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
33 #include <linux/module.h>
37 static int db_delay_usecs
= 1;
38 module_param(db_delay_usecs
, int, 0644);
39 MODULE_PARM_DESC(db_delay_usecs
, "Usecs to delay awaiting db fifo to drain");
41 static int ocqp_support
= 1;
42 module_param(ocqp_support
, int, 0644);
43 MODULE_PARM_DESC(ocqp_support
, "Support on-chip SQs (default=1)");
45 int db_fc_threshold
= 1000;
46 module_param(db_fc_threshold
, int, 0644);
47 MODULE_PARM_DESC(db_fc_threshold
,
48 "QP count/threshold that triggers"
49 " automatic db flow control mode (default = 1000)");
51 int db_coalescing_threshold
;
52 module_param(db_coalescing_threshold
, int, 0644);
53 MODULE_PARM_DESC(db_coalescing_threshold
,
54 "QP count/threshold that triggers"
55 " disabling db coalescing (default = 0)");
57 static int max_fr_immd
= T4_MAX_FR_IMMD
;
58 module_param(max_fr_immd
, int, 0644);
59 MODULE_PARM_DESC(max_fr_immd
, "fastreg threshold for using DSGL instead of immedate");
61 static int alloc_ird(struct c4iw_dev
*dev
, u32 ird
)
65 spin_lock_irq(&dev
->lock
);
66 if (ird
<= dev
->avail_ird
)
67 dev
->avail_ird
-= ird
;
70 spin_unlock_irq(&dev
->lock
);
73 dev_warn(&dev
->rdev
.lldi
.pdev
->dev
,
74 "device IRD resources exhausted\n");
79 static void free_ird(struct c4iw_dev
*dev
, int ird
)
81 spin_lock_irq(&dev
->lock
);
82 dev
->avail_ird
+= ird
;
83 spin_unlock_irq(&dev
->lock
);
86 static void set_state(struct c4iw_qp
*qhp
, enum c4iw_qp_state state
)
89 spin_lock_irqsave(&qhp
->lock
, flag
);
90 qhp
->attr
.state
= state
;
91 spin_unlock_irqrestore(&qhp
->lock
, flag
);
94 static void dealloc_oc_sq(struct c4iw_rdev
*rdev
, struct t4_sq
*sq
)
96 c4iw_ocqp_pool_free(rdev
, sq
->dma_addr
, sq
->memsize
);
99 static void dealloc_host_sq(struct c4iw_rdev
*rdev
, struct t4_sq
*sq
)
101 dma_free_coherent(&(rdev
->lldi
.pdev
->dev
), sq
->memsize
, sq
->queue
,
102 pci_unmap_addr(sq
, mapping
));
105 static void dealloc_sq(struct c4iw_rdev
*rdev
, struct t4_sq
*sq
)
107 if (t4_sq_onchip(sq
))
108 dealloc_oc_sq(rdev
, sq
);
110 dealloc_host_sq(rdev
, sq
);
113 static int alloc_oc_sq(struct c4iw_rdev
*rdev
, struct t4_sq
*sq
)
115 if (!ocqp_support
|| !ocqp_supported(&rdev
->lldi
))
117 sq
->dma_addr
= c4iw_ocqp_pool_alloc(rdev
, sq
->memsize
);
120 sq
->phys_addr
= rdev
->oc_mw_pa
+ sq
->dma_addr
-
121 rdev
->lldi
.vr
->ocq
.start
;
122 sq
->queue
= (__force
union t4_wr
*)(rdev
->oc_mw_kva
+ sq
->dma_addr
-
123 rdev
->lldi
.vr
->ocq
.start
);
124 sq
->flags
|= T4_SQ_ONCHIP
;
128 static int alloc_host_sq(struct c4iw_rdev
*rdev
, struct t4_sq
*sq
)
130 sq
->queue
= dma_alloc_coherent(&(rdev
->lldi
.pdev
->dev
), sq
->memsize
,
131 &(sq
->dma_addr
), GFP_KERNEL
);
134 sq
->phys_addr
= virt_to_phys(sq
->queue
);
135 pci_unmap_addr_set(sq
, mapping
, sq
->dma_addr
);
139 static int alloc_sq(struct c4iw_rdev
*rdev
, struct t4_sq
*sq
, int user
)
143 ret
= alloc_oc_sq(rdev
, sq
);
145 ret
= alloc_host_sq(rdev
, sq
);
149 static int destroy_qp(struct c4iw_rdev
*rdev
, struct t4_wq
*wq
,
150 struct c4iw_dev_ucontext
*uctx
)
153 * uP clears EQ contexts when the connection exits rdma mode,
154 * so no need to post a RESET WR for these EQs.
156 dma_free_coherent(&(rdev
->lldi
.pdev
->dev
),
157 wq
->rq
.memsize
, wq
->rq
.queue
,
158 dma_unmap_addr(&wq
->rq
, mapping
));
159 dealloc_sq(rdev
, &wq
->sq
);
160 c4iw_rqtpool_free(rdev
, wq
->rq
.rqt_hwaddr
, wq
->rq
.rqt_size
);
163 c4iw_put_qpid(rdev
, wq
->rq
.qid
, uctx
);
164 c4iw_put_qpid(rdev
, wq
->sq
.qid
, uctx
);
169 * Determine the BAR2 virtual address and qid. If pbar2_pa is not NULL,
170 * then this is a user mapping so compute the page-aligned physical address
173 void __iomem
*c4iw_bar2_addrs(struct c4iw_rdev
*rdev
, unsigned int qid
,
174 enum cxgb4_bar2_qtype qtype
,
175 unsigned int *pbar2_qid
, u64
*pbar2_pa
)
180 ret
= cxgb4_bar2_sge_qregs(rdev
->lldi
.ports
[0], qid
, qtype
,
182 &bar2_qoffset
, pbar2_qid
);
187 *pbar2_pa
= (rdev
->bar2_pa
+ bar2_qoffset
) & PAGE_MASK
;
188 return rdev
->bar2_kva
+ bar2_qoffset
;
191 static int create_qp(struct c4iw_rdev
*rdev
, struct t4_wq
*wq
,
192 struct t4_cq
*rcq
, struct t4_cq
*scq
,
193 struct c4iw_dev_ucontext
*uctx
)
195 int user
= (uctx
!= &rdev
->uctx
);
196 struct fw_ri_res_wr
*res_wr
;
197 struct fw_ri_res
*res
;
199 struct c4iw_wr_wait wr_wait
;
204 wq
->sq
.qid
= c4iw_get_qpid(rdev
, uctx
);
208 wq
->rq
.qid
= c4iw_get_qpid(rdev
, uctx
);
215 wq
->sq
.sw_sq
= kzalloc(wq
->sq
.size
* sizeof *wq
->sq
.sw_sq
,
222 wq
->rq
.sw_rq
= kzalloc(wq
->rq
.size
* sizeof *wq
->rq
.sw_rq
,
231 * RQT must be a power of 2 and at least 16 deep.
233 wq
->rq
.rqt_size
= roundup_pow_of_two(max_t(u16
, wq
->rq
.size
, 16));
234 wq
->rq
.rqt_hwaddr
= c4iw_rqtpool_alloc(rdev
, wq
->rq
.rqt_size
);
235 if (!wq
->rq
.rqt_hwaddr
) {
240 ret
= alloc_sq(rdev
, &wq
->sq
, user
);
243 memset(wq
->sq
.queue
, 0, wq
->sq
.memsize
);
244 dma_unmap_addr_set(&wq
->sq
, mapping
, wq
->sq
.dma_addr
);
246 wq
->rq
.queue
= dma_alloc_coherent(&(rdev
->lldi
.pdev
->dev
),
247 wq
->rq
.memsize
, &(wq
->rq
.dma_addr
),
253 PDBG("%s sq base va 0x%p pa 0x%llx rq base va 0x%p pa 0x%llx\n",
254 __func__
, wq
->sq
.queue
,
255 (unsigned long long)virt_to_phys(wq
->sq
.queue
),
257 (unsigned long long)virt_to_phys(wq
->rq
.queue
));
258 memset(wq
->rq
.queue
, 0, wq
->rq
.memsize
);
259 dma_unmap_addr_set(&wq
->rq
, mapping
, wq
->rq
.dma_addr
);
261 wq
->db
= rdev
->lldi
.db_reg
;
263 wq
->sq
.bar2_va
= c4iw_bar2_addrs(rdev
, wq
->sq
.qid
, T4_BAR2_QTYPE_EGRESS
,
265 user
? &wq
->sq
.bar2_pa
: NULL
);
266 wq
->rq
.bar2_va
= c4iw_bar2_addrs(rdev
, wq
->rq
.qid
, T4_BAR2_QTYPE_EGRESS
,
268 user
? &wq
->rq
.bar2_pa
: NULL
);
271 * User mode must have bar2 access.
273 if (user
&& (!wq
->sq
.bar2_va
|| !wq
->rq
.bar2_va
)) {
274 pr_warn(MOD
"%s: sqid %u or rqid %u not in BAR2 range.\n",
275 pci_name(rdev
->lldi
.pdev
), wq
->sq
.qid
, wq
->rq
.qid
);
282 /* build fw_ri_res_wr */
283 wr_len
= sizeof *res_wr
+ 2 * sizeof *res
;
285 skb
= alloc_skb(wr_len
, GFP_KERNEL
);
290 set_wr_txq(skb
, CPL_PRIORITY_CONTROL
, 0);
292 res_wr
= (struct fw_ri_res_wr
*)__skb_put(skb
, wr_len
);
293 memset(res_wr
, 0, wr_len
);
294 res_wr
->op_nres
= cpu_to_be32(
295 FW_WR_OP_V(FW_RI_RES_WR
) |
296 FW_RI_RES_WR_NRES_V(2) |
298 res_wr
->len16_pkd
= cpu_to_be32(DIV_ROUND_UP(wr_len
, 16));
299 res_wr
->cookie
= (uintptr_t)&wr_wait
;
301 res
->u
.sqrq
.restype
= FW_RI_RES_TYPE_SQ
;
302 res
->u
.sqrq
.op
= FW_RI_RES_OP_WRITE
;
305 * eqsize is the number of 64B entries plus the status page size.
