2 * Copyright (c) 2006 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
32 #include "iwch_provider.h"
36 #include "cxio_resource.h"
40 static int build_rdma_send(union t3_wr
*wqe
, struct ib_send_wr
*wr
,
48 if (wr
->send_flags
& IB_SEND_SOLICITED
)
49 wqe
->send
.rdmaop
= T3_SEND_WITH_SE
;
51 wqe
->send
.rdmaop
= T3_SEND
;
52 wqe
->send
.rem_stag
= 0;
54 case IB_WR_SEND_WITH_INV
:
55 if (wr
->send_flags
& IB_SEND_SOLICITED
)
56 wqe
->send
.rdmaop
= T3_SEND_WITH_SE_INV
;
58 wqe
->send
.rdmaop
= T3_SEND_WITH_INV
;
59 wqe
->send
.rem_stag
= cpu_to_be32(wr
->ex
.invalidate_rkey
);
64 if (wr
->num_sge
> T3_MAX_SGE
)
66 wqe
->send
.reserved
[0] = 0;
67 wqe
->send
.reserved
[1] = 0;
68 wqe
->send
.reserved
[2] = 0;
70 for (i
= 0; i
< wr
->num_sge
; i
++) {
71 if ((plen
+ wr
->sg_list
[i
].length
) < plen
)
74 plen
+= wr
->sg_list
[i
].length
;
75 wqe
->send
.sgl
[i
].stag
= cpu_to_be32(wr
->sg_list
[i
].lkey
);
76 wqe
->send
.sgl
[i
].len
= cpu_to_be32(wr
->sg_list
[i
].length
);
77 wqe
->send
.sgl
[i
].to
= cpu_to_be64(wr
->sg_list
[i
].addr
);
79 wqe
->send
.num_sgle
= cpu_to_be32(wr
->num_sge
);
80 *flit_cnt
= 4 + ((wr
->num_sge
) << 1);
81 wqe
->send
.plen
= cpu_to_be32(plen
);
85 static int build_rdma_write(union t3_wr
*wqe
, struct ib_send_wr
*wr
,
90 if (wr
->num_sge
> T3_MAX_SGE
)
92 wqe
->write
.rdmaop
= T3_RDMA_WRITE
;
93 wqe
->write
.reserved
[0] = 0;
94 wqe
->write
.reserved
[1] = 0;
95 wqe
->write
.reserved
[2] = 0;
96 wqe
->write
.stag_sink
= cpu_to_be32(wr
->wr
.rdma
.rkey
);
97 wqe
->write
.to_sink
= cpu_to_be64(wr
->wr
.rdma
.remote_addr
);
99 if (wr
->opcode
== IB_WR_RDMA_WRITE_WITH_IMM
) {
101 wqe
->write
.sgl
[0].stag
= wr
->ex
.imm_data
;
102 wqe
->write
.sgl
[0].len
= cpu_to_be32(0);
103 wqe
->write
.num_sgle
= cpu_to_be32(0);
107 for (i
= 0; i
< wr
->num_sge
; i
++) {
108 if ((plen
+ wr
->sg_list
[i
].length
) < plen
) {
111 plen
+= wr
->sg_list
[i
].length
;
112 wqe
->write
.sgl
[i
].stag
=
113 cpu_to_be32(wr
->sg_list
[i
].lkey
);
114 wqe
->write
.sgl
[i
].len
=
115 cpu_to_be32(wr
->sg_list
[i
].length
);
116 wqe
->write
.sgl
[i
].to
=
117 cpu_to_be64(wr
->sg_list
[i
].addr
);
119 wqe
->write
.num_sgle
= cpu_to_be32(wr
->num_sge
);
120 *flit_cnt
= 5 + ((wr
->num_sge
) << 1);
122 wqe
->write
.plen
= cpu_to_be32(plen
);
126 static int build_rdma_read(union t3_wr
*wqe
, struct ib_send_wr
*wr
,
131 wqe
->read
.rdmaop
= T3_READ_REQ
;
132 if (wr
->opcode
== IB_WR_RDMA_READ_WITH_INV
)
133 wqe
->read
.local_inv
= 1;
135 wqe
->read
.local_inv
= 0;
136 wqe
->read
.reserved
[0] = 0;
137 wqe
->read
.reserved
[1] = 0;
138 wqe
->read
.rem_stag
= cpu_to_be32(wr
->wr
.rdma
.rkey
);
139 wqe
->read
.rem_to
= cpu_to_be64(wr
->wr
.rdma
.remote_addr
);
140 wqe
->read
.local_stag
= cpu_to_be32(wr
->sg_list
[0].lkey
);
141 wqe
->read
.local_len
= cpu_to_be32(wr
->sg_list
[0].length
);
142 wqe
->read
.local_to
= cpu_to_be64(wr
->sg_list
[0].addr
);
143 *flit_cnt
= sizeof(struct t3_rdma_read_wr
) >> 3;
147 static int build_fastreg(union t3_wr
*wqe
, struct ib_send_wr
*wr
,
148 u8
*flit_cnt
, int *wr_cnt
, struct t3_wq
*wq
)
153 if (wr
->wr
.fast_reg
.page_list_len
> T3_MAX_FASTREG_DEPTH
)
156 wqe
->fastreg
.stag
= cpu_to_be32(wr
->wr
.fast_reg
.rkey
);
157 wqe
->fastreg
.len
= cpu_to_be32(wr
->wr
.fast_reg
.length
);
158 wqe
->fastreg
.va_base_hi
= cpu_to_be32(wr
->wr
.fast_reg
.iova_start
>> 32);
159 wqe
->fastreg
.va_base_lo_fbo
=
160 cpu_to_be32(wr
->wr
.fast_reg
.iova_start
& 0xffffffff);
161 wqe
->fastreg
.page_type_perms
= cpu_to_be32(
162 V_FR_PAGE_COUNT(wr
->wr
.fast_reg
.page_list_len
) |
163 V_FR_PAGE_SIZE(wr
->wr
.fast_reg
.page_shift
-12) |
164 V_FR_TYPE(TPT_VATO
) |
