t4_main.c revision 308321
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 308321 2016-11-04 21:59:27Z 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);
103static device_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;
132static d_open_t t4_open;
133static d_close_t t4_close;
134
135static struct cdevsw t4_cdevsw = {
136       .d_version = D_VERSION,
137       .d_flags = 0,
138       .d_open = t4_open,
139       .d_close = t4_close,
140       .d_ioctl = t4_ioctl,
141       .d_name = "t4nex",
142};
143
144/* T5 bus driver interface */
145static int t5_probe(device_t);
146static device_method_t t5_methods[] = {
147	DEVMETHOD(device_probe,		t5_probe),
148	DEVMETHOD(device_attach,	t4_attach),
149	DEVMETHOD(device_detach,	t4_detach),
150
151	DEVMETHOD_END
152};
153static driver_t t5_driver = {
154	"t5nex",
155	t5_methods,
156	sizeof(struct adapter)
157};
158
159
160/* T5 port (cxl) interface */
161static driver_t cxl_driver = {
162	"cxl",
163	cxgbe_methods,
164	sizeof(struct port_info)
165};
166
167/* T5 VI (vcxl) interface */
168static driver_t vcxl_driver = {
169	"vcxl",
170	vcxgbe_methods,
171	sizeof(struct vi_info)
172};
173
174static struct cdevsw t5_cdevsw = {
175       .d_version = D_VERSION,
176       .d_flags = 0,
177       .d_open = t4_open,
178       .d_close = t4_close,
179       .d_ioctl = t4_ioctl,
180       .d_name = "t5nex",
181};
182
183/* ifnet + media interface */
184static void cxgbe_init(void *);
185static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
186static int cxgbe_transmit(struct ifnet *, struct mbuf *);
187static void cxgbe_qflush(struct ifnet *);
188static int cxgbe_media_change(struct ifnet *);
189static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
190
191MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
192
193/*
194 * Correct lock order when you need to acquire multiple locks is t4_list_lock,
195 * then ADAPTER_LOCK, then t4_uld_list_lock.
196 */
197static struct sx t4_list_lock;
198SLIST_HEAD(, adapter) t4_list;
199#ifdef TCP_OFFLOAD
200static struct sx t4_uld_list_lock;
201SLIST_HEAD(, uld_info) t4_uld_list;
202#endif
203
204/*
205 * Tunables.  See tweak_tunables() too.
206 *
207 * Each tunable is set to a default value here if it's known at compile-time.
208 * Otherwise it is set to -1 as an indication to tweak_tunables() that it should
209 * provide a reasonable default when the driver is loaded.
210 *
211 * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
212 * T5 are under hw.cxl.
213 */
214
215/*
216 * Number of queues for tx and rx, 10G and 1G, NIC and offload.
217 */
218#define NTXQ_10G 16
219static int t4_ntxq10g = -1;
220TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq10g);
221
222#define NRXQ_10G 8
223static int t4_nrxq10g = -1;
224TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq10g);
225
226#define NTXQ_1G 4
227static int t4_ntxq1g = -1;
228TUNABLE_INT("hw.cxgbe.ntxq1g", &t4_ntxq1g);
229
230#define NRXQ_1G 2
231static int t4_nrxq1g = -1;
232TUNABLE_INT("hw.cxgbe.nrxq1g", &t4_nrxq1g);
233
234#define NTXQ_VI 1
235static int t4_ntxq_vi = -1;
236TUNABLE_INT("hw.cxgbe.ntxq_vi", &t4_ntxq_vi);
237
238#define NRXQ_VI 1
239static int t4_nrxq_vi = -1;
240TUNABLE_INT("hw.cxgbe.nrxq_vi", &t4_nrxq_vi);
241
242static int t4_rsrv_noflowq = 0;
243TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
244
245#ifdef TCP_OFFLOAD
246#define NOFLDTXQ_10G 8
247static int t4_nofldtxq10g = -1;
248TUNABLE_INT("hw.cxgbe.nofldtxq10g", &t4_nofldtxq10g);
249
250#define NOFLDRXQ_10G 2
251static int t4_nofldrxq10g = -1;
252TUNABLE_INT("hw.cxgbe.nofldrxq10g", &t4_nofldrxq10g);
253
254#define NOFLDTXQ_1G 2
255static int t4_nofldtxq1g = -1;
256TUNABLE_INT("hw.cxgbe.nofldtxq1g", &t4_nofldtxq1g);
257
258#define NOFLDRXQ_1G 1
259static int t4_nofldrxq1g = -1;
260TUNABLE_INT("hw.cxgbe.nofldrxq1g", &t4_nofldrxq1g);
261
262#define NOFLDTXQ_VI 1
263static int t4_nofldtxq_vi = -1;
264TUNABLE_INT("hw.cxgbe.nofldtxq_vi", &t4_nofldtxq_vi);
265
266#define NOFLDRXQ_VI 1
267static int t4_nofldrxq_vi = -1;
268TUNABLE_INT("hw.cxgbe.nofldrxq_vi", &t4_nofldrxq_vi);
269#endif
270
271#ifdef DEV_NETMAP
272#define NNMTXQ_VI 2
273static int t4_nnmtxq_vi = -1;
274TUNABLE_INT("hw.cxgbe.nnmtxq_vi", &t4_nnmtxq_vi);
275
276#define NNMRXQ_VI 2
277static int t4_nnmrxq_vi = -1;
278TUNABLE_INT("hw.cxgbe.nnmrxq_vi", &t4_nnmrxq_vi);
279#endif
280
281/*
282 * Holdoff parameters for 10G and 1G ports.
283 */
284#define TMR_IDX_10G 1
285static int t4_tmr_idx_10g = TMR_IDX_10G;
286TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx_10g);
287
288#define PKTC_IDX_10G (-1)
289static int t4_pktc_idx_10g = PKTC_IDX_10G;
290TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx_10g);
291
292#define TMR_IDX_1G 1
293static int t4_tmr_idx_1g = TMR_IDX_1G;
294TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_1G", &t4_tmr_idx_1g);
295
296#define PKTC_IDX_1G (-1)
297static int t4_pktc_idx_1g = PKTC_IDX_1G;
298TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_1G", &t4_pktc_idx_1g);
299
300/*
301 * Size (# of entries) of each tx and rx queue.
302 */
303static unsigned int t4_qsize_txq = TX_EQ_QSIZE;
304TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
305
306static unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
307TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
308
309/*
310 * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
311 */
312static int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
313TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
314
315/*
316 * Configuration file.
317 */
318#define DEFAULT_CF	"default"
319#define FLASH_CF	"flash"
320#define UWIRE_CF	"uwire"
321#define FPGA_CF		"fpga"
322static char t4_cfg_file[32] = DEFAULT_CF;
323TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
324
325/*
326 * PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
327 * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
328 * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
329 *            mark or when signalled to do so, 0 to never emit PAUSE.
330 */
331static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
332TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
333
334/*
335 * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
336 * encouraged respectively).
337 */
338static unsigned int t4_fw_install = 1;
339TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
340
341/*
342 * ASIC features that will be used.  Disable the ones you don't want so that the
343 * chip resources aren't wasted on features that will not be used.
344 */
345static int t4_nbmcaps_allowed = 0;
346TUNABLE_INT("hw.cxgbe.nbmcaps_allowed", &t4_nbmcaps_allowed);
347
348static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
349TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
350
351static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
352    FW_CAPS_CONFIG_SWITCH_EGRESS;
353TUNABLE_INT("hw.cxgbe.switchcaps_allowed", &t4_switchcaps_allowed);
354
355static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC;
356TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
357
358static int t4_toecaps_allowed = -1;
359TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
360
361static int t4_rdmacaps_allowed = -1;
362TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
363
364static int t4_tlscaps_allowed = 0;
365TUNABLE_INT("hw.cxgbe.tlscaps_allowed", &t4_tlscaps_allowed);
366
367static int t4_iscsicaps_allowed = -1;
368TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
369
370static int t4_fcoecaps_allowed = 0;
371TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
372
373static int t5_write_combine = 0;
374TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
375
376static int t4_num_vis = 1;
377TUNABLE_INT("hw.cxgbe.num_vis", &t4_num_vis);
378
379/* Functions used by extra VIs to obtain unique MAC addresses for each VI. */
380static int vi_mac_funcs[] = {
381	FW_VI_FUNC_OFLD,
382	FW_VI_FUNC_IWARP,
383	FW_VI_FUNC_OPENISCSI,
384	FW_VI_FUNC_OPENFCOE,
385	FW_VI_FUNC_FOISCSI,
386	FW_VI_FUNC_FOFCOE,
387};
388
389struct intrs_and_queues {
390	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
391	uint16_t nirq;		/* Total # of vectors */
392	uint16_t intr_flags_10g;/* Interrupt flags for each 10G port */
393	uint16_t intr_flags_1g;	/* Interrupt flags for each 1G port */
394	uint16_t ntxq10g;	/* # of NIC txq's for each 10G port */
395	uint16_t nrxq10g;	/* # of NIC rxq's for each 10G port */
396	uint16_t ntxq1g;	/* # of NIC txq's for each 1G port */
397	uint16_t nrxq1g;	/* # of NIC rxq's for each 1G port */
398	uint16_t rsrv_noflowq;	/* Flag whether to reserve queue 0 */
399	uint16_t nofldtxq10g;	/* # of TOE txq's for each 10G port */
400	uint16_t nofldrxq10g;	/* # of TOE rxq's for each 10G port */
401	uint16_t nofldtxq1g;	/* # of TOE txq's for each 1G port */
402	uint16_t nofldrxq1g;	/* # of TOE rxq's for each 1G port */
403
404	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
405	uint16_t ntxq_vi;	/* # of NIC txq's */
406	uint16_t nrxq_vi;	/* # of NIC rxq's */
407	uint16_t nofldtxq_vi;	/* # of TOE txq's */
408	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
409	uint16_t nnmtxq_vi;	/* # of netmap txq's */
410	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
411};
412
413struct filter_entry {
414        uint32_t valid:1;	/* filter allocated and valid */
415        uint32_t locked:1;	/* filter is administratively locked */
416        uint32_t pending:1;	/* filter action is pending firmware reply */
417	uint32_t smtidx:8;	/* Source MAC Table index for smac */
418	struct l2t_entry *l2t;	/* Layer Two Table entry for dmac */
419
420        struct t4_filter_specification fs;
421};
422
423static int map_bars_0_and_4(struct adapter *);
424static int map_bar_2(struct adapter *);
425static void setup_memwin(struct adapter *);
426static void position_memwin(struct adapter *, int, uint32_t);
427static int rw_via_memwin(struct adapter *, int, uint32_t, uint32_t *, int, int);
428static inline int read_via_memwin(struct adapter *, int, uint32_t, uint32_t *,
429    int);
430static inline int write_via_memwin(struct adapter *, int, uint32_t,
431    const uint32_t *, int);
432static int validate_mem_range(struct adapter *, uint32_t, int);
433static int fwmtype_to_hwmtype(int);
434static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
435    uint32_t *);
436static int fixup_devlog_params(struct adapter *);
437static int cfg_itype_and_nqueues(struct adapter *, int, int, int,
438    struct intrs_and_queues *);
439static int prep_firmware(struct adapter *);
440static int partition_resources(struct adapter *, const struct firmware *,
441    const char *);
442static int get_params__pre_init(struct adapter *);
443static int get_params__post_init(struct adapter *);
444static int set_params__post_init(struct adapter *);
445static void t4_set_desc(struct adapter *);
446static void build_medialist(struct port_info *, struct ifmedia *);
447static int cxgbe_init_synchronized(struct vi_info *);
448static int cxgbe_uninit_synchronized(struct vi_info *);
449static int setup_intr_handlers(struct adapter *);
450static void quiesce_txq(struct adapter *, struct sge_txq *);
451static void quiesce_wrq(struct adapter *, struct sge_wrq *);
452static void quiesce_iq(struct adapter *, struct sge_iq *);
453static void quiesce_fl(struct adapter *, struct sge_fl *);
454static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
455    driver_intr_t *, void *, char *);
456static int t4_free_irq(struct adapter *, struct irq *);
457static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
458static void vi_refresh_stats(struct adapter *, struct vi_info *);
459static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
460static void cxgbe_tick(void *);
461static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
462static int cpl_not_handled(struct sge_iq *, const struct rss_header *,
463    struct mbuf *);
464static int an_not_handled(struct sge_iq *, const struct rsp_ctrl *);
465static int fw_msg_not_handled(struct adapter *, const __be64 *);
466static void t4_sysctls(struct adapter *);
467static void cxgbe_sysctls(struct port_info *);
468static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
469static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
470static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
471static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
472static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
473static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
474static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
475static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
476static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
477static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
478static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
479#ifdef SBUF_DRAIN
480static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
481static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
482static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
483static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
484static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
485static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
486static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
487static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
488static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
489static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
490static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
491static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
492static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
493static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
494static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
495static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
496static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
497static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
498static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
499static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
500static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
501static int sysctl_tids(SYSCTL_HANDLER_ARGS);
502static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
503static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
504static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
505static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
506static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
507static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
508static int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
509#endif
510#ifdef TCP_OFFLOAD
511static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
512static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
513static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
514#endif
515static uint32_t fconf_iconf_to_mode(uint32_t, uint32_t);
516static uint32_t mode_to_fconf(uint32_t);
517static uint32_t mode_to_iconf(uint32_t);
518static int check_fspec_against_fconf_iconf(struct adapter *,
519    struct t4_filter_specification *);
520static int get_filter_mode(struct adapter *, uint32_t *);
521static int set_filter_mode(struct adapter *, uint32_t);
522static inline uint64_t get_filter_hits(struct adapter *, uint32_t);
523static int get_filter(struct adapter *, struct t4_filter *);
524static int set_filter(struct adapter *, struct t4_filter *);
525static int del_filter(struct adapter *, struct t4_filter *);
526static void clear_filter(struct filter_entry *);
527static int set_filter_wr(struct adapter *, int);
528static int del_filter_wr(struct adapter *, int);
529static int get_sge_context(struct adapter *, struct t4_sge_context *);
530static int load_fw(struct adapter *, struct t4_data *);
531static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
532static int read_i2c(struct adapter *, struct t4_i2c_data *);
533static int set_sched_class(struct adapter *, struct t4_sched_params *);
534static int set_sched_queue(struct adapter *, struct t4_sched_queue *);
535#ifdef TCP_OFFLOAD
536static int toe_capability(struct vi_info *, int);
537#endif
538static int mod_event(module_t, int, void *);
539
540struct {
541	uint16_t device;
542	char *desc;
543} t4_pciids[] = {
544	{0xa000, "Chelsio Terminator 4 FPGA"},
545	{0x4400, "Chelsio T440-dbg"},
546	{0x4401, "Chelsio T420-CR"},
547	{0x4402, "Chelsio T422-CR"},
548	{0x4403, "Chelsio T440-CR"},
549	{0x4404, "Chelsio T420-BCH"},
550	{0x4405, "Chelsio T440-BCH"},
551	{0x4406, "Chelsio T440-CH"},
552	{0x4407, "Chelsio T420-SO"},
553	{0x4408, "Chelsio T420-CX"},
554	{0x4409, "Chelsio T420-BT"},
555	{0x440a, "Chelsio T404-BT"},
556	{0x440e, "Chelsio T440-LP-CR"},
557}, t5_pciids[] = {
558	{0xb000, "Chelsio Terminator 5 FPGA"},
559	{0x5400, "Chelsio T580-dbg"},
560	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
561	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
562	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
563	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
564	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
565	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
566	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
567	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
568	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
569	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
570	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
571	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
572	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
573#ifdef notyet
574	{0x5404,  "Chelsio T520-BCH"},
575	{0x5405,  "Chelsio T540-BCH"},
576	{0x5406,  "Chelsio T540-CH"},
577	{0x5408,  "Chelsio T520-CX"},
578	{0x540b,  "Chelsio B520-SR"},
579	{0x540c,  "Chelsio B504-BT"},
580	{0x540f,  "Chelsio Amsterdam"},
581	{0x5413,  "Chelsio T580-CHR"},
582#endif
583};
584
585#ifdef TCP_OFFLOAD
586/*
587 * service_iq() has an iq and needs the fl.  Offset of fl from the iq should be
588 * exactly the same for both rxq and ofld_rxq.
589 */
590CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
591CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
592#endif
593
594/* No easy way to include t4_msg.h before adapter.h so we check this way */
595CTASSERT(nitems(((struct adapter *)0)->cpl_handler) == NUM_CPL_CMDS);
596CTASSERT(nitems(((struct adapter *)0)->fw_msg_handler) == NUM_FW6_TYPES);
597
598CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
599
600static int
601t4_probe(device_t dev)
602{
603	int i;
604	uint16_t v = pci_get_vendor(dev);
605	uint16_t d = pci_get_device(dev);
606	uint8_t f = pci_get_function(dev);
607
608	if (v != PCI_VENDOR_ID_CHELSIO)
609		return (ENXIO);
610
611	/* Attach only to PF0 of the FPGA */
612	if (d == 0xa000 && f != 0)
613		return (ENXIO);
614
615	for (i = 0; i < nitems(t4_pciids); i++) {
616		if (d == t4_pciids[i].device) {
617			device_set_desc(dev, t4_pciids[i].desc);
618			return (BUS_PROBE_DEFAULT);
619		}
620	}
621
622	return (ENXIO);
623}
624
625static int
626t5_probe(device_t dev)
627{
628	int i;
629	uint16_t v = pci_get_vendor(dev);
630	uint16_t d = pci_get_device(dev);
631	uint8_t f = pci_get_function(dev);
632
633	if (v != PCI_VENDOR_ID_CHELSIO)
634		return (ENXIO);
635
636	/* Attach only to PF0 of the FPGA */
637	if (d == 0xb000 && f != 0)
638		return (ENXIO);
639
640	for (i = 0; i < nitems(t5_pciids); i++) {
641		if (d == t5_pciids[i].device) {
642			device_set_desc(dev, t5_pciids[i].desc);
643			return (BUS_PROBE_DEFAULT);
644		}
645	}
646
647	return (ENXIO);
648}
649
650static void
651t5_attribute_workaround(device_t dev)
652{
653	device_t root_port;
654	uint32_t v;
655
656	/*
657	 * The T5 chips do not properly echo the No Snoop and Relaxed
658	 * Ordering attributes when replying to a TLP from a Root
659	 * Port.  As a workaround, find the parent Root Port and
660	 * disable No Snoop and Relaxed Ordering.  Note that this
661	 * affects all devices under this root port.
662	 */
663	root_port = pci_find_pcie_root_port(dev);
664	if (root_port == NULL) {
665		device_printf(dev, "Unable to find parent root port\n");
666		return;
667	}
668
669	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
670	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
671	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
672	    0)
673		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
674		    device_get_nameunit(root_port));
675}
676
677static int
678t4_attach(device_t dev)
679{
680	struct adapter *sc;
681	int rc = 0, i, j, n10g, n1g, rqidx, tqidx;
682	struct intrs_and_queues iaq;
683	struct sge *s;
684	uint8_t *buf;
685#ifdef TCP_OFFLOAD
686	int ofld_rqidx, ofld_tqidx;
687#endif
688#ifdef DEV_NETMAP
689	int nm_rqidx, nm_tqidx;
690#endif
691	int num_vis;
692
693	sc = device_get_softc(dev);
694	sc->dev = dev;
695	TUNABLE_INT_FETCH("hw.cxgbe.debug_flags", &sc->debug_flags);
696
697	if ((pci_get_device(dev) & 0xff00) == 0x5400)
698		t5_attribute_workaround(dev);
699	pci_enable_busmaster(dev);
700	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
701		uint32_t v;
702
703		pci_set_max_read_req(dev, 4096);
704		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
705		v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
706		pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
707
708		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
709	}
710
711	sc->traceq = -1;
712	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
713	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
714	    device_get_nameunit(dev));
715
716	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
717	    device_get_nameunit(dev));
718	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
719	sx_xlock(&t4_list_lock);
720	SLIST_INSERT_HEAD(&t4_list, sc, link);
721	sx_xunlock(&t4_list_lock);
722
723	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
724	TAILQ_INIT(&sc->sfl);
725	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
726
727	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
728
729	rc = map_bars_0_and_4(sc);
730	if (rc != 0)
731		goto done; /* error message displayed already */
732
733	/*
734	 * This is the real PF# to which we're attaching.  Works from within PCI
735	 * passthrough environments too, where pci_get_function() could return a
736	 * different PF# depending on the passthrough configuration.  We need to
737	 * use the real PF# in all our communication with the firmware.
738	 */
739	sc->pf = G_SOURCEPF(t4_read_reg(sc, A_PL_WHOAMI));
740	sc->mbox = sc->pf;
741
742	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
743	sc->an_handler = an_not_handled;
744	for (i = 0; i < nitems(sc->cpl_handler); i++)
745		sc->cpl_handler[i] = cpl_not_handled;
746	for (i = 0; i < nitems(sc->fw_msg_handler); i++)
747		sc->fw_msg_handler[i] = fw_msg_not_handled;
748	t4_register_cpl_handler(sc, CPL_SET_TCB_RPL, t4_filter_rpl);
749	t4_register_cpl_handler(sc, CPL_TRACE_PKT, t4_trace_pkt);
750	t4_register_cpl_handler(sc, CPL_T5_TRACE_PKT, t5_trace_pkt);
751	t4_init_sge_cpl_handlers(sc);
752
753	/* Prepare the adapter for operation. */
754	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
755	rc = -t4_prep_adapter(sc, buf);
756	free(buf, M_CXGBE);
757	if (rc != 0) {
758		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
759		goto done;
760	}
761
762	/*
763	 * Do this really early, with the memory windows set up even before the
764	 * character device.  The userland tool's register i/o and mem read
765	 * will work even in "recovery mode".
766	 */
767	setup_memwin(sc);
768	if (t4_init_devlog_params(sc, 0) == 0)
769		fixup_devlog_params(sc);
770	sc->cdev = make_dev(is_t4(sc) ? &t4_cdevsw : &t5_cdevsw,
771	    device_get_unit(dev), UID_ROOT, GID_WHEEL, 0600, "%s",
772	    device_get_nameunit(dev));
773	if (sc->cdev == NULL)
774		device_printf(dev, "failed to create nexus char device.\n");
775	else
776		sc->cdev->si_drv1 = sc;
777
778	/* Go no further if recovery mode has been requested. */
779	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
780		device_printf(dev, "recovery mode.\n");
781		goto done;
782	}
783
784#if defined(__i386__)
785	if ((cpu_feature & CPUID_CX8) == 0) {
786		device_printf(dev, "64 bit atomics not available.\n");
787		rc = ENOTSUP;
788		goto done;
789	}
790#endif
791
792	/* Prepare the firmware for operation */
793	rc = prep_firmware(sc);
794	if (rc != 0)
795		goto done; /* error message displayed already */
796
797	rc = get_params__post_init(sc);
798	if (rc != 0)
799		goto done; /* error message displayed already */
800
801	rc = set_params__post_init(sc);
802	if (rc != 0)
803		goto done; /* error message displayed already */
804
805	rc = map_bar_2(sc);
806	if (rc != 0)
807		goto done; /* error message displayed already */
808
809	rc = t4_create_dma_tag(sc);
810	if (rc != 0)
811		goto done; /* error message displayed already */
812
813	/*
814	 * Number of VIs to create per-port.  The first VI is the "main" regular
815	 * VI for the port.  The rest are additional virtual interfaces on the
816	 * same physical port.  Note that the main VI does not have native
817	 * netmap support but the extra VIs do.
818	 *
819	 * Limit the number of VIs per port to the number of available
820	 * MAC addresses per port.
821	 */
822	if (t4_num_vis >= 1)
823		num_vis = t4_num_vis;
824	else
825		num_vis = 1;
826	if (num_vis > nitems(vi_mac_funcs)) {
827		num_vis = nitems(vi_mac_funcs);
828		device_printf(dev, "Number of VIs limited to %d\n", num_vis);
829	}
830
831	/*
832	 * First pass over all the ports - allocate VIs and initialize some
833	 * basic parameters like mac address, port type, etc.  We also figure
834	 * out whether a port is 10G or 1G and use that information when
835	 * calculating how many interrupts to attempt to allocate.
836	 */
837	n10g = n1g = 0;
838	for_each_port(sc, i) {
839		struct port_info *pi;
840
841		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
842		sc->port[i] = pi;
843
844		/* These must be set before t4_port_init */
845		pi->adapter = sc;
846		pi->port_id = i;
847		/*
848		 * XXX: vi[0] is special so we can't delay this allocation until
849		 * pi->nvi's final value is known.
850		 */
851		pi->vi = malloc(sizeof(struct vi_info) * num_vis, M_CXGBE,
852		    M_ZERO | M_WAITOK);
853
854		/*
855		 * Allocate the "main" VI and initialize parameters
856		 * like mac addr.
857		 */
858		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
859		if (rc != 0) {
860			device_printf(dev, "unable to initialize port %d: %d\n",
861			    i, rc);
862			free(pi->vi, M_CXGBE);
863			free(pi, M_CXGBE);
864			sc->port[i] = NULL;
865			goto done;
866		}
867
868		pi->link_cfg.requested_fc &= ~(PAUSE_TX | PAUSE_RX);
869		pi->link_cfg.requested_fc |= t4_pause_settings;
870		pi->link_cfg.fc &= ~(PAUSE_TX | PAUSE_RX);
871		pi->link_cfg.fc |= t4_pause_settings;
872
873		rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, &pi->link_cfg);
874		if (rc != 0) {
875			device_printf(dev, "port %d l1cfg failed: %d\n", i, rc);
876			free(pi->vi, M_CXGBE);
877			free(pi, M_CXGBE);
878			sc->port[i] = NULL;
879			goto done;
880		}
881
882		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
883		    device_get_nameunit(dev), i);
884		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
885		sc->chan_map[pi->tx_chan] = i;
886
887		pi->tc = malloc(sizeof(struct tx_sched_class) *
888		    sc->chip_params->nsched_cls, M_CXGBE, M_ZERO | M_WAITOK);
889
890		if (is_10G_port(pi) || is_40G_port(pi)) {
891			n10g++;
892		} else {
893			n1g++;
894		}
895
896		pi->linkdnrc = -1;
897
898		pi->dev = device_add_child(dev, is_t4(sc) ? "cxgbe" : "cxl", -1);
899		if (pi->dev == NULL) {
900			device_printf(dev,
901			    "failed to add device for port %d.\n", i);
902			rc = ENXIO;
903			goto done;
904		}
905		pi->vi[0].dev = pi->dev;
906		device_set_softc(pi->dev, pi);
907	}
908
909	/*
910	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
911	 */
912	rc = cfg_itype_and_nqueues(sc, n10g, n1g, num_vis, &iaq);
913	if (rc != 0)
914		goto done; /* error message displayed already */
915	if (iaq.nrxq_vi + iaq.nofldrxq_vi + iaq.nnmrxq_vi == 0)
916		num_vis = 1;
917
918	sc->intr_type = iaq.intr_type;
919	sc->intr_count = iaq.nirq;
920
921	s = &sc->sge;
922	s->nrxq = n10g * iaq.nrxq10g + n1g * iaq.nrxq1g;
923	s->ntxq = n10g * iaq.ntxq10g + n1g * iaq.ntxq1g;
924	if (num_vis > 1) {
925		s->nrxq += (n10g + n1g) * (num_vis - 1) * iaq.nrxq_vi;
926		s->ntxq += (n10g + n1g) * (num_vis - 1) * iaq.ntxq_vi;
927	}
928	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
929	s->neq += sc->params.nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
930	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
931#ifdef TCP_OFFLOAD
932	if (is_offload(sc)) {
933		s->nofldrxq = n10g * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
934		s->nofldtxq = n10g * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
935		if (num_vis > 1) {
936			s->nofldrxq += (n10g + n1g) * (num_vis - 1) *
937			    iaq.nofldrxq_vi;
938			s->nofldtxq += (n10g + n1g) * (num_vis - 1) *
939			    iaq.nofldtxq_vi;
940		}
941		s->neq += s->nofldtxq + s->nofldrxq;
942		s->niq += s->nofldrxq;
943
944		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
945		    M_CXGBE, M_ZERO | M_WAITOK);
946		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
947		    M_CXGBE, M_ZERO | M_WAITOK);
948	}
949#endif
950#ifdef DEV_NETMAP
951	if (num_vis > 1) {
952		s->nnmrxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmrxq_vi;
953		s->nnmtxq = (n10g + n1g) * (num_vis - 1) * iaq.nnmtxq_vi;
954	}
955	s->neq += s->nnmtxq + s->nnmrxq;
956	s->niq += s->nnmrxq;
957
958	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
959	    M_CXGBE, M_ZERO | M_WAITOK);
960	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
961	    M_CXGBE, M_ZERO | M_WAITOK);
962#endif
963
964	s->ctrlq = malloc(sc->params.nports * sizeof(struct sge_wrq), M_CXGBE,
965	    M_ZERO | M_WAITOK);
966	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
967	    M_ZERO | M_WAITOK);
968	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
969	    M_ZERO | M_WAITOK);
970	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
971	    M_ZERO | M_WAITOK);
972	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
973	    M_ZERO | M_WAITOK);
974
975	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
976	    M_ZERO | M_WAITOK);
977
978	t4_init_l2t(sc, M_WAITOK);
979
980	/*
981	 * Second pass over the ports.  This time we know the number of rx and
982	 * tx queues that each port should get.
983	 */
984	rqidx = tqidx = 0;
985#ifdef TCP_OFFLOAD
986	ofld_rqidx = ofld_tqidx = 0;
987#endif
988#ifdef DEV_NETMAP
989	nm_rqidx = nm_tqidx = 0;
990#endif
991	for_each_port(sc, i) {
992		struct port_info *pi = sc->port[i];
993		struct vi_info *vi;
994
995		if (pi == NULL)
996			continue;
997
998		pi->nvi = num_vis;
999		for_each_vi(pi, j, vi) {
1000			vi->pi = pi;
1001			vi->qsize_rxq = t4_qsize_rxq;
1002			vi->qsize_txq = t4_qsize_txq;
1003
1004			vi->first_rxq = rqidx;
1005			vi->first_txq = tqidx;
1006			if (is_10G_port(pi) || is_40G_port(pi)) {
1007				vi->tmr_idx = t4_tmr_idx_10g;
1008				vi->pktc_idx = t4_pktc_idx_10g;
1009				vi->flags |= iaq.intr_flags_10g & INTR_RXQ;
1010				vi->nrxq = j == 0 ? iaq.nrxq10g : iaq.nrxq_vi;
1011				vi->ntxq = j == 0 ? iaq.ntxq10g : iaq.ntxq_vi;
1012			} else {
1013				vi->tmr_idx = t4_tmr_idx_1g;
1014				vi->pktc_idx = t4_pktc_idx_1g;
1015				vi->flags |= iaq.intr_flags_1g & INTR_RXQ;
1016				vi->nrxq = j == 0 ? iaq.nrxq1g : iaq.nrxq_vi;
1017				vi->ntxq = j == 0 ? iaq.ntxq1g : iaq.ntxq_vi;
1018			}
1019			rqidx += vi->nrxq;
1020			tqidx += vi->ntxq;
1021
1022			if (j == 0 && vi->ntxq > 1)
1023				vi->rsrv_noflowq = iaq.rsrv_noflowq ? 1 : 0;
1024			else
1025				vi->rsrv_noflowq = 0;
1026
1027#ifdef TCP_OFFLOAD
1028			vi->first_ofld_rxq = ofld_rqidx;
1029			vi->first_ofld_txq = ofld_tqidx;
1030			if (is_10G_port(pi) || is_40G_port(pi)) {
1031				vi->flags |= iaq.intr_flags_10g & INTR_OFLD_RXQ;
1032				vi->nofldrxq = j == 0 ? iaq.nofldrxq10g :
1033				    iaq.nofldrxq_vi;
1034				vi->nofldtxq = j == 0 ? iaq.nofldtxq10g :
1035				    iaq.nofldtxq_vi;
1036			} else {
1037				vi->flags |= iaq.intr_flags_1g & INTR_OFLD_RXQ;
1038				vi->nofldrxq = j == 0 ? iaq.nofldrxq1g :
1039				    iaq.nofldrxq_vi;
1040				vi->nofldtxq = j == 0 ? iaq.nofldtxq1g :
1041				    iaq.nofldtxq_vi;
1042			}
1043			ofld_rqidx += vi->nofldrxq;
1044			ofld_tqidx += vi->nofldtxq;
1045#endif
1046#ifdef DEV_NETMAP
1047			if (j > 0) {
1048				vi->first_nm_rxq = nm_rqidx;
1049				vi->first_nm_txq = nm_tqidx;
1050				vi->nnmrxq = iaq.nnmrxq_vi;
1051				vi->nnmtxq = iaq.nnmtxq_vi;
1052				nm_rqidx += vi->nnmrxq;
1053				nm_tqidx += vi->nnmtxq;
1054			}
1055#endif
1056		}
1057	}
1058
1059	rc = setup_intr_handlers(sc);
1060	if (rc != 0) {
1061		device_printf(dev,
1062		    "failed to setup interrupt handlers: %d\n", rc);
1063		goto done;
1064	}
1065
1066	rc = bus_generic_attach(dev);
1067	if (rc != 0) {
1068		device_printf(dev,
1069		    "failed to attach all child ports: %d\n", rc);
1070		goto done;
1071	}
1072
1073	device_printf(dev,
1074	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1075	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1076	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1077	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1078	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1079
1080	t4_set_desc(sc);
1081
1082done:
1083	if (rc != 0 && sc->cdev) {
1084		/* cdev was created and so cxgbetool works; recover that way. */
1085		device_printf(dev,
1086		    "error during attach, adapter is now in recovery mode.\n");
1087		rc = 0;
1088	}
1089
1090	if (rc != 0)
1091		t4_detach(dev);
1092	else
1093		t4_sysctls(sc);
1094
1095	return (rc);
1096}
1097
1098/*
1099 * Idempotent
1100 */
1101static int
1102t4_detach(device_t dev)
1103{
1104	struct adapter *sc;
1105	struct port_info *pi;
1106	int i, rc;
1107
1108	sc = device_get_softc(dev);
1109
1110	if (sc->flags & FULL_INIT_DONE)
1111		t4_intr_disable(sc);
1112
1113	if (sc->cdev) {
1114		destroy_dev(sc->cdev);
1115		sc->cdev = NULL;
1116	}
1117
1118	rc = bus_generic_detach(dev);
1119	if (rc) {
1120		device_printf(dev,
1121		    "failed to detach child devices: %d\n", rc);
1122		return (rc);
1123	}
1124
1125	for (i = 0; i < sc->intr_count; i++)
1126		t4_free_irq(sc, &sc->irq[i]);
1127
1128	for (i = 0; i < MAX_NPORTS; i++) {
1129		pi = sc->port[i];
1130		if (pi) {
1131			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1132			if (pi->dev)
1133				device_delete_child(dev, pi->dev);
1134
1135			mtx_destroy(&pi->pi_lock);
1136			free(pi->vi, M_CXGBE);
1137			free(pi->tc, M_CXGBE);
1138			free(pi, M_CXGBE);
1139		}
1140	}
1141
1142	if (sc->flags & FULL_INIT_DONE)
1143		adapter_full_uninit(sc);
1144
1145	if (sc->flags & FW_OK)
1146		t4_fw_bye(sc, sc->mbox);
1147
1148	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1149		pci_release_msi(dev);
1150
1151	if (sc->regs_res)
1152		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1153		    sc->regs_res);
1154
1155	if (sc->udbs_res)
1156		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1157		    sc->udbs_res);
1158
1159	if (sc->msix_res)
1160		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1161		    sc->msix_res);
1162
1163	if (sc->l2t)
1164		t4_free_l2t(sc->l2t);
1165
1166#ifdef TCP_OFFLOAD
1167	free(sc->sge.ofld_rxq, M_CXGBE);
1168	free(sc->sge.ofld_txq, M_CXGBE);
1169#endif
1170#ifdef DEV_NETMAP
1171	free(sc->sge.nm_rxq, M_CXGBE);
1172	free(sc->sge.nm_txq, M_CXGBE);
1173#endif
1174	free(sc->irq, M_CXGBE);
1175	free(sc->sge.rxq, M_CXGBE);
1176	free(sc->sge.txq, M_CXGBE);
1177	free(sc->sge.ctrlq, M_CXGBE);
1178	free(sc->sge.iqmap, M_CXGBE);
1179	free(sc->sge.eqmap, M_CXGBE);
1180	free(sc->tids.ftid_tab, M_CXGBE);
1181	t4_destroy_dma_tag(sc);
1182	if (mtx_initialized(&sc->sc_lock)) {
1183		sx_xlock(&t4_list_lock);
1184		SLIST_REMOVE(&t4_list, sc, adapter, link);
1185		sx_xunlock(&t4_list_lock);
1186		mtx_destroy(&sc->sc_lock);
1187	}
1188
1189	callout_drain(&sc->sfl_callout);
1190	if (mtx_initialized(&sc->tids.ftid_lock))
1191		mtx_destroy(&sc->tids.ftid_lock);
1192	if (mtx_initialized(&sc->sfl_lock))
1193		mtx_destroy(&sc->sfl_lock);
1194	if (mtx_initialized(&sc->ifp_lock))
1195		mtx_destroy(&sc->ifp_lock);
1196	if (mtx_initialized(&sc->reg_lock))
1197		mtx_destroy(&sc->reg_lock);
1198
1199	for (i = 0; i < NUM_MEMWIN; i++) {
1200		struct memwin *mw = &sc->memwin[i];
1201
1202		if (rw_initialized(&mw->mw_lock))
1203			rw_destroy(&mw->mw_lock);
1204	}
1205
1206	bzero(sc, sizeof(*sc));
1207
1208	return (0);
1209}
1210
1211static int
1212cxgbe_probe(device_t dev)
1213{
1214	char buf[128];
1215	struct port_info *pi = device_get_softc(dev);
1216
1217	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1218	device_set_desc_copy(dev, buf);
1219
1220	return (BUS_PROBE_DEFAULT);
1221}
1222
1223#define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1224    IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1225    IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
1226#define T4_CAP_ENABLE (T4_CAP)
1227
1228static int
1229cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1230{
1231	struct ifnet *ifp;
1232	struct sbuf *sb;
1233
1234	vi->xact_addr_filt = -1;
1235	callout_init(&vi->tick, 1);
1236
1237	/* Allocate an ifnet and set it up */
1238	ifp = if_alloc(IFT_ETHER);
1239	if (ifp == NULL) {
1240		device_printf(dev, "Cannot allocate ifnet\n");
1241		return (ENOMEM);
1242	}
1243	vi->ifp = ifp;
1244	ifp->if_softc = vi;
1245
1246	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1247	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1248
1249	ifp->if_init = cxgbe_init;
1250	ifp->if_ioctl = cxgbe_ioctl;
1251	ifp->if_transmit = cxgbe_transmit;
1252	ifp->if_qflush = cxgbe_qflush;
1253
1254	ifp->if_capabilities = T4_CAP;
1255#ifdef TCP_OFFLOAD
1256	if (vi->nofldrxq != 0)
1257		ifp->if_capabilities |= IFCAP_TOE;
1258#endif
1259	ifp->if_capenable = T4_CAP_ENABLE;
1260	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1261	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1262
1263	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1264	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
1265	ifp->if_hw_tsomaxsegsize = 65536;
1266
1267	/* Initialize ifmedia for this VI */
1268	ifmedia_init(&vi->media, IFM_IMASK, cxgbe_media_change,
1269	    cxgbe_media_status);
1270	build_medialist(vi->pi, &vi->media);
1271
1272	vi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
1273	    EVENTHANDLER_PRI_ANY);
1274
1275	ether_ifattach(ifp, vi->hw_addr);
1276#ifdef DEV_NETMAP
1277	if (vi->nnmrxq != 0)
1278		cxgbe_nm_attach(vi);
1279#endif
1280	sb = sbuf_new_auto();
1281	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1282#ifdef TCP_OFFLOAD
1283	if (ifp->if_capabilities & IFCAP_TOE)
1284		sbuf_printf(sb, "; %d txq, %d rxq (TOE)",
1285		    vi->nofldtxq, vi->nofldrxq);
1286#endif
1287#ifdef DEV_NETMAP
1288	if (ifp->if_capabilities & IFCAP_NETMAP)
1289		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1290		    vi->nnmtxq, vi->nnmrxq);
1291#endif
1292	sbuf_finish(sb);
1293	device_printf(dev, "%s\n", sbuf_data(sb));
1294	sbuf_delete(sb);
1295
1296	vi_sysctls(vi);
1297
1298	return (0);
1299}
1300
1301static int
1302cxgbe_attach(device_t dev)
1303{
1304	struct port_info *pi = device_get_softc(dev);
1305	struct vi_info *vi;
1306	int i, rc;
1307
1308	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1309
1310	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1311	if (rc)
1312		return (rc);
1313
1314	for_each_vi(pi, i, vi) {
1315		if (i == 0)
1316			continue;
1317		vi->dev = device_add_child(dev, is_t4(pi->adapter) ?
