t4_main.c revision 309447
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 309447 2016-12-02 22:53:33Z jhb $");
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/* ifnet + media interface */
170static void cxgbe_init(void *);
171static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
172static int cxgbe_transmit(struct ifnet *, struct mbuf *);
173static void cxgbe_qflush(struct ifnet *);
174static int cxgbe_media_change(struct ifnet *);
175static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
176
177MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
178
179/*
180 * Correct lock order when you need to acquire multiple locks is t4_list_lock,
181 * then ADAPTER_LOCK, then t4_uld_list_lock.
182 */
183static struct sx t4_list_lock;
184SLIST_HEAD(, adapter) t4_list;
185#ifdef TCP_OFFLOAD
186static struct sx t4_uld_list_lock;
187SLIST_HEAD(, uld_info) t4_uld_list;
188#endif
189
190/*
191 * Tunables.  See tweak_tunables() too.
192 *
193 * Each tunable is set to a default value here if it's known at compile-time.
194 * Otherwise it is set to -1 as an indication to tweak_tunables() that it should
195 * provide a reasonable default when the driver is loaded.
196 *
197 * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
198 * T5 are under hw.cxl.
199 */
200
201/*
202 * Number of queues for tx and rx, 10G and 1G, NIC and offload.
203 */
204#define NTXQ_10G 16
205int t4_ntxq10g = -1;
206TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq10g);
207
208#define NRXQ_10G 8
209int t4_nrxq10g = -1;
210TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq10g);
211
212#define NTXQ_1G 4
213int t4_ntxq1g = -1;
214TUNABLE_INT("hw.cxgbe.ntxq1g", &t4_ntxq1g);
215
216#define NRXQ_1G 2
217int t4_nrxq1g = -1;
218TUNABLE_INT("hw.cxgbe.nrxq1g", &t4_nrxq1g);
219
220#define NTXQ_VI 1
221static int t4_ntxq_vi = -1;
222TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
223
224#define NRXQ_VI 1
225static int t4_nrxq_vi = -1;
226TUNABLE_INT("hw.cxgbe.nrxq_vi", &t4_nrxq_vi);
227
228static int t4_rsrv_noflowq = 0;
229TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
230
231#ifdef TCP_OFFLOAD
232#define NOFLDTXQ_10G 8
233static int t4_nofldtxq10g = -1;
234TUNABLE_INT("hw.cxgbe.nofldtxq10g", &t4_nofldtxq10g);
235
236#define NOFLDRXQ_10G 2
237static int t4_nofldrxq10g = -1;
238TUNABLE_INT("hw.cxgbe.nofldrxq10g", &t4_nofldrxq10g);
239
240#define NOFLDTXQ_1G 2
241static int t4_nofldtxq1g = -1;
242TUNABLE_INT("hw.cxgbe.nofldtxq1g", &t4_nofldtxq1g);
243
244#define NOFLDRXQ_1G 1
245static int t4_nofldrxq1g = -1;
246TUNABLE_INT("hw.cxgbe.nofldrxq1g", &t4_nofldrxq1g);
247
248#define NOFLDTXQ_VI 1
249static int t4_nofldtxq_vi = -1;
250TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
251
252#define NOFLDRXQ_VI 1
253static int t4_nofldrxq_vi = -1;
254TUNABLE_INT("hw.cxgbe.nofldrxq_vi", &t4_nofldrxq_vi);
255#endif
256
257#ifdef DEV_NETMAP
258#define NNMTXQ_VI 2
259static int t4_nnmtxq_vi = -1;
260TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
261
262#define NNMRXQ_VI 2
263static int t4_nnmrxq_vi = -1;
264TUNABLE_INT("hw.cxgbe.nnmrxq_vi", &t4_nnmrxq_vi);
265#endif
266
267/*
268 * Holdoff parameters for 10G and 1G ports.
269 */
270#define TMR_IDX_10G 1
271int t4_tmr_idx_10g = TMR_IDX_10G;
272TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx_10g);
273
274#define PKTC_IDX_10G (-1)
275int t4_pktc_idx_10g = PKTC_IDX_10G;
276TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx_10g);
277
278#define TMR_IDX_1G 1
279int t4_tmr_idx_1g = TMR_IDX_1G;
280TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_1G", &t4_tmr_idx_1g);
281
282#define PKTC_IDX_1G (-1)
283int t4_pktc_idx_1g = PKTC_IDX_1G;
284TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_1G", &t4_pktc_idx_1g);
285
286/*
287 * Size (# of entries) of each tx and rx queue.
288 */
289unsigned int t4_qsize_txq = TX_EQ_QSIZE;
290TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
291
292unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
293TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
294
295/*
296 * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
297 */
298int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
299TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
300
301/*
302 * Configuration file.
303 */
304#define DEFAULT_CF	"default"
305#define FLASH_CF	"flash"
306#define UWIRE_CF	"uwire"
307#define FPGA_CF		"fpga"
308static char t4_cfg_file[32] = DEFAULT_CF;
309TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
310
311/*
312 * PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
313 * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
314 * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
315 *            mark or when signalled to do so, 0 to never emit PAUSE.
316 */
317static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
318TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
319
320/*
321 * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
322 * encouraged respectively).
323 */
324static unsigned int t4_fw_install = 1;
325TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
326
327/*
328 * ASIC features that will be used.  Disable the ones you don't want so that the
329 * chip resources aren't wasted on features that will not be used.
330 */
331static int t4_nbmcaps_allowed = 0;
332TUNABLE_INT("hw.cxgbe.nbmcaps_allowed", &t4_nbmcaps_allowed);
333
334static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
335TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
336
337static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
338    FW_CAPS_CONFIG_SWITCH_EGRESS;
339TUNABLE_INT("hw.cxgbe.switchcaps_allowed", &t4_switchcaps_allowed);
340
341static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC;
342TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
343
344static int t4_toecaps_allowed = -1;
345TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
346
347static int t4_rdmacaps_allowed = -1;
348TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
349
350static int t4_tlscaps_allowed = 0;
351TUNABLE_INT("hw.cxgbe.tlscaps_allowed", &t4_tlscaps_allowed);
352
353static int t4_iscsicaps_allowed = -1;
354TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
355
356static int t4_fcoecaps_allowed = 0;
357TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
358
359static int t5_write_combine = 0;
360TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
361
362static int t4_num_vis = 1;
363TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
364
365/* Functions used by extra VIs to obtain unique MAC addresses for each VI. */
366static int vi_mac_funcs[] = {
367	FW_VI_FUNC_OFLD,
368	FW_VI_FUNC_IWARP,
369	FW_VI_FUNC_OPENISCSI,
370	FW_VI_FUNC_OPENFCOE,
371	FW_VI_FUNC_FOISCSI,
372	FW_VI_FUNC_FOFCOE,
373};
374
375struct intrs_and_queues {
376	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
377	uint16_t nirq;		/* Total # of vectors */
378	uint16_t intr_flags_10g;/* Interrupt flags for each 10G port */
379	uint16_t intr_flags_1g;	/* Interrupt flags for each 1G port */
380	uint16_t ntxq10g;	/* # of NIC txq's for each 10G port */
381	uint16_t nrxq10g;	/* # of NIC rxq's for each 10G port */
382	uint16_t ntxq1g;	/* # of NIC txq's for each 1G port */
383	uint16_t nrxq1g;	/* # of NIC rxq's for each 1G port */
384	uint16_t rsrv_noflowq;	/* Flag whether to reserve queue 0 */
385	uint16_t nofldtxq10g;	/* # of TOE txq's for each 10G port */
386	uint16_t nofldrxq10g;	/* # of TOE rxq's for each 10G port */
387	uint16_t nofldtxq1g;	/* # of TOE txq's for each 1G port */
388	uint16_t nofldrxq1g;	/* # of TOE rxq's for each 1G port */
389
390	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
391	uint16_t ntxq_vi;	/* # of NIC txq's */
392	uint16_t nrxq_vi;	/* # of NIC rxq's */
393	uint16_t nofldtxq_vi;	/* # of TOE txq's */
394	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
395	uint16_t nnmtxq_vi;	/* # of netmap txq's */
396	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
397};
398
399struct filter_entry {
400        uint32_t valid:1;	/* filter allocated and valid */
401        uint32_t locked:1;	/* filter is administratively locked */
402        uint32_t pending:1;	/* filter action is pending firmware reply */
403	uint32_t smtidx:8;	/* Source MAC Table index for smac */
404	struct l2t_entry *l2t;	/* Layer Two Table entry for dmac */
405
406        struct t4_filter_specification fs;
407};
408
409static void setup_memwin(struct adapter *);
410static void position_memwin(struct adapter *, int, uint32_t);
411static int rw_via_memwin(struct adapter *, int, uint32_t, uint32_t *, int, int);
412static inline int read_via_memwin(struct adapter *, int, uint32_t, uint32_t *,
413    int);
414static inline int write_via_memwin(struct adapter *, int, uint32_t,
415    const uint32_t *, int);
416static int validate_mem_range(struct adapter *, uint32_t, int);
417static int fwmtype_to_hwmtype(int);
418static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
419    uint32_t *);
420static int fixup_devlog_params(struct adapter *);
421static int cfg_itype_and_nqueues(struct adapter *, int, int, int,
422    struct intrs_and_queues *);
423static int prep_firmware(struct adapter *);
424static int partition_resources(struct adapter *, const struct firmware *,
425    const char *);
426static int get_params__pre_init(struct adapter *);
427static int get_params__post_init(struct adapter *);
428static int set_params__post_init(struct adapter *);
429static void t4_set_desc(struct adapter *);
430static void build_medialist(struct port_info *, struct ifmedia *);
431static int cxgbe_init_synchronized(struct vi_info *);
432static int cxgbe_uninit_synchronized(struct vi_info *);
433static void quiesce_txq(struct adapter *, struct sge_txq *);
434static void quiesce_wrq(struct adapter *, struct sge_wrq *);
435static void quiesce_iq(struct adapter *, struct sge_iq *);
436static void quiesce_fl(struct adapter *, struct sge_fl *);
437static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
438    driver_intr_t *, void *, char *);
439static int t4_free_irq(struct adapter *, struct irq *);
440static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
441static void vi_refresh_stats(struct adapter *, struct vi_info *);
442static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
443static void cxgbe_tick(void *);
444static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
445static void cxgbe_sysctls(struct port_info *);
446static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
447static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
448static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
449static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
450static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
451static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
452static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
453static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
454static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
455static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
456static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
457#ifdef SBUF_DRAIN
458static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
459static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
460static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
461static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
462static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
463static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
464static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
465static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
466static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
467static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
468static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
469static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
470static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
471static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
472static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
473static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
474static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
475static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
476static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
477static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
478static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
479static int sysctl_tids(SYSCTL_HANDLER_ARGS);
480static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
481static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
482static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
483static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
484static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
485static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
486static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
487#endif
488#ifdef TCP_OFFLOAD
489static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
490static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
491static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
492#endif
493static uint32_t fconf_iconf_to_mode(uint32_t, uint32_t);
494static uint32_t mode_to_fconf(uint32_t);
495static uint32_t mode_to_iconf(uint32_t);
496static int check_fspec_against_fconf_iconf(struct adapter *,
497    struct t4_filter_specification *);
498static int get_filter_mode(struct adapter *, uint32_t *);
499static int set_filter_mode(struct adapter *, uint32_t);
500static inline uint64_t get_filter_hits(struct adapter *, uint32_t);
501static int get_filter(struct adapter *, struct t4_filter *);
502static int set_filter(struct adapter *, struct t4_filter *);
503static int del_filter(struct adapter *, struct t4_filter *);
504static void clear_filter(struct filter_entry *);
505static int set_filter_wr(struct adapter *, int);
506static int del_filter_wr(struct adapter *, int);
507static int set_tcb_rpl(struct sge_iq *, const struct rss_header *,
508    struct mbuf *);
509static int get_sge_context(struct adapter *, struct t4_sge_context *);
510static int load_fw(struct adapter *, struct t4_data *);
511static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
512static int read_i2c(struct adapter *, struct t4_i2c_data *);
513#ifdef TCP_OFFLOAD
514static int toe_capability(struct vi_info *, int);
515#endif
516static int mod_event(module_t, int, void *);
517
518struct {
519	uint16_t device;
520	char *desc;
521} t4_pciids[] = {
522	{0xa000, "Chelsio Terminator 4 FPGA"},
523	{0x4400, "Chelsio T440-dbg"},
524	{0x4401, "Chelsio T420-CR"},
525	{0x4402, "Chelsio T422-CR"},
526	{0x4403, "Chelsio T440-CR"},
527	{0x4404, "Chelsio T420-BCH"},
528	{0x4405, "Chelsio T440-BCH"},
529	{0x4406, "Chelsio T440-CH"},
530	{0x4407, "Chelsio T420-SO"},
531	{0x4408, "Chelsio T420-CX"},
532	{0x4409, "Chelsio T420-BT"},
533	{0x440a, "Chelsio T404-BT"},
534	{0x440e, "Chelsio T440-LP-CR"},
535}, t5_pciids[] = {
536	{0xb000, "Chelsio Terminator 5 FPGA"},
537	{0x5400, "Chelsio T580-dbg"},
538	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
539	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
540	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
541	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
542	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
543	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
544	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
545	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
546	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
547	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
548	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
549	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
550	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
551#ifdef notyet
552	{0x5404,  "Chelsio T520-BCH"},
553	{0x5405,  "Chelsio T540-BCH"},
554	{0x5406,  "Chelsio T540-CH"},
555	{0x5408,  "Chelsio T520-CX"},
556	{0x540b,  "Chelsio B520-SR"},
557	{0x540c,  "Chelsio B504-BT"},
558	{0x540f,  "Chelsio Amsterdam"},
559	{0x5413,  "Chelsio T580-CHR"},
560#endif
561};
562
563#ifdef TCP_OFFLOAD
564/*
565 * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
566 * exactly the same for both rxq and ofld_rxq.
567 */
568CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
569CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
570#endif
571CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
572
573static int
574t4_probe(device_t dev)
575{
576	int i;
577	uint16_t v = pci_get_vendor(dev);
578	uint16_t d = pci_get_device(dev);
579	uint8_t f = pci_get_function(dev);
580
581	if (v != PCI_VENDOR_ID_CHELSIO)
582		return (ENXIO);
583
584	/* Attach only to PF0 of the FPGA */
585	if (d == 0xa000 && f != 0)
586		return (ENXIO);
587
588	for (i = 0; i < nitems(t4_pciids); i++) {
589		if (d == t4_pciids[i].device) {
590			device_set_desc(dev, t4_pciids[i].desc);
591			return (BUS_PROBE_DEFAULT);
592		}
593	}
594
595	return (ENXIO);
596}
597
598static int
599t5_probe(device_t dev)
600{
601	int i;
602	uint16_t v = pci_get_vendor(dev);
603	uint16_t d = pci_get_device(dev);
604	uint8_t f = pci_get_function(dev);
605
606	if (v != PCI_VENDOR_ID_CHELSIO)
607		return (ENXIO);
608
609	/* Attach only to PF0 of the FPGA */
610	if (d == 0xb000 && f != 0)
611		return (ENXIO);
612
613	for (i = 0; i < nitems(t5_pciids); i++) {
614		if (d == t5_pciids[i].device) {
615			device_set_desc(dev, t5_pciids[i].desc);
616			return (BUS_PROBE_DEFAULT);
617		}
618	}
619
620	return (ENXIO);
621}
622
623static void
624t5_attribute_workaround(device_t dev)
625{
626	device_t root_port;
627	uint32_t v;
628
629	/*
630	 * The T5 chips do not properly echo the No Snoop and Relaxed
631	 * Ordering attributes when replying to a TLP from a Root
632	 * Port.  As a workaround, find the parent Root Port and
633	 * disable No Snoop and Relaxed Ordering.  Note that this
634	 * affects all devices under this root port.
635	 */
636	root_port = pci_find_pcie_root_port(dev);
637	if (root_port == NULL) {
638		device_printf(dev, "Unable to find parent root port\n");
639		return;
640	}
641
642	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
643	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
644	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
645	    0)
646		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
647		    device_get_nameunit(root_port));
648}
649
650static int
651t4_attach(device_t dev)
652{
653	struct adapter *sc;
654	int rc = 0, i, j, n10g, n1g, rqidx, tqidx;
655	struct make_dev_args mda;
656	struct intrs_and_queues iaq;
657	struct sge *s;
658	uint8_t *buf;
659#ifdef TCP_OFFLOAD
660	int ofld_rqidx, ofld_tqidx;
661#endif
662#ifdef DEV_NETMAP
663	int nm_rqidx, nm_tqidx;
664#endif
665	int num_vis;
666
667	sc = device_get_softc(dev);
668	sc->dev = dev;
669	TUNABLE_INT_FETCH("hw.cxgbe.debug_flags", &sc->debug_flags);
670
671	if ((pci_get_device(dev) & 0xff00) == 0x5400)
672		t5_attribute_workaround(dev);
673	pci_enable_busmaster(dev);
674	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
675		uint32_t v;
676
677		pci_set_max_read_req(dev, 4096);
678		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
679		v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
680		pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
681
682		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
683	}
684
685	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
686	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
687	sc->traceq = -1;
688	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
689	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
690	    device_get_nameunit(dev));
691
692	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
693	    device_get_nameunit(dev));
694	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
695	t4_add_adapter(sc);
696
697	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
698	TAILQ_INIT(&sc->sfl);
699	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
700
701	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
702
703	rc = t4_map_bars_0_and_4(sc);
704	if (rc != 0)
705		goto done; /* error message displayed already */
706
707	/*
708	 * This is the real PF# to which we're attaching.  Works from within PCI
709	 * passthrough environments too, where pci_get_function() could return a
710	 * different PF# depending on the passthrough configuration.  We need to
711	 * use the real PF# in all our communication with the firmware.
712	 */
713	sc->pf = G_SOURCEPF(t4_read_reg(sc, A_PL_WHOAMI));
714	sc->mbox = sc->pf;
715
716	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
717
718	/* Prepare the adapter for operation. */
719	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
720	rc = -t4_prep_adapter(sc, buf);
721	free(buf, M_CXGBE);
722	if (rc != 0) {
723		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
724		goto done;
725	}
726
727	/*
728	 * Do this really early, with the memory windows set up even before the
729	 * character device.  The userland tool's register i/o and mem read
730	 * will work even in "recovery mode".
731	 */
732	setup_memwin(sc);
733	if (t4_init_devlog_params(sc, 0) == 0)
734		fixup_devlog_params(sc);
735	make_dev_args_init(&mda);
736	mda.mda_devsw = &t4_cdevsw;
737	mda.mda_uid = UID_ROOT;
738	mda.mda_gid = GID_WHEEL;
739	mda.mda_mode = 0600;
740	mda.mda_si_drv1 = sc;
741	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
742	if (rc != 0)
743		device_printf(dev, "failed to create nexus char device: %d.\n",
744		    rc);
745
746	/* Go no further if recovery mode has been requested. */
747	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
748		device_printf(dev, "recovery mode.\n");
749		goto done;
750	}
751
752#if defined(__i386__)
753	if ((cpu_feature & CPUID_CX8) == 0) {
754		device_printf(dev, "64 bit atomics not available.\n");
755		rc = ENOTSUP;
756		goto done;
757	}
758#endif
759
760	/* Prepare the firmware for operation */
761	rc = prep_firmware(sc);
762	if (rc != 0)
763		goto done; /* error message displayed already */
764
765	rc = get_params__post_init(sc);
766	if (rc != 0)
767		goto done; /* error message displayed already */
768
769	rc = set_params__post_init(sc);
770	if (rc != 0)
771		goto done; /* error message displayed already */
772
773	rc = t4_map_bar_2(sc);
774	if (rc != 0)
775		goto done; /* error message displayed already */
776
777	rc = t4_create_dma_tag(sc);
778	if (rc != 0)
779		goto done; /* error message displayed already */
780
781	/*
782	 * Number of VIs to create per-port.  The first VI is the "main" regular
783	 * VI for the port.  The rest are additional virtual interfaces on the
784	 * same physical port.  Note that the main VI does not have native
785	 * netmap support but the extra VIs do.
786	 *
787	 * Limit the number of VIs per port to the number of available
788	 * MAC addresses per port.
789	 */
790	if (t4_num_vis >= 1)
791		num_vis = t4_num_vis;
792	else
793		num_vis = 1;
794	if (num_vis > nitems(vi_mac_funcs)) {
795		num_vis = nitems(vi_mac_funcs);
796		device_printf(dev, "Number of VIs limited to %d\n", num_vis);
797	}
798
799	/*
800	 * First pass over all the ports - allocate VIs and initialize some
801	 * basic parameters like mac address, port type, etc.  We also figure
802	 * out whether a port is 10G or 1G and use that information when
803	 * calculating how many interrupts to attempt to allocate.
804	 */
805	n10g = n1g = 0;
806	for_each_port(sc, i) {
807		struct port_info *pi;
808
809		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
810		sc->port[i] = pi;
811
812		/* These must be set before t4_port_init */
813		pi->adapter = sc;
814		pi->port_id = i;
815		/*
816		 * XXX: vi[0] is special so we can't delay this allocation until
817		 * pi->nvi's final value is known.
818		 */
819		pi->vi = malloc(sizeof(struct vi_info) * num_vis, M_CXGBE,
820		    M_ZERO | M_WAITOK);
821
822		/*
823		 * Allocate the "main" VI and initialize parameters
824		 * like mac addr.
825		 */
826		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
827		if (rc != 0) {
828			device_printf(dev, "unable to initialize port %d: %d\n",
829			    i, rc);
830			free(pi->vi, M_CXGBE);
831			free(pi, M_CXGBE);
832			sc->port[i] = NULL;
833			goto done;
834		}
835
836		pi->link_cfg.requested_fc &= ~(PAUSE_TX | PAUSE_RX);
837		pi->link_cfg.requested_fc |= t4_pause_settings;
838		pi->link_cfg.fc &= ~(PAUSE_TX | PAUSE_RX);
839		pi->link_cfg.fc |= t4_pause_settings;
840
841		rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, &pi->link_cfg);
842		if (rc != 0) {
843			device_printf(dev, "port %d l1cfg failed: %d\n", i, rc);
844			free(pi->vi, M_CXGBE);
845			free(pi, M_CXGBE);
846			sc->port[i] = NULL;
847			goto done;
848		}
849
850		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
851		    device_get_nameunit(dev), i);
852		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
853		sc->chan_map[pi->tx_chan] = i;
854
855		pi->tc = malloc(sizeof(struct tx_sched_class) *
856		    sc->chip_params->nsched_cls, M_CXGBE, M_ZERO | M_WAITOK);
857
858		if (is_10G_port(pi) || is_40G_port(pi)) {
859			n10g++;
860		} else {
861			n1g++;
862		}
863
864		pi->linkdnrc = -1;
865
866		pi->dev = device_add_child(dev, is_t4(sc) ? "cxgbe" : "cxl", -1);
867		if (pi->dev == NULL) {
868			device_printf(dev,
869			    "failed to add device for port %d.\n", i);
870			rc = ENXIO;
871			goto done;
872		}
873		pi->vi[0].dev = pi->dev;
874		device_set_softc(pi->dev, pi);
875	}
876
877	/*
878	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
879	 */
880	rc = cfg_itype_and_nqueues(sc, n10g, n1g, num_vis, &iaq);
881	if (rc != 0)
882		goto done; /* error message displayed already */
883	if (iaq.nrxq_vi + iaq.nofldrxq_vi + iaq.nnmrxq_vi == 0)
884		num_vis = 1;
885
886	sc->intr_type = iaq.intr_type;
887	sc->intr_count = iaq.nirq;
888
889	s = &sc->sge;
890	s->nrxq = n10g * iaq.nrxq10g + n1g * iaq.nrxq1g;
891	s->ntxq = n10g * iaq.ntxq10g + n1g * iaq.ntxq1g;
892	if (num_vis > 1) {
893		s->nrxq += (n10g + n1g) * (num_vis - 1) * iaq.nrxq_vi;
894		s->ntxq += (n10g + n1g) * (num_vis - 1) * iaq.ntxq_vi;
895	}
896	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
897	s->neq += sc->params.nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
898	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
899#ifdef TCP_OFFLOAD
900	if (is_offload(sc)) {
901		s->nofldrxq = n10g * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
902		s->nofldtxq = n10g * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
903		if (num_vis > 1) {
904			s->nofldrxq += (n10g + n1g) * (num_vis - 1) *
905			    iaq.nofldrxq_vi;
906			s->nofldtxq += (n10g + n1g) * (num_vis - 1) *
907			    iaq.nofldtxq_vi;
908		}
909		s->neq += s->nofldtxq + s->nofldrxq;
910		s->niq += s->nofldrxq;
911
912		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
913		    M_CXGBE, M_ZERO | M_WAITOK);
914		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
915		    M_CXGBE, M_ZERO | M_WAITOK);
916	}
917#endif
918#ifdef DEV_NETMAP
919	if (num_vis > 1) {
920		s->nnmrxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmrxq_vi;
921		s->nnmtxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmtxq_vi;
922	}
923	s->neq += s->nnmtxq + s->nnmrxq;
924	s->niq += s->nnmrxq;
925
926	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
927	    M_CXGBE, M_ZERO | M_WAITOK);
928	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
929	    M_CXGBE, M_ZERO | M_WAITOK);
930#endif
931
932	s->ctrlq = malloc(sc->params.nports * sizeof(struct sge_wrq), M_CXGBE,
933	    M_ZERO | M_WAITOK);
934	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
935	    M_ZERO | M_WAITOK);
936	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
937	    M_ZERO | M_WAITOK);
938	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
939	    M_ZERO | M_WAITOK);
940	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
941	    M_ZERO | M_WAITOK);
942
943	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
944	    M_ZERO | M_WAITOK);
945
946	t4_init_l2t(sc, M_WAITOK);
947
948	/*
949	 * Second pass over the ports.  This time we know the number of rx and
950	 * tx queues that each port should get.
951	 */
952	rqidx = tqidx = 0;
953#ifdef TCP_OFFLOAD
954	ofld_rqidx = ofld_tqidx = 0;
955#endif
956#ifdef DEV_NETMAP
957	nm_rqidx = nm_tqidx = 0;
958#endif
959	for_each_port(sc, i) {
960		struct port_info *pi = sc->port[i];
961		struct vi_info *vi;
962
963		if (pi == NULL)
964			continue;
965
966		pi->nvi = num_vis;
967		for_each_vi(pi, j, vi) {
968			vi->pi = pi;
969			vi->qsize_rxq = t4_qsize_rxq;
970			vi->qsize_txq = t4_qsize_txq;
971
972			vi->first_rxq = rqidx;
973			vi->first_txq = tqidx;
974			if (is_10G_port(pi) || is_40G_port(pi)) {
975				vi->tmr_idx = t4_tmr_idx_10g;
976				vi->pktc_idx = t4_pktc_idx_10g;
977				vi->flags |= iaq.intr_flags_10g & INTR_RXQ;
978				vi->nrxq = j == 0 ? iaq.nrxq10g : iaq.nrxq_vi;
979				vi->ntxq = j == 0 ? iaq.ntxq10g : iaq.ntxq_vi;
980			} else {
981				vi->tmr_idx = t4_tmr_idx_1g;
982				vi->pktc_idx = t4_pktc_idx_1g;
983				vi->flags |= iaq.intr_flags_1g & INTR_RXQ;
984				vi->nrxq = j == 0 ? iaq.nrxq1g : iaq.nrxq_vi;
985				vi->ntxq = j == 0 ? iaq.ntxq1g : iaq.ntxq_vi;
986			}
987			rqidx += vi->nrxq;
988			tqidx += vi->ntxq;
989
990			if (j == 0 && vi->ntxq > 1)
991				vi->rsrv_noflowq = iaq.rsrv_noflowq ? 1 : 0;
992			else
993				vi->rsrv_noflowq = 0;
994
995#ifdef TCP_OFFLOAD
996			vi->first_ofld_rxq = ofld_rqidx;
997			vi->first_ofld_txq = ofld_tqidx;
998			if (is_10G_port(pi) || is_40G_port(pi)) {
999				vi->flags |= iaq.intr_flags_10g & INTR_OFLD_RXQ;
1000				vi->nofldrxq = j == 0 ? iaq.nofldrxq10g :
1001				    iaq.nofldrxq_vi;
1002				vi->nofldtxq = j == 0 ? iaq.nofldtxq10g :
1003				    iaq.nofldtxq_vi;
1004			} else {
1005				vi->flags |= iaq.intr_flags_1g & INTR_OFLD_RXQ;
1006				vi->nofldrxq = j == 0 ? iaq.nofldrxq1g :
1007				    iaq.nofldrxq_vi;
1008				vi->nofldtxq = j == 0 ? iaq.nofldtxq1g :
1009				    iaq.nofldtxq_vi;
1010			}
1011			ofld_rqidx += vi->nofldrxq;
1012			ofld_tqidx += vi->nofldtxq;
1013#endif
1014#ifdef DEV_NETMAP
1015			if (j > 0) {
1016				vi->first_nm_rxq = nm_rqidx;
1017				vi->first_nm_txq = nm_tqidx;
1018				vi->nnmrxq = iaq.nnmrxq_vi;
1019				vi->nnmtxq = iaq.nnmtxq_vi;
1020				nm_rqidx += vi->nnmrxq;
1021				nm_tqidx += vi->nnmtxq;
1022			}
1023#endif
1024		}
1025	}
1026
1027	rc = t4_setup_intr_handlers(sc);
1028	if (rc != 0) {
1029		device_printf(dev,
1030		    "failed to setup interrupt handlers: %d\n", rc);
1031		goto done;
1032	}
1033
1034	rc = bus_generic_attach(dev);
1035	if (rc != 0) {
1036		device_printf(dev,
1037		    "failed to attach all child ports: %d\n", rc);
1038		goto done;
1039	}
1040
1041	device_printf(dev,
1042	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1043	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1044	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1045	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1046	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1047
1048	t4_set_desc(sc);
1049
1050done:
1051	if (rc != 0 && sc->cdev) {
1052		/* cdev was created and so cxgbetool works; recover that way. */
1053		device_printf(dev,
1054		    "error during attach, adapter is now in recovery mode.\n");
1055		rc = 0;
1056	}
1057
1058	if (rc != 0)
1059		t4_detach_common(dev);
1060	else
1061		t4_sysctls(sc);
1062
1063	return (rc);
1064}
1065
1066/*
1067 * Idempotent
1068 */
1069static int
1070t4_detach(device_t dev)
1071{
1072	struct adapter *sc;
1073
1074	sc = device_get_softc(dev);
1075
1076	return (t4_detach_common(dev));
1077}
1078
1079int
1080t4_detach_common(device_t dev)
1081{
1082	struct adapter *sc;
1083	struct port_info *pi;
1084	int i, rc;
1085
1086	sc = device_get_softc(dev);
1087
1088	if (sc->flags & FULL_INIT_DONE) {
1089		if (!(sc->flags & IS_VF))
1090			t4_intr_disable(sc);
1091	}
1092
1093	if (sc->cdev) {
1094		destroy_dev(sc->cdev);
1095		sc->cdev = NULL;
1096	}
1097
1098	if (device_is_attached(dev)) {
1099		rc = bus_generic_detach(dev);
1100		if (rc) {
1101			device_printf(dev,
1102			    "failed to detach child devices: %d\n", rc);
1103			return (rc);
1104		}
1105	}
1106
1107	for (i = 0; i < sc->intr_count; i++)
1108		t4_free_irq(sc, &sc->irq[i]);
1109
1110	for (i = 0; i < MAX_NPORTS; i++) {
1111		pi = sc->port[i];
1112		if (pi) {
1113			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1114			if (pi->dev)
1115				device_delete_child(dev, pi->dev);
1116
1117			mtx_destroy(&pi->pi_lock);
1118			free(pi->vi, M_CXGBE);
1119			free(pi->tc, M_CXGBE);
1120			free(pi, M_CXGBE);
1121		}
1122	}
1123
1124	if (sc->flags & FULL_INIT_DONE)
1125		adapter_full_uninit(sc);
1126
1127	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1128		t4_fw_bye(sc, sc->mbox);
1129
1130	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1131		pci_release_msi(dev);
1132
1133	if (sc->regs_res)
1134		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1135		    sc->regs_res);
1136
1137	if (sc->udbs_res)
1138		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1139		    sc->udbs_res);
1140
1141	if (sc->msix_res)
1142		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1143		    sc->msix_res);
1144
1145	if (sc->l2t)
1146		t4_free_l2t(sc->l2t);
1147
1148#ifdef TCP_OFFLOAD
1149	free(sc->sge.ofld_rxq, M_CXGBE);
1150	free(sc->sge.ofld_txq, M_CXGBE);
1151#endif
1152#ifdef DEV_NETMAP
1153	free(sc->sge.nm_rxq, M_CXGBE);
1154	free(sc->sge.nm_txq, M_CXGBE);
1155#endif
1156	free(sc->irq, M_CXGBE);
1157	free(sc->sge.rxq, M_CXGBE);
1158	free(sc->sge.txq, M_CXGBE);
1159	free(sc->sge.ctrlq, M_CXGBE);
1160	free(sc->sge.iqmap, M_CXGBE);
1161	free(sc->sge.eqmap, M_CXGBE);
1162	free(sc->tids.ftid_tab, M_CXGBE);
1163	t4_destroy_dma_tag(sc);
1164	if (mtx_initialized(&sc->sc_lock)) {
1165		sx_xlock(&t4_list_lock);
1166		SLIST_REMOVE(&t4_list, sc, adapter, link);
1167		sx_xunlock(&t4_list_lock);
1168		mtx_destroy(&sc->sc_lock);
1169	}
1170
1171	callout_drain(&sc->sfl_callout);
1172	if (mtx_initialized(&sc->tids.ftid_lock))
1173		mtx_destroy(&sc->tids.ftid_lock);
1174	if (mtx_initialized(&sc->sfl_lock))
1175		mtx_destroy(&sc->sfl_lock);
1176	if (mtx_initialized(&sc->ifp_lock))
1177		mtx_destroy(&sc->ifp_lock);
1178	if (mtx_initialized(&sc->reg_lock))
1179		mtx_destroy(&sc->reg_lock);
1180
1181	for (i = 0; i < NUM_MEMWIN; i++) {
1182		struct memwin *mw = &sc->memwin[i];
1183
1184		if (rw_initialized(&mw->mw_lock))
1185			rw_destroy(&mw->mw_lock);
1186	}
1187
1188	bzero(sc, sizeof(*sc));
1189
1190	return (0);
1191}
1192
1193static int
1194cxgbe_probe(device_t dev)
1195{
1196	char buf[128];
1197	struct port_info *pi = device_get_softc(dev);
1198
1199	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1200	device_set_desc_copy(dev, buf);
1201
1202	return (BUS_PROBE_DEFAULT);
1203}
1204
1205#define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1206    IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1207    IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1208#define T4_CAP_ENABLE (T4_CAP)
1209
1210static int
1211cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1212{
1213	struct ifnet *ifp;
1214	struct sbuf *sb;
1215
1216	vi->xact_addr_filt = -1;
1217	callout_init(&vi->tick, 1);
1218
1219	/* Allocate an ifnet and set it up */
1220	ifp = if_alloc(IFT_ETHER);
1221	if (ifp == NULL) {
1222		device_printf(dev, "Cannot allocate ifnet\n");
1223		return (ENOMEM);
1224	}
1225	vi->ifp = ifp;
1226	ifp->if_softc = vi;
1227
1228	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1229	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1230
1231	ifp->if_init = cxgbe_init;
1232	ifp->if_ioctl = cxgbe_ioctl;
1233	ifp->if_transmit = cxgbe_transmit;
1234	ifp->if_qflush = cxgbe_qflush;
1235
1236	ifp->if_capabilities = T4_CAP;
1237#ifdef TCP_OFFLOAD
1238	if (vi->nofldrxq != 0)
1239		ifp->if_capabilities |= IFCAP_TOE;
1240#endif
1241	ifp->if_capenable = T4_CAP_ENABLE;
1242	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1243	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1244
1245	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1246	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1247	ifp->if_hw_tsomaxsegsize = 65536;
1248
1249	/* Initialize ifmedia for this VI */
1250	ifmedia_init(&vi->media, IFM_IMASK, cxgbe_media_change,
1251	    cxgbe_media_status);
1252	build_medialist(vi->pi, &vi->media);
1253
1254	vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1255	    EVENTHANDLER_PRI_ANY);
1256
1257	ether_ifattach(ifp, vi->hw_addr);
1258#ifdef DEV_NETMAP
1259	if (vi->nnmrxq != 0)
1260		cxgbe_nm_attach(vi);
1261#endif
1262	sb = sbuf_new_auto();
1263	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1264#ifdef TCP_OFFLOAD
1265	if (ifp->if_capabilities & IFCAP_TOE)
1266		sbuf_printf(sb, "; %d txq, %d rxq (TOE)",
1267		    vi->nofldtxq, vi->nofldrxq);
1268#endif
1269#ifdef DEV_NETMAP
1270	if (ifp->if_capabilities & IFCAP_NETMAP)
1271		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1272		    vi->nnmtxq, vi->nnmrxq);
1273#endif
1274	sbuf_finish(sb);
1275	device_printf(dev, "%s\n", sbuf_data(sb));
1276	sbuf_delete(sb);
1277
1278	vi_sysctls(vi);
1279
1280	return (0);
1281}
1282
1283static int
1284cxgbe_attach(device_t dev)
1285{
1286	struct port_info *pi = device_get_softc(dev);
1287	struct vi_info *vi;
1288	int i, rc;
1289
1290	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1291
1292	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1293	if (rc)
1294		return (rc);
1295
1296	for_each_vi(pi, i, vi) {
1297		if (i == 0)
1298			continue;
1299		vi->dev = device_add_child(dev, is_t4(pi->adapter) ?
