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
58 #define QPN_MAX (1 << 24)
59 #define QPNMAP_ENTRIES (QPN_MAX / PAGE_SIZE / BITS_PER_BYTE)
62 * Increment this value if any changes that break userspace ABI
63 * compatibility are made.
65 #define QIB_UVERBS_ABI_VERSION 2
68 * Define an ib_cq_notify value that is not valid so we know when CQ
69 * notifications are armed.
71 #define IB_CQ_NONE (IB_CQ_NEXT_COMP + 1)
73 #define IB_SEQ_NAK (3 << 29)
75 /* AETH NAK opcode values */
76 #define IB_RNR_NAK 0x20
77 #define IB_NAK_PSN_ERROR 0x60
78 #define IB_NAK_INVALID_REQUEST 0x61
79 #define IB_NAK_REMOTE_ACCESS_ERROR 0x62
80 #define IB_NAK_REMOTE_OPERATIONAL_ERROR 0x63
81 #define IB_NAK_INVALID_RD_REQUEST 0x64
83 /* Flags for checking QP state (see ib_qib_state_ops[]) */
84 #define QIB_POST_SEND_OK 0x01
85 #define QIB_POST_RECV_OK 0x02
86 #define QIB_PROCESS_RECV_OK 0x04
87 #define QIB_PROCESS_SEND_OK 0x08
88 #define QIB_PROCESS_NEXT_SEND_OK 0x10
89 #define QIB_FLUSH_SEND 0x20
90 #define QIB_FLUSH_RECV 0x40
91 #define QIB_PROCESS_OR_FLUSH_SEND \
92 (QIB_PROCESS_SEND_OK | QIB_FLUSH_SEND)
94 /* IB Performance Manager status values */
95 #define IB_PMA_SAMPLE_STATUS_DONE 0x00
96 #define IB_PMA_SAMPLE_STATUS_STARTED 0x01
97 #define IB_PMA_SAMPLE_STATUS_RUNNING 0x02
99 /* Mandatory IB performance counter select values. */
100 #define IB_PMA_PORT_XMIT_DATA cpu_to_be16(0x0001)
101 #define IB_PMA_PORT_RCV_DATA cpu_to_be16(0x0002)
102 #define IB_PMA_PORT_XMIT_PKTS cpu_to_be16(0x0003)
103 #define IB_PMA_PORT_RCV_PKTS cpu_to_be16(0x0004)
104 #define IB_PMA_PORT_XMIT_WAIT cpu_to_be16(0x0005)
106 #define QIB_VENDOR_IPG cpu_to_be16(0xFFA0)
108 #define IB_BTH_REQ_ACK (1 << 31)
109 #define IB_BTH_SOLICITED (1 << 23)
110 #define IB_BTH_MIG_REQ (1 << 22)
112 /* XXX Should be defined in ib_verbs.h enum ib_port_cap_flags */
113 #define IB_PORT_OTHER_LOCAL_CHANGES_SUP (1 << 26)
115 #define IB_GRH_VERSION 6
116 #define IB_GRH_VERSION_MASK 0xF
117 #define IB_GRH_VERSION_SHIFT 28
118 #define IB_GRH_TCLASS_MASK 0xFF
119 #define IB_GRH_TCLASS_SHIFT 20
120 #define IB_GRH_FLOW_MASK 0xFFFFF
121 #define IB_GRH_FLOW_SHIFT 0
122 #define IB_GRH_NEXT_HDR 0x1B
124 #define IB_DEFAULT_GID_PREFIX cpu_to_be64(0xfe80000000000000ULL)
126 /* Values for set/get portinfo VLCap OperationalVLs */
128 #define IB_VL_VL0_1 2
129 #define IB_VL_VL0_3 3
130 #define IB_VL_VL0_7 4
131 #define IB_VL_VL0_14 5
133 static inline int qib_num_vls(int vls)
156 struct ib_atomic_eth {
157 __be32 vaddr[2]; /* unaligned so access as 2 32-bit words */
163 struct qib_other_headers {
176 __be32 atomic_ack_eth[2];
180 struct ib_atomic_eth atomic_eth;
185 * Note that UD packets with a GRH header are 8+40+12+8 = 68 bytes
186 * long (72 w/ imm_data). Only the first 56 bytes of the IB header
187 * will be in the eager header buffer. The remaining 12 or 16 bytes
188 * are in the data buffer.
190 struct qib_ib_header {
195 struct qib_other_headers oth;
197 struct qib_other_headers oth;
201 struct qib_pio_header {
203 struct qib_ib_header hdr;
207 * There is one struct qib_mcast for each multicast GID.
208 * All attached QPs are then stored as a list of
209 * struct qib_mcast_qp.
