xref: /freebsd/sys/dev/cxgbe/adapter.h (revision dd41de95a84d979615a2ef11df6850622bf6184e)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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  * $FreeBSD$
30  *
31  */
32 
33 #ifndef __T4_ADAPTER_H__
34 #define __T4_ADAPTER_H__
35 
36 #include <sys/kernel.h>
37 #include <sys/bus.h>
38 #include <sys/counter.h>
39 #include <sys/rman.h>
40 #include <sys/types.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/rwlock.h>
44 #include <sys/sx.h>
45 #include <sys/vmem.h>
46 #include <vm/uma.h>
47 
48 #include <dev/pci/pcivar.h>
49 #include <dev/pci/pcireg.h>
50 #include <machine/bus.h>
51 #include <sys/socket.h>
52 #include <sys/sysctl.h>
53 #include <sys/taskqueue.h>
54 #include <net/ethernet.h>
55 #include <net/if.h>
56 #include <net/if_var.h>
57 #include <net/if_media.h>
58 #include <net/pfil.h>
59 #include <netinet/in.h>
60 #include <netinet/tcp_lro.h>
61 
62 #include "offload.h"
63 #include "t4_ioctl.h"
64 #include "common/t4_msg.h"
65 #include "firmware/t4fw_interface.h"
66 
67 #define KTR_CXGBE	KTR_SPARE3
68 MALLOC_DECLARE(M_CXGBE);
69 #define CXGBE_UNIMPLEMENTED(s) \
70     panic("%s (%s, line %d) not implemented yet.", s, __FILE__, __LINE__)
71 
72 /*
73  * Same as LIST_HEAD from queue.h.  This is to avoid conflict with LinuxKPI's
74  * LIST_HEAD when building iw_cxgbe.
75  */
76 #define	CXGBE_LIST_HEAD(name, type)					\
77 struct name {								\
78 	struct type *lh_first;	/* first element */			\
79 }
80 
81 #ifndef SYSCTL_ADD_UQUAD
82 #define SYSCTL_ADD_UQUAD SYSCTL_ADD_QUAD
83 #define sysctl_handle_64 sysctl_handle_quad
84 #define CTLTYPE_U64 CTLTYPE_QUAD
85 #endif
86 
87 SYSCTL_DECL(_hw_cxgbe);
88 
89 struct adapter;
90 typedef struct adapter adapter_t;
91 
92 enum {
93 	/*
94 	 * All ingress queues use this entry size.  Note that the firmware event
95 	 * queue and any iq expecting CPL_RX_PKT in the descriptor needs this to
96 	 * be at least 64.
97 	 */
98 	IQ_ESIZE = 64,
99 
100 	/* Default queue sizes for all kinds of ingress queues */
101 	FW_IQ_QSIZE = 256,
102 	RX_IQ_QSIZE = 1024,
103 
104 	/* All egress queues use this entry size */
105 	EQ_ESIZE = 64,
106 
107 	/* Default queue sizes for all kinds of egress queues */
108 	CTRL_EQ_QSIZE = 1024,
109 	TX_EQ_QSIZE = 1024,
110 
111 #if MJUMPAGESIZE != MCLBYTES
112 	SW_ZONE_SIZES = 4,	/* cluster, jumbop, jumbo9k, jumbo16k */
113 #else
114 	SW_ZONE_SIZES = 3,	/* cluster, jumbo9k, jumbo16k */
115 #endif
116 	CL_METADATA_SIZE = CACHE_LINE_SIZE,
117 
118 	SGE_MAX_WR_NDESC = SGE_MAX_WR_LEN / EQ_ESIZE, /* max WR size in desc */
119 	TX_SGL_SEGS = 39,
120 	TX_SGL_SEGS_TSO = 38,
121 	TX_SGL_SEGS_VM = 38,
122 	TX_SGL_SEGS_VM_TSO = 37,
123 	TX_SGL_SEGS_EO_TSO = 30,	/* XXX: lower for IPv6. */
124 	TX_SGL_SEGS_VXLAN_TSO = 37,
125 	TX_WR_FLITS = SGE_MAX_WR_LEN / 8
126 };
127 
128 enum {
129 	/* adapter intr_type */
130 	INTR_INTX	= (1 << 0),
131 	INTR_MSI 	= (1 << 1),
132 	INTR_MSIX	= (1 << 2)
133 };
134 
135 enum {
136 	XGMAC_MTU	= (1 << 0),
137 	XGMAC_PROMISC	= (1 << 1),
138 	XGMAC_ALLMULTI	= (1 << 2),
139 	XGMAC_VLANEX	= (1 << 3),
140 	XGMAC_UCADDR	= (1 << 4),
141 	XGMAC_MCADDRS	= (1 << 5),
142 
143 	XGMAC_ALL	= 0xffff
144 };
145 
146 enum {
147 	/* flags understood by begin_synchronized_op */
148 	HOLD_LOCK	= (1 << 0),
149 	SLEEP_OK	= (1 << 1),
150 	INTR_OK		= (1 << 2),
151 
152 	/* flags understood by end_synchronized_op */
153 	LOCK_HELD	= HOLD_LOCK,
154 };
155 
156 enum {
157 	/* adapter flags */
158 	FULL_INIT_DONE	= (1 << 0),
159 	FW_OK		= (1 << 1),
160 	CHK_MBOX_ACCESS	= (1 << 2),
161 	MASTER_PF	= (1 << 3),
162 	ADAP_SYSCTL_CTX	= (1 << 4),
163 	ADAP_ERR	= (1 << 5),
164 	BUF_PACKING_OK	= (1 << 6),
165 	IS_VF		= (1 << 7),
166 	KERN_TLS_ON	= (1 << 8),	/* HW is configured for KERN_TLS */
167 	CXGBE_BUSY	= (1 << 9),
168 	HW_OFF_LIMITS	= (1 << 10),	/* off limits to all except reset_thread */
169 
170 	/* port flags */
171 	HAS_TRACEQ	= (1 << 3),
172 	FIXED_IFMEDIA	= (1 << 4),	/* ifmedia list doesn't change. */
173 
174 	/* VI flags */
175 	DOOMED		= (1 << 0),
176 	VI_INIT_DONE	= (1 << 1),
177 	VI_SYSCTL_CTX	= (1 << 2),
178 	TX_USES_VM_WR 	= (1 << 3),
179 	VI_SKIP_STATS 	= (1 << 4),
180 
181 	/* adapter debug_flags */
182 	DF_DUMP_MBOX		= (1 << 0),	/* Log all mbox cmd/rpl. */
183 	DF_LOAD_FW_ANYTIME	= (1 << 1),	/* Allow LOAD_FW after init */
184 	DF_DISABLE_TCB_CACHE	= (1 << 2),	/* Disable TCB cache (T6+) */
185 	DF_DISABLE_CFG_RETRY	= (1 << 3),	/* Disable fallback config */
186 	DF_VERBOSE_SLOWINTR	= (1 << 4),	/* Chatty slow intr handler */
187 };
188 
189 #define IS_DOOMED(vi)	((vi)->flags & DOOMED)
190 #define SET_DOOMED(vi)	do {(vi)->flags |= DOOMED;} while (0)
191 #define IS_BUSY(sc)	((sc)->flags & CXGBE_BUSY)
192 #define SET_BUSY(sc)	do {(sc)->flags |= CXGBE_BUSY;} while (0)
193 #define CLR_BUSY(sc)	do {(sc)->flags &= ~CXGBE_BUSY;} while (0)
194 
195 struct vi_info {
196 	device_t dev;
197 	struct port_info *pi;
198 	struct adapter *adapter;
199 
200 	struct ifnet *ifp;
201 	struct pfil_head *pfil;
202 
203 	unsigned long flags;
204 	int if_flags;
205 
206 	uint16_t *rss, *nm_rss;
207 	uint16_t viid;		/* opaque VI identifier */
208 	uint16_t smt_idx;
209 	uint16_t vin;
210 	uint8_t vfvld;
211 	int16_t  xact_addr_filt;/* index of exact MAC address filter */
212 	uint16_t rss_size;	/* size of VI's RSS table slice */
213 	uint16_t rss_base;	/* start of VI's RSS table slice */
