2 * Copyright (c) 2012, 2013 Intel Corporation. All rights reserved.
3 * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
4 * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
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:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
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.
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
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 <linux/kthread.h>
45 #include <linux/completion.h>
46 #include <rdma/ib_pack.h>
47 #include <rdma/ib_user_verbs.h>
48 #include <rdma/rdma_vt.h>
53 struct qib_verbs_txreq;
55 #define QIB_MAX_RDMA_ATOMIC 16
56 #define QIB_GUIDS_PER_PORT 5
59 * Increment this value if any changes that break userspace ABI
60 * compatibility are made.
62 #define QIB_UVERBS_ABI_VERSION 2
65 * Define an ib_cq_notify value that is not valid so we know when CQ
66 * notifications are armed.
68 #define IB_CQ_NONE (IB_CQ_NEXT_COMP + 1)
70 #define IB_SEQ_NAK (3 << 29)
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
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)
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
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)
103 #define QIB_VENDOR_IPG cpu_to_be16(0xFFA0)
105 #define IB_BTH_REQ_ACK (1 << 31)
106 #define IB_BTH_SOLICITED (1 << 23)
107 #define IB_BTH_MIG_REQ (1 << 22)
109 /* XXX Should be defined in ib_verbs.h enum ib_port_cap_flags */
110 #define IB_PORT_OTHER_LOCAL_CHANGES_SUP (1 << 26)
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
121 #define IB_DEFAULT_GID_PREFIX cpu_to_be64(0xfe80000000000000ULL)
123 /* Values for set/get portinfo VLCap OperationalVLs */
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
130 static inline int qib_num_vls(int vls)
153 struct ib_atomic_eth {
154 __be32 vaddr[2]; /* unaligned so access as 2 32-bit words */
160 struct qib_other_headers {
173 __be32 atomic_ack_eth[2];
177 struct ib_atomic_eth atomic_eth;
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.
187 struct qib_ib_header {
192 struct qib_other_headers oth;
194 struct qib_other_headers oth;
198 struct qib_pio_header {
200 struct qib_ib_header hdr;
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.
208 struct qib_mcast_qp {
209 struct list_head list;
214 struct rb_node rb_node;
216 struct list_head qp_list;
217 wait_queue_head_t wait;
223 * This structure is used to contain the head pointer, tail pointer,
224 * and completion queue entries as a single memory allocation so
225 * it can be mmap'ed into user space.
228 u32 head; /* index of next entry to fill */
229 u32 tail; /* index of next ib_poll_cq() entry */
231 /* these are actually size ibcq.cqe + 1 */
232 struct ib_uverbs_wc uqueue[0];
233 struct ib_wc kqueue[0];
238 * The completion queue structure.
242 struct kthread_work comptask;
243 struct qib_devdata *dd;
244 spinlock_t lock; /* protect changes in this struct */
247 struct qib_cq_wc *queue;
248 struct rvt_mmap_info *ip;
252 * qib specific data structure that will be hidden from rvt after the queue pair
256 struct qib_ib_header *s_hdr; /* next packet header to send */
257 struct list_head iowait; /* link for wait PIO buf */
259 struct qib_verbs_txreq *s_tx;
260 struct work_struct s_work;
261 wait_queue_head_t wait_dma;
262 struct rvt_qp *owner;
265 #define QIB_PSN_CREDIT 16
268 * Since struct rvt_swqe is not a fixed size, we can't simply index into
269 * struct rvt_qp.s_wq. This function does the array index computation.
271 static inline struct rvt_swqe *get_swqe_ptr(struct rvt_qp *qp,
274 return (struct rvt_swqe *)((char *)qp->s_wq +
275 (sizeof(struct rvt_swqe) +
277 sizeof(struct rvt_sge)) * n);
281 * Since struct rvt_rwqe is not a fixed size, we can't simply index into
282 * struct rvt_rwq.wq. This function does the array index computation.
