Loading...
1/*
2 * Copyright (c) 2006, 2007, 2008, 2009, 2010 QLogic Corporation.
3 * All rights reserved.
4 * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
5 *
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
11 *
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
14 * conditions are met:
15 *
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
18 * disclaimer.
19 *
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
24 *
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 * SOFTWARE.
33 */
34
35#ifndef QIB_VERBS_H
36#define QIB_VERBS_H
37
38#include <linux/types.h>
39#include <linux/spinlock.h>
40#include <linux/kernel.h>
41#include <linux/interrupt.h>
42#include <linux/kref.h>
43#include <linux/workqueue.h>
44#include <rdma/ib_pack.h>
45#include <rdma/ib_user_verbs.h>
46
47struct qib_ctxtdata;
48struct qib_pportdata;
49struct qib_devdata;
50struct qib_verbs_txreq;
51
52#define QIB_MAX_RDMA_ATOMIC 16
53#define QIB_GUIDS_PER_PORT 5
54
55#define QPN_MAX (1 << 24)
56#define QPNMAP_ENTRIES (QPN_MAX / PAGE_SIZE / BITS_PER_BYTE)
57
58/*
59 * Increment this value if any changes that break userspace ABI
60 * compatibility are made.
61 */
62#define QIB_UVERBS_ABI_VERSION 2
63
64/*
65 * Define an ib_cq_notify value that is not valid so we know when CQ
66 * notifications are armed.
67 */
68#define IB_CQ_NONE (IB_CQ_NEXT_COMP + 1)
69
70#define IB_SEQ_NAK (3 << 29)
71
72/* AETH NAK opcode values */
73#define IB_RNR_NAK 0x20
74#define IB_NAK_PSN_ERROR 0x60
75#define IB_NAK_INVALID_REQUEST 0x61
76#define IB_NAK_REMOTE_ACCESS_ERROR 0x62
77#define IB_NAK_REMOTE_OPERATIONAL_ERROR 0x63
78#define IB_NAK_INVALID_RD_REQUEST 0x64
79
80/* Flags for checking QP state (see ib_qib_state_ops[]) */
81#define QIB_POST_SEND_OK 0x01
82#define QIB_POST_RECV_OK 0x02
83#define QIB_PROCESS_RECV_OK 0x04
84#define QIB_PROCESS_SEND_OK 0x08
85#define QIB_PROCESS_NEXT_SEND_OK 0x10
86#define QIB_FLUSH_SEND 0x20
87#define QIB_FLUSH_RECV 0x40
88#define QIB_PROCESS_OR_FLUSH_SEND \
89 (QIB_PROCESS_SEND_OK | QIB_FLUSH_SEND)
90
91/* IB Performance Manager status values */
92#define IB_PMA_SAMPLE_STATUS_DONE 0x00
93#define IB_PMA_SAMPLE_STATUS_STARTED 0x01
94#define IB_PMA_SAMPLE_STATUS_RUNNING 0x02
95
96/* Mandatory IB performance counter select values. */
97#define IB_PMA_PORT_XMIT_DATA cpu_to_be16(0x0001)
98#define IB_PMA_PORT_RCV_DATA cpu_to_be16(0x0002)
99#define IB_PMA_PORT_XMIT_PKTS cpu_to_be16(0x0003)
100#define IB_PMA_PORT_RCV_PKTS cpu_to_be16(0x0004)
101#define IB_PMA_PORT_XMIT_WAIT cpu_to_be16(0x0005)
102
103#define QIB_VENDOR_IPG cpu_to_be16(0xFFA0)
104
105#define IB_BTH_REQ_ACK (1 << 31)
106#define IB_BTH_SOLICITED (1 << 23)
107#define IB_BTH_MIG_REQ (1 << 22)
108
109/* XXX Should be defined in ib_verbs.h enum ib_port_cap_flags */
110#define IB_PORT_OTHER_LOCAL_CHANGES_SUP (1 << 26)
111
112#define IB_GRH_VERSION 6
113#define IB_GRH_VERSION_MASK 0xF
114#define IB_GRH_VERSION_SHIFT 28
115#define IB_GRH_TCLASS_MASK 0xFF
116#define IB_GRH_TCLASS_SHIFT 20
117#define IB_GRH_FLOW_MASK 0xFFFFF
118#define IB_GRH_FLOW_SHIFT 0
119#define IB_GRH_NEXT_HDR 0x1B
120
121#define IB_DEFAULT_GID_PREFIX cpu_to_be64(0xfe80000000000000ULL)
122
123/* Values for set/get portinfo VLCap OperationalVLs */
124#define IB_VL_VL0 1
125#define IB_VL_VL0_1 2
126#define IB_VL_VL0_3 3
127#define IB_VL_VL0_7 4
128#define IB_VL_VL0_14 5
129
130static inline int qib_num_vls(int vls)
131{
132 switch (vls) {
133 default:
134 case IB_VL_VL0:
135 return 1;
136 case IB_VL_VL0_1:
137 return 2;
138 case IB_VL_VL0_3:
139 return 4;
140 case IB_VL_VL0_7:
141 return 8;
142 case IB_VL_VL0_14:
143 return 15;
144 }
145}
146
147struct ib_reth {
148 __be64 vaddr;
149 __be32 rkey;
150 __be32 length;
151} __attribute__ ((packed));
152
153struct ib_atomic_eth {
154 __be32 vaddr[2]; /* unaligned so access as 2 32-bit words */
155 __be32 rkey;
156 __be64 swap_data;
157 __be64 compare_data;
158} __attribute__ ((packed));
159
160struct qib_other_headers {
161 __be32 bth[3];
162 union {
163 struct {
164 __be32 deth[2];
165 __be32 imm_data;
166 } ud;
167 struct {
168 struct ib_reth reth;
169 __be32 imm_data;
170 } rc;
171 struct {
172 __be32 aeth;
173 __be32 atomic_ack_eth[2];
174 } at;
175 __be32 imm_data;
176 __be32 aeth;
177 struct ib_atomic_eth atomic_eth;
178 } u;
179} __attribute__ ((packed));
180
181/*
182 * Note that UD packets with a GRH header are 8+40+12+8 = 68 bytes
183 * long (72 w/ imm_data). Only the first 56 bytes of the IB header
184 * will be in the eager header buffer. The remaining 12 or 16 bytes
185 * are in the data buffer.