214 	int hashen;
215 
216 	int nintr;
217 	int first_intr;
218 
219 	/* These need to be int as they are used in sysctl */
220 	int ntxq;		/* # of tx queues */
221 	int first_txq;		/* index of first tx queue */
222 	int rsrv_noflowq; 	/* Reserve queue 0 for non-flowid packets */
223 	int nrxq;		/* # of rx queues */
224 	int first_rxq;		/* index of first rx queue */
225 	int nofldtxq;		/* # of offload tx queues */
226 	int first_ofld_txq;	/* index of first offload tx queue */
227 	int nofldrxq;		/* # of offload rx queues */
228 	int first_ofld_rxq;	/* index of first offload rx queue */
229 	int nnmtxq;
230 	int first_nm_txq;
231 	int nnmrxq;
232 	int first_nm_rxq;
233 	int tmr_idx;
234 	int ofld_tmr_idx;
235 	int pktc_idx;
236 	int ofld_pktc_idx;
237 	int qsize_rxq;
238 	int qsize_txq;
239 
240 	struct timeval last_refreshed;
241 	struct fw_vi_stats_vf stats;
242 	struct mtx tick_mtx;
243 	struct callout tick;
244 
245 	struct sysctl_ctx_list ctx;
246 	struct sysctl_oid *rxq_oid;
247 	struct sysctl_oid *txq_oid;
248 	struct sysctl_oid *nm_rxq_oid;
249 	struct sysctl_oid *nm_txq_oid;
250 	struct sysctl_oid *ofld_rxq_oid;
251 	struct sysctl_oid *ofld_txq_oid;
252 
253 	uint8_t hw_addr[ETHER_ADDR_LEN]; /* factory MAC address, won't change */
254 };
255 
256 struct tx_ch_rl_params {
257 	enum fw_sched_params_rate ratemode;	/* %port (REL) or kbps (ABS) */
258 	uint32_t maxrate;
259 };
260 
261 enum {
262 	CLRL_USER	= (1 << 0),	/* allocated manually. */
263 	CLRL_SYNC	= (1 << 1),	/* sync hw update in progress. */
264 	CLRL_ASYNC	= (1 << 2),	/* async hw update requested. */
265 	CLRL_ERR	= (1 << 3),	/* last hw setup ended in error. */
266 };
267 
268 struct tx_cl_rl_params {
269 	int refcount;
270 	uint8_t flags;
271 	enum fw_sched_params_rate ratemode;	/* %port REL or ABS value */
272 	enum fw_sched_params_unit rateunit;	/* kbps or pps (when ABS) */
273 	enum fw_sched_params_mode mode;		/* aggr or per-flow */
274 	uint32_t maxrate;
275 	uint16_t pktsize;
276 	uint16_t burstsize;
277 };
278 
279 /* Tx scheduler parameters for a channel/port */
280 struct tx_sched_params {
281 	/* Channel Rate Limiter */
282 	struct tx_ch_rl_params ch_rl;
283 
284 	/* Class WRR */
285 	/* XXX */
286 
287 	/* Class Rate Limiter (including the default pktsize and burstsize). */
288 	int pktsize;
289 	int burstsize;
290 	struct tx_cl_rl_params cl_rl[];
291 };
292 
293 struct port_info {
294 	device_t dev;
295 	struct adapter *adapter;
296 
297 	struct vi_info *vi;
298 	int nvi;
299 	int up_vis;
300 	int uld_vis;
301 	bool vxlan_tcam_entry;
302 
303 	struct tx_sched_params *sched_params;
304 
305 	struct mtx pi_lock;
306 	char lockname[16];
307 	unsigned long flags;
308 
309 	uint8_t  lport;		/* associated offload logical port */
310 	int8_t   mdio_addr;
311 	uint8_t  port_type;
312 	uint8_t  mod_type;
313 	uint8_t  port_id;
314 	uint8_t  tx_chan;
315 	uint8_t  mps_bg_map;	/* rx MPS buffer group bitmap */
316 	uint8_t  rx_e_chan_map;	/* rx TP e-channel bitmap */
317 	uint8_t  rx_c_chan;	/* rx TP c-channel */
318 
319 	struct link_config link_cfg;
320 	struct ifmedia media;
321 
322  	struct port_stats stats;
323 	u_int tnl_cong_drops;
324 	u_int tx_parse_error;
325 	int fcs_reg;
326 	uint64_t fcs_base;
327 };
328 
329 #define	IS_MAIN_VI(vi)		((vi) == &((vi)->pi->vi[0]))
330 
331 struct cluster_metadata {
332 	uma_zone_t zone;
333 	caddr_t cl;
334 	u_int refcount;
335 };
336 
337 struct fl_sdesc {
338 	caddr_t cl;
339 	uint16_t nmbuf;	/* # of driver originated mbufs with ref on cluster */
340 	int16_t moff;	/* offset of metadata from cl */
341 	uint8_t zidx;
342 };
343 
344 struct tx_desc {
345 	__be64 flit[8];
346 };
347 
348 struct tx_sdesc {
349 	struct mbuf *m;		/* m_nextpkt linked chain of frames */
350 	uint8_t desc_used;	/* # of hardware descriptors used by the WR */
351 };
352 
353 
354 #define IQ_PAD (IQ_ESIZE - sizeof(struct rsp_ctrl) - sizeof(struct rss_header))
355 struct iq_desc {
356 	struct rss_header rss;
357 	uint8_t cpl[IQ_PAD];
358 	struct rsp_ctrl rsp;
359 };
360 #undef IQ_PAD
361 CTASSERT(sizeof(struct iq_desc) == IQ_ESIZE);
362 
363 enum {
364 	/* iq flags */
365 	IQ_SW_ALLOCATED	= (1 << 0),	/* sw resources allocated */
366 	IQ_HAS_FL	= (1 << 1),	/* iq associated with a freelist */
367 	IQ_RX_TIMESTAMP	= (1 << 2),	/* provide the SGE rx timestamp */
368 	IQ_LRO_ENABLED	= (1 << 3),	/* iq is an eth rxq with LRO enabled */
369 	IQ_ADJ_CREDIT	= (1 << 4),	/* hw is off by 1 credit for this iq */
370 	IQ_HW_ALLOCATED	= (1 << 5),	/* fw/hw resources allocated */
371 
372 	/* iq state */
373 	IQS_DISABLED	= 0,
374 	IQS_BUSY	= 1,
375 	IQS_IDLE	= 2,
376 
377 	/* netmap related flags */
378 	NM_OFF	= 0,
379 	NM_ON	= 1,
380 	NM_BUSY	= 2,
381 };
382 
383 enum {
384 	CPL_COOKIE_RESERVED = 0,
385 	CPL_COOKIE_FILTER,
386 	CPL_COOKIE_DDP0,
387 	CPL_COOKIE_DDP1,
388 	CPL_COOKIE_TOM,
389 	CPL_COOKIE_HASHFILTER,
390 	CPL_COOKIE_ETHOFLD,
391 	CPL_COOKIE_KERN_TLS,
392 
393 	NUM_CPL_COOKIES = 8	/* Limited by M_COOKIE.  Do not increase. */
394 };
395 
396 struct sge_iq;
397 struct rss_header;
398 typedef int (*cpl_handler_t)(struct sge_iq *, const struct rss_header *,
399     struct mbuf *);
400 typedef int (*an_handler_t)(struct sge_iq *, const struct rsp_ctrl *);
401 typedef int (*fw_msg_handler_t)(struct adapter *, const __be64 *);
402 
403 /*
404  * Ingress Queue: T4 is producer, driver is consumer.