165 V_FR_PERMS(iwch_ib_to_tpt_access(wr
->wr
.fast_reg
.access_flags
)));
166 p
= &wqe
->fastreg
.pbl_addrs
[0];
167 for (i
= 0; i
< wr
->wr
.fast_reg
.page_list_len
; i
++, p
++) {
169 /* If we need a 2nd WR, then set it up */
170 if (i
== T3_MAX_FASTREG_FRAG
) {
172 wqe
= (union t3_wr
*)(wq
->queue
+
173 Q_PTR2IDX((wq
->wptr
+1), wq
->size_log2
));
174 build_fw_riwrh((void *)wqe
, T3_WR_FASTREG
, 0,
175 Q_GENBIT(wq
->wptr
+ 1, wq
->size_log2
),
176 0, 1 + wr
->wr
.fast_reg
.page_list_len
- T3_MAX_FASTREG_FRAG
,
179 p
= &wqe
->pbl_frag
.pbl_addrs
[0];
181 *p
= cpu_to_be64((u64
)wr
->wr
.fast_reg
.page_list
->page_list
[i
]);
183 *flit_cnt
= 5 + wr
->wr
.fast_reg
.page_list_len
;
189 static int build_inv_stag(union t3_wr
*wqe
, struct ib_send_wr
*wr
,
192 wqe
->local_inv
.stag
= cpu_to_be32(wr
->ex
.invalidate_rkey
);
193 wqe
->local_inv
.reserved
= 0;
194 *flit_cnt
= sizeof(struct t3_local_inv_wr
) >> 3;
198 static int iwch_sgl2pbl_map(struct iwch_dev
*rhp
, struct ib_sge
*sg_list
,
199 u32 num_sgle
, u32
* pbl_addr
, u8
* page_size
)
204 for (i
= 0; i
< num_sgle
; i
++) {
206 mhp
= get_mhp(rhp
, (sg_list
[i
].lkey
) >> 8);
208 PDBG("%s %d\n", __func__
, __LINE__
);
211 if (!mhp
->attr
.state
) {
212 PDBG("%s %d\n", __func__
, __LINE__
);
215 if (mhp
->attr
.zbva
) {
216 PDBG("%s %d\n", __func__
, __LINE__
);
220 if (sg_list
[i
].addr
< mhp
->attr
.va_fbo
) {
221 PDBG("%s %d\n", __func__
, __LINE__
);
224 if (sg_list
[i
].addr
+ ((u64
) sg_list
[i
].length
) <
226 PDBG("%s %d\n", __func__
, __LINE__
);
229 if (sg_list
[i
].addr
+ ((u64
) sg_list
[i
].length
) >
230 mhp
->attr
.va_fbo
+ ((u64
) mhp
->attr
.len
)) {
231 PDBG("%s %d\n", __func__
, __LINE__
);
234 offset
= sg_list
[i
].addr
- mhp
->attr
.va_fbo
;
235 offset
+= mhp
->attr
.va_fbo
&
236 ((1UL << (12 + mhp
->attr
.page_size
)) - 1);
237 pbl_addr
[i
] = ((mhp
->attr
.pbl_addr
-
238 rhp
->rdev
.rnic_info
.pbl_base
) >> 3) +
239 (offset
>> (12 + mhp
->attr
.page_size
));
240 page_size
[i
] = mhp
->attr
.page_size
;
245 static int build_rdma_recv(struct iwch_qp
*qhp
, union t3_wr
*wqe
,
246 struct ib_recv_wr
*wr
)
249 u32 pbl_addr
[T3_MAX_SGE
];
250 u8 page_size
[T3_MAX_SGE
];
252 err
= iwch_sgl2pbl_map(qhp
->rhp
, wr
->sg_list
, wr
->num_sge
, pbl_addr
,
256 wqe
->recv
.pagesz
[0] = page_size
[0];
257 wqe
->recv
.pagesz
[1] = page_size
[1];
258 wqe
->recv
.pagesz
[2] = page_size
[2];
259 wqe
->recv
.pagesz
[3] = page_size
[3];
260 wqe
->recv
.num_sgle
= cpu_to_be32(wr
->num_sge
);
261 for (i
= 0; i
< wr
->num_sge
; i
++) {
262 wqe
->recv
.sgl
[i
].stag
= cpu_to_be32(wr
->sg_list
[i
].lkey
);
263 wqe
->recv
.sgl
[i
].len
= cpu_to_be32(wr
->sg_list
[i
].length
);
265 /* to in the WQE == the offset into the page */
266 wqe
->recv
.sgl
[i
].to
= cpu_to_be64(((u32
)wr
->sg_list
[i
].addr
) &
267 ((1UL << (12 + page_size
[i
])) - 1));
269 /* pbl_addr is the adapters address in the PBL */
270 wqe
->recv
.pbl_addr
[i
] = cpu_to_be32(pbl_addr
[i
]);
272 for (; i
< T3_MAX_SGE
; i
++) {
273 wqe
->recv
.sgl
[i
].stag
= 0;
274 wqe
->recv
.sgl
[i
].len
= 0;
275 wqe
->recv
.sgl
[i
].to
= 0;
276 wqe
->recv
.pbl_addr
[i
] = 0;
278 qhp
->wq
.rq
[Q_PTR2IDX(qhp
->wq
.rq_wptr
,
279 qhp
->wq
.rq_size_log2
)].wr_id
= wr
->wr_id
;
280 qhp
->wq
.rq
[Q_PTR2IDX(qhp
->wq
.rq_wptr
,
281 qhp
->wq
.rq_size_log2
)].pbl_addr
= 0;
285 static int build_zero_stag_recv(struct iwch_qp
*qhp
, union t3_wr
*wqe
,
286 struct ib_recv_wr
*wr
)
294 * The T3 HW requires the PBL in the HW recv descriptor to reference
295 * a PBL entry. So we allocate the max needed PBL memory here and pass
296 * it to the uP in the recv WR. The uP will build the PBL and setup
297 * the HW recv descriptor.