1318		    "vcxgbe" : "vcxl", -1);
1319		if (vi->dev == NULL) {
1320			device_printf(dev, "failed to add VI %d\n", i);
1321			continue;
1322		}
1323		device_set_softc(vi->dev, vi);
1324	}
1325
1326	cxgbe_sysctls(pi);
1327
1328	bus_generic_attach(dev);
1329
1330	return (0);
1331}
1332
1333static void
1334cxgbe_vi_detach(struct vi_info *vi)
1335{
1336	struct ifnet *ifp = vi->ifp;
1337
1338	ether_ifdetach(ifp);
1339
1340	if (vi->vlan_c)
1341		EVENTHANDLER_DEREGISTER(vlan_config, vi->vlan_c);
1342
1343	/* Let detach proceed even if these fail. */
1344#ifdef DEV_NETMAP
1345	if (ifp->if_capabilities & IFCAP_NETMAP)
1346		cxgbe_nm_detach(vi);
1347#endif
1348	cxgbe_uninit_synchronized(vi);
1349	callout_drain(&vi->tick);
1350	vi_full_uninit(vi);
1351
1352	ifmedia_removeall(&vi->media);
1353	if_free(vi->ifp);
1354	vi->ifp = NULL;
1355}
1356
1357static int
1358cxgbe_detach(device_t dev)
1359{
1360	struct port_info *pi = device_get_softc(dev);
1361	struct adapter *sc = pi->adapter;
1362	int rc;
1363
1364	/* Detach the extra VIs first. */
1365	rc = bus_generic_detach(dev);
1366	if (rc)
1367		return (rc);
1368	device_delete_children(dev);
1369
1370	doom_vi(sc, &pi->vi[0]);
1371
1372	if (pi->flags & HAS_TRACEQ) {
1373		sc->traceq = -1;	/* cloner should not create ifnet */
1374		t4_tracer_port_detach(sc);
1375	}
1376
1377	cxgbe_vi_detach(&pi->vi[0]);
1378	callout_drain(&pi->tick);
1379
1380	end_synchronized_op(sc, 0);
1381
1382	return (0);
1383}
1384
1385static void
1386cxgbe_init(void *arg)
1387{
1388	struct vi_info *vi = arg;
1389	struct adapter *sc = vi->pi->adapter;
1390
1391	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1392		return;
1393	cxgbe_init_synchronized(vi);
1394	end_synchronized_op(sc, 0);
1395}
1396
1397static int
1398cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1399{
1400	int rc = 0, mtu, flags, can_sleep;
1401	struct vi_info *vi = ifp->if_softc;
1402	struct adapter *sc = vi->pi->adapter;
1403	struct ifreq *ifr = (struct ifreq *)data;
1404	uint32_t mask;
1405
1406	switch (cmd) {
1407	case SIOCSIFMTU:
1408		mtu = ifr->ifr_mtu;
1409		if ((mtu < ETHERMIN) || (mtu > ETHERMTU_JUMBO))
1410			return (EINVAL);
1411
1412		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1413		if (rc)
1414			return (rc);
1415		ifp->if_mtu = mtu;
1416		if (vi->flags & VI_INIT_DONE) {
1417			t4_update_fl_bufsize(ifp);
1418			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1419				rc = update_mac_settings(ifp, XGMAC_MTU);
1420		}
1421		end_synchronized_op(sc, 0);
1422		break;
1423
1424	case SIOCSIFFLAGS:
1425		can_sleep = 0;
1426redo_sifflags:
1427		rc = begin_synchronized_op(sc, vi,
1428		    can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
1429		if (rc)
1430			return (rc);
1431
1432		if (ifp->if_flags & IFF_UP) {
1433			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1434				flags = vi->if_flags;
1435				if ((ifp->if_flags ^ flags) &
1436				    (IFF_PROMISC | IFF_ALLMULTI)) {
1437					if (can_sleep == 1) {
1438						end_synchronized_op(sc, 0);
1439						can_sleep = 0;
1440						goto redo_sifflags;
1441					}
1442					rc = update_mac_settings(ifp,
1443					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1444				}
1445			} else {
1446				if (can_sleep == 0) {
1447					end_synchronized_op(sc, LOCK_HELD);
1448					can_sleep = 1;
1449					goto redo_sifflags;
1450				}
1451				rc = cxgbe_init_synchronized(vi);
1452			}
1453			vi->if_flags = ifp->if_flags;
1454		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1455			if (can_sleep == 0) {
1456				end_synchronized_op(sc, LOCK_HELD);
1457				can_sleep = 1;
1458				goto redo_sifflags;
1459			}
1460			rc = cxgbe_uninit_synchronized(vi);
1461		}
1462		end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
1463		break;
1464
1465	case SIOCADDMULTI:
1466	case SIOCDELMULTI: /* these two are called with a mutex held :-( */
1467		rc = begin_synchronized_op(sc, vi, HOLD_LOCK, "t4multi");
1468		if (rc)
1469			return (rc);
1470		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1471			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1472		end_synchronized_op(sc, LOCK_HELD);
1473		break;
1474
1475	case SIOCSIFCAP:
1476		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1477		if (rc)
1478			return (rc);
1479
1480		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1481		if (mask & IFCAP_TXCSUM) {
1482			ifp->if_capenable ^= IFCAP_TXCSUM;
1483			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1484
1485			if (IFCAP_TSO4 & ifp->if_capenable &&
1486			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1487				ifp->if_capenable &= ~IFCAP_TSO4;
1488				if_printf(ifp,
1489				    "tso4 disabled due to -txcsum.\n");
1490			}
1491		}
1492		if (mask & IFCAP_TXCSUM_IPV6) {
1493			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1494			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1495
1496			if (IFCAP_TSO6 & ifp->if_capenable &&
1497			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1498				ifp->if_capenable &= ~IFCAP_TSO6;
1499				if_printf(ifp,
1500				    "tso6 disabled due to -txcsum6.\n");
1501			}
1502		}
1503		if (mask & IFCAP_RXCSUM)
1504			ifp->if_capenable ^= IFCAP_RXCSUM;
1505		if (mask & IFCAP_RXCSUM_IPV6)
1506			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1507
1508		/*
1509		 * Note that we leave CSUM_TSO alone (it is always set).  The
1510		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1511		 * sending a TSO request our way, so it's sufficient to toggle
1512		 * IFCAP_TSOx only.
1513		 */
1514		if (mask & IFCAP_TSO4) {
1515			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1516			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1517				if_printf(ifp, "enable txcsum first.\n");
1518				rc = EAGAIN;
1519				goto fail;
1520			}
1521			ifp->if_capenable ^= IFCAP_TSO4;
1522		}
1523		if (mask & IFCAP_TSO6) {
1524			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1525			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1526				if_printf(ifp, "enable txcsum6 first.\n");
1527				rc = EAGAIN;
1528				goto fail;
1529			}
1530			ifp->if_capenable ^= IFCAP_TSO6;
1531		}
1532		if (mask & IFCAP_LRO) {
1533#if defined(INET) || defined(INET6)
1534			int i;
1535			struct sge_rxq *rxq;
1536
1537			ifp->if_capenable ^= IFCAP_LRO;
1538			for_each_rxq(vi, i, rxq) {
1539				if (ifp->if_capenable & IFCAP_LRO)
1540					rxq->iq.flags |= IQ_LRO_ENABLED;
1541				else
1542					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1543			}
1544#endif
1545		}
1546#ifdef TCP_OFFLOAD
1547		if (mask & IFCAP_TOE) {
1548			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1549
1550			rc = toe_capability(vi, enable);
1551			if (rc != 0)
1552				goto fail;
1553
1554			ifp->if_capenable ^= mask;
1555		}
1556#endif
1557		if (mask & IFCAP_VLAN_HWTAGGING) {
1558			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1559			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1560				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1561		}
1562		if (mask & IFCAP_VLAN_MTU) {
1563			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1564
1565			/* Need to find out how to disable auto-mtu-inflation */
1566		}
1567		if (mask & IFCAP_VLAN_HWTSO)
1568			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1569		if (mask & IFCAP_VLAN_HWCSUM)
1570			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1571
1572#ifdef VLAN_CAPABILITIES
1573		VLAN_CAPABILITIES(ifp);
1574#endif
1575fail:
1576		end_synchronized_op(sc, 0);
1577		break;
1578
1579	case SIOCSIFMEDIA:
1580	case SIOCGIFMEDIA:
1581		ifmedia_ioctl(ifp, ifr, &vi->media, cmd);
1582		break;
1583
1584	case SIOCGI2C: {
1585		struct ifi2creq i2c;
1586
1587		rc = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
1588		if (rc != 0)
1589			break;
1590		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1591			rc = EPERM;
1592			break;
1593		}
1594		if (i2c.len > sizeof(i2c.data)) {
1595			rc = EINVAL;
1596			break;
1597		}
1598		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
1599		if (rc)
1600			return (rc);
1601		rc = -t4_i2c_rd(sc, sc->mbox, vi->pi->port_id, i2c.dev_addr,
1602		    i2c.offset, i2c.len, &i2c.data[0]);
1603		end_synchronized_op(sc, 0);
1604		if (rc == 0)
1605			rc = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
1606		break;
1607	}
1608
1609	default:
1610		rc = ether_ioctl(ifp, cmd, data);
1611	}
1612
1613	return (rc);
1614}
1615
1616static int
1617cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
1618{
1619	struct vi_info *vi = ifp->if_softc;
1620	struct port_info *pi = vi->pi;
1621	struct adapter *sc = pi->adapter;
1622	struct sge_txq *txq;
1623	void *items[1];
1624	int rc;
1625
1626	M_ASSERTPKTHDR(m);
1627	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
1628
1629	if (__predict_false(pi->link_cfg.link_ok == 0)) {
1630		m_freem(m);
1631		return (ENETDOWN);
1632	}
1633
1634	rc = parse_pkt(&m);
1635	if (__predict_false(rc != 0)) {
1636		MPASS(m == NULL);			/* was freed already */
1637		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
1638		return (rc);
1639	}
1640
1641	/* Select a txq. */
1642	txq = &sc->sge.txq[vi->first_txq];
1643	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
1644		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
1645		    vi->rsrv_noflowq);
1646
1647	items[0] = m;
1648	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
1649	if (__predict_false(rc != 0))
1650		m_freem(m);
1651
1652	return (rc);
1653}
1654
1655static void
1656cxgbe_qflush(struct ifnet *ifp)
1657{
1658	struct vi_info *vi = ifp->if_softc;
1659	struct sge_txq *txq;
1660	int i;
1661
1662	/* queues do not exist if !VI_INIT_DONE. */
1663	if (vi->flags & VI_INIT_DONE) {
1664		for_each_txq(vi, i, txq) {
1665			TXQ_LOCK(txq);
1666			txq->eq.flags &= ~EQ_ENABLED;
1667			TXQ_UNLOCK(txq);
1668			while (!mp_ring_is_idle(txq->r)) {
1669				mp_ring_check_drainage(txq->r, 0);
1670				pause("qflush", 1);
1671			}
1672		}
1673	}
1674	if_qflush(ifp);
1675}
1676
1677static int
1678cxgbe_media_change(struct ifnet *ifp)
1679{
1680	struct vi_info *vi = ifp->if_softc;
1681
1682	device_printf(vi->dev, "%s unimplemented.\n", __func__);
1683
1684	return (EOPNOTSUPP);
1685}
1686
1687static void
1688cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1689{
1690	struct vi_info *vi = ifp->if_softc;
1691	struct port_info *pi = vi->pi;
1692	struct ifmedia_entry *cur;
1693	int speed = pi->link_cfg.speed;
1694
1695	cur = vi->media.ifm_cur;
1696
1697	ifmr->ifm_status = IFM_AVALID;
1698	if (!pi->link_cfg.link_ok)
1699		return;
1700
1701	ifmr->ifm_status |= IFM_ACTIVE;
1702
1703	/* active and current will differ iff current media is autoselect. */
1704	if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
1705		return;
1706
1707	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
1708	if (speed == 10000)
1709		ifmr->ifm_active |= IFM_10G_T;
1710	else if (speed == 1000)
1711		ifmr->ifm_active |= IFM_1000_T;
1712	else if (speed == 100)
1713		ifmr->ifm_active |= IFM_100_TX;
1714	else if (speed == 10)
1715		ifmr->ifm_active |= IFM_10_T;
1716	else
1717		KASSERT(0, ("%s: link up but speed unknown (%u)", __func__,
1718			    speed));
1719}
1720
1721static int
1722vcxgbe_probe(device_t dev)
1723{
1724	char buf[128];
1725	struct vi_info *vi = device_get_softc(dev);
1726
1727	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
1728	    vi - vi->pi->vi);
1729	device_set_desc_copy(dev, buf);
1730
1731	return (BUS_PROBE_DEFAULT);
1732}
1733
1734static int
1735vcxgbe_attach(device_t dev)
1736{
1737	struct vi_info *vi;
1738	struct port_info *pi;
1739	struct adapter *sc;
1740	int func, index, rc;
1741	u32 param, val;
1742
1743	vi = device_get_softc(dev);
1744	pi = vi->pi;
1745	sc = pi->adapter;
1746
1747	index = vi - pi->vi;
1748	KASSERT(index < nitems(vi_mac_funcs),
1749	    ("%s: VI %s doesn't have a MAC func", __func__,
1750	    device_get_nameunit(dev)));
1751	func = vi_mac_funcs[index];
1752	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
1753	    vi->hw_addr, &vi->rss_size, func, 0);
1754	if (rc < 0) {
1755		device_printf(dev, "Failed to allocate virtual interface "
1756		    "for port %d: %d\n", pi->port_id, -rc);
1757		return (-rc);
1758	}
1759	vi->viid = rc;
1760
1761	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
1762	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
1763	    V_FW_PARAMS_PARAM_YZ(vi->viid);
1764	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
1765	if (rc)
1766		vi->rss_base = 0xffff;
1767	else {
1768		/* MPASS((val >> 16) == rss_size); */
1769		vi->rss_base = val & 0xffff;
1770	}
1771
1772	rc = cxgbe_vi_attach(dev, vi);
1773	if (rc) {
1774		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
1775		return (rc);
1776	}
1777	return (0);
1778}
1779
1780static int
1781vcxgbe_detach(device_t dev)
1782{
1783	struct vi_info *vi;
1784	struct adapter *sc;
1785
1786	vi = device_get_softc(dev);
1787	sc = vi->pi->adapter;
1788
1789	doom_vi(sc, vi);
1790
1791	cxgbe_vi_detach(vi);
1792	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
1793
1794	end_synchronized_op(sc, 0);
1795
1796	return (0);
1797}
1798
1799void
1800t4_fatal_err(struct adapter *sc)
1801{
1802	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
1803	t4_intr_disable(sc);
1804	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
1805	    device_get_nameunit(sc->dev));
1806}
1807
1808static int
1809map_bars_0_and_4(struct adapter *sc)
1810{
1811	sc->regs_rid = PCIR_BAR(0);
1812	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1813	    &sc->regs_rid, RF_ACTIVE);
1814	if (sc->regs_res == NULL) {
1815		device_printf(sc->dev, "cannot map registers.\n");
1816		return (ENXIO);
1817	}
1818	sc->bt = rman_get_bustag(sc->regs_res);
1819	sc->bh = rman_get_bushandle(sc->regs_res);
1820	sc->mmio_len = rman_get_size(sc->regs_res);
1821	setbit(&sc->doorbells, DOORBELL_KDB);
1822
1823	sc->msix_rid = PCIR_BAR(4);
1824	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1825	    &sc->msix_rid, RF_ACTIVE);
1826	if (sc->msix_res == NULL) {
1827		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
1828		return (ENXIO);
1829	}
1830
1831	return (0);
1832}
1833
1834static int
1835map_bar_2(struct adapter *sc)
1836{
1837
1838	/*
1839	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
1840	 * to map it if RDMA is disabled.
1841	 */
1842	if (is_t4(sc) && sc->rdmacaps == 0)
1843		return (0);
1844
1845	sc->udbs_rid = PCIR_BAR(2);
1846	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
1847	    &sc->udbs_rid, RF_ACTIVE);
1848	if (sc->udbs_res == NULL) {
1849		device_printf(sc->dev, "cannot map doorbell BAR.\n");
1850		return (ENXIO);
1851	}
1852	sc->udbs_base = rman_get_virtual(sc->udbs_res);
1853
1854	if (is_t5(sc)) {
1855		setbit(&sc->doorbells, DOORBELL_UDB);
1856#if defined(__i386__) || defined(__amd64__)
1857		if (t5_write_combine) {
1858			int rc;
1859
1860			/*
1861			 * Enable write combining on BAR2.  This is the
1862			 * userspace doorbell BAR and is split into 128B
1863			 * (UDBS_SEG_SIZE) doorbell regions, each associated
1864			 * with an egress queue.  The first 64B has the doorbell
1865			 * and the second 64B can be used to submit a tx work
1866			 * request with an implicit doorbell.
1867			 */
1868
1869			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
1870			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
1871			if (rc == 0) {
1872				clrbit(&sc->doorbells, DOORBELL_UDB);
1873				setbit(&sc->doorbells, DOORBELL_WCWR);
1874				setbit(&sc->doorbells, DOORBELL_UDBWC);
1875			} else {
1876				device_printf(sc->dev,
1877				    "couldn't enable write combining: %d\n",
1878				    rc);
1879			}
1880
1881			t4_write_reg(sc, A_SGE_STAT_CFG,
1882			    V_STATSOURCE_T5(7) | V_STATMODE(0));
1883		}
1884#endif
1885	}
1886
1887	return (0);
1888}
1889
1890struct memwin_init {
1891	uint32_t base;
1892	uint32_t aperture;
1893};
1894
1895static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
1896	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1897	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1898	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
1899};
1900
1901static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
1902	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
1903	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
1904	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
1905};
1906
1907static void
1908setup_memwin(struct adapter *sc)
1909{
1910	const struct memwin_init *mw_init;
1911	struct memwin *mw;
1912	int i;
1913	uint32_t bar0;
1914
1915	if (is_t4(sc)) {
1916		/*
1917		 * Read low 32b of bar0 indirectly via the hardware backdoor
1918		 * mechanism.  Works from within PCI passthrough environments
1919		 * too, where rman_get_start() can return a different value.  We
1920		 * need to program the T4 memory window decoders with the actual
1921		 * addresses that will be coming across the PCIe link.
1922		 */
1923		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
1924		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
1925
1926		mw_init = &t4_memwin[0];
1927	} else {
1928		/* T5+ use the relative offset inside the PCIe BAR */
1929		bar0 = 0;
1930
1931		mw_init = &t5_memwin[0];
1932	}
1933
1934	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
1935		rw_init(&mw->mw_lock, "memory window access");
1936		mw->mw_base = mw_init->base;
1937		mw->mw_aperture = mw_init->aperture;
1938		mw->mw_curpos = 0;
1939		t4_write_reg(sc,
1940		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
1941		    (mw->mw_base + bar0) | V_BIR(0) |
1942		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
1943		rw_wlock(&mw->mw_lock);
1944		position_memwin(sc, i, 0);
1945		rw_wunlock(&mw->mw_lock);
1946	}
1947
1948	/* flush */
1949	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
1950}
1951
1952/*
1953 * Positions the memory window at the given address in the card's address space.
1954 * There are some alignment requirements and the actual position may be at an
1955 * address prior to the requested address.  mw->mw_curpos always has the actual
1956 * position of the window.
1957 */
1958static void
1959position_memwin(struct adapter *sc, int idx, uint32_t addr)
1960{
1961	struct memwin *mw;
1962	uint32_t pf;
1963	uint32_t reg;
1964
1965	MPASS(idx >= 0 && idx < NUM_MEMWIN);
1966	mw = &sc->memwin[idx];
1967	rw_assert(&mw->mw_lock, RA_WLOCKED);
1968
1969	if (is_t4(sc)) {
1970		pf = 0;
1971		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
1972	} else {
1973		pf = V_PFNUM(sc->pf);
1974		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
1975	}
1976	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
1977	t4_write_reg(sc, reg, mw->mw_curpos | pf);
1978	t4_read_reg(sc, reg);	/* flush */
1979}
1980
1981static int
1982rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
1983    int len, int rw)
1984{
1985	struct memwin *mw;
1986	uint32_t mw_end, v;
1987
1988	MPASS(idx >= 0 && idx < NUM_MEMWIN);
1989
1990	/* Memory can only be accessed in naturally aligned 4 byte units */
1991	if (addr & 3 || len & 3 || len <= 0)
1992		return (EINVAL);
1993
1994	mw = &sc->memwin[idx];
1995	while (len > 0) {
1996		rw_rlock(&mw->mw_lock);
1997		mw_end = mw->mw_curpos + mw->mw_aperture;
1998		if (addr >= mw_end || addr < mw->mw_curpos) {
1999			/* Will need to reposition the window */
2000			if (!rw_try_upgrade(&mw->mw_lock)) {
2001				rw_runlock(&mw->mw_lock);
2002				rw_wlock(&mw->mw_lock);
2003			}
2004			rw_assert(&mw->mw_lock, RA_WLOCKED);
2005			position_memwin(sc, idx, addr);
2006			rw_downgrade(&mw->mw_lock);
2007			mw_end = mw->mw_curpos + mw->mw_aperture;
2008		}
2009		rw_assert(&mw->mw_lock, RA_RLOCKED);
2010		while (addr < mw_end && len > 0) {
2011			if (rw == 0) {
2012				v = t4_read_reg(sc, mw->mw_base + addr -
2013				    mw->mw_curpos);
2014				*val++ = le32toh(v);
2015			} else {
2016				v = *val++;
2017				t4_write_reg(sc, mw->mw_base + addr -
2018				    mw->mw_curpos, htole32(v));;
2019			}
2020			addr += 4;
2021			len -= 4;
2022		}
2023		rw_runlock(&mw->mw_lock);
2024	}
2025
2026	return (0);
2027}
2028
2029static inline int
2030read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2031    int len)
2032{
2033
2034	return (rw_via_memwin(sc, idx, addr, val, len, 0));
2035}
2036
2037static inline int
2038write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
2039    const uint32_t *val, int len)
2040{
2041
2042	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
2043}
2044
2045static int
2046t4_range_cmp(const void *a, const void *b)
2047{
2048	return ((const struct t4_range *)a)->start -
2049	       ((const struct t4_range *)b)->start;
2050}
2051
2052/*
2053 * Verify that the memory range specified by the addr/len pair is valid within
2054 * the card's address space.
2055 */
2056static int
2057validate_mem_range(struct adapter *sc, uint32_t addr, int len)
2058{
2059	struct t4_range mem_ranges[4], *r, *next;
2060	uint32_t em, addr_len;
2061	int i, n, remaining;
2062
2063	/* Memory can only be accessed in naturally aligned 4 byte units */
2064	if (addr & 3 || len & 3 || len <= 0)
2065		return (EINVAL);
2066
2067	/* Enabled memories */
2068	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2069
2070	r = &mem_ranges[0];
2071	n = 0;
2072	bzero(r, sizeof(mem_ranges));
2073	if (em & F_EDRAM0_ENABLE) {
2074		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2075		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2076		if (r->size > 0) {
2077			r->start = G_EDRAM0_BASE(addr_len) << 20;
2078			if (addr >= r->start &&
2079			    addr + len <= r->start + r->size)
2080				return (0);
2081			r++;
2082			n++;
2083		}
2084	}
2085	if (em & F_EDRAM1_ENABLE) {
2086		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2087		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2088		if (r->size > 0) {
2089			r->start = G_EDRAM1_BASE(addr_len) << 20;
2090			if (addr >= r->start &&
2091			    addr + len <= r->start + r->size)
2092				return (0);
2093			r++;
2094			n++;
2095		}
2096	}
2097	if (em & F_EXT_MEM_ENABLE) {
2098		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2099		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2100		if (r->size > 0) {
2101			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2102			if (addr >= r->start &&
2103			    addr + len <= r->start + r->size)
2104				return (0);
2105			r++;
2106			n++;
2107		}
2108	}
2109	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2110		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2111		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2112		if (r->size > 0) {
2113			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2114			if (addr >= r->start &&
2115			    addr + len <= r->start + r->size)
2116				return (0);
2117			r++;
2118			n++;
2119		}
2120	}
2121	MPASS(n <= nitems(mem_ranges));
2122
2123	if (n > 1) {
2124		/* Sort and merge the ranges. */
2125		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2126
2127		/* Start from index 0 and examine the next n - 1 entries. */
2128		r = &mem_ranges[0];
2129		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2130
2131			MPASS(r->size > 0);	/* r is a valid entry. */
2132			next = r + 1;
2133			MPASS(next->size > 0);	/* and so is the next one. */
2134
2135			while (r->start + r->size >= next->start) {
2136				/* Merge the next one into the current entry. */
2137				r->size = max(r->start + r->size,
2138				    next->start + next->size) - r->start;
2139				n--;	/* One fewer entry in total. */
2140				if (--remaining == 0)
2141					goto done;	/* short circuit */
2142				next++;
2143			}
2144			if (next != r + 1) {
2145				/*
2146				 * Some entries were merged into r and next
2147				 * points to the first valid entry that couldn't
2148				 * be merged.
2149				 */
2150				MPASS(next->size > 0);	/* must be valid */
2151				memcpy(r + 1, next, remaining * sizeof(*r));
2152#ifdef INVARIANTS
2153				/*
2154				 * This so that the foo->size assertion in the
2155				 * next iteration of the loop do the right
2156				 * thing for entries that were pulled up and are
2157				 * no longer valid.
2158				 */
2159				MPASS(n < nitems(mem_ranges));
2160				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
2161				    sizeof(struct t4_range));
2162#endif
2163			}
2164		}
2165done:
2166		/* Done merging the ranges. */
2167		MPASS(n > 0);
2168		r = &mem_ranges[0];
2169		for (i = 0; i < n; i++, r++) {
2170			if (addr >= r->start &&
2171			    addr + len <= r->start + r->size)
2172				return (0);
2173		}
2174	}
2175
2176	return (EFAULT);
2177}
2178
2179static int
2180fwmtype_to_hwmtype(int mtype)
2181{
2182
2183	switch (mtype) {
2184	case FW_MEMTYPE_EDC0:
2185		return (MEM_EDC0);
2186	case FW_MEMTYPE_EDC1:
2187		return (MEM_EDC1);
2188	case FW_MEMTYPE_EXTMEM:
2189		return (MEM_MC0);
2190	case FW_MEMTYPE_EXTMEM1:
2191		return (MEM_MC1);
2192	default:
2193		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
2194	}
2195}
2196
2197/*
2198 * Verify that the memory range specified by the memtype/offset/len pair is
2199 * valid and lies entirely within the memtype specified.  The global address of
2200 * the start of the range is returned in addr.
2201 */
2202static int
2203validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
2204    uint32_t *addr)
2205{
2206	uint32_t em, addr_len, maddr;
2207
2208	/* Memory can only be accessed in naturally aligned 4 byte units */
2209	if (off & 3 || len & 3 || len == 0)
2210		return (EINVAL);
2211
2212	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2213	switch (fwmtype_to_hwmtype(mtype)) {
2214	case MEM_EDC0:
2215		if (!(em & F_EDRAM0_ENABLE))
2216			return (EINVAL);
2217		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2218		maddr = G_EDRAM0_BASE(addr_len) << 20;
2219		break;
2220	case MEM_EDC1:
2221		if (!(em & F_EDRAM1_ENABLE))
2222			return (EINVAL);
2223		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2224		maddr = G_EDRAM1_BASE(addr_len) << 20;
2225		break;
2226	case MEM_MC:
2227		if (!(em & F_EXT_MEM_ENABLE))
2228			return (EINVAL);
2229		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2230		maddr = G_EXT_MEM_BASE(addr_len) << 20;
2231		break;
2232	case MEM_MC1:
2233		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
2234			return (EINVAL);
2235		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2236		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
2237		break;
2238	default:
2239		return (EINVAL);
2240	}
2241
2242	*addr = maddr + off;	/* global address */
2243	return (validate_mem_range(sc, *addr, len));
2244}
2245
2246static int
2247fixup_devlog_params(struct adapter *sc)
2248{
2249	struct devlog_params *dparams = &sc->params.devlog;
2250	int rc;
2251
2252	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
2253	    dparams->size, &dparams->addr);
2254
2255	return (rc);
2256}
2257
2258static int
2259cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g, int num_vis,
2260    struct intrs_and_queues *iaq)
2261{
2262	int rc, itype, navail, nrxq10g, nrxq1g, n;
2263	int nofldrxq10g = 0, nofldrxq1g = 0;
2264
2265	bzero(iaq, sizeof(*iaq));
2266
2267	iaq->ntxq10g = t4_ntxq10g;
2268	iaq->ntxq1g = t4_ntxq1g;
2269	iaq->ntxq_vi = t4_ntxq_vi;
2270	iaq->nrxq10g = nrxq10g = t4_nrxq10g;
2271	iaq->nrxq1g = nrxq1g = t4_nrxq1g;
2272	iaq->nrxq_vi = t4_nrxq_vi;
2273	iaq->rsrv_noflowq = t4_rsrv_noflowq;
2274#ifdef TCP_OFFLOAD
2275	if (is_offload(sc)) {
2276		iaq->nofldtxq10g = t4_nofldtxq10g;
2277		iaq->nofldtxq1g = t4_nofldtxq1g;
2278		iaq->nofldtxq_vi = t4_nofldtxq_vi;
2279		iaq->nofldrxq10g = nofldrxq10g = t4_nofldrxq10g;
2280		iaq->nofldrxq1g = nofldrxq1g = t4_nofldrxq1g;
2281		iaq->nofldrxq_vi = t4_nofldrxq_vi;
2282	}
2283#endif
2284#ifdef DEV_NETMAP
2285	iaq->nnmtxq_vi = t4_nnmtxq_vi;
2286	iaq->nnmrxq_vi = t4_nnmrxq_vi;
2287#endif
2288
2289	for (itype = INTR_MSIX; itype; itype >>= 1) {
2290
2291		if ((itype & t4_intr_types) == 0)
2292			continue;	/* not allowed */
2293
2294		if (itype == INTR_MSIX)
2295			navail = pci_msix_count(sc->dev);
2296		else if (itype == INTR_MSI)
2297			navail = pci_msi_count(sc->dev);
2298		else
2299			navail = 1;
2300restart:
2301		if (navail == 0)
2302			continue;
2303
2304		iaq->intr_type = itype;
2305		iaq->intr_flags_10g = 0;
2306		iaq->intr_flags_1g = 0;
2307
2308		/*
2309		 * Best option: an interrupt vector for errors, one for the
2310		 * firmware event queue, and one for every rxq (NIC and TOE) of
2311		 * every VI.  The VIs that support netmap use the same
2312		 * interrupts for the NIC rx queues and the netmap rx queues
2313		 * because only one set of queues is active at a time.