405  */
406 struct sge_iq {
407 	uint32_t flags;
408 	volatile int state;
409 	struct adapter *adapter;
410 	struct iq_desc  *desc;	/* KVA of descriptor ring */
411 	int8_t   intr_pktc_idx;	/* packet count threshold index */
412 	uint8_t  gen;		/* generation bit */
413 	uint8_t  intr_params;	/* interrupt holdoff parameters */
414 	int8_t   cong;		/* congestion settings */
415 	uint16_t qsize;		/* size (# of entries) of the queue */
416 	uint16_t sidx;		/* index of the entry with the status page */
417 	uint16_t cidx;		/* consumer index */
418 	uint16_t cntxt_id;	/* SGE context id for the iq */
419 	uint16_t abs_id;	/* absolute SGE id for the iq */
420 	int16_t intr_idx;	/* interrupt used by the queue */
421 
422 	STAILQ_ENTRY(sge_iq) link;
423 
424 	bus_dma_tag_t desc_tag;
425 	bus_dmamap_t desc_map;
426 	bus_addr_t ba;		/* bus address of descriptor ring */
427 };
428 
429 enum {
430 	/* eq type */
431 	EQ_CTRL		= 1,
432 	EQ_ETH		= 2,
433 	EQ_OFLD		= 3,
434 
435 	/* eq flags */
436 	EQ_SW_ALLOCATED	= (1 << 0),	/* sw resources allocated */
437 	EQ_HW_ALLOCATED	= (1 << 1),	/* hw/fw resources allocated */
438 	EQ_ENABLED	= (1 << 3),	/* open for business */
439 	EQ_QFLUSH	= (1 << 4),	/* if_qflush in progress */
440 };
441 
442 /* Listed in order of preference.  Update t4_sysctls too if you change these */
443 enum {DOORBELL_UDB, DOORBELL_WCWR, DOORBELL_UDBWC, DOORBELL_KDB};
444 
445 /*
446  * Egress Queue: driver is producer, T4 is consumer.
447  *
448  * Note: A free list is an egress queue (driver produces the buffers and T4
449  * consumes them) but it's special enough to have its own struct (see sge_fl).
450  */
451 struct sge_eq {
452 	unsigned int flags;	/* MUST be first */
453 	unsigned int cntxt_id;	/* SGE context id for the eq */
454 	unsigned int abs_id;	/* absolute SGE id for the eq */
455 	uint8_t type;		/* EQ_CTRL/EQ_ETH/EQ_OFLD */
456 	uint8_t doorbells;
457 	uint8_t tx_chan;	/* tx channel used by the eq */
458 	struct mtx eq_lock;
459 
460 	struct tx_desc *desc;	/* KVA of descriptor ring */
461 	volatile uint32_t *udb;	/* KVA of doorbell (lies within BAR2) */
462 	u_int udb_qid;		/* relative qid within the doorbell page */
463 	uint16_t sidx;		/* index of the entry with the status page */
464 	uint16_t cidx;		/* consumer idx (desc idx) */
465 	uint16_t pidx;		/* producer idx (desc idx) */
466 	uint16_t equeqidx;	/* EQUEQ last requested at this pidx */
467 	uint16_t dbidx;		/* pidx of the most recent doorbell */
468 	uint16_t iqid;		/* cached iq->cntxt_id (see iq below) */
469 	volatile u_int equiq;	/* EQUIQ outstanding */
470 	struct sge_iq *iq;	/* iq that receives egr_update for the eq */
471 
472 	bus_dma_tag_t desc_tag;
473 	bus_dmamap_t desc_map;
474 	bus_addr_t ba;		/* bus address of descriptor ring */
475 	char lockname[16];
476 };
477 
478 struct rx_buf_info {
479 	uma_zone_t zone;	/* zone that this cluster comes from */
480 	uint16_t size1;		/* same as size of cluster: 2K/4K/9K/16K.
481 				 * hwsize[hwidx1] = size1.  No spare. */
482 	uint16_t size2;		/* hwsize[hwidx2] = size2.
483 				 * spare in cluster = size1 - size2. */
484 	int8_t hwidx1;		/* SGE bufsize idx for size1 */
485 	int8_t hwidx2;		/* SGE bufsize idx for size2 */
486 	uint8_t type;		/* EXT_xxx type of the cluster */
487 };
488 
489 enum {
490 	NUM_MEMWIN = 3,
491 
492 	MEMWIN0_APERTURE = 2048,
493 	MEMWIN0_BASE     = 0x1b800,
494 
495 	MEMWIN1_APERTURE = 32768,
496 	MEMWIN1_BASE     = 0x28000,
497 
498 	MEMWIN2_APERTURE_T4 = 65536,
499 	MEMWIN2_BASE_T4     = 0x30000,
500 
501 	MEMWIN2_APERTURE_T5 = 128 * 1024,
502 	MEMWIN2_BASE_T5     = 0x60000,
503 };
504 
505 struct memwin {
506 	struct rwlock mw_lock __aligned(CACHE_LINE_SIZE);
507 	uint32_t mw_base;	/* constant after setup_memwin */
508 	uint32_t mw_aperture;	/* ditto */
509 	uint32_t mw_curpos;	/* protected by mw_lock */
510 };
511 
512 enum {
513 	FL_STARVING	= (1 << 0), /* on the adapter's list of starving fl's */
514 	FL_DOOMED	= (1 << 1), /* about to be destroyed */
515 	FL_BUF_PACKING	= (1 << 2), /* buffer packing enabled */
516 	FL_BUF_RESUME	= (1 << 3), /* resume from the middle of the frame */
517 };
518 
519 #define FL_RUNNING_LOW(fl) \
520     (IDXDIFF(fl->dbidx * 8, fl->cidx, fl->sidx * 8) <= fl->lowat)
521 #define FL_NOT_RUNNING_LOW(fl) \
522     (IDXDIFF(fl->dbidx * 8, fl->cidx, fl->sidx * 8) >= 2 * fl->lowat)
523 
524 struct sge_fl {
525 	struct mtx fl_lock;
526 	__be64 *desc;		/* KVA of descriptor ring, ptr to addresses */
527 	struct fl_sdesc *sdesc;	/* KVA of software descriptor ring */
528 	uint16_t zidx;		/* refill zone idx */
529 	uint16_t safe_zidx;
530 	uint16_t lowat;		/* # of buffers <= this means fl needs help */
531 	int flags;
532 	uint16_t buf_boundary;
533 
534 	/* The 16b idx all deal with hw descriptors */
535 	uint16_t dbidx;		/* hw pidx after last doorbell */
536 	uint16_t sidx;		/* index of status page */
537 	volatile uint16_t hw_cidx;
538 
539 	/* The 32b idx are all buffer idx, not hardware descriptor idx */
540 	uint32_t cidx;		/* consumer index */
541 	uint32_t pidx;		/* producer index */
542 
543 	uint32_t dbval;
544 	u_int rx_offset;	/* offset in fl buf (when buffer packing) */
545 	volatile uint32_t *udb;
546 
547 	uint64_t cl_allocated;	/* # of clusters allocated */
548 	uint64_t cl_recycled;	/* # of clusters recycled */
549 	uint64_t cl_fast_recycled; /* # of clusters recycled (fast) */
550 
551 	/* These 3 are valid when FL_BUF_RESUME is set, stale otherwise. */
552 	struct mbuf *m0;
553 	struct mbuf **pnext;
554 	u_int remaining;
555 
556 	uint16_t qsize;		/* # of hw descriptors (status page included) */
557 	uint16_t cntxt_id;	/* SGE context id for the freelist */
558 	TAILQ_ENTRY(sge_fl) link; /* All starving freelists */
559 	bus_dma_tag_t desc_tag;
560 	bus_dmamap_t desc_map;
561 	char lockname[16];
562 	bus_addr_t ba;		/* bus address of descriptor ring */
563 };
564 
565 struct mp_ring;
566 
567 struct txpkts {
568 	uint8_t wr_type;	/* type 0 or type 1 */
569 	uint8_t npkt;		/* # of packets in this work request */
570 	uint8_t len16;		/* # of 16B pieces used by this work request */
571 	uint8_t score;
572 	uint8_t max_npkt;	/* maximum number of packets allowed */
573 	uint16_t plen;		/* total payload (sum of all packets) */
574 
575 	/* straight from fw_eth_tx_pkts_vm_wr. */
