This source file includes following definitions.
- alloc_ird
- free_ird
- set_state
- dealloc_oc_sq
- dealloc_host_sq
- dealloc_sq
- alloc_oc_sq
- alloc_host_sq
- alloc_sq
- destroy_qp
- c4iw_bar2_addrs
- create_qp
- build_immd
- build_isgl
- build_rdma_send
- build_rdma_write
- build_immd_cmpl
- build_rdma_write_cmpl
- build_rdma_read
- post_write_cmpl
- build_rdma_recv
- build_srq_recv
- build_tpte_memreg
- build_memreg
- build_inv_stag
- c4iw_qp_add_ref
- c4iw_qp_rem_ref
- add_to_fc_list
- ring_kernel_sq_db
- ring_kernel_rq_db
- ib_to_fw_opcode
- complete_sq_drain_wr
- complete_sq_drain_wrs
- complete_rq_drain_wr
- complete_rq_drain_wrs
- c4iw_post_send
- c4iw_post_receive
- defer_srq_wr
- c4iw_post_srq_recv
- build_term_codes
- post_terminate
- __flush_qp
- flush_qp
- rdma_fini
- build_rtr_msg
- rdma_init
- c4iw_modify_qp
- c4iw_destroy_qp
- c4iw_create_qp
- c4iw_ib_modify_qp
- c4iw_get_qp
- c4iw_dispatch_srq_limit_reached_event
- c4iw_modify_srq
- c4iw_ib_query_qp
- free_srq_queue
- alloc_srq_queue
- c4iw_copy_wr_to_srq
- c4iw_create_srq
- c4iw_destroy_srq
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33 #include <linux/module.h>
34 #include <rdma/uverbs_ioctl.h>
35
36 #include "iw_cxgb4.h"
37
38 static int db_delay_usecs = 1;
39 module_param(db_delay_usecs, int, 0644);
40 MODULE_PARM_DESC(db_delay_usecs, "Usecs to delay awaiting db fifo to drain");
41
42 static int ocqp_support = 1;
43 module_param(ocqp_support, int, 0644);
44 MODULE_PARM_DESC(ocqp_support, "Support on-chip SQs (default=1)");
45
46 int db_fc_threshold = 1000;
47 module_param(db_fc_threshold, int, 0644);
48 MODULE_PARM_DESC(db_fc_threshold,
49 "QP count/threshold that triggers"
50 " automatic db flow control mode (default = 1000)");
51
52 int db_coalescing_threshold;
53 module_param(db_coalescing_threshold, int, 0644);
54 MODULE_PARM_DESC(db_coalescing_threshold,
55 "QP count/threshold that triggers"
56 " disabling db coalescing (default = 0)");
57
58 static int max_fr_immd = T4_MAX_FR_IMMD;
59 module_param(max_fr_immd, int, 0644);
60 MODULE_PARM_DESC(max_fr_immd, "fastreg threshold for using DSGL instead of immediate");
61
62 static int alloc_ird(struct c4iw_dev *dev, u32 ird)
63 {
64 int ret = 0;
65
66 xa_lock_irq(&dev->qps);
67 if (ird <= dev->avail_ird)
68 dev->avail_ird -= ird;
69 else
70 ret = -ENOMEM;
71 xa_unlock_irq(&dev->qps);
72
73 if (ret)
74 dev_warn(&dev->rdev.lldi.pdev->dev,
75 "device IRD resources exhausted\n");
76
77 return ret;
78 }
79
80 static void free_ird(struct c4iw_dev *dev, int ird)
81 {
82 xa_lock_irq(&dev->qps);
83 dev->avail_ird += ird;
84 xa_unlock_irq(&dev->qps);
85 }
86
87 static void set_state(struct c4iw_qp *qhp, enum c4iw_qp_state state)
88 {
89 unsigned long flag;
90 spin_lock_irqsave(&qhp->lock, flag);
91 qhp->attr.state = state;
92 spin_unlock_irqrestore(&qhp->lock, flag);
93 }
94
95 static void dealloc_oc_sq(struct c4iw_rdev *rdev, struct t4_sq *sq)
96 {
97 c4iw_ocqp_pool_free(rdev, sq->dma_addr, sq->memsize);
98 }
99
100 static void dealloc_host_sq(struct c4iw_rdev *rdev, struct t4_sq *sq)
101 {
102 dma_free_coherent(&(rdev->lldi.pdev->dev), sq->memsize, sq->queue,
103 dma_unmap_addr(sq, mapping));
104 }
105
106 static void dealloc_sq(struct c4iw_rdev *rdev, struct t4_sq *sq)
107 {
108 if (t4_sq_onchip(sq))
109 dealloc_oc_sq(rdev, sq);
110 else
111 dealloc_host_sq(rdev, sq);
112 }
113
114 static int alloc_oc_sq(struct c4iw_rdev *rdev, struct t4_sq *sq)
115 {
116 if (!ocqp_support || !ocqp_supported(&rdev->lldi))
117 return -ENOSYS;
118 sq->dma_addr = c4iw_ocqp_pool_alloc(rdev, sq->memsize);
119 if (!sq->dma_addr)
120 return -ENOMEM;
121 sq->phys_addr = rdev->oc_mw_pa + sq->dma_addr -
122 rdev->lldi.vr->ocq.start;
123 sq->queue = (__force union t4_wr *)(rdev->oc_mw_kva + sq->dma_addr -
124 rdev->lldi.vr->ocq.start);
125 sq->flags |= T4_SQ_ONCHIP;
126 return 0;
127 }
128
129 static int alloc_host_sq(struct c4iw_rdev *rdev, struct t4_sq *sq)
130 {
131 sq->queue = dma_alloc_coherent(&(rdev->lldi.pdev->dev), sq->memsize,
132 &(sq->dma_addr), GFP_KERNEL);
133 if (!sq->queue)
134 return -ENOMEM;
135 sq->phys_addr = virt_to_phys(sq->queue);
136 dma_unmap_addr_set(sq, mapping, sq->dma_addr);
137 return 0;
138 }
139
140 static int alloc_sq(struct c4iw_rdev *rdev, struct t4_sq *sq, int user)
141 {
142 int ret = -ENOSYS;
143 if (user)
144 ret = alloc_oc_sq(rdev, sq);
145 if (ret)
146 ret = alloc_host_sq(rdev, sq);
147 return ret;
148 }
149
150 static int destroy_qp(struct c4iw_rdev *rdev, struct t4_wq *wq,
151 struct c4iw_dev_ucontext *uctx, int has_rq)
152 {
153
154
155
156
157 dealloc_sq(rdev, &wq->sq);
158 kfree(wq->sq.sw_sq);
159 c4iw_put_qpid(rdev, wq->sq.qid, uctx);
160
161 if (has_rq) {
162 dma_free_coherent(&rdev->lldi.pdev->dev,
163 wq->rq.memsize, wq->rq.queue,
164 dma_unmap_addr(&wq->rq, mapping));
165 c4iw_rqtpool_free(rdev, wq->rq.rqt_hwaddr, wq->rq.rqt_size);
166 kfree(wq->rq.sw_rq);
167 c4iw_put_qpid(rdev, wq->rq.qid, uctx);
168 }
169 return 0;
170 }
171
172
173
174
175
176
177 void __iomem *c4iw_bar2_addrs(struct c4iw_rdev *rdev, unsigned int qid,
178 enum cxgb4_bar2_qtype qtype,
179 unsigned int *pbar2_qid, u64 *pbar2_pa)
180 {
181 u64 bar2_qoffset;
182 int ret;
183
184 ret = cxgb4_bar2_sge_qregs(rdev->lldi.ports[0], qid, qtype,
185 pbar2_pa ? 1 : 0,
186 &bar2_qoffset, pbar2_qid);
187 if (ret)
188 return NULL;
189
190 if (pbar2_pa)
191 *pbar2_pa = (rdev->bar2_pa + bar2_qoffset) & PAGE_MASK;
192
193 if (is_t4(rdev->lldi.adapter_type))
194 return NULL;
195
196 return rdev->bar2_kva + bar2_qoffset;
197 }
198
199 static int create_qp(struct c4iw_rdev *rdev, struct t4_wq *wq,
200 struct t4_cq *rcq, struct t4_cq *scq,
201 struct c4iw_dev_ucontext *uctx,
202 struct c4iw_wr_wait *wr_waitp,
203 int need_rq)
204 {
205 int user = (uctx != &rdev->uctx);
206 struct fw_ri_res_wr *res_wr;
207 struct fw_ri_res *res;
208 int wr_len;
209 struct sk_buff *skb;
210 int ret = 0;
211 int eqsize;
212
213 wq->sq.qid = c4iw_get_qpid(rdev, uctx);
214 if (!wq->sq.qid)
215 return -ENOMEM;
216
217 if (need_rq) {
218 wq->rq.qid = c4iw_get_qpid(rdev, uctx);
219 if (!wq->rq.qid) {
220 ret = -ENOMEM;
221 goto free_sq_qid;
222 }
223 }
224
225 if (!user) {
226 wq->sq.sw_sq = kcalloc(wq->sq.size, sizeof(*wq->sq.sw_sq),
227 GFP_KERNEL);
228 if (!wq->sq.sw_sq) {
229 ret = -ENOMEM;
230 goto free_rq_qid;
231 }
232
233 if (need_rq) {
234 wq->rq.sw_rq = kcalloc(wq->rq.size,
235 sizeof(*wq->rq.sw_rq),
236 GFP_KERNEL);
237 if (!wq->rq.sw_rq) {
238 ret = -ENOMEM;
239 goto free_sw_sq;
240 }
241 }
242 }
243
244 if (need_rq) {
245
246
247
248 wq->rq.rqt_size =
249 roundup_pow_of_two(max_t(u16, wq->rq.size, 16));
250 wq->rq.rqt_hwaddr = c4iw_rqtpool_alloc(rdev, wq->rq.rqt_size);
251 if (!wq->rq.rqt_hwaddr) {
252 ret = -ENOMEM;
253 goto free_sw_rq;
254 }
255 }
256
257 ret = alloc_sq(rdev, &wq->sq, user);
258 if (ret)
259 goto free_hwaddr;
260 memset(wq->sq.queue, 0, wq->sq.memsize);
261 dma_unmap_addr_set(&wq->sq, mapping, wq->sq.dma_addr);
262
263 if (need_rq) {
264 wq->rq.queue = dma_alloc_coherent(&rdev->lldi.pdev->dev,
265 wq->rq.memsize,
266 &wq->rq.dma_addr,
267 GFP_KERNEL);
268 if (!wq->rq.queue) {
269 ret = -ENOMEM;
270 goto free_sq;
271 }
272 pr_debug("sq base va 0x%p pa 0x%llx rq base va 0x%p pa 0x%llx\n",
273 wq->sq.queue,
274 (unsigned long long)virt_to_phys(wq->sq.queue),
275 wq->rq.queue,
276 (unsigned long long)virt_to_phys(wq->rq.queue));
277 dma_unmap_addr_set(&wq->rq, mapping, wq->rq.dma_addr);
278 }
279
280 wq->db = rdev->lldi.db_reg;
281
282 wq->sq.bar2_va = c4iw_bar2_addrs(rdev, wq->sq.qid,
283 CXGB4_BAR2_QTYPE_EGRESS,
284 &wq->sq.bar2_qid,
285 user ? &wq->sq.bar2_pa : NULL);
286 if (need_rq)
287 wq->rq.bar2_va = c4iw_bar2_addrs(rdev, wq->rq.qid,
288 CXGB4_BAR2_QTYPE_EGRESS,
289 &wq->rq.bar2_qid,
290 user ? &wq->rq.bar2_pa : NULL);
291
292
293
294
295 if (user && (!wq->sq.bar2_pa || (need_rq && !wq->rq.bar2_pa))) {
296 pr_warn("%s: sqid %u or rqid %u not in BAR2 range\n",
297 pci_name(rdev->lldi.pdev), wq->sq.qid, wq->rq.qid);
298 goto free_dma;
299 }
300
301 wq->rdev = rdev;
302 wq->rq.msn = 1;
303
304
305 wr_len = sizeof(*res_wr) + 2 * sizeof(*res);
306 if (need_rq)
307 wr_len += sizeof(*res);
308 skb = alloc_skb(wr_len, GFP_KERNEL);
309 if (!skb) {
310 ret = -ENOMEM;
311 goto free_dma;
312 }
313 set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
314
315 res_wr = __skb_put_zero(skb, wr_len);
316 res_wr->op_nres = cpu_to_be32(
317 FW_WR_OP_V(FW_RI_RES_WR) |
318 FW_RI_RES_WR_NRES_V(need_rq ? 2 : 1) |
319 FW_WR_COMPL_F);
320 res_wr->len16_pkd = cpu_to_be32(DIV_ROUND_UP(wr_len, 16));
321 res_wr->cookie = (uintptr_t)wr_waitp;
322 res = res_wr->res;
323 res->u.sqrq.restype = FW_RI_RES_TYPE_SQ;
324 res->u.sqrq.op = FW_RI_RES_OP_WRITE;
325
326
327
328
329 eqsize = wq->sq.size * T4_SQ_NUM_SLOTS +
330 rdev->hw_queue.t4_eq_status_entries;
331
332 res->u.sqrq.fetchszm_to_iqid = cpu_to_be32(
333 FW_RI_RES_WR_HOSTFCMODE_V(0) |
334 FW_RI_RES_WR_CPRIO_V(0) |
335 FW_RI_RES_WR_PCIECHN_V(0) |
336 (t4_sq_onchip(&wq->sq) ? FW_RI_RES_WR_ONCHIP_F : 0) |
337 FW_RI_RES_WR_IQID_V(scq->cqid));
338 res->u.sqrq.dcaen_to_eqsize = cpu_to_be32(
339 FW_RI_RES_WR_DCAEN_V(0) |
340 FW_RI_RES_WR_DCACPU_V(0) |
341 FW_RI_RES_WR_FBMIN_V(2) |
342 (t4_sq_onchip(&wq->sq) ? FW_RI_RES_WR_FBMAX_V(2) :
343 FW_RI_RES_WR_FBMAX_V(3)) |
344 FW_RI_RES_WR_CIDXFTHRESHO_V(0) |
345 FW_RI_RES_WR_CIDXFTHRESH_V(0) |
346 FW_RI_RES_WR_EQSIZE_V(eqsize));
347 res->u.sqrq.eqid = cpu_to_be32(wq->sq.qid);
348 res->u.sqrq.eqaddr = cpu_to_be64(wq->sq.dma_addr);
349
350 if (need_rq) {
351 res++;
352 res->u.sqrq.restype = FW_RI_RES_TYPE_RQ;
353 res->u.sqrq.op = FW_RI_RES_OP_WRITE;
354
355
356
357
358 eqsize = wq->rq.size * T4_RQ_NUM_SLOTS +
359 rdev->hw_queue.t4_eq_status_entries;
360 res->u.sqrq.fetchszm_to_iqid =
361
362 cpu_to_be32(FW_RI_RES_WR_HOSTFCMODE_V(0) |
363
364 FW_RI_RES_WR_CPRIO_V(0) |
365
366 FW_RI_RES_WR_PCIECHN_V(0) |
367 FW_RI_RES_WR_IQID_V(rcq->cqid));
368 res->u.sqrq.dcaen_to_eqsize =
369 cpu_to_be32(FW_RI_RES_WR_DCAEN_V(0) |
370 FW_RI_RES_WR_DCACPU_V(0) |
371 FW_RI_RES_WR_FBMIN_V(2) |
372 FW_RI_RES_WR_FBMAX_V(3) |
373 FW_RI_RES_WR_CIDXFTHRESHO_V(0) |
