPASS(m == NULL);			/* was freed already */
1751		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1752		return (rc);
1753	}
1754
1755	/* Select a txq. */
1756	txq = &sc->sge.txq[vi->first_txq];
1757	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1758		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1759		    vi->rsrv_noflowq);
1760
1761	items[0] = m;
1762	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1763	if (__predict_false(rc != 0))
1764		m_freem(m);
1765
1766	return (rc);
1767}
1768
1769static void
1770cxgbe_qflush(struct ifnet *ifp)
1771{
1772	struct vi_info *vi = ifp->if_softc;
1773	struct sge_txq *txq;
1774	int i;
1775
1776	/* queues do not exist if !VI_INIT_DONE. */
1777	if (vi->flags & VI_INIT_DONE) {
1778		for_each_txq(vi, i, txq) {
1779			TXQ_LOCK(txq);
1780			txq->eq.flags &= ~EQ_ENABLED;
1781			TXQ_UNLOCK(txq);
1782			while (!mp_ring_is_idle(txq->r)) {
1783				mp_ring_check_drainage(txq->r, 0);
1784				pause("qflush", 1);
1785			}
1786		}
1787	}
1788	if_qflush(ifp);
1789}
1790
1791static int
1792cxgbe_media_change(struct ifnet *ifp)
1793{
1794	struct vi_info *vi = ifp->if_softc;
1795
1796	device_printf(vi->dev, "%s unimplemented.\n", __func__);
1797
1798	return (EOPNOTSUPP);
1799}
1800
1801static void
1802cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1803{
1804	struct vi_info *vi = ifp->if_softc;
1805	struct port_info *pi = vi->pi;
1806	struct ifmedia_entry *cur;
1807	int speed = pi->link_cfg.speed;
1808
1809	cur = vi->media.ifm_cur;
1810
1811	ifmr->ifm_status = IFM_AVALID;
1812	if (!pi->link_cfg.link_ok)
1813		return;
1814
1815	ifmr->ifm_status |= IFM_ACTIVE;
1816
1817	/* active and current will differ iff current media is autoselect. */
1818	if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
1819		return;
1820
1821	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
1822	if (speed == 10000)
1823		ifmr->ifm_active |= IFM_10G_T;
1824	else if (speed == 1000)
1825		ifmr->ifm_active |= IFM_1000_T;
1826	else if (speed == 100)
1827		ifmr->ifm_active |= IFM_100_TX;
1828	else if (speed == 10)
1829		ifmr->ifm_active |= IFM_10_T;
1830	else
1831		KASSERT(0, ("%s: link up but speed unknown (%u)", __func__,
1832			    speed));
1833}
1834
1835static int
1836vcxgbe_probe(device_t dev)
1837{
1838	char buf[128];
1839	struct vi_info *vi = device_get_softc(dev);
1840
1841	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
1842	    vi - vi->pi->vi);
1843	device_set_desc_copy(dev, buf);
1844
1845	return (BUS_PROBE_DEFAULT);
1846}
1847
1848static int
1849vcxgbe_attach(device_t dev)
1850{
1851	struct vi_info *vi;
1852	struct port_info *pi;
1853	struct adapter *sc;
1854	int func, index, rc;
1855	u32 param, val;
1856
1857	vi = device_get_softc(dev);
1858	pi = vi->pi;
1859	sc = pi->adapter;
1860
1861	index = vi - pi->vi;
1862	KASSERT(index < nitems(vi_mac_funcs),
1863	    ("%s: VI %s doesn't have a MAC func", __func__,
1864	    device_get_nameunit(dev)));
1865	func = vi_mac_funcs[index];
1866	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
1867	    vi->hw_addr, &vi->rss_size, func, 0);
1868	if (rc < 0) {
1869		device_printf(dev, "Failed to allocate virtual interface "
1870		    "for port %d: %d\n", pi->port_id, -rc);
1871		return (-rc);
1872	}
1873	vi->viid = rc;
1874	if (chip_id(sc) <= CHELSIO_T5)
1875		vi->smt_idx = (rc & 0x7f) << 1;
1876	else
1877		vi->smt_idx = (rc & 0x7f);
1878
1879	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
1880	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
1881	    V_FW_PARAMS_PARAM_YZ(vi->viid);
1882	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
1883	if (rc)
1884		vi->rss_base = 0xffff;
1885	else {
1886		/* MPASS((val >> 16) == rss_size); */
1887		vi->rss_base = val & 0xffff;
1888	}
1889
1890	rc = cxgbe_vi_attach(dev, vi);
1891	if (rc) {
1892		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
1893		return (rc);
1894	}
1895	return (0);
1896}
1897
1898static int
1899vcxgbe_detach(device_t dev)
1900{
1901	struct vi_info *vi;
1902	struct adapter *sc;
1903
1904	vi = device_get_softc(dev);
1905	sc = vi->pi->adapter;
1906
1907	doom_vi(sc, vi);
1908
1909	cxgbe_vi_detach(vi);
1910	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
1911
1912	end_synchronized_op(sc, 0);
1913
1914	return (0);
1915}
1916
1917void
1918t4_fatal_err(struct adapter *sc)
1919{
1920	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
1921	t4_intr_disable(sc);
1922	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
1923	    device_get_nameunit(sc->dev));
1924}
1925
1926void
1927t4_add_adapter(struct adapter *sc)
1928{
1929	sx_xlock(&t4_list_lock);
1930	SLIST_INSERT_HEAD(&t4_list, sc, link);
1931	sx_xunlock(&t4_list_lock);
1932}
1933
1934int
1935t4_map_bars_0_and_4(struct adapter *sc)
1936{
1937	sc->regs_rid = PCIR_BAR(0);
1938	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1939	    &sc->regs_rid, RF_ACTIVE);
1940	if (sc->regs_res == NULL) {
1941		device_printf(sc->dev, "cannot map registers.\n");
1942		return (ENXIO);
1943	}
1944	sc->bt = rman_get_bustag(sc->regs_res);
1945	sc->bh = rman_get_bushandle(sc->regs_res);
1946	sc->mmio_len = rman_get_size(sc->regs_res);
1947	setbit(&sc->doorbells, DOORBELL_KDB);
1948
1949	sc->msix_rid = PCIR_BAR(4);
1950	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1951	    &sc->msix_rid, RF_ACTIVE);
1952	if (sc->msix_res == NULL) {
1953		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
1954		return (ENXIO);
1955	}
1956
1957	return (0);
1958}
1959
1960int
1961t4_map_bar_2(struct adapter *sc)
1962{
1963
1964	/*
1965	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
1966	 * to map it if RDMA is disabled.
1967	 */
1968	if (is_t4(sc) && sc->rdmacaps == 0)
1969		return (0);
1970
1971	sc->udbs_rid = PCIR_BAR(2);
1972	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1973	    &sc->udbs_rid, RF_ACTIVE);
1974	if (sc->udbs_res == NULL) {
1975		device_printf(sc->dev, "cannot map doorbell BAR.\n");
1976		return (ENXIO);
1977	}
1978	sc->udbs_base = rman_get_virtual(sc->udbs_res);
1979
1980	if (chip_id(sc) >= CHELSIO_T5) {
1981		setbit(&sc->doorbells, DOORBELL_UDB);
1982#if defined(__i386__) || defined(__amd64__)
1983		if (t5_write_combine) {
1984			int rc, mode;
1985
1986			/*
1987			 * Enable write combining on BAR2.  This is the
1988			 * userspace doorbell BAR and is split into 128B
1989			 * (UDBS_SEG_SIZE) doorbell regions, each associated
1990			 * with an egress queue.  The first 64B has the doorbell
1991			 * and the second 64B can be used to submit a tx work
1992			 * request with an implicit doorbell.
1993			 */
1994
1995			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
1996			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
1997			if (rc == 0) {
1998				clrbit(&sc->doorbells, DOORBELL_UDB);
1999				setbit(&sc->doorbells, DOORBELL_WCWR);
2000				setbit(&sc->doorbells, DOORBELL_UDBWC);
2001			} else {
2002				device_printf(sc->dev,
2003				    "couldn't enable write combining: %d\n",
2004				    rc);
2005			}
2006
2007			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2008			t4_write_reg(sc, A_SGE_STAT_CFG,
2009			    V_STATSOURCE_T5(7) | mode);
2010		}
2011#endif
2012	}
2013
2014	return (0);
2015}
2016
2017struct memwin_init {
2018	uint32_t base;
2019	uint32_t aperture;
2020};
2021
2022static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2023	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2024	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2025	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2026};
2027
2028static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2029	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2030	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2031	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2032};
2033
2034static void
2035setup_memwin(struct adapter *sc)
2036{
2037	const struct memwin_init *mw_init;
2038	struct memwin *mw;
2039	int i;
2040	uint32_t bar0;
2041
2042	if (is_t4(sc)) {
2043		/*
2044		 * Read low 32b of bar0 indirectly via the hardware backdoor
2045		 * mechanism.  Works from within PCI passthrough environments
2046		 * too, where rman_get_start() can return a different value.  We
2047		 * need to program the T4 memory window decoders with the actual
2048		 * addresses that will be coming across the PCIe link.
2049		 */
2050		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2051		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2052
2053		mw_init = &t4_memwin[0];
2054	} else {
2055		/* T5+ use the relative offset inside the PCIe BAR */
2056		bar0 = 0;
2057
2058		mw_init = &t5_memwin[0];
2059	}
2060
2061	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2062		rw_init(&mw->mw_lock, "memory window access");
2063		mw->mw_base = mw_init->base;
2064		mw->mw_aperture = mw_init->aperture;
2065		mw->mw_curpos = 0;
2066		t4_write_reg(sc,
2067		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2068		    (mw->mw_base + bar0) | V_BIR(0) |
2069		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2070		rw_wlock(&mw->mw_lock);
2071		position_memwin(sc, i, 0);
2072		rw_wunlock(&mw->mw_lock);
2073	}
2074
2075	/* flush */
2076	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2077}
2078
2079/*
2080 * Positions the memory window at the given address in the card's address space.
2081 * There are some alignment requirements and the actual position may be at an
2082 * address prior to the requested address.  mw->mw_curpos always has the actual
2083 * position of the window.
2084 */
2085static void
2086position_memwin(struct adapter *sc, int idx, uint32_t addr)
2087{
2088	struct memwin *mw;
2089	uint32_t pf;
2090	uint32_t reg;
2091
2092	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2093	mw = &sc->memwin[idx];
2094	rw_assert(&mw->mw_lock, RA_WLOCKED);
2095
2096	if (is_t4(sc)) {
2097		pf = 0;
2098		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2099	} else {
2100		pf = V_PFNUM(sc->pf);
2101		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2102	}
2103	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2104	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2105	t4_read_reg(sc, reg);	/* flush */
2106}
2107
2108static int
2109rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2110    int len, int rw)
2111{
2112	struct memwin *mw;
2113	uint32_t mw_end, v;
2114
2115	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2116
2117	/* Memory can only be accessed in naturally aligned 4 byte units */
2118	if (addr & 3 || len & 3 || len <= 0)
2119		return (EINVAL);
2120
2121	mw = &sc->memwin[idx];
2122	while (len > 0) {
2123		rw_rlock(&mw->mw_lock);
2124		mw_end = mw->mw_curpos + mw->mw_aperture;
2125		if (addr >= mw_end || addr < mw->mw_curpos) {
2126			/* Will need to reposition the window */
2127			if (!rw_try_upgrade(&mw->mw_lock)) {
2128				rw_runlock(&mw->mw_lock);
2129				rw_wlock(&mw->mw_lock);
2130			}
2131			rw_assert(&mw->mw_lock, RA_WLOCKED);
2132			position_memwin(sc, idx, addr);
2133			rw_downgrade(&mw->mw_lock);
2134			mw_end = mw->mw_curpos + mw->mw_aperture;
2135		}
2136		rw_assert(&mw->mw_lock, RA_RLOCKED);
2137		while (addr < mw_end && len > 0) {
2138			if (rw == 0) {
2139				v = t4_read_reg(sc, mw->mw_base + addr -
2140				    mw->mw_curpos);
2141				*val++ = le32toh(v);
2142			} else {
2143				v = *val++;
2144				t4_write_reg(sc, mw->mw_base + addr -
2145				    mw->mw_curpos, htole32(v));;
2146			}
2147			addr += 4;
2148			len -= 4;
2149		}
2150		rw_runlock(&mw->mw_lock);
2151	}
2152
2153	return (0);
2154}
2155
2156static inline int
2157read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2158    int len)
2159{
2160
2161	return (rw_via_memwin(sc, idx, addr, val, len, 0));
2162}
2163
2164static inline int
2165write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
2166    const uint32_t *val, int len)
2167{
2168
2169	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
2170}
2171
2172static int
2173t4_range_cmp(const void *a, const void *b)
2174{
2175	return ((const struct t4_range *)a)->start -
2176	       ((const struct t4_range *)b)->start;
2177}
2178
2179/*
2180 * Verify that the memory range specified by the addr/len pair is valid within
2181 * the card's address space.
2182 */
2183static int
2184validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2185{
2186	struct t4_range mem_ranges[4], *r, *next;
2187	uint32_t em, addr_len;
2188	int i, n, remaining;
2189
2190	/* Memory can only be accessed in naturally aligned 4 byte units */
2191	if (addr & 3 || len & 3 || len <= 0)
2192		return (EINVAL);
2193
2194	/* Enabled memories */
2195	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2196
2197	r = &mem_ranges[0];
2198	n = 0;
2199	bzero(r, sizeof(mem_ranges));
2200	if (em & F_EDRAM0_ENABLE) {
2201		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2202		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2203		if (r->size > 0) {
2204			r->start = G_EDRAM0_BASE(addr_len) << 20;
2205			if (addr >= r->start &&
2206			    addr + len <= r->start + r->size)
2207				return (0);
2208			r++;
2209			n++;
2210		}
2211	}
2212	if (em & F_EDRAM1_ENABLE) {
2213		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2214		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2215		if (r->size > 0) {
2216			r->start = G_EDRAM1_BASE(addr_len) << 20;
2217			if (addr >= r->start &&
2218			    addr + len <= r->start + r->size)
2219				return (0);
2220			r++;
2221			n++;
2222		}
2223	}
2224	if (em & F_EXT_MEM_ENABLE) {
2225		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2226		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2227		if (r->size > 0) {
2228			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2229			if (addr >= r->start &&
2230			    addr + len <= r->start + r->size)
2231				return (0);
2232			r++;
2233			n++;
2234		}
2235	}
2236	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2237		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2238		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2239		if (r->size > 0) {
2240			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2241			if (addr >= r->start &&
2242			    addr + len <= r->start + r->size)
2243				return (0);
2244			r++;
2245			n++;
2246		}
2247	}
2248	MPASS(n <= nitems(mem_ranges));
2249
2250	if (n > 1) {
2251		/* Sort and merge the ranges. */
2252		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2253
2254		/* Start from index 0 and examine the next n - 1 entries. */
2255		r = &mem_ranges[0];
2256		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2257
2258			MPASS(r->size > 0);	/* r is a valid entry. */
2259			next = r + 1;
2260			MPASS(next->size > 0);	/* and so is the next one. */
2261
2262			while (r->start + r->size >= next->start) {
2263				/* Merge the next one into the current entry. */
2264				r->size = max(r->start + r->size,
2265				    next->start + next->size) - r->start;
2266				n--;	/* One fewer entry in total. */
2267				if (--remaining == 0)
2268					goto done;	/* short circuit */
2269				next++;
2270			}
2271			if (next != r + 1) {
2272				/*
2273				 * Some entries were merged into r and next
2274				 * points to the first valid entry that couldn't
2275				 * be merged.
2276				 */
2277				MPASS(next->size > 0);	/* must be valid */
2278				memcpy(r + 1, next, remaining * sizeof(*r));
2279#ifdef INVARIANTS
2280				/*
2281				 * This so that the foo->size assertion in the
2282				 * next iteration of the loop do the right
2283				 * thing for entries that were pulled up and are
2284				 * no longer valid.
2285				 */
2286				MPASS(n < nitems(mem_ranges));
2287				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2288				    sizeof(struct t4_range));
2289#endif
2290			}
2291		}
2292done:
2293		/* Done merging the ranges. */
2294		MPASS(n > 0);
2295		r = &mem_ranges[0];
2296		for (i = 0; i < n; i++, r++) {
2297			if (addr >= r->start &&
2298			    addr + len <= r->start + r->size)
2299				return (0);
2300		}
2301	}
2302
2303	return (EFAULT);
2304}
2305
2306static int
2307fwmtype_to_hwmtype(int mtype)
2308{
2309
2310	switch (mtype) {
2311	case FW_MEMTYPE_EDC0:
2312		return (MEM_EDC0);
2313	case FW_MEMTYPE_EDC1:
2314		return (MEM_EDC1);
2315	case FW_MEMTYPE_EXTMEM:
2316		return (MEM_MC0);
2317	case FW_MEMTYPE_EXTMEM1:
2318		return (MEM_MC1);
2319	default:
2320		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2321	}
2322}
2323
2324/*
2325 * Verify that the memory range specified by the memtype/offset/len pair is
2326 * valid and lies entirely within the memtype specified.  The global address of
2327 * the start of the range is returned in addr.
2328 */
2329static int
2330validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2331    uint32_t *addr)
2332{
2333	uint32_t em, addr_len, maddr;
2334
2335	/* Memory can only be accessed in naturally aligned 4 byte units */
2336	if (off & 3 || len & 3 || len == 0)
2337		return (EINVAL);
2338
2339	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2340	switch (fwmtype_to_hwmtype(mtype)) {
2341	case MEM_EDC0:
2342		if (!(em & F_EDRAM0_ENABLE))
2343			return (EINVAL);
2344		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2345		maddr = G_EDRAM0_BASE(addr_len) << 20;
2346		break;
2347	case MEM_EDC1:
2348		if (!(em & F_EDRAM1_ENABLE))
2349			return (EINVAL);
2350		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2351		maddr = G_EDRAM1_BASE(addr_len) << 20;
2352		break;
2353	case MEM_MC:
2354		if (!(em & F_EXT_MEM_ENABLE))
2355			return (EINVAL);
2356		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2357		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2358		break;
2359	case MEM_MC1:
2360		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2361			return (EINVAL);
2362		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2363		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2364		break;
2365	default:
2366		return (EINVAL);
2367	}
2368
2369	*addr = maddr + off;	/* global address */
2370	return (validate_mem_range(sc, *addr, len));
2371}
2372
2373static int
2374fixup_devlog_params(struct adapter *sc)
2375{
2376	struct devlog_params *dparams = &sc->params.devlog;
2377	int rc;
2378
2379	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2380	    dparams->size, &dparams->addr);
2381
2382	return (rc);
2383}
2384
2385static int
2386cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g, int num_vis,
2387    struct intrs_and_queues *iaq)
2388{
2389	int rc, itype, navail, nrxq10g, nrxq1g, n;
2390	int nofldrxq10g = 0, nofldrxq1g = 0;
2391
2392	bzero(iaq, sizeof(*iaq));
2393
2394	iaq->ntxq10g = t4_ntxq10g;
2395	iaq->ntxq1g = t4_ntxq1g;
2396	iaq->ntxq_vi = t4_ntxq_vi;
2397	iaq->nrxq10g = nrxq10g = t4_nrxq10g;
2398	iaq->nrxq1g = nrxq1g = t4_nrxq1g;
2399	iaq->nrxq_vi = t4_nrxq_vi;
2400	iaq->rsrv_noflowq = t4_rsrv_noflowq;
2401#ifdef TCP_OFFLOAD
2402	if (is_offload(sc)) {
2403		iaq->nofldtxq10g = t4_nofldtxq10g;
2404		iaq->nofldtxq1g = t4_nofldtxq1g;
2405		iaq->nofldtxq_vi = t4_nofldtxq_vi;
2406		iaq->nofldrxq10g = nofldrxq10g = t4_nofldrxq10g;
2407		iaq->nofldrxq1g = nofldrxq1g = t4_nofldrxq1g;
2408		iaq->nofldrxq_vi = t4_nofldrxq_vi;
2409	}
2410#endif
2411#ifdef DEV_NETMAP
2412	iaq->nnmtxq_vi = t4_nnmtxq_vi;
2413	iaq->nnmrxq_vi = t4_nnmrxq_vi;
2414#endif
2415
2416	for (itype = INTR_MSIX; itype; itype >>= 1) {
2417
2418		if ((itype & t4_intr_types) == 0)
2419			continue;	/* not allowed */
2420
2421		if (itype == INTR_MSIX)
2422			navail = pci_msix_count(sc->dev);
2423		else if (itype == INTR_MSI)
2424			navail = pci_msi_count(sc->dev);
2425		else
2426			navail = 1;
2427restart:
2428		if (navail == 0)
2429			continue;
2430
2431		iaq->intr_type = itype;
2432		iaq->intr_flags_10g = 0;
2433		iaq->intr_flags_1g = 0;
2434
2435		/*
2436		 * Best option: an interrupt vector for errors, one for the
2437		 * firmware event queue, and one for every rxq (NIC and TOE) of
2438		 * every VI.  The VIs that support netmap use the same
2439		 * interrupts for the NIC rx queues and the netmap rx queues
2440		 * because only one set of queues is active at a time.
2441		 */
2442		iaq->nirq = T4_EXTRA_INTR;
2443		iaq->nirq += n10g * (nrxq10g + nofldrxq10g);
2444		iaq->nirq += n1g * (nrxq1g + nofldrxq1g);
2445		iaq->nirq += (n10g + n1g) * (num_vis - 1) *
2446		    max(iaq->nrxq_vi, iaq->nnmrxq_vi);	/* See comment above. */
2447		iaq->nirq += (n10g + n1g) * (num_vis - 1) * iaq->nofldrxq_vi;
2448		if (iaq->nirq <= navail &&
2449		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2450			iaq->intr_flags_10g = INTR_ALL;
2451			iaq->intr_flags_1g = INTR_ALL;
2452			goto allocate;
2453		}
2454
2455		/* Disable the VIs (and netmap) if there aren't enough intrs */
2456		if (num_vis > 1) {
2457			device_printf(sc->dev, "virtual interfaces disabled "
2458			    "because num_vis=%u with current settings "
2459			    "(nrxq10g=%u, nrxq1g=%u, nofldrxq10g=%u, "
2460			    "nofldrxq1g=%u, nrxq_vi=%u nofldrxq_vi=%u, "
2461			    "nnmrxq_vi=%u) would need %u interrupts but "
2462			    "only %u are available.\n", num_vis, nrxq10g,
2463			    nrxq1g, nofldrxq10g, nofldrxq1g, iaq->nrxq_vi,
2464			    iaq->nofldrxq_vi, iaq->nnmrxq_vi, iaq->nirq,
2465			    navail);
2466			num_vis = 1;
2467			iaq->ntxq_vi = iaq->nrxq_vi = 0;
2468			iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
2469			iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
2470			goto restart;
2471		}
2472
2473		/*
2474		 * Second best option: a vector for errors, one for the firmware
2475		 * event queue, and vectors for either all the NIC rx queues or
2476		 * all the TOE rx queues.  The queues that don't get vectors
2477		 * will forward their interrupts to those that do.
2478		 */
2479		iaq->nirq = T4_EXTRA_INTR;
2480		if (nrxq10g >= nofldrxq10g) {
2481			iaq->intr_flags_10g = INTR_RXQ;
2482			iaq->nirq += n10g * nrxq10g;
2483		} else {
2484			iaq->intr_flags_10g = INTR_OFLD_RXQ;
2485			iaq->nirq += n10g * nofldrxq10g;
2486		}
2487		if (nrxq1g >= nofldrxq1g) {
2488			iaq->intr_flags_1g = INTR_RXQ;
2489			iaq->nirq += n1g * nrxq1g;
2490		} else {
2491			iaq->intr_flags_1g = INTR_OFLD_RXQ;
2492			iaq->nirq += n1g * nofldrxq1g;
2493		}
2494		if (iaq->nirq <= navail &&
2495		    (itype != INTR_MSI || powerof2(iaq->nirq)))
2496			goto allocate;
2497
2498		/*
2499		 * Next best option: an interrupt vector for errors, one for the
2500		 * firmware event queue, and at least one per main-VI.  At this
2501		 * point we know we'll have to downsize nrxq and/or nofldrxq to
2502		 * fit what's available to us.
2503		 */
2504		iaq->nirq = T4_EXTRA_INTR;
2505		iaq->nirq += n10g + n1g;
2506		if (iaq->nirq <= navail) {
2507			int leftover = navail - iaq->nirq;
2508
2509			if (n10g > 0) {
2510				int target = max(nrxq10g, nofldrxq10g);
2511
2512				iaq->intr_flags_10g = nrxq10g >= nofldrxq10g ?
2513				    INTR_RXQ : INTR_OFLD_RXQ;
2514
2515				n = 1;
2516				while (n < target && leftover >= n10g) {
2517					leftover -= n10g;
2518					iaq->nirq += n10g;
2519					n++;
2520				}
2521				iaq->nrxq10g = min(n, nrxq10g);
2522#ifdef TCP_OFFLOAD
2523				iaq->nofldrxq10g = min(n, nofldrxq10g);
2524#endif
2525			}
2526
2527			if (n1g > 0) {
2528				int target = max(nrxq1g, nofldrxq1g);
2529
2530				iaq->intr_flags_1g = nrxq1g >= nofldrxq1g ?
2531				    INTR_RXQ : INTR_OFLD_RXQ;
2532
2533				n = 1;
2534				while (n < target && leftover >= n1g) {
2535					leftover -= n1g;
2536					iaq->nirq += n1g;
2537					n++;
2538				}
2539				iaq->nrxq1g = min(n, nrxq1g);
2540#ifdef TCP_OFFLOAD
2541				iaq->nofldrxq1g = min(n, nofldrxq1g);
2542#endif
2543			}
2544
2545			if (itype != INTR_MSI || powerof2(iaq->nirq))
2546				goto allocate;
2547		}
2548
2549		/*
2550		 * Least desirable option: one interrupt vector for everything.
2551		 */
2552		iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2553		iaq->intr_flags_10g = iaq->intr_flags_1g = 0;
2554#ifdef TCP_OFFLOAD
2555		if (is_offload(sc))
2556			iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2557#endif
2558allocate:
2559		navail = iaq->nirq;
2560		rc = 0;
2561		if (itype == INTR_MSIX)
2562			rc = pci_alloc_msix(sc->dev, &navail);
2563		else if (itype == INTR_MSI)
2564			rc = pci_alloc_msi(sc->dev, &navail);
2565
2566		if (rc == 0) {
2567			if (navail == iaq->nirq)
2568				return (0);
2569
2570			/*
2571			 * Didn't get the number requested.  Use whatever number
2572			 * the kernel is willing to allocate (it's in navail).
2573			 */
2574			device_printf(sc->dev, "fewer vectors than requested, "
2575			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2576			    itype, iaq->nirq, navail);
2577			pci_release_msi(sc->dev);
2578			goto restart;
2579		}
2580
2581		device_printf(sc->dev,
2582		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2583		    itype, rc, iaq->nirq, navail);
2584	}
2585
2586	device_printf(sc->dev,
2587	    "failed to find a usable interrupt type.  "
2588	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2589	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2590
2591	return (ENXIO);
2592}
2593
2594#define FW_VERSION(chip) ( \
2595    V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2596    V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2597    V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2598    V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2599#define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2600
2601struct fw_info {
2602	uint8_t chip;
2603	char *kld_name;
2604	char *fw_mod_name;
2605	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
2606} fw_info[] = {
2607	{
2608		.chip = CHELSIO_T4,
2609		.kld_name = "t4fw_cfg",
2610		.fw_mod_name = "t4fw",
2611		.fw_hdr = {
2612			.chip = FW_HDR_CHIP_T4,
2613			.fw_ver = htobe32_const(FW_VERSION(T4)),
2614			.intfver_nic = FW_INTFVER(T4, NIC),
2615			.intfver_vnic = FW_INTFVER(T4, VNIC),
2616			.intfver_ofld = FW_INTFVER(T4, OFLD),
2617			.intfver_ri = FW_INTFVER(T4, RI),
2618			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2619			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
2620			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2621			.intfver_fcoe = FW_INTFVER(T4, FCOE),
2622		},
2623	}, {
2624		.chip = CHELSIO_T5,
2625		.kld_name = "t5fw_cfg",
2626		.fw_mod_name = "t5fw",
2627		.fw_hdr = {
2628			.chip = FW_HDR_CHIP_T5,
2629			.fw_ver = htobe32_const(FW_VERSION(T5)),
2630			.intfver_nic = FW_INTFVER(T5, NIC),
2631			.intfver_vnic = FW_INTFVER(T5, VNIC),
2632			.intfver_ofld = FW_INTFVER(T5, OFLD),
2633			.intfver_ri = FW_INTFVER(T5, RI),
2634			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
2635			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
2636			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
2637			.intfver_fcoe = FW_INTFVER(T5, FCOE),
2638		},
2639	}, {
2640		.chip = CHELSIO_T6,
2641		.kld_name = "t6fw_cfg",
2642		.fw_mod_name = "t6fw",
2643		.fw_hdr = {
2644			.chip = FW_HDR_CHIP_T6,
2645			.fw_ver = htobe32_const(FW_VERSION(T6)),
2646			.intfver_nic = FW_INTFVER(T6, NIC),
2647			.intfver_vnic = FW_INTFVER(T6, VNIC),
2648			.intfver_ofld = FW_INTFVER(T6, OFLD),
2649			.intfver_ri = FW_INTFVER(T6, RI),
2650			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
2651			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
2652			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
2653			.intfver_fcoe = FW_INTFVER(T6, FCOE),
2654		},
2655	}
2656};
2657
2658static struct fw_info *
2659find_fw_info(int chip)
2660{
2661	int i;
2662
2663	for (i = 0; i < nitems(fw_info); i++) {
2664		if (fw_info[i].chip == chip)
2665			return (&fw_info[i]);
2666	}
2667	return (NULL);
2668}
2669
2670/*
2671 * Is the given firmware API compatible with the one the driver was compiled
2672 * with?
