xref: /freebsd/sys/dev/cxgbe/adapter.h (revision b59017c5cad90d0f09a59e68c00457b7faf93e7c)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2011 Chelsio Communications, Inc.
5  * All rights reserved.
6  * Written by: Navdeep Parhar <np@FreeBSD.org>
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  */
30 
31 #ifndef __T4_ADAPTER_H__
32 #define __T4_ADAPTER_H__
33 
34 #include <sys/kernel.h>
35 #include <sys/bus.h>
36 #include <sys/counter.h>
37 #include <sys/rman.h>
38 #include <sys/types.h>
39 #include <sys/lock.h>
40 #include <sys/malloc.h>
41 #include <sys/rwlock.h>
42 #include <sys/seqc.h>
43 #include <sys/sx.h>
44 #include <sys/vmem.h>
45 #include <vm/uma.h>
46 
47 #include <dev/pci/pcivar.h>
48 #include <dev/pci/pcireg.h>
49 #include <machine/bus.h>
50 #include <sys/socket.h>
51 #include <sys/sysctl.h>
52 #include <sys/taskqueue.h>
53 #include <net/ethernet.h>
54 #include <net/if.h>
55 #include <net/if_var.h>
56 #include <net/if_media.h>
57 #include <net/pfil.h>
58 #include <netinet/in.h>
59 #include <netinet/tcp_lro.h>
60 
61 #include "offload.h"
62 #include "t4_ioctl.h"
63 #include "common/t4_msg.h"
64 #include "firmware/t4fw_interface.h"
65 
66 #define KTR_CXGBE	KTR_SPARE3
67 MALLOC_DECLARE(M_CXGBE);
68 #define CXGBE_UNIMPLEMENTED(s) \
69     panic("%s (%s, line %d) not implemented yet.", s, __FILE__, __LINE__)
70 
71 /*
72  * Same as LIST_HEAD from queue.h.  This is to avoid conflict with LinuxKPI's
73  * LIST_HEAD when building iw_cxgbe.
74  */
75 #define	CXGBE_LIST_HEAD(name, type)					\
76 struct name {								\
77 	struct type *lh_first;	/* first element */			\
78 }
79 
80 #ifndef SYSCTL_ADD_UQUAD
81 #define SYSCTL_ADD_UQUAD SYSCTL_ADD_QUAD
82 #define sysctl_handle_64 sysctl_handle_quad
83 #define CTLTYPE_U64 CTLTYPE_QUAD
84 #endif
85 
86 SYSCTL_DECL(_hw_cxgbe);
87 
88 struct adapter;
89 typedef struct adapter adapter_t;
90 
91 enum {
92 	/*
93 	 * All ingress queues use this entry size.  Note that the firmware event
94 	 * queue and any iq expecting CPL_RX_PKT in the descriptor needs this to
95 	 * be at least 64.
96 	 */
97 	IQ_ESIZE = 64,
98 
99 	/* Default queue sizes for all kinds of ingress queues */
100 	FW_IQ_QSIZE = 256,
101 	RX_IQ_QSIZE = 1024,
102 
103 	/* All egress queues use this entry size */
104 	EQ_ESIZE = 64,
105 
106 	/* Default queue sizes for all kinds of egress queues */
107 	CTRL_EQ_QSIZE = 1024,
108 	TX_EQ_QSIZE = 1024,
109 
110 	SW_ZONE_SIZES = 4,	/* cluster, jumbop, jumbo9k, jumbo16k */
111 	CL_METADATA_SIZE = CACHE_LINE_SIZE,
112 
113 	SGE_MAX_WR_NDESC = SGE_MAX_WR_LEN / EQ_ESIZE, /* max WR size in desc */
114 	TX_SGL_SEGS = 39,
115 	TX_SGL_SEGS_TSO = 38,
116 	TX_SGL_SEGS_VM = 38,
117 	TX_SGL_SEGS_VM_TSO = 37,
118 	TX_SGL_SEGS_EO_TSO = 30,	/* XXX: lower for IPv6. */
119 	TX_SGL_SEGS_VXLAN_TSO = 37,
120 	TX_WR_FLITS = SGE_MAX_WR_LEN / 8
121 };
122 
123 enum {
124 	/* adapter intr_type */
125 	INTR_INTX	= (1 << 0),
126 	INTR_MSI	= (1 << 1),
127 	INTR_MSIX	= (1 << 2)
128 };
129 
130 enum {
131 	XGMAC_MTU	= (1 << 0),
132 	XGMAC_PROMISC	= (1 << 1),
133 	XGMAC_ALLMULTI	= (1 << 2),
134 	XGMAC_VLANEX	= (1 << 3),
135 	XGMAC_UCADDR	= (1 << 4),
136 	XGMAC_MCADDRS	= (1 << 5),
137 
138 	XGMAC_ALL	= 0xffff
139 };
140 
141 enum {
142 	/* flags understood by begin_synchronized_op */
143 	HOLD_LOCK	= (1 << 0),
144 	SLEEP_OK	= (1 << 1),
145 	INTR_OK		= (1 << 2),
146 
147 	/* flags understood by end_synchronized_op */
148 	LOCK_HELD	= HOLD_LOCK,
149 };
150 
151 enum {
152 	/* adapter flags.  synch_op or adapter_lock. */
153 	FULL_INIT_DONE	= (1 << 0),
154 	FW_OK		= (1 << 1),
155 	CHK_MBOX_ACCESS	= (1 << 2),
156 	MASTER_PF	= (1 << 3),
157 	BUF_PACKING_OK	= (1 << 6),
158 	IS_VF		= (1 << 7),
159 	KERN_TLS_ON	= (1 << 8),	/* HW is configured for KERN_TLS */
160 	CXGBE_BUSY	= (1 << 9),
161 
162 	/* adapter error_flags.  reg_lock for HW_OFF_LIMITS, atomics for the rest. */
163 	ADAP_STOPPED	= (1 << 0),	/* Adapter has been stopped. */
164 	ADAP_FATAL_ERR	= (1 << 1),	/* Encountered a fatal error. */
165 	HW_OFF_LIMITS	= (1 << 2),	/* off limits to all except reset_thread */
166 	ADAP_CIM_ERR	= (1 << 3),	/* Error was related to FW/CIM. */
167 
168 	/* port flags */
169 	HAS_TRACEQ	= (1 << 3),
170 	FIXED_IFMEDIA	= (1 << 4),	/* ifmedia list doesn't change. */
171 
172 	/* VI flags */
173 	VI_DETACHING	= (1 << 0),
174 	VI_INIT_DONE	= (1 << 1),
175 	/* 1 << 2 is unused, was VI_SYSCTL_CTX */
176 	TX_USES_VM_WR	= (1 << 3),
177 	VI_SKIP_STATS	= (1 << 4),
178 
179 	/* adapter debug_flags */
180 	DF_DUMP_MBOX		= (1 << 0),	/* Log all mbox cmd/rpl. */
181 	DF_LOAD_FW_ANYTIME	= (1 << 1),	/* Allow LOAD_FW after init */
182 	DF_DISABLE_TCB_CACHE	= (1 << 2),	/* Disable TCB cache (T6+) */
183 	DF_DISABLE_CFG_RETRY	= (1 << 3),	/* Disable fallback config */
184 	DF_VERBOSE_SLOWINTR	= (1 << 4),	/* Chatty slow intr handler */
185 };
186 
187 #define IS_DETACHING(vi)	((vi)->flags & VI_DETACHING)
188 #define SET_DETACHING(vi)	do {(vi)->flags |= VI_DETACHING;} while (0)
189 #define CLR_DETACHING(vi)	do {(vi)->flags &= ~VI_DETACHING;} while (0)
190 #define IS_BUSY(sc)	((sc)->flags & CXGBE_BUSY)
191 #define SET_BUSY(sc)	do {(sc)->flags |= CXGBE_BUSY;} while (0)
192 #define CLR_BUSY(sc)	do {(sc)->flags &= ~CXGBE_BUSY;} while (0)
193 
194 struct vi_info {
195 	device_t dev;
196 	struct port_info *pi;
197 	struct adapter *adapter;
198 
199 	if_t ifp;
200 	struct pfil_head *pfil;
201 
202 	unsigned long flags;
203 	int if_flags;
204 
205 	uint16_t *rss, *nm_rss;
206 	uint16_t viid;		/* opaque VI identifier */
207 	uint16_t smt_idx;
208 	uint16_t vin;
209 	uint8_t vfvld;
210 	int16_t  xact_addr_filt;/* index of exact MAC address filter */
211 	uint16_t rss_size;	/* size of VI's RSS table slice */
212 	uint16_t rss_base;	/* start of VI's RSS table slice */
213 	int hashen;