1300		    "vcxgbe" : "vcxl", -1);
1301		if (vi->dev == NULL) {
1302			device_printf(dev, "failed to add VI %d\n", i);
1303			continue;
1304		}
1305		device_set_softc(vi->dev, vi);
1306	}
1307
1308	cxgbe_sysctls(pi);
1309
1310	bus_generic_attach(dev);
1311
1312	return (0);
1313}
1314
1315static void
1316cxgbe_vi_detach(struct vi_info *vi)
1317{
1318	struct ifnet *ifp = vi->ifp;
1319
1320	ether_ifdetach(ifp);
1321
1322	if (vi->vlan_c)
1323		EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1324
1325	/* Let detach proceed even if these fail. */
1326#ifdef DEV_NETMAP
1327	if (ifp->if_capabilities & IFCAP_NETMAP)
1328		cxgbe_nm_detach(vi);
1329#endif
1330	cxgbe_uninit_synchronized(vi);
1331	callout_drain(&vi->tick);
1332	vi_full_uninit(vi);
1333
1334	ifmedia_removeall(&vi->media);
1335	if_free(vi->ifp);
1336	vi->ifp = NULL;
1337}
1338
1339static int
1340cxgbe_detach(device_t dev)
1341{
1342	struct port_info *pi = device_get_softc(dev);
1343	struct adapter *sc = pi->adapter;
1344	int rc;
1345
1346	/* Detach the extra VIs first. */
1347	rc = bus_generic_detach(dev);
1348	if (rc)
1349		return (rc);
1350	device_delete_children(dev);
1351
1352	doom_vi(sc, &pi->vi[0]);
1353
1354	if (pi->flags & HAS_TRACEQ) {
1355		sc->traceq = -1;	/* cloner should not create ifnet */
1356		t4_tracer_port_detach(sc);
1357	}
1358
1359	cxgbe_vi_detach(&pi->vi[0]);
1360	callout_drain(&pi->tick);
1361
1362	end_synchronized_op(sc, 0);
1363
1364	return (0);
1365}
1366
1367static void
1368cxgbe_init(void *arg)
1369{
1370	struct vi_info *vi = arg;
1371	struct adapter *sc = vi->pi->adapter;
1372
1373	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1374		return;
1375	cxgbe_init_synchronized(vi);
1376	end_synchronized_op(sc, 0);
1377}
1378
1379static int
1380cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1381{
1382	int rc = 0, mtu, flags, can_sleep;
1383	struct vi_info *vi = ifp->if_softc;
1384	struct adapter *sc = vi->pi->adapter;
1385	struct ifreq *ifr = (struct ifreq *)data;
1386	uint32_t mask;
1387
1388	switch (cmd) {
1389	case SIOCSIFMTU:
1390		mtu = ifr->ifr_mtu;
1391		if ((mtu < ETHERMIN) || (mtu > ETHERMTU_JUMBO))
1392			return (EINVAL);
1393
1394		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1395		if (rc)
1396			return (rc);
1397		ifp->if_mtu = mtu;
1398		if (vi->flags & VI_INIT_DONE) {
1399			t4_update_fl_bufsize(ifp);
1400			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1401				rc = update_mac_settings(ifp, XGMAC_MTU);
1402		}
1403		end_synchronized_op(sc, 0);
1404		break;
1405
1406	case SIOCSIFFLAGS:
1407		can_sleep = 0;
1408redo_sifflags:
1409		rc = begin_synchronized_op(sc, vi,
1410		    can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1411		if (rc)
1412			return (rc);
1413
1414		if (ifp->if_flags & IFF_UP) {
1415			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1416				flags = vi->if_flags;
1417				if ((ifp->if_flags ^ flags) &
1418				    (IFF_PROMISC | IFF_ALLMULTI)) {
1419					if (can_sleep == 1) {
1420						end_synchronized_op(sc, 0);
1421						can_sleep = 0;
1422						goto redo_sifflags;
1423					}
1424					rc = update_mac_settings(ifp,
1425					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1426				}
1427			} else {
1428				if (can_sleep == 0) {
1429					end_synchronized_op(sc, LOCK_HELD);
1430					can_sleep = 1;
1431					goto redo_sifflags;
1432				}
1433				rc = cxgbe_init_synchronized(vi);
1434			}
1435			vi->if_flags = ifp->if_flags;
1436		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1437			if (can_sleep == 0) {
1438				end_synchronized_op(sc, LOCK_HELD);
1439				can_sleep = 1;
1440				goto redo_sifflags;
1441			}
1442			rc = cxgbe_uninit_synchronized(vi);
1443		}
1444		end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1445		break;
1446
1447	case SIOCADDMULTI:
1448	case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1449		rc = begin_synchronized_op(sc, vi, HOLD_LOCK, "t4multi");
1450		if (rc)
1451			return (rc);
1452		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1453			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1454		end_synchronized_op(sc, LOCK_HELD);
1455		break;
1456
1457	case SIOCSIFCAP:
1458		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1459		if (rc)
1460			return (rc);
1461
1462		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1463		if (mask & IFCAP_TXCSUM) {
1464			ifp->if_capenable ^= IFCAP_TXCSUM;
1465			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1466
1467			if (IFCAP_TSO4 & ifp->if_capenable &&
1468			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1469				ifp->if_capenable &= ~IFCAP_TSO4;
1470				if_printf(ifp,
1471				    "tso4 disabled due to -txcsum.\n");
1472			}
1473		}
1474		if (mask & IFCAP_TXCSUM_IPV6) {
1475			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1476			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1477
1478			if (IFCAP_TSO6 & ifp->if_capenable &&
1479			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1480				ifp->if_capenable &= ~IFCAP_TSO6;
1481				if_printf(ifp,
1482				    "tso6 disabled due to -txcsum6.\n");
1483			}
1484		}
1485		if (mask & IFCAP_RXCSUM)
1486			ifp->if_capenable ^= IFCAP_RXCSUM;
1487		if (mask & IFCAP_RXCSUM_IPV6)
1488			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1489
1490		/*
1491		 * Note that we leave CSUM_TSO alone (it is always set).  The
1492		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1493		 * sending a TSO request our way, so it's sufficient to toggle
1494		 * IFCAP_TSOx only.
1495		 */
1496		if (mask & IFCAP_TSO4) {
1497			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1498			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1499				if_printf(ifp, "enable txcsum first.\n");
1500				rc = EAGAIN;
1501				goto fail;
1502			}
1503			ifp->if_capenable ^= IFCAP_TSO4;
1504		}
1505		if (mask & IFCAP_TSO6) {
1506			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1507			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1508				if_printf(ifp, "enable txcsum6 first.\n");
1509				rc = EAGAIN;
1510				goto fail;
1511			}
1512			ifp->if_capenable ^= IFCAP_TSO6;
1513		}
1514		if (mask & IFCAP_LRO) {
1515#if defined(INET) || defined(INET6)
1516			int i;
1517			struct sge_rxq *rxq;
1518
1519			ifp->if_capenable ^= IFCAP_LRO;
1520			for_each_rxq(vi, i, rxq) {
1521				if (ifp->if_capenable & IFCAP_LRO)
1522					rxq->iq.flags |= IQ_LRO_ENABLED;
1523				else
1524					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1525			}
1526#endif
1527		}
1528#ifdef TCP_OFFLOAD
1529		if (mask & IFCAP_TOE) {
1530			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1531
1532			rc = toe_capability(vi, enable);
1533			if (rc != 0)
1534				goto fail;
1535
1536			ifp->if_capenable ^= mask;
1537		}
1538#endif
1539		if (mask & IFCAP_VLAN_HWTAGGING) {
1540			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1541			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1542				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1543		}
1544		if (mask & IFCAP_VLAN_MTU) {
1545			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1546
1547			/* Need to find out how to disable auto-mtu-inflation */
1548		}
1549		if (mask & IFCAP_VLAN_HWTSO)
1550			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1551		if (mask & IFCAP_VLAN_HWCSUM)
1552			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1553
1554#ifdef VLAN_CAPABILITIES
1555		VLAN_CAPABILITIES(ifp);
1556#endif
1557fail:
1558		end_synchronized_op(sc, 0);
1559		break;
1560
1561	case SIOCSIFMEDIA:
1562	case SIOCGIFMEDIA:
1563		ifmedia_ioctl(ifp, ifr, &vi->media, cmd);
1564		break;
1565
1566	case SIOCGI2C: {
1567		struct ifi2creq i2c;
1568
1569		rc = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
1570		if (rc != 0)
1571			break;
1572		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1573			rc = EPERM;
1574			break;
1575		}
1576		if (i2c.len > sizeof(i2c.data)) {
1577			rc = EINVAL;
1578			break;
1579		}
1580		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1581		if (rc)
1582			return (rc);
1583		rc = -t4_i2c_rd(sc, sc->mbox, vi->pi->port_id, i2c.dev_addr,
1584		    i2c.offset, i2c.len, &i2c.data[0]);
1585		end_synchronized_op(sc, 0);
1586		if (rc == 0)
1587			rc = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
1588		break;
1589	}
1590
1591	default:
1592		rc = ether_ioctl(ifp, cmd, data);
1593	}
1594
1595	return (rc);
1596}
1597
1598static int
1599cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1600{
1601	struct vi_info *vi = ifp->if_softc;
1602	struct port_info *pi = vi->pi;
1603	struct adapter *sc = pi->adapter;
1604	struct sge_txq *txq;
1605	void *items[1];
1606	int rc;
1607
1608	M_ASSERTPKTHDR(m);
1609	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1610
1611	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1612		m_freem(m);
1613		return (ENETDOWN);
1614	}
1615
1616	rc = parse_pkt(sc, &m);
1617	if (__predict_false(rc != 0)) {
1618		MPASS(m == NULL);			/* was freed already */
1619		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1620		return (rc);
1621	}
1622
1623	/* Select a txq. */
1624	txq = &sc->sge.txq[vi->first_txq];
1625	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1626		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1627		    vi->rsrv_noflowq);
1628
1629	items[0] = m;
1630	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1631	if (__predict_false(rc != 0))
1632		m_freem(m);
1633
1634	return (rc);
1635}
1636
1637static void
1638cxgbe_qflush(struct ifnet *ifp)
1639{
1640	struct vi_info *vi = ifp->if_softc;
1641	struct sge_txq *txq;
1642	int i;
1643
1644	/* queues do not exist if !VI_INIT_DONE. */
1645	if (vi->flags & VI_INIT_DONE) {
1646		for_each_txq(vi, i, txq) {
1647			TXQ_LOCK(txq);
1648			txq->eq.flags &= ~EQ_ENABLED;
1649			TXQ_UNLOCK(txq);
1650			while (!mp_ring_is_idle(txq->r)) {
1651				mp_ring_check_drainage(txq->r, 0);
1652				pause("qflush", 1);
1653			}
1654		}
1655	}
1656	if_qflush(ifp);
1657}
1658
1659static int
1660cxgbe_media_change(struct ifnet *ifp)
1661{
1662	struct vi_info *vi = ifp->if_softc;
1663
1664	device_printf(vi->dev, "%s unimplemented.\n", __func__);
1665
1666	return (EOPNOTSUPP);
1667}
1668
1669static void
1670cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1671{
1672	struct vi_info *vi = ifp->if_softc;
1673	struct port_info *pi = vi->pi;
1674	struct ifmedia_entry *cur;
1675	int speed = pi->link_cfg.speed;
1676
1677	cur = vi->media.ifm_cur;
1678
1679	ifmr->ifm_status = IFM_AVALID;
1680	if (!pi->link_cfg.link_ok)
1681		return;
1682
1683	ifmr->ifm_status |= IFM_ACTIVE;
1684
1685	/* active and current will differ iff current media is autoselect. */
1686	if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
1687		return;
1688
1689	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
1690	if (speed == 10000)
1691		ifmr->ifm_active |= IFM_10G_T;
1692	else if (speed == 1000)
1693		ifmr->ifm_active |= IFM_1000_T;
1694	else if (speed == 100)
1695		ifmr->ifm_active |= IFM_100_TX;
1696	else if (speed == 10)
1697		ifmr->ifm_active |= IFM_10_T;
1698	else
1699		KASSERT(0, ("%s: link up but speed unknown (%u)", __func__,
1700			    speed));
1701}
1702
1703static int
1704vcxgbe_probe(device_t dev)
1705{
1706	char buf[128];
1707	struct vi_info *vi = device_get_softc(dev);
1708
1709	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
1710	    vi - vi->pi->vi);
1711	device_set_desc_copy(dev, buf);
1712
1713	return (BUS_PROBE_DEFAULT);
1714}
1715
1716static int
1717vcxgbe_attach(device_t dev)
1718{
1719	struct vi_info *vi;
1720	struct port_info *pi;
1721	struct adapter *sc;
1722	int func, index, rc;
1723	u32 param, val;
1724
1725	vi = device_get_softc(dev);
1726	pi = vi->pi;
1727	sc = pi->adapter;
1728
1729	index = vi - pi->vi;
1730	KASSERT(index < nitems(vi_mac_funcs),
1731	    ("%s: VI %s doesn't have a MAC func", __func__,
1732	    device_get_nameunit(dev)));
1733	func = vi_mac_funcs[index];
1734	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
1735	    vi->hw_addr, &vi->rss_size, func, 0);
1736	if (rc < 0) {
1737		device_printf(dev, "Failed to allocate virtual interface "
1738		    "for port %d: %d\n", pi->port_id, -rc);
1739		return (-rc);
1740	}
1741	vi->viid = rc;
1742
1743	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
1744	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
1745	    V_FW_PARAMS_PARAM_YZ(vi->viid);
1746	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
1747	if (rc)
1748		vi->rss_base = 0xffff;
1749	else {
1750		/* MPASS((val >> 16) == rss_size); */
1751		vi->rss_base = val & 0xffff;
1752	}
1753
1754	rc = cxgbe_vi_attach(dev, vi);
1755	if (rc) {
1756		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
1757		return (rc);
1758	}
1759	return (0);
1760}
1761
1762static int
1763vcxgbe_detach(device_t dev)
1764{
1765	struct vi_info *vi;
1766	struct adapter *sc;
1767
1768	vi = device_get_softc(dev);
1769	sc = vi->pi->adapter;
1770
1771	doom_vi(sc, vi);
1772
1773	cxgbe_vi_detach(vi);
1774	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
1775
1776	end_synchronized_op(sc, 0);
1777
1778	return (0);
1779}
1780
1781void
1782t4_fatal_err(struct adapter *sc)
1783{
1784	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
1785	t4_intr_disable(sc);
1786	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
1787	    device_get_nameunit(sc->dev));
1788}
1789
1790void
1791t4_add_adapter(struct adapter *sc)
1792{
1793	sx_xlock(&t4_list_lock);
1794	SLIST_INSERT_HEAD(&t4_list, sc, link);
1795	sx_xunlock(&t4_list_lock);
1796}
1797
1798int
1799t4_map_bars_0_and_4(struct adapter *sc)
1800{
1801	sc->regs_rid = PCIR_BAR(0);
1802	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1803	    &sc->regs_rid, RF_ACTIVE);
1804	if (sc->regs_res == NULL) {
1805		device_printf(sc->dev, "cannot map registers.\n");
1806		return (ENXIO);
1807	}
1808	sc->bt = rman_get_bustag(sc->regs_res);
1809	sc->bh = rman_get_bushandle(sc->regs_res);
1810	sc->mmio_len = rman_get_size(sc->regs_res);
1811	setbit(&sc->doorbells, DOORBELL_KDB);
1812
1813	sc->msix_rid = PCIR_BAR(4);
1814	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1815	    &sc->msix_rid, RF_ACTIVE);
1816	if (sc->msix_res == NULL) {
1817		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
1818		return (ENXIO);
1819	}
1820
1821	return (0);
1822}
1823
1824int
1825t4_map_bar_2(struct adapter *sc)
1826{
1827
1828	/*
1829	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
1830	 * to map it if RDMA is disabled.
1831	 */
1832	if (is_t4(sc) && sc->rdmacaps == 0)
1833		return (0);
1834
1835	sc->udbs_rid = PCIR_BAR(2);
1836	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1837	    &sc->udbs_rid, RF_ACTIVE);
1838	if (sc->udbs_res == NULL) {
1839		device_printf(sc->dev, "cannot map doorbell BAR.\n");
1840		return (ENXIO);
1841	}
1842	sc->udbs_base = rman_get_virtual(sc->udbs_res);
1843
1844	if (is_t5(sc)) {
1845		setbit(&sc->doorbells, DOORBELL_UDB);
1846#if defined(__i386__) || defined(__amd64__)
1847		if (t5_write_combine) {
1848			int rc;
1849
1850			/*
1851			 * Enable write combining on BAR2.  This is the
1852			 * userspace doorbell BAR and is split into 128B
1853			 * (UDBS_SEG_SIZE) doorbell regions, each associated
1854			 * with an egress queue.  The first 64B has the doorbell
1855			 * and the second 64B can be used to submit a tx work
1856			 * request with an implicit doorbell.
1857			 */
1858
1859			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
1860			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
1861			if (rc == 0) {
1862				clrbit(&sc->doorbells, DOORBELL_UDB);
1863				setbit(&sc->doorbells, DOORBELL_WCWR);
1864				setbit(&sc->doorbells, DOORBELL_UDBWC);
1865			} else {
1866				device_printf(sc->dev,
1867				    "couldn't enable write combining: %d\n",
1868				    rc);
1869			}
1870
1871			t4_write_reg(sc, A_SGE_STAT_CFG,
1872			    V_STATSOURCE_T5(7) | V_STATMODE(0));
1873		}
1874#endif
1875	}
1876
1877	return (0);
1878}
1879
1880struct memwin_init {
1881	uint32_t base;
1882	uint32_t aperture;
1883};
1884
1885static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
1886	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1887	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1888	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
1889};
1890
1891static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
1892	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1893	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1894	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
1895};
1896
1897static void
1898setup_memwin(struct adapter *sc)
1899{
1900	const struct memwin_init *mw_init;
1901	struct memwin *mw;
1902	int i;
1903	uint32_t bar0;
1904
1905	if (is_t4(sc)) {
1906		/*
1907		 * Read low 32b of bar0 indirectly via the hardware backdoor
1908		 * mechanism.  Works from within PCI passthrough environments
1909		 * too, where rman_get_start() can return a different value.  We
1910		 * need to program the T4 memory window decoders with the actual
1911		 * addresses that will be coming across the PCIe link.
1912		 */
1913		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
1914		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
1915
1916		mw_init = &t4_memwin[0];
1917	} else {
1918		/* T5+ use the relative offset inside the PCIe BAR */
1919		bar0 = 0;
1920
1921		mw_init = &t5_memwin[0];
1922	}
1923
1924	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
1925		rw_init(&mw->mw_lock, "memory window access");
1926		mw->mw_base = mw_init->base;
1927		mw->mw_aperture = mw_init->aperture;
1928		mw->mw_curpos = 0;
1929		t4_write_reg(sc,
1930		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
1931		    (mw->mw_base + bar0) | V_BIR(0) |
1932		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
1933		rw_wlock(&mw->mw_lock);
1934		position_memwin(sc, i, 0);
1935		rw_wunlock(&mw->mw_lock);
1936	}
1937
1938	/* flush */
1939	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
1940}
1941
1942/*
1943 * Positions the memory window at the given address in the card's address space.
1944 * There are some alignment requirements and the actual position may be at an
1945 * address prior to the requested address.  mw->mw_curpos always has the actual
1946 * position of the window.
1947 */
1948static void
1949position_memwin(struct adapter *sc, int idx, uint32_t addr)
1950{
1951	struct memwin *mw;
1952	uint32_t pf;
1953	uint32_t reg;
1954
1955	MPASS(idx >= 0 && idx < NUM_MEMWIN);
1956	mw = &sc->memwin[idx];
1957	rw_assert(&mw->mw_lock, RA_WLOCKED);
1958
1959	if (is_t4(sc)) {
1960		pf = 0;
1961		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
1962	} else {
1963		pf = V_PFNUM(sc->pf);
1964		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
1965	}
1966	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
1967	t4_write_reg(sc, reg, mw->mw_curpos | pf);
1968	t4_read_reg(sc, reg);	/* flush */
1969}
1970
1971static int
1972rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
1973    int len, int rw)
1974{
1975	struct memwin *mw;
1976	uint32_t mw_end, v;
1977
1978	MPASS(idx >= 0 && idx < NUM_MEMWIN);
1979
1980	/* Memory can only be accessed in naturally aligned 4 byte units */
1981	if (addr & 3 || len & 3 || len <= 0)
1982		return (EINVAL);
1983
1984	mw = &sc->memwin[idx];
1985	while (len > 0) {
1986		rw_rlock(&mw->mw_lock);
1987		mw_end = mw->mw_curpos + mw->mw_aperture;
1988		if (addr >= mw_end || addr < mw->mw_curpos) {
1989			/* Will need to reposition the window */
1990			if (!rw_try_upgrade(&mw->mw_lock)) {
1991				rw_runlock(&mw->mw_lock);
1992				rw_wlock(&mw->mw_lock);
1993			}
1994			rw_assert(&mw->mw_lock, RA_WLOCKED);
1995			position_memwin(sc, idx, addr);
1996			rw_downgrade(&mw->mw_lock);
1997			mw_end = mw->mw_curpos + mw->mw_aperture;
1998		}
1999		rw_assert(&mw->mw_lock, RA_RLOCKED);
2000		while (addr < mw_end && len > 0) {
2001			if (rw == 0) {
2002				v = t4_read_reg(sc, mw->mw_base + addr -
2003				    mw->mw_curpos);
2004				*val++ = le32toh(v);
2005			} else {
2006				v = *val++;
2007				t4_write_reg(sc, mw->mw_base + addr -
2008				    mw->mw_curpos, htole32(v));;
2009			}
2010			addr += 4;
2011			len -= 4;
2012		}
2013		rw_runlock(&mw->mw_lock);
2014	}
2015
2016	return (0);
2017}
2018
2019static inline int
2020read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2021    int len)
2022{
2023
2024	return (rw_via_memwin(sc, idx, addr, val, len, 0));
2025}
2026
2027static inline int
2028write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
2029    const uint32_t *val, int len)
2030{
2031
2032	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
2033}
2034
2035static int
2036t4_range_cmp(const void *a, const void *b)
2037{
2038	return ((const struct t4_range *)a)->start -
2039	       ((const struct t4_range *)b)->start;
2040}
2041
2042/*
2043 * Verify that the memory range specified by the addr/len pair is valid within
2044 * the card's address space.
2045 */
2046static int
2047validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2048{
2049	struct t4_range mem_ranges[4], *r, *next;
2050	uint32_t em, addr_len;
2051	int i, n, remaining;
2052
2053	/* Memory can only be accessed in naturally aligned 4 byte units */
2054	if (addr & 3 || len & 3 || len <= 0)
2055		return (EINVAL);
2056
2057	/* Enabled memories */
2058	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2059
2060	r = &mem_ranges[0];
2061	n = 0;
2062	bzero(r, sizeof(mem_ranges));
2063	if (em & F_EDRAM0_ENABLE) {
2064		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2065		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2066		if (r->size > 0) {
2067			r->start = G_EDRAM0_BASE(addr_len) << 20;
2068			if (addr >= r->start &&
2069			    addr + len <= r->start + r->size)
2070				return (0);
2071			r++;
2072			n++;
2073		}
2074	}
2075	if (em & F_EDRAM1_ENABLE) {
2076		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2077		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2078		if (r->size > 0) {
2079			r->start = G_EDRAM1_BASE(addr_len) << 20;
2080			if (addr >= r->start &&
2081			    addr + len <= r->start + r->size)
2082				return (0);
2083			r++;
2084			n++;
2085		}
2086	}
2087	if (em & F_EXT_MEM_ENABLE) {
2088		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2089		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2090		if (r->size > 0) {
2091			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2092			if (addr >= r->start &&
2093			    addr + len <= r->start + r->size)
2094				return (0);
2095			r++;
2096			n++;
2097		}
2098	}
2099	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2100		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2101		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2102		if (r->size > 0) {
2103			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2104			if (addr >= r->start &&
2105			    addr + len <= r->start + r->size)
2106				return (0);
2107			r++;
2108			n++;
2109		}
2110	}
2111	MPASS(n <= nitems(mem_ranges));
2112
2113	if (n > 1) {
2114		/* Sort and merge the ranges. */
2115		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2116
2117		/* Start from index 0 and examine the next n - 1 entries. */
2118		r = &mem_ranges[0];
2119		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2120
2121			MPASS(r->size > 0);	/* r is a valid entry. */
2122			next = r + 1;
2123			MPASS(next->size > 0);	/* and so is the next one. */
2124
2125			while (r->start + r->size >= next->start) {
2126				/* Merge the next one into the current entry. */
2127				r->size = max(r->start + r->size,
2128				    next->start + next->size) - r->start;
2129				n--;	/* One fewer entry in total. */
2130				if (--remaining == 0)
2131					goto done;	/* short circuit */
2132				next++;
2133			}
2134			if (next != r + 1) {
2135				/*
2136				 * Some entries were merged into r and next
2137				 * points to the first valid entry that couldn't
2138				 * be merged.
2139				 */
2140				MPASS(next->size > 0);	/* must be valid */
2141				memcpy(r + 1, next, remaining * sizeof(*r));
2142#ifdef INVARIANTS
2143				/*
2144				 * This so that the foo->size assertion in the
2145				 * next iteration of the loop do the right
2146				 * thing for entries that were pulled up and are
2147				 * no longer valid.
2148				 */
2149				MPASS(n < nitems(mem_ranges));
2150				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2151				    sizeof(struct t4_range));
2152#endif
2153			}
2154		}
2155done:
2156		/* Done merging the ranges. */
2157		MPASS(n > 0);
2158		r = &mem_ranges[0];
2159		for (i = 0; i < n; i++, r++) {
2160			if (addr >= r->start &&
2161			    addr + len <= r->start + r->size)
2162				return (0);
2163		}
2164	}
2165
2166	return (EFAULT);
2167}
2168
2169static int
2170fwmtype_to_hwmtype(int mtype)
2171{
2172
2173	switch (mtype) {
2174	case FW_MEMTYPE_EDC0:
2175		return (MEM_EDC0);
2176	case FW_MEMTYPE_EDC1:
2177		return (MEM_EDC1);
2178	case FW_MEMTYPE_EXTMEM:
2179		return (MEM_MC0);
2180	case FW_MEMTYPE_EXTMEM1:
2181		return (MEM_MC1);
2182	default:
2183		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2184	}
2185}
2186
2187/*
2188 * Verify that the memory range specified by the memtype/offset/len pair is
2189 * valid and lies entirely within the memtype specified.  The global address of
2190 * the start of the range is returned in addr.
2191 */
2192static int
2193validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2194    uint32_t *addr)
2195{
2196	uint32_t em, addr_len, maddr;
2197
2198	/* Memory can only be accessed in naturally aligned 4 byte units */
2199	if (off & 3 || len & 3 || len == 0)
2200		return (EINVAL);
2201
2202	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2203	switch (fwmtype_to_hwmtype(mtype)) {
2204	case MEM_EDC0:
2205		if (!(em & F_EDRAM0_ENABLE))
2206			return (EINVAL);
2207		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2208		maddr = G_EDRAM0_BASE(addr_len) << 20;
2209		break;
2210	case MEM_EDC1:
2211		if (!(em & F_EDRAM1_ENABLE))
2212			return (EINVAL);
2213		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2214		maddr = G_EDRAM1_BASE(addr_len) << 20;
2215		break;
2216	case MEM_MC:
2217		if (!(em & F_EXT_MEM_ENABLE))
2218			return (EINVAL);
2219		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2220		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2221		break;
2222	case MEM_MC1:
2223		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2224			return (EINVAL);
2225		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2226		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2227		break;
2228	default:
2229		return (EINVAL);
2230	}
2231
2232	*addr = maddr + off;	/* global address */
2233	return (validate_mem_range(sc, *addr, len));
2234}
2235
2236static int
2237fixup_devlog_params(struct adapter *sc)
2238{
2239	struct devlog_params *dparams = &sc->params.devlog;
2240	int rc;
2241
2242	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2243	    dparams->size, &dparams->addr);
2244
2245	return (rc);
2246}
2247
2248static int
2249cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g, int num_vis,
2250    struct intrs_and_queues *iaq)
2251{
2252	int rc, itype, navail, nrxq10g, nrxq1g, n;
2253	int nofldrxq10g = 0, nofldrxq1g = 0;
2254
2255	bzero(iaq, sizeof(*iaq));
2256
2257	iaq->ntxq10g = t4_ntxq10g;
2258	iaq->ntxq1g = t4_ntxq1g;
2259	iaq->ntxq_vi = t4_ntxq_vi;
2260	iaq->nrxq10g = nrxq10g = t4_nrxq10g;
2261	iaq->nrxq1g = nrxq1g = t4_nrxq1g;
2262	iaq->nrxq_vi = t4_nrxq_vi;
2263	iaq->rsrv_noflowq = t4_rsrv_noflowq;
2264#ifdef TCP_OFFLOAD
2265	if (is_offload(sc)) {
2266		iaq->nofldtxq10g = t4_nofldtxq10g;
2267		iaq->nofldtxq1g = t4_nofldtxq1g;
2268		iaq->nofldtxq_vi = t4_nofldtxq_vi;
2269		iaq->nofldrxq10g = nofldrxq10g = t4_nofldrxq10g;
2270		iaq->nofldrxq1g = nofldrxq1g = t4_nofldrxq1g;
2271		iaq->nofldrxq_vi = t4_nofldrxq_vi;
2272	}
2273#endif
2274#ifdef DEV_NETMAP
2275	iaq->nnmtxq_vi = t4_nnmtxq_vi;
2276	iaq->nnmrxq_vi = t4_nnmrxq_vi;
2277#endif
2278
2279	for (itype = INTR_MSIX; itype; itype >>= 1) {
2280
2281		if ((itype & t4_intr_types) == 0)
2282			continue;	/* not allowed */
2283
2284		if (itype == INTR_MSIX)
2285			navail = pci_msix_count(sc->dev);
2286		else if (itype == INTR_MSI)
2287			navail = pci_msi_count(sc->dev);
2288		else
2289			navail = 1;
2290restart:
2291		if (navail == 0)
2292			continue;
2293
2294		iaq->intr_type = itype;
2295		iaq->intr_flags_10g = 0;
2296		iaq->intr_flags_1g = 0;
2297
2298		/*
2299		 * Best option: an interrupt vector for errors, one for the
2300		 * firmware event queue, and one for every rxq (NIC and TOE) of
2301		 * every VI.  The VIs that support netmap use the same
2302		 * interrupts for the NIC rx queues and the netmap rx queues
2303		 * because only one set of queues is active at a time.
2304		 */
2305		iaq->nirq = T4_EXTRA_INTR;
2306		iaq->nirq += n10g * (nrxq10g + nofldrxq10g);
2307		iaq->nirq += n1g * (nrxq1g + nofldrxq1g);
2308		iaq->nirq += (n10g + n1g) * (num_vis - 1) *
2309		    max(iaq->nrxq_vi, iaq->nnmrxq_vi);	/* See comment above. */
2310		iaq->nirq += (n10g + n1g) * (num_vis - 1) * iaq->nofldrxq_vi;
2311		if (iaq->nirq <= navail &&
2312		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2313			iaq->intr_flags_10g = INTR_ALL;
2314			iaq->intr_flags_1g = INTR_ALL;
2315			goto allocate;
2316		}
2317
2318		/* Disable the VIs (and netmap) if there aren't enough intrs */
2319		if (num_vis > 1) {
2320			device_printf(sc->dev, "virtual interfaces disabled "
2321			    "because num_vis=%u with current settings "
2322			    "(nrxq10g=%u, nrxq1g=%u, nofldrxq10g=%u, "
2323			    "nofldrxq1g=%u, nrxq_vi=%u nofldrxq_vi=%u, "
2324			    "nnmrxq_vi=%u) would need %u interrupts but "
2325			    "only %u are available.\n", num_vis, nrxq10g,
2326			    nrxq1g, nofldrxq10g, nofldrxq1g, iaq->nrxq_vi,
2327			    iaq->nofldrxq_vi, iaq->nnmrxq_vi, iaq->nirq,
2328			    navail);
2329			num_vis = 1;
2330			iaq->ntxq_vi = iaq->nrxq_vi = 0;
2331			iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
2332			iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
2333			goto restart;
2334		}
2335
2336		/*
2337		 * Second best option: a vector for errors, one for the firmware
2338		 * event queue, and vectors for either all the NIC rx queues or
2339		 * all the TOE rx queues.  The queues that don't get vectors
2340		 * will forward their interrupts to those that do.