2314		 */
2315		iaq->nirq = T4_EXTRA_INTR;
2316		iaq->nirq += n10g * (nrxq10g + nofldrxq10g);
2317		iaq->nirq += n1g * (nrxq1g + nofldrxq1g);
2318		iaq->nirq += (n10g + n1g) * (num_vis - 1) *
2319		    max(iaq->nrxq_vi, iaq->nnmrxq_vi);	/* See comment above. */
2320		iaq->nirq += (n10g + n1g) * (num_vis - 1) * iaq->nofldrxq_vi;
2321		if (iaq->nirq <= navail &&
2322		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
2323			iaq->intr_flags_10g = INTR_ALL;
2324			iaq->intr_flags_1g = INTR_ALL;
2325			goto allocate;
2326		}
2327
2328		/* Disable the VIs (and netmap) if there aren't enough intrs */
2329		if (num_vis > 1) {
2330			device_printf(sc->dev, "virtual interfaces disabled "
2331			    "because num_vis=%u with current settings "
2332			    "(nrxq10g=%u, nrxq1g=%u, nofldrxq10g=%u, "
2333			    "nofldrxq1g=%u, nrxq_vi=%u nofldrxq_vi=%u, "
2334			    "nnmrxq_vi=%u) would need %u interrupts but "
2335			    "only %u are available.\n", num_vis, nrxq10g,
2336			    nrxq1g, nofldrxq10g, nofldrxq1g, iaq->nrxq_vi,
2337			    iaq->nofldrxq_vi, iaq->nnmrxq_vi, iaq->nirq,
2338			    navail);
2339			num_vis = 1;
2340			iaq->ntxq_vi = iaq->nrxq_vi = 0;
2341			iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
2342			iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
2343			goto restart;
2344		}
2345
2346		/*
2347		 * Second best option: a vector for errors, one for the firmware
2348		 * event queue, and vectors for either all the NIC rx queues or
2349		 * all the TOE rx queues.  The queues that don't get vectors
2350		 * will forward their interrupts to those that do.
2351		 */
2352		iaq->nirq = T4_EXTRA_INTR;
2353		if (nrxq10g >= nofldrxq10g) {
2354			iaq->intr_flags_10g = INTR_RXQ;
2355			iaq->nirq += n10g * nrxq10g;
2356		} else {
2357			iaq->intr_flags_10g = INTR_OFLD_RXQ;
2358			iaq->nirq += n10g * nofldrxq10g;
2359		}
2360		if (nrxq1g >= nofldrxq1g) {
2361			iaq->intr_flags_1g = INTR_RXQ;
2362			iaq->nirq += n1g * nrxq1g;
2363		} else {
2364			iaq->intr_flags_1g = INTR_OFLD_RXQ;
2365			iaq->nirq += n1g * nofldrxq1g;
2366		}
2367		if (iaq->nirq <= navail &&
2368		    (itype != INTR_MSI || powerof2(iaq->nirq)))
2369			goto allocate;
2370
2371		/*
2372		 * Next best option: an interrupt vector for errors, one for the
2373		 * firmware event queue, and at least one per main-VI.  At this
2374		 * point we know we'll have to downsize nrxq and/or nofldrxq to
2375		 * fit what's available to us.
2376		 */
2377		iaq->nirq = T4_EXTRA_INTR;
2378		iaq->nirq += n10g + n1g;
2379		if (iaq->nirq <= navail) {
2380			int leftover = navail - iaq->nirq;
2381
2382			if (n10g > 0) {
2383				int target = max(nrxq10g, nofldrxq10g);
2384
2385				iaq->intr_flags_10g = nrxq10g >= nofldrxq10g ?
2386				    INTR_RXQ : INTR_OFLD_RXQ;
2387
2388				n = 1;
2389				while (n < target && leftover >= n10g) {
2390					leftover -= n10g;
2391					iaq->nirq += n10g;
2392					n++;
2393				}
2394				iaq->nrxq10g = min(n, nrxq10g);
2395#ifdef TCP_OFFLOAD
2396				iaq->nofldrxq10g = min(n, nofldrxq10g);
2397#endif
2398			}
2399
2400			if (n1g > 0) {
2401				int target = max(nrxq1g, nofldrxq1g);
2402
2403				iaq->intr_flags_1g = nrxq1g >= nofldrxq1g ?
2404				    INTR_RXQ : INTR_OFLD_RXQ;
2405
2406				n = 1;
2407				while (n < target && leftover >= n1g) {
2408					leftover -= n1g;
2409					iaq->nirq += n1g;
2410					n++;
2411				}
2412				iaq->nrxq1g = min(n, nrxq1g);
2413#ifdef TCP_OFFLOAD
2414				iaq->nofldrxq1g = min(n, nofldrxq1g);
2415#endif
2416			}
2417
2418			if (itype != INTR_MSI || powerof2(iaq->nirq))
2419				goto allocate;
2420		}
2421
2422		/*
2423		 * Least desirable option: one interrupt vector for everything.
2424		 */
2425		iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
2426		iaq->intr_flags_10g = iaq->intr_flags_1g = 0;
2427#ifdef TCP_OFFLOAD
2428		if (is_offload(sc))
2429			iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
2430#endif
2431allocate:
2432		navail = iaq->nirq;
2433		rc = 0;
2434		if (itype == INTR_MSIX)
2435			rc = pci_alloc_msix(sc->dev, &navail);
2436		else if (itype == INTR_MSI)
2437			rc = pci_alloc_msi(sc->dev, &navail);
2438
2439		if (rc == 0) {
2440			if (navail == iaq->nirq)
2441				return (0);
2442
2443			/*
2444			 * Didn't get the number requested.  Use whatever number
2445			 * the kernel is willing to allocate (it's in navail).
2446			 */
2447			device_printf(sc->dev, "fewer vectors than requested, "
2448			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
2449			    itype, iaq->nirq, navail);
2450			pci_release_msi(sc->dev);
2451			goto restart;
2452		}
2453
2454		device_printf(sc->dev,
2455		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
2456		    itype, rc, iaq->nirq, navail);
2457	}
2458
2459	device_printf(sc->dev,
2460	    "failed to find a usable interrupt type.  "
2461	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
2462	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
2463
2464	return (ENXIO);
2465}
2466
2467#define FW_VERSION(chip) ( \
2468    V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
2469    V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
2470    V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
2471    V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
2472#define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
2473
2474struct fw_info {
2475	uint8_t chip;
2476	char *kld_name;
2477	char *fw_mod_name;
2478	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
2479} fw_info[] = {
2480	{
2481		.chip = CHELSIO_T4,
2482		.kld_name = "t4fw_cfg",
2483		.fw_mod_name = "t4fw",
2484		.fw_hdr = {
2485			.chip = FW_HDR_CHIP_T4,
2486			.fw_ver = htobe32_const(FW_VERSION(T4)),
2487			.intfver_nic = FW_INTFVER(T4, NIC),
2488			.intfver_vnic = FW_INTFVER(T4, VNIC),
2489			.intfver_ofld = FW_INTFVER(T4, OFLD),
2490			.intfver_ri = FW_INTFVER(T4, RI),
2491			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
2492			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
2493			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
2494			.intfver_fcoe = FW_INTFVER(T4, FCOE),
2495		},
2496	}, {
2497		.chip = CHELSIO_T5,
2498		.kld_name = "t5fw_cfg",
2499		.fw_mod_name = "t5fw",
2500		.fw_hdr = {
2501			.chip = FW_HDR_CHIP_T5,
2502			.fw_ver = htobe32_const(FW_VERSION(T5)),
2503			.intfver_nic = FW_INTFVER(T5, NIC),
2504			.intfver_vnic = FW_INTFVER(T5, VNIC),
2505			.intfver_ofld = FW_INTFVER(T5, OFLD),
2506			.intfver_ri = FW_INTFVER(T5, RI),
2507			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
2508			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
2509			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
2510			.intfver_fcoe = FW_INTFVER(T5, FCOE),
2511		},
2512	}
2513};
2514
2515static struct fw_info *
2516find_fw_info(int chip)
2517{
2518	int i;
2519
2520	for (i = 0; i < nitems(fw_info); i++) {
2521		if (fw_info[i].chip == chip)
2522			return (&fw_info[i]);
2523	}
2524	return (NULL);
2525}
2526
2527/*
2528 * Is the given firmware API compatible with the one the driver was compiled
2529 * with?
2530 */
2531static int
2532fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
2533{
2534
2535	/* short circuit if it's the exact same firmware version */
2536	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
2537		return (1);
2538
2539	/*
2540	 * XXX: Is this too conservative?  Perhaps I should limit this to the
2541	 * features that are supported in the driver.
2542	 */
2543#define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
2544	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
2545	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
2546	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
2547		return (1);
2548#undef SAME_INTF
2549
2550	return (0);
2551}
2552
2553/*
2554 * The firmware in the KLD is usable, but should it be installed?  This routine
2555 * explains itself in detail if it indicates the KLD firmware should be
2556 * installed.
2557 */
2558static int
2559should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
2560{
2561	const char *reason;
2562
2563	if (!card_fw_usable) {
2564		reason = "incompatible or unusable";
2565		goto install;
2566	}
2567
2568	if (k > c) {
2569		reason = "older than the version bundled with this driver";
2570		goto install;
2571	}
2572
2573	if (t4_fw_install == 2 && k != c) {
2574		reason = "different than the version bundled with this driver";
2575		goto install;
2576	}
2577
2578	return (0);
2579
2580install:
2581	if (t4_fw_install == 0) {
2582		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2583		    "but the driver is prohibited from installing a different "
2584		    "firmware on the card.\n",
2585		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2586		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
2587
2588		return (0);
2589	}
2590
2591	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
2592	    "installing firmware %u.%u.%u.%u on card.\n",
2593	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2594	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
2595	    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2596	    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2597
2598	return (1);
2599}
2600/*
2601 * Establish contact with the firmware and determine if we are the master driver
2602 * or not, and whether we are responsible for chip initialization.
2603 */
2604static int
2605prep_firmware(struct adapter *sc)
2606{
2607	const struct firmware *fw = NULL, *default_cfg;
2608	int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
2609	enum dev_state state;
2610	struct fw_info *fw_info;
2611	struct fw_hdr *card_fw;		/* fw on the card */
2612	const struct fw_hdr *kld_fw;	/* fw in the KLD */
2613	const struct fw_hdr *drv_fw;	/* fw header the driver was compiled
2614					   against */
2615
2616	/* Contact firmware. */
2617	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
2618	if (rc < 0 || state == DEV_STATE_ERR) {
2619		rc = -rc;
2620		device_printf(sc->dev,
2621		    "failed to connect to the firmware: %d, %d.\n", rc, state);
2622		return (rc);
2623	}
2624	pf = rc;
2625	if (pf == sc->mbox)
2626		sc->flags |= MASTER_PF;
2627	else if (state == DEV_STATE_UNINIT) {
2628		/*
2629		 * We didn't get to be the master so we definitely won't be
2630		 * configuring the chip.  It's a bug if someone else hasn't
2631		 * configured it already.
2632		 */
2633		device_printf(sc->dev, "couldn't be master(%d), "
2634		    "device not already initialized either(%d).\n", rc, state);
2635		return (EDOOFUS);
2636	}
2637
2638	/* This is the firmware whose headers the driver was compiled against */
2639	fw_info = find_fw_info(chip_id(sc));
2640	if (fw_info == NULL) {
2641		device_printf(sc->dev,
2642		    "unable to look up firmware information for chip %d.\n",
2643		    chip_id(sc));
2644		return (EINVAL);
2645	}
2646	drv_fw = &fw_info->fw_hdr;
2647
2648	/*
2649	 * The firmware KLD contains many modules.  The KLD name is also the
2650	 * name of the module that contains the default config file.
2651	 */
2652	default_cfg = firmware_get(fw_info->kld_name);
2653
2654	/* Read the header of the firmware on the card */
2655	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
2656	rc = -t4_read_flash(sc, FLASH_FW_START,
2657	    sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
2658	if (rc == 0)
2659		card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
2660	else {
2661		device_printf(sc->dev,
2662		    "Unable to read card's firmware header: %d\n", rc);
2663		card_fw_usable = 0;
2664	}
2665
2666	/* This is the firmware in the KLD */
2667	fw = firmware_get(fw_info->fw_mod_name);
2668	if (fw != NULL) {
2669		kld_fw = (const void *)fw->data;
2670		kld_fw_usable = fw_compatible(drv_fw, kld_fw);
2671	} else {
2672		kld_fw = NULL;
2673		kld_fw_usable = 0;
2674	}
2675
2676	if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
2677	    (!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
2678		/*
2679		 * Common case: the firmware on the card is an exact match and
2680		 * the KLD is an exact match too, or the KLD is
2681		 * absent/incompatible.  Note that t4_fw_install = 2 is ignored
2682		 * here -- use cxgbetool loadfw if you want to reinstall the
2683		 * same firmware as the one on the card.
2684		 */
2685	} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
2686	    should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
2687	    be32toh(card_fw->fw_ver))) {
2688
2689		rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
2690		if (rc != 0) {
2691			device_printf(sc->dev,
2692			    "failed to install firmware: %d\n", rc);
2693			goto done;
2694		}
2695
2696		/* Installed successfully, update the cached header too. */
2697		memcpy(card_fw, kld_fw, sizeof(*card_fw));
2698		card_fw_usable = 1;
2699		need_fw_reset = 0;	/* already reset as part of load_fw */
2700	}
2701
2702	if (!card_fw_usable) {
2703		uint32_t d, c, k;
2704
2705		d = ntohl(drv_fw->fw_ver);
2706		c = ntohl(card_fw->fw_ver);
2707		k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
2708
2709		device_printf(sc->dev, "Cannot find a usable firmware: "
2710		    "fw_install %d, chip state %d, "
2711		    "driver compiled with %d.%d.%d.%d, "
2712		    "card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
2713		    t4_fw_install, state,
2714		    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
2715		    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
2716		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
2717		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
2718		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
2719		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
2720		rc = EINVAL;
2721		goto done;
2722	}
2723
2724	/* We're using whatever's on the card and it's known to be good. */
2725	sc->params.fw_vers = ntohl(card_fw->fw_ver);
2726	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
2727	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
2728	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
2729	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
2730	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
2731
2732	t4_get_tp_version(sc, &sc->params.tp_vers);
2733	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
2734	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
2735	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
2736	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
2737	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
2738
2739	if (t4_get_exprom_version(sc, &sc->params.exprom_vers) != 0)
2740		sc->params.exprom_vers = 0;
2741	else {
2742		snprintf(sc->exprom_version, sizeof(sc->exprom_version),
2743		    "%u.%u.%u.%u",
2744		    G_FW_HDR_FW_VER_MAJOR(sc->params.exprom_vers),
2745		    G_FW_HDR_FW_VER_MINOR(sc->params.exprom_vers),
2746		    G_FW_HDR_FW_VER_MICRO(sc->params.exprom_vers),
2747		    G_FW_HDR_FW_VER_BUILD(sc->params.exprom_vers));
2748	}
2749
2750	/* Reset device */
2751	if (need_fw_reset &&
2752	    (rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
2753		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
2754		if (rc != ETIMEDOUT && rc != EIO)
2755			t4_fw_bye(sc, sc->mbox);
2756		goto done;
2757	}
2758	sc->flags |= FW_OK;
2759
2760	rc = get_params__pre_init(sc);
2761	if (rc != 0)
2762		goto done; /* error message displayed already */
2763
2764	/* Partition adapter resources as specified in the config file. */
2765	if (state == DEV_STATE_UNINIT) {
2766
2767		KASSERT(sc->flags & MASTER_PF,
2768		    ("%s: trying to change chip settings when not master.",
2769		    __func__));
2770
2771		rc = partition_resources(sc, default_cfg, fw_info->kld_name);
2772		if (rc != 0)
2773			goto done;	/* error message displayed already */
2774
2775		t4_tweak_chip_settings(sc);
2776
2777		/* get basic stuff going */
2778		rc = -t4_fw_initialize(sc, sc->mbox);
2779		if (rc != 0) {
2780			device_printf(sc->dev, "fw init failed: %d.\n", rc);
2781			goto done;
2782		}
2783	} else {
2784		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
2785		sc->cfcsum = 0;
2786	}
2787
2788done:
2789	free(card_fw, M_CXGBE);
2790	if (fw != NULL)
2791		firmware_put(fw, FIRMWARE_UNLOAD);
2792	if (default_cfg != NULL)
2793		firmware_put(default_cfg, FIRMWARE_UNLOAD);
2794
2795	return (rc);
2796}
2797
2798#define FW_PARAM_DEV(param) \
2799	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
2800	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
2801#define FW_PARAM_PFVF(param) \
2802	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
2803	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
2804
2805/*
2806 * Partition chip resources for use between various PFs, VFs, etc.
2807 */
2808static int
2809partition_resources(struct adapter *sc, const struct firmware *default_cfg,
2810    const char *name_prefix)
2811{
2812	const struct firmware *cfg = NULL;
2813	int rc = 0;
2814	struct fw_caps_config_cmd caps;
2815	uint32_t mtype, moff, finicsum, cfcsum;
2816
2817	/*
2818	 * Figure out what configuration file to use.  Pick the default config
2819	 * file for the card if the user hasn't specified one explicitly.
2820	 */
2821	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
2822	if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
2823		/* Card specific overrides go here. */
2824		if (pci_get_device(sc->dev) == 0x440a)
2825			snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
2826		if (is_fpga(sc))
2827			snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
2828	}
2829
2830	/*
2831	 * We need to load another module if the profile is anything except
2832	 * "default" or "flash".
2833	 */
2834	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
2835	    strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
2836		char s[32];
2837
2838		snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
2839		cfg = firmware_get(s);
2840		if (cfg == NULL) {
2841			if (default_cfg != NULL) {
2842				device_printf(sc->dev,
2843				    "unable to load module \"%s\" for "
2844				    "configuration profile \"%s\", will use "
2845				    "the default config file instead.\n",
2846				    s, sc->cfg_file);
2847				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
2848				    "%s", DEFAULT_CF);
2849			} else {
2850				device_printf(sc->dev,
2851				    "unable to load module \"%s\" for "
2852				    "configuration profile \"%s\", will use "
2853				    "the config file on the card's flash "
2854				    "instead.\n", s, sc->cfg_file);
2855				snprintf(sc->cfg_file, sizeof(sc->cfg_file),
2856				    "%s", FLASH_CF);
2857			}
2858		}
2859	}
2860
2861	if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
2862	    default_cfg == NULL) {
2863		device_printf(sc->dev,
2864		    "default config file not available, will use the config "
2865		    "file on the card's flash instead.\n");
2866		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
2867	}
2868
2869	if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
2870		u_int cflen;
2871		const uint32_t *cfdata;
2872		uint32_t param, val, addr;
2873
2874		KASSERT(cfg != NULL || default_cfg != NULL,
2875		    ("%s: no config to upload", __func__));
2876
2877		/*
2878		 * Ask the firmware where it wants us to upload the config file.
2879		 */
2880		param = FW_PARAM_DEV(CF);
2881		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2882		if (rc != 0) {
2883			/* No support for config file?  Shouldn't happen. */
2884			device_printf(sc->dev,
2885			    "failed to query config file location: %d.\n", rc);
2886			goto done;
2887		}
2888		mtype = G_FW_PARAMS_PARAM_Y(val);
2889		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
2890
2891		/*
2892		 * XXX: sheer laziness.  We deliberately added 4 bytes of
2893		 * useless stuffing/comments at the end of the config file so
2894		 * it's ok to simply throw away the last remaining bytes when
2895		 * the config file is not an exact multiple of 4.  This also
2896		 * helps with the validate_mt_off_len check.
2897		 */
2898		if (cfg != NULL) {
2899			cflen = cfg->datasize & ~3;
2900			cfdata = cfg->data;
2901		} else {
2902			cflen = default_cfg->datasize & ~3;
2903			cfdata = default_cfg->data;
2904		}
2905
2906		if (cflen > FLASH_CFG_MAX_SIZE) {
2907			device_printf(sc->dev,
2908			    "config file too long (%d, max allowed is %d).  "
2909			    "Will try to use the config on the card, if any.\n",
2910			    cflen, FLASH_CFG_MAX_SIZE);
2911			goto use_config_on_flash;
2912		}
2913
2914		rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
2915		if (rc != 0) {
2916			device_printf(sc->dev,
2917			    "%s: addr (%d/0x%x) or len %d is not valid: %d.  "
2918			    "Will try to use the config on the card, if any.\n",
2919			    __func__, mtype, moff, cflen, rc);
2920			goto use_config_on_flash;
2921		}
2922		write_via_memwin(sc, 2, addr, cfdata, cflen);
2923	} else {
2924use_config_on_flash:
2925		mtype = FW_MEMTYPE_FLASH;
2926		moff = t4_flash_cfg_addr(sc);
2927	}
2928
2929	bzero(&caps, sizeof(caps));
2930	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2931	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
2932	caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
2933	    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
2934	    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
2935	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
2936	if (rc != 0) {
2937		device_printf(sc->dev,
2938		    "failed to pre-process config file: %d "
2939		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
2940		goto done;
2941	}
2942
2943	finicsum = be32toh(caps.finicsum);
2944	cfcsum = be32toh(caps.cfcsum);
2945	if (finicsum != cfcsum) {
2946		device_printf(sc->dev,
2947		    "WARNING: config file checksum mismatch: %08x %08x\n",
2948		    finicsum, cfcsum);
2949	}
2950	sc->cfcsum = cfcsum;
2951
2952#define LIMIT_CAPS(x) do { \
2953	caps.x &= htobe16(t4_##x##_allowed); \
2954} while (0)
2955
2956	/*
2957	 * Let the firmware know what features will (not) be used so it can tune
2958	 * things accordingly.
2959	 */
2960	LIMIT_CAPS(nbmcaps);
2961	LIMIT_CAPS(linkcaps);
2962	LIMIT_CAPS(switchcaps);
2963	LIMIT_CAPS(niccaps);
2964	LIMIT_CAPS(toecaps);
2965	LIMIT_CAPS(rdmacaps);
2966	LIMIT_CAPS(tlscaps);
2967	LIMIT_CAPS(iscsicaps);
2968	LIMIT_CAPS(fcoecaps);
2969#undef LIMIT_CAPS
2970
2971	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
2972	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
2973	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
2974	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
2975	if (rc != 0) {
2976		device_printf(sc->dev,
2977		    "failed to process config file: %d.\n", rc);
2978	}
2979done:
2980	if (cfg != NULL)
2981		firmware_put(cfg, FIRMWARE_UNLOAD);
2982	return (rc);
2983}
2984
2985/*
2986 * Retrieve parameters that are needed (or nice to have) very early.
2987 */
2988static int
2989get_params__pre_init(struct adapter *sc)
2990{
2991	int rc;
2992	uint32_t param[2], val[2];
2993
2994	param[0] = FW_PARAM_DEV(PORTVEC);
2995	param[1] = FW_PARAM_DEV(CCLK);
2996	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
2997	if (rc != 0) {
2998		device_printf(sc->dev,
2999		    "failed to query parameters (pre_init): %d.\n", rc);
3000		return (rc);
3001	}
3002
3003	sc->params.portvec = val[0];
3004	sc->params.nports = bitcount32(val[0]);
3005	sc->params.vpd.cclk = val[1];
3006
3007	/* Read device log parameters. */
3008	rc = -t4_init_devlog_params(sc, 1);
3009	if (rc == 0)
3010		fixup_devlog_params(sc);
3011	else {
3012		device_printf(sc->dev,
3013		    "failed to get devlog parameters: %d.\n", rc);
3014		rc = 0;	/* devlog isn't critical for device operation */
3015	}
3016
3017	return (rc);
3018}
3019
3020/*
3021 * Retrieve various parameters that are of interest to the driver.  The device
3022 * has been initialized by the firmware at this point.
3023 */
3024static int
3025get_params__post_init(struct adapter *sc)
3026{
3027	int rc;
3028	uint32_t param[7], val[7];
3029	struct fw_caps_config_cmd caps;
3030
3031	param[0] = FW_PARAM_PFVF(IQFLINT_START);
3032	param[1] = FW_PARAM_PFVF(EQ_START);
3033	param[2] = FW_PARAM_PFVF(FILTER_START);
3034	param[3] = FW_PARAM_PFVF(FILTER_END);
3035	param[4] = FW_PARAM_PFVF(L2T_START);
3036	param[5] = FW_PARAM_PFVF(L2T_END);
3037	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3038	if (rc != 0) {
3039		device_printf(sc->dev,
3040		    "failed to query parameters (post_init): %d.\n", rc);
3041		return (rc);
3042	}
3043
3044	sc->sge.iq_start = val[0];
3045	sc->sge.eq_start = val[1];
3046	sc->tids.ftid_base = val[2];
3047	sc->tids.nftids = val[3] - val[2] + 1;
3048	sc->params.ftid_min = val[2];
3049	sc->params.ftid_max = val[3];
3050	sc->vres.l2t.start = val[4];
3051	sc->vres.l2t.size = val[5] - val[4] + 1;
3052	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
3053	    ("%s: L2 table size (%u) larger than expected (%u)",
3054	    __func__, sc->vres.l2t.size, L2T_SIZE));
3055
3056	/* get capabilites */
3057	bzero(&caps, sizeof(caps));
3058	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3059	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3060	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3061	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3062	if (rc != 0) {
3063		device_printf(sc->dev,
3064		    "failed to get card capabilities: %d.\n", rc);
3065		return (rc);
3066	}
3067
3068#define READ_CAPS(x) do { \
3069	sc->x = htobe16(caps.x); \
3070} while (0)
3071	READ_CAPS(nbmcaps);
3072	READ_CAPS(linkcaps);
3073	READ_CAPS(switchcaps);
3074	READ_CAPS(niccaps);
3075	READ_CAPS(toecaps);
3076	READ_CAPS(rdmacaps);
3077	READ_CAPS(tlscaps);
3078	READ_CAPS(iscsicaps);
3079	READ_CAPS(fcoecaps);
3080
3081	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
3082		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
3083		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
3084		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3085		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
3086		if (rc != 0) {
3087			device_printf(sc->dev,
3088			    "failed to query NIC parameters: %d.\n", rc);
3089			return (rc);
3090		}
3091		sc->tids.etid_base = val[0];
3092		sc->params.etid_min = val[0];
3093		sc->tids.netids = val[1] - val[0] + 1;
3094		sc->params.netids = sc->tids.netids;
3095		sc->params.eo_wr_cred = val[2];
3096		sc->params.ethoffload = 1;
3097	}
3098
3099	if (sc->toecaps) {
3100		/* query offload-related parameters */
3101		param[0] = FW_PARAM_DEV(NTID);
3102		param[1] = FW_PARAM_PFVF(SERVER_START);
3103		param[2] = FW_PARAM_PFVF(SERVER_END);
3104		param[3] = FW_PARAM_PFVF(TDDP_START);
3105		param[4] = FW_PARAM_PFVF(TDDP_END);
3106		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3107		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3108		if (rc != 0) {
3109			device_printf(sc->dev,
3110			    "failed to query TOE parameters: %d.\n", rc);
3111			return (rc);
3112		}
3113		sc->tids.ntids = val[0];
3114		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
3115		sc->tids.stid_base = val[1];
3116		sc->tids.nstids = val[2] - val[1] + 1;
3117		sc->vres.ddp.start = val[3];
3118		sc->vres.ddp.size = val[4] - val[3] + 1;
3119		sc->params.ofldq_wr_cred = val[5];
3120		sc->params.offload = 1;
3121	}
3122	if (sc->rdmacaps) {
3123		param[0] = FW_PARAM_PFVF(STAG_START);
3124		param[1] = FW_PARAM_PFVF(STAG_END);
3125		param[2] = FW_PARAM_PFVF(RQ_START);
3126		param[3] = FW_PARAM_PFVF(RQ_END);
3127		param[4] = FW_PARAM_PFVF(PBL_START);
3128		param[5] = FW_PARAM_PFVF(PBL_END);
3129		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3130		if (rc != 0) {
3131			device_printf(sc->dev,
3132			    "failed to query RDMA parameters(1): %d.\n", rc);
3133			return (rc);
3134		}
3135		sc->vres.stag.start = val[0];
3136		sc->vres.stag.size = val[1] - val[0] + 1;
3137		sc->vres.rq.start = val[2];
3138		sc->vres.rq.size = val[3] - val[2] + 1;
3139		sc->vres.pbl.start = val[4];
3140		sc->vres.pbl.size = val[5] - val[4] + 1;
3141
3142		param[0] = FW_PARAM_PFVF(SQRQ_START);
3143		param[1] = FW_PARAM_PFVF(SQRQ_END);
3144		param[2] = FW_PARAM_PFVF(CQ_START);
3145		param[3] = FW_PARAM_PFVF(CQ_END);
3146		param[4] = FW_PARAM_PFVF(OCQ_START);
3147		param[5] = FW_PARAM_PFVF(OCQ_END);
3148		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
3149		if (rc != 0) {
3150			device_printf(sc->dev,
3151			    "failed to query RDMA parameters(2): %d.\n", rc);
3152			return (rc);
3153		}
3154		sc->vres.qp.start = val[0];
3155		sc->vres.qp.size = val[1] - val[0] + 1;
3156		sc->vres.cq.start = val[2];
3157		sc->vres.cq.size = val[3] - val[2] + 1;
3158		sc->vres.ocq.start = val[4];
3159		sc->vres.ocq.size = val[5] - val[4] + 1;
3160	}
3161	if (sc->iscsicaps) {
3162		param[0] = FW_PARAM_PFVF(ISCSI_START);
3163		param[1] = FW_PARAM_PFVF(ISCSI_END);
3164		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3165		if (rc != 0) {
3166			device_printf(sc->dev,
3167			    "failed to query iSCSI parameters: %d.\n", rc);
3168			return (rc);
3169		}
3170		sc->vres.iscsi.start = val[0];
3171		sc->vres.iscsi.size = val[1] - val[0] + 1;
3172	}
3173
3174	/*
3175	 * We've got the params we wanted to query via the firmware.  Now grab
3176	 * some others directly from the chip.
3177	 */
3178	rc = t4_read_chip_settings(sc);
3179
3180	return (rc);
3181}
3182
3183static int
3184set_params__post_init(struct adapter *sc)
3185{
3186	uint32_t param, val;
3187
3188	/* ask for encapsulated CPLs */
3189	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3190	val = 1;
3191	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3192
3193	return (0);
3194}
3195
3196#undef FW_PARAM_PFVF
3197#undef FW_PARAM_DEV
3198
3199static void
3200t4_set_desc(struct adapter *sc)
3201{
3202	char buf[128];
3203	struct adapter_params *p = &sc->params;
3204
3205	snprintf(buf, sizeof(buf), "Chelsio %s %sNIC (rev %d), S/N:%s, "
3206	    "P/N:%s, E/C:%s", p->vpd.id, is_offload(sc) ? "R" : "",
3207	    chip_rev(sc), p->vpd.sn, p->vpd.pn, p->vpd.ec);
3208
3209	device_set_desc_copy(sc->dev, buf);
3210}
3211
3212static void
3213build_medialist(struct port_info *pi, struct ifmedia *media)
3214{
3215	int m;
3216
3217	PORT_LOCK(pi);
3218
3219	ifmedia_removeall(media);
3220
3221	m = IFM_ETHER | IFM_FDX;
3222
3223	switch(pi->port_type) {
3224	case FW_PORT_TYPE_BT_XFI:
3225	case FW_PORT_TYPE_BT_XAUI:
3226		ifmedia_add(media, m | IFM_10G_T, 0, NULL);
3227		/* fall through */
3228
3229	case FW_PORT_TYPE_BT_SGMII:
3230		ifmedia_add(media, m | IFM_1000_T, 0, NULL);
3231		ifmedia_add(media, m | IFM_100_TX, 0, NULL);
3232		ifmedia_add(media, IFM_ETHER | IFM_AUTO, 0, NULL);
3233		ifmedia_set(media, IFM_ETHER | IFM_AUTO);
3234		break;
3235
3236	case FW_PORT_TYPE_CX4:
3237		ifmedia_add(media, m | IFM_10G_CX4, 0, NULL);
3238		ifmedia_set(media, m | IFM_10G_CX4);
3239		break;
3240
3241	case FW_PORT_TYPE_QSFP_10G:
3242	case FW_PORT_TYPE_SFP:
3243	case FW_PORT_TYPE_FIBER_XFI:
3244	case FW_PORT_TYPE_FIBER_XAUI:
3245		switch (pi->mod_type) {
3246
3247		case FW_PORT_MOD_TYPE_LR:
3248			ifmedia_add(media, m | IFM_10G_LR, 0, NULL);
3249			ifmedia_set(media, m | IFM_10G_LR);
3250			break;
3251
3252		case FW_PORT_MOD_TYPE_SR:
3253			ifmedia_add(media, m | IFM_10G_SR, 0, NULL);
3254			ifmedia_set(media, m | IFM_10G_SR);
3255			break;
3256
3257		case FW_PORT_MOD_TYPE_LRM:
3258			ifmedia_add(media, m | IFM_10G_LRM, 0, NULL);
3259			ifmedia_set(media, m | IFM_10G_LRM);
3260			break;
3261
3262		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3263		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3264			ifmedia_add(media, m | IFM_10G_TWINAX, 0, NULL);
3265			ifmedia_set(media, m | IFM_10G_TWINAX);
3266			break;
3267
3268		case FW_PORT_MOD_TYPE_NONE:
3269			m &= ~IFM_FDX;
3270			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3271			ifmedia_set(media, m | IFM_NONE);
3272			break;
3273
3274		case FW_PORT_MOD_TYPE_NA:
3275		case FW_PORT_MOD_TYPE_ER:
3276		default:
3277			device_printf(pi->dev,
3278			    "unknown port_type (%d), mod_type (%d)\n",
3279			    pi->port_type, pi->mod_type);
3280			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3281			ifmedia_set(media, m | IFM_UNKNOWN);
3282			break;
3283		}
3284		break;
3285
3286	case FW_PORT_TYPE_QSFP:
3287		switch (pi->mod_type) {
3288
3289		case FW_PORT_MOD_TYPE_LR:
3290			ifmedia_add(media, m | IFM_40G_LR4, 0, NULL);
3291			ifmedia_set(media, m | IFM_40G_LR4);
3292			break;
3293
3294		case FW_PORT_MOD_TYPE_SR:
3295			ifmedia_add(media, m | IFM_40G_SR4, 0, NULL);
3296			ifmedia_set(media, m | IFM_40G_SR4);
3297			break;
3298
3299		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
3300		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
3301			ifmedia_add(media, m | IFM_40G_CR4, 0, NULL);
3302			ifmedia_set(media, m | IFM_40G_CR4);
3303			break;
3304
3305		case FW_PORT_MOD_TYPE_NONE:
3306			m &= ~IFM_FDX;
3307			ifmedia_add(media, m | IFM_NONE, 0, NULL);
3308			ifmedia_set(media, m | IFM_NONE);
3309			break;
3310
3311		default:
3312			device_printf(pi->dev,
3313			    "unknown port_type (%d), mod_type (%d)\n",
3314			    pi->port_type, pi->mod_type);
3315			ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3316			ifmedia_set(media, m | IFM_UNKNOWN);
3317			break;
3318		}
3319		break;
3320
3321	default:
3322		device_printf(pi->dev,
3323		    "unknown port_type (%d), mod_type (%d)\n", pi->port_type,
3324		    pi->mod_type);
3325		ifmedia_add(media, m | IFM_UNKNOWN, 0, NULL);
3326		ifmedia_set(media, m | IFM_UNKNOWN);
3327		break;
3328	}
3329
3330	PORT_UNLOCK(pi);
3331}
3332
3333#define FW_MAC_EXACT_CHUNK	7
3334
3335/*
3336 * Program the port's XGMAC based on parameters in ifnet.  The caller also
3337 * indicates which parameters should be programmed (the rest are left alone).