186 */
187struct qib_ib_header {
188 __be16 lrh[4];
189 union {
190 struct {
191 struct ib_grh grh;
192 struct qib_other_headers oth;
193 } l;
194 struct qib_other_headers oth;
195 } u;
196} __attribute__ ((packed));
197
198struct qib_pio_header {
199 __le32 pbc[2];
200 struct qib_ib_header hdr;
201} __attribute__ ((packed));
202
203/*
204 * There is one struct qib_mcast for each multicast GID.
205 * All attached QPs are then stored as a list of
206 * struct qib_mcast_qp.
207 */
208struct qib_mcast_qp {
209 struct list_head list;
210 struct qib_qp *qp;
211};
212
213struct qib_mcast {
214 struct rb_node rb_node;
215 union ib_gid mgid;
216 struct list_head qp_list;
217 wait_queue_head_t wait;
218 atomic_t refcount;
219 int n_attached;
220};
221
222/* Protection domain */
223struct qib_pd {
224 struct ib_pd ibpd;
225 int user; /* non-zero if created from user space */
226};
227
228/* Address Handle */
229struct qib_ah {
230 struct ib_ah ibah;
231 struct ib_ah_attr attr;
232 atomic_t refcount;
233};
234
235/*
236 * This structure is used by qib_mmap() to validate an offset
237 * when an mmap() request is made. The vm_area_struct then uses
238 * this as its vm_private_data.
239 */
240struct qib_mmap_info {
241 struct list_head pending_mmaps;
242 struct ib_ucontext *context;
243 void *obj;
244 __u64 offset;
245 struct kref ref;
246 unsigned size;
247};
248
249/*
250 * This structure is used to contain the head pointer, tail pointer,
251 * and completion queue entries as a single memory allocation so
252 * it can be mmap'ed into user space.
253 */
254struct qib_cq_wc {
255 u32 head; /* index of next entry to fill */
256 u32 tail; /* index of next ib_poll_cq() entry */
257 union {
258 /* these are actually size ibcq.cqe + 1 */
259 struct ib_uverbs_wc uqueue[0];
260 struct ib_wc kqueue[0];
261 };
262};
263
264/*
265 * The completion queue structure.
266 */
267struct qib_cq {
268 struct ib_cq ibcq;
269 struct work_struct comptask;
270 spinlock_t lock; /* protect changes in this struct */
271 u8 notify;
272 u8 triggered;
273 struct qib_cq_wc *queue;
274 struct qib_mmap_info *ip;
275};
276
277/*
278 * A segment is a linear region of low physical memory.
279 * XXX Maybe we should use phys addr here and kmap()/kunmap().
280 * Used by the verbs layer.
281 */
282struct qib_seg {
283 void *vaddr;
284 size_t length;
285};
286
287/* The number of qib_segs that fit in a page. */
288#define QIB_SEGSZ (PAGE_SIZE / sizeof(struct qib_seg))
289
290struct qib_segarray {
291 struct qib_seg segs[QIB_SEGSZ];
292};
293
294struct qib_mregion {
295 struct ib_pd *pd; /* shares refcnt of ibmr.pd */
296 u64 user_base; /* User's address for this region */
297 u64 iova; /* IB start address of this region */
298 size_t length;
299 u32 lkey;
300 u32 offset; /* offset (bytes) to start of region */
301 int access_flags;
302 u32 max_segs; /* number of qib_segs in all the arrays */
303 u32 mapsz; /* size of the map array */
304 u8 page_shift; /* 0 - non unform/non powerof2 sizes */
305 atomic_t refcount;
306 struct qib_segarray *map[0]; /* the segments */
307};
308
309/*
310 * These keep track of the copy progress within a memory region.
311 * Used by the verbs layer.
312 */
313struct qib_sge {
314 struct qib_mregion *mr;
315 void *vaddr; /* kernel virtual address of segment */
316 u32 sge_length; /* length of the SGE */
317 u32 length; /* remaining length of the segment */
318 u16 m; /* current index: mr->map[m] */
319 u16 n; /* current index: mr->map[m]->segs[n] */
320};
321
322/* Memory region */
323struct qib_mr {
324 struct ib_mr ibmr;
325 struct ib_umem *umem;
326 struct qib_mregion mr; /* must be last */
327};
328
329/*
330 * Send work request queue entry.
331 * The size of the sg_list is determined when the QP is created and stored
332 * in qp->s_max_sge.