2341		 */
2342		iaq->nirq = T4_EXTRA_INTR;
2343		if (nrxq10g >= nofldrxq10g) {
2344			iaq->intr_flags_10g = INTR_RXQ;
2345			iaq->nirq += n10g * nrxq10g;
2346		} else {
2347			iaq->intr_flags_10g = INTR_OFLD_RXQ;
2348			iaq->nirq += n10g * nofldrxq10g;
2349		}
2350		if (nrxq1g >= nofldrxq1g) {
2351			iaq->intr_flags_1g = INTR_RXQ;
2352			iaq->nirq += n1g * nrxq1g;
2353		} else {
2354			iaq->intr_flags_1g = INTR_OFLD_RXQ;
2355			iaq->nirq += n1g * nofldrxq1g;
2356		}
2357		if (iaq->nirq <= navail &&
2358		    (itype != INTR_MSI || powerof2(iaq->nirq)))
2359			goto allocate;
2360
2361		/*
2362		 * Next best option: an interrupt vector for errors, one for the
2363		 * firmware event queue, and at least one per main-VI.  At this
2364		 * point we know we'll have to downsize nrxq and/or nofldrxq to
2365		 * fit what's available to us.
2366		 */
2367		iaq->nirq = T4_EXTRA_INTR;
2368		iaq->nirq += n10g + n1g;
2369		if (iaq->nirq <= navail) {
2370			int leftover = navail - iaq->nirq;
2371
2372			if (n10g > 0) {
2373				int target = max(nrxq10g, nofldrxq10g);
2374
2375				iaq->intr_flags_10g = nrxq10g >= nofldrxq10g ?
2376				    INTR_RXQ : INTR_OFLD_RXQ;
2377
2378				n = 1;
2379				while (n < target && leftover >= n10g) {
2380					leftover -= n10g;
2381					iaq->nirq += n10g;
2382					n++;
2383				}
2384				iaq->nrxq10g = min(n, nrxq10g);
2385#ifdef TCP_OFFLOAD
2386				iaq->nofldrxq10g = min(n, nofldrxq10g);
2387#endif
2388			}
2389
2390			if (n1g > 0) {
2391				int target = max(nrxq1g, nofldrxq1g);
2392
2393				iaq->intr_flags_1g = nrxq1g >= nofldrxq1g ?
2394				    INTR_RXQ : INTR_OFLD_RXQ;
2395
2396				n = 1;
2397				while (n < target && leftover >= n1g) {
2398					leftover -= n1g;
2399					iaq->nirq += n1g;
2400					n++;
2401				}
2402				iaq->nrxq1g = min(n, nrxq1g);
2403#ifdef TCP_OFFLOAD
2404				iaq->nofldrxq1g = min(n, nofldrxq1g);
2405#endif
2406			}
2407
2408			if (itype != INTR_MSI || powerof2(iaq->nirq))
2409				goto allocate;
2410		}
2411
2412		/*
2413		 * Least desirable option: one interrupt vector for everything.
2414		 */
2415		iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2416		iaq->intr_flags_10g = iaq->intr_flags_1g = 0;
2417#ifdef TCP_OFFLOAD
2418		if (is_offload(sc))
2419			iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2420#endif
2421allocate:
2422		navail = iaq->nirq;
2423		rc = 0;
2424		if (itype == INTR_MSIX)
2425			rc = pci_alloc_msix(sc->dev, &navail);
2426		else if (itype == INTR_MSI)
2427			rc = pci_alloc_msi(sc->dev, &navail);
2428
2429		if (rc == 0) {
2430			if (navail == iaq->nirq)
2431				return (0);
2432
2433			/*
2434			 * Didn't get the number requested.  Use whatever number
2435			 * the kernel is willing to allocate (it's in navail).
2436			 */
2437			device_printf(sc->dev, "fewer vectors than requested, "
2438			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2439			    itype, iaq->nirq, navail);
2440			pci_release_msi(sc->dev);
2441			goto restart;
2442		}
2443
2444		device_printf(sc->dev,
2445		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2446		    itype, rc, iaq->nirq, navail);
2447	}
2448
2449	device_printf(sc->dev,
2450	    "failed to find a usable interrupt type.  "
2451	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2452	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2453
2454	return (ENXIO);
2455}
2456
2457#define FW_VERSION(chip) ( \
2458    V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2459    V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2460    V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2461    V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2462#define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2463
2464struct fw_info {
2465	uint8_t chip;
2466	char *kld_name;
2467	char *fw_mod_name;
2468	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
2469} fw_info[] = {
2470	{
2471		.chip = CHELSIO_T4,
2472		.kld_name = "t4fw_cfg",
2473		.fw_mod_name = "t4fw",
2474		.fw_hdr = {
2475			.chip = FW_HDR_CHIP_T4,
2476			.fw_ver = htobe32_const(FW_VERSION(T4)),
2477			.intfver_nic = FW_INTFVER(T4, NIC),
2478			.intfver_vnic = FW_INTFVER(T4, VNIC),
2479			.intfver_ofld = FW_INTFVER(T4, OFLD),
2480			.intfver_ri = FW_INTFVER(T4, RI),
2481			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2482			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
2483			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2484			.intfver_fcoe = FW_INTFVER(T4, FCOE),
2485		},
2486	}, {
2487		.chip = CHELSIO_T5,
2488		.kld_name = "t5fw_cfg",
2489		.fw_mod_name = "t5fw",
2490		.fw_hdr = {
2491			.chip = FW_HDR_CHIP_T5,
2492			.fw_ver = htobe32_const(FW_VERSION(T5)),
2493			.intfver_nic = FW_INTFVER(T5, NIC),
2494			.intfver_vnic = FW_INTFVER(T5, VNIC),
2495			.intfver_ofld = FW_INTFVER(T5, OFLD),
2496			.intfver_ri = FW_INTFVER(T5, RI),
2497			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
2498			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
2499			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
2500			.intfver_fcoe = FW_INTFVER(T5, FCOE),
2501		},
2502	}
2503};
2504
2505static struct fw_info *
2506find_fw_info(int chip)
2507{
2508	int i;
2509
2510	for (i = 0; i < nitems(fw_info); i++) {
2511		if (fw_info[i].chip == chip)
2512			return (&fw_info[i]);
2513	}
2514	return (NULL);
2515}
2516
2517/*
2518 * Is the given firmware API compatible with the one the driver was compiled
2519 * with?
2520 */
2521static int
2522fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
2523{
2524
2525	/* short circuit if it's the exact same firmware version */
2526	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
2527		return (1);
2528
2529	/*
2530	 * XXX: Is this too conservative?  Perhaps I should limit this to the
2531	 * features that are supported in the driver.
2532	 */
2533#define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
2534	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
2535	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
2536	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
2537		return (1);
2538#undef SAME_INTF
2539
2540	return (0);
2541}
2542
2543/*
2544 * The firmware in the KLD is usable, but should it be installed?  This routine
2545 * explains itself in detail if it indicates the KLD firmware should be
2546 * installed.
2547 */
2548static int
2549should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
2550{
2551	const char *reason;
2552
2553	if (!card_fw_usable) {
2554		reason = "incompatible or unusable";
2555		goto install;
2556	}
2557
2558	if (k > c) {
2559		reason = "older than the version bundled with this driver";
2560		goto install;
2561	}
2562
2563	if (t4_fw_install == 2 && k != c) {
2564		reason = "different than the version bundled with this driver";
2565		goto install;
2566	}
2567
2568	return (0);
2569
2570install:
2571	if (t4_fw_install == 0) {
2572		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2573		    "but the driver is prohibited from installing a different "
2574		    "firmware on the card.\n",
2575		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2576		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
2577
2578		return (0);
2579	}
2580
2581	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2582	    "installing firmware %u.%u.%u.%u on card.\n",
2583	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2584	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
2585	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2586	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2587
2588	return (1);
2589}
2590/*
2591 * Establish contact with the firmware and determine if we are the master driver
2592 * or not, and whether we are responsible for chip initialization.
2593 */
2594static int
2595prep_firmware(struct adapter *sc)
2596{
2597	const struct firmware *fw = NULL, *default_cfg;
2598	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
2599	enum dev_state state;
2600	struct fw_info *fw_info;
2601	struct fw_hdr *card_fw;		/* fw on the card */
2602	const struct fw_hdr *kld_fw;	/* fw in the KLD */
2603	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
2604					   against */
2605
2606	/* Contact firmware. */
2607	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
2608	if (rc < 0 || state == DEV_STATE_ERR) {
2609		rc = -rc;
2610		device_printf(sc->dev,
2611		    "failed to connect to the firmware: %d, %d.\n", rc, state);
2612		return (rc);
2613	}
2614	pf = rc;
2615	if (pf == sc->mbox)
2616		sc->flags |= MASTER_PF;
2617	else if (state == DEV_STATE_UNINIT) {
2618		/*
2619		 * We didn't get to be the master so we definitely won't be
2620		 * configuring the chip.  It's a bug if someone else hasn't
2621		 * configured it already.
2622		 */
2623		device_printf(sc->dev, "couldn't be master(%d), "
2624		    "device not already initialized either(%d).\n", rc, state);
2625		return (EDOOFUS);
2626	}
2627
2628	/* This is the firmware whose headers the driver was compiled against */
2629	fw_info = find_fw_info(chip_id(sc));
2630	if (fw_info == NULL) {
2631		device_printf(sc->dev,
2632		    "unable to look up firmware information for chip %d.\n",
2633		    chip_id(sc));
2634		return (EINVAL);
2635	}
2636	drv_fw = &fw_info->fw_hdr;
2637
2638	/*
2639	 * The firmware KLD contains many modules.  The KLD name is also the
2640	 * name of the module that contains the default config file.
2641	 */
2642	default_cfg = firmware_get(fw_info->kld_name);
2643
2644	/* Read the header of the firmware on the card */
2645	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
2646	rc = -t4_read_flash(sc, FLASH_FW_START,
2647	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
2648	if (rc == 0)
2649		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
2650	else {
2651		device_printf(sc->dev,
2652		    "Unable to read card's firmware header: %d\n", rc);
2653		card_fw_usable = 0;
2654	}
2655
2656	/* This is the firmware in the KLD */
2657	fw = firmware_get(fw_info->fw_mod_name);
2658	if (fw != NULL) {
2659		kld_fw = (const void *)fw->data;
2660		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
2661	} else {
2662		kld_fw = NULL;
2663		kld_fw_usable = 0;
2664	}
2665
2666	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
2667	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
2668		/*
2669		 * Common case: the firmware on the card is an exact match and
2670		 * the KLD is an exact match too, or the KLD is
2671		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
2672		 * here -- use cxgbetool loadfw if you want to reinstall the
2673		 * same firmware as the one on the card.
2674		 */
2675	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
2676	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
2677	    be32toh(card_fw->fw_ver))) {
2678
2679		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
2680		if (rc != 0) {
2681			device_printf(sc->dev,
2682			    "failed to install firmware: %d\n", rc);
2683			goto done;
2684		}
2685
2686		/* Installed successfully, update the cached header too. */
2687		memcpy(card_fw, kld_fw, sizeof(*card_fw));
2688		card_fw_usable = 1;
2689		need_fw_reset = 0;	/* already reset as part of load_fw */
2690	}
2691
2692	if (!card_fw_usable) {
2693		uint32_t d, c, k;
2694
2695		d = ntohl(drv_fw->fw_ver);
2696		c = ntohl(card_fw->fw_ver);
2697		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
2698
2699		device_printf(sc->dev, "Cannot find a usable firmware: "
2700		    "fw_install %d, chip state %d, "
2701		    "driver compiled with %d.%d.%d.%d, "
2702		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
2703		    t4_fw_install, state,
2704		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
2705		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
2706		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2707		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
2708		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2709		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2710		rc = EINVAL;
2711		goto done;
2712	}
2713
2714	/* We're using whatever's on the card and it's known to be good. */
2715	sc->params.fw_vers = ntohl(card_fw->fw_ver);
2716	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
2717	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
2718	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
2719	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
2720	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
2721
2722	t4_get_tp_version(sc, &sc->params.tp_vers);
2723	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
2724	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
2725	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
2726	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
2727	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
2728
2729	if (t4_get_exprom_version(sc, &sc->params.exprom_vers) != 0)
2730		sc->params.exprom_vers = 0;
2731	else {
2732		snprintf(sc->exprom_version, sizeof(sc->exprom_version),
2733		    "%u.%u.%u.%u",
2734		    G_FW_HDR_FW_VER_MAJOR(sc->params.exprom_vers),
2735		    G_FW_HDR_FW_VER_MINOR(sc->params.exprom_vers),
2736		    G_FW_HDR_FW_VER_MICRO(sc->params.exprom_vers),
2737		    G_FW_HDR_FW_VER_BUILD(sc->params.exprom_vers));
2738	}
2739
2740	/* Reset device */
2741	if (need_fw_reset &&
2742	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
2743		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
2744		if (rc != ETIMEDOUT && rc != EIO)
2745			t4_fw_bye(sc, sc->mbox);
2746		goto done;
2747	}
2748	sc->flags |= FW_OK;
2749
2750	rc = get_params__pre_init(sc);
2751	if (rc != 0)
2752		goto done; /* error message displayed already */
2753
2754	/* Partition adapter resources as specified in the config file. */
2755	if (state == DEV_STATE_UNINIT) {
2756
2757		KASSERT(sc->flags & MASTER_PF,
2758		    ("%s: trying to change chip settings when not master.",
2759		    __func__));
2760
2761		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
2762		if (rc != 0)
2763			goto done;	/* error message displayed already */
2764
2765		t4_tweak_chip_settings(sc);
2766
2767		/* get basic stuff going */
2768		rc = -t4_fw_initialize(sc, sc->mbox);
2769		if (rc != 0) {
2770			device_printf(sc->dev, "fw init failed: %d.\n", rc);
2771			goto done;
2772		}
2773	} else {
2774		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
2775		sc->cfcsum = 0;
2776	}
2777
2778done:
2779	free(card_fw, M_CXGBE);
2780	if (fw != NULL)
2781		firmware_put(fw, FIRMWARE_UNLOAD);
2782	if (default_cfg != NULL)
2783		firmware_put(default_cfg, FIRMWARE_UNLOAD);
2784
2785	return (rc);
2786}
2787
2788#define FW_PARAM_DEV(param) \
2789	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
2790	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
2791#define FW_PARAM_PFVF(param) \
2792	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
2793	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
2794
2795/*
2796 * Partition chip resources for use between various PFs, VFs, etc.
2797 */
2798static int
2799partition_resources(struct adapter *sc, const struct firmware *default_cfg,
2800    const char *name_prefix)
2801{
2802	const struct firmware *cfg = NULL;
2803	int rc = 0;
2804	struct fw_caps_config_cmd caps;
2805	uint32_t mtype, moff, finicsum, cfcsum;
2806
2807	/*
2808	 * Figure out what configuration file to use.  Pick the default config
2809	 * file for the card if the user hasn't specified one explicitly.
2810	 */
2811	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
2812	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
2813		/* Card specific overrides go here. */
2814		if (pci_get_device(sc->dev) == 0x440a)
2815			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
2816		if (is_fpga(sc))
2817			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
2818	}
2819
2820	/*
2821	 * We need to load another module if the profile is anything except
2822	 * "default" or "flash".
2823	 */
2824	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
2825	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
2826		char s[32];
2827
2828		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
2829		cfg = firmware_get(s);
2830		if (cfg == NULL) {
2831			if (default_cfg != NULL) {
2832				device_printf(sc->dev,
2833				    "unable to load module \"%s\" for "
2834				    "configuration profile \"%s\", will use "
2835				    "the default config file instead.\n",
2836				    s, sc->cfg_file);
2837				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
2838				    "%s", DEFAULT_CF);
2839			} else {
2840				device_printf(sc->dev,
2841				    "unable to load module \"%s\" for "
2842				    "configuration profile \"%s\", will use "
2843				    "the config file on the card's flash "
2844				    "instead.\n", s, sc->cfg_file);
2845				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
2846				    "%s", FLASH_CF);
2847			}
2848		}
2849	}
2850
2851	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
2852	    default_cfg == NULL) {
2853		device_printf(sc->dev,
2854		    "default config file not available, will use the config "
2855		    "file on the card's flash instead.\n");
2856		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
2857	}
2858
2859	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
2860		u_int cflen;
2861		const uint32_t *cfdata;
2862		uint32_t param, val, addr;
2863
2864		KASSERT(cfg != NULL || default_cfg != NULL,
2865		    ("%s: no config to upload", __func__));
2866
2867		/*
2868		 * Ask the firmware where it wants us to upload the config file.
2869		 */
2870		param = FW_PARAM_DEV(CF);
2871		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2872		if (rc != 0) {
2873			/* No support for config file?  Shouldn't happen. */
2874			device_printf(sc->dev,
2875			    "failed to query config file location: %d.\n", rc);
2876			goto done;
2877		}
2878		mtype = G_FW_PARAMS_PARAM_Y(val);
2879		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
2880
2881		/*
2882		 * XXX: sheer laziness.  We deliberately added 4 bytes of
2883		 * useless stuffing/comments at the end of the config file so
2884		 * it's ok to simply throw away the last remaining bytes when
2885		 * the config file is not an exact multiple of 4.  This also
2886		 * helps with the validate_mt_off_len check.
2887		 */
2888		if (cfg != NULL) {
2889			cflen = cfg->datasize & ~3;
2890			cfdata = cfg->data;
2891		} else {
2892			cflen = default_cfg->datasize & ~3;
2893			cfdata = default_cfg->data;
2894		}
2895
2896		if (cflen > FLASH_CFG_MAX_SIZE) {
2897			device_printf(sc->dev,
2898			    "config file too long (%d, max allowed is %d).  "
2899			    "Will try to use the config on the card, if any.\n",
2900			    cflen, FLASH_CFG_MAX_SIZE);
2901			goto use_config_on_flash;
2902		}
2903
2904		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
2905		if (rc != 0) {
2906			device_printf(sc->dev,
2907			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
2908			    "Will try to use the config on the card, if any.\n",
2909			    __func__, mtype, moff, cflen, rc);
2910			goto use_config_on_flash;
2911		}
2912		write_via_memwin(sc, 2, addr, cfdata, cflen);
2913	} else {
2914use_config_on_flash:
2915		mtype = FW_MEMTYPE_FLASH;
2916		moff = t4_flash_cfg_addr(sc);
2917	}
2918
2919	bzero(&caps, sizeof(caps));
2920	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2921	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
2922	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
2923	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
2924	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
2925	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
2926	if (rc != 0) {
2927		device_printf(sc->dev,
2928		    "failed to pre-process config file: %d "
2929		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
2930		goto done;
2931	}
2932
2933	finicsum = be32toh(caps.finicsum);
2934	cfcsum = be32toh(caps.cfcsum);
2935	if (finicsum != cfcsum) {
2936		device_printf(sc->dev,
2937		    "WARNING: config file checksum mismatch: %08x %08x\n",
2938		    finicsum, cfcsum);
2939	}
2940	sc->cfcsum = cfcsum;
2941
2942#define LIMIT_CAPS(x) do { \
2943	caps.x &= htobe16(t4_##x##_allowed); \
2944} while (0)
2945
2946	/*
2947	 * Let the firmware know what features will (not) be used so it can tune
2948	 * things accordingly.
2949	 */
2950	LIMIT_CAPS(nbmcaps);
2951	LIMIT_CAPS(linkcaps);
2952	LIMIT_CAPS(switchcaps);
2953	LIMIT_CAPS(niccaps);
2954	LIMIT_CAPS(toecaps);
2955	LIMIT_CAPS(rdmacaps);
2956	LIMIT_CAPS(tlscaps);
2957	LIMIT_CAPS(iscsicaps);
2958	LIMIT_CAPS(fcoecaps);
2959#undef LIMIT_CAPS
2960
2961	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2962	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
2963	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
2964	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
2965	if (rc != 0) {
2966		device_printf(sc->dev,
2967		    "failed to process config file: %d.\n", rc);
2968	}
2969done:
2970	if (cfg != NULL)
2971		firmware_put(cfg, FIRMWARE_UNLOAD);
2972	return (rc);
2973}
2974
2975/*
2976 * Retrieve parameters that are needed (or nice to have) very early.
2977 */
2978static int
2979get_params__pre_init(struct adapter *sc)
2980{
2981	int rc;
2982	uint32_t param[2], val[2];
2983
2984	param[0] = FW_PARAM_DEV(PORTVEC);
2985	param[1] = FW_PARAM_DEV(CCLK);
2986	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
2987	if (rc != 0) {
2988		device_printf(sc->dev,
2989		    "failed to query parameters (pre_init): %d.\n", rc);
2990		return (rc);
2991	}
2992
2993	sc->params.portvec = val[0];
2994	sc->params.nports = bitcount32(val[0]);
2995	sc->params.vpd.cclk = val[1];
2996
2997	/* Read device log parameters. */
2998	rc = -t4_init_devlog_params(sc, 1);
2999	if (rc == 0)
3000		fixup_devlog_params(sc);
3001	else {
3002		device_printf(sc->dev,
3003		    "failed to get devlog parameters: %d.\n", rc);
3004		rc = 0;	/* devlog isn't critical for device operation */
3005	}
3006
3007	return (rc);
3008}
3009
3010/*
3011 * Retrieve various parameters that are of interest to the driver.  The device
3012 * has been initialized by the firmware at this point.
3013 */
3014static int
3015get_params__post_init(struct adapter *sc)
3016{
3017	int rc;
3018	uint32_t param[7], val[7];
3019	struct fw_caps_config_cmd caps;
3020
3021	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3022	param[1] = FW_PARAM_PFVF(EQ_START);
3023	param[2] = FW_PARAM_PFVF(FILTER_START);
3024	param[3] = FW_PARAM_PFVF(FILTER_END);
3025	param[4] = FW_PARAM_PFVF(L2T_START);
3026	param[5] = FW_PARAM_PFVF(L2T_END);
3027	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3028	if (rc != 0) {
3029		device_printf(sc->dev,
3030		    "failed to query parameters (post_init): %d.\n", rc);
3031		return (rc);
3032	}
3033
3034	sc->sge.iq_start = val[0];
3035	sc->sge.eq_start = val[1];
3036	sc->tids.ftid_base = val[2];
3037	sc->tids.nftids = val[3] - val[2] + 1;
3038	sc->params.ftid_min = val[2];
3039	sc->params.ftid_max = val[3];
3040	sc->vres.l2t.start = val[4];
3041	sc->vres.l2t.size = val[5] - val[4] + 1;
3042	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3043	    ("%s: L2 table size (%u) larger than expected (%u)",
3044	    __func__, sc->vres.l2t.size, L2T_SIZE));
3045
3046	/* get capabilites */
3047	bzero(&caps, sizeof(caps));
3048	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3049	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3050	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3051	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3052	if (rc != 0) {
3053		device_printf(sc->dev,
3054		    "failed to get card capabilities: %d.\n", rc);
3055		return (rc);
3056	}
3057
3058#define READ_CAPS(x) do { \
3059	sc->x = htobe16(caps.x); \
3060} while (0)
3061	READ_CAPS(nbmcaps);
3062	READ_CAPS(linkcaps);
3063	READ_CAPS(switchcaps);
3064	READ_CAPS(niccaps);
3065	READ_CAPS(toecaps);
3066	READ_CAPS(rdmacaps);
3067	READ_CAPS(tlscaps);
3068	READ_CAPS(iscsicaps);
3069	READ_CAPS(fcoecaps);
3070
3071	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3072		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3073		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3074		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3075		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3076		if (rc != 0) {
3077			device_printf(sc->dev,
3078			    "failed to query NIC parameters: %d.\n", rc);
3079			return (rc);
3080		}
3081		sc->tids.etid_base = val[0];
3082		sc->params.etid_min = val[0];
3083		sc->tids.netids = val[1] - val[0] + 1;
3084		sc->params.netids = sc->tids.netids;
3085		sc->params.eo_wr_cred = val[2];
3086		sc->params.ethoffload = 1;
3087	}
3088
3089	if (sc->toecaps) {
3090		/* query offload-related parameters */
3091		param[0] = FW_PARAM_DEV(NTID);
3092		param[1] = FW_PARAM_PFVF(SERVER_START);
3093		param[2] = FW_PARAM_PFVF(SERVER_END);
3094		param[3] = FW_PARAM_PFVF(TDDP_START);
3095		param[4] = FW_PARAM_PFVF(TDDP_END);
3096		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3097		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3098		if (rc != 0) {
3099			device_printf(sc->dev,
3100			    "failed to query TOE parameters: %d.\n", rc);
3101			return (rc);
3102		}
3103		sc->tids.ntids = val[0];
3104		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3105		sc->tids.stid_base = val[1];
3106		sc->tids.nstids = val[2] - val[1] + 1;
3107		sc->vres.ddp.start = val[3];
3108		sc->vres.ddp.size = val[4] - val[3] + 1;
3109		sc->params.ofldq_wr_cred = val[5];
3110		sc->params.offload = 1;
3111	}
3112	if (sc->rdmacaps) {
3113		param[0] = FW_PARAM_PFVF(STAG_START);
3114		param[1] = FW_PARAM_PFVF(STAG_END);
3115		param[2] = FW_PARAM_PFVF(RQ_START);
3116		param[3] = FW_PARAM_PFVF(RQ_END);
3117		param[4] = FW_PARAM_PFVF(PBL_START);
3118		param[5] = FW_PARAM_PFVF(PBL_END);
3119		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3120		if (rc != 0) {
3121			device_printf(sc->dev,
3122			    "failed to query RDMA parameters(1): %d.\n", rc);
3123			return (rc);
3124		}
3125		sc->vres.stag.start = val[0];
3126		sc->vres.stag.size = val[1] - val[0] + 1;
3127		sc->vres.rq.start = val[2];
3128		sc->vres.rq.size = val[3] - val[2] + 1;
3129		sc->vres.pbl.start = val[4];
3130		sc->vres.pbl.size = val[5] - val[4] + 1;
3131
3132		param[0] = FW_PARAM_PFVF(SQRQ_START);
3133		param[1] = FW_PARAM_PFVF(SQRQ_END);
3134		param[2] = FW_PARAM_PFVF(CQ_START);
3135		param[3] = FW_PARAM_PFVF(CQ_END);
3136		param[4] = FW_PARAM_PFVF(OCQ_START);
3137		param[5] = FW_PARAM_PFVF(OCQ_END);
3138		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3139		if (rc != 0) {
3140			device_printf(sc->dev,
3141			    "failed to query RDMA parameters(2): %d.\n", rc);
3142			return (rc);
3143		}
3144		sc->vres.qp.start = val[0];
3145		sc->vres.qp.size = val[1] - val[0] + 1;
3146		sc->vres.cq.start = val[2];
3147		sc->vres.cq.size = val[3] - val[2] + 1;
3148		sc->vres.ocq.start = val[4];
3149		sc->vres.ocq.size = val[5] - val[4] + 1;
3150	}
3151	if (sc->iscsicaps) {
3152		param[0] = FW_PARAM_PFVF(ISCSI_START);
3153		param[1] = FW_PARAM_PFVF(ISCSI_END);
3154		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3155		if (rc != 0) {
3156			device_printf(sc->dev,
3157			    "failed to query iSCSI parameters: %d.\n", rc);
3158			return (rc);
3159		}
3160		sc->vres.iscsi.start = val[0];
3161		sc->vres.iscsi.size = val[1] - val[0] + 1;
3162	}
3163
3164	t4_init_sge_params(sc);
3165
3166	/*
3167	 * We've got the params we wanted to query via the firmware.  Now grab
3168	 * some others directly from the chip.
3169	 */
3170	rc = t4_read_chip_settings(sc);
3171
3172	return (rc);
3173}
3174
3175static int
3176set_params__post_init(struct adapter *sc)
3177{
3178	uint32_t param, val;
3179
3180	/* ask for encapsulated CPLs */
3181	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3182	val = 1;
3183	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3184
3185	return (0);
3186}
3187
3188#undef FW_PARAM_PFVF
3189#undef FW_PARAM_DEV
3190
3191static void
3192t4_set_desc(struct adapter *sc)
3193{
3194	char buf[128];
3195	struct adapter_params *p = &sc->params;
3196
3197	snprintf(buf, sizeof(buf), "Chelsio %s %sNIC (rev %d), S/N:%s, "
3198	    "P/N:%s, E/C:%s", p->vpd.id, is_offload(sc) ? "R" : "",
3199	    chip_rev(sc), p->vpd.sn, p->vpd.pn, p->vpd.ec);
3200
3201	device_set_desc_copy(sc->dev, buf);
3202}
3203
3204static void
3205build_medialist(struct port_info *pi, struct ifmedia *media)
3206{
3207	int m;
3208
3209	PORT_LOCK(pi);
3210
3211	ifmedia_removeall(media);
3212
3213	m = IFM_ETHER | IFM_FDX;
3214
3215	switch(pi->port_type) {
3216	case FW_PORT_TYPE_BT_XFI:
3217	case FW_PORT_TYPE_BT_XAUI:
3218		ifmedia_add(media, m | IFM_10G_T, 0, NULL);
3219		/* fall through */
3220
3221	case FW_PORT_TYPE_BT_SGMII:
3222		ifmedia_add(media, m | IFM_1000_T, 0, NULL);
3223		ifmedia_add(media, m | IFM_100_TX, 0, NULL);
3224		ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL);
3225		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
3226		break;
3227
3228	case FW_PORT_TYPE_CX4:
3229		ifmedia_add(media, m | IFM_10G_CX4, 0, NULL);
3230		ifmedia_set(media, m | IFM_10G_CX4);
3231		break;
3232
3233	case FW_PORT_TYPE_QSFP_10G:
3234	case FW_PORT_TYPE_SFP:
3235	case FW_PORT_TYPE_FIBER_XFI:
3236	case FW_PORT_TYPE_FIBER_XAUI:
3237		switch (pi->mod_type) {
3238
3239		case FW_PORT_MOD_TYPE_LR:
3240			ifmedia_add(media, m | IFM_10G_LR, 0, NULL);
3241			ifmedia_set(media, m | IFM_10G_LR);
3242			break;
3243
3244		case FW_PORT_MOD_TYPE_SR:
3245			ifmedia_add(media, m | IFM_10G_SR, 0, NULL);
3246			ifmedia_set(media, m | IFM_10G_SR);
3247			break;
3248
3249		case FW_PORT_MOD_TYPE_LRM:
3250			ifmedia_add(media, m | IFM_10G_LRM, 0, NULL);
3251			ifmedia_set(media, m | IFM_10G_LRM);
3252			break;
3253
3254		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3255		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3256			ifmedia_add(media, m | IFM_10G_TWINAX, 0, NULL);
3257			ifmedia_set(media, m | IFM_10G_TWINAX);
3258			break;
3259
3260		case FW_PORT_MOD_TYPE_NONE:
3261			m &= ~IFM_FDX;
3262			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3263			ifmedia_set(media, m | IFM_NONE);
3264			break;
3265
3266		case FW_PORT_MOD_TYPE_NA:
3267		case FW_PORT_MOD_TYPE_ER:
3268		default:
3269			device_printf(pi->dev,
3270			    "unknown port_type (%d), mod_type (%d)\n",
3271			    pi->port_type, pi->mod_type);
3272			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3273			ifmedia_set(media, m | IFM_UNKNOWN);
3274			break;
3275		}
3276		break;
3277
3278	case FW_PORT_TYPE_QSFP:
3279		switch (pi->mod_type) {
3280
3281		case FW_PORT_MOD_TYPE_LR:
3282			ifmedia_add(media, m | IFM_40G_LR4, 0, NULL);
3283			ifmedia_set(media, m | IFM_40G_LR4);
3284			break;
3285
3286		case FW_PORT_MOD_TYPE_SR:
3287			ifmedia_add(media, m | IFM_40G_SR4, 0, NULL);
3288			ifmedia_set(media, m | IFM_40G_SR4);
3289			break;
3290
3291		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3292		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3293			ifmedia_add(media, m | IFM_40G_CR4, 0, NULL);
3294			ifmedia_set(media, m | IFM_40G_CR4);
3295			break;
3296
3297		case FW_PORT_MOD_TYPE_NONE:
3298			m &= ~IFM_FDX;
3299			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3300			ifmedia_set(media, m | IFM_NONE);
3301			break;
3302
3303		default:
3304			device_printf(pi->dev,
3305			    "unknown port_type (%d), mod_type (%d)\n",
3306			    pi->port_type, pi->mod_type);
3307			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3308			ifmedia_set(media, m | IFM_UNKNOWN);
3309			break;
3310		}
3311		break;
3312
3313	default:
3314		device_printf(pi->dev,
3315		    "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
3316		    pi->mod_type);
3317		ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3318		ifmedia_set(media, m | IFM_UNKNOWN);
3319		break;
3320	}
3321
3322	PORT_UNLOCK(pi);
3323}
3324
3325#define FW_MAC_EXACT_CHUNK	7
3326
3327/*
3328 * Program the port's XGMAC based on parameters in ifnet.  The caller also
3329 * indicates which parameters should be programmed (the rest are left alone).
