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 353418 2019-10-10 23:27:02Z 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 -n as an indication to tweak_tunables() that it should
225 * provide a reasonable default (upto n) 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 = -NTXQ_10G;
236TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq10g);
237
238#define NRXQ_10G 8
239int t4_nrxq10g = -NRXQ_10G;
240TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq10g);
241
242#define NTXQ_1G 4
243int t4_ntxq1g = -NTXQ_1G;
244TUNABLE_INT("hw.cxgbe.ntxq1g", &t4_ntxq1g);
245
246#define NRXQ_1G 2
247int t4_nrxq1g = -NRXQ_1G;
248TUNABLE_INT("hw.cxgbe.nrxq1g", &t4_nrxq1g);
249
250#define NTXQ_VI 1
251static int t4_ntxq_vi = -NTXQ_VI;
252TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
253
254#define NRXQ_VI 1
255static int t4_nrxq_vi = -NRXQ_VI;
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 = -NOFLDTXQ_10G;
264TUNABLE_INT("hw.cxgbe.nofldtxq10g", &t4_nofldtxq10g);
265
266#define NOFLDRXQ_10G 2
267static int t4_nofldrxq10g = -NOFLDRXQ_10G;
268TUNABLE_INT("hw.cxgbe.nofldrxq10g", &t4_nofldrxq10g);
269
270#define NOFLDTXQ_1G 2
271static int t4_nofldtxq1g = -NOFLDTXQ_1G;
272TUNABLE_INT("hw.cxgbe.nofldtxq1g", &t4_nofldtxq1g);
273
274#define NOFLDRXQ_1G 1
275static int t4_nofldrxq1g = -NOFLDRXQ_1G;
276TUNABLE_INT("hw.cxgbe.nofldrxq1g", &t4_nofldrxq1g);
277
278#define NOFLDTXQ_VI 1
279static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
280TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
281
282#define NOFLDRXQ_VI 1
283static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
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 = -NNMTXQ_VI;
290TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
291
292#define NNMRXQ_VI 2
293static int t4_nnmrxq_vi = -NNMRXQ_VI;
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 void init_l1cfg(struct port_info *);
480static int apply_l1cfg(struct port_info *);
481static int cxgbe_init_synchronized(struct vi_info *);
482static int cxgbe_uninit_synchronized(struct vi_info *);
483static void quiesce_txq(struct adapter *, struct sge_txq *);
484static void quiesce_wrq(struct adapter *, struct sge_wrq *);
485static void quiesce_iq(struct adapter *, struct sge_iq *);
486static void quiesce_fl(struct adapter *, struct sge_fl *);
487static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
488    driver_intr_t *, void *, char *);
489static int t4_free_irq(struct adapter *, struct irq *);
490static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
491static void vi_refresh_stats(struct adapter *, struct vi_info *);
492static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
493static void cxgbe_tick(void *);
494static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
495static void cxgbe_sysctls(struct port_info *);
496static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
497static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
498static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
499static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
500static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
501static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
502static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
503static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
504static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
505static int sysctl_fec(SYSCTL_HANDLER_ARGS);
506static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
507static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
508static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
509#ifdef SBUF_DRAIN
510static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
511static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
512static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
513static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
514static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
515static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
516static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
517static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
518static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
519static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
520static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
521static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
522static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
523static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
524static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
525static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
526static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
527static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
528static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
529static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
530static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
531static int sysctl_tids(SYSCTL_HANDLER_ARGS);
532static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
533static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
534static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
535static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
536static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
537static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
538static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
539#endif
540#ifdef TCP_OFFLOAD
541static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
542static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
543static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
544#endif
545static uint32_t fconf_iconf_to_mode(uint32_t, uint32_t);
546static uint32_t mode_to_fconf(uint32_t);
547static uint32_t mode_to_iconf(uint32_t);
548static int check_fspec_against_fconf_iconf(struct adapter *,
549    struct t4_filter_specification *);
550static int get_filter_mode(struct adapter *, uint32_t *);
551static int set_filter_mode(struct adapter *, uint32_t);
552static inline uint64_t get_filter_hits(struct adapter *, uint32_t);
553static int get_filter(struct adapter *, struct t4_filter *);
554static int set_filter(struct adapter *, struct t4_filter *);
555static int del_filter(struct adapter *, struct t4_filter *);
556static void clear_filter(struct filter_entry *);
557static int set_filter_wr(struct adapter *, int);
558static int del_filter_wr(struct adapter *, int);
559static int set_tcb_rpl(struct sge_iq *, const struct rss_header *,
560    struct mbuf *);
561static int get_sge_context(struct adapter *, struct t4_sge_context *);
562static int load_fw(struct adapter *, struct t4_data *);
563static int load_cfg(struct adapter *, struct t4_data *);
564static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
565static int read_i2c(struct adapter *, struct t4_i2c_data *);
566#ifdef TCP_OFFLOAD
567static int toe_capability(struct vi_info *, int);
568#endif
569static int mod_event(module_t, int, void *);
570
571struct {
572	uint16_t device;
573	char *desc;
574} t4_pciids[] = {
575	{0xa000, "Chelsio Terminator 4 FPGA"},
576	{0x4400, "Chelsio T440-dbg"},
577	{0x4401, "Chelsio T420-CR"},
578	{0x4402, "Chelsio T422-CR"},
579	{0x4403, "Chelsio T440-CR"},
580	{0x4404, "Chelsio T420-BCH"},
581	{0x4405, "Chelsio T440-BCH"},
582	{0x4406, "Chelsio T440-CH"},
583	{0x4407, "Chelsio T420-SO"},
584	{0x4408, "Chelsio T420-CX"},
585	{0x4409, "Chelsio T420-BT"},
586	{0x440a, "Chelsio T404-BT"},
587	{0x440e, "Chelsio T440-LP-CR"},
588}, t5_pciids[] = {
589	{0xb000, "Chelsio Terminator 5 FPGA"},
590	{0x5400, "Chelsio T580-dbg"},
591	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
592	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
593	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
594	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
595	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
596	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
597	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
598	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
599	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
600	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
601	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
602	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
603	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
604	{0x5418,  "Chelsio T540-BT"},		/* 4 x 10GBaseT */
605	{0x5419,  "Chelsio T540-LP-BT"},	/* 4 x 10GBaseT */
606	{0x541a,  "Chelsio T540-SO-BT"},	/* 4 x 10GBaseT, nomem */
607	{0x541b,  "Chelsio T540-SO-CR"},	/* 4 x 10G, nomem */
608
609	/* Custom */
610	{0x5483, "Custom T540-CR"},
611	{0x5484, "Custom T540-BT"},
612}, t6_pciids[] = {
613	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
614	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
615	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
616	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
617	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
618	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
619	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
620	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
621	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
622	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
623	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
624	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
625	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
626	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
627	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
628	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
629
630	/* Custom */
631	{0x6480, "Custom T6225-CR"},
632	{0x6481, "Custom T62100-CR"},
633	{0x6482, "Custom T6225-CR"},
634	{0x6483, "Custom T62100-CR"},
635	{0x6484, "Custom T64100-CR"},
636	{0x6485, "Custom T6240-SO"},
637	{0x6486, "Custom T6225-SO-CR"},
638	{0x6487, "Custom T6225-CR"},
639};
640
641#ifdef TCP_OFFLOAD
642/*
643 * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
644 * exactly the same for both rxq and ofld_rxq.
645 */
646CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
647CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
648#endif
649CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
650
651static int
652t4_probe(device_t dev)
653{
654	int i;
655	uint16_t v = pci_get_vendor(dev);
656	uint16_t d = pci_get_device(dev);
657	uint8_t f = pci_get_function(dev);
658
659	if (v != PCI_VENDOR_ID_CHELSIO)
660		return (ENXIO);
661
662	/* Attach only to PF0 of the FPGA */
663	if (d == 0xa000 && f != 0)
664		return (ENXIO);
665
666	for (i = 0; i < nitems(t4_pciids); i++) {
667		if (d == t4_pciids[i].device) {
668			device_set_desc(dev, t4_pciids[i].desc);
669			return (BUS_PROBE_DEFAULT);
670		}
671	}
672
673	return (ENXIO);
674}
675
676static int
677t5_probe(device_t dev)
678{
679	int i;
680	uint16_t v = pci_get_vendor(dev);
681	uint16_t d = pci_get_device(dev);
682	uint8_t f = pci_get_function(dev);
683
684	if (v != PCI_VENDOR_ID_CHELSIO)
685		return (ENXIO);
686
687	/* Attach only to PF0 of the FPGA */
688	if (d == 0xb000 && f != 0)
689		return (ENXIO);
690
691	for (i = 0; i < nitems(t5_pciids); i++) {
692		if (d == t5_pciids[i].device) {
693			device_set_desc(dev, t5_pciids[i].desc);
694			return (BUS_PROBE_DEFAULT);
695		}
696	}
697
698	return (ENXIO);
699}
700
701static int
702t6_probe(device_t dev)
703{
704	int i;
705	uint16_t v = pci_get_vendor(dev);
706	uint16_t d = pci_get_device(dev);
707
708	if (v != PCI_VENDOR_ID_CHELSIO)
709		return (ENXIO);
710
711	for (i = 0; i < nitems(t6_pciids); i++) {
712		if (d == t6_pciids[i].device) {
713			device_set_desc(dev, t6_pciids[i].desc);
714			return (BUS_PROBE_DEFAULT);
715		}
716	}
717
718	return (ENXIO);
719}
720
721static void
722t5_attribute_workaround(device_t dev)
723{
724	device_t root_port;
725	uint32_t v;
726
727	/*
728	 * The T5 chips do not properly echo the No Snoop and Relaxed
729	 * Ordering attributes when replying to a TLP from a Root
730	 * Port.  As a workaround, find the parent Root Port and
731	 * disable No Snoop and Relaxed Ordering.  Note that this
732	 * affects all devices under this root port.
733	 */
734	root_port = pci_find_pcie_root_port(dev);
735	if (root_port == NULL) {
736		device_printf(dev, "Unable to find parent root port\n");
737		return;
738	}
739
740	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
741	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
742	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
743	    0)
744		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
745		    device_get_nameunit(root_port));
746}
747
748static const struct devnames devnames[] = {
749	{
750		.nexus_name = "t4nex",
751		.ifnet_name = "cxgbe",
752		.vi_ifnet_name = "vcxgbe",
753		.pf03_drv_name = "t4iov",
754		.vf_nexus_name = "t4vf",
755		.vf_ifnet_name = "cxgbev"
756	}, {
757		.nexus_name = "t5nex",
758		.ifnet_name = "cxl",
759		.vi_ifnet_name = "vcxl",
760		.pf03_drv_name = "t5iov",
761		.vf_nexus_name = "t5vf",
762		.vf_ifnet_name = "cxlv"
763	}, {
764		.nexus_name = "t6nex",
765		.ifnet_name = "cc",
766		.vi_ifnet_name = "vcc",
767		.pf03_drv_name = "t6iov",
768		.vf_nexus_name = "t6vf",
769		.vf_ifnet_name = "ccv"
770	}
771};
772
773void
774t4_init_devnames(struct adapter *sc)
775{
776	int id;
777
778	id = chip_id(sc);
779	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
780		sc->names = &devnames[id - CHELSIO_T4];
781	else {
782		device_printf(sc->dev, "chip id %d is not supported.\n", id);
783		sc->names = NULL;
784	}
785}
786
787static int
788t4_attach(device_t dev)
789{
790	struct adapter *sc;
791	int rc = 0, i, j, n10g, n1g, rqidx, tqidx;
792	struct make_dev_args mda;
793	struct intrs_and_queues iaq;
794	struct sge *s;
795	uint8_t *buf;
796#ifdef TCP_OFFLOAD
797	int ofld_rqidx, ofld_tqidx;
798#endif
799#ifdef DEV_NETMAP
800	int nm_rqidx, nm_tqidx;
801#endif
802	int num_vis;
803
804	sc = device_get_softc(dev);
805	sc->dev = dev;
806	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
807
808	if ((pci_get_device(dev) & 0xff00) == 0x5400)
809		t5_attribute_workaround(dev);
810	pci_enable_busmaster(dev);
811	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
812		uint32_t v;
813
814		pci_set_max_read_req(dev, 4096);
815		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
816		v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
817		pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
818
819		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
820	}
821
822	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
823	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
824	sc->traceq = -1;
825	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
826	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
827	    device_get_nameunit(dev));
828
829	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
830	    device_get_nameunit(dev));
831	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
832	t4_add_adapter(sc);
833
834	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
835	TAILQ_INIT(&sc->sfl);
836	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
837
838	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
839
840	rc = t4_map_bars_0_and_4(sc);
841	if (rc != 0)
842		goto done; /* error message displayed already */
843
844	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
845
846	/* Prepare the adapter for operation. */
847	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
848	rc = -t4_prep_adapter(sc, buf);
849	free(buf, M_CXGBE);
850	if (rc != 0) {
851		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
852		goto done;
853	}
854
855	/*
856	 * This is the real PF# to which we're attaching.  Works from within PCI
857	 * passthrough environments too, where pci_get_function() could return a
858	 * different PF# depending on the passthrough configuration.  We need to
859	 * use the real PF# in all our communication with the firmware.
860	 */
861	j = t4_read_reg(sc, A_PL_WHOAMI);
862	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
863	sc->mbox = sc->pf;
864
865	t4_init_devnames(sc);
866	if (sc->names == NULL) {
867		rc = ENOTSUP;
868		goto done; /* error message displayed already */
869	}
870
871	/*
872	 * Do this really early, with the memory windows set up even before the
873	 * character device.  The userland tool's register i/o and mem read
874	 * will work even in "recovery mode".
875	 */
876	setup_memwin(sc);
877	if (t4_init_devlog_params(sc, 0) == 0)
878		fixup_devlog_params(sc);
879	make_dev_args_init(&mda);
880	mda.mda_devsw = &t4_cdevsw;
881	mda.mda_uid = UID_ROOT;
882	mda.mda_gid = GID_WHEEL;
883	mda.mda_mode = 0600;
884	mda.mda_si_drv1 = sc;
885	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
886	if (rc != 0)
887		device_printf(dev, "failed to create nexus char device: %d.\n",
888		    rc);
889
890	/* Go no further if recovery mode has been requested. */
891	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
892		device_printf(dev, "recovery mode.\n");
893		goto done;
894	}
895
896#if defined(__i386__)
897	if ((cpu_feature & CPUID_CX8) == 0) {
898		device_printf(dev, "64 bit atomics not available.\n");
899		rc = ENOTSUP;
900		goto done;
901	}
902#endif
903
904	/* Prepare the firmware for operation */
905	rc = prep_firmware(sc);
906	if (rc != 0)
907		goto done; /* error message displayed already */
908
909	rc = get_params__post_init(sc);
910	if (rc != 0)
911		goto done; /* error message displayed already */
912
913	rc = set_params__post_init(sc);
914	if (rc != 0)
915		goto done; /* error message displayed already */
916
917	rc = t4_map_bar_2(sc);
918	if (rc != 0)
919		goto done; /* error message displayed already */
920
921	rc = t4_create_dma_tag(sc);
922	if (rc != 0)
923		goto done; /* error message displayed already */
924
925	/*
926	 * Number of VIs to create per-port.  The first VI is the "main" regular
927	 * VI for the port.  The rest are additional virtual interfaces on the
928	 * same physical port.  Note that the main VI does not have native
929	 * netmap support but the extra VIs do.
930	 *
931	 * Limit the number of VIs per port to the number of available
932	 * MAC addresses per port.
933	 */
934	if (t4_num_vis >= 1)
935		num_vis = t4_num_vis;
936	else
937		num_vis = 1;
938	if (num_vis > nitems(vi_mac_funcs)) {
939		num_vis = nitems(vi_mac_funcs);
940		device_printf(dev, "Number of VIs limited to %d\n", num_vis);
941	}
942
943	/*
944	 * First pass over all the ports - allocate VIs and initialize some
945	 * basic parameters like mac address, port type, etc.  We also figure
946	 * out whether a port is 10G or 1G and use that information when
947	 * calculating how many interrupts to attempt to allocate.
948	 */
949	n10g = n1g = 0;
950	for_each_port(sc, i) {
951		struct port_info *pi;
952
953		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
954		sc->port[i] = pi;
955
956		/* These must be set before t4_port_init */
957		pi->adapter = sc;
958		pi->port_id = i;
959		/*
960		 * XXX: vi[0] is special so we can't delay this allocation until
961		 * pi->nvi's final value is known.
962		 */
963		pi->vi = malloc(sizeof(struct vi_info) * num_vis, M_CXGBE,
964		    M_ZERO | M_WAITOK);
965
966		/*
967		 * Allocate the "main" VI and initialize parameters
968		 * like mac addr.
969		 */
970		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
971		if (rc != 0) {
972			device_printf(dev, "unable to initialize port %d: %d\n",
973			    i, rc);
974			free(pi->vi, M_CXGBE);
975			free(pi, M_CXGBE);
976			sc->port[i] = NULL;
977			goto done;
978		}
979
980		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
981		    device_get_nameunit(dev), i);
982		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
983		sc->chan_map[pi->tx_chan] = i;
984
985		if (port_top_speed(pi) >= 10) {
986			n10g++;
987		} else {
988			n1g++;
989		}
990
991		/* All VIs on this port share this media. */
992		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
993		    cxgbe_media_status);
994
995		pi->dev = device_add_child(dev, sc->names->ifnet_name, -1);
996		if (pi->dev == NULL) {
997			device_printf(dev,
998			    "failed to add device for port %d.\n", i);
999			rc = ENXIO;
1000			goto done;
1001		}
1002		pi->vi[0].dev = pi->dev;
1003		device_set_softc(pi->dev, pi);
1004	}
1005
1006	/*
1007	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1008	 */
1009	rc = cfg_itype_and_nqueues(sc, n10g, n1g, num_vis, &iaq);
1010	if (rc != 0)
1011		goto done; /* error message displayed already */
1012	if (iaq.nrxq_vi + iaq.nofldrxq_vi + iaq.nnmrxq_vi == 0)
1013		num_vis = 1;
1014
1015	sc->intr_type = iaq.intr_type;
1016	sc->intr_count = iaq.nirq;
1017
1018	s = &sc->sge;
1019	s->nrxq = n10g * iaq.nrxq10g + n1g * iaq.nrxq1g;
1020	s->ntxq = n10g * iaq.ntxq10g + n1g * iaq.ntxq1g;
1021	if (num_vis > 1) {
1022		s->nrxq += (n10g + n1g) * (num_vis - 1) * iaq.nrxq_vi;
1023		s->ntxq += (n10g + n1g) * (num_vis - 1) * iaq.ntxq_vi;
1024	}
1025	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1026	s->neq += sc->params.nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
1027	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1028#ifdef TCP_OFFLOAD
1029	if (is_offload(sc)) {
1030		s->nofldrxq = n10g * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
1031		s->nofldtxq = n10g * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
1032		if (num_vis > 1) {
1033			s->nofldrxq += (n10g + n1g) * (num_vis - 1) *
1034			    iaq.nofldrxq_vi;
1035			s->nofldtxq += (n10g + n1g) * (num_vis - 1) *
1036			    iaq.nofldtxq_vi;
1037		}
1038		s->neq += s->nofldtxq + s->nofldrxq;
1039		s->niq += s->nofldrxq;
1040
1041		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1042		    M_CXGBE, M_ZERO | M_WAITOK);
1043		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1044		    M_CXGBE, M_ZERO | M_WAITOK);
1045	}
1046#endif
1047#ifdef DEV_NETMAP
1048	if (num_vis > 1) {
1049		s->nnmrxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmrxq_vi;
1050		s->nnmtxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmtxq_vi;
1051	}
1052	s->neq += s->nnmtxq + s->nnmrxq;
1053	s->niq += s->nnmrxq;
1054
1055	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1056	    M_CXGBE, M_ZERO | M_WAITOK);
1057	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1058	    M_CXGBE, M_ZERO | M_WAITOK);
1059#endif
1060
1061	s->ctrlq = malloc(sc->params.nports * sizeof(struct sge_wrq), M_CXGBE,
1062	    M_ZERO | M_WAITOK);
1063	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1064	    M_ZERO | M_WAITOK);
1065	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1066	    M_ZERO | M_WAITOK);
1067	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1068	    M_ZERO | M_WAITOK);
1069	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1070	    M_ZERO | M_WAITOK);
1071
1072	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1073	    M_ZERO | M_WAITOK);
1074
1075	t4_init_l2t(sc, M_WAITOK);
1076	t4_init_tx_sched(sc);
1077
1078	/*
1079	 * Second pass over the ports.  This time we know the number of rx and
1080	 * tx queues that each port should get.
1081	 */
1082	rqidx = tqidx = 0;
1083#ifdef TCP_OFFLOAD
1084	ofld_rqidx = ofld_tqidx = 0;
1085#endif
1086#ifdef DEV_NETMAP
1087	nm_rqidx = nm_tqidx = 0;
1088#endif
1089	for_each_port(sc, i) {
1090		struct port_info *pi = sc->port[i];
1091		struct vi_info *vi;
1092
1093		if (pi == NULL)
1094			continue;
1095
1096		pi->nvi = num_vis;
1097		for_each_vi(pi, j, vi) {
1098			vi->pi = pi;
1099			vi->qsize_rxq = t4_qsize_rxq;
1100			vi->qsize_txq = t4_qsize_txq;
1101
1102			vi->first_rxq = rqidx;
1103			vi->first_txq = tqidx;
1104			if (port_top_speed(pi) >= 10) {
1105				vi->tmr_idx = t4_tmr_idx_10g;
1106				vi->pktc_idx = t4_pktc_idx_10g;
1107				vi->flags |= iaq.intr_flags_10g & INTR_RXQ;
1108				vi->nrxq = j == 0 ? iaq.nrxq10g : iaq.nrxq_vi;
1109				vi->ntxq = j == 0 ? iaq.ntxq10g : iaq.ntxq_vi;
1110			} else {
1111				vi->tmr_idx = t4_tmr_idx_1g;
1112				vi->pktc_idx = t4_pktc_idx_1g;
1113				vi->flags |= iaq.intr_flags_1g & INTR_RXQ;
1114				vi->nrxq = j == 0 ? iaq.nrxq1g : iaq.nrxq_vi;
1115				vi->ntxq = j == 0 ? iaq.ntxq1g : iaq.ntxq_vi;
1116			}
1117			rqidx += vi->nrxq;
1118			tqidx += vi->ntxq;
1119
1120			if (j == 0 && vi->ntxq > 1)
1121				vi->rsrv_noflowq = iaq.rsrv_noflowq ? 1 : 0;
1122			else
1123				vi->rsrv_noflowq = 0;
1124
1125#ifdef TCP_OFFLOAD
1126			vi->first_ofld_rxq = ofld_rqidx;
1127			vi->first_ofld_txq = ofld_tqidx;
1128			if (port_top_speed(pi) >= 10) {
1129				vi->flags |= iaq.intr_flags_10g & INTR_OFLD_RXQ;
1130				vi->nofldrxq = j == 0 ? iaq.nofldrxq10g :
1131				    iaq.nofldrxq_vi;
1132				vi->nofldtxq = j == 0 ? iaq.nofldtxq10g :
1133				    iaq.nofldtxq_vi;
1134			} else {
1135				vi->flags |= iaq.intr_flags_1g & INTR_OFLD_RXQ;
1136				vi->nofldrxq = j == 0 ? iaq.nofldrxq1g :
1137				    iaq.nofldrxq_vi;
1138				vi->nofldtxq = j == 0 ? iaq.nofldtxq1g :
1139				    iaq.nofldtxq_vi;
1140			}
1141			ofld_rqidx += vi->nofldrxq;
1142			ofld_tqidx += vi->nofldtxq;
1143#endif
1144#ifdef DEV_NETMAP
1145			if (j > 0) {
1146				vi->first_nm_rxq = nm_rqidx;
1147				vi->first_nm_txq = nm_tqidx;
1148				vi->nnmrxq = iaq.nnmrxq_vi;
1149				vi->nnmtxq = iaq.nnmtxq_vi;
1150				nm_rqidx += vi->nnmrxq;
1151				nm_tqidx += vi->nnmtxq;
1152			}
1153#endif
1154		}
1155	}
1156
1157	rc = t4_setup_intr_handlers(sc);
1158	if (rc != 0) {
1159		device_printf(dev,
1160		    "failed to setup interrupt handlers: %d\n", rc);
1161		goto done;
1162	}
1163
1164	rc = bus_generic_probe(dev);
1165	if (rc != 0) {
1166		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1167		goto done;
1168	}
1169
1170	rc = bus_generic_attach(dev);
1171	if (rc != 0) {
1172		device_printf(dev,
1173		    "failed to attach all child ports: %d\n", rc);
1174		goto done;
1175	}
1176
1177	device_printf(dev,
1178	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1179	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1180	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1181	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1182	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1183
1184	t4_set_desc(sc);
1185
1186done:
1187	if (rc != 0 && sc->cdev) {
1188		/* cdev was created and so cxgbetool works; recover that way. */
1189		device_printf(dev,
1190		    "error during attach, adapter is now in recovery mode.\n");
1191		rc = 0;
1192	}
1193
1194	if (rc != 0)
1195		t4_detach_common(dev);
1196	else
1197		t4_sysctls(sc);
1198
1199	return (rc);
1200}
1201
1202/*
1203 * Idempotent
1204 */
1205static int
1206t4_detach(device_t dev)
1207{
1208	struct adapter *sc;
1209
1210	sc = device_get_softc(dev);
1211
1212	return (t4_detach_common(dev));
1213}
1214
1215int
1216t4_detach_common(device_t dev)
1217{
1218	struct adapter *sc;
1219	struct port_info *pi;
1220	int i, rc;
1221
1222	sc = device_get_softc(dev);
1223
1224	if (sc->flags & FULL_INIT_DONE) {
1225		if (!(sc->flags & IS_VF))
1226			t4_intr_disable(sc);
1227	}
1228
1229	if (sc->cdev) {
1230		destroy_dev(sc->cdev);
1231		sc->cdev = NULL;
1232	}
1233
1234	if (device_is_attached(dev)) {
1235		rc = bus_generic_detach(dev);
1236		if (rc) {
1237			device_printf(dev,
1238			    "failed to detach child devices: %d\n", rc);
1239			return (rc);
1240		}
1241	}
1242
1243	for (i = 0; i < sc->intr_count; i++)
1244		t4_free_irq(sc, &sc->irq[i]);
1245
1246	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1247		t4_free_tx_sched(sc);
1248
1249	for (i = 0; i < MAX_NPORTS; i++) {
1250		pi = sc->port[i];
1251		if (pi) {
1252			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1253			if (pi->dev)
1254				device_delete_child(dev, pi->dev);
1255
1256			mtx_destroy(&pi->pi_lock);
1257			free(pi->vi, M_CXGBE);
1258			free(pi, M_CXGBE);
1259		}
1260	}
1261
1262	device_delete_children(dev);
1263
1264	if (sc->flags & FULL_INIT_DONE)
1265		adapter_full_uninit(sc);
1266
1267	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1268		t4_fw_bye(sc, sc->mbox);
1269
1270	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1271		pci_release_msi(dev);
1272
1273	if (sc->regs_res)
1274		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1275		    sc->regs_res);
1276
1277	if (sc->udbs_res)
1278		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1279		    sc->udbs_res);
1280
1281	if (sc->msix_res)
1282		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1283		    sc->msix_res);
1284
1285	if (sc->l2t)
1286		t4_free_l2t(sc->l2t);
1287
1288#ifdef TCP_OFFLOAD
1289	free(sc->sge.ofld_rxq, M_CXGBE);
1290	free(sc->sge.ofld_txq, M_CXGBE);
1291#endif
1292#ifdef DEV_NETMAP
1293	free(sc->sge.nm_rxq, M_CXGBE);
1294	free(sc->sge.nm_txq, M_CXGBE);
1295#endif
1296	free(sc->irq, M_CXGBE);
1297	free(sc->sge.rxq, M_CXGBE);
1298	free(sc->sge.txq, M_CXGBE);
1299	free(sc->sge.ctrlq, M_CXGBE);
1300	free(sc->sge.iqmap, M_CXGBE);
1301	free(sc->sge.eqmap, M_CXGBE);
1302	free(sc->tids.ftid_tab, M_CXGBE);
1303	t4_destroy_dma_tag(sc);
1304	if (mtx_initialized(&sc->sc_lock)) {
1305		sx_xlock(&t4_list_lock);
1306		SLIST_REMOVE(&t4_list, sc, adapter, link);
1307		sx_xunlock(&t4_list_lock);
1308		mtx_destroy(&sc->sc_lock);
1309	}
1310
1311	callout_drain(&sc->sfl_callout);
1312	if (mtx_initialized(&sc->tids.ftid_lock))
1313		mtx_destroy(&sc->tids.ftid_lock);
1314	if (mtx_initialized(&sc->sfl_lock))
1315		mtx_destroy(&sc->sfl_lock);
1316	if (mtx_initialized(&sc->ifp_lock))
1317		mtx_destroy(&sc->ifp_lock);
1318	if (mtx_initialized(&sc->reg_lock))
1319		mtx_destroy(&sc->reg_lock);
1320
1321	for (i = 0; i < NUM_MEMWIN; i++) {
1322		struct memwin *mw = &sc->memwin[i];
1323
1324		if (rw_initialized(&mw->mw_lock))
1325			rw_destroy(&mw->mw_lock);
1326	}
1327
1328	bzero(sc, sizeof(*sc));
1329
1330	return (0);
1331}
1332
1333static int
1334cxgbe_probe(device_t dev)
1335{
1336	char buf[128];
1337	struct port_info *pi = device_get_softc(dev);
1338
1339	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1340	device_set_desc_copy(dev, buf);
1341
1342	return (BUS_PROBE_DEFAULT);
1343}
1344
1345#define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1346    IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1347    IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1348#define T4_CAP_ENABLE (T4_CAP)
1349
1350static int
1351cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1352{
1353	struct ifnet *ifp;
1354	struct sbuf *sb;
1355
1356	vi->xact_addr_filt = -1;
1357	callout_init(&vi->tick, 1);
1358
1359	/* Allocate an ifnet and set it up */
1360	ifp = if_alloc(IFT_ETHER);
1361	if (ifp == NULL) {
1362		device_printf(dev, "Cannot allocate ifnet\n");
1363		return (ENOMEM);
1364	}
1365	vi->ifp = ifp;
1366	ifp->if_softc = vi;
1367
1368	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1369	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1370
1371	ifp->if_init = cxgbe_init;
1372	ifp->if_ioctl = cxgbe_ioctl;
1373	ifp->if_transmit = cxgbe_transmit;
1374	ifp->if_qflush = cxgbe_qflush;
1375
1376	ifp->if_capabilities = T4_CAP;
1377#ifdef TCP_OFFLOAD
1378	if (vi->nofldrxq != 0)
1379		ifp->if_capabilities |= IFCAP_TOE;
1380#endif
1381#ifdef DEV_NETMAP
1382	if (vi->nnmrxq != 0)
1383		ifp->if_capabilities |= IFCAP_NETMAP;
1384#endif
1385	ifp->if_capenable = T4_CAP_ENABLE;
1386	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1387	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1388
1389	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1390	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1391	ifp->if_hw_tsomaxsegsize = 65536;
1392
1393	vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1394	    EVENTHANDLER_PRI_ANY);
1395
1396	ether_ifattach(ifp, vi->hw_addr);
1397#ifdef DEV_NETMAP
1398	if (ifp->if_capabilities & IFCAP_NETMAP)
1399		cxgbe_nm_attach(vi);
1400#endif
1401	sb = sbuf_new_auto();
1402	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1403#ifdef TCP_OFFLOAD
1404	if (ifp->if_capabilities & IFCAP_TOE)
1405		sbuf_printf(sb, "; %d txq, %d rxq (TOE)",
1406		    vi->nofldtxq, vi->nofldrxq);
1407#endif
1408#ifdef DEV_NETMAP
1409	if (ifp->if_capabilities & IFCAP_NETMAP)
1410		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1411		    vi->nnmtxq, vi->nnmrxq);
1412#endif
1413	sbuf_finish(sb);
1414	device_printf(dev, "%s\n", sbuf_data(sb));
1415	sbuf_delete(sb);
1416
1417	vi_sysctls(vi);
1418
1419	return (0);
1420}
1421
1422static int
1423cxgbe_attach(device_t dev)
1424{
1425	struct port_info *pi = device_get_softc(dev);
1426	struct adapter *sc = pi->adapter;
1427	struct vi_info *vi;
1428	int i, rc;
1429
1430	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1431
1432	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1433	if (rc)
1434		return (rc);
1435
1436	for_each_vi(pi, i, vi) {
1437		if (i == 0)
1438			continue;
1439		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1440		if (vi->dev == NULL) {
1441			device_printf(dev, "failed to add VI %d\n", i);
1442			continue;
1443		}
1444		device_set_softc(vi->dev, vi);
1445	}
1446
1447	cxgbe_sysctls(pi);
1448
1449	bus_generic_attach(dev);
1450
1451	return (0);
1452}
1453
1454static void
1455cxgbe_vi_detach(struct vi_info *vi)
1456{
1457	struct ifnet *ifp = vi->ifp;
1458
1459	ether_ifdetach(ifp);
1460
1461	if (vi->vlan_c)
1462		EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1463
1464	/* Let detach proceed even if these fail. */
1465#ifdef DEV_NETMAP
1466	if (ifp->if_capabilities & IFCAP_NETMAP)
1467		cxgbe_nm_detach(vi);
1468#endif
1469	cxgbe_uninit_synchronized(vi);
1470	callout_drain(&vi->tick);
1471	vi_full_uninit(vi);
1472
1473	if_free(vi->ifp);
1474	vi->ifp = NULL;
1475}
1476
1477static int
1478cxgbe_detach(device_t dev)
1479{
1480	struct port_info *pi = device_get_softc(dev);
1481	struct adapter *sc = pi->adapter;
1482	int rc;
1483
1484	/* Detach the extra VIs first. */
1485	rc = bus_generic_detach(dev);
1486	if (rc)
1487		return (rc);
1488	device_delete_children(dev);
1489
1490	doom_vi(sc, &pi->vi[0]);
1491
1492	if (pi->flags & HAS_TRACEQ) {
1493		sc->traceq = -1;	/* cloner should not create ifnet */
1494		t4_tracer_port_detach(sc);
1495	}
1496
1497	cxgbe_vi_detach(&pi->vi[0]);
1498	callout_drain(&pi->tick);
1499	ifmedia_removeall(&pi->media);
1500
1501	end_synchronized_op(sc, 0);
1502
1503	return (0);
1504}
1505
1506static void
1507cxgbe_init(void *arg)
1508{
1509	struct vi_info *vi = arg;
1510	struct adapter *sc = vi->pi->adapter;
1511
1512	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1513		return;
1514	cxgbe_init_synchronized(vi);
1515	end_synchronized_op(sc, 0);
1516}
1517
1518static int
1519cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1520{
1521	int rc = 0, mtu, flags, can_sleep;
1522	struct vi_info *vi = ifp->if_softc;
1523	struct port_info *pi = vi->pi;
1524	struct adapter *sc = pi->adapter;
1525	struct ifreq *ifr = (struct ifreq *)data;
1526	uint32_t mask;
1527
1528	switch (cmd) {
1529	case SIOCSIFMTU:
1530		mtu = ifr->ifr_mtu;
1531		if (mtu < ETHERMIN || mtu > MAX_MTU)
1532			return (EINVAL);
1533
1534		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1535		if (rc)
1536			return (rc);
1537		ifp->if_mtu = mtu;
1538		if (vi->flags & VI_INIT_DONE) {
1539			t4_update_fl_bufsize(ifp);
1540			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1541				rc = update_mac_settings(ifp, XGMAC_MTU);
1542		}
1543		end_synchronized_op(sc, 0);
1544		break;
1545
1546	case SIOCSIFFLAGS:
1547		can_sleep = 0;
1548redo_sifflags:
1549		rc = begin_synchronized_op(sc, vi,
1550		    can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1551		if (rc)
1552			return (rc);
1553
1554		if (ifp->if_flags & IFF_UP) {
1555			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1556				flags = vi->if_flags;
1557				if ((ifp->if_flags ^ flags) &
1558				    (IFF_PROMISC | IFF_ALLMULTI)) {
1559					if (can_sleep == 1) {
1560						end_synchronized_op(sc, 0);
1561						can_sleep = 0;
1562						goto redo_sifflags;
1563					}
1564					rc = update_mac_settings(ifp,
1565					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1566				}
1567			} else {
1568				if (can_sleep == 0) {
1569					end_synchronized_op(sc, LOCK_HELD);
1570					can_sleep = 1;
1571					goto redo_sifflags;
1572				}
1573				rc = cxgbe_init_synchronized(vi);
1574			}
1575			vi->if_flags = ifp->if_flags;
1576		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1577			if (can_sleep == 0) {
1578				end_synchronized_op(sc, LOCK_HELD);
1579				can_sleep = 1;
1580				goto redo_sifflags;
1581			}
1582			rc = cxgbe_uninit_synchronized(vi);
1583		}
1584		end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1585		break;
1586
1587	case SIOCADDMULTI:
1588	case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1589		rc = begin_synchronized_op(sc, vi, HOLD_LOCK, "t4multi");
1590		if (rc)
1591			return (rc);
1592		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1593			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1594		end_synchronized_op(sc, LOCK_HELD);
1595		break;
1596
1597	case SIOCSIFCAP:
1598		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1599		if (rc)
1600			return (rc);
1601
1602		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1603		if (mask & IFCAP_TXCSUM) {
1604			ifp->if_capenable ^= IFCAP_TXCSUM;
1605			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1606
1607			if (IFCAP_TSO4 & ifp->if_capenable &&
1608			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1609				ifp->if_capenable &= ~IFCAP_TSO4;
1610				if_printf(ifp,
1611				    "tso4 disabled due to -txcsum.\n");
1612			}
1613		}
1614		if (mask & IFCAP_TXCSUM_IPV6) {
1615			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1616			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1617
1618			if (IFCAP_TSO6 & ifp->if_capenable &&
1619			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1620				ifp->if_capenable &= ~IFCAP_TSO6;
1621				if_printf(ifp,
1622				    "tso6 disabled due to -txcsum6.\n");
1623			}
1624		}
1625		if (mask & IFCAP_RXCSUM)
1626			ifp->if_capenable ^= IFCAP_RXCSUM;
1627		if (mask & IFCAP_RXCSUM_IPV6)
1628			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1629
1630		/*
1631		 * Note that we leave CSUM_TSO alone (it is always set).  The
1632		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1633		 * sending a TSO request our way, so it's sufficient to toggle
1634		 * IFCAP_TSOx only.
