?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
IB/mlx4: Add support for modifying CQ moderation parameters
[linux-2.6/libata-dev.git] / drivers / infiniband / hw / mlx4 / cq.c
blobe4fb64b118e3f73cef3f556d307aa1a52caa6ae0
1 /*
2 * Copyright (c) 2007 Cisco Systems, Inc. All rights reserved.
3 *
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:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
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.
22 *
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.
31 */
32
33 #include <linux/mlx4/cq.h>
34 #include <linux/mlx4/qp.h>
35
36 #include "mlx4_ib.h"
37 #include "user.h"
38
39 static void mlx4_ib_cq_comp(struct mlx4_cq *cq)
40 {
41 struct ib_cq *ibcq = &to_mibcq(cq)->ibcq;
42 ibcq->comp_handler(ibcq, ibcq->cq_context);
43 }
44
45 static void mlx4_ib_cq_event(struct mlx4_cq *cq, enum mlx4_event type)
46 {
47 struct ib_event event;
48 struct ib_cq *ibcq;
49
50 if (type != MLX4_EVENT_TYPE_CQ_ERROR) {
51 printk(KERN_WARNING "mlx4_ib: Unexpected event type %d "
52 "on CQ %06x\n", type, cq->cqn);
53 return;
54 }
55
56 ibcq = &to_mibcq(cq)->ibcq;
57 if (ibcq->event_handler) {
58 event.device = ibcq->device;
59 event.event = IB_EVENT_CQ_ERR;
60 event.element.cq = ibcq;
61 ibcq->event_handler(&event, ibcq->cq_context);
62 }
63 }
64
65 static void *get_cqe_from_buf(struct mlx4_ib_cq_buf *buf, int n)
66 {
67 return mlx4_buf_offset(&buf->buf, n * sizeof (struct mlx4_cqe));
68 }
69
70 static void *get_cqe(struct mlx4_ib_cq *cq, int n)
71 {
72 return get_cqe_from_buf(&cq->buf, n);
73 }
74
75 static void *get_sw_cqe(struct mlx4_ib_cq *cq, int n)
76 {
77 struct mlx4_cqe *cqe = get_cqe(cq, n & cq->ibcq.cqe);
78
79 return (!!(cqe->owner_sr_opcode & MLX4_CQE_OWNER_MASK) ^
80 !!(n & (cq->ibcq.cqe + 1))) ? NULL : cqe;
81 }
82
83 static struct mlx4_cqe *next_cqe_sw(struct mlx4_ib_cq *cq)
84 {
85 return get_sw_cqe(cq, cq->mcq.cons_index);
86 }
87
88 int mlx4_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
89 {
90 struct mlx4_ib_cq *mcq = to_mcq(cq);
91 struct mlx4_ib_dev *dev = to_mdev(cq->device);
92
93 return mlx4_cq_modify(dev->dev, &mcq->mcq, cq_count, cq_period);
94 }
95
96 struct ib_cq *mlx4_ib_create_cq(struct ib_device *ibdev, int entries, int vector,
97 struct ib_ucontext *context,
98 struct ib_udata *udata)
99 {
100 struct mlx4_ib_dev *dev = to_mdev(ibdev);
101 struct mlx4_ib_cq *cq;
102 struct mlx4_uar *uar;
103 int buf_size;
104 int err;
105
106 if (entries < 1 || entries > dev->dev->caps.max_cqes)
107 return ERR_PTR(-EINVAL);
108
109 cq = kmalloc(sizeof *cq, GFP_KERNEL);
110 if (!cq)
111 return ERR_PTR(-ENOMEM);
112
113 entries = roundup_pow_of_two(entries + 1);
114 cq->ibcq.cqe = entries - 1;
115 buf_size = entries * sizeof (struct mlx4_cqe);
116 spin_lock_init(&cq->lock);
117
118 if (context) {
119 struct mlx4_ib_create_cq ucmd;
120
121 if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