2673 */
2674static int
2675fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
2676{
2677
2678	/* short circuit if it's the exact same firmware version */
2679	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
2680		return (1);
2681
2682	/*
2683	 * XXX: Is this too conservative?  Perhaps I should limit this to the
2684	 * features that are supported in the driver.
2685	 */
2686#define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
2687	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
2688	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
2689	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
2690		return (1);
2691#undef SAME_INTF
2692
2693	return (0);
2694}
2695
2696/*
2697 * The firmware in the KLD is usable, but should it be installed?  This routine
2698 * explains itself in detail if it indicates the KLD firmware should be
2699 * installed.
2700 */
2701static int
2702should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
2703{
2704	const char *reason;
2705
2706	if (!card_fw_usable) {
2707		reason = "incompatible or unusable";
2708		goto install;
2709	}
2710
2711	if (k > c) {
2712		reason = "older than the version bundled with this driver";
2713		goto install;
2714	}
2715
2716	if (t4_fw_install == 2 && k != c) {
2717		reason = "different than the version bundled with this driver";
2718		goto install;
2719	}
2720
2721	return (0);
2722
2723install:
2724	if (t4_fw_install == 0) {
2725		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2726		    "but the driver is prohibited from installing a different "
2727		    "firmware on the card.\n",
2728		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2729		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
2730
2731		return (0);
2732	}
2733
2734	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2735	    "installing firmware %u.%u.%u.%u on card.\n",
2736	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2737	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
2738	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2739	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2740
2741	return (1);
2742}
2743/*
2744 * Establish contact with the firmware and determine if we are the master driver
2745 * or not, and whether we are responsible for chip initialization.
2746 */
2747static int
2748prep_firmware(struct adapter *sc)
2749{
2750	const struct firmware *fw = NULL, *default_cfg;
2751	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
2752	enum dev_state state;
2753	struct fw_info *fw_info;
2754	struct fw_hdr *card_fw;		/* fw on the card */
2755	const struct fw_hdr *kld_fw;	/* fw in the KLD */
2756	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
2757					   against */
2758
2759	/* Contact firmware. */
2760	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
2761	if (rc < 0 || state == DEV_STATE_ERR) {
2762		rc = -rc;
2763		device_printf(sc->dev,
2764		    "failed to connect to the firmware: %d, %d.\n", rc, state);
2765		return (rc);
2766	}
2767	pf = rc;
2768	if (pf == sc->mbox)
2769		sc->flags |= MASTER_PF;
2770	else if (state == DEV_STATE_UNINIT) {
2771		/*
2772		 * We didn't get to be the master so we definitely won't be
2773		 * configuring the chip.  It's a bug if someone else hasn't
2774		 * configured it already.
2775		 */
2776		device_printf(sc->dev, "couldn't be master(%d), "
2777		    "device not already initialized either(%d).\n", rc, state);
2778		return (EDOOFUS);
2779	}
2780
2781	/* This is the firmware whose headers the driver was compiled against */
2782	fw_info = find_fw_info(chip_id(sc));
2783	if (fw_info == NULL) {
2784		device_printf(sc->dev,
2785		    "unable to look up firmware information for chip %d.\n",
2786		    chip_id(sc));
2787		return (EINVAL);
2788	}
2789	drv_fw = &fw_info->fw_hdr;
2790
2791	/*
2792	 * The firmware KLD contains many modules.  The KLD name is also the
2793	 * name of the module that contains the default config file.
2794	 */
2795	default_cfg = firmware_get(fw_info->kld_name);
2796
2797	/* Read the header of the firmware on the card */
2798	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
2799	rc = -t4_read_flash(sc, FLASH_FW_START,
2800	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
2801	if (rc == 0)
2802		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
2803	else {
2804		device_printf(sc->dev,
2805		    "Unable to read card's firmware header: %d\n", rc);
2806		card_fw_usable = 0;
2807	}
2808
2809	/* This is the firmware in the KLD */
2810	fw = firmware_get(fw_info->fw_mod_name);
2811	if (fw != NULL) {
2812		kld_fw = (const void *)fw->data;
2813		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
2814	} else {
2815		kld_fw = NULL;
2816		kld_fw_usable = 0;
2817	}
2818
2819	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
2820	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
2821		/*
2822		 * Common case: the firmware on the card is an exact match and
2823		 * the KLD is an exact match too, or the KLD is
2824		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
2825		 * here -- use cxgbetool loadfw if you want to reinstall the
2826		 * same firmware as the one on the card.
2827		 */
2828	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
2829	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
2830	    be32toh(card_fw->fw_ver))) {
2831
2832		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
2833		if (rc != 0) {
2834			device_printf(sc->dev,
2835			    "failed to install firmware: %d\n", rc);
2836			goto done;
2837		}
2838
2839		/* Installed successfully, update the cached header too. */
2840		memcpy(card_fw, kld_fw, sizeof(*card_fw));
2841		card_fw_usable = 1;
2842		need_fw_reset = 0;	/* already reset as part of load_fw */
2843	}
2844
2845	if (!card_fw_usable) {
2846		uint32_t d, c, k;
2847
2848		d = ntohl(drv_fw->fw_ver);
2849		c = ntohl(card_fw->fw_ver);
2850		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
2851
2852		device_printf(sc->dev, "Cannot find a usable firmware: "
2853		    "fw_install %d, chip state %d, "
2854		    "driver compiled with %d.%d.%d.%d, "
2855		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
2856		    t4_fw_install, state,
2857		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
2858		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
2859		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2860		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
2861		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2862		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2863		rc = EINVAL;
2864		goto done;
2865	}
2866
2867	/* Reset device */
2868	if (need_fw_reset &&
2869	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
2870		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
2871		if (rc != ETIMEDOUT && rc != EIO)
2872			t4_fw_bye(sc, sc->mbox);
2873		goto done;
2874	}
2875	sc->flags |= FW_OK;
2876
2877	rc = get_params__pre_init(sc);
2878	if (rc != 0)
2879		goto done; /* error message displayed already */
2880
2881	/* Partition adapter resources as specified in the config file. */
2882	if (state == DEV_STATE_UNINIT) {
2883
2884		KASSERT(sc->flags & MASTER_PF,
2885		    ("%s: trying to change chip settings when not master.",
2886		    __func__));
2887
2888		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
2889		if (rc != 0)
2890			goto done;	/* error message displayed already */
2891
2892		t4_tweak_chip_settings(sc);
2893
2894		/* get basic stuff going */
2895		rc = -t4_fw_initialize(sc, sc->mbox);
2896		if (rc != 0) {
2897			device_printf(sc->dev, "fw init failed: %d.\n", rc);
2898			goto done;
2899		}
2900	} else {
2901		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
2902		sc->cfcsum = 0;
2903	}
2904
2905done:
2906	free(card_fw, M_CXGBE);
2907	if (fw != NULL)
2908		firmware_put(fw, FIRMWARE_UNLOAD);
2909	if (default_cfg != NULL)
2910		firmware_put(default_cfg, FIRMWARE_UNLOAD);
2911
2912	return (rc);
2913}
2914
2915#define FW_PARAM_DEV(param) \
2916	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
2917	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
2918#define FW_PARAM_PFVF(param) \
2919	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
2920	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
2921
2922/*
2923 * Partition chip resources for use between various PFs, VFs, etc.
2924 */
2925static int
2926partition_resources(struct adapter *sc, const struct firmware *default_cfg,
2927    const char *name_prefix)
2928{
2929	const struct firmware *cfg = NULL;
2930	int rc = 0;
2931	struct fw_caps_config_cmd caps;
2932	uint32_t mtype, moff, finicsum, cfcsum;
2933
2934	/*
2935	 * Figure out what configuration file to use.  Pick the default config
2936	 * file for the card if the user hasn't specified one explicitly.
2937	 */
2938	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
2939	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
2940		/* Card specific overrides go here. */
2941		if (pci_get_device(sc->dev) == 0x440a)
2942			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
2943		if (is_fpga(sc))
2944			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
2945	}
2946
2947	/*
2948	 * We need to load another module if the profile is anything except
2949	 * "default" or "flash".
2950	 */
2951	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
2952	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
2953		char s[32];
2954
2955		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
2956		cfg = firmware_get(s);
2957		if (cfg == NULL) {
2958			if (default_cfg != NULL) {
2959				device_printf(sc->dev,
2960				    "unable to load module \"%s\" for "
2961				    "configuration profile \"%s\", will use "
2962				    "the default config file instead.\n",
2963				    s, sc->cfg_file);
2964				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
2965				    "%s", DEFAULT_CF);
2966			} else {
2967				device_printf(sc->dev,
2968				    "unable to load module \"%s\" for "
2969				    "configuration profile \"%s\", will use "
2970				    "the config file on the card's flash "
2971				    "instead.\n", s, sc->cfg_file);
2972				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
2973				    "%s", FLASH_CF);
2974			}
2975		}
2976	}
2977
2978	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
2979	    default_cfg == NULL) {
2980		device_printf(sc->dev,
2981		    "default config file not available, will use the config "
2982		    "file on the card's flash instead.\n");
2983		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
2984	}
2985
2986	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
2987		u_int cflen;
2988		const uint32_t *cfdata;
2989		uint32_t param, val, addr;
2990
2991		KASSERT(cfg != NULL || default_cfg != NULL,
2992		    ("%s: no config to upload", __func__));
2993
2994		/*
2995		 * Ask the firmware where it wants us to upload the config file.
2996		 */
2997		param = FW_PARAM_DEV(CF);
2998		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2999		if (rc != 0) {
3000			/* No support for config file?  Shouldn't happen. */
3001			device_printf(sc->dev,
3002			    "failed to query config file location: %d.\n", rc);
3003			goto done;
3004		}
3005		mtype = G_FW_PARAMS_PARAM_Y(val);
3006		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3007
3008		/*
3009		 * XXX: sheer laziness.  We deliberately added 4 bytes of
3010		 * useless stuffing/comments at the end of the config file so
3011		 * it's ok to simply throw away the last remaining bytes when
3012		 * the config file is not an exact multiple of 4.  This also
3013		 * helps with the validate_mt_off_len check.
3014		 */
3015		if (cfg != NULL) {
3016			cflen = cfg->datasize & ~3;
3017			cfdata = cfg->data;
3018		} else {
3019			cflen = default_cfg->datasize & ~3;
3020			cfdata = default_cfg->data;
3021		}
3022
3023		if (cflen > FLASH_CFG_MAX_SIZE) {
3024			device_printf(sc->dev,
3025			    "config file too long (%d, max allowed is %d).  "
3026			    "Will try to use the config on the card, if any.\n",
3027			    cflen, FLASH_CFG_MAX_SIZE);
3028			goto use_config_on_flash;
3029		}
3030
3031		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3032		if (rc != 0) {
3033			device_printf(sc->dev,
3034			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3035			    "Will try to use the config on the card, if any.\n",
3036			    __func__, mtype, moff, cflen, rc);
3037			goto use_config_on_flash;
3038		}
3039		write_via_memwin(sc, 2, addr, cfdata, cflen);
3040	} else {
3041use_config_on_flash:
3042		mtype = FW_MEMTYPE_FLASH;
3043		moff = t4_flash_cfg_addr(sc);
3044	}
3045
3046	bzero(&caps, sizeof(caps));
3047	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3048	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3049	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3050	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3051	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3052	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3053	if (rc != 0) {
3054		device_printf(sc->dev,
3055		    "failed to pre-process config file: %d "
3056		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3057		goto done;
3058	}
3059
3060	finicsum = be32toh(caps.finicsum);
3061	cfcsum = be32toh(caps.cfcsum);
3062	if (finicsum != cfcsum) {
3063		device_printf(sc->dev,
3064		    "WARNING: config file checksum mismatch: %08x %08x\n",
3065		    finicsum, cfcsum);
3066	}
3067	sc->cfcsum = cfcsum;
3068
3069#define LIMIT_CAPS(x) do { \
3070	caps.x &= htobe16(t4_##x##_allowed); \
3071} while (0)
3072
3073	/*
3074	 * Let the firmware know what features will (not) be used so it can tune
3075	 * things accordingly.
3076	 */
3077	LIMIT_CAPS(nbmcaps);
3078	LIMIT_CAPS(linkcaps);
3079	LIMIT_CAPS(switchcaps);
3080	LIMIT_CAPS(niccaps);
3081	LIMIT_CAPS(toecaps);
3082	LIMIT_CAPS(rdmacaps);
3083	LIMIT_CAPS(cryptocaps);
3084	LIMIT_CAPS(iscsicaps);
3085	LIMIT_CAPS(fcoecaps);
3086#undef LIMIT_CAPS
3087
3088	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3089	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3090	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3091	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3092	if (rc != 0) {
3093		device_printf(sc->dev,
3094		    "failed to process config file: %d.\n", rc);
3095	}
3096done:
3097	if (cfg != NULL)
3098		firmware_put(cfg, FIRMWARE_UNLOAD);
3099	return (rc);
3100}
3101
3102/*
3103 * Retrieve parameters that are needed (or nice to have) very early.
3104 */
3105static int
3106get_params__pre_init(struct adapter *sc)
3107{
3108	int rc;
3109	uint32_t param[2], val[2];
3110
3111	t4_get_version_info(sc);
3112
3113	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3114	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3115	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3116	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3117	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3118
3119	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3120	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3121	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3122	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3123	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3124
3125	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3126	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3127	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3128	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3129	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3130
3131	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3132	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3133	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3134	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3135	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3136
3137	param[0] = FW_PARAM_DEV(PORTVEC);
3138	param[1] = FW_PARAM_DEV(CCLK);
3139	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3140	if (rc != 0) {
3141		device_printf(sc->dev,
3142		    "failed to query parameters (pre_init): %d.\n", rc);
3143		return (rc);
3144	}
3145
3146	sc->params.portvec = val[0];
3147	sc->params.nports = bitcount32(val[0]);
3148	sc->params.vpd.cclk = val[1];
3149
3150	/* Read device log parameters. */
3151	rc = -t4_init_devlog_params(sc, 1);
3152	if (rc == 0)
3153		fixup_devlog_params(sc);
3154	else {
3155		device_printf(sc->dev,
3156		    "failed to get devlog parameters: %d.\n", rc);
3157		rc = 0;	/* devlog isn't critical for device operation */
3158	}
3159
3160	return (rc);
3161}
3162
3163/*
3164 * Retrieve various parameters that are of interest to the driver.  The device
3165 * has been initialized by the firmware at this point.
3166 */
3167static int
3168get_params__post_init(struct adapter *sc)
3169{
3170	int rc;
3171	uint32_t param[7], val[7];
3172	struct fw_caps_config_cmd caps;
3173
3174	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3175	param[1] = FW_PARAM_PFVF(EQ_START);
3176	param[2] = FW_PARAM_PFVF(FILTER_START);
3177	param[3] = FW_PARAM_PFVF(FILTER_END);
3178	param[4] = FW_PARAM_PFVF(L2T_START);
3179	param[5] = FW_PARAM_PFVF(L2T_END);
3180	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3181	if (rc != 0) {
3182		device_printf(sc->dev,
3183		    "failed to query parameters (post_init): %d.\n", rc);
3184		return (rc);
3185	}
3186
3187	sc->sge.iq_start = val[0];
3188	sc->sge.eq_start = val[1];
3189	sc->tids.ftid_base = val[2];
3190	sc->tids.nftids = val[3] - val[2] + 1;
3191	sc->params.ftid_min = val[2];
3192	sc->params.ftid_max = val[3];
3193	sc->vres.l2t.start = val[4];
3194	sc->vres.l2t.size = val[5] - val[4] + 1;
3195	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3196	    ("%s: L2 table size (%u) larger than expected (%u)",
3197	    __func__, sc->vres.l2t.size, L2T_SIZE));
3198
3199	/* get capabilites */
3200	bzero(&caps, sizeof(caps));
3201	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3202	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3203	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3204	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3205	if (rc != 0) {
3206		device_printf(sc->dev,
3207		    "failed to get card capabilities: %d.\n", rc);
3208		return (rc);
3209	}
3210
3211#define READ_CAPS(x) do { \
3212	sc->x = htobe16(caps.x); \
3213} while (0)
3214	READ_CAPS(nbmcaps);
3215	READ_CAPS(linkcaps);
3216	READ_CAPS(switchcaps);
3217	READ_CAPS(niccaps);
3218	READ_CAPS(toecaps);
3219	READ_CAPS(rdmacaps);
3220	READ_CAPS(cryptocaps);
3221	READ_CAPS(iscsicaps);
3222	READ_CAPS(fcoecaps);
3223
3224	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3225		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3226		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3227		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3228		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3229		if (rc != 0) {
3230			device_printf(sc->dev,
3231			    "failed to query NIC parameters: %d.\n", rc);
3232			return (rc);
3233		}
3234		sc->tids.etid_base = val[0];
3235		sc->params.etid_min = val[0];
3236		sc->tids.netids = val[1] - val[0] + 1;
3237		sc->params.netids = sc->tids.netids;
3238		sc->params.eo_wr_cred = val[2];
3239		sc->params.ethoffload = 1;
3240	}
3241
3242	if (sc->toecaps) {
3243		/* query offload-related parameters */
3244		param[0] = FW_PARAM_DEV(NTID);
3245		param[1] = FW_PARAM_PFVF(SERVER_START);
3246		param[2] = FW_PARAM_PFVF(SERVER_END);
3247		param[3] = FW_PARAM_PFVF(TDDP_START);
3248		param[4] = FW_PARAM_PFVF(TDDP_END);
3249		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3250		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3251		if (rc != 0) {
3252			device_printf(sc->dev,
3253			    "failed to query TOE parameters: %d.\n", rc);
3254			return (rc);
3255		}
3256		sc->tids.ntids = val[0];
3257		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3258		sc->tids.stid_base = val[1];
3259		sc->tids.nstids = val[2] - val[1] + 1;
3260		sc->vres.ddp.start = val[3];
3261		sc->vres.ddp.size = val[4] - val[3] + 1;
3262		sc->params.ofldq_wr_cred = val[5];
3263		sc->params.offload = 1;
3264	}
3265	if (sc->rdmacaps) {
3266		param[0] = FW_PARAM_PFVF(STAG_START);
3267		param[1] = FW_PARAM_PFVF(STAG_END);
3268		param[2] = FW_PARAM_PFVF(RQ_START);
3269		param[3] = FW_PARAM_PFVF(RQ_END);
3270		param[4] = FW_PARAM_PFVF(PBL_START);
3271		param[5] = FW_PARAM_PFVF(PBL_END);
3272		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3273		if (rc != 0) {
3274			device_printf(sc->dev,
3275			    "failed to query RDMA parameters(1): %d.\n", rc);
3276			return (rc);
3277		}
3278		sc->vres.stag.start = val[0];
3279		sc->vres.stag.size = val[1] - val[0] + 1;
3280		sc->vres.rq.start = val[2];
3281		sc->vres.rq.size = val[3] - val[2] + 1;
3282		sc->vres.pbl.start = val[4];
3283		sc->vres.pbl.size = val[5] - val[4] + 1;
3284
3285		param[0] = FW_PARAM_PFVF(SQRQ_START);
3286		param[1] = FW_PARAM_PFVF(SQRQ_END);
3287		param[2] = FW_PARAM_PFVF(CQ_START);
3288		param[3] = FW_PARAM_PFVF(CQ_END);
3289		param[4] = FW_PARAM_PFVF(OCQ_START);
3290		param[5] = FW_PARAM_PFVF(OCQ_END);
3291		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3292		if (rc != 0) {
3293			device_printf(sc->dev,
3294			    "failed to query RDMA parameters(2): %d.\n", rc);
3295			return (rc);
3296		}
3297		sc->vres.qp.start = val[0];
3298		sc->vres.qp.size = val[1] - val[0] + 1;
3299		sc->vres.cq.start = val[2];
3300		sc->vres.cq.size = val[3] - val[2] + 1;
3301		sc->vres.ocq.start = val[4];
3302		sc->vres.ocq.size = val[5] - val[4] + 1;
3303	}
3304	if (sc->iscsicaps) {
3305		param[0] = FW_PARAM_PFVF(ISCSI_START);
3306		param[1] = FW_PARAM_PFVF(ISCSI_END);
3307		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3308		if (rc != 0) {
3309			device_printf(sc->dev,
3310			    "failed to query iSCSI parameters: %d.\n", rc);
3311			return (rc);
3312		}
3313		sc->vres.iscsi.start = val[0];
3314		sc->vres.iscsi.size = val[1] - val[0] + 1;
3315	}
3316
3317	t4_init_sge_params(sc);
3318
3319	/*
3320	 * We've got the params we wanted to query via the firmware.  Now grab
3321	 * some others directly from the chip.
3322	 */
3323	rc = t4_read_chip_settings(sc);
3324
3325	return (rc);
3326}
3327
3328static int
3329set_params__post_init(struct adapter *sc)
3330{
3331	uint32_t param, val;
3332
3333	/* ask for encapsulated CPLs */
3334	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3335	val = 1;
3336	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3337
3338	return (0);
3339}
3340
3341#undef FW_PARAM_PFVF
3342#undef FW_PARAM_DEV
3343
3344static void
3345t4_set_desc(struct adapter *sc)
3346{
3347	char buf[128];
3348	struct adapter_params *p = &sc->params;
3349
3350	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
3351
3352	device_set_desc_copy(sc->dev, buf);
3353}
3354
3355static void
3356build_medialist(struct port_info *pi, struct ifmedia *media)
3357{
3358	int m;
3359
3360	PORT_LOCK(pi);
3361
3362	ifmedia_removeall(media);
3363
3364	m = IFM_ETHER | IFM_FDX;
3365
3366	switch(pi->port_type) {
3367	case FW_PORT_TYPE_BT_XFI:
3368	case FW_PORT_TYPE_BT_XAUI:
3369		ifmedia_add(media, m | IFM_10G_T, 0, NULL);
3370		/* fall through */
3371
3372	case FW_PORT_TYPE_BT_SGMII:
3373		ifmedia_add(media, m | IFM_1000_T, 0, NULL);
3374		ifmedia_add(media, m | IFM_100_TX, 0, NULL);
3375		ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL);
3376		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
3377		break;
3378
3379	case FW_PORT_TYPE_CX4:
3380		ifmedia_add(media, m | IFM_10G_CX4, 0, NULL);
3381		ifmedia_set(media, m | IFM_10G_CX4);
3382		break;
3383
3384	case FW_PORT_TYPE_QSFP_10G:
3385	case FW_PORT_TYPE_SFP:
3386	case FW_PORT_TYPE_FIBER_XFI:
3387	case FW_PORT_TYPE_FIBER_XAUI:
3388		switch (pi->mod_type) {
3389
3390		case FW_PORT_MOD_TYPE_LR:
3391			ifmedia_add(media, m | IFM_10G_LR, 0, NULL);
3392			ifmedia_set(media, m | IFM_10G_LR);
3393			break;
3394
3395		case FW_PORT_MOD_TYPE_SR:
3396			ifmedia_add(media, m | IFM_10G_SR, 0, NULL);
3397			ifmedia_set(media, m | IFM_10G_SR);
3398			break;
3399
3400		case FW_PORT_MOD_TYPE_LRM:
3401			ifmedia_add(media, m | IFM_10G_LRM, 0, NULL);
3402			ifmedia_set(media, m | IFM_10G_LRM);
3403			break;
3404
3405		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3406		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3407			ifmedia_add(media, m | IFM_10G_TWINAX, 0, NULL);
3408			ifmedia_set(media, m | IFM_10G_TWINAX);
3409			break;
3410
3411		case FW_PORT_MOD_TYPE_NONE:
3412			m &= ~IFM_FDX;
3413			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3414			ifmedia_set(media, m | IFM_NONE);
3415			break;
3416
3417		case FW_PORT_MOD_TYPE_NA:
3418		case FW_PORT_MOD_TYPE_ER:
3419		default:
3420			device_printf(pi->dev,
3421			    "unknown port_type (%d), mod_type (%d)\n",
3422			    pi->port_type, pi->mod_type);
3423			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3424			ifmedia_set(media, m | IFM_UNKNOWN);
3425			break;
3426		}
3427		break;
3428
3429	case FW_PORT_TYPE_CR_QSFP:
3430	case FW_PORT_TYPE_SFP28:
3431	case FW_PORT_TYPE_KR_SFP28:
3432		switch (pi->mod_type) {
3433
3434		case FW_PORT_MOD_TYPE_SR:
3435			ifmedia_add(media, m | IFM_25G_SR, 0, NULL);
3436			ifmedia_set(media, m | IFM_25G_SR);
3437			break;
3438
3439		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3440		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3441			ifmedia_add(media, m | IFM_25G_CR, 0, NULL);
3442			ifmedia_set(media, m | IFM_25G_CR);
3443			break;
3444
3445		case FW_PORT_MOD_TYPE_NONE:
3446			m &= ~IFM_FDX;
3447			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3448			ifmedia_set(media, m | IFM_NONE);
3449			break;
3450
3451		default:
3452			device_printf(pi->dev,
3453			    "unknown port_type (%d), mod_type (%d)\n",
3454			    pi->port_type, pi->mod_type);
3455			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3456			ifmedia_set(media, m | IFM_UNKNOWN);
3457			break;
3458		}
3459		break;
3460
3461	case FW_PORT_TYPE_QSFP:
3462		switch (pi->mod_type) {
3463
3464		case FW_PORT_MOD_TYPE_LR:
3465			ifmedia_add(media, m | IFM_40G_LR4, 0, NULL);
3466			ifmedia_set(media, m | IFM_40G_LR4);
3467			break;
3468
3469		case FW_PORT_MOD_TYPE_SR:
3470			ifmedia_add(media, m | IFM_40G_SR4, 0, NULL);
3471			ifmedia_set(media, m | IFM_40G_SR4);
3472			break;
3473
3474		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3475		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3476			ifmedia_add(media, m | IFM_40G_CR4, 0, NULL);
3477			ifmedia_set(media, m | IFM_40G_CR4);
3478			break;
3479
3480		case FW_PORT_MOD_TYPE_NONE:
3481			m &= ~IFM_FDX;
3482			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3483			ifmedia_set(media, m | IFM_NONE);
3484			break;
3485
3486		default:
3487			device_printf(pi->dev,
3488			    "unknown port_type (%d), mod_type (%d)\n",
3489			    pi->port_type, pi->mod_type);
3490			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3491			ifmedia_set(media, m | IFM_UNKNOWN);
3492			break;
3493		}
3494		break;
3495
3496	case FW_PORT_TYPE_KR4_100G:
3497	case FW_PORT_TYPE_CR4_QSFP:
3498		switch (pi->mod_type) {
3499
3500		case FW_PORT_MOD_TYPE_LR:
3501			ifmedia_add(media, m | IFM_100G_LR4, 0, NULL);
3502			ifmedia_set(media, m | IFM_100G_LR4);
3503			break;
3504
3505		case FW_PORT_MOD_TYPE_SR:
3506			ifmedia_add(media, m | IFM_100G_SR4, 0, NULL);
3507			ifmedia_set(media, m | IFM_100G_SR4);
3508			break;
3509
3510		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3511		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3512			ifmedia_add(media, m | IFM_100G_CR4, 0, NULL);
3513			ifmedia_set(media, m | IFM_100G_CR4);
3514			break;
3515
3516		case FW_PORT_MOD_TYPE_NONE:
3517			m &= ~IFM_FDX;
3518			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3519			ifmedia_set(media, m | IFM_NONE);
3520			break;
3521
3522		default:
3523			device_printf(pi->dev,
3524			    "unknown port_type (%d), mod_type (%d)\n",
3525			    pi->port_type, pi->mod_type);
3526			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3527			ifmedia_set(media, m | IFM_UNKNOWN);
3528			break;
3529		}
3530		break;
3531
3532	default:
3533		device_printf(pi->dev,
3534		    "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
3535		    pi->mod_type);
3536		ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3537		ifmedia_set(media, m | IFM_UNKNOWN);
3538		break;
3539	}
3540
3541	PORT_UNLOCK(pi);
3542}
3543
3544#define FW_MAC_EXACT_CHUNK	7
3545
3546/*
3547 * Program the port's XGMAC based on parameters in ifnet.  The caller also
3548 * indicates which parameters should be programmed (the rest are left alone).
