2 * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 static int destroy_qp(struct c4iw_rdev
*rdev
, struct t4_wq
*wq
,
35 struct c4iw_dev_ucontext
*uctx
)
38 * uP clears EQ contexts when the connection exits rdma mode,
39 * so no need to post a RESET WR for these EQs.
41 dma_free_coherent(&(rdev
->lldi
.pdev
->dev
),
42 wq
->rq
.memsize
, wq
->rq
.queue
,
43 dma_unmap_addr(&wq
->rq
, mapping
));
44 dma_free_coherent(&(rdev
->lldi
.pdev
->dev
),
45 wq
->sq
.memsize
, wq
->sq
.queue
,
46 dma_unmap_addr(&wq
->sq
, mapping
));
47 c4iw_rqtpool_free(rdev
, wq
->rq
.rqt_hwaddr
, wq
->rq
.rqt_size
);
50 c4iw_put_qpid(rdev
, wq
->rq
.qid
, uctx
);
51 c4iw_put_qpid(rdev
, wq
->sq
.qid
, uctx
);
55 static int create_qp(struct c4iw_rdev
*rdev
, struct t4_wq
*wq
,
56 struct t4_cq
*rcq
, struct t4_cq
*scq
,
57 struct c4iw_dev_ucontext
*uctx
)
59 int user
= (uctx
!= &rdev
->uctx
);
60 struct fw_ri_res_wr
*res_wr
;
61 struct fw_ri_res
*res
;
63 struct c4iw_wr_wait wr_wait
;
68 wq
->sq
.qid
= c4iw_get_qpid(rdev
, uctx
);
72 wq
->rq
.qid
= c4iw_get_qpid(rdev
, uctx
);
77 wq
->sq
.sw_sq
= kzalloc(wq
->sq
.size
* sizeof *wq
->sq
.sw_sq
,
82 wq
->rq
.sw_rq
= kzalloc(wq
->rq
.size
* sizeof *wq
->rq
.sw_rq
,
89 * RQT must be a power of 2.
91 wq
->rq
.rqt_size
= roundup_pow_of_two(wq
->rq
.size
);
92 wq
->rq
.rqt_hwaddr
= c4iw_rqtpool_alloc(rdev
, wq
->rq
.rqt_size
);
93 if (!wq
->rq
.rqt_hwaddr
)
96 wq
->sq
.queue
= dma_alloc_coherent(&(rdev
->lldi
.pdev
->dev
),
97 wq
->sq
.memsize
, &(wq
->sq
.dma_addr
),
101 memset(wq
->sq
.queue
, 0, wq
->sq
.memsize
);
102 dma_unmap_addr_set(&wq
->sq
, mapping
, wq
->sq
.dma_addr
);
104 wq
->rq
.queue
= dma_alloc_coherent(&(rdev
->lldi
.pdev
->dev
),
105 wq
->rq
.memsize
, &(wq
->rq
.dma_addr
),
109 PDBG("%s sq base va 0x%p pa 0x%llx rq base va 0x%p pa 0x%llx\n",
110 __func__
, wq
->sq
.queue
,
111 (unsigned long long)virt_to_phys(wq
->sq
.queue
),
113 (unsigned long long)virt_to_phys(wq
->rq
.queue
));
114 memset(wq
->rq
.queue
, 0, wq
->rq
.memsize
);
115 dma_unmap_addr_set(&wq
->rq
, mapping
, wq
->rq
.dma_addr
);
117 wq
->db
= rdev
->lldi
.db_reg
;
118 wq
->gts
= rdev
->lldi
.gts_reg
;
120 wq
->sq
.udb
= (u64
)pci_resource_start(rdev
->lldi
.pdev
, 2) +
121 (wq
->sq
.qid
<< rdev
->qpshift
);
122 wq
->sq
.udb
&= PAGE_MASK
;
123 wq
->rq
.udb
= (u64
)pci_resource_start(rdev
->lldi
.pdev
, 2) +
124 (wq
->rq
.qid
<< rdev
->qpshift
);
125 wq
->rq
.udb
&= PAGE_MASK
;
130 /* build fw_ri_res_wr */
131 wr_len
= sizeof *res_wr
+ 2 * sizeof *res
;
133 skb
= alloc_skb(wr_len
, GFP_KERNEL
);
138 set_wr_txq(skb
, CPL_PRIORITY_CONTROL
, 0);
140 res_wr
= (struct fw_ri_res_wr
*)__skb_put(skb
, wr_len
);
141 memset(res_wr
, 0, wr_len
);
142 res_wr
->op_nres
= cpu_to_be32(
143 FW_WR_OP(FW_RI_RES_WR
) |
144 V_FW_RI_RES_WR_NRES(2) |
146 res_wr
->len16_pkd
= cpu_to_be32(DIV_ROUND_UP(wr_len
, 16));
147 res_wr
->cookie
= (u64
)&wr_wait
;
149 res
->u
.sqrq
.restype
= FW_RI_RES_TYPE_SQ
;
150 res
->u
.sqrq
.op
= FW_RI_RES_OP_WRITE
;
153 * eqsize is the number of 64B entries plus the status page size.
155 eqsize
= wq
->sq
.size
* T4_SQ_NUM_SLOTS
+ T4_EQ_STATUS_ENTRIES
;
157 res
->u
.sqrq
.fetchszm_to_iqid
= cpu_to_be32(
158 V_FW_RI_RES_WR_HOSTFCMODE(0) | /* no host cidx updates */
159 V_FW_RI_RES_WR_CPRIO(0) | /* don't keep in chip cache */
160 V_FW_RI_RES_WR_PCIECHN(0) | /* set by uP at ri_init time */
161 V_FW_RI_RES_WR_IQID(scq
->cqid
));
162 res
->u
.sqrq
.dcaen_to_eqsize
= cpu_to_be32(
163 V_FW_RI_RES_WR_DCAEN(0) |
164 V_FW_RI_RES_WR_DCACPU(0) |
165 V_FW_RI_RES_WR_FBMIN(2) |
166 V_FW_RI_RES_WR_FBMAX(3) |
167 V_FW_RI_RES_WR_CIDXFTHRESHO(0) |
168 V_FW_RI_RES_WR_CIDXFTHRESH(0) |
169 V_FW_RI_RES_WR_EQSIZE(eqsize
));
170 res
->u
.sqrq
.eqid
= cpu_to_be32(wq
->sq
.qid
);
171 res
->u
.sqrq
.eqaddr
= cpu_to_be64(wq
->sq
.dma_addr
);
173 res
->u
.sqrq
.restype
= FW_RI_RES_TYPE_RQ
;
174 res
->u
.sqrq
.op
= FW_RI_RES_OP_WRITE
;
177 * eqsize is the number of 64B entries plus the status page size.