3330 */
3331int
3332update_mac_settings(struct ifnet *ifp, int flags)
3333{
3334	int rc = 0;
3335	struct vi_info *vi = ifp->if_softc;
3336	struct port_info *pi = vi->pi;
3337	struct adapter *sc = pi->adapter;
3338	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
3339
3340	ASSERT_SYNCHRONIZED_OP(sc);
3341	KASSERT(flags, ("%s: not told what to update.", __func__));
3342
3343	if (flags & XGMAC_MTU)
3344		mtu = ifp->if_mtu;
3345
3346	if (flags & XGMAC_PROMISC)
3347		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
3348
3349	if (flags & XGMAC_ALLMULTI)
3350		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
3351
3352	if (flags & XGMAC_VLANEX)
3353		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
3354
3355	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
3356		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
3357		    allmulti, 1, vlanex, false);
3358		if (rc) {
3359			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
3360			    rc);
3361			return (rc);
3362		}
3363	}
3364
3365	if (flags & XGMAC_UCADDR) {
3366		uint8_t ucaddr[ETHER_ADDR_LEN];
3367
3368		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
3369		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
3370		    ucaddr, true, true);
3371		if (rc < 0) {
3372			rc = -rc;
3373			if_printf(ifp, "change_mac failed: %d\n", rc);
3374			return (rc);
3375		} else {
3376			vi->xact_addr_filt = rc;
3377			rc = 0;
3378		}
3379	}
3380
3381	if (flags & XGMAC_MCADDRS) {
3382		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
3383		int del = 1;
3384		uint64_t hash = 0;
3385		struct ifmultiaddr *ifma;
3386		int i = 0, j;
3387
3388		if_maddr_rlock(ifp);
3389		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3390			if (ifma->ifma_addr->sa_family != AF_LINK)
3391				continue;
3392			mcaddr[i] =
3393			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
3394			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
3395			i++;
3396
3397			if (i == FW_MAC_EXACT_CHUNK) {
3398				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
3399				    del, i, mcaddr, NULL, &hash, 0);
3400				if (rc < 0) {
3401					rc = -rc;
3402					for (j = 0; j < i; j++) {
3403						if_printf(ifp,
3404						    "failed to add mc address"
3405						    " %02x:%02x:%02x:"
3406						    "%02x:%02x:%02x rc=%d\n",
3407						    mcaddr[j][0], mcaddr[j][1],
3408						    mcaddr[j][2], mcaddr[j][3],
3409						    mcaddr[j][4], mcaddr[j][5],
3410						    rc);
3411					}
3412					goto mcfail;
3413				}
3414				del = 0;
3415				i = 0;
3416			}
3417		}
3418		if (i > 0) {
3419			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
3420			    mcaddr, NULL, &hash, 0);
3421			if (rc < 0) {
3422				rc = -rc;
3423				for (j = 0; j < i; j++) {
3424					if_printf(ifp,
3425					    "failed to add mc address"
3426					    " %02x:%02x:%02x:"
3427					    "%02x:%02x:%02x rc=%d\n",
3428					    mcaddr[j][0], mcaddr[j][1],
3429					    mcaddr[j][2], mcaddr[j][3],
3430					    mcaddr[j][4], mcaddr[j][5],
3431					    rc);
3432				}
3433				goto mcfail;
3434			}
3435		}
3436
3437		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
3438		if (rc != 0)
3439			if_printf(ifp, "failed to set mc address hash: %d", rc);
3440mcfail:
3441		if_maddr_runlock(ifp);
3442	}
3443
3444	return (rc);
3445}
3446
3447/*
3448 * {begin|end}_synchronized_op must be called from the same thread.
3449 */
3450int
3451begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
3452    char *wmesg)
3453{
3454	int rc, pri;
3455
3456#ifdef WITNESS
3457	/* the caller thinks it's ok to sleep, but is it really? */
3458	if (flags & SLEEP_OK)
3459		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
3460		    "begin_synchronized_op");
3461#endif
3462
3463	if (INTR_OK)
3464		pri = PCATCH;
3465	else
3466		pri = 0;
3467
3468	ADAPTER_LOCK(sc);
3469	for (;;) {
3470
3471		if (vi && IS_DOOMED(vi)) {
3472			rc = ENXIO;
3473			goto done;
3474		}
3475
3476		if (!IS_BUSY(sc)) {
3477			rc = 0;
3478			break;
3479		}
3480
3481		if (!(flags & SLEEP_OK)) {
3482			rc = EBUSY;
3483			goto done;
3484		}
3485
3486		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
3487			rc = EINTR;
3488			goto done;
3489		}
3490	}
3491
3492	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
3493	SET_BUSY(sc);
3494#ifdef INVARIANTS
3495	sc->last_op = wmesg;
3496	sc->last_op_thr = curthread;
3497	sc->last_op_flags = flags;
3498#endif
3499
3500done:
3501	if (!(flags & HOLD_LOCK) || rc)
3502		ADAPTER_UNLOCK(sc);
3503
3504	return (rc);
3505}
3506
3507/*
3508 * Tell if_ioctl and if_init that the VI is going away.  This is
3509 * special variant of begin_synchronized_op and must be paired with a
3510 * call to end_synchronized_op.
3511 */
3512void
3513doom_vi(struct adapter *sc, struct vi_info *vi)
3514{
3515
3516	ADAPTER_LOCK(sc);
3517	SET_DOOMED(vi);
3518	wakeup(&sc->flags);
3519	while (IS_BUSY(sc))
3520		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
3521	SET_BUSY(sc);
3522#ifdef INVARIANTS
3523	sc->last_op = "t4detach";
3524	sc->last_op_thr = curthread;
3525	sc->last_op_flags = 0;
3526#endif
3527	ADAPTER_UNLOCK(sc);
3528}
3529
3530/*
3531 * {begin|end}_synchronized_op must be called from the same thread.
3532 */
3533void
3534end_synchronized_op(struct adapter *sc, int flags)
3535{
3536
3537	if (flags & LOCK_HELD)
3538		ADAPTER_LOCK_ASSERT_OWNED(sc);
3539	else
3540		ADAPTER_LOCK(sc);
3541
3542	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
3543	CLR_BUSY(sc);
3544	wakeup(&sc->flags);
3545	ADAPTER_UNLOCK(sc);
3546}
3547
3548static int
3549cxgbe_init_synchronized(struct vi_info *vi)
3550{
3551	struct port_info *pi = vi->pi;
3552	struct adapter *sc = pi->adapter;
3553	struct ifnet *ifp = vi->ifp;
3554	int rc = 0, i;
3555	struct sge_txq *txq;
3556
3557	ASSERT_SYNCHRONIZED_OP(sc);
3558
3559	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3560		return (0);	/* already running */
3561
3562	if (!(sc->flags & FULL_INIT_DONE) &&
3563	    ((rc = adapter_full_init(sc)) != 0))
3564		return (rc);	/* error message displayed already */
3565
3566	if (!(vi->flags & VI_INIT_DONE) &&
3567	    ((rc = vi_full_init(vi)) != 0))
3568		return (rc); /* error message displayed already */
3569
3570	rc = update_mac_settings(ifp, XGMAC_ALL);
3571	if (rc)
3572		goto done;	/* error message displayed already */
3573
3574	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
3575	if (rc != 0) {
3576		if_printf(ifp, "enable_vi failed: %d\n", rc);
3577		goto done;
3578	}
3579
3580	/*
3581	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
3582	 * if this changes.
3583	 */
3584
3585	for_each_txq(vi, i, txq) {
3586		TXQ_LOCK(txq);
3587		txq->eq.flags |= EQ_ENABLED;
3588		TXQ_UNLOCK(txq);
3589	}
3590
3591	/*
3592	 * The first iq of the first port to come up is used for tracing.
3593	 */
3594	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
3595		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
3596		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
3597		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
3598		    V_QUEUENUMBER(sc->traceq));
3599		pi->flags |= HAS_TRACEQ;
3600	}
3601
3602	/* all ok */
3603	PORT_LOCK(pi);
3604	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3605	pi->up_vis++;
3606
3607	if (pi->nvi > 1 || sc->flags & IS_VF)
3608		callout_reset(&vi->tick, hz, vi_tick, vi);
3609	else
3610		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
3611	PORT_UNLOCK(pi);
3612done:
3613	if (rc != 0)
3614		cxgbe_uninit_synchronized(vi);
3615
3616	return (rc);
3617}
3618
3619/*
3620 * Idempotent.
3621 */
3622static int
3623cxgbe_uninit_synchronized(struct vi_info *vi)
3624{
3625	struct port_info *pi = vi->pi;
3626	struct adapter *sc = pi->adapter;
3627	struct ifnet *ifp = vi->ifp;
3628	int rc, i;
3629	struct sge_txq *txq;
3630
3631	ASSERT_SYNCHRONIZED_OP(sc);
3632
3633	if (!(vi->flags & VI_INIT_DONE)) {
3634		KASSERT(!(ifp->if_drv_flags & IFF_DRV_RUNNING),
3635		    ("uninited VI is running"));
3636		return (0);
3637	}
3638
3639	/*
3640	 * Disable the VI so that all its data in either direction is discarded
3641	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
3642	 * tick) intact as the TP can deliver negative advice or data that it's
3643	 * holding in its RAM (for an offloaded connection) even after the VI is
3644	 * disabled.
3645	 */
3646	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
3647	if (rc) {
3648		if_printf(ifp, "disable_vi failed: %d\n", rc);
3649		return (rc);
3650	}
3651
3652	for_each_txq(vi, i, txq) {
3653		TXQ_LOCK(txq);
3654		txq->eq.flags &= ~EQ_ENABLED;
3655		TXQ_UNLOCK(txq);
3656	}
3657
3658	PORT_LOCK(pi);
3659	if (pi->nvi > 1 || sc->flags & IS_VF)
3660		callout_stop(&vi->tick);
3661	else
3662		callout_stop(&pi->tick);
3663	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
3664		PORT_UNLOCK(pi);
3665		return (0);
3666	}
3667	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3668	pi->up_vis--;
3669	if (pi->up_vis > 0) {
3670		PORT_UNLOCK(pi);
3671		return (0);
3672	}
3673	PORT_UNLOCK(pi);
3674
3675	pi->link_cfg.link_ok = 0;
3676	pi->link_cfg.speed = 0;
3677	pi->linkdnrc = -1;
3678	t4_os_link_changed(sc, pi->port_id, 0, -1);
3679
3680	return (0);
3681}
3682
3683/*
3684 * It is ok for this function to fail midway and return right away.  t4_detach
3685 * will walk the entire sc->irq list and clean up whatever is valid.
3686 */
3687int
3688t4_setup_intr_handlers(struct adapter *sc)
3689{
3690	int rc, rid, p, q, v;
3691	char s[8];
3692	struct irq *irq;
3693	struct port_info *pi;
3694	struct vi_info *vi;
3695	struct sge *sge = &sc->sge;
3696	struct sge_rxq *rxq;
3697#ifdef TCP_OFFLOAD
3698	struct sge_ofld_rxq *ofld_rxq;
3699#endif
3700#ifdef DEV_NETMAP
3701	struct sge_nm_rxq *nm_rxq;
3702#endif
3703
3704	/*
3705	 * Setup interrupts.
3706	 */
3707	irq = &sc->irq[0];
3708	rid = sc->intr_type == INTR_INTX ? 0 : 1;
3709	if (sc->intr_count == 1)
3710		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
3711
3712	/* Multiple interrupts. */
3713	if (sc->flags & IS_VF)
3714		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
3715		    ("%s: too few intr.", __func__));
3716	else
3717		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
3718		    ("%s: too few intr.", __func__));
3719
3720	/* The first one is always error intr on PFs */
3721	if (!(sc->flags & IS_VF)) {
3722		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
3723		if (rc != 0)
3724			return (rc);
3725		irq++;
3726		rid++;
3727	}
3728
3729	/* The second one is always the firmware event queue (first on VFs) */
3730	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
3731	if (rc != 0)
3732		return (rc);
3733	irq++;
3734	rid++;
3735
3736	for_each_port(sc, p) {
3737		pi = sc->port[p];
3738		for_each_vi(pi, v, vi) {
3739			vi->first_intr = rid - 1;
3740
3741			if (vi->nnmrxq > 0) {
3742				int n = max(vi->nrxq, vi->nnmrxq);
3743
3744				MPASS(vi->flags & INTR_RXQ);
3745
3746				rxq = &sge->rxq[vi->first_rxq];
3747#ifdef DEV_NETMAP
3748				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
3749#endif
3750				for (q = 0; q < n; q++) {
3751					snprintf(s, sizeof(s), "%x%c%x", p,
3752					    'a' + v, q);
3753					if (q < vi->nrxq)
3754						irq->rxq = rxq++;
3755#ifdef DEV_NETMAP
3756					if (q < vi->nnmrxq)
3757						irq->nm_rxq = nm_rxq++;
3758#endif
3759					rc = t4_alloc_irq(sc, irq, rid,
3760					    t4_vi_intr, irq, s);
3761					if (rc != 0)
3762						return (rc);
3763					irq++;
3764					rid++;
3765					vi->nintr++;
3766				}
3767			} else if (vi->flags & INTR_RXQ) {
3768				for_each_rxq(vi, q, rxq) {
3769					snprintf(s, sizeof(s), "%x%c%x", p,
3770					    'a' + v, q);
3771					rc = t4_alloc_irq(sc, irq, rid,
3772					    t4_intr, rxq, s);
3773					if (rc != 0)
3774						return (rc);
3775					irq++;
3776					rid++;
3777					vi->nintr++;
3778				}
3779			}
3780#ifdef TCP_OFFLOAD
3781			if (vi->flags & INTR_OFLD_RXQ) {
3782				for_each_ofld_rxq(vi, q, ofld_rxq) {
3783					snprintf(s, sizeof(s), "%x%c%x", p,
3784					    'A' + v, q);
3785					rc = t4_alloc_irq(sc, irq, rid,
3786					    t4_intr, ofld_rxq, s);
3787					if (rc != 0)
3788						return (rc);
3789					irq++;
3790					rid++;
3791					vi->nintr++;
3792				}
3793			}
3794#endif
3795		}
3796	}
3797	MPASS(irq == &sc->irq[sc->intr_count]);
3798
3799	return (0);
3800}
3801
3802int
3803adapter_full_init(struct adapter *sc)
3804{
3805	int rc, i;
3806
3807	ASSERT_SYNCHRONIZED_OP(sc);
3808	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
3809	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
3810	    ("%s: FULL_INIT_DONE already", __func__));
3811
3812	/*
3813	 * queues that belong to the adapter (not any particular port).
3814	 */
3815	rc = t4_setup_adapter_queues(sc);
3816	if (rc != 0)
3817		goto done;
3818
3819	for (i = 0; i < nitems(sc->tq); i++) {
3820		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
3821		    taskqueue_thread_enqueue, &sc->tq[i]);
3822		if (sc->tq[i] == NULL) {
3823			device_printf(sc->dev,
3824			    "failed to allocate task queue %d\n", i);
3825			rc = ENOMEM;
3826			goto done;
3827		}
3828		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
3829		    device_get_nameunit(sc->dev), i);
3830	}
3831
3832	if (!(sc->flags & IS_VF))
3833		t4_intr_enable(sc);
3834	sc->flags |= FULL_INIT_DONE;
3835done:
3836	if (rc != 0)
3837		adapter_full_uninit(sc);
3838
3839	return (rc);
3840}
3841
3842int
3843adapter_full_uninit(struct adapter *sc)
3844{
3845	int i;
3846
3847	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
3848
3849	t4_teardown_adapter_queues(sc);
3850
3851	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
3852		taskqueue_free(sc->tq[i]);
3853		sc->tq[i] = NULL;
3854	}
3855
3856	sc->flags &= ~FULL_INIT_DONE;
3857
3858	return (0);
3859}
3860
3861#ifdef RSS
3862#define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
3863    RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
3864    RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
3865    RSS_HASHTYPE_RSS_UDP_IPV6)
3866
3867/* Translates kernel hash types to hardware. */
3868static int
3869hashconfig_to_hashen(int hashconfig)
3870{
3871	int hashen = 0;
3872
3873	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
3874		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
3875	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
3876		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
3877	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
3878		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
3879		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
3880	}
3881	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
3882		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
3883		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
3884	}
3885	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
3886		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
3887	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
3888		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
3889
3890	return (hashen);
3891}
3892
3893/* Translates hardware hash types to kernel. */
3894static int
3895hashen_to_hashconfig(int hashen)
3896{
3897	int hashconfig = 0;
3898
3899	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
3900		/*
3901		 * If UDP hashing was enabled it must have been enabled for
3902		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
3903		 * enabling any 4-tuple hash is nonsense configuration.
3904		 */
3905		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
3906		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
3907
3908		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
3909			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
3910		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
3911			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
3912	}
3913	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
3914		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
3915	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
3916		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
3917	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
3918		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
3919	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
3920		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
3921
3922	return (hashconfig);
3923}
3924#endif
3925
3926int
3927vi_full_init(struct vi_info *vi)
3928{
3929	struct adapter *sc = vi->pi->adapter;
3930	struct ifnet *ifp = vi->ifp;
3931	uint16_t *rss;
3932	struct sge_rxq *rxq;
3933	int rc, i, j, hashen;
3934#ifdef RSS
3935	int nbuckets = rss_getnumbuckets();
3936	int hashconfig = rss_gethashconfig();
3937	int extra;
3938	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
3939	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
3940#endif
3941
3942	ASSERT_SYNCHRONIZED_OP(sc);
3943	KASSERT((vi->flags & VI_INIT_DONE) == 0,
3944	    ("%s: VI_INIT_DONE already", __func__));
3945
3946	sysctl_ctx_init(&vi->ctx);
3947	vi->flags |= VI_SYSCTL_CTX;
3948
3949	/*
3950	 * Allocate tx/rx/fl queues for this VI.
3951	 */
3952	rc = t4_setup_vi_queues(vi);
3953	if (rc != 0)
3954		goto done;	/* error message displayed already */
3955
3956	/*
3957	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
3958	 */
3959	if (vi->nrxq > vi->rss_size) {
3960		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
3961		    "some queues will never receive traffic.\n", vi->nrxq,
3962		    vi->rss_size);
3963	} else if (vi->rss_size % vi->nrxq) {
3964		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
3965		    "expect uneven traffic distribution.\n", vi->nrxq,
3966		    vi->rss_size);
3967	}
3968#ifdef RSS
3969	MPASS(RSS_KEYSIZE == 40);
3970	if (vi->nrxq != nbuckets) {
3971		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
3972		    "performance will be impacted.\n", vi->nrxq, nbuckets);
3973	}
3974
3975	rss_getkey((void *)&raw_rss_key[0]);
3976	for (i = 0; i < nitems(rss_key); i++) {
3977		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
3978	}
3979	t4_write_rss_key(sc, &rss_key[0], -1);
3980#endif
3981	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
3982	for (i = 0; i < vi->rss_size;) {
3983#ifdef RSS
3984		j = rss_get_indirection_to_bucket(i);
3985		j %= vi->nrxq;
3986		rxq = &sc->sge.rxq[vi->first_rxq + j];
3987		rss[i++] = rxq->iq.abs_id;
3988#else
3989		for_each_rxq(vi, j, rxq) {
3990			rss[i++] = rxq->iq.abs_id;
3991			if (i == vi->rss_size)
3992				break;
3993		}
3994#endif
3995	}
3996
3997	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
3998	    vi->rss_size);
3999	if (rc != 0) {
4000		if_printf(ifp, "rss_config failed: %d\n", rc);
4001		goto done;
4002	}
4003
4004#ifdef RSS
4005	hashen = hashconfig_to_hashen(hashconfig);
4006
4007	/*
4008	 * We may have had to enable some hashes even though the global config
4009	 * wants them disabled.  This is a potential problem that must be
4010	 * reported to the user.
4011	 */
4012	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
4013
4014	/*
4015	 * If we consider only the supported hash types, then the enabled hashes
4016	 * are a superset of the requested hashes.  In other words, there cannot
4017	 * be any supported hash that was requested but not enabled, but there
4018	 * can be hashes that were not requested but had to be enabled.
4019	 */
4020	extra &= SUPPORTED_RSS_HASHTYPES;
4021	MPASS((extra & hashconfig) == 0);
4022
4023	if (extra) {
4024		if_printf(ifp,
4025		    "global RSS config (0x%x) cannot be accomodated.\n",
4026		    hashconfig);
4027	}
4028	if (extra & RSS_HASHTYPE_RSS_IPV4)
4029		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
4030	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
4031		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
4032	if (extra & RSS_HASHTYPE_RSS_IPV6)
4033		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
4034	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
4035		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
4036	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
4037		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
4038	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
4039		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
4040#else
4041	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
4042	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
4043	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4044	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
4045#endif
4046	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0]);
4047	if (rc != 0) {
4048		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
4049		goto done;
4050	}
4051
4052	vi->rss = rss;
4053	vi->flags |= VI_INIT_DONE;
4054done:
4055	if (rc != 0)
4056		vi_full_uninit(vi);
4057
4058	return (rc);
4059}
4060
4061/*
4062 * Idempotent.
4063 */
4064int
4065vi_full_uninit(struct vi_info *vi)
4066{
4067	struct port_info *pi = vi->pi;
4068	struct adapter *sc = pi->adapter;
4069	int i;
4070	struct sge_rxq *rxq;
4071	struct sge_txq *txq;
4072#ifdef TCP_OFFLOAD
4073	struct sge_ofld_rxq *ofld_rxq;
4074	struct sge_wrq *ofld_txq;
4075#endif
4076
4077	if (vi->flags & VI_INIT_DONE) {
4078
4079		/* Need to quiesce queues.  */
4080
4081		/* XXX: Only for the first VI? */
4082		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
4083			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
4084
4085		for_each_txq(vi, i, txq) {
4086			quiesce_txq(sc, txq);
4087		}
4088
4089#ifdef TCP_OFFLOAD
4090		for_each_ofld_txq(vi, i, ofld_txq) {
4091			quiesce_wrq(sc, ofld_txq);
4092		}
4093#endif
4094
4095		for_each_rxq(vi, i, rxq) {
4096			quiesce_iq(sc, &rxq->iq);
4097			quiesce_fl(sc, &rxq->fl);
4098		}
4099
4100#ifdef TCP_OFFLOAD
4101		for_each_ofld_rxq(vi, i, ofld_rxq) {
4102			quiesce_iq(sc, &ofld_rxq->iq);
4103			quiesce_fl(sc, &ofld_rxq->fl);
4104		}
4105#endif
4106		free(vi->rss, M_CXGBE);
4107		free(vi->nm_rss, M_CXGBE);
4108	}
4109
4110	t4_teardown_vi_queues(vi);
4111	vi->flags &= ~VI_INIT_DONE;
4112
4113	return (0);
4114}
4115
4116static void
4117quiesce_txq(struct adapter *sc, struct sge_txq *txq)
4118{
4119	struct sge_eq *eq = &txq->eq;
4120	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
4121
4122	(void) sc;	/* unused */
4123
4124#ifdef INVARIANTS
4125	TXQ_LOCK(txq);
4126	MPASS((eq->flags & EQ_ENABLED) == 0);
4127	TXQ_UNLOCK(txq);
4128#endif
4129
4130	/* Wait for the mp_ring to empty. */
4131	while (!mp_ring_is_idle(txq->r)) {
4132		mp_ring_check_drainage(txq->r, 0);
4133		pause("rquiesce", 1);
4134	}
4135
4136	/* Then wait for the hardware to finish. */
4137	while (spg->cidx != htobe16(eq->pidx))
4138		pause("equiesce", 1);
4139
4140	/* Finally, wait for the driver to reclaim all descriptors. */
4141	while (eq->cidx != eq->pidx)
4142		pause("dquiesce", 1);
4143}
4144
4145static void
4146quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
4147{
4148
4149	/* XXXTX */
4150}
4151
4152static void
4153quiesce_iq(struct adapter *sc, struct sge_iq *iq)
4154{
4155	(void) sc;	/* unused */
4156
4157	/* Synchronize with the interrupt handler */
4158	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
4159		pause("iqfree", 1);
4160}
4161
4162static void
4163quiesce_fl(struct adapter *sc, struct sge_fl *fl)
4164{
4165	mtx_lock(&sc->sfl_lock);
4166	FL_LOCK(fl);
4167	fl->flags |= FL_DOOMED;
4168	FL_UNLOCK(fl);
4169	callout_stop(&sc->sfl_callout);
4170	mtx_unlock(&sc->sfl_lock);
4171
4172	KASSERT((fl->flags & FL_STARVING) == 0,
4173	    ("%s: still starving", __func__));
4174}
4175
4176static int
4177t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
4178    driver_intr_t *handler, void *arg, char *name)
4179{
4180	int rc;
4181
4182	irq->rid = rid;
4183	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
4184	    RF_SHAREABLE | RF_ACTIVE);
4185	if (irq->res == NULL) {
4186		device_printf(sc->dev,
4187		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
4188		return (ENOMEM);
4189	}
4190
4191	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
4192	    NULL, handler, arg, &irq->tag);
4193	if (rc != 0) {
4194		device_printf(sc->dev,
4195		    "failed to setup interrupt for rid %d, name %s: %d\n",
4196		    rid, name, rc);
4197	} else if (name)
4198		bus_describe_intr(sc->dev, irq->res, irq->tag, name);
4199
4200	return (rc);
4201}
4202
4203static int
4204t4_free_irq(struct adapter *sc, struct irq *irq)
4205{
4206	if (irq->tag)
4207		bus_teardown_intr(sc->dev, irq->res, irq->tag);
4208	if (irq->res)
4209		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
4210
4211	bzero(irq, sizeof(*irq));
4212
4213	return (0);
4214}
4215
4216static void
4217get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
4218{
4219
4220	regs->version = chip_id(sc) | chip_rev(sc) << 10;
4221	t4_get_regs(sc, buf, regs->len);
4222}
4223
4224#define	A_PL_INDIR_CMD	0x1f8
4225
4226#define	S_PL_AUTOINC	31
4227#define	M_PL_AUTOINC	0x1U
4228#define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
4229#define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
4230
4231#define	S_PL_VFID	20
4232#define	M_PL_VFID	0xffU
4233#define	V_PL_VFID(x)	((x) << S_PL_VFID)
4234#define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
4235
4236#define	S_PL_ADDR	0
4237#define	M_PL_ADDR	0xfffffU
4238#define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
4239#define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
4240
4241#define	A_PL_INDIR_DATA	0x1fc
4242
4243static uint64_t
4244read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
4245{
4246	u32 stats[2];
4247
4248	mtx_assert(&sc->reg_lock, MA_OWNED);
4249	if (sc->flags & IS_VF) {
4250		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
4251		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
4252	} else {
4253		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4254		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4255		    V_PL_ADDR(VF_MPS_REG(reg)));
4256		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
4257		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
4258	}
4259	return (((uint64_t)stats[1]) << 32 | stats[0]);
4260}
4261
4262static void
4263t4_get_vi_stats(struct adapter *sc, unsigned int viid,
4264    struct fw_vi_stats_vf *stats)
4265{
4266
4267#define GET_STAT(name) \
4268	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
4269
4270	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
4271	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
4272	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
4273	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
4274	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
4275	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
4276	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
4277	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
4278	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
4279	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
4280	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
4281	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
4282	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
4283	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
4284	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
4285	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
4286
4287#undef GET_STAT
4288}
4289
4290static void
4291t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
4292{
4293	int reg;
4294
4295	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4296	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4297	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
4298	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
4299	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
4300		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
4301}
4302
4303static void
4304vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
4305{
4306	struct ifnet *ifp = vi->ifp;
4307	struct sge_txq *txq;
4308	int i, drops;
4309	struct fw_vi_stats_vf *s = &vi->stats;
4310	struct timeval tv;
4311	const struct timeval interval = {0, 250000};	/* 250ms */
4312
4313	if (!(vi->flags & VI_INIT_DONE))
4314		return;
4315
4316	getmicrotime(&tv);
4317	timevalsub(&tv, &interval);
4318	if (timevalcmp(&tv, &vi->last_refreshed, <))
4319		return;
4320
4321	mtx_lock(&sc->reg_lock);
4322	t4_get_vi_stats(sc, vi->viid, &vi->stats);
4323
4324	ifp->if_ipackets = s->rx_bcast_frames + s->rx_mcast_frames +
4325	    s->rx_ucast_frames;
4326	ifp->if_ierrors = s->rx_err_frames;
4327	ifp->if_opackets = s->tx_bcast_frames + s->tx_mcast_frames +
4328	    s->tx_ucast_frames + s->tx_offload_frames;
4329	ifp->if_oerrors = s->tx_drop_frames;
4330	ifp->if_ibytes = s->rx_bcast_bytes + s->rx_mcast_bytes +
4331	    s->rx_ucast_bytes;
4332	ifp->if_obytes = s->tx_bcast_bytes + s->tx_mcast_bytes +
4333	    s->tx_ucast_bytes + s->tx_offload_bytes;
4334	ifp->if_imcasts = s->rx_mcast_frames;
4335	ifp->if_omcasts = s->tx_mcast_frames;
4336
4337	drops = 0;
4338	for_each_txq(vi, i, txq)
4339		drops += counter_u64_fetch(txq->r->drops);
4340	ifp->if_snd.ifq_drops = drops;
4341
4342	getmicrotime(&vi->last_refreshed);
4343	mtx_unlock(&sc->reg_lock);
4344}
4345
4346static void
4347cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
4348{
4349	struct vi_info *vi = &pi->vi[0];
4350	struct ifnet *ifp = vi->ifp;
4351	struct sge_txq *txq;
4352	int i, drops;
4353	struct port_stats *s = &pi->stats;
4354	struct timeval tv;
4355	const struct timeval interval = {0, 250000};	/* 250ms */
4356
4357	getmicrotime(&tv);
4358	timevalsub(&tv, &interval);
4359	if (timevalcmp(&tv, &pi->last_refreshed, <))
4360		return;
4361
4362	t4_get_port_stats(sc, pi->tx_chan, s);
4363
4364	ifp->if_opackets = s->tx_frames;
4365	ifp->if_ipackets = s->rx_frames;
4366	ifp->if_obytes = s->tx_octets;
4367	ifp->if_ibytes = s->rx_octets;
4368	ifp->if_omcasts = s->tx_mcast_frames;
4369	ifp->if_imcasts = s->rx_mcast_frames;
4370	ifp->if_iqdrops = s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
4371	    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
4372	    s->rx_trunc3;
4373	for (i = 0; i < sc->chip_params->nchan; i++) {
4374		if (pi->rx_chan_map & (1 << i)) {
4375			uint32_t v;
4376
4377			mtx_lock(&sc->reg_lock);
4378			t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
4379			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
4380			mtx_unlock(&sc->reg_lock);
4381			ifp->if_iqdrops += v;
4382		}
4383	}
4384
4385	drops = s->tx_drop;
4386	for_each_txq(vi, i, txq)
4387		drops += counter_u64_fetch(txq->r->drops);
4388	ifp->if_snd.ifq_drops = drops;
4389
4390	ifp->if_oerrors = s->tx_error_frames;
4391	ifp->if_ierrors = s->rx_jabber + s->rx_runt + s->rx_too_long +
4392	    s->rx_fcs_err + s->rx_len_err;
4393
4394	getmicrotime(&pi->last_refreshed);
4395}
4396
4397static void
4398cxgbe_tick(void *arg)
4399{
4400	struct port_info *pi = arg;
4401	struct adapter *sc = pi->adapter;
4402
4403	PORT_LOCK_ASSERT_OWNED(pi);
4404	cxgbe_refresh_stats(sc, pi);
4405
4406	callout_schedule(&pi->tick, hz);
4407}
4408
4409void
4410vi_tick(void *arg)
4411{
4412	struct vi_info *vi = arg;
4413	struct adapter *sc = vi->pi->adapter;
4414
4415	vi_refresh_stats(sc, vi);
4416
4417	callout_schedule(&vi->tick, hz);
4418}
4419
4420static void
4421cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
4422{
4423	struct ifnet *vlan;
4424
4425	if (arg != ifp || ifp->if_type != IFT_ETHER)
4426		return;
4427
4428	vlan = VLAN_DEVAT(ifp, vid);
4429	VLAN_SETCOOKIE(vlan, ifp);
4430}
4431
4432/*
4433 * Should match fw_caps_config_<foo> enums in t4fw_interface.h
4434 */
4435static char *caps_decoder[] = {
4436	"\20\001IPMI\002NCSI",				/* 0: NBM */
4437	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
4438	"\20\001INGRESS\002EGRESS",			/* 2: switch */
4439	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
4440	    "\006HASHFILTER\007ETHOFLD",
4441	"\20\001TOE",					/* 4: TOE */
4442	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
4443	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
4444	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
4445	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
4446	    "\007T10DIF"
4447	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
4448	"\20\00KEYS",					/* 7: TLS */
4449	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
4450		    "\004PO_INITIATOR\005PO_TARGET",
4451};
4452
4453void
4454t4_sysctls(struct adapter *sc)
4455{
4456	struct sysctl_ctx_list *ctx;
4457	struct sysctl_oid *oid;
4458	struct sysctl_oid_list *children, *c0;
4459	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
4460
4461	ctx = device_get_sysctl_ctx(sc->dev);
4462
4463	/*
4464	 * dev.t4nex.X.
4465	 */
4466	oid = device_get_sysctl_tree(sc->dev);
4467	c0 = children = SYSCTL_CHILDREN(oid);
4468
4469	sc->sc_do_rxcopy = 1;
4470	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
4471	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
4472
4473	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
4474	    sc->params.nports, "# of ports");
4475
4476	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
4477	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
4478	    sysctl_bitfield, "A", "available doorbells");
4479
4480	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
4481	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
4482
4483	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
4484	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
4485	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
4486	    "interrupt holdoff timer values (us)");
4487
4488	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
4489	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
4490	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
4491	    "interrupt holdoff packet counter values");
4492
4493	t4_sge_sysctls(sc, ctx, children);
4494
4495	sc->lro_timeout = 100;
4496	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
4497	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
4498
4499	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "debug_flags", CTLFLAG_RW,
4500	    &sc->debug_flags, 0, "flags to enable runtime debugging");
4501
4502	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
4503	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
4504
4505	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
4506	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
4507
4508	if (sc->flags & IS_VF)
4509		return;
4510
4511	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
4512	    NULL, chip_rev(sc), "chip hardware revision");
4513
4514	if (sc->params.exprom_vers != 0) {
4515		SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "exprom_version",
4516		    CTLFLAG_RD, sc->exprom_version, 0, "expansion ROM version");
4517	}
4518
4519	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
4520	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
4521
4522	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
4523	    sc->cfcsum, "config file checksum");
4524
4525#define SYSCTL_CAP(name, n, text) \
4526	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
4527	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
4528	    sysctl_bitfield, "A", "available " text "capabilities")
4529
4530	SYSCTL_CAP(nbmcaps, 0, "NBM");
4531	SYSCTL_CAP(linkcaps, 1, "link");
4532	SYSCTL_CAP(switchcaps, 2, "switch");
4533	SYSCTL_CAP(niccaps, 3, "NIC");
4534	SYSCTL_CAP(toecaps, 4, "TCP offload");
4535	SYSCTL_CAP(rdmacaps, 5, "RDMA");
4536	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
4537	SYSCTL_CAP(tlscaps, 7, "TLS");
4538	SYSCTL_CAP(fcoecaps, 8, "FCoE");
4539#undef SYSCTL_CAP
4540
4541	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
4542	    NULL, sc->tids.nftids, "number of filters");
4543
4544	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
4545	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
4546	    "chip temperature (in Celsius)");
4547
4548#ifdef SBUF_DRAIN
4549	/*
4550	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
4551	 */
4552	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
4553	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
4554	    "logs and miscellaneous information");
4555	children = SYSCTL_CHILDREN(oid);
4556
4557	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
4558	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4559	    sysctl_cctrl, "A", "congestion control");
4560
4561	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
4562	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4563	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
4564
4565	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
4566	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
4567	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
4568
4569	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
4570	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
4571	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
4572
4573	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
4574	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
4575	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
4576
4577	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
4578	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
4579	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
4580
4581	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
4582	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
4583	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
4584
4585	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
4586	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4587	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
4588	    "A", "CIM logic analyzer");
4589
4590	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
4591	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4592	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
4593
4594	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
4595	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
4596	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
4597
4598	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
4599	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
4600	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
4601
4602	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
4603	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
4604	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
4605
4606	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
4607	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
4608	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
4609
4610	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
4611	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
4612	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
4613
4614	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
4615	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
4616	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
4617
4618	if (chip_id(sc) > CHELSIO_T4) {
4619		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
4620		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
4621		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
4622
4623		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
4624		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
4625		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
4626	}
4627
4628	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
4629	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4630	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
4631
4632	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
4633	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4634	    sysctl_cim_qcfg, "A", "CIM queue configuration");
4635
4636	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
4637	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4638	    sysctl_cpl_stats, "A", "CPL statistics");
4639
4640	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
4641	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4642	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
4643
4644	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
4645	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4646	    sysctl_devlog, "A", "firmware's device log");
4647
4648	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
4649	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4650	    sysctl_fcoe_stats, "A", "FCoE statistics");
4651
4652	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
4653	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4654	    sysctl_hw_sched, "A", "hardware scheduler ");
4655
4656	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
4657	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4658	    sysctl_l2t, "A", "hardware L2 table");
4659
4660	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
4661	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4662	    sysctl_lb_stats, "A", "loopback statistics");
4663
4664	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
4665	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4666	    sysctl_meminfo, "A", "memory regions");
4667
4668	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
4669	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4670	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
4671	    "A", "MPS TCAM entries");
4672
4673	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
4674	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4675	    sysctl_path_mtus, "A", "path MTUs");
4676
4677	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
4678	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4679	    sysctl_pm_stats, "A", "PM statistics");
4680
4681	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
4682	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4683	    sysctl_rdma_stats, "A", "RDMA statistics");
4684
4685	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
4686	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4687	    sysctl_tcp_stats, "A", "TCP statistics");
4688
4689	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
4690	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4691	    sysctl_tids, "A", "TID information");
4692
4693	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
4694	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4695	    sysctl_tp_err_stats, "A", "TP error statistics");
4696
4697	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
4698	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
4699	    "TP logic analyzer event capture mask");
4700
4701	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
4702	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4703	    sysctl_tp_la, "A", "TP logic analyzer");
4704
4705	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
4706	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4707	    sysctl_tx_rate, "A", "Tx rate");
4708
4709	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
4710	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4711	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
4712
4713	if (is_t5(sc)) {
4714		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
4715		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4716		    sysctl_wcwr_stats, "A", "write combined work requests");
4717	}
4718#endif
4719
4720#ifdef TCP_OFFLOAD
4721	if (is_offload(sc)) {
4722		/*
4723		 * dev.t4nex.X.toe.
