Linux 3.11-rc3
[cris-mirror.git] / drivers / infiniband / hw / cxgb4 / cq.c
blob0f1607c8325a5bc8a03e3a5d5471d39ed5741e8b
1 /*
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
12 * conditions are met:
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
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
30 * SOFTWARE.
33 #include "iw_cxgb4.h"
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;
40 int wr_len;
41 struct c4iw_wr_wait wr_wait;
42 struct sk_buff *skb;
43 int ret;
45 wr_len = sizeof *res_wr + sizeof *res;
46 skb = alloc_skb(wr_len, GFP_KERNEL);
47 if (!skb)
48 return -ENOMEM;
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(FW_RI_RES_WR) |
55 V_FW_RI_RES_WR_NRES(1) |
56 FW_WR_COMPL(1));
57 res_wr->len16_pkd = cpu_to_be32(DIV_ROUND_UP(wr_len, 16));
58 res_wr->cookie = (unsigned long) &wr_wait;
59 res = res_wr->res;
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);
66 if (!ret) {
67 ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, 0, __func__);
70 kfree(cq->sw_queue);
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);
75 return ret;
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;
83 int wr_len;
84 int user = (uctx != &rdev->uctx);
85 struct c4iw_wr_wait wr_wait;
86 int ret;
87 struct sk_buff *skb;
89 cq->cqid = c4iw_get_cqid(rdev, uctx);
90 if (!cq->cqid) {
91 ret = -ENOMEM;
92 goto err1;
95 if (!user) {
96 cq->sw_queue = kzalloc(cq->memsize, GFP_KERNEL);
97 if (!cq->sw_queue) {
98 ret = -ENOMEM;
99 goto err2;
102 cq->queue = dma_alloc_coherent(&rdev->lldi.pdev->dev, cq->memsize,
103 &cq->dma_addr, GFP_KERNEL);
104 if (!cq->queue) {
105 ret = -ENOMEM;
106 goto err3;
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);
115 if (!skb) {
116 ret = -ENOMEM;
117 goto err4;
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(FW_RI_RES_WR) |
125 V_FW_RI_RES_WR_NRES(1) |
126 FW_WR_COMPL(1));
127 res_wr->len16_pkd = cpu_to_be32(DIV_ROUND_UP(wr_len, 16));
128 res_wr->cookie = (unsigned long) &wr_wait;
129 res = res_wr->res;
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 V_FW_RI_RES_WR_IQANUS(0) |
135 V_FW_RI_RES_WR_IQANUD(1) |
136 F_FW_RI_RES_WR_IQANDST |
137 V_FW_RI_RES_WR_IQANDSTINDEX(*rdev->lldi.rxq_ids));
138 res->u.cq.iqdroprss_to_iqesize = cpu_to_be16(
139 F_FW_RI_RES_WR_IQDROPRSS |
140 V_FW_RI_RES_WR_IQPCIECH(2) |
141 V_FW_RI_RES_WR_IQINTCNTTHRESH(0) |
142 F_FW_RI_RES_WR_IQO |
143 V_FW_RI_RES_WR_IQESIZE(1));
144 res->u.cq.iqsize = cpu_to_be16(cq->size);
145 res->u.cq.iqaddr = cpu_to_be64(cq->dma_addr);
147 c4iw_init_wr_wait(&wr_wait);
149 ret = c4iw_ofld_send(rdev, skb);
