2 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
3 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <linux/kernel.h>
34 #include <linux/module.h>
35 #include <linux/slab.h>
36 #include <linux/delay.h>
38 #include "iscsi_iser.h"
40 #define ISCSI_ISER_MAX_CONN 8
41 #define ISER_MAX_RX_CQ_LEN (ISER_QP_MAX_RECV_DTOS * ISCSI_ISER_MAX_CONN)
42 #define ISER_MAX_TX_CQ_LEN (ISER_QP_MAX_REQ_DTOS * ISCSI_ISER_MAX_CONN)
44 static void iser_cq_tasklet_fn(unsigned long data);
45 static void iser_cq_callback(struct ib_cq *cq, void *cq_context);
47 static void iser_cq_event_callback(struct ib_event *cause, void *context)
49 iser_err("got cq event %d \n", cause->event);
52 static void iser_qp_event_callback(struct ib_event *cause, void *context)
54 iser_err("got qp event %d\n",cause->event);
57 static void iser_event_handler(struct ib_event_handler *handler,
58 struct ib_event *event)
60 iser_err("async event %d on device %s port %d\n", event->event,
61 event->device->name, event->element.port_num);
65 * iser_create_device_ib_res - creates Protection Domain (PD), Completion
66 * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with
69 * returns 0 on success, -1 on failure
71 static int iser_create_device_ib_res(struct iser_device *device)
74 struct iser_cq_desc *cq_desc;
76 device->cqs_used = min(ISER_MAX_CQ, device->ib_device->num_comp_vectors);
77 iser_info("using %d CQs, device %s supports %d vectors\n",
78 device->cqs_used, device->ib_device->name,
79 device->ib_device->num_comp_vectors);
81 device->cq_desc = kmalloc(sizeof(struct iser_cq_desc) * device->cqs_used,
83 if (device->cq_desc == NULL)
85 cq_desc = device->cq_desc;
87 device->pd = ib_alloc_pd(device->ib_device);
88 if (IS_ERR(device->pd))
91 for (i = 0; i < device->cqs_used; i++) {
92 cq_desc[i].device = device;
93 cq_desc[i].cq_index = i;
95 device->rx_cq[i] = ib_create_cq(device->ib_device,
97 iser_cq_event_callback,
99 ISER_MAX_RX_CQ_LEN, i);
100 if (IS_ERR(device->rx_cq[i]))
103 device->tx_cq[i] = ib_create_cq(device->ib_device,
104 NULL, iser_cq_event_callback,
106 ISER_MAX_TX_CQ_LEN, i);
108 if (IS_ERR(device->tx_cq[i]))
111 if (ib_req_notify_cq(device->rx_cq[i], IB_CQ_NEXT_COMP))
114 tasklet_init(&device->cq_tasklet[i],
116 (unsigned long)&cq_desc[i]);
119 device->mr = ib_get_dma_mr(device->pd, IB_ACCESS_LOCAL_WRITE |
120 IB_ACCESS_REMOTE_WRITE |
121 IB_ACCESS_REMOTE_READ);
122 if (IS_ERR(device->mr))
125 INIT_IB_EVENT_HANDLER(&device->event_handler, device->ib_device,
127 if (ib_register_event_handler(&device->event_handler))
133 ib_dereg_mr(device->mr);
135 for (j = 0; j < device->cqs_used; j++)
136 tasklet_kill(&device->cq_tasklet[j]);
138 for (j = 0; j < i; j++) {
139 if (device->tx_cq[j])
140 ib_destroy_cq(device->tx_cq[j]);
141 if (device->rx_cq[j])
142 ib_destroy_cq(device->rx_cq[j]);
144 ib_dealloc_pd(device->pd);
146 kfree(device->cq_desc);
148 iser_err("failed to allocate an IB resource\n");
153 * iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR,
154 * CQ and PD created with the device associated with the adapator.