374 FW_RI_RES_WR_CIDXFTHRESH_V(0) |
375 FW_RI_RES_WR_EQSIZE_V(eqsize));
376 res->u.sqrq.eqid = cpu_to_be32(wq->rq.qid);
377 res->u.sqrq.eqaddr = cpu_to_be64(wq->rq.dma_addr);
378 }
379
380 c4iw_init_wr_wait(wr_waitp);
381 ret = c4iw_ref_send_wait(rdev, skb, wr_waitp, 0, wq->sq.qid, __func__);
382 if (ret)
383 goto free_dma;
384
385 pr_debug("sqid 0x%x rqid 0x%x kdb 0x%p sq_bar2_addr %p rq_bar2_addr %p\n",
386 wq->sq.qid, wq->rq.qid, wq->db,
387 wq->sq.bar2_va, wq->rq.bar2_va);
388
389 return 0;
390 free_dma:
391 if (need_rq)
392 dma_free_coherent(&rdev->lldi.pdev->dev,
393 wq->rq.memsize, wq->rq.queue,
394 dma_unmap_addr(&wq->rq, mapping));
395 free_sq:
396 dealloc_sq(rdev, &wq->sq);
397 free_hwaddr:
398 if (need_rq)
399 c4iw_rqtpool_free(rdev, wq->rq.rqt_hwaddr, wq->rq.rqt_size);
400 free_sw_rq:
401 if (need_rq)
402 kfree(wq->rq.sw_rq);
403 free_sw_sq:
404 kfree(wq->sq.sw_sq);
405 free_rq_qid:
406 if (need_rq)
407 c4iw_put_qpid(rdev, wq->rq.qid, uctx);
408 free_sq_qid:
409 c4iw_put_qpid(rdev, wq->sq.qid, uctx);
410 return ret;
411 }
412
413 static int build_immd(struct t4_sq *sq, struct fw_ri_immd *immdp,
414 const struct ib_send_wr *wr, int max, u32 *plenp)
415 {
416 u8 *dstp, *srcp;
417 u32 plen = 0;
418 int i;
419 int rem, len;
420
421 dstp = (u8 *)immdp->data;
422 for (i = 0; i < wr->num_sge; i++) {
423 if ((plen + wr->sg_list[i].length) > max)
424 return -EMSGSIZE;
425 srcp = (u8 *)(unsigned long)wr->sg_list[i].addr;
426 plen += wr->sg_list[i].length;
427 rem = wr->sg_list[i].length;
428 while (rem) {
429 if (dstp == (u8 *)&sq->queue[sq->size])
430 dstp = (u8 *)sq->queue;
431 if (rem <= (u8 *)&sq->queue[sq->size] - dstp)
432 len = rem;
433 else
434 len = (u8 *)&sq->queue[sq->size] - dstp;
435 memcpy(dstp, srcp, len);
436 dstp += len;
437 srcp += len;
438 rem -= len;
439 }
440 }
441 len = roundup(plen + sizeof(*immdp), 16) - (plen + sizeof(*immdp));
442 if (len)
443 memset(dstp, 0, len);
444 immdp->op = FW_RI_DATA_IMMD;
445 immdp->r1 = 0;
446 immdp->r2 = 0;
447 immdp->immdlen = cpu_to_be32(plen);
448 *plenp = plen;
449 return 0;
450 }
451
452 static int build_isgl(__be64 *queue_start, __be64 *queue_end,
453 struct fw_ri_isgl *isglp, struct ib_sge *sg_list,
454 int num_sge, u32 *plenp)
455
456 {
457 int i;
458 u32 plen = 0;
459 __be64 *flitp;
460
461 if ((__be64 *)isglp == queue_end)
462 isglp = (struct fw_ri_isgl *)queue_start;
463
464 flitp = (__be64 *)isglp->sge;
465
466 for (i = 0; i < num_sge; i++) {
467 if ((plen + sg_list[i].length) < plen)
468 return -EMSGSIZE;
469 plen += sg_list[i].length;
470 *flitp = cpu_to_be64(((u64)sg_list[i].lkey << 32) |
471 sg_list[i].length);
472 if (++flitp == queue_end)
473 flitp = queue_start;
474 *flitp = cpu_to_be64(sg_list[i].addr);
475 if (++flitp == queue_end)
476 flitp = queue_start;
477 }
478 *flitp = (__force __be64)0;
479 isglp->op = FW_RI_DATA_ISGL;
480 isglp->r1 = 0;
481 isglp->nsge = cpu_to_be16(num_sge);
482 isglp->r2 = 0;
483 if (plenp)
484 *plenp = plen;
485 return 0;
486 }
487
488 static int build_rdma_send(struct t4_sq *sq, union t4_wr *wqe,
489 const struct ib_send_wr *wr, u8 *len16)
490 {
491 u32 plen;
492 int size;
493 int ret;
494
495 if (wr->num_sge > T4_MAX_SEND_SGE)
496 return -EINVAL;
497 switch (wr->opcode) {
498 case IB_WR_SEND:
499 if (wr->send_flags & IB_SEND_SOLICITED)
500 wqe->send.sendop_pkd = cpu_to_be32(
501 FW_RI_SEND_WR_SENDOP_V(FW_RI_SEND_WITH_SE));
502 else
503 wqe->send.sendop_pkd = cpu_to_be32(
504 FW_RI_SEND_WR_SENDOP_V(FW_RI_SEND));
505 wqe->send.stag_inv = 0;
506 break;
507 case IB_WR_SEND_WITH_INV:
508 if (wr->send_flags & IB_SEND_SOLICITED)
509 wqe->send.sendop_pkd = cpu_to_be32(
510 FW_RI_SEND_WR_SENDOP_V(FW_RI_SEND_WITH_SE_INV));
511 else
512 wqe->send.sendop_pkd = cpu_to_be32(
513 FW_RI_SEND_WR_SENDOP_V(FW_RI_SEND_WITH_INV));
514 wqe->send.stag_inv = cpu_to_be32(wr->ex.invalidate_rkey);
515 break;
516
517 default:
518 return -EINVAL;
519 }
520 wqe->send.r3 = 0;
521 wqe->send.r4 = 0;
522
523 plen = 0;
524 if (wr->num_sge) {
525 if (wr->send_flags & IB_SEND_INLINE) {
526 ret = build_immd(sq, wqe->send.u.immd_src, wr,
527 T4_MAX_SEND_INLINE, &plen);
528 if (ret)
529 return ret;
530 size = sizeof(wqe->send) + sizeof(struct fw_ri_immd) +
531 plen;
532 } else {
533 ret = build_isgl((__be64 *)sq->queue,
534 (__be64 *)&sq->queue[sq->size],
535 wqe->send.u.isgl_src,
536 wr->sg_list, wr->num_sge, &plen);
537 if (ret)
538 return ret;
539 size = sizeof(wqe->send) + sizeof(struct fw_ri_isgl) +
540 wr->num_sge * sizeof(struct fw_ri_sge);
541 }
542 } else {
543 wqe->send.u.immd_src[0].op = FW_RI_DATA_IMMD;
544 wqe->send.u.immd_src[0].r1 = 0;
545 wqe->send.u.immd_src[0].r2 = 0;
546 wqe->send.u.immd_src[0].immdlen = 0;
547 size = sizeof(wqe->send) + sizeof(struct fw_ri_immd);
548 plen = 0;
549 }
550 *len16 = DIV_ROUND_UP(size, 16);
551 wqe->send.plen = cpu_to_be32(plen);
552 return 0;
553 }
554
555 static int build_rdma_write(struct t4_sq *sq, union t4_wr *wqe,
556 const struct ib_send_wr *wr, u8 *len16)
557 {
558 u32 plen;
559 int size;
560 int ret;
561
562 if (wr->num_sge > T4_MAX_SEND_SGE)
563 return -EINVAL;
564
565
566
567
568
569 if (wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM)
570 wqe->write.iw_imm_data.ib_imm_data.imm_data32 = wr->ex.imm_data;
571 else
572 wqe->write.iw_imm_data.ib_imm_data.imm_data32 = 0;
573 wqe->write.stag_sink = cpu_to_be32(rdma_wr(wr)->rkey);
574 wqe->write.to_sink = cpu_to_be64(rdma_wr(wr)->remote_addr);
575 if (wr->num_sge) {
576 if (wr->send_flags & IB_SEND_INLINE) {
577 ret = build_immd(sq, wqe->write.u.immd_src, wr,
578 T4_MAX_WRITE_INLINE, &plen);
579 if (ret)
580 return ret;
581 size = sizeof(wqe->write) + sizeof(struct fw_ri_immd) +
582 plen;
583 } else {
584 ret = build_isgl((__be64 *)sq->queue,
585 (__be64 *)&sq->queue[sq->size],
586 wqe->write.u.isgl_src,
587 wr->sg_list, wr->num_sge, &plen);
588 if (ret)
589 return ret;
590 size = sizeof(wqe->write) + sizeof(struct fw_ri_isgl) +
591 wr->num_sge * sizeof(struct fw_ri_sge);
592 }
593 } else {
594 wqe->write.u.immd_src[0].op = FW_RI_DATA_IMMD;
595 wqe->write.u.immd_src[0].r1 = 0;
596 wqe->write.u.immd_src[0].r2 = 0;
597 wqe->write.u.immd_src[0].immdlen = 0;
598 size = sizeof(wqe->write) + sizeof(struct fw_ri_immd);
599 plen = 0;
600 }
601 *len16 = DIV_ROUND_UP(size, 16);
602 wqe->write.plen = cpu_to_be32(plen);
603 return 0;
604 }
605
606 static void build_immd_cmpl(struct t4_sq *sq, struct fw_ri_immd_cmpl *immdp,
607 struct ib_send_wr *wr)
608 {
609 memcpy((u8 *)immdp->data, (u8 *)(uintptr_t)wr->sg_list->addr, 16);
610 memset(immdp->r1, 0, 6);
611 immdp->op = FW_RI_DATA_IMMD;
612 immdp->immdlen = 16;
613 }
614
615 static void build_rdma_write_cmpl(struct t4_sq *sq,
616 struct fw_ri_rdma_write_cmpl_wr *wcwr,
617 const struct ib_send_wr *wr, u8 *len16)
618 {
619 u32 plen;
620 int size;
621
622
623
624
625
626
627
628
629
630
631 BUILD_BUG_ON(offsetof(struct fw_ri_rdma_write_cmpl_wr, u) > 64);
632
633 wcwr->stag_sink = cpu_to_be32(rdma_wr(wr)->rkey);
634 wcwr->to_sink = cpu_to_be64(rdma_wr(wr)->remote_addr);
635 if (wr->next->opcode == IB_WR_SEND)
636 wcwr->stag_inv = 0;
637 else
638 wcwr->stag_inv = cpu_to_be32(wr->next->ex.invalidate_rkey);
639 wcwr->r2 = 0;
640 wcwr->r3 = 0;
641
642
643 if (wr->next->send_flags & IB_SEND_INLINE)
644 build_immd_cmpl(sq, &wcwr->u_cmpl.immd_src, wr->next);
645 else
646 build_isgl((__be64 *)sq->queue, (__be64 *)&sq->queue[sq->size],
647 &wcwr->u_cmpl.isgl_src, wr->next->sg_list, 1, NULL);
648
649
650 build_isgl((__be64 *)sq->queue, (__be64 *)&sq->queue[sq->size],
651 wcwr->u.isgl_src, wr->sg_list, wr->num_sge, &plen);
652
653 size = sizeof(*wcwr) + sizeof(struct fw_ri_isgl) +
654 wr->num_sge * sizeof(struct fw_ri_sge);
655 wcwr->plen = cpu_to_be32(plen);
656 *len16 = DIV_ROUND_UP(size, 16);
657 }
658
659 static int build_rdma_read(union t4_wr *wqe, const struct ib_send_wr *wr,
660 u8 *len16)
661 {
662 if (wr->num_sge > 1)
663 return -EINVAL;
664 if (wr->num_sge && wr->sg_list[0].length) {
665 wqe->read.stag_src = cpu_to_be32(rdma_wr(wr)->rkey);
666 wqe->read.to_src_hi = cpu_to_be32((u32)(rdma_wr(wr)->remote_addr
667 >> 32));
668 wqe->read.to_src_lo = cpu_to_be32((u32)rdma_wr(wr)->remote_addr);
669 wqe->read.stag_sink = cpu_to_be32(wr->sg_list[0].lkey);
670 wqe->read.plen = cpu_to_be32(wr->sg_list[0].length);
671 wqe->read.to_sink_hi = cpu_to_be32((u32)(wr->sg_list[0].addr
672 >> 32));
673 wqe->read.to_sink_lo = cpu_to_be32((u32)(wr->sg_list[0].addr));
674 } else {
675 wqe->read.stag_src = cpu_to_be32(2);
676 wqe->read.to_src_hi = 0;
677 wqe->read.to_src_lo = 0;
678 wqe->read.stag_sink = cpu_to_be32(2);
679 wqe->read.plen = 0;
680 wqe->read.to_sink_hi = 0;
681 wqe->read.to_sink_lo = 0;
682 }
683 wqe->read.r2 = 0;
684 wqe->read.r5 = 0;
685 *len16 = DIV_ROUND_UP(sizeof(wqe->read), 16);
686 return 0;
687 }
688
689 static void post_write_cmpl(struct c4iw_qp *qhp, const struct ib_send_wr *wr)
690 {
691 bool send_signaled = (wr->next->send_flags & IB_SEND_SIGNALED) ||
692 qhp->sq_sig_all;
693 bool write_signaled = (wr->send_flags & IB_SEND_SIGNALED) ||
694 qhp->sq_sig_all;
695 struct t4_swsqe *swsqe;
696 union t4_wr *wqe;
697 u16 write_wrid;
698 u8 len16;
699 u16 idx;
700
701
702
703
704
705 wqe = (union t4_wr *)((u8 *)qhp->wq.sq.queue +
706 qhp->wq.sq.wq_pidx * T4_EQ_ENTRY_SIZE);
707 build_rdma_write_cmpl(&qhp->wq.sq, &wqe->write_cmpl, wr, &len16);
708
709
710 swsqe = &qhp->wq.sq.sw_sq[qhp->wq.sq.pidx];
711 swsqe->opcode = FW_RI_RDMA_WRITE;
712 swsqe->idx = qhp->wq.sq.pidx;
713 swsqe->complete = 0;
714 swsqe->signaled = write_signaled;
715 swsqe->flushed = 0;
716 swsqe->wr_id = wr->wr_id;
717 if (c4iw_wr_log) {
718 swsqe->sge_ts =
719 cxgb4_read_sge_timestamp(qhp->rhp->rdev.lldi.ports[0]);
720 swsqe->host_time = ktime_get();
721 }
722
723 write_wrid = qhp->wq.sq.pidx;
724
725
726 qhp->wq.sq.in_use++;
727 if (++qhp->wq.sq.pidx == qhp->wq.sq.size)
728 qhp->wq.sq.pidx = 0;
729
730
731 swsqe = &qhp->wq.sq.sw_sq[qhp->wq.sq.pidx];
732 if (wr->next->opcode == IB_WR_SEND)
733 swsqe->opcode = FW_RI_SEND;
734 else
735 swsqe->opcode = FW_RI_SEND_WITH_INV;
736 swsqe->idx = qhp->wq.sq.pidx;
737 swsqe->complete = 0;
738 swsqe->signaled = send_signaled;