179 eqsize
= wq
->rq
.size
* T4_RQ_NUM_SLOTS
+ T4_EQ_STATUS_ENTRIES
;
180 res
->u
.sqrq
.fetchszm_to_iqid
= cpu_to_be32(
181 V_FW_RI_RES_WR_HOSTFCMODE(0) | /* no host cidx updates */
182 V_FW_RI_RES_WR_CPRIO(0) | /* don't keep in chip cache */
183 V_FW_RI_RES_WR_PCIECHN(0) | /* set by uP at ri_init time */
184 V_FW_RI_RES_WR_IQID(rcq
->cqid
));
185 res
->u
.sqrq
.dcaen_to_eqsize
= cpu_to_be32(
186 V_FW_RI_RES_WR_DCAEN(0) |
187 V_FW_RI_RES_WR_DCACPU(0) |
188 V_FW_RI_RES_WR_FBMIN(2) |
189 V_FW_RI_RES_WR_FBMAX(3) |
190 V_FW_RI_RES_WR_CIDXFTHRESHO(0) |
191 V_FW_RI_RES_WR_CIDXFTHRESH(0) |
192 V_FW_RI_RES_WR_EQSIZE(eqsize
));
193 res
->u
.sqrq
.eqid
= cpu_to_be32(wq
->rq
.qid
);
194 res
->u
.sqrq
.eqaddr
= cpu_to_be64(wq
->rq
.dma_addr
);
196 c4iw_init_wr_wait(&wr_wait
);
198 ret
= c4iw_ofld_send(rdev
, skb
);
201 wait_event_timeout(wr_wait
.wait
, wr_wait
.done
, C4IW_WR_TO
);
203 printk(KERN_ERR MOD
"Device %s not responding!\n",
204 pci_name(rdev
->lldi
.pdev
));
205 rdev
->flags
= T4_FATAL_ERROR
;
212 PDBG("%s sqid 0x%x rqid 0x%x kdb 0x%p squdb 0x%llx rqudb 0x%llx\n",
213 __func__
, wq
->sq
.qid
, wq
->rq
.qid
, wq
->db
,
214 (unsigned long long)wq
->sq
.udb
, (unsigned long long)wq
->rq
.udb
);
218 dma_free_coherent(&(rdev
->lldi
.pdev
->dev
),
219 wq
->rq
.memsize
, wq
->rq
.queue
,
220 dma_unmap_addr(&wq
->rq
, mapping
));
222 dma_free_coherent(&(rdev
->lldi
.pdev
->dev
),
223 wq
->sq
.memsize
, wq
->sq
.queue
,
224 dma_unmap_addr(&wq
->sq
, mapping
));
226 c4iw_rqtpool_free(rdev
, wq
->rq
.rqt_hwaddr
, wq
->rq
.rqt_size
);
232 c4iw_put_qpid(rdev
, wq
->rq
.qid
, uctx
);
234 c4iw_put_qpid(rdev
, wq
->sq
.qid
, uctx
);
238 static int build_immd(struct t4_sq
*sq
, struct fw_ri_immd
*immdp
,
239 struct ib_send_wr
*wr
, int max
, u32
*plenp
)
246 dstp
= (u8
*)immdp
->data
;
247 for (i
= 0; i
< wr
->num_sge
; i
++) {
248 if ((plen
+ wr
->sg_list
[i
].length
) > max
)
250 srcp
= (u8
*)(unsigned long)wr
->sg_list
[i
].addr
;
251 plen
+= wr
->sg_list
[i
].length
;
252 rem
= wr
->sg_list
[i
].length
;
254 if (dstp
== (u8
*)&sq
->queue
[sq
->size
])
255 dstp
= (u8
*)sq
->queue
;
256 if (rem
<= (u8
*)&sq
->queue
[sq
->size
] - dstp
)
259 len
= (u8
*)&sq
->queue
[sq
->size
] - dstp
;
260 memcpy(dstp
, srcp
, len
);
266 immdp
->op
= FW_RI_DATA_IMMD
;
269 immdp
->immdlen
= cpu_to_be32(plen
);
274 static int build_isgl(__be64
*queue_start
, __be64
*queue_end
,
275 struct fw_ri_isgl
*isglp
, struct ib_sge
*sg_list
,
276 int num_sge
, u32
*plenp
)
281 __be64
*flitp
= (__be64
*)isglp
->sge
;
283 for (i
= 0; i
< num_sge
; i
++) {
284 if ((plen
+ sg_list
[i
].length
) < plen
)
286 plen
+= sg_list
[i
].length
;
287 *flitp
= cpu_to_be64(((u64
)sg_list
[i
].lkey
<< 32) |
289 if (++flitp
== queue_end
)
291 *flitp
= cpu_to_be64(sg_list
[i
].addr
);
292 if (++flitp
== queue_end
)
295 isglp
->op
= FW_RI_DATA_ISGL
;
297 isglp
->nsge
= cpu_to_be16(num_sge
);
304 static int build_rdma_send(struct t4_sq
*sq
, union t4_wr
*wqe
,
305 struct ib_send_wr
*wr
, u8
*len16
)
311 if (wr
->num_sge
> T4_MAX_SEND_SGE
)
313 switch (wr
->opcode
) {
315 if (wr
->send_flags
& IB_SEND_SOLICITED
)
316 wqe
->send
.sendop_pkd
= cpu_to_be32(
317 V_FW_RI_SEND_WR_SENDOP(FW_RI_SEND_WITH_SE
));
319 wqe
->send
.sendop_pkd
= cpu_to_be32(
320 V_FW_RI_SEND_WR_SENDOP(FW_RI_SEND
));
321 wqe
->send
.stag_inv
= 0;
323 case IB_WR_SEND_WITH_INV
:
324 if (wr
->send_flags
& IB_SEND_SOLICITED
)
325 wqe
->send
.sendop_pkd
= cpu_to_be32(
326 V_FW_RI_SEND_WR_SENDOP(FW_RI_SEND_WITH_SE_INV
));
328 wqe
->send
.sendop_pkd
= cpu_to_be32(
329 V_FW_RI_SEND_WR_SENDOP(FW_RI_SEND_WITH_INV
));
330 wqe
->send
.stag_inv
= cpu_to_be32(wr
->ex
.invalidate_rkey
);
339 if (wr
->send_flags
& IB_SEND_INLINE
) {
340 ret
= build_immd(sq
, wqe
->send
.u
.immd_src
, wr
,
341 T4_MAX_SEND_INLINE
, &plen
);
344 size
= sizeof wqe
->send
+ sizeof(struct fw_ri_immd
) +
347 ret
= build_isgl((__be64
*)sq
->queue
,
348 (__be64
*)&sq
->queue
[sq
->size
],
349 wqe
->send
.u
.isgl_src
,
350 wr
->sg_list
, wr
->num_sge
, &plen
);
353 size
= sizeof wqe
->send
+ sizeof(struct fw_ri_isgl
) +
354 wr
->num_sge
* sizeof(struct fw_ri_sge
);
357 wqe
->send
.u
.immd_src
[0].op
= FW_RI_DATA_IMMD
;
358 wqe
->send
.u
.immd_src
[0].r1
= 0;
359 wqe
->send
.u
.immd_src
[0].r2
= 0;
360 wqe
->send
.u
.immd_src
[0].immdlen
= 0;
361 size
= sizeof wqe
->send
+ sizeof(struct fw_ri_immd
);
364 *len16
= DIV_ROUND_UP(size
, 16);
365 wqe
->send
.plen
= cpu_to_be32(plen
);
369 static int build_rdma_write(struct t4_sq
*sq
, union t4_wr
*wqe
,
370 struct ib_send_wr
*wr
, u8
*len16
)
376 if (wr
->num_sge
> T4_MAX_SEND_SGE
)
379 wqe
->write
.stag_sink
= cpu_to_be32(wr
->wr
.rdma
.rkey
);
380 wqe
->write
.to_sink
= cpu_to_be64(wr
->wr
.rdma
.remote_addr
);
382 if (wr
->send_flags
& IB_SEND_INLINE
) {
383 ret
= build_immd(sq
, wqe
->write
.u
.immd_src
, wr
,
384 T4_MAX_WRITE_INLINE
, &plen
);
387 size
= sizeof wqe
->write
+ sizeof(struct fw_ri_immd
) +
390 ret
= build_isgl((__be64
*)sq
->queue
,
391 (__be64
*)&sq
->queue
[sq
->size
],
392 wqe
->write
.u
.isgl_src
,
393 wr
->sg_list
, wr
->num_sge
, &plen
);
396 size
= sizeof wqe
->write
+ sizeof(struct fw_ri_isgl
) +
397 wr
->num_sge
* sizeof(struct fw_ri_sge
);
400 wqe
->write
.u
.immd_src
[0].op
= FW_RI_DATA_IMMD
;
401 wqe
->write
.u
.immd_src
[0].r1
= 0;
402 wqe
->write
.u
.immd_src
[0].r2
= 0;
403 wqe
->write
.u
.immd_src
[0].immdlen
= 0;
404 size
= sizeof wqe
->write
+ sizeof(struct fw_ri_immd
);
407 *len16
= DIV_ROUND_UP(size
, 16);
408 wqe
->write
.plen
= cpu_to_be32(plen
);
412 static int build_rdma_read(union t4_wr
*wqe
, struct ib_send_wr
*wr
, u8
*len16
)
417 wqe
->read
.stag_src
= cpu_to_be32(wr
->wr
.rdma
.rkey
);
418 wqe
->read
.to_src_hi
= cpu_to_be32((u32
)(wr
->wr
.rdma
.remote_addr
420 wqe
->read
.to_src_lo
= cpu_to_be32((u32
)wr
->wr