211 struct qib_mcast_qp {
212 struct list_head list;
217 struct rb_node rb_node;
219 struct list_head qp_list;
220 wait_queue_head_t wait;
228 struct ib_ah_attr attr;
233 * This structure is used by qib_mmap() to validate an offset
234 * when an mmap() request is made. The vm_area_struct then uses
235 * this as its vm_private_data.
237 struct qib_mmap_info {
238 struct list_head pending_mmaps;
239 struct ib_ucontext *context;
247 * This structure is used to contain the head pointer, tail pointer,
248 * and completion queue entries as a single memory allocation so
249 * it can be mmap'ed into user space.
252 u32 head; /* index of next entry to fill */
253 u32 tail; /* index of next ib_poll_cq() entry */
255 /* these are actually size ibcq.cqe + 1 */
256 struct ib_uverbs_wc uqueue[0];
257 struct ib_wc kqueue[0];
262 * The completion queue structure.
266 struct kthread_work comptask;
267 struct qib_devdata *dd;
268 spinlock_t lock; /* protect changes in this struct */
271 struct qib_cq_wc *queue;
272 struct qib_mmap_info *ip;
276 * A segment is a linear region of low physical memory.
277 * XXX Maybe we should use phys addr here and kmap()/kunmap().
278 * Used by the verbs layer.
285 /* The number of qib_segs that fit in a page. */
286 #define QIB_SEGSZ (PAGE_SIZE / sizeof(struct qib_seg))
288 struct qib_segarray {
289 struct qib_seg segs[QIB_SEGSZ];
293 struct ib_pd *pd; /* shares refcnt of ibmr.pd */
294 u64 user_base; /* User's address for this region */
295 u64 iova; /* IB start address of this region */
298 u32 offset; /* offset (bytes) to start of region */
300 u32 max_segs; /* number of qib_segs in all the arrays */
301 u32 mapsz; /* size of the map array */
302 u8 page_shift; /* 0 - non unform/non powerof2 sizes */
303 u8 lkey_published; /* in global table */
304 struct completion comp; /* complete when refcount goes to zero */
305 struct rcu_head list;
307 struct qib_segarray *map[0]; /* the segments */
311 * These keep track of the copy progress within a memory region.
312 * Used by the verbs layer.
315 struct qib_mregion *mr;
316 void *vaddr; /* kernel virtual address of segment */
317 u32 sge_length; /* length of the SGE */
318 u32 length; /* remaining length of the segment */
319 u16 m; /* current index: mr->map[m] */
320 u16 n; /* current index: mr->map[m]->segs[n] */
326 struct ib_umem *umem;
329 struct qib_mregion mr; /* must be last */
333 * Send work request queue entry.
334 * The size of the sg_list is determined when the QP is created and stored
339 struct ib_send_wr wr; /* don't use wr.sg_list */
340 struct ib_ud_wr ud_wr;
341 struct ib_reg_wr reg_wr;
342 struct ib_rdma_wr rdma_wr;
343 struct ib_atomic_wr atomic_wr;
345 u32 psn; /* first packet sequence number */
346 u32 lpsn; /* last packet sequence number */
347 u32 ssn; /* send sequence number */
348 u32 length; /* total length of data in sg_list */
349 struct qib_sge sg_list[0];
353 * Receive work request queue entry.
354 * The size of the sg_list is determined when the QP (or SRQ) is created
355 * and stored in qp->r_rq.max_sge (or srq->rq.max_sge).
360 struct ib_sge sg_list[0];
364 * This structure is used to contain the head pointer, tail pointer,
365 * and receive work queue entries as a single memory allocation so
366 * it can be mmap'ed into user space.
367 * Note that the wq array elements are variable size so you can't
368 * just index into the array to get the N'th element;
369 * use get_rwqe_ptr() instead.
372 u32 head; /* new work requests posted to the head */
373 u32 tail; /* receives pull requests from here. */
374 struct qib_rwqe wq[0];
379 u32 size; /* size of RWQE array */
381 spinlock_t lock /* protect changes in this struct */
382 ____cacheline_aligned_in_smp;
388 struct qib_mmap_info *ip;
389 /* send signal when number of RWQEs < limit */
393 struct qib_sge_state {
394 struct qib_sge *sg_list; /* next SGE to be used if any */
395 struct qib_sge sge; /* progress state for the current SGE */
401 * This structure holds the information that the send tasklet needs
402 * to send a RDMA read response or atomic operation.
404 struct qib_ack_entry {
410 struct qib_sge rdma_sge;
416 * qib specific data structure that will be hidden from rvt after the queue pair
421 struct qib_ib_header *s_hdr; /* next packet header to send */
422 struct list_head iowait; /* link for wait PIO buf */
424 struct qib_verbs_txreq *s_tx;
425 struct work_struct s_work;
426 wait_queue_head_t wait_dma;
427 struct qib_qp *owner;
431 * Variables prefixed with s_ are for the requester (sender).
432 * Variables prefixed with r_ are for the responder (receiver).