307 eqsize
= wq
->sq
.size
* T4_SQ_NUM_SLOTS
+
308 rdev
->hw_queue
.t4_eq_status_entries
;
310 res
->u
.sqrq
.fetchszm_to_iqid
= cpu_to_be32(
311 FW_RI_RES_WR_HOSTFCMODE_V(0) | /* no host cidx updates */
312 FW_RI_RES_WR_CPRIO_V(0) | /* don't keep in chip cache */
313 FW_RI_RES_WR_PCIECHN_V(0) | /* set by uP at ri_init time */
314 (t4_sq_onchip(&wq
->sq
) ? FW_RI_RES_WR_ONCHIP_F
: 0) |
315 FW_RI_RES_WR_IQID_V(scq
->cqid
));
316 res
->u
.sqrq
.dcaen_to_eqsize
= cpu_to_be32(
317 FW_RI_RES_WR_DCAEN_V(0) |
318 FW_RI_RES_WR_DCACPU_V(0) |
319 FW_RI_RES_WR_FBMIN_V(2) |
320 FW_RI_RES_WR_FBMAX_V(2) |
321 FW_RI_RES_WR_CIDXFTHRESHO_V(0) |
322 FW_RI_RES_WR_CIDXFTHRESH_V(0) |
323 FW_RI_RES_WR_EQSIZE_V(eqsize
));
324 res
->u
.sqrq
.eqid
= cpu_to_be32(wq
->sq
.qid
);
325 res
->u
.sqrq
.eqaddr
= cpu_to_be64(wq
->sq
.dma_addr
);
327 res
->u
.sqrq
.restype
= FW_RI_RES_TYPE_RQ
;
328 res
->u
.sqrq
.op
= FW_RI_RES_OP_WRITE
;
331 * eqsize is the number of 64B entries plus the status page size.
333 eqsize
= wq
->rq
.size
* T4_RQ_NUM_SLOTS
+
334 rdev
->hw_queue
.t4_eq_status_entries
;
335 res
->u
.sqrq
.fetchszm_to_iqid
= cpu_to_be32(
336 FW_RI_RES_WR_HOSTFCMODE_V(0) | /* no host cidx updates */
337 FW_RI_RES_WR_CPRIO_V(0) | /* don't keep in chip cache */
338 FW_RI_RES_WR_PCIECHN_V(0) | /* set by uP at ri_init time */
339 FW_RI_RES_WR_IQID_V(rcq
->cqid
));
340 res
->u
.sqrq
.dcaen_to_eqsize
= cpu_to_be32(
341 FW_RI_RES_WR_DCAEN_V(0) |
342 FW_RI_RES_WR_DCACPU_V(0) |
343 FW_RI_RES_WR_FBMIN_V(2) |
344 FW_RI_RES_WR_FBMAX_V(2) |
345 FW_RI_RES_WR_CIDXFTHRESHO_V(0) |
346 FW_RI_RES_WR_CIDXFTHRESH_V(0) |
347 FW_RI_RES_WR_EQSIZE_V(eqsize
));
348 res
->u
.sqrq
.eqid
= cpu_to_be32(wq
->rq
.qid
);
349 res
->u
.sqrq
.eqaddr
= cpu_to_be64(wq
->rq
.dma_addr
);
351 c4iw_init_wr_wait(&wr_wait
);
353 ret
= c4iw_ofld_send(rdev
, skb
);
356 ret
= c4iw_wait_for_reply(rdev
, &wr_wait
, 0, wq
->sq
.qid
, __func__
);
360 PDBG("%s sqid 0x%x rqid 0x%x kdb 0x%p sq_bar2_addr %p rq_bar2_addr %p\n",
361 __func__
, wq
->sq
.qid
, wq
->rq
.qid
, wq
->db
,
362 wq
->sq
.bar2_va
, wq
->rq
.bar2_va
);
366 dma_free_coherent(&(rdev
->lldi
.pdev
->dev
),
367 wq
->rq
.memsize
, wq
->rq
.queue
,
368 dma_unmap_addr(&wq
->rq
, mapping
));
370 dealloc_sq(rdev
, &wq
->sq
);
372 c4iw_rqtpool_free(rdev
, wq
->rq
.rqt_hwaddr
, wq
->rq
.rqt_size
);
378 c4iw_put_qpid(rdev
, wq
->rq
.qid
, uctx
);
380 c4iw_put_qpid(rdev
, wq
->sq
.qid
, uctx
);
384 static int build_immd(struct t4_sq
*sq
, struct fw_ri_immd
*immdp
,
385 struct ib_send_wr
*wr
, int max
, u32
*plenp
)
392 dstp
= (u8
*)immdp
->data
;
393 for (i
= 0; i
< wr
->num_sge
; i
++) {
394 if ((plen
+ wr
->sg_list
[i
].length
) > max
)
396 srcp
= (u8
*)(unsigned long)wr
->sg_list
[i
].addr
;
397 plen
+= wr
->sg_list
[i
].length
;
398 rem
= wr
->sg_list
[i
].length
;
400 if (dstp
== (u8
*)&sq
->queue
[sq
->size
])
401 dstp
= (u8
*)sq
->queue
;
402 if (rem
<= (u8
*)&sq
->queue
[sq
->size
] - dstp
)
405 len
= (u8
*)&sq
->queue
[sq
->size
] - dstp
;
406 memcpy(dstp
, srcp
, len
);
412 len
= roundup(plen
+ sizeof *immdp
, 16) - (plen
+ sizeof *immdp
);
414 memset(dstp
, 0, len
);
415 immdp
->op
= FW_RI_DATA_IMMD
;
418 immdp
->immdlen
= cpu_to_be32(plen
);
423 static int build_isgl(__be64
*queue_start
, __be64
*queue_end
,
424 struct fw_ri_isgl
*isglp
, struct ib_sge
*sg_list
,
425 int num_sge
, u32
*plenp
)
430 __be64
*flitp
= (__be64
*)isglp
->sge
;
432 for (i
= 0; i
< num_sge
; i
++) {
433 if ((plen
+ sg_list
[i
].length
) < plen
)
435 plen
+= sg_list
[i
].length
;
436 *flitp
= cpu_to_be64(((u64
)sg_list
[i
].lkey
<< 32) |
438 if (++flitp
== queue_end
)
440 *flitp
= cpu_to_be64(sg_list
[i
].addr
);
441 if (++flitp
== queue_end
)
444 *flitp
= (__force __be64
)0;
445 isglp
->op
= FW_RI_DATA_ISGL
;
447 isglp
->nsge
= cpu_to_be16(num_sge
);
454 static int build_rdma_send(struct t4_sq
*sq
, union t4_wr
*wqe
,
455 struct ib_send_wr
*wr
, u8
*len16
)
461 if (wr
->num_sge
> T4_MAX_SEND_SGE
)
463 switch (wr
->opcode
) {
465 if (wr
->send_flags
& IB_SEND_SOLICITED
)
466 wqe
->send
.sendop_pkd
= cpu_to_be32(
467 FW_RI_SEND_WR_SENDOP_V(FW_RI_SEND_WITH_SE
));
469 wqe
->send
.sendop_pkd
= cpu_to_be32(
470 FW_RI_SEND_WR_SENDOP_V(FW_RI_SEND
));
471 wqe
->send
.stag_inv
= 0;
473 case IB_WR_SEND_WITH_INV
:
474 if (wr
->send_flags
& IB_SEND_SOLICITED
)
475 wqe
->send
.sendop_pkd
= cpu_to_be32(
476 FW_RI_SEND_WR_SENDOP_V(FW_RI_SEND_WITH_SE_INV
));
478 wqe
->send
.sendop_pkd
= cpu_to_be32(
479 FW_RI_SEND_WR_SENDOP_V(FW_RI_SEND_WITH_INV
));
480 wqe
->send
.stag_inv
= cpu_to_be32(wr
->ex
.invalidate_rkey
);
491 if (wr
->send_flags
& IB_SEND_INLINE
) {
492 ret
= build_immd(sq
, wqe
->send
.u
.immd_src
, wr
,
493 T4_MAX_SEND_INLINE
, &plen
);
496 size
= sizeof wqe
->send
+ sizeof(struct fw_ri_immd
) +
499 ret
= build_isgl((__be64
*)sq
->queue
,
500 (__be64
*)&sq
->queue
[sq
->size
],
501 wqe
->send
.u
.isgl_src
,
502 wr
->sg_list
, wr
->num_sge
, &plen
);
505 size
= sizeof wqe
->send
+ sizeof(struct fw_ri_isgl
) +
506 wr
->num_sge
* sizeof(struct fw_ri_sge
);
509 wqe
->send
.u
.immd_src
[0].op
= FW_RI_DATA_IMMD
;
510 wqe
->send
.u
.immd_src
[0].r1
= 0;
511 wqe
->send
.u
.immd_src
[0].r2
= 0;
512 wqe
->send
.u
.immd_src
[0].immdlen
= 0;
513 size
= sizeof wqe
->send
+ sizeof(struct fw_ri_immd
);
516 *len16
= DIV_ROUND_UP(size
, 16);
517 wqe
->send
.plen
= cpu_to_be32(plen
);
521 static int build_rdma_write(struct t4_sq
*sq
, union t4_wr
*wqe
,
522 struct ib_send_wr
*wr
, u8
*len16
)
528 if (wr
->num_sge
> T4_MAX_SEND_SGE
)
531 wqe
->write
.stag_sink
= cpu_to_be32(rdma_wr(wr
)->rkey
);
532 wqe
->write
.to_sink
= cpu_to_be64(rdma_wr(wr
)->remote_addr
);
534 if (wr
->send_flags
& IB_SEND_INLINE
) {
535 ret
= build_immd(sq
, wqe
->write
.u
.immd_src
, wr
,
536 T4_MAX_WRITE_INLINE
, &plen
);
539 size
= sizeof wqe
->write
+ sizeof(struct fw_ri_immd
) +
542 ret
= build_isgl((__be64
*)sq
->queue
,
543 (__be64
*)&sq
->queue
[sq
->size
],
544 wqe
->write
.u
.isgl_src
,
545 wr
->sg_list
, wr
->num_sge
, &plen
);
548 size
= sizeof wqe
->write
+ sizeof(struct fw_ri_isgl
) +
549 wr
->num_sge
* sizeof(struct fw_ri_sge
);
552 wqe
->write
.u
.immd_src
[0].op
= FW_RI_DATA_IMMD