4724		 */
4725		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
4726		    NULL, "TOE parameters");
4727		children = SYSCTL_CHILDREN(oid);
4728
4729		sc->tt.sndbuf = 256 * 1024;
4730		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
4731		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
4732
4733		sc->tt.ddp = 0;
4734		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
4735		    &sc->tt.ddp, 0, "DDP allowed");
4736
4737		sc->tt.indsz = G_INDICATESIZE(t4_read_reg(sc, A_TP_PARA_REG5));
4738		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "indsz", CTLFLAG_RW,
4739		    &sc->tt.indsz, 0, "DDP max indicate size allowed");
4740
4741		sc->tt.ddp_thres =
4742		    G_RXCOALESCESIZE(t4_read_reg(sc, A_TP_PARA_REG2));
4743		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp_thres", CTLFLAG_RW,
4744		    &sc->tt.ddp_thres, 0, "DDP threshold");
4745
4746		sc->tt.rx_coalesce = 1;
4747		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
4748		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
4749
4750		sc->tt.tx_align = 1;
4751		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
4752		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
4753
4754		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
4755		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
4756		    "TP timer tick (us)");
4757
4758		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
4759		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
4760		    "TCP timestamp tick (us)");
4761
4762		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
4763		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
4764		    "DACK tick (us)");
4765
4766		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
4767		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
4768		    "IU", "DACK timer (us)");
4769
4770		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
4771		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
4772		    sysctl_tp_timer, "LU", "Retransmit min (us)");
4773
4774		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
4775		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
4776		    sysctl_tp_timer, "LU", "Retransmit max (us)");
4777
4778		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
4779		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
4780		    sysctl_tp_timer, "LU", "Persist timer min (us)");
4781
4782		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
4783		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
4784		    sysctl_tp_timer, "LU", "Persist timer max (us)");
4785
4786		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
4787		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
4788		    sysctl_tp_timer, "LU", "Keepidle idle timer (us)");
4789
4790		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_intvl",
4791		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
4792		    sysctl_tp_timer, "LU", "Keepidle interval (us)");
4793
4794		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
4795		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
4796		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
4797
4798		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
4799		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
4800		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
4801	}
4802#endif
4803}
4804
4805void
4806vi_sysctls(struct vi_info *vi)
4807{
4808	struct sysctl_ctx_list *ctx;
4809	struct sysctl_oid *oid;
4810	struct sysctl_oid_list *children;
4811
4812	ctx = device_get_sysctl_ctx(vi->dev);
4813
4814	/*
4815	 * dev.v?(cxgbe|cxl).X.
4816	 */
4817	oid = device_get_sysctl_tree(vi->dev);
4818	children = SYSCTL_CHILDREN(oid);
4819
4820	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
4821	    vi->viid, "VI identifer");
4822	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
4823	    &vi->nrxq, 0, "# of rx queues");
4824	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
4825	    &vi->ntxq, 0, "# of tx queues");
4826	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
4827	    &vi->first_rxq, 0, "index of first rx queue");
4828	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
4829	    &vi->first_txq, 0, "index of first tx queue");
4830
4831	if (IS_MAIN_VI(vi)) {
4832		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
4833		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
4834		    "Reserve queue 0 for non-flowid packets");
4835	}
4836
4837#ifdef TCP_OFFLOAD
4838	if (vi->nofldrxq != 0) {
4839		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
4840		    &vi->nofldrxq, 0,
4841		    "# of rx queues for offloaded TCP connections");
4842		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
4843		    &vi->nofldtxq, 0,
4844		    "# of tx queues for offloaded TCP connections");
4845		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
4846		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
4847		    "index of first TOE rx queue");
4848		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
4849		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
4850		    "index of first TOE tx queue");
4851	}
4852#endif
4853#ifdef DEV_NETMAP
4854	if (vi->nnmrxq != 0) {
4855		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
4856		    &vi->nnmrxq, 0, "# of netmap rx queues");
4857		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
4858		    &vi->nnmtxq, 0, "# of netmap tx queues");
4859		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
4860		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
4861		    "index of first netmap rx queue");
4862		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
4863		    CTLFLAG_RD, &vi->first_nm_txq, 0,
4864		    "index of first netmap tx queue");
4865	}
4866#endif
4867
4868	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
4869	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
4870	    "holdoff timer index");
4871	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
4872	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
4873	    "holdoff packet counter index");
4874
4875	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
4876	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
4877	    "rx queue size");
4878	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
4879	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
4880	    "tx queue size");
4881}
4882
4883static void
4884cxgbe_sysctls(struct port_info *pi)
4885{
4886	struct sysctl_ctx_list *ctx;
4887	struct sysctl_oid *oid;
4888	struct sysctl_oid_list *children, *children2;
4889	struct adapter *sc = pi->adapter;
4890	int i;
4891	char name[16];
4892
4893	ctx = device_get_sysctl_ctx(pi->dev);
4894
4895	/*
4896	 * dev.cxgbe.X.
4897	 */
4898	oid = device_get_sysctl_tree(pi->dev);
4899	children = SYSCTL_CHILDREN(oid);
4900
4901	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
4902	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
4903	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
4904		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
4905		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
4906		    "PHY temperature (in Celsius)");
4907		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
4908		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
4909		    "PHY firmware version");
4910	}
4911
4912	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
4913	    CTLTYPE_STRING | CTLFLAG_RW, pi, PAUSE_TX, sysctl_pause_settings,
4914	    "A", "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
4915
4916	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
4917	    port_top_speed(pi), "max speed (in Gbps)");
4918
4919	if (sc->flags & IS_VF)
4920		return;
4921
4922	/*
4923	 * dev.(cxgbe|cxl).X.tc.
4924	 */
4925	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
4926	    "Tx scheduler traffic classes");
4927	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
4928		struct tx_sched_class *tc = &pi->tc[i];
4929
4930		snprintf(name, sizeof(name), "%d", i);
4931		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
4932		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
4933		    "traffic class"));
4934		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
4935		    &tc->flags, 0, "flags");
4936		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
4937		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
4938#ifdef SBUF_DRAIN
4939		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
4940		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
4941		    sysctl_tc_params, "A", "traffic class parameters");
4942#endif
4943	}
4944
4945	/*
4946	 * dev.cxgbe.X.stats.
4947	 */
4948	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
4949	    NULL, "port statistics");
4950	children = SYSCTL_CHILDREN(oid);
4951	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
4952	    &pi->tx_parse_error, 0,
4953	    "# of tx packets with invalid length or # of segments");
4954
4955#define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
4956	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
4957	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
4958	    sysctl_handle_t4_reg64, "QU", desc)
4959
4960	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
4961	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
4962	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
4963	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
4964	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
4965	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
4966	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
4967	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
4968	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
4969	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
4970	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
4971	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
4972	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
4973	    "# of tx frames in this range",
4974	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
4975	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
4976	    "# of tx frames in this range",
4977	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
4978	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
4979	    "# of tx frames in this range",
4980	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
4981	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
4982	    "# of tx frames in this range",
4983	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
4984	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
4985	    "# of tx frames in this range",
4986	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
4987	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
4988	    "# of tx frames in this range",
4989	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
4990	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
4991	    "# of tx frames in this range",
4992	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
4993	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
4994	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
4995	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
4996	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
4997	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
4998	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
4999	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5000	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5001	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5002	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5003	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5004	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5005	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5006	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5007	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5008	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5009	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5010	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5011	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5012	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5013
5014	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5015	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5016	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5017	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5018	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5019	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5020	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5021	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5022	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5023	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5024	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5025	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5026	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5027	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5028	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5029	    "# of frames received with bad FCS",
5030	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5031	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5032	    "# of frames received with length error",
5033	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5034	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5035	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5036	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5037	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5038	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5039	    "# of rx frames in this range",
5040	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5041	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5042	    "# of rx frames in this range",
5043	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5044	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5045	    "# of rx frames in this range",
5046	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5047	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5048	    "# of rx frames in this range",
5049	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5050	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5051	    "# of rx frames in this range",
5052	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5053	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5054	    "# of rx frames in this range",
5055	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5056	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5057	    "# of rx frames in this range",
5058	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5059	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5060	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5061	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5062	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5063	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5064	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5065	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5066	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5067	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5068	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5069	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5070	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5071	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5072	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5073	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5074	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5075	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5076	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5077
5078#undef SYSCTL_ADD_T4_REG64
5079
5080#define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5081	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5082	    &pi->stats.name, desc)
5083
5084	/* We get these from port_stats and they may be stale by upto 1s */
5085	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5086	    "# drops due to buffer-group 0 overflows");
5087	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5088	    "# drops due to buffer-group 1 overflows");
5089	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5090	    "# drops due to buffer-group 2 overflows");
5091	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5092	    "# drops due to buffer-group 3 overflows");
5093	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5094	    "# of buffer-group 0 truncated packets");
5095	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5096	    "# of buffer-group 1 truncated packets");
5097	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5098	    "# of buffer-group 2 truncated packets");
5099	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5100	    "# of buffer-group 3 truncated packets");
5101
5102#undef SYSCTL_ADD_T4_PORTSTAT
5103}
5104
5105static int
5106sysctl_int_array(SYSCTL_HANDLER_ARGS)
5107{
5108	int rc, *i, space = 0;
5109	struct sbuf sb;
5110
5111	sbuf_new(&sb, NULL, 32, SBUF_AUTOEXTEND);
5112	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
5113		if (space)
5114			sbuf_printf(&sb, " ");
5115		sbuf_printf(&sb, "%d", *i);
5116		space = 1;
5117	}
5118	sbuf_finish(&sb);
5119	rc = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
5120	sbuf_delete(&sb);
5121	return (rc);
5122}
5123
5124static int
5125sysctl_bitfield(SYSCTL_HANDLER_ARGS)
5126{
5127	int rc;
5128	struct sbuf *sb;
5129
5130	rc = sysctl_wire_old_buffer(req, 0);
5131	if (rc != 0)
5132		return(rc);
5133
5134	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5135	if (sb == NULL)
5136		return (ENOMEM);
5137
5138	sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
5139	rc = sbuf_finish(sb);
5140	sbuf_delete(sb);
5141
5142	return (rc);
5143}
5144
5145static int
5146sysctl_btphy(SYSCTL_HANDLER_ARGS)
5147{
5148	struct port_info *pi = arg1;
5149	int op = arg2;
5150	struct adapter *sc = pi->adapter;
5151	u_int v;
5152	int rc;
5153
5154	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
5155	if (rc)
5156		return (rc);
5157	/* XXX: magic numbers */
5158	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
5159	    &v);
5160	end_synchronized_op(sc, 0);
5161	if (rc)
5162		return (rc);
5163	if (op == 0)
5164		v /= 256;
5165
5166	rc = sysctl_handle_int(oidp, &v, 0, req);
5167	return (rc);
5168}
5169
5170static int
5171sysctl_noflowq(SYSCTL_HANDLER_ARGS)
5172{
5173	struct vi_info *vi = arg1;
5174	int rc, val;
5175
5176	val = vi->rsrv_noflowq;
5177	rc = sysctl_handle_int(oidp, &val, 0, req);
5178	if (rc != 0 || req->newptr == NULL)
5179		return (rc);
5180
5181	if ((val >= 1) && (vi->ntxq > 1))
5182		vi->rsrv_noflowq = 1;
5183	else
5184		vi->rsrv_noflowq = 0;
5185
5186	return (rc);
5187}
5188
5189static int
5190sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
5191{
5192	struct vi_info *vi = arg1;
5193	struct adapter *sc = vi->pi->adapter;
5194	int idx, rc, i;
5195	struct sge_rxq *rxq;
5196#ifdef TCP_OFFLOAD
5197	struct sge_ofld_rxq *ofld_rxq;
5198#endif
5199	uint8_t v;
5200
5201	idx = vi->tmr_idx;
5202
5203	rc = sysctl_handle_int(oidp, &idx, 0, req);
5204	if (rc != 0 || req->newptr == NULL)
5205		return (rc);
5206
5207	if (idx < 0 || idx >= SGE_NTIMERS)
5208		return (EINVAL);
5209
5210	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5211	    "t4tmr");
5212	if (rc)
5213		return (rc);
5214
5215	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
5216	for_each_rxq(vi, i, rxq) {
5217#ifdef atomic_store_rel_8
5218		atomic_store_rel_8(&rxq->iq.intr_params, v);
5219#else
5220		rxq->iq.intr_params = v;
5221#endif
5222	}
5223#ifdef TCP_OFFLOAD
5224	for_each_ofld_rxq(vi, i, ofld_rxq) {
5225#ifdef atomic_store_rel_8
5226		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
5227#else
5228		ofld_rxq->iq.intr_params = v;
5229#endif
5230	}
5231#endif
5232	vi->tmr_idx = idx;
5233
5234	end_synchronized_op(sc, LOCK_HELD);
5235	return (0);
5236}
5237
5238static int
5239sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
5240{
5241	struct vi_info *vi = arg1;
5242	struct adapter *sc = vi->pi->adapter;
5243	int idx, rc;
5244
5245	idx = vi->pktc_idx;
5246
5247	rc = sysctl_handle_int(oidp, &idx, 0, req);
5248	if (rc != 0 || req->newptr == NULL)
5249		return (rc);
5250
5251	if (idx < -1 || idx >= SGE_NCOUNTERS)
5252		return (EINVAL);
5253
5254	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5255	    "t4pktc");
5256	if (rc)
5257		return (rc);
5258
5259	if (vi->flags & VI_INIT_DONE)
5260		rc = EBUSY; /* cannot be changed once the queues are created */
5261	else
5262		vi->pktc_idx = idx;
5263
5264	end_synchronized_op(sc, LOCK_HELD);
5265	return (rc);
5266}
5267
5268static int
5269sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
5270{
5271	struct vi_info *vi = arg1;
5272	struct adapter *sc = vi->pi->adapter;
5273	int qsize, rc;
5274
5275	qsize = vi->qsize_rxq;
5276
5277	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5278	if (rc != 0 || req->newptr == NULL)
5279		return (rc);
5280
5281	if (qsize < 128 || (qsize & 7))
5282		return (EINVAL);
5283
5284	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5285	    "t4rxqs");
5286	if (rc)
5287		return (rc);
5288
5289	if (vi->flags & VI_INIT_DONE)
5290		rc = EBUSY; /* cannot be changed once the queues are created */
5291	else
5292		vi->qsize_rxq = qsize;
5293
5294	end_synchronized_op(sc, LOCK_HELD);
5295	return (rc);
5296}
5297
5298static int
5299sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
5300{
5301	struct vi_info *vi = arg1;
5302	struct adapter *sc = vi->pi->adapter;
5303	int qsize, rc;
5304
5305	qsize = vi->qsize_txq;
5306
5307	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5308	if (rc != 0 || req->newptr == NULL)
5309		return (rc);
5310
5311	if (qsize < 128 || qsize > 65536)
5312		return (EINVAL);
5313
5314	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5315	    "t4txqs");
5316	if (rc)
5317		return (rc);
5318
5319	if (vi->flags & VI_INIT_DONE)
5320		rc = EBUSY; /* cannot be changed once the queues are created */
5321	else
5322		vi->qsize_txq = qsize;
5323
5324	end_synchronized_op(sc, LOCK_HELD);
5325	return (rc);
5326}
5327
5328static int
5329sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
5330{
5331	struct port_info *pi = arg1;
5332	struct adapter *sc = pi->adapter;
5333	struct link_config *lc = &pi->link_cfg;
5334	int rc;
5335
5336	if (req->newptr == NULL) {
5337		struct sbuf *sb;
5338		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
5339
5340		rc = sysctl_wire_old_buffer(req, 0);
5341		if (rc != 0)
5342			return(rc);
5343
5344		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5345		if (sb == NULL)
5346			return (ENOMEM);
5347
5348		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
5349		rc = sbuf_finish(sb);
5350		sbuf_delete(sb);
5351	} else {
5352		char s[2];
5353		int n;
5354
5355		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
5356		s[1] = 0;
5357
5358		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5359		if (rc != 0)
5360			return(rc);
5361
5362		if (s[1] != 0)
5363			return (EINVAL);
5364		if (s[0] < '0' || s[0] > '9')
5365			return (EINVAL);	/* not a number */
5366		n = s[0] - '0';
5367		if (n & ~(PAUSE_TX | PAUSE_RX))
5368			return (EINVAL);	/* some other bit is set too */
5369
5370		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5371		    "t4PAUSE");
5372		if (rc)
5373			return (rc);
5374		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
5375			int link_ok = lc->link_ok;
5376
5377			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
5378			lc->requested_fc |= n;
5379			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5380			lc->link_ok = link_ok;	/* restore */
5381		}
5382		end_synchronized_op(sc, 0);
5383	}
5384
5385	return (rc);
5386}
5387
5388static int
5389sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
5390{
5391	struct adapter *sc = arg1;
5392	int reg = arg2;
5393	uint64_t val;
5394
5395	val = t4_read_reg64(sc, reg);
5396
5397	return (sysctl_handle_64(oidp, &val, 0, req));
5398}
5399
5400static int
5401sysctl_temperature(SYSCTL_HANDLER_ARGS)
5402{
5403	struct adapter *sc = arg1;
5404	int rc, t;
5405	uint32_t param, val;
5406
5407	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
5408	if (rc)
5409		return (rc);
5410	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5411	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
5412	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
5413	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
5414	end_synchronized_op(sc, 0);
5415	if (rc)
5416		return (rc);
5417
5418	/* unknown is returned as 0 but we display -1 in that case */
5419	t = val == 0 ? -1 : val;
5420
5421	rc = sysctl_handle_int(oidp, &t, 0, req);
5422	return (rc);
5423}
5424
5425#ifdef SBUF_DRAIN
5426static int
5427sysctl_cctrl(SYSCTL_HANDLER_ARGS)
5428{
5429	struct adapter *sc = arg1;
5430	struct sbuf *sb;
5431	int rc, i;
5432	uint16_t incr[NMTUS][NCCTRL_WIN];
5433	static const char *dec_fac[] = {
5434		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
5435		"0.9375"
5436	};
5437
5438	rc = sysctl_wire_old_buffer(req, 0);
5439	if (rc != 0)
5440		return (rc);
5441
5442	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5443	if (sb == NULL)
5444		return (ENOMEM);
5445
5446	t4_read_cong_tbl(sc, incr);
5447
5448	for (i = 0; i < NCCTRL_WIN; ++i) {
5449		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
5450		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
5451		    incr[5][i], incr[6][i], incr[7][i]);
5452		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
5453		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
5454		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
5455		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
5456	}
5457
5458	rc = sbuf_finish(sb);
5459	sbuf_delete(sb);
5460
5461	return (rc);
5462}
5463
5464static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
5465	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
5466	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
5467	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
5468};
5469
5470static int
5471sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
5472{
5473	struct adapter *sc = arg1;
5474	struct sbuf *sb;
5475	int rc, i, n, qid = arg2;
5476	uint32_t *buf, *p;
5477	char *qtype;
5478	u_int cim_num_obq = sc->chip_params->cim_num_obq;
5479
5480	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
5481	    ("%s: bad qid %d\n", __func__, qid));
5482
5483	if (qid < CIM_NUM_IBQ) {
5484		/* inbound queue */
5485		qtype = "IBQ";
5486		n = 4 * CIM_IBQ_SIZE;
5487		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5488		rc = t4_read_cim_ibq(sc, qid, buf, n);
5489	} else {
5490		/* outbound queue */
5491		qtype = "OBQ";
5492		qid -= CIM_NUM_IBQ;
5493		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
5494		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5495		rc = t4_read_cim_obq(sc, qid, buf, n);
5496	}
5497
5498	if (rc < 0) {
5499		rc = -rc;
5500		goto done;
5501	}
5502	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
5503
5504	rc = sysctl_wire_old_buffer(req, 0);
5505	if (rc != 0)
5506		goto done;
5507
5508	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
5509	if (sb == NULL) {
5510		rc = ENOMEM;
5511		goto done;
5512	}
5513
5514	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
5515	for (i = 0, p = buf; i < n; i += 16, p += 4)
5516		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
5517		    p[2], p[3]);
5518
5519	rc = sbuf_finish(sb);
5520	sbuf_delete(sb);
5521done:
5522	free(buf, M_CXGBE);
5523	return (rc);
5524}
5525
5526static int
5527sysctl_cim_la(SYSCTL_HANDLER_ARGS)
5528{
5529	struct adapter *sc = arg1;
5530	u_int cfg;
5531	struct sbuf *sb;
5532	uint32_t *buf, *p;
5533	int rc;
5534
5535	MPASS(chip_id(sc) <= CHELSIO_T5);
5536
5537	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
5538	if (rc != 0)
5539		return (rc);
5540
5541	rc = sysctl_wire_old_buffer(req, 0);
5542	if (rc != 0)
5543		return (rc);
5544
5545	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5546	if (sb == NULL)
5547		return (ENOMEM);
5548
5549	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
5550	    M_ZERO | M_WAITOK);
5551
5552	rc = -t4_cim_read_la(sc, buf, NULL);
5553	if (rc != 0)
5554		goto done;
5555
5556	sbuf_printf(sb, "Status   Data      PC%s",
5557	    cfg & F_UPDBGLACAPTPCONLY ? "" :
5558	    "     LS0Stat  LS0Addr             LS0Data");
5559
5560	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
5561		if (cfg & F_UPDBGLACAPTPCONLY) {
5562			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
5563			    p[6], p[7]);
5564			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
5565			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
5566			    p[4] & 0xff, p[5] >> 8);
5567			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
5568			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
5569			    p[1] & 0xf, p[2] >> 4);
5570		} else {
5571			sbuf_printf(sb,
5572			    "\n  %02x   %x%07x %x%07x %08x %08x "
5573			    "%08x%08x%08x%08x",
5574			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
5575			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
5576			    p[6], p[7]);
5577		}
5578	}
5579
5580	rc = sbuf_finish(sb);
5581	sbuf_delete(sb);
5582done:
5583	free(buf, M_CXGBE);
5584	return (rc);
5585}
5586
5587static int
5588sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
5589{
5590	struct adapter *sc = arg1;
5591	u_int cfg;
5592	struct sbuf *sb;
5593	uint32_t *buf, *p;
5594	int rc;
5595
5596	MPASS(chip_id(sc) > CHELSIO_T5);
5597
5598	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
5599	if (rc != 0)
5600		return (rc);
5601
5602	rc = sysctl_wire_old_buffer(req, 0);
5603	if (rc != 0)
5604		return (rc);
5605
5606	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5607	if (sb == NULL)
5608		return (ENOMEM);
5609
5610	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
5611	    M_ZERO | M_WAITOK);
5612
5613	rc = -t4_cim_read_la(sc, buf, NULL);
5614	if (rc != 0)
5615		goto done;
5616
5617	sbuf_printf(sb, "Status   Inst    Data      PC%s",
5618	    cfg & F_UPDBGLACAPTPCONLY ? "" :
5619	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
5620
5621	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
5622		if (cfg & F_UPDBGLACAPTPCONLY) {
5623			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
5624			    p[3] & 0xff, p[2], p[1], p[0]);
5625			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
5626			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
5627			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
5628			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
5629			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
5630			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
5631			    p[6] >> 16);
5632		} else {
5633			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
5634			    "%08x %08x %08x %08x %08x %08x",
5635			    (p[9] >> 16) & 0xff,
5636			    p[9] & 0xffff, p[8] >> 16,
5637			    p[8] & 0xffff, p[7] >> 16,
5638			    p[7] & 0xffff, p[6] >> 16,
5639			    p[2], p[1], p[0], p[5], p[4], p[3]);
5640		}
5641	}
5642
5643	rc = sbuf_finish(sb);
5644	sbuf_delete(sb);
5645done:
5646	free(buf, M_CXGBE);
5647	return (rc);
5648}
5649
5650static int
5651sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
5652{
5653	struct adapter *sc = arg1;
5654	u_int i;
5655	struct sbuf *sb;
5656	uint32_t *buf, *p;
5657	int rc;
5658
5659	rc = sysctl_wire_old_buffer(req, 0);
5660	if (rc != 0)
5661		return (rc);
5662
5663	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5664	if (sb == NULL)
5665		return (ENOMEM);
5666
5667	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
5668	    M_ZERO | M_WAITOK);
5669
5670	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
5671	p = buf;
5672
5673	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
5674		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
5675		    p[1], p[0]);
5676	}
5677
5678	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
5679	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
5680		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
5681		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
5682		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
5683		    (p[1] >> 2) | ((p[2] & 3) << 30),
5684		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
5685		    p[0] & 1);
5686	}
5687
5688	rc = sbuf_finish(sb);
5689	sbuf_delete(sb);
5690	free(buf, M_CXGBE);
5691	return (rc);
5692}
5693
5694static int
5695sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
5696{
5697	struct adapter *sc = arg1;
5698	u_int i;
5699	struct sbuf *sb;
5700	uint32_t *buf, *p;
5701	int rc;
5702
5703	rc = sysctl_wire_old_buffer(req, 0);
5704	if (rc != 0)
5705		return (rc);
5706
5707	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5708	if (sb == NULL)
5709		return (ENOMEM);
5710
5711	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
5712	    M_ZERO | M_WAITOK);
5713
5714	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
5715	p = buf;
5716
5717	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
5718	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
5719		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
5720		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
5721		    p[4], p[3], p[2], p[1], p[0]);
5722	}
5723
5724	sbuf_printf(sb, "\n\nCntl ID               Data");
5725	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
5726		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
5727		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
5728	}
5729
5730	rc = sbuf_finish(sb);
5731	sbuf_delete(sb);
5732	free(buf, M_CXGBE);
5733	return (rc);
5734}
5735
5736static int
5737sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
5738{
5739	struct adapter *sc = arg1;
5740	struct sbuf *sb;
5741	int rc, i;
5742	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
5743	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
5744	uint16_t thres[CIM_NUM_IBQ];
5745	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
5746	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
5747	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
5748
5749	cim_num_obq = sc->chip_params->cim_num_obq;
5750	if (is_t4(sc)) {
5751		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
5752		obq_rdaddr = A_UP_OBQ_0_REALADDR;
5753	} else {
5754		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
5755		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
5756	}
5757	nq = CIM_NUM_IBQ + cim_num_obq;
5758
5759	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
5760	if (rc == 0)
5761		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
5762	if (rc != 0)
5763		return (rc);
5764
5765	t4_read_cimq_cfg(sc, base, size, thres);
5766
5767	rc = sysctl_wire_old_buffer(req, 0);
5768	if (rc != 0)
5769		return (rc);
5770
5771	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
5772	if (sb == NULL)
5773		return (ENOMEM);
5774
5775	sbuf_printf(sb, "Queue  Base  Size Thres RdPtr WrPtr  SOP  EOP Avail");
5776
5777	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
5778		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
5779		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
5780		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
5781		    G_QUEREMFLITS(p[2]) * 16);
5782	for ( ; i < nq; i++, p += 4, wr += 2)
5783		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
5784		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
5785		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
5786		    G_QUEREMFLITS(p[2]) * 16);
5787
5788	rc = sbuf_finish(sb);
5789	sbuf_delete(sb);
5790
5791	return (rc);
5792}
5793
5794static int
5795sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
5796{
5797	struct adapter *sc = arg1;
5798	struct sbuf *sb;
5799	int rc;
5800	struct tp_cpl_stats stats;
5801
5802	rc = sysctl_wire_old_buffer(req, 0);
5803	if (rc != 0)
5804		return (rc);
5805
5806	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
5807	if (sb == NULL)
5808		return (ENOMEM);
5809
5810	mtx_lock(&sc->reg_lock);
5811	t4_tp_get_cpl_stats(sc, &stats);
5812	mtx_unlock(&sc->reg_lock);
5813
5814	if (sc->chip_params->nchan > 2) {
5815		sbuf_printf(sb, "                 channel 0  channel 1"
5816		    "  channel 2  channel 3");
5817		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
5818		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
5819		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
5820		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
5821	} else {
5822		sbuf_printf(sb, "                 channel 0  channel 1");
5823		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
5824		    stats.req[0], stats.req[1]);
5825		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
5826		    stats.rsp[0], stats.rsp[1]);
5827	}
5828
5829	rc = sbuf_finish(sb);
5830	sbuf_delete(sb);
5831
5832	return (rc);
5833}
5834
5835static int
5836sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
5837{
5838	struct adapter *sc = arg1;
5839	struct sbuf *sb;
5840	int rc;
5841	struct tp_usm_stats stats;
5842
5843	rc = sysctl_wire_old_buffer(req, 0);
5844	if (rc != 0)
5845		return(rc);
5846
5847	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
5848	if (sb == NULL)
5849		return (ENOMEM);
5850
5851	t4_get_usm_stats(sc, &stats);
5852
5853	sbuf_printf(sb, "Frames: %u\n", stats.frames);
5854	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
5855	sbuf_printf(sb, "Drops:  %u", stats.drops);
5856
5857	rc = sbuf_finish(sb);
5858	sbuf_delete(sb);
5859
5860	return (rc);
5861}
5862
5863static const char * const devlog_level_strings[] = {
5864	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
5865	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
5866	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
5867	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
5868	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
5869	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
5870};
5871
5872static const char * const devlog_facility_strings[] = {
5873	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
5874	[FW_DEVLOG_FACILITY_CF]		= "CF",
5875	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
5876	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
5877	[FW_DEVLOG_FACILITY_RES]	= "RES",
5878	[FW_DEVLOG_FACILITY_HW]		= "HW",
5879	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
5880	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
5881	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
5882	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
5883	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
5884	[FW_DEVLOG_FACILITY_VI]		= "VI",
5885	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
5886	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
5887	[FW_DEVLOG_FACILITY_TM]		= "TM",
5888	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
5889	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
5890	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
5891	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
5892	[FW_DEVLOG_FACILITY_RI]		= "RI",
5893	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
5894	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
5895	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
5896	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
5897	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
5898};
5899
5900static int
5901sysctl_devlog(SYSCTL_HANDLER_ARGS)
5902{
5903	struct adapter *sc = arg1;
5904	struct devlog_params *dparams = &sc->params.devlog;
5905	struct fw_devlog_e *buf, *e;
5906	int i, j, rc, nentries, first = 0;
5907	struct sbuf *sb;
5908	uint64_t ftstamp = UINT64_MAX;
5909
5910	if (dparams->addr == 0)
5911		return (ENXIO);
5912
5913	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
5914	if (buf == NULL)
5915		return (ENOMEM);
5916
5917	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
5918	if (rc != 0)
5919		goto done;
5920
5921	nentries = dparams->size / sizeof(struct fw_devlog_e);
5922	for (i = 0; i < nentries; i++) {
5923		e = &buf[i];
5924
5925		if (e->timestamp == 0)
5926			break;	/* end */
5927
5928		e->timestamp = be64toh(e->timestamp);
5929		e->seqno = be32toh(e->seqno);
5930		for (j = 0; j < 8; j++)
5931			e->params[j] = be32toh(e->params[j]);
5932
5933		if (e->timestamp < ftstamp) {
5934			ftstamp = e->timestamp;
5935			first = i;
5936		}
5937	}
5938
5939	if (buf[first].timestamp == 0)
5940		goto done;	/* nothing in the log */
5941
5942	rc = sysctl_wire_old_buffer(req, 0);
5943	if (rc != 0)
5944		goto done;
5945
5946	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5947	if (sb == NULL) {
5948		rc = ENOMEM;
5949		goto done;
5950	}
5951	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
5952	    "Seq#", "Tstamp", "Level", "Facility", "Message");
5953
5954	i = first;
5955	do {
5956		e = &buf[i];
5957		if (e->timestamp == 0)
5958			break;	/* end */
5959
5960		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
5961		    e->seqno, e->timestamp,
5962		    (e->level < nitems(devlog_level_strings) ?
5963			devlog_level_strings[e->level] : "UNKNOWN"),
5964		    (e->facility < nitems(devlog_facility_strings) ?