.rdma
.remote_addr
);
421 wqe
->read
.stag_sink
= cpu_to_be32(wr
->sg_list
[0].lkey
);
422 wqe
->read
.plen
= cpu_to_be32(wr
->sg_list
[0].length
);
423 wqe
->read
.to_sink_hi
= cpu_to_be32((u32
)(wr
->sg_list
[0].addr
425 wqe
->read
.to_sink_lo
= cpu_to_be32((u32
)(wr
->sg_list
[0].addr
));
427 wqe
->read
.stag_src
= cpu_to_be32(2);
428 wqe
->read
.to_src_hi
= 0;
429 wqe
->read
.to_src_lo
= 0;
430 wqe
->read
.stag_sink
= cpu_to_be32(2);
432 wqe
->read
.to_sink_hi
= 0;
433 wqe
->read
.to_sink_lo
= 0;
437 *len16
= DIV_ROUND_UP(sizeof wqe
->read
, 16);
441 static int build_rdma_recv(struct c4iw_qp
*qhp
, union t4_recv_wr
*wqe
,
442 struct ib_recv_wr
*wr
, u8
*len16
)
446 ret
= build_isgl((__be64
*)qhp
->wq
.rq
.queue
,
447 (__be64
*)&qhp
->wq
.rq
.queue
[qhp
->wq
.rq
.size
],
448 &wqe
->recv
.isgl
, wr
->sg_list
, wr
->num_sge
, NULL
);
451 *len16
= DIV_ROUND_UP(sizeof wqe
->recv
+
452 wr
->num_sge
* sizeof(struct fw_ri_sge
), 16);
456 static int build_fastreg(union t4_wr
*wqe
, struct ib_send_wr
*wr
, u8
*len16
)
459 struct fw_ri_immd
*imdp
;
462 int pbllen
= roundup(wr
->wr
.fast_reg
.page_list_len
* sizeof(u64
), 32);
464 if (wr
->wr
.fast_reg
.page_list_len
> T4_MAX_FR_DEPTH
)
467 wqe
->fr
.qpbinde_to_dcacpu
= 0;
468 wqe
->fr
.pgsz_shift
= wr
->wr
.fast_reg
.page_shift
- 12;
469 wqe
->fr
.addr_type
= FW_RI_VA_BASED_TO
;
470 wqe
->fr
.mem_perms
= c4iw_ib_to_tpt_access(wr
->wr
.fast_reg
.access_flags
);
472 wqe
->fr
.len_lo
= cpu_to_be32(wr
->wr
.fast_reg
.length
);
473 wqe
->fr
.stag
= cpu_to_be32(wr
->wr
.fast_reg
.rkey
);
474 wqe
->fr
.va_hi
= cpu_to_be32(wr
->wr
.fast_reg
.iova_start
>> 32);
475 wqe
->fr
.va_lo_fbo
= cpu_to_be32(wr
->wr
.fast_reg
.iova_start
&
477 if (pbllen
> T4_MAX_FR_IMMD
) {
478 struct c4iw_fr_page_list
*c4pl
=
479 to_c4iw_fr_page_list(wr
->wr
.fast_reg
.page_list
);
480 struct fw_ri_dsgl
*sglp
;
482 sglp
= (struct fw_ri_dsgl
*)(&wqe
->fr
+ 1);
483 sglp
->op
= FW_RI_DATA_DSGL
;
485 sglp
->nsge
= cpu_to_be16(1);
486 sglp
->addr0
= cpu_to_be64(c4pl
->dma_addr
);
487 sglp
->len0
= cpu_to_be32(pbllen
);
489 *len16
= DIV_ROUND_UP(sizeof wqe
->fr
+ sizeof *sglp
, 16);
491 imdp
= (struct fw_ri_immd
*)(&wqe
->fr
+ 1);
492 imdp
->op
= FW_RI_DATA_IMMD
;
495 imdp
->immdlen
= cpu_to_be32(pbllen
);
496 p
= (__be64
*)(imdp
+ 1);
497 for (i
= 0; i
< wr
->wr
.fast_reg
.page_list_len
; i
++, p
++)
499 (u64
)wr
->wr
.fast_reg
.page_list
->page_list
[i
]);
500 *len16
= DIV_ROUND_UP(sizeof wqe
->fr
+ sizeof *imdp
+ pbllen
,
506 static int build_inv_stag(union t4_wr
*wqe
, struct ib_send_wr
*wr
,
509 wqe
->inv
.stag_inv
= cpu_to_be32(wr
->ex
.invalidate_rkey
);
511 *len16
= DIV_ROUND_UP(sizeof wqe
->inv
, 16);
515 void c4iw_qp_add_ref(struct ib_qp
*qp
)
517 PDBG("%s ib_qp %p\n", __func__
, qp
);
518 atomic_inc(&(to_c4iw_qp(qp
)->refcnt
));
521 void c4iw_qp_rem_ref(struct ib_qp
*qp
)
523 PDBG("%s ib_qp %p\n", __func__
, qp
);
524 if (atomic_dec_and_test(&(to_c4iw_qp(qp
)->refcnt
)))
525 wake_up(&(to_c4iw_qp(qp
)->wait
));
528 int c4iw_post_send(struct ib_qp
*ibqp
, struct ib_send_wr
*wr
,
529 struct ib_send_wr
**bad_wr
)
533 enum fw_wr_opcodes fw_opcode
= 0;
534 enum fw_ri_wr_flags fw_flags
;
538 struct t4_swsqe
*swsqe
;
542 qhp
= to_c4iw_qp(ibqp
);
543 spin_lock_irqsave(&qhp
->lock
, flag
);
544 if (t4_wq_in_error(&qhp
->wq
)) {
545 spin_unlock_irqrestore(&qhp
->lock
, flag
);
548 num_wrs
= t4_sq_avail(&qhp
->wq
);
550 spin_unlock_irqrestore(&qhp
->lock
, flag
);
559 wqe
= (union t4_wr
*)((u8
*)qhp
->wq
.sq
.queue
+
560 qhp
->wq
.sq
.wq_pidx
* T4_EQ_ENTRY_SIZE
);
563 if (wr
->send_flags
& IB_SEND_SOLICITED
)
564 fw_flags
|= FW_RI_SOLICITED_EVENT_FLAG
;
565 if (wr
->send_flags
& IB_SEND_SIGNALED
)
566 fw_flags
|= FW_RI_COMPLETION_FLAG
;
567 swsqe
= &qhp
->wq
.sq
.sw_sq
[qhp
->wq
.sq
.pidx
];
568 switch (wr
->opcode
) {
569 case IB_WR_SEND_WITH_INV
:
571 if (wr
->send_flags
& IB_SEND_FENCE
)
572 fw_flags
|= FW_RI_READ_FENCE_FLAG
;
573 fw_opcode
= FW_RI_SEND_WR
;
574 if (wr
->opcode
== IB_WR_SEND
)
575 swsqe
->opcode
= FW_RI_SEND
;
577 swsqe
->opcode
= FW_RI_SEND_WITH_INV
;
578 err
= build_rdma_send(&qhp
->wq
.sq
, wqe
, wr
, &len16
);
580 case IB_WR_RDMA_WRITE
:
581 fw_opcode
= FW_RI_RDMA_WRITE_WR
;
582 swsqe
->opcode
= FW_RI_RDMA_WRITE
;
583 err
= build_rdma_write(&qhp
->wq
.sq
, wqe
, wr
, &len16
);
585 case IB_WR_RDMA_READ
:
586 case IB_WR_RDMA_READ_WITH_INV
:
587 fw_opcode
= FW_RI_RDMA_READ_WR
;
588 swsqe
->opcode
= FW_RI_READ_REQ
;
589 if (wr
->opcode
== IB_WR_RDMA_READ_WITH_INV
)
590 fw_flags
|= FW_RI_RDMA_READ_INVALIDATE
;
593 err
= build_rdma_read(wqe
, wr
, &len16
);
596 swsqe
->read_len
= wr
->sg_list
[0].length
;
597 if (!qhp
->wq
.sq
.oldest_read
)
598 qhp
->wq
.sq
.oldest_read
= swsqe
;
600 case IB_WR_FAST_REG_MR
:
601 fw_opcode
= FW_RI_FR_NSMR_WR
;
602 swsqe
->opcode
= FW_RI_FAST_REGISTER
;
603 err
= build_fastreg(wqe
, wr
, &len16
);
605 case IB_WR_LOCAL_INV
:
606 if (wr
->send_flags
& IB_SEND_FENCE
)
607 fw_flags
|= FW_RI_LOCAL_FENCE_FLAG
;
608 fw_opcode
= FW_RI_INV_LSTAG_WR
;
609 swsqe
->opcode
= FW_RI_LOCAL_INV
;
610 err
= build_inv_stag(wqe
, wr
, &len16
);
613 PDBG("%s post of type=%d TBD!\n", __func__
,
621 swsqe
->idx
= qhp
->wq
.sq
.pidx
;
623 swsqe
->signaled
= (wr
->send_flags
& IB_SEND_SIGNALED
);
624 swsqe
->wr_id
= wr
->wr_id
;
626 init_wr_hdr(wqe
, qhp
->wq
.sq
.pidx
, fw_opcode
, fw_flags
, len16
);
628 PDBG("%s cookie 0x%llx pidx 0x%x opcode 0x%x read_len %u\n",
629 __func__
, (unsigned long long)wr
->wr_id
, qhp
->wq
.sq
.pidx
,
630 swsqe
->opcode
, swsqe
->read_len
);
633 t4_sq_produce(&qhp
->wq
, len16
);
634 idx
+= DIV_ROUND_UP(len16
*16, T4_EQ_ENTRY_SIZE
);
636 if (t4_wq_db_enabled(&qhp
->wq
))
637 t4_ring_sq_db(&qhp
->wq
, idx
);
638 spin_unlock_irqrestore(&qhp
->lock
, flag
);
642 int c4iw_post_receive(struct ib_qp
*ibqp
, struct ib_recv_wr