156 static void iser_free_device_ib_res(struct iser_device *device)
159 BUG_ON(device->mr == NULL);
161 for (i = 0; i < device->cqs_used; i++) {
162 tasklet_kill(&device->cq_tasklet[i]);
163 (void)ib_destroy_cq(device->tx_cq[i]);
164 (void)ib_destroy_cq(device->rx_cq[i]);
165 device->tx_cq[i] = NULL;
166 device->rx_cq[i] = NULL;
169 (void)ib_unregister_event_handler(&device->event_handler);
170 (void)ib_dereg_mr(device->mr);
171 (void)ib_dealloc_pd(device->pd);
173 kfree(device->cq_desc);
180 * iser_create_ib_conn_res - Creates FMR pool and Queue-Pair (QP)
182 * returns 0 on success, -1 on failure
184 static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
186 struct iser_device *device;
187 struct ib_qp_init_attr init_attr;
188 int req_err, resp_err, ret = -ENOMEM;
189 struct ib_fmr_pool_param params;
190 int index, min_index = 0;
192 BUG_ON(ib_conn->device == NULL);
194 device = ib_conn->device;
196 ib_conn->login_buf = kmalloc(ISCSI_DEF_MAX_RECV_SEG_LEN +
197 ISER_RX_LOGIN_SIZE, GFP_KERNEL);
198 if (!ib_conn->login_buf)
201 ib_conn->login_req_buf = ib_conn->login_buf;
202 ib_conn->login_resp_buf = ib_conn->login_buf + ISCSI_DEF_MAX_RECV_SEG_LEN;
204 ib_conn->login_req_dma = ib_dma_map_single(ib_conn->device->ib_device,
205 (void *)ib_conn->login_req_buf,
206 ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE);
208 ib_conn->login_resp_dma = ib_dma_map_single(ib_conn->device->ib_device,
209 (void *)ib_conn->login_resp_buf,
210 ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE);
212 req_err = ib_dma_mapping_error(device->ib_device, ib_conn->login_req_dma);
213 resp_err = ib_dma_mapping_error(device->ib_device, ib_conn->login_resp_dma);
215 if (req_err || resp_err) {
217 ib_conn->login_req_dma = 0;
219 ib_conn->login_resp_dma = 0;
223 ib_conn->page_vec = kmalloc(sizeof(struct iser_page_vec) +
224 (sizeof(u64) * (ISCSI_ISER_SG_TABLESIZE +1)),
226 if (!ib_conn->page_vec)
229 ib_conn->page_vec->pages = (u64 *) (ib_conn->page_vec + 1);
231 params.page_shift = SHIFT_4K;
232 /* when the first/last SG element are not start/end *
233 * page aligned, the map whould be of N+1 pages */
234 params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1;
235 /* make the pool size twice the max number of SCSI commands *
236 * the ML is expected to queue, watermark for unmap at 50% */
237 params.pool_size = ISCSI_DEF_XMIT_CMDS_MAX * 2;
238 params.dirty_watermark = ISCSI_DEF_XMIT_CMDS_MAX;
240 params.flush_function = NULL;
241 params.access = (IB_ACCESS_LOCAL_WRITE |
242 IB_ACCESS_REMOTE_WRITE |
243 IB_ACCESS_REMOTE_READ);
245 ib_conn->fmr_pool = ib_create_fmr_pool(device->pd, ¶ms);
246 ret = PTR_ERR(ib_conn->fmr_pool);
247 if (IS_ERR(ib_conn->fmr_pool) && ret != -ENOSYS) {
248 ib_conn->fmr_pool = NULL;
250 } else if (ret == -ENOSYS) {
251 ib_conn->fmr_pool = NULL;
252 iser_warn("FMRs are not supported, using unaligned mode\n");
256 memset(&init_attr, 0, sizeof init_attr);
258 mutex_lock(&ig.connlist_mutex);
259 /* select the CQ with the minimal number of usages */
260 for (index = 0; index < device->cqs_used; index++)
261 if (device->cq_active_qps[index] <
262 device->cq_active_qps[min_index])
264 device->cq_active_qps[min_index]++;
265 mutex_unlock(&ig.connlist_mutex);
266 iser_info("cq index %d used for ib_conn %p\n", min_index, ib_conn);
268 init_attr.event_handler = iser_qp_event_callback;
269 init_attr.qp_context = (void *)ib_conn;
270 init_attr.send_cq = device->tx_cq[min_index];
271 init_attr.recv_cq = device->rx_cq[min_index];
272 init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS;
273 init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS;
274 init_attr.cap.max_send_sge = 2;
275 init_attr.cap.max_recv_sge = 1;
276 init_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
