2 * Copyright (c) 2006 QLogic, Inc. All rights reserved.
3 * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 #include <rdma/ib_mad.h>
35 #include <rdma/ib_user_verbs.h>
36 #include <linux/utsname.h>
38 #include "ipath_kernel.h"
39 #include "ipath_verbs.h"
40 #include "ipath_common.h"
42 /* Not static, because we don't want the compiler removing it */
43 const char ipath_verbs_version[] = "ipath_verbs " IPATH_IDSTR;
45 static unsigned int ib_ipath_qp_table_size = 251;
46 module_param_named(qp_table_size, ib_ipath_qp_table_size, uint, S_IRUGO);
47 MODULE_PARM_DESC(qp_table_size, "QP table size");
49 unsigned int ib_ipath_lkey_table_size = 12;
50 module_param_named(lkey_table_size, ib_ipath_lkey_table_size, uint,
52 MODULE_PARM_DESC(lkey_table_size,
53 "LKEY table size in bits (2^n, 1 <= n <= 23)");
55 unsigned int ib_ipath_debug; /* debug mask */
56 module_param_named(debug, ib_ipath_debug, uint, S_IWUSR | S_IRUGO);
57 MODULE_PARM_DESC(debug, "Verbs debug mask");
59 static unsigned int ib_ipath_max_pds = 0xFFFF;
60 module_param_named(max_pds, ib_ipath_max_pds, uint, S_IWUSR | S_IRUGO);
61 MODULE_PARM_DESC(max_pds,
62 "Maximum number of protection domains to support");
64 static unsigned int ib_ipath_max_ahs = 0xFFFF;
65 module_param_named(max_ahs, ib_ipath_max_ahs, uint, S_IWUSR | S_IRUGO);
66 MODULE_PARM_DESC(max_ahs, "Maximum number of address handles to support");
68 unsigned int ib_ipath_max_cqes = 0x2FFFF;
69 module_param_named(max_cqes, ib_ipath_max_cqes, uint, S_IWUSR | S_IRUGO);
70 MODULE_PARM_DESC(max_cqes,
71 "Maximum number of completion queue entries to support");
73 unsigned int ib_ipath_max_cqs = 0x1FFFF;
74 module_param_named(max_cqs, ib_ipath_max_cqs, uint, S_IWUSR | S_IRUGO);
75 MODULE_PARM_DESC(max_cqs, "Maximum number of completion queues to support");
77 unsigned int ib_ipath_max_qp_wrs = 0x3FFF;
78 module_param_named(max_qp_wrs, ib_ipath_max_qp_wrs, uint,
80 MODULE_PARM_DESC(max_qp_wrs, "Maximum number of QP WRs to support");
82 unsigned int ib_ipath_max_sges = 0x60;
83 module_param_named(max_sges, ib_ipath_max_sges, uint, S_IWUSR | S_IRUGO);
84 MODULE_PARM_DESC(max_sges, "Maximum number of SGEs to support");
86 unsigned int ib_ipath_max_mcast_grps = 16384;
87 module_param_named(max_mcast_grps, ib_ipath_max_mcast_grps, uint,
89 MODULE_PARM_DESC(max_mcast_grps,
90 "Maximum number of multicast groups to support");
92 unsigned int ib_ipath_max_mcast_qp_attached = 16;
93 module_param_named(max_mcast_qp_attached, ib_ipath_max_mcast_qp_attached,
94 uint, S_IWUSR | S_IRUGO);
95 MODULE_PARM_DESC(max_mcast_qp_attached,
96 "Maximum number of attached QPs to support");
98 unsigned int ib_ipath_max_srqs = 1024;
99 module_param_named(max_srqs, ib_ipath_max_srqs, uint, S_IWUSR | S_IRUGO);
100 MODULE_PARM_DESC(max_srqs, "Maximum number of SRQs to support");
102 unsigned int ib_ipath_max_srq_sges = 128;
103 module_param_named(max_srq_sges, ib_ipath_max_srq_sges,
104 uint, S_IWUSR | S_IRUGO);
105 MODULE_PARM_DESC(max_srq_sges, "Maximum number of SRQ SGEs to support");
107 unsigned int ib_ipath_max_srq_wrs = 0x1FFFF;
108 module_param_named(max_srq_wrs, ib_ipath_max_srq_wrs,
109 uint, S_IWUSR | S_IRUGO);
110 MODULE_PARM_DESC(max_srq_wrs, "Maximum number of SRQ WRs support");
112 MODULE_LICENSE("GPL");
113 MODULE_AUTHOR("QLogic <support@pathscale.com>");
114 MODULE_DESCRIPTION("QLogic InfiniPath driver");
116 const int ib_ipath_state_ops[IB_QPS_ERR + 1] = {
118 [IB_QPS_INIT] = IPATH_POST_RECV_OK,
119 [IB_QPS_RTR] = IPATH_POST_RECV_OK | IPATH_PROCESS_RECV_OK,
120 [IB_QPS_RTS] = IPATH_POST_RECV_OK | IPATH_PROCESS_RECV_OK |
121 IPATH_POST_SEND_OK | IPATH_PROCESS_SEND_OK,
122 [IB_QPS_SQD] = IPATH_POST_RECV_OK | IPATH_PROCESS_RECV_OK |
124 [IB_QPS_SQE] = IPATH_POST_RECV_OK | IPATH_PROCESS_RECV_OK,
129 * Translate ib_wr_opcode into ib_wc_opcode.