*wr
,
643 struct ib_recv_wr
**bad_wr
)
647 union t4_recv_wr
*wqe
;
653 qhp
= to_c4iw_qp(ibqp
);
654 spin_lock_irqsave(&qhp
->lock
, flag
);
655 if (t4_wq_in_error(&qhp
->wq
)) {
656 spin_unlock_irqrestore(&qhp
->lock
, flag
);
659 num_wrs
= t4_rq_avail(&qhp
->wq
);
661 spin_unlock_irqrestore(&qhp
->lock
, flag
);
665 if (wr
->num_sge
> T4_MAX_RECV_SGE
) {
670 wqe
= (union t4_recv_wr
*)((u8
*)qhp
->wq
.rq
.queue
+
674 err
= build_rdma_recv(qhp
, wqe
, wr
, &len16
);
682 qhp
->wq
.rq
.sw_rq
[qhp
->wq
.rq
.pidx
].wr_id
= wr
->wr_id
;
684 wqe
->recv
.opcode
= FW_RI_RECV_WR
;
686 wqe
->recv
.wrid
= qhp
->wq
.rq
.pidx
;
690 wqe
->recv
.len16
= len16
;
691 PDBG("%s cookie 0x%llx pidx %u\n", __func__
,
692 (unsigned long long) wr
->wr_id
, qhp
->wq
.rq
.pidx
);
693 t4_rq_produce(&qhp
->wq
, len16
);
694 idx
+= DIV_ROUND_UP(len16
*16, T4_EQ_ENTRY_SIZE
);
698 if (t4_wq_db_enabled(&qhp
->wq
))
699 t4_ring_rq_db(&qhp
->wq
, idx
);
700 spin_unlock_irqrestore(&qhp
->lock
, flag
);
704 int c4iw_bind_mw(struct ib_qp
*qp
, struct ib_mw
*mw
, struct ib_mw_bind
*mw_bind
)
709 static inline void build_term_codes(struct t4_cqe
*err_cqe
, u8
*layer_type
,
719 *layer_type
= LAYER_RDMAP
|DDP_LOCAL_CATA
;
724 status
= CQE_STATUS(err_cqe
);
725 opcode
= CQE_OPCODE(err_cqe
);
726 rqtype
= RQ_TYPE(err_cqe
);
727 send_inv
= (opcode
== FW_RI_SEND_WITH_INV
) ||
728 (opcode
== FW_RI_SEND_WITH_SE_INV
);
729 tagged
= (opcode
== FW_RI_RDMA_WRITE
) ||
730 (rqtype
&& (opcode
== FW_RI_READ_RESP
));
735 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_OP
;
736 *ecode
= RDMAP_CANT_INV_STAG
;
738 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
739 *ecode
= RDMAP_INV_STAG
;
743 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
744 if ((opcode
== FW_RI_SEND_WITH_INV
) ||
745 (opcode
== FW_RI_SEND_WITH_SE_INV
))
746 *ecode
= RDMAP_CANT_INV_STAG
;
748 *ecode
= RDMAP_STAG_NOT_ASSOC
;
751 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
752 *ecode
= RDMAP_STAG_NOT_ASSOC
;
755 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
756 *ecode
= RDMAP_ACC_VIOL
;
759 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
760 *ecode
= RDMAP_TO_WRAP
;
764 *layer_type
= LAYER_DDP
|DDP_TAGGED_ERR
;
765 *ecode
= DDPT_BASE_BOUNDS
;
767 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
768 *ecode
= RDMAP_BASE_BOUNDS
;
771 case T4_ERR_INVALIDATE_SHARED_MR
:
772 case T4_ERR_INVALIDATE_MR_WITH_MW_BOUND
:
773 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_OP
;
774 *ecode
= RDMAP_CANT_INV_STAG
;
777 case T4_ERR_ECC_PSTAG
:
778 case T4_ERR_INTERNAL_ERR
:
779 *layer_type
= LAYER_RDMAP
|RDMAP_LOCAL_CATA
;
782 case T4_ERR_OUT_OF_RQE
:
783 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
784 *ecode
= DDPU_INV_MSN_NOBUF
;
786 case T4_ERR_PBL_ADDR_BOUND
:
787 *layer_type
= LAYER_DDP
|DDP_TAGGED_ERR
;
788 *ecode
= DDPT_BASE_BOUNDS
;
791 *layer_type
= LAYER_MPA
|DDP_LLP
;
792 *ecode
= MPA_CRC_ERR
;
795 *layer_type
= LAYER_MPA
|DDP_LLP
;
796 *ecode
= MPA_MARKER_ERR
;
798 case T4_ERR_PDU_LEN_ERR
:
799 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
800 *ecode
= DDPU_MSG_TOOBIG
;
802 case T4_ERR_DDP_VERSION
:
804 *layer_type
= LAYER_DDP
|DDP_TAGGED_ERR
;
805 *ecode
= DDPT_INV_VERS
;
807 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
808 *ecode
= DDPU_INV_VERS
;
811 case T4_ERR_RDMA_VERSION
:
812 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_OP
;
813 *ecode
= RDMAP_INV_VERS
;
816 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_OP
;
817 *ecode
= RDMAP_INV_OPCODE
;
819 case T4_ERR_DDP_QUEUE_NUM
:
820 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
821 *ecode
= DDPU_INV_QN
;
825 case T4_ERR_MSN_RANGE
:
826 case T4_ERR_IRD_OVERFLOW
:
827 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
828 *ecode
= DDPU_INV_MSN_RANGE
;
831 *layer_type
= LAYER_DDP
|DDP_LOCAL_CATA
;
835 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
836 *ecode
= DDPU_INV_MO
;
839 *layer_type
= LAYER_RDMAP
|DDP_LOCAL_CATA
;
845 int c4iw_post_zb_read(struct c4iw_qp
*qhp
)
851 PDBG("%s enter\n", __func__
);
852 skb
= alloc_skb(40, GFP_KERNEL
);
854 printk(KERN_ERR
"%s cannot send zb_read!!\n", __func__
);
857 set_wr_txq(skb
, CPL_PRIORITY_DATA
, qhp
->ep
->txq_idx
);
859 wqe
= (union t4_wr
*)skb_put(skb
, sizeof wqe
->read
);
860 memset(wqe
, 0, sizeof wqe
->read
);
861 wqe
->read
.r2
= cpu_to_be64(0);
862 wqe
->read
.stag_sink
= cpu_to_be32(1);
863 wqe
->read
.to_sink_hi
= cpu_to_be32(0);
864 wqe
->read
.to_sink_lo
= cpu_to_be32(1);
865 wqe
->read
.stag_src
= cpu_to_be32(1);
866 wqe
->read
.plen
= cpu_to_be32(0);
867 wqe
->read
.to_src_hi
= cpu_to_be32(0);
868 wqe
->read
.to_src_lo
= cpu_to_be32(1);
869 len16
= DIV_ROUND_UP(sizeof wqe
->read
, 16);
870 init_wr_hdr(wqe
, 0, FW_RI_RDMA_READ_WR
, FW_RI_COMPLETION_FLAG
, len16
);
872 return c4iw_ofld_send(&qhp
->rhp
->rdev
, skb
);
875 static void post_terminate(struct c4iw_qp
*qhp
, struct t4_cqe
*err_cqe
,
878 struct fw_ri_wr
*wqe
;
880 struct terminate_message
*term
;
882 PDBG("%s qhp %p qid 0x%x tid %u\n", __func__
, qhp
, qhp
->wq
.sq
.qid
,
885 skb
= alloc_skb(sizeof *wqe
, gfp
);
888 set_wr_txq(skb
, CPL_PRIORITY_DATA
, qhp
->ep
->txq_idx
);
890 wqe
= (struct fw_ri_wr
*)__skb_put(skb
, sizeof(*wqe
));
891 memset(wqe
, 0, sizeof *wqe
);
892 wqe
->op_compl
= cpu_to_be32(FW_WR_OP(FW_RI_INIT_WR
));
893 wqe
->flowid_len16
= cpu_to_be32(
894 FW_WR_FLOWID(qhp
->ep
->hwtid
) |
895 FW_WR_LEN16(DIV_ROUND_UP(sizeof *wqe
, 16)));
897 wqe
->u
.terminate
.type
= FW_RI_TYPE_TERMINATE
;
898 wqe
->u
.terminate
.immdlen
= cpu_to_be32(sizeof *term
);
899 term
= (struct terminate_message
*)wqe
->u
.terminate
.termmsg
;
900 build_term_codes(err_cqe
, &term
->layer_etype
, &term
->ecode
);
901 c4iw_ofld_send(&qhp
->rhp
->rdev
, skb
);
905 * Assumes qhp lock is held.