277 init_attr.qp_type = IB_QPT_RC;
279 ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr);
283 ib_conn->qp = ib_conn->cma_id->qp;
284 iser_info("setting conn %p cma_id %p: fmr_pool %p qp %p\n",
285 ib_conn, ib_conn->cma_id,
286 ib_conn->fmr_pool, ib_conn->cma_id->qp);
290 iser_err("unable to alloc mem or create resource, err %d\n", ret);
295 * releases the FMR pool and QP objects, returns 0 on success,
298 static int iser_free_ib_conn_res(struct iser_conn *ib_conn)
301 BUG_ON(ib_conn == NULL);
303 iser_info("freeing conn %p cma_id %p fmr pool %p qp %p\n",
304 ib_conn, ib_conn->cma_id,
305 ib_conn->fmr_pool, ib_conn->qp);
307 /* qp is created only once both addr & route are resolved */
308 if (ib_conn->fmr_pool != NULL)
309 ib_destroy_fmr_pool(ib_conn->fmr_pool);
311 if (ib_conn->qp != NULL) {
312 cq_index = ((struct iser_cq_desc *)ib_conn->qp->recv_cq->cq_context)->cq_index;
313 ib_conn->device->cq_active_qps[cq_index]--;
315 rdma_destroy_qp(ib_conn->cma_id);
318 ib_conn->fmr_pool = NULL;
320 kfree(ib_conn->page_vec);
322 if (ib_conn->login_buf) {
323 if (ib_conn->login_req_dma)
324 ib_dma_unmap_single(ib_conn->device->ib_device,
325 ib_conn->login_req_dma,
326 ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE);
327 if (ib_conn->login_resp_dma)
328 ib_dma_unmap_single(ib_conn->device->ib_device,
329 ib_conn->login_resp_dma,
330 ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE);
331 kfree(ib_conn->login_buf);
338 * based on the resolved device node GUID see if there already allocated
339 * device for this device. If there's no such, create one.
342 struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id)
344 struct iser_device *device;
346 mutex_lock(&ig.device_list_mutex);
348 list_for_each_entry(device, &ig.device_list, ig_list)
349 /* find if there's a match using the node GUID */
350 if (device->ib_device->node_guid == cma_id->device->node_guid)
353 device = kzalloc(sizeof *device, GFP_KERNEL);
357 /* assign this device to the device */
358 device->ib_device = cma_id->device;
359 /* init the device and link it into ig device list */
360 if (iser_create_device_ib_res(device)) {
365 list_add(&device->ig_list, &ig.device_list);
370 mutex_unlock(&ig.device_list_mutex);
374 /* if there's no demand for this device, release it */
375 static void iser_device_try_release(struct iser_device *device)
377 mutex_lock(&ig.device_list_mutex);
379 iser_info("device %p refcount %d\n", device, device->refcount);
380 if (!device->refcount) {
381 iser_free_device_ib_res(device);
382 list_del(&device->ig_list);
385 mutex_unlock(&ig.device_list_mutex);
388 static int iser_conn_state_comp_exch(struct iser_conn *ib_conn,
389 enum iser_ib_conn_state comp,
390 enum iser_ib_conn_state exch)
394 spin_lock_bh(&ib_conn->lock);
395 if ((ret = (ib_conn->state == comp)))
396 ib_conn->state = exch;
397 spin_unlock_bh(&ib_conn->lock);
402 * Frees all conn objects and deallocs conn descriptor
404 static void iser_conn_release(struct iser_conn *ib_conn, int can_destroy_id)
406 struct iser_device *device = ib_conn->device;
408 BUG_ON(ib_conn->state != ISER_CONN_DOWN);
410 mutex_lock(&ig.connlist_mutex);
411 list_del(&ib_conn->conn_list);
412 mutex_unlock(&ig.connlist_mutex);
413 iser_free_rx_descriptors(ib_conn);
414 iser_free_ib_conn_res(ib_conn);
415 ib_conn->device = NULL;
416 /* on EVENT_ADDR_ERROR there's no device yet for this conn */
418 iser_device_try_release(device);
419 /* if cma handler context, the caller actually destroy the id */
420 if (ib_conn->cma_id != NULL && can_destroy_id) {
421 rdma_destroy_id(ib_conn->cma_id);
422 ib_conn->cma_id = NULL;
424 iscsi_destroy_endpoint(ib_conn->ep);
427 void iser_conn_get(struct iser_conn *ib_conn)
429 atomic_inc(&ib_conn->refcount);
432 int iser_conn_put(struct iser_conn *ib_conn, int can_destroy_id)