299 pbl_addr
= cxio_hal_pblpool_alloc(&qhp
->rhp
->rdev
, T3_STAG0_PBL_SIZE
);
304 * Compute the 8B aligned offset.
306 pbl_offset
= (pbl_addr
- qhp
->rhp
->rdev
.rnic_info
.pbl_base
) >> 3;
308 wqe
->recv
.num_sgle
= cpu_to_be32(wr
->num_sge
);
310 for (i
= 0; i
< wr
->num_sge
; i
++) {
313 * Use a 128MB page size. This and an imposed 128MB
314 * sge length limit allows us to require only a 2-entry HW
315 * PBL for each SGE. This restriction is acceptable since
316 * since it is not possible to allocate 128MB of contiguous
317 * DMA coherent memory!
319 if (wr
->sg_list
[i
].length
> T3_STAG0_MAX_PBE_LEN
)
321 wqe
->recv
.pagesz
[i
] = T3_STAG0_PAGE_SHIFT
;
324 * T3 restricts a recv to all zero-stag or all non-zero-stag.
326 if (wr
->sg_list
[i
].lkey
!= 0)
328 wqe
->recv
.sgl
[i
].stag
= 0;
329 wqe
->recv
.sgl
[i
].len
= cpu_to_be32(wr
->sg_list
[i
].length
);
330 wqe
->recv
.sgl
[i
].to
= cpu_to_be64(wr
->sg_list
[i
].addr
);
331 wqe
->recv
.pbl_addr
[i
] = cpu_to_be32(pbl_offset
);
334 for (; i
< T3_MAX_SGE
; i
++) {
335 wqe
->recv
.pagesz
[i
] = 0;
336 wqe
->recv
.sgl
[i
].stag
= 0;
337 wqe
->recv
.sgl
[i
].len
= 0;
338 wqe
->recv
.sgl
[i
].to
= 0;
339 wqe
->recv
.pbl_addr
[i
] = 0;
341 qhp
->wq
.rq
[Q_PTR2IDX(qhp
->wq
.rq_wptr
,
342 qhp
->wq
.rq_size_log2
)].wr_id
= wr
->wr_id
;
343 qhp
->wq
.rq
[Q_PTR2IDX(qhp
->wq
.rq_wptr
,
344 qhp
->wq
.rq_size_log2
)].pbl_addr
= pbl_addr
;
348 int iwch_post_send(struct ib_qp
*ibqp
, struct ib_send_wr
*wr
,
349 struct ib_send_wr
**bad_wr
)
352 u8
uninitialized_var(t3_wr_flit_cnt
);
353 enum t3_wr_opcode t3_wr_opcode
= 0;
354 enum t3_wr_flags t3_wr_flags
;
363 qhp
= to_iwch_qp(ibqp
);
364 spin_lock_irqsave(&qhp
->lock
, flag
);
365 if (qhp
->attr
.state
> IWCH_QP_STATE_RTS
) {
366 spin_unlock_irqrestore(&qhp
->lock
, flag
);
369 num_wrs
= Q_FREECNT(qhp
->wq
.sq_rptr
, qhp
->wq
.sq_wptr
,
370 qhp
->wq
.sq_size_log2
);
372 spin_unlock_irqrestore(&qhp
->lock
, flag
);
381 idx
= Q_PTR2IDX(qhp
->wq
.wptr
, qhp
->wq
.size_log2
);
382 wqe
= (union t3_wr
*) (qhp
->wq
.queue
+ idx
);
384 if (wr
->send_flags
& IB_SEND_SOLICITED
)
385 t3_wr_flags
|= T3_SOLICITED_EVENT_FLAG
;
386 if (wr
->send_flags
& IB_SEND_SIGNALED
)
387 t3_wr_flags
|= T3_COMPLETION_FLAG
;
389 Q_PTR2IDX(qhp
->wq
.sq_wptr
, qhp
->wq
.sq_size_log2
);
390 switch (wr
->opcode
) {
392 case IB_WR_SEND_WITH_INV
:
393 if (wr
->send_flags
& IB_SEND_FENCE
)
394 t3_wr_flags
|= T3_READ_FENCE_FLAG
;
395 t3_wr_opcode
= T3_WR_SEND
;
396 err
= build_rdma_send(wqe
, wr
, &t3_wr_flit_cnt
);
398 case IB_WR_RDMA_WRITE
:
399 case IB_WR_RDMA_WRITE_WITH_IMM
:
400 t3_wr_opcode
= T3_WR_WRITE
;
401 err
= build_rdma_write(wqe
, wr
, &t3_wr_flit_cnt
);
403 case IB_WR_RDMA_READ
:
404 case IB_WR_RDMA_READ_WITH_INV
:
405 t3_wr_opcode
= T3_WR_READ
;
406 t3_wr_flags
= 0; /* T3 reads are always signaled */
407 err
= build_rdma_read(wqe
, wr
, &t3_wr_flit_cnt
);
410 sqp
->read_len
= wqe
->read
.local_len
;
411 if (!qhp
->wq
.oldest_read
)
412 qhp
->wq
.oldest_read
= sqp
;
414 case IB_WR_FAST_REG_MR
:
415 t3_wr_opcode
= T3_WR_FASTREG
;
416 err
= build_fastreg(wqe
, wr
, &t3_wr_flit_cnt
,
419 case IB_WR_LOCAL_INV
:
420 if (wr
->send_flags
& IB_SEND_FENCE
)
421 t3_wr_flags
|= T3_LOCAL_FENCE_FLAG
;
422 t3_wr_opcode
= T3_WR_INV_STAG
;
423 err
= build_inv_stag(wqe
, wr
, &t3_wr_flit_cnt
);
426 PDBG("%s post of type=%d TBD!\n", __func__
,
434 wqe
->send
.wrid
.id0
.hi
= qhp
->wq
.sq_wptr
;
435 sqp
->wr_id
= wr
->wr_id
;
436 sqp
->opcode