131 const enum ib_wc_opcode ib_ipath_wc_opcode[] = {
132 [IB_WR_RDMA_WRITE] = IB_WC_RDMA_WRITE,
133 [IB_WR_RDMA_WRITE_WITH_IMM] = IB_WC_RDMA_WRITE,
134 [IB_WR_SEND] = IB_WC_SEND,
135 [IB_WR_SEND_WITH_IMM] = IB_WC_SEND,
136 [IB_WR_RDMA_READ] = IB_WC_RDMA_READ,
137 [IB_WR_ATOMIC_CMP_AND_SWP] = IB_WC_COMP_SWAP,
138 [IB_WR_ATOMIC_FETCH_AND_ADD] = IB_WC_FETCH_ADD
144 static __be64 sys_image_guid;
147 * ipath_copy_sge - copy data to SGE memory
149 * @data: the data to copy
150 * @length: the length of the data
152 void ipath_copy_sge(struct ipath_sge_state *ss, void *data, u32 length)
154 struct ipath_sge *sge = &ss->sge;
157 u32 len = sge->length;
162 memcpy(sge->vaddr, data, len);
165 sge->sge_length -= len;
166 if (sge->sge_length == 0) {
168 *sge = *ss->sg_list++;
169 } else if (sge->length == 0 && sge->mr != NULL) {
170 if (++sge->n >= IPATH_SEGSZ) {
171 if (++sge->m >= sge->mr->mapsz)
176 sge->mr->map[sge->m]->segs[sge->n].vaddr;
178 sge->mr->map[sge->m]->segs[sge->n].length;
186 * ipath_skip_sge - skip over SGE memory - XXX almost dup of prev func
188 * @length: the number of bytes to skip
190 void ipath_skip_sge(struct ipath_sge_state *ss, u32 length)
192 struct ipath_sge *sge = &ss->sge;
195 u32 len = sge->length;
202 sge->sge_length -= len;
203 if (sge->sge_length == 0) {
205 *sge = *ss->sg_list++;
206 } else if (sge->length == 0 && sge->mr != NULL) {
207 if (++sge->n >= IPATH_SEGSZ) {
208 if (++sge->m >= sge->mr->mapsz)
213 sge->mr->map[sge->m]->segs[sge->n].vaddr;
215 sge->mr->map[sge->m]->segs[sge->n].length;
222 * ipath_post_send - post a send on a QP
223 * @ibqp: the QP to post the send on
224 * @wr: the list of work requests to post
225 * @bad_wr: the first bad WR is put here
227 * This may be called from interrupt context.
229 static int ipath_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
230 struct ib_send_wr **bad_wr)
232 struct ipath_qp *qp = to_iqp(ibqp);
235 /* Check that state is OK to post send. */
236 if (!(ib_ipath_state_ops[qp->state] & IPATH_POST_SEND_OK)) {
242 for (; wr; wr = wr->next) {
243 switch (qp->ibqp.qp_type) {
246 err = ipath_post_ruc_send(qp, wr);
252 err = ipath_post_ud_send(qp, wr);
269 * ipath_post_receive - post a receive on a QP
270 * @ibqp: the QP to post the receive on
271 * @wr: the WR to post
272 * @bad_wr: the first bad WR is put here
274 * This may be called from interrupt context.
276 static int ipath_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
277 struct ib_recv_wr **bad_wr)
279 struct ipath_qp *qp = to_iqp(ibqp);
280 struct ipath_rwq *wq = qp->r_rq.wq;
284 /* Check that state is OK to post receive. */
285 if (!(ib_ipath_state_ops[qp->state] & IPATH_POST_RECV_OK) || !wq) {
291 for (; wr; wr = wr->next) {
292 struct ipath_rwqe *wqe;
296 if ((unsigned) wr->num_sge > qp->r_rq.max_sge) {
302 spin_lock_irqsave(&qp->r_rq.lock, flags);
304 if (next >= qp->r_rq.size)
306 if (next == wq->tail) {
307 spin_unlock_irqrestore(&qp->r_rq.lock, flags);
313 wqe = get_rwqe_ptr(&qp->r_rq, wq->head);
314 wqe->wr_id = wr->wr_id;
315 wqe->num_sge = wr->num_sge;
316 for (i = 0; i < wr->num_sge; i++)
317 wqe->sg_list[i] = wr->sg_list[i];
319 spin_unlock_irqrestore(&qp->r_rq.lock, flags);
328 * ipath_qp_rcv - processing an incoming packet on a QP
329 * @dev: the device the packet came on
330 * @hdr: the packet header
331 * @has_grh: true if the packet has a GRH
332 * @data: the packet data
333 * @tlen: the packet length
334 * @qp: the QP the packet came on
336 * This is called from ipath_ib_rcv() to process an incoming packet
338 * Called at interrupt level.