333 */
334struct qib_swqe {
335 struct ib_send_wr wr; /* don't use wr.sg_list */
336 u32 psn; /* first packet sequence number */
337 u32 lpsn; /* last packet sequence number */
338 u32 ssn; /* send sequence number */
339 u32 length; /* total length of data in sg_list */
340 struct qib_sge sg_list[0];
341};
342
343/*
344 * Receive work request queue entry.
345 * The size of the sg_list is determined when the QP (or SRQ) is created
346 * and stored in qp->r_rq.max_sge (or srq->rq.max_sge).
347 */
348struct qib_rwqe {
349 u64 wr_id;
350 u8 num_sge;
351 struct ib_sge sg_list[0];
352};
353
354/*
355 * This structure is used to contain the head pointer, tail pointer,
356 * and receive work queue entries as a single memory allocation so
357 * it can be mmap'ed into user space.
358 * Note that the wq array elements are variable size so you can't
359 * just index into the array to get the N'th element;
360 * use get_rwqe_ptr() instead.
361 */
362struct qib_rwq {
363 u32 head; /* new work requests posted to the head */
364 u32 tail; /* receives pull requests from here. */
365 struct qib_rwqe wq[0];
366};
367
368struct qib_rq {
369 struct qib_rwq *wq;
370 u32 size; /* size of RWQE array */
371 u8 max_sge;
372 spinlock_t lock /* protect changes in this struct */
373 ____cacheline_aligned_in_smp;
374};
375
376struct qib_srq {
377 struct ib_srq ibsrq;
378 struct qib_rq rq;
379 struct qib_mmap_info *ip;
380 /* send signal when number of RWQEs < limit */
381 u32 limit;
382};
383
384struct qib_sge_state {
385 struct qib_sge *sg_list; /* next SGE to be used if any */
386 struct qib_sge sge; /* progress state for the current SGE */
387 u32 total_len;
388 u8 num_sge;
389};
390
391/*
392 * This structure holds the information that the send tasklet needs
393 * to send a RDMA read response or atomic operation.
394 */
395struct qib_ack_entry {
396 u8 opcode;
397 u8 sent;
398 u32 psn;
399 u32 lpsn;
400 union {
401 struct qib_sge rdma_sge;
402 u64 atomic_data;
403 };
404};
405
406/*
407 * Variables prefixed with s_ are for the requester (sender).
408 * Variables prefixed with r_ are for the responder (receiver).
409 * Variables prefixed with ack_ are for responder replies.
410 *
411 * Common variables are protected by both r_rq.lock and s_lock in that order
412 * which only happens in modify_qp() or changing the QP 'state'.
413 */
414struct qib_qp {
415 struct ib_qp ibqp;
416 /* read mostly fields above and below */
417 struct ib_ah_attr remote_ah_attr;
418 struct ib_ah_attr alt_ah_attr;
419 struct qib_qp *next; /* link list for QPN hash table */
420 struct qib_swqe *s_wq; /* send work queue */
421 struct qib_mmap_info *ip;
422 struct qib_ib_header *s_hdr; /* next packet header to send */
423 unsigned long timeout_jiffies; /* computed from timeout */
424
425 enum ib_mtu path_mtu;
426 u32 remote_qpn;
427 u32 pmtu; /* decoded from path_mtu */
428 u32 qkey; /* QKEY for this QP (for UD or RD) */
429 u32 s_size; /* send work queue size */
430 u32 s_rnr_timeout; /* number of milliseconds for RNR timeout */
431
432 u8 state; /* QP state */
433 u8 qp_access_flags;
434 u8 alt_timeout; /* Alternate path timeout for this QP */
435 u8 timeout; /* Timeout for this QP */
436 u8 s_srate;
437 u8 s_mig_state;
438 u8 port_num;
439 u8 s_pkey_index; /* PKEY index to use */
440 u8 s_alt_pkey_index; /* Alternate path PKEY index to use */
441 u8 r_max_rd_atomic; /* max number of RDMA read/atomic to receive */
442 u8 s_max_rd_atomic; /* max number of RDMA read/atomic to send */
443 u8 s_retry_cnt; /* number of times to retry */
444 u8 s_rnr_retry_cnt;
445 u8 r_min_rnr_timer; /* retry timeout value for RNR NAKs */
446 u8 s_max_sge; /* size of s_wq->sg_list */
447 u8 s_draining;
448
449 /* start of read/write fields */
450
451 atomic_t refcount ____cacheline_aligned_in_smp;
452 wait_queue_head_t wait;
453
454
455 struct qib_ack_entry s_ack_queue[QIB_MAX_RDMA_ATOMIC + 1]
456 ____cacheline_aligned_in_smp;
457 struct qib_sge_state s_rdma_read_sge;
458
459 spinlock_t r_lock ____cacheline_aligned_in_smp; /* used for APM */
460 unsigned long r_aflags;
461 u64 r_wr_id; /* ID for current receive WQE */
462 u32 r_ack_psn; /* PSN for next ACK or atomic ACK */
463 u32 r_len; /* total length of r_sge */
464 u32 r_rcv_len; /* receive data len processed */
465 u32 r_psn; /* expected rcv packet sequence number */
466 u32 r_msn; /* message sequence number */
467
468 u8 r_state; /* opcode of last packet received */
469 u8 r_flags;
470 u8 r_head_ack_queue; /* index into s_ack_queue[] */