;
553 wqe
->write
.u
.immd_src
[0].r1
= 0;
554 wqe
->write
.u
.immd_src
[0].r2
= 0;
555 wqe
->write
.u
.immd_src
[0].immdlen
= 0;
556 size
= sizeof wqe
->write
+ sizeof(struct fw_ri_immd
);
559 *len16
= DIV_ROUND_UP(size
, 16);
560 wqe
->write
.plen
= cpu_to_be32(plen
);
564 static int build_rdma_read(union t4_wr
*wqe
, struct ib_send_wr
*wr
, u8
*len16
)
569 wqe
->read
.stag_src
= cpu_to_be32(rdma_wr(wr
)->rkey
);
570 wqe
->read
.to_src_hi
= cpu_to_be32((u32
)(rdma_wr(wr
)->remote_addr
572 wqe
->read
.to_src_lo
= cpu_to_be32((u32
)rdma_wr(wr
)->remote_addr
);
573 wqe
->read
.stag_sink
= cpu_to_be32(wr
->sg_list
[0].lkey
);
574 wqe
->read
.plen
= cpu_to_be32(wr
->sg_list
[0].length
);
575 wqe
->read
.to_sink_hi
= cpu_to_be32((u32
)(wr
->sg_list
[0].addr
577 wqe
->read
.to_sink_lo
= cpu_to_be32((u32
)(wr
->sg_list
[0].addr
));
579 wqe
->read
.stag_src
= cpu_to_be32(2);
580 wqe
->read
.to_src_hi
= 0;
581 wqe
->read
.to_src_lo
= 0;
582 wqe
->read
.stag_sink
= cpu_to_be32(2);
584 wqe
->read
.to_sink_hi
= 0;
585 wqe
->read
.to_sink_lo
= 0;
589 *len16
= DIV_ROUND_UP(sizeof wqe
->read
, 16);
593 static int build_rdma_recv(struct c4iw_qp
*qhp
, union t4_recv_wr
*wqe
,
594 struct ib_recv_wr
*wr
, u8
*len16
)
598 ret
= build_isgl((__be64
*)qhp
->wq
.rq
.queue
,
599 (__be64
*)&qhp
->wq
.rq
.queue
[qhp
->wq
.rq
.size
],
600 &wqe
->recv
.isgl
, wr
->sg_list
, wr
->num_sge
, NULL
);
603 *len16
= DIV_ROUND_UP(sizeof wqe
->recv
+
604 wr
->num_sge
* sizeof(struct fw_ri_sge
), 16);
608 static int build_memreg(struct t4_sq
*sq
, union t4_wr
*wqe
,
609 struct ib_reg_wr
*wr
, u8
*len16
, u8 t5dev
)
611 struct c4iw_mr
*mhp
= to_c4iw_mr(wr
->mr
);
612 struct fw_ri_immd
*imdp
;
615 int pbllen
= roundup(mhp
->mpl_len
* sizeof(u64
), 32);
618 if (mhp
->mpl_len
> t4_max_fr_depth(use_dsgl
))
621 wqe
->fr
.qpbinde_to_dcacpu
= 0;
622 wqe
->fr
.pgsz_shift
= ilog2(wr
->mr
->page_size
) - 12;
623 wqe
->fr
.addr_type
= FW_RI_VA_BASED_TO
;
624 wqe
->fr
.mem_perms
= c4iw_ib_to_tpt_access(wr
->access
);
626 wqe
->fr
.len_lo
= cpu_to_be32(mhp
->ibmr
.length
);
627 wqe
->fr
.stag
= cpu_to_be32(wr
->key
);
628 wqe
->fr
.va_hi
= cpu_to_be32(mhp
->ibmr
.iova
>> 32);
629 wqe
->fr
.va_lo_fbo
= cpu_to_be32(mhp
->ibmr
.iova
&
632 if (t5dev
&& use_dsgl
&& (pbllen
> max_fr_immd
)) {
633 struct fw_ri_dsgl
*sglp
;
635 for (i
= 0; i
< mhp
->mpl_len
; i
++)
636 mhp
->mpl
[i
] = (__force u64
)cpu_to_be64((u64
)mhp
->mpl
[i
]);
638 sglp
= (struct fw_ri_dsgl
*)(&wqe
->fr
+ 1);
639 sglp
->op
= FW_RI_DATA_DSGL
;
641 sglp
->nsge
= cpu_to_be16(1);
642 sglp
->addr0
= cpu_to_be64(mhp
->mpl_addr
);
643 sglp
->len0
= cpu_to_be32(pbllen
);
645 *len16
= DIV_ROUND_UP(sizeof(wqe
->fr
) + sizeof(*sglp
), 16);
647 imdp
= (struct fw_ri_immd
*)(&wqe
->fr
+ 1);
648 imdp
->op
= FW_RI_DATA_IMMD
;
651 imdp
->immdlen
= cpu_to_be32(pbllen
);
652 p
= (__be64
*)(imdp
+ 1);
654 for (i
= 0; i
< mhp
->mpl_len
; i
++) {
655 *p
= cpu_to_be64((u64
)mhp
->mpl
[i
]);
657 if (++p
== (__be64
*)&sq
->queue
[sq
->size
])
658 p
= (__be64
*)sq
->queue
;
664 if (++p
== (__be64
*)&sq
->queue
[sq
->size
])
665 p
= (__be64
*)sq
->queue
;
667 *len16
= DIV_ROUND_UP(sizeof(wqe
->fr
) + sizeof(*imdp
)
673 static int build_fastreg(struct t4_sq
*sq
, union t4_wr
*wqe
,
674 struct ib_send_wr
*send_wr
, u8
*len16
, u8 t5dev
)
676 struct ib_fast_reg_wr
*wr
= fast_reg_wr(send_wr
);
678 struct fw_ri_immd
*imdp
;
681 int pbllen
= roundup(wr
->page_list_len
* sizeof(u64
), 32);
684 if (wr
->page_list_len
> t4_max_fr_depth(use_dsgl
))
687 wqe
->fr
.qpbinde_to_dcacpu
= 0;
688 wqe
->fr
.pgsz_shift
= wr
->page_shift
- 12;
689 wqe
->fr
.addr_type
= FW_RI_VA_BASED_TO
;
690 wqe
->fr
.mem_perms
= c4iw_ib_to_tpt_access(wr
->access_flags
);
692 wqe
->fr
.len_lo
= cpu_to_be32(wr
->length
);
693 wqe
->fr
.stag
= cpu_to_be32(wr
->rkey
);
694 wqe
->fr
.va_hi
= cpu_to_be32(wr
->iova_start
>> 32);
695 wqe
->fr
.va_lo_fbo
= cpu_to_be32(wr
->iova_start
& 0xffffffff);
697 if (t5dev
&& use_dsgl
&& (pbllen
> max_fr_immd
)) {
698 struct c4iw_fr_page_list
*c4pl
=
699 to_c4iw_fr_page_list(wr
->page_list
);
700 struct fw_ri_dsgl
*sglp
;
702 for (i
= 0; i
< wr
->page_list_len
; i
++) {
703 wr
->page_list
->page_list
[i
] = (__force u64
)
704 cpu_to_be64((u64
)wr
->page_list
->page_list
[i
]);
707 sglp
= (struct fw_ri_dsgl
*)(&wqe
->fr
+ 1);
708 sglp
->op
= FW_RI_DATA_DSGL
;
710 sglp
->nsge
= cpu_to_be16(1);
711 sglp
->addr0
= cpu_to_be64(c4pl
->dma_addr
);
712 sglp
->len0
= cpu_to_be32(pbllen
);
714 *len16
= DIV_ROUND_UP(sizeof(wqe
->fr
) + sizeof(*sglp
), 16);
716 imdp
= (struct fw_ri_immd
*)(&wqe
->fr
+ 1);
717 imdp
->op
= FW_RI_DATA_IMMD
;
720 imdp
->immdlen
= cpu_to_be32(pbllen
);
721 p
= (__be64
*)(imdp
+ 1);
723 for (i
= 0; i
< wr
->page_list_len
; i
++) {
724 *p
= cpu_to_be64((u64
)wr
->page_list
->page_list
[i
]);
726 if (++p
== (__be64
*)&sq
->queue
[sq
->size
])
727 p
= (__be64
*)sq
->queue
;
733 if (++p
== (__be64
*)&sq
->queue
[sq
->size
])
734 p
= (__be64
*)sq
->queue
;
736 *len16
= DIV_ROUND_UP(sizeof(wqe
->fr
) + sizeof(*imdp
)
742 static int build_inv_stag(union t4_wr
*wqe
, struct ib_send_wr
*wr
,
745 wqe
->inv
.stag_inv
= cpu_to_be32(wr
->ex
.invalidate_rkey
);
747 *len16
= DIV_ROUND_UP(sizeof wqe
->inv
, 16);
751 void c4iw_qp_add_ref(struct ib_qp
*qp
)
753 PDBG("%s ib_qp %p\n", __func__
, qp
);
754 atomic_inc(&(to_c4iw_qp(qp
)->refcnt
));
757 void c4iw_qp_rem_ref(struct ib_qp
*qp
)
759 PDBG("%s ib_qp %p\n", __func__
, qp
);
760 if (atomic_dec_and_test(&(to_c4iw_qp(qp
)->refcnt
)))
761 wake_up(&(to_c4iw_qp(qp
)->wait
));
764 static void add_to_fc_list(struct list_head
*head
, struct list_head
*entry
)
766 if (list_empty(entry
))
767 list_add_tail(entry
, head
);
770 static int ring_kernel_sq_db(struct c4iw_qp
*qhp
, u16 inc
)
774 spin_lock_irqsave(&qhp
->rhp
->lock
, flags
);
775 spin_lock(&qhp
->lock
);
776 if (qhp
->rhp
->db_state
== NORMAL
)
777 t4_ring_sq_db(&qhp
->wq
, inc
, NULL
);
779 add_to_fc_list(&qhp
->rhp
->db_fc_list
, &qhp
->db_fc_entry
);
780 qhp
->wq
.sq
.wq_pidx_inc
+= inc
;
782 spin_unlock(&qhp
->lock
);
783 spin_unlock_irqrestore(&qhp
->rhp
->lock
, flags
);
787 static int ring_kernel_rq_db(struct c4iw_qp
