xref: /freebsd/sys/dev/cxgbe/adapter.h (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
1 /*-
2  * Copyright (c) 2011 Chelsio Communications, Inc.
3  * All rights reserved.
4  * Written by: Navdeep Parhar <np@FreeBSD.org>
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  *
27  * $FreeBSD$
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/rman.h>
37 #include <sys/types.h>
38 #include <sys/lock.h>
39 #include <sys/malloc.h>
40 #include <sys/rwlock.h>
41 #include <sys/sx.h>
42 #include <vm/uma.h>
43 
44 #include <dev/pci/pcivar.h>
45 #include <dev/pci/pcireg.h>
46 #include <machine/bus.h>
47 #include <sys/socket.h>
48 #include <sys/sysctl.h>
49 #include <net/ethernet.h>
50 #include <net/if.h>
51 #include <net/if_var.h>
52 #include <net/if_media.h>
53 #include <netinet/in.h>
54 #include <netinet/tcp_lro.h>
55 
56 #include "offload.h"
57 #include "common/t4_msg.h"
58 #include "firmware/t4fw_interface.h"
59 
60 #define KTR_CXGBE	KTR_SPARE3
61 MALLOC_DECLARE(M_CXGBE);
62 #define CXGBE_UNIMPLEMENTED(s) \
63     panic("%s (%s, line %d) not implemented yet.", s, __FILE__, __LINE__)
64 
65 #if defined(__i386__) || defined(__amd64__)
66 static __inline void
67 prefetch(void *x)
68 {
69 	__asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x));
70 }
71 #else
72 #define prefetch(x)
73 #endif
74 
75 #ifndef SYSCTL_ADD_UQUAD
76 #define SYSCTL_ADD_UQUAD SYSCTL_ADD_QUAD
77 #define sysctl_handle_64 sysctl_handle_quad
78 #define CTLTYPE_U64 CTLTYPE_QUAD
79 #endif
80 
81 #if (__FreeBSD_version >= 900030) || \
82     ((__FreeBSD_version >= 802507) && (__FreeBSD_version < 900000))
83 #define SBUF_DRAIN 1
84 #endif
85 
86 #ifdef __amd64__
87 /* XXX: need systemwide bus_space_read_8/bus_space_write_8 */
88 static __inline uint64_t
89 t4_bus_space_read_8(bus_space_tag_t tag, bus_space_handle_t handle,
90     bus_size_t offset)
91 {
92 	KASSERT(tag == X86_BUS_SPACE_MEM,
93 	    ("%s: can only handle mem space", __func__));
94 
95 	return (*(volatile uint64_t *)(handle + offset));
96 }
97 
98 static __inline void
99 t4_bus_space_write_8(bus_space_tag_t tag, bus_space_handle_t bsh,
100     bus_size_t offset, uint64_t value)
101 {
102 	KASSERT(tag == X86_BUS_SPACE_MEM,
103 	    ("%s: can only handle mem space", __func__));
104 
105 	*(volatile uint64_t *)(bsh + offset) = value;
106 }
107 #else
108 static __inline uint64_t
109 t4_bus_space_read_8(bus_space_tag_t tag, bus_space_handle_t handle,
110     bus_size_t offset)
111 {
112 	return (uint64_t)bus_space_read_4(tag, handle, offset) +
113 	    ((uint64_t)bus_space_read_4(tag, handle, offset + 4) << 32);
114 }
115 
116 static __inline void
117 t4_bus_space_write_8(bus_space_tag_t tag, bus_space_handle_t bsh,
118     bus_size_t offset, uint64_t value)
119 {
120 	bus_space_write_4(tag, bsh, offset, value);
121 	bus_space_write_4(tag, bsh, offset + 4, value >> 32);
122 }
123 #endif
124 
125 struct adapter;
126 typedef struct adapter adapter_t;
127 
128 enum {
129 	/*
130 	 * All ingress queues use this entry size.  Note that the firmware event
131 	 * queue and any iq expecting CPL_RX_PKT in the descriptor needs this to
132 	 * be at least 64.
133 	 */
134 	IQ_ESIZE = 64,
135 
136 	/* Default queue sizes for all kinds of ingress queues */
137 	FW_IQ_QSIZE = 256,
138 	RX_IQ_QSIZE = 1024,
139 
140 	/* All egress queues use this entry size */
141 	EQ_ESIZE = 64,
142 
143 	/* Default queue sizes for all kinds of egress queues */
144 	CTRL_EQ_QSIZE = 128,
145 	TX_EQ_QSIZE = 1024,
146 
147 #if MJUMPAGESIZE != MCLBYTES
148 	SW_ZONE_SIZES = 4,	/* cluster, jumbop, jumbo9k, jumbo16k */
149 #else
150 	SW_ZONE_SIZES = 3,	/* cluster, jumbo9k, jumbo16k */
151 #endif
152 	CL_METADATA_SIZE = CACHE_LINE_SIZE,
153 
154 	SGE_MAX_WR_NDESC = SGE_MAX_WR_LEN / EQ_ESIZE, /* max WR size in desc */
155 	TX_SGL_SEGS = 39,
156 	TX_SGL_SEGS_TSO = 38,
157 	TX_WR_FLITS = SGE_MAX_WR_LEN / 8
158 };
159 
160 enum {
161 	/* adapter intr_type */
162 	INTR_INTX	= (1 << 0),
163 	INTR_MSI 	= (1 << 1),
164 	INTR_MSIX	= (1 << 2)
165 };
166 
167 enum {
168 	XGMAC_MTU	= (1 << 0),
169 	XGMAC_PROMISC	= (1 << 1),
170 	XGMAC_ALLMULTI	= (1 << 2),
171 	XGMAC_VLANEX	= (1 << 3),
172 	XGMAC_UCADDR	= (1 << 4),
173 	XGMAC_MCADDRS	= (1 << 5),
174 
175 	XGMAC_ALL	= 0xffff
176 };
177 
178 enum {
179 	/* flags understood by begin_synchronized_op */
180 	HOLD_LOCK	= (1 << 0),
181 	SLEEP_OK	= (1 << 1),
182 	INTR_OK		= (1 << 2),
183 
184 	/* flags understood by end_synchronized_op */
185 	LOCK_HELD	= HOLD_LOCK,
186 };
187 
188 enum {
189 	/* adapter flags */
190 	FULL_INIT_DONE	= (1 << 0),
191 	FW_OK		= (1 << 1),
192 	/* INTR_DIRECT	= (1 << 2),	No longer used. */
193 	MASTER_PF	= (1 << 3),
194 	ADAP_SYSCTL_CTX	= (1 << 4),