150 if (ret)
151 goto err4;
152 PDBG("%s wait_event wr_wait %p\n", __func__, &wr_wait);
153 ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, 0, __func__);
154 if (ret)
155 goto err4;
157 cq->gen = 1;
158 cq->gts = rdev->lldi.gts_reg;
159 cq->rdev = rdev;
160 if (user) {
161 cq->ugts = (u64)pci_resource_start(rdev->lldi.pdev, 2) +
162 (cq->cqid << rdev->cqshift);
163 cq->ugts &= PAGE_MASK;
165 return 0;
166 err4:
167 dma_free_coherent(&rdev->lldi.pdev->dev, cq->memsize, cq->queue,
168 dma_unmap_addr(cq, mapping));
169 err3:
170 kfree(cq->sw_queue);
171 err2:
172 c4iw_put_cqid(rdev, cq->cqid, uctx);
173 err1:
174 return ret;
177 static void insert_recv_cqe(struct t4_wq *wq, struct t4_cq *cq)
179 struct t4_cqe cqe;
181 PDBG("%s wq %p cq %p sw_cidx %u sw_pidx %u\n", __func__,
182 wq, cq, cq->sw_cidx, cq->sw_pidx);
183 memset(&cqe, 0, sizeof(cqe));
184 cqe.header = cpu_to_be32(V_CQE_STATUS(T4_ERR_SWFLUSH) |
185 V_CQE_OPCODE(FW_RI_SEND) |
186 V_CQE_TYPE(0) |
187 V_CQE_SWCQE(1) |
188 V_CQE_QPID(wq->sq.qid));
189 cqe.bits_type_ts = cpu_to_be64(V_CQE_GENBIT((u64)cq->gen));
190 cq->sw_queue[cq->sw_pidx] = cqe;
191 t4_swcq_produce(cq);
194 int c4iw_flush_rq(struct t4_wq *wq, struct t4_cq *cq, int count)
196 int flushed = 0;
197 int in_use = wq->rq.in_use - count;
199 BUG_ON(in_use < 0);
200 PDBG("%s wq %p cq %p rq.in_use %u skip count %u\n", __func__,
201 wq, cq, wq->rq.in_use, count);
202 while (in_use--) {
203 insert_recv_cqe(wq, cq);
204 flushed++;
206 return flushed;
209 static void insert_sq_cqe(struct t4_wq *wq, struct t4_cq *cq,
210 struct t4_swsqe *swcqe)
212 struct t4_cqe cqe;
214 PDBG("%s wq %p cq %p sw_cidx %u sw_pidx %u\n", __func__,
215 wq, cq, cq->sw_cidx, cq->sw_pidx);
216 memset(&cqe, 0, sizeof(cqe));
217 cqe.header = cpu_to_be32(V_CQE_STATUS(T4_ERR_SWFLUSH) |
218 V_CQE_OPCODE(swcqe->opcode) |
219 V_CQE_TYPE(1) |
220 V_CQE_SWCQE(1) |
221 V_CQE_QPID(wq->sq.qid));
222 CQE_WRID_SQ_IDX(&cqe) = swcqe->idx;
223 cqe.bits_type_ts = cpu_to_be64(V_CQE_GENBIT((u64)cq->gen));
224 cq->sw_queue[cq->sw_pidx] = cqe;
225 t4_swcq_produce(cq);
228 int c4iw_flush_sq(struct t4_wq *wq, struct t4_cq *cq, int count)
230 int flushed = 0;
231 struct t4_swsqe *swsqe = &wq->sq.sw_sq[wq->sq.cidx + count];
232 int in_use = wq->sq.in_use - count;
234 BUG_ON(in_use < 0);
235 while (in_use--) {
236 swsqe->signaled = 0;
237 insert_sq_cqe(wq, cq, swsqe);
238 swsqe++;
239 if (swsqe == (wq->sq.sw_sq + wq->sq.size))
240 swsqe = wq->sq.sw_sq;
241 flushed++;
243 return flushed;
247 * Move all CQEs from the HWCQ into the SWCQ.