*qhp
, u16 inc
)
791 spin_lock_irqsave(&qhp
->rhp
->lock
, flags
);
792 spin_lock(&qhp
->lock
);
793 if (qhp
->rhp
->db_state
== NORMAL
)
794 t4_ring_rq_db(&qhp
->wq
, inc
, NULL
);
796 add_to_fc_list(&qhp
->rhp
->db_fc_list
, &qhp
->db_fc_entry
);
797 qhp
->wq
.rq
.wq_pidx_inc
+= inc
;
799 spin_unlock(&qhp
->lock
);
800 spin_unlock_irqrestore(&qhp
->rhp
->lock
, flags
);
804 int c4iw_post_send(struct ib_qp
*ibqp
, struct ib_send_wr
*wr
,
805 struct ib_send_wr
**bad_wr
)
809 enum fw_wr_opcodes fw_opcode
= 0;
810 enum fw_ri_wr_flags fw_flags
;
812 union t4_wr
*wqe
= NULL
;
814 struct t4_swsqe
*swsqe
;
818 qhp
= to_c4iw_qp(ibqp
);
819 spin_lock_irqsave(&qhp
->lock
, flag
);
820 if (t4_wq_in_error(&qhp
->wq
)) {
821 spin_unlock_irqrestore(&qhp
->lock
, flag
);
824 num_wrs
= t4_sq_avail(&qhp
->wq
);
826 spin_unlock_irqrestore(&qhp
->lock
, flag
);
835 wqe
= (union t4_wr
*)((u8
*)qhp
->wq
.sq
.queue
+
836 qhp
->wq
.sq
.wq_pidx
* T4_EQ_ENTRY_SIZE
);
839 if (wr
->send_flags
& IB_SEND_SOLICITED
)
840 fw_flags
|= FW_RI_SOLICITED_EVENT_FLAG
;
841 if (wr
->send_flags
& IB_SEND_SIGNALED
|| qhp
->sq_sig_all
)
842 fw_flags
|= FW_RI_COMPLETION_FLAG
;
843 swsqe
= &qhp
->wq
.sq
.sw_sq
[qhp
->wq
.sq
.pidx
];
844 switch (wr
->opcode
) {
845 case IB_WR_SEND_WITH_INV
:
847 if (wr
->send_flags
& IB_SEND_FENCE
)
848 fw_flags
|= FW_RI_READ_FENCE_FLAG
;
849 fw_opcode
= FW_RI_SEND_WR
;
850 if (wr
->opcode
== IB_WR_SEND
)
851 swsqe
->opcode
= FW_RI_SEND
;
853 swsqe
->opcode
= FW_RI_SEND_WITH_INV
;
854 err
= build_rdma_send(&qhp
->wq
.sq
, wqe
, wr
, &len16
);
856 case IB_WR_RDMA_WRITE
:
857 fw_opcode
= FW_RI_RDMA_WRITE_WR
;
858 swsqe
->opcode
= FW_RI_RDMA_WRITE
;
859 err
= build_rdma_write(&qhp
->wq
.sq
, wqe
, wr
, &len16
);
861 case IB_WR_RDMA_READ
:
862 case IB_WR_RDMA_READ_WITH_INV
:
863 fw_opcode
= FW_RI_RDMA_READ_WR
;
864 swsqe
->opcode
= FW_RI_READ_REQ
;
865 if (wr
->opcode
== IB_WR_RDMA_READ_WITH_INV
)
866 fw_flags
= FW_RI_RDMA_READ_INVALIDATE
;
869 err
= build_rdma_read(wqe
, wr
, &len16
);
872 swsqe
->read_len
= wr
->sg_list
[0].length
;
873 if (!qhp
->wq
.sq
.oldest_read
)
874 qhp
->wq
.sq
.oldest_read
= swsqe
;
876 case IB_WR_FAST_REG_MR
:
877 fw_opcode
= FW_RI_FR_NSMR_WR
;
878 swsqe
->opcode
= FW_RI_FAST_REGISTER
;
879 err
= build_fastreg(&qhp
->wq
.sq
, wqe
, wr
, &len16
,
881 qhp
->rhp
->rdev
.lldi
.adapter_type
) ?
885 fw_opcode
= FW_RI_FR_NSMR_WR
;
886 swsqe
->opcode
= FW_RI_FAST_REGISTER
;
887 err
= build_memreg(&qhp
->wq
.sq
, wqe
, reg_wr(wr
), &len16
,
889 qhp
->rhp
->rdev
.lldi
.adapter_type
) ?
892 case IB_WR_LOCAL_INV
:
893 if (wr
->send_flags
& IB_SEND_FENCE
)
894 fw_flags
|= FW_RI_LOCAL_FENCE_FLAG
;
895 fw_opcode
= FW_RI_INV_LSTAG_WR
;
896 swsqe
->opcode
= FW_RI_LOCAL_INV
;
897 err
= build_inv_stag(wqe
, wr
, &len16
);
900 PDBG("%s post of type=%d TBD!\n", __func__
,
908 swsqe
->idx
= qhp
->wq
.sq
.pidx
;
910 swsqe
->signaled
= (wr
->send_flags
& IB_SEND_SIGNALED
) ||
913 swsqe
->wr_id
= wr
->wr_id
;
915 swsqe
->sge_ts
= cxgb4_read_sge_timestamp(
916 qhp
->rhp
->rdev
.lldi
.ports
[0]);
917 getnstimeofday(&swsqe
->host_ts
);
920 init_wr_hdr(wqe
, qhp
->wq
.sq
.pidx
, fw_opcode
, fw_flags
, len16
);
922 PDBG("%s cookie 0x%llx pidx 0x%x opcode 0x%x read_len %u\n",
923 __func__
, (unsigned long long)wr
->wr_id
, qhp
->wq
.sq
.pidx
,
924 swsqe
->opcode
, swsqe
->read_len
);
927 t4_sq_produce(&qhp
->wq
, len16
);
928 idx
+= DIV_ROUND_UP(len16
*16, T4_EQ_ENTRY_SIZE
);
930 if (!qhp
->rhp
->rdev
.status_page
->db_off
) {
931 t4_ring_sq_db(&qhp
->wq
, idx
, wqe
);
932 spin_unlock_irqrestore(&qhp
->lock
, flag
);
934 spin_unlock_irqrestore(&qhp
->lock
, flag
);
935 ring_kernel_sq_db(qhp
, idx
);
940 int c4iw_post_receive(struct ib_qp
*ibqp
, struct ib_recv_wr
*wr
,
941 struct ib_recv_wr
**bad_wr
)
945 union t4_recv_wr
*wqe
= NULL
;
951 qhp
= to_c4iw_qp(ibqp
);
952 spin_lock_irqsave(&qhp
->lock
, flag
);
953 if (t4_wq_in_error(&qhp
->wq
)) {
954 spin_unlock_irqrestore(&qhp
->lock
, flag
);
957 num_wrs
= t4_rq_avail(&qhp
->wq
);
959 spin_unlock_irqrestore(&qhp
->lock
, flag
);
963 if (wr
->num_sge
> T4_MAX_RECV_SGE
) {
968 wqe
= (union t4_recv_wr
*)((u8
*)qhp
->wq
.rq
.queue
+
972 err
= build_rdma_recv(qhp
, wqe
, wr
, &len16
);
980 qhp
->wq
.rq
.sw_rq
[qhp
->wq
.rq
.pidx
].wr_id
= wr
->wr_id
;
982 qhp
->wq
.rq
.sw_rq
[qhp
->wq
.rq
.pidx
].sge_ts
=
983 cxgb4_read_sge_timestamp(
984 qhp
->rhp
->rdev
.lldi
.ports
[0]);
986 &qhp
->wq
.rq
.sw_rq
[qhp
->wq
.rq
.pidx
].host_ts
);
989 wqe
->recv
.opcode
= FW_RI_RECV_WR
;
991 wqe
->recv
.wrid
= qhp
->wq
.rq
.pidx
;
995 wqe
->recv
.len16
= len16
;
996 PDBG("%s cookie 0x%llx pidx %u\n", __func__
,
997 (unsigned long long) wr
->wr_id
, qhp
->wq
.rq
.pidx
);
998 t4_rq_produce(&qhp
->wq
, len16
);
999 idx
+= DIV_ROUND_UP(len16
*16, T4_EQ_ENTRY_SIZE
);
1003 if (!qhp
->rhp
->rdev
.status_page
->db_off
) {
1004 t4_ring_rq_db(&qhp
->wq
, idx
, wqe
);
1005 spin_unlock_irqrestore(&qhp
->lock
, flag
);
1007 spin_unlock_irqrestore(&qhp
->lock
, flag
);
1008 ring_kernel_rq_db(qhp
, idx
);
1013 int c4iw_bind_mw(struct ib_qp
*qp
, struct ib_mw
*mw
, struct ib_mw_bind
*mw_bind
)
1018 static inline void build_term_codes(struct t4_cqe
*err_cqe
, u8
*layer_type
,
1028 *layer_type
= LAYER_RDMAP
|DDP_LOCAL_CATA
;
1033 status
= CQE_STATUS(err_cqe
);
1034 opcode
= CQE_OPCODE(err_cqe
);
1035 rqtype
= RQ_TYPE(err_cqe
);
1036 send_inv
= (opcode
== FW_RI_SEND_WITH_INV
) ||
1037 (opcode
== FW_RI_SEND_WITH_SE_INV
);
1038 tagged
= (opcode
== FW_RI_RDMA_WRITE
) ||
1039 (rqtype
&& (opcode
== FW_RI_READ_RESP
));
1044 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_OP
;
1045 *ecode
= RDMAP_CANT_INV_STAG
;
1047 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
1048 *ecode
= RDMAP_INV_STAG
;
1052 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
1053 if ((opcode
== FW_RI_SEND_WITH_INV
) ||
1054 (opcode
== FW_RI_SEND_WITH_SE_INV
))
1055 *ecode
= RDMAP_CANT_INV_STAG
;
1057 *ecode
= RDMAP_STAG_NOT_ASSOC
;
1060 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
1061 *ecode
= RDMAP_STAG_NOT_ASSOC
;
1064 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
1065 *ecode
= RDMAP_ACC_VIOL
;