122 err = -EFAULT;
123 goto err_cq;
124 }
125
126 cq->umem = ib_umem_get(context, ucmd.buf_addr, buf_size,
127 IB_ACCESS_LOCAL_WRITE);
128 if (IS_ERR(cq->umem)) {
129 err = PTR_ERR(cq->umem);
130 goto err_cq;
131 }
132
133 err = mlx4_mtt_init(dev->dev, ib_umem_page_count(cq->umem),
134 ilog2(cq->umem->page_size), &cq->buf.mtt);
135 if (err)
136 goto err_buf;
137
138 err = mlx4_ib_umem_write_mtt(dev, &cq->buf.mtt, cq->umem);
139 if (err)
140 goto err_mtt;
141
142 err = mlx4_ib_db_map_user(to_mucontext(context), ucmd.db_addr,
143 &cq->db);
144 if (err)
145 goto err_mtt;
146
147 uar = &to_mucontext(context)->uar;
148 } else {
149 err = mlx4_ib_db_alloc(dev, &cq->db, 1);
150 if (err)
151 goto err_cq;
152
153 cq->mcq.set_ci_db = cq->db.db;
154 cq->mcq.arm_db = cq->db.db + 1;
155 *cq->mcq.set_ci_db = 0;
156 *cq->mcq.arm_db = 0;
157
158 if (mlx4_buf_alloc(dev->dev, buf_size, PAGE_SIZE * 2, &cq->buf.buf)) {
159 err = -ENOMEM;
160 goto err_db;
161 }
162
163 err = mlx4_mtt_init(dev->dev, cq->buf.buf.npages, cq->buf.buf.page_shift,
164 &cq->buf.mtt);
165 if (err)
166 goto err_buf;
167
168 err = mlx4_buf_write_mtt(dev->dev, &cq->buf.mtt, &cq->buf.buf);
169 if (err)
170 goto err_mtt;
171
172 uar = &dev->priv_uar;
173 }
174
175 err = mlx4_cq_alloc(dev->dev, entries, &cq->buf.mtt, uar,
176 cq->db.dma, &cq->mcq);
177 if (err)
178 goto err_dbmap;
179
180 cq->mcq.comp = mlx4_ib_cq_comp;
181 cq->mcq.event = mlx4_ib_cq_event;
182
183 if (context)
184 if (ib_copy_to_udata(udata, &cq->mcq.cqn, sizeof (__u32))) {
185 err = -EFAULT;
186 goto err_dbmap;
187 }
188
189 return &cq->ibcq;
190
191 err_dbmap:
192 if (context)
193 mlx4_ib_db_unmap_user(to_mucontext(context), &cq->db);
194
195 err_mtt:
196 mlx4_mtt_cleanup(dev->dev, &cq->buf.mtt);
197
198 err_buf:
199 if (context)
200 ib_umem_release(cq->umem);
201 else
202 mlx4_buf_free(dev->dev, entries * sizeof (struct mlx4_cqe),
203 &cq->buf.buf);
204
205 err_db:
206 if (!context)
207 mlx4_ib_db_free(dev, &cq->db);
208
209 err_cq:
210 kfree(cq);
211
212 return ERR_PTR(err);
213 }
214
215 int mlx4_ib_destroy_cq(struct ib_cq *cq)
216 {
217 struct mlx4_ib_dev *dev = to_mdev(cq->device);
218 struct mlx4_ib_cq *mcq = to_mcq(cq);
219
220 mlx4_cq_free(dev->dev, &mcq->mcq);
221 mlx4_mtt_cleanup(dev->dev, &mcq->buf.mtt);
222
223 if (cq->uobject) {
224 mlx4_ib_db_unmap_user(to_mucontext(cq->uobject->context), &mcq->db);
225 ib_umem_release(mcq->umem);
226 } else {
227 mlx4_buf_free(dev->dev, (cq->cqe + 1) * sizeof (struct mlx4_cqe),
228 &mcq->buf.buf);
229 mlx4_ib_db_free(dev, &mcq->db);
230 }
231
232 kfree(mcq);
233
234 return 0;
235 }
236
237 static void dump_cqe(void *cqe)
238 {
239 __be32 *buf = cqe;
240
241 printk(KERN_DEBUG "CQE contents %08x %08x %08x %08x %08x %08x %08x %08x\n",
242 be32_to_cpu(buf[0]), be32_to_cpu(buf[1]), be32_to_cpu(buf[2]),
243 be32_to_cpu(buf[3]), be32_to_cpu(buf[4]), be32_to_cpu(buf[5]),
