2 * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 static int destroy_cq(struct c4iw_rdev *rdev, struct t4_cq *cq,
36 struct c4iw_dev_ucontext *uctx)
38 struct fw_ri_res_wr *res_wr;
39 struct fw_ri_res *res;
41 struct c4iw_wr_wait wr_wait;
45 wr_len = sizeof *res_wr + sizeof *res;
46 skb = alloc_skb(wr_len, GFP_KERNEL);
49 set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
51 res_wr = (struct fw_ri_res_wr *)__skb_put(skb, wr_len);
52 memset(res_wr, 0, wr_len);
53 res_wr->op_nres = cpu_to_be32(
54 FW_WR_OP_V(FW_RI_RES_WR) |
55 FW_RI_RES_WR_NRES_V(1) |
57 res_wr->len16_pkd = cpu_to_be32(DIV_ROUND_UP(wr_len, 16));
58 res_wr->cookie = (uintptr_t)&wr_wait;
60 res->u.cq.restype = FW_RI_RES_TYPE_CQ;
61 res->u.cq.op = FW_RI_RES_OP_RESET;
62 res->u.cq.iqid = cpu_to_be32(cq->cqid);
64 c4iw_init_wr_wait(&wr_wait);
65 ret = c4iw_ofld_send(rdev, skb);
67 ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, 0, __func__);
71 dma_free_coherent(&(rdev->lldi.pdev->dev),
72 cq->memsize, cq->queue,
73 dma_unmap_addr(cq, mapping));
74 c4iw_put_cqid(rdev, cq->cqid, uctx);
78 static int create_cq(struct c4iw_rdev *rdev, struct t4_cq *cq,
79 struct c4iw_dev_ucontext *uctx)
81 struct fw_ri_res_wr *res_wr;
82 struct fw_ri_res *res;
84 int user = (uctx != &rdev->uctx);
85 struct c4iw_wr_wait wr_wait;
89 cq->cqid = c4iw_get_cqid(rdev, uctx);
96 cq->sw_queue = kzalloc(cq->memsize, GFP_KERNEL);
102 cq->queue = dma_alloc_coherent(&rdev->lldi.pdev->dev, cq->memsize,
103 &cq->dma_addr, GFP_KERNEL);
108 dma_unmap_addr_set(cq, mapping, cq->dma_addr);
109 memset(cq->queue, 0, cq->memsize);
111 /* build fw_ri_res_wr */
112 wr_len = sizeof *res_wr + sizeof *res;
114 skb = alloc_skb(wr_len, GFP_KERNEL);
119 set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
121 res_wr = (struct fw_ri_res_wr *)__skb_put(skb, wr_len);
122 memset(res_wr, 0, wr_len);
123 res_wr->op_nres = cpu_to_be32(
124 FW_WR_OP_V(FW_RI_RES_WR) |
125 FW_RI_RES_WR_NRES_V(1) |
127 res_wr->len16_pkd = cpu_to_be32(DIV_ROUND_UP(wr_len, 16));
128 res_wr->cookie = (uintptr_t)&wr_wait;
130 res->u.cq.restype = FW_RI_RES_TYPE_CQ;
131 res->u.cq.op = FW_RI_RES_OP_WRITE;
132 res->u.cq.iqid = cpu_to_be32(cq->cqid);
133 res->u.cq.iqandst_to_iqandstindex = cpu_to_be32(
134 FW_RI_RES_WR_IQANUS_V(0) |
135 FW_RI_RES_WR_IQANUD_V(1) |
136 FW_RI_RES_WR_IQANDST_F |
137 FW_RI_RES_WR_IQANDSTINDEX_V(
138 rdev->lldi.ciq_ids[cq->vector]));
139 res->u.cq.iqdroprss_to_iqesize = cpu_to_be16(
140 FW_RI_RES_WR_IQDROPRSS_F |
141 FW_RI_RES_WR_IQPCIECH_V(2) |
142 FW_RI_RES_WR_IQINTCNTTHRESH_V(0) |
144 FW_RI_RES_WR_IQESIZE_V(1));
145 res->u.cq.iqsize = cpu_to_be16(cq->size);
146 res->u.cq.iqaddr = cpu_to_be64(cq->dma_addr);
148 c4iw_init_wr_wait(&wr_wait);
150 ret = c4iw_ofld_send(rdev, skb);
153 PDBG("%s wait_event wr_wait %p\n", __func__, &wr_wait);