340 static void ipath_qp_rcv(struct ipath_ibdev *dev,
341 struct ipath_ib_header *hdr, int has_grh,
342 void *data, u32 tlen, struct ipath_qp *qp)
344 /* Check for valid receive state. */
345 if (!(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_RECV_OK)) {
350 switch (qp->ibqp.qp_type) {
354 ipath_ud_rcv(dev, hdr, has_grh, data, tlen, qp);
358 ipath_rc_rcv(dev, hdr, has_grh, data, tlen, qp);
362 ipath_uc_rcv(dev, hdr, has_grh, data, tlen, qp);
371 * ipath_ib_rcv - process and incoming packet
372 * @arg: the device pointer
373 * @rhdr: the header of the packet
374 * @data: the packet data
375 * @tlen: the packet length
377 * This is called from ipath_kreceive() to process an incoming packet at
378 * interrupt level. Tlen is the length of the header + data + CRC in bytes.
380 static void ipath_ib_rcv(void *arg, void *rhdr, void *data, u32 tlen)
382 struct ipath_ibdev *dev = (struct ipath_ibdev *) arg;
383 struct ipath_ib_header *hdr = rhdr;
384 struct ipath_other_headers *ohdr;
391 if (unlikely(dev == NULL))
394 if (unlikely(tlen < 24)) { /* LRH+BTH+CRC */
399 /* Check for a valid destination LID (see ch. 7.11.1). */
400 lid = be16_to_cpu(hdr->lrh[1]);
401 if (lid < IPATH_MULTICAST_LID_BASE) {
402 lid &= ~((1 << (dev->mkeyprot_resv_lmc & 7)) - 1);
403 if (unlikely(lid != ipath_layer_get_lid(dev->dd))) {
410 lnh = be16_to_cpu(hdr->lrh[0]) & 3;
411 if (lnh == IPATH_LRH_BTH)
413 else if (lnh == IPATH_LRH_GRH)
414 ohdr = &hdr->u.l.oth;
420 opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
421 dev->opstats[opcode].n_bytes += tlen;
422 dev->opstats[opcode].n_packets++;
424 /* Get the destination QP number. */
425 qp_num = be32_to_cpu(ohdr->bth[1]) & IPATH_QPN_MASK;
426 if (qp_num == IPATH_MULTICAST_QPN) {
427 struct ipath_mcast *mcast;
428 struct ipath_mcast_qp *p;
430 mcast = ipath_mcast_find(&hdr->u.l.grh.dgid);
435 dev->n_multicast_rcv++;
436 list_for_each_entry_rcu(p, &mcast->qp_list, list)
437 ipath_qp_rcv(dev, hdr, lnh == IPATH_LRH_GRH, data,
440 * Notify ipath_multicast_detach() if it is waiting for us
443 if (atomic_dec_return(&mcast->refcount) <= 1)
444 wake_up(&mcast->wait);
446 qp = ipath_lookup_qpn(&dev->qp_table, qp_num);
448 dev->n_unicast_rcv++;
449 ipath_qp_rcv(dev, hdr, lnh == IPATH_LRH_GRH, data,
452 * Notify ipath_destroy_qp() if it is waiting
455 if (atomic_dec_and_test(&qp->refcount))
465 * ipath_ib_timer - verbs timer
466 * @arg: the device pointer
468 * This is called from ipath_do_rcv_timer() at interrupt level to check for
469 * QPs which need retransmits and to collect performance numbers.
471 static void ipath_ib_timer(void *arg)
473 struct ipath_ibdev *dev = (struct ipath_ibdev *) arg;
474 struct ipath_qp *resend = NULL;
475 struct list_head *last;
482 spin_lock_irqsave(&dev->pending_lock, flags);
483 /* Start filling the next pending queue. */
484 if (++dev->pending_index >= ARRAY_SIZE(dev->pending))
485 dev->pending_index = 0;
486 /* Save any requests still in the new queue, they have timed out. */
487 last = &dev->pending[dev->pending_index];
488 while (!list_empty(last)) {
489 qp = list_entry(last->next, struct ipath_qp, timerwait);
490 list_del_init(&qp->timerwait);
491 qp->timer_next = resend;
493 atomic_inc(&qp->refcount);
495 last = &dev->rnrwait;
496 if (!list_empty(last)) {
497 qp = list_entry(last->next, struct ipath_qp, timerwait);
498 if (--qp->s_rnr_timeout == 0) {
500 list_del_init(&qp->timerwait);
501 tasklet_hi_schedule(&qp->s_task);
502 if (list_empty(last))
504 qp = list_entry(last->next, struct ipath_qp,
506 } while (qp->s_rnr_timeout == 0);
510 * We should only be in the started state if pma_sample_start != 0
512 if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_STARTED &&
513 --dev->pma_sample_start == 0) {
514 dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_RUNNING;
515 ipath_layer_snapshot_counters(dev->dd, &dev->ipath_sword,