284 static inline struct rvt_rwqe *get_rwqe_ptr(struct rvt_rq *rq, unsigned n)
286 return (struct rvt_rwqe *)
287 ((char *) rq->wq->wq +
288 (sizeof(struct rvt_rwqe) +
289 rq->max_sge * sizeof(struct ib_sge)) * n);
292 struct qib_opcode_stats {
293 u64 n_packets; /* number of packets */
294 u64 n_bytes; /* total number of bytes */
297 struct qib_opcode_stats_perctx {
298 struct qib_opcode_stats stats[128];
301 struct qib_pma_counters {
302 u64 n_unicast_xmit; /* total unicast packets sent */
303 u64 n_unicast_rcv; /* total unicast packets received */
304 u64 n_multicast_xmit; /* total multicast packets sent */
305 u64 n_multicast_rcv; /* total multicast packets received */
309 struct rvt_ibport rvp;
310 struct rvt_ah *sm_ah;
311 struct rvt_ah *smi_ah;
312 __be64 guids[QIB_GUIDS_PER_PORT - 1]; /* writable GUIDs */
313 struct qib_pma_counters __percpu *pmastats;
314 u64 z_unicast_xmit; /* starting count for PMA */
315 u64 z_unicast_rcv; /* starting count for PMA */
316 u64 z_multicast_xmit; /* starting count for PMA */
317 u64 z_multicast_rcv; /* starting count for PMA */
318 u64 z_symbol_error_counter; /* starting count for PMA */
319 u64 z_link_error_recovery_counter; /* starting count for PMA */
320 u64 z_link_downed_counter; /* starting count for PMA */
321 u64 z_port_rcv_errors; /* starting count for PMA */
322 u64 z_port_rcv_remphys_errors; /* starting count for PMA */
323 u64 z_port_xmit_discards; /* starting count for PMA */
324 u64 z_port_xmit_data; /* starting count for PMA */
325 u64 z_port_rcv_data; /* starting count for PMA */
326 u64 z_port_xmit_packets; /* starting count for PMA */
327 u64 z_port_rcv_packets; /* starting count for PMA */
328 u32 z_local_link_integrity_errors; /* starting count for PMA */
329 u32 z_excessive_buffer_overrun_errors; /* starting count for PMA */
330 u32 z_vl15_dropped; /* starting count for PMA */
335 struct rvt_dev_info rdi;
337 struct list_head piowait; /* list for wait PIO buf */
338 struct list_head dmawait; /* list for wait DMA */
339 struct list_head txwait; /* list for wait qib_verbs_txreq */
340 struct list_head memwait; /* list for wait kernel memory */
341 struct list_head txreq_free;
342 struct timer_list mem_timer;
343 struct qib_pio_header *pio_hdrs;
344 dma_addr_t pio_hdrs_phys;
345 u32 qp_rnd; /* random bytes for hash */
350 u32 n_cqs_allocated; /* number of CQs allocated for device */
351 spinlock_t n_cqs_lock;
352 u32 n_qps_allocated; /* number of QPs allocated for device */
353 spinlock_t n_qps_lock;
354 u32 n_srqs_allocated; /* number of SRQs allocated for device */
355 spinlock_t n_srqs_lock;
356 u32 n_mcast_grps_allocated; /* number of mcast groups allocated */
357 spinlock_t n_mcast_grps_lock;
358 #ifdef CONFIG_DEBUG_FS
359 /* per HCA debugfs */
360 struct dentry *qib_ibdev_dbg;
364 struct qib_verbs_counters {
365 u64 symbol_error_counter;
366 u64 link_error_recovery_counter;
367 u64 link_downed_counter;
369 u64 port_rcv_remphys_errors;
370 u64 port_xmit_discards;
373 u64 port_xmit_packets;
374 u64 port_rcv_packets;
375 u32 local_link_integrity_errors;
376 u32 excessive_buffer_overrun_errors;
380 static inline struct qib_cq *to_icq(struct ib_cq *ibcq)
382 return container_of(ibcq, struct qib_cq, ibcq);
385 static inline struct rvt_qp *to_iqp(struct ib_qp *ibqp)
387 return container_of(ibqp, struct rvt_qp, ibqp);
390 static inline struct qib_ibdev *to_idev(struct ib_device *ibdev)
392 struct rvt_dev_info *rdi;
394 rdi = container_of(ibdev, struct rvt_dev_info, ibdev);
395 return container_of(rdi, struct qib_ibdev, rdi);
399 * Send if not busy or waiting for I/O and either
400 * a RC response is pending or we can process send work requests.
402 static inline int qib_send_ok(struct rvt_qp *qp)
404 return !(qp->s_flags & (RVT_S_BUSY | RVT_S_ANY_WAIT_IO)) &&
405 (qp->s_hdrwords || (qp->s_flags & RVT_S_RESP_PENDING) ||
406 !(qp->s_flags & RVT_S_ANY_WAIT_SEND));
410 * This must be called with s_lock held.
412 void qib_schedule_send(struct rvt_qp *qp);
414 static inline int qib_pkey_ok(u16 pkey1, u16 pkey2)
416 u16 p1 = pkey1 & 0x7FFF;
417 u16 p2 = pkey2 & 0x7FFF;
420 * Low 15 bits must be non-zero and match, and
421 * one of the two must be a full member.