214 
215 	int nintr;
216 	int first_intr;
217 
218 	/* These need to be int as they are used in sysctl */
219 	int ntxq;		/* # of tx queues */
220 	int first_txq;		/* index of first tx queue */
221 	int rsrv_noflowq;	/* Reserve queue 0 for non-flowid packets */
222 	int nrxq;		/* # of rx queues */
223 	int first_rxq;		/* index of first rx queue */
224 	int nofldtxq;		/* # of offload tx queues */
225 	int first_ofld_txq;	/* index of first offload tx queue */
226 	int nofldrxq;		/* # of offload rx queues */
227 	int first_ofld_rxq;	/* index of first offload rx queue */
228 	int nnmtxq;
229 	int first_nm_txq;
230 	int nnmrxq;
231 	int first_nm_rxq;
232 	int tmr_idx;
233 	int ofld_tmr_idx;
234 	int pktc_idx;
235 	int ofld_pktc_idx;
236 	int qsize_rxq;
237 	int qsize_txq;
238 
239 	struct timeval last_refreshed;
240 	struct fw_vi_stats_vf stats;
241 	struct mtx tick_mtx;
242 	struct callout tick;
243 
244 	struct sysctl_ctx_list ctx;
245 	struct sysctl_oid *rxq_oid;
246 	struct sysctl_oid *txq_oid;
247 	struct sysctl_oid *nm_rxq_oid;
248 	struct sysctl_oid *nm_txq_oid;
249 	struct sysctl_oid *ofld_rxq_oid;
250 	struct sysctl_oid *ofld_txq_oid;
251 
252 	uint8_t hw_addr[ETHER_ADDR_LEN]; /* factory MAC address, won't change */
253 	u_int txq_rr;
254 	u_int rxq_rr;
255 };
256 
257 struct tx_ch_rl_params {
258 	enum fw_sched_params_rate ratemode;	/* %port (REL) or kbps (ABS) */
259 	uint32_t maxrate;
260 };
261 
262 /* CLRL state */
263 enum clrl_state {
264 	CS_UNINITIALIZED = 0,
265 	CS_PARAMS_SET,			/* sw parameters have been set. */
266 	CS_HW_UPDATE_REQUESTED,		/* async HW update requested. */
267 	CS_HW_UPDATE_IN_PROGRESS,	/* sync hw update in progress. */
268 	CS_HW_CONFIGURED		/* configured in the hardware. */
269 };
270 
271 /* CLRL flags */
272 enum {
273 	CF_USER		= (1 << 0),	/* was configured by driver ioctl. */
274 };
275 
276 struct tx_cl_rl_params {
277 	enum clrl_state state;
278 	int refcount;
279 	uint8_t flags;
280 	enum fw_sched_params_rate ratemode;	/* %port REL or ABS value */
281 	enum fw_sched_params_unit rateunit;	/* kbps or pps (when ABS) */
282 	enum fw_sched_params_mode mode;		/* aggr or per-flow */
283 	uint32_t maxrate;
284 	uint16_t pktsize;
285 	uint16_t burstsize;
286 };
287 
288 /* Tx scheduler parameters for a channel/port */
289 struct tx_sched_params {
290 	/* Channel Rate Limiter */
291 	struct tx_ch_rl_params ch_rl;
292 
293 	/* Class WRR */
294 	/* XXX */
295 
296 	/* Class Rate Limiter (including the default pktsize and burstsize). */
297 	int pktsize;
298 	int burstsize;
299 	struct tx_cl_rl_params cl_rl[];
300 };
301 
302 struct port_info {
303 	device_t dev;
304 	struct adapter *adapter;
305 
306 	struct vi_info *vi;
307 	int nvi;
308 	int up_vis;
309 	int uld_vis;
310 	bool vxlan_tcam_entry;
311 
312 	struct tx_sched_params *sched_params;
313 
314 	struct mtx pi_lock;
315 	char lockname[16];
316 	unsigned long flags;
317 
318 	uint8_t  lport;		/* associated offload logical port */
319 	int8_t   mdio_addr;
320 	uint8_t  port_type;
321 	uint8_t  mod_type;
322 	uint8_t  port_id;
323 	uint8_t  tx_chan;	/* tx TP c-channel */
324 	uint8_t  rx_chan;	/* rx TP c-channel */
325 	uint8_t  mps_bg_map;	/* rx MPS buffer group bitmap */
326 	uint8_t  rx_e_chan_map;	/* rx TP e-channel bitmap */
327 
328 	struct link_config link_cfg;
329 	struct ifmedia media;
330 
331 	struct port_stats stats;
332 	u_int tnl_cong_drops;
333 	u_int tx_parse_error;
334 	int fcs_reg;
335 	uint64_t fcs_base;
336 
337 	struct sysctl_ctx_list ctx;
338 };
339 
340 #define	IS_MAIN_VI(vi)		((vi) == &((vi)->pi->vi[0]))
341 
342 struct cluster_metadata {
343 	uma_zone_t zone;
344 	caddr_t cl;
345 	u_int refcount;
346 };
347 
348 struct fl_sdesc {
349 	caddr_t cl;
350 	uint16_t nmbuf;	/* # of driver originated mbufs with ref on cluster */
351 	int16_t moff;	/* offset of metadata from cl */
352 	uint8_t zidx;
353 };
354 
355 struct tx_desc {
356 	__be64 flit[8];
357 };
358 
359 struct tx_sdesc {
360 	struct mbuf *m;		/* m_nextpkt linked chain of frames */
361 	uint8_t desc_used;	/* # of hardware descriptors used by the WR */
362 };
363 
364 
365 #define IQ_PAD (IQ_ESIZE - sizeof(struct rsp_ctrl) - sizeof(struct rss_header))
366 struct iq_desc {
367 	struct rss_header rss;
368 	uint8_t cpl[IQ_PAD];
369 	struct rsp_ctrl rsp;
370 };
371 #undef IQ_PAD
372 CTASSERT(sizeof(struct iq_desc) == IQ_ESIZE);
373 
374 enum {
375 	/* iq type */
376 	IQ_OTHER	= FW_IQ_IQTYPE_OTHER,
377 	IQ_ETH		= FW_IQ_IQTYPE_NIC,
378 	IQ_OFLD		= FW_IQ_IQTYPE_OFLD,
379 
380 	/* iq flags */
381 	IQ_SW_ALLOCATED	= (1 << 0),	/* sw resources allocated */
382 	IQ_HAS_FL	= (1 << 1),	/* iq associated with a freelist */
383 	IQ_RX_TIMESTAMP	= (1 << 2),	/* provide the SGE rx timestamp */
384 	IQ_LRO_ENABLED	= (1 << 3),	/* iq is an eth rxq with LRO enabled */
385 	IQ_ADJ_CREDIT	= (1 << 4),	/* hw is off by 1 credit for this iq */
386 	IQ_HW_ALLOCATED	= (1 << 5),	/* fw/hw resources allocated */
387 
388 	/* iq state */
389 	IQS_DISABLED	= 0,
390 	IQS_BUSY	= 1,
391 	IQS_IDLE	= 2,
392 
393 	/* netmap related flags */
394 	NM_OFF	= 0,
395 	NM_ON	= 1,
396 	NM_BUSY	= 2,
397 };
398 
399 enum {
400 	CPL_COOKIE_RESERVED = 0,
401 	CPL_COOKIE_FILTER,
402 	CPL_COOKIE_DDP0,
403 	CPL_COOKIE_DDP1,
404 	CPL_COOKIE_TOM,
405 	CPL_COOKIE_HASHFILTER,
406 	CPL_COOKIE_ETHOFLD,
407 	CPL_COOKIE_KERN_TLS,
408 
409 	NUM_CPL_COOKIES = 8	/* Limited by M_COOKIE.  Do not increase. */
410 };
411 
412 struct sge_iq;
413 struct rss_header;
414 typedef int (*cpl_handler_t)(struct sge_iq *, const struct rss_header *,
415     struct mbuf *);
416 typedef int (*an_handler_t)(struct sge_iq *, const struct rsp_ctrl *);
417 typedef int (*fw_msg_handler_t)(struct adapter *, const __be64 *);
418 
419 /*
420  * Ingress Queue: T4 is producer, driver is consumer.