249 void c4iw_flush_hw_cq(struct t4_cq *cq)
251 struct t4_cqe *cqe = NULL, *swcqe;
252 int ret;
254 PDBG("%s cq %p cqid 0x%x\n", __func__, cq, cq->cqid);
255 ret = t4_next_hw_cqe(cq, &cqe);
256 while (!ret) {
257 PDBG("%s flushing hwcq cidx 0x%x swcq pidx 0x%x\n",
258 __func__, cq->cidx, cq->sw_pidx);
259 swcqe = &cq->sw_queue[cq->sw_pidx];
260 *swcqe = *cqe;
261 swcqe->header |= cpu_to_be32(V_CQE_SWCQE(1));
262 t4_swcq_produce(cq);
263 t4_hwcq_consume(cq);
264 ret = t4_next_hw_cqe(cq, &cqe);
268 static int cqe_completes_wr(struct t4_cqe *cqe, struct t4_wq *wq)
270 if (CQE_OPCODE(cqe) == FW_RI_TERMINATE)
271 return 0;
273 if ((CQE_OPCODE(cqe) == FW_RI_RDMA_WRITE) && RQ_TYPE(cqe))
274 return 0;
276 if ((CQE_OPCODE(cqe) == FW_RI_READ_RESP) && SQ_TYPE(cqe))
277 return 0;
279 if (CQE_SEND_OPCODE(cqe) && RQ_TYPE(cqe) && t4_rq_empty(wq))
280 return 0;
281 return 1;
284 void c4iw_count_scqes(struct t4_cq *cq, struct t4_wq *wq, int *count)
286 struct t4_cqe *cqe;
287 u32 ptr;
289 *count = 0;
290 ptr = cq->sw_cidx;
291 while (ptr != cq->sw_pidx) {
292 cqe = &cq->sw_queue[ptr];
293 if ((SQ_TYPE(cqe) || ((CQE_OPCODE(cqe) == FW_RI_READ_RESP) &&
294 wq->sq.oldest_read)) &&
295 (CQE_QPID(cqe) == wq->sq.qid))
296 (*count)++;
297 if (++ptr == cq->size)
298 ptr = 0;
300 PDBG("%s cq %p count %d\n", __func__, cq, *count);
303 void c4iw_count_rcqes(struct t4_cq *cq, struct t4_wq *wq, int *count)
305 struct t4_cqe *cqe;
306 u32 ptr;
308 *count = 0;
309 PDBG("%s count zero %d\n", __func__, *count);
310 ptr = cq->sw_cidx;
311 while (ptr != cq->sw_pidx) {
312 cqe = &cq->sw_queue[ptr];
313 if (RQ_TYPE(cqe) && (CQE_OPCODE(cqe) != FW_RI_READ_RESP) &&
314 (CQE_QPID(cqe) == wq->sq.qid) && cqe_completes_wr(cqe, wq))
315 (*count)++;
316 if (++ptr == cq->size)
317 ptr = 0;
319 PDBG("%s cq %p count %d\n", __func__, cq, *count);
322 static void flush_completed_wrs(struct t4_wq *wq, struct t4_cq *cq)
324 struct t4_swsqe *swsqe;
325 u16 ptr = wq->sq.cidx;
326 int count = wq->sq.in_use;
327 int unsignaled = 0;
329 swsqe = &wq->sq.sw_sq[ptr];
330 while (count--)
331 if (!swsqe->signaled) {
332 if (++ptr == wq->sq.size)
333 ptr = 0;
334 swsqe = &wq->sq.sw_sq[ptr];
335 unsignaled++;
336 } else if (swsqe->complete) {
339 * Insert this completed cqe into the swcq.
341 PDBG("%s moving cqe into swcq sq idx %u cq idx %u\n",
342 __func__, ptr, cq->sw_pidx);
343 swsqe->cqe.header |= htonl(V_CQE_SWCQE(1));
344 cq->sw_queue[cq->sw_pidx] = swsqe->cqe;
345 t4_swcq_produce(cq);
346 swsqe->signaled = 0;
347 wq->sq.in_use -= unsignaled;
348 break;
349 } else
350 break;
353 static void create_read_req_cqe(struct t4_wq *wq, struct t4_cqe *hw_cqe,
354 struct t4_cqe *read_cqe)
356 read_cqe->u.scqe.cidx = wq->sq.oldest_read->idx;
357 read_cqe->len = cpu_to_be32(wq->sq.oldest_read->read_len);
358 read_cqe->header = htonl(V_CQE_QPID(CQE_QPID(hw_cqe)) |
359 V_CQE_SWCQE(SW_CQE(hw_cqe)) |
360 V_CQE_OPCODE(FW_RI_READ_REQ) |
361 V_CQE_TYPE(1));
362 read_cqe->bits_type_ts = hw_cqe->bits_type_ts;
366 * Return a ptr to the next read wr in the SWSQ or NULL.