1068 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
1069 *ecode
= RDMAP_TO_WRAP
;
1073 *layer_type
= LAYER_DDP
|DDP_TAGGED_ERR
;
1074 *ecode
= DDPT_BASE_BOUNDS
;
1076 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_PROT
;
1077 *ecode
= RDMAP_BASE_BOUNDS
;
1080 case T4_ERR_INVALIDATE_SHARED_MR
:
1081 case T4_ERR_INVALIDATE_MR_WITH_MW_BOUND
:
1082 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_OP
;
1083 *ecode
= RDMAP_CANT_INV_STAG
;
1086 case T4_ERR_ECC_PSTAG
:
1087 case T4_ERR_INTERNAL_ERR
:
1088 *layer_type
= LAYER_RDMAP
|RDMAP_LOCAL_CATA
;
1091 case T4_ERR_OUT_OF_RQE
:
1092 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
1093 *ecode
= DDPU_INV_MSN_NOBUF
;
1095 case T4_ERR_PBL_ADDR_BOUND
:
1096 *layer_type
= LAYER_DDP
|DDP_TAGGED_ERR
;
1097 *ecode
= DDPT_BASE_BOUNDS
;
1100 *layer_type
= LAYER_MPA
|DDP_LLP
;
1101 *ecode
= MPA_CRC_ERR
;
1104 *layer_type
= LAYER_MPA
|DDP_LLP
;
1105 *ecode
= MPA_MARKER_ERR
;
1107 case T4_ERR_PDU_LEN_ERR
:
1108 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
1109 *ecode
= DDPU_MSG_TOOBIG
;
1111 case T4_ERR_DDP_VERSION
:
1113 *layer_type
= LAYER_DDP
|DDP_TAGGED_ERR
;
1114 *ecode
= DDPT_INV_VERS
;
1116 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
1117 *ecode
= DDPU_INV_VERS
;
1120 case T4_ERR_RDMA_VERSION
:
1121 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_OP
;
1122 *ecode
= RDMAP_INV_VERS
;
1125 *layer_type
= LAYER_RDMAP
|RDMAP_REMOTE_OP
;
1126 *ecode
= RDMAP_INV_OPCODE
;
1128 case T4_ERR_DDP_QUEUE_NUM
:
1129 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
1130 *ecode
= DDPU_INV_QN
;
1133 case T4_ERR_MSN_GAP
:
1134 case T4_ERR_MSN_RANGE
:
1135 case T4_ERR_IRD_OVERFLOW
:
1136 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
1137 *ecode
= DDPU_INV_MSN_RANGE
;
1140 *layer_type
= LAYER_DDP
|DDP_LOCAL_CATA
;
1144 *layer_type
= LAYER_DDP
|DDP_UNTAGGED_ERR
;
1145 *ecode
= DDPU_INV_MO
;
1148 *layer_type
= LAYER_RDMAP
|DDP_LOCAL_CATA
;
1154 static void post_terminate(struct c4iw_qp
*qhp
, struct t4_cqe
*err_cqe
,
1157 struct fw_ri_wr
*wqe
;
1158 struct sk_buff
*skb
;
1159 struct terminate_message
*term
;
1161 PDBG("%s qhp %p qid 0x%x tid %u\n", __func__
, qhp
, qhp
->wq
.sq
.qid
,
1164 skb
= alloc_skb(sizeof *wqe
, gfp
);
1167 set_wr_txq(skb
, CPL_PRIORITY_DATA
, qhp
->ep
->txq_idx
);
1169 wqe
= (struct fw_ri_wr
*)__skb_put(skb
, sizeof(*wqe
));
1170 memset(wqe
, 0, sizeof *wqe
);
1171 wqe
->op_compl
= cpu_to_be32(FW_WR_OP_V(FW_RI_INIT_WR
));
1172 wqe
->flowid_len16
= cpu_to_be32(
1173 FW_WR_FLOWID_V(qhp
->ep
->hwtid
) |
1174 FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*wqe
), 16)));
1176 wqe
->u
.terminate
.type
= FW_RI_TYPE_TERMINATE
;
1177 wqe
->u
.terminate
.immdlen
= cpu_to_be32(sizeof *term
);
1178 term
= (struct terminate_message
*)wqe
->u
.terminate
.termmsg
;
1179 if (qhp
->attr
.layer_etype
== (LAYER_MPA
|DDP_LLP
)) {
1180 term
->layer_etype
= qhp
->attr
.layer_etype
;
1181 term
->ecode
= qhp
->attr
.ecode
;
1183 build_term_codes(err_cqe
, &term
->layer_etype
, &term
->ecode
);
1184 c4iw_ofld_send(&qhp
->rhp
->rdev
, skb
);
1188 * Assumes qhp lock is held.
1190 static void __flush_qp(struct c4iw_qp
*qhp
, struct c4iw_cq
*rchp
,
1191 struct c4iw_cq
*schp
)
1194 int rq_flushed
, sq_flushed
;
1197 PDBG("%s qhp %p rchp %p schp %p\n", __func__
, qhp
, rchp
, schp
);
1199 /* locking hierarchy: cq lock first, then qp lock. */
1200 spin_lock_irqsave(&rchp
->lock
, flag
);
1201 spin_lock(&qhp
->lock
);
1203 if (qhp
->wq
.flushed
) {
1204 spin_unlock(&qhp
->lock
);
1205 spin_unlock_irqrestore(&rchp
->lock
, flag
);
1208 qhp
->wq
.flushed
= 1;
1210 c4iw_flush_hw_cq(rchp
);
1211 c4iw_count_rcqes(&rchp
->cq
, &qhp
->wq
, &count
);
1212 rq_flushed
= c4iw_flush_rq(&qhp
->wq
, &rchp
->cq
, count
);
1213 spin_unlock(&qhp
->lock
);
1214 spin_unlock_irqrestore(&rchp
->lock
, flag
);
1216 /* locking hierarchy: cq lock first, then qp lock. */
1217 spin_lock_irqsave(&schp
->lock
, flag
);
1218 spin_lock(&qhp
->lock
);
1220 c4iw_flush_hw_cq(schp
);
1221 sq_flushed
= c4iw_flush_sq(qhp
);
1222 spin_unlock(&qhp
->lock
);
1223 spin_unlock_irqrestore(&schp
->lock
, flag
);
1226 if (t4_clear_cq_armed(&rchp
->cq
) &&
1227 (rq_flushed
|| sq_flushed
)) {
1228 spin_lock_irqsave(&rchp
->comp_handler_lock
, flag
);
1229 (*rchp
->ibcq
.comp_handler
)(&rchp
->ibcq
,
1230 rchp
->ibcq
.cq_context
);
1231 spin_unlock_irqrestore(&rchp
->comp_handler_lock
, flag
);
1234 if (t4_clear_cq_armed(&rchp
->cq
) && rq_flushed
) {
1235 spin_lock_irqsave(&rchp
->comp_handler_lock
, flag
);
1236 (*rchp
->ibcq
.comp_handler
)(&rchp
->ibcq
,
1237 rchp
->ibcq
.cq_context
);
1238 spin_unlock_irqrestore(&rchp
->comp_handler_lock
, flag
);
1240 if (t4_clear_cq_armed(&schp
->cq
) && sq_flushed
) {
1241 spin_lock_irqsave(&schp
->comp_handler_lock
, flag
);
1242 (*schp
->ibcq
.comp_handler
)(&schp
->ibcq
,
1243 schp
->ibcq
.cq_context
);
1244 spin_unlock_irqrestore(&schp
->comp_handler_lock
, flag
);
1249 static void flush_qp(struct c4iw_qp
*qhp
)
1251 struct c4iw_cq
*rchp
, *schp
;
1254 rchp
= to_c4iw_cq(qhp
->ibqp
.recv_cq
);
1255 schp
= to_c4iw_cq(qhp
->ibqp
.send_cq
);
1257 t4_set_wq_in_error(&qhp
->wq
);
1258 if (qhp
->ibqp
.uobject
) {
1259 t4_set_cq_in_error(&rchp
->cq
);
1260 spin_lock_irqsave(&rchp
->comp_handler_lock
, flag
);
1261 (*rchp
->ibcq
.comp_handler
)(&rchp
->ibcq
, rchp
->ibcq
.cq_context
);
1262 spin_unlock_irqrestore(&rchp
->comp_handler_lock
, flag
);
1264 t4_set_cq_in_error(&schp
->cq
);
1265 spin_lock_irqsave(&schp
->comp_handler_lock
, flag
);
1266 (*schp
->ibcq
.comp_handler
)(&schp
->ibcq
,
1267 schp
->ibcq
.cq_context
);
1268 spin_unlock_irqrestore(&schp
->comp_handler_lock
, flag
);
1272 __flush_qp(qhp
, rchp
, schp
);
1275 static int rdma_fini(struct c4iw_dev
*rhp
, struct c4iw_qp
*qhp
,
1278 struct fw_ri_wr
*wqe
;
1280 struct sk_buff
*skb
;
1282 PDBG("%s qhp %p qid 0x%x tid %u\n", __func__
, qhp
, qhp
->wq
.sq
.qid
,
1285 skb
= alloc_skb(sizeof *wqe
, GFP_KERNEL
);
1288 set_wr_txq(skb
, CPL_PRIORITY_DATA
, ep
->txq_idx
);
1290 wqe
= (struct fw_ri_wr
*)__skb_put(skb
, sizeof(*wqe
));
1291 memset(wqe
, 0, sizeof *wqe
);
1292 wqe
->op_compl