1635		 */
1636		if (mask & IFCAP_TSO4) {
1637			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1638			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1639				if_printf(ifp, "enable txcsum first.\n");
1640				rc = EAGAIN;
1641				goto fail;
1642			}
1643			ifp->if_capenable ^= IFCAP_TSO4;
1644		}
1645		if (mask & IFCAP_TSO6) {
1646			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1647			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1648				if_printf(ifp, "enable txcsum6 first.\n");
1649				rc = EAGAIN;
1650				goto fail;
1651			}
1652			ifp->if_capenable ^= IFCAP_TSO6;
1653		}
1654		if (mask & IFCAP_LRO) {
1655#if defined(INET) || defined(INET6)
1656			int i;
1657			struct sge_rxq *rxq;
1658
1659			ifp->if_capenable ^= IFCAP_LRO;
1660			for_each_rxq(vi, i, rxq) {
1661				if (ifp->if_capenable & IFCAP_LRO)
1662					rxq->iq.flags |= IQ_LRO_ENABLED;
1663				else
1664					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1665			}
1666#endif
1667		}
1668#ifdef TCP_OFFLOAD
1669		if (mask & IFCAP_TOE) {
1670			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1671
1672			rc = toe_capability(vi, enable);
1673			if (rc != 0)
1674				goto fail;
1675
1676			ifp->if_capenable ^= mask;
1677		}
1678#endif
1679		if (mask & IFCAP_VLAN_HWTAGGING) {
1680			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1681			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1682				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1683		}
1684		if (mask & IFCAP_VLAN_MTU) {
1685			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1686
1687			/* Need to find out how to disable auto-mtu-inflation */
1688		}
1689		if (mask & IFCAP_VLAN_HWTSO)
1690			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1691		if (mask & IFCAP_VLAN_HWCSUM)
1692			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1693
1694#ifdef VLAN_CAPABILITIES
1695		VLAN_CAPABILITIES(ifp);
1696#endif
1697fail:
1698		end_synchronized_op(sc, 0);
1699		break;
1700
1701	case SIOCSIFMEDIA:
1702	case SIOCGIFMEDIA:
1703	case SIOCGIFXMEDIA:
1704		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1705		break;
1706
1707	case SIOCGI2C: {
1708		struct ifi2creq i2c;
1709
1710		rc = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
1711		if (rc != 0)
1712			break;
1713		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1714			rc = EPERM;
1715			break;
1716		}
1717		if (i2c.len > sizeof(i2c.data)) {
1718			rc = EINVAL;
1719			break;
1720		}
1721		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1722		if (rc)
1723			return (rc);
1724		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
1725		    i2c.offset, i2c.len, &i2c.data[0]);
1726		end_synchronized_op(sc, 0);
1727		if (rc == 0)
1728			rc = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
1729		break;
1730	}
1731
1732	default:
1733		rc = ether_ioctl(ifp, cmd, data);
1734	}
1735
1736	return (rc);
1737}
1738
1739static int
1740cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1741{
1742	struct vi_info *vi = ifp->if_softc;
1743	struct port_info *pi = vi->pi;
1744	struct adapter *sc = pi->adapter;
1745	struct sge_txq *txq;
1746	void *items[1];
1747	int rc;
1748
1749	M_ASSERTPKTHDR(m);
1750	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1751
1752	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1753		m_freem(m);
1754		return (ENETDOWN);
1755	}
1756
1757	rc = parse_pkt(sc, &m);
1758	if (__predict_false(rc != 0)) {
1759		MPASS(m == NULL);			/* was freed already */
1760		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1761		return (rc);
1762	}
1763
1764	/* Select a txq. */
1765	txq = &sc->sge.txq[vi->first_txq];
1766	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1767		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1768		    vi->rsrv_noflowq);
1769
1770	items[0] = m;
1771	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1772	if (__predict_false(rc != 0))
1773		m_freem(m);
1774
1775	return (rc);
1776}
1777
1778static void
1779cxgbe_qflush(struct ifnet *ifp)
1780{
1781	struct vi_info *vi = ifp->if_softc;
1782	struct sge_txq *txq;
1783	int i;
1784
1785	/* queues do not exist if !VI_INIT_DONE. */
1786	if (vi->flags & VI_INIT_DONE) {
1787		for_each_txq(vi, i, txq) {
1788			TXQ_LOCK(txq);
1789			txq->eq.flags |= EQ_QFLUSH;
1790			TXQ_UNLOCK(txq);
1791			while (!mp_ring_is_idle(txq->r)) {
1792				mp_ring_check_drainage(txq->r, 0);
1793				pause("qflush", 1);
1794			}
1795			TXQ_LOCK(txq);
1796			txq->eq.flags &= ~EQ_QFLUSH;
1797			TXQ_UNLOCK(txq);
1798		}
1799	}
1800	if_qflush(ifp);
1801}
1802
1803/*
1804 * The kernel picks a media from the list we had provided so we do not have to
1805 * validate the request.
1806 */
1807static int
1808cxgbe_media_change(struct ifnet *ifp)
1809{
1810	struct vi_info *vi = ifp->if_softc;
1811	struct port_info *pi = vi->pi;
1812	struct ifmedia *ifm = &pi->media;
1813	struct link_config *lc = &pi->link_cfg;
1814	struct adapter *sc = pi->adapter;
1815	int rc;
1816
1817	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mec");
1818	if (rc != 0)
1819		return (rc);
1820	PORT_LOCK(pi);
1821	if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
1822		MPASS(lc->supported & FW_PORT_CAP_ANEG);
1823		lc->requested_aneg = AUTONEG_ENABLE;
1824	} else {
1825		lc->requested_aneg = AUTONEG_DISABLE;
1826		lc->requested_speed =
1827		    ifmedia_baudrate(ifm->ifm_media) / 1000000;
1828		lc->requested_fc = 0;
1829		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
1830			lc->requested_fc |= PAUSE_RX;
1831		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
1832			lc->requested_fc |= PAUSE_TX;
1833	}
1834	if (pi->up_vis > 0)
1835		rc = apply_l1cfg(pi);
1836	PORT_UNLOCK(pi);
1837	end_synchronized_op(sc, 0);
1838	return (rc);
1839}
1840
1841/*
1842 * Mbps to FW_PORT_CAP_SPEED_* bit.
1843 */
1844static uint16_t
1845speed_to_fwspeed(int speed)
1846{
1847
1848	switch (speed) {
1849	case 100000:
1850		return (FW_PORT_CAP_SPEED_100G);
1851	case 40000:
1852		return (FW_PORT_CAP_SPEED_40G);
1853	case 25000:
1854		return (FW_PORT_CAP_SPEED_25G);
1855	case 10000:
1856		return (FW_PORT_CAP_SPEED_10G);
1857	case 1000:
1858		return (FW_PORT_CAP_SPEED_1G);
1859	case 100:
1860		return (FW_PORT_CAP_SPEED_100M);
1861	}
1862
1863	return (0);
1864}
1865
1866/*
1867 * Base media word (without ETHER, pause, link active, etc.) for the port at the
1868 * given speed.
1869 */
1870static int
1871port_mword(struct port_info *pi, uint16_t speed)
1872{
1873
1874	MPASS(speed & M_FW_PORT_CAP_SPEED);
1875	MPASS(powerof2(speed));
1876
1877	switch(pi->port_type) {
1878	case FW_PORT_TYPE_BT_SGMII:
1879	case FW_PORT_TYPE_BT_XFI:
1880	case FW_PORT_TYPE_BT_XAUI:
1881		/* BaseT */
1882		switch (speed) {
1883		case FW_PORT_CAP_SPEED_100M:
1884			return (IFM_100_T);
1885		case FW_PORT_CAP_SPEED_1G:
1886			return (IFM_1000_T);
1887		case FW_PORT_CAP_SPEED_10G:
1888			return (IFM_10G_T);
1889		}
1890		break;
1891	case FW_PORT_TYPE_KX4:
1892		if (speed == FW_PORT_CAP_SPEED_10G)
1893			return (IFM_10G_KX4);
1894		break;
1895	case FW_PORT_TYPE_CX4:
1896		if (speed == FW_PORT_CAP_SPEED_10G)
1897			return (IFM_10G_CX4);
1898		break;
1899	case FW_PORT_TYPE_KX:
1900		if (speed == FW_PORT_CAP_SPEED_1G)
1901			return (IFM_1000_KX);
1902		break;
1903	case FW_PORT_TYPE_KR:
1904	case FW_PORT_TYPE_BP_AP:
1905	case FW_PORT_TYPE_BP4_AP:
1906	case FW_PORT_TYPE_BP40_BA:
1907	case FW_PORT_TYPE_KR4_100G:
1908	case FW_PORT_TYPE_KR_SFP28:
1909	case FW_PORT_TYPE_KR_XLAUI:
1910		switch (speed) {
1911		case FW_PORT_CAP_SPEED_1G:
1912			return (IFM_1000_KX);
1913		case FW_PORT_CAP_SPEED_10G:
1914			return (IFM_10G_KR);
1915		case FW_PORT_CAP_SPEED_25G:
1916			return (IFM_25G_KR);
1917		case FW_PORT_CAP_SPEED_40G:
1918			return (IFM_40G_KR4);
1919		case FW_PORT_CAP_SPEED_100G:
1920			return (IFM_100G_KR4);
1921		}
1922		break;
1923	case FW_PORT_TYPE_FIBER_XFI:
1924	case FW_PORT_TYPE_FIBER_XAUI:
1925	case FW_PORT_TYPE_SFP:
1926	case FW_PORT_TYPE_QSFP_10G:
1927	case FW_PORT_TYPE_QSA:
1928	case FW_PORT_TYPE_QSFP:
1929	case FW_PORT_TYPE_CR4_QSFP:
1930	case FW_PORT_TYPE_CR_QSFP:
1931	case FW_PORT_TYPE_CR2_QSFP:
1932	case FW_PORT_TYPE_SFP28:
1933		/* Pluggable transceiver */
1934		switch (pi->mod_type) {
1935		case FW_PORT_MOD_TYPE_LR:
1936			switch (speed) {
1937			case FW_PORT_CAP_SPEED_1G:
1938				return (IFM_1000_LX);
1939			case FW_PORT_CAP_SPEED_10G:
1940				return (IFM_10G_LR);
1941			case FW_PORT_CAP_SPEED_25G:
1942				return (IFM_25G_LR);
1943			case FW_PORT_CAP_SPEED_40G:
1944				return (IFM_40G_LR4);
1945			case FW_PORT_CAP_SPEED_100G:
1946				return (IFM_100G_LR4);
1947			}
1948			break;
1949		case FW_PORT_MOD_TYPE_SR:
1950			switch (speed) {
1951			case FW_PORT_CAP_SPEED_1G:
1952				return (IFM_1000_SX);
1953			case FW_PORT_CAP_SPEED_10G:
1954				return (IFM_10G_SR);
1955			case FW_PORT_CAP_SPEED_25G:
1956				return (IFM_25G_SR);
1957			case FW_PORT_CAP_SPEED_40G:
1958				return (IFM_40G_SR4);
1959			case FW_PORT_CAP_SPEED_100G:
1960				return (IFM_100G_SR4);
1961			}
1962			break;
1963		case FW_PORT_MOD_TYPE_ER:
1964			if (speed == FW_PORT_CAP_SPEED_10G)
1965				return (IFM_10G_ER);
1966			break;
1967		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
1968		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
1969			switch (speed) {
1970			case FW_PORT_CAP_SPEED_1G:
1971				return (IFM_1000_CX);
1972			case FW_PORT_CAP_SPEED_10G:
1973				return (IFM_10G_TWINAX);
1974			case FW_PORT_CAP_SPEED_25G:
1975				return (IFM_25G_CR);
1976			case FW_PORT_CAP_SPEED_40G:
1977				return (IFM_40G_CR4);
1978			case FW_PORT_CAP_SPEED_100G:
1979				return (IFM_100G_CR4);
1980			}
1981			break;
1982		case FW_PORT_MOD_TYPE_LRM:
1983			if (speed == FW_PORT_CAP_SPEED_10G)
1984				return (IFM_10G_LRM);
1985			break;
1986		case FW_PORT_MOD_TYPE_NA:
1987			MPASS(0);	/* Not pluggable? */
1988			/* fall throough */
1989		case FW_PORT_MOD_TYPE_ERROR:
1990		case FW_PORT_MOD_TYPE_UNKNOWN:
1991		case FW_PORT_MOD_TYPE_NOTSUPPORTED:
1992			break;
1993		case FW_PORT_MOD_TYPE_NONE:
1994			return (IFM_NONE);
1995		}
1996		break;
1997	case FW_PORT_TYPE_NONE:
1998		return (IFM_NONE);
1999	}
2000
2001	return (IFM_UNKNOWN);
2002}
2003
2004static void
2005cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2006{
2007	struct vi_info *vi = ifp->if_softc;
2008	struct port_info *pi = vi->pi;
2009	struct adapter *sc = pi->adapter;
2010	struct link_config *lc = &pi->link_cfg;
2011
2012	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4med") != 0)
2013		return;
2014	PORT_LOCK(pi);
2015
2016	if (pi->up_vis == 0) {
2017		/*
2018		 * If all the interfaces are administratively down the firmware
2019		 * does not report transceiver changes.  Refresh port info here
2020		 * so that ifconfig displays accurate ifmedia at all times.
2021		 * This is the only reason we have a synchronized op in this
2022		 * function.  Just PORT_LOCK would have been enough otherwise.
2023		 */
2024		t4_update_port_info(pi);
2025		build_medialist(pi, &pi->media);
2026	}
2027
2028	/* ifm_status */
2029	ifmr->ifm_status = IFM_AVALID;
2030	if (lc->link_ok == 0)
2031		goto done;
2032	ifmr->ifm_status |= IFM_ACTIVE;
2033
2034	/* ifm_active */
2035	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2036	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2037	if (lc->fc & PAUSE_RX)
2038		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2039	if (lc->fc & PAUSE_TX)
2040		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2041	ifmr->ifm_active |= port_mword(pi, speed_to_fwspeed(lc->speed));
2042done:
2043	PORT_UNLOCK(pi);
2044	end_synchronized_op(sc, 0);
2045}
2046
2047static int
2048vcxgbe_probe(device_t dev)
2049{
2050	char buf[128];
2051	struct vi_info *vi = device_get_softc(dev);
2052
2053	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2054	    vi - vi->pi->vi);
2055	device_set_desc_copy(dev, buf);
2056
2057	return (BUS_PROBE_DEFAULT);
2058}
2059
2060static int
2061vcxgbe_attach(device_t dev)
2062{
2063	struct vi_info *vi;
2064	struct port_info *pi;
2065	struct adapter *sc;
2066	int func, index, rc;
2067	u32 param, val;
2068
2069	vi = device_get_softc(dev);
2070	pi = vi->pi;
2071	sc = pi->adapter;
2072
2073	index = vi - pi->vi;
2074	KASSERT(index < nitems(vi_mac_funcs),
2075	    ("%s: VI %s doesn't have a MAC func", __func__,
2076	    device_get_nameunit(dev)));
2077	func = vi_mac_funcs[index];
2078	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2079	    vi->hw_addr, &vi->rss_size, func, 0);
2080	if (rc < 0) {
2081		device_printf(dev, "Failed to allocate virtual interface "
2082		    "for port %d: %d\n", pi->port_id, -rc);
2083		return (-rc);
2084	}
2085	vi->viid = rc;
2086	if (chip_id(sc) <= CHELSIO_T5)
2087		vi->smt_idx = (rc & 0x7f) << 1;
2088	else
2089		vi->smt_idx = (rc & 0x7f);
2090
2091	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2092	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2093	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2094	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2095	if (rc)
2096		vi->rss_base = 0xffff;
2097	else {
2098		/* MPASS((val >> 16) == rss_size); */
2099		vi->rss_base = val & 0xffff;
2100	}
2101
2102	rc = cxgbe_vi_attach(dev, vi);
2103	if (rc) {
2104		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2105		return (rc);
2106	}
2107	return (0);
2108}
2109
2110static int
2111vcxgbe_detach(device_t dev)
2112{
2113	struct vi_info *vi;
2114	struct adapter *sc;
2115
2116	vi = device_get_softc(dev);
2117	sc = vi->pi->adapter;
2118
2119	doom_vi(sc, vi);
2120
2121	cxgbe_vi_detach(vi);
2122	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2123
2124	end_synchronized_op(sc, 0);
2125
2126	return (0);
2127}
2128
2129void
2130t4_fatal_err(struct adapter *sc)
2131{
2132	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2133	t4_intr_disable(sc);
2134	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2135	    device_get_nameunit(sc->dev));
2136}
2137
2138void
2139t4_add_adapter(struct adapter *sc)
2140{
2141	sx_xlock(&t4_list_lock);
2142	SLIST_INSERT_HEAD(&t4_list, sc, link);
2143	sx_xunlock(&t4_list_lock);
2144}
2145
2146int
2147t4_map_bars_0_and_4(struct adapter *sc)
2148{
2149	sc->regs_rid = PCIR_BAR(0);
2150	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2151	    &sc->regs_rid, RF_ACTIVE);
2152	if (sc->regs_res == NULL) {
2153		device_printf(sc->dev, "cannot map registers.\n");
2154		return (ENXIO);
2155	}
2156	sc->bt = rman_get_bustag(sc->regs_res);
2157	sc->bh = rman_get_bushandle(sc->regs_res);
2158	sc->mmio_len = rman_get_size(sc->regs_res);
2159	setbit(&sc->doorbells, DOORBELL_KDB);
2160
2161	sc->msix_rid = PCIR_BAR(4);
2162	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2163	    &sc->msix_rid, RF_ACTIVE);
2164	if (sc->msix_res == NULL) {
2165		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2166		return (ENXIO);
2167	}
2168
2169	return (0);
2170}
2171
2172int
2173t4_map_bar_2(struct adapter *sc)
2174{
2175
2176	/*
2177	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2178	 * to map it if RDMA is disabled.
2179	 */
2180	if (is_t4(sc) && sc->rdmacaps == 0)
2181		return (0);
2182
2183	sc->udbs_rid = PCIR_BAR(2);
2184	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2185	    &sc->udbs_rid, RF_ACTIVE);
2186	if (sc->udbs_res == NULL) {
2187		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2188		return (ENXIO);
2189	}
2190	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2191
2192	if (chip_id(sc) >= CHELSIO_T5) {
2193		setbit(&sc->doorbells, DOORBELL_UDB);
2194#if defined(__i386__) || defined(__amd64__)
2195		if (t5_write_combine) {
2196			int rc, mode;
2197
2198			/*
2199			 * Enable write combining on BAR2.  This is the
2200			 * userspace doorbell BAR and is split into 128B
2201			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2202			 * with an egress queue.  The first 64B has the doorbell
2203			 * and the second 64B can be used to submit a tx work
2204			 * request with an implicit doorbell.
2205			 */
2206
2207			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2208			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2209			if (rc == 0) {
2210				clrbit(&sc->doorbells, DOORBELL_UDB);
2211				setbit(&sc->doorbells, DOORBELL_WCWR);
2212				setbit(&sc->doorbells, DOORBELL_UDBWC);
2213			} else {
2214				device_printf(sc->dev,
2215				    "couldn't enable write combining: %d\n",
2216				    rc);
2217			}
2218
2219			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2220			t4_write_reg(sc, A_SGE_STAT_CFG,
2221			    V_STATSOURCE_T5(7) | mode);
2222		}
2223#endif
2224	}
2225
2226	return (0);
2227}
2228
2229struct memwin_init {
2230	uint32_t base;
2231	uint32_t aperture;
2232};
2233
2234static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2235	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2236	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2237	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2238};
2239
2240static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2241	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2242	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2243	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2244};
2245
2246static void
2247setup_memwin(struct adapter *sc)
2248{
2249	const struct memwin_init *mw_init;
2250	struct memwin *mw;
2251	int i;
2252	uint32_t bar0;
2253
2254	if (is_t4(sc)) {
2255		/*
2256		 * Read low 32b of bar0 indirectly via the hardware backdoor
2257		 * mechanism.  Works from within PCI passthrough environments
2258		 * too, where rman_get_start() can return a different value.  We
2259		 * need to program the T4 memory window decoders with the actual
2260		 * addresses that will be coming across the PCIe link.
2261		 */
2262		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2263		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2264
2265		mw_init = &t4_memwin[0];
2266	} else {
2267		/* T5+ use the relative offset inside the PCIe BAR */
2268		bar0 = 0;
2269
2270		mw_init = &t5_memwin[0];
2271	}
2272
2273	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2274		rw_init(&mw->mw_lock, "memory window access");
2275		mw->mw_base = mw_init->base;
2276		mw->mw_aperture = mw_init->aperture;
2277		mw->mw_curpos = 0;
2278		t4_write_reg(sc,
2279		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2280		    (mw->mw_base + bar0) | V_BIR(0) |
2281		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2282		rw_wlock(&mw->mw_lock);
2283		position_memwin(sc, i, 0);
2284		rw_wunlock(&mw->mw_lock);
2285	}
2286
2287	/* flush */
2288	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2289}
2290
2291/*
2292 * Positions the memory window at the given address in the card's address space.
2293 * There are some alignment requirements and the actual position may be at an
2294 * address prior to the requested address.  mw->mw_curpos always has the actual
2295 * position of the window.
2296 */
2297static void
2298position_memwin(struct adapter *sc, int idx, uint32_t addr)
2299{
2300	struct memwin *mw;
2301	uint32_t pf;
2302	uint32_t reg;
2303
2304	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2305	mw = &sc->memwin[idx];
2306	rw_assert(&mw->mw_lock, RA_WLOCKED);
2307
2308	if (is_t4(sc)) {
2309		pf = 0;
2310		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2311	} else {
2312		pf = V_PFNUM(sc->pf);
2313		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2314	}
2315	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2316	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2317	t4_read_reg(sc, reg);	/* flush */
2318}
2319
2320static int
2321rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2322    int len, int rw)
2323{
2324	struct memwin *mw;
2325	uint32_t mw_end, v;
2326
2327	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2328
2329	/* Memory can only be accessed in naturally aligned 4 byte units */
2330	if (addr & 3 || len & 3 || len <= 0)
2331		return (EINVAL);
2332
2333	mw = &sc->memwin[idx];
2334	while (len > 0) {
2335		rw_rlock(&mw->mw_lock);
2336		mw_end = mw->mw_curpos + mw->mw_aperture;
2337		if (addr >= mw_end || addr < mw->mw_curpos) {
2338			/* Will need to reposition the window */
2339			if (!rw_try_upgrade(&mw->mw_lock)) {
2340				rw_runlock(&mw->mw_lock);
2341				rw_wlock(&mw->mw_lock);
2342			}
2343			rw_assert(&mw->mw_lock, RA_WLOCKED);
2344			position_memwin(sc, idx, addr);
2345			rw_downgrade(&mw->mw_lock);
2346			mw_end = mw->mw_curpos + mw->mw_aperture;
2347		}
2348		rw_assert(&mw->mw_lock, RA_RLOCKED);
2349		while (addr < mw_end && len > 0) {
2350			if (rw == 0) {
2351				v = t4_read_reg(sc, mw->mw_base + addr -
2352				    mw->mw_curpos);
2353				*val++ = le32toh(v);
2354			} else {
2355				v = *val++;
2356				t4_write_reg(sc, mw->mw_base + addr -
2357				    mw->mw_curpos, htole32(v));;
2358			}
2359			addr += 4;
2360			len -= 4;
2361		}
2362		rw_runlock(&mw->mw_lock);
2363	}
2364
2365	return (0);
2366}
2367
2368static inline int
2369read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2370    int len)
2371{
2372
2373	return (rw_via_memwin(sc, idx, addr, val, len, 0));
2374}
2375
2376static inline int
2377write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
2378    const uint32_t *val, int len)
2379{
2380
2381	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
2382}
2383
2384static int
2385t4_range_cmp(const void *a, const void *b)
2386{
2387	return ((const struct t4_range *)a)->start -
2388	       ((const struct t4_range *)b)->start;
2389}
2390
2391/*
2392 * Verify that the memory range specified by the addr/len pair is valid within
2393 * the card's address space.
2394 */
2395static int
2396validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2397{
2398	struct t4_range mem_ranges[4], *r, *next;
2399	uint32_t em, addr_len;
2400	int i, n, remaining;
2401
2402	/* Memory can only be accessed in naturally aligned 4 byte units */
2403	if (addr & 3 || len & 3 || len <= 0)
2404		return (EINVAL);
2405
2406	/* Enabled memories */
2407	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2408
2409	r = &mem_ranges[0];
2410	n = 0;
2411	bzero(r, sizeof(mem_ranges));
2412	if (em & F_EDRAM0_ENABLE) {
2413		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2414		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2415		if (r->size > 0) {
2416			r->start = G_EDRAM0_BASE(addr_len) << 20;
2417			if (addr >= r->start &&
2418			    addr + len <= r->start + r->size)
2419				return (0);
2420			r++;
2421			n++;
2422		}
2423	}
2424	if (em & F_EDRAM1_ENABLE) {
2425		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2426		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2427		if (r->size > 0) {
2428			r->start = G_EDRAM1_BASE(addr_len) << 20;
2429			if (addr >= r->start &&
2430			    addr + len <= r->start + r->size)
2431				return (0);
2432			r++;
2433			n++;
2434		}
2435	}
2436	if (em & F_EXT_MEM_ENABLE) {
2437		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2438		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2439		if (r->size > 0) {
2440			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2441			if (addr >= r->start &&
2442			    addr + len <= r->start + r->size)
2443				return (0);
2444			r++;
2445			n++;
2446		}
2447	}
2448	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2449		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2450		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2451		if (r->size > 0) {
2452			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2453			if (addr >= r->start &&
2454			    addr + len <= r->start + r->size)
2455				return (0);
2456			r++;
2457			n++;
2458		}
2459	}
2460	MPASS(n <= nitems(mem_ranges));
2461
2462	if (n > 1) {
2463		/* Sort and merge the ranges. */
2464		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2465
2466		/* Start from index 0 and examine the next n - 1 entries. */
2467		r = &mem_ranges[0];
2468		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2469
2470			MPASS(r->size > 0);	/* r is a valid entry. */
2471			next = r + 1;
2472			MPASS(next->size > 0);	/* and so is the next one. */
2473
2474			while (r->start + r->size >= next->start) {
2475				/* Merge the next one into the current entry. */
2476				r->size = max(r->start + r->size,
2477				    next->start + next->size) - r->start;
2478				n--;	/* One fewer entry in total. */
2479				if (--remaining == 0)
2480					goto done;	/* short circuit */
2481				next++;
2482			}
2483			if (next != r + 1) {
2484				/*
2485				 * Some entries were merged into r and next
2486				 * points to the first valid entry that couldn't
2487				 * be merged.
2488				 */
2489				MPASS(next->size > 0);	/* must be valid */
2490				memcpy(r + 1, next, remaining * sizeof(*r));
2491#ifdef INVARIANTS
2492				/*
2493				 * This so that the foo->size assertion in the
2494				 * next iteration of the loop do the right
2495				 * thing for entries that were pulled up and are
2496				 * no longer valid.
2497				 */
2498				MPASS(n < nitems(mem_ranges));
2499				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2500				    sizeof(struct t4_range));
2501#endif
2502			}
2503		}
2504done:
2505		/* Done merging the ranges. */
2506		MPASS(n > 0);
2507		r = &mem_ranges[0];
2508		for (i = 0; i < n; i++, r++) {
2509			if (addr >= r->start &&
2510			    addr + len <= r->start + r->size)
2511				return (0);
2512		}
2513	}
2514
2515	return (EFAULT);
2516}
2517
2518static int
2519fwmtype_to_hwmtype(int mtype)
2520{
2521
2522	switch (mtype) {
2523	case FW_MEMTYPE_EDC0:
2524		return (MEM_EDC0);
2525	case FW_MEMTYPE_EDC1:
2526		return (MEM_EDC1);
2527	case FW_MEMTYPE_EXTMEM:
2528		return (MEM_MC0);
2529	case FW_MEMTYPE_EXTMEM1:
2530		return (MEM_MC1);
2531	default:
2532		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2533	}
2534}
2535
2536/*
2537 * Verify that the memory range specified by the memtype/offset/len pair is
2538 * valid and lies entirely within the memtype specified.  The global address of
2539 * the start of the range is returned in addr.
2540 */
2541static int
2542validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2543    uint32_t *addr)
2544{
2545	uint32_t em, addr_len, maddr;
2546
2547	/* Memory can only be accessed in naturally aligned 4 byte units */
2548	if (off & 3 || len & 3 || len == 0)
2549		return (EINVAL);
2550
2551	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2552	switch (fwmtype_to_hwmtype(mtype)) {
2553	case MEM_EDC0:
2554		if (!(em & F_EDRAM0_ENABLE))
2555			return (EINVAL);
2556		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2557		maddr = G_EDRAM0_BASE(addr_len) << 20;
2558		break;
2559	case MEM_EDC1:
2560		if (!(em & F_EDRAM1_ENABLE))
2561			return (EINVAL);
2562		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2563		maddr = G_EDRAM1_BASE(addr_len) << 20;
2564		break;
2565	case MEM_MC:
2566		if (!(em & F_EXT_MEM_ENABLE))
2567			return (EINVAL);
2568		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2569		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2570		break;
2571	case MEM_MC1:
2572		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2573			return (EINVAL);
2574		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2575		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2576		break;
2577	default:
2578		return (EINVAL);
2579	}
2580
2581	*addr = maddr + off;	/* global address */
2582	return (validate_mem_range(sc, *addr, len));
2583}
2584
2585static int
2586fixup_devlog_params(struct adapter *sc)
2587{
2588	struct devlog_params *dparams = &sc->params.devlog;
2589	int rc;
2590
2591	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2592	    dparams->size, &dparams->addr);
2593
2594	return (rc);
2595}
2596
2597static int
2598cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g, int num_vis,
2599    struct intrs_and_queues *iaq)
2600{
2601	int rc, itype, navail, nrxq10g, nrxq1g, n;
2602	int nofldrxq10g = 0, nofldrxq1g = 0;
2603
2604	bzero(iaq, sizeof(*iaq));
2605
2606	iaq->ntxq10g = t4_ntxq10g;
2607	iaq->ntxq1g = t4_ntxq1g;
2608	iaq->ntxq_vi = t4_ntxq_vi;
2609	iaq->nrxq10g = nrxq10g = t4_nrxq10g;
2610	iaq->nrxq1g = nrxq1g = t4_nrxq1g;
2611	iaq->nrxq_vi = t4_nrxq_vi;
2612	iaq->rsrv_noflowq = t4_rsrv_noflowq;
2613#ifdef TCP_OFFLOAD
2614	if (is_offload(sc)) {
2615		iaq->nofldtxq10g = t4_nofldtxq10g;
2616		iaq->nofldtxq1g = t4_nofldtxq1g;
2617		iaq->nofldtxq_vi = t4_nofldtxq_vi;
2618		iaq->nofldrxq10g = nofldrxq10g = t4_nofldrxq10g;
2619		iaq->nofldrxq1g = nofldrxq1g = t4_nofldrxq1g;
2620		iaq->nofldrxq_vi = t4_nofldrxq_vi;
2621	}
2622#endif
2623#ifdef DEV_NETMAP
2624	iaq->nnmtxq_vi = t4_nnmtxq_vi;
2625	iaq->nnmrxq_vi = t4_nnmrxq_vi;
2626#endif
2627
2628	for (itype = INTR_MSIX; itype; itype >>= 1) {
2629
2630		if ((itype & t4_intr_types) == 0)
2631			continue;	/* not allowed */
2632
2633		if (itype == INTR_MSIX)
2634			navail = pci_msix_count(sc->dev);
2635		else if (itype == INTR_MSI)
2636			navail = pci_msi_count(sc->dev);
2637		else
2638			navail = 1;
2639restart:
2640		if (navail == 0)
2641			continue;
2642
2643		iaq->intr_type = itype;
2644		iaq->intr_flags_10g = 0;
2645		iaq->intr_flags_1g = 0;
2646
2647		/*
2648		 * Best option: an interrupt vector for errors, one for the
2649		 * firmware event queue, and one for every rxq (NIC and TOE) of
2650		 * every VI.  The VIs that support netmap use the same
2651		 * interrupts for the NIC rx queues and the netmap rx queues
2652		 * because only one set of queues is active at a time.