368 static void advance_oldest_read(struct t4_wq *wq)
371 u32 rptr = wq->sq.oldest_read - wq->sq.sw_sq + 1;
373 if (rptr == wq->sq.size)
374 rptr = 0;
375 while (rptr != wq->sq.pidx) {
376 wq->sq.oldest_read = &wq->sq.sw_sq[rptr];
378 if (wq->sq.oldest_read->opcode == FW_RI_READ_REQ)
379 return;
380 if (++rptr == wq->sq.size)
381 rptr = 0;
383 wq->sq.oldest_read = NULL;
387 * poll_cq
389 * Caller must:
390 * check the validity of the first CQE,
391 * supply the wq assicated with the qpid.
393 * credit: cq credit to return to sge.
394 * cqe_flushed: 1 iff the CQE is flushed.
395 * cqe: copy of the polled CQE.
397 * return value:
398 * 0 CQE returned ok.
399 * -EAGAIN CQE skipped, try again.
400 * -EOVERFLOW CQ overflow detected.
402 static int poll_cq(struct t4_wq *wq, struct t4_cq *cq, struct t4_cqe *cqe,
403 u8 *cqe_flushed, u64 *cookie, u32 *credit)
405 int ret = 0;
406 struct t4_cqe *hw_cqe, read_cqe;
408 *cqe_flushed = 0;
409 *credit = 0;
410 ret = t4_next_cqe(cq, &hw_cqe);
411 if (ret)
412 return ret;
414 PDBG("%s CQE OVF %u qpid 0x%0x genbit %u type %u status 0x%0x"
415 " opcode 0x%0x len 0x%0x wrid_hi_stag 0x%x wrid_low_msn 0x%x\n",
416 __func__, CQE_OVFBIT(hw_cqe), CQE_QPID(hw_cqe),
417 CQE_GENBIT(hw_cqe), CQE_TYPE(hw_cqe), CQE_STATUS(hw_cqe),
418 CQE_OPCODE(hw_cqe), CQE_LEN(hw_cqe), CQE_WRID_HI(hw_cqe),
419 CQE_WRID_LOW(hw_cqe));
422 * skip cqe's not affiliated with a QP.
424 if (wq == NULL) {
425 ret = -EAGAIN;
426 goto skip_cqe;
430 * Gotta tweak READ completions:
431 * 1) the cqe doesn't contain the sq_wptr from the wr.
432 * 2) opcode not reflected from the wr.
433 * 3) read_len not reflected from the wr.
434 * 4) cq_type is RQ_TYPE not SQ_TYPE.
436 if (RQ_TYPE(hw_cqe) && (CQE_OPCODE(hw_cqe) == FW_RI_READ_RESP)) {
439 * If this is an unsolicited read response, then the read
440 * was generated by the kernel driver as part of peer-2-peer
441 * connection setup. So ignore the completion.
443 if (!wq->sq.oldest_read) {
444 if (CQE_STATUS(hw_cqe))
445 t4_set_wq_in_error(wq);
446 ret = -EAGAIN;
447 goto skip_cqe;
451 * Don't write to the HWCQ, so create a new read req CQE
452 * in local memory.
454 create_read_req_cqe(wq, hw_cqe, &read_cqe);
455 hw_cqe = &read_cqe;
456 advance_oldest_read(wq);
459 if (CQE_STATUS(hw_cqe) || t4_wq_in_error(wq)) {
460 *cqe_flushed = t4_wq_in_error(wq);
461 t4_set_wq_in_error(wq);
462 goto proc_cqe;
465 if (CQE_OPCODE(hw_cqe) == FW_RI_TERMINATE) {
466 ret = -EAGAIN;
467 goto skip_cqe;
471 * RECV completion.
