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"
39 static int iwch_build_rdma_send(union t3_wr *wqe, struct ib_send_wr *wr,
47 case IB_WR_SEND_WITH_IMM:
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 #if 0 /* Not currently supported */
55 case TYPE_SEND_INVALIDATE:
56 case TYPE_SEND_INVALIDATE_IMMEDIATE:
57 wqe->send.rdmaop = T3_SEND_WITH_INV;
58 wqe->send.rem_stag = cpu_to_be32(wr->wr.rdma.rkey);
60 case TYPE_SEND_SE_INVALIDATE:
61 wqe->send.rdmaop = T3_SEND_WITH_SE_INV;
62 wqe->send.rem_stag = cpu_to_be32(wr->wr.rdma.rkey);
68 if (wr->num_sge > T3_MAX_SGE)
70 wqe->send.reserved[0] = 0;
71 wqe->send.reserved[1] = 0;
72 wqe->send.reserved[2] = 0;
73 if (wr->opcode == IB_WR_SEND_WITH_IMM) {
75 wqe->send.sgl[0].stag = wr->ex.imm_data;
76 wqe->send.sgl[0].len = __constant_cpu_to_be32(0);
77 wqe->send.num_sgle = __constant_cpu_to_be32(0);
81 for (i = 0; i < wr->num_sge; i++) {
82 if ((plen + wr->sg_list[i].length) < plen) {
85 plen += wr->sg_list[i].length;
86 wqe->send.sgl[i].stag =
87 cpu_to_be32(wr->sg_list[i].lkey);
88 wqe->send.sgl[i].len =
89 cpu_to_be32(wr->sg_list[i].length);
90 wqe->send.sgl[i].to = cpu_to_be64(wr->sg_list[i].addr);
92 wqe->send.num_sgle = cpu_to_be32(wr->num_sge);
93 *flit_cnt = 4 + ((wr->num_sge) << 1);
95 wqe->send.plen = cpu_to_be32(plen);
99 static int iwch_build_rdma_write(union t3_wr *wqe, struct ib_send_wr *wr,
104 if (wr->num_sge > T3_MAX_SGE)
106 wqe->write.rdmaop = T3_RDMA_WRITE;
107 wqe->write.reserved[0] = 0;
108 wqe->write.reserved[1] = 0;
109 wqe->write.reserved[2] = 0;
110 wqe->write.stag_sink = cpu_to_be32(wr->wr.rdma.rkey);
111 wqe->write.to_sink = cpu_to_be64(wr->wr.rdma.remote_addr);
113 if (wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) {
115 wqe->write.sgl[0].stag = wr->ex.imm_data;
116 wqe->write.sgl[0].len = __constant_cpu_to_be32(0);
117 wqe->write.num_sgle = __constant_cpu_to_be32(0);
121 for (i = 0; i < wr->num_sge; i++) {
122 if ((plen + wr->sg_list[i].length) < plen) {
125 plen += wr->sg_list[i].length;
126 wqe->write.sgl[i].stag =
127 cpu_to_be32(wr->sg_list[i].lkey);
128 wqe->write.sgl[i].len =
129 cpu_to_be32(wr->sg_list[i].length);
130 wqe->write.sgl[i].to =
131 cpu_to_be64(wr->sg_list[i].addr);
133 wqe->write.num_sgle = cpu_to_be32(wr->num_sge);
134 *flit_cnt = 5 + ((wr->num_sge) << 1);
136 wqe->write.plen = cpu_to_be32(plen);
140 static int iwch_build_rdma_read(union t3_wr *wqe, struct ib_send_wr *wr,
145 wqe->read.rdmaop = T3_READ_REQ;
146 wqe->read.reserved[0] = 0;
147 wqe->read.reserved[1] = 0;
148 wqe->read.reserved[2] = 0;
149 wqe->read.rem_stag = cpu_to_be32(wr->wr.rdma.rkey);
150 wqe->read.rem_to = cpu_to_be64(wr->wr.rdma.remote_addr);
151 wqe->read.local_stag = cpu_to_be32(wr->sg_list[0].lkey);
152 wqe->read.local_len = cpu_to_be32(wr->sg_list[0].length);
153 wqe->read.local_to = cpu_to_be64(wr->sg_list[0].addr);
154 *flit_cnt = sizeof(struct t3_rdma_read_wr) >> 3;
159 * TBD: this is going to be moved to firmware. Missing pdid/qpid check for now.