2653		 */
2654		iaq->nirq = T4_EXTRA_INTR;
2655		iaq->nirq += n10g * (nrxq10g + nofldrxq10g);
2656		iaq->nirq += n1g * (nrxq1g + nofldrxq1g);
2657		iaq->nirq += (n10g + n1g) * (num_vis - 1) *
2658		    max(iaq->nrxq_vi, iaq->nnmrxq_vi);	/* See comment above. */
2659		iaq->nirq += (n10g + n1g) * (num_vis - 1) * iaq->nofldrxq_vi;
2660		if (iaq->nirq <= navail &&
2661		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2662			iaq->intr_flags_10g = INTR_ALL;
2663			iaq->intr_flags_1g = INTR_ALL;
2664			goto allocate;
2665		}
2666
2667		/* Disable the VIs (and netmap) if there aren't enough intrs */
2668		if (num_vis > 1) {
2669			device_printf(sc->dev, "virtual interfaces disabled "
2670			    "because num_vis=%u with current settings "
2671			    "(nrxq10g=%u, nrxq1g=%u, nofldrxq10g=%u, "
2672			    "nofldrxq1g=%u, nrxq_vi=%u nofldrxq_vi=%u, "
2673			    "nnmrxq_vi=%u) would need %u interrupts but "
2674			    "only %u are available.\n", num_vis, nrxq10g,
2675			    nrxq1g, nofldrxq10g, nofldrxq1g, iaq->nrxq_vi,
2676			    iaq->nofldrxq_vi, iaq->nnmrxq_vi, iaq->nirq,
2677			    navail);
2678			num_vis = 1;
2679			iaq->ntxq_vi = iaq->nrxq_vi = 0;
2680			iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
2681			iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
2682			goto restart;
2683		}
2684
2685		/*
2686		 * Second best option: a vector for errors, one for the firmware
2687		 * event queue, and vectors for either all the NIC rx queues or
2688		 * all the TOE rx queues.  The queues that don't get vectors
2689		 * will forward their interrupts to those that do.
2690		 */
2691		iaq->nirq = T4_EXTRA_INTR;
2692		if (nrxq10g >= nofldrxq10g) {
2693			iaq->intr_flags_10g = INTR_RXQ;
2694			iaq->nirq += n10g * nrxq10g;
2695		} else {
2696			iaq->intr_flags_10g = INTR_OFLD_RXQ;
2697			iaq->nirq += n10g * nofldrxq10g;
2698		}
2699		if (nrxq1g >= nofldrxq1g) {
2700			iaq->intr_flags_1g = INTR_RXQ;
2701			iaq->nirq += n1g * nrxq1g;
2702		} else {
2703			iaq->intr_flags_1g = INTR_OFLD_RXQ;
2704			iaq->nirq += n1g * nofldrxq1g;
2705		}
2706		if (iaq->nirq <= navail &&
2707		    (itype != INTR_MSI || powerof2(iaq->nirq)))
2708			goto allocate;
2709
2710		/*
2711		 * Next best option: an interrupt vector for errors, one for the
2712		 * firmware event queue, and at least one per main-VI.  At this
2713		 * point we know we'll have to downsize nrxq and/or nofldrxq to
2714		 * fit what's available to us.
2715		 */
2716		iaq->nirq = T4_EXTRA_INTR;
2717		iaq->nirq += n10g + n1g;
2718		if (iaq->nirq <= navail) {
2719			int leftover = navail - iaq->nirq;
2720
2721			if (n10g > 0) {
2722				int target = max(nrxq10g, nofldrxq10g);
2723
2724				iaq->intr_flags_10g = nrxq10g >= nofldrxq10g ?
2725				    INTR_RXQ : INTR_OFLD_RXQ;
2726
2727				n = 1;
2728				while (n < target && leftover >= n10g) {
2729					leftover -= n10g;
2730					iaq->nirq += n10g;
2731					n++;
2732				}
2733				iaq->nrxq10g = min(n, nrxq10g);
2734#ifdef TCP_OFFLOAD
2735				iaq->nofldrxq10g = min(n, nofldrxq10g);
2736#endif
2737			}
2738
2739			if (n1g > 0) {
2740				int target = max(nrxq1g, nofldrxq1g);
2741
2742				iaq->intr_flags_1g = nrxq1g >= nofldrxq1g ?
2743				    INTR_RXQ : INTR_OFLD_RXQ;
2744
2745				n = 1;
2746				while (n < target && leftover >= n1g) {
2747					leftover -= n1g;
2748					iaq->nirq += n1g;
2749					n++;
2750				}
2751				iaq->nrxq1g = min(n, nrxq1g);
2752#ifdef TCP_OFFLOAD
2753				iaq->nofldrxq1g = min(n, nofldrxq1g);
2754#endif
2755			}
2756
2757			if (itype != INTR_MSI || powerof2(iaq->nirq))
2758				goto allocate;
2759		}
2760
2761		/*
2762		 * Least desirable option: one interrupt vector for everything.
2763		 */
2764		iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2765		iaq->intr_flags_10g = iaq->intr_flags_1g = 0;
2766#ifdef TCP_OFFLOAD
2767		if (is_offload(sc))
2768			iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2769#endif
2770allocate:
2771		navail = iaq->nirq;
2772		rc = 0;
2773		if (itype == INTR_MSIX)
2774			rc = pci_alloc_msix(sc->dev, &navail);
2775		else if (itype == INTR_MSI)
2776			rc = pci_alloc_msi(sc->dev, &navail);
2777
2778		if (rc == 0) {
2779			if (navail == iaq->nirq)
2780				return (0);
2781
2782			/*
2783			 * Didn't get the number requested.  Use whatever number
2784			 * the kernel is willing to allocate (it's in navail).
2785			 */
2786			device_printf(sc->dev, "fewer vectors than requested, "
2787			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2788			    itype, iaq->nirq, navail);
2789			pci_release_msi(sc->dev);
2790			goto restart;
2791		}
2792
2793		device_printf(sc->dev,
2794		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2795		    itype, rc, iaq->nirq, navail);
2796	}
2797
2798	device_printf(sc->dev,
2799	    "failed to find a usable interrupt type.  "
2800	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2801	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2802
2803	return (ENXIO);
2804}
2805
2806#define FW_VERSION(chip) ( \
2807    V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2808    V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2809    V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2810    V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2811#define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2812
2813struct fw_info {
2814	uint8_t chip;
2815	char *kld_name;
2816	char *fw_mod_name;
2817	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
2818} fw_info[] = {
2819	{
2820		.chip = CHELSIO_T4,
2821		.kld_name = "t4fw_cfg",
2822		.fw_mod_name = "t4fw",
2823		.fw_hdr = {
2824			.chip = FW_HDR_CHIP_T4,
2825			.fw_ver = htobe32_const(FW_VERSION(T4)),
2826			.intfver_nic = FW_INTFVER(T4, NIC),
2827			.intfver_vnic = FW_INTFVER(T4, VNIC),
2828			.intfver_ofld = FW_INTFVER(T4, OFLD),
2829			.intfver_ri = FW_INTFVER(T4, RI),
2830			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2831			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
2832			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2833			.intfver_fcoe = FW_INTFVER(T4, FCOE),
2834		},
2835	}, {
2836		.chip = CHELSIO_T5,
2837		.kld_name = "t5fw_cfg",
2838		.fw_mod_name = "t5fw",
2839		.fw_hdr = {
2840			.chip = FW_HDR_CHIP_T5,
2841			.fw_ver = htobe32_const(FW_VERSION(T5)),
2842			.intfver_nic = FW_INTFVER(T5, NIC),
2843			.intfver_vnic = FW_INTFVER(T5, VNIC),
2844			.intfver_ofld = FW_INTFVER(T5, OFLD),
2845			.intfver_ri = FW_INTFVER(T5, RI),
2846			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
2847			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
2848			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
2849			.intfver_fcoe = FW_INTFVER(T5, FCOE),
2850		},
2851	}, {
2852		.chip = CHELSIO_T6,
2853		.kld_name = "t6fw_cfg",
2854		.fw_mod_name = "t6fw",
2855		.fw_hdr = {
2856			.chip = FW_HDR_CHIP_T6,
2857			.fw_ver = htobe32_const(FW_VERSION(T6)),
2858			.intfver_nic = FW_INTFVER(T6, NIC),
2859			.intfver_vnic = FW_INTFVER(T6, VNIC),
2860			.intfver_ofld = FW_INTFVER(T6, OFLD),
2861			.intfver_ri = FW_INTFVER(T6, RI),
2862			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
2863			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
2864			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
2865			.intfver_fcoe = FW_INTFVER(T6, FCOE),
2866		},
2867	}
2868};
2869
2870static struct fw_info *
2871find_fw_info(int chip)
2872{
2873	int i;
2874
2875	for (i = 0; i < nitems(fw_info); i++) {
2876		if (fw_info[i].chip == chip)
2877			return (&fw_info[i]);
2878	}
2879	return (NULL);
2880}
2881
2882/*
2883 * Is the given firmware API compatible with the one the driver was compiled
2884 * with?
2885 */
2886static int
2887fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
2888{
2889
2890	/* short circuit if it's the exact same firmware version */
2891	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
2892		return (1);
2893
2894	/*
2895	 * XXX: Is this too conservative?  Perhaps I should limit this to the
2896	 * features that are supported in the driver.
2897	 */
2898#define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
2899	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
2900	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
2901	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
2902		return (1);
2903#undef SAME_INTF
2904
2905	return (0);
2906}
2907
2908/*
2909 * The firmware in the KLD is usable, but should it be installed?  This routine
2910 * explains itself in detail if it indicates the KLD firmware should be
2911 * installed.
2912 */
2913static int
2914should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
2915{
2916	const char *reason;
2917
2918	if (!card_fw_usable) {
2919		reason = "incompatible or unusable";
2920		goto install;
2921	}
2922
2923	if (k > c) {
2924		reason = "older than the version bundled with this driver";
2925		goto install;
2926	}
2927
2928	if (t4_fw_install == 2 && k != c) {
2929		reason = "different than the version bundled with this driver";
2930		goto install;
2931	}
2932
2933	return (0);
2934
2935install:
2936	if (t4_fw_install == 0) {
2937		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2938		    "but the driver is prohibited from installing a different "
2939		    "firmware on the card.\n",
2940		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2941		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
2942
2943		return (0);
2944	}
2945
2946	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2947	    "installing firmware %u.%u.%u.%u on card.\n",
2948	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2949	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
2950	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2951	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2952
2953	return (1);
2954}
2955/*
2956 * Establish contact with the firmware and determine if we are the master driver
2957 * or not, and whether we are responsible for chip initialization.
2958 */
2959static int
2960prep_firmware(struct adapter *sc)
2961{
2962	const struct firmware *fw = NULL, *default_cfg;
2963	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
2964	enum dev_state state;
2965	struct fw_info *fw_info;
2966	struct fw_hdr *card_fw;		/* fw on the card */
2967	const struct fw_hdr *kld_fw;	/* fw in the KLD */
2968	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
2969					   against */
2970
2971	/* Contact firmware. */
2972	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
2973	if (rc < 0 || state == DEV_STATE_ERR) {
2974		rc = -rc;
2975		device_printf(sc->dev,
2976		    "failed to connect to the firmware: %d, %d.\n", rc, state);
2977		return (rc);
2978	}
2979	pf = rc;
2980	if (pf == sc->mbox)
2981		sc->flags |= MASTER_PF;
2982	else if (state == DEV_STATE_UNINIT) {
2983		/*
2984		 * We didn't get to be the master so we definitely won't be
2985		 * configuring the chip.  It's a bug if someone else hasn't
2986		 * configured it already.
2987		 */
2988		device_printf(sc->dev, "couldn't be master(%d), "
2989		    "device not already initialized either(%d).\n", rc, state);
2990		return (EDOOFUS);
2991	}
2992
2993	/* This is the firmware whose headers the driver was compiled against */
2994	fw_info = find_fw_info(chip_id(sc));
2995	if (fw_info == NULL) {
2996		device_printf(sc->dev,
2997		    "unable to look up firmware information for chip %d.\n",
2998		    chip_id(sc));
2999		return (EINVAL);
3000	}
3001	drv_fw = &fw_info->fw_hdr;
3002
3003	/*
3004	 * The firmware KLD contains many modules.  The KLD name is also the
3005	 * name of the module that contains the default config file.
3006	 */
3007	default_cfg = firmware_get(fw_info->kld_name);
3008
3009	/* Read the header of the firmware on the card */
3010	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3011	rc = -t4_read_flash(sc, FLASH_FW_START,
3012	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
3013	if (rc == 0)
3014		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
3015	else {
3016		device_printf(sc->dev,
3017		    "Unable to read card's firmware header: %d\n", rc);
3018		card_fw_usable = 0;
3019	}
3020
3021	/* This is the firmware in the KLD */
3022	fw = firmware_get(fw_info->fw_mod_name);
3023	if (fw != NULL) {
3024		kld_fw = (const void *)fw->data;
3025		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
3026	} else {
3027		kld_fw = NULL;
3028		kld_fw_usable = 0;
3029	}
3030
3031	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
3032	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
3033		/*
3034		 * Common case: the firmware on the card is an exact match and
3035		 * the KLD is an exact match too, or the KLD is
3036		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
3037		 * here -- use cxgbetool loadfw if you want to reinstall the
3038		 * same firmware as the one on the card.
3039		 */
3040	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
3041	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
3042	    be32toh(card_fw->fw_ver))) {
3043
3044		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3045		if (rc != 0) {
3046			device_printf(sc->dev,
3047			    "failed to install firmware: %d\n", rc);
3048			goto done;
3049		}
3050
3051		/* Installed successfully, update the cached header too. */
3052		memcpy(card_fw, kld_fw, sizeof(*card_fw));
3053		card_fw_usable = 1;
3054		need_fw_reset = 0;	/* already reset as part of load_fw */
3055	}
3056
3057	if (!card_fw_usable) {
3058		uint32_t d, c, k;
3059
3060		d = ntohl(drv_fw->fw_ver);
3061		c = ntohl(card_fw->fw_ver);
3062		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
3063
3064		device_printf(sc->dev, "Cannot find a usable firmware: "
3065		    "fw_install %d, chip state %d, "
3066		    "driver compiled with %d.%d.%d.%d, "
3067		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
3068		    t4_fw_install, state,
3069		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3070		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
3071		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3072		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
3073		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3074		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3075		rc = EINVAL;
3076		goto done;
3077	}
3078
3079	/* Reset device */
3080	if (need_fw_reset &&
3081	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
3082		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3083		if (rc != ETIMEDOUT && rc != EIO)
3084			t4_fw_bye(sc, sc->mbox);
3085		goto done;
3086	}
3087	sc->flags |= FW_OK;
3088
3089	rc = get_params__pre_init(sc);
3090	if (rc != 0)
3091		goto done; /* error message displayed already */
3092
3093	/* Partition adapter resources as specified in the config file. */
3094	if (state == DEV_STATE_UNINIT) {
3095
3096		KASSERT(sc->flags & MASTER_PF,
3097		    ("%s: trying to change chip settings when not master.",
3098		    __func__));
3099
3100		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
3101		if (rc != 0)
3102			goto done;	/* error message displayed already */
3103
3104		t4_tweak_chip_settings(sc);
3105
3106		/* get basic stuff going */
3107		rc = -t4_fw_initialize(sc, sc->mbox);
3108		if (rc != 0) {
3109			device_printf(sc->dev, "fw init failed: %d.\n", rc);
3110			goto done;
3111		}
3112	} else {
3113		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
3114		sc->cfcsum = 0;
3115	}
3116
3117done:
3118	free(card_fw, M_CXGBE);
3119	if (fw != NULL)
3120		firmware_put(fw, FIRMWARE_UNLOAD);
3121	if (default_cfg != NULL)
3122		firmware_put(default_cfg, FIRMWARE_UNLOAD);
3123
3124	return (rc);
3125}
3126
3127#define FW_PARAM_DEV(param) \
3128	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3129	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3130#define FW_PARAM_PFVF(param) \
3131	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3132	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3133
3134/*
3135 * Partition chip resources for use between various PFs, VFs, etc.
3136 */
3137static int
3138partition_resources(struct adapter *sc, const struct firmware *default_cfg,
3139    const char *name_prefix)
3140{
3141	const struct firmware *cfg = NULL;
3142	int rc = 0;
3143	struct fw_caps_config_cmd caps;
3144	uint32_t mtype, moff, finicsum, cfcsum;
3145
3146	/*
3147	 * Figure out what configuration file to use.  Pick the default config
3148	 * file for the card if the user hasn't specified one explicitly.
3149	 */
3150	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
3151	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3152		/* Card specific overrides go here. */
3153		if (pci_get_device(sc->dev) == 0x440a)
3154			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
3155		if (is_fpga(sc))
3156			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
3157	}
3158
3159	/*
3160	 * We need to load another module if the profile is anything except
3161	 * "default" or "flash".
3162	 */
3163	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
3164	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3165		char s[32];
3166
3167		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
3168		cfg = firmware_get(s);
3169		if (cfg == NULL) {
3170			if (default_cfg != NULL) {
3171				device_printf(sc->dev,
3172				    "unable to load module \"%s\" for "
3173				    "configuration profile \"%s\", will use "
3174				    "the default config file instead.\n",
3175				    s, sc->cfg_file);
3176				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3177				    "%s", DEFAULT_CF);
3178			} else {
3179				device_printf(sc->dev,
3180				    "unable to load module \"%s\" for "
3181				    "configuration profile \"%s\", will use "
3182				    "the config file on the card's flash "
3183				    "instead.\n", s, sc->cfg_file);
3184				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
3185				    "%s", FLASH_CF);
3186			}
3187		}
3188	}
3189
3190	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
3191	    default_cfg == NULL) {
3192		device_printf(sc->dev,
3193		    "default config file not available, will use the config "
3194		    "file on the card's flash instead.\n");
3195		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
3196	}
3197
3198	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
3199		u_int cflen;
3200		const uint32_t *cfdata;
3201		uint32_t param, val, addr;
3202
3203		KASSERT(cfg != NULL || default_cfg != NULL,
3204		    ("%s: no config to upload", __func__));
3205
3206		/*
3207		 * Ask the firmware where it wants us to upload the config file.
3208		 */
3209		param = FW_PARAM_DEV(CF);
3210		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3211		if (rc != 0) {
3212			/* No support for config file?  Shouldn't happen. */
3213			device_printf(sc->dev,
3214			    "failed to query config file location: %d.\n", rc);
3215			goto done;
3216		}
3217		mtype = G_FW_PARAMS_PARAM_Y(val);
3218		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3219
3220		/*
3221		 * XXX: sheer laziness.  We deliberately added 4 bytes of
3222		 * useless stuffing/comments at the end of the config file so
3223		 * it's ok to simply throw away the last remaining bytes when
3224		 * the config file is not an exact multiple of 4.  This also
3225		 * helps with the validate_mt_off_len check.
3226		 */
3227		if (cfg != NULL) {
3228			cflen = cfg->datasize & ~3;
3229			cfdata = cfg->data;
3230		} else {
3231			cflen = default_cfg->datasize & ~3;
3232			cfdata = default_cfg->data;
3233		}
3234
3235		if (cflen > FLASH_CFG_MAX_SIZE) {
3236			device_printf(sc->dev,
3237			    "config file too long (%d, max allowed is %d).  "
3238			    "Will try to use the config on the card, if any.\n",
3239			    cflen, FLASH_CFG_MAX_SIZE);
3240			goto use_config_on_flash;
3241		}
3242
3243		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3244		if (rc != 0) {
3245			device_printf(sc->dev,
3246			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
3247			    "Will try to use the config on the card, if any.\n",
3248			    __func__, mtype, moff, cflen, rc);
3249			goto use_config_on_flash;
3250		}
3251		write_via_memwin(sc, 2, addr, cfdata, cflen);
3252	} else {
3253use_config_on_flash:
3254		mtype = FW_MEMTYPE_FLASH;
3255		moff = t4_flash_cfg_addr(sc);
3256	}
3257
3258	bzero(&caps, sizeof(caps));
3259	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3260	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3261	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3262	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3263	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
3264	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3265	if (rc != 0) {
3266		device_printf(sc->dev,
3267		    "failed to pre-process config file: %d "
3268		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3269		goto done;
3270	}
3271
3272	finicsum = be32toh(caps.finicsum);
3273	cfcsum = be32toh(caps.cfcsum);
3274	if (finicsum != cfcsum) {
3275		device_printf(sc->dev,
3276		    "WARNING: config file checksum mismatch: %08x %08x\n",
3277		    finicsum, cfcsum);
3278	}
3279	sc->cfcsum = cfcsum;
3280
3281#define LIMIT_CAPS(x) do { \
3282	caps.x &= htobe16(t4_##x##_allowed); \
3283} while (0)
3284
3285	/*
3286	 * Let the firmware know what features will (not) be used so it can tune
3287	 * things accordingly.
3288	 */
3289	LIMIT_CAPS(nbmcaps);
3290	LIMIT_CAPS(linkcaps);
3291	LIMIT_CAPS(switchcaps);
3292	LIMIT_CAPS(niccaps);
3293	LIMIT_CAPS(toecaps);
3294	LIMIT_CAPS(rdmacaps);
3295	LIMIT_CAPS(cryptocaps);
3296	LIMIT_CAPS(iscsicaps);
3297	LIMIT_CAPS(fcoecaps);
3298#undef LIMIT_CAPS
3299
3300	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3301	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3302	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3303	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3304	if (rc != 0) {
3305		device_printf(sc->dev,
3306		    "failed to process config file: %d.\n", rc);
3307	}
3308done:
3309	if (cfg != NULL)
3310		firmware_put(cfg, FIRMWARE_UNLOAD);
3311	return (rc);
3312}
3313
3314/*
3315 * Retrieve parameters that are needed (or nice to have) very early.
3316 */
3317static int
3318get_params__pre_init(struct adapter *sc)
3319{
3320	int rc;
3321	uint32_t param[2], val[2];
3322
3323	t4_get_version_info(sc);
3324
3325	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3326	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3327	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3328	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3329	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3330
3331	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3332	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3333	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3334	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3335	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3336
3337	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3338	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3339	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3340	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3341	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3342
3343	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3344	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3345	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3346	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3347	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3348
3349	param[0] = FW_PARAM_DEV(PORTVEC);
3350	param[1] = FW_PARAM_DEV(CCLK);
3351	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3352	if (rc != 0) {
3353		device_printf(sc->dev,
3354		    "failed to query parameters (pre_init): %d.\n", rc);
3355		return (rc);
3356	}
3357
3358	sc->params.portvec = val[0];
3359	sc->params.nports = bitcount32(val[0]);
3360	sc->params.vpd.cclk = val[1];
3361
3362	/* Read device log parameters. */
3363	rc = -t4_init_devlog_params(sc, 1);
3364	if (rc == 0)
3365		fixup_devlog_params(sc);
3366	else {
3367		device_printf(sc->dev,
3368		    "failed to get devlog parameters: %d.\n", rc);
3369		rc = 0;	/* devlog isn't critical for device operation */
3370	}
3371
3372	return (rc);
3373}
3374
3375/*
3376 * Retrieve various parameters that are of interest to the driver.  The device
3377 * has been initialized by the firmware at this point.
3378 */
3379static int
3380get_params__post_init(struct adapter *sc)
3381{
3382	int rc;
3383	uint32_t param[7], val[7];
3384	struct fw_caps_config_cmd caps;
3385
3386	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3387	param[1] = FW_PARAM_PFVF(EQ_START);
3388	param[2] = FW_PARAM_PFVF(FILTER_START);
3389	param[3] = FW_PARAM_PFVF(FILTER_END);
3390	param[4] = FW_PARAM_PFVF(L2T_START);
3391	param[5] = FW_PARAM_PFVF(L2T_END);
3392	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3393	if (rc != 0) {
3394		device_printf(sc->dev,
3395		    "failed to query parameters (post_init): %d.\n", rc);
3396		return (rc);
3397	}
3398
3399	sc->sge.iq_start = val[0];
3400	sc->sge.eq_start = val[1];
3401	sc->tids.ftid_base = val[2];
3402	sc->tids.nftids = val[3] - val[2] + 1;
3403	sc->params.ftid_min = val[2];
3404	sc->params.ftid_max = val[3];
3405	sc->vres.l2t.start = val[4];
3406	sc->vres.l2t.size = val[5] - val[4] + 1;
3407	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3408	    ("%s: L2 table size (%u) larger than expected (%u)",
3409	    __func__, sc->vres.l2t.size, L2T_SIZE));
3410
3411	/* get capabilites */
3412	bzero(&caps, sizeof(caps));
3413	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3414	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3415	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3416	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3417	if (rc != 0) {
3418		device_printf(sc->dev,
3419		    "failed to get card capabilities: %d.\n", rc);
3420		return (rc);
3421	}
3422
3423#define READ_CAPS(x) do { \
3424	sc->x = htobe16(caps.x); \
3425} while (0)
3426	READ_CAPS(nbmcaps);
3427	READ_CAPS(linkcaps);
3428	READ_CAPS(switchcaps);
3429	READ_CAPS(niccaps);
3430	READ_CAPS(toecaps);
3431	READ_CAPS(rdmacaps);
3432	READ_CAPS(cryptocaps);
3433	READ_CAPS(iscsicaps);
3434	READ_CAPS(fcoecaps);
3435
3436	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3437		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3438		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3439		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3440		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3441		if (rc != 0) {
3442			device_printf(sc->dev,
3443			    "failed to query NIC parameters: %d.\n", rc);
3444			return (rc);
3445		}
3446		sc->tids.etid_base = val[0];
3447		sc->params.etid_min = val[0];
3448		sc->tids.netids = val[1] - val[0] + 1;
3449		sc->params.netids = sc->tids.netids;
3450		sc->params.eo_wr_cred = val[2];
3451		sc->params.ethoffload = 1;
3452	}
3453
3454	if (sc->toecaps) {
3455		/* query offload-related parameters */
3456		param[0] = FW_PARAM_DEV(NTID);
3457		param[1] = FW_PARAM_PFVF(SERVER_START);
3458		param[2] = FW_PARAM_PFVF(SERVER_END);
3459		param[3] = FW_PARAM_PFVF(TDDP_START);
3460		param[4] = FW_PARAM_PFVF(TDDP_END);
3461		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3462		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3463		if (rc != 0) {
3464			device_printf(sc->dev,
3465			    "failed to query TOE parameters: %d.\n", rc);
3466			return (rc);
3467		}
3468		sc->tids.ntids = val[0];
3469		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3470		sc->tids.stid_base = val[1];
3471		sc->tids.nstids = val[2] - val[1] + 1;
3472		sc->vres.ddp.start = val[3];
3473		sc->vres.ddp.size = val[4] - val[3] + 1;
3474		sc->params.ofldq_wr_cred = val[5];
3475		sc->params.offload = 1;
3476	}
3477	if (sc->rdmacaps) {
3478		param[0] = FW_PARAM_PFVF(STAG_START);
3479		param[1] = FW_PARAM_PFVF(STAG_END);
3480		param[2] = FW_PARAM_PFVF(RQ_START);
3481		param[3] = FW_PARAM_PFVF(RQ_END);
3482		param[4] = FW_PARAM_PFVF(PBL_START);
3483		param[5] = FW_PARAM_PFVF(PBL_END);
3484		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3485		if (rc != 0) {
3486			device_printf(sc->dev,
3487			    "failed to query RDMA parameters(1): %d.\n", rc);
3488			return (rc);
3489		}
3490		sc->vres.stag.start = val[0];
3491		sc->vres.stag.size = val[1] - val[0] + 1;
3492		sc->vres.rq.start = val[2];
3493		sc->vres.rq.size = val[3] - val[2] + 1;
3494		sc->vres.pbl.start = val[4];
3495		sc->vres.pbl.size = val[5] - val[4] + 1;
3496
3497		param[0] = FW_PARAM_PFVF(SQRQ_START);
3498		param[1] = FW_PARAM_PFVF(SQRQ_END);
3499		param[2] = FW_PARAM_PFVF(CQ_START);
3500		param[3] = FW_PARAM_PFVF(CQ_END);
3501		param[4] = FW_PARAM_PFVF(OCQ_START);
3502		param[5] = FW_PARAM_PFVF(OCQ_END);
3503		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3504		if (rc != 0) {
3505			device_printf(sc->dev,
3506			    "failed to query RDMA parameters(2): %d.\n", rc);
3507			return (rc);
3508		}
3509		sc->vres.qp.start = val[0];
3510		sc->vres.qp.size = val[1] - val[0] + 1;
3511		sc->vres.cq.start = val[2];
3512		sc->vres.cq.size = val[3] - val[2] + 1;
3513		sc->vres.ocq.start = val[4];
3514		sc->vres.ocq.size = val[5] - val[4] + 1;
3515
3516		param[0] = FW_PARAM_PFVF(SRQ_START);
3517		param[1] = FW_PARAM_PFVF(SRQ_END);
3518		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
3519		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
3520		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
3521		if (rc != 0) {
3522			device_printf(sc->dev,
3523			    "failed to query RDMA parameters(3): %d.\n", rc);
3524			return (rc);
3525		}
3526		sc->vres.srq.start = val[0];
3527		sc->vres.srq.size = val[1] - val[0] + 1;
3528		sc->params.max_ordird_qp = val[2];
3529		sc->params.max_ird_adapter = val[3];
3530	}
3531	if (sc->iscsicaps) {
3532		param[0] = FW_PARAM_PFVF(ISCSI_START);
3533		param[1] = FW_PARAM_PFVF(ISCSI_END);
3534		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3535		if (rc != 0) {
3536			device_printf(sc->dev,
3537			    "failed to query iSCSI parameters: %d.\n", rc);
3538			return (rc);
3539		}
3540		sc->vres.iscsi.start = val[0];
3541		sc->vres.iscsi.size = val[1] - val[0] + 1;
3542	}
3543
3544	t4_init_sge_params(sc);
3545
3546	/*
3547	 * We've got the params we wanted to query via the firmware.  Now grab
3548	 * some others directly from the chip.
3549	 */
3550	rc = t4_read_chip_settings(sc);
3551
3552	return (rc);
3553}
3554
3555static int
3556set_params__post_init(struct adapter *sc)
3557{
3558	uint32_t param, val;
3559
3560	/* ask for encapsulated CPLs */
3561	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3562	val = 1;
3563	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3564
3565	return (0);
3566}
3567
3568#undef FW_PARAM_PFVF
3569#undef FW_PARAM_DEV
3570
3571static void
3572t4_set_desc(struct adapter *sc)
3573{
3574	char buf[128];
3575	struct adapter_params *p = &sc->params;
3576
3577	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
3578
3579	device_set_desc_copy(sc->dev, buf);
3580}
3581
3582static inline void
3583ifmedia_add4(struct ifmedia *ifm, int m)
3584{
3585
3586	ifmedia_add(ifm, m, 0, NULL);
3587	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE, 0, NULL);
3588	ifmedia_add(ifm, m | IFM_ETH_RXPAUSE, 0, NULL);
3589	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE, 0, NULL);
3590}
3591
3592static void
3593set_current_media(struct port_info *pi, struct ifmedia *ifm)
3594{
3595	struct link_config *lc;
3596	int mword;
3597
3598	PORT_LOCK_ASSERT_OWNED(pi);
3599
3600	/* Leave current media alone if it's already set to IFM_NONE. */
3601	if (ifm->ifm_cur != NULL &&
3602	    IFM_SUBTYPE(ifm->ifm_cur->ifm_media) == IFM_NONE)
3603		return;
3604
3605	lc = &pi->link_cfg;
3606	if (lc->requested_aneg == AUTONEG_ENABLE &&
3607	    lc->supported & FW_PORT_CAP_ANEG) {
3608		ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
3609		return;
3610	}
3611	mword = IFM_ETHER | IFM_FDX;
3612	if (lc->requested_fc & PAUSE_TX)
3613		mword |= IFM_ETH_TXPAUSE;
3614	if (lc->requested_fc & PAUSE_RX)
3615		mword |= IFM_ETH_RXPAUSE;
3616	mword |= port_mword(pi, speed_to_fwspeed(lc->requested_speed));
3617	ifmedia_set(ifm, mword);
3618}
3619
3620static void
3621build_medialist(struct port_info *pi, struct ifmedia *ifm)
3622{
3623	uint16_t ss, speed;
3624	int unknown, mword, bit;
3625	struct link_config *lc;
3626
3627	PORT_LOCK_ASSERT_OWNED(pi);
3628
3629	if (pi->flags & FIXED_IFMEDIA)
3630		return;
3631
3632	/*
3633	 * First setup all the requested_ fields so that they comply with what's
3634	 * supported by the port + transceiver.  Note that this clobbers any
3635	 * user preferences set via sysctl_pause_settings or sysctl_autoneg.
3636	 */
3637	init_l1cfg(pi);
3638
3639	/*
3640	 * Now (re)build the ifmedia list.
3641	 */
3642	ifmedia_removeall(ifm);
3643	lc = &pi->link_cfg;
3644	ss = G_FW_PORT_CAP_SPEED(lc->supported); /* Supported Speeds */
3645	if (__predict_false(ss == 0)) {	/* not supposed to happen. */
3646		MPASS(ss != 0);
3647no_media:
3648		MPASS(LIST_EMPTY(&ifm->ifm_list));
3649		ifmedia_add(ifm, IFM_ETHER | IFM_NONE, 0, NULL);
3650		ifmedia_set(ifm, IFM_ETHER | IFM_NONE);
3651		return;
3652	}
3653
3654	unknown = 0;
3655	for (bit = 0; bit < fls(ss); bit++) {
3656		speed = 1 << bit;
3657		MPASS(speed & M_FW_PORT_CAP_SPEED);
3658		if (ss & speed) {
3659			mword = port_mword(pi, speed);
3660			if (mword == IFM_NONE) {
3661				goto no_media;
3662			} else if (mword == IFM_UNKNOWN)
3663				unknown++;
3664			else
3665				ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | mword);
3666		}
3667	}
3668	if (unknown > 0) /* Add one unknown for all unknown media types. */
3669		ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | IFM_UNKNOWN);
3670	if (lc->supported & FW_PORT_CAP_ANEG)
3671		ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
3672
3673	set_current_media(pi, ifm);
3674}
3675
3676/*
3677 * Update all the requested_* fields in the link config to something valid (and
3678 * reasonable).