433 * Variables prefixed with ack_ are for responder replies.
435 * Common variables are protected by both r_rq.lock and s_lock in that order
436 * which only happens in modify_qp() or changing the QP 'state'.
440 struct qib_qp_priv *priv;
441 /* read mostly fields above and below */
442 struct ib_ah_attr remote_ah_attr;
443 struct ib_ah_attr alt_ah_attr;
444 struct qib_qp __rcu *next; /* link list for QPN hash table */
445 struct qib_swqe *s_wq; /* send work queue */
446 struct qib_mmap_info *ip;
447 unsigned long timeout_jiffies; /* computed from timeout */
449 enum ib_mtu path_mtu;
451 u32 pmtu; /* decoded from path_mtu */
452 u32 qkey; /* QKEY for this QP (for UD or RD) */
453 u32 s_size; /* send work queue size */
454 u32 s_rnr_timeout; /* number of milliseconds for RNR timeout */
456 u8 state; /* QP state */
458 u8 alt_timeout; /* Alternate path timeout for this QP */
459 u8 timeout; /* Timeout for this QP */
463 u8 s_pkey_index; /* PKEY index to use */
464 u8 s_alt_pkey_index; /* Alternate path PKEY index to use */
465 u8 r_max_rd_atomic; /* max number of RDMA read/atomic to receive */
466 u8 s_max_rd_atomic; /* max number of RDMA read/atomic to send */
467 u8 s_retry_cnt; /* number of times to retry */
469 u8 r_min_rnr_timer; /* retry timeout value for RNR NAKs */
470 u8 s_max_sge; /* size of s_wq->sg_list */
473 /* start of read/write fields */
475 atomic_t refcount ____cacheline_aligned_in_smp;
476 wait_queue_head_t wait;
479 struct qib_ack_entry s_ack_queue[QIB_MAX_RDMA_ATOMIC + 1]
480 ____cacheline_aligned_in_smp;
481 struct qib_sge_state s_rdma_read_sge;
483 spinlock_t r_lock ____cacheline_aligned_in_smp; /* used for APM */
484 unsigned long r_aflags;
485 u64 r_wr_id; /* ID for current receive WQE */
486 u32 r_ack_psn; /* PSN for next ACK or atomic ACK */
487 u32 r_len; /* total length of r_sge */
488 u32 r_rcv_len; /* receive data len processed */
489 u32 r_psn; /* expected rcv packet sequence number */
490 u32 r_msn; /* message sequence number */
492 u8 r_state; /* opcode of last packet received */
494 u8 r_head_ack_queue; /* index into s_ack_queue[] */
496 struct list_head rspwait; /* link for waititing to respond */
498 struct qib_sge_state r_sge; /* current receive data */
499 struct qib_rq r_rq; /* receive work queue */
501 spinlock_t s_lock ____cacheline_aligned_in_smp;
502 struct qib_sge_state *s_cur_sge;
505 struct qib_swqe *s_wqe;
506 struct qib_sge_state s_sge; /* current send request data */
507 struct qib_mregion *s_rdma_mr;
509 u32 s_cur_size; /* size of send packet in bytes */
510 u32 s_len; /* total length of s_sge */
511 u32 s_rdma_read_len; /* total length of s_rdma_read_sge */
512 u32 s_next_psn; /* PSN for next request */
513 u32 s_last_psn; /* last response PSN processed */
514 u32 s_sending_psn; /* lowest PSN that is being sent */
515 u32 s_sending_hpsn; /* highest PSN that is being sent */
516 u32 s_psn; /* current packet sequence number */
517 u32 s_ack_rdma_psn; /* PSN for sending RDMA read responses */
518 u32 s_ack_psn; /* PSN for acking sends and RDMA writes */
519 u32 s_head; /* new entries added here */
520 u32 s_tail; /* next entry to process */
521 u32 s_cur; /* current work queue entry */
522 u32 s_acked; /* last un-ACK'ed entry */
523 u32 s_last; /* last completed entry */
524 u32 s_ssn; /* SSN of tail entry */
525 u32 s_lsn; /* limit sequence number (credit) */
526 u16 s_hdrwords; /* size of s_hdr in 32 bit words */
528 u8 s_state; /* opcode of last packet sent */
529 u8 s_ack_state; /* opcode of packet to ACK */
530 u8 s_nak_state; /* non-zero if NAK is pending */
531 u8 r_nak_state; /* non-zero if NAK is pending */
532 u8 s_retry; /* requester retry counter */
533 u8 s_rnr_retry; /* requester RNR retry counter */
534 u8 s_num_rd_atomic; /* number of RDMA read/atomic pending */
535 u8 s_tail_ack_queue; /* index into s_ack_queue[] */
537 struct qib_sge_state s_ack_rdma_sge;
538 struct timer_list s_timer;
540 struct qib_sge r_sg_list[0] /* verified SGEs */
541 ____cacheline_aligned_in_smp;
545 * Atomic bit definitions for r_aflags.