519 &dev->ipath_xmit_wait);
521 if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_RUNNING) {
522 if (dev->pma_sample_interval == 0) {
523 u64 ta, tb, tc, td, te;
525 dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_DONE;
526 ipath_layer_snapshot_counters(dev->dd, &ta, &tb,
529 dev->ipath_sword = ta - dev->ipath_sword;
530 dev->ipath_rword = tb - dev->ipath_rword;
531 dev->ipath_spkts = tc - dev->ipath_spkts;
532 dev->ipath_rpkts = td - dev->ipath_rpkts;
533 dev->ipath_xmit_wait = te - dev->ipath_xmit_wait;
536 dev->pma_sample_interval--;
538 spin_unlock_irqrestore(&dev->pending_lock, flags);
540 /* XXX What if timer fires again while this is running? */
541 for (qp = resend; qp != NULL; qp = qp->timer_next) {
544 spin_lock_irqsave(&qp->s_lock, flags);
545 if (qp->s_last != qp->s_tail && qp->state == IB_QPS_RTS) {
547 ipath_restart_rc(qp, qp->s_last_psn + 1, &wc);
549 spin_unlock_irqrestore(&qp->s_lock, flags);
551 /* Notify ipath_destroy_qp() if it is waiting. */
552 if (atomic_dec_and_test(&qp->refcount))
558 * ipath_ib_piobufavail - callback when a PIO buffer is available
559 * @arg: the device pointer
561 * This is called from ipath_intr() at interrupt level when a PIO buffer is
562 * available after ipath_verbs_send() returned an error that no buffers were
563 * available. Return 1 if we consumed all the PIO buffers and we still have
564 * QPs waiting for buffers (for now, just do a tasklet_hi_schedule and
567 static int ipath_ib_piobufavail(void *arg)
569 struct ipath_ibdev *dev = (struct ipath_ibdev *) arg;
576 spin_lock_irqsave(&dev->pending_lock, flags);
577 while (!list_empty(&dev->piowait)) {
578 qp = list_entry(dev->piowait.next, struct ipath_qp,
580 list_del_init(&qp->piowait);
581 tasklet_hi_schedule(&qp->s_task);
583 spin_unlock_irqrestore(&dev->pending_lock, flags);
589 static int ipath_query_device(struct ib_device *ibdev,
590 struct ib_device_attr *props)
592 struct ipath_ibdev *dev = to_idev(ibdev);
594 memset(props, 0, sizeof(*props));
596 props->device_cap_flags = IB_DEVICE_BAD_PKEY_CNTR |
597 IB_DEVICE_BAD_QKEY_CNTR | IB_DEVICE_SHUTDOWN_PORT |
598 IB_DEVICE_SYS_IMAGE_GUID;
599 props->page_size_cap = PAGE_SIZE;
600 props->vendor_id = ipath_layer_get_vendorid(dev->dd);
601 props->vendor_part_id = ipath_layer_get_deviceid(dev->dd);
602 props->hw_ver = ipath_layer_get_pcirev(dev->dd);
604 props->sys_image_guid = dev->sys_image_guid;
606 props->max_mr_size = ~0ull;
607 props->max_qp = dev->qp_table.max;
608 props->max_qp_wr = ib_ipath_max_qp_wrs;
609 props->max_sge = ib_ipath_max_sges;
610 props->max_cq = ib_ipath_max_cqs;
611 props->max_ah = ib_ipath_max_ahs;
612 props->max_cqe = ib_ipath_max_cqes;
613 props->max_mr = dev->lk_table.max;
614 props->max_pd = ib_ipath_max_pds;
615 props->max_qp_rd_atom = 1;
616 props->max_qp_init_rd_atom = 1;
617 /* props->max_res_rd_atom */
618 props->max_srq = ib_ipath_max_srqs;
619 props->max_srq_wr = ib_ipath_max_srq_wrs;
620 props->max_srq_sge = ib_ipath_max_srq_sges;
621 /* props->local_ca_ack_delay */
622 props->atomic_cap = IB_ATOMIC_HCA;
623 props->max_pkeys = ipath_layer_get_npkeys(dev->dd);
624 props->max_mcast_grp = ib_ipath_max_mcast_grps;
625 props->max_mcast_qp_attach = ib_ipath_max_mcast_qp_attached;
626 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
627 props->max_mcast_grp;
632 const u8 ipath_cvt_physportstate[16] = {
633 [INFINIPATH_IBCS_LT_STATE_DISABLED] = 3,
634 [INFINIPATH_IBCS_LT_STATE_LINKUP] = 5,
635 [INFINIPATH_IBCS_LT_STATE_POLLACTIVE] = 2,
636 [INFINIPATH_IBCS_LT_STATE_POLLQUIET] = 2,
637 [INFINIPATH_IBCS_LT_STATE_SLEEPDELAY] = 1,
638 [INFINIPATH_IBCS_LT_STATE_SLEEPQUIET] = 1,
639 [INFINIPATH_IBCS_LT_STATE_CFGDEBOUNCE] = 4,
640 [INFINIPATH_IBCS_LT_STATE_CFGRCVFCFG] = 4,
641 [INFINIPATH_IBCS_LT_STATE_CFGWAITRMT] = 4,
642 [INFINIPATH_IBCS_LT_STATE_CFGIDLE] = 4,
643 [INFINIPATH_IBCS_LT_STATE_RECOVERRETRAIN] = 6,