195 	/* TOM_INIT_DONE= (1 << 5),	No longer used */
196 	BUF_PACKING_OK	= (1 << 6),
197 
198 	CXGBE_BUSY	= (1 << 9),
199 
200 	/* port flags */
201 	DOOMED		= (1 << 0),
202 	PORT_INIT_DONE	= (1 << 1),
203 	PORT_SYSCTL_CTX	= (1 << 2),
204 	HAS_TRACEQ	= (1 << 3),
205 	INTR_RXQ	= (1 << 4),	/* All NIC rxq's take interrupts */
206 	INTR_OFLD_RXQ	= (1 << 5),	/* All TOE rxq's take interrupts */
207 	INTR_NM_RXQ	= (1 << 6),	/* All netmap rxq's take interrupts */
208 	INTR_ALL	= (INTR_RXQ | INTR_OFLD_RXQ | INTR_NM_RXQ),
209 
210 	/* adapter debug_flags */
211 	DF_DUMP_MBOX	= (1 << 0),
212 };
213 
214 #define IS_DOOMED(pi)	((pi)->flags & DOOMED)
215 #define SET_DOOMED(pi)	do {(pi)->flags |= DOOMED;} while (0)
216 #define IS_BUSY(sc)	((sc)->flags & CXGBE_BUSY)
217 #define SET_BUSY(sc)	do {(sc)->flags |= CXGBE_BUSY;} while (0)
218 #define CLR_BUSY(sc)	do {(sc)->flags &= ~CXGBE_BUSY;} while (0)
219 
220 struct port_info {
221 	device_t dev;
222 	struct adapter *adapter;
223 
224 	struct ifnet *ifp;
225 	struct ifmedia media;
226 
227 	struct mtx pi_lock;
228 	char lockname[16];
229 	unsigned long flags;
230 	int if_flags;
231 
232 	uint16_t *rss;
233 	uint16_t viid;
234 	int16_t  xact_addr_filt;/* index of exact MAC address filter */
235 	uint16_t rss_size;	/* size of VI's RSS table slice */
236 	uint16_t rss_base;	/* start of VI's RSS table slice */
237 	uint8_t  lport;		/* associated offload logical port */
238 	int8_t   mdio_addr;
239 	uint8_t  port_type;
240 	uint8_t  mod_type;
241 	uint8_t  port_id;
242 	uint8_t  tx_chan;
243 	uint8_t  rx_chan_map;	/* rx MPS channel bitmap */
244 
245 	/* These need to be int as they are used in sysctl */
246 	int ntxq;	/* # of tx queues */
247 	int first_txq;	/* index of first tx queue */
248 	int rsrv_noflowq; /* Reserve queue 0 for non-flowid packets */
249 	int nrxq;	/* # of rx queues */
250 	int first_rxq;	/* index of first rx queue */
251 #ifdef TCP_OFFLOAD
252 	int nofldtxq;		/* # of offload tx queues */
253 	int first_ofld_txq;	/* index of first offload tx queue */
254 	int nofldrxq;		/* # of offload rx queues */
255 	int first_ofld_rxq;	/* index of first offload rx queue */
256 #endif
257 #ifdef DEV_NETMAP
258 	int nnmtxq;		/* # of netmap tx queues */
259 	int first_nm_txq;	/* index of first netmap tx queue */
260 	int nnmrxq;		/* # of netmap rx queues */
261 	int first_nm_rxq;	/* index of first netmap rx queue */
262 
263 	struct ifnet *nm_ifp;
264 	struct ifmedia nm_media;
265 	int nmif_flags;
266 	uint16_t nm_viid;
267 	int16_t nm_xact_addr_filt;
268 	uint16_t nm_rss_size;	/* size of netmap VI's RSS table slice */
269 #endif
270 	int tmr_idx;
271 	int pktc_idx;
272 	int qsize_rxq;
273 	int qsize_txq;
274 
275 	int linkdnrc;
276 	struct link_config link_cfg;
277 
278 	struct timeval last_refreshed;
279  	struct port_stats stats;
280 	u_int tnl_cong_drops;
281 	u_int tx_parse_error;
282 
283 	eventhandler_tag vlan_c;
284 
285 	struct callout tick;
286 	struct sysctl_ctx_list ctx;	/* from ifconfig up to driver detach */
287 
288 	uint8_t hw_addr[ETHER_ADDR_LEN]; /* factory MAC address, won't change */
289 };
290 
291 /* Where the cluster came from, how it has been carved up. */
292 struct cluster_layout {
293 	int8_t zidx;
294 	int8_t hwidx;
295 	uint16_t region1;	/* mbufs laid out within this region */
296 				/* region2 is the DMA region */
297 	uint16_t region3;	/* cluster_metadata within this region */
298 };
299 
300 struct cluster_metadata {
301 	u_int refcount;
302 #ifdef INVARIANTS
303 	struct fl_sdesc *sd;	/* For debug only.  Could easily be stale */
304 #endif
305 };
306 
307 struct fl_sdesc {
308 	caddr_t cl;
309 	uint16_t nmbuf;	/* # of driver originated mbufs with ref on cluster */
310 	struct cluster_layout cll;
311 };
312 
313 struct tx_desc {
314 	__be64 flit[8];
315 };
316 
317 struct tx_sdesc {
318 	struct mbuf *m;		/* m_nextpkt linked chain of frames */
319 	uint8_t desc_used;	/* # of hardware descriptors used by the WR */
320 };
321 
322 
323 #define IQ_PAD (IQ_ESIZE - sizeof(struct rsp_ctrl) - sizeof(struct rss_header))
324 struct iq_desc {
325 	struct rss_header rss;
326 	uint8_t cpl[IQ_PAD];
327 	struct rsp_ctrl rsp;
328 };
329 #undef IQ_PAD
330 CTASSERT(sizeof(struct iq_desc) == IQ_ESIZE);
331 
332 enum {
333 	/* iq flags */
334 	IQ_ALLOCATED	= (1 << 0),	/* firmware resources allocated */
335 	IQ_HAS_FL	= (1 << 1),	/* iq associated with a freelist */
336 	IQ_INTR		= (1 << 2),	/* iq takes direct interrupt */
337 	IQ_LRO_ENABLED	= (1 << 3),	/* iq is an eth rxq with LRO enabled */
338 
339 	/* iq state */
340 	IQS_DISABLED	= 0,
341 	IQS_BUSY	= 1,
342 	IQS_IDLE	= 2,
343 };
344 
345 /*
346  * Ingress Queue: T4 is producer, driver is consumer.