471
472 struct list_head rspwait; /* link for waititing to respond */
473
474 struct qib_sge_state r_sge; /* current receive data */
475 struct qib_rq r_rq; /* receive work queue */
476
477 spinlock_t s_lock ____cacheline_aligned_in_smp;
478 struct qib_sge_state *s_cur_sge;
479 u32 s_flags;
480 struct qib_verbs_txreq *s_tx;
481 struct qib_swqe *s_wqe;
482 struct qib_sge_state s_sge; /* current send request data */
483 struct qib_mregion *s_rdma_mr;
484 atomic_t s_dma_busy;
485 u32 s_cur_size; /* size of send packet in bytes */
486 u32 s_len; /* total length of s_sge */
487 u32 s_rdma_read_len; /* total length of s_rdma_read_sge */
488 u32 s_next_psn; /* PSN for next request */
489 u32 s_last_psn; /* last response PSN processed */
490 u32 s_sending_psn; /* lowest PSN that is being sent */
491 u32 s_sending_hpsn; /* highest PSN that is being sent */
492 u32 s_psn; /* current packet sequence number */
493 u32 s_ack_rdma_psn; /* PSN for sending RDMA read responses */
494 u32 s_ack_psn; /* PSN for acking sends and RDMA writes */
495 u32 s_head; /* new entries added here */
496 u32 s_tail; /* next entry to process */
497 u32 s_cur; /* current work queue entry */
498 u32 s_acked; /* last un-ACK'ed entry */
499 u32 s_last; /* last completed entry */
500 u32 s_ssn; /* SSN of tail entry */
501 u32 s_lsn; /* limit sequence number (credit) */
502 u16 s_hdrwords; /* size of s_hdr in 32 bit words */
503 u16 s_rdma_ack_cnt;
504 u8 s_state; /* opcode of last packet sent */
505 u8 s_ack_state; /* opcode of packet to ACK */
506 u8 s_nak_state; /* non-zero if NAK is pending */
507 u8 r_nak_state; /* non-zero if NAK is pending */
508 u8 s_retry; /* requester retry counter */
509 u8 s_rnr_retry; /* requester RNR retry counter */
510 u8 s_num_rd_atomic; /* number of RDMA read/atomic pending */
511 u8 s_tail_ack_queue; /* index into s_ack_queue[] */
512
513 struct qib_sge_state s_ack_rdma_sge;
514 struct timer_list s_timer;
515 struct list_head iowait; /* link for wait PIO buf */
516
517 struct work_struct s_work;
518
519 wait_queue_head_t wait_dma;
520
521 struct qib_sge r_sg_list[0] /* verified SGEs */
522 ____cacheline_aligned_in_smp;
523};
524
525/*
526 * Atomic bit definitions for r_aflags.
527 */
528#define QIB_R_WRID_VALID 0
529#define QIB_R_REWIND_SGE 1
530
531/*
532 * Bit definitions for r_flags.
533 */
534#define QIB_R_REUSE_SGE 0x01
535#define QIB_R_RDMAR_SEQ 0x02
536#define QIB_R_RSP_NAK 0x04
537#define QIB_R_RSP_SEND 0x08
538#define QIB_R_COMM_EST 0x10
539
540/*
541 * Bit definitions for s_flags.
542 *
543 * QIB_S_SIGNAL_REQ_WR - set if QP send WRs contain completion signaled
544 * QIB_S_BUSY - send tasklet is processing the QP
545 * QIB_S_TIMER - the RC retry timer is active
546 * QIB_S_ACK_PENDING - an ACK is waiting to be sent after RDMA read/atomics
547 * QIB_S_WAIT_FENCE - waiting for all prior RDMA read or atomic SWQEs
548 * before processing the next SWQE
549 * QIB_S_WAIT_RDMAR - waiting for a RDMA read or atomic SWQE to complete
550 * before processing the next SWQE
551 * QIB_S_WAIT_RNR - waiting for RNR timeout
552 * QIB_S_WAIT_SSN_CREDIT - waiting for RC credits to process next SWQE
553 * QIB_S_WAIT_DMA - waiting for send DMA queue to drain before generating
554 * next send completion entry not via send DMA
555 * QIB_S_WAIT_PIO - waiting for a send buffer to be available
556 * QIB_S_WAIT_TX - waiting for a struct qib_verbs_txreq to be available
557 * QIB_S_WAIT_DMA_DESC - waiting for DMA descriptors to be available
558 * QIB_S_WAIT_KMEM - waiting for kernel memory to be available
559 * QIB_S_WAIT_PSN - waiting for a packet to exit the send DMA queue
560 * QIB_S_WAIT_ACK - waiting for an ACK packet before sending more requests
561 * QIB_S_SEND_ONE - send one packet, request ACK, then wait for ACK
562 */
563#define QIB_S_SIGNAL_REQ_WR 0x0001
564#define QIB_S_BUSY 0x0002
565#define QIB_S_TIMER 0x0004
566#define QIB_S_RESP_PENDING 0x0008
567#define QIB_S_ACK_PENDING 0x0010
568#define QIB_S_WAIT_FENCE 0x0020
569#define QIB_S_WAIT_RDMAR 0x0040
570#define QIB_S_WAIT_RNR 0x0080
571#define QIB_S_WAIT_SSN_CREDIT 0x0100
572#define QIB_S_WAIT_DMA 0x0200
573#define QIB_S_WAIT_PIO 0x0400
574#define QIB_S_WAIT_TX 0x0800
575#define QIB_S_WAIT_DMA_DESC 0x1000
576#define QIB_S_WAIT_KMEM 0x2000
577#define QIB_S_WAIT_PSN 0x4000
578#define QIB_S_WAIT_ACK 0x8000
579#define QIB_S_SEND_ONE 0x10000
580#define QIB_S_UNLIMITED_CREDIT 0x20000
581
582/*
583 * Wait flags that would prevent any packet type from being sent.