3549 */
3550int
3551update_mac_settings(struct ifnet *ifp, int flags)
3552{
3553	int rc = 0;
3554	struct vi_info *vi = ifp->if_softc;
3555	struct port_info *pi = vi->pi;
3556	struct adapter *sc = pi->adapter;
3557	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
3558
3559	ASSERT_SYNCHRONIZED_OP(sc);
3560	KASSERT(flags, ("%s: not told what to update.", __func__));
3561
3562	if (flags & XGMAC_MTU)
3563		mtu = ifp->if_mtu;
3564
3565	if (flags & XGMAC_PROMISC)
3566		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
3567
3568	if (flags & XGMAC_ALLMULTI)
3569		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
3570
3571	if (flags & XGMAC_VLANEX)
3572		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
3573
3574	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
3575		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
3576		    allmulti, 1, vlanex, false);
3577		if (rc) {
3578			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
3579			    rc);
3580			return (rc);
3581		}
3582	}
3583
3584	if (flags & XGMAC_UCADDR) {
3585		uint8_t ucaddr[ETHER_ADDR_LEN];
3586
3587		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
3588		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
3589		    ucaddr, true, true);
3590		if (rc < 0) {
3591			rc = -rc;
3592			if_printf(ifp, "change_mac failed: %d\n", rc);
3593			return (rc);
3594		} else {
3595			vi->xact_addr_filt = rc;
3596			rc = 0;
3597		}
3598	}
3599
3600	if (flags & XGMAC_MCADDRS) {
3601		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
3602		int del = 1;
3603		uint64_t hash = 0;
3604		struct ifmultiaddr *ifma;
3605		int i = 0, j;
3606
3607		if_maddr_rlock(ifp);
3608		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3609			if (ifma->ifma_addr->sa_family != AF_LINK)
3610				continue;
3611			mcaddr[i] =
3612			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
3613			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
3614			i++;
3615
3616			if (i == FW_MAC_EXACT_CHUNK) {
3617				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
3618				    del, i, mcaddr, NULL, &hash, 0);
3619				if (rc < 0) {
3620					rc = -rc;
3621					for (j = 0; j < i; j++) {
3622						if_printf(ifp,
3623						    "failed to add mc address"
3624						    " %02x:%02x:%02x:"
3625						    "%02x:%02x:%02x rc=%d\n",
3626						    mcaddr[j][0], mcaddr[j][1],
3627						    mcaddr[j][2], mcaddr[j][3],
3628						    mcaddr[j][4], mcaddr[j][5],
3629						    rc);
3630					}
3631					goto mcfail;
3632				}
3633				del = 0;
3634				i = 0;
3635			}
3636		}
3637		if (i > 0) {
3638			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
3639			    mcaddr, NULL, &hash, 0);
3640			if (rc < 0) {
3641				rc = -rc;
3642				for (j = 0; j < i; j++) {
3643					if_printf(ifp,
3644					    "failed to add mc address"
3645					    " %02x:%02x:%02x:"
3646					    "%02x:%02x:%02x rc=%d\n",
3647					    mcaddr[j][0], mcaddr[j][1],
3648					    mcaddr[j][2], mcaddr[j][3],
3649					    mcaddr[j][4], mcaddr[j][5],
3650					    rc);
3651				}
3652				goto mcfail;
3653			}
3654		}
3655
3656		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
3657		if (rc != 0)
3658			if_printf(ifp, "failed to set mc address hash: %d", rc);
3659mcfail:
3660		if_maddr_runlock(ifp);
3661	}
3662
3663	return (rc);
3664}
3665
3666/*
3667 * {begin|end}_synchronized_op must be called from the same thread.
3668 */
3669int
3670begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
3671    char *wmesg)
3672{
3673	int rc, pri;
3674
3675#ifdef WITNESS
3676	/* the caller thinks it's ok to sleep, but is it really? */
3677	if (flags & SLEEP_OK)
3678		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
3679		    "begin_synchronized_op");
3680#endif
3681
3682	if (INTR_OK)
3683		pri = PCATCH;
3684	else
3685		pri = 0;
3686
3687	ADAPTER_LOCK(sc);
3688	for (;;) {
3689
3690		if (vi && IS_DOOMED(vi)) {
3691			rc = ENXIO;
3692			goto done;
3693		}
3694
3695		if (!IS_BUSY(sc)) {
3696			rc = 0;
3697			break;
3698		}
3699
3700		if (!(flags & SLEEP_OK)) {
3701			rc = EBUSY;
3702			goto done;
3703		}
3704
3705		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
3706			rc = EINTR;
3707			goto done;
3708		}
3709	}
3710
3711	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
3712	SET_BUSY(sc);
3713#ifdef INVARIANTS
3714	sc->last_op = wmesg;
3715	sc->last_op_thr = curthread;
3716	sc->last_op_flags = flags;
3717#endif
3718
3719done:
3720	if (!(flags & HOLD_LOCK) || rc)
3721		ADAPTER_UNLOCK(sc);
3722
3723	return (rc);
3724}
3725
3726/*
3727 * Tell if_ioctl and if_init that the VI is going away.  This is
3728 * special variant of begin_synchronized_op and must be paired with a
3729 * call to end_synchronized_op.
3730 */
3731void
3732doom_vi(struct adapter *sc, struct vi_info *vi)
3733{
3734
3735	ADAPTER_LOCK(sc);
3736	SET_DOOMED(vi);
3737	wakeup(&sc->flags);
3738	while (IS_BUSY(sc))
3739		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
3740	SET_BUSY(sc);
3741#ifdef INVARIANTS
3742	sc->last_op = "t4detach";
3743	sc->last_op_thr = curthread;
3744	sc->last_op_flags = 0;
3745#endif
3746	ADAPTER_UNLOCK(sc);
3747}
3748
3749/*
3750 * {begin|end}_synchronized_op must be called from the same thread.
3751 */
3752void
3753end_synchronized_op(struct adapter *sc, int flags)
3754{
3755
3756	if (flags & LOCK_HELD)
3757		ADAPTER_LOCK_ASSERT_OWNED(sc);
3758	else
3759		ADAPTER_LOCK(sc);
3760
3761	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
3762	CLR_BUSY(sc);
3763	wakeup(&sc->flags);
3764	ADAPTER_UNLOCK(sc);
3765}
3766
3767static int
3768cxgbe_init_synchronized(struct vi_info *vi)
3769{
3770	struct port_info *pi = vi->pi;
3771	struct adapter *sc = pi->adapter;
3772	struct ifnet *ifp = vi->ifp;
3773	int rc = 0, i;
3774	struct sge_txq *txq;
3775
3776	ASSERT_SYNCHRONIZED_OP(sc);
3777
3778	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3779		return (0);	/* already running */
3780
3781	if (!(sc->flags & FULL_INIT_DONE) &&
3782	    ((rc = adapter_full_init(sc)) != 0))
3783		return (rc);	/* error message displayed already */
3784
3785	if (!(vi->flags & VI_INIT_DONE) &&
3786	    ((rc = vi_full_init(vi)) != 0))
3787		return (rc); /* error message displayed already */
3788
3789	rc = update_mac_settings(ifp, XGMAC_ALL);
3790	if (rc)
3791		goto done;	/* error message displayed already */
3792
3793	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
3794	if (rc != 0) {
3795		if_printf(ifp, "enable_vi failed: %d\n", rc);
3796		goto done;
3797	}
3798
3799	/*
3800	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
3801	 * if this changes.
3802	 */
3803
3804	for_each_txq(vi, i, txq) {
3805		TXQ_LOCK(txq);
3806		txq->eq.flags |= EQ_ENABLED;
3807		TXQ_UNLOCK(txq);
3808	}
3809
3810	/*
3811	 * The first iq of the first port to come up is used for tracing.
3812	 */
3813	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
3814		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
3815		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
3816		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
3817		    V_QUEUENUMBER(sc->traceq));
3818		pi->flags |= HAS_TRACEQ;
3819	}
3820
3821	/* all ok */
3822	PORT_LOCK(pi);
3823	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3824	pi->up_vis++;
3825
3826	if (pi->nvi > 1 || sc->flags & IS_VF)
3827		callout_reset(&vi->tick, hz, vi_tick, vi);
3828	else
3829		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
3830	PORT_UNLOCK(pi);
3831done:
3832	if (rc != 0)
3833		cxgbe_uninit_synchronized(vi);
3834
3835	return (rc);
3836}
3837
3838/*
3839 * Idempotent.
3840 */
3841static int
3842cxgbe_uninit_synchronized(struct vi_info *vi)
3843{
3844	struct port_info *pi = vi->pi;
3845	struct adapter *sc = pi->adapter;
3846	struct ifnet *ifp = vi->ifp;
3847	int rc, i;
3848	struct sge_txq *txq;
3849
3850	ASSERT_SYNCHRONIZED_OP(sc);
3851
3852	if (!(vi->flags & VI_INIT_DONE)) {
3853		KASSERT(!(ifp->if_drv_flags & IFF_DRV_RUNNING),
3854		    ("uninited VI is running"));
3855		return (0);
3856	}
3857
3858	/*
3859	 * Disable the VI so that all its data in either direction is discarded
3860	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
3861	 * tick) intact as the TP can deliver negative advice or data that it's
3862	 * holding in its RAM (for an offloaded connection) even after the VI is
3863	 * disabled.
3864	 */
3865	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
3866	if (rc) {
3867		if_printf(ifp, "disable_vi failed: %d\n", rc);
3868		return (rc);
3869	}
3870
3871	for_each_txq(vi, i, txq) {
3872		TXQ_LOCK(txq);
3873		txq->eq.flags &= ~EQ_ENABLED;
3874		TXQ_UNLOCK(txq);
3875	}
3876
3877	PORT_LOCK(pi);
3878	if (pi->nvi > 1 || sc->flags & IS_VF)
3879		callout_stop(&vi->tick);
3880	else
3881		callout_stop(&pi->tick);
3882	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
3883		PORT_UNLOCK(pi);
3884		return (0);
3885	}
3886	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3887	pi->up_vis--;
3888	if (pi->up_vis > 0) {
3889		PORT_UNLOCK(pi);
3890		return (0);
3891	}
3892	PORT_UNLOCK(pi);
3893
3894	pi->link_cfg.link_ok = 0;
3895	pi->link_cfg.speed = 0;
3896	pi->link_cfg.link_down_rc = 255;
3897	t4_os_link_changed(sc, pi->port_id, 0);
3898
3899	return (0);
3900}
3901
3902/*
3903 * It is ok for this function to fail midway and return right away.  t4_detach
3904 * will walk the entire sc->irq list and clean up whatever is valid.
3905 */
3906int
3907t4_setup_intr_handlers(struct adapter *sc)
3908{
3909	int rc, rid, p, q, v;
3910	char s[8];
3911	struct irq *irq;
3912	struct port_info *pi;
3913	struct vi_info *vi;
3914	struct sge *sge = &sc->sge;
3915	struct sge_rxq *rxq;
3916#ifdef TCP_OFFLOAD
3917	struct sge_ofld_rxq *ofld_rxq;
3918#endif
3919#ifdef DEV_NETMAP
3920	struct sge_nm_rxq *nm_rxq;
3921#endif
3922
3923	/*
3924	 * Setup interrupts.
3925	 */
3926	irq = &sc->irq[0];
3927	rid = sc->intr_type == INTR_INTX ? 0 : 1;
3928	if (sc->intr_count == 1)
3929		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
3930
3931	/* Multiple interrupts. */
3932	if (sc->flags & IS_VF)
3933		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
3934		    ("%s: too few intr.", __func__));
3935	else
3936		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
3937		    ("%s: too few intr.", __func__));
3938
3939	/* The first one is always error intr on PFs */
3940	if (!(sc->flags & IS_VF)) {
3941		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
3942		if (rc != 0)
3943			return (rc);
3944		irq++;
3945		rid++;
3946	}
3947
3948	/* The second one is always the firmware event queue (first on VFs) */
3949	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
3950	if (rc != 0)
3951		return (rc);
3952	irq++;
3953	rid++;
3954
3955	for_each_port(sc, p) {
3956		pi = sc->port[p];
3957		for_each_vi(pi, v, vi) {
3958			vi->first_intr = rid - 1;
3959
3960			if (vi->nnmrxq > 0) {
3961				int n = max(vi->nrxq, vi->nnmrxq);
3962
3963				MPASS(vi->flags & INTR_RXQ);
3964
3965				rxq = &sge->rxq[vi->first_rxq];
3966#ifdef DEV_NETMAP
3967				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
3968#endif
3969				for (q = 0; q < n; q++) {
3970					snprintf(s, sizeof(s), "%x%c%x", p,
3971					    'a' + v, q);
3972					if (q < vi->nrxq)
3973						irq->rxq = rxq++;
3974#ifdef DEV_NETMAP
3975					if (q < vi->nnmrxq)
3976						irq->nm_rxq = nm_rxq++;
3977#endif
3978					rc = t4_alloc_irq(sc, irq, rid,
3979					    t4_vi_intr, irq, s);
3980					if (rc != 0)
3981						return (rc);
3982					irq++;
3983					rid++;
3984					vi->nintr++;
3985				}
3986			} else if (vi->flags & INTR_RXQ) {
3987				for_each_rxq(vi, q, rxq) {
3988					snprintf(s, sizeof(s), "%x%c%x", p,
3989					    'a' + v, q);
3990					rc = t4_alloc_irq(sc, irq, rid,
3991					    t4_intr, rxq, s);
3992					if (rc != 0)
3993						return (rc);
3994					irq++;
3995					rid++;
3996					vi->nintr++;
3997				}
3998			}
3999#ifdef TCP_OFFLOAD
4000			if (vi->flags & INTR_OFLD_RXQ) {
4001				for_each_ofld_rxq(vi, q, ofld_rxq) {
4002					snprintf(s, sizeof(s), "%x%c%x", p,
4003					    'A' + v, q);
4004					rc = t4_alloc_irq(sc, irq, rid,
4005					    t4_intr, ofld_rxq, s);
4006					if (rc != 0)
4007						return (rc);
4008					irq++;
4009					rid++;
4010					vi->nintr++;
4011				}
4012			}
4013#endif
4014		}
4015	}
4016	MPASS(irq == &sc->irq[sc->intr_count]);
4017
4018	return (0);
4019}
4020
4021int
4022adapter_full_init(struct adapter *sc)
4023{
4024	int rc, i;
4025
4026	ASSERT_SYNCHRONIZED_OP(sc);
4027	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4028	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
4029	    ("%s: FULL_INIT_DONE already", __func__));
4030
4031	/*
4032	 * queues that belong to the adapter (not any particular port).
4033	 */
4034	rc = t4_setup_adapter_queues(sc);
4035	if (rc != 0)
4036		goto done;
4037
4038	for (i = 0; i < nitems(sc->tq); i++) {
4039		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
4040		    taskqueue_thread_enqueue, &sc->tq[i]);
4041		if (sc->tq[i] == NULL) {
4042			device_printf(sc->dev,
4043			    "failed to allocate task queue %d\n", i);
4044			rc = ENOMEM;
4045			goto done;
4046		}
4047		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
4048		    device_get_nameunit(sc->dev), i);
4049	}
4050
4051	if (!(sc->flags & IS_VF))
4052		t4_intr_enable(sc);
4053	sc->flags |= FULL_INIT_DONE;
4054done:
4055	if (rc != 0)
4056		adapter_full_uninit(sc);
4057
4058	return (rc);
4059}
4060
4061int
4062adapter_full_uninit(struct adapter *sc)
4063{
4064	int i;
4065
4066	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4067
4068	t4_teardown_adapter_queues(sc);
4069
4070	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
4071		taskqueue_free(sc->tq[i]);
4072		sc->tq[i] = NULL;
4073	}
4074
4075	sc->flags &= ~FULL_INIT_DONE;
4076
4077	return (0);
4078}
4079
4080#ifdef RSS
4081#define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
4082    RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
4083    RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
4084    RSS_HASHTYPE_RSS_UDP_IPV6)
4085
4086/* Translates kernel hash types to hardware. */
4087static int
4088hashconfig_to_hashen(int hashconfig)
4089{
4090	int hashen = 0;
4091
4092	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
4093		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
4094	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
4095		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
4096	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
4097		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4098		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4099	}
4100	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
4101		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4102		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4103	}
4104	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
4105		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4106	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
4107		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4108
4109	return (hashen);
4110}
4111
4112/* Translates hardware hash types to kernel. */
4113static int
4114hashen_to_hashconfig(int hashen)
4115{
4116	int hashconfig = 0;
4117
4118	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
4119		/*
4120		 * If UDP hashing was enabled it must have been enabled for
4121		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
4122		 * enabling any 4-tuple hash is nonsense configuration.
4123		 */
4124		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4125		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
4126
4127		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4128			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
4129		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4130			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
4131	}
4132	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4133		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
4134	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4135		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
4136	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
4137		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
4138	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
4139		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
4140
4141	return (hashconfig);
4142}
4143#endif
4144
4145int
4146vi_full_init(struct vi_info *vi)
4147{
4148	struct adapter *sc = vi->pi->adapter;
4149	struct ifnet *ifp = vi->ifp;
4150	uint16_t *rss;
4151	struct sge_rxq *rxq;
4152	int rc, i, j, hashen;
4153#ifdef RSS
4154	int nbuckets = rss_getnumbuckets();
4155	int hashconfig = rss_gethashconfig();
4156	int extra;
4157	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4158	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4159#endif
4160
4161	ASSERT_SYNCHRONIZED_OP(sc);
4162	KASSERT((vi->flags & VI_INIT_DONE) == 0,
4163	    ("%s: VI_INIT_DONE already", __func__));
4164
4165	sysctl_ctx_init(&vi->ctx);
4166	vi->flags |= VI_SYSCTL_CTX;
4167
4168	/*
4169	 * Allocate tx/rx/fl queues for this VI.
4170	 */
4171	rc = t4_setup_vi_queues(vi);
4172	if (rc != 0)
4173		goto done;	/* error message displayed already */
4174
4175	/*
4176	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
4177	 */
4178	if (vi->nrxq > vi->rss_size) {
4179		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
4180		    "some queues will never receive traffic.\n", vi->nrxq,
4181		    vi->rss_size);
4182	} else if (vi->rss_size % vi->nrxq) {
4183		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
4184		    "expect uneven traffic distribution.\n", vi->nrxq,
4185		    vi->rss_size);
4186	}
4187#ifdef RSS
4188	MPASS(RSS_KEYSIZE == 40);
4189	if (vi->nrxq != nbuckets) {
4190		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
4191		    "performance will be impacted.\n", vi->nrxq, nbuckets);
4192	}
4193
4194	rss_getkey((void *)&raw_rss_key[0]);
4195	for (i = 0; i < nitems(rss_key); i++) {
4196		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
4197	}
4198	t4_write_rss_key(sc, &rss_key[0], -1);
4199#endif
4200	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
4201	for (i = 0; i < vi->rss_size;) {
4202#ifdef RSS
4203		j = rss_get_indirection_to_bucket(i);
4204		j %= vi->nrxq;
4205		rxq = &sc->sge.rxq[vi->first_rxq + j];
4206		rss[i++] = rxq->iq.abs_id;
4207#else
4208		for_each_rxq(vi, j, rxq) {
4209			rss[i++] = rxq->iq.abs_id;
4210			if (i == vi->rss_size)
4211				break;
4212		}
4213#endif
4214	}
4215
4216	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
4217	    vi->rss_size);
4218	if (rc != 0) {
4219		if_printf(ifp, "rss_config failed: %d\n", rc);
4220		goto done;
4221	}
4222
4223#ifdef RSS
4224	hashen = hashconfig_to_hashen(hashconfig);
4225
4226	/*
4227	 * We may have had to enable some hashes even though the global config
4228	 * wants them disabled.  This is a potential problem that must be
4229	 * reported to the user.
4230	 */
4231	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
4232
4233	/*
4234	 * If we consider only the supported hash types, then the enabled hashes
4235	 * are a superset of the requested hashes.  In other words, there cannot
4236	 * be any supported hash that was requested but not enabled, but there
4237	 * can be hashes that were not requested but had to be enabled.
4238	 */
4239	extra &= SUPPORTED_RSS_HASHTYPES;
4240	MPASS((extra & hashconfig) == 0);
4241
4242	if (extra) {
4243		if_printf(ifp,
4244		    "global RSS config (0x%x) cannot be accomodated.\n",
4245		    hashconfig);
4246	}
4247	if (extra & RSS_HASHTYPE_RSS_IPV4)
4248		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
4249	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
4250		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
4251	if (extra & RSS_HASHTYPE_RSS_IPV6)
4252		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
4253	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
4254		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
4255	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
4256		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
4257	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
4258		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
4259#else
4260	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
4261	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
4262	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4263	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
4264#endif
4265	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
4266	if (rc != 0) {
4267		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
4268		goto done;
4269	}
4270
4271	vi->rss = rss;
4272	vi->flags |= VI_INIT_DONE;
4273done:
4274	if (rc != 0)
4275		vi_full_uninit(vi);
4276
4277	return (rc);
4278}
4279
4280/*
4281 * Idempotent.
4282 */
4283int
4284vi_full_uninit(struct vi_info *vi)
4285{
4286	struct port_info *pi = vi->pi;
4287	struct adapter *sc = pi->adapter;
4288	int i;
4289	struct sge_rxq *rxq;
4290	struct sge_txq *txq;
4291#ifdef TCP_OFFLOAD
4292	struct sge_ofld_rxq *ofld_rxq;
4293	struct sge_wrq *ofld_txq;
4294#endif
4295
4296	if (vi->flags & VI_INIT_DONE) {
4297
4298		/* Need to quiesce queues.  */
4299
4300		/* XXX: Only for the first VI? */
4301		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
4302			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
4303
4304		for_each_txq(vi, i, txq) {
4305			quiesce_txq(sc, txq);
4306		}
4307
4308#ifdef TCP_OFFLOAD
4309		for_each_ofld_txq(vi, i, ofld_txq) {
4310			quiesce_wrq(sc, ofld_txq);
4311		}
4312#endif
4313
4314		for_each_rxq(vi, i, rxq) {
4315			quiesce_iq(sc, &rxq->iq);
4316			quiesce_fl(sc, &rxq->fl);
4317		}
4318
4319#ifdef TCP_OFFLOAD
4320		for_each_ofld_rxq(vi, i, ofld_rxq) {
4321			quiesce_iq(sc, &ofld_rxq->iq);
4322			quiesce_fl(sc, &ofld_rxq->fl);
4323		}
4324#endif
4325		free(vi->rss, M_CXGBE);
4326		free(vi->nm_rss, M_CXGBE);
4327	}
4328
4329	t4_teardown_vi_queues(vi);
4330	vi->flags &= ~VI_INIT_DONE;
4331
4332	return (0);
4333}
4334
4335static void
4336quiesce_txq(struct adapter *sc, struct sge_txq *txq)
4337{
4338	struct sge_eq *eq = &txq->eq;
4339	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
4340
4341	(void) sc;	/* unused */
4342
4343#ifdef INVARIANTS
4344	TXQ_LOCK(txq);
4345	MPASS((eq->flags & EQ_ENABLED) == 0);
4346	TXQ_UNLOCK(txq);
4347#endif
4348
4349	/* Wait for the mp_ring to empty. */
4350	while (!mp_ring_is_idle(txq->r)) {
4351		mp_ring_check_drainage(txq->r, 0);
4352		pause("rquiesce", 1);
4353	}
4354
4355	/* Then wait for the hardware to finish. */
4356	while (spg->cidx != htobe16(eq->pidx))
4357		pause("equiesce", 1);
4358
4359	/* Finally, wait for the driver to reclaim all descriptors. */
4360	while (eq->cidx != eq->pidx)
4361		pause("dquiesce", 1);
4362}
4363
4364static void
4365quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
4366{
4367
4368	/* XXXTX */
4369}
4370
4371static void
4372quiesce_iq(struct adapter *sc, struct sge_iq *iq)
4373{
4374	(void) sc;	/* unused */
4375
4376	/* Synchronize with the interrupt handler */
4377	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
4378		pause("iqfree", 1);
4379}
4380
4381static void
4382quiesce_fl(struct adapter *sc, struct sge_fl *fl)
4383{
4384	mtx_lock(&sc->sfl_lock);
4385	FL_LOCK(fl);
4386	fl->flags |= FL_DOOMED;
4387	FL_UNLOCK(fl);
4388	callout_stop(&sc->sfl_callout);
4389	mtx_unlock(&sc->sfl_lock);
4390
4391	KASSERT((fl->flags & FL_STARVING) == 0,
4392	    ("%s: still starving", __func__));
4393}
4394
4395static int
4396t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
4397    driver_intr_t *handler, void *arg, char *name)
4398{
4399	int rc;
4400
4401	irq->rid = rid;
4402	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
4403	    RF_SHAREABLE | RF_ACTIVE);
4404	if (irq->res == NULL) {
4405		device_printf(sc->dev,
4406		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
4407		return (ENOMEM);
4408	}
4409
4410	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
4411	    NULL, handler, arg, &irq->tag);
4412	if (rc != 0) {
4413		device_printf(sc->dev,
4414		    "failed to setup interrupt for rid %d, name %s: %d\n",
4415		    rid, name, rc);
4416	} else if (name)
4417		bus_describe_intr(sc->dev, irq->res, irq->tag, name);
4418
4419	return (rc);
4420}
4421
4422static int
4423t4_free_irq(struct adapter *sc, struct irq *irq)
4424{
4425	if (irq->tag)
4426		bus_teardown_intr(sc->dev, irq->res, irq->tag);
4427	if (irq->res)
4428		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
4429
4430	bzero(irq, sizeof(*irq));
4431
4432	return (0);
4433}
4434
4435static void
4436get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
4437{
4438
4439	regs->version = chip_id(sc) | chip_rev(sc) << 10;
4440	t4_get_regs(sc, buf, regs->len);
4441}
4442
4443#define	A_PL_INDIR_CMD	0x1f8
4444
4445#define	S_PL_AUTOINC	31
4446#define	M_PL_AUTOINC	0x1U
4447#define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
4448#define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
4449
4450#define	S_PL_VFID	20
4451#define	M_PL_VFID	0xffU
4452#define	V_PL_VFID(x)	((x) << S_PL_VFID)
4453#define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
4454
4455#define	S_PL_ADDR	0
4456#define	M_PL_ADDR	0xfffffU
4457#define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
4458#define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
4459
4460#define	A_PL_INDIR_DATA	0x1fc
4461
4462static uint64_t
4463read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
4464{
4465	u32 stats[2];
4466
4467	mtx_assert(&sc->reg_lock, MA_OWNED);
4468	if (sc->flags & IS_VF) {
4469		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
4470		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
4471	} else {
4472		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4473		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4474		    V_PL_ADDR(VF_MPS_REG(reg)));
4475		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
4476		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
4477	}
4478	return (((uint64_t)stats[1]) << 32 | stats[0]);
4479}
4480
4481static void
4482t4_get_vi_stats(struct adapter *sc, unsigned int viid,
4483    struct fw_vi_stats_vf *stats)
4484{
4485
4486#define GET_STAT(name) \
4487	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
4488
4489	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
4490	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
4491	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
4492	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
4493	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
4494	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
4495	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
4496	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
4497	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
4498	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
4499	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
4500	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
4501	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
4502	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
4503	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
4504	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
4505
4506#undef GET_STAT
4507}
4508
4509static void
4510t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
4511{
4512	int reg;
4513
4514	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4515	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4516	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
4517	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
4518	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
4519		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
4520}
4521
4522static void
4523vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
4524{
4525	struct ifnet *ifp = vi->ifp;
4526	struct sge_txq *txq;
4527	int i, drops;
4528	struct fw_vi_stats_vf *s = &vi->stats;
4529	struct timeval tv;
4530	const struct timeval interval = {0, 250000};	/* 250ms */
4531
4532	if (!(vi->flags & VI_INIT_DONE))
4533		return;
4534
4535	getmicrotime(&tv);
4536	timevalsub(&tv, &interval);
4537	if (timevalcmp(&tv, &vi->last_refreshed, <))
4538		return;
4539
4540	mtx_lock(&sc->reg_lock);
4541	t4_get_vi_stats(sc, vi->viid, &vi->stats);
4542
4543	ifp->if_ipackets = s->rx_bcast_frames + s->rx_mcast_frames +
4544	    s->rx_ucast_frames;
4545	ifp->if_ierrors = s->rx_err_frames;
4546	ifp->if_opackets = s->tx_bcast_frames + s->tx_mcast_frames +
4547	    s->tx_ucast_frames + s->tx_offload_frames;
4548	ifp->if_oerrors = s->tx_drop_frames;
4549	ifp->if_ibytes = s->rx_bcast_bytes + s->rx_mcast_bytes +
4550	    s->rx_ucast_bytes;
4551	ifp->if_obytes = s->tx_bcast_bytes + s->tx_mcast_bytes +
4552	    s->tx_ucast_bytes + s->tx_offload_bytes;
4553	ifp->if_imcasts = s->rx_mcast_frames;
4554	ifp->if_omcasts = s->tx_mcast_frames;
4555
4556	drops = 0;
4557	for_each_txq(vi, i, txq)
4558		drops += counter_u64_fetch(txq->r->drops);
4559	ifp->if_snd.ifq_drops = drops;
4560
4561	getmicrotime(&vi->last_refreshed);
4562	mtx_unlock(&sc->reg_lock);
4563}
4564
4565static void
4566cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
4567{
4568	struct vi_info *vi = &pi->vi[0];
4569	struct ifnet *ifp = vi->ifp;
4570	struct sge_txq *txq;
4571	int i, drops;
4572	struct port_stats *s = &pi->stats;
4573	struct timeval tv;
4574	const struct timeval interval = {0, 250000};	/* 250ms */
4575
4576	getmicrotime(&tv);
4577	timevalsub(&tv, &interval);
4578	if (timevalcmp(&tv, &pi->last_refreshed, <))
4579		return;
4580
4581	t4_get_port_stats(sc, pi->tx_chan, s);
4582
4583	ifp->if_opackets = s->tx_frames;
4584	ifp->if_ipackets = s->rx_frames;
4585	ifp->if_obytes = s->tx_octets;
4586	ifp->if_ibytes = s->rx_octets;
4587	ifp->if_omcasts = s->tx_mcast_frames;
4588	ifp->if_imcasts = s->rx_mcast_frames;
4589	ifp->if_iqdrops = s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
4590	    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
4591	    s->rx_trunc3;
4592	for (i = 0; i < sc->chip_params->nchan; i++) {
4593		if (pi->rx_chan_map & (1 << i)) {
4594			uint32_t v;
4595
4596			mtx_lock(&sc->reg_lock);
4597			t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
4598			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
4599			mtx_unlock(&sc->reg_lock);
4600			ifp->if_iqdrops += v;
4601		}
4602	}
4603
4604	drops = s->tx_drop;
4605	for_each_txq(vi, i, txq)
4606		drops += counter_u64_fetch(txq->r->drops);
4607	ifp->if_snd.ifq_drops = drops;
4608
4609	ifp->if_oerrors = s->tx_error_frames;
4610	ifp->if_ierrors = s->rx_jabber + s->rx_runt + s->rx_too_long +
4611	    s->rx_fcs_err + s->rx_len_err;
4612
4613	getmicrotime(&pi->last_refreshed);
4614}
4615
4616static void
4617cxgbe_tick(void *arg)
4618{
4619	struct port_info *pi = arg;
4620	struct adapter *sc = pi->adapter;
4621
4622	PORT_LOCK_ASSERT_OWNED(pi);
4623	cxgbe_refresh_stats(sc, pi);
4624
4625	callout_schedule(&pi->tick, hz);
4626}
4627
4628void
4629vi_tick(void *arg)
4630{
4631	struct vi_info *vi = arg;
4632	struct adapter *sc = vi->pi->adapter;
4633
4634	vi_refresh_stats(sc, vi);
4635
4636	callout_schedule(&vi->tick, hz);
4637}
4638
4639static void
4640cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
4641{
4642	struct ifnet *vlan;
4643
4644	if (arg != ifp || ifp->if_type != IFT_ETHER)
4645		return;
4646
4647	vlan = VLAN_DEVAT(ifp, vid);
4648	VLAN_SETCOOKIE(vlan, ifp);
4649}
4650
4651/*
4652 * Should match fw_caps_config_<foo> enums in t4fw_interface.h
4653 */
4654static char *caps_decoder[] = {
4655	"\20\001IPMI\002NCSI",				/* 0: NBM */
4656	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
4657	"\20\001INGRESS\002EGRESS",			/* 2: switch */
4658	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
4659	    "\006HASHFILTER\007ETHOFLD",
4660	"\20\001TOE",					/* 4: TOE */
4661	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
4662	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
4663	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
4664	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
4665	    "\007T10DIF"
4666	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
4667	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
4668	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
4669		    "\004PO_INITIATOR\005PO_TARGET",
4670};
4671
4672void
4673t4_sysctls(struct adapter *sc)
4674{
4675	struct sysctl_ctx_list *ctx;
4676	struct sysctl_oid *oid;
4677	struct sysctl_oid_list *children, *c0;
4678	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
4679
4680	ctx = device_get_sysctl_ctx(sc->dev);
4681
4682	/*
4683	 * dev.t4nex.X.