347  */
348 struct sge_iq {
349 	uint32_t flags;
350 	volatile int state;
351 	struct adapter *adapter;
352 	struct iq_desc  *desc;	/* KVA of descriptor ring */
353 	int8_t   intr_pktc_idx;	/* packet count threshold index */
354 	uint8_t  gen;		/* generation bit */
355 	uint8_t  intr_params;	/* interrupt holdoff parameters */
356 	uint8_t  intr_next;	/* XXX: holdoff for next interrupt */
357 	uint16_t qsize;		/* size (# of entries) of the queue */
358 	uint16_t sidx;		/* index of the entry with the status page */
359 	uint16_t cidx;		/* consumer index */
360 	uint16_t cntxt_id;	/* SGE context id for the iq */
361 	uint16_t abs_id;	/* absolute SGE id for the iq */
362 
363 	STAILQ_ENTRY(sge_iq) link;
364 
365 	bus_dma_tag_t desc_tag;
366 	bus_dmamap_t desc_map;
367 	bus_addr_t ba;		/* bus address of descriptor ring */
368 };
369 
370 enum {
371 	EQ_CTRL		= 1,
372 	EQ_ETH		= 2,
373 	EQ_OFLD		= 3,
374 
375 	/* eq flags */
376 	EQ_TYPEMASK	= 0x3,		/* 2 lsbits hold the type (see above) */
377 	EQ_ALLOCATED	= (1 << 2),	/* firmware resources allocated */
378 	EQ_ENABLED	= (1 << 3),	/* open for business */
379 };
380 
381 /* Listed in order of preference.  Update t4_sysctls too if you change these */
382 enum {DOORBELL_UDB, DOORBELL_WCWR, DOORBELL_UDBWC, DOORBELL_KDB};
383 
384 /*
385  * Egress Queue: driver is producer, T4 is consumer.
386  *
387  * Note: A free list is an egress queue (driver produces the buffers and T4
388  * consumes them) but it's special enough to have its own struct (see sge_fl).
389  */
390 struct sge_eq {
391 	unsigned int flags;	/* MUST be first */
392 	unsigned int cntxt_id;	/* SGE context id for the eq */
393 	struct mtx eq_lock;
394 
395 	struct tx_desc *desc;	/* KVA of descriptor ring */
396 	uint16_t doorbells;
397 	volatile uint32_t *udb;	/* KVA of doorbell (lies within BAR2) */
398 	u_int udb_qid;		/* relative qid within the doorbell page */
399 	uint16_t sidx;		/* index of the entry with the status page */
400 	uint16_t cidx;		/* consumer idx (desc idx) */
401 	uint16_t pidx;		/* producer idx (desc idx) */
402 	uint16_t equeqidx;	/* EQUEQ last requested at this pidx */
403 	uint16_t dbidx;		/* pidx of the most recent doorbell */
404 	uint16_t iqid;		/* iq that gets egr_update for the eq */
405 	uint8_t tx_chan;	/* tx channel used by the eq */
406 	volatile u_int equiq;	/* EQUIQ outstanding */
407 
408 	bus_dma_tag_t desc_tag;
409 	bus_dmamap_t desc_map;
410 	bus_addr_t ba;		/* bus address of descriptor ring */
411 	char lockname[16];
412 };
413 
414 struct sw_zone_info {
415 	uma_zone_t zone;	/* zone that this cluster comes from */
416 	int size;		/* size of cluster: 2K, 4K, 9K, 16K, etc. */
417 	int type;		/* EXT_xxx type of the cluster */
418 	int8_t head_hwidx;
419 	int8_t tail_hwidx;
420 };
421 
422 struct hw_buf_info {
423 	int8_t zidx;		/* backpointer to zone; -ve means unused */
424 	int8_t next;		/* next hwidx for this zone; -1 means no more */
425 	int size;
426 };
427 
428 enum {
429 	FL_STARVING	= (1 << 0), /* on the adapter's list of starving fl's */
430 	FL_DOOMED	= (1 << 1), /* about to be destroyed */
431 	FL_BUF_PACKING	= (1 << 2), /* buffer packing enabled */
432 	FL_BUF_RESUME	= (1 << 3), /* resume from the middle of the frame */
433 };
434 
435 #define FL_RUNNING_LOW(fl) \
436     (IDXDIFF(fl->dbidx * 8, fl->cidx, fl->sidx * 8) <= fl->lowat)
437 #define FL_NOT_RUNNING_LOW(fl) \
438     (IDXDIFF(fl->dbidx * 8, fl->cidx, fl->sidx * 8) >= 2 * fl->lowat)
439 
440 struct sge_fl {
441 	struct mtx fl_lock;
442 	__be64 *desc;		/* KVA of descriptor ring, ptr to addresses */
443 	struct fl_sdesc *sdesc;	/* KVA of software descriptor ring */
444 	struct cluster_layout cll_def;	/* default refill zone, layout */
445 	uint16_t lowat;		/* # of buffers <= this means fl needs help */
446 	int flags;
447 	uint16_t buf_boundary;
448 
449 	/* The 16b idx all deal with hw descriptors */
450 	uint16_t dbidx;		/* hw pidx after last doorbell */
451 	uint16_t sidx;		/* index of status page */
452 	volatile uint16_t hw_cidx;