244 be32_to_cpu(buf[6]), be32_to_cpu(buf[7]));
245 }
246
247 static void mlx4_ib_handle_error_cqe(struct mlx4_err_cqe *cqe,
248 struct ib_wc *wc)
249 {
250 if (cqe->syndrome == MLX4_CQE_SYNDROME_LOCAL_QP_OP_ERR) {
251 printk(KERN_DEBUG "local QP operation err "
252 "(QPN %06x, WQE index %x, vendor syndrome %02x, "
253 "opcode = %02x)\n",
254 be32_to_cpu(cqe->my_qpn), be16_to_cpu(cqe->wqe_index),
255 cqe->vendor_err_syndrome,
256 cqe->owner_sr_opcode & ~MLX4_CQE_OWNER_MASK);
257 dump_cqe(cqe);
258 }
259
260 switch (cqe->syndrome) {
261 case MLX4_CQE_SYNDROME_LOCAL_LENGTH_ERR:
262 wc->status = IB_WC_LOC_LEN_ERR;
263 break;
264 case MLX4_CQE_SYNDROME_LOCAL_QP_OP_ERR:
265 wc->status = IB_WC_LOC_QP_OP_ERR;
266 break;
267 case MLX4_CQE_SYNDROME_LOCAL_PROT_ERR:
268 wc->status = IB_WC_LOC_PROT_ERR;
269 break;
270 case MLX4_CQE_SYNDROME_WR_FLUSH_ERR:
271 wc->status = IB_WC_WR_FLUSH_ERR;
272 break;
273 case MLX4_CQE_SYNDROME_MW_BIND_ERR:
274 wc->status = IB_WC_MW_BIND_ERR;
275 break;
276 case MLX4_CQE_SYNDROME_BAD_RESP_ERR:
277 wc->status = IB_WC_BAD_RESP_ERR;
278 break;
279 case MLX4_CQE_SYNDROME_LOCAL_ACCESS_ERR:
280 wc->status = IB_WC_LOC_ACCESS_ERR;
281 break;
282 case MLX4_CQE_SYNDROME_REMOTE_INVAL_REQ_ERR:
283 wc->status = IB_WC_REM_INV_REQ_ERR;
284 break;
285 case MLX4_CQE_SYNDROME_REMOTE_ACCESS_ERR:
286 wc->status = IB_WC_REM_ACCESS_ERR;
287 break;
288 case MLX4_CQE_SYNDROME_REMOTE_OP_ERR:
289 wc->status = IB_WC_REM_OP_ERR;
290 break;
291 case MLX4_CQE_SYNDROME_TRANSPORT_RETRY_EXC_ERR:
292 wc->status = IB_WC_RETRY_EXC_ERR;
293 break;
294 case MLX4_CQE_SYNDROME_RNR_RETRY_EXC_ERR:
295 wc->status = IB_WC_RNR_RETRY_EXC_ERR;
296 break;
297 case MLX4_CQE_SYNDROME_REMOTE_ABORTED_ERR:
298 wc->status = IB_WC_REM_ABORT_ERR;
299 break;
300 default:
301 wc->status = IB_WC_GENERAL_ERR;
302 break;
303 }
304
305 wc->vendor_err = cqe->vendor_err_syndrome;
306 }
307
308 static int mlx4_ib_ipoib_csum_ok(__be32 status, __be16 checksum)
309 {
310 return ((status & cpu_to_be32(MLX4_CQE_IPOIB_STATUS_IPV4 |
311 MLX4_CQE_IPOIB_STATUS_IPV4F |
312 MLX4_CQE_IPOIB_STATUS_IPV4OPT |
313 MLX4_CQE_IPOIB_STATUS_IPV6 |
314 MLX4_CQE_IPOIB_STATUS_IPOK)) ==
315 cpu_to_be32(MLX4_CQE_IPOIB_STATUS_IPV4 |
316 MLX4_CQE_IPOIB_STATUS_IPOK)) &&
317 (status & cpu_to_be32(MLX4_CQE_IPOIB_STATUS_UDP |
318 MLX4_CQE_IPOIB_STATUS_TCP)) &&
319 checksum == cpu_to_be16(0xffff);
320 }
321
322 static int mlx4_ib_poll_one(struct mlx4_ib_cq *cq,
323 struct mlx4_ib_qp **cur_qp,
324 struct ib_wc *wc)
325 {
326 struct mlx4_cqe *cqe;
327 struct mlx4_qp *mqp;
328 struct mlx4_ib_wq *wq;
329 struct mlx4_ib_srq *srq;
330 int is_send;
331 int is_error;
332 u32 g_mlpath_rqpn;
333 u16 wqe_ctr;
334
335 cqe = next_cqe_sw(cq);
336 if (!cqe)
337 return -EAGAIN;
338
339 ++cq->mcq.cons_index;
340
341 /*
342 * Make sure we read CQ entry contents after we've checked the
343 * ownership bit.