644 [INFINIPATH_IBCS_LT_STATE_RECOVERWAITRMT] = 6,
645 [INFINIPATH_IBCS_LT_STATE_RECOVERIDLE] = 6,
648 static int ipath_query_port(struct ib_device *ibdev,
649 u8 port, struct ib_port_attr *props)
651 struct ipath_ibdev *dev = to_idev(ibdev);
653 u16 lid = ipath_layer_get_lid(dev->dd);
656 memset(props, 0, sizeof(*props));
657 props->lid = lid ? lid : __constant_be16_to_cpu(IB_LID_PERMISSIVE);
658 props->lmc = dev->mkeyprot_resv_lmc & 7;
659 props->sm_lid = dev->sm_lid;
660 props->sm_sl = dev->sm_sl;
661 ibcstat = ipath_layer_get_lastibcstat(dev->dd);
662 props->state = ((ibcstat >> 4) & 0x3) + 1;
663 /* See phys_state_show() */
664 props->phys_state = ipath_cvt_physportstate[
665 ipath_layer_get_lastibcstat(dev->dd) & 0xf];
666 props->port_cap_flags = dev->port_cap_flags;
667 props->gid_tbl_len = 1;
668 props->max_msg_sz = 0x80000000;
669 props->pkey_tbl_len = ipath_layer_get_npkeys(dev->dd);
670 props->bad_pkey_cntr = ipath_layer_get_cr_errpkey(dev->dd) -
671 dev->z_pkey_violations;
672 props->qkey_viol_cntr = dev->qkey_violations;
673 props->active_width = IB_WIDTH_4X;
674 /* See rate_show() */
675 props->active_speed = 1; /* Regular 10Mbs speed. */
676 props->max_vl_num = 1; /* VLCap = VL0 */
677 props->init_type_reply = 0;
679 props->max_mtu = IB_MTU_4096;
680 switch (ipath_layer_get_ibmtu(dev->dd)) {
699 props->active_mtu = mtu;
700 props->subnet_timeout = dev->subnet_timeout;
705 static int ipath_modify_device(struct ib_device *device,
706 int device_modify_mask,
707 struct ib_device_modify *device_modify)
711 if (device_modify_mask & ~(IB_DEVICE_MODIFY_SYS_IMAGE_GUID |
712 IB_DEVICE_MODIFY_NODE_DESC)) {
717 if (device_modify_mask & IB_DEVICE_MODIFY_NODE_DESC)
718 memcpy(device->node_desc, device_modify->node_desc, 64);
720 if (device_modify_mask & IB_DEVICE_MODIFY_SYS_IMAGE_GUID)
721 to_idev(device)->sys_image_guid =
722 cpu_to_be64(device_modify->sys_image_guid);
730 static int ipath_modify_port(struct ib_device *ibdev,
731 u8 port, int port_modify_mask,
732 struct ib_port_modify *props)
734 struct ipath_ibdev *dev = to_idev(ibdev);
736 dev->port_cap_flags |= props->set_port_cap_mask;
737 dev->port_cap_flags &= ~props->clr_port_cap_mask;
738 if (port_modify_mask & IB_PORT_SHUTDOWN)
739 ipath_layer_set_linkstate(dev->dd, IPATH_IB_LINKDOWN);
740 if (port_modify_mask & IB_PORT_RESET_QKEY_CNTR)
741 dev->qkey_violations = 0;
745 static int ipath_query_gid(struct ib_device *ibdev, u8 port,
746 int index, union ib_gid *gid)
748 struct ipath_ibdev *dev = to_idev(ibdev);
755 gid->global.subnet_prefix = dev->gid_prefix;
756 gid->global.interface_id = ipath_layer_get_guid(dev->dd);
764 static struct ib_pd *ipath_alloc_pd(struct ib_device *ibdev,
765 struct ib_ucontext *context,
766 struct ib_udata *udata)
768 struct ipath_ibdev *dev = to_idev(ibdev);
773 * This is actually totally arbitrary. Some correctness tests
774 * assume there's a maximum number of PDs that can be allocated.
775 * We don't actually have this limit, but we fail the test if
776 * we allow allocations of more than we report for this value.
779 pd = kmalloc(sizeof *pd, GFP_KERNEL);
781 ret = ERR_PTR(-ENOMEM);
785 spin_lock(&dev->n_pds_lock);
786 if (dev->n_pds_allocated == ib_ipath_max_pds) {
787 spin_unlock(&dev->n_pds_lock);
789 ret = ERR_PTR(-ENOMEM);
793 dev->n_pds_allocated++;
794 spin_unlock(&dev->n_pds_lock);
796 /* ib_alloc_pd() will initialize pd->ibpd. */
797 pd->user = udata != NULL;
805 static int ipath_dealloc_pd(struct ib_pd *ibpd)
807 struct ipath_pd *pd = to_ipd(ibpd);
808 struct ipath_ibdev *dev = to_idev(ibpd->device);
810 spin_lock(&dev->n_pds_lock);
811 dev->n_pds_allocated--;
812 spin_unlock(&dev->n_pds_lock);
820 * ipath_create_ah - create an address handle
821 * @pd: the protection domain
822 * @ah_attr: the attributes of the AH
824 * This may be called from interrupt context.