453 
454 	/* The 32b idx are all buffer idx, not hardware descriptor idx */
455 	uint32_t cidx;		/* consumer index */
456 	uint32_t pidx;		/* producer index */
457 
458 	uint32_t dbval;
459 	u_int rx_offset;	/* offset in fl buf (when buffer packing) */
460 	volatile uint32_t *udb;
461 
462 	uint64_t mbuf_allocated;/* # of mbuf allocated from zone_mbuf */
463 	uint64_t mbuf_inlined;	/* # of mbuf created within clusters */
464 	uint64_t cl_allocated;	/* # of clusters allocated */
465 	uint64_t cl_recycled;	/* # of clusters recycled */
466 	uint64_t cl_fast_recycled; /* # of clusters recycled (fast) */
467 
468 	/* These 3 are valid when FL_BUF_RESUME is set, stale otherwise. */
469 	struct mbuf *m0;
470 	struct mbuf **pnext;
471 	u_int remaining;
472 
473 	uint16_t qsize;		/* # of hw descriptors (status page included) */
474 	uint16_t cntxt_id;	/* SGE context id for the freelist */
475 	TAILQ_ENTRY(sge_fl) link; /* All starving freelists */
476 	bus_dma_tag_t desc_tag;
477 	bus_dmamap_t desc_map;
478 	char lockname[16];
479 	bus_addr_t ba;		/* bus address of descriptor ring */
480 	struct cluster_layout cll_alt;	/* alternate refill zone, layout */
481 };
482 
483 struct mp_ring;
484 
485 /* txq: SGE egress queue + what's needed for Ethernet NIC */
486 struct sge_txq {
487 	struct sge_eq eq;	/* MUST be first */
488 
489 	struct ifnet *ifp;	/* the interface this txq belongs to */
490 	struct mp_ring *r;	/* tx software ring */
491 	struct tx_sdesc *sdesc;	/* KVA of software descriptor ring */
492 	struct sglist *gl;
493 	__be32 cpl_ctrl0;	/* for convenience */
494 
495 	struct task tx_reclaim_task;
496 	/* stats for common events first */
497 
498 	uint64_t txcsum;	/* # of times hardware assisted with checksum */
499 	uint64_t tso_wrs;	/* # of TSO work requests */
500 	uint64_t vlan_insertion;/* # of times VLAN tag was inserted */
501 	uint64_t imm_wrs;	/* # of work requests with immediate data */
502 	uint64_t sgl_wrs;	/* # of work requests with direct SGL */
503 	uint64_t txpkt_wrs;	/* # of txpkt work requests (not coalesced) */
504 	uint64_t txpkts0_wrs;	/* # of type0 coalesced tx work requests */
505 	uint64_t txpkts1_wrs;	/* # of type1 coalesced tx work requests */
506 	uint64_t txpkts0_pkts;	/* # of frames in type0 coalesced tx WRs */
507 	uint64_t txpkts1_pkts;	/* # of frames in type1 coalesced tx WRs */
508 
509 	/* stats for not-that-common events */
510 } __aligned(CACHE_LINE_SIZE);
511 
512 /* rxq: SGE ingress queue + SGE free list + miscellaneous items */
513 struct sge_rxq {
514 	struct sge_iq iq;	/* MUST be first */
515 	struct sge_fl fl;	/* MUST follow iq */
516 
517 	struct ifnet *ifp;	/* the interface this rxq belongs to */
518 #if defined(INET) || defined(INET6)
519 	struct lro_ctrl lro;	/* LRO state */
520 #endif
521 
522 	/* stats for common events first */
523 
524 	uint64_t rxcsum;	/* # of times hardware assisted with checksum */
525 	uint64_t vlan_extraction;/* # of times VLAN tag was extracted */
526 
527 	/* stats for not-that-common events */
528 
529 } __aligned(CACHE_LINE_SIZE);
530 
531 static inline struct sge_rxq *
532 iq_to_rxq(struct sge_iq *iq)
533 {
534 
535 	return (__containerof(iq, struct sge_rxq, iq));
536 }
537 
538 
539 #ifdef TCP_OFFLOAD
540 /* ofld_rxq: SGE ingress queue + SGE free list + miscellaneous items */
541 struct sge_ofld_rxq {
542 	struct sge_iq iq;	/* MUST be first */
543 	struct sge_fl fl;	/* MUST follow iq */
544 } __aligned(CACHE_LINE_SIZE);
545 
546 static inline struct sge_ofld_rxq *
547 iq_to_ofld_rxq(struct sge_iq *iq)
548 {
549 
550 	return (__containerof(iq, struct sge_ofld_rxq, iq));
551 }
552 #endif
553 
554 struct wrqe {
555 	STAILQ_ENTRY(wrqe) link;
556 	struct sge_wrq *wrq;
557 	int wr_len;
558 	char wr[] __aligned(16);
559 };
560 
561 struct wrq_cookie {
562 	TAILQ_ENTRY(wrq_cookie) link;
563 	int ndesc;
564 	int pidx;
565 };
566 
567 /*
568  * wrq: SGE egress queue that is given prebuilt work requests.  Both the control
569  * and offload tx queues are of this type.