576 	__u8   ethmacdst[6];
577 	__u8   ethmacsrc[6];
578 	__be16 ethtype;
579 	__be16 vlantci;
580 
581 	struct mbuf *mb[15];
582 };
583 
584 /* txq: SGE egress queue + what's needed for Ethernet NIC */
585 struct sge_txq {
586 	struct sge_eq eq;	/* MUST be first */
587 
588 	struct ifnet *ifp;	/* the interface this txq belongs to */
589 	struct mp_ring *r;	/* tx software ring */
590 	struct tx_sdesc *sdesc;	/* KVA of software descriptor ring */
591 	struct sglist *gl;
592 	__be32 cpl_ctrl0;	/* for convenience */
593 	int tc_idx;		/* traffic class */
594 	uint64_t last_tx;	/* cycle count when eth_tx was last called */
595 	struct txpkts txp;
596 
597 	struct task tx_reclaim_task;
598 	/* stats for common events first */
599 
600 	uint64_t txcsum;	/* # of times hardware assisted with checksum */
601 	uint64_t tso_wrs;	/* # of TSO work requests */
602 	uint64_t vlan_insertion;/* # of times VLAN tag was inserted */
603 	uint64_t imm_wrs;	/* # of work requests with immediate data */
604 	uint64_t sgl_wrs;	/* # of work requests with direct SGL */
605 	uint64_t txpkt_wrs;	/* # of txpkt work requests (not coalesced) */
606 	uint64_t txpkts0_wrs;	/* # of type0 coalesced tx work requests */
607 	uint64_t txpkts1_wrs;	/* # of type1 coalesced tx work requests */
608 	uint64_t txpkts0_pkts;	/* # of frames in type0 coalesced tx WRs */
609 	uint64_t txpkts1_pkts;	/* # of frames in type1 coalesced tx WRs */
610 	uint64_t txpkts_flush;	/* # of times txp had to be sent by tx_update */
611 	uint64_t raw_wrs;	/* # of raw work requests (alloc_wr_mbuf) */
612 	uint64_t vxlan_tso_wrs;	/* # of VXLAN TSO work requests */
613 	uint64_t vxlan_txcsum;
614 
615 	uint64_t kern_tls_records;
616 	uint64_t kern_tls_short;
617 	uint64_t kern_tls_partial;
618 	uint64_t kern_tls_full;
619 	uint64_t kern_tls_octets;
620 	uint64_t kern_tls_waste;
621 	uint64_t kern_tls_options;
622 	uint64_t kern_tls_header;
623 	uint64_t kern_tls_fin;
624 	uint64_t kern_tls_fin_short;
625 	uint64_t kern_tls_cbc;
626 	uint64_t kern_tls_gcm;
627 
628 	/* stats for not-that-common events */
629 
630 	/* Optional scratch space for constructing work requests. */
631 	uint8_t ss[SGE_MAX_WR_LEN] __aligned(16);
632 } __aligned(CACHE_LINE_SIZE);
633 
634 /* rxq: SGE ingress queue + SGE free list + miscellaneous items */
635 struct sge_rxq {
636 	struct sge_iq iq;	/* MUST be first */
637 	struct sge_fl fl;	/* MUST follow iq */
638 
639 	struct ifnet *ifp;	/* the interface this rxq belongs to */
640 	struct lro_ctrl lro;	/* LRO state */
641 
642 	/* stats for common events first */
643 
644 	uint64_t rxcsum;	/* # of times hardware assisted with checksum */
645 	uint64_t vlan_extraction;/* # of times VLAN tag was extracted */
646 	uint64_t vxlan_rxcsum;
647 
648 	/* stats for not-that-common events */
649 
650 } __aligned(CACHE_LINE_SIZE);
651 
652 static inline struct sge_rxq *
653 iq_to_rxq(struct sge_iq *iq)
654 {
655 
656 	return (__containerof(iq, struct sge_rxq, iq));
657 }
658 
659 /* ofld_rxq: SGE ingress queue + SGE free list + miscellaneous items */
660 struct sge_ofld_rxq {
661 	struct sge_iq iq;	/* MUST be first */
662 	struct sge_fl fl;	/* MUST follow iq */
663 	counter_u64_t rx_iscsi_ddp_setup_ok;
664 	counter_u64_t rx_iscsi_ddp_setup_error;
665 	uint64_t rx_iscsi_ddp_pdus;
666 	uint64_t rx_iscsi_ddp_octets;
667 	uint64_t rx_iscsi_fl_pdus;
668 	uint64_t rx_iscsi_fl_octets;
669 	u_long	rx_toe_tls_records;
670 	u_long	rx_toe_tls_octets;
671 } __aligned(CACHE_LINE_SIZE);
672 
673 static inline struct sge_ofld_rxq *
674 iq_to_ofld_rxq(struct sge_iq *iq)
675 {
676 
677 	return (__containerof(iq, struct sge_ofld_rxq, iq));
678 }
679 
680 struct wrqe {
681 	STAILQ_ENTRY(wrqe) link;
682 	struct sge_wrq *wrq;
683 	int wr_len;
684 	char wr[] __aligned(16);
685 };
686 
687 struct wrq_cookie {
688 	TAILQ_ENTRY(wrq_cookie) link;
689 	int ndesc;
690 	int pidx;
691 };
692 
693 /*
694  * wrq: SGE egress queue that is given prebuilt work requests.  Control queues
695  * are of this type.
696  */
697 struct sge_wrq {
698 	struct sge_eq eq;	/* MUST be first */
699 
700 	struct adapter *adapter;
701 	struct task wrq_tx_task;
702 
703 	/* Tx desc reserved but WR not "committed" yet. */
704 	TAILQ_HEAD(wrq_incomplete_wrs , wrq_cookie) incomplete_wrs;
705 
706 	/* List of WRs ready to go out as soon as descriptors are available. */
707 	STAILQ_HEAD(, wrqe) wr_list;
708 	u_int nwr_pending;
709 	u_int ndesc_needed;
710 
711 	/* stats for common events first */
712 
713 	uint64_t tx_wrs_direct;	/* # of WRs written directly to desc ring. */
714 	uint64_t tx_wrs_ss;	/* # of WRs copied from scratch space. */
715 	uint64_t tx_wrs_copied;	/* # of WRs queued and copied to desc ring. */
716 
717 	/* stats for not-that-common events */
718 
719 	/*
720 	 * Scratch space for work requests that wrap around after reaching the
721 	 * status page, and some information about the last WR that used it.
722 	 */
723 	uint16_t ss_pidx;
724 	uint16_t ss_len;
725 	uint8_t ss[SGE_MAX_WR_LEN];
726 
727 } __aligned(CACHE_LINE_SIZE);
728 
729 /* ofld_txq: SGE egress queue + miscellaneous items */
730 struct sge_ofld_txq {
731 	struct sge_wrq wrq;
732 	counter_u64_t tx_iscsi_pdus;
733 	counter_u64_t tx_iscsi_octets;
734 	counter_u64_t tx_toe_tls_records;
735 	counter_u64_t tx_toe_tls_octets;
736 } __aligned(CACHE_LINE_SIZE);
737 
738 #define INVALID_NM_RXQ_CNTXT_ID ((uint16_t)(-1))
739 struct sge_nm_rxq {
740 	/* Items used by the driver rx ithread are in this cacheline. */
741 	volatile int nm_state __aligned(CACHE_LINE_SIZE);	/* NM_OFF, NM_ON, or NM_BUSY */
742 	u_int nid;		/* netmap ring # for this queue */
743 	struct vi_info *vi;
744 
745 	struct iq_desc *iq_desc;
746 	uint16_t iq_abs_id;
747 	uint16_t iq_cntxt_id;
748 	uint16_t iq_cidx;
749 	uint16_t iq_sidx;
750 	uint8_t iq_gen;
751 	uint32_t fl_sidx;
752 
753 	/* Items used by netmap rxsync are in this cacheline. */
754 	__be64  *fl_desc __aligned(CACHE_LINE_SIZE);
755 	uint16_t fl_cntxt_id;
756 	uint32_t fl_pidx;
757 	uint32_t fl_sidx2;	/* copy of fl_sidx */
758 	uint32_t fl_db_val;
759 	u_int fl_db_saved;
760 	u_int fl_db_threshold;	/* in descriptors */
761 	u_int fl_hwidx:4;
762 
763 	/*
764 	 * fl_cidx is used by both the ithread and rxsync, the rest are not used
765 	 * in the rx fast path.