344 */
345 rmb();
346
347 is_send = cqe->owner_sr_opcode & MLX4_CQE_IS_SEND_MASK;
348 is_error = (cqe->owner_sr_opcode & MLX4_CQE_OPCODE_MASK) ==
349 MLX4_CQE_OPCODE_ERROR;
350
351 if (unlikely((cqe->owner_sr_opcode & MLX4_CQE_OPCODE_MASK) == MLX4_OPCODE_NOP &&
352 is_send)) {
353 printk(KERN_WARNING "Completion for NOP opcode detected!\n");
354 return -EINVAL;
355 }
356
357 if (!*cur_qp ||
358 (be32_to_cpu(cqe->my_qpn) & 0xffffff) != (*cur_qp)->mqp.qpn) {
359 /*
360 * We do not have to take the QP table lock here,
361 * because CQs will be locked while QPs are removed
362 * from the table.
363 */
364 mqp = __mlx4_qp_lookup(to_mdev(cq->ibcq.device)->dev,
365 be32_to_cpu(cqe->my_qpn));
366 if (unlikely(!mqp)) {
367 printk(KERN_WARNING "CQ %06x with entry for unknown QPN %06x\n",
368 cq->mcq.cqn, be32_to_cpu(cqe->my_qpn) & 0xffffff);
369 return -EINVAL;
370 }
371
372 *cur_qp = to_mibqp(mqp);
373 }
374
375 wc->qp = &(*cur_qp)->ibqp;
376
377 if (is_send) {
378 wq = &(*cur_qp)->sq;
379 if (!(*cur_qp)->sq_signal_bits) {
380 wqe_ctr = be16_to_cpu(cqe->wqe_index);
381 wq->tail += (u16) (wqe_ctr - (u16) wq->tail);
382 }
383 wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
384 ++wq->tail;
385 } else if ((*cur_qp)->ibqp.srq) {
386 srq = to_msrq((*cur_qp)->ibqp.srq);
387 wqe_ctr = be16_to_cpu(cqe->wqe_index);
388 wc->wr_id = srq->wrid[wqe_ctr];
389 mlx4_ib_free_srq_wqe(srq, wqe_ctr);
390 } else {
391 wq = &(*cur_qp)->rq;
392 wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
393 ++wq->tail;
394 }
395
396 if (unlikely(is_error)) {
397 mlx4_ib_handle_error_cqe((struct mlx4_err_cqe *) cqe, wc);
398 return 0;
399 }
400
401 wc->status = IB_WC_SUCCESS;
402
403 if (is_send) {
404 wc->wc_flags = 0;
405 switch (cqe->owner_sr_opcode & MLX4_CQE_OPCODE_MASK) {
406 case MLX4_OPCODE_RDMA_WRITE_IMM:
407 wc->wc_flags |= IB_WC_WITH_IMM;
408 case MLX4_OPCODE_RDMA_WRITE:
409 wc->opcode = IB_WC_RDMA_WRITE;
410 break;
411 case MLX4_OPCODE_SEND_IMM:
412 wc->wc_flags |= IB_WC_WITH_IMM;
413 case MLX4_OPCODE_SEND:
414 wc->opcode = IB_WC_SEND;
415 break;
416 case MLX4_OPCODE_RDMA_READ:
417 wc->opcode = IB_WC_RDMA_READ;
418 wc->byte_len = be32_to_cpu(cqe->byte_cnt);
419 break;
420 case MLX4_OPCODE_ATOMIC_CS:
421 wc->opcode = IB_WC_COMP_SWAP;
422 wc->byte_len = 8;
423 break;
424 case MLX4_OPCODE_ATOMIC_FA:
425 wc->opcode = IB_WC_FETCH_ADD;
426 wc->byte_len = 8;
427 break;
428 case MLX4_OPCODE_BIND_MW:
429 wc->opcode = IB_WC_BIND_MW;
430 break;
431 case MLX4_OPCODE_LSO:
432 wc->opcode = IB_WC_LSO;
433 break;
434 }
435 } else {
436 wc->byte_len = be32_to_cpu(cqe->byte_cnt);
437
438 switch (cqe->owner_sr_opcode & MLX4_CQE_OPCODE_MASK) {
439 case MLX4_RECV_OPCODE_RDMA_WRITE_IMM:
440 wc->opcode = IB_WC_RECV_RDMA_WITH_IMM;
441 wc->wc_flags = IB_WC_WITH_IMM;
442 wc->imm_data = cqe->immed_rss_invalid;
443 break;
444 case MLX4_RECV_OPCODE_SEND:
445 wc->opcode = IB_WC_RECV;
446 wc->wc_flags = 0;
447 break;
448 case MLX4_RECV_OPCODE_SEND_IMM:
449 wc->opcode = IB_WC_RECV;
450 wc->wc_flags = IB_WC_WITH_IMM;
451 wc->imm_data = cqe->immed_rss_invalid;
452 break;
453 }
454
455 wc->slid = be16_to_cpu(cqe->rlid);
456 wc->sl = cqe->sl >> 4;