570  */
571 struct sge_wrq {
572 	struct sge_eq eq;	/* MUST be first */
573 
574 	struct adapter *adapter;
575 	struct task wrq_tx_task;
576 
577 	/* Tx desc reserved but WR not "committed" yet. */
578 	TAILQ_HEAD(wrq_incomplete_wrs , wrq_cookie) incomplete_wrs;
579 
580 	/* List of WRs ready to go out as soon as descriptors are available. */
581 	STAILQ_HEAD(, wrqe) wr_list;
582 	u_int nwr_pending;
583 	u_int ndesc_needed;
584 
585 	/* stats for common events first */
586 
587 	uint64_t tx_wrs_direct;	/* # of WRs written directly to desc ring. */
588 	uint64_t tx_wrs_ss;	/* # of WRs copied from scratch space. */
589 	uint64_t tx_wrs_copied;	/* # of WRs queued and copied to desc ring. */
590 
591 	/* stats for not-that-common events */
592 
593 	/*
594 	 * Scratch space for work requests that wrap around after reaching the
595 	 * status page, and some infomation about the last WR that used it.
596 	 */
597 	uint16_t ss_pidx;
598 	uint16_t ss_len;
599 	uint8_t ss[SGE_MAX_WR_LEN];
600 
601 } __aligned(CACHE_LINE_SIZE);
602 
603 
604 #ifdef DEV_NETMAP
605 struct sge_nm_rxq {
606 	struct port_info *pi;
607 
608 	struct iq_desc *iq_desc;
609 	uint16_t iq_abs_id;
610 	uint16_t iq_cntxt_id;
611 	uint16_t iq_cidx;
612 	uint16_t iq_sidx;
613 	uint8_t iq_gen;
614 
615 	__be64  *fl_desc;
616 	uint16_t fl_cntxt_id;
617 	uint32_t fl_cidx;
618 	uint32_t fl_pidx;
619 	uint32_t fl_sidx;
620 	uint32_t fl_db_val;
621 	u_int fl_hwidx:4;
622 
623 	u_int nid;		/* netmap ring # for this queue */
624 
625 	/* infrequently used items after this */
626 
627 	bus_dma_tag_t iq_desc_tag;
628 	bus_dmamap_t iq_desc_map;
629 	bus_addr_t iq_ba;
630 	int intr_idx;
631 
632 	bus_dma_tag_t fl_desc_tag;
633 	bus_dmamap_t fl_desc_map;
634 	bus_addr_t fl_ba;
635 } __aligned(CACHE_LINE_SIZE);
636 
637 struct sge_nm_txq {
638 	struct tx_desc *desc;
639 	uint16_t cidx;
640 	uint16_t pidx;
641 	uint16_t sidx;
642 	uint16_t equiqidx;	/* EQUIQ last requested at this pidx */
643 	uint16_t equeqidx;	/* EQUEQ last requested at this pidx */
644 	uint16_t dbidx;		/* pidx of the most recent doorbell */
645 	uint16_t doorbells;
646 	volatile uint32_t *udb;
647 	u_int udb_qid;
648 	u_int cntxt_id;
649 	__be32 cpl_ctrl0;	/* for convenience */
650 	u_int nid;		/* netmap ring # for this queue */
651 
652 	/* infrequently used items after this */
653 
654 	bus_dma_tag_t desc_tag;
655 	bus_dmamap_t desc_map;
656 	bus_addr_t ba;
657 	int iqidx;
658 } __aligned(CACHE_LINE_SIZE);
659 #endif
660 
661 struct sge {
662 	int timer_val[SGE_NTIMERS];
663 	int counter_val[SGE_NCOUNTERS];
664 	int fl_starve_threshold;
665 	int fl_starve_threshold2;
666 	int eq_s_qpp;
667 	int iq_s_qpp;
668 
669 	int nrxq;	/* total # of Ethernet rx queues */
670 	int ntxq;	/* total # of Ethernet tx tx queues */
671 #ifdef TCP_OFFLOAD
672 	int nofldrxq;	/* total # of TOE rx queues */
673 	int nofldtxq;	/* total # of TOE tx queues */
674 #endif
675 #ifdef DEV_NETMAP
676 	int nnmrxq;	/* total # of netmap rx queues */
677 	int nnmtxq;	/* total # of netmap tx queues */
678 #endif
679 	int niq;	/* total # of ingress queues */
680 	int neq;	/* total # of egress queues */
681 
682 	struct sge_iq fwq;	/* Firmware event queue */
683 	struct sge_wrq mgmtq;	/* Management queue (control queue) */
684 	struct sge_wrq *ctrlq;	/* Control queues */
685 	struct sge_txq *txq;	/* NIC tx queues */
686 	struct sge_rxq *rxq;	/* NIC rx queues */
687 #ifdef TCP_OFFLOAD
688 	struct sge_wrq *ofld_txq;	/* TOE tx queues */
689 	struct sge_ofld_rxq *ofld_rxq;	/* TOE rx queues */
690 #endif
691 #ifdef DEV_NETMAP
692 	struct sge_nm_txq *nm_txq;	/* netmap tx queues */
693 	struct sge_nm_rxq *nm_rxq;	/* netmap rx queues */
694 #endif
695 
696 	uint16_t iq_start;
697 	int eq_start;
698 	struct sge_iq **iqmap;	/* iq->cntxt_id to iq mapping */
699 	struct sge_eq **eqmap;	/* eq->cntxt_id to eq mapping */
700 
701 	int pad_boundary;
702 	int pack_boundary;
703 	int8_t safe_hwidx1;	/* may not have room for metadata */
704 	int8_t safe_hwidx2;	/* with room for metadata and maybe more */
705 	struct sw_zone_info sw_zone_info[SW_ZONE_SIZES];
706 	struct hw_buf_info hw_buf_info[SGE_FLBUF_SIZES];
707 };
708 
709 struct rss_header;
710 typedef int (*cpl_handler_t)(struct sge_iq *, const struct rss_header *,
711     struct mbuf *);