= wr2opcode(t3_wr_opcode
);
437 sqp
->sq_wptr
= qhp
->wq
.sq_wptr
;
439 sqp
->signaled
= (wr
->send_flags
& IB_SEND_SIGNALED
);
441 build_fw_riwrh((void *) wqe
, t3_wr_opcode
, t3_wr_flags
,
442 Q_GENBIT(qhp
->wq
.wptr
, qhp
->wq
.size_log2
),
444 (wr_cnt
== 1) ? T3_SOPEOP
: T3_SOP
);
445 PDBG("%s cookie 0x%llx wq idx 0x%x swsq idx %ld opcode %d\n",
446 __func__
, (unsigned long long) wr
->wr_id
, idx
,
447 Q_PTR2IDX(qhp
->wq
.sq_wptr
, qhp
->wq
.sq_size_log2
),
451 qhp
->wq
.wptr
+= wr_cnt
;
454 spin_unlock_irqrestore(&qhp
->lock
, flag
);
455 ring_doorbell(qhp
->wq
.doorbell
, qhp
->wq
.qpid
);
459 int iwch_post_receive(struct ib_qp
*ibqp
, struct ib_recv_wr
*wr
,
460 struct ib_recv_wr
**bad_wr
)
469 qhp
= to_iwch_qp(ibqp
);
470 spin_lock_irqsave(&qhp
->lock
, flag
);
471 if (qhp
->attr
.state
> IWCH_QP_STATE_RTS
) {
472 spin_unlock_irqrestore(&qhp
->lock
, flag
);
475 num_wrs
= Q_FREECNT(qhp
->wq
.rq_rptr
, qhp
->wq
.rq_wptr
,
476 qhp
->wq
.rq_size_log2
) - 1;
478 spin_unlock_irqrestore(&qhp
->lock
, flag
);
482 if (wr
->num_sge
> T3_MAX_SGE
) {
487 idx
= Q_PTR2IDX(qhp
->wq
.wptr
, qhp
->wq
.size_log2
);
488 wqe
= (union t3_wr
*) (qhp
->wq
.queue
+ idx
);
490 if (wr
->sg_list
[0].lkey
)
491 err
= build_rdma_recv(qhp
, wqe
, wr
);
493 err
= build_zero_stag_recv(qhp
, wqe
, wr
);
500 build_fw_riwrh((void *) wqe
, T3_WR_RCV
, T3_COMPLETION_FLAG
,
501 Q_GENBIT(qhp
->wq
.wptr
, qhp
->wq
.size_log2
),
502 0, sizeof(struct t3_receive_wr
) >> 3, T3_SOPEOP
);
503 PDBG("%s cookie 0x%llx idx 0x%x rq_wptr 0x%x rw_rptr 0x%x "
504 "wqe %p \n", __func__
, (unsigned long long) wr
->wr_id
,
505 idx
, qhp
->wq
.rq_wptr
, qhp
->wq
.rq_rptr
, wqe
);
511 spin_unlock_irqrestore(&qhp
->lock
, flag
);
512 ring_doorbell(qhp
->wq
.doorbell
, qhp
->wq
.qpid
);
516 int iwch_bind_mw(struct ib_qp
*qp
,
518 struct ib_mw_bind
*mw_bind
)
520 struct iwch_dev
*rhp
;
530 enum t3_wr_flags t3_wr_flags
;
534 qhp
= to_iwch_qp(qp
);
535 mhp
= to_iwch_mw(mw
);
538 spin_lock_irqsave(&qhp
->lock
, flag
);
539 if (qhp
->attr
.state
> IWCH_QP_STATE_RTS
) {
540 spin_unlock_irqrestore(&qhp
->lock
, flag
);
543 num_wrs
= Q_FREECNT(qhp
->wq
.sq_rptr
, qhp
->wq
.sq_wptr
,
544 qhp
->wq
.sq_size_log2
);
545 if ((num_wrs
) <= 0) {
546 spin_unlock_irqrestore(&qhp
->lock
, flag
);
549 idx
= Q_PTR2IDX(qhp
->wq
.wptr
, qhp
->wq
.size_log2
);
550 PDBG("%s: idx 0x%0x, mw 0x%p, mw_bind 0x%p\n", __func__
, idx
,
552 wqe
= (union t3_wr
*) (qhp
->wq
.queue
+ idx
);
555 if (mw_bind
->send_flags
& IB_SEND_SIGNALED
)
556 t3_wr_flags
= T3_COMPLETION_FLAG
;
558 sgl
.addr
= mw_bind
->addr
;
559 sgl
.lkey
= mw_bind
->mr
->lkey
;
560 sgl
.length
= mw_bind
->length
;
561 wqe
->bind
.reserved
= 0;
562 wqe
->bind
.type
= TPT_VATO
;
564 /* TBD: check perms */
565 wqe
->bind
.perms
= iwch_ib_to_tpt_bind_access(mw_bind
->mw_access_flags
);
566 wqe
->bind
.mr_stag
= cpu_to_be32(mw_bind
->mr
->lkey
);
567 wqe
->bind
.mw_stag
= cpu_to_be32(mw
->rkey
);
568 wqe
->bind
.mw_len
= cpu_to_be32(mw_bind
->length
);
569 wqe
->bind
.mw_va
= cpu_to_be64(mw_bind
->addr
);
570 err
= iwch_sgl2pbl_map(rhp
, &sgl
, 1, &pbl_addr
, &page_size
);
572 spin_unlock_irqrestore(&qhp
->lock
, flag
);
575 wqe
->send
.wrid
.id0
.hi
= qhp
->wq
.sq_wptr
;
576 sqp
= qhp
->wq
.sq
+ Q_PTR2IDX(qhp
->wq
.sq_wptr
, qhp
->wq
.sq_size_log2
);
577 sqp
->wr_id
= mw_bind
->wr_id
;
578 sqp
->opcode
= T3_BIND_MW
;
579 sqp
->sq_wptr
= qhp
->wq
.sq_wptr
;
581 sqp
->signaled
= (mw_bind
->send_flags
& IB_SEND_SIGNALED
);
582 wqe
->bind