547 #define QIB_R_WRID_VALID 0
548 #define QIB_R_REWIND_SGE 1
551 * Bit definitions for r_flags.
553 #define QIB_R_REUSE_SGE 0x01
554 #define QIB_R_RDMAR_SEQ 0x02
555 #define QIB_R_RSP_NAK 0x04
556 #define QIB_R_RSP_SEND 0x08
557 #define QIB_R_COMM_EST 0x10
560 * Bit definitions for s_flags.
562 * QIB_S_SIGNAL_REQ_WR - set if QP send WRs contain completion signaled
563 * QIB_S_BUSY - send tasklet is processing the QP
564 * QIB_S_TIMER - the RC retry timer is active
565 * QIB_S_ACK_PENDING - an ACK is waiting to be sent after RDMA read/atomics
566 * QIB_S_WAIT_FENCE - waiting for all prior RDMA read or atomic SWQEs
567 * before processing the next SWQE
568 * QIB_S_WAIT_RDMAR - waiting for a RDMA read or atomic SWQE to complete
569 * before processing the next SWQE
570 * QIB_S_WAIT_RNR - waiting for RNR timeout
571 * QIB_S_WAIT_SSN_CREDIT - waiting for RC credits to process next SWQE
572 * QIB_S_WAIT_DMA - waiting for send DMA queue to drain before generating
573 * next send completion entry not via send DMA
574 * QIB_S_WAIT_PIO - waiting for a send buffer to be available
575 * QIB_S_WAIT_TX - waiting for a struct qib_verbs_txreq to be available
576 * QIB_S_WAIT_DMA_DESC - waiting for DMA descriptors to be available
577 * QIB_S_WAIT_KMEM - waiting for kernel memory to be available
578 * QIB_S_WAIT_PSN - waiting for a packet to exit the send DMA queue
579 * QIB_S_WAIT_ACK - waiting for an ACK packet before sending more requests
580 * QIB_S_SEND_ONE - send one packet, request ACK, then wait for ACK
582 #define QIB_S_SIGNAL_REQ_WR 0x0001
583 #define QIB_S_BUSY 0x0002
584 #define QIB_S_TIMER 0x0004
585 #define QIB_S_RESP_PENDING 0x0008
586 #define QIB_S_ACK_PENDING 0x0010
587 #define QIB_S_WAIT_FENCE 0x0020
588 #define QIB_S_WAIT_RDMAR 0x0040
589 #define QIB_S_WAIT_RNR 0x0080
590 #define QIB_S_WAIT_SSN_CREDIT 0x0100
591 #define QIB_S_WAIT_DMA 0x0200
592 #define QIB_S_WAIT_PIO 0x0400
593 #define QIB_S_WAIT_TX 0x0800
594 #define QIB_S_WAIT_DMA_DESC 0x1000
595 #define QIB_S_WAIT_KMEM 0x2000
596 #define QIB_S_WAIT_PSN 0x4000
597 #define QIB_S_WAIT_ACK 0x8000
598 #define QIB_S_SEND_ONE 0x10000
599 #define QIB_S_UNLIMITED_CREDIT 0x20000
602 * Wait flags that would prevent any packet type from being sent.
604 #define QIB_S_ANY_WAIT_IO (QIB_S_WAIT_PIO | QIB_S_WAIT_TX | \
605 QIB_S_WAIT_DMA_DESC | QIB_S_WAIT_KMEM)
608 * Wait flags that would prevent send work requests from making progress.
610 #define QIB_S_ANY_WAIT_SEND (QIB_S_WAIT_FENCE | QIB_S_WAIT_RDMAR | \
611 QIB_S_WAIT_RNR | QIB_S_WAIT_SSN_CREDIT | QIB_S_WAIT_DMA | \
612 QIB_S_WAIT_PSN | QIB_S_WAIT_ACK)
614 #define QIB_S_ANY_WAIT (QIB_S_ANY_WAIT_IO | QIB_S_ANY_WAIT_SEND)
616 #define QIB_PSN_CREDIT 16
619 * Since struct qib_swqe is not a fixed size, we can't simply index into
620 * struct qib_qp.s_wq. This function does the array index computation.
622 static inline struct qib_swqe *get_swqe_ptr(struct qib_qp *qp,
625 return (struct qib_swqe *)((char *)qp->s_wq +
626 (sizeof(struct qib_swqe) +
628 sizeof(struct qib_sge)) * n);
632 * Since struct qib_rwqe is not a fixed size, we can't simply index into
633 * struct qib_rwq.wq. This function does the array index computation.