584 */
585#define QIB_S_ANY_WAIT_IO (QIB_S_WAIT_PIO | QIB_S_WAIT_TX | \
586 QIB_S_WAIT_DMA_DESC | QIB_S_WAIT_KMEM)
587
588/*
589 * Wait flags that would prevent send work requests from making progress.
590 */
591#define QIB_S_ANY_WAIT_SEND (QIB_S_WAIT_FENCE | QIB_S_WAIT_RDMAR | \
592 QIB_S_WAIT_RNR | QIB_S_WAIT_SSN_CREDIT | QIB_S_WAIT_DMA | \
593 QIB_S_WAIT_PSN | QIB_S_WAIT_ACK)
594
595#define QIB_S_ANY_WAIT (QIB_S_ANY_WAIT_IO | QIB_S_ANY_WAIT_SEND)
596
597#define QIB_PSN_CREDIT 16
598
599/*
600 * Since struct qib_swqe is not a fixed size, we can't simply index into
601 * struct qib_qp.s_wq. This function does the array index computation.
602 */
603static inline struct qib_swqe *get_swqe_ptr(struct qib_qp *qp,
604 unsigned n)
605{
606 return (struct qib_swqe *)((char *)qp->s_wq +
607 (sizeof(struct qib_swqe) +
608 qp->s_max_sge *
609 sizeof(struct qib_sge)) * n);
610}
611
612/*
613 * Since struct qib_rwqe is not a fixed size, we can't simply index into
614 * struct qib_rwq.wq. This function does the array index computation.
615 */
616static inline struct qib_rwqe *get_rwqe_ptr(struct qib_rq *rq, unsigned n)
617{
618 return (struct qib_rwqe *)
619 ((char *) rq->wq->wq +
620 (sizeof(struct qib_rwqe) +
621 rq->max_sge * sizeof(struct ib_sge)) * n);
622}
623
624/*
625 * QPN-map pages start out as NULL, they get allocated upon
626 * first use and are never deallocated. This way,
627 * large bitmaps are not allocated unless large numbers of QPs are used.
628 */
629struct qpn_map {
630 void *page;
631};
632
633struct qib_qpn_table {
634 spinlock_t lock; /* protect changes in this struct */
635 unsigned flags; /* flags for QP0/1 allocated for each port */
636 u32 last; /* last QP number allocated */
637 u32 nmaps; /* size of the map table */
638 u16 limit;
639 u16 mask;
640 /* bit map of free QP numbers other than 0/1 */
641 struct qpn_map map[QPNMAP_ENTRIES];
642};
643
644struct qib_lkey_table {
645 spinlock_t lock; /* protect changes in this struct */
646 u32 next; /* next unused index (speeds search) */
647 u32 gen; /* generation count */
648 u32 max; /* size of the table */
649 struct qib_mregion **table;
650};
651
652struct qib_opcode_stats {
653 u64 n_packets; /* number of packets */
654 u64 n_bytes; /* total number of bytes */
655};
656
657struct qib_ibport {
658 struct qib_qp *qp0;
659 struct qib_qp *qp1;
660 struct ib_mad_agent *send_agent; /* agent for SMI (traps) */
661 struct qib_ah *sm_ah;
662 struct qib_ah *smi_ah;
663 struct rb_root mcast_tree;
664 spinlock_t lock; /* protect changes in this struct */
665
666 /* non-zero when timer is set */
667 unsigned long mkey_lease_timeout;
668 unsigned long trap_timeout;
669 __be64 gid_prefix; /* in network order */
670 __be64 mkey;
671 __be64 guids[QIB_GUIDS_PER_PORT - 1]; /* writable GUIDs */
672 u64 tid; /* TID for traps */
673 u64 n_unicast_xmit; /* total unicast packets sent */
674 u64 n_unicast_rcv; /* total unicast packets received */
675 u64 n_multicast_xmit; /* total multicast packets sent */
676 u64 n_multicast_rcv; /* total multicast packets received */
677 u64 z_symbol_error_counter; /* starting count for PMA */
678 u64 z_link_error_recovery_counter; /* starting count for PMA */
679 u64 z_link_downed_counter; /* starting count for PMA */
680 u64 z_port_rcv_errors; /* starting count for PMA */
681 u64 z_port_rcv_remphys_errors; /* starting count for PMA */
682 u64 z_port_xmit_discards; /* starting count for PMA */
683 u64 z_port_xmit_data; /* starting count for PMA */
684 u64 z_port_rcv_data; /* starting count for PMA */
685 u64 z_port_xmit_packets; /* starting count for PMA */
686 u64 z_port_rcv_packets; /* starting count for PMA */
687 u32 z_local_link_integrity_errors; /* starting count for PMA */
688 u32 z_excessive_buffer_overrun_errors; /* starting count for PMA */
689 u32 z_vl15_dropped; /* starting count for PMA */
690 u32 n_rc_resends;
691 u32 n_rc_acks;
692 u32 n_rc_qacks;
693 u32 n_rc_delayed_comp;
694 u32 n_seq_naks;
695 u32 n_rdma_seq;
696 u32 n_rnr_naks;
697 u32 n_other_naks;
698 u32 n_loop_pkts;
699 u32 n_pkt_drops;
700 u32 n_vl15_dropped;
701 u32 n_rc_timeouts;
702 u32 n_dmawait;
703 u32 n_unaligned;
704 u32 n_rc_dupreq;