421  */
422 struct sge_iq {
423 	uint16_t flags;
424 	uint8_t qtype;
425 	volatile int state;
426 	struct adapter *adapter;
427 	struct iq_desc  *desc;	/* KVA of descriptor ring */
428 	int8_t   intr_pktc_idx;	/* packet count threshold index */
429 	uint8_t  gen;		/* generation bit */
430 	uint8_t  intr_params;	/* interrupt holdoff parameters */
431 	int8_t   cong_drop;	/* congestion drop settings for the queue */
432 	uint16_t qsize;		/* size (# of entries) of the queue */
433 	uint16_t sidx;		/* index of the entry with the status page */
434 	uint16_t cidx;		/* consumer index */
435 	uint16_t cntxt_id;	/* SGE context id for the iq */
436 	uint16_t abs_id;	/* absolute SGE id for the iq */
437 	int16_t intr_idx;	/* interrupt used by the queue */
438 
439 	STAILQ_ENTRY(sge_iq) link;
440 
441 	bus_dma_tag_t desc_tag;
442 	bus_dmamap_t desc_map;
443 	bus_addr_t ba;		/* bus address of descriptor ring */
444 };
445 
446 enum {
447 	/* eq type */
448 	EQ_CTRL		= 1,
449 	EQ_ETH		= 2,
450 	EQ_OFLD		= 3,
451 
452 	/* eq flags */
453 	EQ_SW_ALLOCATED	= (1 << 0),	/* sw resources allocated */
454 	EQ_HW_ALLOCATED	= (1 << 1),	/* hw/fw resources allocated */
455 	EQ_ENABLED	= (1 << 3),	/* open for business */
456 	EQ_QFLUSH	= (1 << 4),	/* if_qflush in progress */
457 };
458 
459 /* Listed in order of preference.  Update t4_sysctls too if you change these */
460 enum {DOORBELL_UDB, DOORBELL_WCWR, DOORBELL_UDBWC, DOORBELL_KDB};
461 
462 /*
463  * Egress Queue: driver is producer, T4 is consumer.
464  *
465  * Note: A free list is an egress queue (driver produces the buffers and T4
466  * consumes them) but it's special enough to have its own struct (see sge_fl).
467  */
468 struct sge_eq {
469 	unsigned int flags;	/* MUST be first */
470 	unsigned int cntxt_id;	/* SGE context id for the eq */
471 	unsigned int abs_id;	/* absolute SGE id for the eq */
472 	uint8_t type;		/* EQ_CTRL/EQ_ETH/EQ_OFLD */
473 	uint8_t doorbells;
474 	uint8_t port_id;	/* port_id of the port associated with the eq */
475 	uint8_t tx_chan;	/* tx channel used by the eq */
476 	struct mtx eq_lock;
477 
478 	struct tx_desc *desc;	/* KVA of descriptor ring */
479 	volatile uint32_t *udb;	/* KVA of doorbell (lies within BAR2) */
480 	u_int udb_qid;		/* relative qid within the doorbell page */
481 	uint16_t sidx;		/* index of the entry with the status page */
482 	uint16_t cidx;		/* consumer idx (desc idx) */
483 	uint16_t pidx;		/* producer idx (desc idx) */
484 	uint16_t equeqidx;	/* EQUEQ last requested at this pidx */
485 	uint16_t dbidx;		/* pidx of the most recent doorbell */
486 	uint16_t iqid;		/* cached iq->cntxt_id (see iq below) */
487 	volatile u_int equiq;	/* EQUIQ outstanding */
488 	struct sge_iq *iq;	/* iq that receives egr_update for the eq */
489 
490 	bus_dma_tag_t desc_tag;
491 	bus_dmamap_t desc_map;
492 	bus_addr_t ba;		/* bus address of descriptor ring */
493 	char lockname[16];
494 };
495 
496 struct rx_buf_info {
497 	uma_zone_t zone;	/* zone that this cluster comes from */
498 	uint16_t size1;		/* same as size of cluster: 2K/4K/9K/16K.
499 				 * hwsize[hwidx1] = size1.  No spare. */
500 	uint16_t size2;		/* hwsize[hwidx2] = size2.
501 				 * spare in cluster = size1 - size2. */
502 	int8_t hwidx1;		/* SGE bufsize idx for size1 */
503 	int8_t hwidx2;		/* SGE bufsize idx for size2 */
504 	uint8_t type;		/* EXT_xxx type of the cluster */
505 };
506 
507 enum {
508 	NUM_MEMWIN = 3,
509 
510 	MEMWIN0_APERTURE = 2048,
511 	MEMWIN0_BASE     = 0x1b800,
512 
513 	MEMWIN1_APERTURE = 32768,
514 	MEMWIN1_BASE     = 0x28000,
515 
516 	MEMWIN2_APERTURE_T4 = 65536,
517 	MEMWIN2_BASE_T4     = 0x30000,
518 
519 	MEMWIN2_APERTURE_T5 = 128 * 1024,
520 	MEMWIN2_BASE_T5     = 0x60000,
521 };
522 
523 struct memwin {
524 	struct rwlock mw_lock __aligned(CACHE_LINE_SIZE);
525 	uint32_t mw_base;	/* constant after setup_memwin */
526 	uint32_t mw_aperture;	/* ditto */
527 	uint32_t mw_curpos;	/* protected by mw_lock */
528 };
529 
530 enum {
531 	FL_STARVING	= (1 << 0), /* on the adapter's list of starving fl's */
532 	FL_DOOMED	= (1 << 1), /* about to be destroyed */
533 	FL_BUF_PACKING	= (1 << 2), /* buffer packing enabled */
534 	FL_BUF_RESUME	= (1 << 3), /* resume from the middle of the frame */
535 };
536 
537 #define FL_RUNNING_LOW(fl) \
538     (IDXDIFF(fl->dbidx * 8, fl->cidx, fl->sidx * 8) <= fl->lowat)
539 #define FL_NOT_RUNNING_LOW(fl) \
540     (IDXDIFF(fl->dbidx * 8, fl->cidx, fl->sidx * 8) >= 2 * fl->lowat)
541 
542 struct sge_fl {
543 	struct mtx fl_lock;
544 	__be64 *desc;		/* KVA of descriptor ring, ptr to addresses */
545 	struct fl_sdesc *sdesc;	/* KVA of software descriptor ring */
546 	uint16_t zidx;		/* refill zone idx */
547 	uint16_t safe_zidx;
548 	uint16_t lowat;		/* # of buffers <= this means fl needs help */
549 	int flags;
550 	uint16_t buf_boundary;
551 
552 	/* The 16b idx all deal with hw descriptors */
553 	uint16_t dbidx;		/* hw pidx after last doorbell */
554 	uint16_t sidx;		/* index of status page */
555 	volatile uint16_t hw_cidx;
556 
557 	/* The 32b idx are all buffer idx, not hardware descriptor idx */
558 	uint32_t cidx;		/* consumer index */
559 	uint32_t pidx;		/* producer index */
560 
561 	uint32_t dbval;
562 	u_int rx_offset;	/* offset in fl buf (when buffer packing) */
563 	volatile uint32_t *udb;
564 
565 	uint64_t cl_allocated;	/* # of clusters allocated */
566 	uint64_t cl_recycled;	/* # of clusters recycled */
567 	uint64_t cl_fast_recycled; /* # of clusters recycled (fast) */
568 
569 	/* These 3 are valid when FL_BUF_RESUME is set, stale otherwise. */
570 	struct mbuf *m0;
571 	struct mbuf **pnext;
572 	u_int remaining;
573 
574 	uint16_t qsize;		/* # of hw descriptors (status page included) */
575 	uint16_t cntxt_id;	/* SGE context id for the freelist */
576 	TAILQ_ENTRY(sge_fl) link; /* All starving freelists */
577 	bus_dma_tag_t desc_tag;
578 	bus_dmamap_t desc_map;
579 	char lockname[16];
580 	bus_addr_t ba;		/* bus address of descriptor ring */
581 };
582 
583 struct mp_ring;
584 
585 struct txpkts {
586 	uint8_t wr_type;	/* type 0 or type 1 */
587 	uint8_t npkt;		/* # of packets in this work request */
588 	uint8_t len16;		/* # of 16B pieces used by this work request */
589 	uint8_t score;
590 	uint8_t max_npkt;	/* maximum number of packets allowed */
591 	uint16_t plen;		/* total payload (sum of all packets) */
592 
593 	/* straight from fw_eth_tx_pkts_vm_wr. */
594 	__u8   ethmacdst[6];
595 	__u8   ethmacsrc[6];
596 	__be16 ethtype;
597 	__be16 vlantci;
598 
599 	struct mbuf *mb[15];
600 };
601 