423 return p1 && p1 == p2 && ((__s16)pkey1 < 0 || (__s16)pkey2 < 0);
426 void qib_bad_pqkey(struct qib_ibport *ibp, __be16 trap_num, u32 key, u32 sl,
427 u32 qp1, u32 qp2, __be16 lid1, __be16 lid2);
428 void qib_cap_mask_chg(struct qib_ibport *ibp);
429 void qib_sys_guid_chg(struct qib_ibport *ibp);
430 void qib_node_desc_chg(struct qib_ibport *ibp);
431 int qib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
432 const struct ib_wc *in_wc, const struct ib_grh *in_grh,
433 const struct ib_mad_hdr *in, size_t in_mad_size,
434 struct ib_mad_hdr *out, size_t *out_mad_size,
435 u16 *out_mad_pkey_index);
436 int qib_create_agents(struct qib_ibdev *dev);
437 void qib_free_agents(struct qib_ibdev *dev);
440 * Compare the lower 24 bits of the two values.
441 * Returns an integer <, ==, or > than zero.
443 static inline int qib_cmp24(u32 a, u32 b)
445 return (((int) a) - ((int) b)) << 8;
448 struct qib_mcast *qib_mcast_find(struct qib_ibport *ibp, union ib_gid *mgid);
450 int qib_snapshot_counters(struct qib_pportdata *ppd, u64 *swords,
451 u64 *rwords, u64 *spkts, u64 *rpkts,
454 int qib_get_counters(struct qib_pportdata *ppd,
455 struct qib_verbs_counters *cntrs);
457 int qib_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
459 int qib_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
461 int qib_mcast_tree_empty(struct qib_ibport *ibp);
463 __be32 qib_compute_aeth(struct rvt_qp *qp);
465 struct rvt_qp *qib_lookup_qpn(struct qib_ibport *ibp, u32 qpn);
467 int qib_destroy_qp(struct ib_qp *ibqp);
469 int qib_error_qp(struct rvt_qp *qp, enum ib_wc_status err);
471 int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
472 int attr_mask, struct ib_udata *udata);
474 int qib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
475 int attr_mask, struct ib_qp_init_attr *init_attr);
477 * Functions provided by qib driver for rdmavt to use
479 unsigned qib_free_all_qps(struct rvt_dev_info *rdi);
480 void *qp_priv_alloc(struct rvt_dev_info *rdi, struct rvt_qp *qp, gfp_t gfp);
481 void qp_priv_free(struct rvt_dev_info *rdi, struct rvt_qp *qp);
482 void notify_qp_reset(struct rvt_qp *qp);
483 int alloc_qpn(struct rvt_dev_info *rdi, struct rvt_qpn_table *qpt,
484 enum ib_qp_type type, u8 port, gfp_t gfp);
486 #ifdef CONFIG_DEBUG_FS
490 struct qib_qp_iter *qib_qp_iter_init(struct qib_ibdev *dev);
492 int qib_qp_iter_next(struct qib_qp_iter *iter);
494 void qib_qp_iter_print(struct seq_file *s, struct qib_qp_iter *iter);
498 void qib_get_credit(struct rvt_qp *qp, u32 aeth);
500 unsigned qib_pkt_delay(u32 plen, u8 snd_mult, u8 rcv_mult);
502 void qib_verbs_sdma_desc_avail(struct qib_pportdata *ppd, unsigned avail);
504 void qib_put_txreq(struct qib_verbs_txreq *tx);
506 int qib_verbs_send(struct rvt_qp *qp, struct qib_ib_header *hdr,
507 u32 hdrwords, struct rvt_sge_state *ss, u32 len);
509 void qib_copy_sge(struct rvt_sge_state *ss, void *data, u32 length,
512 void qib_skip_sge(struct rvt_sge_state *ss, u32 length, int release);
514 void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
515 int has_grh, void *data, u32 tlen, struct rvt_qp *qp);
517 void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
518 int has_grh, void *data, u32 tlen, struct rvt_qp *qp);
520 int qib_check_ah(struct ib_device *ibdev, struct ib_ah_attr *ah_attr);
522 struct ib_ah *qib_create_qp0_ah(struct qib_ibport *ibp, u16 dlid);
524 void qib_rc_rnr_retry(unsigned long arg);
526 void qib_rc_send_complete(struct rvt_qp *qp, struct qib_ib_header *hdr);
528 void qib_rc_error(struct rvt_qp *qp, enum ib_wc_status err);
530 int qib_post_ud_send(struct rvt_qp *qp, struct ib_send_wr *wr);
532 void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
533 int has_grh, void *data, u32 tlen, struct rvt_qp *qp);