766 	 */
767 	uint32_t fl_cidx __aligned(CACHE_LINE_SIZE);
768 
769 	bus_dma_tag_t iq_desc_tag;
770 	bus_dmamap_t iq_desc_map;
771 	bus_addr_t iq_ba;
772 	int intr_idx;
773 
774 	bus_dma_tag_t fl_desc_tag;
775 	bus_dmamap_t fl_desc_map;
776 	bus_addr_t fl_ba;
777 };
778 
779 #define INVALID_NM_TXQ_CNTXT_ID ((u_int)(-1))
780 struct sge_nm_txq {
781 	struct tx_desc *desc;
782 	uint16_t cidx;
783 	uint16_t pidx;
784 	uint16_t sidx;
785 	uint16_t equiqidx;	/* EQUIQ last requested at this pidx */
786 	uint16_t equeqidx;	/* EQUEQ last requested at this pidx */
787 	uint16_t dbidx;		/* pidx of the most recent doorbell */
788 	uint8_t doorbells;
789 	volatile uint32_t *udb;
790 	u_int udb_qid;
791 	u_int cntxt_id;
792 	__be32 cpl_ctrl0;	/* for convenience */
793 	__be32 op_pkd;		/* ditto */
794 	u_int nid;		/* netmap ring # for this queue */
795 
796 	/* infrequently used items after this */
797 
798 	bus_dma_tag_t desc_tag;
799 	bus_dmamap_t desc_map;
800 	bus_addr_t ba;
801 	int iqidx;
802 } __aligned(CACHE_LINE_SIZE);
803 
804 struct sge {
805 	int nrxq;	/* total # of Ethernet rx queues */
806 	int ntxq;	/* total # of Ethernet tx queues */
807 	int nofldrxq;	/* total # of TOE rx queues */
808 	int nofldtxq;	/* total # of TOE tx queues */
809 	int nnmrxq;	/* total # of netmap rx queues */
810 	int nnmtxq;	/* total # of netmap tx queues */
811 	int niq;	/* total # of ingress queues */
812 	int neq;	/* total # of egress queues */
813 
814 	struct sge_iq fwq;	/* Firmware event queue */
815 	struct sge_wrq *ctrlq;	/* Control queues */
816 	struct sge_txq *txq;	/* NIC tx queues */
817 	struct sge_rxq *rxq;	/* NIC rx queues */
818 	struct sge_ofld_txq *ofld_txq;	/* TOE tx queues */
819 	struct sge_ofld_rxq *ofld_rxq;	/* TOE rx queues */
820 	struct sge_nm_txq *nm_txq;	/* netmap tx queues */
821 	struct sge_nm_rxq *nm_rxq;	/* netmap rx queues */
822 
823 	uint16_t iq_start;	/* first cntxt_id */
824 	uint16_t iq_base;	/* first abs_id */
825 	int eq_start;		/* first cntxt_id */
826 	int eq_base;		/* first abs_id */
827 	int iqmap_sz;
828 	int eqmap_sz;
829 	struct sge_iq **iqmap;	/* iq->cntxt_id to iq mapping */
830 	struct sge_eq **eqmap;	/* eq->cntxt_id to eq mapping */
831 
832 	int8_t safe_zidx;
833 	struct rx_buf_info rx_buf_info[SW_ZONE_SIZES];
834 };
835 
836 struct devnames {
837 	const char *nexus_name;
838 	const char *ifnet_name;
839 	const char *vi_ifnet_name;
840 	const char *pf03_drv_name;
841 	const char *vf_nexus_name;
842 	const char *vf_ifnet_name;
843 };
844 
845 struct clip_entry;
846 
847 struct adapter {
848 	SLIST_ENTRY(adapter) link;
849 	device_t dev;
850 	struct cdev *cdev;
851 	const struct devnames *names;
852 
853 	/* PCIe register resources */
854 	int regs_rid;
855 	struct resource *regs_res;
856 	int msix_rid;
857 	struct resource *msix_res;
858 	bus_space_handle_t bh;
859 	bus_space_tag_t bt;
860 	bus_size_t mmio_len;
861 	int udbs_rid;
862 	struct resource *udbs_res;
863 	volatile uint8_t *udbs_base;
864 
865 	unsigned int pf;
866 	unsigned int mbox;
867 	unsigned int vpd_busy;
868 	unsigned int vpd_flag;
869 
870 	/* Interrupt information */
871 	int intr_type;
872 	int intr_count;
873 	struct irq {
874 		struct resource *res;
875 		int rid;
876 		void *tag;
877 		struct sge_rxq *rxq;
878 		struct sge_nm_rxq *nm_rxq;
879 	} __aligned(CACHE_LINE_SIZE) *irq;
880 	int sge_gts_reg;
881 	int sge_kdoorbell_reg;
882 
883 	bus_dma_tag_t dmat;	/* Parent DMA tag */
884 
885 	struct sge sge;
886 	int lro_timeout;
887 	int sc_do_rxcopy;
888 
889 	int vxlan_port;
890 	u_int vxlan_refcount;
891 	int rawf_base;
892 	int nrawf;
893 
894 	struct taskqueue *tq[MAX_NCHAN];	/* General purpose taskqueues */
895 	struct task async_event_task;
896 	struct port_info *port[MAX_NPORTS];
897 	uint8_t chan_map[MAX_NCHAN];		/* channel -> port */
898 
899 	CXGBE_LIST_HEAD(, clip_entry) *clip_table;
900 	TAILQ_HEAD(, clip_entry) clip_pending;	/* these need hw update. */
901 	u_long clip_mask;
902 	int clip_gen;
903 	struct timeout_task clip_task;
904 
905 	void *tom_softc;	/* (struct tom_data *) */
906 	struct tom_tunables tt;
907 	struct t4_offload_policy *policy;
908 	struct rwlock policy_lock;
909 
910 	void *iwarp_softc;	/* (struct c4iw_dev *) */
911 	struct iw_tunables iwt;
912 	void *iscsi_ulp_softc;	/* (struct cxgbei_data *) */
913 	void *ccr_softc;	/* (struct ccr_softc *) */
914 	struct l2t_data *l2t;	/* L2 table */
915 	struct smt_data *smt;	/* Source MAC Table */
916 	struct tid_info tids;
917 	vmem_t *key_map;
918 	struct tls_tunables tlst;
919 
920 	uint8_t doorbells;
921 	int offload_map;	/* ports with IFCAP_TOE enabled */
922 	int active_ulds;	/* ULDs activated on this adapter */
923 	int flags;
924 	int debug_flags;
925 
926 	char ifp_lockname[16];
927 	struct mtx ifp_lock;
928 	struct ifnet *ifp;	/* tracer ifp */
929 	struct ifmedia media;
930 	int traceq;		/* iq used by all tracers, -1 if none */
931 	int tracer_valid;	/* bitmap of valid tracers */
932 	int tracer_enabled;	/* bitmap of enabled tracers */
933 
934 	char fw_version[16];
935 	char tp_version[16];
936 	char er_version[16];
937 	char bs_version[16];
938 	char cfg_file[32];
939 	u_int cfcsum;
940 	struct adapter_params params;
941 	const struct chip_params *chip_params;
942 	struct t4_virt_res vres;
943 
944 	uint16_t nbmcaps;
945 	uint16_t linkcaps;
946 	uint16_t switchcaps;
947 	uint16_t niccaps;
948 	uint16_t toecaps;
949 	uint16_t rdmacaps;
950 	uint16_t cryptocaps;
951 	uint16_t iscsicaps;
952 	uint16_t fcoecaps;
953 
954 	struct sysctl_ctx_list ctx;
955 	struct sysctl_oid *ctrlq_oid;
956 	struct sysctl_oid *fwq_oid;
957 
958 	struct mtx sc_lock;
959 	char lockname[16];
960 
961 	/* Starving free lists */
962 	struct mtx sfl_lock;	/* same cache-line as sc_lock? but that's ok */
963 	TAILQ_HEAD(, sge_fl) sfl;
964 	struct callout sfl_callout;
965 
966 	/*
967 	 * Driver code that can run when the adapter is suspended must use this
968 	 * lock or a synchronized_op and check for HW_OFF_LIMITS before
969 	 * accessing hardware.
970 	 *
971 	 * XXX: could be changed to rwlock.  wlock in suspend/resume and for
972 	 * indirect register access, rlock everywhere else.