= cpu_to_be32(
1293 FW_WR_OP_V(FW_RI_INIT_WR
) |
1295 wqe
->flowid_len16
= cpu_to_be32(
1296 FW_WR_FLOWID_V(ep
->hwtid
) |
1297 FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*wqe
), 16)));
1298 wqe
->cookie
= (uintptr_t)&ep
->com
.wr_wait
;
1300 wqe
->u
.fini
.type
= FW_RI_TYPE_FINI
;
1301 ret
= c4iw_ofld_send(&rhp
->rdev
, skb
);
1305 ret
= c4iw_wait_for_reply(&rhp
->rdev
, &ep
->com
.wr_wait
, qhp
->ep
->hwtid
,
1306 qhp
->wq
.sq
.qid
, __func__
);
1308 PDBG("%s ret %d\n", __func__
, ret
);
1312 static void build_rtr_msg(u8 p2p_type
, struct fw_ri_init
*init
)
1314 PDBG("%s p2p_type = %d\n", __func__
, p2p_type
);
1315 memset(&init
->u
, 0, sizeof init
->u
);
1317 case FW_RI_INIT_P2PTYPE_RDMA_WRITE
:
1318 init
->u
.write
.opcode
= FW_RI_RDMA_WRITE_WR
;
1319 init
->u
.write
.stag_sink
= cpu_to_be32(1);
1320 init
->u
.write
.to_sink
= cpu_to_be64(1);
1321 init
->u
.write
.u
.immd_src
[0].op
= FW_RI_DATA_IMMD
;
1322 init
->u
.write
.len16
= DIV_ROUND_UP(sizeof init
->u
.write
+
1323 sizeof(struct fw_ri_immd
),
1326 case FW_RI_INIT_P2PTYPE_READ_REQ
:
1327 init
->u
.write
.opcode
= FW_RI_RDMA_READ_WR
;
1328 init
->u
.read
.stag_src
= cpu_to_be32(1);
1329 init
->u
.read
.to_src_lo
= cpu_to_be32(1);
1330 init
->u
.read
.stag_sink
= cpu_to_be32(1);
1331 init
->u
.read
.to_sink_lo
= cpu_to_be32(1);
1332 init
->u
.read
.len16
= DIV_ROUND_UP(sizeof init
->u
.read
, 16);
1337 static int rdma_init(struct c4iw_dev
*rhp
, struct c4iw_qp
*qhp
)
1339 struct fw_ri_wr
*wqe
;
1341 struct sk_buff
*skb
;
1343 PDBG("%s qhp %p qid 0x%x tid %u ird %u ord %u\n", __func__
, qhp
,
1344 qhp
->wq
.sq
.qid
, qhp
->ep
->hwtid
, qhp
->ep
->ird
, qhp
->ep
->ord
);
1346 skb
= alloc_skb(sizeof *wqe
, GFP_KERNEL
);
1351 ret
= alloc_ird(rhp
, qhp
->attr
.max_ird
);
1353 qhp
->attr
.max_ird
= 0;
1357 set_wr_txq(skb
, CPL_PRIORITY_DATA
, qhp
->ep
->txq_idx
);
1359 wqe
= (struct fw_ri_wr
*)__skb_put(skb
, sizeof(*wqe
));
1360 memset(wqe
, 0, sizeof *wqe
);
1361 wqe
->op_compl
= cpu_to_be32(
1362 FW_WR_OP_V(FW_RI_INIT_WR
) |
1364 wqe
->flowid_len16
= cpu_to_be32(
1365 FW_WR_FLOWID_V(qhp
->ep
->hwtid
) |
1366 FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*wqe
), 16)));
1368 wqe
->cookie
= (uintptr_t)&qhp
->ep
->com
.wr_wait
;
1370 wqe
->u
.init
.type
= FW_RI_TYPE_INIT
;
1371 wqe
->u
.init
.mpareqbit_p2ptype
=
1372 FW_RI_WR_MPAREQBIT_V(qhp
->attr
.mpa_attr
.initiator
) |
1373 FW_RI_WR_P2PTYPE_V(qhp
->attr
.mpa_attr
.p2p_type
);
1374 wqe
->u
.init
.mpa_attrs
= FW_RI_MPA_IETF_ENABLE
;
1375 if (qhp
->attr
.mpa_attr
.recv_marker_enabled
)
1376 wqe
->u
.init
.mpa_attrs
|= FW_RI_MPA_RX_MARKER_ENABLE
;
1377 if (qhp
->attr
.mpa_attr
.xmit_marker_enabled
)
1378 wqe
->u
.init
.mpa_attrs
|= FW_RI_MPA_TX_MARKER_ENABLE
;
1379 if (qhp
->attr
.mpa_attr
.crc_enabled
)
1380 wqe
->u
.init
.mpa_attrs
|= FW_RI_MPA_CRC_ENABLE
;
1382 wqe
->u
.init
.qp_caps
= FW_RI_QP_RDMA_READ_ENABLE
|
1383 FW_RI_QP_RDMA_WRITE_ENABLE
|
1384 FW_RI_QP_BIND_ENABLE
;
1385 if (!qhp
->ibqp
.uobject
)
1386 wqe
->u
.init
.qp_caps
|= FW_RI_QP_FAST_REGISTER_ENABLE
|
1387 FW_RI_QP_STAG0_ENABLE
;
1388 wqe
->u
.init
.nrqe
= cpu_to_be16(t4_rqes_posted(&qhp
->wq
));
1389 wqe
->u
.init
.pdid
= cpu_to_be32(qhp
->attr
.pd
);
1390 wqe
->u
.init
.qpid
= cpu_to_be32(qhp
->wq
.sq
.qid
);
1391 wqe
->u
.init
.sq_eqid
= cpu_to_be32(qhp
->wq
.sq
.qid
);
1392 wqe
->u
.init
.rq_eqid
= cpu_to_be32(qhp
->wq
.rq
.qid
);
1393 wqe
->u
.init
.scqid
= cpu_to_be32(qhp
->attr
.scq
);
1394 wqe
->u
.init
.rcqid
= cpu_to_be32(qhp
->attr
.rcq
);
1395 wqe
->u
.init
.ord_max
= cpu_to_be32(qhp
->attr
.max_ord
);
1396 wqe
->u
.init
.ird_max
= cpu_to_be32(qhp
->attr
.max_ird
);
1397 wqe
->u
.init
.iss
= cpu_to_be32(qhp
->ep
->snd_seq
);
1398 wqe
->u
.init
.irs
= cpu_to_be32(qhp
->ep
->rcv_seq
);
1399 wqe
->u
.init
.hwrqsize
= cpu_to_be32(qhp
->wq
.rq
.rqt_size
);
1400 wqe
->u
.init
.hwrqaddr
= cpu_to_be32(qhp
->wq
.rq
.rqt_hwaddr
-
1401 rhp
->rdev
.lldi
.vr
->rq
.start
);
1402 if (qhp
->attr
.mpa_attr
.initiator
)
1403 build_rtr_msg(qhp
->attr
.mpa_attr
.p2p_type
, &wqe
->u
.init
);
1405 ret
= c4iw_ofld_send(&rhp
->rdev
, skb
);
1409 ret
= c4iw_wait_for_reply(&rhp
->rdev
, &qhp
->ep
->com
.wr_wait
,
1410 qhp
->ep
->hwtid
, qhp
->wq
.sq
.qid
, __func__
);
1414 free_ird(rhp
, qhp
->attr
.max_ird
);
1416 PDBG("%s ret %d\n", __func__
, ret
);
1420 int c4iw_modify_qp(struct c4iw_dev
*rhp
, struct c4iw_qp
*qhp
,
1421 enum c4iw_qp_attr_mask mask
,
1422 struct c4iw_qp_attributes
*attrs
,
1426 struct c4iw_qp_attributes newattr
= qhp
->attr
;
1431 struct c4iw_ep
*ep
= NULL
;
1433 PDBG("%s qhp %p sqid 0x%x rqid 0x%x ep %p state %d -> %d\n", __func__
,
1434 qhp
, qhp
->wq
.sq
.qid
, qhp
->wq
.rq
.qid
, qhp
->ep
, qhp
->attr
.state
,
1435 (mask
& C4IW_QP_ATTR_NEXT_STATE
) ? attrs
->next_state
: -1);
1437 mutex_lock(&qhp
->mutex
);
1439 /* Process attr changes if in IDLE */
1440 if (mask
& C4IW_QP_ATTR_VALID_MODIFY
) {
1441 if (qhp
->attr
.state
!= C4IW_QP_STATE_IDLE
) {
1445 if (mask
& C4IW_QP_ATTR_ENABLE_RDMA_READ
)
1446 newattr
.enable_rdma_read
= attrs
->enable_rdma_read
;
1447 if (mask
& C4IW_QP_ATTR_ENABLE_RDMA_WRITE
)
1448 newattr
.enable_rdma_write
= attrs
->enable_rdma_write
;
1449 if (mask
& C4IW_QP_ATTR_ENABLE_RDMA_BIND
)
1450 newattr
.enable_bind
= attrs
->enable_bind
;
1451 if (mask
& C4IW_QP_ATTR_MAX_ORD
) {
1452 if (attrs
->max_ord
> c4iw_max_read_depth
) {
1456 newattr
.max_ord
= attrs
->max_ord
;
1458 if (mask
& C4IW_QP_ATTR_MAX_IRD
) {
1459 if (attrs
->max_ird
> cur_max_read_depth(rhp
)) {
1463 newattr
.max_ird
= attrs
->max_ird
;
1465 qhp
->attr
= newattr
;
1468 if (mask
& C4IW_QP_ATTR_SQ_DB
) {
1469 ret
= ring_kernel_sq_db(qhp
, attrs
->sq_db_inc
);
1472 if (mask
& C4IW_QP_ATTR_RQ_DB
) {
1473 ret
= ring_kernel_rq_db(qhp
, attrs
->rq_db_inc
);
1477 if (!(mask
& C4IW_QP_ATTR_NEXT_STATE
))
1479 if (qhp
->attr
.state
== attrs
->next_state
)
1482 switch (qhp
->attr
.state
) {
1483 case C4IW_QP_STATE_IDLE
:
1484 switch (attrs
->next_state
) {
1485 case C4IW_QP_STATE_RTS
:
1486 if (!(mask
& C4IW_QP_ATTR_LLP_STREAM_HANDLE
)) {
1490 if (!(mask
& C4IW_QP_ATTR_MPA_ATTR
)) {
1494 qhp
->attr
.mpa_attr
= attrs