739 swsqe->flushed = 0;
740 swsqe->wr_id = wr->next->wr_id;
741 if (c4iw_wr_log) {
742 swsqe->sge_ts =
743 cxgb4_read_sge_timestamp(qhp->rhp->rdev.lldi.ports[0]);
744 swsqe->host_time = ktime_get();
745 }
746
747 wqe->write_cmpl.flags_send = send_signaled ? FW_RI_COMPLETION_FLAG : 0;
748 wqe->write_cmpl.wrid_send = qhp->wq.sq.pidx;
749
750 init_wr_hdr(wqe, write_wrid, FW_RI_RDMA_WRITE_CMPL_WR,
751 write_signaled ? FW_RI_COMPLETION_FLAG : 0, len16);
752 t4_sq_produce(&qhp->wq, len16);
753 idx = DIV_ROUND_UP(len16 * 16, T4_EQ_ENTRY_SIZE);
754
755 t4_ring_sq_db(&qhp->wq, idx, wqe);
756 }
757
758 static int build_rdma_recv(struct c4iw_qp *qhp, union t4_recv_wr *wqe,
759 const struct ib_recv_wr *wr, u8 *len16)
760 {
761 int ret;
762
763 ret = build_isgl((__be64 *)qhp->wq.rq.queue,
764 (__be64 *)&qhp->wq.rq.queue[qhp->wq.rq.size],
765 &wqe->recv.isgl, wr->sg_list, wr->num_sge, NULL);
766 if (ret)
767 return ret;
768 *len16 = DIV_ROUND_UP(
769 sizeof(wqe->recv) + wr->num_sge * sizeof(struct fw_ri_sge), 16);
770 return 0;
771 }
772
773 static int build_srq_recv(union t4_recv_wr *wqe, const struct ib_recv_wr *wr,
774 u8 *len16)
775 {
776 int ret;
777
778 ret = build_isgl((__be64 *)wqe, (__be64 *)(wqe + 1),
779 &wqe->recv.isgl, wr->sg_list, wr->num_sge, NULL);
780 if (ret)
781 return ret;
782 *len16 = DIV_ROUND_UP(sizeof(wqe->recv) +
783 wr->num_sge * sizeof(struct fw_ri_sge), 16);
784 return 0;
785 }
786
787 static void build_tpte_memreg(struct fw_ri_fr_nsmr_tpte_wr *fr,
788 const struct ib_reg_wr *wr, struct c4iw_mr *mhp,
789 u8 *len16)
790 {
791 __be64 *p = (__be64 *)fr->pbl;
792
793 fr->r2 = cpu_to_be32(0);
794 fr->stag = cpu_to_be32(mhp->ibmr.rkey);
795
796 fr->tpte.valid_to_pdid = cpu_to_be32(FW_RI_TPTE_VALID_F |
797 FW_RI_TPTE_STAGKEY_V((mhp->ibmr.rkey & FW_RI_TPTE_STAGKEY_M)) |
798 FW_RI_TPTE_STAGSTATE_V(1) |
799 FW_RI_TPTE_STAGTYPE_V(FW_RI_STAG_NSMR) |
800 FW_RI_TPTE_PDID_V(mhp->attr.pdid));
801 fr->tpte.locread_to_qpid = cpu_to_be32(
802 FW_RI_TPTE_PERM_V(c4iw_ib_to_tpt_access(wr->access)) |
803 FW_RI_TPTE_ADDRTYPE_V(FW_RI_VA_BASED_TO) |
804 FW_RI_TPTE_PS_V(ilog2(wr->mr->page_size) - 12));
805 fr->tpte.nosnoop_pbladdr = cpu_to_be32(FW_RI_TPTE_PBLADDR_V(
806 PBL_OFF(&mhp->rhp->rdev, mhp->attr.pbl_addr)>>3));
807 fr->tpte.dca_mwbcnt_pstag = cpu_to_be32(0);
808 fr->tpte.len_hi = cpu_to_be32(0);
809 fr->tpte.len_lo = cpu_to_be32(mhp->ibmr.length);
810 fr->tpte.va_hi = cpu_to_be32(mhp->ibmr.iova >> 32);
811 fr->tpte.va_lo_fbo = cpu_to_be32(mhp->ibmr.iova & 0xffffffff);
812
813 p[0] = cpu_to_be64((u64)mhp->mpl[0]);
814 p[1] = cpu_to_be64((u64)mhp->mpl[1]);
815
816 *len16 = DIV_ROUND_UP(sizeof(*fr), 16);
817 }
818
819 static int build_memreg(struct t4_sq *sq, union t4_wr *wqe,
820 const struct ib_reg_wr *wr, struct c4iw_mr *mhp,
821 u8 *len16, bool dsgl_supported)
822 {
823 struct fw_ri_immd *imdp;
824 __be64 *p;
825 int i;
826 int pbllen = roundup(mhp->mpl_len * sizeof(u64), 32);
827 int rem;
828
829 if (mhp->mpl_len > t4_max_fr_depth(dsgl_supported && use_dsgl))
830 return -EINVAL;
831
832 wqe->fr.qpbinde_to_dcacpu = 0;
833 wqe->fr.pgsz_shift = ilog2(wr->mr->page_size) - 12;
834 wqe->fr.addr_type = FW_RI_VA_BASED_TO;
835 wqe->fr.mem_perms = c4iw_ib_to_tpt_access(wr->access);
836 wqe->fr.len_hi = 0;
837 wqe->fr.len_lo = cpu_to_be32(mhp->ibmr.length);
838 wqe->fr.stag = cpu_to_be32(wr->key);
839 wqe->fr.va_hi = cpu_to_be32(mhp->ibmr.iova >> 32);
840 wqe->fr.va_lo_fbo = cpu_to_be32(mhp->ibmr.iova &
841 0xffffffff);
842
843 if (dsgl_supported && use_dsgl && (pbllen > max_fr_immd)) {
844 struct fw_ri_dsgl *sglp;
845
846 for (i = 0; i < mhp->mpl_len; i++)
847 mhp->mpl[i] = (__force u64)cpu_to_be64((u64)mhp->mpl[i]);
848
849 sglp = (struct fw_ri_dsgl *)(&wqe->fr + 1);
850 sglp->op = FW_RI_DATA_DSGL;
851 sglp->r1 = 0;
852 sglp->nsge = cpu_to_be16(1);
853 sglp->addr0 = cpu_to_be64(mhp->mpl_addr);
854 sglp->len0 = cpu_to_be32(pbllen);
855
856 *len16 = DIV_ROUND_UP(sizeof(wqe->fr) + sizeof(*sglp), 16);
857 } else {
858 imdp = (struct fw_ri_immd *)(&wqe->fr + 1);
859 imdp->op = FW_RI_DATA_IMMD;
860 imdp->r1 = 0;
861 imdp->r2 = 0;
862 imdp->immdlen = cpu_to_be32(pbllen);
863 p = (__be64 *)(imdp + 1);
864 rem = pbllen;
865 for (i = 0; i < mhp->mpl_len; i++) {
866 *p = cpu_to_be64((u64)mhp->mpl[i]);
867 rem -= sizeof(*p);
868 if (++p == (__be64 *)&sq->queue[sq->size])
869 p = (__be64 *)sq->queue;
870 }
871 while (rem) {
872 *p = 0;
873 rem -= sizeof(*p);
874 if (++p == (__be64 *)&sq->queue[sq->size])
875 p = (__be64 *)sq->queue;
876 }
877 *len16 = DIV_ROUND_UP(sizeof(wqe->fr) + sizeof(*imdp)
878 + pbllen, 16);
879 }
880 return 0;
881 }
882
883 static int build_inv_stag(union t4_wr *wqe, const struct ib_send_wr *wr,
884 u8 *len16)
885 {
886 wqe->inv.stag_inv = cpu_to_be32(wr->ex.invalidate_rkey);
887 wqe->inv.r2 = 0;
888 *len16 = DIV_ROUND_UP(sizeof(wqe->inv), 16);
889 return 0;
890 }
891
892 void c4iw_qp_add_ref(struct ib_qp *qp)
893 {
894 pr_debug("ib_qp %p\n", qp);
895 refcount_inc(&to_c4iw_qp(qp)->qp_refcnt);
896 }
897
898 void c4iw_qp_rem_ref(struct ib_qp *qp)
899 {
900 pr_debug("ib_qp %p\n", qp);
901 if (refcount_dec_and_test(&to_c4iw_qp(qp)->qp_refcnt))
902 complete(&to_c4iw_qp(qp)->qp_rel_comp);
903 }
904
905 static void add_to_fc_list(struct list_head *head, struct list_head *entry)
906 {
907 if (list_empty(entry))
908 list_add_tail(entry, head);
909 }
910
911 static int ring_kernel_sq_db(struct c4iw_qp *qhp, u16 inc)
912 {
913 unsigned long flags;
914
915 xa_lock_irqsave(&qhp->rhp->qps, flags);
916 spin_lock(&qhp->lock);
917 if (qhp->rhp->db_state == NORMAL)
918 t4_ring_sq_db(&qhp->wq, inc, NULL);
919 else {
920 add_to_fc_list(&qhp->rhp->db_fc_list, &qhp->db_fc_entry);
921 qhp->wq.sq.wq_pidx_inc += inc;
922 }
923 spin_unlock(&qhp->lock);
924 xa_unlock_irqrestore(&qhp->rhp->qps, flags);
925 return 0;
926 }
927
928 static int ring_kernel_rq_db(struct c4iw_qp *qhp, u16 inc)
929 {
930 unsigned long flags;
931
932 xa_lock_irqsave(&qhp->rhp->qps, flags);
933 spin_lock(&qhp->lock);
934 if (qhp->rhp->db_state == NORMAL)
935 t4_ring_rq_db(&qhp->wq, inc, NULL);
936 else {
937 add_to_fc_list(&qhp->rhp->db_fc_list, &qhp->db_fc_entry);
938 qhp->wq.rq.wq_pidx_inc += inc;
939 }
940 spin_unlock(&qhp->lock);
941 xa_unlock_irqrestore(&qhp->rhp->qps, flags);
942 return 0;
943 }
944
945 static int ib_to_fw_opcode(int ib_opcode)
946 {
947 int opcode;
948
949 switch (ib_opcode) {
950 case IB_WR_SEND_WITH_INV:
951 opcode = FW_RI_SEND_WITH_INV;
952 break;
953 case IB_WR_SEND:
954 opcode = FW_RI_SEND;
955 break;
956 case IB_WR_RDMA_WRITE:
957 opcode = FW_RI_RDMA_WRITE;
958 break;
959 case IB_WR_RDMA_WRITE_WITH_IMM:
960 opcode = FW_RI_WRITE_IMMEDIATE;
961 break;
962 case IB_WR_RDMA_READ:
963 case IB_WR_RDMA_READ_WITH_INV:
964 opcode = FW_RI_READ_REQ;
965 break;
966 case IB_WR_REG_MR:
967 opcode = FW_RI_FAST_REGISTER;
968 break;
969 case IB_WR_LOCAL_INV:
970 opcode = FW_RI_LOCAL_INV;
971 break;
972 default:
973 opcode = -EINVAL;
974 }
975 return opcode;
976 }
977
978 static int complete_sq_drain_wr(struct c4iw_qp *qhp,
979 const struct ib_send_wr *wr)
980 {
981 struct t4_cqe cqe = {};
982 struct c4iw_cq *schp;
983 unsigned long flag;
984 struct t4_cq *cq;
985 int opcode;
986
987 schp = to_c4iw_cq(qhp->ibqp.send_cq);
988 cq = &schp->cq;
989
990 opcode = ib_to_fw_opcode(wr->opcode);
991 if (opcode < 0)
992 return opcode;
993
994 cqe.u.drain_cookie = wr->wr_id;
995 cqe.header = cpu_to_be32(CQE_STATUS_V(T4_ERR_SWFLUSH) |
996 CQE_OPCODE_V(opcode) |
997 CQE_TYPE_V(1) |
998 CQE_SWCQE_V(1) |
999 CQE_DRAIN_V(1) |
1000 CQE_QPID_V(qhp->wq.sq.qid));
1001
1002 spin_lock_irqsave(&schp->lock, flag);
1003 cqe.bits_type_ts = cpu_to_be64(CQE_GENBIT_V((u64)cq->gen));
1004 cq->sw_queue[cq->sw_pidx] = cqe;
1005 t4_swcq_produce(cq);
1006 spin_unlock_irqrestore(&schp->lock, flag);
1007
1008 if (t4_clear_cq_armed(&schp->cq)) {
1009 spin_lock_irqsave(&schp->comp_handler_lock, flag);
1010 (*schp->ibcq.comp_handler)(&schp->ibcq,
1011 schp->ibcq.cq_context);
1012 spin_unlock_irqrestore(&schp->comp_handler_lock, flag);
1013 }
1014 return 0;
1015 }
1016
1017 static int complete_sq_drain_wrs(struct c4iw_qp *qhp,
1018 const struct ib_send_wr *wr,
1019 const struct ib_send_wr **bad_wr)
1020 {
1021 int ret = 0;
1022
1023 while (wr) {
1024 ret = complete_sq_drain_wr(qhp, wr);
1025 if (ret) {
1026 *bad_wr = wr;
1027 break;
1028 }
1029 wr = wr->next;
1030 }
1031 return ret;
1032 }
1033
1034 static void complete_rq_drain_wr(struct c4iw_qp *qhp,
1035 const struct ib_recv_wr *wr)
1036 {
1037 struct t4_cqe cqe = {};
1038 struct c4iw_cq *rchp;
1039 unsigned long flag;
1040 struct t4_cq *cq;
1041
1042 rchp = to_c4iw_cq(qhp->ibqp.recv_cq);
1043 cq = &rchp->cq;
1044
1045 cqe.u.drain_cookie = wr->wr_id;
1046 cqe.header = cpu_to_be32(CQE_STATUS_V(T4_ERR_SWFLUSH) |
1047 CQE_OPCODE_V(FW_RI_SEND) |
1048 CQE_TYPE_V(0) |
1049 CQE_SWCQE_V(1) |
1050 CQE_DRAIN_V(1) |
1051 CQE_QPID_V(qhp->wq.sq.qid));
1052
1053 spin_lock_irqsave(&rchp->lock, flag);
1054 cqe.bits_type_ts = cpu_to_be64(CQE_GENBIT_V((u64)cq->gen));
1055 cq->sw_queue[cq->sw_pidx] = cqe;
1056 t4_swcq_produce(cq);
1057 spin_unlock_irqrestore(&rchp->lock, flag);
1058
1059 if (t4_clear_cq_armed(&rchp->cq)) {
1060 spin_lock_irqsave(&rchp->comp_handler_lock, flag);
1061 (*rchp->ibcq.comp_handler)(&rchp->ibcq,
1062 rchp->ibcq.cq_context);
1063 spin_unlock_irqrestore(&rchp->comp_handler_lock, flag);
1064 }
1065 }
1066
1067 static void complete_rq_drain_wrs(struct c4iw_qp *qhp,
1068 const struct ib_recv_wr *wr)
1069 {
1070 while (wr) {
1071 complete_rq_drain_wr(qhp, wr);
1072 wr = wr->next;
1073 }
1074 }
1075
1076 int c4iw_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
1077 const struct ib_send_wr **bad_wr)
1078 {
1079 int err = 0;
1080 u8 len16 = 0;
1081 enum fw_wr_opcodes fw_opcode = 0;
1082 enum fw_ri_wr_flags fw_flags;
1083 struct c4iw_qp *qhp;
1084 struct c4iw_dev *rhp;
1085 union t4_wr *wqe = NULL;
1086 u32 num_wrs;
1087 struct t4_swsqe *swsqe;
1088 unsigned long flag;
1089 u16 idx = 0;
1090
1091 qhp = to_c4iw_qp(ibqp);
1092 rhp = qhp->rhp;
1093 spin_lock_irqsave(&qhp->lock, flag);
1094
1095
1096
1097
1098
1099 if (qhp->wq.flushed) {