3679 */
3680static void
3681init_l1cfg(struct port_info *pi)
3682{
3683	struct link_config *lc = &pi->link_cfg;
3684
3685	PORT_LOCK_ASSERT_OWNED(pi);
3686
3687	/* Gbps -> Mbps */
3688	lc->requested_speed = port_top_speed(pi) * 1000;
3689
3690	if (t4_autoneg != 0 && lc->supported & FW_PORT_CAP_ANEG) {
3691		lc->requested_aneg = AUTONEG_ENABLE;
3692	} else {
3693		lc->requested_aneg = AUTONEG_DISABLE;
3694	}
3695
3696	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX);
3697
3698	if (t4_fec != -1) {
3699		if (t4_fec & FEC_RS && lc->supported & FW_PORT_CAP_FEC_RS) {
3700			lc->requested_fec = FEC_RS;
3701		} else if (t4_fec & FEC_BASER_RS &&
3702		    lc->supported & FW_PORT_CAP_FEC_BASER_RS) {
3703			lc->requested_fec = FEC_BASER_RS;
3704		} else {
3705			lc->requested_fec = 0;
3706		}
3707	} else {
3708		/* Use the suggested value provided by the firmware in acaps */
3709		if (lc->advertising & FW_PORT_CAP_FEC_RS &&
3710		    lc->supported & FW_PORT_CAP_FEC_RS) {
3711			lc->requested_fec = FEC_RS;
3712		} else if (lc->advertising & FW_PORT_CAP_FEC_BASER_RS &&
3713		    lc->supported & FW_PORT_CAP_FEC_BASER_RS) {
3714			lc->requested_fec = FEC_BASER_RS;
3715		} else {
3716			lc->requested_fec = 0;
3717		}
3718	}
3719}
3720
3721/*
3722 * Apply the settings in requested_* to the hardware.  The parameters are
3723 * expected to be sane.
3724 */
3725static int
3726apply_l1cfg(struct port_info *pi)
3727{
3728	struct adapter *sc = pi->adapter;
3729	struct link_config *lc = &pi->link_cfg;
3730	int rc;
3731#ifdef INVARIANTS
3732	uint16_t fwspeed;
3733
3734	ASSERT_SYNCHRONIZED_OP(sc);
3735	PORT_LOCK_ASSERT_OWNED(pi);
3736
3737	if (lc->requested_aneg == AUTONEG_ENABLE)
3738		MPASS(lc->supported & FW_PORT_CAP_ANEG);
3739	if (lc->requested_fc & PAUSE_TX)
3740		MPASS(lc->supported & FW_PORT_CAP_FC_TX);
3741	if (lc->requested_fc & PAUSE_RX)
3742		MPASS(lc->supported & FW_PORT_CAP_FC_RX);
3743	if (lc->requested_fec == FEC_RS)
3744		MPASS(lc->supported & FW_PORT_CAP_FEC_RS);
3745	if (lc->requested_fec == FEC_BASER_RS)
3746		MPASS(lc->supported & FW_PORT_CAP_FEC_BASER_RS);
3747	fwspeed = speed_to_fwspeed(lc->requested_speed);
3748	MPASS(fwspeed != 0);
3749	MPASS(lc->supported & fwspeed);
3750#endif
3751	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
3752	if (rc != 0) {
3753		device_printf(pi->dev, "l1cfg failed: %d\n", rc);
3754	} else {
3755		lc->fc = lc->requested_fc;
3756		lc->fec = lc->requested_fec;
3757	}
3758	return (rc);
3759}
3760
3761#define FW_MAC_EXACT_CHUNK	7
3762
3763/*
3764 * Program the port's XGMAC based on parameters in ifnet.  The caller also
3765 * indicates which parameters should be programmed (the rest are left alone).
3766 */
3767int
3768update_mac_settings(struct ifnet *ifp, int flags)
3769{
3770	int rc = 0;
3771	struct vi_info *vi = ifp->if_softc;
3772	struct port_info *pi = vi->pi;
3773	struct adapter *sc = pi->adapter;
3774	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
3775
3776	ASSERT_SYNCHRONIZED_OP(sc);
3777	KASSERT(flags, ("%s: not told what to update.", __func__));
3778
3779	if (flags & XGMAC_MTU)
3780		mtu = ifp->if_mtu;
3781
3782	if (flags & XGMAC_PROMISC)
3783		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
3784
3785	if (flags & XGMAC_ALLMULTI)
3786		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
3787
3788	if (flags & XGMAC_VLANEX)
3789		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
3790
3791	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
3792		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
3793		    allmulti, 1, vlanex, false);
3794		if (rc) {
3795			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
3796			    rc);
3797			return (rc);
3798		}
3799	}
3800
3801	if (flags & XGMAC_UCADDR) {
3802		uint8_t ucaddr[ETHER_ADDR_LEN];
3803
3804		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
3805		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
3806		    ucaddr, true, true);
3807		if (rc < 0) {
3808			rc = -rc;
3809			if_printf(ifp, "change_mac failed: %d\n", rc);
3810			return (rc);
3811		} else {
3812			vi->xact_addr_filt = rc;
3813			rc = 0;
3814		}
3815	}
3816
3817	if (flags & XGMAC_MCADDRS) {
3818		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
3819		int del = 1;
3820		uint64_t hash = 0;
3821		struct ifmultiaddr *ifma;
3822		int i = 0, j;
3823
3824		if_maddr_rlock(ifp);
3825		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3826			if (ifma->ifma_addr->sa_family != AF_LINK)
3827				continue;
3828			mcaddr[i] =
3829			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
3830			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
3831			i++;
3832
3833			if (i == FW_MAC_EXACT_CHUNK) {
3834				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
3835				    del, i, mcaddr, NULL, &hash, 0);
3836				if (rc < 0) {
3837					rc = -rc;
3838					for (j = 0; j < i; j++) {
3839						if_printf(ifp,
3840						    "failed to add mc address"
3841						    " %02x:%02x:%02x:"
3842						    "%02x:%02x:%02x rc=%d\n",
3843						    mcaddr[j][0], mcaddr[j][1],
3844						    mcaddr[j][2], mcaddr[j][3],
3845						    mcaddr[j][4], mcaddr[j][5],
3846						    rc);
3847					}
3848					goto mcfail;
3849				}
3850				del = 0;
3851				i = 0;
3852			}
3853		}
3854		if (i > 0) {
3855			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
3856			    mcaddr, NULL, &hash, 0);
3857			if (rc < 0) {
3858				rc = -rc;
3859				for (j = 0; j < i; j++) {
3860					if_printf(ifp,
3861					    "failed to add mc address"
3862					    " %02x:%02x:%02x:"
3863					    "%02x:%02x:%02x rc=%d\n",
3864					    mcaddr[j][0], mcaddr[j][1],
3865					    mcaddr[j][2], mcaddr[j][3],
3866					    mcaddr[j][4], mcaddr[j][5],
3867					    rc);
3868				}
3869				goto mcfail;
3870			}
3871		}
3872
3873		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
3874		if (rc != 0)
3875			if_printf(ifp, "failed to set mc address hash: %d", rc);
3876mcfail:
3877		if_maddr_runlock(ifp);
3878	}
3879
3880	return (rc);
3881}
3882
3883/*
3884 * {begin|end}_synchronized_op must be called from the same thread.
3885 */
3886int
3887begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
3888    char *wmesg)
3889{
3890	int rc, pri;
3891
3892#ifdef WITNESS
3893	/* the caller thinks it's ok to sleep, but is it really? */
3894	if (flags & SLEEP_OK)
3895		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
3896		    "begin_synchronized_op");
3897#endif
3898
3899	if (INTR_OK)
3900		pri = PCATCH;
3901	else
3902		pri = 0;
3903
3904	ADAPTER_LOCK(sc);
3905	for (;;) {
3906
3907		if (vi && IS_DOOMED(vi)) {
3908			rc = ENXIO;
3909			goto done;
3910		}
3911
3912		if (!IS_BUSY(sc)) {
3913			rc = 0;
3914			break;
3915		}
3916
3917		if (!(flags & SLEEP_OK)) {
3918			rc = EBUSY;
3919			goto done;
3920		}
3921
3922		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
3923			rc = EINTR;
3924			goto done;
3925		}
3926	}
3927
3928	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
3929	SET_BUSY(sc);
3930#ifdef INVARIANTS
3931	sc->last_op = wmesg;
3932	sc->last_op_thr = curthread;
3933	sc->last_op_flags = flags;
3934#endif
3935
3936done:
3937	if (!(flags & HOLD_LOCK) || rc)
3938		ADAPTER_UNLOCK(sc);
3939
3940	return (rc);
3941}
3942
3943/*
3944 * Tell if_ioctl and if_init that the VI is going away.  This is
3945 * special variant of begin_synchronized_op and must be paired with a
3946 * call to end_synchronized_op.
3947 */
3948void
3949doom_vi(struct adapter *sc, struct vi_info *vi)
3950{
3951
3952	ADAPTER_LOCK(sc);
3953	SET_DOOMED(vi);
3954	wakeup(&sc->flags);
3955	while (IS_BUSY(sc))
3956		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
3957	SET_BUSY(sc);
3958#ifdef INVARIANTS
3959	sc->last_op = "t4detach";
3960	sc->last_op_thr = curthread;
3961	sc->last_op_flags = 0;
3962#endif
3963	ADAPTER_UNLOCK(sc);
3964}
3965
3966/*
3967 * {begin|end}_synchronized_op must be called from the same thread.
3968 */
3969void
3970end_synchronized_op(struct adapter *sc, int flags)
3971{
3972
3973	if (flags & LOCK_HELD)
3974		ADAPTER_LOCK_ASSERT_OWNED(sc);
3975	else
3976		ADAPTER_LOCK(sc);
3977
3978	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
3979	CLR_BUSY(sc);
3980	wakeup(&sc->flags);
3981	ADAPTER_UNLOCK(sc);
3982}
3983
3984static int
3985cxgbe_init_synchronized(struct vi_info *vi)
3986{
3987	struct port_info *pi = vi->pi;
3988	struct adapter *sc = pi->adapter;
3989	struct ifnet *ifp = vi->ifp;
3990	int rc = 0, i;
3991	struct sge_txq *txq;
3992
3993	ASSERT_SYNCHRONIZED_OP(sc);
3994
3995	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3996		return (0);	/* already running */
3997
3998	if (!(sc->flags & FULL_INIT_DONE) &&
3999	    ((rc = adapter_full_init(sc)) != 0))
4000		return (rc);	/* error message displayed already */
4001
4002	if (!(vi->flags & VI_INIT_DONE) &&
4003	    ((rc = vi_full_init(vi)) != 0))
4004		return (rc); /* error message displayed already */
4005
4006	rc = update_mac_settings(ifp, XGMAC_ALL);
4007	if (rc)
4008		goto done;	/* error message displayed already */
4009
4010	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
4011	if (rc != 0) {
4012		if_printf(ifp, "enable_vi failed: %d\n", rc);
4013		goto done;
4014	}
4015
4016	/*
4017	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
4018	 * if this changes.
4019	 */
4020
4021	for_each_txq(vi, i, txq) {
4022		TXQ_LOCK(txq);
4023		txq->eq.flags |= EQ_ENABLED;
4024		TXQ_UNLOCK(txq);
4025	}
4026
4027	/*
4028	 * The first iq of the first port to come up is used for tracing.
4029	 */
4030	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4031		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4032		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4033		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4034		    V_QUEUENUMBER(sc->traceq));
4035		pi->flags |= HAS_TRACEQ;
4036	}
4037
4038	/* all ok */
4039	PORT_LOCK(pi);
4040	if (pi->up_vis++ == 0) {
4041		t4_update_port_info(pi);
4042		build_medialist(pi, &pi->media);
4043		apply_l1cfg(pi);
4044	}
4045	ifp->if_drv_flags |= IFF_DRV_RUNNING;
4046
4047	if (pi->nvi > 1 || sc->flags & IS_VF)
4048		callout_reset(&vi->tick, hz, vi_tick, vi);
4049	else
4050		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4051	PORT_UNLOCK(pi);
4052done:
4053	if (rc != 0)
4054		cxgbe_uninit_synchronized(vi);
4055
4056	return (rc);
4057}
4058
4059/*
4060 * Idempotent.
4061 */
4062static int
4063cxgbe_uninit_synchronized(struct vi_info *vi)
4064{
4065	struct port_info *pi = vi->pi;
4066	struct adapter *sc = pi->adapter;
4067	struct ifnet *ifp = vi->ifp;
4068	int rc, i;
4069	struct sge_txq *txq;
4070
4071	ASSERT_SYNCHRONIZED_OP(sc);
4072
4073	if (!(vi->flags & VI_INIT_DONE)) {
4074		KASSERT(!(ifp->if_drv_flags & IFF_DRV_RUNNING),
4075		    ("uninited VI is running"));
4076		return (0);
4077	}
4078
4079	/*
4080	 * Disable the VI so that all its data in either direction is discarded
4081	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4082	 * tick) intact as the TP can deliver negative advice or data that it's
4083	 * holding in its RAM (for an offloaded connection) even after the VI is
4084	 * disabled.
4085	 */
4086	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4087	if (rc) {
4088		if_printf(ifp, "disable_vi failed: %d\n", rc);
4089		return (rc);
4090	}
4091
4092	for_each_txq(vi, i, txq) {
4093		TXQ_LOCK(txq);
4094		txq->eq.flags &= ~EQ_ENABLED;
4095		TXQ_UNLOCK(txq);
4096	}
4097
4098	PORT_LOCK(pi);
4099	if (pi->nvi > 1 || sc->flags & IS_VF)
4100		callout_stop(&vi->tick);
4101	else
4102		callout_stop(&pi->tick);
4103	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4104		PORT_UNLOCK(pi);
4105		return (0);
4106	}
4107	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
4108	pi->up_vis--;
4109	if (pi->up_vis > 0) {
4110		PORT_UNLOCK(pi);
4111		return (0);
4112	}
4113
4114	pi->link_cfg.link_ok = 0;
4115	pi->link_cfg.speed = 0;
4116	pi->link_cfg.link_down_rc = 255;
4117	t4_os_link_changed(pi);
4118	pi->old_link_cfg = pi->link_cfg;
4119	PORT_UNLOCK(pi);
4120
4121	return (0);
4122}
4123
4124/*
4125 * It is ok for this function to fail midway and return right away.  t4_detach
4126 * will walk the entire sc->irq list and clean up whatever is valid.
4127 */
4128int
4129t4_setup_intr_handlers(struct adapter *sc)
4130{
4131	int rc, rid, p, q, v;
4132	char s[8];
4133	struct irq *irq;
4134	struct port_info *pi;
4135	struct vi_info *vi;
4136	struct sge *sge = &sc->sge;
4137	struct sge_rxq *rxq;
4138#ifdef TCP_OFFLOAD
4139	struct sge_ofld_rxq *ofld_rxq;
4140#endif
4141#ifdef DEV_NETMAP
4142	struct sge_nm_rxq *nm_rxq;
4143#endif
4144
4145	/*
4146	 * Setup interrupts.
4147	 */
4148	irq = &sc->irq[0];
4149	rid = sc->intr_type == INTR_INTX ? 0 : 1;
4150	if (sc->intr_count == 1)
4151		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
4152
4153	/* Multiple interrupts. */
4154	if (sc->flags & IS_VF)
4155		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
4156		    ("%s: too few intr.", __func__));
4157	else
4158		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
4159		    ("%s: too few intr.", __func__));
4160
4161	/* The first one is always error intr on PFs */
4162	if (!(sc->flags & IS_VF)) {
4163		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
4164		if (rc != 0)
4165			return (rc);
4166		irq++;
4167		rid++;
4168	}
4169
4170	/* The second one is always the firmware event queue (first on VFs) */
4171	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
4172	if (rc != 0)
4173		return (rc);
4174	irq++;
4175	rid++;
4176
4177	for_each_port(sc, p) {
4178		pi = sc->port[p];
4179		for_each_vi(pi, v, vi) {
4180			vi->first_intr = rid - 1;
4181
4182			if (vi->nnmrxq > 0) {
4183				int n = max(vi->nrxq, vi->nnmrxq);
4184
4185				MPASS(vi->flags & INTR_RXQ);
4186
4187				rxq = &sge->rxq[vi->first_rxq];
4188#ifdef DEV_NETMAP
4189				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
4190#endif
4191				for (q = 0; q < n; q++) {
4192					snprintf(s, sizeof(s), "%x%c%x", p,
4193					    'a' + v, q);
4194					if (q < vi->nrxq)
4195						irq->rxq = rxq++;
4196#ifdef DEV_NETMAP
4197					if (q < vi->nnmrxq)
4198						irq->nm_rxq = nm_rxq++;
4199#endif
4200					rc = t4_alloc_irq(sc, irq, rid,
4201					    t4_vi_intr, irq, s);
4202					if (rc != 0)
4203						return (rc);
4204					irq++;
4205					rid++;
4206					vi->nintr++;
4207				}
4208			} else if (vi->flags & INTR_RXQ) {
4209				for_each_rxq(vi, q, rxq) {
4210					snprintf(s, sizeof(s), "%x%c%x", p,
4211					    'a' + v, q);
4212					rc = t4_alloc_irq(sc, irq, rid,
4213					    t4_intr, rxq, s);
4214					if (rc != 0)
4215						return (rc);
4216					irq++;
4217					rid++;
4218					vi->nintr++;
4219				}
4220			}
4221#ifdef TCP_OFFLOAD
4222			if (vi->flags & INTR_OFLD_RXQ) {
4223				for_each_ofld_rxq(vi, q, ofld_rxq) {
4224					snprintf(s, sizeof(s), "%x%c%x", p,
4225					    'A' + v, q);
4226					rc = t4_alloc_irq(sc, irq, rid,
4227					    t4_intr, ofld_rxq, s);
4228					if (rc != 0)
4229						return (rc);
4230					irq++;
4231					rid++;
4232					vi->nintr++;
4233				}
4234			}
4235#endif
4236		}
4237	}
4238	MPASS(irq == &sc->irq[sc->intr_count]);
4239
4240	return (0);
4241}
4242
4243int
4244adapter_full_init(struct adapter *sc)
4245{
4246	int rc, i;
4247#ifdef RSS
4248	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4249	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
4250#endif
4251
4252	ASSERT_SYNCHRONIZED_OP(sc);
4253	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4254	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
4255	    ("%s: FULL_INIT_DONE already", __func__));
4256
4257	/*
4258	 * queues that belong to the adapter (not any particular port).
4259	 */
4260	rc = t4_setup_adapter_queues(sc);
4261	if (rc != 0)
4262		goto done;
4263
4264	for (i = 0; i < nitems(sc->tq); i++) {
4265		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
4266		    taskqueue_thread_enqueue, &sc->tq[i]);
4267		if (sc->tq[i] == NULL) {
4268			device_printf(sc->dev,
4269			    "failed to allocate task queue %d\n", i);
4270			rc = ENOMEM;
4271			goto done;
4272		}
4273		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
4274		    device_get_nameunit(sc->dev), i);
4275	}
4276#ifdef RSS
4277	MPASS(RSS_KEYSIZE == 40);
4278	rss_getkey((void *)&raw_rss_key[0]);
4279	for (i = 0; i < nitems(rss_key); i++) {
4280		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
4281	}
4282	t4_write_rss_key(sc, &rss_key[0], -1, 1);
4283#endif
4284
4285	if (!(sc->flags & IS_VF))
4286		t4_intr_enable(sc);
4287	sc->flags |= FULL_INIT_DONE;
4288done:
4289	if (rc != 0)
4290		adapter_full_uninit(sc);
4291
4292	return (rc);
4293}
4294
4295int
4296adapter_full_uninit(struct adapter *sc)
4297{
4298	int i;
4299
4300	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
4301
4302	t4_teardown_adapter_queues(sc);
4303
4304	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
4305		taskqueue_free(sc->tq[i]);
4306		sc->tq[i] = NULL;
4307	}
4308
4309	sc->flags &= ~FULL_INIT_DONE;
4310
4311	return (0);
4312}
4313
4314#ifdef RSS
4315#define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
4316    RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
4317    RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
4318    RSS_HASHTYPE_RSS_UDP_IPV6)
4319
4320/* Translates kernel hash types to hardware. */
4321static int
4322hashconfig_to_hashen(int hashconfig)
4323{
4324	int hashen = 0;
4325
4326	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
4327		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
4328	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
4329		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
4330	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
4331		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4332		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4333	}
4334	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
4335		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
4336		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4337	}
4338	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
4339		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
4340	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
4341		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
4342
4343	return (hashen);
4344}
4345
4346/* Translates hardware hash types to kernel. */
4347static int
4348hashen_to_hashconfig(int hashen)
4349{
4350	int hashconfig = 0;
4351
4352	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
4353		/*
4354		 * If UDP hashing was enabled it must have been enabled for
4355		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
4356		 * enabling any 4-tuple hash is nonsense configuration.
4357		 */
4358		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4359		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
4360
4361		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4362			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
4363		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4364			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
4365	}
4366	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
4367		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
4368	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
4369		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
4370	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
4371		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
4372	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
4373		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
4374
4375	return (hashconfig);
4376}
4377#endif
4378
4379int
4380vi_full_init(struct vi_info *vi)
4381{
4382	struct adapter *sc = vi->pi->adapter;
4383	struct ifnet *ifp = vi->ifp;
4384	uint16_t *rss;
4385	struct sge_rxq *rxq;
4386	int rc, i, j, hashen;
4387#ifdef RSS
4388	int nbuckets = rss_getnumbuckets();
4389	int hashconfig = rss_gethashconfig();
4390	int extra;
4391#endif
4392
4393	ASSERT_SYNCHRONIZED_OP(sc);
4394	KASSERT((vi->flags & VI_INIT_DONE) == 0,
4395	    ("%s: VI_INIT_DONE already", __func__));
4396
4397	sysctl_ctx_init(&vi->ctx);
4398	vi->flags |= VI_SYSCTL_CTX;
4399
4400	/*
4401	 * Allocate tx/rx/fl queues for this VI.
4402	 */
4403	rc = t4_setup_vi_queues(vi);
4404	if (rc != 0)
4405		goto done;	/* error message displayed already */
4406
4407	/*
4408	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
4409	 */
4410	if (vi->nrxq > vi->rss_size) {
4411		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
4412		    "some queues will never receive traffic.\n", vi->nrxq,
4413		    vi->rss_size);
4414	} else if (vi->rss_size % vi->nrxq) {
4415		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
4416		    "expect uneven traffic distribution.\n", vi->nrxq,
4417		    vi->rss_size);
4418	}
4419#ifdef RSS
4420	if (vi->nrxq != nbuckets) {
4421		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
4422		    "performance will be impacted.\n", vi->nrxq, nbuckets);
4423	}
4424#endif
4425	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
4426	for (i = 0; i < vi->rss_size;) {
4427#ifdef RSS
4428		j = rss_get_indirection_to_bucket(i);
4429		j %= vi->nrxq;
4430		rxq = &sc->sge.rxq[vi->first_rxq + j];
4431		rss[i++] = rxq->iq.abs_id;
4432#else
4433		for_each_rxq(vi, j, rxq) {
4434			rss[i++] = rxq->iq.abs_id;
4435			if (i == vi->rss_size)
4436				break;
4437		}
4438#endif
4439	}
4440
4441	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
4442	    vi->rss_size);
4443	if (rc != 0) {
4444		if_printf(ifp, "rss_config failed: %d\n", rc);
4445		goto done;
4446	}
4447
4448#ifdef RSS
4449	hashen = hashconfig_to_hashen(hashconfig);
4450
4451	/*
4452	 * We may have had to enable some hashes even though the global config
4453	 * wants them disabled.  This is a potential problem that must be
4454	 * reported to the user.
4455	 */
4456	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
4457
4458	/*
4459	 * If we consider only the supported hash types, then the enabled hashes
4460	 * are a superset of the requested hashes.  In other words, there cannot
4461	 * be any supported hash that was requested but not enabled, but there
4462	 * can be hashes that were not requested but had to be enabled.
4463	 */
4464	extra &= SUPPORTED_RSS_HASHTYPES;
4465	MPASS((extra & hashconfig) == 0);
4466
4467	if (extra) {
4468		if_printf(ifp,
4469		    "global RSS config (0x%x) cannot be accomodated.\n",
4470		    hashconfig);
4471	}
4472	if (extra & RSS_HASHTYPE_RSS_IPV4)
4473		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
4474	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
4475		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
4476	if (extra & RSS_HASHTYPE_RSS_IPV6)
4477		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
4478	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
4479		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
4480	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
4481		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
4482	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
4483		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
4484#else
4485	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
4486	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
4487	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4488	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
4489#endif
4490	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0], 0, 0);
4491	if (rc != 0) {
4492		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
4493		goto done;
4494	}
4495
4496	vi->rss = rss;
4497	vi->flags |= VI_INIT_DONE;
4498done:
4499	if (rc != 0)
4500		vi_full_uninit(vi);
4501
4502	return (rc);
4503}
4504
4505/*
4506 * Idempotent.
4507 */
4508int
4509vi_full_uninit(struct vi_info *vi)
4510{
4511	struct port_info *pi = vi->pi;
4512	struct adapter *sc = pi->adapter;
4513	int i;
4514	struct sge_rxq *rxq;
4515	struct sge_txq *txq;
4516#ifdef TCP_OFFLOAD
4517	struct sge_ofld_rxq *ofld_rxq;
4518	struct sge_wrq *ofld_txq;
4519#endif
4520
4521	if (vi->flags & VI_INIT_DONE) {
4522
4523		/* Need to quiesce queues.  */
4524
4525		/* XXX: Only for the first VI? */
4526		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
4527			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
4528
4529		for_each_txq(vi, i, txq) {
4530			quiesce_txq(sc, txq);
4531		}
4532
4533#ifdef TCP_OFFLOAD
4534		for_each_ofld_txq(vi, i, ofld_txq) {
4535			quiesce_wrq(sc, ofld_txq);
4536		}
4537#endif
4538
4539		for_each_rxq(vi, i, rxq) {
4540			quiesce_iq(sc, &rxq->iq);
4541			quiesce_fl(sc, &rxq->fl);
4542		}
4543
4544#ifdef TCP_OFFLOAD
4545		for_each_ofld_rxq(vi, i, ofld_rxq) {
4546			quiesce_iq(sc, &ofld_rxq->iq);
4547			quiesce_fl(sc, &ofld_rxq->fl);
4548		}
4549#endif
4550		free(vi->rss, M_CXGBE);
4551		free(vi->nm_rss, M_CXGBE);
4552	}
4553
4554	t4_teardown_vi_queues(vi);
4555	vi->flags &= ~VI_INIT_DONE;
4556
4557	return (0);
4558}
4559
4560static void
4561quiesce_txq(struct adapter *sc, struct sge_txq *txq)
4562{
4563	struct sge_eq *eq = &txq->eq;
4564	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
4565
4566	(void) sc;	/* unused */
4567
4568#ifdef INVARIANTS
4569	TXQ_LOCK(txq);
4570	MPASS((eq->flags & EQ_ENABLED) == 0);
4571	TXQ_UNLOCK(txq);
4572#endif
4573
4574	/* Wait for the mp_ring to empty. */
4575	while (!mp_ring_is_idle(txq->r)) {
4576		mp_ring_check_drainage(txq->r, 0);
4577		pause("rquiesce", 1);
4578	}
4579
4580	/* Then wait for the hardware to finish. */
4581	while (spg->cidx != htobe16(eq->pidx))
4582		pause("equiesce", 1);
4583
4584	/* Finally, wait for the driver to reclaim all descriptors. */
4585	while (eq->cidx != eq->pidx)
4586		pause("dquiesce", 1);
4587}
4588
4589static void
4590quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
4591{
4592
4593	/* XXXTX */
4594}
4595
4596static void
4597quiesce_iq(struct adapter *sc, struct sge_iq *iq)
4598{
4599	(void) sc;	/* unused */
4600
4601	/* Synchronize with the interrupt handler */
4602	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
4603		pause("iqfree", 1);
4604}
4605
4606static void
4607quiesce_fl(struct adapter *sc, struct sge_fl *fl)
4608{
4609	mtx_lock(&sc->sfl_lock);
4610	FL_LOCK(fl);
4611	fl->flags |= FL_DOOMED;
4612	FL_UNLOCK(fl);
4613	callout_stop(&sc->sfl_callout);
4614	mtx_unlock(&sc->sfl_lock);
4615
4616	KASSERT((fl->flags & FL_STARVING) == 0,
4617	    ("%s: still starving", __func__));
4618}
4619
4620static int
4621t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
4622    driver_intr_t *handler, void *arg, char *name)
4623{
4624	int rc;
4625
4626	irq->rid = rid;
4627	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
4628	    RF_SHAREABLE | RF_ACTIVE);
4629	if (irq->res == NULL) {
4630		device_printf(sc->dev,
4631		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
4632		return (ENOMEM);
4633	}
4634
4635	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
4636	    NULL, handler, arg, &irq->tag);
4637	if (rc != 0) {
4638		device_printf(sc->dev,
4639		    "failed to setup interrupt for rid %d, name %s: %d\n",
4640		    rid, name, rc);
4641	} else if (name)
4642		bus_describe_intr(sc->dev, irq->res, irq->tag, name);
4643
4644	return (rc);
4645}
4646
4647static int
4648t4_free_irq(struct adapter *sc, struct irq *irq)
4649{
4650	if (irq->tag)
4651		bus_teardown_intr(sc->dev, irq->res, irq->tag);
4652	if (irq->res)
4653		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
4654
4655	bzero(irq, sizeof(*irq));
4656
4657	return (0);
4658}
4659
4660static void
4661get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
4662{
4663
4664	regs->version = chip_id(sc) | chip_rev(sc) << 10;
4665	t4_get_regs(sc, buf, regs->len);
4666}
4667
4668#define	A_PL_INDIR_CMD	0x1f8
4669
4670#define	S_PL_AUTOINC	31
4671#define	M_PL_AUTOINC	0x1U
4672#define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
4673#define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
4674
4675#define	S_PL_VFID	20
4676#define	M_PL_VFID	0xffU
4677#define	V_PL_VFID(x)	((x) << S_PL_VFID)
4678#define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
4679
4680#define	S_PL_ADDR	0
4681#define	M_PL_ADDR	0xfffffU
4682#define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
4683#define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
4684
4685#define	A_PL_INDIR_DATA	0x1fc
4686
4687static uint64_t
4688read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
4689{
4690	u32 stats[2];
4691
4692	mtx_assert(&sc->reg_lock, MA_OWNED);
4693	if (sc->flags & IS_VF) {
4694		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
4695		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
4696	} else {
4697		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4698		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4699		    V_PL_ADDR(VF_MPS_REG(reg)));
4700		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
4701		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
4702	}
4703	return (((uint64_t)stats[1]) << 32 | stats[0]);
4704}
4705
4706static void
4707t4_get_vi_stats(struct adapter *sc, unsigned int viid,
4708    struct fw_vi_stats_vf *stats)
4709{
4710
4711#define GET_STAT(name) \
4712	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
4713
4714	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
4715	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
4716	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
4717	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
4718	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
4719	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
4720	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
4721	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
4722	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
4723	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
4724	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
4725	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
4726	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
4727	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
4728	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
4729	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
4730
4731#undef GET_STAT
4732}
4733
4734static void
4735t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
4736{
4737	int reg;
4738
4739	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4740	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4741	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
4742	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
4743	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
4744		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
4745}
4746
4747static void
4748vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
4749{
4750	struct ifnet *ifp = vi->ifp;
4751	struct sge_txq *txq;
4752	int i, drops;
4753	struct fw_vi_stats_vf *s = &vi->stats;
4754	struct timeval tv;
4755	const struct timeval interval = {0, 250000};	/* 250ms */
4756
4757	if (!(vi->flags & VI_INIT_DONE))
4758		return;
4759
4760	getmicrotime(&tv);
4761	timevalsub(&tv, &interval);
4762	if (timevalcmp(&tv, &vi->last_refreshed, <))
4763		return;
4764
4765	mtx_lock(&sc->reg_lock);
4766	t4_get_vi_stats(sc, vi->viid, &vi->stats);
4767
4768	ifp->if_ipackets = s->rx_bcast_frames + s->rx_mcast_frames +
4769	    s->rx_ucast_frames;
4770	ifp->if_ierrors = s->rx_err_frames;
4771	ifp->if_opackets = s->tx_bcast_frames + s->tx_mcast_frames +
4772	    s->tx_ucast_frames + s->tx_offload_frames;
4773	ifp->if_oerrors = s->tx_drop_frames;
4774	ifp->if_ibytes = s->rx_bcast_bytes + s->rx_mcast_bytes +
4775	    s->rx_ucast_bytes;
4776	ifp->if_obytes = s->tx_bcast_bytes + s->tx_mcast_bytes +
4777	    s->tx_ucast_bytes + s->tx_offload_bytes;
4778	ifp->if_imcasts = s->rx_mcast_frames;
4779	ifp->if_omcasts = s->tx_mcast_frames;
4780
4781	drops = 0;
4782	for_each_txq(vi, i, txq)
4783		drops += counter_u64_fetch(txq->r->drops);
4784	ifp->if_snd.ifq_drops = drops;
4785
4786	getmicrotime(&vi->last_refreshed);
4787	mtx_unlock(&sc->reg_lock);
4788}
4789
4790static void
4791cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
4792{
4793	struct vi_info *vi = &pi->vi[0];
4794	struct ifnet *ifp = vi->ifp;
4795	struct sge_txq *txq;
4796	int i, drops;
4797	struct port_stats *s = &pi->stats;
4798	struct timeval tv;
4799	const struct timeval interval = {0, 250000};	/* 250ms */
4800
4801	getmicrotime(&tv);
4802	timevalsub(&tv, &interval);
4803	if (timevalcmp(&tv, &pi->last_refreshed, <))
4804		return;
4805
4806	t4_get_port_stats(sc, pi->tx_chan, s);
4807
4808	ifp->if_opackets = s->tx_frames;
4809	ifp->if_ipackets = s->rx_frames;
4810	ifp->if_obytes = s->tx_octets;
4811	ifp->if_ibytes = s->rx_octets;
4812	ifp->if_omcasts = s->tx_mcast_frames;
4813	ifp->if_imcasts = s->rx_mcast_frames;
4814	ifp->if_iqdrops = s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
4815	    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
4816	    s->rx_trunc3;
4817	for (i = 0; i < sc->chip_params->nchan; i++) {
4818		if (pi->rx_chan_map & (1 << i)) {
4819			uint32_t v;
4820
4821			mtx_lock(&sc->reg_lock);
4822			t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
4823			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
4824			mtx_unlock(&sc->reg_lock);
4825			ifp->if_iqdrops += v;
4826		}
4827	}
4828
4829	drops = s->tx_drop;
4830	for_each_txq(vi, i, txq)
4831		drops += counter_u64_fetch(txq->r->drops);
4832	ifp->if_snd.ifq_drops = drops;
4833
4834	ifp->if_oerrors = s->tx_error_frames;
4835	ifp->if_ierrors = s->rx_jabber + s->rx_runt + s->rx_too_long +
4836	    s->rx_fcs_err + s->rx_len_err;
4837
4838	getmicrotime(&pi->last_refreshed);
4839}
4840
4841static void
4842cxgbe_tick(void *arg)
4843{
4844	struct port_info *pi = arg;
4845	struct adapter *sc = pi->adapter;
4846
4847	PORT_LOCK_ASSERT_OWNED(pi);
4848	cxgbe_refresh_stats(sc, pi);
4849
4850	callout_schedule(&pi->tick, hz);
4851}
4852
4853void
4854vi_tick(void *arg)
4855{
4856	struct vi_info *vi = arg;
4857	struct adapter *sc = vi->pi->adapter;
4858
4859	vi_refresh_stats(sc, vi);
4860
4861	callout_schedule(&vi->tick, hz);
4862}
4863
4864static void
4865cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
4866{
4867	struct ifnet *vlan;
4868
4869	if (arg != ifp || ifp->if_type != IFT_ETHER)
4870		return;
4871
4872	vlan = VLAN_DEVAT(ifp, vid);
4873	VLAN_SETCOOKIE(vlan, ifp);
4874}
4875
4876/*
4877 * Should match fw_caps_config_<foo> enums in t4fw_interface.h
4878 */
4879static char *caps_decoder[] = {
4880	"\20\001IPMI\002NCSI",				/* 0: NBM */
4881	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
4882	"\20\001INGRESS\002EGRESS",			/* 2: switch */
4883	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
4884	    "\006HASHFILTER\007ETHOFLD",
4885	"\20\001TOE",					/* 4: TOE */
4886	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
4887	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
4888	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
4889	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
4890	    "\007T10DIF"
4891	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
4892	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
4893	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
4894		    "\004PO_INITIATOR\005PO_TARGET",
4895};
4896
4897void
4898t4_sysctls(struct adapter *sc)
4899{
4900	struct sysctl_ctx_list *ctx;
4901	struct sysctl_oid *oid;
4902	struct sysctl_oid_list *children, *c0;
4903	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
4904
4905	ctx = device_get_sysctl_ctx(sc->dev);
4906
4907	/*
4908	 * dev.t4nex.X.