161 static int iwch_sgl2pbl_map(struct iwch_dev *rhp, struct ib_sge *sg_list,
162 u32 num_sgle, u32 * pbl_addr, u8 * page_size)
167 for (i = 0; i < num_sgle; i++) {
169 mhp = get_mhp(rhp, (sg_list[i].lkey) >> 8);
171 PDBG("%s %d\n", __func__, __LINE__);
174 if (!mhp->attr.state) {
175 PDBG("%s %d\n", __func__, __LINE__);
178 if (mhp->attr.zbva) {
179 PDBG("%s %d\n", __func__, __LINE__);
183 if (sg_list[i].addr < mhp->attr.va_fbo) {
184 PDBG("%s %d\n", __func__, __LINE__);
187 if (sg_list[i].addr + ((u64) sg_list[i].length) <
189 PDBG("%s %d\n", __func__, __LINE__);
192 if (sg_list[i].addr + ((u64) sg_list[i].length) >
193 mhp->attr.va_fbo + ((u64) mhp->attr.len)) {
194 PDBG("%s %d\n", __func__, __LINE__);
197 offset = sg_list[i].addr - mhp->attr.va_fbo;
198 offset += ((u32) mhp->attr.va_fbo) %
199 (1UL << (12 + mhp->attr.page_size));
200 pbl_addr[i] = ((mhp->attr.pbl_addr -
201 rhp->rdev.rnic_info.pbl_base) >> 3) +
202 (offset >> (12 + mhp->attr.page_size));
203 page_size[i] = mhp->attr.page_size;
208 static int iwch_build_rdma_recv(struct iwch_dev *rhp, union t3_wr *wqe,
209 struct ib_recv_wr *wr)
212 if (wr->num_sge > T3_MAX_SGE)
214 wqe->recv.num_sgle = cpu_to_be32(wr->num_sge);
215 for (i = 0; i < wr->num_sge; i++) {
216 wqe->recv.sgl[i].stag = cpu_to_be32(wr->sg_list[i].lkey);
217 wqe->recv.sgl[i].len = cpu_to_be32(wr->sg_list[i].length);
218 wqe->recv.sgl[i].to = cpu_to_be64(wr->sg_list[i].addr);
220 for (; i < T3_MAX_SGE; i++) {
221 wqe->recv.sgl[i].stag = 0;
222 wqe->recv.sgl[i].len = 0;
223 wqe->recv.sgl[i].to = 0;
228 int iwch_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
229 struct ib_send_wr **bad_wr)
232 u8 uninitialized_var(t3_wr_flit_cnt);
233 enum t3_wr_opcode t3_wr_opcode = 0;
234 enum t3_wr_flags t3_wr_flags;
242 qhp = to_iwch_qp(ibqp);
243 spin_lock_irqsave(&qhp->lock, flag);
244 if (qhp->attr.state > IWCH_QP_STATE_RTS) {
245 spin_unlock_irqrestore(&qhp->lock, flag);
248 num_wrs = Q_FREECNT(qhp->wq.sq_rptr, qhp->wq.sq_wptr,
249 qhp->wq.sq_size_log2);
251 spin_unlock_irqrestore(&qhp->lock, flag);
260 idx = Q_PTR2IDX(qhp->wq.wptr, qhp->wq.size_log2);
261 wqe = (union t3_wr *) (qhp->wq.queue + idx);
263 if (wr->send_flags & IB_SEND_SOLICITED)
264 t3_wr_flags |= T3_SOLICITED_EVENT_FLAG;
265 if (wr->send_flags & IB_SEND_FENCE)
266 t3_wr_flags |= T3_READ_FENCE_FLAG;
267 if (wr->send_flags & IB_SEND_SIGNALED)
268 t3_wr_flags |= T3_COMPLETION_FLAG;
270 Q_PTR2IDX(qhp->wq.sq_wptr, qhp->wq.sq_size_log2);
271 switch (wr->opcode) {
273 case IB_WR_SEND_WITH_IMM:
274 t3_wr_opcode = T3_WR_SEND;
275 err = iwch_build_rdma_send(wqe, wr, &t3_wr_flit_cnt);
277 case IB_WR_RDMA_WRITE:
278 case IB_WR_RDMA_WRITE_WITH_IMM:
279 t3_wr_opcode = T3_WR_WRITE;
280 err = iwch_build_rdma_write(wqe, wr, &t3_wr_flit_cnt);
282 case IB_WR_RDMA_READ:
283 t3_wr_opcode = T3_WR_READ;
284 t3_wr_flags = 0; /* T3 reads are always signaled */