.mr_pbl_addr
= cpu_to_be32(pbl_addr
);
583 wqe
->bind
.mr_pagesz
= page_size
;
584 build_fw_riwrh((void *)wqe
, T3_WR_BIND
, t3_wr_flags
,
585 Q_GENBIT(qhp
->wq
.wptr
, qhp
->wq
.size_log2
), 0,
586 sizeof(struct t3_bind_mw_wr
) >> 3, T3_SOPEOP
);
589 spin_unlock_irqrestore(&qhp
->lock
, flag
);
591 ring_doorbell(qhp
->wq
.doorbell
, qhp
->wq
.qpid
);
596 static inline void build_term_codes(struct respQ_msg_t
*rsp_msg
,
597 u8
*layer_type
, u8
*ecode
)
599 int status
= TPT_ERR_INTERNAL_ERR
;
606 status
= CQE_STATUS(rsp_msg
->cqe
);
607 opcode
= CQE_OPCODE(rsp_msg
->cqe
);
608 rqtype
= RQ_TYPE(rsp_msg
->cqe
);
609 send_inv
= (opcode
== T3_SEND_WITH_INV
) ||
610 (opcode
== T3_SEND_WITH_SE_INV
);
611 tagged
= (opcode
== T3_RDMA_WRITE
) ||
612 (rqtype
&& (opcode
== T3_READ_RESP
));
618 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_OP
;
619 *ecode
= RDMAP_CANT_INV_STAG
;
621 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
622 *ecode
= RDMAP_INV_STAG
;
626 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
627 if ((opcode
== T3_SEND_WITH_INV
) ||
628 (opcode
== T3_SEND_WITH_SE_INV
))
629 *ecode
= RDMAP_CANT_INV_STAG
;
631 *ecode
= RDMAP_STAG_NOT_ASSOC
;
634 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
635 *ecode
= RDMAP_STAG_NOT_ASSOC
;
638 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
639 *ecode
= RDMAP_ACC_VIOL
;
642 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
643 *ecode
= RDMAP_TO_WRAP
;
647 *layer_type
= LAYER_DDP
|DDP_TAGGED_ERR
;
648 *ecode
= DDPT_BASE_BOUNDS
;
650 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
651 *ecode
= RDMAP_BASE_BOUNDS
;
654 case TPT_ERR_INVALIDATE_SHARED_MR
:
655 case TPT_ERR_INVALIDATE_MR_WITH_MW_BOUND
:
656 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_OP
;
657 *ecode
= RDMAP_CANT_INV_STAG
;
660 case TPT_ERR_ECC_PSTAG
:
661 case TPT_ERR_INTERNAL_ERR
:
662 *layer_type
= LAYER_RDMAP
|RDMAP_LOCAL_CATA
;
665 case TPT_ERR_OUT_OF_RQE
:
666 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
667 *ecode
= DDPU_INV_MSN_NOBUF
;
669 case TPT_ERR_PBL_ADDR_BOUND
:
670 *layer_type
= LAYER_DDP
|DDP_TAGGED_ERR
;
671 *ecode
= DDPT_BASE_BOUNDS
;
674 *layer_type
= LAYER_MPA
|DDP_LLP
;
675 *ecode
= MPA_CRC_ERR
;
678 *layer_type
= LAYER_MPA
|DDP_LLP
;
679 *ecode
= MPA_MARKER_ERR
;
681 case TPT_ERR_PDU_LEN_ERR
:
682 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
683 *ecode
= DDPU_MSG_TOOBIG
;
685 case TPT_ERR_DDP_VERSION
:
687 *layer_type
= LAYER_DDP
|DDP_TAGGED_ERR
;
688 *ecode
= DDPT_INV_VERS
;
690 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
691 *ecode
= DDPU_INV_VERS
;
694 case TPT_ERR_RDMA_VERSION
:
695 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_OP
;
696 *ecode
= RDMAP_INV_VERS
;
699 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_OP
;
700 *ecode
= RDMAP_INV_OPCODE
;
702 case TPT_ERR_DDP_QUEUE_NUM
:
703 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
704 *ecode
= DDPU_INV_QN
;
707 case TPT_ERR_MSN_GAP
:
708 case TPT_ERR_MSN_RANGE
:
709 case TPT_ERR_IRD_OVERFLOW
:
710 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
711 *ecode
= DDPU_INV_MSN_RANGE
;
714 *layer_type
= LAYER_DDP
|DDP_LOCAL_CATA
;
718 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
719 *ecode
= DDPU_INV_MO
;
722 *layer_type
= LAYER_RDMAP
|DDP_LOCAL_CATA
;
728 int iwch_post_zb_read(struct iwch_qp
*qhp
)
732 u8 flit_cnt
= sizeof(struct t3_rdma_read_wr
) >> 3;
734 PDBG("%s enter\n", __func__
);
735 skb
= alloc_skb(40, GFP_KERNEL