907 static void __flush_qp(struct c4iw_qp
*qhp
, struct c4iw_cq
*rchp
,
908 struct c4iw_cq
*schp
, unsigned long *flag
)
913 PDBG("%s qhp %p rchp %p schp %p\n", __func__
, qhp
, rchp
, schp
);
914 /* take a ref on the qhp since we must release the lock */
915 atomic_inc(&qhp
->refcnt
);
916 spin_unlock_irqrestore(&qhp
->lock
, *flag
);
918 /* locking hierarchy: cq lock first, then qp lock. */
919 spin_lock_irqsave(&rchp
->lock
, *flag
);
920 spin_lock(&qhp
->lock
);
921 c4iw_flush_hw_cq(&rchp
->cq
);
922 c4iw_count_rcqes(&rchp
->cq
, &qhp
->wq
, &count
);
923 flushed
= c4iw_flush_rq(&qhp
->wq
, &rchp
->cq
, count
);
924 spin_unlock(&qhp
->lock
);
925 spin_unlock_irqrestore(&rchp
->lock
, *flag
);
927 (*rchp
->ibcq
.comp_handler
)(&rchp
->ibcq
, rchp
->ibcq
.cq_context
);
929 /* locking hierarchy: cq lock first, then qp lock. */
930 spin_lock_irqsave(&schp
->lock
, *flag
);
931 spin_lock(&qhp
->lock
);
932 c4iw_flush_hw_cq(&schp
->cq
);
933 c4iw_count_scqes(&schp
->cq
, &qhp
->wq
, &count
);
934 flushed
= c4iw_flush_sq(&qhp
->wq
, &schp
->cq
, count
);
935 spin_unlock(&qhp
->lock
);
936 spin_unlock_irqrestore(&schp
->lock
, *flag
);
938 (*schp
->ibcq
.comp_handler
)(&schp
->ibcq
, schp
->ibcq
.cq_context
);
941 if (atomic_dec_and_test(&qhp
->refcnt
))
944 spin_lock_irqsave(&qhp
->lock
, *flag
);
947 static void flush_qp(struct c4iw_qp
*qhp
, unsigned long *flag
)
949 struct c4iw_cq
*rchp
, *schp
;
951 rchp
= get_chp(qhp
->rhp
, qhp
->attr
.rcq
);
952 schp
= get_chp(qhp
->rhp
, qhp
->attr
.scq
);
954 if (qhp
->ibqp
.uobject
) {
955 t4_set_wq_in_error(&qhp
->wq
);
956 t4_set_cq_in_error(&rchp
->cq
);
958 t4_set_cq_in_error(&schp
->cq
);
961 __flush_qp(qhp
, rchp
, schp
, flag
);
964 static int rdma_fini(struct c4iw_dev
*rhp
, struct c4iw_qp
*qhp
,
967 struct fw_ri_wr
*wqe
;
969 struct c4iw_wr_wait wr_wait
;
972 PDBG("%s qhp %p qid 0x%x tid %u\n", __func__
, qhp
, qhp
->wq
.sq
.qid
,
975 skb
= alloc_skb(sizeof *wqe
, GFP_KERNEL
);
978 set_wr_txq(skb
, CPL_PRIORITY_DATA
, ep
->txq_idx
);
980 wqe
= (struct fw_ri_wr
*)__skb_put(skb
, sizeof(*wqe
));
981 memset(wqe
, 0, sizeof *wqe
);
982 wqe
->op_compl
= cpu_to_be32(
983 FW_WR_OP(FW_RI_INIT_WR
) |
985 wqe
->flowid_len16
= cpu_to_be32(
986 FW_WR_FLOWID(ep
->hwtid
) |
987 FW_WR_LEN16(DIV_ROUND_UP(sizeof *wqe
, 16)));
988 wqe
->cookie
= (u64
)&wr_wait
;
990 wqe
->u
.fini
.type
= FW_RI_TYPE_FINI
;
991 c4iw_init_wr_wait(&wr_wait
);
992 ret
= c4iw_ofld_send(&rhp
->rdev
, skb
);
996 wait_event_timeout(wr_wait
.wait
, wr_wait
.done
, C4IW_WR_TO
);
998 printk(KERN_ERR MOD
"Device %s not responding!\n",
999 pci_name(rhp
->rdev
.lldi
.pdev
));
1000 rhp
->rdev
.flags
= T4_FATAL_ERROR
;
1005 printk(KERN_WARNING MOD
1006 "%s: Abnormal close qpid %d ret %u\n",
1007 pci_name(rhp
->rdev
.lldi
.pdev
), qhp
->wq
.sq
.qid
,
1011 PDBG("%s ret %d\n", __func__
, ret
);
1015 static void build_rtr_msg(u8 p2p_type
, struct fw_ri_init
*init
)
1017 memset(&init
->u
, 0, sizeof init
->u
);
1019 case FW_RI_INIT_P2PTYPE_RDMA_WRITE
:
1020 init
->u
.write
.opcode
= FW_RI_RDMA_WRITE_WR
;
1021 init
->u
.write
.stag_sink
= cpu_to_be32(1);
1022 init
->u
.write
.to_sink
= cpu_to_be64(1);
1023 init
->u
.write
.u
.immd_src
[0].op
= FW_RI_DATA_IMMD
;
1024 init
->u
.write
.len16
= DIV_ROUND_UP(sizeof init
->u
.write
+
1025 sizeof(struct fw_ri_immd
),
1028 case FW_RI_INIT_P2PTYPE_READ_REQ
:
1029 init
->u
.write
.opcode
= FW_RI_RDMA_READ_WR
;
1030 init
->u
.read
.stag_src
= cpu_to_be32(1);
1031 init
->u
.read
.to_src_lo
= cpu_to_be32(1);
1032 init
->u
.read
.stag_sink
= cpu_to_be32(1);
1033 init
->u
.read
.to_sink_lo
= cpu_to_be32(1);
1034 init
->u
.read
.len16
= DIV_ROUND_UP(sizeof init
->u
.read
, 16);
1039 static int rdma_init(struct c4iw_dev
*rhp
, struct c4iw_qp
*qhp
)
1041 struct fw_ri_wr
*wqe
;
1043 struct c4iw_wr_wait wr_wait
;
1044 struct sk_buff
*skb
;
1046 PDBG("%s qhp %p qid 0x%x tid %u\n", __func__
, qhp
, qhp
->wq
.sq
.qid
,
1049 skb
= alloc_skb(sizeof *wqe
, GFP_KERNEL
);
1052 set_wr_txq(skb
, CPL_PRIORITY_DATA
, qhp
->ep
->txq_idx
);
1054 wqe
= (struct fw_ri_wr
*)__skb_put(skb
, sizeof(*wqe
));
1055 memset(wqe
, 0, sizeof *wqe
);
1056 wqe
->op_compl
= cpu_to_be32(
1057 FW_WR_OP(FW_RI_INIT_WR
) |
1059 wqe
->flowid_len16
= cpu_to_be32(
1060 FW_WR_FLOWID(qhp
->ep
->hwtid
) |
1061 FW_WR_LEN16(DIV_ROUND_UP(sizeof *wqe
, 16)));
1063 wqe
->cookie
= (u64
)&wr_wait
;
1065 wqe
->u
.init
.type
= FW_RI_TYPE_INIT
;
1066 wqe
->u
.init
.mpareqbit_p2ptype
=
1067 V_FW_RI_WR_MPAREQBIT(qhp
->attr
.mpa_attr
.initiator
) |
1068 V_FW_RI_WR_P2PTYPE(qhp