635 static inline struct qib_rwqe *get_rwqe_ptr(struct qib_rq *rq, unsigned n)
637 return (struct qib_rwqe *)
638 ((char *) rq->wq->wq +
639 (sizeof(struct qib_rwqe) +
640 rq->max_sge * sizeof(struct ib_sge)) * n);
644 * QPN-map pages start out as NULL, they get allocated upon
645 * first use and are never deallocated. This way,
646 * large bitmaps are not allocated unless large numbers of QPs are used.
652 struct qib_qpn_table {
653 spinlock_t lock; /* protect changes in this struct */
654 unsigned flags; /* flags for QP0/1 allocated for each port */
655 u32 last; /* last QP number allocated */
656 u32 nmaps; /* size of the map table */
659 /* bit map of free QP numbers other than 0/1 */
660 struct qpn_map map[QPNMAP_ENTRIES];
663 #define MAX_LKEY_TABLE_BITS 23
665 struct qib_lkey_table {
666 spinlock_t lock; /* protect changes in this struct */
667 u32 next; /* next unused index (speeds search) */
668 u32 gen; /* generation count */
669 u32 max; /* size of the table */
670 struct qib_mregion __rcu **table;
673 struct qib_opcode_stats {
674 u64 n_packets; /* number of packets */
675 u64 n_bytes; /* total number of bytes */
678 struct qib_opcode_stats_perctx {
679 struct qib_opcode_stats stats[128];
682 struct qib_pma_counters {
683 u64 n_unicast_xmit; /* total unicast packets sent */
684 u64 n_unicast_rcv; /* total unicast packets received */
685 u64 n_multicast_xmit; /* total multicast packets sent */
686 u64 n_multicast_rcv; /* total multicast packets received */
690 struct qib_qp __rcu *qp0;
691 struct qib_qp __rcu *qp1;
692 struct ib_mad_agent *send_agent; /* agent for SMI (traps) */
693 struct qib_ah *sm_ah;
694 struct qib_ah *smi_ah;
695 struct rb_root mcast_tree;
696 spinlock_t lock; /* protect changes in this struct */
698 /* non-zero when timer is set */
699 unsigned long mkey_lease_timeout;
700 unsigned long trap_timeout;
701 __be64 gid_prefix; /* in network order */
703 __be64 guids[QIB_GUIDS_PER_PORT - 1]; /* writable GUIDs */
704 u64 tid; /* TID for traps */
705 struct qib_pma_counters __percpu *pmastats;
706 u64 z_unicast_xmit; /* starting count for PMA */
707 u64 z_unicast_rcv; /* starting count for PMA */
708 u64 z_multicast_xmit; /* starting count for PMA */
709 u64 z_multicast_rcv; /* starting count for PMA */
710 u64 z_symbol_error_counter; /* starting count for PMA */
711 u64 z_link_error_recovery_counter; /* starting count for PMA */
712 u64 z_link_downed_counter; /* starting count for PMA */
713 u64 z_port_rcv_errors; /* starting count for PMA */
714 u64 z_port_rcv_remphys_errors; /* starting count for PMA */
715 u64 z_port_xmit_discards; /* starting count for PMA */
716 u64 z_port_xmit_data; /* starting count for PMA */
717 u64 z_port_rcv_data; /* starting count for PMA */
718 u64 z_port_xmit_packets; /* starting count for PMA */
719 u64 z_port_rcv_packets; /* starting count for PMA */
720 u32 z_local_link_integrity_errors; /* starting count for PMA */
721 u32 z_excessive_buffer_overrun_errors; /* starting count for PMA */
722 u32 z_vl15_dropped; /* starting count for PMA */
726 u32 n_rc_delayed_comp;
740 u32 pma_sample_start;
741 u32 pma_sample_interval;
742 __be16 pma_counter_select[5];
747 u16 mkey_lease_period;
760 struct rvt_dev_info rdi;
761 struct list_head pending_mmaps;
762 spinlock_t mmap_offset_lock; /* protect mmap_offset */
764 struct qib_mregion __rcu *dma_mr;
766 /* QP numbers are shared by all IB ports */
767 struct qib_qpn_table qpn_table;
768 struct qib_lkey_table lk_table;
769 struct list_head piowait; /* list for wait PIO buf */
770 struct list_head dmawait; /* list for wait DMA */
771 struct list_head txwait; /* list for wait qib_verbs_txreq */