602 /* txq: SGE egress queue + what's needed for Ethernet NIC */
603 struct sge_txq {
604 	struct sge_eq eq;	/* MUST be first */
605 
606 	if_t ifp;		/* the interface this txq belongs to */
607 	struct mp_ring *r;	/* tx software ring */
608 	struct tx_sdesc *sdesc;	/* KVA of software descriptor ring */
609 	struct sglist *gl;
610 	__be32 cpl_ctrl0;	/* for convenience */
611 	int tc_idx;		/* traffic class */
612 	uint64_t last_tx;	/* cycle count when eth_tx was last called */
613 	struct txpkts txp;
614 
615 	struct task tx_reclaim_task;
616 	/* stats for common events first */
617 
618 	uint64_t txcsum;	/* # of times hardware assisted with checksum */
619 	uint64_t tso_wrs;	/* # of TSO work requests */
620 	uint64_t vlan_insertion;/* # of times VLAN tag was inserted */
621 	uint64_t imm_wrs;	/* # of work requests with immediate data */
622 	uint64_t sgl_wrs;	/* # of work requests with direct SGL */
623 	uint64_t txpkt_wrs;	/* # of txpkt work requests (not coalesced) */
624 	uint64_t txpkts0_wrs;	/* # of type0 coalesced tx work requests */
625 	uint64_t txpkts1_wrs;	/* # of type1 coalesced tx work requests */
626 	uint64_t txpkts0_pkts;	/* # of frames in type0 coalesced tx WRs */
627 	uint64_t txpkts1_pkts;	/* # of frames in type1 coalesced tx WRs */
628 	uint64_t txpkts_flush;	/* # of times txp had to be sent by tx_update */
629 	uint64_t raw_wrs;	/* # of raw work requests (alloc_wr_mbuf) */
630 	uint64_t vxlan_tso_wrs;	/* # of VXLAN TSO work requests */
631 	uint64_t vxlan_txcsum;
632 
633 	uint64_t kern_tls_records;
634 	uint64_t kern_tls_short;
635 	uint64_t kern_tls_partial;
636 	uint64_t kern_tls_full;
637 	uint64_t kern_tls_octets;
638 	uint64_t kern_tls_waste;
639 	uint64_t kern_tls_options;
640 	uint64_t kern_tls_header;
641 	uint64_t kern_tls_fin;
642 	uint64_t kern_tls_fin_short;
643 	uint64_t kern_tls_cbc;
644 	uint64_t kern_tls_gcm;
645 
646 	/* stats for not-that-common events */
647 
648 	/* Optional scratch space for constructing work requests. */
649 	uint8_t ss[SGE_MAX_WR_LEN] __aligned(16);
650 } __aligned(CACHE_LINE_SIZE);
651 
652 /* rxq: SGE ingress queue + SGE free list + miscellaneous items */
653 struct sge_rxq {
654 	struct sge_iq iq;	/* MUST be first */
655 	struct sge_fl fl;	/* MUST follow iq */
656 
657 	if_t ifp;		/* the interface this rxq belongs to */
658 	struct lro_ctrl lro;	/* LRO state */
659 
660 	/* stats for common events first */
661 
662 	uint64_t rxcsum;	/* # of times hardware assisted with checksum */
663 	uint64_t vlan_extraction;/* # of times VLAN tag was extracted */
664 	uint64_t vxlan_rxcsum;
665 
666 	/* stats for not-that-common events */
667 
668 } __aligned(CACHE_LINE_SIZE);
669 
670 static inline struct sge_rxq *
671 iq_to_rxq(struct sge_iq *iq)
672 {
673 
674 	return (__containerof(iq, struct sge_rxq, iq));
675 }
676 
677 /* ofld_rxq: SGE ingress queue + SGE free list + miscellaneous items */
678 struct sge_ofld_rxq {
679 	struct sge_iq iq;	/* MUST be first */
680 	struct sge_fl fl;	/* MUST follow iq */
681 	counter_u64_t rx_iscsi_ddp_setup_ok;
682 	counter_u64_t rx_iscsi_ddp_setup_error;
683 	uint64_t rx_iscsi_ddp_pdus;
684 	uint64_t rx_iscsi_ddp_octets;
685 	uint64_t rx_iscsi_fl_pdus;
686 	uint64_t rx_iscsi_fl_octets;
687 	uint64_t rx_iscsi_padding_errors;
688 	uint64_t rx_iscsi_header_digest_errors;
689 	uint64_t rx_iscsi_data_digest_errors;
690 	uint64_t rx_aio_ddp_jobs;
691 	uint64_t rx_aio_ddp_octets;
692 	u_long	rx_toe_tls_records;
693 	u_long	rx_toe_tls_octets;
694 	u_long	rx_toe_ddp_octets;
695 	counter_u64_t ddp_buffer_alloc;
696 	counter_u64_t ddp_buffer_reuse;
697 	counter_u64_t ddp_buffer_free;
698 } __aligned(CACHE_LINE_SIZE);
699 
700 static inline struct sge_ofld_rxq *
701 iq_to_ofld_rxq(struct sge_iq *iq)
702 {
703 
704 	return (__containerof(iq, struct sge_ofld_rxq, iq));
705 }
706 
707 struct wrqe {
708 	STAILQ_ENTRY(wrqe) link;
709 	struct sge_wrq *wrq;
710 	int wr_len;
711 	char wr[] __aligned(16);
712 };
713 
714 struct wrq_cookie {
715 	TAILQ_ENTRY(wrq_cookie) link;
716 	int ndesc;
717 	int pidx;
718 };
719 
720 /*
721  * wrq: SGE egress queue that is given prebuilt work requests.  Control queues
722  * are of this type.
723  */
724 struct sge_wrq {
725 	struct sge_eq eq;	/* MUST be first */
726 
727 	struct adapter *adapter;
728 	struct task wrq_tx_task;
729 
730 	/* Tx desc reserved but WR not "committed" yet. */
731 	TAILQ_HEAD(wrq_incomplete_wrs , wrq_cookie) incomplete_wrs;
732 
733 	/* List of WRs ready to go out as soon as descriptors are available. */
734 	STAILQ_HEAD(, wrqe) wr_list;
735 	u_int nwr_pending;
736 	u_int ndesc_needed;
737 
738 	/* stats for common events first */
739 
740 	uint64_t tx_wrs_direct;	/* # of WRs written directly to desc ring. */
741 	uint64_t tx_wrs_ss;	/* # of WRs copied from scratch space. */
742 	uint64_t tx_wrs_copied;	/* # of WRs queued and copied to desc ring. */
743 
744 	/* stats for not-that-common events */
745 
746 	/*
747 	 * Scratch space for work requests that wrap around after reaching the
748 	 * status page, and some information about the last WR that used it.
749 	 */
750 	uint16_t ss_pidx;
751 	uint16_t ss_len;
752 	uint8_t ss[SGE_MAX_WR_LEN];
753 
754 } __aligned(CACHE_LINE_SIZE);
755 
756 /* ofld_txq: SGE egress queue + miscellaneous items */
757 struct sge_ofld_txq {
758 	struct sge_wrq wrq;
759 	counter_u64_t tx_iscsi_pdus;
760 	counter_u64_t tx_iscsi_octets;
761 	counter_u64_t tx_iscsi_iso_wrs;
762 	counter_u64_t tx_aio_jobs;
763 	counter_u64_t tx_aio_octets;
764 	counter_u64_t tx_toe_tls_records;
765 	counter_u64_t tx_toe_tls_octets;
766 } __aligned(CACHE_LINE_SIZE);
767 
768 #define INVALID_NM_RXQ_CNTXT_ID ((uint16_t)(-1))
769 struct sge_nm_rxq {
770 	/* Items used by the driver rx ithread are in this cacheline. */
771 	volatile int nm_state __aligned(CACHE_LINE_SIZE);	/* NM_OFF, NM_ON, or NM_BUSY */
772 	u_int nid;		/* netmap ring # for this queue */
773 	struct vi_info *vi;
774 
775 	struct iq_desc *iq_desc;
776 	uint16_t iq_abs_id;
777 	uint16_t iq_cntxt_id;
778 	uint16_t iq_cidx;
779 	uint16_t iq_sidx;
780 	uint8_t iq_gen;
781 	uint32_t fl_sidx;
782 
783 	/* Items used by netmap rxsync are in this cacheline. */
784 	__be64  *fl_desc __aligned(CACHE_LINE_SIZE);
785 	uint16_t fl_cntxt_id;
786 	uint32_t fl_pidx;
787 	uint32_t fl_sidx2;	/* copy of fl_sidx */
788 	uint32_t fl_db_val;
789 	u_int fl_db_saved;
790 	u_int fl_db_threshold;	/* in descriptors */
791 	u_int fl_hwidx:4;
792 
793 	/*
794 	 * fl_cidx is used by both the ithread and rxsync, the rest are not used
795 	 * in the rx fast path.