1100 spin_unlock_irqrestore(&qhp->lock, flag);
1101 err = complete_sq_drain_wrs(qhp, wr, bad_wr);
1102 return err;
1103 }
1104 num_wrs = t4_sq_avail(&qhp->wq);
1105 if (num_wrs == 0) {
1106 spin_unlock_irqrestore(&qhp->lock, flag);
1107 *bad_wr = wr;
1108 return -ENOMEM;
1109 }
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120 if (qhp->rhp->rdev.lldi.write_cmpl_support &&
1121 CHELSIO_CHIP_VERSION(qhp->rhp->rdev.lldi.adapter_type) >=
1122 CHELSIO_T5 &&
1123 wr && wr->next && !wr->next->next &&
1124 wr->opcode == IB_WR_RDMA_WRITE &&
1125 wr->sg_list[0].length && wr->num_sge <= T4_WRITE_CMPL_MAX_SGL &&
1126 (wr->next->opcode == IB_WR_SEND ||
1127 wr->next->opcode == IB_WR_SEND_WITH_INV) &&
1128 wr->next->sg_list[0].length == T4_WRITE_CMPL_MAX_CQE &&
1129 wr->next->num_sge == 1 && num_wrs >= 2) {
1130 post_write_cmpl(qhp, wr);
1131 spin_unlock_irqrestore(&qhp->lock, flag);
1132 return 0;
1133 }
1134
1135 while (wr) {
1136 if (num_wrs == 0) {
1137 err = -ENOMEM;
1138 *bad_wr = wr;
1139 break;
1140 }
1141 wqe = (union t4_wr *)((u8 *)qhp->wq.sq.queue +
1142 qhp->wq.sq.wq_pidx * T4_EQ_ENTRY_SIZE);
1143
1144 fw_flags = 0;
1145 if (wr->send_flags & IB_SEND_SOLICITED)
1146 fw_flags |= FW_RI_SOLICITED_EVENT_FLAG;
1147 if (wr->send_flags & IB_SEND_SIGNALED || qhp->sq_sig_all)
1148 fw_flags |= FW_RI_COMPLETION_FLAG;
1149 swsqe = &qhp->wq.sq.sw_sq[qhp->wq.sq.pidx];
1150 switch (wr->opcode) {
1151 case IB_WR_SEND_WITH_INV:
1152 case IB_WR_SEND:
1153 if (wr->send_flags & IB_SEND_FENCE)
1154 fw_flags |= FW_RI_READ_FENCE_FLAG;
1155 fw_opcode = FW_RI_SEND_WR;
1156 if (wr->opcode == IB_WR_SEND)
1157 swsqe->opcode = FW_RI_SEND;
1158 else
1159 swsqe->opcode = FW_RI_SEND_WITH_INV;
1160 err = build_rdma_send(&qhp->wq.sq, wqe, wr, &len16);
1161 break;
1162 case IB_WR_RDMA_WRITE_WITH_IMM:
1163 if (unlikely(!rhp->rdev.lldi.write_w_imm_support)) {
1164 err = -EINVAL;
1165 break;
1166 }
1167 fw_flags |= FW_RI_RDMA_WRITE_WITH_IMMEDIATE;
1168
1169 case IB_WR_RDMA_WRITE:
1170 fw_opcode = FW_RI_RDMA_WRITE_WR;
1171 swsqe->opcode = FW_RI_RDMA_WRITE;
1172 err = build_rdma_write(&qhp->wq.sq, wqe, wr, &len16);
1173 break;
1174 case IB_WR_RDMA_READ:
1175 case IB_WR_RDMA_READ_WITH_INV:
1176 fw_opcode = FW_RI_RDMA_READ_WR;
1177 swsqe->opcode = FW_RI_READ_REQ;
1178 if (wr->opcode == IB_WR_RDMA_READ_WITH_INV) {
1179 c4iw_invalidate_mr(rhp, wr->sg_list[0].lkey);
1180 fw_flags = FW_RI_RDMA_READ_INVALIDATE;
1181 } else {
1182 fw_flags = 0;
1183 }
1184 err = build_rdma_read(wqe, wr, &len16);
1185 if (err)
1186 break;
1187 swsqe->read_len = wr->sg_list[0].length;
1188 if (!qhp->wq.sq.oldest_read)
1189 qhp->wq.sq.oldest_read = swsqe;
1190 break;
1191 case IB_WR_REG_MR: {
1192 struct c4iw_mr *mhp = to_c4iw_mr(reg_wr(wr)->mr);
1193
1194 swsqe->opcode = FW_RI_FAST_REGISTER;
1195 if (rhp->rdev.lldi.fr_nsmr_tpte_wr_support &&
1196 !mhp->attr.state && mhp->mpl_len <= 2) {
1197 fw_opcode = FW_RI_FR_NSMR_TPTE_WR;
1198 build_tpte_memreg(&wqe->fr_tpte, reg_wr(wr),
1199 mhp, &len16);
1200 } else {
1201 fw_opcode = FW_RI_FR_NSMR_WR;
1202 err = build_memreg(&qhp->wq.sq, wqe, reg_wr(wr),
1203 mhp, &len16,
1204 rhp->rdev.lldi.ulptx_memwrite_dsgl);
1205 if (err)
1206 break;
1207 }
1208 mhp->attr.state = 1;
1209 break;
1210 }
1211 case IB_WR_LOCAL_INV:
1212 if (wr->send_flags & IB_SEND_FENCE)
1213 fw_flags |= FW_RI_LOCAL_FENCE_FLAG;
1214 fw_opcode = FW_RI_INV_LSTAG_WR;
1215 swsqe->opcode = FW_RI_LOCAL_INV;
1216 err = build_inv_stag(wqe, wr, &len16);
1217 c4iw_invalidate_mr(rhp, wr->ex.invalidate_rkey);
1218 break;
1219 default:
1220 pr_warn("%s post of type=%d TBD!\n", __func__,
1221 wr->opcode);
1222 err = -EINVAL;
1223 }
1224 if (err) {
1225 *bad_wr = wr;
1226 break;
1227 }
1228 swsqe->idx = qhp->wq.sq.pidx;
1229 swsqe->complete = 0;
1230 swsqe->signaled = (wr->send_flags & IB_SEND_SIGNALED) ||
1231 qhp->sq_sig_all;
1232 swsqe->flushed = 0;
1233 swsqe->wr_id = wr->wr_id;
1234 if (c4iw_wr_log) {
1235 swsqe->sge_ts = cxgb4_read_sge_timestamp(
1236 rhp->rdev.lldi.ports[0]);
1237 swsqe->host_time = ktime_get();
1238 }
1239
1240 init_wr_hdr(wqe, qhp->wq.sq.pidx, fw_opcode, fw_flags, len16);
1241
1242 pr_debug("cookie 0x%llx pidx 0x%x opcode 0x%x read_len %u\n",
1243 (unsigned long long)wr->wr_id, qhp->wq.sq.pidx,
1244 swsqe->opcode, swsqe->read_len);
1245 wr = wr->next;
1246 num_wrs--;
1247 t4_sq_produce(&qhp->wq, len16);
1248 idx += DIV_ROUND_UP(len16*16, T4_EQ_ENTRY_SIZE);
1249 }
1250 if (!rhp->rdev.status_page->db_off) {
1251 t4_ring_sq_db(&qhp->wq, idx, wqe);
1252 spin_unlock_irqrestore(&qhp->lock, flag);
1253 } else {
1254 spin_unlock_irqrestore(&qhp->lock, flag);
1255 ring_kernel_sq_db(qhp, idx);
1256 }
1257 return err;
1258 }
1259
1260 int c4iw_post_receive(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
1261 const struct ib_recv_wr **bad_wr)
1262 {
1263 int err = 0;
1264 struct c4iw_qp *qhp;
1265 union t4_recv_wr *wqe = NULL;
1266 u32 num_wrs;
1267 u8 len16 = 0;
1268 unsigned long flag;
1269 u16 idx = 0;
1270
1271 qhp = to_c4iw_qp(ibqp);
1272 spin_lock_irqsave(&qhp->lock, flag);
1273
1274
1275
1276
1277
1278 if (qhp->wq.flushed) {
1279 spin_unlock_irqrestore(&qhp->lock, flag);
1280 complete_rq_drain_wrs(qhp, wr);
1281 return err;
1282 }
1283 num_wrs = t4_rq_avail(&qhp->wq);
1284 if (num_wrs == 0) {
1285 spin_unlock_irqrestore(&qhp->lock, flag);
1286 *bad_wr = wr;
1287 return -ENOMEM;
1288 }
1289 while (wr) {
1290 if (wr->num_sge > T4_MAX_RECV_SGE) {
1291 err = -EINVAL;
1292 *bad_wr = wr;
1293 break;
1294 }
1295 wqe = (union t4_recv_wr *)((u8 *)qhp->wq.rq.queue +
1296 qhp->wq.rq.wq_pidx *
1297 T4_EQ_ENTRY_SIZE);
1298 if (num_wrs)
1299 err = build_rdma_recv(qhp, wqe, wr, &len16);
1300 else
1301 err = -ENOMEM;
1302 if (err) {
1303 *bad_wr = wr;
1304 break;
1305 }
1306
1307 qhp->wq.rq.sw_rq[qhp->wq.rq.pidx].wr_id = wr->wr_id;
1308 if (c4iw_wr_log) {
1309 qhp->wq.rq.sw_rq[qhp->wq.rq.pidx].sge_ts =
1310 cxgb4_read_sge_timestamp(
1311 qhp->rhp->rdev.lldi.ports[0]);
1312 qhp->wq.rq.sw_rq[qhp->wq.rq.pidx].host_time =
1313 ktime_get();
1314 }
1315
1316 wqe->recv.opcode = FW_RI_RECV_WR;
1317 wqe->recv.r1 = 0;
1318 wqe->recv.wrid = qhp->wq.rq.pidx;
1319 wqe->recv.r2[0] = 0;
1320 wqe->recv.r2[1] = 0;
1321 wqe->recv.r2[2] = 0;
1322 wqe->recv.len16 = len16;
1323 pr_debug("cookie 0x%llx pidx %u\n",
1324 (unsigned long long)wr->wr_id, qhp->wq.rq.pidx);
1325 t4_rq_produce(&qhp->wq, len16);
1326 idx += DIV_ROUND_UP(len16*16, T4_EQ_ENTRY_SIZE);
1327 wr = wr->next;
1328 num_wrs--;
1329 }
1330 if (!qhp->rhp->rdev.status_page->db_off) {
1331 t4_ring_rq_db(&qhp->wq, idx, wqe);
1332 spin_unlock_irqrestore(&qhp->lock, flag);
1333 } else {
1334 spin_unlock_irqrestore(&qhp->lock, flag);
1335 ring_kernel_rq_db(qhp, idx);
1336 }
1337 return err;
1338 }
1339
1340 static void defer_srq_wr(struct t4_srq *srq, union t4_recv_wr *wqe,
1341 u64 wr_id, u8 len16)
1342 {
1343 struct t4_srq_pending_wr *pwr = &srq->pending_wrs[srq->pending_pidx];
1344
1345 pr_debug("%s cidx %u pidx %u wq_pidx %u in_use %u ooo_count %u wr_id 0x%llx pending_cidx %u pending_pidx %u pending_in_use %u\n",
1346 __func__, srq->cidx, srq->pidx, srq->wq_pidx,
1347 srq->in_use, srq->ooo_count,
1348 (unsigned long long)wr_id, srq->pending_cidx,
1349 srq->pending_pidx, srq->pending_in_use);
1350 pwr->wr_id = wr_id;
1351 pwr->len16 = len16;
1352 memcpy(&pwr->wqe, wqe, len16 * 16);
1353 t4_srq_produce_pending_wr(srq);
1354 }
1355
1356 int c4iw_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
1357 const struct ib_recv_wr **bad_wr)
1358 {
1359 union t4_recv_wr *wqe, lwqe;
1360 struct c4iw_srq *srq;
1361 unsigned long flag;
1362 u8 len16 = 0;
1363 u16 idx = 0;
1364 int err = 0;
1365 u32 num_wrs;
1366
1367 srq = to_c4iw_srq(ibsrq);
1368 spin_lock_irqsave(&srq->lock, flag);
1369 num_wrs = t4_srq_avail(&srq->wq);
1370 if (num_wrs == 0) {
1371 spin_unlock_irqrestore(&srq->lock, flag);
1372 return -ENOMEM;
1373 }
1374 while (wr) {
1375 if (wr->num_sge > T4_MAX_RECV_SGE) {
1376 err = -EINVAL;
1377 *bad_wr = wr;
1378 break;
1379 }
1380 wqe = &lwqe;
1381 if (num_wrs)
1382 err = build_srq_recv(wqe, wr, &len16);
1383 else
1384 err = -ENOMEM;
1385 if (err) {
1386 *bad_wr = wr;
1387 break;
1388 }
1389
1390 wqe->recv.opcode = FW_RI_RECV_WR;
1391 wqe->recv.r1 = 0;
1392 wqe->recv.wrid = srq->wq.pidx;
1393 wqe->recv.r2[0] = 0;
1394 wqe->recv.r2[1] = 0;
1395 wqe->recv.r2[2] = 0;
1396 wqe->recv.len16 = len16;
1397
1398 if (srq->wq.ooo_count ||
1399 srq->wq.pending_in_use ||
1400 srq->wq.sw_rq[srq->wq.pidx].valid) {
1401 defer_srq_wr(&srq->wq, wqe, wr->wr_id, len16);
1402 } else {
1403 srq->wq.sw_rq[srq->wq.pidx].wr_id = wr->wr_id;
1404 srq->wq.sw_rq[srq->wq.pidx].valid = 1;
1405 c4iw_copy_wr_to_srq(&srq->wq, wqe, len16);
1406 pr_debug("%s cidx %u pidx %u wq_pidx %u in_use %u wr_id 0x%llx\n",
1407 __func__, srq->wq.cidx,
1408 srq->wq.pidx, srq->wq.wq_pidx,
1409 srq->wq.in_use,
1410 (unsigned long long)wr->wr_id);
1411 t4_srq_produce(&srq->wq, len16);
1412 idx += DIV_ROUND_UP(len16 * 16, T4_EQ_ENTRY_SIZE);
1413 }
1414 wr = wr->next;
1415 num_wrs--;
1416 }
1417 if (idx)
1418 t4_ring_srq_db(&srq->wq, idx, len16, wqe);
1419 spin_unlock_irqrestore(&srq->lock, flag);
1420 return err;
1421 }
1422
1423 static inline void build_term_codes(struct t4_cqe *err_cqe, u8 *layer_type,
1424 u8 *ecode)
1425 {
1426 int status;
1427 int tagged;
1428 int opcode;
1429 int rqtype;
1430 int send_inv;
1431
1432 if (!err_cqe) {
1433 *layer_type = LAYER_RDMAP|DDP_LOCAL_CATA;
1434 *ecode = 0;
1435 return;
1436 }
1437
1438 status = CQE_STATUS(err_cqe);
1439 opcode = CQE_OPCODE(err_cqe);
1440 rqtype = RQ_TYPE(err_cqe);
1441 send_inv = (opcode == FW_RI_SEND_WITH_INV) ||
1442 (opcode == FW_RI_SEND_WITH_SE_INV);
1443 tagged = (opcode == FW_RI_RDMA_WRITE) ||
1444 (rqtype && (opcode == FW_RI_READ_RESP));
1445
1446 switch (status) {
1447 case T4_ERR_STAG:
1448 if (send_inv) {
1449 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP;
1450 *ecode = RDMAP_CANT_INV_STAG;
1451 } else {
1452 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
1453 *ecode = RDMAP_INV_STAG;
1454 }
1455 break;
1456 case T4_ERR_PDID:
1457 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