154 ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, 0, __func__);
159 cq->gts = rdev->lldi.gts_reg;
162 cq->bar2_va = c4iw_bar2_addrs(rdev, cq->cqid, T4_BAR2_QTYPE_INGRESS,
164 user ? &cq->bar2_pa : NULL);
165 if (user && !cq->bar2_va) {
166 pr_warn(MOD "%s: cqid %u not in BAR2 range.\n",
167 pci_name(rdev->lldi.pdev), cq->cqid);
173 dma_free_coherent(&rdev->lldi.pdev->dev, cq->memsize, cq->queue,
174 dma_unmap_addr(cq, mapping));
178 c4iw_put_cqid(rdev, cq->cqid, uctx);
183 static void insert_recv_cqe(struct t4_wq *wq, struct t4_cq *cq)
187 PDBG("%s wq %p cq %p sw_cidx %u sw_pidx %u\n", __func__,
188 wq, cq, cq->sw_cidx, cq->sw_pidx);
189 memset(&cqe, 0, sizeof(cqe));
190 cqe.header = cpu_to_be32(CQE_STATUS_V(T4_ERR_SWFLUSH) |
191 CQE_OPCODE_V(FW_RI_SEND) |
194 CQE_QPID_V(wq->sq.qid));
195 cqe.bits_type_ts = cpu_to_be64(CQE_GENBIT_V((u64)cq->gen));
196 cq->sw_queue[cq->sw_pidx] = cqe;
200 int c4iw_flush_rq(struct t4_wq *wq, struct t4_cq *cq, int count)
203 int in_use = wq->rq.in_use - count;
206 PDBG("%s wq %p cq %p rq.in_use %u skip count %u\n", __func__,
207 wq, cq, wq->rq.in_use, count);
209 insert_recv_cqe(wq, cq);
215 static void insert_sq_cqe(struct t4_wq *wq, struct t4_cq *cq,
216 struct t4_swsqe *swcqe)
220 PDBG("%s wq %p cq %p sw_cidx %u sw_pidx %u\n", __func__,
221 wq, cq, cq->sw_cidx, cq->sw_pidx);
222 memset(&cqe, 0, sizeof(cqe));
223 cqe.header = cpu_to_be32(CQE_STATUS_V(T4_ERR_SWFLUSH) |
224 CQE_OPCODE_V(swcqe->opcode) |
227 CQE_QPID_V(wq->sq.qid));
228 CQE_WRID_SQ_IDX(&cqe) = swcqe->idx;
229 cqe.bits_type_ts = cpu_to_be64(CQE_GENBIT_V((u64)cq->gen));
230 cq->sw_queue[cq->sw_pidx] = cqe;
234 static void advance_oldest_read(struct t4_wq *wq);
236 int c4iw_flush_sq(struct c4iw_qp *qhp)
239 struct t4_wq *wq = &qhp->wq;
240 struct c4iw_cq *chp = to_c4iw_cq(qhp->ibqp.send_cq);
241 struct t4_cq *cq = &chp->cq;
243 struct t4_swsqe *swsqe;
245 if (wq->sq.flush_cidx == -1)
246 wq->sq.flush_cidx = wq->sq.cidx;
247 idx = wq->sq.flush_cidx;
248 BUG_ON(idx >= wq->sq.size);
249 while (idx != wq->sq.pidx) {
250 swsqe = &wq->sq.sw_sq[idx];
251 BUG_ON(swsqe->flushed);
253 insert_sq_cqe(wq, cq, swsqe);
254 if (wq->sq.oldest_read == swsqe) {
255 BUG_ON(swsqe->opcode != FW_RI_READ_REQ);
256 advance_oldest_read(wq);
259 if (++idx == wq->sq.size)
262 wq->sq.flush_cidx += flushed;
263 if (wq->sq.flush_cidx >= wq->sq.size)
264 wq->sq.flush_cidx -= wq->sq.size;
268 static void flush_completed_wrs(struct t4_wq *wq, struct t4_cq *cq)
270 struct t4_swsqe *swsqe;
273 if (wq->sq.flush_cidx == -1)
274 wq->sq.flush_cidx = wq->sq.cidx;
275 cidx = wq->sq.flush_cidx;
276 BUG_ON(cidx > wq->sq.size);
278 while (cidx != wq->sq.pidx) {
279 swsqe = &wq->sq.sw_sq[cidx];
280 if (!swsqe->signaled) {
281 if (++cidx == wq->sq.size)
283 } else if (swsqe->complete) {
285 BUG_ON(swsqe->flushed);
288 * Insert this completed cqe into the swcq.