826 static struct ib_ah *ipath_create_ah(struct ib_pd *pd,
827 struct ib_ah_attr *ah_attr)
831 struct ipath_ibdev *dev = to_idev(pd->device);
833 /* A multicast address requires a GRH (see ch. 8.4.1). */
834 if (ah_attr->dlid >= IPATH_MULTICAST_LID_BASE &&
835 ah_attr->dlid != IPATH_PERMISSIVE_LID &&
836 !(ah_attr->ah_flags & IB_AH_GRH)) {
837 ret = ERR_PTR(-EINVAL);
841 if (ah_attr->dlid == 0) {
842 ret = ERR_PTR(-EINVAL);
846 if (ah_attr->port_num < 1 ||
847 ah_attr->port_num > pd->device->phys_port_cnt) {
848 ret = ERR_PTR(-EINVAL);
852 ah = kmalloc(sizeof *ah, GFP_ATOMIC);
854 ret = ERR_PTR(-ENOMEM);
858 spin_lock(&dev->n_ahs_lock);
859 if (dev->n_ahs_allocated == ib_ipath_max_ahs) {
860 spin_unlock(&dev->n_ahs_lock);
862 ret = ERR_PTR(-ENOMEM);
866 dev->n_ahs_allocated++;
867 spin_unlock(&dev->n_ahs_lock);
869 /* ib_create_ah() will initialize ah->ibah. */
879 * ipath_destroy_ah - destroy an address handle
880 * @ibah: the AH to destroy
882 * This may be called from interrupt context.
884 static int ipath_destroy_ah(struct ib_ah *ibah)
886 struct ipath_ibdev *dev = to_idev(ibah->device);
887 struct ipath_ah *ah = to_iah(ibah);
889 spin_lock(&dev->n_ahs_lock);
890 dev->n_ahs_allocated--;
891 spin_unlock(&dev->n_ahs_lock);
898 static int ipath_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr)
900 struct ipath_ah *ah = to_iah(ibah);
907 static int ipath_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
910 struct ipath_ibdev *dev = to_idev(ibdev);
913 if (index >= ipath_layer_get_npkeys(dev->dd)) {
918 *pkey = ipath_layer_get_pkey(dev->dd, index);
927 * ipath_alloc_ucontext - allocate a ucontest
928 * @ibdev: the infiniband device
929 * @udata: not used by the InfiniPath driver
932 static struct ib_ucontext *ipath_alloc_ucontext(struct ib_device *ibdev,
933 struct ib_udata *udata)
935 struct ipath_ucontext *context;
936 struct ib_ucontext *ret;
938 context = kmalloc(sizeof *context, GFP_KERNEL);
940 ret = ERR_PTR(-ENOMEM);
944 ret = &context->ibucontext;
950 static int ipath_dealloc_ucontext(struct ib_ucontext *context)
952 kfree(to_iucontext(context));
956 static int ipath_verbs_register_sysfs(struct ib_device *dev);
959 * ipath_register_ib_device - register our device with the infiniband core
960 * @unit: the device number to register
961 * @dd: the device data structure
962 * Return the allocated ipath_ibdev pointer or NULL on error.
964 static void *ipath_register_ib_device(int unit, struct ipath_devdata *dd)
966 struct ipath_layer_counters cntrs;
967 struct ipath_ibdev *idev;
968 struct ib_device *dev;
971 idev = (struct ipath_ibdev *)ib_alloc_device(sizeof *idev);
977 /* Only need to initialize non-zero fields. */
978 spin_lock_init(&idev->n_pds_lock);
979 spin_lock_init(&idev->n_ahs_lock);
980 spin_lock_init(&idev->n_cqs_lock);
981 spin_lock_init(&idev->n_srqs_lock);
982 spin_lock_init(&idev->n_mcast_grps_lock);
984 spin_lock_init(&idev->qp_table.lock);
985 spin_lock_init(&idev->lk_table.lock);
986 idev->sm_lid = __constant_be16_to_cpu(IB_LID_PERMISSIVE);
987 /* Set the prefix to the default value (see ch. 4.1.1) */
988 idev->gid_prefix = __constant_cpu_to_be64(0xfe80000000000000ULL);
990 ret = ipath_init_qp_table(idev, ib_ipath_qp_table_size);
995 * The top ib_ipath_lkey_table_size bits are used to index the
996 * table. The lower 8 bits can be owned by the user (copied from
997 * the LKEY). The remaining bits act as a generation number or tag.