285 err = iwch_build_rdma_read(wqe, wr, &t3_wr_flit_cnt);
288 sqp->read_len = wqe->read.local_len;
289 if (!qhp->wq.oldest_read)
290 qhp->wq.oldest_read = sqp;
293 PDBG("%s post of type=%d TBD!\n", __func__,
301 wqe->send.wrid.id0.hi = qhp->wq.sq_wptr;
302 sqp->wr_id = wr->wr_id;
303 sqp->opcode = wr2opcode(t3_wr_opcode);
304 sqp->sq_wptr = qhp->wq.sq_wptr;
306 sqp->signaled = (wr->send_flags & IB_SEND_SIGNALED);
308 build_fw_riwrh((void *) wqe, t3_wr_opcode, t3_wr_flags,
309 Q_GENBIT(qhp->wq.wptr, qhp->wq.size_log2),
311 PDBG("%s cookie 0x%llx wq idx 0x%x swsq idx %ld opcode %d\n",
312 __func__, (unsigned long long) wr->wr_id, idx,
313 Q_PTR2IDX(qhp->wq.sq_wptr, qhp->wq.sq_size_log2),
320 spin_unlock_irqrestore(&qhp->lock, flag);
321 ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid);
325 int iwch_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
326 struct ib_recv_wr **bad_wr)
335 qhp = to_iwch_qp(ibqp);
336 spin_lock_irqsave(&qhp->lock, flag);
337 if (qhp->attr.state > IWCH_QP_STATE_RTS) {
338 spin_unlock_irqrestore(&qhp->lock, flag);
341 num_wrs = Q_FREECNT(qhp->wq.rq_rptr, qhp->wq.rq_wptr,
342 qhp->wq.rq_size_log2) - 1;
344 spin_unlock_irqrestore(&qhp->lock, flag);
348 idx = Q_PTR2IDX(qhp->wq.wptr, qhp->wq.size_log2);
349 wqe = (union t3_wr *) (qhp->wq.queue + idx);
351 err = iwch_build_rdma_recv(qhp->rhp, wqe, wr);
358 qhp->wq.rq[Q_PTR2IDX(qhp->wq.rq_wptr, qhp->wq.rq_size_log2)] =
360 build_fw_riwrh((void *) wqe, T3_WR_RCV, T3_COMPLETION_FLAG,
361 Q_GENBIT(qhp->wq.wptr, qhp->wq.size_log2),
362 0, sizeof(struct t3_receive_wr) >> 3);
363 PDBG("%s cookie 0x%llx idx 0x%x rq_wptr 0x%x rw_rptr 0x%x "
364 "wqe %p \n", __func__, (unsigned long long) wr->wr_id,
365 idx, qhp->wq.rq_wptr, qhp->wq.rq_rptr, wqe);
371 spin_unlock_irqrestore(&qhp->lock, flag);
372 ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid);
376 int iwch_bind_mw(struct ib_qp *qp,
378 struct ib_mw_bind *mw_bind)
380 struct iwch_dev *rhp;
390 enum t3_wr_flags t3_wr_flags;
394 qhp = to_iwch_qp(qp);
395 mhp = to_iwch_mw(mw);
398 spin_lock_irqsave(&qhp->lock, flag);
399 if (qhp->attr.state > IWCH_QP_STATE_RTS) {
400 spin_unlock_irqrestore(&qhp->lock, flag);
403 num_wrs = Q_FREECNT(qhp->wq.sq_rptr, qhp->wq.sq_wptr,
404 qhp->wq.sq_size_log2);
405 if ((num_wrs) <= 0) {
406 spin_unlock_irqrestore(&qhp->lock, flag);
409 idx = Q_PTR2IDX(qhp->wq.wptr, qhp->wq.size_log2);
410 PDBG("%s: idx 0x%0x, mw 0x%p, mw_bind 0x%p\n", __func__, idx,
412 wqe = (union t3_wr *) (qhp->wq.queue + idx);
415 if (mw_bind->send_flags & IB_SEND_SIGNALED)
416 t3_wr_flags = T3_COMPLETION_FLAG;
418 sgl.addr = mw_bind->addr;
419 sgl.lkey = mw_bind->mr->lkey;
420 sgl.length = mw_bind->length;
421 wqe->bind.reserved = 0;
422 wqe->bind.type = T3_VA_BASED_TO;
424 /* TBD: check perms */
425 wqe->bind.perms = iwch_ib_to_mwbind_access(mw_bind->mw_access_flags);
426 wqe->bind.mr_stag = cpu_to_be32(mw_bind->mr->lkey);