712 typedef int (*an_handler_t)(struct sge_iq *, const struct rsp_ctrl *);
713 typedef int (*fw_msg_handler_t)(struct adapter *, const __be64 *);
714 
715 struct adapter {
716 	SLIST_ENTRY(adapter) link;
717 	device_t dev;
718 	struct cdev *cdev;
719 
720 	/* PCIe register resources */
721 	int regs_rid;
722 	struct resource *regs_res;
723 	int msix_rid;
724 	struct resource *msix_res;
725 	bus_space_handle_t bh;
726 	bus_space_tag_t bt;
727 	bus_size_t mmio_len;
728 	int udbs_rid;
729 	struct resource *udbs_res;
730 	volatile uint8_t *udbs_base;
731 
732 	unsigned int pf;
733 	unsigned int mbox;
734 
735 	/* Interrupt information */
736 	int intr_type;
737 	int intr_count;
738 	struct irq {
739 		struct resource *res;
740 		int rid;
741 		void *tag;
742 	} *irq;
743 
744 	bus_dma_tag_t dmat;	/* Parent DMA tag */
745 
746 	struct sge sge;
747 	int lro_timeout;
748 
749 	struct taskqueue *tq[NCHAN];	/* General purpose taskqueues */
750 	struct port_info *port[MAX_NPORTS];
751 	uint8_t chan_map[NCHAN];
752 
753 #ifdef TCP_OFFLOAD
754 	void *tom_softc;	/* (struct tom_data *) */
755 	struct tom_tunables tt;
756 	void *iwarp_softc;	/* (struct c4iw_dev *) */
757 	void *iscsi_softc;
758 #endif
759 	struct l2t_data *l2t;	/* L2 table */
760 	struct tid_info tids;
761 
762 	uint16_t doorbells;
763 	int open_device_map;
764 #ifdef TCP_OFFLOAD
765 	int offload_map;	/* ports with IFCAP_TOE enabled */
766 	int active_ulds;	/* ULDs activated on this adapter */
767 #endif
768 	int flags;
769 	int debug_flags;
770 
771 	char ifp_lockname[16];
772 	struct mtx ifp_lock;
773 	struct ifnet *ifp;	/* tracer ifp */
774 	struct ifmedia media;
775 	int traceq;		/* iq used by all tracers, -1 if none */
776 	int tracer_valid;	/* bitmap of valid tracers */
777 	int tracer_enabled;	/* bitmap of enabled tracers */
778 
779 	char fw_version[32];
780 	char cfg_file[32];
781 	u_int cfcsum;
782 	struct adapter_params params;
783 	struct t4_virt_res vres;
784 
785 	uint16_t linkcaps;
786 	uint16_t niccaps;
787 	uint16_t toecaps;
788 	uint16_t rdmacaps;
789 	uint16_t iscsicaps;
790 	uint16_t fcoecaps;
791 
792 	struct sysctl_ctx_list ctx; /* from adapter_full_init to full_uninit */
793 
794 	struct mtx sc_lock;
795 	char lockname[16];
796 
797 	/* Starving free lists */
798 	struct mtx sfl_lock;	/* same cache-line as sc_lock? but that's ok */
799 	TAILQ_HEAD(, sge_fl) sfl;
800 	struct callout sfl_callout;
801 
802 	struct mtx regwin_lock;	/* for indirect reads and memory windows */
803 
804 	an_handler_t an_handler __aligned(CACHE_LINE_SIZE);
805 	fw_msg_handler_t fw_msg_handler[5];	/* NUM_FW6_TYPES */
806 	cpl_handler_t cpl_handler[0xef];	/* NUM_CPL_CMDS */
807 
808 #ifdef INVARIANTS
809 	const char *last_op;
810 	const void *last_op_thr;
811 #endif
812 
813 	int sc_do_rxcopy;
814 };
815 
816 #define ADAPTER_LOCK(sc)		mtx_lock(&(sc)->sc_lock)
817 #define ADAPTER_UNLOCK(sc)		mtx_unlock(&(sc)->sc_lock)
818 #define ADAPTER_LOCK_ASSERT_OWNED(sc)	mtx_assert(&(sc)->sc_lock, MA_OWNED)
819 #define ADAPTER_LOCK_ASSERT_NOTOWNED(sc) mtx_assert(&(sc)->sc_lock, MA_NOTOWNED)
820 
821 #define ASSERT_SYNCHRONIZED_OP(sc)	\
822     KASSERT(IS_BUSY(sc) && \
823 	(mtx_owned(&(sc)->sc_lock) || sc->last_op_thr == curthread), \
824 	("%s: operation not synchronized.", __func__))
825 
826 #define PORT_LOCK(pi)			mtx_lock(&(pi)->pi_lock)
827 #define PORT_UNLOCK(pi)			mtx_unlock(&(pi)->pi_lock)
828 #define PORT_LOCK_ASSERT_OWNED(pi)	mtx_assert(&(pi)->pi_lock, MA_OWNED)
829 #define PORT_LOCK_ASSERT_NOTOWNED(pi)	mtx_assert(&(pi)->pi_lock, MA_NOTOWNED)
830 
831 #define FL_LOCK(fl)			mtx_lock(&(fl)->fl_lock)
832 #define FL_TRYLOCK(fl)			mtx_trylock(&(fl)->fl_lock)
833 #define FL_UNLOCK(fl)			mtx_unlock(&(fl)->fl_lock)
834 #define FL_LOCK_ASSERT_OWNED(fl)	mtx_assert(&(fl)->fl_lock, MA_OWNED)
835 #define FL_LOCK_ASSERT_NOTOWNED(fl)	mtx_assert(&(fl)->fl_lock, MA_NOTOWNED)
836 
837 #define RXQ_FL_LOCK(rxq)		FL_LOCK(&(rxq)->fl)
838 #define RXQ_FL_UNLOCK(rxq)		FL_UNLOCK(&(rxq)->fl)
839 #define RXQ_FL_LOCK_ASSERT_OWNED(rxq)	FL_LOCK_ASSERT_OWNED(&(rxq)->fl)
840 #define RXQ_FL_LOCK_ASSERT_NOTOWNED(rxq) FL_LOCK_ASSERT_NOTOWNED(&(rxq)->fl)
841 
842 #define EQ_LOCK(eq)			mtx_lock(&(eq)->eq_lock)
843 #define EQ_TRYLOCK(eq)			mtx_trylock(&(eq)->eq_lock)
844 #define EQ_UNLOCK(eq)			mtx_unlock(&(eq)->eq_lock)