1458 if ((opcode == FW_RI_SEND_WITH_INV) ||
1459 (opcode == FW_RI_SEND_WITH_SE_INV))
1460 *ecode = RDMAP_CANT_INV_STAG;
1461 else
1462 *ecode = RDMAP_STAG_NOT_ASSOC;
1463 break;
1464 case T4_ERR_QPID:
1465 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
1466 *ecode = RDMAP_STAG_NOT_ASSOC;
1467 break;
1468 case T4_ERR_ACCESS:
1469 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
1470 *ecode = RDMAP_ACC_VIOL;
1471 break;
1472 case T4_ERR_WRAP:
1473 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
1474 *ecode = RDMAP_TO_WRAP;
1475 break;
1476 case T4_ERR_BOUND:
1477 if (tagged) {
1478 *layer_type = LAYER_DDP|DDP_TAGGED_ERR;
1479 *ecode = DDPT_BASE_BOUNDS;
1480 } else {
1481 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
1482 *ecode = RDMAP_BASE_BOUNDS;
1483 }
1484 break;
1485 case T4_ERR_INVALIDATE_SHARED_MR:
1486 case T4_ERR_INVALIDATE_MR_WITH_MW_BOUND:
1487 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP;
1488 *ecode = RDMAP_CANT_INV_STAG;
1489 break;
1490 case T4_ERR_ECC:
1491 case T4_ERR_ECC_PSTAG:
1492 case T4_ERR_INTERNAL_ERR:
1493 *layer_type = LAYER_RDMAP|RDMAP_LOCAL_CATA;
1494 *ecode = 0;
1495 break;
1496 case T4_ERR_OUT_OF_RQE:
1497 *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
1498 *ecode = DDPU_INV_MSN_NOBUF;
1499 break;
1500 case T4_ERR_PBL_ADDR_BOUND:
1501 *layer_type = LAYER_DDP|DDP_TAGGED_ERR;
1502 *ecode = DDPT_BASE_BOUNDS;
1503 break;
1504 case T4_ERR_CRC:
1505 *layer_type = LAYER_MPA|DDP_LLP;
1506 *ecode = MPA_CRC_ERR;
1507 break;
1508 case T4_ERR_MARKER:
1509 *layer_type = LAYER_MPA|DDP_LLP;
1510 *ecode = MPA_MARKER_ERR;
1511 break;
1512 case T4_ERR_PDU_LEN_ERR:
1513 *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
1514 *ecode = DDPU_MSG_TOOBIG;
1515 break;
1516 case T4_ERR_DDP_VERSION:
1517 if (tagged) {
1518 *layer_type = LAYER_DDP|DDP_TAGGED_ERR;
1519 *ecode = DDPT_INV_VERS;
1520 } else {
1521 *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
1522 *ecode = DDPU_INV_VERS;
1523 }
1524 break;
1525 case T4_ERR_RDMA_VERSION:
1526 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP;
1527 *ecode = RDMAP_INV_VERS;
1528 break;
1529 case T4_ERR_OPCODE:
1530 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP;
1531 *ecode = RDMAP_INV_OPCODE;
1532 break;
1533 case T4_ERR_DDP_QUEUE_NUM:
1534 *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
1535 *ecode = DDPU_INV_QN;
1536 break;
1537 case T4_ERR_MSN:
1538 case T4_ERR_MSN_GAP:
1539 case T4_ERR_MSN_RANGE:
1540 case T4_ERR_IRD_OVERFLOW:
1541 *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
1542 *ecode = DDPU_INV_MSN_RANGE;
1543 break;
1544 case T4_ERR_TBIT:
1545 *layer_type = LAYER_DDP|DDP_LOCAL_CATA;
1546 *ecode = 0;
1547 break;
1548 case T4_ERR_MO:
1549 *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
1550 *ecode = DDPU_INV_MO;
1551 break;
1552 default:
1553 *layer_type = LAYER_RDMAP|DDP_LOCAL_CATA;
1554 *ecode = 0;
1555 break;
1556 }
1557 }
1558
1559 static void post_terminate(struct c4iw_qp *qhp, struct t4_cqe *err_cqe,
1560 gfp_t gfp)
1561 {
1562 struct fw_ri_wr *wqe;
1563 struct sk_buff *skb;
1564 struct terminate_message *term;
1565
1566 pr_debug("qhp %p qid 0x%x tid %u\n", qhp, qhp->wq.sq.qid,
1567 qhp->ep->hwtid);
1568
1569 skb = skb_dequeue(&qhp->ep->com.ep_skb_list);
1570 if (WARN_ON(!skb))
1571 return;
1572
1573 set_wr_txq(skb, CPL_PRIORITY_DATA, qhp->ep->txq_idx);
1574
1575 wqe = __skb_put_zero(skb, sizeof(*wqe));
1576 wqe->op_compl = cpu_to_be32(FW_WR_OP_V(FW_RI_INIT_WR));
1577 wqe->flowid_len16 = cpu_to_be32(
1578 FW_WR_FLOWID_V(qhp->ep->hwtid) |
1579 FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*wqe), 16)));
1580
1581 wqe->u.terminate.type = FW_RI_TYPE_TERMINATE;
1582 wqe->u.terminate.immdlen = cpu_to_be32(sizeof(*term));
1583 term = (struct terminate_message *)wqe->u.terminate.termmsg;
1584 if (qhp->attr.layer_etype == (LAYER_MPA|DDP_LLP)) {
1585 term->layer_etype = qhp->attr.layer_etype;
1586 term->ecode = qhp->attr.ecode;
1587 } else
1588 build_term_codes(err_cqe, &term->layer_etype, &term->ecode);
1589 c4iw_ofld_send(&qhp->rhp->rdev, skb);
1590 }
1591
1592
1593
1594
1595 static void __flush_qp(struct c4iw_qp *qhp, struct c4iw_cq *rchp,
1596 struct c4iw_cq *schp)
1597 {
1598 int count;
1599 int rq_flushed = 0, sq_flushed;
1600 unsigned long flag;
1601
1602 pr_debug("qhp %p rchp %p schp %p\n", qhp, rchp, schp);
1603
1604
1605 spin_lock_irqsave(&rchp->lock, flag);
1606 if (schp != rchp)
1607 spin_lock(&schp->lock);
1608 spin_lock(&qhp->lock);
1609
1610 if (qhp->wq.flushed) {
1611 spin_unlock(&qhp->lock);
1612 if (schp != rchp)
1613 spin_unlock(&schp->lock);
1614 spin_unlock_irqrestore(&rchp->lock, flag);
1615 return;
1616 }
1617 qhp->wq.flushed = 1;
1618 t4_set_wq_in_error(&qhp->wq, 0);
1619
1620 c4iw_flush_hw_cq(rchp, qhp);
1621 if (!qhp->srq) {
1622 c4iw_count_rcqes(&rchp->cq, &qhp->wq, &count);
1623 rq_flushed = c4iw_flush_rq(&qhp->wq, &rchp->cq, count);
1624 }
1625
1626 if (schp != rchp)
1627 c4iw_flush_hw_cq(schp, qhp);
1628 sq_flushed = c4iw_flush_sq(qhp);
1629
1630 spin_unlock(&qhp->lock);
1631 if (schp != rchp)
1632 spin_unlock(&schp->lock);
1633 spin_unlock_irqrestore(&rchp->lock, flag);
1634
1635 if (schp == rchp) {
1636 if ((rq_flushed || sq_flushed) &&
1637 t4_clear_cq_armed(&rchp->cq)) {
1638 spin_lock_irqsave(&rchp->comp_handler_lock, flag);
1639 (*rchp->ibcq.comp_handler)(&rchp->ibcq,
1640 rchp->ibcq.cq_context);
1641 spin_unlock_irqrestore(&rchp->comp_handler_lock, flag);
1642 }
1643 } else {
1644 if (rq_flushed && t4_clear_cq_armed(&rchp->cq)) {
1645 spin_lock_irqsave(&rchp->comp_handler_lock, flag);
1646 (*rchp->ibcq.comp_handler)(&rchp->ibcq,
1647 rchp->ibcq.cq_context);
1648 spin_unlock_irqrestore(&rchp->comp_handler_lock, flag);
1649 }
1650 if (sq_flushed && t4_clear_cq_armed(&schp->cq)) {
1651 spin_lock_irqsave(&schp->comp_handler_lock, flag);
1652 (*schp->ibcq.comp_handler)(&schp->ibcq,
1653 schp->ibcq.cq_context);
1654 spin_unlock_irqrestore(&schp->comp_handler_lock, flag);
1655 }
1656 }
1657 }
1658
1659 static void flush_qp(struct c4iw_qp *qhp)
1660 {
1661 struct c4iw_cq *rchp, *schp;
1662 unsigned long flag;
1663
1664 rchp = to_c4iw_cq(qhp->ibqp.recv_cq);
1665 schp = to_c4iw_cq(qhp->ibqp.send_cq);
1666
1667 if (qhp->ibqp.uobject) {
1668
1669
1670 if (qhp->wq.flushed)
1671 return;
1672
1673 qhp->wq.flushed = 1;
1674 t4_set_wq_in_error(&qhp->wq, 0);
1675 t4_set_cq_in_error(&rchp->cq);
1676 spin_lock_irqsave(&rchp->comp_handler_lock, flag);
1677 (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
1678 spin_unlock_irqrestore(&rchp->comp_handler_lock, flag);
1679 if (schp != rchp) {
1680 t4_set_cq_in_error(&schp->cq);
1681 spin_lock_irqsave(&schp->comp_handler_lock, flag);
1682 (*schp->ibcq.comp_handler)(&schp->ibcq,
1683 schp->ibcq.cq_context);
1684 spin_unlock_irqrestore(&schp->comp_handler_lock, flag);
1685 }
1686 return;
1687 }
1688 __flush_qp(qhp, rchp, schp);
1689 }
1690
1691 static int rdma_fini(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
1692 struct c4iw_ep *ep)
1693 {
1694 struct fw_ri_wr *wqe;
1695 int ret;
1696 struct sk_buff *skb;
1697
1698 pr_debug("qhp %p qid 0x%x tid %u\n", qhp, qhp->wq.sq.qid, ep->hwtid);
1699
1700 skb = skb_dequeue(&ep->com.ep_skb_list);
1701 if (WARN_ON(!skb))
1702 return -ENOMEM;
1703
1704 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1705
1706 wqe = __skb_put_zero(skb, sizeof(*wqe));
1707 wqe->op_compl = cpu_to_be32(
1708 FW_WR_OP_V(FW_RI_INIT_WR) |
1709 FW_WR_COMPL_F);
1710 wqe->flowid_len16 = cpu_to_be32(
1711 FW_WR_FLOWID_V(ep->hwtid) |
1712 FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*wqe), 16)));
1713 wqe->cookie = (uintptr_t)ep->com.wr_waitp;
1714
1715 wqe->u.fini.type = FW_RI_TYPE_FINI;
1716
1717 ret = c4iw_ref_send_wait(&rhp->rdev, skb, ep->com.wr_waitp,
1718 qhp->ep->hwtid, qhp->wq.sq.qid, __func__);
1719
1720 pr_debug("ret %d\n", ret);
1721 return ret;
1722 }
1723
1724 static void build_rtr_msg(u8 p2p_type, struct fw_ri_init *init)
1725 {
1726 pr_debug("p2p_type = %d\n", p2p_type);
1727 memset(&init->u, 0, sizeof(init->u));
1728 switch (p2p_type) {
1729 case FW_RI_INIT_P2PTYPE_RDMA_WRITE:
1730 init->u.write.opcode = FW_RI_RDMA_WRITE_WR;
1731 init->u.write.stag_sink = cpu_to_be32(1);
1732 init->u.write.to_sink = cpu_to_be64(1);
1733 init->u.write.u.immd_src[0].op = FW_RI_DATA_IMMD;
1734 init->u.write.len16 = DIV_ROUND_UP(
1735 sizeof(init->u.write) + sizeof(struct fw_ri_immd), 16);
1736 break;
1737 case FW_RI_INIT_P2PTYPE_READ_REQ:
1738 init->u.write.opcode = FW_RI_RDMA_READ_WR;
1739 init->u.read.stag_src = cpu_to_be32(1);
1740 init->u.read.to_src_lo = cpu_to_be32(1);
1741 init->u.read.stag_sink = cpu_to_be32(1);
1742 init->u.read.to_sink_lo = cpu_to_be32(1);
1743 init->u.read.len16 = DIV_ROUND_UP(sizeof(init->u.read), 16);
1744 break;
1745 }
1746 }
1747
1748 static int rdma_init(struct c4iw_dev *rhp, struct c4iw_qp *qhp)
1749 {
1750 struct fw_ri_wr *wqe;
1751 int ret;
1752 struct sk_buff *skb;
1753
1754 pr_debug("qhp %p qid 0x%x tid %u ird %u ord %u\n", qhp,
1755 qhp->wq.sq.qid, qhp->ep->hwtid, qhp->ep->ird, qhp->ep->ord);
1756
1757 skb = alloc_skb(sizeof(*wqe), GFP_KERNEL);
1758 if (!skb) {
1759 ret = -ENOMEM;
1760 goto out;
1761 }
1762 ret = alloc_ird(rhp, qhp->attr.max_ird);
1763 if (ret) {
1764 qhp->attr.max_ird = 0;
1765 kfree_skb(skb);
1766 goto out;
1767 }
1768 set_wr_txq(skb, CPL_PRIORITY_DATA, qhp->ep->txq_idx);
1769
1770 wqe = __skb_put_zero(skb, sizeof(*wqe));
1771 wqe->op_compl = cpu_to_be32(
1772 FW_WR_OP_V(FW_RI_INIT_WR) |
1773 FW_WR_COMPL_F);
1774 wqe->flowid_len16 = cpu_to_be32(
1775 FW_WR_FLOWID_V(qhp->ep->hwtid) |
1776 FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*wqe), 16)));
1777
1778 wqe->cookie = (uintptr_t)qhp->ep->com.wr_waitp;
1779
1780 wqe->u.init.type = FW_RI_TYPE_INIT;
1781 wqe->u.init.mpareqbit_p2ptype =
1782 FW_RI_WR_MPAREQBIT_V(qhp->attr.mpa_attr.initiator) |
1783 FW_RI_WR_P2PTYPE_V(qhp->attr.mpa_attr.p2p_type);
1784 wqe->u.init.mpa_attrs = FW_RI_MPA_IETF_ENABLE;
1785 if (qhp->attr.mpa_attr.recv_marker_enabled)
1786 wqe->u.init.mpa_attrs |= FW_RI_MPA_RX_MARKER_ENABLE;
1787 if (qhp->attr.mpa_attr.xmit_marker_enabled)