3338 */
3339int
3340update_mac_settings(struct ifnet *ifp, int flags)
3341{
3342	int rc = 0;
3343	struct vi_info *vi = ifp->if_softc;
3344	struct port_info *pi = vi->pi;
3345	struct adapter *sc = pi->adapter;
3346	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
3347
3348	ASSERT_SYNCHRONIZED_OP(sc);
3349	KASSERT(flags, ("%s: not told what to update.", __func__));
3350
3351	if (flags & XGMAC_MTU)
3352		mtu = ifp->if_mtu;
3353
3354	if (flags & XGMAC_PROMISC)
3355		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
3356
3357	if (flags & XGMAC_ALLMULTI)
3358		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
3359
3360	if (flags & XGMAC_VLANEX)
3361		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
3362
3363	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
3364		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
3365		    allmulti, 1, vlanex, false);
3366		if (rc) {
3367			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
3368			    rc);
3369			return (rc);
3370		}
3371	}
3372
3373	if (flags & XGMAC_UCADDR) {
3374		uint8_t ucaddr[ETHER_ADDR_LEN];
3375
3376		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
3377		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
3378		    ucaddr, true, true);
3379		if (rc < 0) {
3380			rc = -rc;
3381			if_printf(ifp, "change_mac failed: %d\n", rc);
3382			return (rc);
3383		} else {
3384			vi->xact_addr_filt = rc;
3385			rc = 0;
3386		}
3387	}
3388
3389	if (flags & XGMAC_MCADDRS) {
3390		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
3391		int del = 1;
3392		uint64_t hash = 0;
3393		struct ifmultiaddr *ifma;
3394		int i = 0, j;
3395
3396		if_maddr_rlock(ifp);
3397		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3398			if (ifma->ifma_addr->sa_family != AF_LINK)
3399				continue;
3400			mcaddr[i] =
3401			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
3402			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
3403			i++;
3404
3405			if (i == FW_MAC_EXACT_CHUNK) {
3406				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
3407				    del, i, mcaddr, NULL, &hash, 0);
3408				if (rc < 0) {
3409					rc = -rc;
3410					for (j = 0; j < i; j++) {
3411						if_printf(ifp,
3412						    "failed to add mc address"
3413						    " %02x:%02x:%02x:"
3414						    "%02x:%02x:%02x rc=%d\n",
3415						    mcaddr[j][0], mcaddr[j][1],
3416						    mcaddr[j][2], mcaddr[j][3],
3417						    mcaddr[j][4], mcaddr[j][5],
3418						    rc);
3419					}
3420					goto mcfail;
3421				}
3422				del = 0;
3423				i = 0;
3424			}
3425		}
3426		if (i > 0) {
3427			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
3428			    mcaddr, NULL, &hash, 0);
3429			if (rc < 0) {
3430				rc = -rc;
3431				for (j = 0; j < i; j++) {
3432					if_printf(ifp,
3433					    "failed to add mc address"
3434					    " %02x:%02x:%02x:"
3435					    "%02x:%02x:%02x rc=%d\n",
3436					    mcaddr[j][0], mcaddr[j][1],
3437					    mcaddr[j][2], mcaddr[j][3],
3438					    mcaddr[j][4], mcaddr[j][5],
3439					    rc);
3440				}
3441				goto mcfail;
3442			}
3443		}
3444
3445		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
3446		if (rc != 0)
3447			if_printf(ifp, "failed to set mc address hash: %d", rc);
3448mcfail:
3449		if_maddr_runlock(ifp);
3450	}
3451
3452	return (rc);
3453}
3454
3455/*
3456 * {begin|end}_synchronized_op must be called from the same thread.
3457 */
3458int
3459begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
3460    char *wmesg)
3461{
3462	int rc, pri;
3463
3464#ifdef WITNESS
3465	/* the caller thinks it's ok to sleep, but is it really? */
3466	if (flags & SLEEP_OK)
3467		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
3468		    "begin_synchronized_op");
3469#endif
3470
3471	if (INTR_OK)
3472		pri = PCATCH;
3473	else
3474		pri = 0;
3475
3476	ADAPTER_LOCK(sc);
3477	for (;;) {
3478
3479		if (vi && IS_DOOMED(vi)) {
3480			rc = ENXIO;
3481			goto done;
3482		}
3483
3484		if (!IS_BUSY(sc)) {
3485			rc = 0;
3486			break;
3487		}
3488
3489		if (!(flags & SLEEP_OK)) {
3490			rc = EBUSY;
3491			goto done;
3492		}
3493
3494		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
3495			rc = EINTR;
3496			goto done;
3497		}
3498	}
3499
3500	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
3501	SET_BUSY(sc);
3502#ifdef INVARIANTS
3503	sc->last_op = wmesg;
3504	sc->last_op_thr = curthread;
3505	sc->last_op_flags = flags;
3506#endif
3507
3508done:
3509	if (!(flags & HOLD_LOCK) || rc)
3510		ADAPTER_UNLOCK(sc);
3511
3512	return (rc);
3513}
3514
3515/*
3516 * Tell if_ioctl and if_init that the VI is going away.  This is
3517 * special variant of begin_synchronized_op and must be paired with a
3518 * call to end_synchronized_op.
3519 */
3520void
3521doom_vi(struct adapter *sc, struct vi_info *vi)
3522{
3523
3524	ADAPTER_LOCK(sc);
3525	SET_DOOMED(vi);
3526	wakeup(&sc->flags);
3527	while (IS_BUSY(sc))
3528		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
3529	SET_BUSY(sc);
3530#ifdef INVARIANTS
3531	sc->last_op = "t4detach";
3532	sc->last_op_thr = curthread;
3533	sc->last_op_flags = 0;
3534#endif
3535	ADAPTER_UNLOCK(sc);
3536}
3537
3538/*
3539 * {begin|end}_synchronized_op must be called from the same thread.
3540 */
3541void
3542end_synchronized_op(struct adapter *sc, int flags)
3543{
3544
3545	if (flags & LOCK_HELD)
3546		ADAPTER_LOCK_ASSERT_OWNED(sc);
3547	else
3548		ADAPTER_LOCK(sc);
3549
3550	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
3551	CLR_BUSY(sc);
3552	wakeup(&sc->flags);
3553	ADAPTER_UNLOCK(sc);
3554}
3555
3556static int
3557cxgbe_init_synchronized(struct vi_info *vi)
3558{
3559	struct port_info *pi = vi->pi;
3560	struct adapter *sc = pi->adapter;
3561	struct ifnet *ifp = vi->ifp;
3562	int rc = 0, i;
3563	struct sge_txq *txq;
3564
3565	ASSERT_SYNCHRONIZED_OP(sc);
3566
3567	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
3568		return (0);	/* already running */
3569
3570	if (!(sc->flags & FULL_INIT_DONE) &&
3571	    ((rc = adapter_full_init(sc)) != 0))
3572		return (rc);	/* error message displayed already */
3573
3574	if (!(vi->flags & VI_INIT_DONE) &&
3575	    ((rc = vi_full_init(vi)) != 0))
3576		return (rc); /* error message displayed already */
3577
3578	rc = update_mac_settings(ifp, XGMAC_ALL);
3579	if (rc)
3580		goto done;	/* error message displayed already */
3581
3582	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
3583	if (rc != 0) {
3584		if_printf(ifp, "enable_vi failed: %d\n", rc);
3585		goto done;
3586	}
3587
3588	/*
3589	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
3590	 * if this changes.
3591	 */
3592
3593	for_each_txq(vi, i, txq) {
3594		TXQ_LOCK(txq);
3595		txq->eq.flags |= EQ_ENABLED;
3596		TXQ_UNLOCK(txq);
3597	}
3598
3599	/*
3600	 * The first iq of the first port to come up is used for tracing.
3601	 */
3602	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
3603		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
3604		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
3605		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
3606		    V_QUEUENUMBER(sc->traceq));
3607		pi->flags |= HAS_TRACEQ;
3608	}
3609
3610	/* all ok */
3611	PORT_LOCK(pi);
3612	ifp->if_drv_flags |= IFF_DRV_RUNNING;
3613	pi->up_vis++;
3614
3615	if (pi->nvi > 1)
3616		callout_reset(&vi->tick, hz, vi_tick, vi);
3617	else
3618		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
3619	PORT_UNLOCK(pi);
3620done:
3621	if (rc != 0)
3622		cxgbe_uninit_synchronized(vi);
3623
3624	return (rc);
3625}
3626
3627/*
3628 * Idempotent.
3629 */
3630static int
3631cxgbe_uninit_synchronized(struct vi_info *vi)
3632{
3633	struct port_info *pi = vi->pi;
3634	struct adapter *sc = pi->adapter;
3635	struct ifnet *ifp = vi->ifp;
3636	int rc, i;
3637	struct sge_txq *txq;
3638
3639	ASSERT_SYNCHRONIZED_OP(sc);
3640
3641	if (!(vi->flags & VI_INIT_DONE)) {
3642		KASSERT(!(ifp->if_drv_flags & IFF_DRV_RUNNING),
3643		    ("uninited VI is running"));
3644		return (0);
3645	}
3646
3647	/*
3648	 * Disable the VI so that all its data in either direction is discarded
3649	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
3650	 * tick) intact as the TP can deliver negative advice or data that it's
3651	 * holding in its RAM (for an offloaded connection) even after the VI is
3652	 * disabled.
3653	 */
3654	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
3655	if (rc) {
3656		if_printf(ifp, "disable_vi failed: %d\n", rc);
3657		return (rc);
3658	}
3659
3660	for_each_txq(vi, i, txq) {
3661		TXQ_LOCK(txq);
3662		txq->eq.flags &= ~EQ_ENABLED;
3663		TXQ_UNLOCK(txq);
3664	}
3665
3666	PORT_LOCK(pi);
3667	if (pi->nvi == 1)
3668		callout_stop(&pi->tick);
3669	else
3670		callout_stop(&vi->tick);
3671	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
3672		PORT_UNLOCK(pi);
3673		return (0);
3674	}
3675	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
3676	pi->up_vis--;
3677	if (pi->up_vis > 0) {
3678		PORT_UNLOCK(pi);
3679		return (0);
3680	}
3681	PORT_UNLOCK(pi);
3682
3683	pi->link_cfg.link_ok = 0;
3684	pi->link_cfg.speed = 0;
3685	pi->linkdnrc = -1;
3686	t4_os_link_changed(sc, pi->port_id, 0, -1);
3687
3688	return (0);
3689}
3690
3691/*
3692 * It is ok for this function to fail midway and return right away.  t4_detach
3693 * will walk the entire sc->irq list and clean up whatever is valid.
3694 */
3695static int
3696setup_intr_handlers(struct adapter *sc)
3697{
3698	int rc, rid, p, q, v;
3699	char s[8];
3700	struct irq *irq;
3701	struct port_info *pi;
3702	struct vi_info *vi;
3703	struct sge *sge = &sc->sge;
3704	struct sge_rxq *rxq;
3705#ifdef TCP_OFFLOAD
3706	struct sge_ofld_rxq *ofld_rxq;
3707#endif
3708#ifdef DEV_NETMAP
3709	struct sge_nm_rxq *nm_rxq;
3710#endif
3711
3712	/*
3713	 * Setup interrupts.
3714	 */
3715	irq = &sc->irq[0];
3716	rid = sc->intr_type == INTR_INTX ? 0 : 1;
3717	if (sc->intr_count == 1)
3718		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
3719
3720	/* Multiple interrupts. */
3721	KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
3722	    ("%s: too few intr.", __func__));
3723
3724	/* The first one is always error intr */
3725	rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
3726	if (rc != 0)
3727		return (rc);
3728	irq++;
3729	rid++;
3730
3731	/* The second one is always the firmware event queue */
3732	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
3733	if (rc != 0)
3734		return (rc);
3735	irq++;
3736	rid++;
3737
3738	for_each_port(sc, p) {
3739		pi = sc->port[p];
3740		for_each_vi(pi, v, vi) {
3741			vi->first_intr = rid - 1;
3742
3743			if (vi->nnmrxq > 0) {
3744				int n = max(vi->nrxq, vi->nnmrxq);
3745
3746				MPASS(vi->flags & INTR_RXQ);
3747
3748				rxq = &sge->rxq[vi->first_rxq];
3749#ifdef DEV_NETMAP
3750				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
3751#endif
3752				for (q = 0; q < n; q++) {
3753					snprintf(s, sizeof(s), "%x%c%x", p,
3754					    'a' + v, q);
3755					if (q < vi->nrxq)
3756						irq->rxq = rxq++;
3757#ifdef DEV_NETMAP
3758					if (q < vi->nnmrxq)
3759						irq->nm_rxq = nm_rxq++;
3760#endif
3761					rc = t4_alloc_irq(sc, irq, rid,
3762					    t4_vi_intr, irq, s);
3763					if (rc != 0)
3764						return (rc);
3765					irq++;
3766					rid++;
3767					vi->nintr++;
3768				}
3769			} else if (vi->flags & INTR_RXQ) {
3770				for_each_rxq(vi, q, rxq) {
3771					snprintf(s, sizeof(s), "%x%c%x", p,
3772					    'a' + v, q);
3773					rc = t4_alloc_irq(sc, irq, rid,
3774					    t4_intr, rxq, s);
3775					if (rc != 0)
3776						return (rc);
3777					irq++;
3778					rid++;
3779					vi->nintr++;
3780				}
3781			}
3782#ifdef TCP_OFFLOAD
3783			if (vi->flags & INTR_OFLD_RXQ) {
3784				for_each_ofld_rxq(vi, q, ofld_rxq) {
3785					snprintf(s, sizeof(s), "%x%c%x", p,
3786					    'A' + v, q);
3787					rc = t4_alloc_irq(sc, irq, rid,
3788					    t4_intr, ofld_rxq, s);
3789					if (rc != 0)
3790						return (rc);
3791					irq++;
3792					rid++;
3793					vi->nintr++;
3794				}
3795			}
3796#endif
3797		}
3798	}
3799	MPASS(irq == &sc->irq[sc->intr_count]);
3800
3801	return (0);
3802}
3803
3804int
3805adapter_full_init(struct adapter *sc)
3806{
3807	int rc, i;
3808
3809	ASSERT_SYNCHRONIZED_OP(sc);
3810	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
3811	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
3812	    ("%s: FULL_INIT_DONE already", __func__));
3813
3814	/*
3815	 * queues that belong to the adapter (not any particular port).
3816	 */
3817	rc = t4_setup_adapter_queues(sc);
3818	if (rc != 0)
3819		goto done;
3820
3821	for (i = 0; i < nitems(sc->tq); i++) {
3822		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
3823		    taskqueue_thread_enqueue, &sc->tq[i]);
3824		if (sc->tq[i] == NULL) {
3825			device_printf(sc->dev,
3826			    "failed to allocate task queue %d\n", i);
3827			rc = ENOMEM;
3828			goto done;
3829		}
3830		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
3831		    device_get_nameunit(sc->dev), i);
3832	}
3833
3834	t4_intr_enable(sc);
3835	sc->flags |= FULL_INIT_DONE;
3836done:
3837	if (rc != 0)
3838		adapter_full_uninit(sc);
3839
3840	return (rc);
3841}
3842
3843int
3844adapter_full_uninit(struct adapter *sc)
3845{
3846	int i;
3847
3848	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
3849
3850	t4_teardown_adapter_queues(sc);
3851
3852	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
3853		taskqueue_free(sc->tq[i]);
3854		sc->tq[i] = NULL;
3855	}
3856
3857	sc->flags &= ~FULL_INIT_DONE;
3858
3859	return (0);
3860}
3861
3862#ifdef RSS
3863#define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
3864    RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
3865    RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
3866    RSS_HASHTYPE_RSS_UDP_IPV6)
3867
3868/* Translates kernel hash types to hardware. */
3869static int
3870hashconfig_to_hashen(int hashconfig)
3871{
3872	int hashen = 0;
3873
3874	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
3875		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
3876	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
3877		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
3878	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
3879		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
3880		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
3881	}
3882	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
3883		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
3884		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
3885	}
3886	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
3887		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
3888	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
3889		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
3890
3891	return (hashen);
3892}
3893
3894/* Translates hardware hash types to kernel. */
3895static int
3896hashen_to_hashconfig(int hashen)
3897{
3898	int hashconfig = 0;
3899
3900	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
3901		/*
3902		 * If UDP hashing was enabled it must have been enabled for
3903		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
3904		 * enabling any 4-tuple hash is nonsense configuration.
3905		 */
3906		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
3907		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
3908
3909		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
3910			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
3911		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
3912			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
3913	}
3914	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
3915		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
3916	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
3917		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
3918	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
3919		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
3920	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
3921		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
3922
3923	return (hashconfig);
3924}
3925#endif
3926
3927int
3928vi_full_init(struct vi_info *vi)
3929{
3930	struct adapter *sc = vi->pi->adapter;
3931	struct ifnet *ifp = vi->ifp;
3932	uint16_t *rss;
3933	struct sge_rxq *rxq;
3934	int rc, i, j, hashen;
3935#ifdef RSS
3936	int nbuckets = rss_getnumbuckets();
3937	int hashconfig = rss_gethashconfig();
3938	int extra;
3939	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
3940	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
3941#endif
3942
3943	ASSERT_SYNCHRONIZED_OP(sc);
3944	KASSERT((vi->flags & VI_INIT_DONE) == 0,
3945	    ("%s: VI_INIT_DONE already", __func__));
3946
3947	sysctl_ctx_init(&vi->ctx);
3948	vi->flags |= VI_SYSCTL_CTX;
3949
3950	/*
3951	 * Allocate tx/rx/fl queues for this VI.
3952	 */
3953	rc = t4_setup_vi_queues(vi);
3954	if (rc != 0)
3955		goto done;	/* error message displayed already */
3956
3957	/*
3958	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
3959	 */
3960	if (vi->nrxq > vi->rss_size) {
3961		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
3962		    "some queues will never receive traffic.\n", vi->nrxq,
3963		    vi->rss_size);
3964	} else if (vi->rss_size % vi->nrxq) {
3965		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
3966		    "expect uneven traffic distribution.\n", vi->nrxq,
3967		    vi->rss_size);
3968	}
3969#ifdef RSS
3970	MPASS(RSS_KEYSIZE == 40);
3971	if (vi->nrxq != nbuckets) {
3972		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
3973		    "performance will be impacted.\n", vi->nrxq, nbuckets);
3974	}
3975
3976	rss_getkey((void *)&raw_rss_key[0]);
3977	for (i = 0; i < nitems(rss_key); i++) {
3978		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
3979	}
3980	t4_write_rss_key(sc, &rss_key[0], -1);
3981#endif
3982	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
3983	for (i = 0; i < vi->rss_size;) {
3984#ifdef RSS
3985		j = rss_get_indirection_to_bucket(i);
3986		j %= vi->nrxq;
3987		rxq = &sc->sge.rxq[vi->first_rxq + j];
3988		rss[i++] = rxq->iq.abs_id;
3989#else
3990		for_each_rxq(vi, j, rxq) {
3991			rss[i++] = rxq->iq.abs_id;
3992			if (i == vi->rss_size)
3993				break;
3994		}
3995#endif
3996	}
3997
3998	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
3999	    vi->rss_size);
4000	if (rc != 0) {
4001		if_printf(ifp, "rss_config failed: %d\n", rc);
4002		goto done;
4003	}
4004
4005#ifdef RSS
4006	hashen = hashconfig_to_hashen(hashconfig);
4007
4008	/*
4009	 * We may have had to enable some hashes even though the global config
4010	 * wants them disabled.  This is a potential problem that must be
4011	 * reported to the user.
4012	 */
4013	extra = hashen_to_hashconfig(hashen) ^ hashconfig;
4014
4015	/*
4016	 * If we consider only the supported hash types, then the enabled hashes
4017	 * are a superset of the requested hashes.  In other words, there cannot
4018	 * be any supported hash that was requested but not enabled, but there
4019	 * can be hashes that were not requested but had to be enabled.
4020	 */
4021	extra &= SUPPORTED_RSS_HASHTYPES;
4022	MPASS((extra & hashconfig) == 0);
4023
4024	if (extra) {
4025		if_printf(ifp,
4026		    "global RSS config (0x%x) cannot be accomodated.\n",
4027		    hashconfig);
4028	}
4029	if (extra & RSS_HASHTYPE_RSS_IPV4)
4030		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
4031	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
4032		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
4033	if (extra & RSS_HASHTYPE_RSS_IPV6)
4034		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
4035	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
4036		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
4037	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
4038		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
4039	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
4040		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
4041#else
4042	hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
4043	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
4044	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
4045	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
4046#endif
4047	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, hashen, rss[0]);
4048	if (rc != 0) {
4049		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
4050		goto done;
4051	}
4052
4053	vi->rss = rss;
4054	vi->flags |= VI_INIT_DONE;
4055done:
4056	if (rc != 0)
4057		vi_full_uninit(vi);
4058
4059	return (rc);
4060}
4061
4062/*
4063 * Idempotent.
4064 */
4065int
4066vi_full_uninit(struct vi_info *vi)
4067{
4068	struct port_info *pi = vi->pi;
4069	struct adapter *sc = pi->adapter;
4070	int i;
4071	struct sge_rxq *rxq;
4072	struct sge_txq *txq;
4073#ifdef TCP_OFFLOAD
4074	struct sge_ofld_rxq *ofld_rxq;
4075	struct sge_wrq *ofld_txq;
4076#endif
4077
4078	if (vi->flags & VI_INIT_DONE) {
4079
4080		/* Need to quiesce queues.  */
4081
4082		/* XXX: Only for the first VI? */
4083		if (IS_MAIN_VI(vi))
4084			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
4085
4086		for_each_txq(vi, i, txq) {
4087			quiesce_txq(sc, txq);
4088		}
4089
4090#ifdef TCP_OFFLOAD
4091		for_each_ofld_txq(vi, i, ofld_txq) {
4092			quiesce_wrq(sc, ofld_txq);
4093		}
4094#endif
4095
4096		for_each_rxq(vi, i, rxq) {
4097			quiesce_iq(sc, &rxq->iq);
4098			quiesce_fl(sc, &rxq->fl);
4099		}
4100
4101#ifdef TCP_OFFLOAD
4102		for_each_ofld_rxq(vi, i, ofld_rxq) {
4103			quiesce_iq(sc, &ofld_rxq->iq);
4104			quiesce_fl(sc, &ofld_rxq->fl);
4105		}
4106#endif
4107		free(vi->rss, M_CXGBE);
4108		free(vi->nm_rss, M_CXGBE);
4109	}
4110
4111	t4_teardown_vi_queues(vi);
4112	vi->flags &= ~VI_INIT_DONE;
4113
4114	return (0);
4115}
4116
4117static void
4118quiesce_txq(struct adapter *sc, struct sge_txq *txq)
4119{
4120	struct sge_eq *eq = &txq->eq;
4121	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
4122
4123	(void) sc;	/* unused */
4124
4125#ifdef INVARIANTS
4126	TXQ_LOCK(txq);
4127	MPASS((eq->flags & EQ_ENABLED) == 0);
4128	TXQ_UNLOCK(txq);
4129#endif
4130
4131	/* Wait for the mp_ring to empty. */
4132	while (!mp_ring_is_idle(txq->r)) {
4133		mp_ring_check_drainage(txq->r, 0);
4134		pause("rquiesce", 1);
4135	}
4136
4137	/* Then wait for the hardware to finish. */
4138	while (spg->cidx != htobe16(eq->pidx))
4139		pause("equiesce", 1);
4140
4141	/* Finally, wait for the driver to reclaim all descriptors. */
4142	while (eq->cidx != eq->pidx)
4143		pause("dquiesce", 1);
4144}
4145
4146static void
4147quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
4148{
4149
4150	/* XXXTX */
4151}
4152
4153static void
4154quiesce_iq(struct adapter *sc, struct sge_iq *iq)
4155{
4156	(void) sc;	/* unused */
4157
4158	/* Synchronize with the interrupt handler */
4159	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
4160		pause("iqfree", 1);
4161}
4162
4163static void
4164quiesce_fl(struct adapter *sc, struct sge_fl *fl)
4165{
4166	mtx_lock(&sc->sfl_lock);
4167	FL_LOCK(fl);
4168	fl->flags |= FL_DOOMED;
4169	FL_UNLOCK(fl);
4170	callout_stop(&sc->sfl_callout);
4171	mtx_unlock(&sc->sfl_lock);
4172
4173	KASSERT((fl->flags & FL_STARVING) == 0,
4174	    ("%s: still starving", __func__));
4175}
4176
4177static int
4178t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
4179    driver_intr_t *handler, void *arg, char *name)
4180{
4181	int rc;
4182
4183	irq->rid = rid;
4184	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
4185	    RF_SHAREABLE | RF_ACTIVE);
4186	if (irq->res == NULL) {
4187		device_printf(sc->dev,
4188		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
4189		return (ENOMEM);
4190	}
4191
4192	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
4193	    NULL, handler, arg, &irq->tag);
4194	if (rc != 0) {
4195		device_printf(sc->dev,
4196		    "failed to setup interrupt for rid %d, name %s: %d\n",
4197		    rid, name, rc);
4198	} else if (name)
4199		bus_describe_intr(sc->dev, irq->res, irq->tag, name);
4200
4201	return (rc);
4202}
4203
4204static int
4205t4_free_irq(struct adapter *sc, struct irq *irq)
4206{
4207	if (irq->tag)
4208		bus_teardown_intr(sc->dev, irq->res, irq->tag);
4209	if (irq->res)
4210		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
4211
4212	bzero(irq, sizeof(*irq));
4213
4214	return (0);
4215}
4216
4217static void
4218get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
4219{
4220
4221	regs->version = chip_id(sc) | chip_rev(sc) << 10;
4222	t4_get_regs(sc, buf, regs->len);
4223}
4224
4225#define	A_PL_INDIR_CMD	0x1f8
4226
4227#define	S_PL_AUTOINC	31
4228#define	M_PL_AUTOINC	0x1U
4229#define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
4230#define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
4231
4232#define	S_PL_VFID	20
4233#define	M_PL_VFID	0xffU
4234#define	V_PL_VFID(x)	((x) << S_PL_VFID)
4235#define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
4236
4237#define	S_PL_ADDR	0
4238#define	M_PL_ADDR	0xfffffU
4239#define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
4240#define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
4241
4242#define	A_PL_INDIR_DATA	0x1fc
4243
4244static uint64_t
4245read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
4246{
4247	u32 stats[2];
4248
4249	mtx_assert(&sc->reg_lock, MA_OWNED);
4250	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4251	    V_PL_VFID(G_FW_VIID_VIN(viid)) | V_PL_ADDR(VF_MPS_REG(reg)));
4252	stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
4253	stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
4254	return (((uint64_t)stats[1]) << 32 | stats[0]);
4255}
4256
4257static void
4258t4_get_vi_stats(struct adapter *sc, unsigned int viid,
4259    struct fw_vi_stats_vf *stats)
4260{
4261
4262#define GET_STAT(name) \
4263	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
4264
4265	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
4266	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
4267	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
4268	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
4269	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
4270	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
4271	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
4272	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
4273	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
4274	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
4275	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
4276	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
4277	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
4278	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
4279	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
4280	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
4281
4282#undef GET_STAT
4283}
4284
4285static void
4286t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
4287{
4288	int reg;
4289
4290	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
4291	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
4292	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
4293	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
4294	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
4295		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
4296}
4297
4298static void
4299vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
4300{
4301	struct ifnet *ifp = vi->ifp;
4302	struct sge_txq *txq;
4303	int i, drops;
4304	struct fw_vi_stats_vf *s = &vi->stats;
4305	struct timeval tv;
4306	const struct timeval interval = {0, 250000};	/* 250ms */
4307
4308	if (!(vi->flags & VI_INIT_DONE))
4309		return;
4310
4311	getmicrotime(&tv);
4312	timevalsub(&tv, &interval);
4313	if (timevalcmp(&tv, &vi->last_refreshed, <))
4314		return;
4315
4316	mtx_lock(&sc->reg_lock);
4317	t4_get_vi_stats(sc, vi->viid, &vi->stats);
4318
4319	ifp->if_ipackets = s->rx_bcast_frames + s->rx_mcast_frames +
4320	    s->rx_ucast_frames;
4321	ifp->if_ierrors = s->rx_err_frames;
4322	ifp->if_opackets = s->tx_bcast_frames + s->tx_mcast_frames +
4323	    s->tx_ucast_frames + s->tx_offload_frames;
4324	ifp->if_oerrors = s->tx_drop_frames;
4325	ifp->if_ibytes = s->rx_bcast_bytes + s->rx_mcast_bytes +
4326	    s->rx_ucast_bytes;
4327	ifp->if_obytes = s->tx_bcast_bytes + s->tx_mcast_bytes +
4328	    s->tx_ucast_bytes + s->tx_offload_bytes;
4329	ifp->if_imcasts = s->rx_mcast_frames;
4330	ifp->if_omcasts = s->tx_mcast_frames;
4331
4332	drops = 0;
4333	for_each_txq(vi, i, txq)
4334		drops += counter_u64_fetch(txq->r->drops);
4335	ifp->if_snd.ifq_drops = drops;
4336
4337	getmicrotime(&vi->last_refreshed);
4338	mtx_unlock(&sc->reg_lock);
4339}
4340
4341static void
4342cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
4343{
4344	struct vi_info *vi = &pi->vi[0];
4345	struct ifnet *ifp = vi->ifp;
4346	struct sge_txq *txq;
4347	int i, drops;
4348	struct port_stats *s = &pi->stats;
4349	struct timeval tv;
4350	const struct timeval interval = {0, 250000};	/* 250ms */
4351
4352	getmicrotime(&tv);
4353	timevalsub(&tv, &interval);
4354	if (timevalcmp(&tv, &pi->last_refreshed, <))
4355		return;
4356
4357	t4_get_port_stats(sc, pi->tx_chan, s);
4358
4359	ifp->if_opackets = s->tx_frames;
4360	ifp->if_ipackets = s->rx_frames;
4361	ifp->if_obytes = s->tx_octets;
4362	ifp->if_ibytes = s->rx_octets;
4363	ifp->if_omcasts = s->tx_mcast_frames;
4364	ifp->if_imcasts = s->rx_mcast_frames;
4365	ifp->if_iqdrops = s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
4366	    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
4367	    s->rx_trunc3;
4368	for (i = 0; i < sc->chip_params->nchan; i++) {
4369		if (pi->rx_chan_map & (1 << i)) {
4370			uint32_t v;
4371
4372			mtx_lock(&sc->reg_lock);
4373			t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
4374			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
4375			mtx_unlock(&sc->reg_lock);
4376			ifp->if_iqdrops += v;
4377		}
4378	}
4379
4380	drops = s->tx_drop;
4381	for_each_txq(vi, i, txq)
4382		drops += counter_u64_fetch(txq->r->drops);
4383	ifp->if_snd.ifq_drops = drops;
4384
4385	ifp->if_oerrors = s->tx_error_frames;
4386	ifp->if_ierrors = s->rx_jabber + s->rx_runt + s->rx_too_long +
4387	    s->rx_fcs_err + s->rx_len_err;
4388
4389	getmicrotime(&pi->last_refreshed);
4390}
4391
4392static void
4393cxgbe_tick(void *arg)
4394{
4395	struct port_info *pi = arg;
4396	struct adapter *sc = pi->adapter;
4397
4398	PORT_LOCK_ASSERT_OWNED(pi);
4399	cxgbe_refresh_stats(sc, pi);
4400
4401	callout_schedule(&pi->tick, hz);
4402}
4403
4404void
4405vi_tick(void *arg)
4406{
4407	struct vi_info *vi = arg;
4408	struct adapter *sc = vi->pi->adapter;
4409
4410	vi_refresh_stats(sc, vi);
4411
4412	callout_schedule(&vi->tick, hz);
4413}
4414
4415static void
4416cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
4417{
4418	struct ifnet *vlan;
4419
4420	if (arg != ifp || ifp->if_type != IFT_ETHER)
4421		return;
4422
4423	vlan = VLAN_DEVAT(ifp, vid);
4424	VLAN_SETCOOKIE(vlan, ifp);
4425}
4426
4427static int
4428cpl_not_handled(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
4429{
4430
4431#ifdef INVARIANTS
4432	panic("%s: opcode 0x%02x on iq %p with payload %p",
4433	    __func__, rss->opcode, iq, m);
4434#else
4435	log(LOG_ERR, "%s: opcode 0x%02x on iq %p with payload %p\n",
4436	    __func__, rss->opcode, iq, m);
4437	m_freem(m);
4438#endif
4439	return (EDOOFUS);
4440}
4441
4442int
4443t4_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h)
4444{
4445	uintptr_t *loc, new;
4446
4447	if (opcode >= nitems(sc->cpl_handler))
4448		return (EINVAL);
4449
4450	new = h ? (uintptr_t)h : (uintptr_t)cpl_not_handled;
4451	loc = (uintptr_t *) &sc->cpl_handler[opcode];
4452	atomic_store_rel_ptr(loc, new);
4453
4454	return (0);
4455}
4456
4457static int
4458an_not_handled(struct sge_iq *iq, const struct rsp_ctrl *ctrl)
4459{
4460
4461#ifdef INVARIANTS
4462	panic("%s: async notification on iq %p (ctrl %p)", __func__, iq, ctrl);
4463#else
4464	log(LOG_ERR, "%s: async notification on iq %p (ctrl %p)\n",
4465	    __func__, iq, ctrl);
4466#endif
4467	return (EDOOFUS);
4468}
4469
4470int
4471t4_register_an_handler(struct adapter *sc, an_handler_t h)
4472{
4473	uintptr_t *loc, new;
4474
4475	new = h ? (uintptr_t)h : (uintptr_t)an_not_handled;
4476	loc = (uintptr_t *) &sc->an_handler;
4477	atomic_store_rel_ptr(loc, new);
4478
4479	return (0);
4480}
4481
4482static int
4483fw_msg_not_handled(struct adapter *sc, const __be64 *rpl)
4484{
4485	const struct cpl_fw6_msg *cpl =
4486	    __containerof(rpl, struct cpl_fw6_msg, data[0]);
4487
4488#ifdef INVARIANTS
4489	panic("%s: fw_msg type %d", __func__, cpl->type);
4490#else
4491	log(LOG_ERR, "%s: fw_msg type %d\n", __func__, cpl->type);
4492#endif
4493	return (EDOOFUS);
4494}
4495
4496int
4497t4_register_fw_msg_handler(struct adapter *sc, int type, fw_msg_handler_t h)
4498{
4499	uintptr_t *loc, new;
4500
4501	if (type >= nitems(sc->fw_msg_handler))
4502		return (EINVAL);
4503
4504	/*
4505	 * These are dispatched by the handler for FW{4|6}_CPL_MSG using the CPL
4506	 * handler dispatch table.  Reject any attempt to install a handler for
4507	 * this subtype.
4508	 */
4509	if (type == FW_TYPE_RSSCPL || type == FW6_TYPE_RSSCPL)
4510		return (EINVAL);
4511
4512	new = h ? (uintptr_t)h : (uintptr_t)fw_msg_not_handled;
4513	loc = (uintptr_t *) &sc->fw_msg_handler[type];
4514	atomic_store_rel_ptr(loc, new);
4515
4516	return (0);
4517}
4518
4519/*
4520 * Should match fw_caps_config_<foo> enums in t4fw_interface.h
4521 */
4522static char *caps_decoder[] = {
4523	"\20\001IPMI\002NCSI",				/* 0: NBM */
4524	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
4525	"\20\001INGRESS\002EGRESS",			/* 2: switch */
4526	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
4527	    "\006HASHFILTER\007ETHOFLD",
4528	"\20\001TOE",					/* 4: TOE */
4529	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
4530	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
4531	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
4532	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
4533	    "\007T10DIF"
4534	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
4535	"\20\00KEYS",					/* 7: TLS */
4536	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
4537		    "\004PO_INITIATOR\005PO_TARGET",
4538};
4539
4540static void
4541t4_sysctls(struct adapter *sc)
4542{
4543	struct sysctl_ctx_list *ctx;
4544	struct sysctl_oid *oid;
4545	struct sysctl_oid_list *children, *c0;
4546	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
4547
4548	ctx = device_get_sysctl_ctx(sc->dev);
4549
4550	/*
4551	 * dev.t4nex.X.