5965			devlog_facility_strings[e->facility] : "UNKNOWN"));
5966		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
5967		    e->params[2], e->params[3], e->params[4],
5968		    e->params[5], e->params[6], e->params[7]);
5969
5970		if (++i == nentries)
5971			i = 0;
5972	} while (i != first);
5973
5974	rc = sbuf_finish(sb);
5975	sbuf_delete(sb);
5976done:
5977	free(buf, M_CXGBE);
5978	return (rc);
5979}
5980
5981static int
5982sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
5983{
5984	struct adapter *sc = arg1;
5985	struct sbuf *sb;
5986	int rc;
5987	struct tp_fcoe_stats stats[MAX_NCHAN];
5988	int i, nchan = sc->chip_params->nchan;
5989
5990	rc = sysctl_wire_old_buffer(req, 0);
5991	if (rc != 0)
5992		return (rc);
5993
5994	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
5995	if (sb == NULL)
5996		return (ENOMEM);
5997
5998	for (i = 0; i < nchan; i++)
5999		t4_get_fcoe_stats(sc, i, &stats[i]);
6000
6001	if (nchan > 2) {
6002		sbuf_printf(sb, "                   channel 0        channel 1"
6003		    "        channel 2        channel 3");
6004		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
6005		    stats[0].octets_ddp, stats[1].octets_ddp,
6006		    stats[2].octets_ddp, stats[3].octets_ddp);
6007		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
6008		    stats[0].frames_ddp, stats[1].frames_ddp,
6009		    stats[2].frames_ddp, stats[3].frames_ddp);
6010		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
6011		    stats[0].frames_drop, stats[1].frames_drop,
6012		    stats[2].frames_drop, stats[3].frames_drop);
6013	} else {
6014		sbuf_printf(sb, "                   channel 0        channel 1");
6015		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
6016		    stats[0].octets_ddp, stats[1].octets_ddp);
6017		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
6018		    stats[0].frames_ddp, stats[1].frames_ddp);
6019		sbuf_printf(sb, "\nframesDrop: %16u %16u",
6020		    stats[0].frames_drop, stats[1].frames_drop);
6021	}
6022
6023	rc = sbuf_finish(sb);
6024	sbuf_delete(sb);
6025
6026	return (rc);
6027}
6028
6029static int
6030sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
6031{
6032	struct adapter *sc = arg1;
6033	struct sbuf *sb;
6034	int rc, i;
6035	unsigned int map, kbps, ipg, mode;
6036	unsigned int pace_tab[NTX_SCHED];
6037
6038	rc = sysctl_wire_old_buffer(req, 0);
6039	if (rc != 0)
6040		return (rc);
6041
6042	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6043	if (sb == NULL)
6044		return (ENOMEM);
6045
6046	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
6047	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
6048	t4_read_pace_tbl(sc, pace_tab);
6049
6050	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
6051	    "Class IPG (0.1 ns)   Flow IPG (us)");
6052
6053	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
6054		t4_get_tx_sched(sc, i, &kbps, &ipg);
6055		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
6056		    (mode & (1 << i)) ? "flow" : "class", map & 3);
6057		if (kbps)
6058			sbuf_printf(sb, "%9u     ", kbps);
6059		else
6060			sbuf_printf(sb, " disabled     ");
6061
6062		if (ipg)
6063			sbuf_printf(sb, "%13u        ", ipg);
6064		else
6065			sbuf_printf(sb, "     disabled        ");
6066
6067		if (pace_tab[i])
6068			sbuf_printf(sb, "%10u", pace_tab[i]);
6069		else
6070			sbuf_printf(sb, "  disabled");
6071	}
6072
6073	rc = sbuf_finish(sb);
6074	sbuf_delete(sb);
6075
6076	return (rc);
6077}
6078
6079static int
6080sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
6081{
6082	struct adapter *sc = arg1;
6083	struct sbuf *sb;
6084	int rc, i, j;
6085	uint64_t *p0, *p1;
6086	struct lb_port_stats s[2];
6087	static const char *stat_name[] = {
6088		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
6089		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
6090		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
6091		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
6092		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
6093		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
6094		"BG2FramesTrunc:", "BG3FramesTrunc:"
6095	};
6096
6097	rc = sysctl_wire_old_buffer(req, 0);
6098	if (rc != 0)
6099		return (rc);
6100
6101	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6102	if (sb == NULL)
6103		return (ENOMEM);
6104
6105	memset(s, 0, sizeof(s));
6106
6107	for (i = 0; i < sc->chip_params->nchan; i += 2) {
6108		t4_get_lb_stats(sc, i, &s[0]);
6109		t4_get_lb_stats(sc, i + 1, &s[1]);
6110
6111		p0 = &s[0].octets;
6112		p1 = &s[1].octets;
6113		sbuf_printf(sb, "%s                       Loopback %u"
6114		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
6115
6116		for (j = 0; j < nitems(stat_name); j++)
6117			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
6118				   *p0++, *p1++);
6119	}
6120
6121	rc = sbuf_finish(sb);
6122	sbuf_delete(sb);
6123
6124	return (rc);
6125}
6126
6127static int
6128sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
6129{
6130	int rc = 0;
6131	struct port_info *pi = arg1;
6132	struct sbuf *sb;
6133
6134	rc = sysctl_wire_old_buffer(req, 0);
6135	if (rc != 0)
6136		return(rc);
6137	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
6138	if (sb == NULL)
6139		return (ENOMEM);
6140
6141	if (pi->linkdnrc < 0)
6142		sbuf_printf(sb, "n/a");
6143	else
6144		sbuf_printf(sb, "%s", t4_link_down_rc_str(pi->linkdnrc));
6145
6146	rc = sbuf_finish(sb);
6147	sbuf_delete(sb);
6148
6149	return (rc);
6150}
6151
6152struct mem_desc {
6153	unsigned int base;
6154	unsigned int limit;
6155	unsigned int idx;
6156};
6157
6158static int
6159mem_desc_cmp(const void *a, const void *b)
6160{
6161	return ((const struct mem_desc *)a)->base -
6162	       ((const struct mem_desc *)b)->base;
6163}
6164
6165static void
6166mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
6167    unsigned int to)
6168{
6169	unsigned int size;
6170
6171	if (from == to)
6172		return;
6173
6174	size = to - from + 1;
6175	if (size == 0)
6176		return;
6177
6178	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
6179	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
6180}
6181
6182static int
6183sysctl_meminfo(SYSCTL_HANDLER_ARGS)
6184{
6185	struct adapter *sc = arg1;
6186	struct sbuf *sb;
6187	int rc, i, n;
6188	uint32_t lo, hi, used, alloc;
6189	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
6190	static const char *region[] = {
6191		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
6192		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
6193		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
6194		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
6195		"RQUDP region:", "PBL region:", "TXPBL region:",
6196		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
6197		"On-chip queues:"
6198	};
6199	struct mem_desc avail[4];
6200	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
6201	struct mem_desc *md = mem;
6202
6203	rc = sysctl_wire_old_buffer(req, 0);
6204	if (rc != 0)
6205		return (rc);
6206
6207	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6208	if (sb == NULL)
6209		return (ENOMEM);
6210
6211	for (i = 0; i < nitems(mem); i++) {
6212		mem[i].limit = 0;
6213		mem[i].idx = i;
6214	}
6215
6216	/* Find and sort the populated memory ranges */
6217	i = 0;
6218	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
6219	if (lo & F_EDRAM0_ENABLE) {
6220		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
6221		avail[i].base = G_EDRAM0_BASE(hi) << 20;
6222		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
6223		avail[i].idx = 0;
6224		i++;
6225	}
6226	if (lo & F_EDRAM1_ENABLE) {
6227		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
6228		avail[i].base = G_EDRAM1_BASE(hi) << 20;
6229		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
6230		avail[i].idx = 1;
6231		i++;
6232	}
6233	if (lo & F_EXT_MEM_ENABLE) {
6234		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
6235		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
6236		avail[i].limit = avail[i].base +
6237		    (G_EXT_MEM_SIZE(hi) << 20);
6238		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
6239		i++;
6240	}
6241	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
6242		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
6243		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
6244		avail[i].limit = avail[i].base +
6245		    (G_EXT_MEM1_SIZE(hi) << 20);
6246		avail[i].idx = 4;
6247		i++;
6248	}
6249	if (!i)                                    /* no memory available */
6250		return 0;
6251	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
6252
6253	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
6254	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
6255	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
6256	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
6257	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
6258	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
6259	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
6260	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
6261	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
6262
6263	/* the next few have explicit upper bounds */
6264	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
6265	md->limit = md->base - 1 +
6266		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
6267		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
6268	md++;
6269
6270	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
6271	md->limit = md->base - 1 +
6272		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
6273		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
6274	md++;
6275
6276	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
6277		if (chip_id(sc) <= CHELSIO_T5)
6278			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
6279		else
6280			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
6281		md->limit = 0;
6282	} else {
6283		md->base = 0;
6284		md->idx = nitems(region);  /* hide it */
6285	}
6286	md++;
6287
6288#define ulp_region(reg) \
6289	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
6290	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
6291
6292	ulp_region(RX_ISCSI);
6293	ulp_region(RX_TDDP);
6294	ulp_region(TX_TPT);
6295	ulp_region(RX_STAG);
6296	ulp_region(RX_RQ);
6297	ulp_region(RX_RQUDP);
6298	ulp_region(RX_PBL);
6299	ulp_region(TX_PBL);
6300#undef ulp_region
6301
6302	md->base = 0;
6303	md->idx = nitems(region);
6304	if (!is_t4(sc)) {
6305		uint32_t size = 0;
6306		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
6307		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
6308
6309		if (is_t5(sc)) {
6310			if (sge_ctrl & F_VFIFO_ENABLE)
6311				size = G_DBVFIFO_SIZE(fifo_size);
6312		} else
6313			size = G_T6_DBVFIFO_SIZE(fifo_size);
6314
6315		if (size) {
6316			md->base = G_BASEADDR(t4_read_reg(sc,
6317			    A_SGE_DBVFIFO_BADDR));
6318			md->limit = md->base + (size << 2) - 1;
6319		}
6320	}
6321	md++;
6322
6323	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
6324	md->limit = 0;
6325	md++;
6326	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
6327	md->limit = 0;
6328	md++;
6329
6330	md->base = sc->vres.ocq.start;
6331	if (sc->vres.ocq.size)
6332		md->limit = md->base + sc->vres.ocq.size - 1;
6333	else
6334		md->idx = nitems(region);  /* hide it */
6335	md++;
6336
6337	/* add any address-space holes, there can be up to 3 */
6338	for (n = 0; n < i - 1; n++)
6339		if (avail[n].limit < avail[n + 1].base)
6340			(md++)->base = avail[n].limit;
6341	if (avail[n].limit)
6342		(md++)->base = avail[n].limit;
6343
6344	n = md - mem;
6345	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
6346
6347	for (lo = 0; lo < i; lo++)
6348		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
6349				avail[lo].limit - 1);
6350
6351	sbuf_printf(sb, "\n");
6352	for (i = 0; i < n; i++) {
6353		if (mem[i].idx >= nitems(region))
6354			continue;                        /* skip holes */
6355		if (!mem[i].limit)
6356			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
6357		mem_region_show(sb, region[mem[i].idx], mem[i].base,
6358				mem[i].limit);
6359	}
6360
6361	sbuf_printf(sb, "\n");
6362	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
6363	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
6364	mem_region_show(sb, "uP RAM:", lo, hi);
6365
6366	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
6367	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
6368	mem_region_show(sb, "uP Extmem2:", lo, hi);
6369
6370	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
6371	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
6372		   G_PMRXMAXPAGE(lo),
6373		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
6374		   (lo & F_PMRXNUMCHN) ? 2 : 1);
6375
6376	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
6377	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
6378	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
6379		   G_PMTXMAXPAGE(lo),
6380		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
6381		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
6382	sbuf_printf(sb, "%u p-structs\n",
6383		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
6384
6385	for (i = 0; i < 4; i++) {
6386		if (chip_id(sc) > CHELSIO_T5)
6387			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
6388		else
6389			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
6390		if (is_t5(sc)) {
6391			used = G_T5_USED(lo);
6392			alloc = G_T5_ALLOC(lo);
6393		} else {
6394			used = G_USED(lo);
6395			alloc = G_ALLOC(lo);
6396		}
6397		/* For T6 these are MAC buffer groups */
6398		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
6399		    i, used, alloc);
6400	}
6401	for (i = 0; i < sc->chip_params->nchan; i++) {
6402		if (chip_id(sc) > CHELSIO_T5)
6403			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
6404		else
6405			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
6406		if (is_t5(sc)) {
6407			used = G_T5_USED(lo);
6408			alloc = G_T5_ALLOC(lo);
6409		} else {
6410			used = G_USED(lo);
6411			alloc = G_ALLOC(lo);
6412		}
6413		/* For T6 these are MAC buffer groups */
6414		sbuf_printf(sb,
6415		    "\nLoopback %d using %u pages out of %u allocated",
6416		    i, used, alloc);
6417	}
6418
6419	rc = sbuf_finish(sb);
6420	sbuf_delete(sb);
6421
6422	return (rc);
6423}
6424
6425static inline void
6426tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
6427{
6428	*mask = x | y;
6429	y = htobe64(y);
6430	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
6431}
6432
6433static int
6434sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
6435{
6436	struct adapter *sc = arg1;
6437	struct sbuf *sb;
6438	int rc, i;
6439
6440	MPASS(chip_id(sc) <= CHELSIO_T5);
6441
6442	rc = sysctl_wire_old_buffer(req, 0);
6443	if (rc != 0)
6444		return (rc);
6445
6446	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6447	if (sb == NULL)
6448		return (ENOMEM);
6449
6450	sbuf_printf(sb,
6451	    "Idx  Ethernet address     Mask     Vld Ports PF"
6452	    "  VF              Replication             P0 P1 P2 P3  ML");
6453	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
6454		uint64_t tcamx, tcamy, mask;
6455		uint32_t cls_lo, cls_hi;
6456		uint8_t addr[ETHER_ADDR_LEN];
6457
6458		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
6459		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
6460		if (tcamx & tcamy)
6461			continue;
6462		tcamxy2valmask(tcamx, tcamy, addr, &mask);
6463		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
6464		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
6465		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
6466			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
6467			   addr[3], addr[4], addr[5], (uintmax_t)mask,
6468			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
6469			   G_PORTMAP(cls_hi), G_PF(cls_lo),
6470			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
6471
6472		if (cls_lo & F_REPLICATE) {
6473			struct fw_ldst_cmd ldst_cmd;
6474
6475			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
6476			ldst_cmd.op_to_addrspace =
6477			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
6478				F_FW_CMD_REQUEST | F_FW_CMD_READ |
6479				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
6480			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
6481			ldst_cmd.u.mps.rplc.fid_idx =
6482			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
6483				V_FW_LDST_CMD_IDX(i));
6484
6485			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
6486			    "t4mps");
6487			if (rc)
6488				break;
6489			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
6490			    sizeof(ldst_cmd), &ldst_cmd);
6491			end_synchronized_op(sc, 0);
6492
6493			if (rc != 0) {
6494				sbuf_printf(sb, "%36d", rc);
6495				rc = 0;
6496			} else {
6497				sbuf_printf(sb, " %08x %08x %08x %08x",
6498				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
6499				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
6500				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
6501				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
6502			}
6503		} else
6504			sbuf_printf(sb, "%36s", "");
6505
6506		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
6507		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
6508		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
6509	}
6510
6511	if (rc)
6512		(void) sbuf_finish(sb);
6513	else
6514		rc = sbuf_finish(sb);
6515	sbuf_delete(sb);
6516
6517	return (rc);
6518}
6519
6520static int
6521sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
6522{
6523	struct adapter *sc = arg1;
6524	struct sbuf *sb;
6525	int rc, i;
6526
6527	MPASS(chip_id(sc) > CHELSIO_T5);
6528
6529	rc = sysctl_wire_old_buffer(req, 0);
6530	if (rc != 0)
6531		return (rc);
6532
6533	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6534	if (sb == NULL)
6535		return (ENOMEM);
6536
6537	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
6538	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
6539	    "                           Replication"
6540	    "                                    P0 P1 P2 P3  ML\n");
6541
6542	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
6543		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
6544		uint16_t ivlan;
6545		uint64_t tcamx, tcamy, val, mask;
6546		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
6547		uint8_t addr[ETHER_ADDR_LEN];
6548
6549		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
6550		if (i < 256)
6551			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
6552		else
6553			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
6554		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
6555		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
6556		tcamy = G_DMACH(val) << 32;
6557		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
6558		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
6559		lookup_type = G_DATALKPTYPE(data2);
6560		port_num = G_DATAPORTNUM(data2);
6561		if (lookup_type && lookup_type != M_DATALKPTYPE) {
6562			/* Inner header VNI */
6563			vniy = ((data2 & F_DATAVIDH2) << 23) |
6564				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
6565			dip_hit = data2 & F_DATADIPHIT;
6566			vlan_vld = 0;
6567		} else {
6568			vniy = 0;
6569			dip_hit = 0;
6570			vlan_vld = data2 & F_DATAVIDH2;
6571			ivlan = G_VIDL(val);
6572		}
6573
6574		ctl |= V_CTLXYBITSEL(1);
6575		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
6576		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
6577		tcamx = G_DMACH(val) << 32;
6578		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
6579		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
6580		if (lookup_type && lookup_type != M_DATALKPTYPE) {
6581			/* Inner header VNI mask */
6582			vnix = ((data2 & F_DATAVIDH2) << 23) |
6583			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
6584		} else
6585			vnix = 0;
6586
6587		if (tcamx & tcamy)
6588			continue;
6589		tcamxy2valmask(tcamx, tcamy, addr, &mask);
6590
6591		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
6592		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
6593
6594		if (lookup_type && lookup_type != M_DATALKPTYPE) {
6595			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
6596			    "%012jx %06x %06x    -    -   %3c"
6597			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
6598			    addr[1], addr[2], addr[3], addr[4], addr[5],
6599			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
6600			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
6601			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
6602			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
6603		} else {
6604			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
6605			    "%012jx    -       -   ", i, addr[0], addr[1],
6606			    addr[2], addr[3], addr[4], addr[5],
6607			    (uintmax_t)mask);
6608
6609			if (vlan_vld)
6610				sbuf_printf(sb, "%4u   Y     ", ivlan);
6611			else
6612				sbuf_printf(sb, "  -    N     ");
6613
6614			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
6615			    lookup_type ? 'I' : 'O', port_num,
6616			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
6617			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
6618			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
6619		}
6620
6621
6622		if (cls_lo & F_T6_REPLICATE) {
6623			struct fw_ldst_cmd ldst_cmd;
6624
6625			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
6626			ldst_cmd.op_to_addrspace =
6627			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
6628				F_FW_CMD_REQUEST | F_FW_CMD_READ |
6629				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
6630			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
6631			ldst_cmd.u.mps.rplc.fid_idx =
6632			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
6633				V_FW_LDST_CMD_IDX(i));
6634
6635			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
6636			    "t6mps");
6637			if (rc)
6638				break;
6639			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
6640			    sizeof(ldst_cmd), &ldst_cmd);
6641			end_synchronized_op(sc, 0);
6642
6643			if (rc != 0) {
6644				sbuf_printf(sb, "%72d", rc);
6645				rc = 0;
6646			} else {
6647				sbuf_printf(sb, " %08x %08x %08x %08x"
6648				    " %08x %08x %08x %08x",
6649				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
6650				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
6651				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
6652				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
6653				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
6654				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
6655				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
6656				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
6657			}
6658		} else
6659			sbuf_printf(sb, "%72s", "");
6660
6661		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
6662		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
6663		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
6664		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
6665	}
6666
6667	if (rc)
6668		(void) sbuf_finish(sb);
6669	else
6670		rc = sbuf_finish(sb);
6671	sbuf_delete(sb);
6672
6673	return (rc);
6674}
6675
6676static int
6677sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
6678{
6679	struct adapter *sc = arg1;
6680	struct sbuf *sb;
6681	int rc;
6682	uint16_t mtus[NMTUS];
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, 256, req);
6689	if (sb == NULL)
6690		return (ENOMEM);
6691
6692	t4_read_mtu_tbl(sc, mtus, NULL);
6693
6694	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
6695	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
6696	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
6697	    mtus[14], mtus[15]);
6698
6699	rc = sbuf_finish(sb);
6700	sbuf_delete(sb);
6701
6702	return (rc);
6703}
6704
6705static int
6706sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
6707{
6708	struct adapter *sc = arg1;
6709	struct sbuf *sb;
6710	int rc, i;
6711	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
6712	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
6713	static const char *tx_stats[MAX_PM_NSTATS] = {
6714		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
6715		"Tx FIFO wait", NULL, "Tx latency"
6716	};
6717	static const char *rx_stats[MAX_PM_NSTATS] = {
6718		"Read:", "Write bypass:", "Write mem:", "Flush:",
6719		" Rx FIFO wait", NULL, "Rx latency"
6720	};
6721
6722	rc = sysctl_wire_old_buffer(req, 0);
6723	if (rc != 0)
6724		return (rc);
6725
6726	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6727	if (sb == NULL)
6728		return (ENOMEM);
6729
6730	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
6731	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
6732
6733	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
6734	for (i = 0; i < 4; i++) {
6735		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
6736		    tx_cyc[i]);
6737	}
6738
6739	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
6740	for (i = 0; i < 4; i++) {
6741		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
6742		    rx_cyc[i]);
6743	}
6744
6745	if (chip_id(sc) > CHELSIO_T5) {
6746		sbuf_printf(sb,
6747		    "\n              Total wait      Total occupancy");
6748		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
6749		    tx_cyc[i]);
6750		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
6751		    rx_cyc[i]);
6752
6753		i += 2;
6754		MPASS(i < nitems(tx_stats));
6755
6756		sbuf_printf(sb,
6757		    "\n                   Reads           Total wait");
6758		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
6759		    tx_cyc[i]);
6760		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
6761		    rx_cyc[i]);
6762	}
6763
6764	rc = sbuf_finish(sb);
6765	sbuf_delete(sb);
6766
6767	return (rc);
6768}
6769
6770static int
6771sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
6772{
6773	struct adapter *sc = arg1;
6774	struct sbuf *sb;
6775	int rc;
6776	struct tp_rdma_stats stats;
6777
6778	rc = sysctl_wire_old_buffer(req, 0);
6779	if (rc != 0)
6780		return (rc);
6781
6782	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6783	if (sb == NULL)
6784		return (ENOMEM);
6785
6786	mtx_lock(&sc->reg_lock);
6787	t4_tp_get_rdma_stats(sc, &stats);
6788	mtx_unlock(&sc->reg_lock);
6789
6790	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
6791	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
6792
6793	rc = sbuf_finish(sb);
6794	sbuf_delete(sb);
6795
6796	return (rc);
6797}
6798
6799static int
6800sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
6801{
6802	struct adapter *sc = arg1;
6803	struct sbuf *sb;
6804	int rc;
6805	struct tp_tcp_stats v4, v6;
6806
6807	rc = sysctl_wire_old_buffer(req, 0);
6808	if (rc != 0)
6809		return (rc);
6810
6811	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6812	if (sb == NULL)
6813		return (ENOMEM);
6814
6815	mtx_lock(&sc->reg_lock);
6816	t4_tp_get_tcp_stats(sc, &v4, &v6);
6817	mtx_unlock(&sc->reg_lock);
6818
6819	sbuf_printf(sb,
6820	    "                                IP                 IPv6\n");
6821	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
6822	    v4.tcp_out_rsts, v6.tcp_out_rsts);
6823	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
6824	    v4.tcp_in_segs, v6.tcp_in_segs);
6825	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
6826	    v4.tcp_out_segs, v6.tcp_out_segs);
6827	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
6828	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
6829
6830	rc = sbuf_finish(sb);
6831	sbuf_delete(sb);
6832
6833	return (rc);
6834}
6835
6836static int
6837sysctl_tids(SYSCTL_HANDLER_ARGS)
6838{
6839	struct adapter *sc = arg1;
6840	struct sbuf *sb;
6841	int rc;
6842	struct tid_info *t = &sc->tids;
6843
6844	rc = sysctl_wire_old_buffer(req, 0);
6845	if (rc != 0)
6846		return (rc);
6847
6848	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6849	if (sb == NULL)
6850		return (ENOMEM);
6851
6852	if (t->natids) {
6853		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
6854		    t->atids_in_use);
6855	}
6856
6857	if (t->ntids) {
6858		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
6859			uint32_t b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
6860
6861			if (b) {
6862				sbuf_printf(sb, "TID range: 0-%u, %u-%u", b - 1,
6863				    t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4,
6864				    t->ntids - 1);
6865			} else {
6866				sbuf_printf(sb, "TID range: %u-%u",
6867				    t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4,
6868				    t->ntids - 1);
6869			}
6870		} else
6871			sbuf_printf(sb, "TID range: 0-%u", t->ntids - 1);
6872		sbuf_printf(sb, ", in use: %u\n",
6873		    atomic_load_acq_int(&t->tids_in_use));
6874	}
6875
6876	if (t->nstids) {
6877		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
6878		    t->stid_base + t->nstids - 1, t->stids_in_use);
6879	}
6880
6881	if (t->nftids) {
6882		sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
6883		    t->ftid_base + t->nftids - 1);
6884	}
6885
6886	if (t->netids) {
6887		sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
6888		    t->etid_base + t->netids - 1);
6889	}
6890
6891	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
6892	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
6893	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
6894
6895	rc = sbuf_finish(sb);
6896	sbuf_delete(sb);
6897
6898	return (rc);
6899}
6900
6901static int
6902sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
6903{
6904	struct adapter *sc = arg1;
6905	struct sbuf *sb;
6906	int rc;
6907	struct tp_err_stats stats;
6908
6909	rc = sysctl_wire_old_buffer(req, 0);
6910	if (rc != 0)
6911		return (rc);
6912
6913	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6914	if (sb == NULL)
6915		return (ENOMEM);
6916
6917	mtx_lock(&sc->reg_lock);
6918	t4_tp_get_err_stats(sc, &stats);
6919	mtx_unlock(&sc->reg_lock);
6920
6921	if (sc->chip_params->nchan > 2) {
6922		sbuf_printf(sb, "                 channel 0  channel 1"
6923		    "  channel 2  channel 3\n");
6924		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
6925		    stats.mac_in_errs[0], stats.mac_in_errs[1],
6926		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
6927		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
6928		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
6929		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
6930		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
6931		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
6932		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
6933		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
6934		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
6935		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
6936		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
6937		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
6938		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
6939		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
6940		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
6941		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
6942		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
6943		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
6944		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
6945		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
6946		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
6947		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
6948	} else {
6949		sbuf_printf(sb, "                 channel 0  channel 1\n");
6950		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
6951		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
6952		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
6953		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
6954		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
6955		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
6956		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
6957		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
6958		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
6959		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
6960		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
6961		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
6962		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
6963		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
6964		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
6965		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
6966	}
6967
6968	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
6969	    stats.ofld_no_neigh, stats.ofld_cong_defer);
6970
6971	rc = sbuf_finish(sb);
6972	sbuf_delete(sb);
6973
6974	return (rc);
6975}
6976
6977static int
6978sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
6979{
6980	struct adapter *sc = arg1;
6981	struct tp_params *tpp = &sc->params.tp;
6982	u_int mask;
6983	int rc;
6984
6985	mask = tpp->la_mask >> 16;
6986	rc = sysctl_handle_int(oidp, &mask, 0, req);
6987	if (rc != 0 || req->newptr == NULL)
6988		return (rc);
6989	if (mask > 0xffff)
6990		return (EINVAL);
6991	tpp->la_mask = mask << 16;
6992	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
6993
6994	return (0);
6995}
6996
6997struct field_desc {
6998	const char *name;
6999	u_int start;
7000	u_int width;
7001};
7002
7003static void
7004field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
7005{
7006	char buf[32];
7007	int line_size = 0;
7008
7009	while (f->name) {
7010		uint64_t mask = (1ULL << f->width) - 1;
7011		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
7012		    ((uintmax_t)v >> f->start) & mask);
7013
7014		if (line_size + len >= 79) {
7015			line_size = 8;
7016			sbuf_printf(sb, "\n        ");
7017		}
7018		sbuf_printf(sb, "%s ", buf);
7019		line_size += len + 1;
7020		f++;
7021	}
7022	sbuf_printf(sb, "\n");
7023}
7024
7025static const struct field_desc tp_la0[] = {
7026	{ "RcfOpCodeOut", 60, 4 },
7027	{ "State", 56, 4 },
7028	{ "WcfState", 52, 4 },
7029	{ "RcfOpcSrcOut", 50, 2 },
7030	{ "CRxError", 49, 1 },
7031	{ "ERxError", 48, 1 },
7032	{ "SanityFailed", 47, 1 },
7033	{ "SpuriousMsg", 46, 1 },
7034	{ "FlushInputMsg", 45, 1 },
7035	{ "FlushInputCpl", 44, 1 },
7036	{ "RssUpBit", 43, 1 },
7037	{ "RssFilterHit", 42, 1 },
7038	{ "Tid", 32, 10 },
7039	{ "InitTcb", 31, 1 },
7040	{ "LineNumber", 24, 7 },
7041	{ "Emsg", 23, 1 },
7042	{ "EdataOut", 22, 1 },
7043	{ "Cmsg", 21, 1 },
7044	{ "CdataOut", 20, 1 },
7045	{ "EreadPdu", 19, 1 },
7046	{ "CreadPdu", 18, 1 },
7047	{ "TunnelPkt", 17, 1 },
7048	{ "RcfPeerFin", 16, 1 },
7049	{ "RcfReasonOut", 12, 4 },
7050	{ "TxCchannel", 10, 2 },
7051	{ "RcfTxChannel", 8, 2 },
7052	{ "RxEchannel", 6, 2 },