535 int qib_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
536 struct ib_recv_wr **bad_wr);
538 struct ib_srq *qib_create_srq(struct ib_pd *ibpd,
539 struct ib_srq_init_attr *srq_init_attr,
540 struct ib_udata *udata);
542 int qib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
543 enum ib_srq_attr_mask attr_mask,
544 struct ib_udata *udata);
546 int qib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr);
548 int qib_destroy_srq(struct ib_srq *ibsrq);
550 int qib_cq_init(struct qib_devdata *dd);
552 void qib_cq_exit(struct qib_devdata *dd);
554 void qib_cq_enter(struct qib_cq *cq, struct ib_wc *entry, int sig);
556 int qib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry);
558 struct ib_cq *qib_create_cq(struct ib_device *ibdev,
559 const struct ib_cq_init_attr *attr,
560 struct ib_ucontext *context,
561 struct ib_udata *udata);
563 int qib_destroy_cq(struct ib_cq *ibcq);
565 int qib_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags);
567 int qib_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata);
569 void mr_rcu_callback(struct rcu_head *list);
571 static inline void qib_put_ss(struct rvt_sge_state *ss)
573 while (ss->num_sge) {
574 rvt_put_mr(ss->sge.mr);
576 ss->sge = *ss->sg_list++;
580 int qib_get_rwqe(struct rvt_qp *qp, int wr_id_only);
582 void qib_migrate_qp(struct rvt_qp *qp);
584 int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr,
585 int has_grh, struct rvt_qp *qp, u32 bth0);
587 u32 qib_make_grh(struct qib_ibport *ibp, struct ib_grh *hdr,
588 struct ib_global_route *grh, u32 hwords, u32 nwords);
590 void qib_make_ruc_header(struct rvt_qp *qp, struct qib_other_headers *ohdr,
593 void qib_do_send(struct work_struct *work);
595 void qib_send_complete(struct rvt_qp *qp, struct rvt_swqe *wqe,
596 enum ib_wc_status status);
598 void qib_send_rc_ack(struct rvt_qp *qp);
600 int qib_make_rc_req(struct rvt_qp *qp);
602 int qib_make_uc_req(struct rvt_qp *qp);
604 int qib_make_ud_req(struct rvt_qp *qp);
606 int qib_register_ib_device(struct qib_devdata *);
608 void qib_unregister_ib_device(struct qib_devdata *);
610 void qib_ib_rcv(struct qib_ctxtdata *, void *, void *, u32);
612 void qib_ib_piobufavail(struct qib_devdata *);
614 unsigned qib_get_npkeys(struct qib_devdata *);
616 unsigned qib_get_pkey(struct qib_ibport *, unsigned);
618 extern const enum ib_wc_opcode ib_qib_wc_opcode[];
621 * Below HCA-independent IB PhysPortState values, returned
622 * by the f_ibphys_portstate() routine.
624 #define IB_PHYSPORTSTATE_SLEEP 1
625 #define IB_PHYSPORTSTATE_POLL 2
626 #define IB_PHYSPORTSTATE_DISABLED 3
627 #define IB_PHYSPORTSTATE_CFG_TRAIN 4
628 #define IB_PHYSPORTSTATE_LINKUP 5
629 #define IB_PHYSPORTSTATE_LINK_ERR_RECOVER 6
630 #define IB_PHYSPORTSTATE_CFG_DEBOUNCE 8
631 #define IB_PHYSPORTSTATE_CFG_IDLE 0xB
632 #define IB_PHYSPORTSTATE_RECOVERY_RETRAIN 0xC
633 #define IB_PHYSPORTSTATE_RECOVERY_WAITRMT 0xE
634 #define IB_PHYSPORTSTATE_RECOVERY_IDLE 0xF
635 #define IB_PHYSPORTSTATE_CFG_ENH 0x10
636 #define IB_PHYSPORTSTATE_CFG_WAIT_ENH 0x13
638 extern const int ib_qib_state_ops[];
640 extern __be64 ib_qib_sys_image_guid; /* in network order */
642 extern unsigned int ib_rvt_lkey_table_size;
644 extern unsigned int ib_qib_max_cqes;
646 extern unsigned int ib_qib_max_cqs;
648 extern unsigned int ib_qib_max_qp_wrs;
650 extern unsigned int ib_qib_max_qps;
652 extern unsigned int ib_qib_max_sges;
654 extern unsigned int ib_qib_max_mcast_grps;
656 extern unsigned int ib_qib_max_mcast_qp_attached;
658 extern unsigned int ib_qib_max_srqs;
660 extern unsigned int ib_qib_max_srq_sges;
662 extern unsigned int ib_qib_max_srq_wrs;
664 extern const u32 ib_qib_rnr_table[];
666 #endif /* QIB_VERBS_H */