4552	 */
4553	oid = device_get_sysctl_tree(sc->dev);
4554	c0 = children = SYSCTL_CHILDREN(oid);
4555
4556	sc->sc_do_rxcopy = 1;
4557	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
4558	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
4559
4560	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
4561	    sc->params.nports, "# of ports");
4562
4563	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
4564	    NULL, chip_rev(sc), "chip hardware revision");
4565
4566	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
4567	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
4568
4569	if (sc->params.exprom_vers != 0) {
4570		SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "exprom_version",
4571		    CTLFLAG_RD, sc->exprom_version, 0, "expansion ROM version");
4572	}
4573
4574	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
4575	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
4576
4577	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
4578	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
4579
4580	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
4581	    sc->cfcsum, "config file checksum");
4582
4583	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
4584	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
4585	    sysctl_bitfield, "A", "available doorbells");
4586
4587#define SYSCTL_CAP(name, n, text) \
4588	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
4589	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], sc->name, \
4590	    sysctl_bitfield, "A", "available " text "capabilities")
4591
4592	SYSCTL_CAP(nbmcaps, 0, "NBM");
4593	SYSCTL_CAP(linkcaps, 1, "link");
4594	SYSCTL_CAP(switchcaps, 2, "switch");
4595	SYSCTL_CAP(niccaps, 3, "NIC");
4596	SYSCTL_CAP(toecaps, 4, "TCP offload");
4597	SYSCTL_CAP(rdmacaps, 5, "RDMA");
4598	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
4599	SYSCTL_CAP(tlscaps, 7, "TLS");
4600	SYSCTL_CAP(fcoecaps, 8, "FCoE");
4601#undef SYSCTL_CAP
4602
4603	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
4604	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
4605
4606	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
4607	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
4608	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
4609	    "interrupt holdoff timer values (us)");
4610
4611	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
4612	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
4613	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
4614	    "interrupt holdoff packet counter values");
4615
4616	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
4617	    NULL, sc->tids.nftids, "number of filters");
4618
4619	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
4620	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
4621	    "chip temperature (in Celsius)");
4622
4623	t4_sge_sysctls(sc, ctx, children);
4624
4625	sc->lro_timeout = 100;
4626	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
4627	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
4628
4629	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "debug_flags", CTLFLAG_RW,
4630	    &sc->debug_flags, 0, "flags to enable runtime debugging");
4631
4632#ifdef SBUF_DRAIN
4633	/*
4634	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
4635	 */
4636	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
4637	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
4638	    "logs and miscellaneous information");
4639	children = SYSCTL_CHILDREN(oid);
4640
4641	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
4642	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4643	    sysctl_cctrl, "A", "congestion control");
4644
4645	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
4646	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4647	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
4648
4649	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
4650	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
4651	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
4652
4653	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
4654	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
4655	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
4656
4657	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
4658	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
4659	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
4660
4661	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
4662	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
4663	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
4664
4665	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
4666	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
4667	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
4668
4669	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
4670	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4671	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
4672	    "A", "CIM logic analyzer");
4673
4674	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
4675	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4676	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
4677
4678	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
4679	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
4680	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
4681
4682	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
4683	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
4684	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
4685
4686	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
4687	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
4688	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
4689
4690	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
4691	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
4692	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
4693
4694	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
4695	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
4696	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
4697
4698	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
4699	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
4700	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
4701
4702	if (chip_id(sc) > CHELSIO_T4) {
4703		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
4704		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
4705		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
4706
4707		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
4708		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
4709		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
4710	}
4711
4712	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
4713	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4714	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
4715
4716	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
4717	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4718	    sysctl_cim_qcfg, "A", "CIM queue configuration");
4719
4720	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
4721	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4722	    sysctl_cpl_stats, "A", "CPL statistics");
4723
4724	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
4725	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4726	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
4727
4728	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
4729	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4730	    sysctl_devlog, "A", "firmware's device log");
4731
4732	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
4733	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4734	    sysctl_fcoe_stats, "A", "FCoE statistics");
4735
4736	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
4737	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4738	    sysctl_hw_sched, "A", "hardware scheduler ");
4739
4740	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
4741	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4742	    sysctl_l2t, "A", "hardware L2 table");
4743
4744	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
4745	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4746	    sysctl_lb_stats, "A", "loopback statistics");
4747
4748	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
4749	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4750	    sysctl_meminfo, "A", "memory regions");
4751
4752	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
4753	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4754	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
4755	    "A", "MPS TCAM entries");
4756
4757	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
4758	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4759	    sysctl_path_mtus, "A", "path MTUs");
4760
4761	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
4762	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4763	    sysctl_pm_stats, "A", "PM statistics");
4764
4765	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
4766	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4767	    sysctl_rdma_stats, "A", "RDMA statistics");
4768
4769	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
4770	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4771	    sysctl_tcp_stats, "A", "TCP statistics");
4772
4773	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
4774	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4775	    sysctl_tids, "A", "TID information");
4776
4777	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
4778	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4779	    sysctl_tp_err_stats, "A", "TP error statistics");
4780
4781	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
4782	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
4783	    "TP logic analyzer event capture mask");
4784
4785	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
4786	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4787	    sysctl_tp_la, "A", "TP logic analyzer");
4788
4789	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
4790	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4791	    sysctl_tx_rate, "A", "Tx rate");
4792
4793	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
4794	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4795	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
4796
4797	if (is_t5(sc)) {
4798		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
4799		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
4800		    sysctl_wcwr_stats, "A", "write combined work requests");
4801	}
4802#endif
4803
4804#ifdef TCP_OFFLOAD
4805	if (is_offload(sc)) {
4806		/*
4807		 * dev.t4nex.X.toe.
4808		 */
4809		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
4810		    NULL, "TOE parameters");
4811		children = SYSCTL_CHILDREN(oid);
4812
4813		sc->tt.sndbuf = 256 * 1024;
4814		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
4815		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
4816
4817		sc->tt.ddp = 0;
4818		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
4819		    &sc->tt.ddp, 0, "DDP allowed");
4820
4821		sc->tt.indsz = G_INDICATESIZE(t4_read_reg(sc, A_TP_PARA_REG5));
4822		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "indsz", CTLFLAG_RW,
4823		    &sc->tt.indsz, 0, "DDP max indicate size allowed");
4824
4825		sc->tt.ddp_thres =
4826		    G_RXCOALESCESIZE(t4_read_reg(sc, A_TP_PARA_REG2));
4827		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp_thres", CTLFLAG_RW,
4828		    &sc->tt.ddp_thres, 0, "DDP threshold");
4829
4830		sc->tt.rx_coalesce = 1;
4831		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
4832		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
4833
4834		sc->tt.tx_align = 1;
4835		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
4836		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
4837
4838		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
4839		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
4840		    "TP timer tick (us)");
4841
4842		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
4843		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
4844		    "TCP timestamp tick (us)");
4845
4846		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
4847		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
4848		    "DACK tick (us)");
4849
4850		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
4851		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
4852		    "IU", "DACK timer (us)");
4853
4854		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
4855		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
4856		    sysctl_tp_timer, "LU", "Retransmit min (us)");
4857
4858		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
4859		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
4860		    sysctl_tp_timer, "LU", "Retransmit max (us)");
4861
4862		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
4863		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
4864		    sysctl_tp_timer, "LU", "Persist timer min (us)");
4865
4866		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
4867		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
4868		    sysctl_tp_timer, "LU", "Persist timer max (us)");
4869
4870		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
4871		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
4872		    sysctl_tp_timer, "LU", "Keepidle idle timer (us)");
4873
4874		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_intvl",
4875		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
4876		    sysctl_tp_timer, "LU", "Keepidle interval (us)");
4877
4878		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
4879		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
4880		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
4881
4882		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
4883		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
4884		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
4885	}
4886#endif
4887}
4888
4889void
4890vi_sysctls(struct vi_info *vi)
4891{
4892	struct sysctl_ctx_list *ctx;
4893	struct sysctl_oid *oid;
4894	struct sysctl_oid_list *children;
4895
4896	ctx = device_get_sysctl_ctx(vi->dev);
4897
4898	/*
4899	 * dev.v?(cxgbe|cxl).X.
4900	 */
4901	oid = device_get_sysctl_tree(vi->dev);
4902	children = SYSCTL_CHILDREN(oid);
4903
4904	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
4905	    vi->viid, "VI identifer");
4906	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
4907	    &vi->nrxq, 0, "# of rx queues");
4908	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
4909	    &vi->ntxq, 0, "# of tx queues");
4910	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
4911	    &vi->first_rxq, 0, "index of first rx queue");
4912	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
4913	    &vi->first_txq, 0, "index of first tx queue");
4914
4915	if (IS_MAIN_VI(vi)) {
4916		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
4917		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
4918		    "Reserve queue 0 for non-flowid packets");
4919	}
4920
4921#ifdef TCP_OFFLOAD
4922	if (vi->nofldrxq != 0) {
4923		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
4924		    &vi->nofldrxq, 0,
4925		    "# of rx queues for offloaded TCP connections");
4926		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
4927		    &vi->nofldtxq, 0,
4928		    "# of tx queues for offloaded TCP connections");
4929		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
4930		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
4931		    "index of first TOE rx queue");
4932		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
4933		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
4934		    "index of first TOE tx queue");
4935	}
4936#endif
4937#ifdef DEV_NETMAP
4938	if (vi->nnmrxq != 0) {
4939		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
4940		    &vi->nnmrxq, 0, "# of netmap rx queues");
4941		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
4942		    &vi->nnmtxq, 0, "# of netmap tx queues");
4943		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
4944		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
4945		    "index of first netmap rx queue");
4946		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
4947		    CTLFLAG_RD, &vi->first_nm_txq, 0,
4948		    "index of first netmap tx queue");
4949	}
4950#endif
4951
4952	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
4953	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
4954	    "holdoff timer index");
4955	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
4956	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
4957	    "holdoff packet counter index");
4958
4959	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
4960	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
4961	    "rx queue size");
4962	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
4963	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
4964	    "tx queue size");
4965}
4966
4967static void
4968cxgbe_sysctls(struct port_info *pi)
4969{
4970	struct sysctl_ctx_list *ctx;
4971	struct sysctl_oid *oid;
4972	struct sysctl_oid_list *children, *children2;
4973	struct adapter *sc = pi->adapter;
4974	int i;
4975	char name[16];
4976
4977	ctx = device_get_sysctl_ctx(pi->dev);
4978
4979	/*
4980	 * dev.cxgbe.X.
4981	 */
4982	oid = device_get_sysctl_tree(pi->dev);
4983	children = SYSCTL_CHILDREN(oid);
4984
4985	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
4986	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
4987	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
4988		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
4989		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
4990		    "PHY temperature (in Celsius)");
4991		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
4992		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
4993		    "PHY firmware version");
4994	}
4995
4996	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
4997	    CTLTYPE_STRING | CTLFLAG_RW, pi, PAUSE_TX, sysctl_pause_settings,
4998	    "A", "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
4999
5000	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
5001	    port_top_speed(pi), "max speed (in Gbps)");
5002
5003	/*
5004	 * dev.(cxgbe|cxl).X.tc.
5005	 */
5006	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
5007	    "Tx scheduler traffic classes");
5008	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
5009		struct tx_sched_class *tc = &pi->tc[i];
5010
5011		snprintf(name, sizeof(name), "%d", i);
5012		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
5013		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
5014		    "traffic class"));
5015		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "flags", CTLFLAG_RD,
5016		    &tc->flags, 0, "flags");
5017		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
5018		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
5019#ifdef SBUF_DRAIN
5020		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
5021		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
5022		    sysctl_tc_params, "A", "traffic class parameters");
5023#endif
5024	}
5025
5026	/*
5027	 * dev.cxgbe.X.stats.
5028	 */
5029	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
5030	    NULL, "port statistics");
5031	children = SYSCTL_CHILDREN(oid);
5032	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
5033	    &pi->tx_parse_error, 0,
5034	    "# of tx packets with invalid length or # of segments");
5035
5036#define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
5037	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
5038	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
5039	    sysctl_handle_t4_reg64, "QU", desc)
5040
5041	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
5042	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
5043	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
5044	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
5045	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
5046	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
5047	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
5048	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
5049	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
5050	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
5051	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
5052	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
5053	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
5054	    "# of tx frames in this range",
5055	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
5056	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
5057	    "# of tx frames in this range",
5058	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
5059	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
5060	    "# of tx frames in this range",
5061	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
5062	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
5063	    "# of tx frames in this range",
5064	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
5065	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
5066	    "# of tx frames in this range",
5067	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
5068	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
5069	    "# of tx frames in this range",
5070	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
5071	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
5072	    "# of tx frames in this range",
5073	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
5074	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
5075	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
5076	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
5077	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
5078	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
5079	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
5080	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
5081	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
5082	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
5083	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
5084	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
5085	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
5086	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
5087	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
5088	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
5089	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
5090	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
5091	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
5092	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
5093	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
5094
5095	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
5096	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
5097	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
5098	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
5099	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
5100	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
5101	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
5102	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
5103	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
5104	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
5105	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
5106	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
5107	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
5108	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
5109	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
5110	    "# of frames received with bad FCS",
5111	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
5112	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
5113	    "# of frames received with length error",
5114	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
5115	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
5116	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
5117	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
5118	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
5119	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
5120	    "# of rx frames in this range",
5121	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
5122	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
5123	    "# of rx frames in this range",
5124	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
5125	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
5126	    "# of rx frames in this range",
5127	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
5128	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
5129	    "# of rx frames in this range",
5130	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
5131	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
5132	    "# of rx frames in this range",
5133	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
5134	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
5135	    "# of rx frames in this range",
5136	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
5137	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
5138	    "# of rx frames in this range",
5139	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
5140	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
5141	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
5142	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
5143	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
5144	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
5145	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
5146	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
5147	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
5148	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
5149	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
5150	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
5151	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
5152	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
5153	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
5154	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
5155	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
5156	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
5157	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
5158
5159#undef SYSCTL_ADD_T4_REG64
5160
5161#define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
5162	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
5163	    &pi->stats.name, desc)
5164
5165	/* We get these from port_stats and they may be stale by upto 1s */
5166	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
5167	    "# drops due to buffer-group 0 overflows");
5168	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
5169	    "# drops due to buffer-group 1 overflows");
5170	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
5171	    "# drops due to buffer-group 2 overflows");
5172	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
5173	    "# drops due to buffer-group 3 overflows");
5174	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
5175	    "# of buffer-group 0 truncated packets");
5176	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
5177	    "# of buffer-group 1 truncated packets");
5178	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
5179	    "# of buffer-group 2 truncated packets");
5180	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
5181	    "# of buffer-group 3 truncated packets");
5182
5183#undef SYSCTL_ADD_T4_PORTSTAT
5184}
5185
5186static int
5187sysctl_int_array(SYSCTL_HANDLER_ARGS)
5188{
5189	int rc, *i, space = 0;
5190	struct sbuf sb;
5191
5192	sbuf_new(&sb, NULL, 32, SBUF_AUTOEXTEND);
5193	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
5194		if (space)
5195			sbuf_printf(&sb, " ");
5196		sbuf_printf(&sb, "%d", *i);
5197		space = 1;
5198	}
5199	sbuf_finish(&sb);
5200	rc = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
5201	sbuf_delete(&sb);
5202	return (rc);
5203}
5204
5205static int
5206sysctl_bitfield(SYSCTL_HANDLER_ARGS)
5207{
5208	int rc;
5209	struct sbuf *sb;
5210
5211	rc = sysctl_wire_old_buffer(req, 0);
5212	if (rc != 0)
5213		return(rc);
5214
5215	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5216	if (sb == NULL)
5217		return (ENOMEM);
5218
5219	sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
5220	rc = sbuf_finish(sb);
5221	sbuf_delete(sb);
5222
5223	return (rc);
5224}
5225
5226static int
5227sysctl_btphy(SYSCTL_HANDLER_ARGS)
5228{
5229	struct port_info *pi = arg1;
5230	int op = arg2;
5231	struct adapter *sc = pi->adapter;
5232	u_int v;
5233	int rc;
5234
5235	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
5236	if (rc)
5237		return (rc);
5238	/* XXX: magic numbers */
5239	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
5240	    &v);
5241	end_synchronized_op(sc, 0);
5242	if (rc)
5243		return (rc);
5244	if (op == 0)
5245		v /= 256;
5246
5247	rc = sysctl_handle_int(oidp, &v, 0, req);
5248	return (rc);
5249}
5250
5251static int
5252sysctl_noflowq(SYSCTL_HANDLER_ARGS)
5253{
5254	struct vi_info *vi = arg1;
5255	int rc, val;
5256
5257	val = vi->rsrv_noflowq;
5258	rc = sysctl_handle_int(oidp, &val, 0, req);
5259	if (rc != 0 || req->newptr == NULL)
5260		return (rc);
5261
5262	if ((val >= 1) && (vi->ntxq > 1))
5263		vi->rsrv_noflowq = 1;
5264	else
5265		vi->rsrv_noflowq = 0;
5266
5267	return (rc);
5268}
5269
5270static int
5271sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
5272{
5273	struct vi_info *vi = arg1;
5274	struct adapter *sc = vi->pi->adapter;
5275	int idx, rc, i;
5276	struct sge_rxq *rxq;
5277#ifdef TCP_OFFLOAD
5278	struct sge_ofld_rxq *ofld_rxq;
5279#endif
5280	uint8_t v;
5281
5282	idx = vi->tmr_idx;
5283
5284	rc = sysctl_handle_int(oidp, &idx, 0, req);
5285	if (rc != 0 || req->newptr == NULL)
5286		return (rc);
5287
5288	if (idx < 0 || idx >= SGE_NTIMERS)
5289		return (EINVAL);
5290
5291	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5292	    "t4tmr");
5293	if (rc)
5294		return (rc);
5295
5296	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
5297	for_each_rxq(vi, i, rxq) {
5298#ifdef atomic_store_rel_8
5299		atomic_store_rel_8(&rxq->iq.intr_params, v);
5300#else
5301		rxq->iq.intr_params = v;
5302#endif
5303	}
5304#ifdef TCP_OFFLOAD
5305	for_each_ofld_rxq(vi, i, ofld_rxq) {
5306#ifdef atomic_store_rel_8
5307		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
5308#else
5309		ofld_rxq->iq.intr_params = v;
5310#endif
5311	}
5312#endif
5313	vi->tmr_idx = idx;
5314
5315	end_synchronized_op(sc, LOCK_HELD);
5316	return (0);
5317}
5318
5319static int
5320sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
5321{
5322	struct vi_info *vi = arg1;
5323	struct adapter *sc = vi->pi->adapter;
5324	int idx, rc;
5325
5326	idx = vi->pktc_idx;
5327
5328	rc = sysctl_handle_int(oidp, &idx, 0, req);
5329	if (rc != 0 || req->newptr == NULL)
5330		return (rc);
5331
5332	if (idx < -1 || idx >= SGE_NCOUNTERS)
5333		return (EINVAL);
5334
5335	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5336	    "t4pktc");
5337	if (rc)
5338		return (rc);
5339
5340	if (vi->flags & VI_INIT_DONE)
5341		rc = EBUSY; /* cannot be changed once the queues are created */
5342	else
5343		vi->pktc_idx = idx;
5344
5345	end_synchronized_op(sc, LOCK_HELD);
5346	return (rc);
5347}
5348
5349static int
5350sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
5351{
5352	struct vi_info *vi = arg1;
5353	struct adapter *sc = vi->pi->adapter;
5354	int qsize, rc;
5355
5356	qsize = vi->qsize_rxq;
5357
5358	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5359	if (rc != 0 || req->newptr == NULL)
5360		return (rc);
5361
5362	if (qsize < 128 || (qsize & 7))
5363		return (EINVAL);
5364
5365	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5366	    "t4rxqs");
5367	if (rc)
5368		return (rc);
5369
5370	if (vi->flags & VI_INIT_DONE)
5371		rc = EBUSY; /* cannot be changed once the queues are created */
5372	else
5373		vi->qsize_rxq = qsize;
5374
5375	end_synchronized_op(sc, LOCK_HELD);
5376	return (rc);
5377}
5378
5379static int
5380sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
5381{
5382	struct vi_info *vi = arg1;
5383	struct adapter *sc = vi->pi->adapter;
5384	int qsize, rc;
5385
5386	qsize = vi->qsize_txq;
5387
5388	rc = sysctl_handle_int(oidp, &qsize, 0, req);
5389	if (rc != 0 || req->newptr == NULL)
5390		return (rc);
5391
5392	if (qsize < 128 || qsize > 65536)
5393		return (EINVAL);
5394
5395	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
5396	    "t4txqs");
5397	if (rc)
5398		return (rc);
5399
5400	if (vi->flags & VI_INIT_DONE)
5401		rc = EBUSY; /* cannot be changed once the queues are created */
5402	else
5403		vi->qsize_txq = qsize;
5404
5405	end_synchronized_op(sc, LOCK_HELD);
5406	return (rc);
5407}
5408
5409static int
5410sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
5411{
5412	struct port_info *pi = arg1;
5413	struct adapter *sc = pi->adapter;
5414	struct link_config *lc = &pi->link_cfg;
5415	int rc;
5416
5417	if (req->newptr == NULL) {
5418		struct sbuf *sb;
5419		static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
5420
5421		rc = sysctl_wire_old_buffer(req, 0);
5422		if (rc != 0)
5423			return(rc);
5424
5425		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
5426		if (sb == NULL)
5427			return (ENOMEM);
5428
5429		sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
5430		rc = sbuf_finish(sb);
5431		sbuf_delete(sb);
5432	} else {
5433		char s[2];
5434		int n;
5435
5436		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
5437		s[1] = 0;
5438
5439		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
5440		if (rc != 0)
5441			return(rc);
5442
5443		if (s[1] != 0)
5444			return (EINVAL);
5445		if (s[0] < '0' || s[0] > '9')
5446			return (EINVAL);	/* not a number */
5447		n = s[0] - '0';
5448		if (n & ~(PAUSE_TX | PAUSE_RX))
5449			return (EINVAL);	/* some other bit is set too */
5450
5451		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
5452		    "t4PAUSE");
5453		if (rc)
5454			return (rc);
5455		if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
5456			int link_ok = lc->link_ok;
5457
5458			lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
5459			lc->requested_fc |= n;
5460			rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
5461			lc->link_ok = link_ok;	/* restore */
5462		}
5463		end_synchronized_op(sc, 0);
5464	}
5465
5466	return (rc);
5467}
5468
5469static int
5470sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
5471{
5472	struct adapter *sc = arg1;
5473	int reg = arg2;
5474	uint64_t val;
5475
5476	val = t4_read_reg64(sc, reg);
5477
5478	return (sysctl_handle_64(oidp, &val, 0, req));
5479}
5480
5481static int
5482sysctl_temperature(SYSCTL_HANDLER_ARGS)
5483{
5484	struct adapter *sc = arg1;
5485	int rc, t;
5486	uint32_t param, val;
5487
5488	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
5489	if (rc)
5490		return (rc);
5491	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5492	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
5493	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
5494	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
5495	end_synchronized_op(sc, 0);
5496	if (rc)
5497		return (rc);
5498
5499	/* unknown is returned as 0 but we display -1 in that case */
5500	t = val == 0 ? -1 : val;
5501
5502	rc = sysctl_handle_int(oidp, &t, 0, req);
5503	return (rc);
5504}
5505
5506#ifdef SBUF_DRAIN
5507static int
5508sysctl_cctrl(SYSCTL_HANDLER_ARGS)
5509{
5510	struct adapter *sc = arg1;
5511	struct sbuf *sb;
5512	int rc, i;
5513	uint16_t incr[NMTUS][NCCTRL_WIN];
5514	static const char *dec_fac[] = {
5515		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
5516		"0.9375"
5517	};
5518
5519	rc = sysctl_wire_old_buffer(req, 0);
5520	if (rc != 0)
5521		return (rc);
5522
5523	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5524	if (sb == NULL)
5525		return (ENOMEM);
5526
5527	t4_read_cong_tbl(sc, incr);
5528
5529	for (i = 0; i < NCCTRL_WIN; ++i) {
5530		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
5531		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
5532		    incr[5][i], incr[6][i], incr[7][i]);
5533		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
5534		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
5535		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
5536		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
5537	}
5538
5539	rc = sbuf_finish(sb);
5540	sbuf_delete(sb);
5541
5542	return (rc);
5543}
5544
5545static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
5546	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
5547	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
5548	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
5549};
5550
5551static int
5552sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
5553{
5554	struct adapter *sc = arg1;
5555	struct sbuf *sb;
5556	int rc, i, n, qid = arg2;
5557	uint32_t *buf, *p;
5558	char *qtype;
5559	u_int cim_num_obq = sc->chip_params->cim_num_obq;
5560
5561	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
5562	    ("%s: bad qid %d\n", __func__, qid));
5563
5564	if (qid < CIM_NUM_IBQ) {
5565		/* inbound queue */
5566		qtype = "IBQ";
5567		n = 4 * CIM_IBQ_SIZE;
5568		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5569		rc = t4_read_cim_ibq(sc, qid, buf, n);
5570	} else {
5571		/* outbound queue */
5572		qtype = "OBQ";
5573		qid -= CIM_NUM_IBQ;
5574		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
5575		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
5576		rc = t4_read_cim_obq(sc, qid, buf, n);
5577	}
5578
5579	if (rc < 0) {
5580		rc = -rc;
5581		goto done;
5582	}
5583	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
5584
5585	rc = sysctl_wire_old_buffer(req, 0);
5586	if (rc != 0)
5587		goto done;
5588
5589	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
5590	if (sb == NULL) {
5591		rc = ENOMEM;
5592		goto done;
5593	}
5594
5595	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
5596	for (i = 0, p = buf; i < n; i += 16, p += 4)
5597		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
5598		    p[2], p[3]);
5599
5600	rc = sbuf_finish(sb);
5601	sbuf_delete(sb);
5602done:
5603	free(buf, M_CXGBE);
5604	return (rc);
5605}
5606
5607static int
5608sysctl_cim_la(SYSCTL_HANDLER_ARGS)
5609{
5610	struct adapter *sc = arg1;
5611	u_int cfg;
5612	struct sbuf *sb;
5613	uint32_t *buf, *p;
5614	int rc;
5615
5616	MPASS(chip_id(sc) <= CHELSIO_T5);
5617
5618	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
5619	if (rc != 0)
5620		return (rc);
5621
5622	rc = sysctl_wire_old_buffer(req, 0);
5623	if (rc != 0)
5624		return (rc);
5625
5626	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5627	if (sb == NULL)
5628		return (ENOMEM);
5629
5630	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
5631	    M_ZERO | M_WAITOK);
5632
5633	rc = -t4_cim_read_la(sc, buf, NULL);
5634	if (rc != 0)
5635		goto done;
5636
5637	sbuf_printf(sb, "Status   Data      PC%s",
5638	    cfg & F_UPDBGLACAPTPCONLY ? "" :
5639	    "     LS0Stat  LS0Addr             LS0Data");
5640
5641	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
5642		if (cfg & F_UPDBGLACAPTPCONLY) {
5643			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
5644			    p[6], p[7]);
5645			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
5646			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
5647			    p[4] & 0xff, p[5] >> 8);
5648			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
5649			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
5650			    p[1] & 0xf, p[2] >> 4);
5651		} else {
5652			sbuf_printf(sb,
5653			    "\n  %02x   %x%07x %x%07x %08x %08x "
5654			    "%08x%08x%08x%08x",
5655			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
5656			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
5657			    p[6], p[7]);
5658		}
5659	}
5660
5661	rc = sbuf_finish(sb);
5662	sbuf_delete(sb);
5663done:
5664	free(buf, M_CXGBE);
5665	return (rc);
5666}
5667
5668static int
5669sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
5670{
5671	struct adapter *sc = arg1;
5672	u_int cfg;
5673	struct sbuf *sb;
5674	uint32_t *buf, *p;
5675	int rc;
5676
5677	MPASS(chip_id(sc) > CHELSIO_T5);
5678
5679	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
5680	if (rc != 0)
5681		return (rc);
5682
5683	rc = sysctl_wire_old_buffer(req, 0);
5684	if (rc != 0)
5685		return (rc);
5686
5687	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5688	if (sb == NULL)
5689		return (ENOMEM);
5690
5691	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
5692	    M_ZERO | M_WAITOK);
5693
5694	rc = -t4_cim_read_la(sc, buf, NULL);
5695	if (rc != 0)
5696		goto done;
5697
5698	sbuf_printf(sb, "Status   Inst    Data      PC%s",
5699	    cfg & F_UPDBGLACAPTPCONLY ? "" :
5700	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
5701
5702	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
5703		if (cfg & F_UPDBGLACAPTPCONLY) {
5704			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
5705			    p[3] & 0xff, p[2], p[1], p[0]);
5706			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
5707			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
5708			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
5709			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
5710			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
5711			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
5712			    p[6] >> 16);
5713		} else {
5714			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
5715			    "%08x %08x %08x %08x %08x %08x",
5716			    (p[9] >> 16) & 0xff,
5717			    p[9] & 0xffff, p[8] >> 16,
5718			    p[8] & 0xffff, p[7] >> 16,
5719			    p[7] & 0xffff, p[6] >> 16,
5720			    p[2], p[1], p[0], p[5], p[4], p[3]);
5721		}
5722	}
5723
5724	rc = sbuf_finish(sb);
5725	sbuf_delete(sb);
5726done:
5727	free(buf, M_CXGBE);
5728	return (rc);
5729}
5730
5731static int
5732sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
5733{
5734	struct adapter *sc = arg1;
5735	u_int i;
5736	struct sbuf *sb;
5737	uint32_t *buf, *p;
5738	int rc;
5739
5740	rc = sysctl_wire_old_buffer(req, 0);
5741	if (rc != 0)
5742		return (rc);
5743
5744	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5745	if (sb == NULL)
5746		return (ENOMEM);
5747
5748	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
5749	    M_ZERO | M_WAITOK);
5750
5751	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
5752	p = buf;
5753
5754	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
5755		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
5756		    p[1], p[0]);
5757	}
5758
5759	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
5760	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
5761		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
5762		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
5763		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
5764		    (p[1] >> 2) | ((p[2] & 3) << 30),
5765		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
5766		    p[0] & 1);
5767	}
5768
5769	rc = sbuf_finish(sb);
5770	sbuf_delete(sb);
5771	free(buf, M_CXGBE);
5772	return (rc);
5773}
5774
5775static int
5776sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
5777{
5778	struct adapter *sc = arg1;
5779	u_int i;
5780	struct sbuf *sb;
5781	uint32_t *buf, *p;
5782	int rc;
5783
5784	rc = sysctl_wire_old_buffer(req, 0);
5785	if (rc != 0)
5786		return (rc);
5787
5788	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
5789	if (sb == NULL)
5790		return (ENOMEM);
5791
5792	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
5793	    M_ZERO | M_WAITOK);
5794
5795	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
5796	p = buf;
5797
5798	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
5799	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
5800		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
5801		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
5802		    p[4], p[3], p[2], p[1], p[0]);
5803	}
5804
5805	sbuf_printf(sb, "\n\nCntl ID               Data");
5806	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
5807		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
5808		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
5809	}
5810
5811	rc = sbuf_finish(sb);
5812	sbuf_delete(sb);
5813	free(buf, M_CXGBE);
5814	return (rc);
5815}
5816
5817static int
5818sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
5819{
5820	struct adapter *sc = arg1;
5821	struct sbuf *sb;
5822	int rc, i;
5823	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
5824	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
5825	uint16_t thres[CIM_NUM_IBQ];
5826	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
5827	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
5828	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
5829
5830	cim_num_obq = sc->chip_params->cim_num_obq;
5831	if (is_t4(sc)) {
5832		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
5833		obq_rdaddr = A_UP_OBQ_0_REALADDR;
5834	} else {
5835		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
5836		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
5837	}
5838	nq = CIM_NUM_IBQ + cim_num_obq;
5839
5840	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
5841	if (rc == 0)
5842		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
5843	if (rc != 0)
5844		return (rc);
5845
5846	t4_read_cimq_cfg(sc, base, size, thres);
5847
5848	rc = sysctl_wire_old_buffer(req, 0);
5849	if (rc != 0)
5850		return (rc);
5851
5852	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
5853	if (sb == NULL)
5854		return (ENOMEM);
5855
5856	sbuf_printf(sb, "Queue  Base  Size Thres RdPtr WrPtr  SOP  EOP Avail");
5857
5858	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
5859		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
5860		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
5861		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
5862		    G_QUEREMFLITS(p[2]) * 16);
5863	for ( ; i < nq; i++, p += 4, wr += 2)
5864		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
5865		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
5866		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
5867		    G_QUEREMFLITS(p[2]) * 16);
5868
5869	rc = sbuf_finish(sb);
5870	sbuf_delete(sb);
5871
5872	return (rc);
5873}
5874
5875static int
5876sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
5877{
5878	struct adapter *sc = arg1;
5879	struct sbuf *sb;
5880	int rc;
5881	struct tp_cpl_stats stats;
5882
5883	rc = sysctl_wire_old_buffer(req, 0);
5884	if (rc != 0)
5885		return (rc);
5886
5887	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
5888	if (sb == NULL)
5889		return (ENOMEM);
5890
5891	mtx_lock(&sc->reg_lock);
5892	t4_tp_get_cpl_stats(sc, &stats);
5893	mtx_unlock(&sc->reg_lock);
5894
5895	if (sc->chip_params->nchan > 2) {
5896		sbuf_printf(sb, "                 channel 0  channel 1"
5897		    "  channel 2  channel 3");
5898		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
5899		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
5900		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
5901		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
5902	} else {
5903		sbuf_printf(sb, "                 channel 0  channel 1");
5904		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
5905		    stats.req[0], stats.req[1]);
5906		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
5907		    stats.rsp[0], stats.rsp[1]);
5908	}
5909
5910	rc = sbuf_finish(sb);
5911	sbuf_delete(sb);
5912
5913	return (rc);
5914}
5915
5916static int
5917sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
5918{
5919	struct adapter *sc = arg1;
5920	struct sbuf *sb;
5921	int rc;
5922	struct tp_usm_stats stats;
5923
5924	rc = sysctl_wire_old_buffer(req, 0);
5925	if (rc != 0)
5926		return(rc);
5927
5928	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
5929	if (sb == NULL)
5930		return (ENOMEM);
5931
5932	t4_get_usm_stats(sc, &stats);
5933
5934	sbuf_printf(sb, "Frames: %u\n", stats.frames);
5935	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
5936	sbuf_printf(sb, "Drops:  %u", stats.drops);
5937
5938	rc = sbuf_finish(sb);
5939	sbuf_delete(sb);
5940
5941	return (rc);
5942}
5943
5944static const char * const devlog_level_strings[] = {
5945	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
5946	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
5947	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
5948	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
5949	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
5950	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
5951};
5952
5953static const char * const devlog_facility_strings[] = {
5954	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
5955	[FW_DEVLOG_FACILITY_CF]		= "CF",
5956	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
5957	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
5958	[FW_DEVLOG_FACILITY_RES]	= "RES",
5959	[FW_DEVLOG_FACILITY_HW]		= "HW",
5960	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
5961	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
5962	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
5963	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
5964	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
5965	[FW_DEVLOG_FACILITY_VI]		= "VI",
5966	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
5967	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
5968	[FW_DEVLOG_FACILITY_TM]		= "TM",
5969	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
5970	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
5971	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
5972	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
5973	[FW_DEVLOG_FACILITY_RI]		= "RI",
5974	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
5975	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
5976	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
5977	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
5978	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
5979};
5980
5981static int
5982sysctl_devlog(SYSCTL_HANDLER_ARGS)
5983{
5984	struct adapter *sc = arg1;
5985	struct devlog_params *dparams = &sc->params.devlog;
5986	struct fw_devlog_e *buf, *e;
5987	int i, j, rc, nentries, first = 0;
5988	struct sbuf *sb;
5989	uint64_t ftstamp = UINT64_MAX;
5990
5991	if (dparams->addr == 0)
5992		return (ENXIO);
5993
5994	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
5995	if (buf == NULL)
5996		return (ENOMEM);
5997
5998	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
5999	if (rc != 0)
6000		goto done;
6001
6002	nentries = dparams->size / sizeof(struct fw_devlog_e);
6003	for (i = 0; i < nentries; i++) {
6004		e = &buf[i];
6005
6006		if (e->timestamp == 0)
6007			break;	/* end */
6008
6009		e->timestamp = be64toh(e->timestamp);
6010		e->seqno = be32toh(e->seqno);
6011		for (j = 0; j < 8; j++)
6012			e->params[j] = be32toh(e->params[j]);
6013
6014		if (e->timestamp < ftstamp) {
6015			ftstamp = e->timestamp;
6016			first = i;
6017		}
6018	}
6019
6020	if (buf[first].timestamp == 0)
6021		goto done;	/* nothing in the log */
6022
6023	rc = sysctl_wire_old_buffer(req, 0);
6024	if (rc != 0)
6025		goto done;
6026
6027	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6028	if (sb == NULL) {
6029		rc = ENOMEM;
6030		goto done;
6031	}
6032	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
6033	    "Seq#", "Tstamp", "Level", "Facility", "Message");
6034
6035	i = first;
6036	do {
6037		e = &buf[i];
6038		if (e->timestamp == 0)
6039			break;	/* end */
6040
6041		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
6042		    e->seqno, e->timestamp,
6043		    (e->level < nitems(devlog_level_strings) ?