->attr
.mpa_attr
.p2p_type
);
1069 wqe
->u
.init
.mpa_attrs
= FW_RI_MPA_IETF_ENABLE
;
1070 if (qhp
->attr
.mpa_attr
.recv_marker_enabled
)
1071 wqe
->u
.init
.mpa_attrs
|= FW_RI_MPA_RX_MARKER_ENABLE
;
1072 if (qhp
->attr
.mpa_attr
.xmit_marker_enabled
)
1073 wqe
->u
.init
.mpa_attrs
|= FW_RI_MPA_TX_MARKER_ENABLE
;
1074 if (qhp
->attr
.mpa_attr
.crc_enabled
)
1075 wqe
->u
.init
.mpa_attrs
|= FW_RI_MPA_CRC_ENABLE
;
1077 wqe
->u
.init
.qp_caps
= FW_RI_QP_RDMA_READ_ENABLE
|
1078 FW_RI_QP_RDMA_WRITE_ENABLE
|
1079 FW_RI_QP_BIND_ENABLE
;
1080 if (!qhp
->ibqp
.uobject
)
1081 wqe
->u
.init
.qp_caps
|= FW_RI_QP_FAST_REGISTER_ENABLE
|
1082 FW_RI_QP_STAG0_ENABLE
;
1083 wqe
->u
.init
.nrqe
= cpu_to_be16(t4_rqes_posted(&qhp
->wq
));
1084 wqe
->u
.init
.pdid
= cpu_to_be32(qhp
->attr
.pd
);
1085 wqe
->u
.init
.qpid
= cpu_to_be32(qhp
->wq
.sq
.qid
);
1086 wqe
->u
.init
.sq_eqid
= cpu_to_be32(qhp
->wq
.sq
.qid
);
1087 wqe
->u
.init
.rq_eqid
= cpu_to_be32(qhp
->wq
.rq
.qid
);
1088 wqe
->u
.init
.scqid
= cpu_to_be32(qhp
->attr
.scq
);
1089 wqe
->u
.init
.rcqid
= cpu_to_be32(qhp
->attr
.rcq
);
1090 wqe
->u
.init
.ord_max
= cpu_to_be32(qhp
->attr
.max_ord
);
1091 wqe
->u
.init
.ird_max
= cpu_to_be32(qhp
->attr
.max_ird
);
1092 wqe
->u
.init
.iss
= cpu_to_be32(qhp
->ep
->snd_seq
);
1093 wqe
->u
.init
.irs
= cpu_to_be32(qhp
->ep
->rcv_seq
);
1094 wqe
->u
.init
.hwrqsize
= cpu_to_be32(qhp
->wq
.rq
.rqt_size
);
1095 wqe
->u
.init
.hwrqaddr
= cpu_to_be32(qhp
->wq
.rq
.rqt_hwaddr
-
1096 rhp
->rdev
.lldi
.vr
->rq
.start
);
1097 if (qhp
->attr
.mpa_attr
.initiator
)
1098 build_rtr_msg(qhp
->attr
.mpa_attr
.p2p_type
, &wqe
->u
.init
);
1100 c4iw_init_wr_wait(&wr_wait
);
1101 ret
= c4iw_ofld_send(&rhp
->rdev
, skb
);
1105 wait_event_timeout(wr_wait
.wait
, wr_wait
.done
, C4IW_WR_TO
);
1106 if (!wr_wait
.done
) {
1107 printk(KERN_ERR MOD
"Device %s not responding!\n",
1108 pci_name(rhp
->rdev
.lldi
.pdev
));
1109 rhp
->rdev
.flags
= T4_FATAL_ERROR
;
1114 PDBG("%s ret %d\n", __func__
, ret
);
1118 int c4iw_modify_qp(struct c4iw_dev
*rhp
, struct c4iw_qp
*qhp
,
1119 enum c4iw_qp_attr_mask mask
,
1120 struct c4iw_qp_attributes
*attrs
,
1124 struct c4iw_qp_attributes newattr
= qhp
->attr
;
1130 struct c4iw_ep
*ep
= NULL
;
1132 PDBG("%s qhp %p sqid 0x%x rqid 0x%x ep %p state %d -> %d\n", __func__
,
1133 qhp
, qhp
->wq
.sq
.qid
, qhp
->wq
.rq
.qid
, qhp
->ep
, qhp
->attr
.state
,
1134 (mask
& C4IW_QP_ATTR_NEXT_STATE
) ? attrs
->next_state
: -1);
1136 spin_lock_irqsave(&qhp
->lock
, flag
);
1138 /* Process attr changes if in IDLE */
1139 if (mask
& C4IW_QP_ATTR_VALID_MODIFY
) {
1140 if (qhp
->attr
.state
!= C4IW_QP_STATE_IDLE
) {
1144 if (mask
& C4IW_QP_ATTR_ENABLE_RDMA_READ
)
1145 newattr
.enable_rdma_read
= attrs
->enable_rdma_read
;
1146 if (mask
& C4IW_QP_ATTR_ENABLE_RDMA_WRITE
)
1147 newattr
.enable_rdma_write
= attrs
->enable_rdma_write
;
1148 if (mask
& C4IW_QP_ATTR_ENABLE_RDMA_BIND
)
1149 newattr
.enable_bind
= attrs
->enable_bind
;
1150 if (mask
& C4IW_QP_ATTR_MAX_ORD
) {
1151 if (attrs
->max_ord
> c4iw_max_read_depth
) {
1155 newattr
.max_ord
= attrs
->max_ord
;
1157 if (mask
& C4IW_QP_ATTR_MAX_IRD
) {
1158 if (attrs
->max_ird
> c4iw_max_read_depth
) {
1162 newattr
.max_ird
= attrs
->max_ird
;
1164 qhp
->attr
= newattr
;
1167 if (!(mask
& C4IW_QP_ATTR_NEXT_STATE
))
1169 if (qhp
->attr
.state
== attrs
->next_state
)
1172 switch (qhp
->attr
.state
) {
1173 case C4IW_QP_STATE_IDLE
:
1174 switch (attrs
->next_state
) {
1175 case C4IW_QP_STATE_RTS
:
1176 if (!(mask
& C4IW_QP_ATTR_LLP_STREAM_HANDLE
)) {
1180 if (!(mask
& C4IW_QP_ATTR_MPA_ATTR
)) {
1184 qhp
->attr
.mpa_attr
= attrs
->mpa_attr
;
1185 qhp
->attr
.llp_stream_handle
= attrs
->llp_stream_handle
;
1186 qhp
->ep
= qhp
->attr
.llp_stream_handle
;
1187 qhp
->attr
.state
= C4IW_QP_STATE_RTS
;
1190 * Ref the endpoint here and deref when we
1191 * disassociate the endpoint from the QP. This
1192 * happens in CLOSING->IDLE transition or *->ERROR
1195 c4iw_get_ep(&qhp
->ep
->com
);
1196 spin_unlock_irqrestore(&qhp
->lock
, flag
);
1197 ret
= rdma_init(rhp
, qhp
);
1198 spin_lock_irqsave(&qhp
->lock
, flag
);
1202 case C4IW_QP_STATE_ERROR
:
1203 qhp
->attr
.state
= C4IW_QP_STATE_ERROR
;
1204 flush_qp(qhp
, &flag
);
1211 case C4IW_QP_STATE_RTS
:
1212 switch (attrs
->next_state
) {
1213 case C4IW_QP_STATE_CLOSING
:
1214 BUG_ON(atomic_read(&qhp
->ep
->com
.kref
.refcount
) < 2);
1215 qhp
->attr
.state
= C4IW_QP_STATE_CLOSING
;
1220 c4iw_get_ep(&ep
->com
);
1222 spin_unlock_irqrestore(&qhp
->lock
, flag
);
1223 ret