473 if (RQ_TYPE(hw_cqe)) {
476 * HW only validates 4 bits of MSN. So we must validate that
477 * the MSN in the SEND is the next expected MSN. If its not,
478 * then we complete this with T4_ERR_MSN and mark the wq in
479 * error.
482 if (t4_rq_empty(wq)) {
483 t4_set_wq_in_error(wq);
484 ret = -EAGAIN;
485 goto skip_cqe;
487 if (unlikely((CQE_WRID_MSN(hw_cqe) != (wq->rq.msn)))) {
488 t4_set_wq_in_error(wq);
489 hw_cqe->header |= htonl(V_CQE_STATUS(T4_ERR_MSN));
490 goto proc_cqe;
492 goto proc_cqe;
496 * If we get here its a send completion.
498 * Handle out of order completion. These get stuffed
499 * in the SW SQ. Then the SW SQ is walked to move any
500 * now in-order completions into the SW CQ. This handles
501 * 2 cases:
502 * 1) reaping unsignaled WRs when the first subsequent
503 * signaled WR is completed.
504 * 2) out of order read completions.
506 if (!SW_CQE(hw_cqe) && (CQE_WRID_SQ_IDX(hw_cqe) != wq->sq.cidx)) {
507 struct t4_swsqe *swsqe;
509 PDBG("%s out of order completion going in sw_sq at idx %u\n",
510 __func__, CQE_WRID_SQ_IDX(hw_cqe));
511 swsqe = &wq->sq.sw_sq[CQE_WRID_SQ_IDX(hw_cqe)];
512 swsqe->cqe = *hw_cqe;
513 swsqe->complete = 1;
514 ret = -EAGAIN;
515 goto flush_wq;
518 proc_cqe:
519 *cqe = *hw_cqe;
522 * Reap the associated WR(s) that are freed up with this
523 * completion.
525 if (SQ_TYPE(hw_cqe)) {
526 wq->sq.cidx = CQE_WRID_SQ_IDX(hw_cqe);
527 PDBG("%s completing sq idx %u\n", __func__, wq->sq.cidx);
528 *cookie = wq->sq.sw_sq[wq->sq.cidx].wr_id;
529 t4_sq_consume(wq);
530 } else {
531 PDBG("%s completing rq idx %u\n", __func__, wq->rq.cidx);
532 *cookie = wq->rq.sw_rq[wq->rq.cidx].wr_id;
533 BUG_ON(t4_rq_empty(wq));
534 t4_rq_consume(wq);
537 flush_wq:
539 * Flush any completed cqes that are now in-order.
541 flush_completed_wrs(wq, cq);
543 skip_cqe:
544 if (SW_CQE(hw_cqe)) {
545 PDBG("%s cq %p cqid 0x%x skip sw cqe cidx %u\n",
546 __func__, cq, cq->cqid, cq->sw_cidx);
547 t4_swcq_consume(cq);
548 } else {
549 PDBG("%s cq %p cqid 0x%x skip hw cqe cidx %u\n",
550 __func__, cq, cq->cqid, cq->cidx);
551 t4_hwcq_consume(cq);
553 return ret;
557 * Get one cq entry from c4iw and map it to openib.