4909	 */
4910	oid = device_get_sysctl_tree(sc->dev);
4911	c0 = children = SYSCTL_CHILDREN(oid);
4912
4913	sc->sc_do_rxcopy = 1;
4914	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
4915	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
4916
4917	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
4918	    sc->params.nports, "# of ports");
4919
4920	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
4921	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
4922	    sysctl_bitfield, "A", "available doorbells");
4923
4924	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
4925	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
4926
4927	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
4928	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
4929	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
4930	    "interrupt holdoff timer values (us)");
4931
4932	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
4933	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
4934	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
4935	    "interrupt holdoff packet counter values");
4936
4937	t4_sge_sysctls(sc, ctx, children);
4938
4939	sc->lro_timeout = 100;
4940	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
4941	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
4942
4943	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
4944	    &sc->debug_flags, 0, "flags to enable runtime debugging");
4945
4946	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
4947	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
4948
4949	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
4950	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
4951
4952	if (sc->flags & IS_VF)
4953		return;
4954
4955	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
4956	    NULL, chip_rev(sc), "chip hardware revision");
4957
4958	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
4959	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
4960
4961	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
4962	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
4963
4964	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
4965	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
4966
4967	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
4968	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
4969
4970	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
4971	    sc->er_version, 0, "expansion ROM version");
4972
4973	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
4974	    sc->bs_version, 0, "bootstrap firmware version");
4975
4976	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
4977	    NULL, sc->params.scfg_vers, "serial config version");
4978
4979	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
4980	    NULL, sc->params.vpd_vers, "VPD version");
4981
4982	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
4983	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
4984
4985	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
4986	    sc->cfcsum, "config file checksum");
4987
4988#define SYSCTL_CAP(name, n, text) \
4989	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
4990	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
4991	    sysctl_bitfield, "A", "available " text " capabilities")
4992
4993	SYSCTL_CAP(nbmcaps, 0, "NBM");
4994	SYSCTL_CAP(linkcaps, 1, "link");
4995	SYSCTL_CAP(switchcaps, 2, "switch");
4996	SYSCTL_CAP(niccaps, 3, "NIC");
4997	SYSCTL_CAP(toecaps, 4, "TCP offload");
4998	SYSCTL_CAP(rdmacaps, 5, "RDMA");
4999	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5000	SYSCTL_CAP(cryptocaps, 7, "crypto");
5001	SYSCTL_CAP(fcoecaps, 8, "FCoE");
5002#undef SYSCTL_CAP
5003
5004	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5005	    NULL, sc->tids.nftids, "number of filters");
5006
5007	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5008	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5009	    "chip temperature (in Celsius)");
5010
5011#ifdef SBUF_DRAIN
5012	/*
5013	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
5014	 */
5015	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5016	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5017	    "logs and miscellaneous information");
5018	children = SYSCTL_CHILDREN(oid);
5019
5020	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
5021	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5022	    sysctl_cctrl, "A", "congestion control");
5023
5024	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
5025	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5026	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
5027
5028	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
5029	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
5030	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
5031
5032	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
5033	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
5034	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
5035
5036	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
5037	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
5038	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
5039
5040	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
5041	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
5042	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
5043
5044	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
5045	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
5046	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
5047
5048	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
5049	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5050	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
5051	    "A", "CIM logic analyzer");
5052
5053	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
5054	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5055	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
5056
5057	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
5058	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
5059	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
5060
5061	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
5062	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
5063	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
5064
5065	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
5066	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
5067	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
5068
5069	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5070	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5071	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5072
5073	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5074	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5075	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5076
5077	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5078	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5079	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5080
5081	if (chip_id(sc) > CHELSIO_T4) {
5082		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5083		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5084		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5085
5086		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5087		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5088		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5089	}
5090
5091	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5092	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5093	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5094
5095	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5096	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5097	    sysctl_cim_qcfg, "A", "CIM queue configuration");
5098
5099	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5100	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5101	    sysctl_cpl_stats, "A", "CPL statistics");
5102
5103	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5104	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5105	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5106
5107	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5108	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5109	    sysctl_devlog, "A", "firmware's device log");
5110
5111	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5112	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5113	    sysctl_fcoe_stats, "A", "FCoE statistics");
5114
5115	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
5116	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5117	    sysctl_hw_sched, "A", "hardware scheduler ");
5118
5119	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
5120	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5121	    sysctl_l2t, "A", "hardware L2 table");
5122
5123	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
5124	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5125	    sysctl_lb_stats, "A", "loopback statistics");
5126
5127	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
5128	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5129	    sysctl_meminfo, "A", "memory regions");
5130
5131	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
5132	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5133	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
5134	    "A", "MPS TCAM entries");
5135
5136	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
5137	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5138	    sysctl_path_mtus, "A", "path MTUs");
5139
5140	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
5141	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5142	    sysctl_pm_stats, "A", "PM statistics");
5143
5144	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
5145	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5146	    sysctl_rdma_stats, "A", "RDMA statistics");
5147
5148	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
5149	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5150	    sysctl_tcp_stats, "A", "TCP statistics");
5151
5152	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
5153	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5154	    sysctl_tids, "A", "TID information");
5155
5156	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
5157	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5158	    sysctl_tp_err_stats, "A", "TP error statistics");
5159
5160	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
5161	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
5162	    "TP logic analyzer event capture mask");
5163
5164	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
5165	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5166	    sysctl_tp_la, "A", "TP logic analyzer");
5167
5168	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
5169	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5170	    sysctl_tx_rate, "A", "Tx rate");
5171
5172	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
5173	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5174	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
5175
5176	if (chip_id(sc) >= CHELSIO_T5) {
5177		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
5178		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5179		    sysctl_wcwr_stats, "A", "write combined work requests");
5180	}
5181#endif
5182
5183#ifdef TCP_OFFLOAD
5184	if (is_offload(sc)) {
5185		/*
5186		 * dev.t4nex.X.toe.
5187		 */
5188		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
5189		    NULL, "TOE parameters");
5190		children = SYSCTL_CHILDREN(oid);
5191
5192		sc->tt.sndbuf = 256 * 1024;
5193		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
5194		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
5195
5196		sc->tt.ddp = 0;
5197		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
5198		    &sc->tt.ddp, 0, "DDP allowed");
5199
5200		sc->tt.indsz = G_INDICATESIZE(t4_read_reg(sc, A_TP_PARA_REG5));
5201		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "indsz", CTLFLAG_RW,
5202		    &sc->tt.indsz, 0, "DDP max indicate size allowed");
5203
5204		sc->tt.ddp_thres =
5205		    G_RXCOALESCESIZE(t4_read_reg(sc, A_TP_PARA_REG2));
5206		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp_thres", CTLFLAG_RW,
5207		    &sc->tt.ddp_thres, 0, "DDP threshold");
5208
5209		sc->tt.rx_coalesce = 1;
5210		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
5211		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
5212
5213		sc->tt.tx_align = 1;
5214		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
5215		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
5216
5217		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
5218		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
5219		    "TP timer tick (us)");
5220
5221		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
5222		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
5223		    "TCP timestamp tick (us)");
5224
5225		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
5226		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
5227		    "DACK tick (us)");
5228
5229		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
5230		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
5231		    "IU", "DACK timer (us)");
5232
5233		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
5234		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
5235		    sysctl_tp_timer, "LU", "Retransmit min (us)");
5236
5237		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
5238		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
5239		    sysctl_tp_timer, "LU", "Retransmit max (us)");
5240
5241		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
5242		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
5243		    sysctl_tp_timer, "LU", "Persist timer min (us)");
5244
5245		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
5246		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
5247		    sysctl_tp_timer, "LU", "Persist timer max (us)");
5248
5249		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
5250		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
5251		    sysctl_tp_timer, "LU", "Keepidle idle timer (us)");
5252
5253		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_intvl",
5254		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
5255		    sysctl_tp_timer, "LU", "Keepidle interval (us)");
5256
5257		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
5258		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
5259		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
5260
5261		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
5262		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
5263		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
5264	}
5265#endif
5266}
5267
5268void
5269vi_sysctls(struct vi_info *vi)
5270{
5271	struct sysctl_ctx_list *ctx;
5272	struct sysctl_oid *oid;
5273	struct sysctl_oid_list *children;
5274
5275	ctx = device_get_sysctl_ctx(vi->dev);
5276
5277	/*
5278	 * dev.v?(cxgbe|cxl).X.
5279	 */
5280	oid = device_get_sysctl_tree(vi->dev);
5281	children = SYSCTL_CHILDREN(oid);
5282
5283	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
5284	    vi->viid, "VI identifer");
5285	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
5286	    &vi->nrxq, 0, "# of rx queues");
5287	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
5288	    &vi->ntxq, 0, "# of tx queues");
5289	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
5290	    &vi->first_rxq, 0, "index of first rx queue");
5291	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
5292	    &vi->first_txq, 0, "index of first tx queue");
5293	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
5294	    vi->rss_size, "size of RSS indirection table");
5295
5296	if (IS_MAIN_VI(vi)) {
5297		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
5298		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
5299		    "Reserve queue 0 for non-flowid packets");
5300	}
5301
5302#ifdef TCP_OFFLOAD
5303	if (vi->nofldrxq != 0) {
5304		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
5305		    &vi->nofldrxq, 0,
5306		    "# of rx queues for offloaded TCP connections");
5307		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
5308		    &vi->nofldtxq, 0,
5309		    "# of tx queues for offloaded TCP connections");
5310		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
5311		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
5312		    "index of first TOE rx queue");
5313		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
5314		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
5315		    "index of first TOE tx queue");
5316	}
5317#endif
5318#ifdef DEV_NETMAP
5319	if (vi->nnmrxq != 0) {
5320		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
5321		    &vi->nnmrxq, 0, "# of netmap rx queues");
5322		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
5323		    &vi->nnmtxq, 0, "# of netmap tx queues");
5324		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
5325		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
5326		    "index of first netmap rx queue");
5327		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
5328		    CTLFLAG_RD, &vi->first_nm_txq, 0,
5329		    "index of first netmap tx queue");
5330	}
5331#endif
5332
5333	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
5334	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
5335	    "holdoff timer index");
5336	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
5337	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
5338	    "holdoff packet counter index");
5339
5340	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
5341	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
5342	    "rx queue size");
5343	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
5344	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
5345	    "tx queue size");
5346}
5347
5348static void
5349cxgbe_sysctls(struct port_info *pi)
5350{
5351	struct sysctl_ctx_list *ctx;
5352	struct sysctl_oid *oid;
5353	struct sysctl_oid_list *children, *children2;
5354	struct adapter *sc = pi->adapter;
5355	int i;
5356	char name[16];
5357
5358	ctx = device_get_sysctl_ctx(pi->dev);
5359
5360	/*
5361	 * dev.cxgbe.X.
5362	 */
5363	oid = device_get_sysctl_tree(pi->dev);
5364	children = SYSCTL_CHILDREN(oid);
5365
5366	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
5367	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
5368	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
5369		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
5370		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
5371		    "PHY temperature (in Celsius)");
5372		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
5373		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
5374		    "PHY firmware version");
5375	}
5376
5377	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
5378	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
5379	    "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
5380	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
5381	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
5382	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
5383	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
5384	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
5385	    "autonegotiation (-1 = not supported)");
5386
5387	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
5388	    port_top_speed(pi), "max speed (in Gbps)");
5389
5390	if (sc->flags & IS_VF)
5391		return;
5392
5393	/*
5394	 * dev.(cxgbe|cxl).X.tc.
5395	 */
5396	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
5397	    "Tx scheduler traffic classes (cl_rl)");
5398	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
5399		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
5400
5401		snprintf(name, sizeof(name), "%d", i);
5402		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
5403		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
5404		    "traffic class"));
5405		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
5406		    &tc->flags, 0, "flags");
5407		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
5408		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
5409#ifdef SBUF_DRAIN
5410		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
5411		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
5412		    sysctl_tc_params, "A", "traffic class parameters");
5413#endif
5414	}
5415
5416	/*
5417	 * dev.cxgbe.X.stats.
5418	 */
5419	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
5420	    NULL, "port statistics");
5421	children = SYSCTL_CHILDREN(oid);
5422	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
5423	    &pi->tx_parse_error, 0,
5424	    "# of tx packets with invalid length or # of segments");
5425
5426#define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
5427	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
5428	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
5429	    sysctl_handle_t4_reg64, "QU", desc)
5430
5431	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
5432	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
5433	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
5434	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
5435	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
5436	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
5437	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
5438	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
5439	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
5440	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
5441	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
5442	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
5443	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
5444	    "# of tx frames in this range",
5445	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
5446	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
5447	    "# of tx frames in this range",
5448	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
5449	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
5450	    "# of tx frames in this range",
5451	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
5452	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
5453	    "# of tx frames in this range",
5454	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
5455	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
5456	    "# of tx frames in this range",
5457	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
5458	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
5459	    "# of tx frames in this range",
5460	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
5461	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
5462	    "# of tx frames in this range",
5463	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
5464	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
5465	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
5466	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
5467	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
5468	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
5469	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
5470	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5471	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5472	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5473	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5474	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5475	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5476	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5477	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5478	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5479	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5480	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5481	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5482	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5483	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5484
5485	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5486	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5487	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5488	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5489	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5490	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5491	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5492	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5493	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5494	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5495	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5496	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5497	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5498	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5499	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5500	    "# of frames received with bad FCS",
5501	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5502	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5503	    "# of frames received with length error",
5504	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5505	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5506	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5507	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5508	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5509	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5510	    "# of rx frames in this range",
5511	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5512	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5513	    "# of rx frames in this range",
5514	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5515	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5516	    "# of rx frames in this range",
5517	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5518	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5519	    "# of rx frames in this range",
5520	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5521	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5522	    "# of rx frames in this range",
5523	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5524	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5525	    "# of rx frames in this range",
5526	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5527	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5528	    "# of rx frames in this range",
5529	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5530	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5531	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5532	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5533	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5534	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5535	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5536	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5537	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5538	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5539	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5540	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5541	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5542	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5543	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5544	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5545	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5546	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5547	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5548
5549#undef SYSCTL_ADD_T4_REG64
5550
5551#define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5552	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5553	    &pi->stats.name, desc)
5554
5555	/* We get these from port_stats and they may be stale by upto 1s */
5556	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5557	    "# drops due to buffer-group 0 overflows");
5558	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5559	    "# drops due to buffer-group 1 overflows");
5560	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5561	    "# drops due to buffer-group 2 overflows");
5562	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5563	    "# drops due to buffer-group 3 overflows");
5564	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5565	    "# of buffer-group 0 truncated packets");
5566	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5567	    "# of buffer-group 1 truncated packets");
5568	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5569	    "# of buffer-group 2 truncated packets");
5570	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5571	    "# of buffer-group 3 truncated packets");
5572
5573#undef SYSCTL_ADD_T4_PORTSTAT
5574}
5575
5576static int
5577sysctl_int_array(SYSCTL_HANDLER_ARGS)
5578{
5579	int rc, *i, space = 0;
5580	struct sbuf sb;
5581
5582	sbuf_new(&sb, NULL, 32, SBUF_AUTOEXTEND);
5583	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
5584		if (space)
5585			sbuf_printf(&sb, " ");
5586		sbuf_printf(&sb, "%d", *i);
5587		space = 1;
5588	}
5589	sbuf_finish(&sb);
5590	rc = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
5591	sbuf_delete(&sb);
5592	return (rc);
5593}
5594
5595static int
5596sysctl_bitfield(SYSCTL_HANDLER_ARGS)
5597{
5598	int rc;
5599	struct sbuf *sb;
5600
5601	rc = sysctl_wire_old_buffer(req, 0);
5602	if (rc != 0)
5603		return(rc);
5604
5605	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5606	if (sb == NULL)
5607		return (ENOMEM);
5608
5609	sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
5610	rc = sbuf_finish(sb);
5611	sbuf_delete(sb);
5612
5613	return (rc);
5614}
5615
5616static int
5617sysctl_btphy(SYSCTL_HANDLER_ARGS)
5618{
5619	struct port_info *pi = arg1;
5620	int op = arg2;
5621	struct adapter *sc = pi->adapter;
5622	u_int v;
5623	int rc;
5624
5625	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
5626	if (rc)
5627		return (rc);
5628	/* XXX: magic numbers */
5629	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
5630	    &v);
5631	end_synchronized_op(sc, 0);
5632	if (rc)
5633		return (rc);
5634	if (op == 0)
5635		v /= 256;
5636
5637	rc = sysctl_handle_int(oidp, &v, 0, req);
5638	return (rc);
5639}
5640
5641static int
5642sysctl_noflowq(SYSCTL_HANDLER_ARGS)
5643{
5644	struct vi_info *vi = arg1;
5645	int rc, val;
5646
5647	val = vi->rsrv_noflowq;
5648	rc = sysctl_handle_int(oidp, &val, 0, req);
5649	if (rc != 0 || req->newptr == NULL)
5650		return (rc);
5651
5652	if ((val >= 1) && (vi->ntxq > 1))
5653		vi->rsrv_noflowq = 1;
5654	else
5655		vi->rsrv_noflowq = 0;
5656
5657	return (rc);
5658}
5659
5660static int
5661sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
5662{
5663	struct vi_info *vi = arg1;
5664	struct adapter *sc = vi->pi->adapter;
5665	int idx, rc, i;
5666	struct sge_rxq *rxq;
5667#ifdef TCP_OFFLOAD
5668	struct sge_ofld_rxq *ofld_rxq;
5669#endif
5670	uint8_t v;
5671
5672	idx = vi->tmr_idx;
5673
5674	rc = sysctl_handle_int(oidp, &idx, 0, req);
5675	if (rc != 0 || req->newptr == NULL)
5676		return (rc);
5677
5678	if (idx < 0 || idx >= SGE_NTIMERS)
5679		return (EINVAL);
5680
5681	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5682	    "t4tmr");
5683	if (rc)
5684		return (rc);
5685
5686	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
5687	for_each_rxq(vi, i, rxq) {
5688#ifdef atomic_store_rel_8
5689		atomic_store_rel_8(&rxq->iq.intr_params, v);
5690#else
5691		rxq->iq.intr_params = v;
5692#endif
5693	}
5694#ifdef TCP_OFFLOAD
5695	for_each_ofld_rxq(vi, i, ofld_rxq) {
5696#ifdef atomic_store_rel_8
5697		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
5698#else
5699		ofld_rxq->iq.intr_params = v;
5700#endif
5701	}
5702#endif
5703	vi->tmr_idx = idx;
5704
5705	end_synchronized_op(sc, LOCK_HELD);
5706	return (0);
5707}
5708
5709static int
5710sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
5711{
5712	struct vi_info *vi = arg1;
5713	struct adapter *sc = vi->pi->adapter;
5714	int idx, rc;
5715
5716	idx = vi->pktc_idx;
5717
5718	rc = sysctl_handle_int(oidp, &idx, 0, req);
5719	if (rc != 0 || req->newptr == NULL)
5720		return (rc);
5721
5722	if (idx < -1 || idx >= SGE_NCOUNTERS)
5723		return (EINVAL);
5724
5725	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5726	    "t4pktc");
5727	if (rc)
5728		return (rc);
5729
5730	if (vi->flags & VI_INIT_DONE)
5731		rc = EBUSY; /* cannot be changed once the queues are created */
5732	else
5733		vi->pktc_idx = idx;
5734
5735	end_synchronized_op(sc, LOCK_HELD);
5736	return (rc);
5737}
5738
5739static int
5740sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
5741{
5742	struct vi_info *vi = arg1;
5743	struct adapter *sc = vi->pi->adapter;
5744	int qsize, rc;
5745
5746	qsize = vi->qsize_rxq;
5747
5748	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5749	if (rc != 0 || req->newptr == NULL)
5750		return (rc);
5751
5752	if (qsize < 128 || (qsize & 7))
5753		return (EINVAL);
5754
5755	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5756	    "t4rxqs");
5757	if (rc)
5758		return (rc);
5759
5760	if (vi->flags & VI_INIT_DONE)
5761		rc = EBUSY; /* cannot be changed once the queues are created */
5762	else
5763		vi->qsize_rxq = qsize;
5764
5765	end_synchronized_op(sc, LOCK_HELD);
5766	return (rc);
5767}
5768
5769static int
5770sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
5771{
5772	struct vi_info *vi = arg1;
5773	struct adapter *sc = vi->pi->adapter;
5774	int qsize, rc;
5775
5776	qsize = vi->qsize_txq;
5777
5778	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5779	if (rc != 0 || req->newptr == NULL)
5780		return (rc);
5781
5782	if (qsize < 128 || qsize > 65536)
5783		return (EINVAL);
5784
5785	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5786	    "t4txqs");
5787	if (rc)
5788		return (rc);
5789
5790	if (vi->flags & VI_INIT_DONE)
5791		rc = EBUSY; /* cannot be changed once the queues are created */
5792	else
5793		vi->qsize_txq = qsize;
5794
5795	end_synchronized_op(sc, LOCK_HELD);
5796	return (rc);
5797}
5798
5799static int
5800sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
5801{
5802	struct port_info *pi = arg1;
5803	struct adapter *sc = pi->adapter;
5804	struct link_config *lc = &pi->link_cfg;
5805	int rc;
5806
5807	if (req->newptr == NULL) {
5808		struct sbuf *sb;
5809		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
5810
5811		rc = sysctl_wire_old_buffer(req, 0);
5812		if (rc != 0)
5813			return(rc);
5814
5815		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5816		if (sb == NULL)
5817			return (ENOMEM);
5818
5819		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
5820		rc = sbuf_finish(sb);
5821		sbuf_delete(sb);
5822	} else {
5823		char s[2];
5824		int n;
5825
5826		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
5827		s[1] = 0;
5828
5829		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5830		if (rc != 0)
5831			return(rc);
5832
5833		if (s[1] != 0)
5834			return (EINVAL);
5835		if (s[0] < '0' || s[0] > '9')
5836			return (EINVAL);	/* not a number */
5837		n = s[0] - '0';
5838		if (n & ~(PAUSE_TX | PAUSE_RX))
5839			return (EINVAL);	/* some other bit is set too */
5840
5841		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5842		    "t4PAUSE");
5843		if (rc)
5844			return (rc);
5845		PORT_LOCK(pi);
5846		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
5847			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
5848			lc->requested_fc |= n;
5849			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5850			if (rc == 0) {
5851				lc->fc = lc->requested_fc;
5852				set_current_media(pi, &pi->media);
5853			}
5854		}
5855		PORT_UNLOCK(pi);
5856		end_synchronized_op(sc, 0);
5857	}
5858
5859	return (rc);
5860}
5861
5862static int
5863sysctl_fec(SYSCTL_HANDLER_ARGS)
5864{
5865	struct port_info *pi = arg1;
5866	struct adapter *sc = pi->adapter;
5867	struct link_config *lc = &pi->link_cfg;
5868	int rc;
5869
5870	if (req->newptr == NULL) {
5871		struct sbuf *sb;
5872		static char *bits = "\20\1RS\2BASER_RS\3RESERVED";
5873
5874		rc = sysctl_wire_old_buffer(req, 0);
5875		if (rc != 0)
5876			return(rc);
5877
5878		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5879		if (sb == NULL)
5880			return (ENOMEM);
5881
5882		sbuf_printf(sb, "%b", lc->fec & M_FW_PORT_CAP_FEC, bits);
5883		rc = sbuf_finish(sb);
5884		sbuf_delete(sb);
5885	} else {
5886		char s[2];
5887		int n;
5888
5889		s[0] = '0' + (lc->requested_fec & M_FW_PORT_CAP_FEC);
5890		s[1] = 0;
5891
5892		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5893		if (rc != 0)
5894			return(rc);
5895
5896		if (s[1] != 0)
5897			return (EINVAL);
5898		if (s[0] < '0' || s[0] > '9')
5899			return (EINVAL);	/* not a number */
5900		n = s[0] - '0';
5901		if (n & ~M_FW_PORT_CAP_FEC)
5902			return (EINVAL);	/* some other bit is set too */
5903		if (!powerof2(n))
5904			return (EINVAL);	/* one bit can be set at most */
5905
5906		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5907		    "t4fec");
5908		if (rc)
5909			return (rc);
5910		PORT_LOCK(pi);
5911		if ((lc->requested_fec & M_FW_PORT_CAP_FEC) != n) {
5912			lc->requested_fec = n &
5913			    G_FW_PORT_CAP_FEC(lc->supported);
5914			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5915			if (rc == 0) {
5916				lc->fec = lc->requested_fec;
5917			}
5918		}
5919		PORT_UNLOCK(pi);
5920		end_synchronized_op(sc, 0);
5921	}
5922
5923	return (rc);
5924}
5925
5926static int
5927sysctl_autoneg(SYSCTL_HANDLER_ARGS)
5928{
5929	struct port_info *pi = arg1;
5930	struct adapter *sc = pi->adapter;
5931	struct link_config *lc = &pi->link_cfg;
5932	int rc, val, old;
5933
5934	if (lc->supported & FW_PORT_CAP_ANEG)
5935		val = lc->requested_aneg == AUTONEG_ENABLE ? 1 : 0;
5936	else
5937		val = -1;
5938	rc = sysctl_handle_int(oidp, &val, 0, req);
5939	if (rc != 0 || req->newptr == NULL)
5940		return (rc);
5941	if (val == 0)
5942		val = AUTONEG_DISABLE;
5943	else if (val == 1)
5944		val = AUTONEG_ENABLE;
5945	else
5946		return (EINVAL);
5947
5948	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5949	    "t4aneg");
5950	if (rc)
5951		return (rc);
5952	PORT_LOCK(pi);
5953	if ((lc->supported & FW_PORT_CAP_ANEG) == 0) {
5954		rc = ENOTSUP;
5955		goto done;
5956	}
5957	if (lc->requested_aneg == val) {
5958		rc = 0;	/* no change, do nothing. */
5959		goto done;
5960	}
5961	old = lc->requested_aneg;
5962	lc->requested_aneg = val;
5963	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5964	if (rc != 0)
5965		lc->requested_aneg = old;
5966	else
5967		set_current_media(pi, &pi->media);
5968done:
5969	PORT_UNLOCK(pi);
5970	end_synchronized_op(sc, 0);
5971	return (rc);
5972}
5973
5974static int
5975sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
5976{
5977	struct adapter *sc = arg1;
5978	int reg = arg2;
5979	uint64_t val;
5980
5981	val = t4_read_reg64(sc, reg);
5982
5983	return (sysctl_handle_64(oidp, &val, 0, req));
5984}
5985
5986static int
5987sysctl_temperature(SYSCTL_HANDLER_ARGS)
5988{
5989	struct adapter *sc = arg1;
5990	int rc, t;
5991	uint32_t param, val;
5992
5993	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
5994	if (rc)
5995		return (rc);
5996	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5997	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
5998	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
5999	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
6000	end_synchronized_op(sc, 0);
6001	if (rc)
6002		return (rc);
6003
6004	/* unknown is returned as 0 but we display -1 in that case */
6005	t = val == 0 ? -1 : val;
6006
6007	rc = sysctl_handle_int(oidp, &t, 0, req);
6008	return (rc);
6009}
6010
6011#ifdef SBUF_DRAIN
6012static int
6013sysctl_cctrl(SYSCTL_HANDLER_ARGS)
6014{
6015	struct adapter *sc = arg1;
6016	struct sbuf *sb;
6017	int rc, i;
6018	uint16_t incr[NMTUS][NCCTRL_WIN];
6019	static const char *dec_fac[] = {
6020		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
6021		"0.9375"
6022	};
6023
6024	rc = sysctl_wire_old_buffer(req, 0);
6025	if (rc != 0)
6026		return (rc);
6027
6028	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6029	if (sb == NULL)
6030		return (ENOMEM);
6031
6032	t4_read_cong_tbl(sc, incr);
6033
6034	for (i = 0; i < NCCTRL_WIN; ++i) {
6035		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
6036		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
6037		    incr[5][i], incr[6][i], incr[7][i]);
6038		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
6039		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
6040		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
6041		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
6042	}
6043
6044	rc = sbuf_finish(sb);
6045	sbuf_delete(sb);
6046
6047	return (rc);
6048}
6049
6050static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
6051	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
6052	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
6053	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
6054};
6055
6056static int
6057sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
6058{
6059	struct adapter *sc = arg1;
6060	struct sbuf *sb;
6061	int rc, i, n, qid = arg2;
6062	uint32_t *buf, *p;
6063	char *qtype;
6064	u_int cim_num_obq = sc->chip_params->cim_num_obq;
6065
6066	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
6067	    ("%s: bad qid %d\n", __func__, qid));
6068
6069	if (qid < CIM_NUM_IBQ) {
6070		/* inbound queue */
6071		qtype = "IBQ";
6072		n = 4 * CIM_IBQ_SIZE;
6073		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6074		rc = t4_read_cim_ibq(sc, qid, buf, n);
6075	} else {
6076		/* outbound queue */
6077		qtype = "OBQ";
6078		qid -= CIM_NUM_IBQ;
6079		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
6080		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
6081		rc = t4_read_cim_obq(sc, qid, buf, n);
6082	}
6083
6084	if (rc < 0) {
6085		rc = -rc;
6086		goto done;
6087	}
6088	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
6089
6090	rc = sysctl_wire_old_buffer(req, 0);
6091	if (rc != 0)
6092		goto done;
6093
6094	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6095	if (sb == NULL) {
6096		rc = ENOMEM;
6097		goto done;
6098	}
6099
6100	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
6101	for (i = 0, p = buf; i < n; i += 16, p += 4)
6102		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
6103		    p[2], p[3]);
6104
6105	rc = sbuf_finish(sb);
6106	sbuf_delete(sb);
6107done:
6108	free(buf, M_CXGBE);
6109	return (rc);
6110}
6111
6112static int
6113sysctl_cim_la(SYSCTL_HANDLER_ARGS)
6114{
6115	struct adapter *sc = arg1;
6116	u_int cfg;
6117	struct sbuf *sb;
6118	uint32_t *buf, *p;
6119	int rc;
6120
6121	MPASS(chip_id(sc) <= CHELSIO_T5);
6122
6123	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6124	if (rc != 0)
6125		return (rc);
6126
6127	rc = sysctl_wire_old_buffer(req, 0);
6128	if (rc != 0)
6129		return (rc);
6130
6131	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6132	if (sb == NULL)
6133		return (ENOMEM);
6134
6135	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6136	    M_ZERO | M_WAITOK);
6137
6138	rc = -t4_cim_read_la(sc, buf, NULL);
6139	if (rc != 0)
6140		goto done;
6141
6142	sbuf_printf(sb, "Status   Data      PC%s",
6143	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6144	    "     LS0Stat  LS0Addr             LS0Data");
6145
6146	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
6147		if (cfg & F_UPDBGLACAPTPCONLY) {
6148			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
6149			    p[6], p[7]);
6150			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
6151			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
6152			    p[4] & 0xff, p[5] >> 8);
6153			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
6154			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6155			    p[1] & 0xf, p[2] >> 4);
6156		} else {
6157			sbuf_printf(sb,
6158			    "\n  %02x   %x%07x %x%07x %08x %08x "
6159			    "%08x%08x%08x%08x",
6160			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
6161			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
6162			    p[6], p[7]);
6163		}
6164	}
6165
6166	rc = sbuf_finish(sb);
6167	sbuf_delete(sb);
6168done:
6169	free(buf, M_CXGBE);
6170	return (rc);
6171}
6172
6173static int
6174sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
6175{
6176	struct adapter *sc = arg1;
6177	u_int cfg;
6178	struct sbuf *sb;
6179	uint32_t *buf, *p;
6180	int rc;
6181
6182	MPASS(chip_id(sc) > CHELSIO_T5);
6183
6184	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
6185	if (rc != 0)
6186		return (rc);
6187
6188	rc = sysctl_wire_old_buffer(req, 0);
6189	if (rc != 0)
6190		return (rc);
6191
6192	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6193	if (sb == NULL)
6194		return (ENOMEM);
6195
6196	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
6197	    M_ZERO | M_WAITOK);
6198
6199	rc = -t4_cim_read_la(sc, buf, NULL);
6200	if (rc != 0)
6201		goto done;
6202
6203	sbuf_printf(sb, "Status   Inst    Data      PC%s",
6204	    cfg & F_UPDBGLACAPTPCONLY ? "" :
6205	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
6206
6207	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
6208		if (cfg & F_UPDBGLACAPTPCONLY) {
6209			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
6210			    p[3] & 0xff, p[2], p[1], p[0]);
6211			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
6212			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
6213			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
6214			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
6215			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
6216			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
6217			    p[6] >> 16);
6218		} else {
6219			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
6220			    "%08x %08x %08x %08x %08x %08x",
6221			    (p[9] >> 16) & 0xff,
6222			    p[9] & 0xffff, p[8] >> 16,
6223			    p[8] & 0xffff, p[7] >> 16,
6224			    p[7] & 0xffff, p[6] >> 16,
6225			    p[2], p[1], p[0], p[5], p[4], p[3]);
6226		}
6227	}
6228
6229	rc = sbuf_finish(sb);
6230	sbuf_delete(sb);
6231done:
6232	free(buf, M_CXGBE);
6233	return (rc);
6234}
6235
6236static int
6237sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
6238{
6239	struct adapter *sc = arg1;
6240	u_int i;
6241	struct sbuf *sb;
6242	uint32_t *buf, *p;
6243	int rc;
6244
6245	rc = sysctl_wire_old_buffer(req, 0);
6246	if (rc != 0)
6247		return (rc);
6248
6249	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6250	if (sb == NULL)
6251		return (ENOMEM);
6252
6253	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
6254	    M_ZERO | M_WAITOK);
6255
6256	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
6257	p = buf;
6258
6259	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6260		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
6261		    p[1], p[0]);
6262	}
6263
6264	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
6265	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
6266		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
6267		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
6268		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
6269		    (p[1] >> 2) | ((p[2] & 3) << 30),
6270		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
6271		    p[0] & 1);
6272	}
6273
6274	rc = sbuf_finish(sb);
6275	sbuf_delete(sb);
6276	free(buf, M_CXGBE);
6277	return (rc);
6278}
6279
6280static int
6281sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
6282{
6283	struct adapter *sc = arg1;
6284	u_int i;
6285	struct sbuf *sb;
6286	uint32_t *buf, *p;
6287	int rc;
6288
6289	rc = sysctl_wire_old_buffer(req, 0);
6290	if (rc != 0)
6291		return (rc);
6292
6293	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6294	if (sb == NULL)
6295		return (ENOMEM);
6296
6297	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
6298	    M_ZERO | M_WAITOK);
6299
6300	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
6301	p = buf;
6302
6303	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
6304	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6305		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
6306		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
6307		    p[4], p[3], p[2], p[1], p[0]);
6308	}
6309
6310	sbuf_printf(sb, "\n\nCntl ID               Data");
6311	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
6312		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
6313		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
6314	}
6315
6316	rc = sbuf_finish(sb);
6317	sbuf_delete(sb);
6318	free(buf, M_CXGBE);
6319	return (rc);
6320}
6321
6322static int
6323sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
6324{
6325	struct adapter *sc = arg1;
6326	struct sbuf *sb;
6327	int rc, i;
6328	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6329	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
6330	uint16_t thres[CIM_NUM_IBQ];
6331	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
6332	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
6333	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
6334
6335	cim_num_obq = sc->chip_params->cim_num_obq;
6336	if (is_t4(sc)) {
6337		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
6338		obq_rdaddr = A_UP_OBQ_0_REALADDR;
6339	} else {
6340		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
6341		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
6342	}
6343	nq = CIM_NUM_IBQ + cim_num_obq;
6344
6345	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
6346	if (rc == 0)
6347		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
6348	if (rc != 0)
6349		return (rc);
6350
6351	t4_read_cimq_cfg(sc, base, size, thres);
6352
6353	rc = sysctl_wire_old_buffer(req, 0);
6354	if (rc != 0)
6355		return (rc);
6356
6357	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
6358	if (sb == NULL)
6359		return (ENOMEM);
6360
6361	sbuf_printf(sb,
6362	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
6363
6364	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
6365		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
6366		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
6367		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6368		    G_QUEREMFLITS(p[2]) * 16);
6369	for ( ; i < nq; i++, p += 4, wr += 2)
6370		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
6371		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
6372		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
6373		    G_QUEREMFLITS(p[2]) * 16);
6374
6375	rc = sbuf_finish(sb);
6376	sbuf_delete(sb);
6377
6378	return (rc);
6379}
6380
6381static int
6382sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
6383{
6384	struct adapter *sc = arg1;
6385	struct sbuf *sb;
6386	int rc;
6387	struct tp_cpl_stats stats;
6388
6389	rc = sysctl_wire_old_buffer(req, 0);
6390	if (rc != 0)
6391		return (rc);
6392
6393	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6394	if (sb == NULL)
6395		return (ENOMEM);
6396
6397	mtx_lock(&sc->reg_lock);
6398	t4_tp_get_cpl_stats(sc, &stats, 0);
6399	mtx_unlock(&sc->reg_lock);
6400
6401	if (sc->chip_params->nchan > 2) {
6402		sbuf_printf(sb, "                 channel 0  channel 1"
6403		    "  channel 2  channel 3");
6404		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
6405		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
6406		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
6407		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
6408	} else {
6409		sbuf_printf(sb, "                 channel 0  channel 1");
6410		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
6411		    stats.req[0], stats.req[1]);
6412		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
6413		    stats.rsp[0], stats.rsp[1]);
6414	}
6415
6416	rc = sbuf_finish(sb);
6417	sbuf_delete(sb);
6418
6419	return (rc);
6420}
6421
6422static int
6423sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
6424{
6425	struct adapter *sc = arg1;
6426	struct sbuf *sb;
6427	int rc;
6428	struct tp_usm_stats stats;
6429
6430	rc = sysctl_wire_old_buffer(req, 0);
6431	if (rc != 0)
6432		return(rc);
6433
6434	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6435	if (sb == NULL)
6436		return (ENOMEM);
6437
6438	t4_get_usm_stats(sc, &stats, 1);
6439
6440	sbuf_printf(sb, "Frames: %u\n", stats.frames);
6441	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
6442	sbuf_printf(sb, "Drops:  %u", stats.drops);
6443
6444	rc = sbuf_finish(sb);
6445	sbuf_delete(sb);
6446
6447	return (rc);
6448}
6449
6450static const char * const devlog_level_strings[] = {
6451	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
6452	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
6453	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
6454	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
6455	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
6456	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
6457};
6458
6459static const char * const devlog_facility_strings[] = {
6460	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
6461	[FW_DEVLOG_FACILITY_CF]		= "CF",
6462	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
6463	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
6464	[FW_DEVLOG_FACILITY_RES]	= "RES",
6465	[FW_DEVLOG_FACILITY_HW]		= "HW",
6466	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
6467	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
6468	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
6469	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
6470	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
6471	[FW_DEVLOG_FACILITY_VI]		= "VI",
6472	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
6473	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
6474	[FW_DEVLOG_FACILITY_TM]		= "TM",
6475	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
6476	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
6477	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
6478	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
6479	[FW_DEVLOG_FACILITY_RI]		= "RI",
6480	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
6481	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
6482	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
6483	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
6484	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
6485};
6486
6487static int
6488sysctl_devlog(SYSCTL_HANDLER_ARGS)
6489{
6490	struct adapter *sc = arg1;
6491	struct devlog_params *dparams = &sc->params.devlog;
6492	struct fw_devlog_e *buf, *e;
6493	int i, j, rc, nentries, first = 0;
6494	struct sbuf *sb;
6495	uint64_t ftstamp = UINT64_MAX;
6496
6497	if (dparams->addr == 0)
6498		return (ENXIO);
6499
6500	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
6501	if (buf == NULL)
6502		return (ENOMEM);
6503
6504	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
6505	if (rc != 0)
6506		goto done;
6507
6508	nentries = dparams->size / sizeof(struct fw_devlog_e);
6509	for (i = 0; i < nentries; i++) {
6510		e = &buf[i];
6511
6512		if (e->timestamp == 0)
6513			break;	/* end */
6514
6515		e->timestamp = be64toh(e->timestamp);
6516		e->seqno = be32toh(e->seqno);
6517		for (j = 0; j < 8; j++)
6518			e->params[j] = be32toh(e->params[j]);
6519
6520		if (e->timestamp < ftstamp) {
6521			ftstamp = e->timestamp;
6522			first = i;
6523		}
6524	}
6525
6526	if (buf[first].timestamp == 0)
6527		goto done;	/* nothing in the log */
6528
6529	rc = sysctl_wire_old_buffer(req, 0);
6530	if (rc != 0)
6531		goto done;
6532
6533	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6534	if (sb == NULL) {
6535		rc = ENOMEM;
6536		goto done;
6537	}
6538	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
6539	    "Seq#", "Tstamp", "Level", "Facility", "Message");
6540
6541	i = first;
6542	do {
6543		e = &buf[i];
6544		if (e->timestamp == 0)
6545			break;	/* end */
6546
6547		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
6548		    e->seqno, e->timestamp,
6549		    (e->level < nitems(devlog_level_strings) ?