427 wqe->bind.mw_stag = cpu_to_be32(mw->rkey);
428 wqe->bind.mw_len = cpu_to_be32(mw_bind->length);
429 wqe->bind.mw_va = cpu_to_be64(mw_bind->addr);
430 err = iwch_sgl2pbl_map(rhp, &sgl, 1, &pbl_addr, &page_size);
432 spin_unlock_irqrestore(&qhp->lock, flag);
435 wqe->send.wrid.id0.hi = qhp->wq.sq_wptr;
436 sqp = qhp->wq.sq + Q_PTR2IDX(qhp->wq.sq_wptr, qhp->wq.sq_size_log2);
437 sqp->wr_id = mw_bind->wr_id;
438 sqp->opcode = T3_BIND_MW;
439 sqp->sq_wptr = qhp->wq.sq_wptr;
441 sqp->signaled = (mw_bind->send_flags & IB_SEND_SIGNALED);
442 wqe->bind.mr_pbl_addr = cpu_to_be32(pbl_addr);
443 wqe->bind.mr_pagesz = page_size;
444 wqe->flit[T3_SQ_COOKIE_FLIT] = mw_bind->wr_id;
445 build_fw_riwrh((void *)wqe, T3_WR_BIND, t3_wr_flags,
446 Q_GENBIT(qhp->wq.wptr, qhp->wq.size_log2), 0,
447 sizeof(struct t3_bind_mw_wr) >> 3);
450 spin_unlock_irqrestore(&qhp->lock, flag);
452 ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid);
457 static inline void build_term_codes(struct respQ_msg_t *rsp_msg,
458 u8 *layer_type, u8 *ecode)
460 int status = TPT_ERR_INTERNAL_ERR;
467 status = CQE_STATUS(rsp_msg->cqe);
468 opcode = CQE_OPCODE(rsp_msg->cqe);
469 rqtype = RQ_TYPE(rsp_msg->cqe);
470 send_inv = (opcode == T3_SEND_WITH_INV) ||
471 (opcode == T3_SEND_WITH_SE_INV);
472 tagged = (opcode == T3_RDMA_WRITE) ||
473 (rqtype && (opcode == T3_READ_RESP));
479 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP;
480 *ecode = RDMAP_CANT_INV_STAG;
482 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
483 *ecode = RDMAP_INV_STAG;
487 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
488 if ((opcode == T3_SEND_WITH_INV) ||
489 (opcode == T3_SEND_WITH_SE_INV))
490 *ecode = RDMAP_CANT_INV_STAG;
492 *ecode = RDMAP_STAG_NOT_ASSOC;
495 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
496 *ecode = RDMAP_STAG_NOT_ASSOC;
499 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
500 *ecode = RDMAP_ACC_VIOL;
503 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
504 *ecode = RDMAP_TO_WRAP;
508 *layer_type = LAYER_DDP|DDP_TAGGED_ERR;
509 *ecode = DDPT_BASE_BOUNDS;
511 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
512 *ecode = RDMAP_BASE_BOUNDS;
515 case TPT_ERR_INVALIDATE_SHARED_MR:
516 case TPT_ERR_INVALIDATE_MR_WITH_MW_BOUND:
517 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP;
518 *ecode = RDMAP_CANT_INV_STAG;
521 case TPT_ERR_ECC_PSTAG:
522 case TPT_ERR_INTERNAL_ERR:
523 *layer_type = LAYER_RDMAP|RDMAP_LOCAL_CATA;
526 case TPT_ERR_OUT_OF_RQE:
527 *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
528 *ecode = DDPU_INV_MSN_NOBUF;
530 case TPT_ERR_PBL_ADDR_BOUND:
531 *layer_type = LAYER_DDP|DDP_TAGGED_ERR;
532 *ecode = DDPT_BASE_BOUNDS;
535 *layer_type = LAYER_MPA|DDP_LLP;
536 *ecode = MPA_CRC_ERR;
539 *layer_type = LAYER_MPA|DDP_LLP;
540 *ecode = MPA_MARKER_ERR;
542 case TPT_ERR_PDU_LEN_ERR:
543 *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
544 *ecode = DDPU_MSG_TOOBIG;
546 case TPT_ERR_DDP_VERSION:
548 *layer_type = LAYER_DDP|DDP_TAGGED_ERR;
549 *ecode = DDPT_INV_VERS;
551 *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
552 *ecode = DDPU_INV_VERS;
555 case TPT_ERR_RDMA_VERSION:
556 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP;
557 *ecode = RDMAP_INV_VERS;
560 *layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP;
561 *ecode = RDMAP_INV_OPCODE;
563 case TPT_ERR_DDP_QUEUE_NUM:
564 *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
565 *ecode = DDPU_INV_QN;
568 case TPT_ERR_MSN_GAP:
569 case TPT_ERR_MSN_RANGE:
570 case TPT_ERR_IRD_OVERFLOW:
571 *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
572 *ecode = DDPU_INV_MSN_RANGE;
575 *layer_type = LAYER_DDP|DDP_LOCAL_CATA;
579 *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
580 *ecode = DDPU_INV_MO;
583 *layer_type = LAYER_RDMAP|DDP_LOCAL_CATA;
589 int iwch_post_zb_read(struct iwch_qp *qhp)
593 u8 flit_cnt = sizeof(struct t3_rdma_read_wr) >> 3;
595 PDBG("%s enter\n", __func__);
596 skb = alloc_skb(40, GFP_KERNEL);
598 printk(KERN_ERR "%s cannot send zb_read!!\n", __func__);
601 wqe = (union t3_wr *)skb_put(skb, sizeof(struct t3_rdma_read_wr));
602 memset(wqe, 0, sizeof(struct t3_rdma_read_wr));
603 wqe->read.rdmaop = T3_READ_REQ;
604 wqe->read.reserved[0] = 0;
605 wqe->read.reserved[1] = 0;
606 wqe->read.reserved[2] = 0;
607 wqe->read.rem_stag = cpu_to_be32(1);
608 wqe->read.rem_to = cpu_to_be64(1);
609 wqe->read.local_stag = cpu_to_be32(1);
610 wqe->read.local_len = cpu_to_be32(0);
611 wqe->read.local_to = cpu_to_be64(1);
612 wqe->send.wrh.op_seop_flags = cpu_to_be32(V_FW_RIWR_OP(T3_WR_READ));
613 wqe->send.wrh.gen_tid_len = cpu_to_be32(V_FW_RIWR_TID(qhp->ep->hwtid)|
614 V_FW_RIWR_LEN(flit_cnt));
615 skb->priority = CPL_PRIORITY_DATA;
616 return cxgb3_ofld_send(qhp->rhp->rdev.t3cdev_p, skb);
620 * This posts a TERMINATE with layer=RDMA, type=catastrophic.
622 int iwch_post_terminate(struct iwch_qp *qhp, struct respQ_msg_t *rsp_msg)
625 struct terminate_message *term;
628 PDBG("%s %d\n", __func__, __LINE__);
629 skb = alloc_skb(40, GFP_ATOMIC);
631 printk(KERN_ERR "%s cannot send TERMINATE!\n", __func__);
634 wqe = (union t3_wr *)skb_put(skb, 40);
636 wqe->send.rdmaop = T3_TERMINATE;
638 /* immediate data length */
639 wqe->send.plen = htonl(4);
641 /* immediate data starts here. */
642 term = (struct terminate_message *)wqe->send.sgl;
643 build_term_codes(rsp_msg, &term->layer_etype, &term->ecode);
644 wqe->send.wrh.op_seop_flags = cpu_to_be32(V_FW_RIWR_OP(T3_WR_SEND) |
645 V_FW_RIWR_FLAGS(T3_COMPLETION_FLAG | T3_NOTIFY_FLAG));
646 wqe->send.wrh.gen_tid_len = cpu_to_be32(V_FW_RIWR_TID(qhp->ep->hwtid));
647 skb->priority = CPL_PRIORITY_DATA;
648 return cxgb3_ofld_send(qhp->rhp->rdev.t3cdev_p, skb);
652 * Assumes qhp lock is held.