457 g_mlpath_rqpn = be32_to_cpu(cqe->g_mlpath_rqpn);
458 wc->src_qp = g_mlpath_rqpn & 0xffffff;
459 wc->dlid_path_bits = (g_mlpath_rqpn >> 24) & 0x7f;
460 wc->wc_flags |= g_mlpath_rqpn & 0x80000000 ? IB_WC_GRH : 0;
461 wc->pkey_index = be32_to_cpu(cqe->immed_rss_invalid) & 0x7f;
462 wc->csum_ok = mlx4_ib_ipoib_csum_ok(cqe->ipoib_status,
463 cqe->checksum);
464 }
465
466 return 0;
467 }
468
469 int mlx4_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
470 {
471 struct mlx4_ib_cq *cq = to_mcq(ibcq);
472 struct mlx4_ib_qp *cur_qp = NULL;
473 unsigned long flags;
474 int npolled;
475 int err = 0;
476
477 spin_lock_irqsave(&cq->lock, flags);
478
479 for (npolled = 0; npolled < num_entries; ++npolled) {
480 err = mlx4_ib_poll_one(cq, &cur_qp, wc + npolled);
481 if (err)
482 break;
483 }
484
485 if (npolled)
486 mlx4_cq_set_ci(&cq->mcq);
487
488 spin_unlock_irqrestore(&cq->lock, flags);
489
490 if (err == 0 || err == -EAGAIN)
491 return npolled;
492 else
493 return err;
494 }
495
496 int mlx4_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
497 {
498 mlx4_cq_arm(&to_mcq(ibcq)->mcq,
499 (flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED ?
500 MLX4_CQ_DB_REQ_NOT_SOL : MLX4_CQ_DB_REQ_NOT,
501 to_mdev(ibcq->device)->uar_map,
502 MLX4_GET_DOORBELL_LOCK(&to_mdev(ibcq->device)->uar_lock));
503
504 return 0;
505 }
506
507 void __mlx4_ib_cq_clean(struct mlx4_ib_cq *cq, u32 qpn, struct mlx4_ib_srq *srq)
508 {
509 u32 prod_index;
510 int nfreed = 0;
511 struct mlx4_cqe *cqe, *dest;
512 u8 owner_bit;
513
514 /*
515 * First we need to find the current producer index, so we
516 * know where to start cleaning from. It doesn't matter if HW
517 * adds new entries after this loop -- the QP we're worried
518 * about is already in RESET, so the new entries won't come
519 * from our QP and therefore don't need to be checked.
520 */
521 for (prod_index = cq->mcq.cons_index; get_sw_cqe(cq, prod_index); ++prod_index)
522 if (prod_index == cq->mcq.cons_index + cq->ibcq.cqe)
523 break;
524
525 /*
526 * Now sweep backwards through the CQ, removing CQ entries
527 * that match our QP by copying older entries on top of them.
528 */
529 while ((int) --prod_index - (int) cq->mcq.cons_index >= 0) {
530 cqe = get_cqe(cq, prod_index & cq->ibcq.cqe);
531 if ((be32_to_cpu(cqe->my_qpn) & 0xffffff) == qpn) {
532 if (srq && !(cqe->owner_sr_opcode & MLX4_CQE_IS_SEND_MASK))
533 mlx4_ib_free_srq_wqe(srq, be16_to_cpu(cqe->wqe_index));
534 ++nfreed;
535 } else if (nfreed) {
536 dest = get_cqe(cq, (prod_index + nfreed) & cq->ibcq.cqe);
537 owner_bit = dest->owner_sr_opcode & MLX4_CQE_OWNER_MASK;
538 memcpy(dest, cqe, sizeof *cqe);
539 dest->owner_sr_opcode = owner_bit |
540 (dest->owner_sr_opcode & ~MLX4_CQE_OWNER_MASK);
541 }
542 }
543
544 if (nfreed) {
545 cq->mcq.cons_index += nfreed;
546 /*
547 * Make sure update of buffer contents is done before
548 * updating consumer index.
549 */
550 wmb();
551 mlx4_cq_set_ci(&cq->mcq);
552 }
553 }
554
555 void mlx4_ib_cq_clean(struct mlx4_ib_cq *cq, u32 qpn, struct mlx4_ib_srq *srq)
556 {
557 spin_lock_irq(&cq->lock);
558 __mlx4_ib_cq_clean(cq, qpn, srq);
559 spin_unlock_irq(&cq->lock);
560 }
Linux 2.6 libata-dev (mirror)