4684	 */
4685	oid = device_get_sysctl_tree(sc->dev);
4686	c0 = children = SYSCTL_CHILDREN(oid);
4687
4688	sc->sc_do_rxcopy = 1;
4689	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
4690	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
4691
4692	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
4693	    sc->params.nports, "# of ports");
4694
4695	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
4696	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
4697	    sysctl_bitfield, "A", "available doorbells");
4698
4699	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
4700	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
4701
4702	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
4703	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
4704	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
4705	    "interrupt holdoff timer values (us)");
4706
4707	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
4708	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
4709	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
4710	    "interrupt holdoff packet counter values");
4711
4712	t4_sge_sysctls(sc, ctx, children);
4713
4714	sc->lro_timeout = 100;
4715	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
4716	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
4717
4718	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
4719	    &sc->debug_flags, 0, "flags to enable runtime debugging");
4720
4721	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
4722	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
4723
4724	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
4725	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
4726
4727	if (sc->flags & IS_VF)
4728		return;
4729
4730	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
4731	    NULL, chip_rev(sc), "chip hardware revision");
4732
4733	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
4734	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
4735
4736	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
4737	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
4738
4739	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
4740	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
4741
4742	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
4743	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
4744
4745	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
4746	    sc->er_version, 0, "expansion ROM version");
4747
4748	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
4749	    sc->bs_version, 0, "bootstrap firmware version");
4750
4751	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
4752	    NULL, sc->params.scfg_vers, "serial config version");
4753
4754	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
4755	    NULL, sc->params.vpd_vers, "VPD version");
4756
4757	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
4758	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
4759
4760	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
4761	    sc->cfcsum, "config file checksum");
4762
4763#define SYSCTL_CAP(name, n, text) \
4764	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
4765	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
4766	    sysctl_bitfield, "A", "available " text " capabilities")
4767
4768	SYSCTL_CAP(nbmcaps, 0, "NBM");
4769	SYSCTL_CAP(linkcaps, 1, "link");
4770	SYSCTL_CAP(switchcaps, 2, "switch");
4771	SYSCTL_CAP(niccaps, 3, "NIC");
4772	SYSCTL_CAP(toecaps, 4, "TCP offload");
4773	SYSCTL_CAP(rdmacaps, 5, "RDMA");
4774	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
4775	SYSCTL_CAP(cryptocaps, 7, "crypto");
4776	SYSCTL_CAP(fcoecaps, 8, "FCoE");
4777#undef SYSCTL_CAP
4778
4779	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
4780	    NULL, sc->tids.nftids, "number of filters");
4781
4782	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
4783	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
4784	    "chip temperature (in Celsius)");
4785
4786#ifdef SBUF_DRAIN
4787	/*
4788	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
4789	 */
4790	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
4791	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
4792	    "logs and miscellaneous information");
4793	children = SYSCTL_CHILDREN(oid);
4794
4795	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
4796	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4797	    sysctl_cctrl, "A", "congestion control");
4798
4799	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
4800	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4801	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
4802
4803	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
4804	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
4805	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
4806
4807	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
4808	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
4809	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
4810
4811	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
4812	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
4813	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
4814
4815	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
4816	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
4817	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
4818
4819	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
4820	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
4821	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
4822
4823	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
4824	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4825	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
4826	    "A", "CIM logic analyzer");
4827
4828	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
4829	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4830	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
4831
4832	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
4833	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
4834	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
4835
4836	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
4837	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
4838	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
4839
4840	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
4841	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
4842	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
4843
4844	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
4845	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
4846	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
4847
4848	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
4849	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
4850	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
4851
4852	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
4853	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
4854	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
4855
4856	if (chip_id(sc) > CHELSIO_T4) {
4857		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
4858		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
4859		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
4860
4861		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
4862		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
4863		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
4864	}
4865
4866	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
4867	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4868	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
4869
4870	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
4871	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4872	    sysctl_cim_qcfg, "A", "CIM queue configuration");
4873
4874	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
4875	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4876	    sysctl_cpl_stats, "A", "CPL statistics");
4877
4878	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
4879	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4880	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
4881
4882	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
4883	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4884	    sysctl_devlog, "A", "firmware's device log");
4885
4886	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
4887	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4888	    sysctl_fcoe_stats, "A", "FCoE statistics");
4889
4890	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
4891	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4892	    sysctl_hw_sched, "A", "hardware scheduler ");
4893
4894	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
4895	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4896	    sysctl_l2t, "A", "hardware L2 table");
4897
4898	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
4899	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4900	    sysctl_lb_stats, "A", "loopback statistics");
4901
4902	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
4903	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4904	    sysctl_meminfo, "A", "memory regions");
4905
4906	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
4907	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4908	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
4909	    "A", "MPS TCAM entries");
4910
4911	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
4912	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4913	    sysctl_path_mtus, "A", "path MTUs");
4914
4915	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
4916	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4917	    sysctl_pm_stats, "A", "PM statistics");
4918
4919	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
4920	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4921	    sysctl_rdma_stats, "A", "RDMA statistics");
4922
4923	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
4924	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4925	    sysctl_tcp_stats, "A", "TCP statistics");
4926
4927	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
4928	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4929	    sysctl_tids, "A", "TID information");
4930
4931	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
4932	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4933	    sysctl_tp_err_stats, "A", "TP error statistics");
4934
4935	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
4936	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
4937	    "TP logic analyzer event capture mask");
4938
4939	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
4940	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4941	    sysctl_tp_la, "A", "TP logic analyzer");
4942
4943	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
4944	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4945	    sysctl_tx_rate, "A", "Tx rate");
4946
4947	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
4948	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4949	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
4950
4951	if (chip_id(sc) >= CHELSIO_T5) {
4952		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
4953		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4954		    sysctl_wcwr_stats, "A", "write combined work requests");
4955	}
4956#endif
4957
4958#ifdef TCP_OFFLOAD
4959	if (is_offload(sc)) {
4960		/*
4961		 * dev.t4nex.X.toe.
4962		 */
4963		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
4964		    NULL, "TOE parameters");
4965		children = SYSCTL_CHILDREN(oid);
4966
4967		sc->tt.sndbuf = 256 * 1024;
4968		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
4969		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
4970
4971		sc->tt.ddp = 0;
4972		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
4973		    &sc->tt.ddp, 0, "DDP allowed");
4974
4975		sc->tt.indsz = G_INDICATESIZE(t4_read_reg(sc, A_TP_PARA_REG5));
4976		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "indsz", CTLFLAG_RW,
4977		    &sc->tt.indsz, 0, "DDP max indicate size allowed");
4978
4979		sc->tt.ddp_thres =
4980		    G_RXCOALESCESIZE(t4_read_reg(sc, A_TP_PARA_REG2));
4981		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp_thres", CTLFLAG_RW,
4982		    &sc->tt.ddp_thres, 0, "DDP threshold");
4983
4984		sc->tt.rx_coalesce = 1;
4985		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
4986		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
4987
4988		sc->tt.tx_align = 1;
4989		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
4990		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
4991
4992		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
4993		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
4994		    "TP timer tick (us)");
4995
4996		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
4997		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
4998		    "TCP timestamp tick (us)");
4999
5000		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5001		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5002		    "DACK tick (us)");
5003
5004		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5005		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5006		    "IU", "DACK timer (us)");
5007
5008		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5009		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5010		    sysctl_tp_timer, "LU", "Retransmit min (us)");
5011
5012		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5013		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5014		    sysctl_tp_timer, "LU", "Retransmit max (us)");
5015
5016		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5017		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5018		    sysctl_tp_timer, "LU", "Persist timer min (us)");
5019
5020		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5021		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5022		    sysctl_tp_timer, "LU", "Persist timer max (us)");
5023
5024		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5025		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5026		    sysctl_tp_timer, "LU", "Keepidle idle timer (us)");
5027
5028		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_intvl",
5029		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
5030		    sysctl_tp_timer, "LU", "Keepidle interval (us)");
5031
5032		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
5033		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
5034		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
5035
5036		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
5037		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
5038		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
5039	}
5040#endif
5041}
5042
5043void
5044vi_sysctls(struct vi_info *vi)
5045{
5046	struct sysctl_ctx_list *ctx;
5047	struct sysctl_oid *oid;
5048	struct sysctl_oid_list *children;
5049
5050	ctx = device_get_sysctl_ctx(vi->dev);
5051
5052	/*
5053	 * dev.v?(cxgbe|cxl).X.
5054	 */
5055	oid = device_get_sysctl_tree(vi->dev);
5056	children = SYSCTL_CHILDREN(oid);
5057
5058	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
5059	    vi->viid, "VI identifer");
5060	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
5061	    &vi->nrxq, 0, "# of rx queues");
5062	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5063	    &vi->ntxq, 0, "# of tx queues");
5064	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5065	    &vi->first_rxq, 0, "index of first rx queue");
5066	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5067	    &vi->first_txq, 0, "index of first tx queue");
5068	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
5069	    vi->rss_size, "size of RSS indirection table");
5070
5071	if (IS_MAIN_VI(vi)) {
5072		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
5073		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
5074		    "Reserve queue 0 for non-flowid packets");
5075	}
5076
5077#ifdef TCP_OFFLOAD
5078	if (vi->nofldrxq != 0) {
5079		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5080		    &vi->nofldrxq, 0,
5081		    "# of rx queues for offloaded TCP connections");
5082		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5083		    &vi->nofldtxq, 0,
5084		    "# of tx queues for offloaded TCP connections");
5085		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5086		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
5087		    "index of first TOE rx queue");
5088		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5089		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
5090		    "index of first TOE tx queue");
5091	}
5092#endif
5093#ifdef DEV_NETMAP
5094	if (vi->nnmrxq != 0) {
5095		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5096		    &vi->nnmrxq, 0, "# of netmap rx queues");
5097		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5098		    &vi->nnmtxq, 0, "# of netmap tx queues");
5099		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5100		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
5101		    "index of first netmap rx queue");
5102		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
5103		    CTLFLAG_RD, &vi->first_nm_txq, 0,
5104		    "index of first netmap tx queue");
5105	}
5106#endif
5107
5108	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
5109	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
5110	    "holdoff timer index");
5111	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
5112	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
5113	    "holdoff packet counter index");
5114
5115	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
5116	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
5117	    "rx queue size");
5118	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
5119	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
5120	    "tx queue size");
5121}
5122
5123static void
5124cxgbe_sysctls(struct port_info *pi)
5125{
5126	struct sysctl_ctx_list *ctx;
5127	struct sysctl_oid *oid;
5128	struct sysctl_oid_list *children, *children2;
5129	struct adapter *sc = pi->adapter;
5130	int i;
5131	char name[16];
5132
5133	ctx = device_get_sysctl_ctx(pi->dev);
5134
5135	/*
5136	 * dev.cxgbe.X.
5137	 */
5138	oid = device_get_sysctl_tree(pi->dev);
5139	children = SYSCTL_CHILDREN(oid);
5140
5141	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
5142	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
5143	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
5144		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
5145		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
5146		    "PHY temperature (in Celsius)");
5147		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
5148		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
5149		    "PHY firmware version");
5150	}
5151
5152	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
5153	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
5154	    "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
5155	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
5156	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
5157	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
5158	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
5159	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
5160	    "autonegotiation (-1 = not supported)");
5161
5162	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
5163	    port_top_speed(pi), "max speed (in Gbps)");
5164
5165	if (sc->flags & IS_VF)
5166		return;
5167
5168	/*
5169	 * dev.(cxgbe|cxl).X.tc.
5170	 */
5171	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
5172	    "Tx scheduler traffic classes");
5173	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
5174		struct tx_sched_class *tc = &pi->tc[i];
5175
5176		snprintf(name, sizeof(name), "%d", i);
5177		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
5178		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
5179		    "traffic class"));
5180		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
5181		    &tc->flags, 0, "flags");
5182		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
5183		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
5184#ifdef SBUF_DRAIN
5185		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
5186		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
5187		    sysctl_tc_params, "A", "traffic class parameters");
5188#endif
5189	}
5190
5191	/*
5192	 * dev.cxgbe.X.stats.
5193	 */
5194	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
5195	    NULL, "port statistics");
5196	children = SYSCTL_CHILDREN(oid);
5197	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
5198	    &pi->tx_parse_error, 0,
5199	    "# of tx packets with invalid length or # of segments");
5200
5201#define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
5202	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
5203	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
5204	    sysctl_handle_t4_reg64, "QU", desc)
5205
5206	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
5207	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
5208	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
5209	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
5210	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
5211	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
5212	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
5213	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
5214	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
5215	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
5216	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
5217	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
5218	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
5219	    "# of tx frames in this range",
5220	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
5221	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
5222	    "# of tx frames in this range",
5223	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
5224	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
5225	    "# of tx frames in this range",
5226	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
5227	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
5228	    "# of tx frames in this range",
5229	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
5230	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
5231	    "# of tx frames in this range",
5232	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
5233	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
5234	    "# of tx frames in this range",
5235	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
5236	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
5237	    "# of tx frames in this range",
5238	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
5239	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
5240	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
5241	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
5242	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
5243	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
5244	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
5245	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5246	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5247	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5248	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5249	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5250	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5251	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5252	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5253	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5254	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5255	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5256	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5257	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5258	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5259
5260	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5261	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5262	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5263	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5264	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5265	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5266	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5267	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5268	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5269	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5270	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5271	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5272	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5273	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5274	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5275	    "# of frames received with bad FCS",
5276	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5277	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5278	    "# of frames received with length error",
5279	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5280	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5281	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5282	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5283	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5284	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5285	    "# of rx frames in this range",
5286	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5287	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5288	    "# of rx frames in this range",
5289	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5290	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5291	    "# of rx frames in this range",
5292	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5293	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5294	    "# of rx frames in this range",
5295	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5296	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5297	    "# of rx frames in this range",
5298	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5299	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5300	    "# of rx frames in this range",
5301	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5302	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5303	    "# of rx frames in this range",
5304	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5305	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5306	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5307	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5308	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5309	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5310	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5311	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5312	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5313	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5314	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5315	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5316	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5317	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5318	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5319	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5320	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5321	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5322	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5323
5324#undef SYSCTL_ADD_T4_REG64
5325
5326#define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5327	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5328	    &pi->stats.name, desc)
5329
5330	/* We get these from port_stats and they may be stale by upto 1s */
5331	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5332	    "# drops due to buffer-group 0 overflows");
5333	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5334	    "# drops due to buffer-group 1 overflows");
5335	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5336	    "# drops due to buffer-group 2 overflows");
5337	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5338	    "# drops due to buffer-group 3 overflows");
5339	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5340	    "# of buffer-group 0 truncated packets");
5341	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5342	    "# of buffer-group 1 truncated packets");
5343	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5344	    "# of buffer-group 2 truncated packets");
5345	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5346	    "# of buffer-group 3 truncated packets");
5347
5348#undef SYSCTL_ADD_T4_PORTSTAT
5349}
5350
5351static int
5352sysctl_int_array(SYSCTL_HANDLER_ARGS)
5353{
5354	int rc, *i, space = 0;
5355	struct sbuf sb;
5356
5357	sbuf_new(&sb, NULL, 32, SBUF_AUTOEXTEND);
5358	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
5359		if (space)
5360			sbuf_printf(&sb, " ");
5361		sbuf_printf(&sb, "%d", *i);
5362		space = 1;
5363	}
5364	sbuf_finish(&sb);
5365	rc = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
5366	sbuf_delete(&sb);
5367	return (rc);
5368}
5369
5370static int
5371sysctl_bitfield(SYSCTL_HANDLER_ARGS)
5372{
5373	int rc;
5374	struct sbuf *sb;
5375
5376	rc = sysctl_wire_old_buffer(req, 0);
5377	if (rc != 0)
5378		return(rc);
5379
5380	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5381	if (sb == NULL)
5382		return (ENOMEM);
5383
5384	sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
5385	rc = sbuf_finish(sb);
5386	sbuf_delete(sb);
5387
5388	return (rc);
5389}
5390
5391static int
5392sysctl_btphy(SYSCTL_HANDLER_ARGS)
5393{
5394	struct port_info *pi = arg1;
5395	int op = arg2;
5396	struct adapter *sc = pi->adapter;
5397	u_int v;
5398	int rc;
5399
5400	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
5401	if (rc)
5402		return (rc);
5403	/* XXX: magic numbers */
5404	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
5405	    &v);
5406	end_synchronized_op(sc, 0);
5407	if (rc)
5408		return (rc);
5409	if (op == 0)
5410		v /= 256;
5411
5412	rc = sysctl_handle_int(oidp, &v, 0, req);
5413	return (rc);
5414}
5415
5416static int
5417sysctl_noflowq(SYSCTL_HANDLER_ARGS)
5418{
5419	struct vi_info *vi = arg1;
5420	int rc, val;
5421
5422	val = vi->rsrv_noflowq;
5423	rc = sysctl_handle_int(oidp, &val, 0, req);
5424	if (rc != 0 || req->newptr == NULL)
5425		return (rc);
5426
5427	if ((val >= 1) && (vi->ntxq > 1))
5428		vi->rsrv_noflowq = 1;
5429	else
5430		vi->rsrv_noflowq = 0;
5431
5432	return (rc);
5433}
5434
5435static int
5436sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
5437{
5438	struct vi_info *vi = arg1;
5439	struct adapter *sc = vi->pi->adapter;
5440	int idx, rc, i;
5441	struct sge_rxq *rxq;
5442#ifdef TCP_OFFLOAD
5443	struct sge_ofld_rxq *ofld_rxq;
5444#endif
5445	uint8_t v;
5446
5447	idx = vi->tmr_idx;
5448
5449	rc = sysctl_handle_int(oidp, &idx, 0, req);
5450	if (rc != 0 || req->newptr == NULL)
5451		return (rc);
5452
5453	if (idx < 0 || idx >= SGE_NTIMERS)
5454		return (EINVAL);
5455
5456	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5457	    "t4tmr");
5458	if (rc)
5459		return (rc);
5460
5461	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
5462	for_each_rxq(vi, i, rxq) {
5463#ifdef atomic_store_rel_8
5464		atomic_store_rel_8(&rxq->iq.intr_params, v);
5465#else
5466		rxq->iq.intr_params = v;
5467#endif
5468	}
5469#ifdef TCP_OFFLOAD
5470	for_each_ofld_rxq(vi, i, ofld_rxq) {
5471#ifdef atomic_store_rel_8
5472		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
5473#else
5474		ofld_rxq->iq.intr_params = v;
5475#endif
5476	}
5477#endif
5478	vi->tmr_idx = idx;
5479
5480	end_synchronized_op(sc, LOCK_HELD);
5481	return (0);
5482}
5483
5484static int
5485sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
5486{
5487	struct vi_info *vi = arg1;
5488	struct adapter *sc = vi->pi->adapter;
5489	int idx, rc;
5490
5491	idx = vi->pktc_idx;
5492
5493	rc = sysctl_handle_int(oidp, &idx, 0, req);
5494	if (rc != 0 || req->newptr == NULL)
5495		return (rc);
5496
5497	if (idx < -1 || idx >= SGE_NCOUNTERS)
5498		return (EINVAL);
5499
5500	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5501	    "t4pktc");
5502	if (rc)
5503		return (rc);
5504
5505	if (vi->flags & VI_INIT_DONE)
5506		rc = EBUSY; /* cannot be changed once the queues are created */
5507	else
5508		vi->pktc_idx = idx;
5509
5510	end_synchronized_op(sc, LOCK_HELD);
5511	return (rc);
5512}
5513
5514static int
5515sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
5516{
5517	struct vi_info *vi = arg1;
5518	struct adapter *sc = vi->pi->adapter;
5519	int qsize, rc;
5520
5521	qsize = vi->qsize_rxq;
5522
5523	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5524	if (rc != 0 || req->newptr == NULL)
5525		return (rc);
5526
5527	if (qsize < 128 || (qsize & 7))
5528		return (EINVAL);
5529
5530	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5531	    "t4rxqs");
5532	if (rc)
5533		return (rc);
5534
5535	if (vi->flags & VI_INIT_DONE)
5536		rc = EBUSY; /* cannot be changed once the queues are created */
5537	else
5538		vi->qsize_rxq = qsize;
5539
5540	end_synchronized_op(sc, LOCK_HELD);
5541	return (rc);
5542}
5543
5544static int
5545sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
5546{
5547	struct vi_info *vi = arg1;
5548	struct adapter *sc = vi->pi->adapter;
5549	int qsize, rc;
5550
5551	qsize = vi->qsize_txq;
5552
5553	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5554	if (rc != 0 || req->newptr == NULL)
5555		return (rc);
5556
5557	if (qsize < 128 || qsize > 65536)
5558		return (EINVAL);
5559
5560	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5561	    "t4txqs");
5562	if (rc)
5563		return (rc);
5564
5565	if (vi->flags & VI_INIT_DONE)
5566		rc = EBUSY; /* cannot be changed once the queues are created */
5567	else
5568		vi->qsize_txq = qsize;
5569
5570	end_synchronized_op(sc, LOCK_HELD);
5571	return (rc);
5572}
5573
5574static int
5575sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
5576{
5577	struct port_info *pi = arg1;
5578	struct adapter *sc = pi->adapter;
5579	struct link_config *lc = &pi->link_cfg;
5580	int rc;
5581
5582	if (req->newptr == NULL) {
5583		struct sbuf *sb;
5584		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
5585
5586		rc = sysctl_wire_old_buffer(req, 0);
5587		if (rc != 0)
5588			return(rc);
5589
5590		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5591		if (sb == NULL)
5592			return (ENOMEM);
5593
5594		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
5595		rc = sbuf_finish(sb);
5596		sbuf_delete(sb);
5597	} else {
5598		char s[2];
5599		int n;
5600
5601		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
5602		s[1] = 0;
5603
5604		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5605		if (rc != 0)
5606			return(rc);
5607
5608		if (s[1] != 0)
5609			return (EINVAL);
5610		if (s[0] < '0' || s[0] > '9')
5611			return (EINVAL);	/* not a number */
5612		n = s[0] - '0';
5613		if (n & ~(PAUSE_TX | PAUSE_RX))
5614			return (EINVAL);	/* some other bit is set too */
5615
5616		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5617		    "t4PAUSE");
5618		if (rc)
5619			return (rc);
5620		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
5621			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
5622			lc->requested_fc |= n;