999 idev->lk_table.max = 1 << ib_ipath_lkey_table_size;
1000 idev->lk_table.table = kzalloc(idev->lk_table.max *
1001 sizeof(*idev->lk_table.table),
1003 if (idev->lk_table.table == NULL) {
1007 spin_lock_init(&idev->pending_lock);
1008 INIT_LIST_HEAD(&idev->pending[0]);
1009 INIT_LIST_HEAD(&idev->pending[1]);
1010 INIT_LIST_HEAD(&idev->pending[2]);
1011 INIT_LIST_HEAD(&idev->piowait);
1012 INIT_LIST_HEAD(&idev->rnrwait);
1013 idev->pending_index = 0;
1014 idev->port_cap_flags =
1015 IB_PORT_SYS_IMAGE_GUID_SUP | IB_PORT_CLIENT_REG_SUP;
1016 idev->pma_counter_select[0] = IB_PMA_PORT_XMIT_DATA;
1017 idev->pma_counter_select[1] = IB_PMA_PORT_RCV_DATA;
1018 idev->pma_counter_select[2] = IB_PMA_PORT_XMIT_PKTS;
1019 idev->pma_counter_select[3] = IB_PMA_PORT_RCV_PKTS;
1020 idev->pma_counter_select[5] = IB_PMA_PORT_XMIT_WAIT;
1021 idev->link_width_enabled = 3; /* 1x or 4x */
1023 /* Snapshot current HW counters to "clear" them. */
1024 ipath_layer_get_counters(dd, &cntrs);
1025 idev->z_symbol_error_counter = cntrs.symbol_error_counter;
1026 idev->z_link_error_recovery_counter =
1027 cntrs.link_error_recovery_counter;
1028 idev->z_link_downed_counter = cntrs.link_downed_counter;
1029 idev->z_port_rcv_errors = cntrs.port_rcv_errors;
1030 idev->z_port_rcv_remphys_errors =
1031 cntrs.port_rcv_remphys_errors;
1032 idev->z_port_xmit_discards = cntrs.port_xmit_discards;
1033 idev->z_port_xmit_data = cntrs.port_xmit_data;
1034 idev->z_port_rcv_data = cntrs.port_rcv_data;
1035 idev->z_port_xmit_packets = cntrs.port_xmit_packets;
1036 idev->z_port_rcv_packets = cntrs.port_rcv_packets;
1037 idev->z_local_link_integrity_errors =
1038 cntrs.local_link_integrity_errors;
1039 idev->z_excessive_buffer_overrun_errors =
1040 cntrs.excessive_buffer_overrun_errors;
1043 * The system image GUID is supposed to be the same for all
1044 * IB HCAs in a single system but since there can be other
1045 * device types in the system, we can't be sure this is unique.
1047 if (!sys_image_guid)
1048 sys_image_guid = ipath_layer_get_guid(dd);
1049 idev->sys_image_guid = sys_image_guid;
1050 idev->ib_unit = unit;
1053 strlcpy(dev->name, "ipath%d", IB_DEVICE_NAME_MAX);
1054 dev->owner = THIS_MODULE;
1055 dev->node_guid = ipath_layer_get_guid(dd);
1056 dev->uverbs_abi_ver = IPATH_UVERBS_ABI_VERSION;
1057 dev->uverbs_cmd_mask =
1058 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1059 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1060 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1061 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1062 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1063 (1ull << IB_USER_VERBS_CMD_CREATE_AH) |
1064 (1ull << IB_USER_VERBS_CMD_DESTROY_AH) |
1065 (1ull << IB_USER_VERBS_CMD_QUERY_AH) |
1066 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1067 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1068 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1069 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1070 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
1071 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1072 (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1073 (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1074 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1075 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
1076 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1077 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1078 (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1079 (1ull << IB_USER_VERBS_CMD_POST_RECV) |
1080 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1081 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
1082 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1083 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1084 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1085 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
1086 (1ull << IB_USER_VERBS_CMD_POST_SRQ_RECV);
1087 dev->node_type = IB_NODE_CA;
1088 dev->phys_port_cnt = 1;
1089 dev->dma_device = ipath_layer_get_device(dd);
1090 dev->class_dev.dev = dev->dma_device;
1091 dev->query_device = ipath_query_device;
1092 dev->modify_device = ipath_modify_device;
1093 dev->query_port = ipath_query_port;
1094 dev->modify_port = ipath_modify_port;
1095 dev->query_pkey = ipath_query_pkey;
1096 dev->query_gid = ipath_query_gid;
1097 dev->alloc_ucontext = ipath_alloc_ucontext;
1098 dev->dealloc_ucontext = ipath_dealloc_ucontext;
1099 dev->alloc_pd = ipath_alloc_pd;
1100 dev->dealloc_pd = ipath_dealloc_pd;
1101 dev->create_ah = ipath_create_ah;
1102 dev->destroy_ah = ipath_destroy_ah;
1103 dev->query_ah = ipath_query_ah;