559 * Returns:
560 * 0 cqe returned
561 * -ENODATA EMPTY;
562 * -EAGAIN caller must try again
563 * any other -errno fatal error
565 static int c4iw_poll_cq_one(struct c4iw_cq *chp, struct ib_wc *wc)
567 struct c4iw_qp *qhp = NULL;
568 struct t4_cqe cqe = {0, 0}, *rd_cqe;
569 struct t4_wq *wq;
570 u32 credit = 0;
571 u8 cqe_flushed;
572 u64 cookie = 0;
573 int ret;
575 ret = t4_next_cqe(&chp->cq, &rd_cqe);
577 if (ret)
578 return ret;
580 qhp = get_qhp(chp->rhp, CQE_QPID(rd_cqe));
581 if (!qhp)
582 wq = NULL;
583 else {
584 spin_lock(&qhp->lock);
585 wq = &(qhp->wq);
587 ret = poll_cq(wq, &(chp->cq), &cqe, &cqe_flushed, &cookie, &credit);
588 if (ret)
589 goto out;
591 wc->wr_id = cookie;
592 wc->qp = &qhp->ibqp;
593 wc->vendor_err = CQE_STATUS(&cqe);
594 wc->wc_flags = 0;
596 PDBG("%s qpid 0x%x type %d opcode %d status 0x%x len %u wrid hi 0x%x "
597 "lo 0x%x cookie 0x%llx\n", __func__, CQE_QPID(&cqe),
598 CQE_TYPE(&cqe), CQE_OPCODE(&cqe), CQE_STATUS(&cqe), CQE_LEN(&cqe),
599 CQE_WRID_HI(&cqe), CQE_WRID_LOW(&cqe), (unsigned long long)cookie);
601 if (CQE_TYPE(&cqe) == 0) {
602 if (!CQE_STATUS(&cqe))
603 wc->byte_len = CQE_LEN(&cqe);
604 else
605 wc->byte_len = 0;
606 wc->opcode = IB_WC_RECV;
607 if (CQE_OPCODE(&cqe) == FW_RI_SEND_WITH_INV ||
608 CQE_OPCODE(&cqe) == FW_RI_SEND_WITH_SE_INV) {
609 wc->ex.invalidate_rkey = CQE_WRID_STAG(&cqe);
610 wc->wc_flags |= IB_WC_WITH_INVALIDATE;
612 } else {
613 switch (CQE_OPCODE(&cqe)) {
614 case FW_RI_RDMA_WRITE:
615 wc->opcode = IB_WC_RDMA_WRITE;
616 break;
617 case FW_RI_READ_REQ:
618 wc->opcode = IB_WC_RDMA_READ;
619 wc->byte_len = CQE_LEN(&cqe);
620 break;
621 case FW_RI_SEND_WITH_INV:
622 case FW_RI_SEND_WITH_SE_INV:
623 wc->opcode = IB_WC_SEND;
624 wc->wc_flags |= IB_WC_WITH_INVALIDATE;
625 break;
626 case FW_RI_SEND:
627 case FW_RI_SEND_WITH_SE:
628 wc->opcode = IB_WC_SEND;
629 break;
630 case FW_RI_BIND_MW:
631 wc->opcode = IB_WC_BIND_MW;
632 break;
634 case FW_RI_LOCAL_INV:
635 wc->opcode = IB_WC_LOCAL_INV;
636 break;
637 case FW_RI_FAST_REGISTER:
638 wc->opcode = IB_WC_FAST_REG_MR;
639 break;
640 default:
641 printk(KERN_ERR MOD "Unexpected opcode %d "
642 "in the CQE received for QPID=0x%0x\n",
643 CQE_OPCODE(&cqe), CQE_QPID(&cqe));
644 ret = -EINVAL;
645 goto out;
649 if (cqe_flushed)
650 wc->status = IB_WC_WR_FLUSH_ERR;
651 else {
653 switch (CQE_STATUS(&cqe)) {
654 case T4_ERR_SUCCESS:
655 wc->status = IB_WC_SUCCESS;
656 break;
657 case T4_ERR_STAG:
658 wc->status = IB_WC_LOC_ACCESS_ERR;
659 break;
660 case T4_ERR_PDID:
661 wc->status = IB_WC_LOC_PROT_ERR;
662 break;
663 case T4_ERR_QPID:
664 case T4_ERR_ACCESS:
665 wc->status = IB_WC_LOC_ACCESS_ERR;
666 break;
667 case T4_ERR_WRAP:
668 wc->status = IB_WC_GENERAL_ERR;