5623			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5624		}
5625		end_synchronized_op(sc, 0);
5626	}
5627
5628	return (rc);
5629}
5630
5631static int
5632sysctl_fec(SYSCTL_HANDLER_ARGS)
5633{
5634	struct port_info *pi = arg1;
5635	struct adapter *sc = pi->adapter;
5636	struct link_config *lc = &pi->link_cfg;
5637	int rc;
5638
5639	if (req->newptr == NULL) {
5640		struct sbuf *sb;
5641		static char *bits = "\20\1RS\2BASER_RS\3RESERVED";
5642
5643		rc = sysctl_wire_old_buffer(req, 0);
5644		if (rc != 0)
5645			return(rc);
5646
5647		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5648		if (sb == NULL)
5649			return (ENOMEM);
5650
5651		sbuf_printf(sb, "%b", lc->fec & M_FW_PORT_CAP_FEC, bits);
5652		rc = sbuf_finish(sb);
5653		sbuf_delete(sb);
5654	} else {
5655		char s[2];
5656		int n;
5657
5658		s[0] = '0' + (lc->requested_fec & M_FW_PORT_CAP_FEC);
5659		s[1] = 0;
5660
5661		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5662		if (rc != 0)
5663			return(rc);
5664
5665		if (s[1] != 0)
5666			return (EINVAL);
5667		if (s[0] < '0' || s[0] > '9')
5668			return (EINVAL);	/* not a number */
5669		n = s[0] - '0';
5670		if (n & ~M_FW_PORT_CAP_FEC)
5671			return (EINVAL);	/* some other bit is set too */
5672
5673		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5674		    "t4fec");
5675		if (rc)
5676			return (rc);
5677		if ((lc->requested_fec & M_FW_PORT_CAP_FEC) != n) {
5678			lc->requested_fec = n &
5679			    G_FW_PORT_CAP_FEC(lc->supported);
5680			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5681		}
5682		end_synchronized_op(sc, 0);
5683	}
5684
5685	return (rc);
5686}
5687
5688static int
5689sysctl_autoneg(SYSCTL_HANDLER_ARGS)
5690{
5691	struct port_info *pi = arg1;
5692	struct adapter *sc = pi->adapter;
5693	struct link_config *lc = &pi->link_cfg;
5694	int rc, val, old;
5695
5696	if (lc->supported & FW_PORT_CAP_ANEG)
5697		val = lc->autoneg == AUTONEG_ENABLE ? 1 : 0;
5698	else
5699		val = -1;
5700	rc = sysctl_handle_int(oidp, &val, 0, req);
5701	if (rc != 0 || req->newptr == NULL)
5702		return (rc);
5703	if ((lc->supported & FW_PORT_CAP_ANEG) == 0)
5704		return (ENOTSUP);
5705
5706	val = val ? AUTONEG_ENABLE : AUTONEG_DISABLE;
5707	if (lc->autoneg == val)
5708		return (0);	/* no change */
5709
5710	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5711	    "t4aneg");
5712	if (rc)
5713		return (rc);
5714	old = lc->autoneg;
5715	lc->autoneg = val;
5716	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5717	if (rc != 0)
5718		lc->autoneg = old;
5719	return (rc);
5720}
5721
5722static int
5723sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
5724{
5725	struct adapter *sc = arg1;
5726	int reg = arg2;
5727	uint64_t val;
5728
5729	val = t4_read_reg64(sc, reg);
5730
5731	return (sysctl_handle_64(oidp, &val, 0, req));
5732}
5733
5734static int
5735sysctl_temperature(SYSCTL_HANDLER_ARGS)
5736{
5737	struct adapter *sc = arg1;
5738	int rc, t;
5739	uint32_t param, val;
5740
5741	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
5742	if (rc)
5743		return (rc);
5744	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5745	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
5746	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
5747	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
5748	end_synchronized_op(sc, 0);
5749	if (rc)
5750		return (rc);
5751
5752	/* unknown is returned as 0 but we display -1 in that case */
5753	t = val == 0 ? -1 : val;
5754
5755	rc = sysctl_handle_int(oidp, &t, 0, req);
5756	return (rc);
5757}
5758
5759#ifdef SBUF_DRAIN
5760static int
5761sysctl_cctrl(SYSCTL_HANDLER_ARGS)
5762{
5763	struct adapter *sc = arg1;
5764	struct sbuf *sb;
5765	int rc, i;
5766	uint16_t incr[NMTUS][NCCTRL_WIN];
5767	static const char *dec_fac[] = {
5768		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
5769		"0.9375"
5770	};
5771
5772	rc = sysctl_wire_old_buffer(req, 0);
5773	if (rc != 0)
5774		return (rc);
5775
5776	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5777	if (sb == NULL)
5778		return (ENOMEM);
5779
5780	t4_read_cong_tbl(sc, incr);
5781
5782	for (i = 0; i < NCCTRL_WIN; ++i) {
5783		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
5784		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
5785		    incr[5][i], incr[6][i], incr[7][i]);
5786		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
5787		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
5788		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
5789		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
5790	}
5791
5792	rc = sbuf_finish(sb);
5793	sbuf_delete(sb);
5794
5795	return (rc);
5796}
5797
5798static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
5799	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
5800	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
5801	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
5802};
5803
5804static int
5805sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
5806{
5807	struct adapter *sc = arg1;
5808	struct sbuf *sb;
5809	int rc, i, n, qid = arg2;
5810	uint32_t *buf, *p;
5811	char *qtype;
5812	u_int cim_num_obq = sc->chip_params->cim_num_obq;
5813
5814	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
5815	    ("%s: bad qid %d\n", __func__, qid));
5816
5817	if (qid < CIM_NUM_IBQ) {
5818		/* inbound queue */
5819		qtype = "IBQ";
5820		n = 4 * CIM_IBQ_SIZE;
5821		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5822		rc = t4_read_cim_ibq(sc, qid, buf, n);
5823	} else {
5824		/* outbound queue */
5825		qtype = "OBQ";
5826		qid -= CIM_NUM_IBQ;
5827		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
5828		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5829		rc = t4_read_cim_obq(sc, qid, buf, n);
5830	}
5831
5832	if (rc < 0) {
5833		rc = -rc;
5834		goto done;
5835	}
5836	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
5837
5838	rc = sysctl_wire_old_buffer(req, 0);
5839	if (rc != 0)
5840		goto done;
5841
5842	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
5843	if (sb == NULL) {
5844		rc = ENOMEM;
5845		goto done;
5846	}
5847
5848	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
5849	for (i = 0, p = buf; i < n; i += 16, p += 4)
5850		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
5851		    p[2], p[3]);
5852
5853	rc = sbuf_finish(sb);
5854	sbuf_delete(sb);
5855done:
5856	free(buf, M_CXGBE);
5857	return (rc);
5858}
5859
5860static int
5861sysctl_cim_la(SYSCTL_HANDLER_ARGS)
5862{
5863	struct adapter *sc = arg1;
5864	u_int cfg;
5865	struct sbuf *sb;
5866	uint32_t *buf, *p;
5867	int rc;
5868
5869	MPASS(chip_id(sc) <= CHELSIO_T5);
5870
5871	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
5872	if (rc != 0)
5873		return (rc);
5874
5875	rc = sysctl_wire_old_buffer(req, 0);
5876	if (rc != 0)
5877		return (rc);
5878
5879	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5880	if (sb == NULL)
5881		return (ENOMEM);
5882
5883	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
5884	    M_ZERO | M_WAITOK);
5885
5886	rc = -t4_cim_read_la(sc, buf, NULL);
5887	if (rc != 0)
5888		goto done;
5889
5890	sbuf_printf(sb, "Status   Data      PC%s",
5891	    cfg & F_UPDBGLACAPTPCONLY ? "" :
5892	    "     LS0Stat  LS0Addr             LS0Data");
5893
5894	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
5895		if (cfg & F_UPDBGLACAPTPCONLY) {
5896			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
5897			    p[6], p[7]);
5898			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
5899			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
5900			    p[4] & 0xff, p[5] >> 8);
5901			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
5902			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
5903			    p[1] & 0xf, p[2] >> 4);
5904		} else {
5905			sbuf_printf(sb,
5906			    "\n  %02x   %x%07x %x%07x %08x %08x "
5907			    "%08x%08x%08x%08x",
5908			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
5909			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
5910			    p[6], p[7]);
5911		}
5912	}
5913
5914	rc = sbuf_finish(sb);
5915	sbuf_delete(sb);
5916done:
5917	free(buf, M_CXGBE);
5918	return (rc);
5919}
5920
5921static int
5922sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
5923{
5924	struct adapter *sc = arg1;
5925	u_int cfg;
5926	struct sbuf *sb;
5927	uint32_t *buf, *p;
5928	int rc;
5929
5930	MPASS(chip_id(sc) > CHELSIO_T5);
5931
5932	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
5933	if (rc != 0)
5934		return (rc);
5935
5936	rc = sysctl_wire_old_buffer(req, 0);
5937	if (rc != 0)
5938		return (rc);
5939
5940	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5941	if (sb == NULL)
5942		return (ENOMEM);
5943
5944	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
5945	    M_ZERO | M_WAITOK);
5946
5947	rc = -t4_cim_read_la(sc, buf, NULL);
5948	if (rc != 0)
5949		goto done;
5950
5951	sbuf_printf(sb, "Status   Inst    Data      PC%s",
5952	    cfg & F_UPDBGLACAPTPCONLY ? "" :
5953	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
5954
5955	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
5956		if (cfg & F_UPDBGLACAPTPCONLY) {
5957			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
5958			    p[3] & 0xff, p[2], p[1], p[0]);
5959			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
5960			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
5961			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
5962			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
5963			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
5964			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
5965			    p[6] >> 16);
5966		} else {
5967			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
5968			    "%08x %08x %08x %08x %08x %08x",
5969			    (p[9] >> 16) & 0xff,
5970			    p[9] & 0xffff, p[8] >> 16,
5971			    p[8] & 0xffff, p[7] >> 16,
5972			    p[7] & 0xffff, p[6] >> 16,
5973			    p[2], p[1], p[0], p[5], p[4], p[3]);
5974		}
5975	}
5976
5977	rc = sbuf_finish(sb);
5978	sbuf_delete(sb);
5979done:
5980	free(buf, M_CXGBE);
5981	return (rc);
5982}
5983
5984static int
5985sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
5986{
5987	struct adapter *sc = arg1;
5988	u_int i;
5989	struct sbuf *sb;
5990	uint32_t *buf, *p;
5991	int rc;
5992
5993	rc = sysctl_wire_old_buffer(req, 0);
5994	if (rc != 0)
5995		return (rc);
5996
5997	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5998	if (sb == NULL)
5999		return (ENOMEM);
6000
6001	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
6002	    M_ZERO | M_WAITOK);
6003
6004	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
6005	p = buf;
6006
6007	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6008		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
6009		    p[1], p[0]);
6010	}
6011
6012	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
6013	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6014		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
6015		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
6016		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
6017		    (p[1] >> 2) | ((p[2] & 3) << 30),
6018		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
6019		    p[0] & 1);
6020	}
6021
6022	rc = sbuf_finish(sb);
6023	sbuf_delete(sb);
6024	free(buf, M_CXGBE);
6025	return (rc);
6026}
6027
6028static int
6029sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
6030{
6031	struct adapter *sc = arg1;
6032	u_int i;
6033	struct sbuf *sb;
6034	uint32_t *buf, *p;
6035	int rc;
6036
6037	rc = sysctl_wire_old_buffer(req, 0);
6038	if (rc != 0)
6039		return (rc);
6040
6041	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6042	if (sb == NULL)
6043		return (ENOMEM);
6044
6045	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
6046	    M_ZERO | M_WAITOK);
6047
6048	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
6049	p = buf;
6050
6051	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
6052	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6053		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
6054		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
6055		    p[4], p[3], p[2], p[1], p[0]);
6056	}
6057
6058	sbuf_printf(sb, "\n\nCntl ID               Data");
6059	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6060		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
6061		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
6062	}
6063
6064	rc = sbuf_finish(sb);
6065	sbuf_delete(sb);
6066	free(buf, M_CXGBE);
6067	return (rc);
6068}
6069
6070static int
6071sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
6072{
6073	struct adapter *sc = arg1;
6074	struct sbuf *sb;
6075	int rc, i;
6076	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6077	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6078	uint16_t thres[CIM_NUM_IBQ];
6079	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
6080	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
6081	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
6082
6083	cim_num_obq = sc->chip_params->cim_num_obq;
6084	if (is_t4(sc)) {
6085		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
6086		obq_rdaddr = A_UP_OBQ_0_REALADDR;
6087	} else {
6088		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
6089		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
6090	}
6091	nq = CIM_NUM_IBQ + cim_num_obq;
6092
6093	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
6094	if (rc == 0)
6095		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
6096	if (rc != 0)
6097		return (rc);
6098
6099	t4_read_cimq_cfg(sc, base, size, thres);
6100
6101	rc = sysctl_wire_old_buffer(req, 0);
6102	if (rc != 0)
6103		return (rc);
6104
6105	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6106	if (sb == NULL)
6107		return (ENOMEM);
6108
6109	sbuf_printf(sb,
6110	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
6111
6112	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
6113		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
6114		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
6115		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6116		    G_QUEREMFLITS(p[2]) * 16);
6117	for ( ; i < nq; i++, p += 4, wr += 2)
6118		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
6119		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
6120		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6121		    G_QUEREMFLITS(p[2]) * 16);
6122
6123	rc = sbuf_finish(sb);
6124	sbuf_delete(sb);
6125
6126	return (rc);
6127}
6128
6129static int
6130sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
6131{
6132	struct adapter *sc = arg1;
6133	struct sbuf *sb;
6134	int rc;
6135	struct tp_cpl_stats stats;
6136
6137	rc = sysctl_wire_old_buffer(req, 0);
6138	if (rc != 0)
6139		return (rc);
6140
6141	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6142	if (sb == NULL)
6143		return (ENOMEM);
6144
6145	mtx_lock(&sc->reg_lock);
6146	t4_tp_get_cpl_stats(sc, &stats);
6147	mtx_unlock(&sc->reg_lock);
6148
6149	if (sc->chip_params->nchan > 2) {
6150		sbuf_printf(sb, "                 channel 0  channel 1"
6151		    "  channel 2  channel 3");
6152		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
6153		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
6154		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
6155		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
6156	} else {
6157		sbuf_printf(sb, "                 channel 0  channel 1");
6158		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
6159		    stats.req[0], stats.req[1]);
6160		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
6161		    stats.rsp[0], stats.rsp[1]);
6162	}
6163
6164	rc = sbuf_finish(sb);
6165	sbuf_delete(sb);
6166
6167	return (rc);
6168}
6169
6170static int
6171sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
6172{
6173	struct adapter *sc = arg1;
6174	struct sbuf *sb;
6175	int rc;
6176	struct tp_usm_stats stats;
6177
6178	rc = sysctl_wire_old_buffer(req, 0);
6179	if (rc != 0)
6180		return(rc);
6181
6182	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6183	if (sb == NULL)
6184		return (ENOMEM);
6185
6186	t4_get_usm_stats(sc, &stats);
6187
6188	sbuf_printf(sb, "Frames: %u\n", stats.frames);
6189	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
6190	sbuf_printf(sb, "Drops:  %u", stats.drops);
6191
6192	rc = sbuf_finish(sb);
6193	sbuf_delete(sb);
6194
6195	return (rc);
6196}
6197
6198static const char * const devlog_level_strings[] = {
6199	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
6200	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
6201	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
6202	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
6203	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
6204	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
6205};
6206
6207static const char * const devlog_facility_strings[] = {
6208	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
6209	[FW_DEVLOG_FACILITY_CF]		= "CF",
6210	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
6211	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
6212	[FW_DEVLOG_FACILITY_RES]	= "RES",
6213	[FW_DEVLOG_FACILITY_HW]		= "HW",
6214	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
6215	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
6216	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
6217	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
6218	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
6219	[FW_DEVLOG_FACILITY_VI]		= "VI",
6220	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
6221	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
6222	[FW_DEVLOG_FACILITY_TM]		= "TM",
6223	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
6224	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
6225	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
6226	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
6227	[FW_DEVLOG_FACILITY_RI]		= "RI",
6228	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
6229	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
6230	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
6231	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
6232	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
6233};
6234
6235static int
6236sysctl_devlog(SYSCTL_HANDLER_ARGS)
6237{
6238	struct adapter *sc = arg1;
6239	struct devlog_params *dparams = &sc->params.devlog;
6240	struct fw_devlog_e *buf, *e;
6241	int i, j, rc, nentries, first = 0;
6242	struct sbuf *sb;
6243	uint64_t ftstamp = UINT64_MAX;
6244
6245	if (dparams->addr == 0)
6246		return (ENXIO);
6247
6248	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
6249	if (buf == NULL)
6250		return (ENOMEM);
6251
6252	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
6253	if (rc != 0)
6254		goto done;
6255
6256	nentries = dparams->size / sizeof(struct fw_devlog_e);
6257	for (i = 0; i < nentries; i++) {
6258		e = &buf[i];
6259
6260		if (e->timestamp == 0)
6261			break;	/* end */
6262
6263		e->timestamp = be64toh(e->timestamp);
6264		e->seqno = be32toh(e->seqno);
6265		for (j = 0; j < 8; j++)
6266			e->params[j] = be32toh(e->params[j]);
6267
6268		if (e->timestamp < ftstamp) {
6269			ftstamp = e->timestamp;
6270			first = i;
6271		}
6272	}
6273
6274	if (buf[first].timestamp == 0)
6275		goto done;	/* nothing in the log */
6276
6277	rc = sysctl_wire_old_buffer(req, 0);
6278	if (rc != 0)
6279		goto done;
6280
6281	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6282	if (sb == NULL) {
6283		rc = ENOMEM;
6284		goto done;
6285	}
6286	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
6287	    "Seq#", "Tstamp", "Level", "Facility", "Message");
6288
6289	i = first;
6290	do {
6291		e = &buf[i];
6292		if (e->timestamp == 0)
6293			break;	/* end */
6294
6295		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
6296		    e->seqno, e->timestamp,
6297		    (e->level < nitems(devlog_level_strings) ?
6298			devlog_level_strings[e->level] : "UNKNOWN"),
6299		    (e->facility < nitems(devlog_facility_strings) ?
6300			devlog_facility_strings[e->facility] : "UNKNOWN"));
6301		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
6302		    e->params[2], e->params[3], e->params[4],
6303		    e->params[5], e->params[6], e->params[7]);
6304
6305		if (++i == nentries)
6306			i = 0;
6307	} while (i != first);
6308
6309	rc = sbuf_finish(sb);
6310	sbuf_delete(sb);
6311done:
6312	free(buf, M_CXGBE);
6313	return (rc);
6314}
6315
6316static int
6317sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
6318{
6319	struct adapter *sc = arg1;
6320	struct sbuf *sb;
6321	int rc;
6322	struct tp_fcoe_stats stats[MAX_NCHAN];
6323	int i, nchan = sc->chip_params->nchan;
6324
6325	rc = sysctl_wire_old_buffer(req, 0);
6326	if (rc != 0)
6327		return (rc);
6328
6329	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6330	if (sb == NULL)
6331		return (ENOMEM);
6332
6333	for (i = 0; i < nchan; i++)
6334		t4_get_fcoe_stats(sc, i, &stats[i]);
6335
6336	if (nchan > 2) {
6337		sbuf_printf(sb, "                   channel 0        channel 1"
6338		    "        channel 2        channel 3");
6339		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
6340		    stats[0].octets_ddp, stats[1].octets_ddp,
6341		    stats[2].octets_ddp, stats[3].octets_ddp);
6342		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
6343		    stats[0].frames_ddp, stats[1].frames_ddp,
6344		    stats[2].frames_ddp, stats[3].frames_ddp);
6345		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
6346		    stats[0].frames_drop, stats[1].frames_drop,
6347		    stats[2].frames_drop, stats[3].frames_drop);
6348	} else {
6349		sbuf_printf(sb, "                   channel 0        channel 1");
6350		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
6351		    stats[0].octets_ddp, stats[1].octets_ddp);
6352		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
6353		    stats[0].frames_ddp, stats[1].frames_ddp);
6354		sbuf_printf(sb, "\nframesDrop: %16u %16u",
6355		    stats[0].frames_drop, stats[1].frames_drop);
6356	}
6357
6358	rc = sbuf_finish(sb);
6359	sbuf_delete(sb);
6360
6361	return (rc);
6362}
6363
6364static int
6365sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
6366{
6367	struct adapter *sc = arg1;
6368	struct sbuf *sb;
6369	int rc, i;
6370	unsigned int map, kbps, ipg, mode;
6371	unsigned int pace_tab[NTX_SCHED];
6372
6373	rc = sysctl_wire_old_buffer(req, 0);
6374	if (rc != 0)
6375		return (rc);
6376
6377	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6378	if (sb == NULL)
6379		return (ENOMEM);
6380
6381	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
6382	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
6383	t4_read_pace_tbl(sc, pace_tab);
6384
6385	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
6386	    "Class IPG (0.1 ns)   Flow IPG (us)");
6387
6388	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
6389		t4_get_tx_sched(sc, i, &kbps, &ipg);
6390		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
6391		    (mode & (1 << i)) ? "flow" : "class", map & 3);
6392		if (kbps)
6393			sbuf_printf(sb, "%9u     ", kbps);
6394		else
6395			sbuf_printf(sb, " disabled     ");
6396
6397		if (ipg)
6398			sbuf_printf(sb, "%13u        ", ipg);
6399		else
6400			sbuf_printf(sb, "     disabled        ");
6401
6402		if (pace_tab[i])
6403			sbuf_printf(sb, "%10u", pace_tab[i]);
6404		else
6405			sbuf_printf(sb, "  disabled");
6406	}
6407
6408	rc = sbuf_finish(sb);
6409	sbuf_delete(sb);
6410
6411	return (rc);
6412}
6413
6414static int
6415sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
6416{
6417	struct adapter *sc = arg1;
6418	struct sbuf *sb;
6419	int rc, i, j;
6420	uint64_t *p0, *p1;
6421	struct lb_port_stats s[2];
6422	static const char *stat_name[] = {
6423		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
6424		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
6425		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
6426		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
6427		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
6428		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
6429		"BG2FramesTrunc:", "BG3FramesTrunc:"
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, 4096, req);
6437	if (sb == NULL)
6438		return (ENOMEM);
6439
6440	memset(s, 0, sizeof(s));
6441
6442	for (i = 0; i < sc->chip_params->nchan; i += 2) {
6443		t4_get_lb_stats(sc, i, &s[0]);
6444		t4_get_lb_stats(sc, i + 1, &s[1]);
6445
6446		p0 = &s[0].octets;
6447		p1 = &s[1].octets;
6448		sbuf_printf(sb, "%s                       Loopback %u"
6449		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
6450
6451		for (j = 0; j < nitems(stat_name); j++)
6452			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
6453				   *p0++, *p1++);
6454	}
6455
6456	rc = sbuf_finish(sb);
6457	sbuf_delete(sb);
6458
6459	return (rc);
6460}
6461
6462static int
6463sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
6464{
6465	int rc = 0;
6466	struct port_info *pi = arg1;
6467	struct link_config *lc = &pi->link_cfg;
6468	struct sbuf *sb;
6469
6470	rc = sysctl_wire_old_buffer(req, 0);
6471	if (rc != 0)
6472		return(rc);
6473	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
6474	if (sb == NULL)
6475		return (ENOMEM);
6476
6477	if (lc->link_ok || lc->link_down_rc == 255)
6478		sbuf_printf(sb, "n/a");
6479	else
6480		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
6481
6482	rc = sbuf_finish(sb);
6483	sbuf_delete(sb);
6484
6485	return (rc);
6486}
6487
6488struct mem_desc {
6489	unsigned int base;
6490	unsigned int limit;
6491	unsigned int idx;
6492};
6493
6494static int
6495mem_desc_cmp(const void *a, const void *b)
6496{
6497	return ((const struct mem_desc *)a)->base -
6498	       ((const struct mem_desc *)b)->base;
6499}
6500
6501static void
6502mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
6503    unsigned int to)
6504{
6505	unsigned int size;
6506
6507	if (from == to)
6508		return;
6509
6510	size = to - from + 1;
6511	if (size == 0)
6512		return;
6513
6514	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
6515	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
6516}
6517
6518static int
6519sysctl_meminfo(SYSCTL_HANDLER_ARGS)
6520{
6521	struct adapter *sc = arg1;
6522	struct sbuf *sb;
6523	int rc, i, n;
6524	uint32_t lo, hi, used, alloc;
6525	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
6526	static const char *region[] = {
6527		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
6528		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
6529		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
6530		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
6531		"RQUDP region:", "PBL region:", "TXPBL region:",
6532		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
6533		"On-chip queues:"
6534	};
6535	struct mem_desc avail[4];
6536	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
6537	struct mem_desc *md = mem;
6538
6539	rc = sysctl_wire_old_buffer(req, 0);
6540	if (rc != 0)
6541		return (rc);
6542
6543	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6544	if (sb == NULL)
6545		return (ENOMEM);
6546
6547	for (i = 0; i < nitems(mem); i++) {
6548		mem[i].limit = 0;
6549		mem[i].idx = i;
6550	}
6551
6552	/* Find and sort the populated memory ranges */
6553	i = 0;
6554	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
6555	if (lo & F_EDRAM0_ENABLE) {
6556		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
6557		avail[i].base = G_EDRAM0_BASE(hi) << 20;
6558		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
6559		avail[i].idx = 0;
6560		i++;
6561	}
6562	if (lo & F_EDRAM1_ENABLE) {
6563		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
6564		avail[i].base = G_EDRAM1_BASE(hi) << 20;
6565		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
6566		avail[i].idx = 1;
6567		i++;
6568	}
6569	if (lo & F_EXT_MEM_ENABLE) {
6570		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
6571		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
6572		avail[i].limit = avail[i].base +
6573		    (G_EXT_MEM_SIZE(hi) << 20);
6574		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
6575		i++;
6576	}
6577	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
6578		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
6579		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
6580		avail[i].limit = avail[i].base +
6581		    (G_EXT_MEM1_SIZE(hi) << 20);
6582		avail[i].idx = 4;
6583		i++;
6584	}
6585	if (!i)                                    /* no memory available */
6586		return 0;
6587	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
6588
6589	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
6590	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
6591	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
6592	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
6593	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
6594	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
6595	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
6596	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
6597	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
6598
6599	/* the next few have explicit upper bounds */
6600	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
6601	md->limit = md->base - 1 +
6602		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
6603		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
6604	md++;
6605
6606	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
6607	md->limit = md->base - 1 +
6608		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
6609		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
6610	md++;
6611
6612	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
6613		if (chip_id(sc) <= CHELSIO_T5)
6614			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
6615		else
6616			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
6617		md->limit = 0;
6618	} else {
6619		md->base = 0;
6620		md->idx = nitems(region);  /* hide it */
6621	}
6622	md++;
6623
6624#define ulp_region(reg) \
6625	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
6626	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
6627
6628	ulp_region(RX_ISCSI);
6629	ulp_region(RX_TDDP);
6630	ulp_region(TX_TPT);
6631	ulp_region(RX_STAG);
6632	ulp_region(RX_RQ);
6633	ulp_region(RX_RQUDP);
6634	ulp_region(RX_PBL);
6635	ulp_region(TX_PBL);
6636#undef ulp_region
6637
6638	md->base = 0;
6639	md->idx = nitems(region);
6640	if (!is_t4(sc)) {
6641		uint32_t size = 0;
6642		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
6643		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
6644
6645		if (is_t5(sc)) {
6646			if (sge_ctrl & F_VFIFO_ENABLE)
6647				size = G_DBVFIFO_SIZE(fifo_size);
6648		} else
6649			size = G_T6_DBVFIFO_SIZE(fifo_size);
6650
6651		if (size) {
6652			md->base = G_BASEADDR(t4_read_reg(sc,
6653			    A_SGE_DBVFIFO_BADDR));
6654			md->limit = md->base + (size << 2) - 1;
6655		}
6656	}
6657	md++;
6658
6659	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
6660	md->limit = 0;
6661	md++;
6662	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
6663	md->limit = 0;
6664	md++;
6665
6666	md->base = sc->vres.ocq.start;
6667	if (sc->vres.ocq.size)
6668		md->limit = md->base + sc->vres.ocq.size - 1;
6669	else
6670		md->idx = nitems(region);  /* hide it */
6671	md++;
6672
6673	/* add any address-space holes, there can be up to 3 */
6674	for (n = 0; n < i - 1; n++)
6675		if (avail[n].limit < avail[n + 1].base)
6676			(md++)->base = avail[n].limit;
6677	if (avail[n].limit)
6678		(md++)->base = avail[n].limit;
6679
6680	n = md - mem;
6681	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
6682
6683	for (lo = 0; lo < i; lo++)
6684		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
6685				avail[lo].limit - 1);
6686
6687	sbuf_printf(sb, "\n");
6688	for (i = 0; i < n; i++) {
6689		if (mem[i].idx >= nitems(region))
6690			continue;                        /* skip holes */
6691		if (!mem[i].limit)
6692			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
6693		mem_region_show(sb, region[mem[i].idx], mem[i].base,
6694				mem[i].limit);
6695	}
6696
6697	sbuf_printf(sb, "\n");
6698	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
6699	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
6700	mem_region_show(sb, "uP RAM:", lo, hi);
6701
6702	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
6703	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
6704	mem_region_show(sb, "uP Extmem2:", lo, hi);
6705
6706	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
6707	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
6708		   G_PMRXMAXPAGE(lo),
6709		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
6710		   (lo & F_PMRXNUMCHN) ? 2 : 1);
6711
6712	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
6713	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
6714	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
6715		   G_PMTXMAXPAGE(lo),
6716		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
6717		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
6718	sbuf_printf(sb, "%u p-structs\n",
6719		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
6720
6721	for (i = 0; i < 4; i++) {
6722		if (chip_id(sc) > CHELSIO_T5)
6723			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
6724		else
6725			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
6726		if (is_t5(sc)) {
6727			used = G_T5_USED(lo);
6728			alloc = G_T5_ALLOC(lo);
6729		} else {
6730			used = G_USED(lo);
6731			alloc = G_ALLOC(lo);
6732		}
6733		/* For T6 these are MAC buffer groups */
6734		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
6735		    i, used, alloc);
6736	}
6737	for (i = 0; i < sc->chip_params->nchan; i++) {
6738		if (chip_id(sc) > CHELSIO_T5)
6739			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
6740		else
6741			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
6742		if (is_t5(sc)) {
6743			used = G_T5_USED(lo);
6744			alloc = G_T5_ALLOC(lo);
6745		} else {
6746			used = G_USED(lo);
6747			alloc = G_ALLOC(lo);
6748		}
6749		/* For T6 these are MAC buffer groups */
6750		sbuf_printf(sb,
6751		    "\nLoopback %d using %u pages out of %u allocated",
6752		    i, used, alloc);
6753	}
6754
6755	rc = sbuf_finish(sb);
6756	sbuf_delete(sb);
6757
6758	return (rc);
6759}
6760
6761static inline void
6762tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
6763{
6764	*mask = x | y;
6765	y = htobe64(y);
6766	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
6767}
6768
6769static int
6770sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
6771{
6772	struct adapter *sc = arg1;
6773	struct sbuf *sb;
6774	int rc, i;
6775
6776	MPASS(chip_id(sc) <= CHELSIO_T5);
6777
6778	rc = sysctl_wire_old_buffer(req, 0);
6779	if (rc != 0)
6780		return (rc);
6781
6782	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6783	if (sb == NULL)
6784		return (ENOMEM);
6785
6786	sbuf_printf(sb,
6787	    "Idx  Ethernet address     Mask     Vld Ports PF"
6788	    "  VF              Replication             P0 P1 P2 P3  ML");
6789	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
6790		uint64_t tcamx, tcamy, mask;
6791		uint32_t cls_lo, cls_hi;
6792		uint8_t addr[ETHER_ADDR_LEN];
6793
6794		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
6795		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
6796		if (tcamx & tcamy)
6797			continue;
6798		tcamxy2valmask(tcamx, tcamy, addr, &mask);
6799		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
6800		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