654 static void __flush_qp(struct iwch_qp *qhp, unsigned long *flag)
656 struct iwch_cq *rchp, *schp;
660 rchp = get_chp(qhp->rhp, qhp->attr.rcq);
661 schp = get_chp(qhp->rhp, qhp->attr.scq);
663 PDBG("%s qhp %p rchp %p schp %p\n", __func__, qhp, rchp, schp);
664 /* take a ref on the qhp since we must release the lock */
665 atomic_inc(&qhp->refcnt);
666 spin_unlock_irqrestore(&qhp->lock, *flag);
668 /* locking heirarchy: cq lock first, then qp lock. */
669 spin_lock_irqsave(&rchp->lock, *flag);
670 spin_lock(&qhp->lock);
671 cxio_flush_hw_cq(&rchp->cq);
672 cxio_count_rcqes(&rchp->cq, &qhp->wq, &count);
673 flushed = cxio_flush_rq(&qhp->wq, &rchp->cq, count);
674 spin_unlock(&qhp->lock);
675 spin_unlock_irqrestore(&rchp->lock, *flag);
677 (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
679 /* locking heirarchy: cq lock first, then qp lock. */
680 spin_lock_irqsave(&schp->lock, *flag);
681 spin_lock(&qhp->lock);
682 cxio_flush_hw_cq(&schp->cq);
683 cxio_count_scqes(&schp->cq, &qhp->wq, &count);
684 flushed = cxio_flush_sq(&qhp->wq, &schp->cq, count);
685 spin_unlock(&qhp->lock);
686 spin_unlock_irqrestore(&schp->lock, *flag);
688 (*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context);
691 if (atomic_dec_and_test(&qhp->refcnt))
694 spin_lock_irqsave(&qhp->lock, *flag);
697 static void flush_qp(struct iwch_qp *qhp, unsigned long *flag)
699 if (qhp->ibqp.uobject)
700 cxio_set_wq_in_error(&qhp->wq);
702 __flush_qp(qhp, flag);
707 * Return count of RECV WRs posted
709 u16 iwch_rqes_posted(struct iwch_qp *qhp)
711 union t3_wr *wqe = qhp->wq.queue;
713 while ((count+1) != 0 && fw_riwrh_opcode((struct fw_riwrh *)wqe) == T3_WR_RCV) {
717 PDBG("%s qhp %p count %u\n", __func__, qhp, count);
721 static int rdma_init(struct iwch_dev *rhp, struct iwch_qp *qhp,
722 enum iwch_qp_attr_mask mask,
723 struct iwch_qp_attributes *attrs)
725 struct t3_rdma_init_attr init_attr;
728 init_attr.tid = qhp->ep->hwtid;
729 init_attr.qpid = qhp->wq.qpid;
730 init_attr.pdid = qhp->attr.pd;
731 init_attr.scqid = qhp->attr.scq;
732 init_attr.rcqid = qhp->attr.rcq;
733 init_attr.rq_addr = qhp->wq.rq_addr;
734 init_attr.rq_size = 1 << qhp->wq.rq_size_log2;
735 init_attr.mpaattrs = uP_RI_MPA_IETF_ENABLE |
736 qhp->attr.mpa_attr.recv_marker_enabled |
737 (qhp->attr.mpa_attr.xmit_marker_enabled << 1) |
738 (qhp->attr.mpa_attr.crc_enabled << 2);
741 * XXX - The IWCM doesn't quite handle getting these
742 * attrs set before going into RTS. For now, just turn
746 init_attr.qpcaps = qhp->attr.enableRdmaRead |
747 (qhp->attr.enableRdmaWrite << 1) |
748 (qhp->attr.enableBind << 2) |
749 (qhp->attr.enable_stag0_fastreg << 3) |
750 (qhp->attr.enable_stag0_fastreg << 4);
752 init_attr.qpcaps = 0x1f;
754 init_attr.tcp_emss = qhp->ep->emss;