7053	{ "RcfRxChannel", 5, 1 },
7054	{ "RcfDataOutSrdy", 4, 1 },
7055	{ "RxDvld", 3, 1 },
7056	{ "RxOoDvld", 2, 1 },
7057	{ "RxCongestion", 1, 1 },
7058	{ "TxCongestion", 0, 1 },
7059	{ NULL }
7060};
7061
7062static const struct field_desc tp_la1[] = {
7063	{ "CplCmdIn", 56, 8 },
7064	{ "CplCmdOut", 48, 8 },
7065	{ "ESynOut", 47, 1 },
7066	{ "EAckOut", 46, 1 },
7067	{ "EFinOut", 45, 1 },
7068	{ "ERstOut", 44, 1 },
7069	{ "SynIn", 43, 1 },
7070	{ "AckIn", 42, 1 },
7071	{ "FinIn", 41, 1 },
7072	{ "RstIn", 40, 1 },
7073	{ "DataIn", 39, 1 },
7074	{ "DataInVld", 38, 1 },
7075	{ "PadIn", 37, 1 },
7076	{ "RxBufEmpty", 36, 1 },
7077	{ "RxDdp", 35, 1 },
7078	{ "RxFbCongestion", 34, 1 },
7079	{ "TxFbCongestion", 33, 1 },
7080	{ "TxPktSumSrdy", 32, 1 },
7081	{ "RcfUlpType", 28, 4 },
7082	{ "Eread", 27, 1 },
7083	{ "Ebypass", 26, 1 },
7084	{ "Esave", 25, 1 },
7085	{ "Static0", 24, 1 },
7086	{ "Cread", 23, 1 },
7087	{ "Cbypass", 22, 1 },
7088	{ "Csave", 21, 1 },
7089	{ "CPktOut", 20, 1 },
7090	{ "RxPagePoolFull", 18, 2 },
7091	{ "RxLpbkPkt", 17, 1 },
7092	{ "TxLpbkPkt", 16, 1 },
7093	{ "RxVfValid", 15, 1 },
7094	{ "SynLearned", 14, 1 },
7095	{ "SetDelEntry", 13, 1 },
7096	{ "SetInvEntry", 12, 1 },
7097	{ "CpcmdDvld", 11, 1 },
7098	{ "CpcmdSave", 10, 1 },
7099	{ "RxPstructsFull", 8, 2 },
7100	{ "EpcmdDvld", 7, 1 },
7101	{ "EpcmdFlush", 6, 1 },
7102	{ "EpcmdTrimPrefix", 5, 1 },
7103	{ "EpcmdTrimPostfix", 4, 1 },
7104	{ "ERssIp4Pkt", 3, 1 },
7105	{ "ERssIp6Pkt", 2, 1 },
7106	{ "ERssTcpUdpPkt", 1, 1 },
7107	{ "ERssFceFipPkt", 0, 1 },
7108	{ NULL }
7109};
7110
7111static const struct field_desc tp_la2[] = {
7112	{ "CplCmdIn", 56, 8 },
7113	{ "MpsVfVld", 55, 1 },
7114	{ "MpsPf", 52, 3 },
7115	{ "MpsVf", 44, 8 },
7116	{ "SynIn", 43, 1 },
7117	{ "AckIn", 42, 1 },
7118	{ "FinIn", 41, 1 },
7119	{ "RstIn", 40, 1 },
7120	{ "DataIn", 39, 1 },
7121	{ "DataInVld", 38, 1 },
7122	{ "PadIn", 37, 1 },
7123	{ "RxBufEmpty", 36, 1 },
7124	{ "RxDdp", 35, 1 },
7125	{ "RxFbCongestion", 34, 1 },
7126	{ "TxFbCongestion", 33, 1 },
7127	{ "TxPktSumSrdy", 32, 1 },
7128	{ "RcfUlpType", 28, 4 },
7129	{ "Eread", 27, 1 },
7130	{ "Ebypass", 26, 1 },
7131	{ "Esave", 25, 1 },
7132	{ "Static0", 24, 1 },
7133	{ "Cread", 23, 1 },
7134	{ "Cbypass", 22, 1 },
7135	{ "Csave", 21, 1 },
7136	{ "CPktOut", 20, 1 },
7137	{ "RxPagePoolFull", 18, 2 },
7138	{ "RxLpbkPkt", 17, 1 },
7139	{ "TxLpbkPkt", 16, 1 },
7140	{ "RxVfValid", 15, 1 },
7141	{ "SynLearned", 14, 1 },
7142	{ "SetDelEntry", 13, 1 },
7143	{ "SetInvEntry", 12, 1 },
7144	{ "CpcmdDvld", 11, 1 },
7145	{ "CpcmdSave", 10, 1 },
7146	{ "RxPstructsFull", 8, 2 },
7147	{ "EpcmdDvld", 7, 1 },
7148	{ "EpcmdFlush", 6, 1 },
7149	{ "EpcmdTrimPrefix", 5, 1 },
7150	{ "EpcmdTrimPostfix", 4, 1 },
7151	{ "ERssIp4Pkt", 3, 1 },
7152	{ "ERssIp6Pkt", 2, 1 },
7153	{ "ERssTcpUdpPkt", 1, 1 },
7154	{ "ERssFceFipPkt", 0, 1 },
7155	{ NULL }
7156};
7157
7158static void
7159tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
7160{
7161
7162	field_desc_show(sb, *p, tp_la0);
7163}
7164
7165static void
7166tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
7167{
7168
7169	if (idx)
7170		sbuf_printf(sb, "\n");
7171	field_desc_show(sb, p[0], tp_la0);
7172	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7173		field_desc_show(sb, p[1], tp_la0);
7174}
7175
7176static void
7177tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
7178{
7179
7180	if (idx)
7181		sbuf_printf(sb, "\n");
7182	field_desc_show(sb, p[0], tp_la0);
7183	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7184		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
7185}
7186
7187static int
7188sysctl_tp_la(SYSCTL_HANDLER_ARGS)
7189{
7190	struct adapter *sc = arg1;
7191	struct sbuf *sb;
7192	uint64_t *buf, *p;
7193	int rc;
7194	u_int i, inc;
7195	void (*show_func)(struct sbuf *, uint64_t *, int);
7196
7197	rc = sysctl_wire_old_buffer(req, 0);
7198	if (rc != 0)
7199		return (rc);
7200
7201	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7202	if (sb == NULL)
7203		return (ENOMEM);
7204
7205	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
7206
7207	t4_tp_read_la(sc, buf, NULL);
7208	p = buf;
7209
7210	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
7211	case 2:
7212		inc = 2;
7213		show_func = tp_la_show2;
7214		break;
7215	case 3:
7216		inc = 2;
7217		show_func = tp_la_show3;
7218		break;
7219	default:
7220		inc = 1;
7221		show_func = tp_la_show;
7222	}
7223
7224	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
7225		(*show_func)(sb, p, i);
7226
7227	rc = sbuf_finish(sb);
7228	sbuf_delete(sb);
7229	free(buf, M_CXGBE);
7230	return (rc);
7231}
7232
7233static int
7234sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
7235{
7236	struct adapter *sc = arg1;
7237	struct sbuf *sb;
7238	int rc;
7239	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
7240
7241	rc = sysctl_wire_old_buffer(req, 0);
7242	if (rc != 0)
7243		return (rc);
7244
7245	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7246	if (sb == NULL)
7247		return (ENOMEM);
7248
7249	t4_get_chan_txrate(sc, nrate, orate);
7250
7251	if (sc->chip_params->nchan > 2) {
7252		sbuf_printf(sb, "              channel 0   channel 1"
7253		    "   channel 2   channel 3\n");
7254		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
7255		    nrate[0], nrate[1], nrate[2], nrate[3]);
7256		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
7257		    orate[0], orate[1], orate[2], orate[3]);
7258	} else {
7259		sbuf_printf(sb, "              channel 0   channel 1\n");
7260		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
7261		    nrate[0], nrate[1]);
7262		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
7263		    orate[0], orate[1]);
7264	}
7265
7266	rc = sbuf_finish(sb);
7267	sbuf_delete(sb);
7268
7269	return (rc);
7270}
7271
7272static int
7273sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
7274{
7275	struct adapter *sc = arg1;
7276	struct sbuf *sb;
7277	uint32_t *buf, *p;
7278	int rc, i;
7279
7280	rc = sysctl_wire_old_buffer(req, 0);
7281	if (rc != 0)
7282		return (rc);
7283
7284	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7285	if (sb == NULL)
7286		return (ENOMEM);
7287
7288	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
7289	    M_ZERO | M_WAITOK);
7290
7291	t4_ulprx_read_la(sc, buf);
7292	p = buf;
7293
7294	sbuf_printf(sb, "      Pcmd        Type   Message"
7295	    "                Data");
7296	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
7297		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
7298		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
7299	}
7300
7301	rc = sbuf_finish(sb);
7302	sbuf_delete(sb);
7303	free(buf, M_CXGBE);
7304	return (rc);
7305}
7306
7307static int
7308sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
7309{
7310	struct adapter *sc = arg1;
7311	struct sbuf *sb;
7312	int rc, v;
7313
7314	rc = sysctl_wire_old_buffer(req, 0);
7315	if (rc != 0)
7316		return (rc);
7317
7318	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7319	if (sb == NULL)
7320		return (ENOMEM);
7321
7322	v = t4_read_reg(sc, A_SGE_STAT_CFG);
7323	if (G_STATSOURCE_T5(v) == 7) {
7324		if (G_STATMODE(v) == 0) {
7325			sbuf_printf(sb, "total %d, incomplete %d",
7326			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7327			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7328		} else if (G_STATMODE(v) == 1) {
7329			sbuf_printf(sb, "total %d, data overflow %d",
7330			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7331			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7332		}
7333	}
7334	rc = sbuf_finish(sb);
7335	sbuf_delete(sb);
7336
7337	return (rc);
7338}
7339
7340static int
7341sysctl_tc_params(SYSCTL_HANDLER_ARGS)
7342{
7343	struct adapter *sc = arg1;
7344	struct tx_sched_class *tc;
7345	struct t4_sched_class_params p;
7346	struct sbuf *sb;
7347	int i, rc, port_id, flags, mbps, gbps;
7348
7349	rc = sysctl_wire_old_buffer(req, 0);
7350	if (rc != 0)
7351		return (rc);
7352
7353	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7354	if (sb == NULL)
7355		return (ENOMEM);
7356
7357	port_id = arg2 >> 16;
7358	MPASS(port_id < sc->params.nports);
7359	MPASS(sc->port[port_id] != NULL);
7360	i = arg2 & 0xffff;
7361	MPASS(i < sc->chip_params->nsched_cls);
7362	tc = &sc->port[port_id]->tc[i];
7363
7364	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7365	    "t4tc_p");
7366	if (rc)
7367		goto done;
7368	flags = tc->flags;
7369	p = tc->params;
7370	end_synchronized_op(sc, LOCK_HELD);
7371
7372	if ((flags & TX_SC_OK) == 0) {
7373		sbuf_printf(sb, "none");
7374		goto done;
7375	}
7376
7377	if (p.level == SCHED_CLASS_LEVEL_CL_WRR) {
7378		sbuf_printf(sb, "cl-wrr weight %u", p.weight);
7379		goto done;
7380	} else if (p.level == SCHED_CLASS_LEVEL_CL_RL)
7381		sbuf_printf(sb, "cl-rl");
7382	else if (p.level == SCHED_CLASS_LEVEL_CH_RL)
7383		sbuf_printf(sb, "ch-rl");
7384	else {
7385		rc = ENXIO;
7386		goto done;
7387	}
7388
7389	if (p.ratemode == SCHED_CLASS_RATEMODE_REL) {
7390		/* XXX: top speed or actual link speed? */
7391		gbps = port_top_speed(sc->port[port_id]);
7392		sbuf_printf(sb, " %u%% of %uGbps", p.maxrate, gbps);
7393	}
7394	else if (p.ratemode == SCHED_CLASS_RATEMODE_ABS) {
7395		switch (p.rateunit) {
7396		case SCHED_CLASS_RATEUNIT_BITS:
7397			mbps = p.maxrate / 1000;
7398			gbps = p.maxrate / 1000000;
7399			if (p.maxrate == gbps * 1000000)
7400				sbuf_printf(sb, " %uGbps", gbps);
7401			else if (p.maxrate == mbps * 1000)
7402				sbuf_printf(sb, " %uMbps", mbps);
7403			else
7404				sbuf_printf(sb, " %uKbps", p.maxrate);
7405			break;
7406		case SCHED_CLASS_RATEUNIT_PKTS:
7407			sbuf_printf(sb, " %upps", p.maxrate);
7408			break;
7409		default:
7410			rc = ENXIO;
7411			goto done;
7412		}
7413	}
7414
7415	switch (p.mode) {
7416	case SCHED_CLASS_MODE_CLASS:
7417		sbuf_printf(sb, " aggregate");
7418		break;
7419	case SCHED_CLASS_MODE_FLOW:
7420		sbuf_printf(sb, " per-flow");
7421		break;
7422	default:
7423		rc = ENXIO;
7424		goto done;
7425	}
7426
7427done:
7428	if (rc == 0)
7429		rc = sbuf_finish(sb);
7430	sbuf_delete(sb);
7431
7432	return (rc);
7433}
7434#endif
7435
7436#ifdef TCP_OFFLOAD
7437static void
7438unit_conv(char *buf, size_t len, u_int val, u_int factor)
7439{
7440	u_int rem = val % factor;
7441
7442	if (rem == 0)
7443		snprintf(buf, len, "%u", val / factor);
7444	else {
7445		while (rem % 10 == 0)
7446			rem /= 10;
7447		snprintf(buf, len, "%u.%u", val / factor, rem);
7448	}
7449}
7450
7451static int
7452sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
7453{
7454	struct adapter *sc = arg1;
7455	char buf[16];
7456	u_int res, re;
7457	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7458
7459	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
7460	switch (arg2) {
7461	case 0:
7462		/* timer_tick */
7463		re = G_TIMERRESOLUTION(res);
7464		break;
7465	case 1:
7466		/* TCP timestamp tick */
7467		re = G_TIMESTAMPRESOLUTION(res);
7468		break;
7469	case 2:
7470		/* DACK tick */
7471		re = G_DELAYEDACKRESOLUTION(res);
7472		break;
7473	default:
7474		return (EDOOFUS);
7475	}
7476
7477	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
7478
7479	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
7480}
7481
7482static int
7483sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
7484{
7485	struct adapter *sc = arg1;
7486	u_int res, dack_re, v;
7487	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7488
7489	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
7490	dack_re = G_DELAYEDACKRESOLUTION(res);
7491	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
7492
7493	return (sysctl_handle_int(oidp, &v, 0, req));
7494}
7495
7496static int
7497sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
7498{
7499	struct adapter *sc = arg1;
7500	int reg = arg2;
7501	u_int tre;
7502	u_long tp_tick_us, v;
7503	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7504
7505	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
7506	    reg == A_TP_PERS_MIN || reg == A_TP_PERS_MAX ||
7507	    reg == A_TP_KEEP_IDLE || A_TP_KEEP_INTVL || reg == A_TP_INIT_SRTT ||
7508	    reg == A_TP_FINWAIT2_TIMER);
7509
7510	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
7511	tp_tick_us = (cclk_ps << tre) / 1000000;
7512
7513	if (reg == A_TP_INIT_SRTT)
7514		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
7515	else
7516		v = tp_tick_us * t4_read_reg(sc, reg);
7517
7518	return (sysctl_handle_long(oidp, &v, 0, req));
7519}
7520#endif
7521
7522static uint32_t
7523fconf_iconf_to_mode(uint32_t fconf, uint32_t iconf)
7524{
7525	uint32_t mode;
7526
7527	mode = T4_FILTER_IPv4 | T4_FILTER_IPv6 | T4_FILTER_IP_SADDR |
7528	    T4_FILTER_IP_DADDR | T4_FILTER_IP_SPORT | T4_FILTER_IP_DPORT;
7529
7530	if (fconf & F_FRAGMENTATION)
7531		mode |= T4_FILTER_IP_FRAGMENT;
7532
7533	if (fconf & F_MPSHITTYPE)
7534		mode |= T4_FILTER_MPS_HIT_TYPE;
7535
7536	if (fconf & F_MACMATCH)
7537		mode |= T4_FILTER_MAC_IDX;
7538
7539	if (fconf & F_ETHERTYPE)
7540		mode |= T4_FILTER_ETH_TYPE;
7541
7542	if (fconf & F_PROTOCOL)
7543		mode |= T4_FILTER_IP_PROTO;
7544
7545	if (fconf & F_TOS)
7546		mode |= T4_FILTER_IP_TOS;
7547
7548	if (fconf & F_VLAN)
7549		mode |= T4_FILTER_VLAN;
7550
7551	if (fconf & F_VNIC_ID) {
7552		mode |= T4_FILTER_VNIC;
7553		if (iconf & F_VNIC)
7554			mode |= T4_FILTER_IC_VNIC;
7555	}
7556
7557	if (fconf & F_PORT)
7558		mode |= T4_FILTER_PORT;
7559
7560	if (fconf & F_FCOE)
7561		mode |= T4_FILTER_FCoE;
7562
7563	return (mode);
7564}
7565
7566static uint32_t
7567mode_to_fconf(uint32_t mode)
7568{
7569	uint32_t fconf = 0;
7570
7571	if (mode & T4_FILTER_IP_FRAGMENT)
7572		fconf |= F_FRAGMENTATION;
7573
7574	if (mode & T4_FILTER_MPS_HIT_TYPE)
7575		fconf |= F_MPSHITTYPE;
7576
7577	if (mode & T4_FILTER_MAC_IDX)
7578		fconf |= F_MACMATCH;
7579
7580	if (mode & T4_FILTER_ETH_TYPE)
7581		fconf |= F_ETHERTYPE;
7582
7583	if (mode & T4_FILTER_IP_PROTO)
7584		fconf |= F_PROTOCOL;
7585
7586	if (mode & T4_FILTER_IP_TOS)
7587		fconf |= F_TOS;
7588
7589	if (mode & T4_FILTER_VLAN)
7590		fconf |= F_VLAN;
7591
7592	if (mode & T4_FILTER_VNIC)
7593		fconf |= F_VNIC_ID;
7594
7595	if (mode & T4_FILTER_PORT)
7596		fconf |= F_PORT;
7597
7598	if (mode & T4_FILTER_FCoE)
7599		fconf |= F_FCOE;
7600
7601	return (fconf);
7602}
7603
7604static uint32_t
7605mode_to_iconf(uint32_t mode)
7606{
7607
7608	if (mode & T4_FILTER_IC_VNIC)
7609		return (F_VNIC);
7610	return (0);
7611}
7612
7613static int check_fspec_against_fconf_iconf(struct adapter *sc,
7614    struct t4_filter_specification *fs)
7615{
7616	struct tp_params *tpp = &sc->params.tp;
7617	uint32_t fconf = 0;
7618
7619	if (fs->val.frag || fs->mask.frag)
7620		fconf |= F_FRAGMENTATION;
7621
7622	if (fs->val.matchtype || fs->mask.matchtype)
7623		fconf |= F_MPSHITTYPE;
7624
7625	if (fs->val.macidx || fs->mask.macidx)
7626		fconf |= F_MACMATCH;
7627
7628	if (fs->val.ethtype || fs->mask.ethtype)
7629		fconf |= F_ETHERTYPE;
7630
7631	if (fs->val.proto || fs->mask.proto)
7632		fconf |= F_PROTOCOL;
7633
7634	if (fs->val.tos || fs->mask.tos)
7635		fconf |= F_TOS;
7636
7637	if (fs->val.vlan_vld || fs->mask.vlan_vld)
7638		fconf |= F_VLAN;
7639
7640	if (fs->val.ovlan_vld || fs->mask.ovlan_vld) {
7641		fconf |= F_VNIC_ID;
7642		if (tpp->ingress_config & F_VNIC)
7643			return (EINVAL);
7644	}
7645
7646	if (fs->val.pfvf_vld || fs->mask.pfvf_vld) {
7647		fconf |= F_VNIC_ID;
7648		if ((tpp->ingress_config & F_VNIC) == 0)
7649			return (EINVAL);
7650	}
7651
7652	if (fs->val.iport || fs->mask.iport)
7653		fconf |= F_PORT;
7654
7655	if (fs->val.fcoe || fs->mask.fcoe)
7656		fconf |= F_FCOE;
7657
7658	if ((tpp->vlan_pri_map | fconf) != tpp->vlan_pri_map)
7659		return (E2BIG);
7660
7661	return (0);
7662}
7663
7664static int
7665get_filter_mode(struct adapter *sc, uint32_t *mode)
7666{
7667	struct tp_params *tpp = &sc->params.tp;
7668
7669	/*
7670	 * We trust the cached values of the relevant TP registers.  This means
7671	 * things work reliably only if writes to those registers are always via
7672	 * t4_set_filter_mode.
7673	 */
7674	*mode = fconf_iconf_to_mode(tpp->vlan_pri_map, tpp->ingress_config);
7675
7676	return (0);
7677}
7678
7679static int
7680set_filter_mode(struct adapter *sc, uint32_t mode)
7681{
7682	struct tp_params *tpp = &sc->params.tp;
7683	uint32_t fconf, iconf;
7684	int rc;
7685
7686	iconf = mode_to_iconf(mode);
7687	if ((iconf ^ tpp->ingress_config) & F_VNIC) {
7688		/*
7689		 * For now we just complain if A_TP_INGRESS_CONFIG is not
7690		 * already set to the correct value for the requested filter
7691		 * mode.  It's not clear if it's safe to write to this register
7692		 * on the fly.  (And we trust the cached value of the register).
7693		 */
7694		return (EBUSY);
7695	}
7696
7697	fconf = mode_to_fconf(mode);
7698
7699	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7700	    "t4setfm");
7701	if (rc)
7702		return (rc);
7703
7704	if (sc->tids.ftids_in_use > 0) {
7705		rc = EBUSY;
7706		goto done;
7707	}
7708
7709#ifdef TCP_OFFLOAD
7710	if (uld_active(sc, ULD_TOM)) {
7711		rc = EBUSY;
7712		goto done;
7713	}
7714#endif
7715
7716	rc = -t4_set_filter_mode(sc, fconf);
7717done:
7718	end_synchronized_op(sc, LOCK_HELD);
7719	return (rc);
7720}
7721
7722static inline uint64_t
7723get_filter_hits(struct adapter *sc, uint32_t fid)
7724{
7725	uint32_t tcb_addr;
7726
7727	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE) +
7728	    (fid + sc->tids.ftid_base) * TCB_SIZE;
7729
7730	if (is_t4(sc)) {
7731		uint64_t hits;
7732
7733		read_via_memwin(sc, 0, tcb_addr + 16, (uint32_t *)&hits, 8);
7734		return (be64toh(hits));
7735	} else {
7736		uint32_t hits;
7737
7738		read_via_memwin(sc, 0, tcb_addr + 24, &hits, 4);
7739		return (be32toh(hits));
7740	}
7741}
7742
7743static int
7744get_filter(struct adapter *sc, struct t4_filter *t)
7745{
7746	int i, rc, nfilters = sc->tids.nftids;
7747	struct filter_entry *f;
7748
7749	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7750	    "t4getf");
7751	if (rc)
7752		return (rc);
7753
7754	if (sc->tids.ftids_in_use == 0 || sc->tids.ftid_tab == NULL ||
7755	    t->idx >= nfilters) {
7756		t->idx = 0xffffffff;
7757		goto done;
7758	}
7759
7760	f = &sc->tids.ftid_tab[t->idx];
7761	for (i = t->idx; i < nfilters; i++, f++) {
7762		if (f->valid) {
7763			t->idx = i;
7764			t->l2tidx = f->l2t ? f->l2t->idx : 0;
7765			t->smtidx = f->smtidx;
7766			if (f->fs.hitcnts)
7767				t->hits = get_filter_hits(sc, t->idx);
7768			else
7769				t->hits = UINT64_MAX;
7770			t->fs = f->fs;
7771
7772			goto done;
7773		}
7774	}
7775
7776	t->idx = 0xffffffff;
7777done:
7778	end_synchronized_op(sc, LOCK_HELD);
7779	return (0);
7780}
7781
7782static int
7783set_filter(struct adapter *sc, struct t4_filter *t)
7784{
7785	unsigned int nfilters, nports;
7786	struct filter_entry *f;
7787	int i, rc;
7788
7789	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setf");
7790	if (rc)
7791		return (rc);
7792
7793	nfilters = sc->tids.nftids;
7794	nports = sc->params.nports;
7795
7796	if (nfilters == 0) {
7797		rc = ENOTSUP;
7798		goto done;
7799	}
7800
7801	if (!(sc->flags & FULL_INIT_DONE)) {
7802		rc = EAGAIN;
7803		goto done;
7804	}
7805
7806	if (t->idx >= nfilters) {
7807		rc = EINVAL;
7808		goto done;
7809	}
7810
7811	/* Validate against the global filter mode and ingress config */
7812	rc = check_fspec_against_fconf_iconf(sc, &t->fs);
7813	if (rc != 0)
7814		goto done;
7815
7816	if (t->fs.action == FILTER_SWITCH && t->fs.eport >= nports) {
7817		rc = EINVAL;
7818		goto done;
7819	}
7820
7821	if (t->fs.val.iport >= nports) {
7822		rc = EINVAL;
7823		goto done;
7824	}
7825
7826	/* Can't specify an iq if not steering to it */
7827	if (!t->fs.dirsteer && t->fs.iq) {
7828		rc = EINVAL;
7829		goto done;
7830	}
7831
7832	/* IPv6 filter idx must be 4 aligned */
7833	if (t->fs.type == 1 &&
7834	    ((t->idx & 0x3) || t->idx + 4 >= nfilters)) {
7835		rc = EINVAL;
7836		goto done;
7837	}
7838
7839	if (sc->tids.ftid_tab == NULL) {
7840		KASSERT(sc->tids.ftids_in_use == 0,
7841		    ("%s: no memory allocated but filters_in_use > 0",
7842		    __func__));
7843
7844		sc->tids.ftid_tab = malloc(sizeof (struct filter_entry) *
7845		    nfilters, M_CXGBE, M_NOWAIT | M_ZERO);
7846		if (sc->tids.ftid_tab == NULL) {
7847			rc = ENOMEM;
7848			goto done;
7849		}
7850		mtx_init(&sc->tids.ftid_lock, "T4 filters", 0, MTX_DEF);
7851	}
7852
7853	for (i = 0; i < 4; i++) {
7854		f = &sc->tids.ftid_tab[t->idx + i];
7855
7856		if (f->pending || f->valid) {
7857			rc = EBUSY;
7858			goto done;
7859		}
7860		if (f->locked) {
7861			rc = EPERM;
7862			goto done;
7863		}
7864
7865		if (t->fs.type == 0)
7866			break;
7867	}
7868
7869	f = &sc->tids.ftid_tab[t->idx];
7870	f->fs = t->fs;
7871
7872	rc = set_filter_wr(sc, t->idx);
7873done:
7874	end_synchronized_op(sc, 0);
7875
7876	if (rc == 0) {
7877		mtx_lock(&sc->tids.ftid_lock);
7878		for (;;) {
7879			if (f->pending == 0) {
7880				rc = f->valid ? 0 : EIO;
7881				break;
7882			}
7883
7884			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
7885			    PCATCH, "t4setfw", 0)) {
7886				rc = EINPROGRESS;
7887				break;
7888			}
7889		}
7890		mtx_unlock(&sc->tids.ftid_lock);
7891	}
7892	return (rc);
7893}
7894
7895static int
7896del_filter(struct adapter *sc, struct t4_filter *t)
7897{
7898	unsigned int nfilters;
7899	struct filter_entry *f;
7900	int rc;
7901
7902	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4delf");
7903	if (rc)
7904		return (rc);
7905
7906	nfilters = sc->tids.nftids;
7907
7908	if (nfilters == 0) {
7909		rc = ENOTSUP;
7910		goto done;
7911	}
7912
7913	if (sc->tids.ftid_tab == NULL || sc->tids.ftids_in_use == 0 ||
7914	    t->idx >= nfilters) {
7915		rc = EINVAL;
7916		goto done;
7917	}
7918
7919	if (!(sc->flags & FULL_INIT_DONE)) {
7920		rc = EAGAIN;
7921		goto done;
7922	}
7923
7924	f = &sc->tids.ftid_tab[t->idx];
7925
7926	if (f->pending) {
7927		rc = EBUSY;
7928		goto done;
7929	}
7930	if (f->locked) {
7931		rc = EPERM;
7932		goto done;
7933	}
7934
7935	if (f->valid) {
7936		t->fs = f->fs;	/* extra info for the caller */
7937		rc = del_filter_wr(sc, t->idx);
7938	}
7939
7940done:
7941	end_synchronized_op(sc, 0);
7942
7943	if (rc == 0) {
7944		mtx_lock(&sc->tids.ftid_lock);
7945		for (;;) {
7946			if (f->pending == 0) {
7947				rc = f->valid ? EIO : 0;
7948				break;
7949			}
7950
7951			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
7952			    PCATCH, "t4delfw", 0)) {
7953				rc = EINPROGRESS;
7954				break;
7955			}
7956		}
7957		mtx_unlock(&sc->tids.ftid_lock);
7958	}
7959
7960	return (rc);
7961}
7962
7963static void
7964clear_filter(struct filter_entry *f)
7965{
7966	if (f->l2t)
7967		t4_l2t_release(f->l2t);
7968
7969	bzero(f, sizeof (*f));
7970}
7971
7972static int
7973set_filter_wr(struct adapter *sc, int fidx)
7974{
7975	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
7976	struct fw_filter_wr *fwr;
7977	unsigned int ftid, vnic_vld, vnic_vld_mask;
7978	struct wrq_cookie cookie;
7979
7980	ASSERT_SYNCHRONIZED_OP(sc);
7981
7982	if (f->fs.newdmac || f->fs.newvlan) {
7983		/* This filter needs an L2T entry; allocate one. */
7984		f->l2t = t4_l2t_alloc_switching(sc->l2t);
7985		if (f->l2t == NULL)
7986			return (EAGAIN);
7987		if (t4_l2t_set_switching(sc, f->l2t, f->fs.vlan, f->fs.eport,
7988		    f->fs.dmac)) {
7989			t4_l2t_release(f->l2t);
7990			f->l2t = NULL;
7991			return (ENOMEM);
7992		}
7993	}
7994
7995	/* Already validated against fconf, iconf */
7996	MPASS((f->fs.val.pfvf_vld & f->fs.val.ovlan_vld) == 0);
7997	MPASS((f->fs.mask.pfvf_vld & f->fs.mask.ovlan_vld) == 0);
7998	if (f->fs.val.pfvf_vld || f->fs.val.ovlan_vld)
7999		vnic_vld = 1;
8000	else
8001		vnic_vld = 0;
8002	if (f->fs.mask.pfvf_vld || f->fs.mask.ovlan_vld)
8003		vnic_vld_mask = 1;
8004	else
8005		vnic_vld_mask = 0;
8006
8007	ftid = sc->tids.ftid_base + fidx;
8008
8009	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8010	if (fwr == NULL)
8011		return (ENOMEM);
8012	bzero(fwr, sizeof(*fwr));
8013
8014	fwr->op_pkd = htobe32(V_FW_WR_OP(FW_FILTER_WR));
8015	fwr->len16_pkd = htobe32(FW_LEN16(*fwr));
8016	fwr->tid_to_iq =
8017	    htobe32(V_FW_FILTER_WR_TID(ftid) |
8018		V_FW_FILTER_WR_RQTYPE(f->fs.type) |
8019		V_FW_FILTER_WR_NOREPLY(0) |
8020		V_FW_FILTER_WR_IQ(f->fs.iq));
8021	fwr->del_filter_to_l2tix =
8022	    htobe32(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
8023		V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
8024		V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
8025		V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
8026		V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
8027		V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
8028		V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
8029		V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
8030		V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
8031		    f->fs.newvlan == VLAN_REWRITE) |
8032		V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
8033		    f->fs.newvlan == VLAN_REWRITE) |
8034		V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
8035		V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
8036		V_FW_FILTER_WR_PRIO(f->fs.prio) |
8037		V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
8038	fwr->ethtype = htobe16(f->fs.val.ethtype);
8039	fwr->ethtypem = htobe16(f->fs.mask.ethtype);
8040	fwr->frag_to_ovlan_vldm =
8041	    (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
8042		V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
8043		V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.vlan_vld) |
8044		V_FW_FILTER_WR_OVLAN_VLD(vnic_vld) |
8045		V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.vlan_vld) |
8046		V_FW_FILTER_WR_OVLAN_VLDM(vnic_vld_mask));
8047	fwr->smac_sel = 0;
8048	fwr->rx_chan_rx_rpl_iq = htobe16(V_FW_FILTER_WR_RX_CHAN(0) |
8049	    V_FW_FILTER_WR_RX_RPL_IQ(sc->sge.fwq.abs_id));
8050	fwr->maci_to_matchtypem =
8051	    htobe32(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
8052		V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
8053		V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
8054		V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
8055		V_FW_FILTER_WR_PORT(f->fs.val.iport) |
8056		V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
8057		V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
8058		V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
8059	fwr->ptcl = f->fs.val.proto;
8060	fwr->ptclm = f->fs.mask.proto;
8061	fwr->ttyp = f->fs.val.tos;
8062	fwr->ttypm = f->fs.mask.tos;
8063	fwr->ivlan = htobe16(f->fs.val.vlan);
8064	fwr->ivlanm = htobe16(f->fs.mask.vlan);
8065	fwr->ovlan = htobe16(f->fs.val.vnic);
8066	fwr->ovlanm = htobe16(f->fs.mask.vnic);
8067	bcopy(f->fs.val.dip, fwr->lip, sizeof (fwr->lip));
8068	bcopy(f->fs.mask.dip, fwr->lipm, sizeof (fwr->lipm));
8069	bcopy(f->fs.val.sip, fwr->fip, sizeof (fwr->fip));
8070	bcopy(f->fs.mask.sip, fwr->fipm, sizeof (fwr->fipm));
8071	fwr->lp = htobe16(f->fs.val.dport);
8072	fwr->lpm = htobe16(f->fs.mask.dport);
8073	fwr->fp = htobe16(f->fs.val.sport);
8074	fwr->fpm = htobe16(f->fs.mask.sport);
8075	if (f->fs.newsmac)
8076		bcopy(f->fs.smac, fwr->sma, sizeof (fwr->sma));
8077
8078	f->pending = 1;
8079	sc->tids.ftids_in_use++;
8080
8081	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8082	return (0);
8083}
8084
8085static int
8086del_filter_wr(struct adapter *sc, int fidx)
8087{
8088	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8089	struct fw_filter_wr *fwr;
8090	unsigned int ftid;
8091	struct wrq_cookie cookie;
8092
8093	ftid = sc->tids.ftid_base + fidx;
8094
8095	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8096	if (fwr == NULL)
8097		return (ENOMEM);
8098	bzero(fwr, sizeof (*fwr));
8099
8100	t4_mk_filtdelwr(ftid, fwr, sc->sge.fwq.abs_id);
8101
8102	f->pending = 1;
8103	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8104	return (0);
8105}
8106
8107int
8108t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8109{
8110	struct adapter *sc = iq->adapter;
8111	const struct cpl_set_tcb_rpl *rpl = (const void *)(rss + 1);
8112	unsigned int idx = GET_TID(rpl);
8113	unsigned int rc;
8114	struct filter_entry *f;
8115
8116	KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
8117	    rss->opcode));
8118	MPASS(iq == &sc->sge.fwq);
8119	MPASS(is_ftid(sc, idx));
8120
8121	idx -= sc->tids.ftid_base;
8122	f = &sc->tids.ftid_tab[idx];
8123	rc = G_COOKIE(rpl->cookie);
8124
8125	mtx_lock(&sc->tids.ftid_lock);
8126	if (rc == FW_FILTER_WR_FLT_ADDED) {
8127		KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
8128		    __func__, idx));
8129		f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
8130		f->pending = 0;  /* asynchronous setup completed */
8131		f->valid = 1;
8132	} else {
8133		if (rc != FW_FILTER_WR_FLT_DELETED) {
8134			/* Add or delete failed, display an error */
8135			log(LOG_ERR,
8136			    "filter %u setup failed with error %u\n",
8137			    idx, rc);
8138		}
8139
8140		clear_filter(f);
8141		sc->tids.ftids_in_use--;
8142	}
8143	wakeup(&sc->tids.ftid_tab);
8144	mtx_unlock(&sc->tids.ftid_lock);
8145
8146	return (0);
8147}
8148
8149static int
8150set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8151{
8152
8153	MPASS(iq->set_tcb_rpl != NULL);
8154	return (iq->set_tcb_rpl(iq, rss, m));
8155}
8156
8157static int
8158l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8159{
8160
8161	MPASS(iq->l2t_write_rpl != NULL);
8162	return (iq->l2t_write_rpl(iq, rss, m));
8163}
8164
8165static int
8166get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
8167{
8168	int rc;
8169
8170	if (cntxt->cid > M_CTXTQID)
8171		return (EINVAL);
8172
8173	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
8174	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
8175		return (EINVAL);
8176
8177	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
8178	if (rc)
8179		return (rc);
8180
8181	if (sc->flags & FW_OK) {
8182		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
8183		    &cntxt->data[0]);
8184		if (rc == 0)
8185			goto done;
8186	}
8187
8188	/*
8189	 * Read via firmware failed or wasn't even attempted.  Read directly via
8190	 * the backdoor.
8191	 */
8192	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
8193done:
8194	end_synchronized_op(sc, 0);
8195	return (rc);
8196}
8197
8198static int
8199load_fw(struct adapter *sc, struct t4_data *fw)
8200{
8201	int rc;
8202	uint8_t *fw_data;
8203
8204	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
8205	if (rc)
8206		return (rc);
8207
8208	if (sc->flags & FULL_INIT_DONE) {
8209		rc = EBUSY;
8210		goto done;
8211	}
8212
8213	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
8214	if (fw_data == NULL) {
8215		rc = ENOMEM;
8216		goto done;
8217	}
8218
8219	rc = copyin(fw->data, fw_data, fw->len);
8220	if (rc == 0)
8221		rc = -t4_load_fw(sc, fw_data, fw->len);
8222
8223	free(fw_data, M_CXGBE);
8224done:
8225	end_synchronized_op(sc, 0);
8226	return (rc);
8227}
8228
8229#define MAX_READ_BUF_SIZE (128 * 1024)
8230static int
8231read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
8232{
8233	uint32_t addr, remaining, n;
8234	uint32_t *buf;
8235	int rc;
8236	uint8_t *dst;
8237
8238	rc = validate_mem_range(sc, mr->addr, mr->len);
8239	if (rc != 0)
8240		return (rc);
8241
8242	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
8243	addr = mr->addr;
8244	remaining = mr->len;
8245	dst = (void *)mr->data;
8246
8247	while (remaining) {
8248		n = min(remaining, MAX_READ_BUF_SIZE);
8249		read_via_memwin(sc, 2, addr, buf, n);
8250
8251		rc = copyout(buf, dst, n);
8252		if (rc != 0)
8253			break;
8254
8255		dst += n;
8256		remaining -= n;
8257		addr += n;
8258	}
8259
8260	free(buf, M_CXGBE);
8261	return (rc);
8262}
8263#undef MAX_READ_BUF_SIZE
8264
8265static int
8266read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
8267{
8268	int rc;
8269
8270	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
8271		return (EINVAL);
8272
8273	if (i2cd->len > sizeof(i2cd->data))
8274		return (EFBIG);
8275
8276	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
8277	if (rc)
8278		return (rc);
8279	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
8280	    i2cd->offset, i2cd->len, &i2cd->data[0]);
8281	end_synchronized_op(sc, 0);
8282
8283	return (rc);
8284}
8285
8286static int
8287in_range(int val, int lo, int hi)
8288{
8289
8290	return (val < 0 || (val <= hi && val >= lo));
8291}
8292
8293static int
8294set_sched_class_config(struct adapter *sc, int minmax)
8295{
8296	int rc;
8297
8298	if (minmax < 0)
8299		return (EINVAL);
8300
8301	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4sscc");
8302	if (rc)
8303		return (rc);
8304	rc = -t4_sched_config(sc, FW_SCHED_TYPE_PKTSCHED, minmax, 1);
8305	end_synchronized_op(sc, 0);
8306
8307	return (rc);
8308}
8309
8310static int
8311set_sched_class_params(struct adapter *sc, struct t4_sched_class_params *p,
8312    int sleep_ok)
8313{
8314	int rc, top_speed, fw_level, fw_mode, fw_rateunit, fw_ratemode;
8315	struct port_info *pi;
8316	struct tx_sched_class *tc;
8317
8318	if (p->level == SCHED_CLASS_LEVEL_CL_RL)
8319		fw_level = FW_SCHED_PARAMS_LEVEL_CL_RL;
8320	else if (p->level == SCHED_CLASS_LEVEL_CL_WRR)
8321		fw_level = FW_SCHED_PARAMS_LEVEL_CL_WRR;
8322	else if (p->level == SCHED_CLASS_LEVEL_CH_RL)
8323		fw_level = FW_SCHED_PARAMS_LEVEL_CH_RL;
8324	else
8325		return (EINVAL);
8326
8327	if (p->mode == SCHED_CLASS_MODE_CLASS)
8328		fw_mode = FW_SCHED_PARAMS_MODE_CLASS;
8329	else if (p->mode == SCHED_CLASS_MODE_FLOW)
8330		fw_mode = FW_SCHED_PARAMS_MODE_FLOW;
8331	else
8332		return (EINVAL);
8333
8334	if (p->rateunit == SCHED_CLASS_RATEUNIT_BITS)
8335		fw_rateunit = FW_SCHED_PARAMS_UNIT_BITRATE;
8336	else if (p->rateunit == SCHED_CLASS_RATEUNIT_PKTS)
8337		fw_rateunit = FW_SCHED_PARAMS_UNIT_PKTRATE;
8338	else
8339		return (EINVAL);
8340
8341	if (p->ratemode == SCHED_CLASS_RATEMODE_REL)
8342		fw_ratemode = FW_SCHED_PARAMS_RATE_REL;
8343	else if (p->ratemode == SCHED_CLASS_RATEMODE_ABS)
8344		fw_ratemode = FW_SCHED_PARAMS_RATE_ABS;
8345	else
8346		return (EINVAL);
8347
8348	/* Vet our parameters ... */
8349	if (!in_range(p->channel, 0, sc->chip_params->nchan - 1))
8350		return (ERANGE);
8351
8352	pi = sc->port[sc->chan_map[p->channel]];
8353	if (pi == NULL)
8354		return (ENXIO);
8355	MPASS(pi->tx_chan == p->channel);
8356	top_speed = port_top_speed(pi) * 1000000; /* Gbps -> Kbps */
8357
8358	if (!in_range(p->cl, 0, sc->chip_params->nsched_cls) ||
8359	    !in_range(p->minrate, 0, top_speed) ||
8360	    !in_range(p->maxrate, 0, top_speed) ||
8361	    !in_range(p->weight, 0, 100))
8362		return (ERANGE);
8363
8364	/*
8365	 * Translate any unset parameters into the firmware's
8366	 * nomenclature and/or fail the call if the parameters
8367	 * are required ...