6801		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
6802			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
6803			   addr[3], addr[4], addr[5], (uintmax_t)mask,
6804			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
6805			   G_PORTMAP(cls_hi), G_PF(cls_lo),
6806			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
6807
6808		if (cls_lo & F_REPLICATE) {
6809			struct fw_ldst_cmd ldst_cmd;
6810
6811			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
6812			ldst_cmd.op_to_addrspace =
6813			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
6814				F_FW_CMD_REQUEST | F_FW_CMD_READ |
6815				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
6816			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
6817			ldst_cmd.u.mps.rplc.fid_idx =
6818			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
6819				V_FW_LDST_CMD_IDX(i));
6820
6821			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
6822			    "t4mps");
6823			if (rc)
6824				break;
6825			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
6826			    sizeof(ldst_cmd), &ldst_cmd);
6827			end_synchronized_op(sc, 0);
6828
6829			if (rc != 0) {
6830				sbuf_printf(sb, "%36d", rc);
6831				rc = 0;
6832			} else {
6833				sbuf_printf(sb, " %08x %08x %08x %08x",
6834				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
6835				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
6836				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
6837				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
6838			}
6839		} else
6840			sbuf_printf(sb, "%36s", "");
6841
6842		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
6843		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
6844		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
6845	}
6846
6847	if (rc)
6848		(void) sbuf_finish(sb);
6849	else
6850		rc = sbuf_finish(sb);
6851	sbuf_delete(sb);
6852
6853	return (rc);
6854}
6855
6856static int
6857sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
6858{
6859	struct adapter *sc = arg1;
6860	struct sbuf *sb;
6861	int rc, i;
6862
6863	MPASS(chip_id(sc) > CHELSIO_T5);
6864
6865	rc = sysctl_wire_old_buffer(req, 0);
6866	if (rc != 0)
6867		return (rc);
6868
6869	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6870	if (sb == NULL)
6871		return (ENOMEM);
6872
6873	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
6874	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
6875	    "                           Replication"
6876	    "                                    P0 P1 P2 P3  ML\n");
6877
6878	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
6879		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
6880		uint16_t ivlan;
6881		uint64_t tcamx, tcamy, val, mask;
6882		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
6883		uint8_t addr[ETHER_ADDR_LEN];
6884
6885		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
6886		if (i < 256)
6887			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
6888		else
6889			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
6890		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
6891		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
6892		tcamy = G_DMACH(val) << 32;
6893		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
6894		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
6895		lookup_type = G_DATALKPTYPE(data2);
6896		port_num = G_DATAPORTNUM(data2);
6897		if (lookup_type && lookup_type != M_DATALKPTYPE) {
6898			/* Inner header VNI */
6899			vniy = ((data2 & F_DATAVIDH2) << 23) |
6900				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
6901			dip_hit = data2 & F_DATADIPHIT;
6902			vlan_vld = 0;
6903		} else {
6904			vniy = 0;
6905			dip_hit = 0;
6906			vlan_vld = data2 & F_DATAVIDH2;
6907			ivlan = G_VIDL(val);
6908		}
6909
6910		ctl |= V_CTLXYBITSEL(1);
6911		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
6912		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
6913		tcamx = G_DMACH(val) << 32;
6914		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
6915		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
6916		if (lookup_type && lookup_type != M_DATALKPTYPE) {
6917			/* Inner header VNI mask */
6918			vnix = ((data2 & F_DATAVIDH2) << 23) |
6919			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
6920		} else
6921			vnix = 0;
6922
6923		if (tcamx & tcamy)
6924			continue;
6925		tcamxy2valmask(tcamx, tcamy, addr, &mask);
6926
6927		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
6928		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
6929
6930		if (lookup_type && lookup_type != M_DATALKPTYPE) {
6931			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
6932			    "%012jx %06x %06x    -    -   %3c"
6933			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
6934			    addr[1], addr[2], addr[3], addr[4], addr[5],
6935			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
6936			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
6937			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
6938			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
6939		} else {
6940			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
6941			    "%012jx    -       -   ", i, addr[0], addr[1],
6942			    addr[2], addr[3], addr[4], addr[5],
6943			    (uintmax_t)mask);
6944
6945			if (vlan_vld)
6946				sbuf_printf(sb, "%4u   Y     ", ivlan);
6947			else
6948				sbuf_printf(sb, "  -    N     ");
6949
6950			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
6951			    lookup_type ? 'I' : 'O', port_num,
6952			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
6953			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
6954			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
6955		}
6956
6957
6958		if (cls_lo & F_T6_REPLICATE) {
6959			struct fw_ldst_cmd ldst_cmd;
6960
6961			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
6962			ldst_cmd.op_to_addrspace =
6963			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
6964				F_FW_CMD_REQUEST | F_FW_CMD_READ |
6965				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
6966			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
6967			ldst_cmd.u.mps.rplc.fid_idx =
6968			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
6969				V_FW_LDST_CMD_IDX(i));
6970
6971			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
6972			    "t6mps");
6973			if (rc)
6974				break;
6975			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
6976			    sizeof(ldst_cmd), &ldst_cmd);
6977			end_synchronized_op(sc, 0);
6978
6979			if (rc != 0) {
6980				sbuf_printf(sb, "%72d", rc);
6981				rc = 0;
6982			} else {
6983				sbuf_printf(sb, " %08x %08x %08x %08x"
6984				    " %08x %08x %08x %08x",
6985				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
6986				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
6987				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
6988				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
6989				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
6990				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
6991				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
6992				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
6993			}
6994		} else
6995			sbuf_printf(sb, "%72s", "");
6996
6997		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
6998		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
6999		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
7000		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
7001	}
7002
7003	if (rc)
7004		(void) sbuf_finish(sb);
7005	else
7006		rc = sbuf_finish(sb);
7007	sbuf_delete(sb);
7008
7009	return (rc);
7010}
7011
7012static int
7013sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
7014{
7015	struct adapter *sc = arg1;
7016	struct sbuf *sb;
7017	int rc;
7018	uint16_t mtus[NMTUS];
7019
7020	rc = sysctl_wire_old_buffer(req, 0);
7021	if (rc != 0)
7022		return (rc);
7023
7024	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7025	if (sb == NULL)
7026		return (ENOMEM);
7027
7028	t4_read_mtu_tbl(sc, mtus, NULL);
7029
7030	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
7031	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
7032	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
7033	    mtus[14], mtus[15]);
7034
7035	rc = sbuf_finish(sb);
7036	sbuf_delete(sb);
7037
7038	return (rc);
7039}
7040
7041static int
7042sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
7043{
7044	struct adapter *sc = arg1;
7045	struct sbuf *sb;
7046	int rc, i;
7047	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
7048	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
7049	static const char *tx_stats[MAX_PM_NSTATS] = {
7050		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
7051		"Tx FIFO wait", NULL, "Tx latency"
7052	};
7053	static const char *rx_stats[MAX_PM_NSTATS] = {
7054		"Read:", "Write bypass:", "Write mem:", "Flush:",
7055		"Rx FIFO wait", NULL, "Rx latency"
7056	};
7057
7058	rc = sysctl_wire_old_buffer(req, 0);
7059	if (rc != 0)
7060		return (rc);
7061
7062	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7063	if (sb == NULL)
7064		return (ENOMEM);
7065
7066	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
7067	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
7068
7069	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
7070	for (i = 0; i < 4; i++) {
7071		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7072		    tx_cyc[i]);
7073	}
7074
7075	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
7076	for (i = 0; i < 4; i++) {
7077		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7078		    rx_cyc[i]);
7079	}
7080
7081	if (chip_id(sc) > CHELSIO_T5) {
7082		sbuf_printf(sb,
7083		    "\n              Total wait      Total occupancy");
7084		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7085		    tx_cyc[i]);
7086		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7087		    rx_cyc[i]);
7088
7089		i += 2;
7090		MPASS(i < nitems(tx_stats));
7091
7092		sbuf_printf(sb,
7093		    "\n                   Reads           Total wait");
7094		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7095		    tx_cyc[i]);
7096		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7097		    rx_cyc[i]);
7098	}
7099
7100	rc = sbuf_finish(sb);
7101	sbuf_delete(sb);
7102
7103	return (rc);
7104}
7105
7106static int
7107sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
7108{
7109	struct adapter *sc = arg1;
7110	struct sbuf *sb;
7111	int rc;
7112	struct tp_rdma_stats stats;
7113
7114	rc = sysctl_wire_old_buffer(req, 0);
7115	if (rc != 0)
7116		return (rc);
7117
7118	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7119	if (sb == NULL)
7120		return (ENOMEM);
7121
7122	mtx_lock(&sc->reg_lock);
7123	t4_tp_get_rdma_stats(sc, &stats);
7124	mtx_unlock(&sc->reg_lock);
7125
7126	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
7127	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
7128
7129	rc = sbuf_finish(sb);
7130	sbuf_delete(sb);
7131
7132	return (rc);
7133}
7134
7135static int
7136sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
7137{
7138	struct adapter *sc = arg1;
7139	struct sbuf *sb;
7140	int rc;
7141	struct tp_tcp_stats v4, v6;
7142
7143	rc = sysctl_wire_old_buffer(req, 0);
7144	if (rc != 0)
7145		return (rc);
7146
7147	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7148	if (sb == NULL)
7149		return (ENOMEM);
7150
7151	mtx_lock(&sc->reg_lock);
7152	t4_tp_get_tcp_stats(sc, &v4, &v6);
7153	mtx_unlock(&sc->reg_lock);
7154
7155	sbuf_printf(sb,
7156	    "                                IP                 IPv6\n");
7157	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
7158	    v4.tcp_out_rsts, v6.tcp_out_rsts);
7159	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
7160	    v4.tcp_in_segs, v6.tcp_in_segs);
7161	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
7162	    v4.tcp_out_segs, v6.tcp_out_segs);
7163	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
7164	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
7165
7166	rc = sbuf_finish(sb);
7167	sbuf_delete(sb);
7168
7169	return (rc);
7170}
7171
7172static int
7173sysctl_tids(SYSCTL_HANDLER_ARGS)
7174{
7175	struct adapter *sc = arg1;
7176	struct sbuf *sb;
7177	int rc;
7178	struct tid_info *t = &sc->tids;
7179
7180	rc = sysctl_wire_old_buffer(req, 0);
7181	if (rc != 0)
7182		return (rc);
7183
7184	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7185	if (sb == NULL)
7186		return (ENOMEM);
7187
7188	if (t->natids) {
7189		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
7190		    t->atids_in_use);
7191	}
7192
7193	if (t->ntids) {
7194		sbuf_printf(sb, "TID range: ");
7195		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7196			uint32_t b, hb;
7197
7198			if (chip_id(sc) <= CHELSIO_T5) {
7199				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
7200				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
7201			} else {
7202				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
7203				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
7204			}
7205
7206			if (b)
7207				sbuf_printf(sb, "0-%u, ", b - 1);
7208			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
7209		} else
7210			sbuf_printf(sb, "0-%u", t->ntids - 1);
7211		sbuf_printf(sb, ", in use: %u\n",
7212		    atomic_load_acq_int(&t->tids_in_use));
7213	}
7214
7215	if (t->nstids) {
7216		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
7217		    t->stid_base + t->nstids - 1, t->stids_in_use);
7218	}
7219
7220	if (t->nftids) {
7221		sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
7222		    t->ftid_base + t->nftids - 1);
7223	}
7224
7225	if (t->netids) {
7226		sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
7227		    t->etid_base + t->netids - 1);
7228	}
7229
7230	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
7231	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
7232	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
7233
7234	rc = sbuf_finish(sb);
7235	sbuf_delete(sb);
7236
7237	return (rc);
7238}
7239
7240static int
7241sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
7242{
7243	struct adapter *sc = arg1;
7244	struct sbuf *sb;
7245	int rc;
7246	struct tp_err_stats stats;
7247
7248	rc = sysctl_wire_old_buffer(req, 0);
7249	if (rc != 0)
7250		return (rc);
7251
7252	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7253	if (sb == NULL)
7254		return (ENOMEM);
7255
7256	mtx_lock(&sc->reg_lock);
7257	t4_tp_get_err_stats(sc, &stats);
7258	mtx_unlock(&sc->reg_lock);
7259
7260	if (sc->chip_params->nchan > 2) {
7261		sbuf_printf(sb, "                 channel 0  channel 1"
7262		    "  channel 2  channel 3\n");
7263		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
7264		    stats.mac_in_errs[0], stats.mac_in_errs[1],
7265		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
7266		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
7267		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
7268		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
7269		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
7270		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
7271		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
7272		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
7273		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
7274		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
7275		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
7276		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
7277		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
7278		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
7279		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
7280		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
7281		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
7282		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
7283		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
7284		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
7285		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
7286		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
7287	} else {
7288		sbuf_printf(sb, "                 channel 0  channel 1\n");
7289		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
7290		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
7291		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
7292		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
7293		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
7294		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
7295		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
7296		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
7297		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
7298		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
7299		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
7300		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
7301		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
7302		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
7303		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
7304		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
7305	}
7306
7307	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
7308	    stats.ofld_no_neigh, stats.ofld_cong_defer);
7309
7310	rc = sbuf_finish(sb);
7311	sbuf_delete(sb);
7312
7313	return (rc);
7314}
7315
7316static int
7317sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
7318{
7319	struct adapter *sc = arg1;
7320	struct tp_params *tpp = &sc->params.tp;
7321	u_int mask;
7322	int rc;
7323
7324	mask = tpp->la_mask >> 16;
7325	rc = sysctl_handle_int(oidp, &mask, 0, req);
7326	if (rc != 0 || req->newptr == NULL)
7327		return (rc);
7328	if (mask > 0xffff)
7329		return (EINVAL);
7330	tpp->la_mask = mask << 16;
7331	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
7332
7333	return (0);
7334}
7335
7336struct field_desc {
7337	const char *name;
7338	u_int start;
7339	u_int width;
7340};
7341
7342static void
7343field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
7344{
7345	char buf[32];
7346	int line_size = 0;
7347
7348	while (f->name) {
7349		uint64_t mask = (1ULL << f->width) - 1;
7350		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
7351		    ((uintmax_t)v >> f->start) & mask);
7352
7353		if (line_size + len >= 79) {
7354			line_size = 8;
7355			sbuf_printf(sb, "\n        ");
7356		}
7357		sbuf_printf(sb, "%s ", buf);
7358		line_size += len + 1;
7359		f++;
7360	}
7361	sbuf_printf(sb, "\n");
7362}
7363
7364static const struct field_desc tp_la0[] = {
7365	{ "RcfOpCodeOut", 60, 4 },
7366	{ "State", 56, 4 },
7367	{ "WcfState", 52, 4 },
7368	{ "RcfOpcSrcOut", 50, 2 },
7369	{ "CRxError", 49, 1 },
7370	{ "ERxError", 48, 1 },
7371	{ "SanityFailed", 47, 1 },
7372	{ "SpuriousMsg", 46, 1 },
7373	{ "FlushInputMsg", 45, 1 },
7374	{ "FlushInputCpl", 44, 1 },
7375	{ "RssUpBit", 43, 1 },
7376	{ "RssFilterHit", 42, 1 },
7377	{ "Tid", 32, 10 },
7378	{ "InitTcb", 31, 1 },
7379	{ "LineNumber", 24, 7 },
7380	{ "Emsg", 23, 1 },
7381	{ "EdataOut", 22, 1 },
7382	{ "Cmsg", 21, 1 },
7383	{ "CdataOut", 20, 1 },
7384	{ "EreadPdu", 19, 1 },
7385	{ "CreadPdu", 18, 1 },
7386	{ "TunnelPkt", 17, 1 },
7387	{ "RcfPeerFin", 16, 1 },
7388	{ "RcfReasonOut", 12, 4 },
7389	{ "TxCchannel", 10, 2 },
7390	{ "RcfTxChannel", 8, 2 },
7391	{ "RxEchannel", 6, 2 },
7392	{ "RcfRxChannel", 5, 1 },
7393	{ "RcfDataOutSrdy", 4, 1 },
7394	{ "RxDvld", 3, 1 },
7395	{ "RxOoDvld", 2, 1 },
7396	{ "RxCongestion", 1, 1 },
7397	{ "TxCongestion", 0, 1 },
7398	{ NULL }
7399};
7400
7401static const struct field_desc tp_la1[] = {
7402	{ "CplCmdIn", 56, 8 },
7403	{ "CplCmdOut", 48, 8 },
7404	{ "ESynOut", 47, 1 },
7405	{ "EAckOut", 46, 1 },
7406	{ "EFinOut", 45, 1 },
7407	{ "ERstOut", 44, 1 },
7408	{ "SynIn", 43, 1 },
7409	{ "AckIn", 42, 1 },
7410	{ "FinIn", 41, 1 },
7411	{ "RstIn", 40, 1 },
7412	{ "DataIn", 39, 1 },
7413	{ "DataInVld", 38, 1 },
7414	{ "PadIn", 37, 1 },
7415	{ "RxBufEmpty", 36, 1 },
7416	{ "RxDdp", 35, 1 },
7417	{ "RxFbCongestion", 34, 1 },
7418	{ "TxFbCongestion", 33, 1 },
7419	{ "TxPktSumSrdy", 32, 1 },
7420	{ "RcfUlpType", 28, 4 },
7421	{ "Eread", 27, 1 },
7422	{ "Ebypass", 26, 1 },
7423	{ "Esave", 25, 1 },
7424	{ "Static0", 24, 1 },
7425	{ "Cread", 23, 1 },
7426	{ "Cbypass", 22, 1 },
7427	{ "Csave", 21, 1 },
7428	{ "CPktOut", 20, 1 },
7429	{ "RxPagePoolFull", 18, 2 },
7430	{ "RxLpbkPkt", 17, 1 },
7431	{ "TxLpbkPkt", 16, 1 },
7432	{ "RxVfValid", 15, 1 },
7433	{ "SynLearned", 14, 1 },
7434	{ "SetDelEntry", 13, 1 },
7435	{ "SetInvEntry", 12, 1 },
7436	{ "CpcmdDvld", 11, 1 },
7437	{ "CpcmdSave", 10, 1 },
7438	{ "RxPstructsFull", 8, 2 },
7439	{ "EpcmdDvld", 7, 1 },
7440	{ "EpcmdFlush", 6, 1 },
7441	{ "EpcmdTrimPrefix", 5, 1 },
7442	{ "EpcmdTrimPostfix", 4, 1 },
7443	{ "ERssIp4Pkt", 3, 1 },
7444	{ "ERssIp6Pkt", 2, 1 },
7445	{ "ERssTcpUdpPkt", 1, 1 },
7446	{ "ERssFceFipPkt", 0, 1 },
7447	{ NULL }
7448};
7449
7450static const struct field_desc tp_la2[] = {
7451	{ "CplCmdIn", 56, 8 },
7452	{ "MpsVfVld", 55, 1 },
7453	{ "MpsPf", 52, 3 },
7454	{ "MpsVf", 44, 8 },
7455	{ "SynIn", 43, 1 },
7456	{ "AckIn", 42, 1 },
7457	{ "FinIn", 41, 1 },
7458	{ "RstIn", 40, 1 },
7459	{ "DataIn", 39, 1 },
7460	{ "DataInVld", 38, 1 },
7461	{ "PadIn", 37, 1 },
7462	{ "RxBufEmpty", 36, 1 },
7463	{ "RxDdp", 35, 1 },
7464	{ "RxFbCongestion", 34, 1 },
7465	{ "TxFbCongestion", 33, 1 },
7466	{ "TxPktSumSrdy", 32, 1 },
7467	{ "RcfUlpType", 28, 4 },
7468	{ "Eread", 27, 1 },
7469	{ "Ebypass", 26, 1 },
7470	{ "Esave", 25, 1 },
7471	{ "Static0", 24, 1 },
7472	{ "Cread", 23, 1 },
7473	{ "Cbypass", 22, 1 },
7474	{ "Csave", 21, 1 },
7475	{ "CPktOut", 20, 1 },
7476	{ "RxPagePoolFull", 18, 2 },
7477	{ "RxLpbkPkt", 17, 1 },
7478	{ "TxLpbkPkt", 16, 1 },
7479	{ "RxVfValid", 15, 1 },
7480	{ "SynLearned", 14, 1 },
7481	{ "SetDelEntry", 13, 1 },
7482	{ "SetInvEntry", 12, 1 },
7483	{ "CpcmdDvld", 11, 1 },
7484	{ "CpcmdSave", 10, 1 },
7485	{ "RxPstructsFull", 8, 2 },
7486	{ "EpcmdDvld", 7, 1 },
7487	{ "EpcmdFlush", 6, 1 },
7488	{ "EpcmdTrimPrefix", 5, 1 },
7489	{ "EpcmdTrimPostfix", 4, 1 },
7490	{ "ERssIp4Pkt", 3, 1 },
7491	{ "ERssIp6Pkt", 2, 1 },
7492	{ "ERssTcpUdpPkt", 1, 1 },
7493	{ "ERssFceFipPkt", 0, 1 },
7494	{ NULL }
7495};
7496
7497static void
7498tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
7499{
7500
7501	field_desc_show(sb, *p, tp_la0);
7502}
7503
7504static void
7505tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
7506{
7507
7508	if (idx)
7509		sbuf_printf(sb, "\n");
7510	field_desc_show(sb, p[0], tp_la0);
7511	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7512		field_desc_show(sb, p[1], tp_la0);
7513}
7514
7515static void
7516tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
7517{
7518
7519	if (idx)
7520		sbuf_printf(sb, "\n");
7521	field_desc_show(sb, p[0], tp_la0);
7522	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7523		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
7524}
7525
7526static int
7527sysctl_tp_la(SYSCTL_HANDLER_ARGS)
7528{
7529	struct adapter *sc = arg1;
7530	struct sbuf *sb;
7531	uint64_t *buf, *p;
7532	int rc;
7533	u_int i, inc;
7534	void (*show_func)(struct sbuf *, uint64_t *, int);
7535
7536	rc = sysctl_wire_old_buffer(req, 0);
7537	if (rc != 0)
7538		return (rc);
7539
7540	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7541	if (sb == NULL)
7542		return (ENOMEM);
7543
7544	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
7545
7546	t4_tp_read_la(sc, buf, NULL);
7547	p = buf;
7548
7549	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
7550	case 2:
7551		inc = 2;
7552		show_func = tp_la_show2;
7553		break;
7554	case 3:
7555		inc = 2;
7556		show_func = tp_la_show3;
7557		break;
7558	default:
7559		inc = 1;
7560		show_func = tp_la_show;
7561	}
7562
7563	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
7564		(*show_func)(sb, p, i);
7565
7566	rc = sbuf_finish(sb);
7567	sbuf_delete(sb);
7568	free(buf, M_CXGBE);
7569	return (rc);
7570}
7571
7572static int
7573sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
7574{
7575	struct adapter *sc = arg1;
7576	struct sbuf *sb;
7577	int rc;
7578	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
7579
7580	rc = sysctl_wire_old_buffer(req, 0);
7581	if (rc != 0)
7582		return (rc);
7583
7584	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7585	if (sb == NULL)
7586		return (ENOMEM);
7587
7588	t4_get_chan_txrate(sc, nrate, orate);
7589
7590	if (sc->chip_params->nchan > 2) {
7591		sbuf_printf(sb, "              channel 0   channel 1"
7592		    "   channel 2   channel 3\n");
7593		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
7594		    nrate[0], nrate[1], nrate[2], nrate[3]);
7595		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
7596		    orate[0], orate[1], orate[2], orate[3]);
7597	} else {
7598		sbuf_printf(sb, "              channel 0   channel 1\n");
7599		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
7600		    nrate[0], nrate[1]);
7601		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
7602		    orate[0], orate[1]);
7603	}
7604
7605	rc = sbuf_finish(sb);
7606	sbuf_delete(sb);
7607
7608	return (rc);
7609}
7610
7611static int
7612sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
7613{
7614	struct adapter *sc = arg1;
7615	struct sbuf *sb;
7616	uint32_t *buf, *p;
7617	int rc, i;
7618
7619	rc = sysctl_wire_old_buffer(req, 0);
7620	if (rc != 0)
7621		return (rc);
7622
7623	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7624	if (sb == NULL)
7625		return (ENOMEM);
7626
7627	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
7628	    M_ZERO | M_WAITOK);
7629
7630	t4_ulprx_read_la(sc, buf);
7631	p = buf;
7632
7633	sbuf_printf(sb, "      Pcmd        Type   Message"
7634	    "                Data");
7635	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
7636		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
7637		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
7638	}
7639
7640	rc = sbuf_finish(sb);
7641	sbuf_delete(sb);
7642	free(buf, M_CXGBE);
7643	return (rc);
7644}
7645
7646static int
7647sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
7648{
7649	struct adapter *sc = arg1;
7650	struct sbuf *sb;
7651	int rc, v;
7652
7653	MPASS(chip_id(sc) >= CHELSIO_T5);
7654
7655	rc = sysctl_wire_old_buffer(req, 0);
7656	if (rc != 0)
7657		return (rc);
7658
7659	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7660	if (sb == NULL)
7661		return (ENOMEM);
7662
7663	v = t4_read_reg(sc, A_SGE_STAT_CFG);
7664	if (G_STATSOURCE_T5(v) == 7) {
7665		int mode;
7666
7667		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
7668		if (mode == 0) {
7669			sbuf_printf(sb, "total %d, incomplete %d",
7670			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7671			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7672		} else if (mode == 1) {
7673			sbuf_printf(sb, "total %d, data overflow %d",
7674			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7675			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7676		} else {
7677			sbuf_printf(sb, "unknown mode %d", mode);
7678		}
7679	}
7680	rc = sbuf_finish(sb);
7681	sbuf_delete(sb);
7682
7683	return (rc);
7684}
7685
7686static int
7687sysctl_tc_params(SYSCTL_HANDLER_ARGS)
7688{
7689	struct adapter *sc = arg1;
7690	struct tx_sched_class *tc;
7691	struct t4_sched_class_params p;
7692	struct sbuf *sb;
7693	int i, rc, port_id, flags, mbps, gbps;
7694
7695	rc = sysctl_wire_old_buffer(req, 0);
7696	if (rc != 0)
7697		return (rc);
7698
7699	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7700	if (sb == NULL)
7701		return (ENOMEM);
7702
7703	port_id = arg2 >> 16;
7704	MPASS(port_id < sc->params.nports);
7705	MPASS(sc->port[port_id] != NULL);
7706	i = arg2 & 0xffff;
7707	MPASS(i < sc->chip_params->nsched_cls);
7708	tc = &sc->port[port_id]->tc[i];
7709
7710	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7711	    "t4tc_p");
7712	if (rc)
7713		goto done;
7714	flags = tc->flags;
7715	p = tc->params;
7716	end_synchronized_op(sc, LOCK_HELD);
7717
7718	if ((flags & TX_SC_OK) == 0) {
7719		sbuf_printf(sb, "none");
7720		goto done;
7721	}
7722
7723	if (p.level == SCHED_CLASS_LEVEL_CL_WRR) {
7724		sbuf_printf(sb, "cl-wrr weight %u", p.weight);
7725		goto done;
7726	} else if (p.level == SCHED_CLASS_LEVEL_CL_RL)
7727		sbuf_printf(sb, "cl-rl");
7728	else if (p.level == SCHED_CLASS_LEVEL_CH_RL)
7729		sbuf_printf(sb, "ch-rl");
7730	else {
7731		rc = ENXIO;
7732		goto done;
7733	}
7734
7735	if (p.ratemode == SCHED_CLASS_RATEMODE_REL) {
7736		/* XXX: top speed or actual link speed? */
7737		gbps = port_top_speed(sc->port[port_id]);
7738		sbuf_printf(sb, " %u%% of %uGbps", p.maxrate, gbps);
7739	}
7740	else if (p.ratemode == SCHED_CLASS_RATEMODE_ABS) {
7741		switch (p.rateunit) {
7742		case SCHED_CLASS_RATEUNIT_BITS:
7743			mbps = p.maxrate / 1000;
7744			gbps = p.maxrate / 1000000;
7745			if (p.maxrate == gbps * 1000000)
7746				sbuf_printf(sb, " %uGbps", gbps);
7747			else if (p.maxrate == mbps * 1000)
7748				sbuf_printf(sb, " %uMbps", mbps);
7749			else
7750				sbuf_printf(sb, " %uKbps", p.maxrate);
7751			break;
7752		case SCHED_CLASS_RATEUNIT_PKTS:
7753			sbuf_printf(sb, " %upps", p.maxrate);
7754			break;
7755		default:
7756			rc = ENXIO;
7757			goto done;
7758		}
7759	}
7760
7761	switch (p.mode) {
7762	case SCHED_CLASS_MODE_CLASS:
7763		sbuf_printf(sb, " aggregate");
7764		break;
7765	case SCHED_CLASS_MODE_FLOW:
7766		sbuf_printf(sb, " per-flow");
7767		break;
7768	default:
7769		rc = ENXIO;
7770		goto done;
7771	}
7772
7773done:
7774	if (rc == 0)
7775		rc = sbuf_finish(sb);
7776	sbuf_delete(sb);
7777
7778	return (rc);
7779}
7780#endif
7781
7782#ifdef TCP_OFFLOAD
7783static void
7784unit_conv(char *buf, size_t len, u_int val, u_int factor)
7785{
7786	u_int rem = val % factor;
7787
7788	if (rem == 0)
7789		snprintf(buf, len, "%u", val / factor);
7790	else {
7791		while (rem % 10 == 0)
7792			rem /= 10;
7793		snprintf(buf, len, "%u.%u", val / factor, rem);
7794	}
7795}
7796
7797static int
7798sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
7799{
7800	struct adapter *sc = arg1;
7801	char buf[16];
7802	u_int res, re;
7803	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7804
7805	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
7806	switch (arg2) {
7807	case 0:
7808		/* timer_tick */
7809		re = G_TIMERRESOLUTION(res);
7810		break;
7811	case 1:
7812		/* TCP timestamp tick */
7813		re = G_TIMESTAMPRESOLUTION(res);
7814		break;
7815	case 2:
7816		/* DACK tick */
7817		re = G_DELAYEDACKRESOLUTION(res);
7818		break;
7819	default:
7820		return (EDOOFUS);
7821	}
7822
7823	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
7824
7825	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
7826}
7827
7828static int
7829sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
7830{
7831	struct adapter *sc = arg1;
7832	u_int res, dack_re, v;
7833	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7834
7835	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
7836	dack_re = G_DELAYEDACKRESOLUTION(res);
7837	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
7838
7839	return (sysctl_handle_int(oidp, &v, 0, req));
7840}
7841
7842static int
7843sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
7844{
7845	struct adapter *sc = arg1;
7846	int reg = arg2;
7847	u_int tre;
7848	u_long tp_tick_us, v;
7849	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7850
7851	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
7852	    reg == A_TP_PERS_MIN || reg == A_TP_PERS_MAX ||
7853	    reg == A_TP_KEEP_IDLE || A_TP_KEEP_INTVL || reg == A_TP_INIT_SRTT ||
7854	    reg == A_TP_FINWAIT2_TIMER);
7855
7856	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
7857	tp_tick_us = (cclk_ps << tre) / 1000000;
7858
7859	if (reg == A_TP_INIT_SRTT)
7860		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
7861	else
7862		v = tp_tick_us * t4_read_reg(sc, reg);
7863
7864	return (sysctl_handle_long(oidp, &v, 0, req));
7865}
7866#endif
7867
7868static uint32_t
7869fconf_iconf_to_mode(uint32_t fconf, uint32_t iconf)
7870{
7871	uint32_t mode;
7872
7873	mode = T4_FILTER_IPv4 | T4_FILTER_IPv6 | T4_FILTER_IP_SADDR |
7874	    T4_FILTER_IP_DADDR | T4_FILTER_IP_SPORT | T4_FILTER_IP_DPORT;
7875
7876	if (fconf & F_FRAGMENTATION)
7877		mode |= T4_FILTER_IP_FRAGMENT;
7878
7879	if (fconf & F_MPSHITTYPE)
7880		mode |= T4_FILTER_MPS_HIT_TYPE;
7881
7882	if (fconf & F_MACMATCH)
7883		mode |= T4_FILTER_MAC_IDX;
7884
7885	if (fconf & F_ETHERTYPE)
7886		mode |= T4_FILTER_ETH_TYPE;
7887
7888	if (fconf & F_PROTOCOL)
7889		mode |= T4_FILTER_IP_PROTO;
7890
7891	if (fconf & F_TOS)
7892		mode |= T4_FILTER_IP_TOS;
7893
7894	if (fconf & F_VLAN)
7895		mode |= T4_FILTER_VLAN;
7896
7897	if (fconf & F_VNIC_ID) {
7898		mode |= T4_FILTER_VNIC;
7899		if (iconf & F_VNIC)
7900			mode |= T4_FILTER_IC_VNIC;
7901	}
7902
7903	if (fconf & F_PORT)
7904		mode |= T4_FILTER_PORT;
7905
7906	if (fconf & F_FCOE)
7907		mode |= T4_FILTER_FCoE;
7908
7909	return (mode);
7910}
7911
7912static uint32_t
7913mode_to_fconf(uint32_t mode)
7914{
7915	uint32_t fconf = 0;
7916
7917	if (mode & T4_FILTER_IP_FRAGMENT)
7918		fconf |= F_FRAGMENTATION;
7919
7920	if (mode & T4_FILTER_MPS_HIT_TYPE)
7921		fconf |= F_MPSHITTYPE;
7922
7923	if (mode & T4_FILTER_MAC_IDX)
7924		fconf |= F_MACMATCH;
7925
7926	if (mode & T4_FILTER_ETH_TYPE)
7927		fconf |= F_ETHERTYPE;
7928
7929	if (mode & T4_FILTER_IP_PROTO)
7930		fconf |= F_PROTOCOL;
7931
7932	if (mode & T4_FILTER_IP_TOS)
7933		fconf |= F_TOS;
7934
7935	if (mode & T4_FILTER_VLAN)
7936		fconf |= F_VLAN;
7937
7938	if (mode & T4_FILTER_VNIC)
7939		fconf |= F_VNIC_ID;
7940
7941	if (mode & T4_FILTER_PORT)
7942		fconf |= F_PORT;
7943
7944	if (mode & T4_FILTER_FCoE)
7945		fconf |= F_FCOE;
7946
7947	return (fconf);
7948}
7949
7950static uint32_t
7951mode_to_iconf(uint32_t mode)
7952{
7953
7954	if (mode & T4_FILTER_IC_VNIC)
7955		return (F_VNIC);
7956	return (0);
7957}
7958
7959static int check_fspec_against_fconf_iconf(struct adapter *sc,
7960    struct t4_filter_specification *fs)
7961{
7962	struct tp_params *tpp = &sc->params.tp;
7963	uint32_t fconf = 0;
7964
7965	if (fs->val.frag || fs->mask.frag)
7966		fconf |= F_FRAGMENTATION;
7967
7968	if (fs->val.matchtype || fs->mask.matchtype)
7969		fconf |= F_MPSHITTYPE;
7970
7971	if (fs->val.macidx || fs->mask.macidx)
7972		fconf |= F_MACMATCH;
7973
7974	if (fs->val.ethtype || fs->mask.ethtype)
7975		fconf |= F_ETHERTYPE;
7976
7977	if (fs->val.proto || fs->mask.proto)
7978		fconf |= F_PROTOCOL;
7979
7980	if (fs->val.tos || fs->mask.tos)
7981		fconf |= F_TOS;
7982
7983	if (fs->val.vlan_vld || fs->mask.vlan_vld)
7984		fconf |= F_VLAN;
7985
7986	if (fs->val.ovlan_vld || fs->mask.ovlan_vld) {
7987		fconf |= F_VNIC_ID;
7988		if (tpp->ingress_config & F_VNIC)
7989			return (EINVAL);
7990	}
7991
7992	if (fs->val.pfvf_vld || fs->mask.pfvf_vld) {
7993		fconf |= F_VNIC_ID;
7994		if ((tpp->ingress_config & F_VNIC) == 0)
7995			return (EINVAL);
7996	}
7997
7998	if (fs->val.iport || fs->mask.iport)
7999		fconf |= F_PORT;
8000
8001	if (fs->val.fcoe || fs->mask.fcoe)
8002		fconf |= F_FCOE;
8003
8004	if ((tpp->vlan_pri_map | fconf) != tpp->vlan_pri_map)
8005		return (E2BIG);
8006
8007	return (0);
8008}
8009
8010static int
8011get_filter_mode(struct adapter *sc, uint32_t *mode)
8012{
8013	struct tp_params *tpp = &sc->params.tp;
8014
8015	/*
8016	 * We trust the cached values of the relevant TP registers.  This means
8017	 * things work reliably only if writes to those registers are always via
8018	 * t4_set_filter_mode.