796 	 */
797 	uint32_t fl_cidx __aligned(CACHE_LINE_SIZE);
798 
799 	bus_dma_tag_t iq_desc_tag;
800 	bus_dmamap_t iq_desc_map;
801 	bus_addr_t iq_ba;
802 	int intr_idx;
803 
804 	bus_dma_tag_t fl_desc_tag;
805 	bus_dmamap_t fl_desc_map;
806 	bus_addr_t fl_ba;
807 };
808 
809 #define INVALID_NM_TXQ_CNTXT_ID ((u_int)(-1))
810 struct sge_nm_txq {
811 	struct tx_desc *desc;
812 	uint16_t cidx;
813 	uint16_t pidx;
814 	uint16_t sidx;
815 	uint16_t equiqidx;	/* EQUIQ last requested at this pidx */
816 	uint16_t equeqidx;	/* EQUEQ last requested at this pidx */
817 	uint16_t dbidx;		/* pidx of the most recent doorbell */
818 	uint8_t doorbells;
819 	volatile uint32_t *udb;
820 	u_int udb_qid;
821 	u_int cntxt_id;
822 	__be32 cpl_ctrl0;	/* for convenience */
823 	__be32 op_pkd;		/* ditto */
824 	u_int nid;		/* netmap ring # for this queue */
825 
826 	/* infrequently used items after this */
827 
828 	bus_dma_tag_t desc_tag;
829 	bus_dmamap_t desc_map;
830 	bus_addr_t ba;
831 	int iqidx;
832 } __aligned(CACHE_LINE_SIZE);
833 
834 struct sge {
835 	int nrxq;	/* total # of Ethernet rx queues */
836 	int ntxq;	/* total # of Ethernet tx queues */
837 	int nofldrxq;	/* total # of TOE rx queues */
838 	int nofldtxq;	/* total # of TOE tx queues */
839 	int nnmrxq;	/* total # of netmap rx queues */
840 	int nnmtxq;	/* total # of netmap tx queues */
841 	int niq;	/* total # of ingress queues */
842 	int neq;	/* total # of egress queues */
843 
844 	struct sge_iq fwq;	/* Firmware event queue */
845 	struct sge_wrq *ctrlq;	/* Control queues */
846 	struct sge_txq *txq;	/* NIC tx queues */
847 	struct sge_rxq *rxq;	/* NIC rx queues */
848 	struct sge_ofld_txq *ofld_txq;	/* TOE tx queues */
849 	struct sge_ofld_rxq *ofld_rxq;	/* TOE rx queues */
850 	struct sge_nm_txq *nm_txq;	/* netmap tx queues */
851 	struct sge_nm_rxq *nm_rxq;	/* netmap rx queues */
852 
853 	uint16_t iq_start;	/* first cntxt_id */
854 	uint16_t iq_base;	/* first abs_id */
855 	int eq_start;		/* first cntxt_id */
856 	int eq_base;		/* first abs_id */
857 	int iqmap_sz;
858 	int eqmap_sz;
859 	struct sge_iq **iqmap;	/* iq->cntxt_id to iq mapping */
860 	struct sge_eq **eqmap;	/* eq->cntxt_id to eq mapping */
861 
862 	int8_t safe_zidx;
863 	struct rx_buf_info rx_buf_info[SW_ZONE_SIZES];
864 };
865 
866 struct devnames {
867 	const char *nexus_name;
868 	const char *ifnet_name;
869 	const char *vi_ifnet_name;
870 	const char *pf03_drv_name;
871 	const char *vf_nexus_name;
872 	const char *vf_ifnet_name;
873 };
874 
875 struct clip_entry;
876 
877 #define CNT_CAL_INFO 3
878 struct clock_sync {
879 	uint64_t hw_cur;
880 	uint64_t hw_prev;
881 	sbintime_t sbt_cur;
882 	sbintime_t sbt_prev;
883 	seqc_t gen;
884 };
885 
886 struct adapter {
887 	SLIST_ENTRY(adapter) link;
888 	device_t dev;
889 	struct cdev *cdev;
890 	const struct devnames *names;
891 
892 	/* PCIe register resources */
893 	int regs_rid;
894 	struct resource *regs_res;
895 	int msix_rid;
896 	struct resource *msix_res;
897 	bus_space_handle_t bh;
898 	bus_space_tag_t bt;
899 	bus_size_t mmio_len;
900 	int udbs_rid;
901 	struct resource *udbs_res;
902 	volatile uint8_t *udbs_base;
903 
904 	unsigned int pf;
905 	unsigned int mbox;
906 	unsigned int vpd_busy;
907 	unsigned int vpd_flag;
908 
909 	/* Interrupt information */
910 	int intr_type;
911 	int intr_count;
912 	struct irq {
913 		struct resource *res;
914 		int rid;
915 		void *tag;
916 		struct sge_rxq *rxq;
917 		struct sge_nm_rxq *nm_rxq;
918 	} __aligned(CACHE_LINE_SIZE) *irq;
919 	int sge_gts_reg;
920 	int sge_kdoorbell_reg;
921 
922 	bus_dma_tag_t dmat;	/* Parent DMA tag */
923 
924 	struct sge sge;
925 	int lro_timeout;
926 	int sc_do_rxcopy;
927 
928 	int vxlan_port;
929 	u_int vxlan_refcount;
930 	int rawf_base;
931 	int nrawf;
932 	u_int vlan_id;
933 
934 	struct taskqueue *tq[MAX_NPORTS];	/* General purpose taskqueues */
935 	struct port_info *port[MAX_NPORTS];
936 	uint8_t chan_map[MAX_NCHAN];		/* channel -> port */
937 
938 	CXGBE_LIST_HEAD(, clip_entry) *clip_table;
939 	TAILQ_HEAD(, clip_entry) clip_pending;	/* these need hw update. */
940 	u_long clip_mask;
941 	int clip_gen;
942 	struct timeout_task clip_task;
943 
944 	void *tom_softc;	/* (struct tom_data *) */
945 	struct tom_tunables tt;
946 	struct t4_offload_policy *policy;
947 	struct rwlock policy_lock;
948 
949 	void *iwarp_softc;	/* (struct c4iw_dev *) */
950 	struct iw_tunables iwt;
951 	void *iscsi_ulp_softc;	/* (struct cxgbei_data *) */
952 	struct l2t_data *l2t;	/* L2 table */
953 	struct smt_data *smt;	/* Source MAC Table */
954 	struct tid_info tids;
955 	vmem_t *key_map;
956 	struct tls_tunables tlst;
957 
958 	uint8_t doorbells;
959 	int offload_map;	/* port_id's with IFCAP_TOE enabled */
960 	int bt_map;		/* tx_chan's with BASE-T */
961 	int active_ulds;	/* ULDs activated on this adapter */
962 	int flags;
963 	int debug_flags;
964 	int error_flags;	/* Used by error handler and live reset. */
965 
966 	char ifp_lockname[16];
967 	struct mtx ifp_lock;
968 	if_t ifp;		/* tracer ifp */
969 	struct ifmedia media;
970 	int traceq;		/* iq used by all tracers, -1 if none */
971 	int tracer_valid;	/* bitmap of valid tracers */
972 	int tracer_enabled;	/* bitmap of enabled tracers */
973 
974 	char fw_version[16];
975 	char tp_version[16];
976 	char er_version[16];
977 	char bs_version[16];
978 	char cfg_file[32];
979 	u_int cfcsum;
980 	struct adapter_params params;
981 	const struct chip_params *chip_params;
982 	struct t4_virt_res vres;
983 
984 	uint16_t nbmcaps;
985 	uint16_t linkcaps;
986 	uint16_t switchcaps;
987 	uint16_t niccaps;
988 	uint16_t toecaps;
989 	uint16_t rdmacaps;
990 	uint16_t cryptocaps;
991 	uint16_t iscsicaps;
992 	uint16_t fcoecaps;
993 
994 	struct sysctl_ctx_list ctx;
995 	struct sysctl_oid *ctrlq_oid;
996 	struct sysctl_oid *fwq_oid;
997 
998 	struct mtx sc_lock;
999 	char lockname[16];
1000 
1001 	/* Starving free lists */
1002 	struct mtx sfl_lock;	/* same cache-line as sc_lock? but that's ok */
1003 	TAILQ_HEAD(, sge_fl) sfl;
1004 	struct callout sfl_callout;
1005 	struct callout cal_callout;
1006 	struct clock_sync cal_info[CNT_CAL_INFO];
1007 	int cal_current;
1008 	int cal_count;
1009 	uint32_t cal_gen;
1010 
1011 	/*
1012 	 * Driver code that can run when the adapter is suspended must use this
1013 	 * lock or a synchronized_op and check for HW_OFF_LIMITS before
1014 	 * accessing hardware.
1015 	 *
1016 	 * XXX: could be changed to rwlock.  wlock in suspend/resume and for
1017 	 * indirect register access, rlock everywhere else.