8368	 */
8369	if (p->rateunit < 0 || p->ratemode < 0 || p->channel < 0 || p->cl < 0)
8370		return (EINVAL);
8371
8372	if (p->minrate < 0)
8373		p->minrate = 0;
8374	if (p->maxrate < 0) {
8375		if (p->level == SCHED_CLASS_LEVEL_CL_RL ||
8376		    p->level == SCHED_CLASS_LEVEL_CH_RL)
8377			return (EINVAL);
8378		else
8379			p->maxrate = 0;
8380	}
8381	if (p->weight < 0) {
8382		if (p->level == SCHED_CLASS_LEVEL_CL_WRR)
8383			return (EINVAL);
8384		else
8385			p->weight = 0;
8386	}
8387	if (p->pktsize < 0) {
8388		if (p->level == SCHED_CLASS_LEVEL_CL_RL ||
8389		    p->level == SCHED_CLASS_LEVEL_CH_RL)
8390			return (EINVAL);
8391		else
8392			p->pktsize = 0;
8393	}
8394
8395	rc = begin_synchronized_op(sc, NULL,
8396	    sleep_ok ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4sscp");
8397	if (rc)
8398		return (rc);
8399	tc = &pi->tc[p->cl];
8400	tc->params = *p;
8401	rc = -t4_sched_params(sc, FW_SCHED_TYPE_PKTSCHED, fw_level, fw_mode,
8402	    fw_rateunit, fw_ratemode, p->channel, p->cl, p->minrate, p->maxrate,
8403	    p->weight, p->pktsize, sleep_ok);
8404	if (rc == 0)
8405		tc->flags |= TX_SC_OK;
8406	else {
8407		/*
8408		 * Unknown state at this point, see tc->params for what was
8409		 * attempted.
8410		 */
8411		tc->flags &= ~TX_SC_OK;
8412	}
8413	end_synchronized_op(sc, sleep_ok ? 0 : LOCK_HELD);
8414
8415	return (rc);
8416}
8417
8418int
8419t4_set_sched_class(struct adapter *sc, struct t4_sched_params *p)
8420{
8421
8422	if (p->type != SCHED_CLASS_TYPE_PACKET)
8423		return (EINVAL);
8424
8425	if (p->subcmd == SCHED_CLASS_SUBCMD_CONFIG)
8426		return (set_sched_class_config(sc, p->u.config.minmax));
8427
8428	if (p->subcmd == SCHED_CLASS_SUBCMD_PARAMS)
8429		return (set_sched_class_params(sc, &p->u.params, 1));
8430
8431	return (EINVAL);
8432}
8433
8434int
8435t4_set_sched_queue(struct adapter *sc, struct t4_sched_queue *p)
8436{
8437	struct port_info *pi = NULL;
8438	struct vi_info *vi;
8439	struct sge_txq *txq;
8440	uint32_t fw_mnem, fw_queue, fw_class;
8441	int i, rc;
8442
8443	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setsq");
8444	if (rc)
8445		return (rc);
8446
8447	if (p->port >= sc->params.nports) {
8448		rc = EINVAL;
8449		goto done;
8450	}
8451
8452	/* XXX: Only supported for the main VI. */
8453	pi = sc->port[p->port];
8454	vi = &pi->vi[0];
8455	if (!(vi->flags & VI_INIT_DONE)) {
8456		/* tx queues not set up yet */
8457		rc = EAGAIN;
8458		goto done;
8459	}
8460
8461	if (!in_range(p->queue, 0, vi->ntxq - 1) ||
8462	    !in_range(p->cl, 0, sc->chip_params->nsched_cls - 1)) {
8463		rc = EINVAL;
8464		goto done;
8465	}
8466
8467	/*
8468	 * Create a template for the FW_PARAMS_CMD mnemonic and value (TX
8469	 * Scheduling Class in this case).
8470	 */
8471	fw_mnem = (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
8472	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_EQ_SCHEDCLASS_ETH));
8473	fw_class = p->cl < 0 ? 0xffffffff : p->cl;
8474
8475	/*
8476	 * If op.queue is non-negative, then we're only changing the scheduling
8477	 * on a single specified TX queue.
8478	 */
8479	if (p->queue >= 0) {
8480		txq = &sc->sge.txq[vi->first_txq + p->queue];
8481		fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
8482		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
8483		    &fw_class);
8484		goto done;
8485	}
8486
8487	/*
8488	 * Change the scheduling on all the TX queues for the
8489	 * interface.
8490	 */
8491	for_each_txq(vi, i, txq) {
8492		fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
8493		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
8494		    &fw_class);
8495		if (rc)
8496			goto done;
8497	}
8498
8499	rc = 0;
8500done:
8501	end_synchronized_op(sc, 0);
8502	return (rc);
8503}
8504
8505int
8506t4_os_find_pci_capability(struct adapter *sc, int cap)
8507{
8508	int i;
8509
8510	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
8511}
8512
8513int
8514t4_os_pci_save_state(struct adapter *sc)
8515{
8516	device_t dev;
8517	struct pci_devinfo *dinfo;
8518
8519	dev = sc->dev;
8520	dinfo = device_get_ivars(dev);
8521
8522	pci_cfg_save(dev, dinfo, 0);
8523	return (0);
8524}
8525
8526int
8527t4_os_pci_restore_state(struct adapter *sc)
8528{
8529	device_t dev;
8530	struct pci_devinfo *dinfo;
8531
8532	dev = sc->dev;
8533	dinfo = device_get_ivars(dev);
8534
8535	pci_cfg_restore(dev, dinfo);
8536	return (0);
8537}
8538
8539void
8540t4_os_portmod_changed(const struct adapter *sc, int idx)
8541{
8542	struct port_info *pi = sc->port[idx];
8543	struct vi_info *vi;
8544	struct ifnet *ifp;
8545	int v;
8546	static const char *mod_str[] = {
8547		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
8548	};
8549
8550	for_each_vi(pi, v, vi) {
8551		build_medialist(pi, &vi->media);
8552	}
8553
8554	ifp = pi->vi[0].ifp;
8555	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
8556		if_printf(ifp, "transceiver unplugged.\n");
8557	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
8558		if_printf(ifp, "unknown transceiver inserted.\n");
8559	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
8560		if_printf(ifp, "unsupported transceiver inserted.\n");
8561	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
8562		if_printf(ifp, "%s transceiver inserted.\n",
8563		    mod_str[pi->mod_type]);
8564	} else {
8565		if_printf(ifp, "transceiver (type %d) inserted.\n",
8566		    pi->mod_type);
8567	}
8568}
8569
8570void
8571t4_os_link_changed(struct adapter *sc, int idx, int link_stat, int reason)
8572{
8573	struct port_info *pi = sc->port[idx];
8574	struct vi_info *vi;
8575	struct ifnet *ifp;
8576	int v;
8577
8578	if (link_stat)
8579		pi->linkdnrc = -1;
8580	else {
8581		if (reason >= 0)
8582			pi->linkdnrc = reason;
8583	}
8584	for_each_vi(pi, v, vi) {
8585		ifp = vi->ifp;
8586		if (ifp == NULL)
8587			continue;
8588
8589		if (link_stat) {
8590			ifp->if_baudrate = IF_Mbps(pi->link_cfg.speed);
8591			if_link_state_change(ifp, LINK_STATE_UP);
8592		} else {
8593			if_link_state_change(ifp, LINK_STATE_DOWN);
8594		}
8595	}
8596}
8597
8598void
8599t4_iterate(void (*func)(struct adapter *, void *), void *arg)
8600{
8601	struct adapter *sc;
8602
8603	sx_slock(&t4_list_lock);
8604	SLIST_FOREACH(sc, &t4_list, link) {
8605		/*
8606		 * func should not make any assumptions about what state sc is
8607		 * in - the only guarantee is that sc->sc_lock is a valid lock.
8608		 */
8609		func(sc, arg);
8610	}
8611	sx_sunlock(&t4_list_lock);
8612}
8613
8614static int
8615t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
8616    struct thread *td)
8617{
8618	int rc;
8619	struct adapter *sc = dev->si_drv1;
8620
8621	rc = priv_check(td, PRIV_DRIVER);
8622	if (rc != 0)
8623		return (rc);
8624
8625	switch (cmd) {
8626	case CHELSIO_T4_GETREG: {
8627		struct t4_reg *edata = (struct t4_reg *)data;
8628
8629		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
8630			return (EFAULT);
8631
8632		if (edata->size == 4)
8633			edata->val = t4_read_reg(sc, edata->addr);
8634		else if (edata->size == 8)
8635			edata->val = t4_read_reg64(sc, edata->addr);
8636		else
8637			return (EINVAL);
8638
8639		break;
8640	}
8641	case CHELSIO_T4_SETREG: {
8642		struct t4_reg *edata = (struct t4_reg *)data;
8643
8644		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
8645			return (EFAULT);
8646
8647		if (edata->size == 4) {
8648			if (edata->val & 0xffffffff00000000)
8649				return (EINVAL);
8650			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
8651		} else if (edata->size == 8)
8652			t4_write_reg64(sc, edata->addr, edata->val);
8653		else
8654			return (EINVAL);
8655		break;
8656	}
8657	case CHELSIO_T4_REGDUMP: {
8658		struct t4_regdump *regs = (struct t4_regdump *)data;
8659		int reglen = t4_get_regs_len(sc);
8660		uint8_t *buf;
8661
8662		if (regs->len < reglen) {
8663			regs->len = reglen; /* hint to the caller */
8664			return (ENOBUFS);
8665		}
8666
8667		regs->len = reglen;
8668		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
8669		get_regs(sc, regs, buf);
8670		rc = copyout(buf, regs->data, reglen);
8671		free(buf, M_CXGBE);
8672		break;
8673	}
8674	case CHELSIO_T4_GET_FILTER_MODE:
8675		rc = get_filter_mode(sc, (uint32_t *)data);
8676		break;
8677	case CHELSIO_T4_SET_FILTER_MODE:
8678		rc = set_filter_mode(sc, *(uint32_t *)data);
8679		break;
8680	case CHELSIO_T4_GET_FILTER:
8681		rc = get_filter(sc, (struct t4_filter *)data);
8682		break;
8683	case CHELSIO_T4_SET_FILTER:
8684		rc = set_filter(sc, (struct t4_filter *)data);
8685		break;
8686	case CHELSIO_T4_DEL_FILTER:
8687		rc = del_filter(sc, (struct t4_filter *)data);
8688		break;
8689	case CHELSIO_T4_GET_SGE_CONTEXT:
8690		rc = get_sge_context(sc, (struct t4_sge_context *)data);
8691		break;
8692	case CHELSIO_T4_LOAD_FW:
8693		rc = load_fw(sc, (struct t4_data *)data);
8694		break;
8695	case CHELSIO_T4_GET_MEM:
8696		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
8697		break;
8698	case CHELSIO_T4_GET_I2C:
8699		rc = read_i2c(sc, (struct t4_i2c_data *)data);
8700		break;
8701	case CHELSIO_T4_CLEAR_STATS: {
8702		int i, v;
8703		u_int port_id = *(uint32_t *)data;
8704		struct port_info *pi;
8705		struct vi_info *vi;
8706
8707		if (port_id >= sc->params.nports)
8708			return (EINVAL);
8709		pi = sc->port[port_id];
8710
8711		/* MAC stats */
8712		t4_clr_port_stats(sc, pi->tx_chan);
8713		pi->tx_parse_error = 0;
8714		mtx_lock(&sc->reg_lock);
8715		for_each_vi(pi, v, vi) {
8716			if (vi->flags & VI_INIT_DONE)
8717				t4_clr_vi_stats(sc, vi->viid);
8718		}
8719		mtx_unlock(&sc->reg_lock);
8720
8721		/*
8722		 * Since this command accepts a port, clear stats for
8723		 * all VIs on this port.
8724		 */
8725		for_each_vi(pi, v, vi) {
8726			if (vi->flags & VI_INIT_DONE) {
8727				struct sge_rxq *rxq;
8728				struct sge_txq *txq;
8729				struct sge_wrq *wrq;
8730
8731				for_each_rxq(vi, i, rxq) {
8732#if defined(INET) || defined(INET6)
8733					rxq->lro.lro_queued = 0;
8734					rxq->lro.lro_flushed = 0;
8735#endif
8736					rxq->rxcsum = 0;
8737					rxq->vlan_extraction = 0;
8738				}
8739
8740				for_each_txq(vi, i, txq) {
8741					txq->txcsum = 0;
8742					txq->tso_wrs = 0;
8743					txq->vlan_insertion = 0;
8744					txq->imm_wrs = 0;
8745					txq->sgl_wrs = 0;
8746					txq->txpkt_wrs = 0;
8747					txq->txpkts0_wrs = 0;
8748					txq->txpkts1_wrs = 0;
8749					txq->txpkts0_pkts = 0;
8750					txq->txpkts1_pkts = 0;
8751					mp_ring_reset_stats(txq->r);
8752				}
8753
8754#ifdef TCP_OFFLOAD
8755				/* nothing to clear for each ofld_rxq */
8756
8757				for_each_ofld_txq(vi, i, wrq) {
8758					wrq->tx_wrs_direct = 0;
8759					wrq->tx_wrs_copied = 0;
8760				}
8761#endif
8762
8763				if (IS_MAIN_VI(vi)) {
8764					wrq = &sc->sge.ctrlq[pi->port_id];
8765					wrq->tx_wrs_direct = 0;
8766					wrq->tx_wrs_copied = 0;
8767				}
8768			}
8769		}
8770		break;
8771	}
8772	case CHELSIO_T4_SCHED_CLASS:
8773		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
8774		break;
8775	case CHELSIO_T4_SCHED_QUEUE:
8776		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
8777		break;
8778	case CHELSIO_T4_GET_TRACER:
8779		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
8780		break;
8781	case CHELSIO_T4_SET_TRACER:
8782		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
8783		break;
8784	default:
8785		rc = ENOTTY;
8786	}
8787
8788	return (rc);
8789}
8790
8791void
8792t4_db_full(struct adapter *sc)
8793{
8794
8795	CXGBE_UNIMPLEMENTED(__func__);
8796}
8797
8798void
8799t4_db_dropped(struct adapter *sc)
8800{
8801
8802	CXGBE_UNIMPLEMENTED(__func__);
8803}
8804
8805#ifdef TCP_OFFLOAD
8806void
8807t4_iscsi_init(struct adapter *sc, u_int tag_mask, const u_int *pgsz_order)
8808{
8809
8810	t4_write_reg(sc, A_ULP_RX_ISCSI_TAGMASK, tag_mask);
8811	t4_write_reg(sc, A_ULP_RX_ISCSI_PSZ, V_HPZ0(pgsz_order[0]) |
8812		V_HPZ1(pgsz_order[1]) | V_HPZ2(pgsz_order[2]) |
8813		V_HPZ3(pgsz_order[3]));
8814}
8815
8816static int
8817toe_capability(struct vi_info *vi, int enable)
8818{
8819	int rc;
8820	struct port_info *pi = vi->pi;
8821	struct adapter *sc = pi->adapter;
8822
8823	ASSERT_SYNCHRONIZED_OP(sc);
8824
8825	if (!is_offload(sc))
8826		return (ENODEV);
8827
8828	if (enable) {
8829		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
8830			/* TOE is already enabled. */
8831			return (0);
8832		}
8833
8834		/*
8835		 * We need the port's queues around so that we're able to send
8836		 * and receive CPLs to/from the TOE even if the ifnet for this
8837		 * port has never been UP'd administratively.
8838		 */
8839		if (!(vi->flags & VI_INIT_DONE)) {
8840			rc = vi_full_init(vi);
8841			if (rc)
8842				return (rc);
8843		}
8844		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
8845			rc = vi_full_init(&pi->vi[0]);
8846			if (rc)
8847				return (rc);
8848		}
8849
8850		if (isset(&sc->offload_map, pi->port_id)) {
8851			/* TOE is enabled on another VI of this port. */
8852			pi->uld_vis++;
8853			return (0);
8854		}
8855
8856		if (!uld_active(sc, ULD_TOM)) {
8857			rc = t4_activate_uld(sc, ULD_TOM);
8858			if (rc == EAGAIN) {
8859				log(LOG_WARNING,
8860				    "You must kldload t4_tom.ko before trying "
8861				    "to enable TOE on a cxgbe interface.\n");
8862			}
8863			if (rc != 0)
8864				return (rc);
8865			KASSERT(sc->tom_softc != NULL,
8866			    ("%s: TOM activated but softc NULL", __func__));
8867			KASSERT(uld_active(sc, ULD_TOM),
8868			    ("%s: TOM activated but flag not set", __func__));
8869		}
8870
8871		/* Activate iWARP and iSCSI too, if the modules are loaded. */
8872		if (!uld_active(sc, ULD_IWARP))
8873			(void) t4_activate_uld(sc, ULD_IWARP);
8874		if (!uld_active(sc, ULD_ISCSI))
8875			(void) t4_activate_uld(sc, ULD_ISCSI);
8876
8877		pi->uld_vis++;
8878		setbit(&sc->offload_map, pi->port_id);
8879	} else {
8880		pi->uld_vis--;
8881
8882		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
8883			return (0);
8884
8885		KASSERT(uld_active(sc, ULD_TOM),
8886		    ("%s: TOM never initialized?", __func__));
8887		clrbit(&sc->offload_map, pi->port_id);
8888	}
8889
8890	return (0);
8891}
8892
8893/*
8894 * Add an upper layer driver to the global list.
8895 */
8896int
8897t4_register_uld(struct uld_info *ui)
8898{
8899	int rc = 0;
8900	struct uld_info *u;
8901
8902	sx_xlock(&t4_uld_list_lock);
8903	SLIST_FOREACH(u, &t4_uld_list, link) {
8904	    if (u->uld_id == ui->uld_id) {
8905		    rc = EEXIST;
8906		    goto done;
8907	    }
8908	}
8909
8910	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
8911	ui->refcount = 0;
8912done:
8913	sx_xunlock(&t4_uld_list_lock);
8914	return (rc);
8915}
8916
8917int
8918t4_unregister_uld(struct uld_info *ui)
8919{
8920	int rc = EINVAL;
8921	struct uld_info *u;
8922
8923	sx_xlock(&t4_uld_list_lock);
8924
8925	SLIST_FOREACH(u, &t4_uld_list, link) {
8926	    if (u == ui) {
8927		    if (ui->refcount > 0) {
8928			    rc = EBUSY;
8929			    goto done;
8930		    }
8931
8932		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
8933		    rc = 0;
8934		    goto done;
8935	    }
8936	}
8937done:
8938	sx_xunlock(&t4_uld_list_lock);
8939	return (rc);
8940}
8941
8942int
8943t4_activate_uld(struct adapter *sc, int id)
8944{
8945	int rc;
8946	struct uld_info *ui;
8947
8948	ASSERT_SYNCHRONIZED_OP(sc);
8949
8950	if (id < 0 || id > ULD_MAX)
8951		return (EINVAL);
8952	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
8953
8954	sx_slock(&t4_uld_list_lock);
8955
8956	SLIST_FOREACH(ui, &t4_uld_list, link) {
8957		if (ui->uld_id == id) {
8958			if (!(sc->flags & FULL_INIT_DONE)) {
8959				rc = adapter_full_init(sc);
8960				if (rc != 0)
8961					break;
8962			}
8963
8964			rc = ui->activate(sc);
8965			if (rc == 0) {
8966				setbit(&sc->active_ulds, id);
8967				ui->refcount++;
8968			}
8969			break;
8970		}
8971	}
8972
8973	sx_sunlock(&t4_uld_list_lock);
8974
8975	return (rc);
8976}
8977
8978int
8979t4_deactivate_uld(struct adapter *sc, int id)
8980{
8981	int rc;
8982	struct uld_info *ui;
8983
8984	ASSERT_SYNCHRONIZED_OP(sc);
8985
8986	if (id < 0 || id > ULD_MAX)
8987		return (EINVAL);
8988	rc = ENXIO;
8989
8990	sx_slock(&t4_uld_list_lock);
8991
8992	SLIST_FOREACH(ui, &t4_uld_list, link) {
8993		if (ui->uld_id == id) {
8994			rc = ui->deactivate(sc);
8995			if (rc == 0) {
8996				clrbit(&sc->active_ulds, id);
8997				ui->refcount--;
8998			}
8999			break;
9000		}
9001	}
9002
9003	sx_sunlock(&t4_uld_list_lock);
9004
9005	return (rc);
9006}
9007
9008int
9009uld_active(struct adapter *sc, int uld_id)
9010{
9011
9012	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9013
9014	return (isset(&sc->active_ulds, uld_id));
9015}
9016#endif
9017
9018/*
9019 * Come up with reasonable defaults for some of the tunables, provided they're
9020 * not set by the user (in which case we'll use the values as is).
9021 */
9022static void
9023tweak_tunables(void)
9024{
9025	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
9026
9027	if (t4_ntxq10g < 1) {
9028#ifdef RSS
9029		t4_ntxq10g = rss_getnumbuckets();
9030#else
9031		t4_ntxq10g = min(nc, NTXQ_10G);
9032#endif
9033	}
9034
9035	if (t4_ntxq1g < 1) {
9036#ifdef RSS
9037		/* XXX: way too many for 1GbE? */
9038		t4_ntxq1g = rss_getnumbuckets();
9039#else
9040		t4_ntxq1g = min(nc, NTXQ_1G);
9041#endif
9042	}
9043
9044	if (t4_ntxq_vi < 1)
9045		t4_ntxq_vi = min(nc, NTXQ_VI);
9046
9047	if (t4_nrxq10g < 1) {
9048#ifdef RSS
9049		t4_nrxq10g = rss_getnumbuckets();
9050#else
9051		t4_nrxq10g = min(nc, NRXQ_10G);
9052#endif
9053	}
9054
9055	if (t4_nrxq1g < 1) {
9056#ifdef RSS
9057		/* XXX: way too many for 1GbE? */
9058		t4_nrxq1g = rss_getnumbuckets();
9059#else
9060		t4_nrxq1g = min(nc, NRXQ_1G);
9061#endif
9062	}
9063
9064	if (t4_nrxq_vi < 1)
9065		t4_nrxq_vi = min(nc, NRXQ_VI);
9066
9067#ifdef TCP_OFFLOAD
9068	if (t4_nofldtxq10g < 1)
9069		t4_nofldtxq10g = min(nc, NOFLDTXQ_10G);
9070
9071	if (t4_nofldtxq1g < 1)
9072		t4_nofldtxq1g = min(nc, NOFLDTXQ_1G);
9073
9074	if (t4_nofldtxq_vi < 1)
9075		t4_nofldtxq_vi = min(nc, NOFLDTXQ_VI);
9076
9077	if (t4_nofldrxq10g < 1)
9078		t4_nofldrxq10g = min(nc, NOFLDRXQ_10G);
9079
9080	if (t4_nofldrxq1g < 1)
9081		t4_nofldrxq1g = min(nc, NOFLDRXQ_1G);
9082
9083	if (t4_nofldrxq_vi < 1)
9084		t4_nofldrxq_vi = min(nc, NOFLDRXQ_VI);
9085
9086	if (t4_toecaps_allowed == -1)
9087		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9088
9089	if (t4_rdmacaps_allowed == -1) {
9090		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9091		    FW_CAPS_CONFIG_RDMA_RDMAC;
9092	}
9093
9094	if (t4_iscsicaps_allowed == -1) {
9095		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9096		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9097		    FW_CAPS_CONFIG_ISCSI_T10DIF;
9098	}
9099#else
9100	if (t4_toecaps_allowed == -1)
9101		t4_toecaps_allowed = 0;
9102
9103	if (t4_rdmacaps_allowed == -1)
9104		t4_rdmacaps_allowed = 0;
9105
9106	if (t4_iscsicaps_allowed == -1)
9107		t4_iscsicaps_allowed = 0;
9108#endif
9109
9110#ifdef DEV_NETMAP
9111	if (t4_nnmtxq_vi < 1)
9112		t4_nnmtxq_vi = min(nc, NNMTXQ_VI);
9113
9114	if (t4_nnmrxq_vi < 1)
9115		t4_nnmrxq_vi = min(nc, NNMRXQ_VI);
9116#endif
9117
9118	if (t4_tmr_idx_10g < 0 || t4_tmr_idx_10g >= SGE_NTIMERS)
9119		t4_tmr_idx_10g = TMR_IDX_10G;
9120
9121	if (t4_pktc_idx_10g < -1 || t4_pktc_idx_10g >= SGE_NCOUNTERS)
9122		t4_pktc_idx_10g = PKTC_IDX_10G;
9123
9124	if (t4_tmr_idx_1g < 0 || t4_tmr_idx_1g >= SGE_NTIMERS)
9125		t4_tmr_idx_1g = TMR_IDX_1G;
9126
9127	if (t4_pktc_idx_1g < -1 || t4_pktc_idx_1g >= SGE_NCOUNTERS)
9128		t4_pktc_idx_1g = PKTC_IDX_1G;
9129
9130	if (t4_qsize_txq < 128)
9131		t4_qsize_txq = 128;
9132
9133	if (t4_qsize_rxq < 128)
9134		t4_qsize_rxq = 128;
9135	while (t4_qsize_rxq & 7)
9136		t4_qsize_rxq++;
9137
9138	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9139}
9140
9141#ifdef DDB
9142static void
9143t4_dump_tcb(struct adapter *sc, int tid)
9144{
9145	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
9146
9147	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
9148	save = t4_read_reg(sc, reg);
9149	base = sc->memwin[2].mw_base;
9150
9151	/* Dump TCB for the tid */
9152	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
9153	tcb_addr += tid * TCB_SIZE;
9154
9155	if (is_t4(sc)) {
9156		pf = 0;
9157		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
9158	} else {
9159		pf = V_PFNUM(sc->pf);
9160		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
9161	}
9162	t4_write_reg(sc, reg, win_pos | pf);
9163	t4_read_reg(sc, reg);
9164
9165	off = tcb_addr - win_pos;
9166	for (i = 0; i < 4; i++) {
9167		uint32_t buf[8];
9168		for (j = 0; j < 8; j++, off += 4)
9169			buf[j] = htonl(t4_read_reg(sc, base + off));
9170
9171		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
9172		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
9173		    buf[7]);
9174	}
9175
9176	t4_write_reg(sc, reg, save);
9177	t4_read_reg(sc, reg);
9178}
9179
9180static void
9181t4_dump_devlog(struct adapter *sc)
9182{
9183	struct devlog_params *dparams = &sc->params.devlog;
9184	struct fw_devlog_e e;
9185	int i, first, j, m, nentries, rc;
9186	uint64_t ftstamp = UINT64_MAX;
9187
9188	if (dparams->start == 0) {
9189		db_printf("devlog params not valid\n");
9190		return;
9191	}
9192
9193	nentries = dparams->size / sizeof(struct fw_devlog_e);
9194	m = fwmtype_to_hwmtype(dparams->memtype);
9195
9196	/* Find the first entry. */
9197	first = -1;
9198	for (i = 0; i < nentries && !db_pager_quit; i++) {
9199		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9200		    sizeof(e), (void *)&e);
9201		if (rc != 0)
9202			break;
9203
9204		if (e.timestamp == 0)
9205			break;
9206
9207		e.timestamp = be64toh(e.timestamp);
9208		if (e.timestamp < ftstamp) {
9209			ftstamp = e.timestamp;
9210			first = i;
9211		}
9212	}
9213
9214	if (first == -1)
9215		return;
9216
9217	i = first;
9218	do {
9219		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9220		    sizeof(e), (void *)&e);
9221		if (rc != 0)
9222			return;
9223
9224		if (e.timestamp == 0)
9225			return;
9226
9227		e.timestamp = be64toh(e.timestamp);
9228		e.seqno = be32toh(e.seqno);
9229		for (j = 0; j < 8; j++)
9230			e.params[j] = be32toh(e.params[j]);
9231
9232		db_printf("%10d  %15ju  %8s  %8s  ",
9233		    e.seqno, e.timestamp,
9234		    (e.level < nitems(devlog_level_strings) ?
9235			devlog_level_strings[e.level] : "UNKNOWN"),
9236		    (e.facility < nitems(devlog_facility_strings) ?
9237			devlog_facility_strings[e.facility] : "UNKNOWN"));
9238		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
9239		    e.params[3], e.params[4], e.params[5], e.params[6],
9240		    e.params[7]);
9241
9242		if (++i == nentries)
9243			i = 0;
9244	} while (i != first && !db_pager_quit);
9245}
9246
9247static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
9248_DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
9249
9250DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
9251{
9252	device_t dev;
9253	int t;
9254	bool valid;
9255
9256	valid = false;
9257	t = db_read_token();
9258	if (t == tIDENT) {
9259		dev = device_lookup_by_name(db_tok_string);
9260		valid = true;
9261	}
9262	db_skip_to_eol();
9263	if (!valid) {
9264		db_printf("usage: show t4 devlog <nexus>\n");
9265		return;
9266	}
9267
9268	if (dev == NULL) {
9269		db_printf("device not found\n");
9270		return;
9271	}
9272
9273	t4_dump_devlog(device_get_softc(dev));
9274}
9275
9276DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
9277{
9278	device_t dev;
9279	int radix, tid, t;
9280	bool valid;
9281
9282	valid = false;
9283	radix = db_radix;
9284	db_radix = 10;
9285	t = db_read_token();
9286	if (t == tIDENT) {
9287		dev = device_lookup_by_name(db_tok_string);
9288		t = db_read_token();
9289		if (t == tNUMBER) {
9290			tid = db_tok_number;
9291			valid = true;
9292		}
9293	}
9294	db_radix = radix;
9295	db_skip_to_eol();
9296	if (!valid) {
9297		db_printf("usage: show t4 tcb <nexus> <tid>\n");
9298		return;
9299	}
9300
9301	if (dev == NULL) {
9302		db_printf("device not found\n");
9303		return;
9304	}
9305	if (tid < 0) {
9306		db_printf("invalid tid\n");
9307		return;
9308	}
9309
9310	t4_dump_tcb(device_get_softc(dev), tid);
9311}
9312#endif
9313
9314static struct sx mlu;	/* mod load unload */
9315SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
9316
9317static int
9318mod_event(module_t mod, int cmd, void *arg)
9319{
9320	int rc = 0;
9321	static int loaded = 0;
9322
9323	switch (cmd) {
9324	case MOD_LOAD:
9325		sx_xlock(&mlu);
9326		if (loaded++ == 0) {
9327			t4_sge_modload();
9328			t4_register_cpl_handler(CPL_SET_TCB_RPL, set_tcb_rpl);
9329			t4_register_cpl_handler(CPL_L2T_WRITE_RPL, l2t_write_rpl);
9330			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
9331			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
9332			sx_init(&t4_list_lock, "T4/T5 adapters");
9333			SLIST_INIT(&t4_list);
9334#ifdef TCP_OFFLOAD
9335			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
9336			SLIST_INIT(&t4_uld_list);
9337#endif
9338			t4_tracer_modload();
9339			tweak_tunables();
9340		}
9341		sx_xunlock(&mlu);
9342		break;
9343
9344	case MOD_UNLOAD:
9345		sx_xlock(&mlu);
9346		if (--loaded == 0) {
9347			int tries;
9348
9349			sx_slock(&t4_list_lock);
9350			if (!SLIST_EMPTY(&t4_list)) {
9351				rc = EBUSY;
9352				sx_sunlock(&t4_list_lock);
9353				goto done_unload;
9354			}
9355#ifdef TCP_OFFLOAD
9356			sx_slock(&t4_uld_list_lock);
9357			if (!SLIST_EMPTY(&t4_uld_list)) {
9358				rc = EBUSY;
9359				sx_sunlock(&t4_uld_list_lock);
9360				sx_sunlock(&t4_list_lock);
9361				goto done_unload;
9362			}
9363#endif
9364			tries = 0;
9365			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
9366				uprintf("%ju clusters with custom free routine "
9367				    "still is use.\n", t4_sge_extfree_refs());
9368				pause("t4unload", 2 * hz);
9369			}
9370#ifdef TCP_OFFLOAD
9371			sx_sunlock(&t4_uld_list_lock);
9372#endif
9373			sx_sunlock(&t4_list_lock);
9374
9375			if (t4_sge_extfree_refs() == 0) {
9376				t4_tracer_modunload();
9377#ifdef TCP_OFFLOAD
9378				sx_destroy(&t4_uld_list_lock);
9379#endif
9380				sx_destroy(&t4_list_lock);
9381				t4_sge_modunload();
9382				loaded = 0;
9383			} else {
9384				rc = EBUSY;
9385				loaded++;	/* undo earlier decrement */
9386			}
9387		}
9388done_unload:
9389		sx_xunlock(&mlu);
9390		break;
9391	}
9392
9393	return (rc);
9394}
9395
9396static devclass_t t4_devclass, t5_devclass;
9397static devclass_t cxgbe_devclass, cxl_devclass;
9398static devclass_t vcxgbe_devclass, vcxl_devclass;
9399
9400DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
9401MODULE_VERSION(t4nex, 1);
9402MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
9403
9404DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
9405MODULE_VERSION(t5nex, 1);
9406MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
9407
9408DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
9409MODULE_VERSION(cxgbe, 1);
9410
9411DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
9412MODULE_VERSION(cxl, 1);
9413
9414DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
9415MODULE_VERSION(vcxgbe, 1);
9416
9417DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
9418MODULE_VERSION(vcxl, 1);
9419