705 u32 n_rc_seqnak;
706 u32 port_cap_flags;
707 u32 pma_sample_start;
708 u32 pma_sample_interval;
709 __be16 pma_counter_select[5];
710 u16 pma_tag;
711 u16 pkey_violations;
712 u16 qkey_violations;
713 u16 mkey_violations;
714 u16 mkey_lease_period;
715 u16 sm_lid;
716 u16 repress_traps;
717 u8 sm_sl;
718 u8 mkeyprot;
719 u8 subnet_timeout;
720 u8 vl_high_limit;
721 u8 sl_to_vl[16];
722
723 struct qib_opcode_stats opstats[128];
724};
725
726struct qib_ibdev {
727 struct ib_device ibdev;
728 struct list_head pending_mmaps;
729 spinlock_t mmap_offset_lock; /* protect mmap_offset */
730 u32 mmap_offset;
731 struct qib_mregion *dma_mr;
732
733 /* QP numbers are shared by all IB ports */
734 struct qib_qpn_table qpn_table;
735 struct qib_lkey_table lk_table;
736 struct list_head piowait; /* list for wait PIO buf */
737 struct list_head dmawait; /* list for wait DMA */
738 struct list_head txwait; /* list for wait qib_verbs_txreq */
739 struct list_head memwait; /* list for wait kernel memory */
740 struct list_head txreq_free;
741 struct timer_list mem_timer;
742 struct qib_qp **qp_table;
743 struct qib_pio_header *pio_hdrs;
744 dma_addr_t pio_hdrs_phys;
745 /* list of QPs waiting for RNR timer */
746 spinlock_t pending_lock; /* protect wait lists, PMA counters, etc. */
747 u32 qp_table_size; /* size of the hash table */
748 u32 qp_rnd; /* random bytes for hash */
749 spinlock_t qpt_lock;
750
751 u32 n_piowait;
752 u32 n_txwait;
753
754 u32 n_pds_allocated; /* number of PDs allocated for device */
755 spinlock_t n_pds_lock;
756 u32 n_ahs_allocated; /* number of AHs allocated for device */
757 spinlock_t n_ahs_lock;
758 u32 n_cqs_allocated; /* number of CQs allocated for device */
759 spinlock_t n_cqs_lock;
760 u32 n_qps_allocated; /* number of QPs allocated for device */
761 spinlock_t n_qps_lock;
762 u32 n_srqs_allocated; /* number of SRQs allocated for device */
763 spinlock_t n_srqs_lock;
764 u32 n_mcast_grps_allocated; /* number of mcast groups allocated */
765 spinlock_t n_mcast_grps_lock;
766};
767
768struct qib_verbs_counters {
769 u64 symbol_error_counter;
770 u64 link_error_recovery_counter;
771 u64 link_downed_counter;
772 u64 port_rcv_errors;
773 u64 port_rcv_remphys_errors;
774 u64 port_xmit_discards;
775 u64 port_xmit_data;
776 u64 port_rcv_data;
777 u64 port_xmit_packets;
778 u64 port_rcv_packets;
779 u32 local_link_integrity_errors;
780 u32 excessive_buffer_overrun_errors;
781 u32 vl15_dropped;
782};
783
784static inline struct qib_mr *to_imr(struct ib_mr *ibmr)
785{
786 return container_of(ibmr, struct qib_mr, ibmr);
787}
788
789static inline struct qib_pd *to_ipd(struct ib_pd *ibpd)
790{
791 return container_of(ibpd, struct qib_pd, ibpd);
792}
793
794static inline struct qib_ah *to_iah(struct ib_ah *ibah)
795{
796 return container_of(ibah, struct qib_ah, ibah);
797}
798
799static inline struct qib_cq *to_icq(struct ib_cq *ibcq)
800{
801 return container_of(ibcq, struct qib_cq, ibcq);
802}
803
804static inline struct qib_srq *to_isrq(struct ib_srq *ibsrq)
805{
806 return container_of(ibsrq, struct qib_srq, ibsrq);
807}
808
809static inline struct qib_qp *to_iqp(struct ib_qp *ibqp)
810{
811 return container_of(ibqp, struct qib_qp, ibqp);
812}
813
814static inline struct qib_ibdev *to_idev(struct ib_device *ibdev)
815{
816 return container_of(ibdev, struct qib_ibdev, ibdev);
817}
818
819/*
820 * Send if not busy or waiting for I/O and either
821 * a RC response is pending or we can process send work requests.
822 */
823static inline int qib_send_ok(struct qib_qp *qp)
824{
825 return !(qp->s_flags & (QIB_S_BUSY | QIB_S_ANY_WAIT_IO)) &&
826 (qp->s_hdrwords || (qp->s_flags & QIB_S_RESP_PENDING) ||
827 !(qp->s_flags & QIB_S_ANY_WAIT_SEND));
828}
829
830extern struct workqueue_struct *qib_cq_wq;
831
832/*
833 * This must be called with s_lock held.
834 */
835static inline void qib_schedule_send(struct qib_qp *qp)
836{
837 if (qib_send_ok(qp))
838 queue_work(ib_wq, &qp->s_work);
839}
840
841static inline int qib_pkey_ok(u16 pkey1, u16 pkey2)
842{
843 u16 p1 = pkey1 & 0x7FFF;
844 u16 p2 = pkey2 & 0x7FFF;
845
846 /*
847 * Low 15 bits must be non-zero and match, and
848 * one of the two must be a full member.