290 PDBG("%s moving cqe into swcq sq idx %u cq idx %u\n",
291 __func__, cidx, cq->sw_pidx);
292 swsqe->cqe.header |= htonl(CQE_SWCQE_V(1));
293 cq->sw_queue[cq->sw_pidx] = swsqe->cqe;
296 if (++cidx == wq->sq.size)
298 wq->sq.flush_cidx = cidx;
304 static void create_read_req_cqe(struct t4_wq *wq, struct t4_cqe *hw_cqe,
305 struct t4_cqe *read_cqe)
307 read_cqe->u.scqe.cidx = wq->sq.oldest_read->idx;
308 read_cqe->len = htonl(wq->sq.oldest_read->read_len);
309 read_cqe->header = htonl(CQE_QPID_V(CQE_QPID(hw_cqe)) |
310 CQE_SWCQE_V(SW_CQE(hw_cqe)) |
311 CQE_OPCODE_V(FW_RI_READ_REQ) |
313 read_cqe->bits_type_ts = hw_cqe->bits_type_ts;
316 static void advance_oldest_read(struct t4_wq *wq)
319 u32 rptr = wq->sq.oldest_read - wq->sq.sw_sq + 1;
321 if (rptr == wq->sq.size)
323 while (rptr != wq->sq.pidx) {
324 wq->sq.oldest_read = &wq->sq.sw_sq[rptr];
326 if (wq->sq.oldest_read->opcode == FW_RI_READ_REQ)
328 if (++rptr == wq->sq.size)
331 wq->sq.oldest_read = NULL;
335 * Move all CQEs from the HWCQ into the SWCQ.
336 * Deal with out-of-order and/or completions that complete
337 * prior unsignalled WRs.
339 void c4iw_flush_hw_cq(struct c4iw_cq *chp)
341 struct t4_cqe *hw_cqe, *swcqe, read_cqe;
343 struct t4_swsqe *swsqe;
346 PDBG("%s cqid 0x%x\n", __func__, chp->cq.cqid);
347 ret = t4_next_hw_cqe(&chp->cq, &hw_cqe);
350 * This logic is similar to poll_cq(), but not quite the same
351 * unfortunately. Need to move pertinent HW CQEs to the SW CQ but
352 * also do any translation magic that poll_cq() normally does.
355 qhp = get_qhp(chp->rhp, CQE_QPID(hw_cqe));
358 * drop CQEs with no associated QP
363 if (CQE_OPCODE(hw_cqe) == FW_RI_TERMINATE)
366 if (CQE_OPCODE(hw_cqe) == FW_RI_READ_RESP) {
368 /* If we have reached here because of async
369 * event or other error, and have egress error
372 if (CQE_TYPE(hw_cqe) == 1)
375 /* drop peer2peer RTR reads.
377 if (CQE_WRID_STAG(hw_cqe) == 1)
381 * Eat completions for unsignaled read WRs.
383 if (!qhp->wq.sq.oldest_read->signaled) {
384 advance_oldest_read(&qhp->wq);
389 * Don't write to the HWCQ, create a new read req CQE
390 * in local memory and move it into the swcq.
392 create_read_req_cqe(&qhp->wq, hw_cqe, &read_cqe);
394 advance_oldest_read(&qhp->wq);
397 /* if its a SQ completion, then do the magic to move all the
398 * unsignaled and now in-order completions into the swcq.