973 	 */
974 	struct mtx reg_lock;
975 
976 	struct memwin memwin[NUM_MEMWIN];	/* memory windows */
977 
978 	struct mtx tc_lock;
979 	struct task tc_task;
980 
981 	struct task reset_task;
982 	const void *reset_thread;
983 	int num_resets;
984 	int incarnation;
985 
986 	const char *last_op;
987 	const void *last_op_thr;
988 	int last_op_flags;
989 
990 	int swintr;
991 	int sensor_resets;
992 
993 	struct callout ktls_tick;
994 };
995 
996 #define ADAPTER_LOCK(sc)		mtx_lock(&(sc)->sc_lock)
997 #define ADAPTER_UNLOCK(sc)		mtx_unlock(&(sc)->sc_lock)
998 #define ADAPTER_LOCK_ASSERT_OWNED(sc)	mtx_assert(&(sc)->sc_lock, MA_OWNED)
999 #define ADAPTER_LOCK_ASSERT_NOTOWNED(sc) mtx_assert(&(sc)->sc_lock, MA_NOTOWNED)
1000 
1001 #define ASSERT_SYNCHRONIZED_OP(sc)	\
1002     KASSERT(IS_BUSY(sc) && \
1003 	(mtx_owned(&(sc)->sc_lock) || sc->last_op_thr == curthread), \
1004 	("%s: operation not synchronized.", __func__))
1005 
1006 #define PORT_LOCK(pi)			mtx_lock(&(pi)->pi_lock)
1007 #define PORT_UNLOCK(pi)			mtx_unlock(&(pi)->pi_lock)
1008 #define PORT_LOCK_ASSERT_OWNED(pi)	mtx_assert(&(pi)->pi_lock, MA_OWNED)
1009 #define PORT_LOCK_ASSERT_NOTOWNED(pi)	mtx_assert(&(pi)->pi_lock, MA_NOTOWNED)
1010 
1011 #define FL_LOCK(fl)			mtx_lock(&(fl)->fl_lock)
1012 #define FL_TRYLOCK(fl)			mtx_trylock(&(fl)->fl_lock)
1013 #define FL_UNLOCK(fl)			mtx_unlock(&(fl)->fl_lock)
1014 #define FL_LOCK_ASSERT_OWNED(fl)	mtx_assert(&(fl)->fl_lock, MA_OWNED)
1015 #define FL_LOCK_ASSERT_NOTOWNED(fl)	mtx_assert(&(fl)->fl_lock, MA_NOTOWNED)
1016 
1017 #define RXQ_FL_LOCK(rxq)		FL_LOCK(&(rxq)->fl)
1018 #define RXQ_FL_UNLOCK(rxq)		FL_UNLOCK(&(rxq)->fl)
1019 #define RXQ_FL_LOCK_ASSERT_OWNED(rxq)	FL_LOCK_ASSERT_OWNED(&(rxq)->fl)
1020 #define RXQ_FL_LOCK_ASSERT_NOTOWNED(rxq) FL_LOCK_ASSERT_NOTOWNED(&(rxq)->fl)
1021 
1022 #define EQ_LOCK(eq)			mtx_lock(&(eq)->eq_lock)
1023 #define EQ_TRYLOCK(eq)			mtx_trylock(&(eq)->eq_lock)
1024 #define EQ_UNLOCK(eq)			mtx_unlock(&(eq)->eq_lock)
1025 #define EQ_LOCK_ASSERT_OWNED(eq)	mtx_assert(&(eq)->eq_lock, MA_OWNED)
1026 #define EQ_LOCK_ASSERT_NOTOWNED(eq)	mtx_assert(&(eq)->eq_lock, MA_NOTOWNED)
1027 
1028 #define TXQ_LOCK(txq)			EQ_LOCK(&(txq)->eq)
1029 #define TXQ_TRYLOCK(txq)		EQ_TRYLOCK(&(txq)->eq)
1030 #define TXQ_UNLOCK(txq)			EQ_UNLOCK(&(txq)->eq)
1031 #define TXQ_LOCK_ASSERT_OWNED(txq)	EQ_LOCK_ASSERT_OWNED(&(txq)->eq)
1032 #define TXQ_LOCK_ASSERT_NOTOWNED(txq)	EQ_LOCK_ASSERT_NOTOWNED(&(txq)->eq)
1033 
1034 #define for_each_txq(vi, iter, q) \
1035 	for (q = &vi->adapter->sge.txq[vi->first_txq], iter = 0; \
1036 	    iter < vi->ntxq; ++iter, ++q)
1037 #define for_each_rxq(vi, iter, q) \
1038 	for (q = &vi->adapter->sge.rxq[vi->first_rxq], iter = 0; \
1039 	    iter < vi->nrxq; ++iter, ++q)
1040 #define for_each_ofld_txq(vi, iter, q) \
1041 	for (q = &vi->adapter->sge.ofld_txq[vi->first_ofld_txq], iter = 0; \
1042 	    iter < vi->nofldtxq; ++iter, ++q)
1043 #define for_each_ofld_rxq(vi, iter, q) \
1044 	for (q = &vi->adapter->sge.ofld_rxq[vi->first_ofld_rxq], iter = 0; \
1045 	    iter < vi->nofldrxq; ++iter, ++q)
1046 #define for_each_nm_txq(vi, iter, q) \
1047 	for (q = &vi->adapter->sge.nm_txq[vi->first_nm_txq], iter = 0; \
1048 	    iter < vi->nnmtxq; ++iter, ++q)
1049 #define for_each_nm_rxq(vi, iter, q) \
1050 	for (q = &vi->adapter->sge.nm_rxq[vi->first_nm_rxq], iter = 0; \
1051 	    iter < vi->nnmrxq; ++iter, ++q)
1052 #define for_each_vi(_pi, _iter, _vi) \
1053 	for ((_vi) = (_pi)->vi, (_iter) = 0; (_iter) < (_pi)->nvi; \
1054 	     ++(_iter), ++(_vi))
1055 
1056 #define IDXINCR(idx, incr, wrap) do { \
1057 	idx = wrap - idx > incr ? idx + incr : incr - (wrap - idx); \
1058 } while (0)
1059 #define IDXDIFF(head, tail, wrap) \
1060 	((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head))
1061 
1062 /* One for errors, one for firmware events */
1063 #define T4_EXTRA_INTR 2
1064 
1065 /* One for firmware events */
1066 #define T4VF_EXTRA_INTR 1
1067 
1068 static inline int
1069 forwarding_intr_to_fwq(struct adapter *sc)
1070 {
1071 
1072 	return (sc->intr_count == 1);
1073 }
1074 
1075 /* Works reliably inside a sync_op or with reg_lock held. */
1076 static inline bool
1077 hw_off_limits(struct adapter *sc)
1078 {
1079 	return (__predict_false(sc->flags & HW_OFF_LIMITS));
1080 }
1081 
1082 static inline uint32_t
1083 t4_read_reg(struct adapter *sc, uint32_t reg)
1084 {
1085 	if (hw_off_limits(sc))
1086 		MPASS(curthread == sc->reset_thread);
1087 	return bus_space_read_4(sc->bt, sc->bh, reg);
1088 }
1089 
1090 static inline void
1091 t4_write_reg(struct adapter *sc, uint32_t reg, uint32_t val)
1092 {
1093 	if (hw_off_limits(sc))
1094 		MPASS(curthread == sc->reset_thread);
1095 	bus_space_write_4(sc->bt, sc->bh, reg, val);
1096 }
1097 
1098 static inline uint64_t
1099 t4_read_reg64(struct adapter *sc, uint32_t reg)
1100 {
1101 	if (hw_off_limits(sc))
1102 		MPASS(curthread == sc->reset_thread);
1103 #ifdef __LP64__
1104 	return bus_space_read_8(sc->bt, sc->bh, reg);
1105 #else
1106 	return (uint64_t)bus_space_read_4(sc->bt, sc->bh, reg) +
1107 	    ((uint64_t)bus_space_read_4(sc->bt, sc->bh, reg + 4) << 32);
1108 
1109 #endif
1110 }
1111 
1112 static inline void
1113 t4_write_reg64(struct adapter *sc, uint32_t reg, uint64_t val)
1114 {
1115 	if (hw_off_limits(sc))
1116 		MPASS(curthread == sc->reset_thread);
1117 #ifdef __LP64__
1118 	bus_space_write_8(sc->bt, sc->bh, reg, val);
1119 #else
1120 	bus_space_write_4(sc->bt, sc->bh, reg, val);
1121 	bus_space_write_4(sc->bt, sc->bh, reg + 4, val>> 32);
1122 #endif
1123 }
1124 
1125 static inline void
1126 t4_os_pci_read_cfg1(struct adapter *sc, int reg, uint8_t *val)
1127 {
1128 	if (hw_off_limits(sc))
1129 		MPASS(curthread == sc->reset_thread);
1130 	*val = pci_read_config(sc->dev, reg, 1);
1131 }
1132 
1133 static inline void
1134 t4_os_pci_write_cfg1(struct adapter *sc, int reg, uint8_t val)
1135 {
1136 	if (hw_off_limits(sc))
1137 		MPASS(curthread == sc->reset_thread);