845 #define EQ_LOCK_ASSERT_OWNED(eq)	mtx_assert(&(eq)->eq_lock, MA_OWNED)
846 #define EQ_LOCK_ASSERT_NOTOWNED(eq)	mtx_assert(&(eq)->eq_lock, MA_NOTOWNED)
847 
848 #define TXQ_LOCK(txq)			EQ_LOCK(&(txq)->eq)
849 #define TXQ_TRYLOCK(txq)		EQ_TRYLOCK(&(txq)->eq)
850 #define TXQ_UNLOCK(txq)			EQ_UNLOCK(&(txq)->eq)
851 #define TXQ_LOCK_ASSERT_OWNED(txq)	EQ_LOCK_ASSERT_OWNED(&(txq)->eq)
852 #define TXQ_LOCK_ASSERT_NOTOWNED(txq)	EQ_LOCK_ASSERT_NOTOWNED(&(txq)->eq)
853 
854 #define CH_DUMP_MBOX(sc, mbox, data_reg) \
855 	do { \
856 		if (sc->debug_flags & DF_DUMP_MBOX) { \
857 			log(LOG_NOTICE, \
858 			    "%s mbox %u: %016llx %016llx %016llx %016llx " \
859 			    "%016llx %016llx %016llx %016llx\n", \
860 			    device_get_nameunit(sc->dev), mbox, \
861 			    (unsigned long long)t4_read_reg64(sc, data_reg), \
862 			    (unsigned long long)t4_read_reg64(sc, data_reg + 8), \
863 			    (unsigned long long)t4_read_reg64(sc, data_reg + 16), \
864 			    (unsigned long long)t4_read_reg64(sc, data_reg + 24), \
865 			    (unsigned long long)t4_read_reg64(sc, data_reg + 32), \
866 			    (unsigned long long)t4_read_reg64(sc, data_reg + 40), \
867 			    (unsigned long long)t4_read_reg64(sc, data_reg + 48), \
868 			    (unsigned long long)t4_read_reg64(sc, data_reg + 56)); \
869 		} \
870 	} while (0)
871 
872 #define for_each_txq(pi, iter, q) \
873 	for (q = &pi->adapter->sge.txq[pi->first_txq], iter = 0; \
874 	    iter < pi->ntxq; ++iter, ++q)
875 #define for_each_rxq(pi, iter, q) \
876 	for (q = &pi->adapter->sge.rxq[pi->first_rxq], iter = 0; \
877 	    iter < pi->nrxq; ++iter, ++q)
878 #define for_each_ofld_txq(pi, iter, q) \
879 	for (q = &pi->adapter->sge.ofld_txq[pi->first_ofld_txq], iter = 0; \
880 	    iter < pi->nofldtxq; ++iter, ++q)
881 #define for_each_ofld_rxq(pi, iter, q) \
882 	for (q = &pi->adapter->sge.ofld_rxq[pi->first_ofld_rxq], iter = 0; \
883 	    iter < pi->nofldrxq; ++iter, ++q)
884 #define for_each_nm_txq(pi, iter, q) \
885 	for (q = &pi->adapter->sge.nm_txq[pi->first_nm_txq], iter = 0; \
886 	    iter < pi->nnmtxq; ++iter, ++q)
887 #define for_each_nm_rxq(pi, iter, q) \
888 	for (q = &pi->adapter->sge.nm_rxq[pi->first_nm_rxq], iter = 0; \
889 	    iter < pi->nnmrxq; ++iter, ++q)
890 
891 #define IDXINCR(idx, incr, wrap) do { \
892 	idx = wrap - idx > incr ? idx + incr : incr - (wrap - idx); \
893 } while (0)
894 #define IDXDIFF(head, tail, wrap) \
895 	((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head))
896 
897 /* One for errors, one for firmware events */
898 #define T4_EXTRA_INTR 2
899 
900 static inline uint32_t
901 t4_read_reg(struct adapter *sc, uint32_t reg)
902 {
903 
904 	return bus_space_read_4(sc->bt, sc->bh, reg);
905 }
906 
907 static inline void
908 t4_write_reg(struct adapter *sc, uint32_t reg, uint32_t val)
909 {
910 
911 	bus_space_write_4(sc->bt, sc->bh, reg, val);
912 }
913 
914 static inline uint64_t
915 t4_read_reg64(struct adapter *sc, uint32_t reg)
916 {
917 
918 	return t4_bus_space_read_8(sc->bt, sc->bh, reg);
919 }
920 
921 static inline void
922 t4_write_reg64(struct adapter *sc, uint32_t reg, uint64_t val)
923 {
924 
925 	t4_bus_space_write_8(sc->bt, sc->bh, reg, val);
926 }
927 
928 static inline void
929 t4_os_pci_read_cfg1(struct adapter *sc, int reg, uint8_t *val)
930 {
931 
932 	*val = pci_read_config(sc->dev, reg, 1);
933 }
934 
935 static inline void
936 t4_os_pci_write_cfg1(struct adapter *sc, int reg, uint8_t val)
937 {
938 
939 	pci_write_config(sc->dev, reg, val, 1);
940 }
941 
942 static inline void
943 t4_os_pci_read_cfg2(struct adapter *sc, int reg, uint16_t *val)
944 {
945 
946 	*val = pci_read_config(sc->dev, reg, 2);
947 }
948 
949 static inline void
950 t4_os_pci_write_cfg2(struct adapter *sc, int reg, uint16_t val)
951 {
952 
953 	pci_write_config(sc->dev, reg, val, 2);
954 }
955 
956 static inline void
957 t4_os_pci_read_cfg4(struct adapter *sc, int reg, uint32_t *val)
958 {
959 
960 	*val = pci_read_config(sc->dev, reg, 4);
961 }
962 
963 static inline void
964 t4_os_pci_write_cfg4(struct adapter *sc, int reg, uint32_t val)
965 {
966 
967 	pci_write_config(sc->dev, reg, val, 4);
968 }
969 
970 static inline struct port_info *
971 adap2pinfo(struct adapter *sc, int idx)
972 {
973 
974 	return (sc->port[idx]);
975 }
976 
977 static inline void