6044			devlog_level_strings[e->level] : "UNKNOWN"),
6045		    (e->facility < nitems(devlog_facility_strings) ?
6046			devlog_facility_strings[e->facility] : "UNKNOWN"));
6047		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
6048		    e->params[2], e->params[3], e->params[4],
6049		    e->params[5], e->params[6], e->params[7]);
6050
6051		if (++i == nentries)
6052			i = 0;
6053	} while (i != first);
6054
6055	rc = sbuf_finish(sb);
6056	sbuf_delete(sb);
6057done:
6058	free(buf, M_CXGBE);
6059	return (rc);
6060}
6061
6062static int
6063sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
6064{
6065	struct adapter *sc = arg1;
6066	struct sbuf *sb;
6067	int rc;
6068	struct tp_fcoe_stats stats[MAX_NCHAN];
6069	int i, nchan = sc->chip_params->nchan;
6070
6071	rc = sysctl_wire_old_buffer(req, 0);
6072	if (rc != 0)
6073		return (rc);
6074
6075	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6076	if (sb == NULL)
6077		return (ENOMEM);
6078
6079	for (i = 0; i < nchan; i++)
6080		t4_get_fcoe_stats(sc, i, &stats[i]);
6081
6082	if (nchan > 2) {
6083		sbuf_printf(sb, "                   channel 0        channel 1"
6084		    "        channel 2        channel 3");
6085		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
6086		    stats[0].octets_ddp, stats[1].octets_ddp,
6087		    stats[2].octets_ddp, stats[3].octets_ddp);
6088		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
6089		    stats[0].frames_ddp, stats[1].frames_ddp,
6090		    stats[2].frames_ddp, stats[3].frames_ddp);
6091		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
6092		    stats[0].frames_drop, stats[1].frames_drop,
6093		    stats[2].frames_drop, stats[3].frames_drop);
6094	} else {
6095		sbuf_printf(sb, "                   channel 0        channel 1");
6096		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
6097		    stats[0].octets_ddp, stats[1].octets_ddp);
6098		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
6099		    stats[0].frames_ddp, stats[1].frames_ddp);
6100		sbuf_printf(sb, "\nframesDrop: %16u %16u",
6101		    stats[0].frames_drop, stats[1].frames_drop);
6102	}
6103
6104	rc = sbuf_finish(sb);
6105	sbuf_delete(sb);
6106
6107	return (rc);
6108}
6109
6110static int
6111sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
6112{
6113	struct adapter *sc = arg1;
6114	struct sbuf *sb;
6115	int rc, i;
6116	unsigned int map, kbps, ipg, mode;
6117	unsigned int pace_tab[NTX_SCHED];
6118
6119	rc = sysctl_wire_old_buffer(req, 0);
6120	if (rc != 0)
6121		return (rc);
6122
6123	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6124	if (sb == NULL)
6125		return (ENOMEM);
6126
6127	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
6128	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
6129	t4_read_pace_tbl(sc, pace_tab);
6130
6131	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
6132	    "Class IPG (0.1 ns)   Flow IPG (us)");
6133
6134	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
6135		t4_get_tx_sched(sc, i, &kbps, &ipg);
6136		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
6137		    (mode & (1 << i)) ? "flow" : "class", map & 3);
6138		if (kbps)
6139			sbuf_printf(sb, "%9u     ", kbps);
6140		else
6141			sbuf_printf(sb, " disabled     ");
6142
6143		if (ipg)
6144			sbuf_printf(sb, "%13u        ", ipg);
6145		else
6146			sbuf_printf(sb, "     disabled        ");
6147
6148		if (pace_tab[i])
6149			sbuf_printf(sb, "%10u", pace_tab[i]);
6150		else
6151			sbuf_printf(sb, "  disabled");
6152	}
6153
6154	rc = sbuf_finish(sb);
6155	sbuf_delete(sb);
6156
6157	return (rc);
6158}
6159
6160static int
6161sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
6162{
6163	struct adapter *sc = arg1;
6164	struct sbuf *sb;
6165	int rc, i, j;
6166	uint64_t *p0, *p1;
6167	struct lb_port_stats s[2];
6168	static const char *stat_name[] = {
6169		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
6170		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
6171		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
6172		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
6173		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
6174		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
6175		"BG2FramesTrunc:", "BG3FramesTrunc:"
6176	};
6177
6178	rc = sysctl_wire_old_buffer(req, 0);
6179	if (rc != 0)
6180		return (rc);
6181
6182	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6183	if (sb == NULL)
6184		return (ENOMEM);
6185
6186	memset(s, 0, sizeof(s));
6187
6188	for (i = 0; i < sc->chip_params->nchan; i += 2) {
6189		t4_get_lb_stats(sc, i, &s[0]);
6190		t4_get_lb_stats(sc, i + 1, &s[1]);
6191
6192		p0 = &s[0].octets;
6193		p1 = &s[1].octets;
6194		sbuf_printf(sb, "%s                       Loopback %u"
6195		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
6196
6197		for (j = 0; j < nitems(stat_name); j++)
6198			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
6199				   *p0++, *p1++);
6200	}
6201
6202	rc = sbuf_finish(sb);
6203	sbuf_delete(sb);
6204
6205	return (rc);
6206}
6207
6208static int
6209sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
6210{
6211	int rc = 0;
6212	struct port_info *pi = arg1;
6213	struct sbuf *sb;
6214
6215	rc = sysctl_wire_old_buffer(req, 0);
6216	if (rc != 0)
6217		return(rc);
6218	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
6219	if (sb == NULL)
6220		return (ENOMEM);
6221
6222	if (pi->linkdnrc < 0)
6223		sbuf_printf(sb, "n/a");
6224	else
6225		sbuf_printf(sb, "%s", t4_link_down_rc_str(pi->linkdnrc));
6226
6227	rc = sbuf_finish(sb);
6228	sbuf_delete(sb);
6229
6230	return (rc);
6231}
6232
6233struct mem_desc {
6234	unsigned int base;
6235	unsigned int limit;
6236	unsigned int idx;
6237};
6238
6239static int
6240mem_desc_cmp(const void *a, const void *b)
6241{
6242	return ((const struct mem_desc *)a)->base -
6243	       ((const struct mem_desc *)b)->base;
6244}
6245
6246static void
6247mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
6248    unsigned int to)
6249{
6250	unsigned int size;
6251
6252	if (from == to)
6253		return;
6254
6255	size = to - from + 1;
6256	if (size == 0)
6257		return;
6258
6259	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
6260	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
6261}
6262
6263static int
6264sysctl_meminfo(SYSCTL_HANDLER_ARGS)
6265{
6266	struct adapter *sc = arg1;
6267	struct sbuf *sb;
6268	int rc, i, n;
6269	uint32_t lo, hi, used, alloc;
6270	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
6271	static const char *region[] = {
6272		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
6273		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
6274		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
6275		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
6276		"RQUDP region:", "PBL region:", "TXPBL region:",
6277		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
6278		"On-chip queues:"
6279	};
6280	struct mem_desc avail[4];
6281	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
6282	struct mem_desc *md = mem;
6283
6284	rc = sysctl_wire_old_buffer(req, 0);
6285	if (rc != 0)
6286		return (rc);
6287
6288	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6289	if (sb == NULL)
6290		return (ENOMEM);
6291
6292	for (i = 0; i < nitems(mem); i++) {
6293		mem[i].limit = 0;
6294		mem[i].idx = i;
6295	}
6296
6297	/* Find and sort the populated memory ranges */
6298	i = 0;
6299	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
6300	if (lo & F_EDRAM0_ENABLE) {
6301		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
6302		avail[i].base = G_EDRAM0_BASE(hi) << 20;
6303		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
6304		avail[i].idx = 0;
6305		i++;
6306	}
6307	if (lo & F_EDRAM1_ENABLE) {
6308		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
6309		avail[i].base = G_EDRAM1_BASE(hi) << 20;
6310		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
6311		avail[i].idx = 1;
6312		i++;
6313	}
6314	if (lo & F_EXT_MEM_ENABLE) {
6315		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
6316		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
6317		avail[i].limit = avail[i].base +
6318		    (G_EXT_MEM_SIZE(hi) << 20);
6319		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
6320		i++;
6321	}
6322	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
6323		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
6324		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
6325		avail[i].limit = avail[i].base +
6326		    (G_EXT_MEM1_SIZE(hi) << 20);
6327		avail[i].idx = 4;
6328		i++;
6329	}
6330	if (!i)                                    /* no memory available */
6331		return 0;
6332	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
6333
6334	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
6335	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
6336	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
6337	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
6338	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
6339	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
6340	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
6341	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
6342	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
6343
6344	/* the next few have explicit upper bounds */
6345	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
6346	md->limit = md->base - 1 +
6347		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
6348		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
6349	md++;
6350
6351	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
6352	md->limit = md->base - 1 +
6353		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
6354		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
6355	md++;
6356
6357	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
6358		if (chip_id(sc) <= CHELSIO_T5)
6359			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
6360		else
6361			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
6362		md->limit = 0;
6363	} else {
6364		md->base = 0;
6365		md->idx = nitems(region);  /* hide it */
6366	}
6367	md++;
6368
6369#define ulp_region(reg) \
6370	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
6371	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
6372
6373	ulp_region(RX_ISCSI);
6374	ulp_region(RX_TDDP);
6375	ulp_region(TX_TPT);
6376	ulp_region(RX_STAG);
6377	ulp_region(RX_RQ);
6378	ulp_region(RX_RQUDP);
6379	ulp_region(RX_PBL);
6380	ulp_region(TX_PBL);
6381#undef ulp_region
6382
6383	md->base = 0;
6384	md->idx = nitems(region);
6385	if (!is_t4(sc)) {
6386		uint32_t size = 0;
6387		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
6388		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
6389
6390		if (is_t5(sc)) {
6391			if (sge_ctrl & F_VFIFO_ENABLE)
6392				size = G_DBVFIFO_SIZE(fifo_size);
6393		} else
6394			size = G_T6_DBVFIFO_SIZE(fifo_size);
6395
6396		if (size) {
6397			md->base = G_BASEADDR(t4_read_reg(sc,
6398			    A_SGE_DBVFIFO_BADDR));
6399			md->limit = md->base + (size << 2) - 1;
6400		}
6401	}
6402	md++;
6403
6404	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
6405	md->limit = 0;
6406	md++;
6407	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
6408	md->limit = 0;
6409	md++;
6410
6411	md->base = sc->vres.ocq.start;
6412	if (sc->vres.ocq.size)
6413		md->limit = md->base + sc->vres.ocq.size - 1;
6414	else
6415		md->idx = nitems(region);  /* hide it */
6416	md++;
6417
6418	/* add any address-space holes, there can be up to 3 */
6419	for (n = 0; n < i - 1; n++)
6420		if (avail[n].limit < avail[n + 1].base)
6421			(md++)->base = avail[n].limit;
6422	if (avail[n].limit)
6423		(md++)->base = avail[n].limit;
6424
6425	n = md - mem;
6426	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
6427
6428	for (lo = 0; lo < i; lo++)
6429		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
6430				avail[lo].limit - 1);
6431
6432	sbuf_printf(sb, "\n");
6433	for (i = 0; i < n; i++) {
6434		if (mem[i].idx >= nitems(region))
6435			continue;                        /* skip holes */
6436		if (!mem[i].limit)
6437			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
6438		mem_region_show(sb, region[mem[i].idx], mem[i].base,
6439				mem[i].limit);
6440	}
6441
6442	sbuf_printf(sb, "\n");
6443	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
6444	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
6445	mem_region_show(sb, "uP RAM:", lo, hi);
6446
6447	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
6448	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
6449	mem_region_show(sb, "uP Extmem2:", lo, hi);
6450
6451	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
6452	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
6453		   G_PMRXMAXPAGE(lo),
6454		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
6455		   (lo & F_PMRXNUMCHN) ? 2 : 1);
6456
6457	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
6458	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
6459	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
6460		   G_PMTXMAXPAGE(lo),
6461		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
6462		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
6463	sbuf_printf(sb, "%u p-structs\n",
6464		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
6465
6466	for (i = 0; i < 4; i++) {
6467		if (chip_id(sc) > CHELSIO_T5)
6468			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
6469		else
6470			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
6471		if (is_t5(sc)) {
6472			used = G_T5_USED(lo);
6473			alloc = G_T5_ALLOC(lo);
6474		} else {
6475			used = G_USED(lo);
6476			alloc = G_ALLOC(lo);
6477		}
6478		/* For T6 these are MAC buffer groups */
6479		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
6480		    i, used, alloc);
6481	}
6482	for (i = 0; i < sc->chip_params->nchan; i++) {
6483		if (chip_id(sc) > CHELSIO_T5)
6484			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
6485		else
6486			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
6487		if (is_t5(sc)) {
6488			used = G_T5_USED(lo);
6489			alloc = G_T5_ALLOC(lo);
6490		} else {
6491			used = G_USED(lo);
6492			alloc = G_ALLOC(lo);
6493		}
6494		/* For T6 these are MAC buffer groups */
6495		sbuf_printf(sb,
6496		    "\nLoopback %d using %u pages out of %u allocated",
6497		    i, used, alloc);
6498	}
6499
6500	rc = sbuf_finish(sb);
6501	sbuf_delete(sb);
6502
6503	return (rc);
6504}
6505
6506static inline void
6507tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
6508{
6509	*mask = x | y;
6510	y = htobe64(y);
6511	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
6512}
6513
6514static int
6515sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
6516{
6517	struct adapter *sc = arg1;
6518	struct sbuf *sb;
6519	int rc, i;
6520
6521	MPASS(chip_id(sc) <= CHELSIO_T5);
6522
6523	rc = sysctl_wire_old_buffer(req, 0);
6524	if (rc != 0)
6525		return (rc);
6526
6527	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6528	if (sb == NULL)
6529		return (ENOMEM);
6530
6531	sbuf_printf(sb,
6532	    "Idx  Ethernet address     Mask     Vld Ports PF"
6533	    "  VF              Replication             P0 P1 P2 P3  ML");
6534	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
6535		uint64_t tcamx, tcamy, mask;
6536		uint32_t cls_lo, cls_hi;
6537		uint8_t addr[ETHER_ADDR_LEN];
6538
6539		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
6540		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
6541		if (tcamx & tcamy)
6542			continue;
6543		tcamxy2valmask(tcamx, tcamy, addr, &mask);
6544		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
6545		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
6546		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
6547			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
6548			   addr[3], addr[4], addr[5], (uintmax_t)mask,
6549			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
6550			   G_PORTMAP(cls_hi), G_PF(cls_lo),
6551			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
6552
6553		if (cls_lo & F_REPLICATE) {
6554			struct fw_ldst_cmd ldst_cmd;
6555
6556			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
6557			ldst_cmd.op_to_addrspace =
6558			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
6559				F_FW_CMD_REQUEST | F_FW_CMD_READ |
6560				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
6561			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
6562			ldst_cmd.u.mps.rplc.fid_idx =
6563			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
6564				V_FW_LDST_CMD_IDX(i));
6565
6566			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
6567			    "t4mps");
6568			if (rc)
6569				break;
6570			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
6571			    sizeof(ldst_cmd), &ldst_cmd);
6572			end_synchronized_op(sc, 0);
6573
6574			if (rc != 0) {
6575				sbuf_printf(sb, "%36d", rc);
6576				rc = 0;
6577			} else {
6578				sbuf_printf(sb, " %08x %08x %08x %08x",
6579				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
6580				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
6581				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
6582				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
6583			}
6584		} else
6585			sbuf_printf(sb, "%36s", "");
6586
6587		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
6588		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
6589		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
6590	}
6591
6592	if (rc)
6593		(void) sbuf_finish(sb);
6594	else
6595		rc = sbuf_finish(sb);
6596	sbuf_delete(sb);
6597
6598	return (rc);
6599}
6600
6601static int
6602sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
6603{
6604	struct adapter *sc = arg1;
6605	struct sbuf *sb;
6606	int rc, i;
6607
6608	MPASS(chip_id(sc) > CHELSIO_T5);
6609
6610	rc = sysctl_wire_old_buffer(req, 0);
6611	if (rc != 0)
6612		return (rc);
6613
6614	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
6615	if (sb == NULL)
6616		return (ENOMEM);
6617
6618	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
6619	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
6620	    "                           Replication"
6621	    "                                    P0 P1 P2 P3  ML\n");
6622
6623	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
6624		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
6625		uint16_t ivlan;
6626		uint64_t tcamx, tcamy, val, mask;
6627		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
6628		uint8_t addr[ETHER_ADDR_LEN];
6629
6630		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
6631		if (i < 256)
6632			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
6633		else
6634			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
6635		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
6636		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
6637		tcamy = G_DMACH(val) << 32;
6638		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
6639		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
6640		lookup_type = G_DATALKPTYPE(data2);
6641		port_num = G_DATAPORTNUM(data2);
6642		if (lookup_type && lookup_type != M_DATALKPTYPE) {
6643			/* Inner header VNI */
6644			vniy = ((data2 & F_DATAVIDH2) << 23) |
6645				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
6646			dip_hit = data2 & F_DATADIPHIT;
6647			vlan_vld = 0;
6648		} else {
6649			vniy = 0;
6650			dip_hit = 0;
6651			vlan_vld = data2 & F_DATAVIDH2;
6652			ivlan = G_VIDL(val);
6653		}
6654
6655		ctl |= V_CTLXYBITSEL(1);
6656		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
6657		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
6658		tcamx = G_DMACH(val) << 32;
6659		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
6660		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
6661		if (lookup_type && lookup_type != M_DATALKPTYPE) {
6662			/* Inner header VNI mask */
6663			vnix = ((data2 & F_DATAVIDH2) << 23) |
6664			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
6665		} else
6666			vnix = 0;
6667
6668		if (tcamx & tcamy)
6669			continue;
6670		tcamxy2valmask(tcamx, tcamy, addr, &mask);
6671
6672		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
6673		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
6674
6675		if (lookup_type && lookup_type != M_DATALKPTYPE) {
6676			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
6677			    "%012jx %06x %06x    -    -   %3c"
6678			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
6679			    addr[1], addr[2], addr[3], addr[4], addr[5],
6680			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
6681			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
6682			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
6683			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
6684		} else {
6685			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
6686			    "%012jx    -       -   ", i, addr[0], addr[1],
6687			    addr[2], addr[3], addr[4], addr[5],
6688			    (uintmax_t)mask);
6689
6690			if (vlan_vld)
6691				sbuf_printf(sb, "%4u   Y     ", ivlan);
6692			else
6693				sbuf_printf(sb, "  -    N     ");
6694
6695			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
6696			    lookup_type ? 'I' : 'O', port_num,
6697			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
6698			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
6699			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
6700		}
6701
6702
6703		if (cls_lo & F_T6_REPLICATE) {
6704			struct fw_ldst_cmd ldst_cmd;
6705
6706			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
6707			ldst_cmd.op_to_addrspace =
6708			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
6709				F_FW_CMD_REQUEST | F_FW_CMD_READ |
6710				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
6711			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
6712			ldst_cmd.u.mps.rplc.fid_idx =
6713			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
6714				V_FW_LDST_CMD_IDX(i));
6715
6716			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
6717			    "t6mps");
6718			if (rc)
6719				break;
6720			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
6721			    sizeof(ldst_cmd), &ldst_cmd);
6722			end_synchronized_op(sc, 0);
6723
6724			if (rc != 0) {
6725				sbuf_printf(sb, "%72d", rc);
6726				rc = 0;
6727			} else {
6728				sbuf_printf(sb, " %08x %08x %08x %08x"
6729				    " %08x %08x %08x %08x",
6730				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
6731				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
6732				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
6733				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
6734				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
6735				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
6736				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
6737				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
6738			}
6739		} else
6740			sbuf_printf(sb, "%72s", "");
6741
6742		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
6743		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
6744		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
6745		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
6746	}
6747
6748	if (rc)
6749		(void) sbuf_finish(sb);
6750	else
6751		rc = sbuf_finish(sb);
6752	sbuf_delete(sb);
6753
6754	return (rc);
6755}
6756
6757static int
6758sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
6759{
6760	struct adapter *sc = arg1;
6761	struct sbuf *sb;
6762	int rc;
6763	uint16_t mtus[NMTUS];
6764
6765	rc = sysctl_wire_old_buffer(req, 0);
6766	if (rc != 0)
6767		return (rc);
6768
6769	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6770	if (sb == NULL)
6771		return (ENOMEM);
6772
6773	t4_read_mtu_tbl(sc, mtus, NULL);
6774
6775	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
6776	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
6777	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
6778	    mtus[14], mtus[15]);
6779
6780	rc = sbuf_finish(sb);
6781	sbuf_delete(sb);
6782
6783	return (rc);
6784}
6785
6786static int
6787sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
6788{
6789	struct adapter *sc = arg1;
6790	struct sbuf *sb;
6791	int rc, i;
6792	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
6793	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
6794	static const char *tx_stats[MAX_PM_NSTATS] = {
6795		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
6796		"Tx FIFO wait", NULL, "Tx latency"
6797	};
6798	static const char *rx_stats[MAX_PM_NSTATS] = {
6799		"Read:", "Write bypass:", "Write mem:", "Flush:",
6800		" Rx FIFO wait", NULL, "Rx latency"
6801	};
6802
6803	rc = sysctl_wire_old_buffer(req, 0);
6804	if (rc != 0)
6805		return (rc);
6806
6807	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6808	if (sb == NULL)
6809		return (ENOMEM);
6810
6811	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
6812	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
6813
6814	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
6815	for (i = 0; i < 4; i++) {
6816		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
6817		    tx_cyc[i]);
6818	}
6819
6820	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
6821	for (i = 0; i < 4; i++) {
6822		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
6823		    rx_cyc[i]);
6824	}
6825
6826	if (chip_id(sc) > CHELSIO_T5) {
6827		sbuf_printf(sb,
6828		    "\n              Total wait      Total occupancy");
6829		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
6830		    tx_cyc[i]);
6831		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
6832		    rx_cyc[i]);
6833
6834		i += 2;
6835		MPASS(i < nitems(tx_stats));
6836
6837		sbuf_printf(sb,
6838		    "\n                   Reads           Total wait");
6839		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
6840		    tx_cyc[i]);
6841		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
6842		    rx_cyc[i]);
6843	}
6844
6845	rc = sbuf_finish(sb);
6846	sbuf_delete(sb);
6847
6848	return (rc);
6849}
6850
6851static int
6852sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
6853{
6854	struct adapter *sc = arg1;
6855	struct sbuf *sb;
6856	int rc;
6857	struct tp_rdma_stats stats;
6858
6859	rc = sysctl_wire_old_buffer(req, 0);
6860	if (rc != 0)
6861		return (rc);
6862
6863	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6864	if (sb == NULL)
6865		return (ENOMEM);
6866
6867	mtx_lock(&sc->reg_lock);
6868	t4_tp_get_rdma_stats(sc, &stats);
6869	mtx_unlock(&sc->reg_lock);
6870
6871	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
6872	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
6873
6874	rc = sbuf_finish(sb);
6875	sbuf_delete(sb);
6876
6877	return (rc);
6878}
6879
6880static int
6881sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
6882{
6883	struct adapter *sc = arg1;
6884	struct sbuf *sb;
6885	int rc;
6886	struct tp_tcp_stats v4, v6;
6887
6888	rc = sysctl_wire_old_buffer(req, 0);
6889	if (rc != 0)
6890		return (rc);
6891
6892	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6893	if (sb == NULL)
6894		return (ENOMEM);
6895
6896	mtx_lock(&sc->reg_lock);
6897	t4_tp_get_tcp_stats(sc, &v4, &v6);
6898	mtx_unlock(&sc->reg_lock);
6899
6900	sbuf_printf(sb,
6901	    "                                IP                 IPv6\n");
6902	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
6903	    v4.tcp_out_rsts, v6.tcp_out_rsts);
6904	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
6905	    v4.tcp_in_segs, v6.tcp_in_segs);
6906	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
6907	    v4.tcp_out_segs, v6.tcp_out_segs);
6908	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
6909	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
6910
6911	rc = sbuf_finish(sb);
6912	sbuf_delete(sb);
6913
6914	return (rc);
6915}
6916
6917static int
6918sysctl_tids(SYSCTL_HANDLER_ARGS)
6919{
6920	struct adapter *sc = arg1;
6921	struct sbuf *sb;
6922	int rc;
6923	struct tid_info *t = &sc->tids;
6924
6925	rc = sysctl_wire_old_buffer(req, 0);
6926	if (rc != 0)
6927		return (rc);
6928
6929	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6930	if (sb == NULL)
6931		return (ENOMEM);
6932
6933	if (t->natids) {
6934		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
6935		    t->atids_in_use);
6936	}
6937
6938	if (t->ntids) {
6939		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
6940			uint32_t b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
6941
6942			if (b) {
6943				sbuf_printf(sb, "TID range: 0-%u, %u-%u", b - 1,
6944				    t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4,
6945				    t->ntids - 1);
6946			} else {
6947				sbuf_printf(sb, "TID range: %u-%u",
6948				    t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4,
6949				    t->ntids - 1);
6950			}
6951		} else
6952			sbuf_printf(sb, "TID range: 0-%u", t->ntids - 1);
6953		sbuf_printf(sb, ", in use: %u\n",
6954		    atomic_load_acq_int(&t->tids_in_use));
6955	}
6956
6957	if (t->nstids) {
6958		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
6959		    t->stid_base + t->nstids - 1, t->stids_in_use);
6960	}
6961
6962	if (t->nftids) {
6963		sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
6964		    t->ftid_base + t->nftids - 1);
6965	}
6966
6967	if (t->netids) {
6968		sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
6969		    t->etid_base + t->netids - 1);
6970	}
6971
6972	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
6973	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
6974	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
6975
6976	rc = sbuf_finish(sb);
6977	sbuf_delete(sb);
6978
6979	return (rc);
6980}
6981
6982static int
6983sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
6984{
6985	struct adapter *sc = arg1;
6986	struct sbuf *sb;
6987	int rc;
6988	struct tp_err_stats stats;
6989
6990	rc = sysctl_wire_old_buffer(req, 0);
6991	if (rc != 0)
6992		return (rc);
6993
6994	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
6995	if (sb == NULL)
6996		return (ENOMEM);
6997
6998	mtx_lock(&sc->reg_lock);
6999	t4_tp_get_err_stats(sc, &stats);
7000	mtx_unlock(&sc->reg_lock);
7001
7002	if (sc->chip_params->nchan > 2) {
7003		sbuf_printf(sb, "                 channel 0  channel 1"
7004		    "  channel 2  channel 3\n");
7005		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
7006		    stats.mac_in_errs[0], stats.mac_in_errs[1],
7007		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
7008		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
7009		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
7010		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
7011		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
7012		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
7013		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
7014		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
7015		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
7016		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
7017		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
7018		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
7019		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
7020		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
7021		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
7022		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
7023		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
7024		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
7025		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
7026		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
7027		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
7028		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
7029	} else {
7030		sbuf_printf(sb, "                 channel 0  channel 1\n");
7031		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
7032		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
7033		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
7034		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
7035		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
7036		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
7037		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
7038		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
7039		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
7040		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
7041		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
7042		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
7043		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
7044		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
7045		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
7046		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
7047	}
7048
7049	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
7050	    stats.ofld_no_neigh, stats.ofld_cong_defer);
7051
7052	rc = sbuf_finish(sb);
7053	sbuf_delete(sb);
7054
7055	return (rc);
7056}
7057
7058static int
7059sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
7060{
7061	struct adapter *sc = arg1;
7062	struct tp_params *tpp = &sc->params.tp;
7063	u_int mask;
7064	int rc;
7065
7066	mask = tpp->la_mask >> 16;
7067	rc = sysctl_handle_int(oidp, &mask, 0, req);
7068	if (rc != 0 || req->newptr == NULL)
7069		return (rc);
7070	if (mask > 0xffff)
7071		return (EINVAL);
7072	tpp->la_mask = mask << 16;
7073	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
7074
7075	return (0);
7076}
7077
7078struct field_desc {
7079	const char *name;
7080	u_int start;
7081	u_int width;
7082};
7083
7084static void
7085field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
7086{
7087	char buf[32];
7088	int line_size = 0;
7089
7090	while (f->name) {
7091		uint64_t mask = (1ULL << f->width) - 1;
7092		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
7093		    ((uintmax_t)v >> f->start) & mask);
7094
7095		if (line_size + len >= 79) {
7096			line_size = 8;
7097			sbuf_printf(sb, "\n        ");
7098		}
7099		sbuf_printf(sb, "%s ", buf);
7100		line_size += len + 1;
7101		f++;
7102	}
7103	sbuf_printf(sb, "\n");
7104}
7105
7106static const struct field_desc tp_la0[] = {
7107	{ "RcfOpCodeOut", 60, 4 },
7108	{ "State", 56, 4 },
7109	{ "WcfState", 52, 4 },
7110	{ "RcfOpcSrcOut", 50, 2 },
7111	{ "CRxError", 49, 1 },
7112	{ "ERxError", 48, 1 },
7113	{ "SanityFailed", 47, 1 },
7114	{ "SpuriousMsg", 46, 1 },
7115	{ "FlushInputMsg", 45, 1 },
7116	{ "FlushInputCpl", 44, 1 },
7117	{ "RssUpBit", 43, 1 },
7118	{ "RssFilterHit", 42, 1 },
7119	{ "Tid", 32, 10 },
7120	{ "InitTcb", 31, 1 },
7121	{ "LineNumber", 24, 7 },
7122	{ "Emsg", 23, 1 },
7123	{ "EdataOut", 22, 1 },
7124	{ "Cmsg", 21, 1 },