669 break;
670 case T4_ERR_BOUND:
671 wc->status = IB_WC_LOC_LEN_ERR;
672 break;
673 case T4_ERR_INVALIDATE_SHARED_MR:
674 case T4_ERR_INVALIDATE_MR_WITH_MW_BOUND:
675 wc->status = IB_WC_MW_BIND_ERR;
676 break;
677 case T4_ERR_CRC:
678 case T4_ERR_MARKER:
679 case T4_ERR_PDU_LEN_ERR:
680 case T4_ERR_OUT_OF_RQE:
681 case T4_ERR_DDP_VERSION:
682 case T4_ERR_RDMA_VERSION:
683 case T4_ERR_DDP_QUEUE_NUM:
684 case T4_ERR_MSN:
685 case T4_ERR_TBIT:
686 case T4_ERR_MO:
687 case T4_ERR_MSN_RANGE:
688 case T4_ERR_IRD_OVERFLOW:
689 case T4_ERR_OPCODE:
690 case T4_ERR_INTERNAL_ERR:
691 wc->status = IB_WC_FATAL_ERR;
692 break;
693 case T4_ERR_SWFLUSH:
694 wc->status = IB_WC_WR_FLUSH_ERR;
695 break;
696 default:
697 printk(KERN_ERR MOD
698 "Unexpected cqe_status 0x%x for QPID=0x%0x\n",
699 CQE_STATUS(&cqe), CQE_QPID(&cqe));
700 ret = -EINVAL;
703 out:
704 if (wq)
705 spin_unlock(&qhp->lock);
706 return ret;
709 int c4iw_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
711 struct c4iw_cq *chp;
712 unsigned long flags;
713 int npolled;
714 int err = 0;
716 chp = to_c4iw_cq(ibcq);
718 spin_lock_irqsave(&chp->lock, flags);
719 for (npolled = 0; npolled < num_entries; ++npolled) {
720 do {
721 err = c4iw_poll_cq_one(chp, wc + npolled);
722 } while (err == -EAGAIN);
723 if (err)
724 break;
726 spin_unlock_irqrestore(&chp->lock, flags);
727 return !err || err == -ENODATA ? npolled : err;
730 int c4iw_destroy_cq(struct ib_cq *ib_cq)
732 struct c4iw_cq *chp;
733 struct c4iw_ucontext *ucontext;
735 PDBG("%s ib_cq %p\n", __func__, ib_cq);
736 chp = to_c4iw_cq(ib_cq);
738 remove_handle(chp->rhp, &chp->rhp->cqidr, chp->cq.cqid);
739 atomic_dec(&chp->refcnt);
740 wait_event(chp->wait, !atomic_read(&chp->refcnt));
742 ucontext = ib_cq->uobject ? to_c4iw_ucontext(ib_cq->uobject->context)
743 : NULL;
744 destroy_cq(&chp->rhp->rdev, &chp->cq,
745 ucontext ? &ucontext->uctx : &chp->cq.rdev->uctx);
746 kfree(chp);
747 return 0;
750 struct ib_cq *c4iw_create_cq(struct ib_device *ibdev, int entries,
751 int vector, struct ib_ucontext *ib_context,
752 struct ib_udata *udata)
754 struct c4iw_dev *rhp;
755 struct c4iw_cq *chp;
756 struct c4iw_create_cq_resp uresp;
757 struct c4iw_ucontext *ucontext = NULL;
758 int ret;
759 size_t memsize, hwentries;
760 struct c4iw_mm_entry *mm, *mm2;
762 PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
764 rhp = to_c4iw_dev(ibdev);
766 chp = kzalloc(sizeof(*chp), GFP_KERNEL);
767 if (!chp)
768 return ERR_PTR(-ENOMEM);
770 if (ib_context)
771 ucontext = to_c4iw_ucontext(ib_context);
773 /* account for the status page. */
774 entries++;
776 /* IQ needs one extra entry to differentiate full vs empty. */
777 entries++;
780 * entries must be multiple of 16 for HW.