6550			devlog_level_strings[e->level] : "UNKNOWN"),
6551		    (e->facility < nitems(devlog_facility_strings) ?
6552			devlog_facility_strings[e->facility] : "UNKNOWN"));
6553		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
6554		    e->params[2], e->params[3], e->params[4],
6555		    e->params[5], e->params[6], e->params[7]);
6556
6557		if (++i == nentries)
6558			i = 0;
6559	} while (i != first);
6560
6561	rc = sbuf_finish(sb);
6562	sbuf_delete(sb);
6563done:
6564	free(buf, M_CXGBE);
6565	return (rc);
6566}
6567
6568static int
6569sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
6570{
6571	struct adapter *sc = arg1;
6572	struct sbuf *sb;
6573	int rc;
6574	struct tp_fcoe_stats stats[MAX_NCHAN];
6575	int i, nchan = sc->chip_params->nchan;
6576
6577	rc = sysctl_wire_old_buffer(req, 0);
6578	if (rc != 0)
6579		return (rc);
6580
6581	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6582	if (sb == NULL)
6583		return (ENOMEM);
6584
6585	for (i = 0; i < nchan; i++)
6586		t4_get_fcoe_stats(sc, i, &stats[i], 1);
6587
6588	if (nchan > 2) {
6589		sbuf_printf(sb, "                   channel 0        channel 1"
6590		    "        channel 2        channel 3");
6591		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
6592		    stats[0].octets_ddp, stats[1].octets_ddp,
6593		    stats[2].octets_ddp, stats[3].octets_ddp);
6594		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
6595		    stats[0].frames_ddp, stats[1].frames_ddp,
6596		    stats[2].frames_ddp, stats[3].frames_ddp);
6597		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
6598		    stats[0].frames_drop, stats[1].frames_drop,
6599		    stats[2].frames_drop, stats[3].frames_drop);
6600	} else {
6601		sbuf_printf(sb, "                   channel 0        channel 1");
6602		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
6603		    stats[0].octets_ddp, stats[1].octets_ddp);
6604		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
6605		    stats[0].frames_ddp, stats[1].frames_ddp);
6606		sbuf_printf(sb, "\nframesDrop: %16u %16u",
6607		    stats[0].frames_drop, stats[1].frames_drop);
6608	}
6609
6610	rc = sbuf_finish(sb);
6611	sbuf_delete(sb);
6612
6613	return (rc);
6614}
6615
6616static int
6617sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
6618{
6619	struct adapter *sc = arg1;
6620	struct sbuf *sb;
6621	int rc, i;
6622	unsigned int map, kbps, ipg, mode;
6623	unsigned int pace_tab[NTX_SCHED];
6624
6625	rc = sysctl_wire_old_buffer(req, 0);
6626	if (rc != 0)
6627		return (rc);
6628
6629	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6630	if (sb == NULL)
6631		return (ENOMEM);
6632
6633	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
6634	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
6635	t4_read_pace_tbl(sc, pace_tab);
6636
6637	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
6638	    "Class IPG (0.1 ns)   Flow IPG (us)");
6639
6640	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
6641		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
6642		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
6643		    (mode & (1 << i)) ? "flow" : "class", map & 3);
6644		if (kbps)
6645			sbuf_printf(sb, "%9u     ", kbps);
6646		else
6647			sbuf_printf(sb, " disabled     ");
6648
6649		if (ipg)
6650			sbuf_printf(sb, "%13u        ", ipg);
6651		else
6652			sbuf_printf(sb, "     disabled        ");
6653
6654		if (pace_tab[i])
6655			sbuf_printf(sb, "%10u", pace_tab[i]);
6656		else
6657			sbuf_printf(sb, "  disabled");
6658	}
6659
6660	rc = sbuf_finish(sb);
6661	sbuf_delete(sb);
6662
6663	return (rc);
6664}
6665
6666static int
6667sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
6668{
6669	struct adapter *sc = arg1;
6670	struct sbuf *sb;
6671	int rc, i, j;
6672	uint64_t *p0, *p1;
6673	struct lb_port_stats s[2];
6674	static const char *stat_name[] = {
6675		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
6676		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
6677		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
6678		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
6679		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
6680		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
6681		"BG2FramesTrunc:", "BG3FramesTrunc:"
6682	};
6683
6684	rc = sysctl_wire_old_buffer(req, 0);
6685	if (rc != 0)
6686		return (rc);
6687
6688	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6689	if (sb == NULL)
6690		return (ENOMEM);
6691
6692	memset(s, 0, sizeof(s));
6693
6694	for (i = 0; i < sc->chip_params->nchan; i += 2) {
6695		t4_get_lb_stats(sc, i, &s[0]);
6696		t4_get_lb_stats(sc, i + 1, &s[1]);
6697
6698		p0 = &s[0].octets;
6699		p1 = &s[1].octets;
6700		sbuf_printf(sb, "%s                       Loopback %u"
6701		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
6702
6703		for (j = 0; j < nitems(stat_name); j++)
6704			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
6705				   *p0++, *p1++);
6706	}
6707
6708	rc = sbuf_finish(sb);
6709	sbuf_delete(sb);
6710
6711	return (rc);
6712}
6713
6714static int
6715sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
6716{
6717	int rc = 0;
6718	struct port_info *pi = arg1;
6719	struct link_config *lc = &pi->link_cfg;
6720	struct sbuf *sb;
6721
6722	rc = sysctl_wire_old_buffer(req, 0);
6723	if (rc != 0)
6724		return(rc);
6725	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
6726	if (sb == NULL)
6727		return (ENOMEM);
6728
6729	if (lc->link_ok || lc->link_down_rc == 255)
6730		sbuf_printf(sb, "n/a");
6731	else
6732		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
6733
6734	rc = sbuf_finish(sb);
6735	sbuf_delete(sb);
6736
6737	return (rc);
6738}
6739
6740struct mem_desc {
6741	unsigned int base;
6742	unsigned int limit;
6743	unsigned int idx;
6744};
6745
6746static int
6747mem_desc_cmp(const void *a, const void *b)
6748{
6749	return ((const struct mem_desc *)a)->base -
6750	       ((const struct mem_desc *)b)->base;
6751}
6752
6753static void
6754mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
6755    unsigned int to)
6756{
6757	unsigned int size;
6758
6759	if (from == to)
6760		return;
6761
6762	size = to - from + 1;
6763	if (size == 0)
6764		return;
6765
6766	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
6767	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
6768}
6769
6770static int
6771sysctl_meminfo(SYSCTL_HANDLER_ARGS)
6772{
6773	struct adapter *sc = arg1;
6774	struct sbuf *sb;
6775	int rc, i, n;
6776	uint32_t lo, hi, used, alloc;
6777	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
6778	static const char *region[] = {
6779		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
6780		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
6781		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
6782		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
6783		"RQUDP region:", "PBL region:", "TXPBL region:",
6784		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
6785		"On-chip queues:"
6786	};
6787	struct mem_desc avail[4];
6788	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
6789	struct mem_desc *md = mem;
6790
6791	rc = sysctl_wire_old_buffer(req, 0);
6792	if (rc != 0)
6793		return (rc);
6794
6795	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6796	if (sb == NULL)
6797		return (ENOMEM);
6798
6799	for (i = 0; i < nitems(mem); i++) {
6800		mem[i].limit = 0;
6801		mem[i].idx = i;
6802	}
6803
6804	/* Find and sort the populated memory ranges */
6805	i = 0;
6806	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
6807	if (lo & F_EDRAM0_ENABLE) {
6808		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
6809		avail[i].base = G_EDRAM0_BASE(hi) << 20;
6810		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
6811		avail[i].idx = 0;
6812		i++;
6813	}
6814	if (lo & F_EDRAM1_ENABLE) {
6815		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
6816		avail[i].base = G_EDRAM1_BASE(hi) << 20;
6817		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
6818		avail[i].idx = 1;
6819		i++;
6820	}
6821	if (lo & F_EXT_MEM_ENABLE) {
6822		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
6823		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
6824		avail[i].limit = avail[i].base +
6825		    (G_EXT_MEM_SIZE(hi) << 20);
6826		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
6827		i++;
6828	}
6829	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
6830		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
6831		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
6832		avail[i].limit = avail[i].base +
6833		    (G_EXT_MEM1_SIZE(hi) << 20);
6834		avail[i].idx = 4;
6835		i++;
6836	}
6837	if (!i)                                    /* no memory available */
6838		return 0;
6839	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
6840
6841	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
6842	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
6843	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
6844	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
6845	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
6846	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
6847	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
6848	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
6849	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
6850
6851	/* the next few have explicit upper bounds */
6852	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
6853	md->limit = md->base - 1 +
6854		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
6855		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
6856	md++;
6857
6858	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
6859	md->limit = md->base - 1 +
6860		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
6861		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
6862	md++;
6863
6864	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
6865		if (chip_id(sc) <= CHELSIO_T5)
6866			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
6867		else
6868			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
6869		md->limit = 0;
6870	} else {
6871		md->base = 0;
6872		md->idx = nitems(region);  /* hide it */
6873	}
6874	md++;
6875
6876#define ulp_region(reg) \
6877	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
6878	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
6879
6880	ulp_region(RX_ISCSI);
6881	ulp_region(RX_TDDP);
6882	ulp_region(TX_TPT);
6883	ulp_region(RX_STAG);
6884	ulp_region(RX_RQ);
6885	ulp_region(RX_RQUDP);
6886	ulp_region(RX_PBL);
6887	ulp_region(TX_PBL);
6888#undef ulp_region
6889
6890	md->base = 0;
6891	md->idx = nitems(region);
6892	if (!is_t4(sc)) {
6893		uint32_t size = 0;
6894		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
6895		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
6896
6897		if (is_t5(sc)) {
6898			if (sge_ctrl & F_VFIFO_ENABLE)
6899				size = G_DBVFIFO_SIZE(fifo_size);
6900		} else
6901			size = G_T6_DBVFIFO_SIZE(fifo_size);
6902
6903		if (size) {
6904			md->base = G_BASEADDR(t4_read_reg(sc,
6905			    A_SGE_DBVFIFO_BADDR));
6906			md->limit = md->base + (size << 2) - 1;
6907		}
6908	}
6909	md++;
6910
6911	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
6912	md->limit = 0;
6913	md++;
6914	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
6915	md->limit = 0;
6916	md++;
6917
6918	md->base = sc->vres.ocq.start;
6919	if (sc->vres.ocq.size)
6920		md->limit = md->base + sc->vres.ocq.size - 1;
6921	else
6922		md->idx = nitems(region);  /* hide it */
6923	md++;
6924
6925	/* add any address-space holes, there can be up to 3 */
6926	for (n = 0; n < i - 1; n++)
6927		if (avail[n].limit < avail[n + 1].base)
6928			(md++)->base = avail[n].limit;
6929	if (avail[n].limit)
6930		(md++)->base = avail[n].limit;
6931
6932	n = md - mem;
6933	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
6934
6935	for (lo = 0; lo < i; lo++)
6936		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
6937				avail[lo].limit - 1);
6938
6939	sbuf_printf(sb, "\n");
6940	for (i = 0; i < n; i++) {
6941		if (mem[i].idx >= nitems(region))
6942			continue;                        /* skip holes */
6943		if (!mem[i].limit)
6944			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
6945		mem_region_show(sb, region[mem[i].idx], mem[i].base,
6946				mem[i].limit);
6947	}
6948
6949	sbuf_printf(sb, "\n");
6950	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
6951	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
6952	mem_region_show(sb, "uP RAM:", lo, hi);
6953
6954	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
6955	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
6956	mem_region_show(sb, "uP Extmem2:", lo, hi);
6957
6958	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
6959	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
6960		   G_PMRXMAXPAGE(lo),
6961		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
6962		   (lo & F_PMRXNUMCHN) ? 2 : 1);
6963
6964	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
6965	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
6966	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
6967		   G_PMTXMAXPAGE(lo),
6968		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
6969		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
6970	sbuf_printf(sb, "%u p-structs\n",
6971		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
6972
6973	for (i = 0; i < 4; i++) {
6974		if (chip_id(sc) > CHELSIO_T5)
6975			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
6976		else
6977			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
6978		if (is_t5(sc)) {
6979			used = G_T5_USED(lo);
6980			alloc = G_T5_ALLOC(lo);
6981		} else {
6982			used = G_USED(lo);
6983			alloc = G_ALLOC(lo);
6984		}
6985		/* For T6 these are MAC buffer groups */
6986		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
6987		    i, used, alloc);
6988	}
6989	for (i = 0; i < sc->chip_params->nchan; i++) {
6990		if (chip_id(sc) > CHELSIO_T5)
6991			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
6992		else
6993			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
6994		if (is_t5(sc)) {
6995			used = G_T5_USED(lo);
6996			alloc = G_T5_ALLOC(lo);
6997		} else {
6998			used = G_USED(lo);
6999			alloc = G_ALLOC(lo);
7000		}
7001		/* For T6 these are MAC buffer groups */
7002		sbuf_printf(sb,
7003		    "\nLoopback %d using %u pages out of %u allocated",
7004		    i, used, alloc);
7005	}
7006
7007	rc = sbuf_finish(sb);
7008	sbuf_delete(sb);
7009
7010	return (rc);
7011}
7012
7013static inline void
7014tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
7015{
7016	*mask = x | y;
7017	y = htobe64(y);
7018	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
7019}
7020
7021static int
7022sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
7023{
7024	struct adapter *sc = arg1;
7025	struct sbuf *sb;
7026	int rc, i;
7027
7028	MPASS(chip_id(sc) <= CHELSIO_T5);
7029
7030	rc = sysctl_wire_old_buffer(req, 0);
7031	if (rc != 0)
7032		return (rc);
7033
7034	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7035	if (sb == NULL)
7036		return (ENOMEM);
7037
7038	sbuf_printf(sb,
7039	    "Idx  Ethernet address     Mask     Vld Ports PF"
7040	    "  VF              Replication             P0 P1 P2 P3  ML");
7041	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7042		uint64_t tcamx, tcamy, mask;
7043		uint32_t cls_lo, cls_hi;
7044		uint8_t addr[ETHER_ADDR_LEN];
7045
7046		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
7047		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
7048		if (tcamx & tcamy)
7049			continue;
7050		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7051		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7052		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7053		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
7054			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
7055			   addr[3], addr[4], addr[5], (uintmax_t)mask,
7056			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
7057			   G_PORTMAP(cls_hi), G_PF(cls_lo),
7058			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
7059
7060		if (cls_lo & F_REPLICATE) {
7061			struct fw_ldst_cmd ldst_cmd;
7062
7063			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7064			ldst_cmd.op_to_addrspace =
7065			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7066				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7067				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7068			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7069			ldst_cmd.u.mps.rplc.fid_idx =
7070			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7071				V_FW_LDST_CMD_IDX(i));
7072
7073			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7074			    "t4mps");
7075			if (rc)
7076				break;
7077			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7078			    sizeof(ldst_cmd), &ldst_cmd);
7079			end_synchronized_op(sc, 0);
7080
7081			if (rc != 0) {
7082				sbuf_printf(sb, "%36d", rc);
7083				rc = 0;
7084			} else {
7085				sbuf_printf(sb, " %08x %08x %08x %08x",
7086				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7087				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7088				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7089				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7090			}
7091		} else
7092			sbuf_printf(sb, "%36s", "");
7093
7094		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
7095		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
7096		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
7097	}
7098
7099	if (rc)
7100		(void) sbuf_finish(sb);
7101	else
7102		rc = sbuf_finish(sb);
7103	sbuf_delete(sb);
7104
7105	return (rc);
7106}
7107
7108static int
7109sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
7110{
7111	struct adapter *sc = arg1;
7112	struct sbuf *sb;
7113	int rc, i;
7114
7115	MPASS(chip_id(sc) > CHELSIO_T5);
7116
7117	rc = sysctl_wire_old_buffer(req, 0);
7118	if (rc != 0)
7119		return (rc);
7120
7121	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7122	if (sb == NULL)
7123		return (ENOMEM);
7124
7125	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
7126	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
7127	    "                           Replication"
7128	    "                                    P0 P1 P2 P3  ML\n");
7129
7130	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
7131		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
7132		uint16_t ivlan;
7133		uint64_t tcamx, tcamy, val, mask;
7134		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
7135		uint8_t addr[ETHER_ADDR_LEN];
7136
7137		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
7138		if (i < 256)
7139			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
7140		else
7141			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
7142		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7143		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7144		tcamy = G_DMACH(val) << 32;
7145		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7146		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7147		lookup_type = G_DATALKPTYPE(data2);
7148		port_num = G_DATAPORTNUM(data2);
7149		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7150			/* Inner header VNI */
7151			vniy = ((data2 & F_DATAVIDH2) << 23) |
7152				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7153			dip_hit = data2 & F_DATADIPHIT;
7154			vlan_vld = 0;
7155		} else {
7156			vniy = 0;
7157			dip_hit = 0;
7158			vlan_vld = data2 & F_DATAVIDH2;
7159			ivlan = G_VIDL(val);
7160		}
7161
7162		ctl |= V_CTLXYBITSEL(1);
7163		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
7164		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
7165		tcamx = G_DMACH(val) << 32;
7166		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
7167		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
7168		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7169			/* Inner header VNI mask */
7170			vnix = ((data2 & F_DATAVIDH2) << 23) |
7171			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
7172		} else
7173			vnix = 0;
7174
7175		if (tcamx & tcamy)
7176			continue;
7177		tcamxy2valmask(tcamx, tcamy, addr, &mask);
7178
7179		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
7180		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
7181
7182		if (lookup_type && lookup_type != M_DATALKPTYPE) {
7183			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7184			    "%012jx %06x %06x    -    -   %3c"
7185			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
7186			    addr[1], addr[2], addr[3], addr[4], addr[5],
7187			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
7188			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7189			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7190			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7191		} else {
7192			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
7193			    "%012jx    -       -   ", i, addr[0], addr[1],
7194			    addr[2], addr[3], addr[4], addr[5],
7195			    (uintmax_t)mask);
7196
7197			if (vlan_vld)
7198				sbuf_printf(sb, "%4u   Y     ", ivlan);
7199			else
7200				sbuf_printf(sb, "  -    N     ");
7201
7202			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
7203			    lookup_type ? 'I' : 'O', port_num,
7204			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
7205			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
7206			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
7207		}
7208
7209
7210		if (cls_lo & F_T6_REPLICATE) {
7211			struct fw_ldst_cmd ldst_cmd;
7212
7213			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
7214			ldst_cmd.op_to_addrspace =
7215			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
7216				F_FW_CMD_REQUEST | F_FW_CMD_READ |
7217				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
7218			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
7219			ldst_cmd.u.mps.rplc.fid_idx =
7220			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
7221				V_FW_LDST_CMD_IDX(i));
7222
7223			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
7224			    "t6mps");
7225			if (rc)
7226				break;
7227			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
7228			    sizeof(ldst_cmd), &ldst_cmd);
7229			end_synchronized_op(sc, 0);
7230
7231			if (rc != 0) {
7232				sbuf_printf(sb, "%72d", rc);
7233				rc = 0;
7234			} else {
7235				sbuf_printf(sb, " %08x %08x %08x %08x"
7236				    " %08x %08x %08x %08x",
7237				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
7238				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
7239				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
7240				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
7241				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
7242				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
7243				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
7244				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
7245			}
7246		} else
7247			sbuf_printf(sb, "%72s", "");
7248
7249		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
7250		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
7251		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
7252		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
7253	}
7254
7255	if (rc)
7256		(void) sbuf_finish(sb);
7257	else
7258		rc = sbuf_finish(sb);
7259	sbuf_delete(sb);
7260
7261	return (rc);
7262}
7263
7264static int
7265sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
7266{
7267	struct adapter *sc = arg1;
7268	struct sbuf *sb;
7269	int rc;
7270	uint16_t mtus[NMTUS];
7271
7272	rc = sysctl_wire_old_buffer(req, 0);
7273	if (rc != 0)
7274		return (rc);
7275
7276	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7277	if (sb == NULL)
7278		return (ENOMEM);
7279
7280	t4_read_mtu_tbl(sc, mtus, NULL);
7281
7282	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
7283	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
7284	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
7285	    mtus[14], mtus[15]);
7286
7287	rc = sbuf_finish(sb);
7288	sbuf_delete(sb);
7289
7290	return (rc);
7291}
7292
7293static int
7294sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
7295{
7296	struct adapter *sc = arg1;
7297	struct sbuf *sb;
7298	int rc, i;
7299	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
7300	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
7301	static const char *tx_stats[MAX_PM_NSTATS] = {
7302		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
7303		"Tx FIFO wait", NULL, "Tx latency"
7304	};
7305	static const char *rx_stats[MAX_PM_NSTATS] = {
7306		"Read:", "Write bypass:", "Write mem:", "Flush:",
7307		"Rx FIFO wait", NULL, "Rx latency"
7308	};
7309
7310	rc = sysctl_wire_old_buffer(req, 0);
7311	if (rc != 0)
7312		return (rc);
7313
7314	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7315	if (sb == NULL)
7316		return (ENOMEM);
7317
7318	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
7319	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
7320
7321	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
7322	for (i = 0; i < 4; i++) {
7323		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7324		    tx_cyc[i]);
7325	}
7326
7327	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
7328	for (i = 0; i < 4; i++) {
7329		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7330		    rx_cyc[i]);
7331	}
7332
7333	if (chip_id(sc) > CHELSIO_T5) {
7334		sbuf_printf(sb,
7335		    "\n              Total wait      Total occupancy");
7336		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7337		    tx_cyc[i]);
7338		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7339		    rx_cyc[i]);
7340
7341		i += 2;
7342		MPASS(i < nitems(tx_stats));
7343
7344		sbuf_printf(sb,
7345		    "\n                   Reads           Total wait");
7346		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
7347		    tx_cyc[i]);
7348		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
7349		    rx_cyc[i]);
7350	}
7351
7352	rc = sbuf_finish(sb);
7353	sbuf_delete(sb);
7354
7355	return (rc);
7356}
7357
7358static int
7359sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
7360{
7361	struct adapter *sc = arg1;
7362	struct sbuf *sb;
7363	int rc;
7364	struct tp_rdma_stats stats;
7365
7366	rc = sysctl_wire_old_buffer(req, 0);
7367	if (rc != 0)
7368		return (rc);
7369
7370	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7371	if (sb == NULL)
7372		return (ENOMEM);
7373
7374	mtx_lock(&sc->reg_lock);
7375	t4_tp_get_rdma_stats(sc, &stats, 0);
7376	mtx_unlock(&sc->reg_lock);
7377
7378	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
7379	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
7380
7381	rc = sbuf_finish(sb);
7382	sbuf_delete(sb);
7383
7384	return (rc);
7385}
7386
7387static int
7388sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
7389{
7390	struct adapter *sc = arg1;
7391	struct sbuf *sb;
7392	int rc;
7393	struct tp_tcp_stats v4, v6;
7394
7395	rc = sysctl_wire_old_buffer(req, 0);
7396	if (rc != 0)
7397		return (rc);
7398
7399	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7400	if (sb == NULL)
7401		return (ENOMEM);
7402
7403	mtx_lock(&sc->reg_lock);
7404	t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
7405	mtx_unlock(&sc->reg_lock);
7406
7407	sbuf_printf(sb,
7408	    "                                IP                 IPv6\n");
7409	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
7410	    v4.tcp_out_rsts, v6.tcp_out_rsts);
7411	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
7412	    v4.tcp_in_segs, v6.tcp_in_segs);
7413	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
7414	    v4.tcp_out_segs, v6.tcp_out_segs);
7415	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
7416	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
7417
7418	rc = sbuf_finish(sb);
7419	sbuf_delete(sb);
7420
7421	return (rc);
7422}
7423
7424static int
7425sysctl_tids(SYSCTL_HANDLER_ARGS)
7426{
7427	struct adapter *sc = arg1;
7428	struct sbuf *sb;
7429	int rc;
7430	struct tid_info *t = &sc->tids;
7431
7432	rc = sysctl_wire_old_buffer(req, 0);
7433	if (rc != 0)
7434		return (rc);
7435
7436	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7437	if (sb == NULL)
7438		return (ENOMEM);
7439
7440	if (t->natids) {
7441		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
7442		    t->atids_in_use);
7443	}
7444
7445	if (t->ntids) {
7446		sbuf_printf(sb, "TID range: ");
7447		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7448			uint32_t b, hb;
7449
7450			if (chip_id(sc) <= CHELSIO_T5) {
7451				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
7452				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
7453			} else {
7454				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
7455				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
7456			}
7457
7458			if (b)
7459				sbuf_printf(sb, "0-%u, ", b - 1);
7460			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
7461		} else
7462			sbuf_printf(sb, "0-%u", t->ntids - 1);
7463		sbuf_printf(sb, ", in use: %u\n",
7464		    atomic_load_acq_int(&t->tids_in_use));
7465	}
7466
7467	if (t->nstids) {
7468		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
7469		    t->stid_base + t->nstids - 1, t->stids_in_use);
7470	}
7471
7472	if (t->nftids) {
7473		sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
7474		    t->ftid_base + t->nftids - 1);
7475	}
7476
7477	if (t->netids) {
7478		sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
7479		    t->etid_base + t->netids - 1);
7480	}
7481
7482	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
7483	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
7484	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
7485
7486	rc = sbuf_finish(sb);
7487	sbuf_delete(sb);
7488
7489	return (rc);
7490}
7491
7492static int
7493sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
7494{
7495	struct adapter *sc = arg1;
7496	struct sbuf *sb;
7497	int rc;
7498	struct tp_err_stats stats;
7499
7500	rc = sysctl_wire_old_buffer(req, 0);
7501	if (rc != 0)
7502		return (rc);
7503
7504	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7505	if (sb == NULL)
7506		return (ENOMEM);
7507
7508	mtx_lock(&sc->reg_lock);
7509	t4_tp_get_err_stats(sc, &stats, 0);
7510	mtx_unlock(&sc->reg_lock);
7511
7512	if (sc->chip_params->nchan > 2) {
7513		sbuf_printf(sb, "                 channel 0  channel 1"
7514		    "  channel 2  channel 3\n");
7515		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
7516		    stats.mac_in_errs[0], stats.mac_in_errs[1],
7517		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
7518		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
7519		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
7520		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
7521		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
7522		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
7523		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
7524		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
7525		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
7526		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
7527		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
7528		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
7529		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
7530		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
7531		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
7532		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
7533		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
7534		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
7535		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
7536		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
7537		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
7538		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
7539	} else {
7540		sbuf_printf(sb, "                 channel 0  channel 1\n");
7541		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
7542		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
7543		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
7544		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
7545		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
7546		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
7547		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
7548		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