1018 	 */
1019 	struct mtx reg_lock;
1020 
1021 	struct memwin memwin[NUM_MEMWIN];	/* memory windows */
1022 
1023 	struct mtx tc_lock;
1024 	struct task tc_task;
1025 
1026 	struct task fatal_error_task;
1027 	struct task reset_task;
1028 	const void *reset_thread;
1029 	int num_resets;
1030 	int incarnation;
1031 
1032 	const char *last_op;
1033 	const void *last_op_thr;
1034 	int last_op_flags;
1035 
1036 	int swintr;
1037 	int sensor_resets;
1038 
1039 	struct callout ktls_tick;
1040 };
1041 
1042 #define ADAPTER_LOCK(sc)		mtx_lock(&(sc)->sc_lock)
1043 #define ADAPTER_UNLOCK(sc)		mtx_unlock(&(sc)->sc_lock)
1044 #define ADAPTER_LOCK_ASSERT_OWNED(sc)	mtx_assert(&(sc)->sc_lock, MA_OWNED)
1045 #define ADAPTER_LOCK_ASSERT_NOTOWNED(sc) mtx_assert(&(sc)->sc_lock, MA_NOTOWNED)
1046 
1047 #define ASSERT_SYNCHRONIZED_OP(sc)	\
1048     KASSERT(IS_BUSY(sc) && \
1049 	(mtx_owned(&(sc)->sc_lock) || sc->last_op_thr == curthread), \
1050 	("%s: operation not synchronized.", __func__))
1051 
1052 #define PORT_LOCK(pi)			mtx_lock(&(pi)->pi_lock)
1053 #define PORT_UNLOCK(pi)			mtx_unlock(&(pi)->pi_lock)
1054 #define PORT_LOCK_ASSERT_OWNED(pi)	mtx_assert(&(pi)->pi_lock, MA_OWNED)
1055 #define PORT_LOCK_ASSERT_NOTOWNED(pi)	mtx_assert(&(pi)->pi_lock, MA_NOTOWNED)
1056 
1057 #define FL_LOCK(fl)			mtx_lock(&(fl)->fl_lock)
1058 #define FL_TRYLOCK(fl)			mtx_trylock(&(fl)->fl_lock)
1059 #define FL_UNLOCK(fl)			mtx_unlock(&(fl)->fl_lock)
1060 #define FL_LOCK_ASSERT_OWNED(fl)	mtx_assert(&(fl)->fl_lock, MA_OWNED)
1061 #define FL_LOCK_ASSERT_NOTOWNED(fl)	mtx_assert(&(fl)->fl_lock, MA_NOTOWNED)
1062 
1063 #define RXQ_FL_LOCK(rxq)		FL_LOCK(&(rxq)->fl)
1064 #define RXQ_FL_UNLOCK(rxq)		FL_UNLOCK(&(rxq)->fl)
1065 #define RXQ_FL_LOCK_ASSERT_OWNED(rxq)	FL_LOCK_ASSERT_OWNED(&(rxq)->fl)
1066 #define RXQ_FL_LOCK_ASSERT_NOTOWNED(rxq) FL_LOCK_ASSERT_NOTOWNED(&(rxq)->fl)
1067 
1068 #define EQ_LOCK(eq)			mtx_lock(&(eq)->eq_lock)
1069 #define EQ_TRYLOCK(eq)			mtx_trylock(&(eq)->eq_lock)
1070 #define EQ_UNLOCK(eq)			mtx_unlock(&(eq)->eq_lock)
1071 #define EQ_LOCK_ASSERT_OWNED(eq)	mtx_assert(&(eq)->eq_lock, MA_OWNED)
1072 #define EQ_LOCK_ASSERT_NOTOWNED(eq)	mtx_assert(&(eq)->eq_lock, MA_NOTOWNED)
1073 
1074 #define TXQ_LOCK(txq)			EQ_LOCK(&(txq)->eq)
1075 #define TXQ_TRYLOCK(txq)		EQ_TRYLOCK(&(txq)->eq)
1076 #define TXQ_UNLOCK(txq)			EQ_UNLOCK(&(txq)->eq)
1077 #define TXQ_LOCK_ASSERT_OWNED(txq)	EQ_LOCK_ASSERT_OWNED(&(txq)->eq)
1078 #define TXQ_LOCK_ASSERT_NOTOWNED(txq)	EQ_LOCK_ASSERT_NOTOWNED(&(txq)->eq)
1079 
1080 #define for_each_txq(vi, iter, q) \
1081 	for (q = &vi->adapter->sge.txq[vi->first_txq], iter = 0; \
1082 	    iter < vi->ntxq; ++iter, ++q)
1083 #define for_each_rxq(vi, iter, q) \
1084 	for (q = &vi->adapter->sge.rxq[vi->first_rxq], iter = 0; \
1085 	    iter < vi->nrxq; ++iter, ++q)
1086 #define for_each_ofld_txq(vi, iter, q) \
1087 	for (q = &vi->adapter->sge.ofld_txq[vi->first_ofld_txq], iter = 0; \
1088 	    iter < vi->nofldtxq; ++iter, ++q)
1089 #define for_each_ofld_rxq(vi, iter, q) \
1090 	for (q = &vi->adapter->sge.ofld_rxq[vi->first_ofld_rxq], iter = 0; \
1091 	    iter < vi->nofldrxq; ++iter, ++q)
1092 #define for_each_nm_txq(vi, iter, q) \
1093 	for (q = &vi->adapter->sge.nm_txq[vi->first_nm_txq], iter = 0; \
1094 	    iter < vi->nnmtxq; ++iter, ++q)
1095 #define for_each_nm_rxq(vi, iter, q) \
1096 	for (q = &vi->adapter->sge.nm_rxq[vi->first_nm_rxq], iter = 0; \
1097 	    iter < vi->nnmrxq; ++iter, ++q)
1098 #define for_each_vi(_pi, _iter, _vi) \
1099 	for ((_vi) = (_pi)->vi, (_iter) = 0; (_iter) < (_pi)->nvi; \
1100 	     ++(_iter), ++(_vi))
1101 
1102 #define IDXINCR(idx, incr, wrap) do { \
1103 	idx = wrap - idx > incr ? idx + incr : incr - (wrap - idx); \
1104 } while (0)
1105 #define IDXDIFF(head, tail, wrap) \
1106 	((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head))
1107 
1108 /* One for errors, one for firmware events */
1109 #define T4_EXTRA_INTR 2
1110 
1111 /* One for firmware events */
1112 #define T4VF_EXTRA_INTR 1
1113 
1114 static inline int
1115 forwarding_intr_to_fwq(struct adapter *sc)
1116 {
1117 
1118 	return (sc->intr_count == 1);
1119 }
1120 
1121 /* Works reliably inside a sync_op or with reg_lock held. */
1122 static inline bool
1123 hw_off_limits(struct adapter *sc)
1124 {
1125 	const int off_limits = atomic_load_int(&sc->error_flags) & HW_OFF_LIMITS;
1126 
1127 	return (__predict_false(off_limits != 0));
1128 }
1129 
1130 static inline bool
1131 adapter_stopped(struct adapter *sc)
1132 {
1133 	const int stopped = atomic_load_int(&sc->error_flags) & ADAP_STOPPED;
1134 
1135 	return (__predict_false(stopped != 0));
1136 }
1137 
1138 static inline int
1139 mbuf_nsegs(struct mbuf *m)
1140 {
1141 	M_ASSERTPKTHDR(m);
1142 	KASSERT(m->m_pkthdr.inner_l5hlen > 0,
1143 	    ("%s: mbuf %p missing information on # of segments.", __func__, m));
1144 
1145 	return (m->m_pkthdr.inner_l5hlen);
1146 }
1147 
1148 static inline void
1149 set_mbuf_nsegs(struct mbuf *m, uint8_t nsegs)
1150 {
1151 	M_ASSERTPKTHDR(m);
1152 	m->m_pkthdr.inner_l5hlen = nsegs;
1153 }
1154 
1155 /* Internal mbuf flags stored in PH_loc.eight[1]. */
1156 #define	MC_NOMAP		0x01
1157 #define	MC_RAW_WR		0x02
1158 #define	MC_TLS			0x04
1159 
1160 static inline int
1161 mbuf_cflags(struct mbuf *m)
1162 {
1163 	M_ASSERTPKTHDR(m);
1164 	return (m->m_pkthdr.PH_loc.eight[4]);
1165 }
1166 
1167 static inline void
1168 set_mbuf_cflags(struct mbuf *m, uint8_t flags)
1169 {
1170 	M_ASSERTPKTHDR(m);
1171 	m->m_pkthdr.PH_loc.eight[4] = flags;
1172 }
1173 
1174 static inline int
1175 mbuf_len16(struct mbuf *m)
1176 {
1177 	int n;
1178 
1179 	M_ASSERTPKTHDR(m);
1180 	n = m->m_pkthdr.PH_loc.eight[0];
1181 	if (!(mbuf_cflags(m) & MC_TLS))
1182 		MPASS(n > 0 && n <= SGE_MAX_WR_LEN / 16);
1183 
1184 	return (n);
1185 }
1186 
1187 static inline void
1188 set_mbuf_len16(struct mbuf *m, uint8_t len16)
1189 {
1190 	M_ASSERTPKTHDR(m);
1191 	if (!(mbuf_cflags(m) & MC_TLS))
1192 		MPASS(len16 > 0 && len16 <= SGE_MAX_WR_LEN / 16);
1193 	m->m_pkthdr.PH_loc.eight[0] = len16;
1194 }
1195 
1196 static inline uint32_t
1197 t4_read_reg(struct adapter *sc, uint32_t reg)
1198 {
1199 	if (hw_off_limits(sc))
1200 		MPASS(curthread == sc->reset_thread);
1201 	return bus_space_read_4(sc->bt, sc->bh, reg);
1202 }
1203 
1204 static inline void
1205 t4_write_reg(struct adapter *sc, uint32_t reg, uint32_t val)
1206 {
1207 	if (hw_off_limits(sc))
1208 		MPASS(curthread == sc->reset_thread);
1209 	bus_space_write_4(sc->bt, sc->bh, reg, val);
1210 }
1211 
1212 static inline uint64_t
1213 t4_read_reg64(struct adapter *sc, uint32_t reg)
1214 {
1215 	if (hw_off_limits(sc))
1216 		MPASS(curthread == sc->reset_thread);
1217 #ifdef __LP64__
1218 	return bus_space_read_8(sc->bt, sc->bh, reg);
1219 #else