782 entries = roundup(entries, 16);
785 * Make actual HW queue 2x to avoid cdix_inc overflows.
787 hwentries = entries * 2;
790 * Make HW queue at least 64 entries so GTS updates aren't too
791 * frequent.
793 if (hwentries < 64)
794 hwentries = 64;
796 memsize = hwentries * sizeof *chp->cq.queue;
799 * memsize must be a multiple of the page size if its a user cq.
801 if (ucontext) {
802 memsize = roundup(memsize, PAGE_SIZE);
803 hwentries = memsize / sizeof *chp->cq.queue;
804 while (hwentries > T4_MAX_IQ_SIZE) {
805 memsize -= PAGE_SIZE;
806 hwentries = memsize / sizeof *chp->cq.queue;
809 chp->cq.size = hwentries;
810 chp->cq.memsize = memsize;
812 ret = create_cq(&rhp->rdev, &chp->cq,
813 ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
814 if (ret)
815 goto err1;
817 chp->rhp = rhp;
818 chp->cq.size--; /* status page */
819 chp->ibcq.cqe = entries - 2;
820 spin_lock_init(&chp->lock);
821 spin_lock_init(&chp->comp_handler_lock);
822 atomic_set(&chp->refcnt, 1);
823 init_waitqueue_head(&chp->wait);
824 ret = insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid);
825 if (ret)
826 goto err2;
828 if (ucontext) {
829 mm = kmalloc(sizeof *mm, GFP_KERNEL);
830 if (!mm)
831 goto err3;
832 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
833 if (!mm2)
834 goto err4;
836 uresp.qid_mask = rhp->rdev.cqmask;
837 uresp.cqid = chp->cq.cqid;
838 uresp.size = chp->cq.size;
839 uresp.memsize = chp->cq.memsize;
840 spin_lock(&ucontext->mmap_lock);
841 uresp.key = ucontext->key;
842 ucontext->key += PAGE_SIZE;
843 uresp.gts_key = ucontext->key;
844 ucontext->key += PAGE_SIZE;
845 spin_unlock(&ucontext->mmap_lock);
846 ret = ib_copy_to_udata(udata, &uresp, sizeof uresp);
847 if (ret)
848 goto err5;
850 mm->key = uresp.key;
851 mm->addr = virt_to_phys(chp->cq.queue);
852 mm->len = chp->cq.memsize;
853 insert_mmap(ucontext, mm);
855 mm2->key = uresp.gts_key;
856 mm2->addr = chp->cq.ugts;
857 mm2->len = PAGE_SIZE;
858 insert_mmap(ucontext, mm2);
860 PDBG("%s cqid 0x%0x chp %p size %u memsize %zu, dma_addr 0x%0llx\n",
861 __func__, chp->cq.cqid, chp, chp->cq.size,
862 chp->cq.memsize,
863 (unsigned long long) chp->cq.dma_addr);
864 return &chp->ibcq;
865 err5:
866 kfree(mm2);
867 err4:
868 kfree(mm);
869 err3:
870 remove_handle(rhp, &rhp->cqidr, chp->cq.cqid);
871 err2:
872 destroy_cq(&chp->rhp->rdev, &chp->cq,
873 ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
874 err1:
875 kfree(chp);
876 return ERR_PTR(ret);
879 int c4iw_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
881 return -ENOSYS;
884 int c4iw_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
886 struct c4iw_cq *chp;
887 int ret;
888 unsigned long flag;
890 chp = to_c4iw_cq(ibcq);
891 spin_lock_irqsave(&chp->lock, flag);
892 ret = t4_arm_cq(&chp->cq,
893 (flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED);
894 spin_unlock_irqrestore(&chp->lock, flag);
895 if (ret && !(flags & IB_CQ_REPORT_MISSED_EVENTS))
896 ret = 0;
897 return ret;