);
737 printk(KERN_ERR
"%s cannot send zb_read!!\n", __func__
);
740 wqe
= (union t3_wr
*)skb_put(skb
, sizeof(struct t3_rdma_read_wr
));
741 memset(wqe
, 0, sizeof(struct t3_rdma_read_wr
));
742 wqe
->read
.rdmaop
= T3_READ_REQ
;
743 wqe
->read
.reserved
[0] = 0;
744 wqe
->read
.reserved
[1] = 0;
745 wqe
->read
.rem_stag
= cpu_to_be32(1);
746 wqe
->read
.rem_to
= cpu_to_be64(1);
747 wqe
->read
.local_stag
= cpu_to_be32(1);
748 wqe
->read
.local_len
= cpu_to_be32(0);
749 wqe
->read
.local_to
= cpu_to_be64(1);
750 wqe
->send
.wrh
.op_seop_flags
= cpu_to_be32(V_FW_RIWR_OP(T3_WR_READ
));
751 wqe
->send
.wrh
.gen_tid_len
= cpu_to_be32(V_FW_RIWR_TID(qhp
->ep
->hwtid
)|
752 V_FW_RIWR_LEN(flit_cnt
));
753 skb
->priority
= CPL_PRIORITY_DATA
;
754 return iwch_cxgb3_ofld_send(qhp
->rhp
->rdev
.t3cdev_p
, skb
);
758 * This posts a TERMINATE with layer=RDMA, type=catastrophic.
760 int iwch_post_terminate(struct iwch_qp
*qhp
, struct respQ_msg_t
*rsp_msg
)
763 struct terminate_message
*term
;
766 PDBG("%s %d\n", __func__
, __LINE__
);
767 skb
= alloc_skb(40, GFP_ATOMIC
);
769 printk(KERN_ERR
"%s cannot send TERMINATE!\n", __func__
);
772 wqe
= (union t3_wr
*)skb_put(skb
, 40);
774 wqe
->send
.rdmaop
= T3_TERMINATE
;
776 /* immediate data length */
777 wqe
->send
.plen
= htonl(4);
779 /* immediate data starts here. */
780 term
= (struct terminate_message
*)wqe
->send
.sgl
;
781 build_term_codes(rsp_msg
, &term
->layer_etype
, &term
->ecode
);
782 wqe
->send
.wrh
.op_seop_flags
= cpu_to_be32(V_FW_RIWR_OP(T3_WR_SEND
) |
783 V_FW_RIWR_FLAGS(T3_COMPLETION_FLAG
| T3_NOTIFY_FLAG
));
784 wqe
->send
.wrh
.gen_tid_len
= cpu_to_be32(V_FW_RIWR_TID(qhp
->ep
->hwtid
));
785 skb
->priority
= CPL_PRIORITY_DATA
;
786 return iwch_cxgb3_ofld_send(qhp
->rhp
->rdev
.t3cdev_p
, skb
);
790 * Assumes qhp lock is held.
792 static void __flush_qp(struct iwch_qp
*qhp
, unsigned long *flag
)
794 struct iwch_cq
*rchp
, *schp
;
798 rchp
= get_chp(qhp
->rhp
, qhp
->attr
.rcq
);
799 schp
= get_chp(qhp
->rhp
, qhp
->attr
.scq
);
801 PDBG("%s qhp %p rchp %p schp %p\n", __func__
, qhp
, rchp
, schp
);
802 /* take a ref on the qhp since we must release the lock */
803 atomic_inc(&qhp
->refcnt
);
804 spin_unlock_irqrestore(&qhp
->lock
, *flag
);
806 /* locking heirarchy: cq lock first, then qp lock. */
807 spin_lock_irqsave(&rchp
->lock
, *flag
);
808 spin_lock(&qhp
->lock
);
809 cxio_flush_hw_cq(&rchp
->cq
);
810 cxio_count_rcqes(&rchp
->cq
, &qhp
->wq
, &count
);
811 flushed
= cxio_flush_rq(&qhp
->wq
, &rchp
->cq
, count
);
812 spin_unlock(&qhp
->lock
);
813 spin_unlock_irqrestore(&rchp
->lock
, *flag
);
815 (*rchp
->ibcq
.comp_handler
)(&rchp
->ibcq
, rchp
->ibcq
.cq_context
);
817 /* locking heirarchy: cq lock first, then qp lock. */
818 spin_lock_irqsave(&schp
->lock
, *flag
);
819 spin_lock(&qhp
->lock
);
820 cxio_flush_hw_cq(&schp
->cq
);
821 cxio_count_scqes(&schp
->cq
, &qhp
->wq
, &count
);
822 flushed
= cxio_flush_sq(&qhp
->wq
, &schp
->cq
, count
);
823 spin_unlock(&qhp
->lock
);
824 spin_unlock_irqrestore(&schp
->lock
, *flag
);
826 (*schp
->ibcq
.comp_handler
)(&schp
->ibcq
, schp
->ibcq
.cq_context
);
829 if (atomic_dec_and_test(&qhp
->refcnt
))
832 spin_lock_irqsave(&qhp
->lock
, *flag
);
835 static void flush_qp(struct iwch_qp
*qhp
, unsigned long *flag
)
837 if (qhp
->ibqp
.uobject
)
838 cxio_set_wq_in_error(&qhp
->wq
);
840 __flush_qp(qhp
, flag
);
845 * Return count of RECV WRs posted
847 u16
iwch_rqes_posted(struct iwch_qp
*qhp