400 if (SQ_TYPE(hw_cqe)) {
401 swsqe = &qhp->wq.sq.sw_sq[CQE_WRID_SQ_IDX(hw_cqe)];
402 swsqe->cqe = *hw_cqe;
404 flush_completed_wrs(&qhp->wq, &chp->cq);
406 swcqe = &chp->cq.sw_queue[chp->cq.sw_pidx];
408 swcqe->header |= cpu_to_be32(CQE_SWCQE_V(1));
409 t4_swcq_produce(&chp->cq);
412 t4_hwcq_consume(&chp->cq);
413 ret = t4_next_hw_cqe(&chp->cq, &hw_cqe);
417 static int cqe_completes_wr(struct t4_cqe *cqe, struct t4_wq *wq)
419 if (CQE_OPCODE(cqe) == FW_RI_TERMINATE)
422 if ((CQE_OPCODE(cqe) == FW_RI_RDMA_WRITE) && RQ_TYPE(cqe))
425 if ((CQE_OPCODE(cqe) == FW_RI_READ_RESP) && SQ_TYPE(cqe))
428 if (CQE_SEND_OPCODE(cqe) && RQ_TYPE(cqe) && t4_rq_empty(wq))
433 void c4iw_count_rcqes(struct t4_cq *cq, struct t4_wq *wq, int *count)
439 PDBG("%s count zero %d\n", __func__, *count);
441 while (ptr != cq->sw_pidx) {
442 cqe = &cq->sw_queue[ptr];
443 if (RQ_TYPE(cqe) && (CQE_OPCODE(cqe) != FW_RI_READ_RESP) &&
444 (CQE_QPID(cqe) == wq->sq.qid) && cqe_completes_wr(cqe, wq))
446 if (++ptr == cq->size)
449 PDBG("%s cq %p count %d\n", __func__, cq, *count);
456 * check the validity of the first CQE,
457 * supply the wq assicated with the qpid.
459 * credit: cq credit to return to sge.
460 * cqe_flushed: 1 iff the CQE is flushed.
461 * cqe: copy of the polled CQE.
465 * -EAGAIN CQE skipped, try again.
466 * -EOVERFLOW CQ overflow detected.
468 static int poll_cq(struct t4_wq *wq, struct t4_cq *cq, struct t4_cqe *cqe,
469 u8 *cqe_flushed, u64 *cookie, u32 *credit)
472 struct t4_cqe *hw_cqe, read_cqe;
476 ret = t4_next_cqe(cq, &hw_cqe);
480 PDBG("%s CQE OVF %u qpid 0x%0x genbit %u type %u status 0x%0x"
481 " opcode 0x%0x len 0x%0x wrid_hi_stag 0x%x wrid_low_msn 0x%x\n",
482 __func__, CQE_OVFBIT(hw_cqe), CQE_QPID(hw_cqe),
483 CQE_GENBIT(hw_cqe), CQE_TYPE(hw_cqe), CQE_STATUS(hw_cqe),
484 CQE_OPCODE(hw_cqe), CQE_LEN(hw_cqe), CQE_WRID_HI(hw_cqe),
485 CQE_WRID_LOW(hw_cqe));
488 * skip cqe's not affiliated with a QP.
496 * skip hw cqe's if the wq is flushed.
498 if (wq->flushed && !SW_CQE(hw_cqe)) {
504 * skip TERMINATE cqes...
506 if (CQE_OPCODE(hw_cqe) == FW_RI_TERMINATE) {
512 * Gotta tweak READ completions:
513 * 1) the cqe doesn't contain the sq_wptr from the wr.
514 * 2) opcode not reflected from the wr.
515 * 3) read_len not reflected from the wr.
516 * 4) cq_type is RQ_TYPE not SQ_TYPE.
518 if (RQ_TYPE(hw_cqe) && (CQE_OPCODE(hw_cqe) == FW_RI_READ_RESP)) {
520 /* If we have reached here because of async
521 * event or other error, and have egress error
524 if (CQE_TYPE(hw_cqe) == 1) {
525 if (CQE_STATUS(hw_cqe))
526 t4_set_wq_in_error(wq);
531 /* If this is an unsolicited read response, then the read
532 * was generated by the kernel driver as part of peer-2-peer
533 * connection setup. So ignore the completion.
535 if (CQE_WRID_STAG(hw_cqe) == 1) {
536 if (CQE_STATUS(hw_cqe))
537 t4_set_wq_in_error(wq);
543 * Eat completions for unsignaled read WRs.