= rdma_fini(rhp
, qhp
, ep
);
1224 spin_lock_irqsave(&qhp
->lock
, flag
);
1226 c4iw_get_ep(&ep
->com
);
1227 disconnect
= abort
= 1;
1231 case C4IW_QP_STATE_TERMINATE
:
1232 qhp
->attr
.state
= C4IW_QP_STATE_TERMINATE
;
1233 if (qhp
->ibqp
.uobject
)
1234 t4_set_wq_in_error(&qhp
->wq
);
1236 c4iw_get_ep(&ep
->com
);
1240 case C4IW_QP_STATE_ERROR
:
1241 qhp
->attr
.state
= C4IW_QP_STATE_ERROR
;
1246 c4iw_get_ep(&ep
->com
);
1255 case C4IW_QP_STATE_CLOSING
:
1260 switch (attrs
->next_state
) {
1261 case C4IW_QP_STATE_IDLE
:
1262 flush_qp(qhp
, &flag
);
1263 qhp
->attr
.state
= C4IW_QP_STATE_IDLE
;
1264 qhp
->attr
.llp_stream_handle
= NULL
;
1265 c4iw_put_ep(&qhp
->ep
->com
);
1267 wake_up(&qhp
->wait
);
1269 case C4IW_QP_STATE_ERROR
:
1276 case C4IW_QP_STATE_ERROR
:
1277 if (attrs
->next_state
!= C4IW_QP_STATE_IDLE
) {
1281 if (!t4_sq_empty(&qhp
->wq
) || !t4_rq_empty(&qhp
->wq
)) {
1285 qhp
->attr
.state
= C4IW_QP_STATE_IDLE
;
1287 case C4IW_QP_STATE_TERMINATE
:
1295 printk(KERN_ERR
"%s in a bad state %d\n",
1296 __func__
, qhp
->attr
.state
);
1303 PDBG("%s disassociating ep %p qpid 0x%x\n", __func__
, qhp
->ep
,
1306 /* disassociate the LLP connection */
1307 qhp
->attr
.llp_stream_handle
= NULL
;
1310 qhp
->attr
.state
= C4IW_QP_STATE_ERROR
;
1312 wake_up(&qhp
->wait
);
1314 flush_qp(qhp
, &flag
);
1316 spin_unlock_irqrestore(&qhp
->lock
, flag
);
1319 post_terminate(qhp
, NULL
, internal
? GFP_ATOMIC
: GFP_KERNEL
);
1322 * If disconnect is 1, then we need to initiate a disconnect
1323 * on the EP. This can be a normal close (RTS->CLOSING) or
1324 * an abnormal close (RTS/CLOSING->ERROR).
1327 c4iw_ep_disconnect(ep
, abort
, internal
? GFP_ATOMIC
:
1329 c4iw_put_ep(&ep
->com
);
1333 * If free is 1, then we've disassociated the EP from the QP
1334 * and we need to dereference the EP.
1337 c4iw_put_ep(&ep
->com
);
1339 PDBG("%s exit state %d\n", __func__
, qhp
->attr
.state
);
1343 int c4iw_destroy_qp(struct ib_qp
*ib_qp
)
1345 struct c4iw_dev
*rhp
;
1346 struct c4iw_qp
*qhp
;
1347 struct c4iw_qp_attributes attrs
;
1348 struct c4iw_ucontext
*ucontext
;
1350 qhp
= to_c4iw_qp(ib_qp
);
1353 attrs
.next_state
= C4IW_QP_STATE_ERROR
;
1354 c4iw_modify_qp(rhp
, qhp
, C4IW_QP_ATTR_NEXT_STATE
, &attrs
, 0);
1355 wait_event(qhp
->wait
, !qhp
->ep
);
1357 remove_handle(rhp
, &rhp
->qpidr
, qhp
->wq
.sq
.qid
);
1358 atomic_dec(&qhp
->refcnt
);
1359 wait_event(qhp
->wait
, !atomic_read(&qhp
->refcnt
));
1361 ucontext
= ib_qp
->uobject
?
1362 to_c4iw_ucontext(ib_qp
->uobject
->context
) : NULL
;
1363 destroy_qp(&rhp
->rdev
, &qhp
->wq
,
1364 ucontext
? &ucontext
->uctx
: &rhp
->rdev
.uctx
);
1366 PDBG("%s ib_qp %p qpid 0x%0x\n", __func__
, ib_qp
, qhp
->wq
.sq
.qid
);
1371 struct ib_qp
*c4iw_create_qp(struct ib_pd
*pd
, struct ib_qp_init_attr
*attrs
,
1372 struct ib_udata
*udata
)
1374 struct c4iw_dev
*rhp
;
1375 struct c4iw_qp
*qhp
;
1376 struct c4iw_pd
*php
;
1377 struct c4iw_cq
*schp
;
1378 struct c4iw_cq
*rchp
;
1379 struct c4iw_create_qp_resp uresp
;
1381 struct c4iw_ucontext
*ucontext
;
1383 struct c4iw_mm_entry
*mm1
, *mm2
, *mm3
, *mm4
;
1385 PDBG("%s ib_pd %p\n", __func__
, pd
);
1387 if (attrs
->qp_type
!= IB_QPT_RC
)
1388 return ERR_PTR(-EINVAL
);
1390 php
= to_c4iw_pd(pd
);
1392 schp
= get_chp(rhp
, ((struct c4iw_cq
*)attrs
->send_cq
)->cq
.cqid
);
1393 rchp
= get_chp(rhp
, ((struct c4iw_cq
*)attrs
->recv_cq
)->cq
.cqid
);
1395 return ERR_PTR(-EINVAL
);
1397 if (attrs
->cap
.max_inline_data
> T4_MAX_SEND_INLINE
)
1398 return ERR_PTR(-EINVAL
);
1400 rqsize
= roundup(attrs
->cap
.max_recv_wr
+ 1, 16);
1401 if (rqsize
> T4_MAX_RQ_SIZE
)
1402 return ERR_PTR(-E2BIG
);
1404 sqsize
= roundup(attrs
->cap
.max_send_wr
+ 1, 16);
1405 if (sqsize
> T4_MAX_SQ_SIZE
)
1406 return ERR_PTR(-E2BIG
);
1408 ucontext
= pd
->uobject
? to_c4iw_ucontext(pd
->uobject
->context
) : NULL
;
1411 qhp
= kzalloc(sizeof(*qhp
), GFP_KERNEL
);
1413 return ERR_PTR(-ENOMEM
);
1414 qhp
->wq
.sq
.size
= sqsize
;
1415 qhp
->wq
.sq
.memsize
= (sqsize
+ 1) * sizeof *qhp
->wq
.sq
.queue
;
1416 qhp
->wq
.rq
.size
= rqsize
;
1417 qhp
->wq
.rq
.memsize
= (rqsize
+ 1) * sizeof *qhp
->wq
.rq
.queue
;
1420 qhp
->wq
.sq
.memsize
= roundup(qhp
->wq
.sq
.memsize
, PAGE_SIZE
);
1421 qhp
->wq
.rq
.memsize
= roundup(qhp
->wq
.rq
.memsize
, PAGE_SIZE
);
1424 PDBG("%s sqsize %u sqmemsize %zu rqsize %u rqmemsize %zu\n",
1425 __func__
, sqsize
, qhp
->wq
.sq
.memsize
, rqsize
, qhp
->wq
.rq
.memsize
);
1427 ret
= create_qp(&rhp
->rdev
, &qhp
->wq
, &schp
->cq
, &rchp
->cq
,
1428 ucontext