)
849 union t3_wr
*wqe
= qhp
->wq
.queue
;
851 while ((count
+1) != 0 && fw_riwrh_opcode((struct fw_riwrh
*)wqe
) == T3_WR_RCV
) {
855 PDBG("%s qhp %p count %u\n", __func__
, qhp
, count
);
859 static int rdma_init(struct iwch_dev
*rhp
, struct iwch_qp
*qhp
,
860 enum iwch_qp_attr_mask mask
,
861 struct iwch_qp_attributes
*attrs
)
863 struct t3_rdma_init_attr init_attr
;
866 init_attr
.tid
= qhp
->ep
->hwtid
;
867 init_attr
.qpid
= qhp
->wq
.qpid
;
868 init_attr
.pdid
= qhp
->attr
.pd
;
869 init_attr
.scqid
= qhp
->attr
.scq
;
870 init_attr
.rcqid
= qhp
->attr
.rcq
;
871 init_attr
.rq_addr
= qhp
->wq
.rq_addr
;
872 init_attr
.rq_size
= 1 << qhp
->wq
.rq_size_log2
;
873 init_attr
.mpaattrs
= uP_RI_MPA_IETF_ENABLE
|
874 qhp
->attr
.mpa_attr
.recv_marker_enabled
|
875 (qhp
->attr
.mpa_attr
.xmit_marker_enabled
<< 1) |
876 (qhp
->attr
.mpa_attr
.crc_enabled
<< 2);
878 init_attr
.qpcaps
= uP_RI_QP_RDMA_READ_ENABLE
|
879 uP_RI_QP_RDMA_WRITE_ENABLE
|
880 uP_RI_QP_BIND_ENABLE
;
881 if (!qhp
->ibqp
.uobject
)
882 init_attr
.qpcaps
|= uP_RI_QP_STAG0_ENABLE
|
883 uP_RI_QP_FAST_REGISTER_ENABLE
;
885 init_attr
.tcp_emss
= qhp
->ep
->emss
;
886 init_attr
.ord
= qhp
->attr
.max_ord
;
887 init_attr
.ird
= qhp
->attr
.max_ird
;
888 init_attr
.qp_dma_addr
= qhp
->wq
.dma_addr
;
889 init_attr
.qp_dma_size
= (1UL << qhp
->wq
.size_log2
);
890 init_attr
.rqe_count
= iwch_rqes_posted(qhp
);
891 init_attr
.flags
= qhp
->attr
.mpa_attr
.initiator
? MPA_INITIATOR
: 0;
893 init_attr
.rtr_type
= RTR_READ
;
894 if (init_attr
.ord
== 0 && qhp
->attr
.mpa_attr
.initiator
)
896 if (init_attr
.ird
== 0 && !qhp
->attr
.mpa_attr
.initiator
)
899 init_attr
.rtr_type
= 0;
900 init_attr
.irs
= qhp
->ep
->rcv_seq
;
901 PDBG("%s init_attr.rq_addr 0x%x init_attr.rq_size = %d "
902 "flags 0x%x qpcaps 0x%x\n", __func__
,
903 init_attr
.rq_addr
, init_attr
.rq_size
,
904 init_attr
.flags
, init_attr
.qpcaps
);
905 ret
= cxio_rdma_init(&rhp
->rdev
, &init_attr
);
906 PDBG("%s ret %d\n", __func__
, ret
);
910 int iwch_modify_qp(struct iwch_dev
*rhp
, struct iwch_qp
*qhp
,
911 enum iwch_qp_attr_mask mask
,
912 struct iwch_qp_attributes
*attrs
,
916 struct iwch_qp_attributes newattr
= qhp
->attr
;
922 struct iwch_ep
*ep
= NULL
;
924 PDBG("%s qhp %p qpid 0x%x ep %p state %d -> %d\n", __func__
,
925 qhp
, qhp
->wq
.qpid
, qhp
->ep
, qhp
->attr
.state
,
926 (mask
& IWCH_QP_ATTR_NEXT_STATE
) ? attrs
->next_state
: -1);
928 spin_lock_irqsave(&qhp
->lock
, flag
);
930 /* Process attr changes if in IDLE */
931 if (mask
& IWCH_QP_ATTR_VALID_MODIFY
) {
932 if (qhp
->attr
.state
!= IWCH_QP_STATE_IDLE
) {
936 if (mask
& IWCH_QP_ATTR_ENABLE_RDMA_READ
)
937 newattr
.enable_rdma_read
= attrs
->enable_rdma_read
;
938 if (mask
& IWCH_QP_ATTR_ENABLE_RDMA_WRITE
)
939 newattr
.enable_rdma_write
= attrs
->enable_rdma_write
;
940 if (mask
& IWCH_QP_ATTR_ENABLE_RDMA_BIND
)
941 newattr
.enable_bind
= attrs
->enable_bind
;
942 if (mask
& IWCH_QP_ATTR_MAX_ORD
) {
944 rhp
->attr
.max_rdma_read_qp_depth
) {
948 newattr
.max_ord
= attrs
->max_ord
;
950 if (mask
& IWCH_QP_ATTR_MAX_IRD
) {
952 rhp
->attr
.max_rdma_reads_per_qp
) {
956 newattr
.max_ird
= attrs
->max_ird
;
961 if (!(mask
& IWCH_QP_ATTR_NEXT_STATE
))
963 if (qhp
->attr
.state
== attrs
->next_state
)
966 switch (qhp
->attr
.state
) {
967 case IWCH_QP_STATE_IDLE
:
968 switch (attrs
->next_state
) {
969 case IWCH_QP_STATE_RTS
:
970 if (!(mask
& IWCH_QP_ATTR_LLP_STREAM_HANDLE
)) {
974 if (!(mask
& IWCH_QP_ATTR_MPA_ATTR