545 if (!wq->sq.oldest_read->signaled) {
546 advance_oldest_read(wq);
552 * Don't write to the HWCQ, so create a new read req CQE
555 create_read_req_cqe(wq, hw_cqe, &read_cqe);
557 advance_oldest_read(wq);
560 if (CQE_STATUS(hw_cqe) || t4_wq_in_error(wq)) {
561 *cqe_flushed = (CQE_STATUS(hw_cqe) == T4_ERR_SWFLUSH);
562 t4_set_wq_in_error(wq);
568 if (RQ_TYPE(hw_cqe)) {
571 * HW only validates 4 bits of MSN. So we must validate that
572 * the MSN in the SEND is the next expected MSN. If its not,
573 * then we complete this with T4_ERR_MSN and mark the wq in
577 if (t4_rq_empty(wq)) {
578 t4_set_wq_in_error(wq);
582 if (unlikely((CQE_WRID_MSN(hw_cqe) != (wq->rq.msn)))) {
583 t4_set_wq_in_error(wq);
584 hw_cqe->header |= htonl(CQE_STATUS_V(T4_ERR_MSN));
591 * If we get here its a send completion.
593 * Handle out of order completion. These get stuffed
594 * in the SW SQ. Then the SW SQ is walked to move any
595 * now in-order completions into the SW CQ. This handles
597 * 1) reaping unsignaled WRs when the first subsequent
598 * signaled WR is completed.
599 * 2) out of order read completions.
601 if (!SW_CQE(hw_cqe) && (CQE_WRID_SQ_IDX(hw_cqe) != wq->sq.cidx)) {
602 struct t4_swsqe *swsqe;
604 PDBG("%s out of order completion going in sw_sq at idx %u\n",
605 __func__, CQE_WRID_SQ_IDX(hw_cqe));
606 swsqe = &wq->sq.sw_sq[CQE_WRID_SQ_IDX(hw_cqe)];
607 swsqe->cqe = *hw_cqe;
617 * Reap the associated WR(s) that are freed up with this
620 if (SQ_TYPE(hw_cqe)) {
621 int idx = CQE_WRID_SQ_IDX(hw_cqe);
622 BUG_ON(idx >= wq->sq.size);
625 * Account for any unsignaled completions completed by
626 * this signaled completion. In this case, cidx points
627 * to the first unsignaled one, and idx points to the
628 * signaled one. So adjust in_use based on this delta.
629 * if this is not completing any unsigned wrs, then the
630 * delta will be 0. Handle wrapping also!
632 if (idx < wq->sq.cidx)
633 wq->sq.in_use -= wq->sq.size + idx - wq->sq.cidx;
635 wq->sq.in_use -= idx - wq->sq.cidx;
636 BUG_ON(wq->sq.in_use <= 0 && wq->sq.in_use >= wq->sq.size);
638 wq->sq.cidx = (uint16_t)idx;
639 PDBG("%s completing sq idx %u\n", __func__, wq->sq.cidx);
640 *cookie = wq->sq.sw_sq[wq->sq.cidx].wr_id;
642 c4iw_log_wr_stats(wq, hw_cqe);
645 PDBG("%s completing rq idx %u\n", __func__, wq->rq.cidx);
646 *cookie = wq->rq.sw_rq[wq->rq.cidx].wr_id;
647 BUG_ON(t4_rq_empty(wq));
649 c4iw_log_wr_stats(wq, hw_cqe);
656 * Flush any completed cqes that are now in-order.
658 flush_completed_wrs(wq, cq);
661 if (SW_CQE(hw_cqe)) {
662 PDBG("%s cq %p cqid 0x%x skip sw cqe cidx %u\n",
663 __func__, cq, cq->cqid, cq->sw_cidx);
666 PDBG("%s cq %p cqid 0x%x skip hw cqe cidx %u\n",
667 __func__, cq, cq->cqid, cq->cidx);
674 * Get one cq entry from c4iw and map it to openib.