7125	{ "CdataOut", 20, 1 },
7126	{ "EreadPdu", 19, 1 },
7127	{ "CreadPdu", 18, 1 },
7128	{ "TunnelPkt", 17, 1 },
7129	{ "RcfPeerFin", 16, 1 },
7130	{ "RcfReasonOut", 12, 4 },
7131	{ "TxCchannel", 10, 2 },
7132	{ "RcfTxChannel", 8, 2 },
7133	{ "RxEchannel", 6, 2 },
7134	{ "RcfRxChannel", 5, 1 },
7135	{ "RcfDataOutSrdy", 4, 1 },
7136	{ "RxDvld", 3, 1 },
7137	{ "RxOoDvld", 2, 1 },
7138	{ "RxCongestion", 1, 1 },
7139	{ "TxCongestion", 0, 1 },
7140	{ NULL }
7141};
7142
7143static const struct field_desc tp_la1[] = {
7144	{ "CplCmdIn", 56, 8 },
7145	{ "CplCmdOut", 48, 8 },
7146	{ "ESynOut", 47, 1 },
7147	{ "EAckOut", 46, 1 },
7148	{ "EFinOut", 45, 1 },
7149	{ "ERstOut", 44, 1 },
7150	{ "SynIn", 43, 1 },
7151	{ "AckIn", 42, 1 },
7152	{ "FinIn", 41, 1 },
7153	{ "RstIn", 40, 1 },
7154	{ "DataIn", 39, 1 },
7155	{ "DataInVld", 38, 1 },
7156	{ "PadIn", 37, 1 },
7157	{ "RxBufEmpty", 36, 1 },
7158	{ "RxDdp", 35, 1 },
7159	{ "RxFbCongestion", 34, 1 },
7160	{ "TxFbCongestion", 33, 1 },
7161	{ "TxPktSumSrdy", 32, 1 },
7162	{ "RcfUlpType", 28, 4 },
7163	{ "Eread", 27, 1 },
7164	{ "Ebypass", 26, 1 },
7165	{ "Esave", 25, 1 },
7166	{ "Static0", 24, 1 },
7167	{ "Cread", 23, 1 },
7168	{ "Cbypass", 22, 1 },
7169	{ "Csave", 21, 1 },
7170	{ "CPktOut", 20, 1 },
7171	{ "RxPagePoolFull", 18, 2 },
7172	{ "RxLpbkPkt", 17, 1 },
7173	{ "TxLpbkPkt", 16, 1 },
7174	{ "RxVfValid", 15, 1 },
7175	{ "SynLearned", 14, 1 },
7176	{ "SetDelEntry", 13, 1 },
7177	{ "SetInvEntry", 12, 1 },
7178	{ "CpcmdDvld", 11, 1 },
7179	{ "CpcmdSave", 10, 1 },
7180	{ "RxPstructsFull", 8, 2 },
7181	{ "EpcmdDvld", 7, 1 },
7182	{ "EpcmdFlush", 6, 1 },
7183	{ "EpcmdTrimPrefix", 5, 1 },
7184	{ "EpcmdTrimPostfix", 4, 1 },
7185	{ "ERssIp4Pkt", 3, 1 },
7186	{ "ERssIp6Pkt", 2, 1 },
7187	{ "ERssTcpUdpPkt", 1, 1 },
7188	{ "ERssFceFipPkt", 0, 1 },
7189	{ NULL }
7190};
7191
7192static const struct field_desc tp_la2[] = {
7193	{ "CplCmdIn", 56, 8 },
7194	{ "MpsVfVld", 55, 1 },
7195	{ "MpsPf", 52, 3 },
7196	{ "MpsVf", 44, 8 },
7197	{ "SynIn", 43, 1 },
7198	{ "AckIn", 42, 1 },
7199	{ "FinIn", 41, 1 },
7200	{ "RstIn", 40, 1 },
7201	{ "DataIn", 39, 1 },
7202	{ "DataInVld", 38, 1 },
7203	{ "PadIn", 37, 1 },
7204	{ "RxBufEmpty", 36, 1 },
7205	{ "RxDdp", 35, 1 },
7206	{ "RxFbCongestion", 34, 1 },
7207	{ "TxFbCongestion", 33, 1 },
7208	{ "TxPktSumSrdy", 32, 1 },
7209	{ "RcfUlpType", 28, 4 },
7210	{ "Eread", 27, 1 },
7211	{ "Ebypass", 26, 1 },
7212	{ "Esave", 25, 1 },
7213	{ "Static0", 24, 1 },
7214	{ "Cread", 23, 1 },
7215	{ "Cbypass", 22, 1 },
7216	{ "Csave", 21, 1 },
7217	{ "CPktOut", 20, 1 },
7218	{ "RxPagePoolFull", 18, 2 },
7219	{ "RxLpbkPkt", 17, 1 },
7220	{ "TxLpbkPkt", 16, 1 },
7221	{ "RxVfValid", 15, 1 },
7222	{ "SynLearned", 14, 1 },
7223	{ "SetDelEntry", 13, 1 },
7224	{ "SetInvEntry", 12, 1 },
7225	{ "CpcmdDvld", 11, 1 },
7226	{ "CpcmdSave", 10, 1 },
7227	{ "RxPstructsFull", 8, 2 },
7228	{ "EpcmdDvld", 7, 1 },
7229	{ "EpcmdFlush", 6, 1 },
7230	{ "EpcmdTrimPrefix", 5, 1 },
7231	{ "EpcmdTrimPostfix", 4, 1 },
7232	{ "ERssIp4Pkt", 3, 1 },
7233	{ "ERssIp6Pkt", 2, 1 },
7234	{ "ERssTcpUdpPkt", 1, 1 },
7235	{ "ERssFceFipPkt", 0, 1 },
7236	{ NULL }
7237};
7238
7239static void
7240tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
7241{
7242
7243	field_desc_show(sb, *p, tp_la0);
7244}
7245
7246static void
7247tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
7248{
7249
7250	if (idx)
7251		sbuf_printf(sb, "\n");
7252	field_desc_show(sb, p[0], tp_la0);
7253	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7254		field_desc_show(sb, p[1], tp_la0);
7255}
7256
7257static void
7258tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
7259{
7260
7261	if (idx)
7262		sbuf_printf(sb, "\n");
7263	field_desc_show(sb, p[0], tp_la0);
7264	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
7265		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
7266}
7267
7268static int
7269sysctl_tp_la(SYSCTL_HANDLER_ARGS)
7270{
7271	struct adapter *sc = arg1;
7272	struct sbuf *sb;
7273	uint64_t *buf, *p;
7274	int rc;
7275	u_int i, inc;
7276	void (*show_func)(struct sbuf *, uint64_t *, int);
7277
7278	rc = sysctl_wire_old_buffer(req, 0);
7279	if (rc != 0)
7280		return (rc);
7281
7282	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7283	if (sb == NULL)
7284		return (ENOMEM);
7285
7286	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
7287
7288	t4_tp_read_la(sc, buf, NULL);
7289	p = buf;
7290
7291	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
7292	case 2:
7293		inc = 2;
7294		show_func = tp_la_show2;
7295		break;
7296	case 3:
7297		inc = 2;
7298		show_func = tp_la_show3;
7299		break;
7300	default:
7301		inc = 1;
7302		show_func = tp_la_show;
7303	}
7304
7305	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
7306		(*show_func)(sb, p, i);
7307
7308	rc = sbuf_finish(sb);
7309	sbuf_delete(sb);
7310	free(buf, M_CXGBE);
7311	return (rc);
7312}
7313
7314static int
7315sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
7316{
7317	struct adapter *sc = arg1;
7318	struct sbuf *sb;
7319	int rc;
7320	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
7321
7322	rc = sysctl_wire_old_buffer(req, 0);
7323	if (rc != 0)
7324		return (rc);
7325
7326	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7327	if (sb == NULL)
7328		return (ENOMEM);
7329
7330	t4_get_chan_txrate(sc, nrate, orate);
7331
7332	if (sc->chip_params->nchan > 2) {
7333		sbuf_printf(sb, "              channel 0   channel 1"
7334		    "   channel 2   channel 3\n");
7335		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
7336		    nrate[0], nrate[1], nrate[2], nrate[3]);
7337		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
7338		    orate[0], orate[1], orate[2], orate[3]);
7339	} else {
7340		sbuf_printf(sb, "              channel 0   channel 1\n");
7341		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
7342		    nrate[0], nrate[1]);
7343		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
7344		    orate[0], orate[1]);
7345	}
7346
7347	rc = sbuf_finish(sb);
7348	sbuf_delete(sb);
7349
7350	return (rc);
7351}
7352
7353static int
7354sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
7355{
7356	struct adapter *sc = arg1;
7357	struct sbuf *sb;
7358	uint32_t *buf, *p;
7359	int rc, i;
7360
7361	rc = sysctl_wire_old_buffer(req, 0);
7362	if (rc != 0)
7363		return (rc);
7364
7365	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7366	if (sb == NULL)
7367		return (ENOMEM);
7368
7369	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
7370	    M_ZERO | M_WAITOK);
7371
7372	t4_ulprx_read_la(sc, buf);
7373	p = buf;
7374
7375	sbuf_printf(sb, "      Pcmd        Type   Message"
7376	    "                Data");
7377	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
7378		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
7379		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
7380	}
7381
7382	rc = sbuf_finish(sb);
7383	sbuf_delete(sb);
7384	free(buf, M_CXGBE);
7385	return (rc);
7386}
7387
7388static int
7389sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
7390{
7391	struct adapter *sc = arg1;
7392	struct sbuf *sb;
7393	int rc, v;
7394
7395	rc = sysctl_wire_old_buffer(req, 0);
7396	if (rc != 0)
7397		return (rc);
7398
7399	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7400	if (sb == NULL)
7401		return (ENOMEM);
7402
7403	v = t4_read_reg(sc, A_SGE_STAT_CFG);
7404	if (G_STATSOURCE_T5(v) == 7) {
7405		if (G_STATMODE(v) == 0) {
7406			sbuf_printf(sb, "total %d, incomplete %d",
7407			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7408			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7409		} else if (G_STATMODE(v) == 1) {
7410			sbuf_printf(sb, "total %d, data overflow %d",
7411			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
7412			    t4_read_reg(sc, A_SGE_STAT_MATCH));
7413		}
7414	}
7415	rc = sbuf_finish(sb);
7416	sbuf_delete(sb);
7417
7418	return (rc);
7419}
7420
7421static int
7422sysctl_tc_params(SYSCTL_HANDLER_ARGS)
7423{
7424	struct adapter *sc = arg1;
7425	struct tx_sched_class *tc;
7426	struct t4_sched_class_params p;
7427	struct sbuf *sb;
7428	int i, rc, port_id, flags, mbps, gbps;
7429
7430	rc = sysctl_wire_old_buffer(req, 0);
7431	if (rc != 0)
7432		return (rc);
7433
7434	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7435	if (sb == NULL)
7436		return (ENOMEM);
7437
7438	port_id = arg2 >> 16;
7439	MPASS(port_id < sc->params.nports);
7440	MPASS(sc->port[port_id] != NULL);
7441	i = arg2 & 0xffff;
7442	MPASS(i < sc->chip_params->nsched_cls);
7443	tc = &sc->port[port_id]->tc[i];
7444
7445	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7446	    "t4tc_p");
7447	if (rc)
7448		goto done;
7449	flags = tc->flags;
7450	p = tc->params;
7451	end_synchronized_op(sc, LOCK_HELD);
7452
7453	if ((flags & TX_SC_OK) == 0) {
7454		sbuf_printf(sb, "none");
7455		goto done;
7456	}
7457
7458	if (p.level == SCHED_CLASS_LEVEL_CL_WRR) {
7459		sbuf_printf(sb, "cl-wrr weight %u", p.weight);
7460		goto done;
7461	} else if (p.level == SCHED_CLASS_LEVEL_CL_RL)
7462		sbuf_printf(sb, "cl-rl");
7463	else if (p.level == SCHED_CLASS_LEVEL_CH_RL)
7464		sbuf_printf(sb, "ch-rl");
7465	else {
7466		rc = ENXIO;
7467		goto done;
7468	}
7469
7470	if (p.ratemode == SCHED_CLASS_RATEMODE_REL) {
7471		/* XXX: top speed or actual link speed? */
7472		gbps = port_top_speed(sc->port[port_id]);
7473		sbuf_printf(sb, " %u%% of %uGbps", p.maxrate, gbps);
7474	}
7475	else if (p.ratemode == SCHED_CLASS_RATEMODE_ABS) {
7476		switch (p.rateunit) {
7477		case SCHED_CLASS_RATEUNIT_BITS:
7478			mbps = p.maxrate / 1000;
7479			gbps = p.maxrate / 1000000;
7480			if (p.maxrate == gbps * 1000000)
7481				sbuf_printf(sb, " %uGbps", gbps);
7482			else if (p.maxrate == mbps * 1000)
7483				sbuf_printf(sb, " %uMbps", mbps);
7484			else
7485				sbuf_printf(sb, " %uKbps", p.maxrate);
7486			break;
7487		case SCHED_CLASS_RATEUNIT_PKTS:
7488			sbuf_printf(sb, " %upps", p.maxrate);
7489			break;
7490		default:
7491			rc = ENXIO;
7492			goto done;
7493		}
7494	}
7495
7496	switch (p.mode) {
7497	case SCHED_CLASS_MODE_CLASS:
7498		sbuf_printf(sb, " aggregate");
7499		break;
7500	case SCHED_CLASS_MODE_FLOW:
7501		sbuf_printf(sb, " per-flow");
7502		break;
7503	default:
7504		rc = ENXIO;
7505		goto done;
7506	}
7507
7508done:
7509	if (rc == 0)
7510		rc = sbuf_finish(sb);
7511	sbuf_delete(sb);
7512
7513	return (rc);
7514}
7515#endif
7516
7517#ifdef TCP_OFFLOAD
7518static void
7519unit_conv(char *buf, size_t len, u_int val, u_int factor)
7520{
7521	u_int rem = val % factor;
7522
7523	if (rem == 0)
7524		snprintf(buf, len, "%u", val / factor);
7525	else {
7526		while (rem % 10 == 0)
7527			rem /= 10;
7528		snprintf(buf, len, "%u.%u", val / factor, rem);
7529	}
7530}
7531
7532static int
7533sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
7534{
7535	struct adapter *sc = arg1;
7536	char buf[16];
7537	u_int res, re;
7538	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7539
7540	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
7541	switch (arg2) {
7542	case 0:
7543		/* timer_tick */
7544		re = G_TIMERRESOLUTION(res);
7545		break;
7546	case 1:
7547		/* TCP timestamp tick */
7548		re = G_TIMESTAMPRESOLUTION(res);
7549		break;
7550	case 2:
7551		/* DACK tick */
7552		re = G_DELAYEDACKRESOLUTION(res);
7553		break;
7554	default:
7555		return (EDOOFUS);
7556	}
7557
7558	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
7559
7560	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
7561}
7562
7563static int
7564sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
7565{
7566	struct adapter *sc = arg1;
7567	u_int res, dack_re, v;
7568	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7569
7570	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
7571	dack_re = G_DELAYEDACKRESOLUTION(res);
7572	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
7573
7574	return (sysctl_handle_int(oidp, &v, 0, req));
7575}
7576
7577static int
7578sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
7579{
7580	struct adapter *sc = arg1;
7581	int reg = arg2;
7582	u_int tre;
7583	u_long tp_tick_us, v;
7584	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
7585
7586	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
7587	    reg == A_TP_PERS_MIN || reg == A_TP_PERS_MAX ||
7588	    reg == A_TP_KEEP_IDLE || A_TP_KEEP_INTVL || reg == A_TP_INIT_SRTT ||
7589	    reg == A_TP_FINWAIT2_TIMER);
7590
7591	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
7592	tp_tick_us = (cclk_ps << tre) / 1000000;
7593
7594	if (reg == A_TP_INIT_SRTT)
7595		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
7596	else
7597		v = tp_tick_us * t4_read_reg(sc, reg);
7598
7599	return (sysctl_handle_long(oidp, &v, 0, req));
7600}
7601#endif
7602
7603static uint32_t
7604fconf_iconf_to_mode(uint32_t fconf, uint32_t iconf)
7605{
7606	uint32_t mode;
7607
7608	mode = T4_FILTER_IPv4 | T4_FILTER_IPv6 | T4_FILTER_IP_SADDR |
7609	    T4_FILTER_IP_DADDR | T4_FILTER_IP_SPORT | T4_FILTER_IP_DPORT;
7610
7611	if (fconf & F_FRAGMENTATION)
7612		mode |= T4_FILTER_IP_FRAGMENT;
7613
7614	if (fconf & F_MPSHITTYPE)
7615		mode |= T4_FILTER_MPS_HIT_TYPE;
7616
7617	if (fconf & F_MACMATCH)
7618		mode |= T4_FILTER_MAC_IDX;
7619
7620	if (fconf & F_ETHERTYPE)
7621		mode |= T4_FILTER_ETH_TYPE;
7622
7623	if (fconf & F_PROTOCOL)
7624		mode |= T4_FILTER_IP_PROTO;
7625
7626	if (fconf & F_TOS)
7627		mode |= T4_FILTER_IP_TOS;
7628
7629	if (fconf & F_VLAN)
7630		mode |= T4_FILTER_VLAN;
7631
7632	if (fconf & F_VNIC_ID) {
7633		mode |= T4_FILTER_VNIC;
7634		if (iconf & F_VNIC)
7635			mode |= T4_FILTER_IC_VNIC;
7636	}
7637
7638	if (fconf & F_PORT)
7639		mode |= T4_FILTER_PORT;
7640
7641	if (fconf & F_FCOE)
7642		mode |= T4_FILTER_FCoE;
7643
7644	return (mode);
7645}
7646
7647static uint32_t
7648mode_to_fconf(uint32_t mode)
7649{
7650	uint32_t fconf = 0;
7651
7652	if (mode & T4_FILTER_IP_FRAGMENT)
7653		fconf |= F_FRAGMENTATION;
7654
7655	if (mode & T4_FILTER_MPS_HIT_TYPE)
7656		fconf |= F_MPSHITTYPE;
7657
7658	if (mode & T4_FILTER_MAC_IDX)
7659		fconf |= F_MACMATCH;
7660
7661	if (mode & T4_FILTER_ETH_TYPE)
7662		fconf |= F_ETHERTYPE;
7663
7664	if (mode & T4_FILTER_IP_PROTO)
7665		fconf |= F_PROTOCOL;
7666
7667	if (mode & T4_FILTER_IP_TOS)
7668		fconf |= F_TOS;
7669
7670	if (mode & T4_FILTER_VLAN)
7671		fconf |= F_VLAN;
7672
7673	if (mode & T4_FILTER_VNIC)
7674		fconf |= F_VNIC_ID;
7675
7676	if (mode & T4_FILTER_PORT)
7677		fconf |= F_PORT;
7678
7679	if (mode & T4_FILTER_FCoE)
7680		fconf |= F_FCOE;
7681
7682	return (fconf);
7683}
7684
7685static uint32_t
7686mode_to_iconf(uint32_t mode)
7687{
7688
7689	if (mode & T4_FILTER_IC_VNIC)
7690		return (F_VNIC);
7691	return (0);
7692}
7693
7694static int check_fspec_against_fconf_iconf(struct adapter *sc,
7695    struct t4_filter_specification *fs)
7696{
7697	struct tp_params *tpp = &sc->params.tp;
7698	uint32_t fconf = 0;
7699
7700	if (fs->val.frag || fs->mask.frag)
7701		fconf |= F_FRAGMENTATION;
7702
7703	if (fs->val.matchtype || fs->mask.matchtype)
7704		fconf |= F_MPSHITTYPE;
7705
7706	if (fs->val.macidx || fs->mask.macidx)
7707		fconf |= F_MACMATCH;
7708
7709	if (fs->val.ethtype || fs->mask.ethtype)
7710		fconf |= F_ETHERTYPE;
7711
7712	if (fs->val.proto || fs->mask.proto)
7713		fconf |= F_PROTOCOL;
7714
7715	if (fs->val.tos || fs->mask.tos)
7716		fconf |= F_TOS;
7717
7718	if (fs->val.vlan_vld || fs->mask.vlan_vld)
7719		fconf |= F_VLAN;
7720
7721	if (fs->val.ovlan_vld || fs->mask.ovlan_vld) {
7722		fconf |= F_VNIC_ID;
7723		if (tpp->ingress_config & F_VNIC)
7724			return (EINVAL);
7725	}
7726
7727	if (fs->val.pfvf_vld || fs->mask.pfvf_vld) {
7728		fconf |= F_VNIC_ID;
7729		if ((tpp->ingress_config & F_VNIC) == 0)
7730			return (EINVAL);
7731	}
7732
7733	if (fs->val.iport || fs->mask.iport)
7734		fconf |= F_PORT;
7735
7736	if (fs->val.fcoe || fs->mask.fcoe)
7737		fconf |= F_FCOE;
7738
7739	if ((tpp->vlan_pri_map | fconf) != tpp->vlan_pri_map)
7740		return (E2BIG);
7741
7742	return (0);
7743}
7744
7745static int
7746get_filter_mode(struct adapter *sc, uint32_t *mode)
7747{
7748	struct tp_params *tpp = &sc->params.tp;
7749
7750	/*
7751	 * We trust the cached values of the relevant TP registers.  This means
7752	 * things work reliably only if writes to those registers are always via
7753	 * t4_set_filter_mode.
7754	 */
7755	*mode = fconf_iconf_to_mode(tpp->vlan_pri_map, tpp->ingress_config);
7756
7757	return (0);
7758}
7759
7760static int
7761set_filter_mode(struct adapter *sc, uint32_t mode)
7762{
7763	struct tp_params *tpp = &sc->params.tp;
7764	uint32_t fconf, iconf;
7765	int rc;
7766
7767	iconf = mode_to_iconf(mode);
7768	if ((iconf ^ tpp->ingress_config) & F_VNIC) {
7769		/*
7770		 * For now we just complain if A_TP_INGRESS_CONFIG is not
7771		 * already set to the correct value for the requested filter
7772		 * mode.  It's not clear if it's safe to write to this register
7773		 * on the fly.  (And we trust the cached value of the register).
7774		 */
7775		return (EBUSY);
7776	}
7777
7778	fconf = mode_to_fconf(mode);
7779
7780	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7781	    "t4setfm");
7782	if (rc)
7783		return (rc);
7784
7785	if (sc->tids.ftids_in_use > 0) {
7786		rc = EBUSY;
7787		goto done;
7788	}
7789
7790#ifdef TCP_OFFLOAD
7791	if (uld_active(sc, ULD_TOM)) {
7792		rc = EBUSY;
7793		goto done;
7794	}
7795#endif
7796
7797	rc = -t4_set_filter_mode(sc, fconf);
7798done:
7799	end_synchronized_op(sc, LOCK_HELD);
7800	return (rc);
7801}
7802
7803static inline uint64_t
7804get_filter_hits(struct adapter *sc, uint32_t fid)
7805{
7806	uint32_t tcb_addr;
7807
7808	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE) +
7809	    (fid + sc->tids.ftid_base) * TCB_SIZE;
7810
7811	if (is_t4(sc)) {
7812		uint64_t hits;
7813
7814		read_via_memwin(sc, 0, tcb_addr + 16, (uint32_t *)&hits, 8);
7815		return (be64toh(hits));
7816	} else {
7817		uint32_t hits;
7818
7819		read_via_memwin(sc, 0, tcb_addr + 24, &hits, 4);
7820		return (be32toh(hits));
7821	}
7822}
7823
7824static int
7825get_filter(struct adapter *sc, struct t4_filter *t)
7826{
7827	int i, rc, nfilters = sc->tids.nftids;
7828	struct filter_entry *f;
7829
7830	rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
7831	    "t4getf");
7832	if (rc)
7833		return (rc);
7834
7835	if (sc->tids.ftids_in_use == 0 || sc->tids.ftid_tab == NULL ||
7836	    t->idx >= nfilters) {
7837		t->idx = 0xffffffff;
7838		goto done;
7839	}
7840
7841	f = &sc->tids.ftid_tab[t->idx];
7842	for (i = t->idx; i < nfilters; i++, f++) {
7843		if (f->valid) {
7844			t->idx = i;
7845			t->l2tidx = f->l2t ? f->l2t->idx : 0;
7846			t->smtidx = f->smtidx;
7847			if (f->fs.hitcnts)
7848				t->hits = get_filter_hits(sc, t->idx);
7849			else
7850				t->hits = UINT64_MAX;
7851			t->fs = f->fs;
7852
7853			goto done;
7854		}
7855	}
7856
7857	t->idx = 0xffffffff;
7858done:
7859	end_synchronized_op(sc, LOCK_HELD);
7860	return (0);
7861}
7862
7863static int
7864set_filter(struct adapter *sc, struct t4_filter *t)
7865{
7866	unsigned int nfilters, nports;
7867	struct filter_entry *f;
7868	int i, rc;
7869
7870	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setf");
7871	if (rc)
7872		return (rc);
7873
7874	nfilters = sc->tids.nftids;
7875	nports = sc->params.nports;
7876
7877	if (nfilters == 0) {
7878		rc = ENOTSUP;
7879		goto done;
7880	}
7881
7882	if (!(sc->flags & FULL_INIT_DONE)) {
7883		rc = EAGAIN;
7884		goto done;
7885	}
7886
7887	if (t->idx >= nfilters) {
7888		rc = EINVAL;
7889		goto done;
7890	}
7891
7892	/* Validate against the global filter mode and ingress config */
7893	rc = check_fspec_against_fconf_iconf(sc, &t->fs);
7894	if (rc != 0)
7895		goto done;
7896
7897	if (t->fs.action == FILTER_SWITCH && t->fs.eport >= nports) {
7898		rc = EINVAL;
7899		goto done;
7900	}
7901
7902	if (t->fs.val.iport >= nports) {
7903		rc = EINVAL;
7904		goto done;
7905	}
7906
7907	/* Can't specify an iq if not steering to it */
7908	if (!t->fs.dirsteer && t->fs.iq) {
7909		rc = EINVAL;
7910		goto done;
7911	}
7912
7913	/* IPv6 filter idx must be 4 aligned */
7914	if (t->fs.type == 1 &&
7915	    ((t->idx & 0x3) || t->idx + 4 >= nfilters)) {
7916		rc = EINVAL;
7917		goto done;
7918	}
7919
7920	if (sc->tids.ftid_tab == NULL) {
7921		KASSERT(sc->tids.ftids_in_use == 0,
7922		    ("%s: no memory allocated but filters_in_use > 0",
7923		    __func__));
7924
7925		sc->tids.ftid_tab = malloc(sizeof (struct filter_entry) *
7926		    nfilters, M_CXGBE, M_NOWAIT | M_ZERO);
7927		if (sc->tids.ftid_tab == NULL) {
7928			rc = ENOMEM;
7929			goto done;
7930		}
7931		mtx_init(&sc->tids.ftid_lock, "T4 filters", 0, MTX_DEF);
7932	}
7933
7934	for (i = 0; i < 4; i++) {
7935		f = &sc->tids.ftid_tab[t->idx + i];
7936
7937		if (f->pending || f->valid) {
7938			rc = EBUSY;
7939			goto done;
7940		}
7941		if (f->locked) {
7942			rc = EPERM;
7943			goto done;
7944		}
7945
7946		if (t->fs.type == 0)
7947			break;
7948	}
7949
7950	f = &sc->tids.ftid_tab[t->idx];
7951	f->fs = t->fs;
7952
7953	rc = set_filter_wr(sc, t->idx);
7954done:
7955	end_synchronized_op(sc, 0);
7956
7957	if (rc == 0) {
7958		mtx_lock(&sc->tids.ftid_lock);
7959		for (;;) {
7960			if (f->pending == 0) {
7961				rc = f->valid ? 0 : EIO;
7962				break;
7963			}
7964
7965			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
7966			    PCATCH, "t4setfw", 0)) {
7967				rc = EINPROGRESS;
7968				break;
7969			}
7970		}
7971		mtx_unlock(&sc->tids.ftid_lock);
7972	}
7973	return (rc);
7974}
7975
7976static int
7977del_filter(struct adapter *sc, struct t4_filter *t)
7978{
7979	unsigned int nfilters;
7980	struct filter_entry *f;
7981	int rc;
7982
7983	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4delf");
7984	if (rc)
7985		return (rc);
7986
7987	nfilters = sc->tids.nftids;
7988
7989	if (nfilters == 0) {
7990		rc = ENOTSUP;
7991		goto done;
7992	}
7993
7994	if (sc->tids.ftid_tab == NULL || sc->tids.ftids_in_use == 0 ||
7995	    t->idx >= nfilters) {
7996		rc = EINVAL;
7997		goto done;
7998	}
7999
8000	if (!(sc->flags & FULL_INIT_DONE)) {
8001		rc = EAGAIN;
8002		goto done;
8003	}
8004
8005	f = &sc->tids.ftid_tab[t->idx];
8006
8007	if (f->pending) {
8008		rc = EBUSY;
8009		goto done;
8010	}
8011	if (f->locked) {
8012		rc = EPERM;
8013		goto done;
8014	}
8015
8016	if (f->valid) {
8017		t->fs = f->fs;	/* extra info for the caller */
8018		rc = del_filter_wr(sc, t->idx);
8019	}
8020
8021done:
8022	end_synchronized_op(sc, 0);
8023
8024	if (rc == 0) {
8025		mtx_lock(&sc->tids.ftid_lock);
8026		for (;;) {
8027			if (f->pending == 0) {
8028				rc = f->valid ? EIO : 0;
8029				break;
8030			}
8031
8032			if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
8033			    PCATCH, "t4delfw", 0)) {
8034				rc = EINPROGRESS;
8035				break;
8036			}
8037		}
8038		mtx_unlock(&sc->tids.ftid_lock);
8039	}
8040
8041	return (rc);
8042}
8043
8044static void
8045clear_filter(struct filter_entry *f)
8046{
8047	if (f->l2t)
8048		t4_l2t_release(f->l2t);
8049
8050	bzero(f, sizeof (*f));
8051}
8052
8053static int
8054set_filter_wr(struct adapter *sc, int fidx)
8055{
8056	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8057	struct fw_filter_wr *fwr;
8058	unsigned int ftid, vnic_vld, vnic_vld_mask;
8059	struct wrq_cookie cookie;
8060
8061	ASSERT_SYNCHRONIZED_OP(sc);
8062
8063	if (f->fs.newdmac || f->fs.newvlan) {
8064		/* This filter needs an L2T entry; allocate one. */
8065		f->l2t = t4_l2t_alloc_switching(sc->l2t);
8066		if (f->l2t == NULL)
8067			return (EAGAIN);
8068		if (t4_l2t_set_switching(sc, f->l2t, f->fs.vlan, f->fs.eport,
8069		    f->fs.dmac)) {
8070			t4_l2t_release(f->l2t);
8071			f->l2t = NULL;
8072			return (ENOMEM);
8073		}
8074	}
8075
8076	/* Already validated against fconf, iconf */
8077	MPASS((f->fs.val.pfvf_vld & f->fs.val.ovlan_vld) == 0);
8078	MPASS((f->fs.mask.pfvf_vld & f->fs.mask.ovlan_vld) == 0);
8079	if (f->fs.val.pfvf_vld || f->fs.val.ovlan_vld)
8080		vnic_vld = 1;
8081	else
8082		vnic_vld = 0;
8083	if (f->fs.mask.pfvf_vld || f->fs.mask.ovlan_vld)
8084		vnic_vld_mask = 1;
8085	else
8086		vnic_vld_mask = 0;
8087
8088	ftid = sc->tids.ftid_base + fidx;
8089
8090	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8091	if (fwr == NULL)
8092		return (ENOMEM);
8093	bzero(fwr, sizeof(*fwr));
8094
8095	fwr->op_pkd = htobe32(V_FW_WR_OP(FW_FILTER_WR));
8096	fwr->len16_pkd = htobe32(FW_LEN16(*fwr));
8097	fwr->tid_to_iq =
8098	    htobe32(V_FW_FILTER_WR_TID(ftid) |
8099		V_FW_FILTER_WR_RQTYPE(f->fs.type) |
8100		V_FW_FILTER_WR_NOREPLY(0) |
8101		V_FW_FILTER_WR_IQ(f->fs.iq));
8102	fwr->del_filter_to_l2tix =
8103	    htobe32(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
8104		V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
8105		V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
8106		V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
8107		V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
8108		V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
8109		V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
8110		V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
8111		V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
8112		    f->fs.newvlan == VLAN_REWRITE) |
8113		V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
8114		    f->fs.newvlan == VLAN_REWRITE) |
8115		V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
8116		V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
8117		V_FW_FILTER_WR_PRIO(f->fs.prio) |
8118		V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
8119	fwr->ethtype = htobe16(f->fs.val.ethtype);
8120	fwr->ethtypem = htobe16(f->fs.mask.ethtype);
8121	fwr->frag_to_ovlan_vldm =
8122	    (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
8123		V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
8124		V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.vlan_vld) |
8125		V_FW_FILTER_WR_OVLAN_VLD(vnic_vld) |
8126		V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.vlan_vld) |
8127		V_FW_FILTER_WR_OVLAN_VLDM(vnic_vld_mask));
8128	fwr->smac_sel = 0;
8129	fwr->rx_chan_rx_rpl_iq = htobe16(V_FW_FILTER_WR_RX_CHAN(0) |
8130	    V_FW_FILTER_WR_RX_RPL_IQ(sc->sge.fwq.abs_id));
8131	fwr->maci_to_matchtypem =
8132	    htobe32(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
8133		V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
8134		V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
8135		V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
8136		V_FW_FILTER_WR_PORT(f->fs.val.iport) |
8137		V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
8138		V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
8139		V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
8140	fwr->ptcl = f->fs.val.proto;
8141	fwr->ptclm = f->fs.mask.proto;
8142	fwr->ttyp = f->fs.val.tos;
8143	fwr->ttypm = f->fs.mask.tos;
8144	fwr->ivlan = htobe16(f->fs.val.vlan);
8145	fwr->ivlanm = htobe16(f->fs.mask.vlan);
8146	fwr->ovlan = htobe16(f->fs.val.vnic);
8147	fwr->ovlanm = htobe16(f->fs.mask.vnic);
8148	bcopy(f->fs.val.dip, fwr->lip, sizeof (fwr->lip));
8149	bcopy(f->fs.mask.dip, fwr->lipm, sizeof (fwr->lipm));
8150	bcopy(f->fs.val.sip, fwr->fip, sizeof (fwr->fip));
8151	bcopy(f->fs.mask.sip, fwr->fipm, sizeof (fwr->fipm));
8152	fwr->lp = htobe16(f->fs.val.dport);
8153	fwr->lpm = htobe16(f->fs.mask.dport);
8154	fwr->fp = htobe16(f->fs.val.sport);
8155	fwr->fpm = htobe16(f->fs.mask.sport);
8156	if (f->fs.newsmac)
8157		bcopy(f->fs.smac, fwr->sma, sizeof (fwr->sma));
8158
8159	f->pending = 1;
8160	sc->tids.ftids_in_use++;
8161
8162	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8163	return (0);
8164}
8165
8166static int
8167del_filter_wr(struct adapter *sc, int fidx)
8168{
8169	struct filter_entry *f = &sc->tids.ftid_tab[fidx];
8170	struct fw_filter_wr *fwr;
8171	unsigned int ftid;
8172	struct wrq_cookie cookie;
8173
8174	ftid = sc->tids.ftid_base + fidx;
8175
8176	fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
8177	if (fwr == NULL)
8178		return (ENOMEM);
8179	bzero(fwr, sizeof (*fwr));
8180
8181	t4_mk_filtdelwr(ftid, fwr, sc->sge.fwq.abs_id);
8182
8183	f->pending = 1;
8184	commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
8185	return (0);
8186}
8187
8188int
8189t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
8190{
8191	struct adapter *sc = iq->adapter;
8192	const struct cpl_set_tcb_rpl *rpl = (const void *)(rss + 1);
8193	unsigned int idx = GET_TID(rpl);
8194	unsigned int rc;
8195	struct filter_entry *f;
8196
8197	KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
8198	    rss->opcode));
8199
8200	if (is_ftid(sc, idx)) {
8201
8202		idx -= sc->tids.ftid_base;
8203		f = &sc->tids.ftid_tab[idx];
8204		rc = G_COOKIE(rpl->cookie);
8205
8206		mtx_lock(&sc->tids.ftid_lock);
8207		if (rc == FW_FILTER_WR_FLT_ADDED) {
8208			KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
8209			    __func__, idx));
8210			f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
8211			f->pending = 0;  /* asynchronous setup completed */
8212			f->valid = 1;
8213		} else {
8214			if (rc != FW_FILTER_WR_FLT_DELETED) {
8215				/* Add or delete failed, display an error */
8216				log(LOG_ERR,
8217				    "filter %u setup failed with error %u\n",
8218				    idx, rc);
8219			}
8220
8221			clear_filter(f);
8222			sc->tids.ftids_in_use--;
8223		}
8224		wakeup(&sc->tids.ftid_tab);
8225		mtx_unlock(&sc->tids.ftid_lock);
8226	}
8227
8228	return (0);
8229}
8230
8231static int
8232get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
8233{
8234	int rc;
8235
8236	if (cntxt->cid > M_CTXTQID)
8237		return (EINVAL);
8238
8239	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
8240	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
8241		return (EINVAL);
8242
8243	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
8244	if (rc)
8245		return (rc);
8246
8247	if (sc->flags & FW_OK) {
8248		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
8249		    &cntxt->data[0]);
8250		if (rc == 0)
8251			goto done;
8252	}
8253
8254	/*
8255	 * Read via firmware failed or wasn't even attempted.  Read directly via
8256	 * the backdoor.
8257	 */
8258	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
8259done:
8260	end_synchronized_op(sc, 0);
8261	return (rc);
8262}
8263
8264static int
8265load_fw(struct adapter *sc, struct t4_data *fw)
8266{
8267	int rc;
8268	uint8_t *fw_data;
8269
8270	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
8271	if (rc)
8272		return (rc);
8273
8274	if (sc->flags & FULL_INIT_DONE) {
8275		rc = EBUSY;
8276		goto done;
8277	}
8278
8279	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
8280	if (fw_data == NULL) {
8281		rc = ENOMEM;
8282		goto done;
8283	}
8284
8285	rc = copyin(fw->data, fw_data, fw->len);
8286	if (rc == 0)
8287		rc = -t4_load_fw(sc, fw_data, fw->len);
8288
8289	free(fw_data, M_CXGBE);
8290done:
8291	end_synchronized_op(sc, 0);
8292	return (rc);
8293}
8294
8295#define MAX_READ_BUF_SIZE (128 * 1024)
8296static int
8297read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
8298{
8299	uint32_t addr, remaining, n;
8300	uint32_t *buf;
8301	int rc;
8302	uint8_t *dst;
8303
8304	rc = validate_mem_range(sc, mr->addr, mr->len);
8305	if (rc != 0)
8306		return (rc);
8307
8308	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
8309	addr = mr->addr;
8310	remaining = mr->len;
8311	dst = (void *)mr->data;
8312
8313	while (remaining) {
8314		n = min(remaining, MAX_READ_BUF_SIZE);
8315		read_via_memwin(sc, 2, addr, buf, n);
8316
8317		rc = copyout(buf, dst, n);
8318		if (rc != 0)
8319			break;
8320
8321		dst += n;
8322		remaining -= n;
8323		addr += n;
8324	}
8325
8326	free(buf, M_CXGBE);
8327	return (rc);
8328}
8329#undef MAX_READ_BUF_SIZE
8330
8331static int
8332read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
8333{
8334	int rc;
8335
8336	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
8337		return (EINVAL);
8338
8339	if (i2cd->len > sizeof(i2cd->data))
8340		return (EFBIG);
8341
8342	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
8343	if (rc)
8344		return (rc);
8345	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
8346	    i2cd->offset, i2cd->len, &i2cd->data[0]);
8347	end_synchronized_op(sc, 0);
8348
8349	return (rc);
8350}
8351
8352static int
8353in_range(int val, int lo, int hi)
8354{
8355
8356	return (val < 0 || (val <= hi && val >= lo));
8357}
8358
8359static int
8360set_sched_class_config(struct adapter *sc, int minmax)
8361{
8362	int rc;
8363
8364	if (minmax < 0)
8365		return (EINVAL);
8366
8367	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4sscc");
8368	if (rc)
8369		return (rc);
8370	rc = -t4_sched_config(sc, FW_SCHED_TYPE_PKTSCHED, minmax, 1);
8371	end_synchronized_op(sc, 0);
8372
8373	return (rc);
8374}
8375
8376static int
8377set_sched_class_params(struct adapter *sc, struct t4_sched_class_params *p,
8378    int sleep_ok)
8379{
8380	int rc, top_speed, fw_level, fw_mode, fw_rateunit, fw_ratemode;
8381	struct port_info *pi;
8382	struct tx_sched_class *tc;
8383
8384	if (p->level == SCHED_CLASS_LEVEL_CL_RL)
8385		fw_level = FW_SCHED_PARAMS_LEVEL_CL_RL;
8386	else if (p->level == SCHED_CLASS_LEVEL_CL_WRR)
8387		fw_level = FW_SCHED_PARAMS_LEVEL_CL_WRR;
8388	else if (p->level == SCHED_CLASS_LEVEL_CH_RL)
8389		fw_level = FW_SCHED_PARAMS_LEVEL_CH_RL;
8390	else
8391		return (EINVAL);
8392
8393	if (p->mode == SCHED_CLASS_MODE_CLASS)
8394		fw_mode = FW_SCHED_PARAMS_MODE_CLASS;
8395	else if (p->mode == SCHED_CLASS_MODE_FLOW)
8396		fw_mode = FW_SCHED_PARAMS_MODE_FLOW;
8397	else
8398		return (EINVAL);
8399
8400	if (p->rateunit == SCHED_CLASS_RATEUNIT_BITS)
8401		fw_rateunit = FW_SCHED_PARAMS_UNIT_BITRATE;
8402	else if (p->rateunit == SCHED_CLASS_RATEUNIT_PKTS)
8403		fw_rateunit = FW_SCHED_PARAMS_UNIT_PKTRATE;
8404	else
8405		return (EINVAL);
8406
8407	if (p->ratemode == SCHED_CLASS_RATEMODE_REL)
8408		fw_ratemode = FW_SCHED_PARAMS_RATE_REL;
8409	else if (p->ratemode == SCHED_CLASS_RATEMODE_ABS)
8410		fw_ratemode = FW_SCHED_PARAMS_RATE_ABS;
8411	else
8412		return (EINVAL);
8413
8414	/* Vet our parameters ... */
8415	if (!in_range(p->channel, 0, sc->chip_params->nchan - 1))
8416		return (ERANGE);
8417
8418	pi = sc->port[sc->chan_map[p->channel]];
8419	if (pi == NULL)
8420		return (ENXIO);
8421	MPASS(pi->tx_chan == p->channel);
8422	top_speed = port_top_speed(pi) * 1000000; /* Gbps -> Kbps */
8423
8424	if (!in_range(p->cl, 0, sc->chip_params->nsched_cls) ||
8425	    !in_range(p->minrate, 0, top_speed) ||
8426	    !in_range(p->maxrate, 0, top_speed) ||
8427	    !in_range(p->weight, 0, 100))
8428		return (ERANGE);
8429
8430	/*
8431	 * Translate any unset parameters into the firmware's
8432	 * nomenclature and/or fail the call if the parameters
8433	 * are required ...