1220 	return (uint64_t)bus_space_read_4(sc->bt, sc->bh, reg) +
1221 	    ((uint64_t)bus_space_read_4(sc->bt, sc->bh, reg + 4) << 32);
1222 
1223 #endif
1224 }
1225 
1226 static inline void
1227 t4_write_reg64(struct adapter *sc, uint32_t reg, uint64_t val)
1228 {
1229 	if (hw_off_limits(sc))
1230 		MPASS(curthread == sc->reset_thread);
1231 #ifdef __LP64__
1232 	bus_space_write_8(sc->bt, sc->bh, reg, val);
1233 #else
1234 	bus_space_write_4(sc->bt, sc->bh, reg, val);
1235 	bus_space_write_4(sc->bt, sc->bh, reg + 4, val>> 32);
1236 #endif
1237 }
1238 
1239 static inline void
1240 t4_os_pci_read_cfg1(struct adapter *sc, int reg, uint8_t *val)
1241 {
1242 	if (hw_off_limits(sc))
1243 		MPASS(curthread == sc->reset_thread);
1244 	*val = pci_read_config(sc->dev, reg, 1);
1245 }
1246 
1247 static inline void
1248 t4_os_pci_write_cfg1(struct adapter *sc, int reg, uint8_t val)
1249 {
1250 	if (hw_off_limits(sc))
1251 		MPASS(curthread == sc->reset_thread);
1252 	pci_write_config(sc->dev, reg, val, 1);
1253 }
1254 
1255 static inline void
1256 t4_os_pci_read_cfg2(struct adapter *sc, int reg, uint16_t *val)
1257 {
1258 
1259 	if (hw_off_limits(sc))
1260 		MPASS(curthread == sc->reset_thread);
1261 	*val = pci_read_config(sc->dev, reg, 2);
1262 }
1263 
1264 static inline void
1265 t4_os_pci_write_cfg2(struct adapter *sc, int reg, uint16_t val)
1266 {
1267 	if (hw_off_limits(sc))
1268 		MPASS(curthread == sc->reset_thread);
1269 	pci_write_config(sc->dev, reg, val, 2);
1270 }
1271 
1272 static inline void
1273 t4_os_pci_read_cfg4(struct adapter *sc, int reg, uint32_t *val)
1274 {
1275 	if (hw_off_limits(sc))
1276 		MPASS(curthread == sc->reset_thread);
1277 	*val = pci_read_config(sc->dev, reg, 4);
1278 }
1279 
1280 static inline void
1281 t4_os_pci_write_cfg4(struct adapter *sc, int reg, uint32_t val)
1282 {
1283 	if (hw_off_limits(sc))
1284 		MPASS(curthread == sc->reset_thread);
1285 	pci_write_config(sc->dev, reg, val, 4);
1286 }
1287 
1288 static inline struct port_info *
1289 adap2pinfo(struct adapter *sc, int idx)
1290 {
1291 
1292 	return (sc->port[idx]);
1293 }
1294 
1295 static inline void
1296 t4_os_set_hw_addr(struct port_info *pi, uint8_t hw_addr[])
1297 {
1298 
1299 	bcopy(hw_addr, pi->vi[0].hw_addr, ETHER_ADDR_LEN);
1300 }
1301 
1302 static inline int
1303 tx_resume_threshold(struct sge_eq *eq)
1304 {
1305 
1306 	/* not quite the same as qsize / 4, but this will do. */
1307 	return (eq->sidx / 4);
1308 }
1309 
1310 static inline int
1311 t4_use_ldst(struct adapter *sc)
1312 {
1313 
1314 #ifdef notyet
1315 	return (sc->flags & FW_OK || !sc->use_bd);
1316 #else
1317 	return (0);
1318 #endif
1319 }
1320 
1321 static inline void
1322 CH_DUMP_MBOX(struct adapter *sc, int mbox, const int reg,
1323     const char *msg, const __be64 *const p, const bool err)
1324 {
1325 
1326 	if (!(sc->debug_flags & DF_DUMP_MBOX) && !err)
1327 		return;
1328 	if (p != NULL) {
1329 		log(err ? LOG_ERR : LOG_DEBUG,
1330 		    "%s: mbox %u %s %016llx %016llx %016llx %016llx "
1331 		    "%016llx %016llx %016llx %016llx\n",
1332 		    device_get_nameunit(sc->dev), mbox, msg,
1333 		    (long long)be64_to_cpu(p[0]), (long long)be64_to_cpu(p[1]),
1334 		    (long long)be64_to_cpu(p[2]), (long long)be64_to_cpu(p[3]),
1335 		    (long long)be64_to_cpu(p[4]), (long long)be64_to_cpu(p[5]),
1336 		    (long long)be64_to_cpu(p[6]), (long long)be64_to_cpu(p[7]));
1337 	} else {
1338 		log(err ? LOG_ERR : LOG_DEBUG,
1339 		    "%s: mbox %u %s %016llx %016llx %016llx %016llx "
1340 		    "%016llx %016llx %016llx %016llx\n",
1341 		    device_get_nameunit(sc->dev), mbox, msg,
1342 		    (long long)t4_read_reg64(sc, reg),
1343 		    (long long)t4_read_reg64(sc, reg + 8),
1344 		    (long long)t4_read_reg64(sc, reg + 16),
1345 		    (long long)t4_read_reg64(sc, reg + 24),
1346 		    (long long)t4_read_reg64(sc, reg + 32),
1347 		    (long long)t4_read_reg64(sc, reg + 40),
1348 		    (long long)t4_read_reg64(sc, reg + 48),
1349 		    (long long)t4_read_reg64(sc, reg + 56));
1350 	}
1351 }
1352 
1353 /* t4_main.c */
1354 extern int t4_ntxq;
1355 extern int t4_nrxq;
1356 extern int t4_intr_types;
1357 extern int t4_tmr_idx;
1358 extern int t4_pktc_idx;
1359 extern unsigned int t4_qsize_rxq;
1360 extern unsigned int t4_qsize_txq;
1361 extern int t4_ddp_rcvbuf_len;
1362 extern unsigned int t4_ddp_rcvbuf_cache;
1363 extern device_method_t cxgbe_methods[];
1364 
1365 int t4_os_find_pci_capability(struct adapter *, int);
1366 int t4_os_pci_save_state(struct adapter *);
1367 int t4_os_pci_restore_state(struct adapter *);
1368 void t4_os_portmod_changed(struct port_info *);
1369 void t4_os_link_changed(struct port_info *);
1370 void t4_iterate(void (*)(struct adapter *, void *), void *);
1371 void t4_init_devnames(struct adapter *);
1372 void t4_add_adapter(struct adapter *);
1373 int t4_detach_common(device_t);
1374 int t4_map_bars_0_and_4(struct adapter *);
1375 int t4_map_bar_2(struct adapter *);
1376 int t4_adj_doorbells(struct adapter *);
1377 int t4_setup_intr_handlers(struct adapter *);
1378 void t4_sysctls(struct adapter *);
1379 int begin_synchronized_op(struct adapter *, struct vi_info *, int, char *);
1380 void end_synchronized_op(struct adapter *, int);
1381 void begin_vi_detach(struct adapter *, struct vi_info *);
1382 void end_vi_detach(struct adapter *, struct vi_info *);
1383 int update_mac_settings(if_t, int);
1384 int adapter_init(struct adapter *);
1385 int vi_init(struct vi_info *);
1386 void vi_sysctls(struct vi_info *);
1387 int rw_via_memwin(struct adapter *, int, uint32_t, uint32_t *, int, int);
1388 int alloc_atid(struct adapter *, void *);
1389 void *lookup_atid(struct adapter *, int);
1390 void free_atid(struct adapter *, int);
1391 void release_tid(struct adapter *, int, struct sge_wrq *);
1392 int cxgbe_media_change(if_t);
1393 void cxgbe_media_status(if_t, struct ifmediareq *);
1394 void t4_os_cim_err(struct adapter *);
1395 int suspend_adapter(struct adapter *);
1396 int resume_adapter(struct adapter *);
1397 int toe_capability(struct vi_info *, bool);
1398 
1399 #ifdef KERN_TLS
1400 /* t6_kern_tls.c */
1401 int t6_tls_tag_alloc(if_t, union if_snd_tag_alloc_params *,
1402     struct m_snd_tag **);
1403 void t6_ktls_modload(void);
1404 void t6_ktls_modunload(void);
1405 int t6_ktls_try(if_t, struct socket *, struct ktls_session *);
1406 int t6_ktls_parse_pkt(struct mbuf *);
1407 int t6_ktls_write_wr(struct sge_txq *, void *, struct mbuf *, u_int);
1408 #endif
1409 
1410 /* t4_keyctx.c */
1411 struct auth_hash;
1412 union authctx;
1413 #ifdef KERN_TLS
1414 struct ktls_session;
1415 struct tls_key_req;
1416 struct tls_keyctx;
1417 #endif
1418 
1419 void t4_aes_getdeckey(void *, const void *, unsigned int);
1420 void t4_copy_partial_hash(int, union authctx *, void *);