1104 dev->create_srq = ipath_create_srq;
1105 dev->modify_srq = ipath_modify_srq;
1106 dev->query_srq = ipath_query_srq;
1107 dev->destroy_srq = ipath_destroy_srq;
1108 dev->create_qp = ipath_create_qp;
1109 dev->modify_qp = ipath_modify_qp;
1110 dev->query_qp = ipath_query_qp;
1111 dev->destroy_qp = ipath_destroy_qp;
1112 dev->post_send = ipath_post_send;
1113 dev->post_recv = ipath_post_receive;
1114 dev->post_srq_recv = ipath_post_srq_receive;
1115 dev->create_cq = ipath_create_cq;
1116 dev->destroy_cq = ipath_destroy_cq;
1117 dev->resize_cq = ipath_resize_cq;
1118 dev->poll_cq = ipath_poll_cq;
1119 dev->req_notify_cq = ipath_req_notify_cq;
1120 dev->get_dma_mr = ipath_get_dma_mr;
1121 dev->reg_phys_mr = ipath_reg_phys_mr;
1122 dev->reg_user_mr = ipath_reg_user_mr;
1123 dev->dereg_mr = ipath_dereg_mr;
1124 dev->alloc_fmr = ipath_alloc_fmr;
1125 dev->map_phys_fmr = ipath_map_phys_fmr;
1126 dev->unmap_fmr = ipath_unmap_fmr;
1127 dev->dealloc_fmr = ipath_dealloc_fmr;
1128 dev->attach_mcast = ipath_multicast_attach;
1129 dev->detach_mcast = ipath_multicast_detach;
1130 dev->process_mad = ipath_process_mad;
1131 dev->mmap = ipath_mmap;
1133 snprintf(dev->node_desc, sizeof(dev->node_desc),
1134 IPATH_IDSTR " %s kernel_SMA", system_utsname.nodename);
1136 ret = ib_register_device(dev);
1140 if (ipath_verbs_register_sysfs(dev))
1143 ipath_layer_enable_timer(dd);
1148 ib_unregister_device(dev);
1150 kfree(idev->lk_table.table);
1152 kfree(idev->qp_table.table);
1154 ib_dealloc_device(dev);
1155 _VERBS_ERROR("ib_ipath%d cannot register verbs (%d)!\n",
1163 static void ipath_unregister_ib_device(void *arg)
1165 struct ipath_ibdev *dev = (struct ipath_ibdev *) arg;
1166 struct ib_device *ibdev = &dev->ibdev;
1168 ipath_layer_disable_timer(dev->dd);
1170 ib_unregister_device(ibdev);
1172 if (!list_empty(&dev->pending[0]) ||
1173 !list_empty(&dev->pending[1]) ||
1174 !list_empty(&dev->pending[2]))
1175 _VERBS_ERROR("ipath%d pending list not empty!\n",
1177 if (!list_empty(&dev->piowait))
1178 _VERBS_ERROR("ipath%d piowait list not empty!\n",
1180 if (!list_empty(&dev->rnrwait))
1181 _VERBS_ERROR("ipath%d rnrwait list not empty!\n",
1183 if (!ipath_mcast_tree_empty())
1184 _VERBS_ERROR("ipath%d multicast table memory leak!\n",
1187 * Note that ipath_unregister_ib_device() can be called before all
1188 * the QPs are destroyed!
1190 ipath_free_all_qps(&dev->qp_table);
1191 kfree(dev->qp_table.table);
1192 kfree(dev->lk_table.table);
1193 ib_dealloc_device(ibdev);
1196 static int __init ipath_verbs_init(void)
1198 return ipath_verbs_register(ipath_register_ib_device,
1199 ipath_unregister_ib_device,
1200 ipath_ib_piobufavail, ipath_ib_rcv,
1204 static void __exit ipath_verbs_cleanup(void)
1206 ipath_verbs_unregister();
1209 static ssize_t show_rev(struct class_device *cdev, char *buf)
1211 struct ipath_ibdev *dev =
1212 container_of(cdev, struct ipath_ibdev, ibdev.class_dev);
1214 return sprintf(buf, "%x\n", ipath_layer_get_pcirev(dev->dd));
1217 static ssize_t show_hca(struct class_device *cdev, char *buf)
1219 struct ipath_ibdev *dev =
1220 container_of(cdev, struct ipath_ibdev, ibdev.class_dev);
1223 ret = ipath_layer_get_boardname(dev->dd, buf, 128);
1233 static ssize_t show_stats(struct class_device *cdev, char *buf)
1235 struct ipath_ibdev *dev =
1236 container_of(cdev, struct ipath_ibdev, ibdev.class_dev);
1254 dev->n_rc_resends, dev->n_rc_qacks, dev->n_rc_acks,
1255 dev->n_seq_naks, dev->n_rdma_seq, dev->n_rnr_naks,
1256 dev->n_other_naks, dev->n_timeouts,
1257 dev->n_rdma_dup_busy, dev->n_piowait,
1258 dev->n_no_piobuf, dev->n_pkt_drops, dev->n_wqe_errs);
1259 for (i = 0; i < ARRAY_SIZE(dev->opstats); i++) {
1260 const struct ipath_opcode_stats *si = &dev->opstats[i];
1262 if (!si->n_packets && !si->n_bytes)
1264 len += sprintf(buf + len, "%02x %llu/%llu\n", i,
1265 (unsigned long long) si->n_packets,
1266 (unsigned long long) si->n_bytes);
1271 static CLASS_DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1272 static CLASS_DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1273 static CLASS_DEVICE_ATTR(board_id, S_IRUGO, show_hca, NULL);
1274 static CLASS_DEVICE_ATTR(stats, S_IRUGO, show_stats, NULL);
1276 static struct class_device_attribute *ipath_class_attributes[] = {
1277 &class_device_attr_hw_rev,
1278 &class_device_attr_hca_type,
1279 &class_device_attr_board_id,
1280 &class_device_attr_stats
1283 static int ipath_verbs_register_sysfs(struct ib_device *dev)
1288 for (i = 0; i < ARRAY_SIZE(ipath_class_attributes); ++i)
1289 if (class_device_create_file(&dev->class_dev,
1290 ipath_class_attributes[i])) {
1301 module_init(ipath_verbs_init);
1302 module_exit(ipath_verbs_cleanup);