772 struct list_head memwait; /* list for wait kernel memory */
773 struct list_head txreq_free;
774 struct timer_list mem_timer;
775 struct qib_qp __rcu **qp_table;
776 struct qib_pio_header *pio_hdrs;
777 dma_addr_t pio_hdrs_phys;
778 /* list of QPs waiting for RNR timer */
779 spinlock_t pending_lock; /* protect wait lists, PMA counters, etc. */
780 u32 qp_table_size; /* size of the hash table */
781 u32 qp_rnd; /* random bytes for hash */
787 u32 n_pds_allocated; /* number of PDs allocated for device */
788 spinlock_t n_pds_lock;
789 u32 n_ahs_allocated; /* number of AHs allocated for device */
790 spinlock_t n_ahs_lock;
791 u32 n_cqs_allocated; /* number of CQs allocated for device */
792 spinlock_t n_cqs_lock;
793 u32 n_qps_allocated; /* number of QPs allocated for device */
794 spinlock_t n_qps_lock;
795 u32 n_srqs_allocated; /* number of SRQs allocated for device */
796 spinlock_t n_srqs_lock;
797 u32 n_mcast_grps_allocated; /* number of mcast groups allocated */
798 spinlock_t n_mcast_grps_lock;
799 #ifdef CONFIG_DEBUG_FS
800 /* per HCA debugfs */
801 struct dentry *qib_ibdev_dbg;
805 struct qib_verbs_counters {
806 u64 symbol_error_counter;
807 u64 link_error_recovery_counter;
808 u64 link_downed_counter;
810 u64 port_rcv_remphys_errors;
811 u64 port_xmit_discards;
814 u64 port_xmit_packets;
815 u64 port_rcv_packets;
816 u32 local_link_integrity_errors;
817 u32 excessive_buffer_overrun_errors;
821 static inline struct qib_mr *to_imr(struct ib_mr *ibmr)
823 return container_of(ibmr, struct qib_mr, ibmr);
826 static inline struct qib_ah *to_iah(struct ib_ah *ibah)
828 return container_of(ibah, struct qib_ah, ibah);
831 static inline struct qib_cq *to_icq(struct ib_cq *ibcq)
833 return container_of(ibcq, struct qib_cq, ibcq);
836 static inline struct qib_srq *to_isrq(struct ib_srq *ibsrq)
838 return container_of(ibsrq, struct qib_srq, ibsrq);
841 static inline struct qib_qp *to_iqp(struct ib_qp *ibqp)
843 return container_of(ibqp, struct qib_qp, ibqp);
846 static inline struct qib_ibdev *to_idev(struct ib_device *ibdev)
848 struct rvt_dev_info *rdi;
850 rdi = container_of(ibdev, struct rvt_dev_info, ibdev);
851 return container_of(rdi, struct qib_ibdev, rdi);
855 * Send if not busy or waiting for I/O and either
856 * a RC response is pending or we can process send work requests.
858 static inline int qib_send_ok(struct qib_qp *qp)
860 return !(qp->s_flags & (QIB_S_BUSY | QIB_S_ANY_WAIT_IO)) &&
861 (qp->s_hdrwords || (qp->s_flags & QIB_S_RESP_PENDING) ||
862 !(qp->s_flags & QIB_S_ANY_WAIT_SEND));
866 * This must be called with s_lock held.
868 void qib_schedule_send(struct qib_qp *qp);
870 static inline int qib_pkey_ok(u16 pkey1, u16 pkey2)
872 u16 p1 = pkey1 & 0x7FFF;
873 u16 p2 = pkey2 & 0x7FFF;
876 * Low 15 bits must be non-zero and match, and
877 * one of the two must be a full member.
879 return p1 && p1 == p2 && ((__s16)pkey1 < 0 || (__s16)pkey2 < 0);
882 void qib_bad_pqkey(struct qib_ibport *ibp, __be16 trap_num, u32 key, u32 sl,
883 u32 qp1, u32 qp2, __be16 lid1, __be16 lid2);
884 void qib_cap_mask_chg(struct qib_ibport *ibp);
885 void qib_sys_guid_chg(struct qib_ibport *ibp);
886 void qib_node_desc_chg(struct qib_ibport *ibp);
887 int qib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
888 const struct ib_wc *in_wc, const struct ib_grh *in_grh,
889 const struct ib_mad_hdr *in, size_t in_mad_size,
890 struct ib_mad_hdr *out, size_t *out_mad_size,
891 u16 *out_mad_pkey_index);
892 int qib_create_agents(struct qib_ibdev *dev);
893 void qib_free_agents(struct qib_ibdev *dev);
896 * Compare the lower 24 bits of the two values.
897 * Returns an integer <, ==, or > than zero.