8019	 */
8020	*mode = fconf_iconf_to_mode(tpp->vlan_pri_map, tpp->ingress_config);
8021
8022	return (0);
8023}
8024
8025static int
8026set_filter_mode(struct adapter *sc, uint32_t mode)
8027{
8028	struct tp_params *tpp = &sc->params.tp;
8029	uint32_t fconf, iconf;
8030	int rc;
8031
8032	iconf = mode_to_iconf(mode);
8033	if ((iconf ^ tpp->ingress_config) & F_VNIC) {
8034		/*
8035		 * For now we just complain if A_TP_INGRESS_CONFIG is not
8036		 * already set to the correct value for the requested filter
8037		 * mode.  It's not clear if it's safe to write to this register
8038		 * on the fly.  (And we trust the cached value of the register).
8039		 */
8040		return (EBUSY);
8041	}
8042
8043	fconf = mode_to_fconf(mode);
8044
8045	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8046	    "t4setfm");
8047	if (rc)
8048		return (rc);
8049
8050	if (sc->tids.ftids_in_use > 0) {
8051		rc = EBUSY;
8052		goto done;
8053	}
8054
8055#ifdef TCP_OFFLOAD
8056	if (uld_active(sc, ULD_TOM)) {
8057		rc = EBUSY;
8058		goto done;
8059	}
8060#endif
8061
8062	rc = -t4_set_filter_mode(sc, fconf);
8063done:
8064	end_synchronized_op(sc, LOCK_HELD);
8065	return (rc);
8066}
8067
8068static inline uint64_t
8069get_filter_hits(struct adapter *sc, uint32_t fid)
8070{
8071	uint32_t tcb_addr;
8072
8073	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE) +
8074	    (fid + sc->tids.ftid_base) * TCB_SIZE;
8075
8076	if (is_t4(sc)) {
8077		uint64_t hits;
8078
8079		read_via_memwin(sc, 0, tcb_addr + 16, (uint32_t *)&hits, 8);
8080		return (be64toh(hits));
8081	} else {
8082		uint32_t hits;
8083
8084		read_via_memwin(sc, 0, tcb_addr + 24, &hits, 4);
8085		return (be32toh(hits));
8086	}
8087}
8088
8089static int
8090get_filter(struct adapter *sc, struct t4_filter *t)
8091{
8092	int i, rc, nfilters = sc->tids.nftids;
8093	struct filter_entry *f;
8094
8095	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8096	    "t4getf");
8097	if (rc)
8098		return (rc);
8099
8100	if (sc->tids.ftids_in_use == 0 || sc->tids.ftid_tab == NULL ||
8101	    t->idx >= nfilters) {
8102		t->idx = 0xffffffff;
8103		goto done;
8104	}
8105
8106	f = &sc->tids.ftid_tab[t->idx];
8107	for (i = t->idx; i < nfilters; i++, f++) {
8108		if (f->valid) {
8109			t->idx = i;
8110			t->l2tidx = f->l2t ? f->l2t->idx : 0;
8111			t->smtidx = f->smtidx;
8112			if (f->fs.hitcnts)
8113				t->hits = get_filter_hits(sc, t->idx);
8114			else
8115				t->hits = UINT64_MAX;
8116			t->fs = f->fs;
8117
8118			goto done;
8119		}
8120	}
8121
8122	t->idx = 0xffffffff;
8123done:
8124	end_synchronized_op(sc, LOCK_HELD);
8125	return (0);
8126}
8127
8128static int
8129set_filter(struct adapter *sc, struct t4_filter *t)
8130{
8131	unsigned int nfilters, nports;
8132	struct filter_entry *f;
8133	int i, rc;
8134
8135	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setf");
8136	if (rc)
8137		return (rc);
8138
8139	nfilters = sc->tids.nftids;
8140	nports = sc->params.nports;
8141
8142	if (nfilters == 0) {
8143		rc = ENOTSUP;
8144		goto done;
8145	}
8146
8147	if (t->idx >= nfilters) {
8148		rc = EINVAL;
8149		goto done;
8150	}
8151
8152	/* Validate against the global filter mode and ingress config */
8153	rc = check_fspec_against_fconf_iconf(sc, &t->fs);
8154	if (rc != 0)
8155		goto done;
8156
8157	if (t->fs.action == FILTER_SWITCH && t->fs.eport >= nports) {
8158		rc = EINVAL;
8159		goto done;
8160	}
8161
8162	if (t->fs.val.iport >= nports) {
8163		rc = EINVAL;
8164		goto done;
8165	}
8166
8167	/* Can't specify an iq if not steering to it */
8168	if (!t->fs.dirsteer && t->fs.iq) {
8169		rc = EINVAL;
8170		goto done;
8171	}
8172
8173	/* IPv6 filter idx must be 4 aligned */
8174	if (t->fs.type == 1 &&
8175	    ((t->idx & 0x3) || t->idx + 4 >= nfilters)) {
8176		rc = EINVAL;
8177		goto done;
8178	}
8179
8180	if (!(sc->flags & FULL_INIT_DONE) &&
8181	    ((rc = adapter_full_init(sc)) != 0))
8182		goto done;
8183
8184	if (sc->tids.ftid_tab == NULL) {
8185		KASSERT(sc->tids.ftids_in_use == 0,
8186		    ("%s: no memory allocated but filters_in_use > 0",
8187		    __func__));
8188
8189		sc->tids.ftid_tab = malloc(sizeof (struct filter_entry) *
8190		    nfilters, M_CXGBE, M_NOWAIT | M_ZERO);
8191		if (sc->tids.ftid_tab == NULL) {
8192			rc = ENOMEM;
8193			goto done;
8194		}
8195		mtx_init(&sc->tids.ftid_lock, "T4 filters", 0, MTX_DEF);
8196	}
8197
8198	for (i = 0; i < 4; i++) {
8199		f = &sc->tids.ftid_tab[t->idx + i];
8200
8201		if (f->pending || f->valid) {
8202			rc = EBUSY;
8203			goto done;
8204		}
8205		if (f->locked) {
8206			rc = EPERM;
8207			goto done;
8208		}
8209
8210		if (t->fs.type == 0)
8211			break;
8212	}
8213
8214	f = &sc->tids.ftid_tab[t->idx];
8215	f->fs = t->fs;
8216
8217	rc = set_filter_wr(sc, t->idx);
8218done:
8219	end_synchronized_op(sc, 0);
8220
8221	if (rc == 0) {
8222		mtx_lock(&sc->tids.ftid_lock);
8223		for (;;) {
8224			if (f->pending == 0) {
8225				rc = f->valid ? 0 : EIO;
8226				break;
8227			}
8228
8229			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8230			    PCATCH, "t4setfw", 0)) {
8231				rc = EINPROGRESS;
8232				break;
8233			}
8234		}
8235		mtx_unlock(&sc->tids.ftid_lock);
8236	}
8237	return (rc);
8238}
8239
8240static int
8241del_filter(struct adapter *sc, struct t4_filter *t)
8242{
8243	unsigned int nfilters;
8244	struct filter_entry *f;
8245	int rc;
8246
8247	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4delf");
8248	if (rc)
8249		return (rc);
8250
8251	nfilters = sc->tids.nftids;
8252
8253	if (nfilters == 0) {
8254		rc = ENOTSUP;
8255		goto done;
8256	}
8257
8258	if (sc->tids.ftid_tab == NULL || sc->tids.ftids_in_use == 0 ||
8259	    t->idx >= nfilters) {
8260		rc = EINVAL;
8261		goto done;
8262	}
8263
8264	if (!(sc->flags & FULL_INIT_DONE)) {
8265		rc = EAGAIN;
8266		goto done;
8267	}
8268
8269	f = &sc->tids.ftid_tab[t->idx];
8270
8271	if (f->pending) {
8272		rc = EBUSY;
8273		goto done;
8274	}
8275	if (f->locked) {
8276		rc = EPERM;
8277		goto done;
8278	}
8279
8280	if (f->valid) {
8281		t->fs = f->fs;	/* extra info for the caller */
8282		rc = del_filter_wr(sc, t->idx);
8283	}
8284
8285done:
8286	end_synchronized_op(sc, 0);
8287
8288	if (rc == 0) {
8289		mtx_lock(&sc->tids.ftid_lock);
8290		for (;;) {
8291			if (f->pending == 0) {
8292				rc = f->valid ? EIO : 0;
8293				break;
8294			}
8295
8296			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8297			    PCATCH, "t4delfw", 0)) {
8298				rc = EINPROGRESS;
8299				break;
8300			}
8301		}
8302		mtx_unlock(&sc->tids.ftid_lock);
8303	}
8304
8305	return (rc);
8306}
8307
8308static void
8309clear_filter(struct filter_entry *f)
8310{
8311	if (f->l2t)
8312		t4_l2t_release(f->l2t);
8313
8314	bzero(f, sizeof (*f));
8315}
8316
8317static int
8318set_filter_wr(struct adapter *sc, int fidx)
8319{
8320	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8321	struct fw_filter_wr *fwr;
8322	unsigned int ftid, vnic_vld, vnic_vld_mask;
8323	struct wrq_cookie cookie;
8324
8325	ASSERT_SYNCHRONIZED_OP(sc);
8326
8327	if (f->fs.newdmac || f->fs.newvlan) {
8328		/* This filter needs an L2T entry; allocate one. */
8329		f->l2t = t4_l2t_alloc_switching(sc->l2t);
8330		if (f->l2t == NULL)
8331			return (EAGAIN);
8332		if (t4_l2t_set_switching(sc, f->l2t, f->fs.vlan, f->fs.eport,
8333		    f->fs.dmac)) {
8334			t4_l2t_release(f->l2t);
8335			f->l2t = NULL;
8336			return (ENOMEM);
8337		}
8338	}
8339
8340	/* Already validated against fconf, iconf */
8341	MPASS((f->fs.val.pfvf_vld & f->fs.val.ovlan_vld) == 0);
8342	MPASS((f->fs.mask.pfvf_vld & f->fs.mask.ovlan_vld) == 0);
8343	if (f->fs.val.pfvf_vld || f->fs.val.ovlan_vld)
8344		vnic_vld = 1;
8345	else
8346		vnic_vld = 0;
8347	if (f->fs.mask.pfvf_vld || f->fs.mask.ovlan_vld)
8348		vnic_vld_mask = 1;
8349	else
8350		vnic_vld_mask = 0;
8351
8352	ftid = sc->tids.ftid_base + fidx;
8353
8354	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8355	if (fwr == NULL)
8356		return (ENOMEM);
8357	bzero(fwr, sizeof(*fwr));
8358
8359	fwr->op_pkd = htobe32(V_FW_WR_OP(FW_FILTER_WR));
8360	fwr->len16_pkd = htobe32(FW_LEN16(*fwr));
8361	fwr->tid_to_iq =
8362	    htobe32(V_FW_FILTER_WR_TID(ftid) |
8363		V_FW_FILTER_WR_RQTYPE(f->fs.type) |
8364		V_FW_FILTER_WR_NOREPLY(0) |
8365		V_FW_FILTER_WR_IQ(f->fs.iq));
8366	fwr->del_filter_to_l2tix =
8367	    htobe32(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
8368		V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
8369		V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
8370		V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
8371		V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
8372		V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
8373		V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
8374		V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
8375		V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
8376		    f->fs.newvlan == VLAN_REWRITE) |
8377		V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
8378		    f->fs.newvlan == VLAN_REWRITE) |
8379		V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
8380		V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
8381		V_FW_FILTER_WR_PRIO(f->fs.prio) |
8382		V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
8383	fwr->ethtype = htobe16(f->fs.val.ethtype);
8384	fwr->ethtypem = htobe16(f->fs.mask.ethtype);
8385	fwr->frag_to_ovlan_vldm =
8386	    (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
8387		V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
8388		V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.vlan_vld) |
8389		V_FW_FILTER_WR_OVLAN_VLD(vnic_vld) |
8390		V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.vlan_vld) |
8391		V_FW_FILTER_WR_OVLAN_VLDM(vnic_vld_mask));
8392	fwr->smac_sel = 0;
8393	fwr->rx_chan_rx_rpl_iq = htobe16(V_FW_FILTER_WR_RX_CHAN(0) |
8394	    V_FW_FILTER_WR_RX_RPL_IQ(sc->sge.fwq.abs_id));
8395	fwr->maci_to_matchtypem =
8396	    htobe32(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
8397		V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
8398		V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
8399		V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
8400		V_FW_FILTER_WR_PORT(f->fs.val.iport) |
8401		V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
8402		V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
8403		V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
8404	fwr->ptcl = f->fs.val.proto;
8405	fwr->ptclm = f->fs.mask.proto;
8406	fwr->ttyp = f->fs.val.tos;
8407	fwr->ttypm = f->fs.mask.tos;
8408	fwr->ivlan = htobe16(f->fs.val.vlan);
8409	fwr->ivlanm = htobe16(f->fs.mask.vlan);
8410	fwr->ovlan = htobe16(f->fs.val.vnic);
8411	fwr->ovlanm = htobe16(f->fs.mask.vnic);
8412	bcopy(f->fs.val.dip, fwr->lip, sizeof (fwr->lip));
8413	bcopy(f->fs.mask.dip, fwr->lipm, sizeof (fwr->lipm));
8414	bcopy(f->fs.val.sip, fwr->fip, sizeof (fwr->fip));
8415	bcopy(f->fs.mask.sip, fwr->fipm, sizeof (fwr->fipm));
8416	fwr->lp = htobe16(f->fs.val.dport);
8417	fwr->lpm = htobe16(f->fs.mask.dport);
8418	fwr->fp = htobe16(f->fs.val.sport);
8419	fwr->fpm = htobe16(f->fs.mask.sport);
8420	if (f->fs.newsmac)
8421		bcopy(f->fs.smac, fwr->sma, sizeof (fwr->sma));
8422
8423	f->pending = 1;
8424	sc->tids.ftids_in_use++;
8425
8426	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8427	return (0);
8428}
8429
8430static int
8431del_filter_wr(struct adapter *sc, int fidx)
8432{
8433	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8434	struct fw_filter_wr *fwr;
8435	unsigned int ftid;
8436	struct wrq_cookie cookie;
8437
8438	ftid = sc->tids.ftid_base + fidx;
8439
8440	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8441	if (fwr == NULL)
8442		return (ENOMEM);
8443	bzero(fwr, sizeof (*fwr));
8444
8445	t4_mk_filtdelwr(ftid, fwr, sc->sge.fwq.abs_id);
8446
8447	f->pending = 1;
8448	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8449	return (0);
8450}
8451
8452int
8453t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8454{
8455	struct adapter *sc = iq->adapter;
8456	const struct cpl_set_tcb_rpl *rpl = (const void *)(rss + 1);
8457	unsigned int idx = GET_TID(rpl);
8458	unsigned int rc;
8459	struct filter_entry *f;
8460
8461	KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
8462	    rss->opcode));
8463	MPASS(iq == &sc->sge.fwq);
8464	MPASS(is_ftid(sc, idx));
8465
8466	idx -= sc->tids.ftid_base;
8467	f = &sc->tids.ftid_tab[idx];
8468	rc = G_COOKIE(rpl->cookie);
8469
8470	mtx_lock(&sc->tids.ftid_lock);
8471	if (rc == FW_FILTER_WR_FLT_ADDED) {
8472		KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
8473		    __func__, idx));
8474		f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
8475		f->pending = 0;  /* asynchronous setup completed */
8476		f->valid = 1;
8477	} else {
8478		if (rc != FW_FILTER_WR_FLT_DELETED) {
8479			/* Add or delete failed, display an error */
8480			log(LOG_ERR,
8481			    "filter %u setup failed with error %u\n",
8482			    idx, rc);
8483		}
8484
8485		clear_filter(f);
8486		sc->tids.ftids_in_use--;
8487	}
8488	wakeup(&sc->tids.ftid_tab);
8489	mtx_unlock(&sc->tids.ftid_lock);
8490
8491	return (0);
8492}
8493
8494static int
8495set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8496{
8497
8498	MPASS(iq->set_tcb_rpl != NULL);
8499	return (iq->set_tcb_rpl(iq, rss, m));
8500}
8501
8502static int
8503l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8504{
8505
8506	MPASS(iq->l2t_write_rpl != NULL);
8507	return (iq->l2t_write_rpl(iq, rss, m));
8508}
8509
8510static int
8511get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
8512{
8513	int rc;
8514
8515	if (cntxt->cid > M_CTXTQID)
8516		return (EINVAL);
8517
8518	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
8519	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
8520		return (EINVAL);
8521
8522	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
8523	if (rc)
8524		return (rc);
8525
8526	if (sc->flags & FW_OK) {
8527		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
8528		    &cntxt->data[0]);
8529		if (rc == 0)
8530			goto done;
8531	}
8532
8533	/*
8534	 * Read via firmware failed or wasn't even attempted.  Read directly via
8535	 * the backdoor.
8536	 */
8537	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
8538done:
8539	end_synchronized_op(sc, 0);
8540	return (rc);
8541}
8542
8543static int
8544load_fw(struct adapter *sc, struct t4_data *fw)
8545{
8546	int rc;
8547	uint8_t *fw_data;
8548
8549	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
8550	if (rc)
8551		return (rc);
8552
8553	if (sc->flags & FULL_INIT_DONE) {
8554		rc = EBUSY;
8555		goto done;
8556	}
8557
8558	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
8559	if (fw_data == NULL) {
8560		rc = ENOMEM;
8561		goto done;
8562	}
8563
8564	rc = copyin(fw->data, fw_data, fw->len);
8565	if (rc == 0)
8566		rc = -t4_load_fw(sc, fw_data, fw->len);
8567
8568	free(fw_data, M_CXGBE);
8569done:
8570	end_synchronized_op(sc, 0);
8571	return (rc);
8572}
8573
8574static int
8575load_cfg(struct adapter *sc, struct t4_data *cfg)
8576{
8577	int rc;
8578	uint8_t *cfg_data = NULL;
8579
8580	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
8581	if (rc)
8582		return (rc);
8583
8584	if (cfg->len == 0) {
8585		/* clear */
8586		rc = -t4_load_cfg(sc, NULL, 0);
8587		goto done;
8588	}
8589
8590	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
8591	if (cfg_data == NULL) {
8592		rc = ENOMEM;
8593		goto done;
8594	}
8595
8596	rc = copyin(cfg->data, cfg_data, cfg->len);
8597	if (rc == 0)
8598		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
8599
8600	free(cfg_data, M_CXGBE);
8601done:
8602	end_synchronized_op(sc, 0);
8603	return (rc);
8604}
8605
8606#define MAX_READ_BUF_SIZE (128 * 1024)
8607static int
8608read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
8609{
8610	uint32_t addr, remaining, n;
8611	uint32_t *buf;
8612	int rc;
8613	uint8_t *dst;
8614
8615	rc = validate_mem_range(sc, mr->addr, mr->len);
8616	if (rc != 0)
8617		return (rc);
8618
8619	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
8620	addr = mr->addr;
8621	remaining = mr->len;
8622	dst = (void *)mr->data;
8623
8624	while (remaining) {
8625		n = min(remaining, MAX_READ_BUF_SIZE);
8626		read_via_memwin(sc, 2, addr, buf, n);
8627
8628		rc = copyout(buf, dst, n);
8629		if (rc != 0)
8630			break;
8631
8632		dst += n;
8633		remaining -= n;
8634		addr += n;
8635	}
8636
8637	free(buf, M_CXGBE);
8638	return (rc);
8639}
8640#undef MAX_READ_BUF_SIZE
8641
8642static int
8643read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
8644{
8645	int rc;
8646
8647	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
8648		return (EINVAL);
8649
8650	if (i2cd->len > sizeof(i2cd->data))
8651		return (EFBIG);
8652
8653	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
8654	if (rc)
8655		return (rc);
8656	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
8657	    i2cd->offset, i2cd->len, &i2cd->data[0]);
8658	end_synchronized_op(sc, 0);
8659
8660	return (rc);
8661}
8662
8663static int
8664in_range(int val, int lo, int hi)
8665{
8666
8667	return (val < 0 || (val <= hi && val >= lo));
8668}
8669
8670static int
8671set_sched_class_config(struct adapter *sc, int minmax)
8672{
8673	int rc;
8674
8675	if (minmax < 0)
8676		return (EINVAL);
8677
8678	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4sscc");
8679	if (rc)
8680		return (rc);
8681	rc = -t4_sched_config(sc, FW_SCHED_TYPE_PKTSCHED, minmax, 1);
8682	end_synchronized_op(sc, 0);
8683
8684	return (rc);
8685}
8686
8687static int
8688set_sched_class_params(struct adapter *sc, struct t4_sched_class_params *p,
8689    int sleep_ok)
8690{
8691	int rc, top_speed, fw_level, fw_mode, fw_rateunit, fw_ratemode;
8692	struct port_info *pi;
8693	struct tx_sched_class *tc;
8694
8695	if (p->level == SCHED_CLASS_LEVEL_CL_RL)
8696		fw_level = FW_SCHED_PARAMS_LEVEL_CL_RL;
8697	else if (p->level == SCHED_CLASS_LEVEL_CL_WRR)
8698		fw_level = FW_SCHED_PARAMS_LEVEL_CL_WRR;
8699	else if (p->level == SCHED_CLASS_LEVEL_CH_RL)
8700		fw_level = FW_SCHED_PARAMS_LEVEL_CH_RL;
8701	else
8702		return (EINVAL);
8703
8704	if (p->mode == SCHED_CLASS_MODE_CLASS)
8705		fw_mode = FW_SCHED_PARAMS_MODE_CLASS;
8706	else if (p->mode == SCHED_CLASS_MODE_FLOW)
8707		fw_mode = FW_SCHED_PARAMS_MODE_FLOW;
8708	else
8709		return (EINVAL);
8710
8711	if (p->rateunit == SCHED_CLASS_RATEUNIT_BITS)
8712		fw_rateunit = FW_SCHED_PARAMS_UNIT_BITRATE;
8713	else if (p->rateunit == SCHED_CLASS_RATEUNIT_PKTS)
8714		fw_rateunit = FW_SCHED_PARAMS_UNIT_PKTRATE;
8715	else
8716		return (EINVAL);
8717
8718	if (p->ratemode == SCHED_CLASS_RATEMODE_REL)
8719		fw_ratemode = FW_SCHED_PARAMS_RATE_REL;
8720	else if (p->ratemode == SCHED_CLASS_RATEMODE_ABS)
8721		fw_ratemode = FW_SCHED_PARAMS_RATE_ABS;
8722	else
8723		return (EINVAL);
8724
8725	/* Vet our parameters ... */
8726	if (!in_range(p->channel, 0, sc->chip_params->nchan - 1))
8727		return (ERANGE);
8728
8729	pi = sc->port[sc->chan_map[p->channel]];
8730	if (pi == NULL)
8731		return (ENXIO);
8732	MPASS(pi->tx_chan == p->channel);
8733	top_speed = port_top_speed(pi) * 1000000; /* Gbps -> Kbps */
8734
8735	if (!in_range(p->cl, 0, sc->chip_params->nsched_cls) ||
8736	    !in_range(p->minrate, 0, top_speed) ||
8737	    !in_range(p->maxrate, 0, top_speed) ||
8738	    !in_range(p->weight, 0, 100))
8739		return (ERANGE);
8740
8741	/*
8742	 * Translate any unset parameters into the firmware's
8743	 * nomenclature and/or fail the call if the parameters
8744	 * are required ...
8745	 */
8746	if (p->rateunit < 0 || p->ratemode < 0 || p->channel < 0 || p->cl < 0)
8747		return (EINVAL);
8748
8749	if (p->minrate < 0)
8750		p->minrate = 0;
8751	if (p->maxrate < 0) {
8752		if (p->level == SCHED_CLASS_LEVEL_CL_RL ||
8753		    p->level == SCHED_CLASS_LEVEL_CH_RL)
8754			return (EINVAL);
8755		else
8756			p->maxrate = 0;
8757	}
8758	if (p->weight < 0) {
8759		if (p->level == SCHED_CLASS_LEVEL_CL_WRR)
8760			return (EINVAL);
8761		else
8762			p->weight = 0;
8763	}
8764	if (p->pktsize < 0) {
8765		if (p->level == SCHED_CLASS_LEVEL_CL_RL ||
8766		    p->level == SCHED_CLASS_LEVEL_CH_RL)
8767			return (EINVAL);
8768		else
8769			p->pktsize = 0;
8770	}
8771
8772	rc = begin_synchronized_op(sc, NULL,
8773	    sleep_ok ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4sscp");
8774	if (rc)
8775		return (rc);
8776	tc = &pi->tc[p->cl];
8777	tc->params = *p;
8778	rc = -t4_sched_params(sc, FW_SCHED_TYPE_PKTSCHED, fw_level, fw_mode,
8779	    fw_rateunit, fw_ratemode, p->channel, p->cl, p->minrate, p->maxrate,
8780	    p->weight, p->pktsize, sleep_ok);
8781	if (rc == 0)
8782		tc->flags |= TX_SC_OK;
8783	else {
8784		/*
8785		 * Unknown state at this point, see tc->params for what was
8786		 * attempted.
8787		 */
8788		tc->flags &= ~TX_SC_OK;
8789	}
8790	end_synchronized_op(sc, sleep_ok ? 0 : LOCK_HELD);
8791
8792	return (rc);
8793}
8794
8795int
8796t4_set_sched_class(struct adapter *sc, struct t4_sched_params *p)
8797{
8798
8799	if (p->type != SCHED_CLASS_TYPE_PACKET)
8800		return (EINVAL);
8801
8802	if (p->subcmd == SCHED_CLASS_SUBCMD_CONFIG)
8803		return (set_sched_class_config(sc, p->u.config.minmax));
8804
8805	if (p->subcmd == SCHED_CLASS_SUBCMD_PARAMS)
8806		return (set_sched_class_params(sc, &p->u.params, 1));
8807
8808	return (EINVAL);
8809}
8810
8811int
8812t4_set_sched_queue(struct adapter *sc, struct t4_sched_queue *p)
8813{
8814	struct port_info *pi = NULL;
8815	struct vi_info *vi;
8816	struct sge_txq *txq;
8817	uint32_t fw_mnem, fw_queue, fw_class;
8818	int i, rc;
8819
8820	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setsq");
8821	if (rc)
8822		return (rc);
8823
8824	if (p->port >= sc->params.nports) {
8825		rc = EINVAL;
8826		goto done;
8827	}
8828
8829	/* XXX: Only supported for the main VI. */
8830	pi = sc->port[p->port];
8831	vi = &pi->vi[0];
8832	if (!(vi->flags & VI_INIT_DONE)) {
8833		/* tx queues not set up yet */
8834		rc = EAGAIN;
8835		goto done;
8836	}
8837
8838	if (!in_range(p->queue, 0, vi->ntxq - 1) ||
8839	    !in_range(p->cl, 0, sc->chip_params->nsched_cls - 1)) {
8840		rc = EINVAL;
8841		goto done;
8842	}
8843
8844	/*
8845	 * Create a template for the FW_PARAMS_CMD mnemonic and value (TX
8846	 * Scheduling Class in this case).