->mpa_attr
;
1495 qhp
->attr
.llp_stream_handle
= attrs
->llp_stream_handle
;
1496 qhp
->ep
= qhp
->attr
.llp_stream_handle
;
1497 set_state(qhp
, C4IW_QP_STATE_RTS
);
1500 * Ref the endpoint here and deref when we
1501 * disassociate the endpoint from the QP. This
1502 * happens in CLOSING->IDLE transition or *->ERROR
1505 c4iw_get_ep(&qhp
->ep
->com
);
1506 ret
= rdma_init(rhp
, qhp
);
1510 case C4IW_QP_STATE_ERROR
:
1511 set_state(qhp
, C4IW_QP_STATE_ERROR
);
1519 case C4IW_QP_STATE_RTS
:
1520 switch (attrs
->next_state
) {
1521 case C4IW_QP_STATE_CLOSING
:
1522 BUG_ON(atomic_read(&qhp
->ep
->com
.kref
.refcount
) < 2);
1523 t4_set_wq_in_error(&qhp
->wq
);
1524 set_state(qhp
, C4IW_QP_STATE_CLOSING
);
1529 c4iw_get_ep(&qhp
->ep
->com
);
1531 ret
= rdma_fini(rhp
, qhp
, ep
);
1535 case C4IW_QP_STATE_TERMINATE
:
1536 t4_set_wq_in_error(&qhp
->wq
);
1537 set_state(qhp
, C4IW_QP_STATE_TERMINATE
);
1538 qhp
->attr
.layer_etype
= attrs
->layer_etype
;
1539 qhp
->attr
.ecode
= attrs
->ecode
;
1542 c4iw_get_ep(&qhp
->ep
->com
);
1546 terminate
= qhp
->attr
.send_term
;
1547 ret
= rdma_fini(rhp
, qhp
, ep
);
1552 case C4IW_QP_STATE_ERROR
:
1553 t4_set_wq_in_error(&qhp
->wq
);
1554 set_state(qhp
, C4IW_QP_STATE_ERROR
);
1559 c4iw_get_ep(&qhp
->ep
->com
);
1568 case C4IW_QP_STATE_CLOSING
:
1573 switch (attrs
->next_state
) {
1574 case C4IW_QP_STATE_IDLE
:
1576 set_state(qhp
, C4IW_QP_STATE_IDLE
);
1577 qhp
->attr
.llp_stream_handle
= NULL
;
1578 c4iw_put_ep(&qhp
->ep
->com
);
1580 wake_up(&qhp
->wait
);
1582 case C4IW_QP_STATE_ERROR
:
1589 case C4IW_QP_STATE_ERROR
:
1590 if (attrs
->next_state
!= C4IW_QP_STATE_IDLE
) {
1594 if (!t4_sq_empty(&qhp
->wq
) || !t4_rq_empty(&qhp
->wq
)) {
1598 set_state(qhp
, C4IW_QP_STATE_IDLE
);
1600 case C4IW_QP_STATE_TERMINATE
:
1608 printk(KERN_ERR
"%s in a bad state %d\n",
1609 __func__
, qhp
->attr
.state
);
1616 PDBG("%s disassociating ep %p qpid 0x%x\n", __func__
, qhp
->ep
,
1619 /* disassociate the LLP connection */
1620 qhp
->attr
.llp_stream_handle
= NULL
;
1624 set_state(qhp
, C4IW_QP_STATE_ERROR
);
1629 wake_up(&qhp
->wait
);
1631 mutex_unlock(&qhp
->mutex
);
1634 post_terminate(qhp
, NULL
, internal
? GFP_ATOMIC
: GFP_KERNEL
);
1637 * If disconnect is 1, then we need to initiate a disconnect
1638 * on the EP. This can be a normal close (RTS->CLOSING) or
1639 * an abnormal close (RTS/CLOSING->ERROR).
1642 c4iw_ep_disconnect(ep
, abort
, internal
? GFP_ATOMIC
:
1644 c4iw_put_ep(&ep
->com
);
1648 * If free is 1, then we've disassociated the EP from the QP
1649 * and we need to dereference the EP.
1652 c4iw_put_ep(&ep
->com
);
1653 PDBG("%s exit state %d\n", __func__
, qhp
->attr
.state
);
1657 int c4iw_destroy_qp(struct ib_qp
*ib_qp
)
1659 struct c4iw_dev
*rhp
;
1660 struct c4iw_qp
*qhp
;
1661 struct c4iw_qp_attributes attrs
;
1662 struct c4iw_ucontext
*ucontext
;
1664 qhp
= to_c4iw_qp(ib_qp
);
1667 attrs
.next_state
= C4IW_QP_STATE_ERROR
;
1668 if (qhp
->attr
.state
== C4IW_QP_STATE_TERMINATE
)
1669 c4iw_modify_qp(rhp
, qhp
, C4IW_QP_ATTR_NEXT_STATE
, &attrs
, 1);
1671 c4iw_modify_qp(rhp
, qhp
, C4IW_QP_ATTR_NEXT_STATE
, &attrs
, 0);
1672 wait_event(qhp
->wait
, !qhp
->ep
);
1674 remove_handle(rhp
, &rhp
->qpidr
, qhp
->wq
.sq
.qid
);
1675 atomic_dec(&qhp
->refcnt
);
1676 wait_event(qhp
->wait
, !atomic_read(&qhp
->refcnt
));
1678 spin_lock_irq(&rhp
->lock
);
1679 if (!list_empty(&qhp
->db_fc_entry
))
1680 list_del_init(&qhp
->db_fc_entry
);
1681 spin_unlock_irq(&rhp
->lock
);
1682 free_ird(rhp
, qhp
->attr
.max_ird
);
1684 ucontext
= ib_qp
->uobject
?
1685 to_c4iw_ucontext(ib_qp
->uobject
->context
) : NULL
;
1686 destroy_qp(&rhp
->rdev
, &qhp
->wq
,
1687 ucontext
? &ucontext
->uctx
: &rhp
->rdev
.uctx
);
1689 PDBG("%s ib_qp %p qpid 0x%0x\n", __func__
, ib_qp
, qhp
->wq
.sq
.qid
);
1694 struct ib_qp
*c4iw_create_qp(struct ib_pd
*pd
, struct ib_qp_init_attr
*attrs
,
1695 struct ib_udata
*udata
)
1697 struct c4iw_dev
*rhp
;
1698 struct c4iw_qp
*qhp
;
1699 struct c4iw_pd
*php
;
1700 struct c4iw_cq
*schp
;
1701 struct c4iw_cq
*rchp
;
1702 struct c4iw_create_qp_resp uresp
;
1703 unsigned int sqsize
, rqsize
;
1704 struct c4iw_ucontext
*ucontext
;
1706 struct c4iw_mm_entry
*mm1
, *mm2
, *mm3
, *mm4
, *mm5
= NULL
;
1708 PDBG("%s ib_pd %p\n", __func__
, pd
);
1710 if (attrs
->qp_type
!= IB_QPT_RC
)
1711 return ERR_PTR(-EINVAL
);
1713 php
= to_c4iw_pd(pd
);
1715 schp
= get_chp(rhp
, ((struct c4iw_cq
*)attrs
->send_cq
)->cq
.cqid
);
1716 rchp
= get_chp(rhp
, ((struct c4iw_cq
*)attrs
->recv_cq
)->cq
.cqid
);
1718 return ERR_PTR(-EINVAL
);
1720 if (attrs
->cap
.max_inline_data
> T4_MAX_SEND_INLINE
)
1721 return ERR_PTR(-EINVAL
);
1723 if (attrs
->cap
.max_recv_wr
> rhp
->rdev
.hw_queue
.t4_max_rq_size
)
1724 return ERR_PTR(-E2BIG
);
1725 rqsize
= attrs
->cap
.max_recv_wr
+ 1;
1729 if (attrs
->cap
.max_send_wr
> rhp
->rdev
.hw_queue
.t4_max_sq_size
)
1730 return ERR_PTR(-E2BIG
);
1731 sqsize
= attrs
->cap
.max_send_wr
+ 1;
1735 ucontext
= pd
->uobject
? to_c4iw_ucontext(pd
->uobject
->context
) : NULL
;
1737 qhp
= kzalloc(sizeof(*qhp
), GFP_KERNEL
);
1739 return ERR_PTR(-ENOMEM
);
1740 qhp
->wq
.sq
.size
= sqsize
;
1741 qhp
->wq
.sq
.memsize
=
1742 (sqsize
+ rhp
->rdev
.hw_queue
.t4_eq_status_entries
) *
1743 sizeof(*qhp
->wq
.sq
.queue
) + 16 * sizeof(__be64
);
1744 qhp
->wq
.sq
.flush_cidx
= -1;
1745 qhp
->wq
.rq
.size
= rqsize
;
1746 qhp
->wq
.rq
.memsize
=
1747 (rqsize
+ rhp
->rdev
.hw_queue
.t4_eq_status_entries
) *
1748 sizeof(*qhp
->wq
.rq
.queue
);
1751 qhp
->wq
.sq
.memsize
= roundup(qhp
->wq
.sq
.memsize
, PAGE_SIZE
);
1752 qhp
->wq
.rq
.memsize
= roundup(qhp
->wq
.rq
.memsize
, PAGE_SIZE
);
1755 ret
= create_qp(&rhp
->rdev
, &qhp
->wq
, &schp
->cq
, &rchp
->cq
,
1756 ucontext
? &ucontext
->uctx
: &rhp
->rdev
.uctx
);
1760 attrs
->cap
.max_recv_wr
= rqsize
- 1;
1761 attrs
->cap
.max_send_wr
= sqsize
- 1;
1762 attrs
->cap
.max_inline_data
= T4_MAX_SEND_INLINE
;
1765 qhp
->attr
.pd
= php
->pdid
;
1766 qhp
->attr
.scq
= ((struct c4iw_cq
*) attrs
->send_cq
)->cq
.cqid
;
1767 qhp
->attr
.rcq
= ((struct c4iw_cq
*) attrs
->recv_cq
)->cq
.cqid
;
1768 qhp
->attr
.sq_num_entries
= attrs