755 init_attr.ord = qhp->attr.max_ord;
756 init_attr.ird = qhp->attr.max_ird;
757 init_attr.qp_dma_addr = qhp->wq.dma_addr;
758 init_attr.qp_dma_size = (1UL << qhp->wq.size_log2);
759 init_attr.rqe_count = iwch_rqes_posted(qhp);
760 init_attr.flags = qhp->attr.mpa_attr.initiator ? MPA_INITIATOR : 0;
761 init_attr.flags |= capable(CAP_NET_BIND_SERVICE) ? PRIV_QP : 0;
763 init_attr.rtr_type = RTR_READ;
764 if (init_attr.ord == 0 && qhp->attr.mpa_attr.initiator)
766 if (init_attr.ird == 0 && !qhp->attr.mpa_attr.initiator)
769 init_attr.rtr_type = 0;
770 init_attr.irs = qhp->ep->rcv_seq;
771 PDBG("%s init_attr.rq_addr 0x%x init_attr.rq_size = %d "
772 "flags 0x%x qpcaps 0x%x\n", __func__,
773 init_attr.rq_addr, init_attr.rq_size,
774 init_attr.flags, init_attr.qpcaps);
775 ret = cxio_rdma_init(&rhp->rdev, &init_attr);
776 PDBG("%s ret %d\n", __func__, ret);
780 int iwch_modify_qp(struct iwch_dev *rhp, struct iwch_qp *qhp,
781 enum iwch_qp_attr_mask mask,
782 struct iwch_qp_attributes *attrs,
786 struct iwch_qp_attributes newattr = qhp->attr;
792 struct iwch_ep *ep = NULL;
794 PDBG("%s qhp %p qpid 0x%x ep %p state %d -> %d\n", __func__,
795 qhp, qhp->wq.qpid, qhp->ep, qhp->attr.state,
796 (mask & IWCH_QP_ATTR_NEXT_STATE) ? attrs->next_state : -1);
798 spin_lock_irqsave(&qhp->lock, flag);
800 /* Process attr changes if in IDLE */
801 if (mask & IWCH_QP_ATTR_VALID_MODIFY) {
802 if (qhp->attr.state != IWCH_QP_STATE_IDLE) {
806 if (mask & IWCH_QP_ATTR_ENABLE_RDMA_READ)
807 newattr.enable_rdma_read = attrs->enable_rdma_read;
808 if (mask & IWCH_QP_ATTR_ENABLE_RDMA_WRITE)
809 newattr.enable_rdma_write = attrs->enable_rdma_write;
810 if (mask & IWCH_QP_ATTR_ENABLE_RDMA_BIND)
811 newattr.enable_bind = attrs->enable_bind;
812 if (mask & IWCH_QP_ATTR_MAX_ORD) {
814 rhp->attr.max_rdma_read_qp_depth) {
818 newattr.max_ord = attrs->max_ord;
820 if (mask & IWCH_QP_ATTR_MAX_IRD) {
822 rhp->attr.max_rdma_reads_per_qp) {
826 newattr.max_ird = attrs->max_ird;
831 if (!(mask & IWCH_QP_ATTR_NEXT_STATE))
833 if (qhp->attr.state == attrs->next_state)
836 switch (qhp->attr.state) {
837 case IWCH_QP_STATE_IDLE:
838 switch (attrs->next_state) {
839 case IWCH_QP_STATE_RTS:
840 if (!(mask & IWCH_QP_ATTR_LLP_STREAM_HANDLE)) {
844 if (!(mask & IWCH_QP_ATTR_MPA_ATTR)) {
848 qhp->attr.mpa_attr = attrs->mpa_attr;
849 qhp->attr.llp_stream_handle = attrs->llp_stream_handle;
850 qhp->ep = qhp->attr.llp_stream_handle;
851 qhp->attr.state = IWCH_QP_STATE_RTS;
854 * Ref the endpoint here and deref when we
855 * disassociate the endpoint from the QP. This
856 * happens in CLOSING->IDLE transition or *->ERROR
859 get_ep(&qhp->ep->com);
860 spin_unlock_irqrestore(&qhp->lock, flag);
861 ret = rdma_init(rhp, qhp, mask, attrs);
862 spin_lock_irqsave(&qhp->lock, flag);