8847	 */
8848	fw_mnem = (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
8849	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_EQ_SCHEDCLASS_ETH));
8850	fw_class = p->cl < 0 ? 0xffffffff : p->cl;
8851
8852	/*
8853	 * If op.queue is non-negative, then we're only changing the scheduling
8854	 * on a single specified TX queue.
8855	 */
8856	if (p->queue >= 0) {
8857		txq = &sc->sge.txq[vi->first_txq + p->queue];
8858		fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
8859		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
8860		    &fw_class);
8861		goto done;
8862	}
8863
8864	/*
8865	 * Change the scheduling on all the TX queues for the
8866	 * interface.
8867	 */
8868	for_each_txq(vi, i, txq) {
8869		fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
8870		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
8871		    &fw_class);
8872		if (rc)
8873			goto done;
8874	}
8875
8876	rc = 0;
8877done:
8878	end_synchronized_op(sc, 0);
8879	return (rc);
8880}
8881
8882int
8883t4_os_find_pci_capability(struct adapter *sc, int cap)
8884{
8885	int i;
8886
8887	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
8888}
8889
8890int
8891t4_os_pci_save_state(struct adapter *sc)
8892{
8893	device_t dev;
8894	struct pci_devinfo *dinfo;
8895
8896	dev = sc->dev;
8897	dinfo = device_get_ivars(dev);
8898
8899	pci_cfg_save(dev, dinfo, 0);
8900	return (0);
8901}
8902
8903int
8904t4_os_pci_restore_state(struct adapter *sc)
8905{
8906	device_t dev;
8907	struct pci_devinfo *dinfo;
8908
8909	dev = sc->dev;
8910	dinfo = device_get_ivars(dev);
8911
8912	pci_cfg_restore(dev, dinfo);
8913	return (0);
8914}
8915
8916void
8917t4_os_portmod_changed(const struct adapter *sc, int idx)
8918{
8919	struct port_info *pi = sc->port[idx];
8920	struct vi_info *vi;
8921	struct ifnet *ifp;
8922	int v;
8923	static const char *mod_str[] = {
8924		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
8925	};
8926
8927	for_each_vi(pi, v, vi) {
8928		build_medialist(pi, &vi->media);
8929	}
8930
8931	ifp = pi->vi[0].ifp;
8932	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
8933		if_printf(ifp, "transceiver unplugged.\n");
8934	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
8935		if_printf(ifp, "unknown transceiver inserted.\n");
8936	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
8937		if_printf(ifp, "unsupported transceiver inserted.\n");
8938	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
8939		if_printf(ifp, "%s transceiver inserted.\n",
8940		    mod_str[pi->mod_type]);
8941	} else {
8942		if_printf(ifp, "transceiver (type %d) inserted.\n",
8943		    pi->mod_type);
8944	}
8945}
8946
8947void
8948t4_os_link_changed(struct adapter *sc, int idx, int link_stat)
8949{
8950	struct port_info *pi = sc->port[idx];
8951	struct vi_info *vi;
8952	struct ifnet *ifp;
8953	int v;
8954
8955	for_each_vi(pi, v, vi) {
8956		ifp = vi->ifp;
8957		if (ifp == NULL)
8958			continue;
8959
8960		if (link_stat) {
8961			ifp->if_baudrate = IF_Mbps(pi->link_cfg.speed);
8962			if_link_state_change(ifp, LINK_STATE_UP);
8963		} else {
8964			if_link_state_change(ifp, LINK_STATE_DOWN);
8965		}
8966	}
8967}
8968
8969void
8970t4_iterate(void (*func)(struct adapter *, void *), void *arg)
8971{
8972	struct adapter *sc;
8973
8974	sx_slock(&t4_list_lock);
8975	SLIST_FOREACH(sc, &t4_list, link) {
8976		/*
8977		 * func should not make any assumptions about what state sc is
8978		 * in - the only guarantee is that sc->sc_lock is a valid lock.
8979		 */
8980		func(sc, arg);
8981	}
8982	sx_sunlock(&t4_list_lock);
8983}
8984
8985static int
8986t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
8987    struct thread *td)
8988{
8989	int rc;
8990	struct adapter *sc = dev->si_drv1;
8991
8992	rc = priv_check(td, PRIV_DRIVER);
8993	if (rc != 0)
8994		return (rc);
8995
8996	switch (cmd) {
8997	case CHELSIO_T4_GETREG: {
8998		struct t4_reg *edata = (struct t4_reg *)data;
8999
9000		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9001			return (EFAULT);
9002
9003		if (edata->size == 4)
9004			edata->val = t4_read_reg(sc, edata->addr);
9005		else if (edata->size == 8)
9006			edata->val = t4_read_reg64(sc, edata->addr);
9007		else
9008			return (EINVAL);
9009
9010		break;
9011	}
9012	case CHELSIO_T4_SETREG: {
9013		struct t4_reg *edata = (struct t4_reg *)data;
9014
9015		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9016			return (EFAULT);
9017
9018		if (edata->size == 4) {
9019			if (edata->val & 0xffffffff00000000)
9020				return (EINVAL);
9021			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9022		} else if (edata->size == 8)
9023			t4_write_reg64(sc, edata->addr, edata->val);
9024		else
9025			return (EINVAL);
9026		break;
9027	}
9028	case CHELSIO_T4_REGDUMP: {
9029		struct t4_regdump *regs = (struct t4_regdump *)data;
9030		int reglen = t4_get_regs_len(sc);
9031		uint8_t *buf;
9032
9033		if (regs->len < reglen) {
9034			regs->len = reglen; /* hint to the caller */
9035			return (ENOBUFS);
9036		}
9037
9038		regs->len = reglen;
9039		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9040		get_regs(sc, regs, buf);
9041		rc = copyout(buf, regs->data, reglen);
9042		free(buf, M_CXGBE);
9043		break;
9044	}
9045	case CHELSIO_T4_GET_FILTER_MODE:
9046		rc = get_filter_mode(sc, (uint32_t *)data);
9047		break;
9048	case CHELSIO_T4_SET_FILTER_MODE:
9049		rc = set_filter_mode(sc, *(uint32_t *)data);
9050		break;
9051	case CHELSIO_T4_GET_FILTER:
9052		rc = get_filter(sc, (struct t4_filter *)data);
9053		break;
9054	case CHELSIO_T4_SET_FILTER:
9055		rc = set_filter(sc, (struct t4_filter *)data);
9056		break;
9057	case CHELSIO_T4_DEL_FILTER:
9058		rc = del_filter(sc, (struct t4_filter *)data);
9059		break;
9060	case CHELSIO_T4_GET_SGE_CONTEXT:
9061		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9062		break;
9063	case CHELSIO_T4_LOAD_FW:
9064		rc = load_fw(sc, (struct t4_data *)data);
9065		break;
9066	case CHELSIO_T4_GET_MEM:
9067		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9068		break;
9069	case CHELSIO_T4_GET_I2C:
9070		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9071		break;
9072	case CHELSIO_T4_CLEAR_STATS: {
9073		int i, v;
9074		u_int port_id = *(uint32_t *)data;
9075		struct port_info *pi;
9076		struct vi_info *vi;
9077
9078		if (port_id >= sc->params.nports)
9079			return (EINVAL);
9080		pi = sc->port[port_id];
9081		if (pi == NULL)
9082			return (EIO);
9083
9084		/* MAC stats */
9085		t4_clr_port_stats(sc, pi->tx_chan);
9086		pi->tx_parse_error = 0;
9087		mtx_lock(&sc->reg_lock);
9088		for_each_vi(pi, v, vi) {
9089			if (vi->flags & VI_INIT_DONE)
9090				t4_clr_vi_stats(sc, vi->viid);
9091		}
9092		mtx_unlock(&sc->reg_lock);
9093
9094		/*
9095		 * Since this command accepts a port, clear stats for
9096		 * all VIs on this port.
9097		 */
9098		for_each_vi(pi, v, vi) {
9099			if (vi->flags & VI_INIT_DONE) {
9100				struct sge_rxq *rxq;
9101				struct sge_txq *txq;
9102				struct sge_wrq *wrq;
9103
9104				for_each_rxq(vi, i, rxq) {
9105#if defined(INET) || defined(INET6)
9106					rxq->lro.lro_queued = 0;
9107					rxq->lro.lro_flushed = 0;
9108#endif
9109					rxq->rxcsum = 0;
9110					rxq->vlan_extraction = 0;
9111				}
9112
9113				for_each_txq(vi, i, txq) {
9114					txq->txcsum = 0;
9115					txq->tso_wrs = 0;
9116					txq->vlan_insertion = 0;
9117					txq->imm_wrs = 0;
9118					txq->sgl_wrs = 0;
9119					txq->txpkt_wrs = 0;
9120					txq->txpkts0_wrs = 0;
9121					txq->txpkts1_wrs = 0;
9122					txq->txpkts0_pkts = 0;
9123					txq->txpkts1_pkts = 0;
9124					mp_ring_reset_stats(txq->r);
9125				}
9126
9127#ifdef TCP_OFFLOAD
9128				/* nothing to clear for each ofld_rxq */
9129
9130				for_each_ofld_txq(vi, i, wrq) {
9131					wrq->tx_wrs_direct = 0;
9132					wrq->tx_wrs_copied = 0;
9133				}
9134#endif
9135
9136				if (IS_MAIN_VI(vi)) {
9137					wrq = &sc->sge.ctrlq[pi->port_id];
9138					wrq->tx_wrs_direct = 0;
9139					wrq->tx_wrs_copied = 0;
9140				}
9141			}
9142		}
9143		break;
9144	}
9145	case CHELSIO_T4_SCHED_CLASS:
9146		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9147		break;
9148	case CHELSIO_T4_SCHED_QUEUE:
9149		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9150		break;
9151	case CHELSIO_T4_GET_TRACER:
9152		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9153		break;
9154	case CHELSIO_T4_SET_TRACER:
9155		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9156		break;
9157	case CHELSIO_T4_LOAD_CFG:
9158		rc = load_cfg(sc, (struct t4_data *)data);
9159		break;
9160	default:
9161		rc = ENOTTY;
9162	}
9163
9164	return (rc);
9165}
9166
9167void
9168t4_db_full(struct adapter *sc)
9169{
9170
9171	CXGBE_UNIMPLEMENTED(__func__);
9172}
9173
9174void
9175t4_db_dropped(struct adapter *sc)
9176{
9177
9178	CXGBE_UNIMPLEMENTED(__func__);
9179}
9180
9181#ifdef TCP_OFFLOAD
9182void
9183t4_iscsi_init(struct adapter *sc, u_int tag_mask, const u_int *pgsz_order)
9184{
9185
9186	t4_write_reg(sc, A_ULP_RX_ISCSI_TAGMASK, tag_mask);
9187	t4_write_reg(sc, A_ULP_RX_ISCSI_PSZ, V_HPZ0(pgsz_order[0]) |
9188		V_HPZ1(pgsz_order[1]) | V_HPZ2(pgsz_order[2]) |
9189		V_HPZ3(pgsz_order[3]));
9190}
9191
9192static int
9193toe_capability(struct vi_info *vi, int enable)
9194{
9195	int rc;
9196	struct port_info *pi = vi->pi;
9197	struct adapter *sc = pi->adapter;
9198
9199	ASSERT_SYNCHRONIZED_OP(sc);
9200
9201	if (!is_offload(sc))
9202		return (ENODEV);
9203
9204	if (enable) {
9205		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9206			/* TOE is already enabled. */
9207			return (0);
9208		}
9209
9210		/*
9211		 * We need the port's queues around so that we're able to send
9212		 * and receive CPLs to/from the TOE even if the ifnet for this
9213		 * port has never been UP'd administratively.
9214		 */
9215		if (!(vi->flags & VI_INIT_DONE)) {
9216			rc = vi_full_init(vi);
9217			if (rc)
9218				return (rc);
9219		}
9220		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9221			rc = vi_full_init(&pi->vi[0]);
9222			if (rc)
9223				return (rc);
9224		}
9225
9226		if (isset(&sc->offload_map, pi->port_id)) {
9227			/* TOE is enabled on another VI of this port. */
9228			pi->uld_vis++;
9229			return (0);
9230		}
9231
9232		if (!uld_active(sc, ULD_TOM)) {
9233			rc = t4_activate_uld(sc, ULD_TOM);
9234			if (rc == EAGAIN) {
9235				log(LOG_WARNING,
9236				    "You must kldload t4_tom.ko before trying "
9237				    "to enable TOE on a cxgbe interface.\n");
9238			}
9239			if (rc != 0)
9240				return (rc);
9241			KASSERT(sc->tom_softc != NULL,
9242			    ("%s: TOM activated but softc NULL", __func__));
9243			KASSERT(uld_active(sc, ULD_TOM),
9244			    ("%s: TOM activated but flag not set", __func__));
9245		}
9246
9247		/* Activate iWARP and iSCSI too, if the modules are loaded. */
9248		if (!uld_active(sc, ULD_IWARP))
9249			(void) t4_activate_uld(sc, ULD_IWARP);
9250		if (!uld_active(sc, ULD_ISCSI))
9251			(void) t4_activate_uld(sc, ULD_ISCSI);
9252
9253		pi->uld_vis++;
9254		setbit(&sc->offload_map, pi->port_id);
9255	} else {
9256		pi->uld_vis--;
9257
9258		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9259			return (0);
9260
9261		KASSERT(uld_active(sc, ULD_TOM),
9262		    ("%s: TOM never initialized?", __func__));
9263		clrbit(&sc->offload_map, pi->port_id);
9264	}
9265
9266	return (0);
9267}
9268
9269/*
9270 * Add an upper layer driver to the global list.
9271 */
9272int
9273t4_register_uld(struct uld_info *ui)
9274{
9275	int rc = 0;
9276	struct uld_info *u;
9277
9278	sx_xlock(&t4_uld_list_lock);
9279	SLIST_FOREACH(u, &t4_uld_list, link) {
9280	    if (u->uld_id == ui->uld_id) {
9281		    rc = EEXIST;
9282		    goto done;
9283	    }
9284	}
9285
9286	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9287	ui->refcount = 0;
9288done:
9289	sx_xunlock(&t4_uld_list_lock);
9290	return (rc);
9291}
9292
9293int
9294t4_unregister_uld(struct uld_info *ui)
9295{
9296	int rc = EINVAL;
9297	struct uld_info *u;
9298
9299	sx_xlock(&t4_uld_list_lock);
9300
9301	SLIST_FOREACH(u, &t4_uld_list, link) {
9302	    if (u == ui) {
9303		    if (ui->refcount > 0) {
9304			    rc = EBUSY;
9305			    goto done;
9306		    }
9307
9308		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9309		    rc = 0;
9310		    goto done;
9311	    }
9312	}
9313done:
9314	sx_xunlock(&t4_uld_list_lock);
9315	return (rc);
9316}
9317
9318int
9319t4_activate_uld(struct adapter *sc, int id)
9320{
9321	int rc;
9322	struct uld_info *ui;
9323
9324	ASSERT_SYNCHRONIZED_OP(sc);
9325
9326	if (id < 0 || id > ULD_MAX)
9327		return (EINVAL);
9328	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
9329
9330	sx_slock(&t4_uld_list_lock);
9331
9332	SLIST_FOREACH(ui, &t4_uld_list, link) {
9333		if (ui->uld_id == id) {
9334			if (!(sc->flags & FULL_INIT_DONE)) {
9335				rc = adapter_full_init(sc);
9336				if (rc != 0)
9337					break;
9338			}
9339
9340			rc = ui->activate(sc);
9341			if (rc == 0) {
9342				setbit(&sc->active_ulds, id);
9343				ui->refcount++;
9344			}
9345			break;
9346		}
9347	}
9348
9349	sx_sunlock(&t4_uld_list_lock);
9350
9351	return (rc);
9352}
9353
9354int
9355t4_deactivate_uld(struct adapter *sc, int id)
9356{
9357	int rc;
9358	struct uld_info *ui;
9359
9360	ASSERT_SYNCHRONIZED_OP(sc);
9361
9362	if (id < 0 || id > ULD_MAX)
9363		return (EINVAL);
9364	rc = ENXIO;
9365
9366	sx_slock(&t4_uld_list_lock);
9367
9368	SLIST_FOREACH(ui, &t4_uld_list, link) {
9369		if (ui->uld_id == id) {
9370			rc = ui->deactivate(sc);
9371			if (rc == 0) {
9372				clrbit(&sc->active_ulds, id);
9373				ui->refcount--;
9374			}
9375			break;
9376		}
9377	}
9378
9379	sx_sunlock(&t4_uld_list_lock);
9380
9381	return (rc);
9382}
9383
9384int
9385uld_active(struct adapter *sc, int uld_id)
9386{
9387
9388	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9389
9390	return (isset(&sc->active_ulds, uld_id));
9391}
9392#endif
9393
9394/*
9395 * Come up with reasonable defaults for some of the tunables, provided they're
9396 * not set by the user (in which case we'll use the values as is).
9397 */
9398static void
9399tweak_tunables(void)
9400{
9401	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
9402
9403	if (t4_ntxq10g < 1) {
9404#ifdef RSS
9405		t4_ntxq10g = rss_getnumbuckets();
9406#else
9407		t4_ntxq10g = min(nc, NTXQ_10G);
9408#endif
9409	}
9410
9411	if (t4_ntxq1g < 1) {
9412#ifdef RSS
9413		/* XXX: way too many for 1GbE? */
9414		t4_ntxq1g = rss_getnumbuckets();
9415#else
9416		t4_ntxq1g = min(nc, NTXQ_1G);
9417#endif
9418	}
9419
9420	if (t4_ntxq_vi < 1)
9421		t4_ntxq_vi = min(nc, NTXQ_VI);
9422
9423	if (t4_nrxq10g < 1) {
9424#ifdef RSS
9425		t4_nrxq10g = rss_getnumbuckets();
9426#else
9427		t4_nrxq10g = min(nc, NRXQ_10G);
9428#endif
9429	}
9430
9431	if (t4_nrxq1g < 1) {
9432#ifdef RSS
9433		/* XXX: way too many for 1GbE? */
9434		t4_nrxq1g = rss_getnumbuckets();
9435#else
9436		t4_nrxq1g = min(nc, NRXQ_1G);
9437#endif
9438	}
9439
9440	if (t4_nrxq_vi < 1)
9441		t4_nrxq_vi = min(nc, NRXQ_VI);
9442
9443#ifdef TCP_OFFLOAD
9444	if (t4_nofldtxq10g < 1)
9445		t4_nofldtxq10g = min(nc, NOFLDTXQ_10G);
9446
9447	if (t4_nofldtxq1g < 1)
9448		t4_nofldtxq1g = min(nc, NOFLDTXQ_1G);
9449
9450	if (t4_nofldtxq_vi < 1)
9451		t4_nofldtxq_vi = min(nc, NOFLDTXQ_VI);
9452
9453	if (t4_nofldrxq10g < 1)
9454		t4_nofldrxq10g = min(nc, NOFLDRXQ_10G);
9455
9456	if (t4_nofldrxq1g < 1)
9457		t4_nofldrxq1g = min(nc, NOFLDRXQ_1G);
9458
9459	if (t4_nofldrxq_vi < 1)
9460		t4_nofldrxq_vi = min(nc, NOFLDRXQ_VI);
9461
9462	if (t4_toecaps_allowed == -1)
9463		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9464
9465	if (t4_rdmacaps_allowed == -1) {
9466		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9467		    FW_CAPS_CONFIG_RDMA_RDMAC;
9468	}
9469
9470	if (t4_iscsicaps_allowed == -1) {
9471		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9472		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9473		    FW_CAPS_CONFIG_ISCSI_T10DIF;
9474	}
9475#else
9476	if (t4_toecaps_allowed == -1)
9477		t4_toecaps_allowed = 0;
9478
9479	if (t4_rdmacaps_allowed == -1)
9480		t4_rdmacaps_allowed = 0;
9481
9482	if (t4_iscsicaps_allowed == -1)
9483		t4_iscsicaps_allowed = 0;
9484#endif
9485
9486#ifdef DEV_NETMAP
9487	if (t4_nnmtxq_vi < 1)
9488		t4_nnmtxq_vi = min(nc, NNMTXQ_VI);
9489
9490	if (t4_nnmrxq_vi < 1)
9491		t4_nnmrxq_vi = min(nc, NNMRXQ_VI);
9492#endif
9493
9494	if (t4_tmr_idx_10g < 0 || t4_tmr_idx_10g >= SGE_NTIMERS)
9495		t4_tmr_idx_10g = TMR_IDX_10G;
9496
9497	if (t4_pktc_idx_10g < -1 || t4_pktc_idx_10g >= SGE_NCOUNTERS)
9498		t4_pktc_idx_10g = PKTC_IDX_10G;
9499
9500	if (t4_tmr_idx_1g < 0 || t4_tmr_idx_1g >= SGE_NTIMERS)
9501		t4_tmr_idx_1g = TMR_IDX_1G;
9502
9503	if (t4_pktc_idx_1g < -1 || t4_pktc_idx_1g >= SGE_NCOUNTERS)
9504		t4_pktc_idx_1g = PKTC_IDX_1G;
9505
9506	if (t4_qsize_txq < 128)
9507		t4_qsize_txq = 128;
9508
9509	if (t4_qsize_rxq < 128)
9510		t4_qsize_rxq = 128;
9511	while (t4_qsize_rxq & 7)
9512		t4_qsize_rxq++;
9513
9514	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9515}
9516
9517#ifdef DDB
9518static void
9519t4_dump_tcb(struct adapter *sc, int tid)
9520{
9521	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
9522
9523	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
9524	save = t4_read_reg(sc, reg);
9525	base = sc->memwin[2].mw_base;
9526
9527	/* Dump TCB for the tid */
9528	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
9529	tcb_addr += tid * TCB_SIZE;
9530
9531	if (is_t4(sc)) {
9532		pf = 0;
9533		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
9534	} else {
9535		pf = V_PFNUM(sc->pf);
9536		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
9537	}
9538	t4_write_reg(sc, reg, win_pos | pf);
9539	t4_read_reg(sc, reg);
9540
9541	off = tcb_addr - win_pos;
9542	for (i = 0; i < 4; i++) {
9543		uint32_t buf[8];
9544		for (j = 0; j < 8; j++, off += 4)
9545			buf[j] = htonl(t4_read_reg(sc, base + off));
9546
9547		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
9548		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
9549		    buf[7]);
9550	}
9551
9552	t4_write_reg(sc, reg, save);
9553	t4_read_reg(sc, reg);
9554}
9555
9556static void
9557t4_dump_devlog(struct adapter *sc)
9558{
9559	struct devlog_params *dparams = &sc->params.devlog;
9560	struct fw_devlog_e e;
9561	int i, first, j, m, nentries, rc;
9562	uint64_t ftstamp = UINT64_MAX;
9563
9564	if (dparams->start == 0) {
9565		db_printf("devlog params not valid\n");
9566		return;
9567	}
9568
9569	nentries = dparams->size / sizeof(struct fw_devlog_e);
9570	m = fwmtype_to_hwmtype(dparams->memtype);
9571
9572	/* Find the first entry. */
9573	first = -1;
9574	for (i = 0; i < nentries && !db_pager_quit; i++) {
9575		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9576		    sizeof(e), (void *)&e);
9577		if (rc != 0)
9578			break;
9579
9580		if (e.timestamp == 0)
9581			break;
9582
9583		e.timestamp = be64toh(e.timestamp);
9584		if (e.timestamp < ftstamp) {
9585			ftstamp = e.timestamp;
9586			first = i;
9587		}
9588	}
9589
9590	if (first == -1)
9591		return;
9592
9593	i = first;
9594	do {
9595		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9596		    sizeof(e), (void *)&e);
9597		if (rc != 0)
9598			return;
9599
9600		if (e.timestamp == 0)
9601			return;
9602
9603		e.timestamp = be64toh(e.timestamp);
9604		e.seqno = be32toh(e.seqno);
9605		for (j = 0; j < 8; j++)
9606			e.params[j] = be32toh(e.params[j]);
9607
9608		db_printf("%10d  %15ju  %8s  %8s  ",
9609		    e.seqno, e.timestamp,
9610		    (e.level < nitems(devlog_level_strings) ?
9611			devlog_level_strings[e.level] : "UNKNOWN"),
9612		    (e.facility < nitems(devlog_facility_strings) ?
9613			devlog_facility_strings[e.facility] : "UNKNOWN"));
9614		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
9615		    e.params[3], e.params[4], e.params[5], e.params[6],
9616		    e.params[7]);
9617
9618		if (++i == nentries)
9619			i = 0;
9620	} while (i != first && !db_pager_quit);
9621}
9622
9623static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
9624_DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
9625
9626DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
9627{
9628	device_t dev;
9629	int t;
9630	bool valid;
9631
9632	valid = false;
9633	t = db_read_token();
9634	if (t == tIDENT) {
9635		dev = device_lookup_by_name(db_tok_string);
9636		valid = true;
9637	}
9638	db_skip_to_eol();
9639	if (!valid) {
9640		db_printf("usage: show t4 devlog <nexus>\n");
9641		return;
9642	}
9643
9644	if (dev == NULL) {
9645		db_printf("device not found\n");
9646		return;
9647	}
9648
9649	t4_dump_devlog(device_get_softc(dev));
9650}
9651
9652DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
9653{
9654	device_t dev;
9655	int radix, tid, t;
9656	bool valid;
9657
9658	valid = false;
9659	radix = db_radix;
9660	db_radix = 10;
9661	t = db_read_token();
9662	if (t == tIDENT) {
9663		dev = device_lookup_by_name(db_tok_string);
9664		t = db_read_token();
9665		if (t == tNUMBER) {
9666			tid = db_tok_number;
9667			valid = true;
9668		}
9669	}
9670	db_radix = radix;
9671	db_skip_to_eol();
9672	if (!valid) {
9673		db_printf("usage: show t4 tcb <nexus> <tid>\n");
9674		return;
9675	}
9676
9677	if (dev == NULL) {
9678		db_printf("device not found\n");
9679		return;
9680	}
9681	if (tid < 0) {
9682		db_printf("invalid tid\n");
9683		return;
9684	}
9685
9686	t4_dump_tcb(device_get_softc(dev), tid);
9687}
9688#endif
9689
9690static struct sx mlu;	/* mod load unload */
9691SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
9692
9693static int
9694mod_event(module_t mod, int cmd, void *arg)
9695{
9696	int rc = 0;
9697	static int loaded = 0;
9698
9699	switch (cmd) {
9700	case MOD_LOAD:
9701		sx_xlock(&mlu);
9702		if (loaded++ == 0) {
9703			t4_sge_modload();
9704			t4_register_cpl_handler(CPL_SET_TCB_RPL, set_tcb_rpl);
9705			t4_register_cpl_handler(CPL_L2T_WRITE_RPL, l2t_write_rpl);
9706			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
9707			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
9708			sx_init(&t4_list_lock, "T4/T5 adapters");
9709			SLIST_INIT(&t4_list);
9710#ifdef TCP_OFFLOAD
9711			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
9712			SLIST_INIT(&t4_uld_list);
9713#endif
9714			t4_tracer_modload();
9715			tweak_tunables();
9716		}
9717		sx_xunlock(&mlu);
9718		break;
9719
9720	case MOD_UNLOAD:
9721		sx_xlock(&mlu);
9722		if (--loaded == 0) {
9723			int tries;
9724
9725			sx_slock(&t4_list_lock);
9726			if (!SLIST_EMPTY(&t4_list)) {
9727				rc = EBUSY;
9728				sx_sunlock(&t4_list_lock);
9729				goto done_unload;
9730			}
9731#ifdef TCP_OFFLOAD
9732			sx_slock(&t4_uld_list_lock);
9733			if (!SLIST_EMPTY(&t4_uld_list)) {
9734				rc = EBUSY;
9735				sx_sunlock(&t4_uld_list_lock);
9736				sx_sunlock(&t4_list_lock);
9737				goto done_unload;
9738			}
9739#endif
9740			tries = 0;
9741			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
9742				uprintf("%ju clusters with custom free routine "
9743				    "still is use.\n", t4_sge_extfree_refs());
9744				pause("t4unload", 2 * hz);
9745			}
9746#ifdef TCP_OFFLOAD
9747			sx_sunlock(&t4_uld_list_lock);
9748#endif
9749			sx_sunlock(&t4_list_lock);
9750
9751			if (t4_sge_extfree_refs() == 0) {
9752				t4_tracer_modunload();
9753#ifdef TCP_OFFLOAD
9754				sx_destroy(&t4_uld_list_lock);
9755#endif
9756				sx_destroy(&t4_list_lock);
9757				t4_sge_modunload();
9758				loaded = 0;
9759			} else {
9760				rc = EBUSY;
9761				loaded++;	/* undo earlier decrement */
9762			}
9763		}
9764done_unload:
9765		sx_xunlock(&mlu);
9766		break;
9767	}
9768
9769	return (rc);
9770}
9771
9772static devclass_t t4_devclass, t5_devclass, t6_devclass;
9773static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
9774static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
9775
9776DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
9777MODULE_VERSION(t4nex, 1);
9778MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
9779
9780DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
9781MODULE_VERSION(t5nex, 1);
9782MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
9783
9784DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
9785MODULE_VERSION(t6nex, 1);
9786MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
9787#ifdef DEV_NETMAP
9788MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
9789#endif /* DEV_NETMAP */
9790
9791DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
9792MODULE_VERSION(cxgbe, 1);
9793
9794DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
9795MODULE_VERSION(cxl, 1);
9796
9797DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
9798MODULE_VERSION(cc, 1);
9799
9800DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
9801MODULE_VERSION(vcxgbe, 1);
9802
9803DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
9804MODULE_VERSION(vcxl, 1);
9805
9806DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
9807MODULE_VERSION(vcc, 1);
9808