849 */
850 return p1 && p1 == p2 && ((__s16)pkey1 < 0 || (__s16)pkey2 < 0);
851}
852
853void qib_bad_pqkey(struct qib_ibport *ibp, __be16 trap_num, u32 key, u32 sl,
854 u32 qp1, u32 qp2, __be16 lid1, __be16 lid2);
855void qib_cap_mask_chg(struct qib_ibport *ibp);
856void qib_sys_guid_chg(struct qib_ibport *ibp);
857void qib_node_desc_chg(struct qib_ibport *ibp);
858int qib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
859 struct ib_wc *in_wc, struct ib_grh *in_grh,
860 struct ib_mad *in_mad, struct ib_mad *out_mad);
861int qib_create_agents(struct qib_ibdev *dev);
862void qib_free_agents(struct qib_ibdev *dev);
863
864/*
865 * Compare the lower 24 bits of the two values.
866 * Returns an integer <, ==, or > than zero.
867 */
868static inline int qib_cmp24(u32 a, u32 b)
869{
870 return (((int) a) - ((int) b)) << 8;
871}
872
873struct qib_mcast *qib_mcast_find(struct qib_ibport *ibp, union ib_gid *mgid);
874
875int qib_snapshot_counters(struct qib_pportdata *ppd, u64 *swords,
876 u64 *rwords, u64 *spkts, u64 *rpkts,
877 u64 *xmit_wait);
878
879int qib_get_counters(struct qib_pportdata *ppd,
880 struct qib_verbs_counters *cntrs);
881
882int qib_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
883
884int qib_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
885
886int qib_mcast_tree_empty(struct qib_ibport *ibp);
887
888__be32 qib_compute_aeth(struct qib_qp *qp);
889
890struct qib_qp *qib_lookup_qpn(struct qib_ibport *ibp, u32 qpn);
891
892struct ib_qp *qib_create_qp(struct ib_pd *ibpd,
893 struct ib_qp_init_attr *init_attr,
894 struct ib_udata *udata);
895
896int qib_destroy_qp(struct ib_qp *ibqp);
897
898int qib_error_qp(struct qib_qp *qp, enum ib_wc_status err);
899
900int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
901 int attr_mask, struct ib_udata *udata);
902
903int qib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
904 int attr_mask, struct ib_qp_init_attr *init_attr);
905
906unsigned qib_free_all_qps(struct qib_devdata *dd);
907
908void qib_init_qpn_table(struct qib_devdata *dd, struct qib_qpn_table *qpt);
909
910void qib_free_qpn_table(struct qib_qpn_table *qpt);
911
912void qib_get_credit(struct qib_qp *qp, u32 aeth);
913
914unsigned qib_pkt_delay(u32 plen, u8 snd_mult, u8 rcv_mult);
915
916void qib_verbs_sdma_desc_avail(struct qib_pportdata *ppd, unsigned avail);
917
918void qib_put_txreq(struct qib_verbs_txreq *tx);
919
920int qib_verbs_send(struct qib_qp *qp, struct qib_ib_header *hdr,
921 u32 hdrwords, struct qib_sge_state *ss, u32 len);
922
923void qib_copy_sge(struct qib_sge_state *ss, void *data, u32 length,
924 int release);
925
926void qib_skip_sge(struct qib_sge_state *ss, u32 length, int release);
927
928void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
929 int has_grh, void *data, u32 tlen, struct qib_qp *qp);
930
931void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
932 int has_grh, void *data, u32 tlen, struct qib_qp *qp);
933
934int qib_check_ah(struct ib_device *ibdev, struct ib_ah_attr *ah_attr);
935
936void qib_rc_rnr_retry(unsigned long arg);
937
938void qib_rc_send_complete(struct qib_qp *qp, struct qib_ib_header *hdr);
939
940void qib_rc_error(struct qib_qp *qp, enum ib_wc_status err);
941
942int qib_post_ud_send(struct qib_qp *qp, struct ib_send_wr *wr);
943
944void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
945 int has_grh, void *data, u32 tlen, struct qib_qp *qp);
946
947int qib_alloc_lkey(struct qib_lkey_table *rkt, struct qib_mregion *mr);
948
949int qib_free_lkey(struct qib_ibdev *dev, struct qib_mregion *mr);
950
951int qib_lkey_ok(struct qib_lkey_table *rkt, struct qib_pd *pd,
952 struct qib_sge *isge, struct ib_sge *sge, int acc);
953
954int qib_rkey_ok(struct qib_qp *qp, struct qib_sge *sge,
955 u32 len, u64 vaddr, u32 rkey, int acc);
956
957int qib_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
958 struct ib_recv_wr **bad_wr);
959
960struct ib_srq *qib_create_srq(struct ib_pd *ibpd,
961 struct ib_srq_init_attr *srq_init_attr,
962 struct ib_udata *udata);
963
964int qib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
965 enum ib_srq_attr_mask attr_mask,
966 struct ib_udata *udata);
967
968int qib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr);
969
970int qib_destroy_srq(struct ib_srq *ibsrq);
971
972void qib_cq_enter(struct qib_cq *cq, struct ib_wc *entry, int sig);
973
974int qib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry);
975
976struct ib_cq *qib_create_cq(struct ib_device *ibdev, int entries,