1138 	pci_write_config(sc->dev, reg, val, 1);
1139 }
1140 
1141 static inline void
1142 t4_os_pci_read_cfg2(struct adapter *sc, int reg, uint16_t *val)
1143 {
1144 
1145 	if (hw_off_limits(sc))
1146 		MPASS(curthread == sc->reset_thread);
1147 	*val = pci_read_config(sc->dev, reg, 2);
1148 }
1149 
1150 static inline void
1151 t4_os_pci_write_cfg2(struct adapter *sc, int reg, uint16_t val)
1152 {
1153 	if (hw_off_limits(sc))
1154 		MPASS(curthread == sc->reset_thread);
1155 	pci_write_config(sc->dev, reg, val, 2);
1156 }
1157 
1158 static inline void
1159 t4_os_pci_read_cfg4(struct adapter *sc, int reg, uint32_t *val)
1160 {
1161 	if (hw_off_limits(sc))
1162 		MPASS(curthread == sc->reset_thread);
1163 	*val = pci_read_config(sc->dev, reg, 4);
1164 }
1165 
1166 static inline void
1167 t4_os_pci_write_cfg4(struct adapter *sc, int reg, uint32_t val)
1168 {
1169 	if (hw_off_limits(sc))
1170 		MPASS(curthread == sc->reset_thread);
1171 	pci_write_config(sc->dev, reg, val, 4);
1172 }
1173 
1174 static inline struct port_info *
1175 adap2pinfo(struct adapter *sc, int idx)
1176 {
1177 
1178 	return (sc->port[idx]);
1179 }
1180 
1181 static inline void
1182 t4_os_set_hw_addr(struct port_info *pi, uint8_t hw_addr[])
1183 {
1184 
1185 	bcopy(hw_addr, pi->vi[0].hw_addr, ETHER_ADDR_LEN);
1186 }
1187 
1188 static inline int
1189 tx_resume_threshold(struct sge_eq *eq)
1190 {
1191 
1192 	/* not quite the same as qsize / 4, but this will do. */
1193 	return (eq->sidx / 4);
1194 }
1195 
1196 static inline int
1197 t4_use_ldst(struct adapter *sc)
1198 {
1199 
1200 #ifdef notyet
1201 	return (sc->flags & FW_OK || !sc->use_bd);
1202 #else
1203 	return (0);
1204 #endif
1205 }
1206 
1207 static inline void
1208 CH_DUMP_MBOX(struct adapter *sc, int mbox, const int reg,
1209     const char *msg, const __be64 *const p, const bool err)
1210 {
1211 
1212 	if (!(sc->debug_flags & DF_DUMP_MBOX) && !err)
1213 		return;
1214 	if (p != NULL) {
1215 		log(err ? LOG_ERR : LOG_DEBUG,
1216 		    "%s: mbox %u %s %016llx %016llx %016llx %016llx "
1217 		    "%016llx %016llx %016llx %016llx\n",
1218 		    device_get_nameunit(sc->dev), mbox, msg,
1219 		    (long long)be64_to_cpu(p[0]), (long long)be64_to_cpu(p[1]),
1220 		    (long long)be64_to_cpu(p[2]), (long long)be64_to_cpu(p[3]),
1221 		    (long long)be64_to_cpu(p[4]), (long long)be64_to_cpu(p[5]),
1222 		    (long long)be64_to_cpu(p[6]), (long long)be64_to_cpu(p[7]));
1223 	} else {
1224 		log(err ? LOG_ERR : LOG_DEBUG,
1225 		    "%s: mbox %u %s %016llx %016llx %016llx %016llx "
1226 		    "%016llx %016llx %016llx %016llx\n",
1227 		    device_get_nameunit(sc->dev), mbox, msg,
1228 		    (long long)t4_read_reg64(sc, reg),
1229 		    (long long)t4_read_reg64(sc, reg + 8),
1230 		    (long long)t4_read_reg64(sc, reg + 16),
1231 		    (long long)t4_read_reg64(sc, reg + 24),
1232 		    (long long)t4_read_reg64(sc, reg + 32),
1233 		    (long long)t4_read_reg64(sc, reg + 40),
1234 		    (long long)t4_read_reg64(sc, reg + 48),
1235 		    (long long)t4_read_reg64(sc, reg + 56));
1236 	}
1237 }
1238 
1239 /* t4_main.c */
1240 extern int t4_ntxq;
1241 extern int t4_nrxq;
1242 extern int t4_intr_types;
1243 extern int t4_tmr_idx;
1244 extern int t4_pktc_idx;
1245 extern unsigned int t4_qsize_rxq;
1246 extern unsigned int t4_qsize_txq;
1247 extern device_method_t cxgbe_methods[];
1248 
1249 int t4_os_find_pci_capability(struct adapter *, int);
1250 int t4_os_pci_save_state(struct adapter *);
1251 int t4_os_pci_restore_state(struct adapter *);
1252 void t4_os_portmod_changed(struct port_info *);
1253 void t4_os_link_changed(struct port_info *);
1254 void t4_iterate(void (*)(struct adapter *, void *), void *);
1255 void t4_init_devnames(struct adapter *);
1256 void t4_add_adapter(struct adapter *);
1257 int t4_detach_common(device_t);
1258 int t4_map_bars_0_and_4(struct adapter *);
1259 int t4_map_bar_2(struct adapter *);
1260 int t4_setup_intr_handlers(struct adapter *);
1261 void t4_sysctls(struct adapter *);
1262 int begin_synchronized_op(struct adapter *, struct vi_info *, int, char *);
1263 void doom_vi(struct adapter *, struct vi_info *);
1264 void end_synchronized_op(struct adapter *, int);
1265 int update_mac_settings(struct ifnet *, int);
1266 int adapter_init(struct adapter *);
1267 int vi_init(struct vi_info *);
1268 void vi_sysctls(struct vi_info *);
1269 int rw_via_memwin(struct adapter *, int, uint32_t, uint32_t *, int, int);
1270 int alloc_atid(struct adapter *, void *);
1271 void *lookup_atid(struct adapter *, int);
1272 void free_atid(struct adapter *, int);
1273 void release_tid(struct adapter *, int, struct sge_wrq *);
1274 int cxgbe_media_change(struct ifnet *);
1275 void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
1276 bool t4_os_dump_cimla(struct adapter *, int, bool);
1277 void t4_os_dump_devlog(struct adapter *);
1278 
1279 #ifdef KERN_TLS
1280 /* t4_kern_tls.c */
1281 int cxgbe_tls_tag_alloc(struct ifnet *, union if_snd_tag_alloc_params *,
1282     struct m_snd_tag **);
1283 void cxgbe_tls_tag_free(struct m_snd_tag *);
1284 void t6_ktls_modload(void);
1285 void t6_ktls_modunload(void);
1286 int t6_ktls_try(struct ifnet *, struct socket *, struct ktls_session *);
1287 int t6_ktls_parse_pkt(struct mbuf *, int *, int *);
1288 int t6_ktls_write_wr(struct sge_txq *, void *, struct mbuf *, u_int, u_int);
1289 #endif
1290 
1291 /* t4_keyctx.c */
1292 struct auth_hash;
1293 union authctx;
1294 
1295 void t4_aes_getdeckey(void *, const void *, unsigned int);
1296 void t4_copy_partial_hash(int, union authctx *, void *);
1297 void t4_init_gmac_hash(const char *, int, char *);
1298 void t4_init_hmac_digest(struct auth_hash *, u_int, const char *, int, char *);
1299 
1300 #ifdef DEV_NETMAP
1301 /* t4_netmap.c */
1302 struct sge_nm_rxq;
1303 void cxgbe_nm_attach(struct vi_info *);
1304 void cxgbe_nm_detach(struct vi_info *);
1305 void service_nm_rxq(struct sge_nm_rxq *);
1306 int alloc_nm_rxq(struct vi_info *, struct sge_nm_rxq *, int, int);