866 case IWCH_QP_STATE_ERROR:
867 qhp->attr.state = IWCH_QP_STATE_ERROR;
868 flush_qp(qhp, &flag);
875 case IWCH_QP_STATE_RTS:
876 switch (attrs->next_state) {
877 case IWCH_QP_STATE_CLOSING:
878 BUG_ON(atomic_read(&qhp->ep->com.kref.refcount) < 2);
879 qhp->attr.state = IWCH_QP_STATE_CLOSING;
887 case IWCH_QP_STATE_TERMINATE:
888 qhp->attr.state = IWCH_QP_STATE_TERMINATE;
889 if (qhp->ibqp.uobject)
890 cxio_set_wq_in_error(&qhp->wq);
894 case IWCH_QP_STATE_ERROR:
895 qhp->attr.state = IWCH_QP_STATE_ERROR;
909 case IWCH_QP_STATE_CLOSING:
914 switch (attrs->next_state) {
915 case IWCH_QP_STATE_IDLE:
916 flush_qp(qhp, &flag);
917 qhp->attr.state = IWCH_QP_STATE_IDLE;
918 qhp->attr.llp_stream_handle = NULL;
919 put_ep(&qhp->ep->com);
923 case IWCH_QP_STATE_ERROR:
930 case IWCH_QP_STATE_ERROR:
931 if (attrs->next_state != IWCH_QP_STATE_IDLE) {
936 if (!Q_EMPTY(qhp->wq.sq_rptr, qhp->wq.sq_wptr) ||
937 !Q_EMPTY(qhp->wq.rq_rptr, qhp->wq.rq_wptr)) {
941 qhp->attr.state = IWCH_QP_STATE_IDLE;
942 memset(&qhp->attr, 0, sizeof(qhp->attr));
944 case IWCH_QP_STATE_TERMINATE:
952 printk(KERN_ERR "%s in a bad state %d\n",
953 __func__, qhp->attr.state);
960 PDBG("%s disassociating ep %p qpid 0x%x\n", __func__, qhp->ep,
963 /* disassociate the LLP connection */
964 qhp->attr.llp_stream_handle = NULL;
967 qhp->attr.state = IWCH_QP_STATE_ERROR;
971 flush_qp(qhp, &flag);
973 spin_unlock_irqrestore(&qhp->lock, flag);
976 iwch_post_terminate(qhp, NULL);
979 * If disconnect is 1, then we need to initiate a disconnect
980 * on the EP. This can be a normal close (RTS->CLOSING) or
981 * an abnormal close (RTS/CLOSING->ERROR).
984 iwch_ep_disconnect(ep, abort, GFP_KERNEL);
989 * If free is 1, then we've disassociated the EP from the QP
990 * and we need to dereference the EP.
995 PDBG("%s exit state %d\n", __func__, qhp->attr.state);
999 static int quiesce_qp(struct iwch_qp *qhp)
1001 spin_lock_irq(&qhp->lock);
1002 iwch_quiesce_tid(qhp->ep);
1003 qhp->flags |= QP_QUIESCED;
1004 spin_unlock_irq(&qhp->lock);
1008 static int resume_qp(struct iwch_qp *qhp)
1010 spin_lock_irq(&qhp->lock);
1011 iwch_resume_tid(qhp->ep);
1012 qhp->flags &= ~QP_QUIESCED;
1013 spin_unlock_irq(&qhp->lock);
1017 int iwch_quiesce_qps(struct iwch_cq *chp)
1020 struct iwch_qp *qhp;
1022 for (i=0; i < T3_MAX_NUM_QP; i++) {
1023 qhp = get_qhp(chp->rhp, i);
1026 if ((qhp->attr.rcq == chp->cq.cqid) && !qp_quiesced(qhp)) {
1030 if ((qhp->attr.scq == chp->cq.cqid) && !qp_quiesced(qhp))
1036 int iwch_resume_qps(struct iwch_cq *chp)
1039 struct iwch_qp *qhp;
1041 for (i=0; i < T3_MAX_NUM_QP; i++) {
1042 qhp = get_qhp(chp->rhp, i);
1045 if ((qhp->attr.rcq == chp->cq.cqid) && qp_quiesced(qhp)) {
1049 if ((qhp->attr.scq == chp->cq.cqid) && qp_quiesced(qhp))