1788 wqe->u.init.mpa_attrs |= FW_RI_MPA_TX_MARKER_ENABLE;
1789 if (qhp->attr.mpa_attr.crc_enabled)
1790 wqe->u.init.mpa_attrs |= FW_RI_MPA_CRC_ENABLE;
1791
1792 wqe->u.init.qp_caps = FW_RI_QP_RDMA_READ_ENABLE |
1793 FW_RI_QP_RDMA_WRITE_ENABLE |
1794 FW_RI_QP_BIND_ENABLE;
1795 if (!qhp->ibqp.uobject)
1796 wqe->u.init.qp_caps |= FW_RI_QP_FAST_REGISTER_ENABLE |
1797 FW_RI_QP_STAG0_ENABLE;
1798 wqe->u.init.nrqe = cpu_to_be16(t4_rqes_posted(&qhp->wq));
1799 wqe->u.init.pdid = cpu_to_be32(qhp->attr.pd);
1800 wqe->u.init.qpid = cpu_to_be32(qhp->wq.sq.qid);
1801 wqe->u.init.sq_eqid = cpu_to_be32(qhp->wq.sq.qid);
1802 if (qhp->srq) {
1803 wqe->u.init.rq_eqid = cpu_to_be32(FW_RI_INIT_RQEQID_SRQ |
1804 qhp->srq->idx);
1805 } else {
1806 wqe->u.init.rq_eqid = cpu_to_be32(qhp->wq.rq.qid);
1807 wqe->u.init.hwrqsize = cpu_to_be32(qhp->wq.rq.rqt_size);
1808 wqe->u.init.hwrqaddr = cpu_to_be32(qhp->wq.rq.rqt_hwaddr -
1809 rhp->rdev.lldi.vr->rq.start);
1810 }
1811 wqe->u.init.scqid = cpu_to_be32(qhp->attr.scq);
1812 wqe->u.init.rcqid = cpu_to_be32(qhp->attr.rcq);
1813 wqe->u.init.ord_max = cpu_to_be32(qhp->attr.max_ord);
1814 wqe->u.init.ird_max = cpu_to_be32(qhp->attr.max_ird);
1815 wqe->u.init.iss = cpu_to_be32(qhp->ep->snd_seq);
1816 wqe->u.init.irs = cpu_to_be32(qhp->ep->rcv_seq);
1817 if (qhp->attr.mpa_attr.initiator)
1818 build_rtr_msg(qhp->attr.mpa_attr.p2p_type, &wqe->u.init);
1819
1820 ret = c4iw_ref_send_wait(&rhp->rdev, skb, qhp->ep->com.wr_waitp,
1821 qhp->ep->hwtid, qhp->wq.sq.qid, __func__);
1822 if (!ret)
1823 goto out;
1824
1825 free_ird(rhp, qhp->attr.max_ird);
1826 out:
1827 pr_debug("ret %d\n", ret);
1828 return ret;
1829 }
1830
1831 int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
1832 enum c4iw_qp_attr_mask mask,
1833 struct c4iw_qp_attributes *attrs,
1834 int internal)
1835 {
1836 int ret = 0;
1837 struct c4iw_qp_attributes newattr = qhp->attr;
1838 int disconnect = 0;
1839 int terminate = 0;
1840 int abort = 0;
1841 int free = 0;
1842 struct c4iw_ep *ep = NULL;
1843
1844 pr_debug("qhp %p sqid 0x%x rqid 0x%x ep %p state %d -> %d\n",
1845 qhp, qhp->wq.sq.qid, qhp->wq.rq.qid, qhp->ep, qhp->attr.state,
1846 (mask & C4IW_QP_ATTR_NEXT_STATE) ? attrs->next_state : -1);
1847
1848 mutex_lock(&qhp->mutex);
1849
1850
1851 if (mask & C4IW_QP_ATTR_VALID_MODIFY) {
1852 if (qhp->attr.state != C4IW_QP_STATE_IDLE) {
1853 ret = -EIO;
1854 goto out;
1855 }
1856 if (mask & C4IW_QP_ATTR_ENABLE_RDMA_READ)
1857 newattr.enable_rdma_read = attrs->enable_rdma_read;
1858 if (mask & C4IW_QP_ATTR_ENABLE_RDMA_WRITE)
1859 newattr.enable_rdma_write = attrs->enable_rdma_write;
1860 if (mask & C4IW_QP_ATTR_ENABLE_RDMA_BIND)
1861 newattr.enable_bind = attrs->enable_bind;
1862 if (mask & C4IW_QP_ATTR_MAX_ORD) {
1863 if (attrs->max_ord > c4iw_max_read_depth) {
1864 ret = -EINVAL;
1865 goto out;
1866 }
1867 newattr.max_ord = attrs->max_ord;
1868 }
1869 if (mask & C4IW_QP_ATTR_MAX_IRD) {
1870 if (attrs->max_ird > cur_max_read_depth(rhp)) {
1871 ret = -EINVAL;
1872 goto out;
1873 }
1874 newattr.max_ird = attrs->max_ird;
1875 }
1876 qhp->attr = newattr;
1877 }
1878
1879 if (mask & C4IW_QP_ATTR_SQ_DB) {
1880 ret = ring_kernel_sq_db(qhp, attrs->sq_db_inc);
1881 goto out;
1882 }
1883 if (mask & C4IW_QP_ATTR_RQ_DB) {
1884 ret = ring_kernel_rq_db(qhp, attrs->rq_db_inc);
1885 goto out;
1886 }
1887
1888 if (!(mask & C4IW_QP_ATTR_NEXT_STATE))
1889 goto out;
1890 if (qhp->attr.state == attrs->next_state)
1891 goto out;
1892
1893 switch (qhp->attr.state) {
1894 case C4IW_QP_STATE_IDLE:
1895 switch (attrs->next_state) {
1896 case C4IW_QP_STATE_RTS:
1897 if (!(mask & C4IW_QP_ATTR_LLP_STREAM_HANDLE)) {
1898 ret = -EINVAL;
1899 goto out;
1900 }
1901 if (!(mask & C4IW_QP_ATTR_MPA_ATTR)) {
1902 ret = -EINVAL;
1903 goto out;
1904 }
1905 qhp->attr.mpa_attr = attrs->mpa_attr;
1906 qhp->attr.llp_stream_handle = attrs->llp_stream_handle;
1907 qhp->ep = qhp->attr.llp_stream_handle;
1908 set_state(qhp, C4IW_QP_STATE_RTS);
1909
1910
1911
1912
1913
1914
1915
1916 c4iw_get_ep(&qhp->ep->com);
1917 ret = rdma_init(rhp, qhp);
1918 if (ret)
1919 goto err;
1920 break;
1921 case C4IW_QP_STATE_ERROR:
1922 set_state(qhp, C4IW_QP_STATE_ERROR);
1923 flush_qp(qhp);
1924 break;
1925 default:
1926 ret = -EINVAL;
1927 goto out;
1928 }
1929 break;
1930 case C4IW_QP_STATE_RTS:
1931 switch (attrs->next_state) {
1932 case C4IW_QP_STATE_CLOSING:
1933 t4_set_wq_in_error(&qhp->wq, 0);
1934 set_state(qhp, C4IW_QP_STATE_CLOSING);
1935 ep = qhp->ep;
1936 if (!internal) {
1937 abort = 0;
1938 disconnect = 1;
1939 c4iw_get_ep(&qhp->ep->com);
1940 }
1941 ret = rdma_fini(rhp, qhp, ep);
1942 if (ret)
1943 goto err;
1944 break;
1945 case C4IW_QP_STATE_TERMINATE:
1946 t4_set_wq_in_error(&qhp->wq, 0);
1947 set_state(qhp, C4IW_QP_STATE_TERMINATE);
1948 qhp->attr.layer_etype = attrs->layer_etype;
1949 qhp->attr.ecode = attrs->ecode;
1950 ep = qhp->ep;
1951 if (!internal) {
1952 c4iw_get_ep(&ep->com);
1953 terminate = 1;
1954 disconnect = 1;
1955 } else {
1956 terminate = qhp->attr.send_term;
1957 ret = rdma_fini(rhp, qhp, ep);
1958 if (ret)
1959 goto err;
1960 }
1961 break;
1962 case C4IW_QP_STATE_ERROR:
1963 t4_set_wq_in_error(&qhp->wq, 0);
1964 set_state(qhp, C4IW_QP_STATE_ERROR);
1965 if (!internal) {
1966 abort = 1;
1967 disconnect = 1;
1968 ep = qhp->ep;
1969 c4iw_get_ep(&qhp->ep->com);
1970 }
1971 goto err;
1972 break;
1973 default:
1974 ret = -EINVAL;
1975 goto out;
1976 }
1977 break;
1978 case C4IW_QP_STATE_CLOSING:
1979
1980
1981
1982
1983 if (!internal && (qhp->ibqp.uobject || attrs->next_state !=
1984 C4IW_QP_STATE_ERROR)) {
1985 ret = -EINVAL;
1986 goto out;
1987 }
1988 switch (attrs->next_state) {
1989 case C4IW_QP_STATE_IDLE:
1990 flush_qp(qhp);
1991 set_state(qhp, C4IW_QP_STATE_IDLE);
1992 qhp->attr.llp_stream_handle = NULL;
1993 c4iw_put_ep(&qhp->ep->com);
1994 qhp->ep = NULL;
1995 wake_up(&qhp->wait);
1996 break;
1997 case C4IW_QP_STATE_ERROR:
1998 goto err;
1999 default:
2000 ret = -EINVAL;
2001 goto err;
2002 }
2003 break;
2004 case C4IW_QP_STATE_ERROR:
2005 if (attrs->next_state != C4IW_QP_STATE_IDLE) {
2006 ret = -EINVAL;
2007 goto out;
2008 }
2009 if (!t4_sq_empty(&qhp->wq) || !t4_rq_empty(&qhp->wq)) {
2010 ret = -EINVAL;
2011 goto out;
2012 }
2013 set_state(qhp, C4IW_QP_STATE_IDLE);
2014 break;
2015 case C4IW_QP_STATE_TERMINATE:
2016 if (!internal) {
2017 ret = -EINVAL;
2018 goto out;
2019 }
2020 goto err;
2021 break;
2022 default:
2023 pr_err("%s in a bad state %d\n", __func__, qhp->attr.state);
2024 ret = -EINVAL;
2025 goto err;
2026 break;
2027 }
2028 goto out;
2029 err:
2030 pr_debug("disassociating ep %p qpid 0x%x\n", qhp->ep,
2031 qhp->wq.sq.qid);
2032
2033
2034 qhp->attr.llp_stream_handle = NULL;
2035 if (!ep)
2036 ep = qhp->ep;
2037 qhp->ep = NULL;
2038 set_state(qhp, C4IW_QP_STATE_ERROR);
2039 free = 1;
2040 abort = 1;
2041 flush_qp(qhp);
2042 wake_up(&qhp->wait);
2043 out:
2044 mutex_unlock(&qhp->mutex);
2045
2046 if (terminate)
2047 post_terminate(qhp, NULL, internal ? GFP_ATOMIC : GFP_KERNEL);
2048
2049
2050
2051
2052
2053
2054 if (disconnect) {
2055 c4iw_ep_disconnect(ep, abort, internal ? GFP_ATOMIC :
2056 GFP_KERNEL);
2057 c4iw_put_ep(&ep->com);
2058 }
2059
2060
2061
2062
2063
2064 if (free)
2065 c4iw_put_ep(&ep->com);
2066 pr_debug("exit state %d\n", qhp->attr.state);
2067 return ret;
2068 }
2069
2070 int c4iw_destroy_qp(struct ib_qp *ib_qp, struct ib_udata *udata)
2071 {
2072 struct c4iw_dev *rhp;
2073 struct c4iw_qp *qhp;
2074 struct c4iw_ucontext *ucontext;
2075 struct c4iw_qp_attributes attrs;
2076
2077 qhp = to_c4iw_qp(ib_qp);
2078 rhp = qhp->rhp;
2079 ucontext = qhp->ucontext;
2080
2081 attrs.next_state = C4IW_QP_STATE_ERROR;
2082 if (qhp->attr.state == C4IW_QP_STATE_TERMINATE)
2083 c4iw_modify_qp(rhp, qhp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2084 else
2085 c4iw_modify_qp(rhp, qhp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
2086 wait_event(qhp->wait, !qhp->ep);
2087
2088 xa_lock_irq(&rhp->qps);
2089 __xa_erase(&rhp->qps, qhp->wq.sq.qid);
2090 if (!list_empty(&qhp->db_fc_entry))
2091 list_del_init(&qhp->db_fc_entry);
2092 xa_unlock_irq(&rhp->qps);
2093 free_ird(rhp, qhp->attr.max_ird);
2094
2095 c4iw_qp_rem_ref(ib_qp);
2096
2097 wait_for_completion(&qhp->qp_rel_comp);
2098
2099 pr_debug("ib_qp %p qpid 0x%0x\n", ib_qp, qhp->wq.sq.qid);
2100 pr_debug("qhp %p ucontext %p\n", qhp, ucontext);
2101
2102 destroy_qp(&rhp->rdev, &qhp->wq,
2103 ucontext ? &ucontext->uctx : &rhp->rdev.uctx, !qhp->srq);
2104
2105 c4iw_put_wr_wait(qhp->wr_waitp);
2106
2107 kfree(qhp);
2108 return 0;
2109 }
2110
2111 struct ib_qp *c4iw_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attrs,
2112 struct ib_udata *udata)
2113 {
2114 struct c4iw_dev *rhp;
2115 struct c4iw_qp *qhp;
2116 struct c4iw_pd *php;
2117 struct c4iw_cq *schp;
2118 struct c4iw_cq *rchp;
2119 struct c4iw_create_qp_resp uresp;
2120 unsigned int sqsize, rqsize = 0;
2121 struct c4iw_ucontext *ucontext = rdma_udata_to_drv_context(
2122 udata, struct c4iw_ucontext, ibucontext);
2123 int ret;
2124 struct c4iw_mm_entry *sq_key_mm, *rq_key_mm = NULL, *sq_db_key_mm;
2125 struct c4iw_mm_entry *rq_db_key_mm = NULL, *ma_sync_key_mm = NULL;
2126
2127 pr_debug("ib_pd %p\n", pd);
2128
2129 if (attrs->qp_type != IB_QPT_RC)
2130 return ERR_PTR(-EINVAL);
2131
2132 php = to_c4iw_pd(pd);
2133 rhp = php->rhp;
2134 schp = get_chp(rhp, ((struct c4iw_cq *)attrs->send_cq)->cq.cqid);
2135 rchp = get_chp(rhp, ((struct c4iw_cq *)attrs->recv_cq)->cq.cqid);
2136 if (!schp || !rchp)
2137 return ERR_PTR(-EINVAL);
2138
2139 if (attrs->cap.max_inline_data > T4_MAX_SEND_INLINE)
2140 return ERR_PTR(-EINVAL);
2141
2142 if (!attrs->srq) {
2143 if (attrs->cap.max_recv_wr > rhp->rdev.hw_queue.t4_max_rq_size)
2144 return ERR_PTR(-E2BIG);
2145 rqsize = attrs->cap.max_recv_wr + 1;
2146 if (rqsize < 8)
2147 rqsize = 8;
2148 }
2149
2150 if (attrs->cap.max_send_wr > rhp->rdev.hw_queue.t4_max_sq_size)
2151 return ERR_PTR(-E2BIG);
2152 sqsize = attrs->cap.max_send_wr + 1;
2153 if (sqsize < 8)
2154 sqsize = 8;
2155
2156 qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
2157 if (!qhp)
2158 return ERR_PTR(-ENOMEM);
2159
2160 qhp->wr_waitp = c4iw_alloc_wr_wait(GFP_KERNEL);