1307 int free_nm_rxq(struct vi_info *, struct sge_nm_rxq *);
1308 int alloc_nm_txq(struct vi_info *, struct sge_nm_txq *, int, int);
1309 int free_nm_txq(struct vi_info *, struct sge_nm_txq *);
1310 #endif
1311 
1312 /* t4_sge.c */
1313 void t4_sge_modload(void);
1314 void t4_sge_modunload(void);
1315 uint64_t t4_sge_extfree_refs(void);
1316 void t4_tweak_chip_settings(struct adapter *);
1317 int t4_verify_chip_settings(struct adapter *);
1318 void t4_init_rx_buf_info(struct adapter *);
1319 int t4_create_dma_tag(struct adapter *);
1320 void t4_sge_sysctls(struct adapter *, struct sysctl_ctx_list *,
1321     struct sysctl_oid_list *);
1322 int t4_destroy_dma_tag(struct adapter *);
1323 int alloc_ring(struct adapter *, size_t, bus_dma_tag_t *, bus_dmamap_t *,
1324     bus_addr_t *, void **);
1325 int free_ring(struct adapter *, bus_dma_tag_t, bus_dmamap_t, bus_addr_t,
1326     void *);
1327 void free_fl_buffers(struct adapter *, struct sge_fl *);
1328 int t4_setup_adapter_queues(struct adapter *);
1329 int t4_teardown_adapter_queues(struct adapter *);
1330 int t4_setup_vi_queues(struct vi_info *);
1331 int t4_teardown_vi_queues(struct vi_info *);
1332 void t4_intr_all(void *);
1333 void t4_intr(void *);
1334 #ifdef DEV_NETMAP
1335 void t4_nm_intr(void *);
1336 void t4_vi_intr(void *);
1337 #endif
1338 void t4_intr_err(void *);
1339 void t4_intr_evt(void *);
1340 void t4_wrq_tx_locked(struct adapter *, struct sge_wrq *, struct wrqe *);
1341 void t4_update_fl_bufsize(struct ifnet *);
1342 struct mbuf *alloc_wr_mbuf(int, int);
1343 int parse_pkt(struct mbuf **, bool);
1344 void *start_wrq_wr(struct sge_wrq *, int, struct wrq_cookie *);
1345 void commit_wrq_wr(struct sge_wrq *, void *, struct wrq_cookie *);
1346 int tnl_cong(struct port_info *, int);
1347 void t4_register_an_handler(an_handler_t);
1348 void t4_register_fw_msg_handler(int, fw_msg_handler_t);
1349 void t4_register_cpl_handler(int, cpl_handler_t);
1350 void t4_register_shared_cpl_handler(int, cpl_handler_t, int);
1351 #ifdef RATELIMIT
1352 int ethofld_transmit(struct ifnet *, struct mbuf *);
1353 void send_etid_flush_wr(struct cxgbe_rate_tag *);
1354 #endif
1355 
1356 /* t4_tracer.c */
1357 struct t4_tracer;
1358 void t4_tracer_modload(void);
1359 void t4_tracer_modunload(void);
1360 void t4_tracer_port_detach(struct adapter *);
1361 int t4_get_tracer(struct adapter *, struct t4_tracer *);
1362 int t4_set_tracer(struct adapter *, struct t4_tracer *);
1363 int t4_trace_pkt(struct sge_iq *, const struct rss_header *, struct mbuf *);
1364 int t5_trace_pkt(struct sge_iq *, const struct rss_header *, struct mbuf *);
1365 
1366 /* t4_sched.c */
1367 int t4_set_sched_class(struct adapter *, struct t4_sched_params *);
1368 int t4_set_sched_queue(struct adapter *, struct t4_sched_queue *);
1369 int t4_init_tx_sched(struct adapter *);
1370 int t4_free_tx_sched(struct adapter *);
1371 void t4_update_tx_sched(struct adapter *);
1372 int t4_reserve_cl_rl_kbps(struct adapter *, int, u_int, int *);
1373 void t4_release_cl_rl(struct adapter *, int, int);
1374 int sysctl_tc(SYSCTL_HANDLER_ARGS);
1375 int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
1376 #ifdef RATELIMIT
1377 void t4_init_etid_table(struct adapter *);
1378 void t4_free_etid_table(struct adapter *);
1379 struct cxgbe_rate_tag *lookup_etid(struct adapter *, int);
1380 int cxgbe_rate_tag_alloc(struct ifnet *, union if_snd_tag_alloc_params *,
1381     struct m_snd_tag **);
1382 int cxgbe_rate_tag_modify(struct m_snd_tag *, union if_snd_tag_modify_params *);
1383 int cxgbe_rate_tag_query(struct m_snd_tag *, union if_snd_tag_query_params *);
1384 void cxgbe_rate_tag_free(struct m_snd_tag *);
1385 void cxgbe_rate_tag_free_locked(struct cxgbe_rate_tag *);
1386 void cxgbe_ratelimit_query(struct ifnet *, struct if_ratelimit_query_results *);
1387 #endif
1388 
1389 /* t4_filter.c */
1390 int get_filter_mode(struct adapter *, uint32_t *);
1391 int set_filter_mode(struct adapter *, uint32_t);
1392 int set_filter_mask(struct adapter *, uint32_t);
1393 int get_filter(struct adapter *, struct t4_filter *);
1394 int set_filter(struct adapter *, struct t4_filter *);
1395 int del_filter(struct adapter *, struct t4_filter *);
1396 int t4_filter_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1397 int t4_hashfilter_ao_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1398 int t4_hashfilter_tcb_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1399 int t4_del_hashfilter_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1400 void free_hftid_hash(struct tid_info *);
1401 
1402 static inline struct wrqe *
1403 alloc_wrqe(int wr_len, struct sge_wrq *wrq)
1404 {
1405 	int len = offsetof(struct wrqe, wr) + wr_len;
1406 	struct wrqe *wr;
1407 
1408 	wr = malloc(len, M_CXGBE, M_NOWAIT);
1409 	if (__predict_false(wr == NULL))
1410 		return (NULL);
1411 	wr->wr_len = wr_len;
1412 	wr->wrq = wrq;
1413 	return (wr);
1414 }
1415 
1416 static inline void *
1417 wrtod(struct wrqe *wr)
1418 {
1419 	return (&wr->wr[0]);
1420 }
1421 
1422 static inline void
1423 free_wrqe(struct wrqe *wr)
1424 {
1425 	free(wr, M_CXGBE);
1426 }
1427 
1428 static inline void
1429 t4_wrq_tx(struct adapter *sc, struct wrqe *wr)
1430 {
1431 	struct sge_wrq *wrq = wr->wrq;
1432 
1433 	TXQ_LOCK(wrq);
1434 	t4_wrq_tx_locked(sc, wrq, wr);
1435 	TXQ_UNLOCK(wrq);
1436 }
1437 
1438 static inline int
1439 read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
1440     int len)
1441 {
1442 
1443 	return (rw_via_memwin(sc, idx, addr, val, len, 0));
1444 }
1445 
1446 static inline int
1447 write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
1448     const uint32_t *val, int len)
1449 {
1450 
1451 	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
1452 }
1453 
1454 /* Number of len16 -> number of descriptors */
1455 static inline int
1456 tx_len16_to_desc(int len16)
1457 {
1458 
1459 	return (howmany(len16, EQ_ESIZE / 16));
1460 }
1461 #endif
1462