)) {
978 qhp
->attr
.mpa_attr
= attrs
->mpa_attr
;
979 qhp
->attr
.llp_stream_handle
= attrs
->llp_stream_handle
;
980 qhp
->ep
= qhp
->attr
.llp_stream_handle
;
981 qhp
->attr
.state
= IWCH_QP_STATE_RTS
;
984 * Ref the endpoint here and deref when we
985 * disassociate the endpoint from the QP. This
986 * happens in CLOSING->IDLE transition or *->ERROR
989 get_ep(&qhp
->ep
->com
);
990 spin_unlock_irqrestore(&qhp
->lock
, flag
);
991 ret
= rdma_init(rhp
, qhp
, mask
, attrs
);
992 spin_lock_irqsave(&qhp
->lock
, flag
);
996 case IWCH_QP_STATE_ERROR
:
997 qhp
->attr
.state
= IWCH_QP_STATE_ERROR
;
998 flush_qp(qhp
, &flag
);
1005 case IWCH_QP_STATE_RTS
:
1006 switch (attrs
->next_state
) {
1007 case IWCH_QP_STATE_CLOSING
:
1008 BUG_ON(atomic_read(&qhp
->ep
->com
.kref
.refcount
) < 2);
1009 qhp
->attr
.state
= IWCH_QP_STATE_CLOSING
;
1017 case IWCH_QP_STATE_TERMINATE
:
1018 qhp
->attr
.state
= IWCH_QP_STATE_TERMINATE
;
1019 if (qhp
->ibqp
.uobject
)
1020 cxio_set_wq_in_error(&qhp
->wq
);
1024 case IWCH_QP_STATE_ERROR
:
1025 qhp
->attr
.state
= IWCH_QP_STATE_ERROR
;
1039 case IWCH_QP_STATE_CLOSING
:
1044 switch (attrs
->next_state
) {
1045 case IWCH_QP_STATE_IDLE
:
1046 flush_qp(qhp
, &flag
);
1047 qhp
->attr
.state
= IWCH_QP_STATE_IDLE
;
1048 qhp
->attr
.llp_stream_handle
= NULL
;
1049 put_ep(&qhp
->ep
->com
);
1051 wake_up(&qhp
->wait
);
1053 case IWCH_QP_STATE_ERROR
:
1060 case IWCH_QP_STATE_ERROR
:
1061 if (attrs
->next_state
!= IWCH_QP_STATE_IDLE
) {
1066 if (!Q_EMPTY(qhp
->wq
.sq_rptr
, qhp
->wq
.sq_wptr
) ||
1067 !Q_EMPTY(qhp
->wq
.rq_rptr
, qhp
->wq
.rq_wptr
)) {
1071 qhp
->attr
.state
= IWCH_QP_STATE_IDLE
;
1073 case IWCH_QP_STATE_TERMINATE
:
1081 printk(KERN_ERR
"%s in a bad state %d\n",
1082 __func__
, qhp
->attr
.state
);
1089 PDBG("%s disassociating ep %p qpid 0x%x\n", __func__
, qhp
->ep
,
1092 /* disassociate the LLP connection */
1093 qhp
->attr
.llp_stream_handle
= NULL
;
1096 qhp
->attr
.state
= IWCH_QP_STATE_ERROR
;
1098 wake_up(&qhp
->wait
);
1100 flush_qp(qhp
, &flag
);
1102 spin_unlock_irqrestore(&qhp
->lock
, flag
);
1105 iwch_post_terminate(qhp
, NULL
);
1108 * If disconnect is 1, then we need to initiate a disconnect
1109 * on the EP. This can be a normal close (RTS->CLOSING) or
1110 * an abnormal close (RTS/CLOSING->ERROR).
1113 iwch_ep_disconnect(ep
, abort
, GFP_KERNEL
);
1118 * If free is 1, then we've disassociated the EP from the QP
1119 * and we need to dereference the EP.
1124 PDBG("%s exit state %d\n", __func__
, qhp
->attr
.state
);
1128 static int quiesce_qp(struct iwch_qp
*qhp
)
1130 spin_lock_irq(&qhp
->lock
);
1131 iwch_quiesce_tid(qhp
->ep
);
1132 qhp
->flags
|= QP_QUIESCED
;
1133 spin_unlock_irq(&qhp
->lock
);
1137 static int resume_qp(struct iwch_qp
*qhp
)
1139 spin_lock_irq(&qhp
->lock
);
1140 iwch_resume_tid(qhp
->ep
);
1141 qhp
->flags
&= ~QP_QUIESCED
;
1142 spin_unlock_irq(&qhp
->lock
);
1146 int iwch_quiesce_qps(struct iwch_cq
*chp
)
1149 struct iwch_qp
*qhp
;
1151 for (i
=0; i
< T3_MAX_NUM_QP
; i
++) {
1152 qhp
= get_qhp(chp
->rhp
, i
);
1155 if ((qhp
->attr
.rcq
== chp
->cq
.cqid
) && !qp_quiesced(qhp
)) {
1159 if ((qhp
->attr
.scq
== chp
->cq
.cqid
) && !qp_quiesced(qhp
))
1165 int iwch_resume_qps(struct iwch_cq
*chp
)
1168 struct iwch_qp
*qhp
;
1170 for (i
=0; i
< T3_MAX_NUM_QP
; i
++) {
1171 qhp
= get_qhp(chp
->rhp
, i
);
1174 if ((qhp
->attr
.rcq
== chp
->cq
.cqid
) && qp_quiesced(qhp
)) {
1178 if ((qhp
->attr
.scq
== chp
->cq
.cqid
) && qp_quiesced(qhp
))