2161 if (!qhp->wr_waitp) {
2162 ret = -ENOMEM;
2163 goto err_free_qhp;
2164 }
2165
2166 qhp->wq.sq.size = sqsize;
2167 qhp->wq.sq.memsize =
2168 (sqsize + rhp->rdev.hw_queue.t4_eq_status_entries) *
2169 sizeof(*qhp->wq.sq.queue) + 16 * sizeof(__be64);
2170 qhp->wq.sq.flush_cidx = -1;
2171 if (!attrs->srq) {
2172 qhp->wq.rq.size = rqsize;
2173 qhp->wq.rq.memsize =
2174 (rqsize + rhp->rdev.hw_queue.t4_eq_status_entries) *
2175 sizeof(*qhp->wq.rq.queue);
2176 }
2177
2178 if (ucontext) {
2179 qhp->wq.sq.memsize = roundup(qhp->wq.sq.memsize, PAGE_SIZE);
2180 if (!attrs->srq)
2181 qhp->wq.rq.memsize =
2182 roundup(qhp->wq.rq.memsize, PAGE_SIZE);
2183 }
2184
2185 ret = create_qp(&rhp->rdev, &qhp->wq, &schp->cq, &rchp->cq,
2186 ucontext ? &ucontext->uctx : &rhp->rdev.uctx,
2187 qhp->wr_waitp, !attrs->srq);
2188 if (ret)
2189 goto err_free_wr_wait;
2190
2191 attrs->cap.max_recv_wr = rqsize - 1;
2192 attrs->cap.max_send_wr = sqsize - 1;
2193 attrs->cap.max_inline_data = T4_MAX_SEND_INLINE;
2194
2195 qhp->rhp = rhp;
2196 qhp->attr.pd = php->pdid;
2197 qhp->attr.scq = ((struct c4iw_cq *) attrs->send_cq)->cq.cqid;
2198 qhp->attr.rcq = ((struct c4iw_cq *) attrs->recv_cq)->cq.cqid;
2199 qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
2200 qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
2201 qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
2202 if (!attrs->srq) {
2203 qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
2204 qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
2205 }
2206 qhp->attr.state = C4IW_QP_STATE_IDLE;
2207 qhp->attr.next_state = C4IW_QP_STATE_IDLE;
2208 qhp->attr.enable_rdma_read = 1;
2209 qhp->attr.enable_rdma_write = 1;
2210 qhp->attr.enable_bind = 1;
2211 qhp->attr.max_ord = 0;
2212 qhp->attr.max_ird = 0;
2213 qhp->sq_sig_all = attrs->sq_sig_type == IB_SIGNAL_ALL_WR;
2214 spin_lock_init(&qhp->lock);
2215 mutex_init(&qhp->mutex);
2216 init_waitqueue_head(&qhp->wait);
2217 init_completion(&qhp->qp_rel_comp);
2218 refcount_set(&qhp->qp_refcnt, 1);
2219
2220 ret = xa_insert_irq(&rhp->qps, qhp->wq.sq.qid, qhp, GFP_KERNEL);
2221 if (ret)
2222 goto err_destroy_qp;
2223
2224 if (udata && ucontext) {
2225 sq_key_mm = kmalloc(sizeof(*sq_key_mm), GFP_KERNEL);
2226 if (!sq_key_mm) {
2227 ret = -ENOMEM;
2228 goto err_remove_handle;
2229 }
2230 if (!attrs->srq) {
2231 rq_key_mm = kmalloc(sizeof(*rq_key_mm), GFP_KERNEL);
2232 if (!rq_key_mm) {
2233 ret = -ENOMEM;
2234 goto err_free_sq_key;
2235 }
2236 }
2237 sq_db_key_mm = kmalloc(sizeof(*sq_db_key_mm), GFP_KERNEL);
2238 if (!sq_db_key_mm) {
2239 ret = -ENOMEM;
2240 goto err_free_rq_key;
2241 }
2242 if (!attrs->srq) {
2243 rq_db_key_mm =
2244 kmalloc(sizeof(*rq_db_key_mm), GFP_KERNEL);
2245 if (!rq_db_key_mm) {
2246 ret = -ENOMEM;
2247 goto err_free_sq_db_key;
2248 }
2249 }
2250 memset(&uresp, 0, sizeof(uresp));
2251 if (t4_sq_onchip(&qhp->wq.sq)) {
2252 ma_sync_key_mm = kmalloc(sizeof(*ma_sync_key_mm),
2253 GFP_KERNEL);
2254 if (!ma_sync_key_mm) {
2255 ret = -ENOMEM;
2256 goto err_free_rq_db_key;
2257 }
2258 uresp.flags = C4IW_QPF_ONCHIP;
2259 }
2260 if (rhp->rdev.lldi.write_w_imm_support)
2261 uresp.flags |= C4IW_QPF_WRITE_W_IMM;
2262 uresp.qid_mask = rhp->rdev.qpmask;
2263 uresp.sqid = qhp->wq.sq.qid;
2264 uresp.sq_size = qhp->wq.sq.size;
2265 uresp.sq_memsize = qhp->wq.sq.memsize;
2266 if (!attrs->srq) {
2267 uresp.rqid = qhp->wq.rq.qid;
2268 uresp.rq_size = qhp->wq.rq.size;
2269 uresp.rq_memsize = qhp->wq.rq.memsize;
2270 }
2271 spin_lock(&ucontext->mmap_lock);
2272 if (ma_sync_key_mm) {
2273 uresp.ma_sync_key = ucontext->key;
2274 ucontext->key += PAGE_SIZE;
2275 }
2276 uresp.sq_key = ucontext->key;
2277 ucontext->key += PAGE_SIZE;
2278 if (!attrs->srq) {
2279 uresp.rq_key = ucontext->key;
2280 ucontext->key += PAGE_SIZE;
2281 }
2282 uresp.sq_db_gts_key = ucontext->key;
2283 ucontext->key += PAGE_SIZE;
2284 if (!attrs->srq) {
2285 uresp.rq_db_gts_key = ucontext->key;
2286 ucontext->key += PAGE_SIZE;
2287 }
2288 spin_unlock(&ucontext->mmap_lock);
2289 ret = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
2290 if (ret)
2291 goto err_free_ma_sync_key;
2292 sq_key_mm->key = uresp.sq_key;
2293 sq_key_mm->addr = qhp->wq.sq.phys_addr;
2294 sq_key_mm->len = PAGE_ALIGN(qhp->wq.sq.memsize);
2295 insert_mmap(ucontext, sq_key_mm);
2296 if (!attrs->srq) {
2297 rq_key_mm->key = uresp.rq_key;
2298 rq_key_mm->addr = virt_to_phys(qhp->wq.rq.queue);
2299 rq_key_mm->len = PAGE_ALIGN(qhp->wq.rq.memsize);
2300 insert_mmap(ucontext, rq_key_mm);
2301 }
2302 sq_db_key_mm->key = uresp.sq_db_gts_key;
2303 sq_db_key_mm->addr = (u64)(unsigned long)qhp->wq.sq.bar2_pa;
2304 sq_db_key_mm->len = PAGE_SIZE;
2305 insert_mmap(ucontext, sq_db_key_mm);
2306 if (!attrs->srq) {
2307 rq_db_key_mm->key = uresp.rq_db_gts_key;
2308 rq_db_key_mm->addr =
2309 (u64)(unsigned long)qhp->wq.rq.bar2_pa;
2310 rq_db_key_mm->len = PAGE_SIZE;
2311 insert_mmap(ucontext, rq_db_key_mm);
2312 }
2313 if (ma_sync_key_mm) {
2314 ma_sync_key_mm->key = uresp.ma_sync_key;
2315 ma_sync_key_mm->addr =
2316 (pci_resource_start(rhp->rdev.lldi.pdev, 0) +
2317 PCIE_MA_SYNC_A) & PAGE_MASK;
2318 ma_sync_key_mm->len = PAGE_SIZE;
2319 insert_mmap(ucontext, ma_sync_key_mm);
2320 }
2321
2322 qhp->ucontext = ucontext;
2323 }
2324 if (!attrs->srq) {
2325 qhp->wq.qp_errp =
2326 &qhp->wq.rq.queue[qhp->wq.rq.size].status.qp_err;
2327 } else {
2328 qhp->wq.qp_errp =
2329 &qhp->wq.sq.queue[qhp->wq.sq.size].status.qp_err;
2330 qhp->wq.srqidxp =
2331 &qhp->wq.sq.queue[qhp->wq.sq.size].status.srqidx;
2332 }
2333
2334 qhp->ibqp.qp_num = qhp->wq.sq.qid;
2335 if (attrs->srq)
2336 qhp->srq = to_c4iw_srq(attrs->srq);
2337 INIT_LIST_HEAD(&qhp->db_fc_entry);
2338 pr_debug("sq id %u size %u memsize %zu num_entries %u rq id %u size %u memsize %zu num_entries %u\n",
2339 qhp->wq.sq.qid, qhp->wq.sq.size, qhp->wq.sq.memsize,
2340 attrs->cap.max_send_wr, qhp->wq.rq.qid, qhp->wq.rq.size,
2341 qhp->wq.rq.memsize, attrs->cap.max_recv_wr);
2342 return &qhp->ibqp;
2343 err_free_ma_sync_key:
2344 kfree(ma_sync_key_mm);
2345 err_free_rq_db_key:
2346 if (!attrs->srq)
2347 kfree(rq_db_key_mm);
2348 err_free_sq_db_key:
2349 kfree(sq_db_key_mm);
2350 err_free_rq_key:
2351 if (!attrs->srq)
2352 kfree(rq_key_mm);
2353 err_free_sq_key:
2354 kfree(sq_key_mm);
2355 err_remove_handle:
2356 xa_erase_irq(&rhp->qps, qhp->wq.sq.qid);
2357 err_destroy_qp:
2358 destroy_qp(&rhp->rdev, &qhp->wq,
2359 ucontext ? &ucontext->uctx : &rhp->rdev.uctx, !attrs->srq);
2360 err_free_wr_wait:
2361 c4iw_put_wr_wait(qhp->wr_waitp);
2362 err_free_qhp:
2363 kfree(qhp);
2364 return ERR_PTR(ret);
2365 }
2366
2367 int c4iw_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
2368 int attr_mask, struct ib_udata *udata)
2369 {
2370 struct c4iw_dev *rhp;
2371 struct c4iw_qp *qhp;
2372 enum c4iw_qp_attr_mask mask = 0;
2373 struct c4iw_qp_attributes attrs = {};
2374
2375 pr_debug("ib_qp %p\n", ibqp);
2376
2377
2378 if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
2379 attr_mask &= ~IB_QP_STATE;
2380
2381
2382 if (!attr_mask)
2383 return 0;
2384
2385 qhp = to_c4iw_qp(ibqp);
2386 rhp = qhp->rhp;
2387
2388 attrs.next_state = c4iw_convert_state(attr->qp_state);
2389 attrs.enable_rdma_read = (attr->qp_access_flags &
2390 IB_ACCESS_REMOTE_READ) ? 1 : 0;
2391 attrs.enable_rdma_write = (attr->qp_access_flags &
2392 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
2393 attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
2394
2395
2396 mask |= (attr_mask & IB_QP_STATE) ? C4IW_QP_ATTR_NEXT_STATE : 0;
2397 mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
2398 (C4IW_QP_ATTR_ENABLE_RDMA_READ |
2399 C4IW_QP_ATTR_ENABLE_RDMA_WRITE |
2400 C4IW_QP_ATTR_ENABLE_RDMA_BIND) : 0;
2401
2402
2403
2404
2405
2406
2407 attrs.sq_db_inc = attr->sq_psn;
2408 attrs.rq_db_inc = attr->rq_psn;
2409 mask |= (attr_mask & IB_QP_SQ_PSN) ? C4IW_QP_ATTR_SQ_DB : 0;
2410 mask |= (attr_mask & IB_QP_RQ_PSN) ? C4IW_QP_ATTR_RQ_DB : 0;
2411 if (!is_t4(to_c4iw_qp(ibqp)->rhp->rdev.lldi.adapter_type) &&
2412 (mask & (C4IW_QP_ATTR_SQ_DB|C4IW_QP_ATTR_RQ_DB)))
2413 return -EINVAL;
2414
2415 return c4iw_modify_qp(rhp, qhp, mask, &attrs, 0);
2416 }
2417
2418 struct ib_qp *c4iw_get_qp(struct ib_device *dev, int qpn)
2419 {
2420 pr_debug("ib_dev %p qpn 0x%x\n", dev, qpn);
2421 return (struct ib_qp *)get_qhp(to_c4iw_dev(dev), qpn);
2422 }
2423
2424 void c4iw_dispatch_srq_limit_reached_event(struct c4iw_srq *srq)
2425 {
2426 struct ib_event event = {};
2427
2428 event.device = &srq->rhp->ibdev;
2429 event.element.srq = &srq->ibsrq;
2430 event.event = IB_EVENT_SRQ_LIMIT_REACHED;
2431 ib_dispatch_event(&event);
2432 }
2433
2434 int c4iw_modify_srq(struct ib_srq *ib_srq, struct ib_srq_attr *attr,
2435 enum ib_srq_attr_mask srq_attr_mask,
2436 struct ib_udata *udata)
2437 {
2438 struct c4iw_srq *srq = to_c4iw_srq(ib_srq);
2439 int ret = 0;
2440
2441
2442
2443
2444 if (udata && !srq_attr_mask) {
2445 c4iw_dispatch_srq_limit_reached_event(srq);
2446 goto out;
2447 }
2448
2449
2450 if (srq_attr_mask & IB_SRQ_MAX_WR) {
2451 ret = -EINVAL;
2452 goto out;
2453 }
2454
2455 if (!udata && (srq_attr_mask & IB_SRQ_LIMIT)) {
2456 srq->armed = true;
2457 srq->srq_limit = attr->srq_limit;
2458 }
2459 out:
2460 return ret;
2461 }
2462
2463 int c4iw_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
2464 int attr_mask, struct ib_qp_init_attr *init_attr)
2465 {
2466 struct c4iw_qp *qhp = to_c4iw_qp(ibqp);
2467
2468 memset(attr, 0, sizeof(*attr));
2469 memset(init_attr, 0, sizeof(*init_attr));
2470 attr->qp_state = to_ib_qp_state(qhp->attr.state);
2471 init_attr->cap.max_send_wr = qhp->attr.sq_num_entries;
2472 init_attr->cap.max_recv_wr = qhp->attr.rq_num_entries;
2473 init_attr->cap.max_send_sge = qhp->attr.sq_max_sges;
2474 init_attr->cap.max_recv_sge = qhp->attr.sq_max_sges;
2475 init_attr->cap.max_inline_data = T4_MAX_SEND_INLINE;
2476 init_attr->sq_sig_type = qhp->sq_sig_all ? IB_SIGNAL_ALL_WR : 0;
2477 return 0;
2478 }
2479
2480 static void free_srq_queue(struct c4iw_srq *srq, struct c4iw_dev_ucontext *uctx,
2481 struct c4iw_wr_wait *wr_waitp)
2482 {
2483 struct c4iw_rdev *rdev = &srq->rhp->rdev;
2484 struct sk_buff *skb = srq->destroy_skb;