7549		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
7550		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
7551		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
7552		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
7553		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
7554		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
7555		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
7556		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
7557	}
7558
7559	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
7560	    stats.ofld_no_neigh, stats.ofld_cong_defer);
7561
7562	rc = sbuf_finish(sb);
7563	sbuf_delete(sb);
7564
7565	return (rc);
7566}
7567
7568static int
7569sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
7570{
7571	struct adapter *sc = arg1;
7572	struct tp_params *tpp = &sc->params.tp;
7573	u_int mask;
7574	int rc;
7575
7576	mask = tpp->la_mask >> 16;
7577	rc = sysctl_handle_int(oidp, &mask, 0, req);
7578	if (rc != 0 || req->newptr == NULL)
7579		return (rc);
7580	if (mask > 0xffff)
7581		return (EINVAL);
7582	tpp->la_mask = mask << 16;
7583	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
7584
7585	return (0);
7586}
7587
7588struct field_desc {
7589	const char *name;
7590	u_int start;
7591	u_int width;
7592};
7593
7594static void
7595field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
7596{
7597	char buf[32];
7598	int line_size = 0;
7599
7600	while (f->name) {
7601		uint64_t mask = (1ULL << f->width) - 1;
7602		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
7603		    ((uintmax_t)v >> f->start) & mask);
7604
7605		if (line_size + len >= 79) {
7606			line_size = 8;
7607			sbuf_printf(sb, "\n        ");
7608		}
7609		sbuf_printf(sb, "%s ", buf);
7610		line_size += len + 1;
7611		f++;
7612	}
7613	sbuf_printf(sb, "\n");
7614}
7615
7616static const struct field_desc tp_la0[] = {
7617	{ "RcfOpCodeOut", 60, 4 },
7618	{ "State", 56, 4 },
7619	{ "WcfState", 52, 4 },
7620	{ "RcfOpcSrcOut", 50, 2 },
7621	{ "CRxError", 49, 1 },
7622	{ "ERxError", 48, 1 },
7623	{ "SanityFailed", 47, 1 },
7624	{ "SpuriousMsg", 46, 1 },
7625	{ "FlushInputMsg", 45, 1 },
7626	{ "FlushInputCpl", 44, 1 },
7627	{ "RssUpBit", 43, 1 },
7628	{ "RssFilterHit", 42, 1 },
7629	{ "Tid", 32, 10 },
7630	{ "InitTcb", 31, 1 },
7631	{ "LineNumber", 24, 7 },
7632	{ "Emsg", 23, 1 },
7633	{ "EdataOut", 22, 1 },
7634	{ "Cmsg", 21, 1 },
7635	{ "CdataOut", 20, 1 },
7636	{ "EreadPdu", 19, 1 },
7637	{ "CreadPdu", 18, 1 },
7638	{ "TunnelPkt", 17, 1 },
7639	{ "RcfPeerFin", 16, 1 },
7640	{ "RcfReasonOut", 12, 4 },
7641	{ "TxCchannel", 10, 2 },
7642	{ "RcfTxChannel", 8, 2 },
7643	{ "RxEchannel", 6, 2 },
7644	{ "RcfRxChannel", 5, 1 },
7645	{ "RcfDataOutSrdy", 4, 1 },
7646	{ "RxDvld", 3, 1 },
7647	{ "RxOoDvld", 2, 1 },
7648	{ "RxCongestion", 1, 1 },
7649	{ "TxCongestion", 0, 1 },
7650	{ NULL }
7651};
7652
7653static const struct field_desc tp_la1[] = {
7654	{ "CplCmdIn", 56, 8 },
7655	{ "CplCmdOut", 48, 8 },
7656	{ "ESynOut", 47, 1 },
7657	{ "EAckOut", 46, 1 },
7658	{ "EFinOut", 45, 1 },
7659	{ "ERstOut", 44, 1 },
7660	{ "SynIn", 43, 1 },
7661	{ "AckIn", 42, 1 },
7662	{ "FinIn", 41, 1 },
7663	{ "RstIn", 40, 1 },
7664	{ "DataIn", 39, 1 },
7665	{ "DataInVld", 38, 1 },
7666	{ "PadIn", 37, 1 },
7667	{ "RxBufEmpty", 36, 1 },
7668	{ "RxDdp", 35, 1 },
7669	{ "RxFbCongestion", 34, 1 },
7670	{ "TxFbCongestion", 33, 1 },
7671	{ "TxPktSumSrdy", 32, 1 },
7672	{ "RcfUlpType", 28, 4 },
7673	{ "Eread", 27, 1 },
7674	{ "Ebypass", 26, 1 },
7675	{ "Esave", 25, 1 },
7676	{ "Static0", 24, 1 },
7677	{ "Cread", 23, 1 },
7678	{ "Cbypass", 22, 1 },
7679	{ "Csave", 21, 1 },
7680	{ "CPktOut", 20, 1 },
7681	{ "RxPagePoolFull", 18, 2 },
7682	{ "RxLpbkPkt", 17, 1 },
7683	{ "TxLpbkPkt", 16, 1 },
7684	{ "RxVfValid", 15, 1 },
7685	{ "SynLearned", 14, 1 },
7686	{ "SetDelEntry", 13, 1 },
7687	{ "SetInvEntry", 12, 1 },
7688	{ "CpcmdDvld", 11, 1 },
7689	{ "CpcmdSave", 10, 1 },
7690	{ "RxPstructsFull", 8, 2 },
7691	{ "EpcmdDvld", 7, 1 },
7692	{ "EpcmdFlush", 6, 1 },
7693	{ "EpcmdTrimPrefix", 5, 1 },
7694	{ "EpcmdTrimPostfix", 4, 1 },
7695	{ "ERssIp4Pkt", 3, 1 },
7696	{ "ERssIp6Pkt", 2, 1 },
7697	{ "ERssTcpUdpPkt", 1, 1 },
7698	{ "ERssFceFipPkt", 0, 1 },
7699	{ NULL }
7700};
7701
7702static const struct field_desc tp_la2[] = {
7703	{ "CplCmdIn", 56, 8 },
7704	{ "MpsVfVld", 55, 1 },
7705	{ "MpsPf", 52, 3 },
7706	{ "MpsVf", 44, 8 },
7707	{ "SynIn", 43, 1 },
7708	{ "AckIn", 42, 1 },
7709	{ "FinIn", 41, 1 },
7710	{ "RstIn", 40, 1 },
7711	{ "DataIn", 39, 1 },
7712	{ "DataInVld", 38, 1 },
7713	{ "PadIn", 37, 1 },
7714	{ "RxBufEmpty", 36, 1 },
7715	{ "RxDdp", 35, 1 },
7716	{ "RxFbCongestion", 34, 1 },
7717	{ "TxFbCongestion", 33, 1 },
7718	{ "TxPktSumSrdy", 32, 1 },
7719	{ "RcfUlpType", 28, 4 },
7720	{ "Eread", 27, 1 },
7721	{ "Ebypass", 26, 1 },
7722	{ "Esave", 25, 1 },
7723	{ "Static0", 24, 1 },
7724	{ "Cread", 23, 1 },
7725	{ "Cbypass", 22, 1 },
7726	{ "Csave", 21, 1 },
7727	{ "CPktOut", 20, 1 },
7728	{ "RxPagePoolFull", 18, 2 },
7729	{ "RxLpbkPkt", 17, 1 },
7730	{ "TxLpbkPkt", 16, 1 },
7731	{ "RxVfValid", 15, 1 },
7732	{ "SynLearned", 14, 1 },
7733	{ "SetDelEntry", 13, 1 },
7734	{ "SetInvEntry", 12, 1 },
7735	{ "CpcmdDvld", 11, 1 },
7736	{ "CpcmdSave", 10, 1 },
7737	{ "RxPstructsFull", 8, 2 },
7738	{ "EpcmdDvld", 7, 1 },
7739	{ "EpcmdFlush", 6, 1 },
7740	{ "EpcmdTrimPrefix", 5, 1 },
7741	{ "EpcmdTrimPostfix", 4, 1 },
7742	{ "ERssIp4Pkt", 3, 1 },
7743	{ "ERssIp6Pkt", 2, 1 },
7744	{ "ERssTcpUdpPkt", 1, 1 },
7745	{ "ERssFceFipPkt", 0, 1 },
7746	{ NULL }
7747};
7748
7749static void
7750tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
7751{
7752
7753	field_desc_show(sb, *p, tp_la0);
7754}
7755
7756static void
7757tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
7758{
7759
7760	if (idx)
7761		sbuf_printf(sb, "\n");
7762	field_desc_show(sb, p[0], tp_la0);
7763	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7764		field_desc_show(sb, p[1], tp_la0);
7765}
7766
7767static void
7768tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
7769{
7770
7771	if (idx)
7772		sbuf_printf(sb, "\n");
7773	field_desc_show(sb, p[0], tp_la0);
7774	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7775		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
7776}
7777
7778static int
7779sysctl_tp_la(SYSCTL_HANDLER_ARGS)
7780{
7781	struct adapter *sc = arg1;
7782	struct sbuf *sb;
7783	uint64_t *buf, *p;
7784	int rc;
7785	u_int i, inc;
7786	void (*show_func)(struct sbuf *, uint64_t *, int);
7787
7788	rc = sysctl_wire_old_buffer(req, 0);
7789	if (rc != 0)
7790		return (rc);
7791
7792	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7793	if (sb == NULL)
7794		return (ENOMEM);
7795
7796	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
7797
7798	t4_tp_read_la(sc, buf, NULL);
7799	p = buf;
7800
7801	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
7802	case 2:
7803		inc = 2;
7804		show_func = tp_la_show2;
7805		break;
7806	case 3:
7807		inc = 2;
7808		show_func = tp_la_show3;
7809		break;
7810	default:
7811		inc = 1;
7812		show_func = tp_la_show;
7813	}
7814
7815	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
7816		(*show_func)(sb, p, i);
7817
7818	rc = sbuf_finish(sb);
7819	sbuf_delete(sb);
7820	free(buf, M_CXGBE);
7821	return (rc);
7822}
7823
7824static int
7825sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
7826{
7827	struct adapter *sc = arg1;
7828	struct sbuf *sb;
7829	int rc;
7830	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
7831
7832	rc = sysctl_wire_old_buffer(req, 0);
7833	if (rc != 0)
7834		return (rc);
7835
7836	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7837	if (sb == NULL)
7838		return (ENOMEM);
7839
7840	t4_get_chan_txrate(sc, nrate, orate);
7841
7842	if (sc->chip_params->nchan > 2) {
7843		sbuf_printf(sb, "              channel 0   channel 1"
7844		    "   channel 2   channel 3\n");
7845		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
7846		    nrate[0], nrate[1], nrate[2], nrate[3]);
7847		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
7848		    orate[0], orate[1], orate[2], orate[3]);
7849	} else {
7850		sbuf_printf(sb, "              channel 0   channel 1\n");
7851		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
7852		    nrate[0], nrate[1]);
7853		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
7854		    orate[0], orate[1]);
7855	}
7856
7857	rc = sbuf_finish(sb);
7858	sbuf_delete(sb);
7859
7860	return (rc);
7861}
7862
7863static int
7864sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
7865{
7866	struct adapter *sc = arg1;
7867	struct sbuf *sb;
7868	uint32_t *buf, *p;
7869	int rc, i;
7870
7871	rc = sysctl_wire_old_buffer(req, 0);
7872	if (rc != 0)
7873		return (rc);
7874
7875	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7876	if (sb == NULL)
7877		return (ENOMEM);
7878
7879	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
7880	    M_ZERO | M_WAITOK);
7881
7882	t4_ulprx_read_la(sc, buf);
7883	p = buf;
7884
7885	sbuf_printf(sb, "      Pcmd        Type   Message"
7886	    "                Data");
7887	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
7888		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
7889		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
7890	}
7891
7892	rc = sbuf_finish(sb);
7893	sbuf_delete(sb);
7894	free(buf, M_CXGBE);
7895	return (rc);
7896}
7897
7898static int
7899sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
7900{
7901	struct adapter *sc = arg1;
7902	struct sbuf *sb;
7903	int rc, v;
7904
7905	MPASS(chip_id(sc) >= CHELSIO_T5);
7906
7907	rc = sysctl_wire_old_buffer(req, 0);
7908	if (rc != 0)
7909		return (rc);
7910
7911	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7912	if (sb == NULL)
7913		return (ENOMEM);
7914
7915	v = t4_read_reg(sc, A_SGE_STAT_CFG);
7916	if (G_STATSOURCE_T5(v) == 7) {
7917		int mode;
7918
7919		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
7920		if (mode == 0) {
7921			sbuf_printf(sb, "total %d, incomplete %d",
7922			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7923			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7924		} else if (mode == 1) {
7925			sbuf_printf(sb, "total %d, data overflow %d",
7926			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7927			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7928		} else {
7929			sbuf_printf(sb, "unknown mode %d", mode);
7930		}
7931	}
7932	rc = sbuf_finish(sb);
7933	sbuf_delete(sb);
7934
7935	return (rc);
7936}
7937
7938static int
7939sysctl_tc_params(SYSCTL_HANDLER_ARGS)
7940{
7941	struct adapter *sc = arg1;
7942	struct tx_cl_rl_params tc;
7943	struct sbuf *sb;
7944	int i, rc, port_id, mbps, gbps;
7945
7946	rc = sysctl_wire_old_buffer(req, 0);
7947	if (rc != 0)
7948		return (rc);
7949
7950	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7951	if (sb == NULL)
7952		return (ENOMEM);
7953
7954	port_id = arg2 >> 16;
7955	MPASS(port_id < sc->params.nports);
7956	MPASS(sc->port[port_id] != NULL);
7957	i = arg2 & 0xffff;
7958	MPASS(i < sc->chip_params->nsched_cls);
7959
7960	mtx_lock(&sc->tc_lock);
7961	tc = sc->port[port_id]->sched_params->cl_rl[i];
7962	mtx_unlock(&sc->tc_lock);
7963
7964	if (tc.flags & TX_CLRL_ERROR) {
7965		sbuf_printf(sb, "error");
7966		goto done;
7967	}
7968
7969	if (tc.ratemode == SCHED_CLASS_RATEMODE_REL) {
7970		/* XXX: top speed or actual link speed? */
7971		gbps = port_top_speed(sc->port[port_id]);
7972		sbuf_printf(sb, " %u%% of %uGbps", tc.maxrate, gbps);
7973	} else if (tc.ratemode == SCHED_CLASS_RATEMODE_ABS) {
7974		switch (tc.rateunit) {
7975		case SCHED_CLASS_RATEUNIT_BITS:
7976			mbps = tc.maxrate / 1000;
7977			gbps = tc.maxrate / 1000000;
7978			if (tc.maxrate == gbps * 1000000)
7979				sbuf_printf(sb, " %uGbps", gbps);
7980			else if (tc.maxrate == mbps * 1000)
7981				sbuf_printf(sb, " %uMbps", mbps);
7982			else
7983				sbuf_printf(sb, " %uKbps", tc.maxrate);
7984			break;
7985		case SCHED_CLASS_RATEUNIT_PKTS:
7986			sbuf_printf(sb, " %upps", tc.maxrate);
7987			break;
7988		default:
7989			rc = ENXIO;
7990			goto done;
7991		}
7992	}
7993
7994	switch (tc.mode) {
7995	case SCHED_CLASS_MODE_CLASS:
7996		sbuf_printf(sb, " aggregate");
7997		break;
7998	case SCHED_CLASS_MODE_FLOW:
7999		sbuf_printf(sb, " per-flow");
8000		break;
8001	default:
8002		rc = ENXIO;
8003		goto done;
8004	}
8005
8006done:
8007	if (rc == 0)
8008		rc = sbuf_finish(sb);
8009	sbuf_delete(sb);
8010
8011	return (rc);
8012}
8013#endif
8014
8015#ifdef TCP_OFFLOAD
8016static void
8017unit_conv(char *buf, size_t len, u_int val, u_int factor)
8018{
8019	u_int rem = val % factor;
8020
8021	if (rem == 0)
8022		snprintf(buf, len, "%u", val / factor);
8023	else {
8024		while (rem % 10 == 0)
8025			rem /= 10;
8026		snprintf(buf, len, "%u.%u", val / factor, rem);
8027	}
8028}
8029
8030static int
8031sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
8032{
8033	struct adapter *sc = arg1;
8034	char buf[16];
8035	u_int res, re;
8036	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8037
8038	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8039	switch (arg2) {
8040	case 0:
8041		/* timer_tick */
8042		re = G_TIMERRESOLUTION(res);
8043		break;
8044	case 1:
8045		/* TCP timestamp tick */
8046		re = G_TIMESTAMPRESOLUTION(res);
8047		break;
8048	case 2:
8049		/* DACK tick */
8050		re = G_DELAYEDACKRESOLUTION(res);
8051		break;
8052	default:
8053		return (EDOOFUS);
8054	}
8055
8056	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
8057
8058	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
8059}
8060
8061static int
8062sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
8063{
8064	struct adapter *sc = arg1;
8065	u_int res, dack_re, v;
8066	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8067
8068	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
8069	dack_re = G_DELAYEDACKRESOLUTION(res);
8070	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
8071
8072	return (sysctl_handle_int(oidp, &v, 0, req));
8073}
8074
8075static int
8076sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
8077{
8078	struct adapter *sc = arg1;
8079	int reg = arg2;
8080	u_int tre;
8081	u_long tp_tick_us, v;
8082	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
8083
8084	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
8085	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
8086	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
8087	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
8088
8089	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
8090	tp_tick_us = (cclk_ps << tre) / 1000000;
8091
8092	if (reg == A_TP_INIT_SRTT)
8093		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
8094	else
8095		v = tp_tick_us * t4_read_reg(sc, reg);
8096
8097	return (sysctl_handle_long(oidp, &v, 0, req));
8098}
8099#endif
8100
8101static uint32_t
8102fconf_iconf_to_mode(uint32_t fconf, uint32_t iconf)
8103{
8104	uint32_t mode;
8105
8106	mode = T4_FILTER_IPv4 | T4_FILTER_IPv6 | T4_FILTER_IP_SADDR |
8107	    T4_FILTER_IP_DADDR | T4_FILTER_IP_SPORT | T4_FILTER_IP_DPORT;
8108
8109	if (fconf & F_FRAGMENTATION)
8110		mode |= T4_FILTER_IP_FRAGMENT;
8111
8112	if (fconf & F_MPSHITTYPE)
8113		mode |= T4_FILTER_MPS_HIT_TYPE;
8114
8115	if (fconf & F_MACMATCH)
8116		mode |= T4_FILTER_MAC_IDX;
8117
8118	if (fconf & F_ETHERTYPE)
8119		mode |= T4_FILTER_ETH_TYPE;
8120
8121	if (fconf & F_PROTOCOL)
8122		mode |= T4_FILTER_IP_PROTO;
8123
8124	if (fconf & F_TOS)
8125		mode |= T4_FILTER_IP_TOS;
8126
8127	if (fconf & F_VLAN)
8128		mode |= T4_FILTER_VLAN;
8129
8130	if (fconf & F_VNIC_ID) {
8131		mode |= T4_FILTER_VNIC;
8132		if (iconf & F_VNIC)
8133			mode |= T4_FILTER_IC_VNIC;
8134	}
8135
8136	if (fconf & F_PORT)
8137		mode |= T4_FILTER_PORT;
8138
8139	if (fconf & F_FCOE)
8140		mode |= T4_FILTER_FCoE;
8141
8142	return (mode);
8143}
8144
8145static uint32_t
8146mode_to_fconf(uint32_t mode)
8147{
8148	uint32_t fconf = 0;
8149
8150	if (mode & T4_FILTER_IP_FRAGMENT)
8151		fconf |= F_FRAGMENTATION;
8152
8153	if (mode & T4_FILTER_MPS_HIT_TYPE)
8154		fconf |= F_MPSHITTYPE;
8155
8156	if (mode & T4_FILTER_MAC_IDX)
8157		fconf |= F_MACMATCH;
8158
8159	if (mode & T4_FILTER_ETH_TYPE)
8160		fconf |= F_ETHERTYPE;
8161
8162	if (mode & T4_FILTER_IP_PROTO)
8163		fconf |= F_PROTOCOL;
8164
8165	if (mode & T4_FILTER_IP_TOS)
8166		fconf |= F_TOS;
8167
8168	if (mode & T4_FILTER_VLAN)
8169		fconf |= F_VLAN;
8170
8171	if (mode & T4_FILTER_VNIC)
8172		fconf |= F_VNIC_ID;
8173
8174	if (mode & T4_FILTER_PORT)
8175		fconf |= F_PORT;
8176
8177	if (mode & T4_FILTER_FCoE)
8178		fconf |= F_FCOE;
8179
8180	return (fconf);
8181}
8182
8183static uint32_t
8184mode_to_iconf(uint32_t mode)
8185{
8186
8187	if (mode & T4_FILTER_IC_VNIC)
8188		return (F_VNIC);
8189	return (0);
8190}
8191
8192static int check_fspec_against_fconf_iconf(struct adapter *sc,
8193    struct t4_filter_specification *fs)
8194{
8195	struct tp_params *tpp = &sc->params.tp;
8196	uint32_t fconf = 0;
8197
8198	if (fs->val.frag || fs->mask.frag)
8199		fconf |= F_FRAGMENTATION;
8200
8201	if (fs->val.matchtype || fs->mask.matchtype)
8202		fconf |= F_MPSHITTYPE;
8203
8204	if (fs->val.macidx || fs->mask.macidx)
8205		fconf |= F_MACMATCH;
8206
8207	if (fs->val.ethtype || fs->mask.ethtype)
8208		fconf |= F_ETHERTYPE;
8209
8210	if (fs->val.proto || fs->mask.proto)
8211		fconf |= F_PROTOCOL;
8212
8213	if (fs->val.tos || fs->mask.tos)
8214		fconf |= F_TOS;
8215
8216	if (fs->val.vlan_vld || fs->mask.vlan_vld)
8217		fconf |= F_VLAN;
8218
8219	if (fs->val.ovlan_vld || fs->mask.ovlan_vld) {
8220		fconf |= F_VNIC_ID;
8221		if (tpp->ingress_config & F_VNIC)
8222			return (EINVAL);
8223	}
8224
8225	if (fs->val.pfvf_vld || fs->mask.pfvf_vld) {
8226		fconf |= F_VNIC_ID;
8227		if ((tpp->ingress_config & F_VNIC) == 0)
8228			return (EINVAL);
8229	}
8230
8231	if (fs->val.iport || fs->mask.iport)
8232		fconf |= F_PORT;
8233
8234	if (fs->val.fcoe || fs->mask.fcoe)
8235		fconf |= F_FCOE;
8236
8237	if ((tpp->vlan_pri_map | fconf) != tpp->vlan_pri_map)
8238		return (E2BIG);
8239
8240	return (0);
8241}
8242
8243static int
8244get_filter_mode(struct adapter *sc, uint32_t *mode)
8245{
8246	struct tp_params *tpp = &sc->params.tp;
8247
8248	/*
8249	 * We trust the cached values of the relevant TP registers.  This means
8250	 * things work reliably only if writes to those registers are always via
8251	 * t4_set_filter_mode.
8252	 */
8253	*mode = fconf_iconf_to_mode(tpp->vlan_pri_map, tpp->ingress_config);
8254
8255	return (0);
8256}
8257
8258static int
8259set_filter_mode(struct adapter *sc, uint32_t mode)
8260{
8261	struct tp_params *tpp = &sc->params.tp;
8262	uint32_t fconf, iconf;
8263	int rc;
8264
8265	iconf = mode_to_iconf(mode);
8266	if ((iconf ^ tpp->ingress_config) & F_VNIC) {
8267		/*
8268		 * For now we just complain if A_TP_INGRESS_CONFIG is not
8269		 * already set to the correct value for the requested filter
8270		 * mode.  It's not clear if it's safe to write to this register
8271		 * on the fly.  (And we trust the cached value of the register).
8272		 */
8273		return (EBUSY);
8274	}
8275
8276	fconf = mode_to_fconf(mode);
8277
8278	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8279	    "t4setfm");
8280	if (rc)
8281		return (rc);
8282
8283	if (sc->tids.ftids_in_use > 0) {
8284		rc = EBUSY;
8285		goto done;
8286	}
8287
8288#ifdef TCP_OFFLOAD
8289	if (uld_active(sc, ULD_TOM)) {
8290		rc = EBUSY;
8291		goto done;
8292	}
8293#endif
8294
8295	rc = -t4_set_filter_mode(sc, fconf, true);
8296done:
8297	end_synchronized_op(sc, LOCK_HELD);
8298	return (rc);
8299}
8300
8301static inline uint64_t
8302get_filter_hits(struct adapter *sc, uint32_t fid)
8303{
8304	uint32_t tcb_addr;
8305
8306	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE) +
8307	    (fid + sc->tids.ftid_base) * TCB_SIZE;
8308
8309	if (is_t4(sc)) {
8310		uint64_t hits;
8311
8312		read_via_memwin(sc, 0, tcb_addr + 16, (uint32_t *)&hits, 8);
8313		return (be64toh(hits));
8314	} else {
8315		uint32_t hits;
8316
8317		read_via_memwin(sc, 0, tcb_addr + 24, &hits, 4);
8318		return (be32toh(hits));
8319	}
8320}
8321
8322static int
8323get_filter(struct adapter *sc, struct t4_filter *t)
8324{
8325	int i, rc, nfilters = sc->tids.nftids;
8326	struct filter_entry *f;
8327
8328	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
8329	    "t4getf");
8330	if (rc)
8331		return (rc);
8332
8333	if (sc->tids.ftids_in_use == 0 || sc->tids.ftid_tab == NULL ||
8334	    t->idx >= nfilters) {
8335		t->idx = 0xffffffff;
8336		goto done;
8337	}
8338
8339	f = &sc->tids.ftid_tab[t->idx];
8340	for (i = t->idx; i < nfilters; i++, f++) {
8341		if (f->valid) {
8342			t->idx = i;
8343			t->l2tidx = f->l2t ? f->l2t->idx : 0;
8344			t->smtidx = f->smtidx;
8345			if (f->fs.hitcnts)
8346				t->hits = get_filter_hits(sc, t->idx);
8347			else
8348				t->hits = UINT64_MAX;
8349			t->fs = f->fs;
8350
8351			goto done;
8352		}
8353	}
8354
8355	t->idx = 0xffffffff;
8356done:
8357	end_synchronized_op(sc, LOCK_HELD);
8358	return (0);
8359}
8360
8361static int
8362set_filter(struct adapter *sc, struct t4_filter *t)
8363{
8364	unsigned int nfilters, nports;
8365	struct filter_entry *f;
8366	int i, rc;
8367
8368	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setf");
8369	if (rc)
8370		return (rc);
8371
8372	nfilters = sc->tids.nftids;
8373	nports = sc->params.nports;
8374
8375	if (nfilters == 0) {
8376		rc = ENOTSUP;
8377		goto done;
8378	}
8379
8380	if (t->idx >= nfilters) {
8381		rc = EINVAL;
8382		goto done;
8383	}
8384
8385	/* Validate against the global filter mode and ingress config */
8386	rc = check_fspec_against_fconf_iconf(sc, &t->fs);
8387	if (rc != 0)
8388		goto done;
8389
8390	if (t->fs.action == FILTER_SWITCH && t->fs.eport >= nports) {
8391		rc = EINVAL;
8392		goto done;
8393	}
8394
8395	if (t->fs.val.iport >= nports) {
8396		rc = EINVAL;
8397		goto done;
8398	}
8399
8400	/* Can't specify an iq if not steering to it */
8401	if (!t->fs.dirsteer && t->fs.iq) {
8402		rc = EINVAL;
8403		goto done;
8404	}
8405
8406	/* IPv6 filter idx must be 4 aligned */
8407	if (t->fs.type == 1 &&
8408	    ((t->idx & 0x3) || t->idx + 4 >= nfilters)) {
8409		rc = EINVAL;
8410		goto done;
8411	}
8412
8413	if (!(sc->flags & FULL_INIT_DONE) &&
8414	    ((rc = adapter_full_init(sc)) != 0))
8415		goto done;
8416
8417	if (sc->tids.ftid_tab == NULL) {
8418		KASSERT(sc->tids.ftids_in_use == 0,
8419		    ("%s: no memory allocated but filters_in_use > 0",
8420		    __func__));
8421
8422		sc->tids.ftid_tab = malloc(sizeof (struct filter_entry) *
8423		    nfilters, M_CXGBE, M_NOWAIT | M_ZERO);
8424		if (sc->tids.ftid_tab == NULL) {
8425			rc = ENOMEM;
8426			goto done;
8427		}
8428		mtx_init(&sc->tids.ftid_lock, "T4 filters", 0, MTX_DEF);
8429	}
8430
8431	for (i = 0; i < 4; i++) {
8432		f = &sc->tids.ftid_tab[t->idx + i];
8433
8434		if (f->pending || f->valid) {
8435			rc = EBUSY;
8436			goto done;
8437		}
8438		if (f->locked) {
8439			rc = EPERM;
8440			goto done;
8441		}
8442
8443		if (t->fs.type == 0)
8444			break;
8445	}
8446
8447	f = &sc->tids.ftid_tab[t->idx];
8448	f->fs = t->fs;
8449
8450	rc = set_filter_wr(sc, t->idx);
8451done:
8452	end_synchronized_op(sc, 0);
8453
8454	if (rc == 0) {
8455		mtx_lock(&sc->tids.ftid_lock);
8456		for (;;) {
8457			if (f->pending == 0) {
8458				rc = f->valid ? 0 : EIO;
8459				break;
8460			}
8461
8462			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8463			    PCATCH, "t4setfw", 0)) {
8464				rc = EINPROGRESS;
8465				break;
8466			}
8467		}
8468		mtx_unlock(&sc->tids.ftid_lock);
8469	}
8470	return (rc);
8471}
8472
8473static int
8474del_filter(struct adapter *sc, struct t4_filter *t)
8475{
8476	unsigned int nfilters;
8477	struct filter_entry *f;
8478	int rc;
8479
8480	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4delf");
8481	if (rc)
8482		return (rc);
8483
8484	nfilters = sc->tids.nftids;
8485
8486	if (nfilters == 0) {
8487		rc = ENOTSUP;
8488		goto done;
8489	}
8490
8491	if (sc->tids.ftid_tab == NULL || sc->tids.ftids_in_use == 0 ||
8492	    t->idx >= nfilters) {
8493		rc = EINVAL;
8494		goto done;
8495	}
8496
8497	if (!(sc->flags & FULL_INIT_DONE)) {
8498		rc = EAGAIN;
8499		goto done;
8500	}
8501
8502	f = &sc->tids.ftid_tab[t->idx];
8503
8504	if (f->pending) {
8505		rc = EBUSY;
8506		goto done;
8507	}
8508	if (f->locked) {
8509		rc = EPERM;
8510		goto done;
8511	}
8512
8513	if (f->valid) {
8514		t->fs = f->fs;	/* extra info for the caller */
8515		rc = del_filter_wr(sc, t->idx);
8516	}
8517
8518done:
8519	end_synchronized_op(sc, 0);
8520
8521	if (rc == 0) {
8522		mtx_lock(&sc->tids.ftid_lock);
8523		for (;;) {
8524			if (f->pending == 0) {
8525				rc = f->valid ? EIO : 0;
8526				break;
8527			}
8528
8529			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8530			    PCATCH, "t4delfw", 0)) {
8531				rc = EINPROGRESS;
8532				break;
8533			}
8534		}
8535		mtx_unlock(&sc->tids.ftid_lock);
8536	}
8537
8538	return (rc);
8539}
8540
8541static void
8542clear_filter(struct filter_entry *f)
8543{
8544	if (f->l2t)
8545		t4_l2t_release(f->l2t);
8546
8547	bzero(f, sizeof (*f));
8548}
8549
8550static int
8551set_filter_wr(struct adapter *sc, int fidx)
8552{
8553	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8554	struct fw_filter_wr *fwr;
8555	unsigned int ftid, vnic_vld, vnic_vld_mask;
8556	struct wrq_cookie cookie;
8557
8558	ASSERT_SYNCHRONIZED_OP(sc);
8559
8560	if (f->fs.newdmac || f->fs.newvlan) {
8561		/* This filter needs an L2T entry; allocate one. */
8562		f->l2t = t4_l2t_alloc_switching(sc->l2t);
8563		if (f->l2t == NULL)
8564			return (EAGAIN);
8565		if (t4_l2t_set_switching(sc, f->l2t, f->fs.vlan, f->fs.eport,
8566		    f->fs.dmac)) {
8567			t4_l2t_release(f->l2t);
8568			f->l2t = NULL;
8569			return (ENOMEM);
8570		}
8571	}
8572
8573	/* Already validated against fconf, iconf */
8574	MPASS((f->fs.val.pfvf_vld & f->fs.val.ovlan_vld) == 0);
8575	MPASS((f->fs.mask.pfvf_vld & f->fs.mask.ovlan_vld) == 0);
8576	if (f->fs.val.pfvf_vld || f->fs.val.ovlan_vld)
8577		vnic_vld = 1;
8578	else
8579		vnic_vld = 0;
8580	if (f->fs.mask.pfvf_vld || f->fs.mask.ovlan_vld)
8581		vnic_vld_mask = 1;
8582	else
8583		vnic_vld_mask = 0;
8584
8585	ftid = sc->tids.ftid_base + fidx;
8586
8587	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8588	if (fwr == NULL)
8589		return (ENOMEM);
8590	bzero(fwr, sizeof(*fwr));
8591
8592	fwr->op_pkd = htobe32(V_FW_WR_OP(FW_FILTER_WR));
8593	fwr->len16_pkd = htobe32(FW_LEN16(*fwr));
8594	fwr->tid_to_iq =
8595	    htobe32(V_FW_FILTER_WR_TID(ftid) |
8596		V_FW_FILTER_WR_RQTYPE(f->fs.type) |
8597		V_FW_FILTER_WR_NOREPLY(0) |
8598		V_FW_FILTER_WR_IQ(f->fs.iq));
8599	fwr->del_filter_to_l2tix =
8600	    htobe32(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
8601		V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
8602		V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
8603		V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
8604		V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
8605		V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
8606		V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
8607		V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
8608		V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
8609		    f->fs.newvlan == VLAN_REWRITE) |
8610		V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
8611		    f->fs.newvlan == VLAN_REWRITE) |
8612		V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
8613		V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
8614		V_FW_FILTER_WR_PRIO(f->fs.prio) |
8615		V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
8616	fwr->ethtype = htobe16(f->fs.val.ethtype);
8617	fwr->ethtypem = htobe16(f->fs.mask.ethtype);
8618	fwr->frag_to_ovlan_vldm =
8619	    (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
8620		V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
8621		V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.vlan_vld) |
8622		V_FW_FILTER_WR_OVLAN_VLD(vnic_vld) |
8623		V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.vlan_vld) |
8624		V_FW_FILTER_WR_OVLAN_VLDM(vnic_vld_mask));
8625	fwr->smac_sel = 0;
8626	fwr->rx_chan_rx_rpl_iq = htobe16(V_FW_FILTER_WR_RX_CHAN(0) |
8627	    V_FW_FILTER_WR_RX_RPL_IQ(sc->sge.fwq.abs_id));
8628	fwr->maci_to_matchtypem =
8629	    htobe32(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
8630		V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
8631		V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
8632		V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
8633		V_FW_FILTER_WR_PORT(f->fs.val.iport) |
8634		V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
8635		V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
8636		V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
8637	fwr->ptcl = f->fs.val.proto;
8638	fwr->ptclm = f->fs.mask.proto;
8639	fwr->ttyp = f->fs.val.tos;
8640	fwr->ttypm = f->fs.mask.tos;
8641	fwr->ivlan = htobe16(f->fs.val.vlan);
8642	fwr->ivlanm = htobe16(f->fs.mask.vlan);
8643	fwr->ovlan = htobe16(f->fs.val.vnic);
8644	fwr->ovlanm = htobe16(f->fs.mask.vnic);
8645	bcopy(f->fs.val.dip, fwr->lip, sizeof (fwr->lip));
8646	bcopy(f->fs.mask.dip, fwr->lipm, sizeof (fwr->lipm));
8647	bcopy(f->fs.val.sip, fwr->fip, sizeof (fwr->fip));
8648	bcopy(f->fs.mask.sip, fwr->fipm, sizeof (fwr->fipm));
8649	fwr->lp = htobe16(f->fs.val.dport);
8650	fwr->lpm = htobe16(f->fs.mask.dport);
8651	fwr->fp = htobe16(f->fs.val.sport);
8652	fwr->fpm = htobe16(f->fs.mask.sport);
8653	if (f->fs.newsmac)
8654		bcopy(f->fs.smac, fwr->sma, sizeof (fwr->sma));
8655
8656	f->pending = 1;
8657	sc->tids.ftids_in_use++;
8658
8659	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8660	return (0);
8661}
8662
8663static int
8664del_filter_wr(struct adapter *sc, int fidx)
8665{
8666	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8667	struct fw_filter_wr *fwr;
8668	unsigned int ftid;
8669	struct wrq_cookie cookie;
8670
8671	ftid = sc->tids.ftid_base + fidx;
8672
8673	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8674	if (fwr == NULL)
8675		return (ENOMEM);
8676	bzero(fwr, sizeof (*fwr));
8677
8678	t4_mk_filtdelwr(ftid, fwr, sc->sge.fwq.abs_id);
8679
8680	f->pending = 1;
8681	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8682	return (0);
8683}
8684
8685int
8686t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8687{
8688	struct adapter *sc = iq->adapter;
8689	const struct cpl_set_tcb_rpl *rpl = (const void *)(rss + 1);
8690	unsigned int idx = GET_TID(rpl);
8691	unsigned int rc;
8692	struct filter_entry *f;
8693
8694	KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
8695	    rss->opcode));
8696	MPASS(iq == &sc->sge.fwq);
8697	MPASS(is_ftid(sc, idx));
8698
8699	idx -= sc->tids.ftid_base;
8700	f = &sc->tids.ftid_tab[idx];
8701	rc = G_COOKIE(rpl->cookie);
8702
8703	mtx_lock(&sc->tids.ftid_lock);
8704	if (rc == FW_FILTER_WR_FLT_ADDED) {
8705		KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
8706		    __func__, idx));
8707		f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
8708		f->pending = 0;  /* asynchronous setup completed */
8709		f->valid = 1;
8710	} else {
8711		if (rc != FW_FILTER_WR_FLT_DELETED) {
8712			/* Add or delete failed, display an error */
8713			log(LOG_ERR,
8714			    "filter %u setup failed with error %u\n",
8715			    idx, rc);
8716		}
8717
8718		clear_filter(f);
8719		sc->tids.ftids_in_use--;
8720	}
8721	wakeup(&sc->tids.ftid_tab);
8722	mtx_unlock(&sc->tids.ftid_lock);
8723
8724	return (0);
8725}
8726
8727static int
8728set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8729{
8730
8731	MPASS(iq->set_tcb_rpl != NULL);
8732	return (iq->set_tcb_rpl(iq, rss, m));
8733}
8734
8735static int
8736l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8737{
8738
8739	MPASS(iq->l2t_write_rpl != NULL);
8740	return (iq->l2t_write_rpl(iq, rss, m));
8741}
8742
8743static int
8744get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
8745{
8746	int rc;
8747
8748	if (cntxt->cid > M_CTXTQID)
8749		return (EINVAL);
8750
8751	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
8752	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
8753		return (EINVAL);
8754
8755	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
8756	if (rc)
8757		return (rc);
8758
8759	if (sc->flags & FW_OK) {
8760		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
8761		    &cntxt->data[0]);
8762		if (rc == 0)
8763			goto done;
8764	}
8765
8766	/*
8767	 * Read via firmware failed or wasn't even attempted.  Read directly via
8768	 * the backdoor.