899 static inline int qib_cmp24(u32 a, u32 b)
901 return (((int) a) - ((int) b)) << 8;
904 struct qib_mcast *qib_mcast_find(struct qib_ibport *ibp, union ib_gid *mgid);
906 int qib_snapshot_counters(struct qib_pportdata *ppd, u64 *swords,
907 u64 *rwords, u64 *spkts, u64 *rpkts,
910 int qib_get_counters(struct qib_pportdata *ppd,
911 struct qib_verbs_counters *cntrs);
913 int qib_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
915 int qib_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
917 int qib_mcast_tree_empty(struct qib_ibport *ibp);
919 __be32 qib_compute_aeth(struct qib_qp *qp);
921 struct qib_qp *qib_lookup_qpn(struct qib_ibport *ibp, u32 qpn);
923 struct ib_qp *qib_create_qp(struct ib_pd *ibpd,
924 struct ib_qp_init_attr *init_attr,
925 struct ib_udata *udata);
927 int qib_destroy_qp(struct ib_qp *ibqp);
929 int qib_error_qp(struct qib_qp *qp, enum ib_wc_status err);
931 int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
932 int attr_mask, struct ib_udata *udata);
934 int qib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
935 int attr_mask, struct ib_qp_init_attr *init_attr);
937 unsigned qib_free_all_qps(struct qib_devdata *dd);
939 void qib_init_qpn_table(struct qib_devdata *dd, struct qib_qpn_table *qpt);
941 void qib_free_qpn_table(struct qib_qpn_table *qpt);
943 #ifdef CONFIG_DEBUG_FS
947 struct qib_qp_iter *qib_qp_iter_init(struct qib_ibdev *dev);
949 int qib_qp_iter_next(struct qib_qp_iter *iter);
951 void qib_qp_iter_print(struct seq_file *s, struct qib_qp_iter *iter);
955 void qib_get_credit(struct qib_qp *qp, u32 aeth);
957 unsigned qib_pkt_delay(u32 plen, u8 snd_mult, u8 rcv_mult);
959 void qib_verbs_sdma_desc_avail(struct qib_pportdata *ppd, unsigned avail);
961 void qib_put_txreq(struct qib_verbs_txreq *tx);
963 int qib_verbs_send(struct qib_qp *qp, struct qib_ib_header *hdr,
964 u32 hdrwords, struct qib_sge_state *ss, u32 len);
966 void qib_copy_sge(struct qib_sge_state *ss, void *data, u32 length,
969 void qib_skip_sge(struct qib_sge_state *ss, u32 length, int release);
971 void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
972 int has_grh, void *data, u32 tlen, struct qib_qp *qp);
974 void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
975 int has_grh, void *data, u32 tlen, struct qib_qp *qp);
977 int qib_check_ah(struct ib_device *ibdev, struct ib_ah_attr *ah_attr);
979 struct ib_ah *qib_create_qp0_ah(struct qib_ibport *ibp, u16 dlid);
981 void qib_rc_rnr_retry(unsigned long arg);
983 void qib_rc_send_complete(struct qib_qp *qp, struct qib_ib_header *hdr);
985 void qib_rc_error(struct qib_qp *qp, enum ib_wc_status err);
987 int qib_post_ud_send(struct qib_qp *qp, struct ib_send_wr *wr);
989 void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
990 int has_grh, void *data, u32 tlen, struct qib_qp *qp);
992 int qib_alloc_lkey(struct qib_mregion *mr, int dma_region);
994 void qib_free_lkey(struct qib_mregion *mr);
996 int qib_lkey_ok(struct qib_lkey_table *rkt, struct rvt_pd *pd,
997 struct qib_sge *isge, struct ib_sge *sge, int acc);
999 int qib_rkey_ok(struct qib_qp *qp, struct qib_sge *sge,
1000 u32 len, u64 vaddr, u32 rkey, int acc);
1002 int qib_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
1003 struct ib_recv_wr **bad_wr);
1005 struct ib_srq *qib_create_srq(struct ib_pd *ibpd,
1006 struct ib_srq_init_attr *srq_init_attr,
1007 struct ib_udata *udata);
1009 int qib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
1010 enum ib_srq_attr_mask attr_mask,
1011 struct ib_udata *udata);
1013 int qib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr);
1015 int qib_destroy_srq(struct ib_srq *ibsrq);
1017 int qib_cq_init(struct qib_devdata *dd);
1019 void qib_cq_exit(struct qib_devdata *dd);
1021 void qib_cq_enter(struct qib_cq *cq, struct ib_wc *entry, int sig);
1023 int qib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry);
1025 struct ib_cq *qib_create_cq(struct ib_device *ibdev,
1026 const struct ib_cq_init_attr *attr,
1027 struct ib_ucontext *context,
1028 struct ib_udata *udata);
1030 int qib_destroy_cq(struct ib_cq *ibcq);
1032 int qib_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags);