977 int comp_vector, struct ib_ucontext *context,
978 struct ib_udata *udata);
979
980int qib_destroy_cq(struct ib_cq *ibcq);
981
982int qib_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags);
983
984int qib_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata);
985
986struct ib_mr *qib_get_dma_mr(struct ib_pd *pd, int acc);
987
988struct ib_mr *qib_reg_phys_mr(struct ib_pd *pd,
989 struct ib_phys_buf *buffer_list,
990 int num_phys_buf, int acc, u64 *iova_start);
991
992struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
993 u64 virt_addr, int mr_access_flags,
994 struct ib_udata *udata);
995
996int qib_dereg_mr(struct ib_mr *ibmr);
997
998struct ib_mr *qib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
999
1000struct ib_fast_reg_page_list *qib_alloc_fast_reg_page_list(
1001 struct ib_device *ibdev, int page_list_len);
1002
1003void qib_free_fast_reg_page_list(struct ib_fast_reg_page_list *pl);
1004
1005int qib_fast_reg_mr(struct qib_qp *qp, struct ib_send_wr *wr);
1006
1007struct ib_fmr *qib_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
1008 struct ib_fmr_attr *fmr_attr);
1009
1010int qib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
1011 int list_len, u64 iova);
1012
1013int qib_unmap_fmr(struct list_head *fmr_list);
1014
1015int qib_dealloc_fmr(struct ib_fmr *ibfmr);
1016
1017void qib_release_mmap_info(struct kref *ref);
1018
1019struct qib_mmap_info *qib_create_mmap_info(struct qib_ibdev *dev, u32 size,
1020 struct ib_ucontext *context,
1021 void *obj);
1022
1023void qib_update_mmap_info(struct qib_ibdev *dev, struct qib_mmap_info *ip,
1024 u32 size, void *obj);
1025
1026int qib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma);
1027
1028int qib_get_rwqe(struct qib_qp *qp, int wr_id_only);
1029
1030void qib_migrate_qp(struct qib_qp *qp);
1031
1032int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr,
1033 int has_grh, struct qib_qp *qp, u32 bth0);
1034
1035u32 qib_make_grh(struct qib_ibport *ibp, struct ib_grh *hdr,
1036 struct ib_global_route *grh, u32 hwords, u32 nwords);
1037
1038void qib_make_ruc_header(struct qib_qp *qp, struct qib_other_headers *ohdr,
1039 u32 bth0, u32 bth2);
1040
1041void qib_do_send(struct work_struct *work);
1042
1043void qib_send_complete(struct qib_qp *qp, struct qib_swqe *wqe,
1044 enum ib_wc_status status);
1045
1046void qib_send_rc_ack(struct qib_qp *qp);
1047
1048int qib_make_rc_req(struct qib_qp *qp);
1049
1050int qib_make_uc_req(struct qib_qp *qp);
1051
1052int qib_make_ud_req(struct qib_qp *qp);
1053
1054int qib_register_ib_device(struct qib_devdata *);
1055
1056void qib_unregister_ib_device(struct qib_devdata *);
1057
1058void qib_ib_rcv(struct qib_ctxtdata *, void *, void *, u32);
1059
1060void qib_ib_piobufavail(struct qib_devdata *);
1061
1062unsigned qib_get_npkeys(struct qib_devdata *);
1063
1064unsigned qib_get_pkey(struct qib_ibport *, unsigned);
1065
1066extern const enum ib_wc_opcode ib_qib_wc_opcode[];
1067
1068/*
1069 * Below HCA-independent IB PhysPortState values, returned
1070 * by the f_ibphys_portstate() routine.
1071 */
1072#define IB_PHYSPORTSTATE_SLEEP 1
1073#define IB_PHYSPORTSTATE_POLL 2
1074#define IB_PHYSPORTSTATE_DISABLED 3
1075#define IB_PHYSPORTSTATE_CFG_TRAIN 4
1076#define IB_PHYSPORTSTATE_LINKUP 5
1077#define IB_PHYSPORTSTATE_LINK_ERR_RECOVER 6
1078#define IB_PHYSPORTSTATE_CFG_DEBOUNCE 8
1079#define IB_PHYSPORTSTATE_CFG_IDLE 0xB
1080#define IB_PHYSPORTSTATE_RECOVERY_RETRAIN 0xC
1081#define IB_PHYSPORTSTATE_RECOVERY_WAITRMT 0xE
1082#define IB_PHYSPORTSTATE_RECOVERY_IDLE 0xF
1083#define IB_PHYSPORTSTATE_CFG_ENH 0x10
1084#define IB_PHYSPORTSTATE_CFG_WAIT_ENH 0x13
1085
1086extern const int ib_qib_state_ops[];
1087
1088extern __be64 ib_qib_sys_image_guid; /* in network order */
1089
1090extern unsigned int ib_qib_lkey_table_size;
1091
1092extern unsigned int ib_qib_max_cqes;
1093
1094extern unsigned int ib_qib_max_cqs;
1095
1096extern unsigned int ib_qib_max_qp_wrs;
1097
1098extern unsigned int ib_qib_max_qps;
1099
1100extern unsigned int ib_qib_max_sges;
1101
1102extern unsigned int ib_qib_max_mcast_grps;
1103
1104extern unsigned int ib_qib_max_mcast_qp_attached;
1105
1106extern unsigned int ib_qib_max_srqs;
1107
1108extern unsigned int ib_qib_max_srq_sges;
1109
1110extern unsigned int ib_qib_max_srq_wrs;
1111
1112extern const u32 ib_qib_rnr_table[];
1113
1114extern struct ib_dma_mapping_ops qib_dma_mapping_ops;
1115
1116#endif /* QIB_VERBS_H */