2485 struct t4_srq *wq = &srq->wq;
2486 struct fw_ri_res_wr *res_wr;
2487 struct fw_ri_res *res;
2488 int wr_len;
2489
2490 wr_len = sizeof(*res_wr) + sizeof(*res);
2491 set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
2492
2493 res_wr = (struct fw_ri_res_wr *)__skb_put(skb, wr_len);
2494 memset(res_wr, 0, wr_len);
2495 res_wr->op_nres = cpu_to_be32(FW_WR_OP_V(FW_RI_RES_WR) |
2496 FW_RI_RES_WR_NRES_V(1) |
2497 FW_WR_COMPL_F);
2498 res_wr->len16_pkd = cpu_to_be32(DIV_ROUND_UP(wr_len, 16));
2499 res_wr->cookie = (uintptr_t)wr_waitp;
2500 res = res_wr->res;
2501 res->u.srq.restype = FW_RI_RES_TYPE_SRQ;
2502 res->u.srq.op = FW_RI_RES_OP_RESET;
2503 res->u.srq.srqid = cpu_to_be32(srq->idx);
2504 res->u.srq.eqid = cpu_to_be32(wq->qid);
2505
2506 c4iw_init_wr_wait(wr_waitp);
2507 c4iw_ref_send_wait(rdev, skb, wr_waitp, 0, 0, __func__);
2508
2509 dma_free_coherent(&rdev->lldi.pdev->dev,
2510 wq->memsize, wq->queue,
2511 dma_unmap_addr(wq, mapping));
2512 c4iw_rqtpool_free(rdev, wq->rqt_hwaddr, wq->rqt_size);
2513 kfree(wq->sw_rq);
2514 c4iw_put_qpid(rdev, wq->qid, uctx);
2515 }
2516
2517 static int alloc_srq_queue(struct c4iw_srq *srq, struct c4iw_dev_ucontext *uctx,
2518 struct c4iw_wr_wait *wr_waitp)
2519 {
2520 struct c4iw_rdev *rdev = &srq->rhp->rdev;
2521 int user = (uctx != &rdev->uctx);
2522 struct t4_srq *wq = &srq->wq;
2523 struct fw_ri_res_wr *res_wr;
2524 struct fw_ri_res *res;
2525 struct sk_buff *skb;
2526 int wr_len;
2527 int eqsize;
2528 int ret = -ENOMEM;
2529
2530 wq->qid = c4iw_get_qpid(rdev, uctx);
2531 if (!wq->qid)
2532 goto err;
2533
2534 if (!user) {
2535 wq->sw_rq = kcalloc(wq->size, sizeof(*wq->sw_rq),
2536 GFP_KERNEL);
2537 if (!wq->sw_rq)
2538 goto err_put_qpid;
2539 wq->pending_wrs = kcalloc(srq->wq.size,
2540 sizeof(*srq->wq.pending_wrs),
2541 GFP_KERNEL);
2542 if (!wq->pending_wrs)
2543 goto err_free_sw_rq;
2544 }
2545
2546 wq->rqt_size = wq->size;
2547 wq->rqt_hwaddr = c4iw_rqtpool_alloc(rdev, wq->rqt_size);
2548 if (!wq->rqt_hwaddr)
2549 goto err_free_pending_wrs;
2550 wq->rqt_abs_idx = (wq->rqt_hwaddr - rdev->lldi.vr->rq.start) >>
2551 T4_RQT_ENTRY_SHIFT;
2552
2553 wq->queue = dma_alloc_coherent(&rdev->lldi.pdev->dev, wq->memsize,
2554 &wq->dma_addr, GFP_KERNEL);
2555 if (!wq->queue)
2556 goto err_free_rqtpool;
2557
2558 dma_unmap_addr_set(wq, mapping, wq->dma_addr);
2559
2560 wq->bar2_va = c4iw_bar2_addrs(rdev, wq->qid, CXGB4_BAR2_QTYPE_EGRESS,
2561 &wq->bar2_qid,
2562 user ? &wq->bar2_pa : NULL);
2563
2564
2565
2566
2567
2568 if (user && !wq->bar2_va) {
2569 pr_warn(MOD "%s: srqid %u not in BAR2 range.\n",
2570 pci_name(rdev->lldi.pdev), wq->qid);
2571 ret = -EINVAL;
2572 goto err_free_queue;
2573 }
2574
2575
2576 wr_len = sizeof(*res_wr) + sizeof(*res);
2577
2578 skb = alloc_skb(wr_len, GFP_KERNEL);
2579 if (!skb)
2580 goto err_free_queue;
2581 set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
2582
2583 res_wr = (struct fw_ri_res_wr *)__skb_put(skb, wr_len);
2584 memset(res_wr, 0, wr_len);
2585 res_wr->op_nres = cpu_to_be32(FW_WR_OP_V(FW_RI_RES_WR) |
2586 FW_RI_RES_WR_NRES_V(1) |
2587 FW_WR_COMPL_F);
2588 res_wr->len16_pkd = cpu_to_be32(DIV_ROUND_UP(wr_len, 16));
2589 res_wr->cookie = (uintptr_t)wr_waitp;
2590 res = res_wr->res;
2591 res->u.srq.restype = FW_RI_RES_TYPE_SRQ;
2592 res->u.srq.op = FW_RI_RES_OP_WRITE;
2593
2594
2595
2596
2597 eqsize = wq->size * T4_RQ_NUM_SLOTS +
2598 rdev->hw_queue.t4_eq_status_entries;
2599 res->u.srq.eqid = cpu_to_be32(wq->qid);
2600 res->u.srq.fetchszm_to_iqid =
2601
2602 cpu_to_be32(FW_RI_RES_WR_HOSTFCMODE_V(0) |
2603 FW_RI_RES_WR_CPRIO_V(0) |
2604 FW_RI_RES_WR_PCIECHN_V(0) |
2605 FW_RI_RES_WR_FETCHRO_V(0));
2606 res->u.srq.dcaen_to_eqsize =
2607 cpu_to_be32(FW_RI_RES_WR_DCAEN_V(0) |
2608 FW_RI_RES_WR_DCACPU_V(0) |
2609 FW_RI_RES_WR_FBMIN_V(2) |
2610 FW_RI_RES_WR_FBMAX_V(3) |
2611 FW_RI_RES_WR_CIDXFTHRESHO_V(0) |
2612 FW_RI_RES_WR_CIDXFTHRESH_V(0) |
2613 FW_RI_RES_WR_EQSIZE_V(eqsize));
2614 res->u.srq.eqaddr = cpu_to_be64(wq->dma_addr);
2615 res->u.srq.srqid = cpu_to_be32(srq->idx);
2616 res->u.srq.pdid = cpu_to_be32(srq->pdid);
2617 res->u.srq.hwsrqsize = cpu_to_be32(wq->rqt_size);
2618 res->u.srq.hwsrqaddr = cpu_to_be32(wq->rqt_hwaddr -
2619 rdev->lldi.vr->rq.start);
2620
2621 c4iw_init_wr_wait(wr_waitp);
2622
2623 ret = c4iw_ref_send_wait(rdev, skb, wr_waitp, 0, wq->qid, __func__);
2624 if (ret)
2625 goto err_free_queue;
2626
2627 pr_debug("%s srq %u eqid %u pdid %u queue va %p pa 0x%llx\n"
2628 " bar2_addr %p rqt addr 0x%x size %d\n",
2629 __func__, srq->idx, wq->qid, srq->pdid, wq->queue,
2630 (u64)virt_to_phys(wq->queue), wq->bar2_va,
2631 wq->rqt_hwaddr, wq->rqt_size);
2632
2633 return 0;
2634 err_free_queue:
2635 dma_free_coherent(&rdev->lldi.pdev->dev,
2636 wq->memsize, wq->queue,
2637 dma_unmap_addr(wq, mapping));
2638 err_free_rqtpool:
2639 c4iw_rqtpool_free(rdev, wq->rqt_hwaddr, wq->rqt_size);
2640 err_free_pending_wrs:
2641 if (!user)
2642 kfree(wq->pending_wrs);
2643 err_free_sw_rq:
2644 if (!user)
2645 kfree(wq->sw_rq);
2646 err_put_qpid:
2647 c4iw_put_qpid(rdev, wq->qid, uctx);
2648 err:
2649 return ret;
2650 }
2651
2652 void c4iw_copy_wr_to_srq(struct t4_srq *srq, union t4_recv_wr *wqe, u8 len16)
2653 {
2654 u64 *src, *dst;
2655
2656 src = (u64 *)wqe;
2657 dst = (u64 *)((u8 *)srq->queue + srq->wq_pidx * T4_EQ_ENTRY_SIZE);
2658 while (len16) {
2659 *dst++ = *src++;
2660 if (dst >= (u64 *)&srq->queue[srq->size])
2661 dst = (u64 *)srq->queue;
2662 *dst++ = *src++;
2663 if (dst >= (u64 *)&srq->queue[srq->size])
2664 dst = (u64 *)srq->queue;
2665 len16--;
2666 }
2667 }
2668
2669 int c4iw_create_srq(struct ib_srq *ib_srq, struct ib_srq_init_attr *attrs,
2670 struct ib_udata *udata)
2671 {
2672 struct ib_pd *pd = ib_srq->pd;
2673 struct c4iw_dev *rhp;
2674 struct c4iw_srq *srq = to_c4iw_srq(ib_srq);
2675 struct c4iw_pd *php;
2676 struct c4iw_create_srq_resp uresp;
2677 struct c4iw_ucontext *ucontext;
2678 struct c4iw_mm_entry *srq_key_mm, *srq_db_key_mm;
2679 int rqsize;
2680 int ret;
2681 int wr_len;
2682
2683 pr_debug("%s ib_pd %p\n", __func__, pd);
2684
2685 php = to_c4iw_pd(pd);
2686 rhp = php->rhp;
2687
2688 if (!rhp->rdev.lldi.vr->srq.size)
2689 return -EINVAL;
2690 if (attrs->attr.max_wr > rhp->rdev.hw_queue.t4_max_rq_size)
2691 return -E2BIG;
2692 if (attrs->attr.max_sge > T4_MAX_RECV_SGE)
2693 return -E2BIG;
2694
2695
2696
2697
2698 rqsize = attrs->attr.max_wr + 1;
2699 rqsize = roundup_pow_of_two(max_t(u16, rqsize, 16));
2700
2701 ucontext = rdma_udata_to_drv_context(udata, struct c4iw_ucontext,
2702 ibucontext);
2703
2704 srq->wr_waitp = c4iw_alloc_wr_wait(GFP_KERNEL);
2705 if (!srq->wr_waitp)
2706 return -ENOMEM;
2707
2708 srq->idx = c4iw_alloc_srq_idx(&rhp->rdev);
2709 if (srq->idx < 0) {
2710 ret = -ENOMEM;
2711 goto err_free_wr_wait;
2712 }
2713
2714 wr_len = sizeof(struct fw_ri_res_wr) + sizeof(struct fw_ri_res);
2715 srq->destroy_skb = alloc_skb(wr_len, GFP_KERNEL);
2716 if (!srq->destroy_skb) {
2717 ret = -ENOMEM;
2718 goto err_free_srq_idx;
2719 }
2720
2721 srq->rhp = rhp;
2722 srq->pdid = php->pdid;
2723
2724 srq->wq.size = rqsize;
2725 srq->wq.memsize =
2726 (rqsize + rhp->rdev.hw_queue.t4_eq_status_entries) *
2727 sizeof(*srq->wq.queue);
2728 if (ucontext)
2729 srq->wq.memsize = roundup(srq->wq.memsize, PAGE_SIZE);
2730
2731 ret = alloc_srq_queue(srq, ucontext ? &ucontext->uctx :
2732 &rhp->rdev.uctx, srq->wr_waitp);
2733 if (ret)
2734 goto err_free_skb;
2735 attrs->attr.max_wr = rqsize - 1;
2736
2737 if (CHELSIO_CHIP_VERSION(rhp->rdev.lldi.adapter_type) > CHELSIO_T6)
2738 srq->flags = T4_SRQ_LIMIT_SUPPORT;
2739
2740 if (udata) {
2741 srq_key_mm = kmalloc(sizeof(*srq_key_mm), GFP_KERNEL);
2742 if (!srq_key_mm) {
2743 ret = -ENOMEM;
2744 goto err_free_queue;
2745 }
2746 srq_db_key_mm = kmalloc(sizeof(*srq_db_key_mm), GFP_KERNEL);
2747 if (!srq_db_key_mm) {
2748 ret = -ENOMEM;
2749 goto err_free_srq_key_mm;
2750 }
2751 memset(&uresp, 0, sizeof(uresp));
2752 uresp.flags = srq->flags;
2753 uresp.qid_mask = rhp->rdev.qpmask;
2754 uresp.srqid = srq->wq.qid;
2755 uresp.srq_size = srq->wq.size;
2756 uresp.srq_memsize = srq->wq.memsize;
2757 uresp.rqt_abs_idx = srq->wq.rqt_abs_idx;
2758 spin_lock(&ucontext->mmap_lock);
2759 uresp.srq_key = ucontext->key;
2760 ucontext->key += PAGE_SIZE;
2761 uresp.srq_db_gts_key = ucontext->key;
2762 ucontext->key += PAGE_SIZE;
2763 spin_unlock(&ucontext->mmap_lock);
2764 ret = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
2765 if (ret)
2766 goto err_free_srq_db_key_mm;
2767 srq_key_mm->key = uresp.srq_key;
2768 srq_key_mm->addr = virt_to_phys(srq->wq.queue);
2769 srq_key_mm->len = PAGE_ALIGN(srq->wq.memsize);
2770 insert_mmap(ucontext, srq_key_mm);
2771 srq_db_key_mm->key = uresp.srq_db_gts_key;
2772 srq_db_key_mm->addr = (u64)(unsigned long)srq->wq.bar2_pa;
2773 srq_db_key_mm->len = PAGE_SIZE;
2774 insert_mmap(ucontext, srq_db_key_mm);
2775 }
2776
2777 pr_debug("%s srq qid %u idx %u size %u memsize %lu num_entries %u\n",
2778 __func__, srq->wq.qid, srq->idx, srq->wq.size,
2779 (unsigned long)srq->wq.memsize, attrs->attr.max_wr);
2780
2781 spin_lock_init(&srq->lock);
2782 return 0;
2783
2784 err_free_srq_db_key_mm:
2785 kfree(srq_db_key_mm);
2786 err_free_srq_key_mm:
2787 kfree(srq_key_mm);
2788 err_free_queue:
2789 free_srq_queue(srq, ucontext ? &ucontext->uctx : &rhp->rdev.uctx,
2790 srq->wr_waitp);
2791 err_free_skb:
2792 kfree_skb(srq->destroy_skb);
2793 err_free_srq_idx:
2794 c4iw_free_srq_idx(&rhp->rdev, srq->idx);
2795 err_free_wr_wait:
2796 c4iw_put_wr_wait(srq->wr_waitp);
2797 return ret;
2798 }
2799
2800 void c4iw_destroy_srq(struct ib_srq *ibsrq, struct ib_udata *udata)
2801 {
2802 struct c4iw_dev *rhp;
2803 struct c4iw_srq *srq;
2804 struct c4iw_ucontext *ucontext;
2805
2806 srq = to_c4iw_srq(ibsrq);
2807 rhp = srq->rhp;
2808
2809 pr_debug("%s id %d\n", __func__, srq->wq.qid);
2810 ucontext = rdma_udata_to_drv_context(udata, struct c4iw_ucontext,
2811 ibucontext);
2812 free_srq_queue(srq, ucontext ? &ucontext->uctx : &rhp->rdev.uctx,
2813 srq->wr_waitp);
2814 c4iw_free_srq_idx(&rhp->rdev, srq->idx);
2815 c4iw_put_wr_wait(srq->wr_waitp);
2816 }