1034 int qib_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata);
1036 struct ib_mr *qib_get_dma_mr(struct ib_pd *pd, int acc);
1038 struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
1039 u64 virt_addr, int mr_access_flags,
1040 struct ib_udata *udata);
1042 int qib_dereg_mr(struct ib_mr *ibmr);
1044 struct ib_mr *qib_alloc_mr(struct ib_pd *pd,
1045 enum ib_mr_type mr_type,
1048 int qib_map_mr_sg(struct ib_mr *ibmr,
1049 struct scatterlist *sg,
1052 int qib_reg_mr(struct qib_qp *qp, struct ib_reg_wr *wr);
1054 struct ib_fmr *qib_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
1055 struct ib_fmr_attr *fmr_attr);
1057 int qib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
1058 int list_len, u64 iova);
1060 int qib_unmap_fmr(struct list_head *fmr_list);
1062 int qib_dealloc_fmr(struct ib_fmr *ibfmr);
1064 static inline void qib_get_mr(struct qib_mregion *mr)
1066 atomic_inc(&mr->refcount);
1069 void mr_rcu_callback(struct rcu_head *list);
1071 static inline void qib_put_mr(struct qib_mregion *mr)
1073 if (unlikely(atomic_dec_and_test(&mr->refcount)))
1074 call_rcu(&mr->list, mr_rcu_callback);
1077 static inline void qib_put_ss(struct qib_sge_state *ss)
1079 while (ss->num_sge) {
1080 qib_put_mr(ss->sge.mr);
1082 ss->sge = *ss->sg_list++;
1087 void qib_release_mmap_info(struct kref *ref);
1089 struct qib_mmap_info *qib_create_mmap_info(struct qib_ibdev *dev, u32 size,
1090 struct ib_ucontext *context,
1093 void qib_update_mmap_info(struct qib_ibdev *dev, struct qib_mmap_info *ip,
1094 u32 size, void *obj);
1096 int qib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma);
1098 int qib_get_rwqe(struct qib_qp *qp, int wr_id_only);
1100 void qib_migrate_qp(struct qib_qp *qp);
1102 int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr,
1103 int has_grh, struct qib_qp *qp, u32 bth0);
1105 u32 qib_make_grh(struct qib_ibport *ibp, struct ib_grh *hdr,
1106 struct ib_global_route *grh, u32 hwords, u32 nwords);
1108 void qib_make_ruc_header(struct qib_qp *qp, struct qib_other_headers *ohdr,
1109 u32 bth0, u32 bth2);
1111 void qib_do_send(struct work_struct *work);
1113 void qib_send_complete(struct qib_qp *qp, struct qib_swqe *wqe,
1114 enum ib_wc_status status);
1116 void qib_send_rc_ack(struct qib_qp *qp);
1118 int qib_make_rc_req(struct qib_qp *qp);
1120 int qib_make_uc_req(struct qib_qp *qp);
1122 int qib_make_ud_req(struct qib_qp *qp);
1124 int qib_register_ib_device(struct qib_devdata *);
1126 void qib_unregister_ib_device(struct qib_devdata *);
1128 void qib_ib_rcv(struct qib_ctxtdata *, void *, void *, u32);
1130 void qib_ib_piobufavail(struct qib_devdata *);
1132 unsigned qib_get_npkeys(struct qib_devdata *);
1134 unsigned qib_get_pkey(struct qib_ibport *, unsigned);
1136 extern const enum ib_wc_opcode ib_qib_wc_opcode[];
1139 * Below HCA-independent IB PhysPortState values, returned
1140 * by the f_ibphys_portstate() routine.
1142 #define IB_PHYSPORTSTATE_SLEEP 1
1143 #define IB_PHYSPORTSTATE_POLL 2
1144 #define IB_PHYSPORTSTATE_DISABLED 3
1145 #define IB_PHYSPORTSTATE_CFG_TRAIN 4
1146 #define IB_PHYSPORTSTATE_LINKUP 5
1147 #define IB_PHYSPORTSTATE_LINK_ERR_RECOVER 6
1148 #define IB_PHYSPORTSTATE_CFG_DEBOUNCE 8
1149 #define IB_PHYSPORTSTATE_CFG_IDLE 0xB
1150 #define IB_PHYSPORTSTATE_RECOVERY_RETRAIN 0xC
1151 #define IB_PHYSPORTSTATE_RECOVERY_WAITRMT 0xE
1152 #define IB_PHYSPORTSTATE_RECOVERY_IDLE 0xF
1153 #define IB_PHYSPORTSTATE_CFG_ENH 0x10
1154 #define IB_PHYSPORTSTATE_CFG_WAIT_ENH 0x13
1156 extern const int ib_qib_state_ops[];
1158 extern __be64 ib_qib_sys_image_guid; /* in network order */
1160 extern unsigned int ib_qib_lkey_table_size;
1162 extern unsigned int ib_qib_max_cqes;
1164 extern unsigned int ib_qib_max_cqs;
1166 extern unsigned int ib_qib_max_qp_wrs;
1168 extern unsigned int ib_qib_max_qps;
1170 extern unsigned int ib_qib_max_sges;
1172 extern unsigned int ib_qib_max_mcast_grps;
1174 extern unsigned int ib_qib_max_mcast_qp_attached;
1176 extern unsigned int ib_qib_max_srqs;
1178 extern unsigned int ib_qib_max_srq_sges;
1180 extern unsigned int ib_qib_max_srq_wrs;
1182 extern const u32 ib_qib_rnr_table[];
1184 #endif /* QIB_VERBS_H */