1421 void t4_init_gmac_hash(const char *, int, char *);
1422 void t4_init_hmac_digest(const struct auth_hash *, u_int, const char *, int,
1423     char *);
1424 #ifdef KERN_TLS
1425 u_int t4_tls_key_info_size(const struct ktls_session *);
1426 int t4_tls_proto_ver(const struct ktls_session *);
1427 int t4_tls_cipher_mode(const struct ktls_session *);
1428 int t4_tls_auth_mode(const struct ktls_session *);
1429 int t4_tls_hmac_ctrl(const struct ktls_session *);
1430 void t4_tls_key_ctx(const struct ktls_session *, int, struct tls_keyctx *);
1431 int t4_alloc_tls_keyid(struct adapter *);
1432 void t4_free_tls_keyid(struct adapter *, int);
1433 void t4_write_tlskey_wr(const struct ktls_session *, int, int, int, int,
1434     struct tls_key_req *);
1435 #endif
1436 
1437 #ifdef DEV_NETMAP
1438 /* t4_netmap.c */
1439 struct sge_nm_rxq;
1440 void cxgbe_nm_attach(struct vi_info *);
1441 void cxgbe_nm_detach(struct vi_info *);
1442 void service_nm_rxq(struct sge_nm_rxq *);
1443 int alloc_nm_rxq(struct vi_info *, struct sge_nm_rxq *, int, int);
1444 int free_nm_rxq(struct vi_info *, struct sge_nm_rxq *);
1445 int alloc_nm_txq(struct vi_info *, struct sge_nm_txq *, int, int);
1446 int free_nm_txq(struct vi_info *, struct sge_nm_txq *);
1447 #endif
1448 
1449 /* t4_sge.c */
1450 void t4_sge_modload(void);
1451 void t4_sge_modunload(void);
1452 uint64_t t4_sge_extfree_refs(void);
1453 void t4_tweak_chip_settings(struct adapter *);
1454 int t4_verify_chip_settings(struct adapter *);
1455 void t4_init_rx_buf_info(struct adapter *);
1456 int t4_create_dma_tag(struct adapter *);
1457 void t4_sge_sysctls(struct adapter *, struct sysctl_ctx_list *,
1458     struct sysctl_oid_list *);
1459 int t4_destroy_dma_tag(struct adapter *);
1460 int alloc_ring(struct adapter *, size_t, bus_dma_tag_t *, bus_dmamap_t *,
1461     bus_addr_t *, void **);
1462 int free_ring(struct adapter *, bus_dma_tag_t, bus_dmamap_t, bus_addr_t,
1463     void *);
1464 void free_fl_buffers(struct adapter *, struct sge_fl *);
1465 int t4_setup_adapter_queues(struct adapter *);
1466 int t4_teardown_adapter_queues(struct adapter *);
1467 int t4_setup_vi_queues(struct vi_info *);
1468 int t4_teardown_vi_queues(struct vi_info *);
1469 void t4_intr_all(void *);
1470 void t4_intr(void *);
1471 #ifdef DEV_NETMAP
1472 void t4_nm_intr(void *);
1473 void t4_vi_intr(void *);
1474 #endif
1475 void t4_intr_err(void *);
1476 void t4_intr_evt(void *);
1477 void t4_wrq_tx_locked(struct adapter *, struct sge_wrq *, struct wrqe *);
1478 void t4_update_fl_bufsize(if_t);
1479 struct mbuf *alloc_wr_mbuf(int, int);
1480 int parse_pkt(struct mbuf **, bool);
1481 void *start_wrq_wr(struct sge_wrq *, int, struct wrq_cookie *);
1482 void commit_wrq_wr(struct sge_wrq *, void *, struct wrq_cookie *);
1483 int t4_sge_set_conm_context(struct adapter *, int, int, int);
1484 void t4_register_an_handler(an_handler_t);
1485 void t4_register_fw_msg_handler(int, fw_msg_handler_t);
1486 void t4_register_cpl_handler(int, cpl_handler_t);
1487 void t4_register_shared_cpl_handler(int, cpl_handler_t, int);
1488 #ifdef RATELIMIT
1489 void send_etid_flush_wr(struct cxgbe_rate_tag *);
1490 #endif
1491 
1492 /* t4_tracer.c */
1493 struct t4_tracer;
1494 void t4_tracer_modload(void);
1495 void t4_tracer_modunload(void);
1496 void t4_tracer_port_detach(struct adapter *);
1497 int t4_get_tracer(struct adapter *, struct t4_tracer *);
1498 int t4_set_tracer(struct adapter *, struct t4_tracer *);
1499 int t4_trace_pkt(struct sge_iq *, const struct rss_header *, struct mbuf *);
1500 int t5_trace_pkt(struct sge_iq *, const struct rss_header *, struct mbuf *);
1501 
1502 /* t4_sched.c */
1503 int t4_set_sched_class(struct adapter *, struct t4_sched_params *);
1504 int t4_set_sched_queue(struct adapter *, struct t4_sched_queue *);
1505 int t4_init_tx_sched(struct adapter *);
1506 int t4_free_tx_sched(struct adapter *);
1507 void t4_update_tx_sched(struct adapter *);
1508 int t4_reserve_cl_rl_kbps(struct adapter *, int, u_int, int *);
1509 void t4_release_cl_rl(struct adapter *, int, int);
1510 int sysctl_tc(SYSCTL_HANDLER_ARGS);
1511 int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
1512 #ifdef RATELIMIT
1513 void t4_init_etid_table(struct adapter *);
1514 void t4_free_etid_table(struct adapter *);
1515 struct cxgbe_rate_tag *lookup_etid(struct adapter *, int);
1516 int cxgbe_rate_tag_alloc(if_t, union if_snd_tag_alloc_params *,
1517     struct m_snd_tag **);
1518 void cxgbe_rate_tag_free_locked(struct cxgbe_rate_tag *);
1519 void cxgbe_ratelimit_query(if_t, struct if_ratelimit_query_results *);
1520 #endif
1521 
1522 /* t4_filter.c */
1523 int get_filter_mode(struct adapter *, uint32_t *);
1524 int set_filter_mode(struct adapter *, uint32_t);
1525 int set_filter_mask(struct adapter *, uint32_t);
1526 int get_filter(struct adapter *, struct t4_filter *);
1527 int set_filter(struct adapter *, struct t4_filter *);
1528 int del_filter(struct adapter *, struct t4_filter *);
1529 int t4_filter_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1530 int t4_hashfilter_ao_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1531 int t4_hashfilter_tcb_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1532 int t4_del_hashfilter_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1533 void free_hftid_hash(struct tid_info *);
1534 
1535 static inline struct wrqe *
1536 alloc_wrqe(int wr_len, struct sge_wrq *wrq)
1537 {
1538 	int len = offsetof(struct wrqe, wr) + wr_len;
1539 	struct wrqe *wr;
1540 
1541 	wr = malloc(len, M_CXGBE, M_NOWAIT);
1542 	if (__predict_false(wr == NULL))
1543 		return (NULL);
1544 	wr->wr_len = wr_len;
1545 	wr->wrq = wrq;
1546 	return (wr);
1547 }
1548 
1549 static inline void *
1550 wrtod(struct wrqe *wr)
1551 {
1552 	return (&wr->wr[0]);
1553 }
1554 
1555 static inline void
1556 free_wrqe(struct wrqe *wr)
1557 {
1558 	free(wr, M_CXGBE);
1559 }
1560 
1561 static inline void
1562 t4_wrq_tx(struct adapter *sc, struct wrqe *wr)
1563 {
1564 	struct sge_wrq *wrq = wr->wrq;
1565 
1566 	TXQ_LOCK(wrq);
1567 	if (__predict_true(wrq->eq.flags & EQ_HW_ALLOCATED))
1568 		t4_wrq_tx_locked(sc, wrq, wr);
1569 	else
1570 		free(wr, M_CXGBE);
1571 	TXQ_UNLOCK(wrq);
1572 }
1573 
1574 static inline int
1575 read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
1576     int len)
1577 {
1578 
1579 	return (rw_via_memwin(sc, idx, addr, val, len, 0));
1580 }
1581 
1582 static inline int
1583 write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
1584     const uint32_t *val, int len)
1585 {
1586 
1587 	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
1588 }
1589 
1590 /* Number of len16 -> number of descriptors */
1591 static inline int
1592 tx_len16_to_desc(int len16)
1593 {
1594 
1595 	return (howmany(len16, EQ_ESIZE / 16));
1596 }
1597 #endif
1598