8769	 */
8770	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
8771done:
8772	end_synchronized_op(sc, 0);
8773	return (rc);
8774}
8775
8776static int
8777load_fw(struct adapter *sc, struct t4_data *fw)
8778{
8779	int rc;
8780	uint8_t *fw_data;
8781
8782	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
8783	if (rc)
8784		return (rc);
8785
8786	if (sc->flags & FULL_INIT_DONE) {
8787		rc = EBUSY;
8788		goto done;
8789	}
8790
8791	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
8792	if (fw_data == NULL) {
8793		rc = ENOMEM;
8794		goto done;
8795	}
8796
8797	rc = copyin(fw->data, fw_data, fw->len);
8798	if (rc == 0)
8799		rc = -t4_load_fw(sc, fw_data, fw->len);
8800
8801	free(fw_data, M_CXGBE);
8802done:
8803	end_synchronized_op(sc, 0);
8804	return (rc);
8805}
8806
8807static int
8808load_cfg(struct adapter *sc, struct t4_data *cfg)
8809{
8810	int rc;
8811	uint8_t *cfg_data = NULL;
8812
8813	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
8814	if (rc)
8815		return (rc);
8816
8817	if (cfg->len == 0) {
8818		/* clear */
8819		rc = -t4_load_cfg(sc, NULL, 0);
8820		goto done;
8821	}
8822
8823	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
8824	if (cfg_data == NULL) {
8825		rc = ENOMEM;
8826		goto done;
8827	}
8828
8829	rc = copyin(cfg->data, cfg_data, cfg->len);
8830	if (rc == 0)
8831		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
8832
8833	free(cfg_data, M_CXGBE);
8834done:
8835	end_synchronized_op(sc, 0);
8836	return (rc);
8837}
8838
8839#define MAX_READ_BUF_SIZE (128 * 1024)
8840static int
8841read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
8842{
8843	uint32_t addr, remaining, n;
8844	uint32_t *buf;
8845	int rc;
8846	uint8_t *dst;
8847
8848	rc = validate_mem_range(sc, mr->addr, mr->len);
8849	if (rc != 0)
8850		return (rc);
8851
8852	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
8853	addr = mr->addr;
8854	remaining = mr->len;
8855	dst = (void *)mr->data;
8856
8857	while (remaining) {
8858		n = min(remaining, MAX_READ_BUF_SIZE);
8859		read_via_memwin(sc, 2, addr, buf, n);
8860
8861		rc = copyout(buf, dst, n);
8862		if (rc != 0)
8863			break;
8864
8865		dst += n;
8866		remaining -= n;
8867		addr += n;
8868	}
8869
8870	free(buf, M_CXGBE);
8871	return (rc);
8872}
8873#undef MAX_READ_BUF_SIZE
8874
8875static int
8876read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
8877{
8878	int rc;
8879
8880	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
8881		return (EINVAL);
8882
8883	if (i2cd->len > sizeof(i2cd->data))
8884		return (EFBIG);
8885
8886	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
8887	if (rc)
8888		return (rc);
8889	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
8890	    i2cd->offset, i2cd->len, &i2cd->data[0]);
8891	end_synchronized_op(sc, 0);
8892
8893	return (rc);
8894}
8895
8896int
8897t4_os_find_pci_capability(struct adapter *sc, int cap)
8898{
8899	int i;
8900
8901	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
8902}
8903
8904int
8905t4_os_pci_save_state(struct adapter *sc)
8906{
8907	device_t dev;
8908	struct pci_devinfo *dinfo;
8909
8910	dev = sc->dev;
8911	dinfo = device_get_ivars(dev);
8912
8913	pci_cfg_save(dev, dinfo, 0);
8914	return (0);
8915}
8916
8917int
8918t4_os_pci_restore_state(struct adapter *sc)
8919{
8920	device_t dev;
8921	struct pci_devinfo *dinfo;
8922
8923	dev = sc->dev;
8924	dinfo = device_get_ivars(dev);
8925
8926	pci_cfg_restore(dev, dinfo);
8927	return (0);
8928}
8929
8930void
8931t4_os_portmod_changed(struct port_info *pi)
8932{
8933	struct adapter *sc = pi->adapter;
8934	struct vi_info *vi;
8935	struct ifnet *ifp;
8936	static const char *mod_str[] = {
8937		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
8938	};
8939
8940	MPASS((pi->flags & FIXED_IFMEDIA) == 0);
8941
8942	vi = &pi->vi[0];
8943	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
8944		PORT_LOCK(pi);
8945		build_medialist(pi, &pi->media);
8946		apply_l1cfg(pi);
8947		PORT_UNLOCK(pi);
8948		end_synchronized_op(sc, LOCK_HELD);
8949	}
8950
8951	ifp = vi->ifp;
8952	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
8953		if_printf(ifp, "transceiver unplugged.\n");
8954	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
8955		if_printf(ifp, "unknown transceiver inserted.\n");
8956	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
8957		if_printf(ifp, "unsupported transceiver inserted.\n");
8958	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
8959		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
8960		    port_top_speed(pi), mod_str[pi->mod_type]);
8961	} else {
8962		if_printf(ifp, "transceiver (type %d) inserted.\n",
8963		    pi->mod_type);
8964	}
8965}
8966
8967void
8968t4_os_link_changed(struct port_info *pi)
8969{
8970	struct vi_info *vi;
8971	struct ifnet *ifp;
8972	struct link_config *lc;
8973	int v;
8974
8975	PORT_LOCK_ASSERT_OWNED(pi);
8976
8977	for_each_vi(pi, v, vi) {
8978		ifp = vi->ifp;
8979		if (ifp == NULL)
8980			continue;
8981
8982		lc = &pi->link_cfg;
8983		if (lc->link_ok) {
8984			ifp->if_baudrate = IF_Mbps(lc->speed);
8985			if_link_state_change(ifp, LINK_STATE_UP);
8986		} else {
8987			if_link_state_change(ifp, LINK_STATE_DOWN);
8988		}
8989	}
8990}
8991
8992void
8993t4_iterate(void (*func)(struct adapter *, void *), void *arg)
8994{
8995	struct adapter *sc;
8996
8997	sx_slock(&t4_list_lock);
8998	SLIST_FOREACH(sc, &t4_list, link) {
8999		/*
9000		 * func should not make any assumptions about what state sc is
9001		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9002		 */
9003		func(sc, arg);
9004	}
9005	sx_sunlock(&t4_list_lock);
9006}
9007
9008static int
9009t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9010    struct thread *td)
9011{
9012	int rc;
9013	struct adapter *sc = dev->si_drv1;
9014
9015	rc = priv_check(td, PRIV_DRIVER);
9016	if (rc != 0)
9017		return (rc);
9018
9019	switch (cmd) {
9020	case CHELSIO_T4_GETREG: {
9021		struct t4_reg *edata = (struct t4_reg *)data;
9022
9023		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9024			return (EFAULT);
9025
9026		if (edata->size == 4)
9027			edata->val = t4_read_reg(sc, edata->addr);
9028		else if (edata->size == 8)
9029			edata->val = t4_read_reg64(sc, edata->addr);
9030		else
9031			return (EINVAL);
9032
9033		break;
9034	}
9035	case CHELSIO_T4_SETREG: {
9036		struct t4_reg *edata = (struct t4_reg *)data;
9037
9038		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9039			return (EFAULT);
9040
9041		if (edata->size == 4) {
9042			if (edata->val & 0xffffffff00000000)
9043				return (EINVAL);
9044			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9045		} else if (edata->size == 8)
9046			t4_write_reg64(sc, edata->addr, edata->val);
9047		else
9048			return (EINVAL);
9049		break;
9050	}
9051	case CHELSIO_T4_REGDUMP: {
9052		struct t4_regdump *regs = (struct t4_regdump *)data;
9053		int reglen = t4_get_regs_len(sc);
9054		uint8_t *buf;
9055
9056		if (regs->len < reglen) {
9057			regs->len = reglen; /* hint to the caller */
9058			return (ENOBUFS);
9059		}
9060
9061		regs->len = reglen;
9062		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9063		get_regs(sc, regs, buf);
9064		rc = copyout(buf, regs->data, reglen);
9065		free(buf, M_CXGBE);
9066		break;
9067	}
9068	case CHELSIO_T4_GET_FILTER_MODE:
9069		rc = get_filter_mode(sc, (uint32_t *)data);
9070		break;
9071	case CHELSIO_T4_SET_FILTER_MODE:
9072		rc = set_filter_mode(sc, *(uint32_t *)data);
9073		break;
9074	case CHELSIO_T4_GET_FILTER:
9075		rc = get_filter(sc, (struct t4_filter *)data);
9076		break;
9077	case CHELSIO_T4_SET_FILTER:
9078		rc = set_filter(sc, (struct t4_filter *)data);
9079		break;
9080	case CHELSIO_T4_DEL_FILTER:
9081		rc = del_filter(sc, (struct t4_filter *)data);
9082		break;
9083	case CHELSIO_T4_GET_SGE_CONTEXT:
9084		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9085		break;
9086	case CHELSIO_T4_LOAD_FW:
9087		rc = load_fw(sc, (struct t4_data *)data);
9088		break;
9089	case CHELSIO_T4_GET_MEM:
9090		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9091		break;
9092	case CHELSIO_T4_GET_I2C:
9093		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9094		break;
9095	case CHELSIO_T4_CLEAR_STATS: {
9096		int i, v;
9097		u_int port_id = *(uint32_t *)data;
9098		struct port_info *pi;
9099		struct vi_info *vi;
9100
9101		if (port_id >= sc->params.nports)
9102			return (EINVAL);
9103		pi = sc->port[port_id];
9104		if (pi == NULL)
9105			return (EIO);
9106
9107		/* MAC stats */
9108		t4_clr_port_stats(sc, pi->tx_chan);
9109		pi->tx_parse_error = 0;
9110		mtx_lock(&sc->reg_lock);
9111		for_each_vi(pi, v, vi) {
9112			if (vi->flags & VI_INIT_DONE)
9113				t4_clr_vi_stats(sc, vi->viid);
9114		}
9115		mtx_unlock(&sc->reg_lock);
9116
9117		/*
9118		 * Since this command accepts a port, clear stats for
9119		 * all VIs on this port.
9120		 */
9121		for_each_vi(pi, v, vi) {
9122			if (vi->flags & VI_INIT_DONE) {
9123				struct sge_rxq *rxq;
9124				struct sge_txq *txq;
9125				struct sge_wrq *wrq;
9126
9127				for_each_rxq(vi, i, rxq) {
9128#if defined(INET) || defined(INET6)
9129					rxq->lro.lro_queued = 0;
9130					rxq->lro.lro_flushed = 0;
9131#endif
9132					rxq->rxcsum = 0;
9133					rxq->vlan_extraction = 0;
9134				}
9135
9136				for_each_txq(vi, i, txq) {
9137					txq->txcsum = 0;
9138					txq->tso_wrs = 0;
9139					txq->vlan_insertion = 0;
9140					txq->imm_wrs = 0;
9141					txq->sgl_wrs = 0;
9142					txq->txpkt_wrs = 0;
9143					txq->txpkts0_wrs = 0;
9144					txq->txpkts1_wrs = 0;
9145					txq->txpkts0_pkts = 0;
9146					txq->txpkts1_pkts = 0;
9147					mp_ring_reset_stats(txq->r);
9148				}
9149
9150#ifdef TCP_OFFLOAD
9151				/* nothing to clear for each ofld_rxq */
9152
9153				for_each_ofld_txq(vi, i, wrq) {
9154					wrq->tx_wrs_direct = 0;
9155					wrq->tx_wrs_copied = 0;
9156				}
9157#endif
9158
9159				if (IS_MAIN_VI(vi)) {
9160					wrq = &sc->sge.ctrlq[pi->port_id];
9161					wrq->tx_wrs_direct = 0;
9162					wrq->tx_wrs_copied = 0;
9163				}
9164			}
9165		}
9166		break;
9167	}
9168	case CHELSIO_T4_SCHED_CLASS:
9169		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9170		break;
9171	case CHELSIO_T4_SCHED_QUEUE:
9172		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9173		break;
9174	case CHELSIO_T4_GET_TRACER:
9175		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9176		break;
9177	case CHELSIO_T4_SET_TRACER:
9178		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9179		break;
9180	case CHELSIO_T4_LOAD_CFG:
9181		rc = load_cfg(sc, (struct t4_data *)data);
9182		break;
9183	default:
9184		rc = ENOTTY;
9185	}
9186
9187	return (rc);
9188}
9189
9190void
9191t4_db_full(struct adapter *sc)
9192{
9193
9194	CXGBE_UNIMPLEMENTED(__func__);
9195}
9196
9197void
9198t4_db_dropped(struct adapter *sc)
9199{
9200
9201	CXGBE_UNIMPLEMENTED(__func__);
9202}
9203
9204#ifdef TCP_OFFLOAD
9205void
9206t4_iscsi_init(struct adapter *sc, u_int tag_mask, const u_int *pgsz_order)
9207{
9208
9209	t4_write_reg(sc, A_ULP_RX_ISCSI_TAGMASK, tag_mask);
9210	t4_write_reg(sc, A_ULP_RX_ISCSI_PSZ, V_HPZ0(pgsz_order[0]) |
9211		V_HPZ1(pgsz_order[1]) | V_HPZ2(pgsz_order[2]) |
9212		V_HPZ3(pgsz_order[3]));
9213}
9214
9215static int
9216toe_capability(struct vi_info *vi, int enable)
9217{
9218	int rc;
9219	struct port_info *pi = vi->pi;
9220	struct adapter *sc = pi->adapter;
9221
9222	ASSERT_SYNCHRONIZED_OP(sc);
9223
9224	if (!is_offload(sc))
9225		return (ENODEV);
9226
9227	if (enable) {
9228		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
9229			/* TOE is already enabled. */
9230			return (0);
9231		}
9232
9233		/*
9234		 * We need the port's queues around so that we're able to send
9235		 * and receive CPLs to/from the TOE even if the ifnet for this
9236		 * port has never been UP'd administratively.
9237		 */
9238		if (!(vi->flags & VI_INIT_DONE)) {
9239			rc = vi_full_init(vi);
9240			if (rc)
9241				return (rc);
9242		}
9243		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
9244			rc = vi_full_init(&pi->vi[0]);
9245			if (rc)
9246				return (rc);
9247		}
9248
9249		if (isset(&sc->offload_map, pi->port_id)) {
9250			/* TOE is enabled on another VI of this port. */
9251			pi->uld_vis++;
9252			return (0);
9253		}
9254
9255		if (!uld_active(sc, ULD_TOM)) {
9256			rc = t4_activate_uld(sc, ULD_TOM);
9257			if (rc == EAGAIN) {
9258				log(LOG_WARNING,
9259				    "You must kldload t4_tom.ko before trying "
9260				    "to enable TOE on a cxgbe interface.\n");
9261			}
9262			if (rc != 0)
9263				return (rc);
9264			KASSERT(sc->tom_softc != NULL,
9265			    ("%s: TOM activated but softc NULL", __func__));
9266			KASSERT(uld_active(sc, ULD_TOM),
9267			    ("%s: TOM activated but flag not set", __func__));
9268		}
9269
9270		/* Activate iWARP and iSCSI too, if the modules are loaded. */
9271		if (!uld_active(sc, ULD_IWARP))
9272			(void) t4_activate_uld(sc, ULD_IWARP);
9273		if (!uld_active(sc, ULD_ISCSI))
9274			(void) t4_activate_uld(sc, ULD_ISCSI);
9275
9276		pi->uld_vis++;
9277		setbit(&sc->offload_map, pi->port_id);
9278	} else {
9279		pi->uld_vis--;
9280
9281		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
9282			return (0);
9283
9284		KASSERT(uld_active(sc, ULD_TOM),
9285		    ("%s: TOM never initialized?", __func__));
9286		clrbit(&sc->offload_map, pi->port_id);
9287	}
9288
9289	return (0);
9290}
9291
9292/*
9293 * Add an upper layer driver to the global list.
9294 */
9295int
9296t4_register_uld(struct uld_info *ui)
9297{
9298	int rc = 0;
9299	struct uld_info *u;
9300
9301	sx_xlock(&t4_uld_list_lock);
9302	SLIST_FOREACH(u, &t4_uld_list, link) {
9303	    if (u->uld_id == ui->uld_id) {
9304		    rc = EEXIST;
9305		    goto done;
9306	    }
9307	}
9308
9309	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
9310	ui->refcount = 0;
9311done:
9312	sx_xunlock(&t4_uld_list_lock);
9313	return (rc);
9314}
9315
9316int
9317t4_unregister_uld(struct uld_info *ui)
9318{
9319	int rc = EINVAL;
9320	struct uld_info *u;
9321
9322	sx_xlock(&t4_uld_list_lock);
9323
9324	SLIST_FOREACH(u, &t4_uld_list, link) {
9325	    if (u == ui) {
9326		    if (ui->refcount > 0) {
9327			    rc = EBUSY;
9328			    goto done;
9329		    }
9330
9331		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9332		    rc = 0;
9333		    goto done;
9334	    }
9335	}
9336done:
9337	sx_xunlock(&t4_uld_list_lock);
9338	return (rc);
9339}
9340
9341int
9342t4_activate_uld(struct adapter *sc, int id)
9343{
9344	int rc;
9345	struct uld_info *ui;
9346
9347	ASSERT_SYNCHRONIZED_OP(sc);
9348
9349	if (id < 0 || id > ULD_MAX)
9350		return (EINVAL);
9351	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
9352
9353	sx_slock(&t4_uld_list_lock);
9354
9355	SLIST_FOREACH(ui, &t4_uld_list, link) {
9356		if (ui->uld_id == id) {
9357			if (!(sc->flags & FULL_INIT_DONE)) {
9358				rc = adapter_full_init(sc);
9359				if (rc != 0)
9360					break;
9361			}
9362
9363			rc = ui->activate(sc);
9364			if (rc == 0) {
9365				setbit(&sc->active_ulds, id);
9366				ui->refcount++;
9367			}
9368			break;
9369		}
9370	}
9371
9372	sx_sunlock(&t4_uld_list_lock);
9373
9374	return (rc);
9375}
9376
9377int
9378t4_deactivate_uld(struct adapter *sc, int id)
9379{
9380	int rc;
9381	struct uld_info *ui;
9382
9383	ASSERT_SYNCHRONIZED_OP(sc);
9384
9385	if (id < 0 || id > ULD_MAX)
9386		return (EINVAL);
9387	rc = ENXIO;
9388
9389	sx_slock(&t4_uld_list_lock);
9390
9391	SLIST_FOREACH(ui, &t4_uld_list, link) {
9392		if (ui->uld_id == id) {
9393			rc = ui->deactivate(sc);
9394			if (rc == 0) {
9395				clrbit(&sc->active_ulds, id);
9396				ui->refcount--;
9397			}
9398			break;
9399		}
9400	}
9401
9402	sx_sunlock(&t4_uld_list_lock);
9403
9404	return (rc);
9405}
9406
9407int
9408uld_active(struct adapter *sc, int uld_id)
9409{
9410
9411	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9412
9413	return (isset(&sc->active_ulds, uld_id));
9414}
9415#endif
9416
9417/*
9418 * t  = ptr to tunable.
9419 * nc = number of CPUs.
9420 * c  = compiled in default for that tunable.
9421 */
9422static void
9423calculate_nqueues(int *t, int nc, const int c)
9424{
9425	int nq;
9426
9427	if (*t > 0)
9428		return;
9429	nq = *t < 0 ? -*t : c;
9430	*t = min(nc, nq);
9431}
9432
9433/*
9434 * Come up with reasonable defaults for some of the tunables, provided they're
9435 * not set by the user (in which case we'll use the values as is).
9436 */
9437static void
9438tweak_tunables(void)
9439{
9440	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
9441
9442	if (t4_ntxq10g < 1) {
9443#ifdef RSS
9444		t4_ntxq10g = rss_getnumbuckets();
9445#else
9446		calculate_nqueues(&t4_ntxq10g, nc, NTXQ_10G);
9447#endif
9448	}
9449
9450	if (t4_ntxq1g < 1) {
9451#ifdef RSS
9452		/* XXX: way too many for 1GbE? */
9453		t4_ntxq1g = rss_getnumbuckets();
9454#else
9455		calculate_nqueues(&t4_ntxq1g, nc, NTXQ_1G);
9456#endif
9457	}
9458
9459	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
9460
9461	if (t4_nrxq10g < 1) {
9462#ifdef RSS
9463		t4_nrxq10g = rss_getnumbuckets();
9464#else
9465		calculate_nqueues(&t4_nrxq10g, nc, NRXQ_10G);
9466#endif
9467	}
9468
9469	if (t4_nrxq1g < 1) {
9470#ifdef RSS
9471		/* XXX: way too many for 1GbE? */
9472		t4_nrxq1g = rss_getnumbuckets();
9473#else
9474		calculate_nqueues(&t4_nrxq1g, nc, NRXQ_1G);
9475#endif
9476	}
9477
9478	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
9479
9480#ifdef TCP_OFFLOAD
9481	calculate_nqueues(&t4_nofldtxq10g, nc, NOFLDTXQ_10G);
9482	calculate_nqueues(&t4_nofldtxq1g, nc, NOFLDTXQ_1G);
9483	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
9484	calculate_nqueues(&t4_nofldrxq10g, nc, NOFLDRXQ_10G);
9485	calculate_nqueues(&t4_nofldrxq1g, nc, NOFLDRXQ_1G);
9486	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
9487
9488	if (t4_toecaps_allowed == -1)
9489		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9490
9491	if (t4_rdmacaps_allowed == -1) {
9492		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9493		    FW_CAPS_CONFIG_RDMA_RDMAC;
9494	}
9495
9496	if (t4_iscsicaps_allowed == -1) {
9497		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9498		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9499		    FW_CAPS_CONFIG_ISCSI_T10DIF;
9500	}
9501#else
9502	if (t4_toecaps_allowed == -1)
9503		t4_toecaps_allowed = 0;
9504
9505	if (t4_rdmacaps_allowed == -1)
9506		t4_rdmacaps_allowed = 0;
9507
9508	if (t4_iscsicaps_allowed == -1)
9509		t4_iscsicaps_allowed = 0;
9510#endif
9511
9512#ifdef DEV_NETMAP
9513	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
9514	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
9515#endif
9516
9517	if (t4_tmr_idx_10g < 0 || t4_tmr_idx_10g >= SGE_NTIMERS)
9518		t4_tmr_idx_10g = TMR_IDX_10G;
9519
9520	if (t4_pktc_idx_10g < -1 || t4_pktc_idx_10g >= SGE_NCOUNTERS)
9521		t4_pktc_idx_10g = PKTC_IDX_10G;
9522
9523	if (t4_tmr_idx_1g < 0 || t4_tmr_idx_1g >= SGE_NTIMERS)
9524		t4_tmr_idx_1g = TMR_IDX_1G;
9525
9526	if (t4_pktc_idx_1g < -1 || t4_pktc_idx_1g >= SGE_NCOUNTERS)
9527		t4_pktc_idx_1g = PKTC_IDX_1G;
9528
9529	if (t4_qsize_txq < 128)
9530		t4_qsize_txq = 128;
9531
9532	if (t4_qsize_rxq < 128)
9533		t4_qsize_rxq = 128;
9534	while (t4_qsize_rxq & 7)
9535		t4_qsize_rxq++;
9536
9537	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9538}
9539
9540#ifdef DDB
9541static void
9542t4_dump_tcb(struct adapter *sc, int tid)
9543{
9544	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
9545
9546	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
9547	save = t4_read_reg(sc, reg);
9548	base = sc->memwin[2].mw_base;
9549
9550	/* Dump TCB for the tid */
9551	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
9552	tcb_addr += tid * TCB_SIZE;
9553
9554	if (is_t4(sc)) {
9555		pf = 0;
9556		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
9557	} else {
9558		pf = V_PFNUM(sc->pf);
9559		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
9560	}
9561	t4_write_reg(sc, reg, win_pos | pf);
9562	t4_read_reg(sc, reg);
9563
9564	off = tcb_addr - win_pos;
9565	for (i = 0; i < 4; i++) {
9566		uint32_t buf[8];
9567		for (j = 0; j < 8; j++, off += 4)
9568			buf[j] = htonl(t4_read_reg(sc, base + off));
9569
9570		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
9571		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
9572		    buf[7]);
9573	}
9574
9575	t4_write_reg(sc, reg, save);
9576	t4_read_reg(sc, reg);
9577}
9578
9579static void
9580t4_dump_devlog(struct adapter *sc)
9581{
9582	struct devlog_params *dparams = &sc->params.devlog;
9583	struct fw_devlog_e e;
9584	int i, first, j, m, nentries, rc;
9585	uint64_t ftstamp = UINT64_MAX;
9586
9587	if (dparams->start == 0) {
9588		db_printf("devlog params not valid\n");
9589		return;
9590	}
9591
9592	nentries = dparams->size / sizeof(struct fw_devlog_e);
9593	m = fwmtype_to_hwmtype(dparams->memtype);
9594
9595	/* Find the first entry. */
9596	first = -1;
9597	for (i = 0; i < nentries && !db_pager_quit; i++) {
9598		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9599		    sizeof(e), (void *)&e);
9600		if (rc != 0)
9601			break;
9602
9603		if (e.timestamp == 0)
9604			break;
9605
9606		e.timestamp = be64toh(e.timestamp);
9607		if (e.timestamp < ftstamp) {
9608			ftstamp = e.timestamp;
9609			first = i;
9610		}
9611	}
9612
9613	if (first == -1)
9614		return;
9615
9616	i = first;
9617	do {
9618		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9619		    sizeof(e), (void *)&e);
9620		if (rc != 0)
9621			return;
9622
9623		if (e.timestamp == 0)
9624			return;
9625
9626		e.timestamp = be64toh(e.timestamp);
9627		e.seqno = be32toh(e.seqno);
9628		for (j = 0; j < 8; j++)
9629			e.params[j] = be32toh(e.params[j]);
9630
9631		db_printf("%10d  %15ju  %8s  %8s  ",
9632		    e.seqno, e.timestamp,
9633		    (e.level < nitems(devlog_level_strings) ?
9634			devlog_level_strings[e.level] : "UNKNOWN"),
9635		    (e.facility < nitems(devlog_facility_strings) ?
9636			devlog_facility_strings[e.facility] : "UNKNOWN"));
9637		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
9638		    e.params[3], e.params[4], e.params[5], e.params[6],
9639		    e.params[7]);
9640
9641		if (++i == nentries)
9642			i = 0;
9643	} while (i != first && !db_pager_quit);
9644}
9645
9646static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
9647_DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
9648
9649DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
9650{
9651	device_t dev;
9652	int t;
9653	bool valid;
9654
9655	valid = false;
9656	t = db_read_token();
9657	if (t == tIDENT) {
9658		dev = device_lookup_by_name(db_tok_string);
9659		valid = true;
9660	}
9661	db_skip_to_eol();
9662	if (!valid) {
9663		db_printf("usage: show t4 devlog <nexus>\n");
9664		return;
9665	}
9666
9667	if (dev == NULL) {
9668		db_printf("device not found\n");
9669		return;
9670	}
9671
9672	t4_dump_devlog(device_get_softc(dev));
9673}
9674
9675DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
9676{
9677	device_t dev;
9678	int radix, tid, t;
9679	bool valid;
9680
9681	valid = false;
9682	radix = db_radix;
9683	db_radix = 10;
9684	t = db_read_token();
9685	if (t == tIDENT) {
9686		dev = device_lookup_by_name(db_tok_string);
9687		t = db_read_token();
9688		if (t == tNUMBER) {
9689			tid = db_tok_number;
9690			valid = true;
9691		}
9692	}
9693	db_radix = radix;
9694	db_skip_to_eol();
9695	if (!valid) {
9696		db_printf("usage: show t4 tcb <nexus> <tid>\n");
9697		return;
9698	}
9699
9700	if (dev == NULL) {
9701		db_printf("device not found\n");
9702		return;
9703	}
9704	if (tid < 0) {
9705		db_printf("invalid tid\n");
9706		return;
9707	}
9708
9709	t4_dump_tcb(device_get_softc(dev), tid);
9710}
9711#endif
9712
9713static struct sx mlu;	/* mod load unload */
9714SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
9715
9716static int
9717mod_event(module_t mod, int cmd, void *arg)
9718{
9719	int rc = 0;
9720	static int loaded = 0;
9721
9722	switch (cmd) {
9723	case MOD_LOAD:
9724		sx_xlock(&mlu);
9725		if (loaded++ == 0) {
9726			t4_sge_modload();
9727			t4_register_cpl_handler(CPL_SET_TCB_RPL, set_tcb_rpl);
9728			t4_register_cpl_handler(CPL_L2T_WRITE_RPL, l2t_write_rpl);
9729			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
9730			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
9731			sx_init(&t4_list_lock, "T4/T5 adapters");
9732			SLIST_INIT(&t4_list);
9733#ifdef TCP_OFFLOAD
9734			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
9735			SLIST_INIT(&t4_uld_list);
9736#endif
9737			t4_tracer_modload();
9738			tweak_tunables();
9739		}
9740		sx_xunlock(&mlu);
9741		break;
9742
9743	case MOD_UNLOAD:
9744		sx_xlock(&mlu);
9745		if (--loaded == 0) {
9746			int tries;
9747
9748			sx_slock(&t4_list_lock);
9749			if (!SLIST_EMPTY(&t4_list)) {
9750				rc = EBUSY;
9751				sx_sunlock(&t4_list_lock);
9752				goto done_unload;
9753			}
9754#ifdef TCP_OFFLOAD
9755			sx_slock(&t4_uld_list_lock);
9756			if (!SLIST_EMPTY(&t4_uld_list)) {
9757				rc = EBUSY;
9758				sx_sunlock(&t4_uld_list_lock);
9759				sx_sunlock(&t4_list_lock);
9760				goto done_unload;
9761			}
9762#endif
9763			tries = 0;
9764			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
9765				uprintf("%ju clusters with custom free routine "
9766				    "still is use.\n", t4_sge_extfree_refs());
9767				pause("t4unload", 2 * hz);
9768			}
9769#ifdef TCP_OFFLOAD
9770			sx_sunlock(&t4_uld_list_lock);
9771#endif
9772			sx_sunlock(&t4_list_lock);
9773
9774			if (t4_sge_extfree_refs() == 0) {
9775				t4_tracer_modunload();
9776#ifdef TCP_OFFLOAD
9777				sx_destroy(&t4_uld_list_lock);
9778#endif
9779				sx_destroy(&t4_list_lock);
9780				t4_sge_modunload();
9781				loaded = 0;
9782			} else {
9783				rc = EBUSY;
9784				loaded++;	/* undo earlier decrement */
9785			}
9786		}
9787done_unload:
9788		sx_xunlock(&mlu);
9789		break;
9790	}
9791
9792	return (rc);
9793}
9794
9795static devclass_t t4_devclass, t5_devclass, t6_devclass;
9796static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
9797static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
9798
9799DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
9800MODULE_VERSION(t4nex, 1);
9801MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
9802
9803DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
9804MODULE_VERSION(t5nex, 1);
9805MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
9806
9807DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
9808MODULE_VERSION(t6nex, 1);
9809MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
9810#ifdef DEV_NETMAP
9811MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
9812#endif /* DEV_NETMAP */
9813
9814DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
9815MODULE_VERSION(cxgbe, 1);
9816
9817DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
9818MODULE_VERSION(cxl, 1);
9819
9820DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
9821MODULE_VERSION(cc, 1);
9822
9823DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
9824MODULE_VERSION(vcxgbe, 1);
9825
9826DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
9827MODULE_VERSION(vcxl, 1);
9828
9829DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
9830MODULE_VERSION(vcc, 1);
9831