? &ucontext
->uctx
: &rhp
->rdev
.uctx
);
1432 attrs
->cap
.max_recv_wr
= rqsize
- 1;
1433 attrs
->cap
.max_send_wr
= sqsize
- 1;
1434 attrs
->cap
.max_inline_data
= T4_MAX_SEND_INLINE
;
1437 qhp
->attr
.pd
= php
->pdid
;
1438 qhp
->attr
.scq
= ((struct c4iw_cq
*) attrs
->send_cq
)->cq
.cqid
;
1439 qhp
->attr
.rcq
= ((struct c4iw_cq
*) attrs
->recv_cq
)->cq
.cqid
;
1440 qhp
->attr
.sq_num_entries
= attrs
->cap
.max_send_wr
;
1441 qhp
->attr
.rq_num_entries
= attrs
->cap
.max_recv_wr
;
1442 qhp
->attr
.sq_max_sges
= attrs
->cap
.max_send_sge
;
1443 qhp
->attr
.sq_max_sges_rdma_write
= attrs
->cap
.max_send_sge
;
1444 qhp
->attr
.rq_max_sges
= attrs
->cap
.max_recv_sge
;
1445 qhp
->attr
.state
= C4IW_QP_STATE_IDLE
;
1446 qhp
->attr
.next_state
= C4IW_QP_STATE_IDLE
;
1447 qhp
->attr
.enable_rdma_read
= 1;
1448 qhp
->attr
.enable_rdma_write
= 1;
1449 qhp
->attr
.enable_bind
= 1;
1450 qhp
->attr
.max_ord
= 1;
1451 qhp
->attr
.max_ird
= 1;
1452 spin_lock_init(&qhp
->lock
);
1453 init_waitqueue_head(&qhp
->wait
);
1454 atomic_set(&qhp
->refcnt
, 1);
1456 ret
= insert_handle(rhp
, &rhp
->qpidr
, qhp
, qhp
->wq
.sq
.qid
);
1461 mm1
= kmalloc(sizeof *mm1
, GFP_KERNEL
);
1466 mm2
= kmalloc(sizeof *mm2
, GFP_KERNEL
);
1471 mm3
= kmalloc(sizeof *mm3
, GFP_KERNEL
);
1476 mm4
= kmalloc(sizeof *mm4
, GFP_KERNEL
);
1482 uresp
.qid_mask
= rhp
->rdev
.qpmask
;
1483 uresp
.sqid
= qhp
->wq
.sq
.qid
;
1484 uresp
.sq_size
= qhp
->wq
.sq
.size
;
1485 uresp
.sq_memsize
= qhp
->wq
.sq
.memsize
;
1486 uresp
.rqid
= qhp
->wq
.rq
.qid
;
1487 uresp
.rq_size
= qhp
->wq
.rq
.size
;
1488 uresp
.rq_memsize
= qhp
->wq
.rq
.memsize
;
1489 spin_lock(&ucontext
->mmap_lock
);
1490 uresp
.sq_key
= ucontext
->key
;
1491 ucontext
->key
+= PAGE_SIZE
;
1492 uresp
.rq_key
= ucontext
->key
;
1493 ucontext
->key
+= PAGE_SIZE
;
1494 uresp
.sq_db_gts_key
= ucontext
->key
;
1495 ucontext
->key
+= PAGE_SIZE
;
1496 uresp
.rq_db_gts_key
= ucontext
->key
;
1497 ucontext
->key
+= PAGE_SIZE
;
1498 spin_unlock(&ucontext
->mmap_lock
);
1499 ret
= ib_copy_to_udata(udata
, &uresp
, sizeof uresp
);
1502 mm1
->key
= uresp
.sq_key
;
1503 mm1
->addr
= virt_to_phys(qhp
->wq
.sq
.queue
);
1504 mm1
->len
= PAGE_ALIGN(qhp
->wq
.sq
.memsize
);
1505 insert_mmap(ucontext
, mm1
);
1506 mm2
->key
= uresp
.rq_key
;
1507 mm2
->addr
= virt_to_phys(qhp
->wq
.rq
.queue
);
1508 mm2
->len
= PAGE_ALIGN(qhp
->wq
.rq
.memsize
);
1509 insert_mmap(ucontext
, mm2
);
1510 mm3
->key
= uresp
.sq_db_gts_key
;
1511 mm3
->addr
= qhp
->wq
.sq
.udb
;
1512 mm3
->len
= PAGE_SIZE
;
1513 insert_mmap(ucontext
, mm3
);
1514 mm4
->key
= uresp
.rq_db_gts_key
;
1515 mm4
->addr
= qhp
->wq
.rq
.udb
;
1516 mm4
->len
= PAGE_SIZE
;
1517 insert_mmap(ucontext
, mm4
);
1519 qhp
->ibqp
.qp_num
= qhp
->wq
.sq
.qid
;
1520 init_timer(&(qhp
->timer
));
1521 PDBG("%s qhp %p sq_num_entries %d, rq_num_entries %d qpid 0x%0x\n",
1522 __func__
, qhp
, qhp
->attr
.sq_num_entries
, qhp
->attr
.rq_num_entries
,
1534 remove_handle(rhp
, &rhp
->qpidr
, qhp
->wq
.sq
.qid
);
1536 destroy_qp(&rhp
->rdev
, &qhp
->wq
,
1537 ucontext
? &ucontext
->uctx
: &rhp
->rdev
.uctx
);
1540 return ERR_PTR(ret
);
1543 int c4iw_ib_modify_qp(struct ib_qp
*ibqp
, struct ib_qp_attr
*attr
,
1544 int attr_mask
, struct ib_udata
*udata
)
1546 struct c4iw_dev
*rhp
;
1547 struct c4iw_qp
*qhp
;
1548 enum c4iw_qp_attr_mask mask
= 0;
1549 struct c4iw_qp_attributes attrs
;
1551 PDBG("%s ib_qp %p\n", __func__
, ibqp
);
1553 /* iwarp does not support the RTR state */
1554 if ((attr_mask
& IB_QP_STATE
) && (attr
->qp_state
== IB_QPS_RTR
))
1555 attr_mask
&= ~IB_QP_STATE
;
1557 /* Make sure we still have something left to do */
1561 memset(&attrs
, 0, sizeof attrs
);
1562 qhp
= to_c4iw_qp(ibqp
);
1565 attrs
.next_state
= c4iw_convert_state(attr
->qp_state
);
1566 attrs
.enable_rdma_read
= (attr
->qp_access_flags
&
1567 IB_ACCESS_REMOTE_READ
) ? 1 : 0;
1568 attrs
.enable_rdma_write
= (attr
->qp_access_flags
&
1569 IB_ACCESS_REMOTE_WRITE
) ? 1 : 0;
1570 attrs
.enable_bind
= (attr
->qp_access_flags
& IB_ACCESS_MW_BIND
) ? 1 : 0;
1573 mask
|= (attr_mask
& IB_QP_STATE
) ? C4IW_QP_ATTR_NEXT_STATE
: 0;
1574 mask
|= (attr_mask
& IB_QP_ACCESS_FLAGS
) ?
1575 (C4IW_QP_ATTR_ENABLE_RDMA_READ
|
1576 C4IW_QP_ATTR_ENABLE_RDMA_WRITE
|
1577 C4IW_QP_ATTR_ENABLE_RDMA_BIND
) : 0;
1579 return c4iw_modify_qp(rhp
, qhp
, mask
, &attrs
, 0);
1582 struct ib_qp
*c4iw_get_qp(struct ib_device
*dev
, int qpn
)
1584 PDBG("%s ib_dev %p qpn 0x%x\n", __func__
, dev
, qpn
);
1585 return (struct ib_qp
*)get_qhp(to_c4iw_dev(dev
), qpn
);