978 t4_os_set_hw_addr(struct adapter *sc, int idx, uint8_t hw_addr[])
979 {
980 
981 	bcopy(hw_addr, sc->port[idx]->hw_addr, ETHER_ADDR_LEN);
982 }
983 
984 static inline bool
985 is_10G_port(const struct port_info *pi)
986 {
987 
988 	return ((pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G) != 0);
989 }
990 
991 static inline bool
992 is_40G_port(const struct port_info *pi)
993 {
994 
995 	return ((pi->link_cfg.supported & FW_PORT_CAP_SPEED_40G) != 0);
996 }
997 
998 static inline int
999 tx_resume_threshold(struct sge_eq *eq)
1000 {
1001 
1002 	/* not quite the same as qsize / 4, but this will do. */
1003 	return (eq->sidx / 4);
1004 }
1005 
1006 /* t4_main.c */
1007 int t4_os_find_pci_capability(struct adapter *, int);
1008 int t4_os_pci_save_state(struct adapter *);
1009 int t4_os_pci_restore_state(struct adapter *);
1010 void t4_os_portmod_changed(const struct adapter *, int);
1011 void t4_os_link_changed(struct adapter *, int, int, int);
1012 void t4_iterate(void (*)(struct adapter *, void *), void *);
1013 int t4_register_cpl_handler(struct adapter *, int, cpl_handler_t);
1014 int t4_register_an_handler(struct adapter *, an_handler_t);
1015 int t4_register_fw_msg_handler(struct adapter *, int, fw_msg_handler_t);
1016 int t4_filter_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1017 int begin_synchronized_op(struct adapter *, struct port_info *, int, char *);
1018 void end_synchronized_op(struct adapter *, int);
1019 int update_mac_settings(struct ifnet *, int);
1020 int adapter_full_init(struct adapter *);
1021 int adapter_full_uninit(struct adapter *);
1022 int port_full_init(struct port_info *);
1023 int port_full_uninit(struct port_info *);
1024 
1025 #ifdef DEV_NETMAP
1026 /* t4_netmap.c */
1027 int create_netmap_ifnet(struct port_info *);
1028 int destroy_netmap_ifnet(struct port_info *);
1029 void t4_nm_intr(void *);
1030 #endif
1031 
1032 /* t4_sge.c */
1033 void t4_sge_modload(void);
1034 void t4_sge_modunload(void);
1035 uint64_t t4_sge_extfree_refs(void);
1036 void t4_init_sge_cpl_handlers(struct adapter *);
1037 void t4_tweak_chip_settings(struct adapter *);
1038 int t4_read_chip_settings(struct adapter *);
1039 int t4_create_dma_tag(struct adapter *);
1040 void t4_sge_sysctls(struct adapter *, struct sysctl_ctx_list *,
1041     struct sysctl_oid_list *);
1042 int t4_destroy_dma_tag(struct adapter *);
1043 int t4_setup_adapter_queues(struct adapter *);
1044 int t4_teardown_adapter_queues(struct adapter *);
1045 int t4_setup_port_queues(struct port_info *);
1046 int t4_teardown_port_queues(struct port_info *);
1047 void t4_intr_all(void *);
1048 void t4_intr(void *);
1049 void t4_intr_err(void *);
1050 void t4_intr_evt(void *);
1051 void t4_wrq_tx_locked(struct adapter *, struct sge_wrq *, struct wrqe *);
1052 void t4_update_fl_bufsize(struct ifnet *);
1053 int parse_pkt(struct mbuf **);
1054 void *start_wrq_wr(struct sge_wrq *, int, struct wrq_cookie *);
1055 void commit_wrq_wr(struct sge_wrq *, void *, struct wrq_cookie *);
1056 int tnl_cong(struct port_info *, int);
1057 
1058 /* t4_tracer.c */
1059 struct t4_tracer;
1060 void t4_tracer_modload(void);
1061 void t4_tracer_modunload(void);
1062 void t4_tracer_port_detach(struct adapter *);
1063 int t4_get_tracer(struct adapter *, struct t4_tracer *);
1064 int t4_set_tracer(struct adapter *, struct t4_tracer *);
1065 int t4_trace_pkt(struct sge_iq *, const struct rss_header *, struct mbuf *);
1066 int t5_trace_pkt(struct sge_iq *, const struct rss_header *, struct mbuf *);
1067 
1068 static inline struct wrqe *
1069 alloc_wrqe(int wr_len, struct sge_wrq *wrq)
1070 {
1071 	int len = offsetof(struct wrqe, wr) + wr_len;
1072 	struct wrqe *wr;
1073 
1074 	wr = malloc(len, M_CXGBE, M_NOWAIT);
1075 	if (__predict_false(wr == NULL))
1076 		return (NULL);
1077 	wr->wr_len = wr_len;
1078 	wr->wrq = wrq;
1079 	return (wr);
1080 }
1081 
1082 static inline void *
1083 wrtod(struct wrqe *wr)
1084 {
1085 	return (&wr->wr[0]);
1086 }
1087 
1088 static inline void
1089 free_wrqe(struct wrqe *wr)
1090 {
1091 	free(wr, M_CXGBE);
1092 }
1093 
1094 static inline void
1095 t4_wrq_tx(struct adapter *sc, struct wrqe *wr)
1096 {
1097 	struct sge_wrq *wrq = wr->wrq;
1098 
1099 	TXQ_LOCK(wrq);
1100 	t4_wrq_tx_locked(sc, wrq, wr);
1101 	TXQ_UNLOCK(wrq);
1102 }
1103 
1104 #endif
1105