8434	 */
8435	if (p->rateunit < 0 || p->ratemode < 0 || p->channel < 0 || p->cl < 0)
8436		return (EINVAL);
8437
8438	if (p->minrate < 0)
8439		p->minrate = 0;
8440	if (p->maxrate < 0) {
8441		if (p->level == SCHED_CLASS_LEVEL_CL_RL ||
8442		    p->level == SCHED_CLASS_LEVEL_CH_RL)
8443			return (EINVAL);
8444		else
8445			p->maxrate = 0;
8446	}
8447	if (p->weight < 0) {
8448		if (p->level == SCHED_CLASS_LEVEL_CL_WRR)
8449			return (EINVAL);
8450		else
8451			p->weight = 0;
8452	}
8453	if (p->pktsize < 0) {
8454		if (p->level == SCHED_CLASS_LEVEL_CL_RL ||
8455		    p->level == SCHED_CLASS_LEVEL_CH_RL)
8456			return (EINVAL);
8457		else
8458			p->pktsize = 0;
8459	}
8460
8461	rc = begin_synchronized_op(sc, NULL,
8462	    sleep_ok ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4sscp");
8463	if (rc)
8464		return (rc);
8465	tc = &pi->tc[p->cl];
8466	tc->params = *p;
8467	rc = -t4_sched_params(sc, FW_SCHED_TYPE_PKTSCHED, fw_level, fw_mode,
8468	    fw_rateunit, fw_ratemode, p->channel, p->cl, p->minrate, p->maxrate,
8469	    p->weight, p->pktsize, sleep_ok);
8470	if (rc == 0)
8471		tc->flags |= TX_SC_OK;
8472	else {
8473		/*
8474		 * Unknown state at this point, see tc->params for what was
8475		 * attempted.
8476		 */
8477		tc->flags &= ~TX_SC_OK;
8478	}
8479	end_synchronized_op(sc, sleep_ok ? 0 : LOCK_HELD);
8480
8481	return (rc);
8482}
8483
8484static int
8485set_sched_class(struct adapter *sc, struct t4_sched_params *p)
8486{
8487
8488	if (p->type != SCHED_CLASS_TYPE_PACKET)
8489		return (EINVAL);
8490
8491	if (p->subcmd == SCHED_CLASS_SUBCMD_CONFIG)
8492		return (set_sched_class_config(sc, p->u.config.minmax));
8493
8494	if (p->subcmd == SCHED_CLASS_SUBCMD_PARAMS)
8495		return (set_sched_class_params(sc, &p->u.params, 1));
8496
8497	return (EINVAL);
8498}
8499
8500static int
8501set_sched_queue(struct adapter *sc, struct t4_sched_queue *p)
8502{
8503	struct port_info *pi = NULL;
8504	struct vi_info *vi;
8505	struct sge_txq *txq;
8506	uint32_t fw_mnem, fw_queue, fw_class;
8507	int i, rc;
8508
8509	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setsq");
8510	if (rc)
8511		return (rc);
8512
8513	if (p->port >= sc->params.nports) {
8514		rc = EINVAL;
8515		goto done;
8516	}
8517
8518	/* XXX: Only supported for the main VI. */
8519	pi = sc->port[p->port];
8520	vi = &pi->vi[0];
8521	if (!(vi->flags & VI_INIT_DONE)) {
8522		/* tx queues not set up yet */
8523		rc = EAGAIN;
8524		goto done;
8525	}
8526
8527	if (!in_range(p->queue, 0, vi->ntxq - 1) ||
8528	    !in_range(p->cl, 0, sc->chip_params->nsched_cls - 1)) {
8529		rc = EINVAL;
8530		goto done;
8531	}
8532
8533	/*
8534	 * Create a template for the FW_PARAMS_CMD mnemonic and value (TX
8535	 * Scheduling Class in this case).
8536	 */
8537	fw_mnem = (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
8538	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_EQ_SCHEDCLASS_ETH));
8539	fw_class = p->cl < 0 ? 0xffffffff : p->cl;
8540
8541	/*
8542	 * If op.queue is non-negative, then we're only changing the scheduling
8543	 * on a single specified TX queue.
8544	 */
8545	if (p->queue >= 0) {
8546		txq = &sc->sge.txq[vi->first_txq + p->queue];
8547		fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
8548		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
8549		    &fw_class);
8550		goto done;
8551	}
8552
8553	/*
8554	 * Change the scheduling on all the TX queues for the
8555	 * interface.
8556	 */
8557	for_each_txq(vi, i, txq) {
8558		fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
8559		rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
8560		    &fw_class);
8561		if (rc)
8562			goto done;
8563	}
8564
8565	rc = 0;
8566done:
8567	end_synchronized_op(sc, 0);
8568	return (rc);
8569}
8570
8571int
8572t4_os_find_pci_capability(struct adapter *sc, int cap)
8573{
8574	int i;
8575
8576	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
8577}
8578
8579int
8580t4_os_pci_save_state(struct adapter *sc)
8581{
8582	device_t dev;
8583	struct pci_devinfo *dinfo;
8584
8585	dev = sc->dev;
8586	dinfo = device_get_ivars(dev);
8587
8588	pci_cfg_save(dev, dinfo, 0);
8589	return (0);
8590}
8591
8592int
8593t4_os_pci_restore_state(struct adapter *sc)
8594{
8595	device_t dev;
8596	struct pci_devinfo *dinfo;
8597
8598	dev = sc->dev;
8599	dinfo = device_get_ivars(dev);
8600
8601	pci_cfg_restore(dev, dinfo);
8602	return (0);
8603}
8604
8605void
8606t4_os_portmod_changed(const struct adapter *sc, int idx)
8607{
8608	struct port_info *pi = sc->port[idx];
8609	struct vi_info *vi;
8610	struct ifnet *ifp;
8611	int v;
8612	static const char *mod_str[] = {
8613		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
8614	};
8615
8616	for_each_vi(pi, v, vi) {
8617		build_medialist(pi, &vi->media);
8618	}
8619
8620	ifp = pi->vi[0].ifp;
8621	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
8622		if_printf(ifp, "transceiver unplugged.\n");
8623	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
8624		if_printf(ifp, "unknown transceiver inserted.\n");
8625	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
8626		if_printf(ifp, "unsupported transceiver inserted.\n");
8627	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
8628		if_printf(ifp, "%s transceiver inserted.\n",
8629		    mod_str[pi->mod_type]);
8630	} else {
8631		if_printf(ifp, "transceiver (type %d) inserted.\n",
8632		    pi->mod_type);
8633	}
8634}
8635
8636void
8637t4_os_link_changed(struct adapter *sc, int idx, int link_stat, int reason)
8638{
8639	struct port_info *pi = sc->port[idx];
8640	struct vi_info *vi;
8641	struct ifnet *ifp;
8642	int v;
8643
8644	if (link_stat)
8645		pi->linkdnrc = -1;
8646	else {
8647		if (reason >= 0)
8648			pi->linkdnrc = reason;
8649	}
8650	for_each_vi(pi, v, vi) {
8651		ifp = vi->ifp;
8652		if (ifp == NULL)
8653			continue;
8654
8655		if (link_stat) {
8656			ifp->if_baudrate = IF_Mbps(pi->link_cfg.speed);
8657			if_link_state_change(ifp, LINK_STATE_UP);
8658		} else {
8659			if_link_state_change(ifp, LINK_STATE_DOWN);
8660		}
8661	}
8662}
8663
8664void
8665t4_iterate(void (*func)(struct adapter *, void *), void *arg)
8666{
8667	struct adapter *sc;
8668
8669	sx_slock(&t4_list_lock);
8670	SLIST_FOREACH(sc, &t4_list, link) {
8671		/*
8672		 * func should not make any assumptions about what state sc is
8673		 * in - the only guarantee is that sc->sc_lock is a valid lock.
8674		 */
8675		func(sc, arg);
8676	}
8677	sx_sunlock(&t4_list_lock);
8678}
8679
8680static int
8681t4_open(struct cdev *dev, int flags, int type, struct thread *td)
8682{
8683       return (0);
8684}
8685
8686static int
8687t4_close(struct cdev *dev, int flags, int type, struct thread *td)
8688{
8689       return (0);
8690}
8691
8692static int
8693t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
8694    struct thread *td)
8695{
8696	int rc;
8697	struct adapter *sc = dev->si_drv1;
8698
8699	rc = priv_check(td, PRIV_DRIVER);
8700	if (rc != 0)
8701		return (rc);
8702
8703	switch (cmd) {
8704	case CHELSIO_T4_GETREG: {
8705		struct t4_reg *edata = (struct t4_reg *)data;
8706
8707		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
8708			return (EFAULT);
8709
8710		if (edata->size == 4)
8711			edata->val = t4_read_reg(sc, edata->addr);
8712		else if (edata->size == 8)
8713			edata->val = t4_read_reg64(sc, edata->addr);
8714		else
8715			return (EINVAL);
8716
8717		break;
8718	}
8719	case CHELSIO_T4_SETREG: {
8720		struct t4_reg *edata = (struct t4_reg *)data;
8721
8722		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
8723			return (EFAULT);
8724
8725		if (edata->size == 4) {
8726			if (edata->val & 0xffffffff00000000)
8727				return (EINVAL);
8728			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
8729		} else if (edata->size == 8)
8730			t4_write_reg64(sc, edata->addr, edata->val);
8731		else
8732			return (EINVAL);
8733		break;
8734	}
8735	case CHELSIO_T4_REGDUMP: {
8736		struct t4_regdump *regs = (struct t4_regdump *)data;
8737		int reglen = is_t4(sc) ? T4_REGDUMP_SIZE : T5_REGDUMP_SIZE;
8738		uint8_t *buf;
8739
8740		if (regs->len < reglen) {
8741			regs->len = reglen; /* hint to the caller */
8742			return (ENOBUFS);
8743		}
8744
8745		regs->len = reglen;
8746		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
8747		get_regs(sc, regs, buf);
8748		rc = copyout(buf, regs->data, reglen);
8749		free(buf, M_CXGBE);
8750		break;
8751	}
8752	case CHELSIO_T4_GET_FILTER_MODE:
8753		rc = get_filter_mode(sc, (uint32_t *)data);
8754		break;
8755	case CHELSIO_T4_SET_FILTER_MODE:
8756		rc = set_filter_mode(sc, *(uint32_t *)data);
8757		break;
8758	case CHELSIO_T4_GET_FILTER:
8759		rc = get_filter(sc, (struct t4_filter *)data);
8760		break;
8761	case CHELSIO_T4_SET_FILTER:
8762		rc = set_filter(sc, (struct t4_filter *)data);
8763		break;
8764	case CHELSIO_T4_DEL_FILTER:
8765		rc = del_filter(sc, (struct t4_filter *)data);
8766		break;
8767	case CHELSIO_T4_GET_SGE_CONTEXT:
8768		rc = get_sge_context(sc, (struct t4_sge_context *)data);
8769		break;
8770	case CHELSIO_T4_LOAD_FW:
8771		rc = load_fw(sc, (struct t4_data *)data);
8772		break;
8773	case CHELSIO_T4_GET_MEM:
8774		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
8775		break;
8776	case CHELSIO_T4_GET_I2C:
8777		rc = read_i2c(sc, (struct t4_i2c_data *)data);
8778		break;
8779	case CHELSIO_T4_CLEAR_STATS: {
8780		int i, v;
8781		u_int port_id = *(uint32_t *)data;
8782		struct port_info *pi;
8783		struct vi_info *vi;
8784
8785		if (port_id >= sc->params.nports)
8786			return (EINVAL);
8787		pi = sc->port[port_id];
8788
8789		/* MAC stats */
8790		t4_clr_port_stats(sc, pi->tx_chan);
8791		pi->tx_parse_error = 0;
8792		mtx_lock(&sc->reg_lock);
8793		for_each_vi(pi, v, vi) {
8794			if (vi->flags & VI_INIT_DONE)
8795				t4_clr_vi_stats(sc, vi->viid);
8796		}
8797		mtx_unlock(&sc->reg_lock);
8798
8799		/*
8800		 * Since this command accepts a port, clear stats for
8801		 * all VIs on this port.
8802		 */
8803		for_each_vi(pi, v, vi) {
8804			if (vi->flags & VI_INIT_DONE) {
8805				struct sge_rxq *rxq;
8806				struct sge_txq *txq;
8807				struct sge_wrq *wrq;
8808
8809				for_each_rxq(vi, i, rxq) {
8810#if defined(INET) || defined(INET6)
8811					rxq->lro.lro_queued = 0;
8812					rxq->lro.lro_flushed = 0;
8813#endif
8814					rxq->rxcsum = 0;
8815					rxq->vlan_extraction = 0;
8816				}
8817
8818				for_each_txq(vi, i, txq) {
8819					txq->txcsum = 0;
8820					txq->tso_wrs = 0;
8821					txq->vlan_insertion = 0;
8822					txq->imm_wrs = 0;
8823					txq->sgl_wrs = 0;
8824					txq->txpkt_wrs = 0;
8825					txq->txpkts0_wrs = 0;
8826					txq->txpkts1_wrs = 0;
8827					txq->txpkts0_pkts = 0;
8828					txq->txpkts1_pkts = 0;
8829					mp_ring_reset_stats(txq->r);
8830				}
8831
8832#ifdef TCP_OFFLOAD
8833				/* nothing to clear for each ofld_rxq */
8834
8835				for_each_ofld_txq(vi, i, wrq) {
8836					wrq->tx_wrs_direct = 0;
8837					wrq->tx_wrs_copied = 0;
8838				}
8839#endif
8840
8841				if (IS_MAIN_VI(vi)) {
8842					wrq = &sc->sge.ctrlq[pi->port_id];
8843					wrq->tx_wrs_direct = 0;
8844					wrq->tx_wrs_copied = 0;
8845				}
8846			}
8847		}
8848		break;
8849	}
8850	case CHELSIO_T4_SCHED_CLASS:
8851		rc = set_sched_class(sc, (struct t4_sched_params *)data);
8852		break;
8853	case CHELSIO_T4_SCHED_QUEUE:
8854		rc = set_sched_queue(sc, (struct t4_sched_queue *)data);
8855		break;
8856	case CHELSIO_T4_GET_TRACER:
8857		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
8858		break;
8859	case CHELSIO_T4_SET_TRACER:
8860		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
8861		break;
8862	default:
8863		rc = EINVAL;
8864	}
8865
8866	return (rc);
8867}
8868
8869void
8870t4_db_full(struct adapter *sc)
8871{
8872
8873	CXGBE_UNIMPLEMENTED(__func__);
8874}
8875
8876void
8877t4_db_dropped(struct adapter *sc)
8878{
8879
8880	CXGBE_UNIMPLEMENTED(__func__);
8881}
8882
8883#ifdef TCP_OFFLOAD
8884void
8885t4_iscsi_init(struct ifnet *ifp, unsigned int tag_mask,
8886    const unsigned int *pgsz_order)
8887{
8888	struct vi_info *vi = ifp->if_softc;
8889	struct adapter *sc = vi->pi->adapter;
8890
8891	t4_write_reg(sc, A_ULP_RX_ISCSI_TAGMASK, tag_mask);
8892	t4_write_reg(sc, A_ULP_RX_ISCSI_PSZ, V_HPZ0(pgsz_order[0]) |
8893		V_HPZ1(pgsz_order[1]) | V_HPZ2(pgsz_order[2]) |
8894		V_HPZ3(pgsz_order[3]));
8895}
8896
8897static int
8898toe_capability(struct vi_info *vi, int enable)
8899{
8900	int rc;
8901	struct port_info *pi = vi->pi;
8902	struct adapter *sc = pi->adapter;
8903
8904	ASSERT_SYNCHRONIZED_OP(sc);
8905
8906	if (!is_offload(sc))
8907		return (ENODEV);
8908
8909	if (enable) {
8910		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
8911			/* TOE is already enabled. */
8912			return (0);
8913		}
8914
8915		/*
8916		 * We need the port's queues around so that we're able to send
8917		 * and receive CPLs to/from the TOE even if the ifnet for this
8918		 * port has never been UP'd administratively.
8919		 */
8920		if (!(vi->flags & VI_INIT_DONE)) {
8921			rc = vi_full_init(vi);
8922			if (rc)
8923				return (rc);
8924		}
8925		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
8926			rc = vi_full_init(&pi->vi[0]);
8927			if (rc)
8928				return (rc);
8929		}
8930
8931		if (isset(&sc->offload_map, pi->port_id)) {
8932			/* TOE is enabled on another VI of this port. */
8933			pi->uld_vis++;
8934			return (0);
8935		}
8936
8937		if (!uld_active(sc, ULD_TOM)) {
8938			rc = t4_activate_uld(sc, ULD_TOM);
8939			if (rc == EAGAIN) {
8940				log(LOG_WARNING,
8941				    "You must kldload t4_tom.ko before trying "
8942				    "to enable TOE on a cxgbe interface.\n");
8943			}
8944			if (rc != 0)
8945				return (rc);
8946			KASSERT(sc->tom_softc != NULL,
8947			    ("%s: TOM activated but softc NULL", __func__));
8948			KASSERT(uld_active(sc, ULD_TOM),
8949			    ("%s: TOM activated but flag not set", __func__));
8950		}
8951
8952		/* Activate iWARP and iSCSI too, if the modules are loaded. */
8953		if (!uld_active(sc, ULD_IWARP))
8954			(void) t4_activate_uld(sc, ULD_IWARP);
8955		if (!uld_active(sc, ULD_ISCSI))
8956			(void) t4_activate_uld(sc, ULD_ISCSI);
8957
8958		pi->uld_vis++;
8959		setbit(&sc->offload_map, pi->port_id);
8960	} else {
8961		pi->uld_vis--;
8962
8963		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
8964			return (0);
8965
8966		KASSERT(uld_active(sc, ULD_TOM),
8967		    ("%s: TOM never initialized?", __func__));
8968		clrbit(&sc->offload_map, pi->port_id);
8969	}
8970
8971	return (0);
8972}
8973
8974/*
8975 * Add an upper layer driver to the global list.
8976 */
8977int
8978t4_register_uld(struct uld_info *ui)
8979{
8980	int rc = 0;
8981	struct uld_info *u;
8982
8983	sx_xlock(&t4_uld_list_lock);
8984	SLIST_FOREACH(u, &t4_uld_list, link) {
8985	    if (u->uld_id == ui->uld_id) {
8986		    rc = EEXIST;
8987		    goto done;
8988	    }
8989	}
8990
8991	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
8992	ui->refcount = 0;
8993done:
8994	sx_xunlock(&t4_uld_list_lock);
8995	return (rc);
8996}
8997
8998int
8999t4_unregister_uld(struct uld_info *ui)
9000{
9001	int rc = EINVAL;
9002	struct uld_info *u;
9003
9004	sx_xlock(&t4_uld_list_lock);
9005
9006	SLIST_FOREACH(u, &t4_uld_list, link) {
9007	    if (u == ui) {
9008		    if (ui->refcount > 0) {
9009			    rc = EBUSY;
9010			    goto done;
9011		    }
9012
9013		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
9014		    rc = 0;
9015		    goto done;
9016	    }
9017	}
9018done:
9019	sx_xunlock(&t4_uld_list_lock);
9020	return (rc);
9021}
9022
9023int
9024t4_activate_uld(struct adapter *sc, int id)
9025{
9026	int rc;
9027	struct uld_info *ui;
9028
9029	ASSERT_SYNCHRONIZED_OP(sc);
9030
9031	if (id < 0 || id > ULD_MAX)
9032		return (EINVAL);
9033	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
9034
9035	sx_slock(&t4_uld_list_lock);
9036
9037	SLIST_FOREACH(ui, &t4_uld_list, link) {
9038		if (ui->uld_id == id) {
9039			if (!(sc->flags & FULL_INIT_DONE)) {
9040				rc = adapter_full_init(sc);
9041				if (rc != 0)
9042					break;
9043			}
9044
9045			rc = ui->activate(sc);
9046			if (rc == 0) {
9047				setbit(&sc->active_ulds, id);
9048				ui->refcount++;
9049			}
9050			break;
9051		}
9052	}
9053
9054	sx_sunlock(&t4_uld_list_lock);
9055
9056	return (rc);
9057}
9058
9059int
9060t4_deactivate_uld(struct adapter *sc, int id)
9061{
9062	int rc;
9063	struct uld_info *ui;
9064
9065	ASSERT_SYNCHRONIZED_OP(sc);
9066
9067	if (id < 0 || id > ULD_MAX)
9068		return (EINVAL);
9069	rc = ENXIO;
9070
9071	sx_slock(&t4_uld_list_lock);
9072
9073	SLIST_FOREACH(ui, &t4_uld_list, link) {
9074		if (ui->uld_id == id) {
9075			rc = ui->deactivate(sc);
9076			if (rc == 0) {
9077				clrbit(&sc->active_ulds, id);
9078				ui->refcount--;
9079			}
9080			break;
9081		}
9082	}
9083
9084	sx_sunlock(&t4_uld_list_lock);
9085
9086	return (rc);
9087}
9088
9089int
9090uld_active(struct adapter *sc, int uld_id)
9091{
9092
9093	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
9094
9095	return (isset(&sc->active_ulds, uld_id));
9096}
9097#endif
9098
9099/*
9100 * Come up with reasonable defaults for some of the tunables, provided they're
9101 * not set by the user (in which case we'll use the values as is).
9102 */
9103static void
9104tweak_tunables(void)
9105{
9106	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
9107
9108	if (t4_ntxq10g < 1) {
9109#ifdef RSS
9110		t4_ntxq10g = rss_getnumbuckets();
9111#else
9112		t4_ntxq10g = min(nc, NTXQ_10G);
9113#endif
9114	}
9115
9116	if (t4_ntxq1g < 1) {
9117#ifdef RSS
9118		/* XXX: way too many for 1GbE? */
9119		t4_ntxq1g = rss_getnumbuckets();
9120#else
9121		t4_ntxq1g = min(nc, NTXQ_1G);
9122#endif
9123	}
9124
9125	if (t4_ntxq_vi < 1)
9126		t4_ntxq_vi = min(nc, NTXQ_VI);
9127
9128	if (t4_nrxq10g < 1) {
9129#ifdef RSS
9130		t4_nrxq10g = rss_getnumbuckets();
9131#else
9132		t4_nrxq10g = min(nc, NRXQ_10G);
9133#endif
9134	}
9135
9136	if (t4_nrxq1g < 1) {
9137#ifdef RSS
9138		/* XXX: way too many for 1GbE? */
9139		t4_nrxq1g = rss_getnumbuckets();
9140#else
9141		t4_nrxq1g = min(nc, NRXQ_1G);
9142#endif
9143	}
9144
9145	if (t4_nrxq_vi < 1)
9146		t4_nrxq_vi = min(nc, NRXQ_VI);
9147
9148#ifdef TCP_OFFLOAD
9149	if (t4_nofldtxq10g < 1)
9150		t4_nofldtxq10g = min(nc, NOFLDTXQ_10G);
9151
9152	if (t4_nofldtxq1g < 1)
9153		t4_nofldtxq1g = min(nc, NOFLDTXQ_1G);
9154
9155	if (t4_nofldtxq_vi < 1)
9156		t4_nofldtxq_vi = min(nc, NOFLDTXQ_VI);
9157
9158	if (t4_nofldrxq10g < 1)
9159		t4_nofldrxq10g = min(nc, NOFLDRXQ_10G);
9160
9161	if (t4_nofldrxq1g < 1)
9162		t4_nofldrxq1g = min(nc, NOFLDRXQ_1G);
9163
9164	if (t4_nofldrxq_vi < 1)
9165		t4_nofldrxq_vi = min(nc, NOFLDRXQ_VI);
9166
9167	if (t4_toecaps_allowed == -1)
9168		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
9169
9170	if (t4_rdmacaps_allowed == -1) {
9171		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
9172		    FW_CAPS_CONFIG_RDMA_RDMAC;
9173	}
9174
9175	if (t4_iscsicaps_allowed == -1) {
9176		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
9177		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
9178		    FW_CAPS_CONFIG_ISCSI_T10DIF;
9179	}
9180#else
9181	if (t4_toecaps_allowed == -1)
9182		t4_toecaps_allowed = 0;
9183
9184	if (t4_rdmacaps_allowed == -1)
9185		t4_rdmacaps_allowed = 0;
9186
9187	if (t4_iscsicaps_allowed == -1)
9188		t4_iscsicaps_allowed = 0;
9189#endif
9190
9191#ifdef DEV_NETMAP
9192	if (t4_nnmtxq_vi < 1)
9193		t4_nnmtxq_vi = min(nc, NNMTXQ_VI);
9194
9195	if (t4_nnmrxq_vi < 1)
9196		t4_nnmrxq_vi = min(nc, NNMRXQ_VI);
9197#endif
9198
9199	if (t4_tmr_idx_10g < 0 || t4_tmr_idx_10g >= SGE_NTIMERS)
9200		t4_tmr_idx_10g = TMR_IDX_10G;
9201
9202	if (t4_pktc_idx_10g < -1 || t4_pktc_idx_10g >= SGE_NCOUNTERS)
9203		t4_pktc_idx_10g = PKTC_IDX_10G;
9204
9205	if (t4_tmr_idx_1g < 0 || t4_tmr_idx_1g >= SGE_NTIMERS)
9206		t4_tmr_idx_1g = TMR_IDX_1G;
9207
9208	if (t4_pktc_idx_1g < -1 || t4_pktc_idx_1g >= SGE_NCOUNTERS)
9209		t4_pktc_idx_1g = PKTC_IDX_1G;
9210
9211	if (t4_qsize_txq < 128)
9212		t4_qsize_txq = 128;
9213
9214	if (t4_qsize_rxq < 128)
9215		t4_qsize_rxq = 128;
9216	while (t4_qsize_rxq & 7)
9217		t4_qsize_rxq++;
9218
9219	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
9220}
9221
9222#ifdef DDB
9223static void
9224t4_dump_tcb(struct adapter *sc, int tid)
9225{
9226	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
9227
9228	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
9229	save = t4_read_reg(sc, reg);
9230	base = sc->memwin[2].mw_base;
9231
9232	/* Dump TCB for the tid */
9233	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
9234	tcb_addr += tid * TCB_SIZE;
9235
9236	if (is_t4(sc)) {
9237		pf = 0;
9238		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
9239	} else {
9240		pf = V_PFNUM(sc->pf);
9241		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
9242	}
9243	t4_write_reg(sc, reg, win_pos | pf);
9244	t4_read_reg(sc, reg);
9245
9246	off = tcb_addr - win_pos;
9247	for (i = 0; i < 4; i++) {
9248		uint32_t buf[8];
9249		for (j = 0; j < 8; j++, off += 4)
9250			buf[j] = htonl(t4_read_reg(sc, base + off));
9251
9252		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
9253		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
9254		    buf[7]);
9255	}
9256
9257	t4_write_reg(sc, reg, save);
9258	t4_read_reg(sc, reg);
9259}
9260
9261static void
9262t4_dump_devlog(struct adapter *sc)
9263{
9264	struct devlog_params *dparams = &sc->params.devlog;
9265	struct fw_devlog_e e;
9266	int i, first, j, m, nentries, rc;
9267	uint64_t ftstamp = UINT64_MAX;
9268
9269	if (dparams->start == 0) {
9270		db_printf("devlog params not valid\n");
9271		return;
9272	}
9273
9274	nentries = dparams->size / sizeof(struct fw_devlog_e);
9275	m = fwmtype_to_hwmtype(dparams->memtype);
9276
9277	/* Find the first entry. */
9278	first = -1;
9279	for (i = 0; i < nentries && !db_pager_quit; i++) {
9280		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9281		    sizeof(e), (void *)&e);
9282		if (rc != 0)
9283			break;
9284
9285		if (e.timestamp == 0)
9286			break;
9287
9288		e.timestamp = be64toh(e.timestamp);
9289		if (e.timestamp < ftstamp) {
9290			ftstamp = e.timestamp;
9291			first = i;
9292		}
9293	}
9294
9295	if (first == -1)
9296		return;
9297
9298	i = first;
9299	do {
9300		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
9301		    sizeof(e), (void *)&e);
9302		if (rc != 0)
9303			return;
9304
9305		if (e.timestamp == 0)
9306			return;
9307
9308		e.timestamp = be64toh(e.timestamp);
9309		e.seqno = be32toh(e.seqno);
9310		for (j = 0; j < 8; j++)
9311			e.params[j] = be32toh(e.params[j]);
9312
9313		db_printf("%10d  %15ju  %8s  %8s  ",
9314		    e.seqno, e.timestamp,
9315		    (e.level < nitems(devlog_level_strings) ?
9316			devlog_level_strings[e.level] : "UNKNOWN"),
9317		    (e.facility < nitems(devlog_facility_strings) ?
9318			devlog_facility_strings[e.facility] : "UNKNOWN"));
9319		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
9320		    e.params[3], e.params[4], e.params[5], e.params[6],
9321		    e.params[7]);
9322
9323		if (++i == nentries)
9324			i = 0;
9325	} while (i != first && !db_pager_quit);
9326}
9327
9328static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
9329_DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
9330
9331DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
9332{
9333	device_t dev;
9334	int t;
9335	bool valid;
9336
9337	valid = false;
9338	t = db_read_token();
9339	if (t == tIDENT) {
9340		dev = device_lookup_by_name(db_tok_string);
9341		valid = true;
9342	}
9343	db_skip_to_eol();
9344	if (!valid) {
9345		db_printf("usage: show t4 devlog <nexus>\n");
9346		return;
9347	}
9348
9349	if (dev == NULL) {
9350		db_printf("device not found\n");
9351		return;
9352	}
9353
9354	t4_dump_devlog(device_get_softc(dev));
9355}
9356
9357DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
9358{
9359	device_t dev;
9360	int radix, tid, t;
9361	bool valid;
9362
9363	valid = false;
9364	radix = db_radix;
9365	db_radix = 10;
9366	t = db_read_token();
9367	if (t == tIDENT) {
9368		dev = device_lookup_by_name(db_tok_string);
9369		t = db_read_token();
9370		if (t == tNUMBER) {
9371			tid = db_tok_number;
9372			valid = true;
9373		}
9374	}
9375	db_radix = radix;
9376	db_skip_to_eol();
9377	if (!valid) {
9378		db_printf("usage: show t4 tcb <nexus> <tid>\n");
9379		return;
9380	}
9381
9382	if (dev == NULL) {
9383		db_printf("device not found\n");
9384		return;
9385	}
9386	if (tid < 0) {
9387		db_printf("invalid tid\n");
9388		return;
9389	}
9390
9391	t4_dump_tcb(device_get_softc(dev), tid);
9392}
9393#endif
9394
9395static struct sx mlu;	/* mod load unload */
9396SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
9397
9398static int
9399mod_event(module_t mod, int cmd, void *arg)
9400{
9401	int rc = 0;
9402	static int loaded = 0;
9403
9404	switch (cmd) {
9405	case MOD_LOAD:
9406		sx_xlock(&mlu);
9407		if (loaded++ == 0) {
9408			t4_sge_modload();
9409			sx_init(&t4_list_lock, "T4/T5 adapters");
9410			SLIST_INIT(&t4_list);
9411#ifdef TCP_OFFLOAD
9412			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
9413			SLIST_INIT(&t4_uld_list);
9414#endif
9415			t4_tracer_modload();
9416			tweak_tunables();
9417		}
9418		sx_xunlock(&mlu);
9419		break;
9420
9421	case MOD_UNLOAD:
9422		sx_xlock(&mlu);
9423		if (--loaded == 0) {
9424			int tries;
9425
9426			sx_slock(&t4_list_lock);
9427			if (!SLIST_EMPTY(&t4_list)) {
9428				rc = EBUSY;
9429				sx_sunlock(&t4_list_lock);
9430				goto done_unload;
9431			}
9432#ifdef TCP_OFFLOAD
9433			sx_slock(&t4_uld_list_lock);
9434			if (!SLIST_EMPTY(&t4_uld_list)) {
9435				rc = EBUSY;
9436				sx_sunlock(&t4_uld_list_lock);
9437				sx_sunlock(&t4_list_lock);
9438				goto done_unload;
9439			}
9440#endif
9441			tries = 0;
9442			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
9443				uprintf("%ju clusters with custom free routine "
9444				    "still is use.\n", t4_sge_extfree_refs());
9445				pause("t4unload", 2 * hz);
9446			}
9447#ifdef TCP_OFFLOAD
9448			sx_sunlock(&t4_uld_list_lock);
9449#endif
9450			sx_sunlock(&t4_list_lock);
9451
9452			if (t4_sge_extfree_refs() == 0) {
9453				t4_tracer_modunload();
9454#ifdef TCP_OFFLOAD
9455				sx_destroy(&t4_uld_list_lock);
9456#endif
9457				sx_destroy(&t4_list_lock);
9458				t4_sge_modunload();
9459				loaded = 0;
9460			} else {
9461				rc = EBUSY;
9462				loaded++;	/* undo earlier decrement */
9463			}
9464		}
9465done_unload:
9466		sx_xunlock(&mlu);
9467		break;
9468	}
9469
9470	return (rc);
9471}
9472
9473static devclass_t t4_devclass, t5_devclass;
9474static devclass_t cxgbe_devclass, cxl_devclass;
9475static devclass_t vcxgbe_devclass, vcxl_devclass;
9476
9477DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
9478MODULE_VERSION(t4nex, 1);
9479MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
9480
9481DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
9482MODULE_VERSION(t5nex, 1);
9483MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
9484
9485DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
9486MODULE_VERSION(cxgbe, 1);
9487
9488DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
9489MODULE_VERSION(cxl, 1);
9490
9491DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
9492MODULE_VERSION(vcxgbe, 1);
9493
9494DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
9495MODULE_VERSION(vcxl, 1);
9496