679 * -EAGAIN caller must try again
680 * any other -errno fatal error
682 static int c4iw_poll_cq_one(struct c4iw_cq *chp, struct ib_wc *wc)
684 struct c4iw_qp *qhp = NULL;
685 struct t4_cqe uninitialized_var(cqe), *rd_cqe;
692 ret = t4_next_cqe(&chp->cq, &rd_cqe);
697 qhp = get_qhp(chp->rhp, CQE_QPID(rd_cqe));
701 spin_lock(&qhp->lock);
704 ret = poll_cq(wq, &(chp->cq), &cqe, &cqe_flushed, &cookie, &credit);
710 wc->vendor_err = CQE_STATUS(&cqe);
713 PDBG("%s qpid 0x%x type %d opcode %d status 0x%x len %u wrid hi 0x%x "
714 "lo 0x%x cookie 0x%llx\n", __func__, CQE_QPID(&cqe),
715 CQE_TYPE(&cqe), CQE_OPCODE(&cqe), CQE_STATUS(&cqe), CQE_LEN(&cqe),
716 CQE_WRID_HI(&cqe), CQE_WRID_LOW(&cqe), (unsigned long long)cookie);
718 if (CQE_TYPE(&cqe) == 0) {
719 if (!CQE_STATUS(&cqe))
720 wc->byte_len = CQE_LEN(&cqe);
723 wc->opcode = IB_WC_RECV;
724 if (CQE_OPCODE(&cqe) == FW_RI_SEND_WITH_INV ||
725 CQE_OPCODE(&cqe) == FW_RI_SEND_WITH_SE_INV) {
726 wc->ex.invalidate_rkey = CQE_WRID_STAG(&cqe);
727 wc->wc_flags |= IB_WC_WITH_INVALIDATE;
730 switch (CQE_OPCODE(&cqe)) {
731 case FW_RI_RDMA_WRITE:
732 wc->opcode = IB_WC_RDMA_WRITE;
735 wc->opcode = IB_WC_RDMA_READ;
736 wc->byte_len = CQE_LEN(&cqe);
738 case FW_RI_SEND_WITH_INV:
739 case FW_RI_SEND_WITH_SE_INV:
740 wc->opcode = IB_WC_SEND;
741 wc->wc_flags |= IB_WC_WITH_INVALIDATE;
744 case FW_RI_SEND_WITH_SE:
745 wc->opcode = IB_WC_SEND;
748 wc->opcode = IB_WC_BIND_MW;
751 case FW_RI_LOCAL_INV:
752 wc->opcode = IB_WC_LOCAL_INV;
754 case FW_RI_FAST_REGISTER:
755 wc->opcode = IB_WC_FAST_REG_MR;
758 printk(KERN_ERR MOD "Unexpected opcode %d "
759 "in the CQE received for QPID=0x%0x\n",
760 CQE_OPCODE(&cqe), CQE_QPID(&cqe));
767 wc->status = IB_WC_WR_FLUSH_ERR;
770 switch (CQE_STATUS(&cqe)) {
772 wc->status = IB_WC_SUCCESS;
775 wc->status = IB_WC_LOC_ACCESS_ERR;
778 wc->status = IB_WC_LOC_PROT_ERR;
782 wc->status = IB_WC_LOC_ACCESS_ERR;
785 wc->status = IB_WC_GENERAL_ERR;
788 wc->status = IB_WC_LOC_LEN_ERR;
790 case T4_ERR_INVALIDATE_SHARED_MR:
791 case T4_ERR_INVALIDATE_MR_WITH_MW_BOUND:
792 wc->status = IB_WC_MW_BIND_ERR;
796 case T4_ERR_PDU_LEN_ERR:
797 case T4_ERR_OUT_OF_RQE:
798 case T4_ERR_DDP_VERSION:
799 case T4_ERR_RDMA_VERSION:
800 case T4_ERR_DDP_QUEUE_NUM:
804 case T4_ERR_MSN_RANGE:
805 case T4_ERR_IRD_OVERFLOW:
807 case T4_ERR_INTERNAL_ERR:
808 wc->status = IB_WC_FATAL_ERR;
811 wc->status = IB_WC_WR_FLUSH_ERR;
815 "Unexpected cqe_status 0x%x for QPID=0x%0x\n",
816 CQE_STATUS(&cqe), CQE_QPID(&cqe));
822 spin_unlock(&qhp->lock);
826 int c4iw_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
833 chp = to_c4iw_cq(ibcq);
835 spin_lock_irqsave(&chp->lock, flags);
836 for (npolled = 0; npolled < num_entries; ++npolled) {
838 err = c4iw_poll_cq_one(chp, wc + npolled);
839 } while (err == -EAGAIN);
843 spin_unlock_irqrestore(&chp->lock, flags);
844 return !err || err == -ENODATA ? npolled : err;
847 int c4iw_destroy_cq(struct ib_cq *ib_cq)
850 struct c4iw_ucontext *ucontext;
852 PDBG("%s ib_cq %p\n", __func__, ib_cq);
853 chp = to_c4iw_cq(ib_cq);
855 remove_handle(chp->rhp, &chp->rhp->cqidr, chp->cq.cqid);
856 atomic_dec(&chp->refcnt);
857 wait_event(chp->wait, !atomic_read(&chp->refcnt));
859 ucontext = ib_cq->uobject ? to_c4iw_ucontext(ib_cq->uobject->context)
861 destroy_cq(&chp->rhp->rdev, &chp->cq,