->cap
.max_send_wr
;
1769 qhp
->attr
.rq_num_entries
= attrs
->cap
.max_recv_wr
;
1770 qhp
->attr
.sq_max_sges
= attrs
->cap
.max_send_sge
;
1771 qhp
->attr
.sq_max_sges_rdma_write
= attrs
->cap
.max_send_sge
;
1772 qhp
->attr
.rq_max_sges
= attrs
->cap
.max_recv_sge
;
1773 qhp
->attr
.state
= C4IW_QP_STATE_IDLE
;
1774 qhp
->attr
.next_state
= C4IW_QP_STATE_IDLE
;
1775 qhp
->attr
.enable_rdma_read
= 1;
1776 qhp
->attr
.enable_rdma_write
= 1;
1777 qhp
->attr
.enable_bind
= 1;
1778 qhp
->attr
.max_ord
= 0;
1779 qhp
->attr
.max_ird
= 0;
1780 qhp
->sq_sig_all
= attrs
->sq_sig_type
== IB_SIGNAL_ALL_WR
;
1781 spin_lock_init(&qhp
->lock
);
1782 mutex_init(&qhp
->mutex
);
1783 init_waitqueue_head(&qhp
->wait
);
1784 atomic_set(&qhp
->refcnt
, 1);
1786 ret
= insert_handle(rhp
, &rhp
->qpidr
, qhp
, qhp
->wq
.sq
.qid
);
1791 mm1
= kmalloc(sizeof *mm1
, GFP_KERNEL
);
1796 mm2
= kmalloc(sizeof *mm2
, GFP_KERNEL
);
1801 mm3
= kmalloc(sizeof *mm3
, GFP_KERNEL
);
1806 mm4
= kmalloc(sizeof *mm4
, GFP_KERNEL
);
1811 if (t4_sq_onchip(&qhp
->wq
.sq
)) {
1812 mm5
= kmalloc(sizeof *mm5
, GFP_KERNEL
);
1817 uresp
.flags
= C4IW_QPF_ONCHIP
;
1820 uresp
.qid_mask
= rhp
->rdev
.qpmask
;
1821 uresp
.sqid
= qhp
->wq
.sq
.qid
;
1822 uresp
.sq_size
= qhp
->wq
.sq
.size
;
1823 uresp
.sq_memsize
= qhp
->wq
.sq
.memsize
;
1824 uresp
.rqid
= qhp
->wq
.rq
.qid
;
1825 uresp
.rq_size
= qhp
->wq
.rq
.size
;
1826 uresp
.rq_memsize
= qhp
->wq
.rq
.memsize
;
1827 spin_lock(&ucontext
->mmap_lock
);
1829 uresp
.ma_sync_key
= ucontext
->key
;
1830 ucontext
->key
+= PAGE_SIZE
;
1832 uresp
.ma_sync_key
= 0;
1834 uresp
.sq_key
= ucontext
->key
;
1835 ucontext
->key
+= PAGE_SIZE
;
1836 uresp
.rq_key
= ucontext
->key
;
1837 ucontext
->key
+= PAGE_SIZE
;
1838 uresp
.sq_db_gts_key
= ucontext
->key
;
1839 ucontext
->key
+= PAGE_SIZE
;
1840 uresp
.rq_db_gts_key
= ucontext
->key
;
1841 ucontext
->key
+= PAGE_SIZE
;
1842 spin_unlock(&ucontext
->mmap_lock
);
1843 ret
= ib_copy_to_udata(udata
, &uresp
, sizeof uresp
);
1846 mm1
->key
= uresp
.sq_key
;
1847 mm1
->addr
= qhp
->wq
.sq
.phys_addr
;
1848 mm1
->len
= PAGE_ALIGN(qhp
->wq
.sq
.memsize
);
1849 insert_mmap(ucontext
, mm1
);
1850 mm2
->key
= uresp
.rq_key
;
1851 mm2
->addr
= virt_to_phys(qhp
->wq
.rq
.queue
);
1852 mm2
->len
= PAGE_ALIGN(qhp
->wq
.rq
.memsize
);
1853 insert_mmap(ucontext
, mm2
);
1854 mm3
->key
= uresp
.sq_db_gts_key
;
1855 mm3
->addr
= (__force
unsigned long)qhp
->wq
.sq
.bar2_pa
;
1856 mm3
->len
= PAGE_SIZE
;
1857 insert_mmap(ucontext
, mm3
);
1858 mm4
->key
= uresp
.rq_db_gts_key
;
1859 mm4
->addr
= (__force
unsigned long)qhp
->wq
.rq
.bar2_pa
;
1860 mm4
->len
= PAGE_SIZE
;
1861 insert_mmap(ucontext
, mm4
);
1863 mm5
->key
= uresp
.ma_sync_key
;
1864 mm5
->addr
= (pci_resource_start(rhp
->rdev
.lldi
.pdev
, 0)
1865 + PCIE_MA_SYNC_A
) & PAGE_MASK
;
1866 mm5
->len
= PAGE_SIZE
;
1867 insert_mmap(ucontext
, mm5
);
1870 qhp
->ibqp
.qp_num
= qhp
->wq
.sq
.qid
;
1871 init_timer(&(qhp
->timer
));
1872 INIT_LIST_HEAD(&qhp
->db_fc_entry
);
1873 PDBG("%s sq id %u size %u memsize %zu num_entries %u "
1874 "rq id %u size %u memsize %zu num_entries %u\n", __func__
,
1875 qhp
->wq
.sq
.qid
, qhp
->wq
.sq
.size
, qhp
->wq
.sq
.memsize
,
1876 attrs
->cap
.max_send_wr
, qhp
->wq
.rq
.qid
, qhp
->wq
.rq
.size
,
1877 qhp
->wq
.rq
.memsize
, attrs
->cap
.max_recv_wr
);
1890 remove_handle(rhp
, &rhp
->qpidr
, qhp
->wq
.sq
.qid
);
1892 destroy_qp(&rhp
->rdev
, &qhp
->wq
,
1893 ucontext
? &ucontext
->uctx
: &rhp
->rdev
.uctx
);
1896 return ERR_PTR(ret
);
1899 int c4iw_ib_modify_qp(struct ib_qp
*ibqp
, struct ib_qp_attr
*attr
,
1900 int attr_mask
, struct ib_udata
*udata
)
1902 struct c4iw_dev
*rhp
;
1903 struct c4iw_qp
*qhp
;
1904 enum c4iw_qp_attr_mask mask
= 0;
1905 struct c4iw_qp_attributes attrs
;
1907 PDBG("%s ib_qp %p\n", __func__
, ibqp
);
1909 /* iwarp does not support the RTR state */
1910 if ((attr_mask
& IB_QP_STATE
) && (attr
->qp_state
== IB_QPS_RTR
))
1911 attr_mask
&= ~IB_QP_STATE
;
1913 /* Make sure we still have something left to do */
1917 memset(&attrs
, 0, sizeof attrs
);
1918 qhp
= to_c4iw_qp(ibqp
);
1921 attrs
.next_state
= c4iw_convert_state(attr
->qp_state
);
1922 attrs
.enable_rdma_read
= (attr
->qp_access_flags
&
1923 IB_ACCESS_REMOTE_READ
) ? 1 : 0;
1924 attrs
.enable_rdma_write
= (attr
->qp_access_flags
&
1925 IB_ACCESS_REMOTE_WRITE
) ? 1 : 0;
1926 attrs
.enable_bind
= (attr
->qp_access_flags
& IB_ACCESS_MW_BIND
) ? 1 : 0;
1929 mask
|= (attr_mask
& IB_QP_STATE
) ? C4IW_QP_ATTR_NEXT_STATE
: 0;
1930 mask
|= (attr_mask
& IB_QP_ACCESS_FLAGS
) ?
1931 (C4IW_QP_ATTR_ENABLE_RDMA_READ
|
1932 C4IW_QP_ATTR_ENABLE_RDMA_WRITE
|
1933 C4IW_QP_ATTR_ENABLE_RDMA_BIND
) : 0;
1936 * Use SQ_PSN and RQ_PSN to pass in IDX_INC values for
1937 * ringing the queue db when we're in DB_FULL mode.
1938 * Only allow this on T4 devices.
1940 attrs
.sq_db_inc
= attr
->sq_psn
;
1941 attrs
.rq_db_inc
= attr
->rq_psn
;
1942 mask
|= (attr_mask
& IB_QP_SQ_PSN
) ? C4IW_QP_ATTR_SQ_DB
: 0;
1943 mask
|= (attr_mask
& IB_QP_RQ_PSN
) ? C4IW_QP_ATTR_RQ_DB
: 0;
1944 if (!is_t4(to_c4iw_qp(ibqp
)->rhp
->rdev
.lldi
.adapter_type
) &&
1945 (mask
& (C4IW_QP_ATTR_SQ_DB
|C4IW_QP_ATTR_RQ_DB
)))
1948 return c4iw_modify_qp(rhp
, qhp
, mask
, &attrs
, 0);
1951 struct ib_qp
*c4iw_get_qp(struct ib_device
*dev
, int qpn
)
1953 PDBG("%s ib_dev %p qpn 0x%x\n", __func__
, dev
, qpn
);
1954 return (struct ib_qp
*)get_qhp(to_c4iw_dev(dev
), qpn
);
1957 int c4iw_ib_query_qp(struct ib_qp
*ibqp
, struct ib_qp_attr
*attr
,
1958 int attr_mask
, struct ib_qp_init_attr
*init_attr
)
1960 struct c4iw_qp
*qhp
= to_c4iw_qp(ibqp
);
1962 memset(attr
, 0, sizeof *attr
);
1963 memset(init_attr
, 0, sizeof *init_attr
);
1964 attr
->qp_state
= to_ib_qp_state(qhp
->attr
.state
);
1965 init_attr
->cap
.max_send_wr
= qhp
->attr
.sq_num_entries
;
1966 init_attr
->cap
.max_recv_wr
= qhp
->attr
.rq_num_entries
;
1967 init_attr
->cap
.max_send_sge
= qhp
->attr
.sq_max_sges
;
1968 init_attr
->cap
.max_recv_sge
= qhp
->attr
.sq_max_sges
;
1969 init_attr
->cap
.max_inline_data
= T4_MAX_SEND_INLINE
;
1970 init_attr
->sq_sig_type
= qhp
->sq_sig_all
? IB_SIGNAL_ALL_WR
: 0;