862 ucontext ? &ucontext->uctx : &chp->cq.rdev->uctx);
867 struct ib_cq *c4iw_create_cq(struct ib_device *ibdev,
868 const struct ib_cq_init_attr *attr,
869 struct ib_ucontext *ib_context,
870 struct ib_udata *udata)
872 int entries = attr->cqe;
873 int vector = attr->comp_vector;
874 struct c4iw_dev *rhp;
876 struct c4iw_create_cq_resp uresp;
877 struct c4iw_ucontext *ucontext = NULL;
879 size_t memsize, hwentries;
880 struct c4iw_mm_entry *mm, *mm2;
882 PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
884 return ERR_PTR(-EINVAL);
886 rhp = to_c4iw_dev(ibdev);
888 if (vector >= rhp->rdev.lldi.nciq)
889 return ERR_PTR(-EINVAL);
891 chp = kzalloc(sizeof(*chp), GFP_KERNEL);
893 return ERR_PTR(-ENOMEM);
896 ucontext = to_c4iw_ucontext(ib_context);
898 /* account for the status page. */
901 /* IQ needs one extra entry to differentiate full vs empty. */
905 * entries must be multiple of 16 for HW.
907 entries = roundup(entries, 16);
910 * Make actual HW queue 2x to avoid cdix_inc overflows.
912 hwentries = min(entries * 2, rhp->rdev.hw_queue.t4_max_iq_size);
915 * Make HW queue at least 64 entries so GTS updates aren't too
921 memsize = hwentries * sizeof *chp->cq.queue;
924 * memsize must be a multiple of the page size if its a user cq.
927 memsize = roundup(memsize, PAGE_SIZE);
928 chp->cq.size = hwentries;
929 chp->cq.memsize = memsize;
930 chp->cq.vector = vector;
932 ret = create_cq(&rhp->rdev, &chp->cq,
933 ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
938 chp->cq.size--; /* status page */
939 chp->ibcq.cqe = entries - 2;
940 spin_lock_init(&chp->lock);
941 spin_lock_init(&chp->comp_handler_lock);
942 atomic_set(&chp->refcnt, 1);
943 init_waitqueue_head(&chp->wait);
944 ret = insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid);
949 mm = kmalloc(sizeof *mm, GFP_KERNEL);
952 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
956 uresp.qid_mask = rhp->rdev.cqmask;
957 uresp.cqid = chp->cq.cqid;
958 uresp.size = chp->cq.size;
959 uresp.memsize = chp->cq.memsize;
960 spin_lock(&ucontext->mmap_lock);
961 uresp.key = ucontext->key;
962 ucontext->key += PAGE_SIZE;
963 uresp.gts_key = ucontext->key;
964 ucontext->key += PAGE_SIZE;
965 spin_unlock(&ucontext->mmap_lock);
966 ret = ib_copy_to_udata(udata, &uresp,
967 sizeof(uresp) - sizeof(uresp.reserved));
972 mm->addr = virt_to_phys(chp->cq.queue);
973 mm->len = chp->cq.memsize;
974 insert_mmap(ucontext, mm);
976 mm2->key = uresp.gts_key;
977 mm2->addr = chp->cq.bar2_pa;
978 mm2->len = PAGE_SIZE;
979 insert_mmap(ucontext, mm2);
981 PDBG("%s cqid 0x%0x chp %p size %u memsize %zu, dma_addr 0x%0llx\n",
982 __func__, chp->cq.cqid, chp, chp->cq.size,
983 chp->cq.memsize, (unsigned long long) chp->cq.dma_addr);
990 remove_handle(rhp, &rhp->cqidr, chp->cq.cqid);
992 destroy_cq(&chp->rhp->rdev, &chp->cq,
993 ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
999 int c4iw_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
1004 int c4iw_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
1006 struct c4iw_cq *chp;
1010 chp = to_c4iw_cq(ibcq);
1011 spin_lock_irqsave(&chp->lock, flag);
1012 ret = t4_arm_cq(&chp->cq,
1013 (flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED);
1014 spin_unlock_irqrestore(&chp->lock, flag);
1015 if (ret && !(flags & IB_CQ_REPORT_MISSED_EVENTS))