434 if (atomic_dec_and_test(&ib_conn->refcount)) {
435 iser_conn_release(ib_conn, can_destroy_id);
442 * triggers start of the disconnect procedures and wait for them to be done
444 void iser_conn_terminate(struct iser_conn *ib_conn)
448 /* change the ib conn state only if the conn is UP, however always call
449 * rdma_disconnect since this is the only way to cause the CMA to change
450 * the QP state to ERROR
453 iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP, ISER_CONN_TERMINATING);
454 err = rdma_disconnect(ib_conn->cma_id);
456 iser_err("Failed to disconnect, conn: 0x%p err %d\n",
459 wait_event_interruptible(ib_conn->wait,
460 ib_conn->state == ISER_CONN_DOWN);
462 iser_conn_put(ib_conn, 1); /* deref ib conn deallocate */
465 static int iser_connect_error(struct rdma_cm_id *cma_id)
467 struct iser_conn *ib_conn;
468 ib_conn = (struct iser_conn *)cma_id->context;
470 ib_conn->state = ISER_CONN_DOWN;
471 wake_up_interruptible(&ib_conn->wait);
472 return iser_conn_put(ib_conn, 0); /* deref ib conn's cma id */
475 static int iser_addr_handler(struct rdma_cm_id *cma_id)
477 struct iser_device *device;
478 struct iser_conn *ib_conn;
481 device = iser_device_find_by_ib_device(cma_id);
483 iser_err("device lookup/creation failed\n");
484 return iser_connect_error(cma_id);
487 ib_conn = (struct iser_conn *)cma_id->context;
488 ib_conn->device = device;
490 ret = rdma_resolve_route(cma_id, 1000);
492 iser_err("resolve route failed: %d\n", ret);
493 return iser_connect_error(cma_id);
499 static int iser_route_handler(struct rdma_cm_id *cma_id)
501 struct rdma_conn_param conn_param;
503 struct iser_cm_hdr req_hdr;
505 ret = iser_create_ib_conn_res((struct iser_conn *)cma_id->context);
509 memset(&conn_param, 0, sizeof conn_param);
510 conn_param.responder_resources = 4;
511 conn_param.initiator_depth = 1;
512 conn_param.retry_count = 7;
513 conn_param.rnr_retry_count = 6;
515 memset(&req_hdr, 0, sizeof(req_hdr));
516 req_hdr.flags = (ISER_ZBVA_NOT_SUPPORTED |
517 ISER_SEND_W_INV_NOT_SUPPORTED);
518 conn_param.private_data = (void *)&req_hdr;
519 conn_param.private_data_len = sizeof(struct iser_cm_hdr);
521 ret = rdma_connect(cma_id, &conn_param);
523 iser_err("failure connecting: %d\n", ret);
529 return iser_connect_error(cma_id);
532 static void iser_connected_handler(struct rdma_cm_id *cma_id)
534 struct iser_conn *ib_conn;
536 ib_conn = (struct iser_conn *)cma_id->context;
537 ib_conn->state = ISER_CONN_UP;
538 wake_up_interruptible(&ib_conn->wait);
541 static int iser_disconnected_handler(struct rdma_cm_id *cma_id)
543 struct iser_conn *ib_conn;
546 ib_conn = (struct iser_conn *)cma_id->context;
548 /* getting here when the state is UP means that the conn is being *
549 * terminated asynchronously from the iSCSI layer's perspective. */
550 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP,
551 ISER_CONN_TERMINATING))
552 iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn,
553 ISCSI_ERR_CONN_FAILED);
555 /* Complete the termination process if no posts are pending */
556 if (ib_conn->post_recv_buf_count == 0 &&
557 (atomic_read(&ib_conn->post_send_buf_count) == 0)) {
558 ib_conn->state = ISER_CONN_DOWN;
559 wake_up_interruptible(&ib_conn->wait);
562 ret = iser_conn_put(ib_conn, 0); /* deref ib conn's cma id */
566 static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
570 iser_info("event %d status %d conn %p id %p\n",
571 event->event, event->status, cma_id->context, cma_id);
573 switch (event->event) {
574 case RDMA_CM_EVENT_ADDR_RESOLVED:
575 ret = iser_addr_handler(cma_id);
577 case RDMA_CM_EVENT_ROUTE_RESOLVED:
578 ret = iser_route_handler(cma_id);
580 case RDMA_CM_EVENT_ESTABLISHED:
581 iser_connected_handler(cma_id);
583 case RDMA_CM_EVENT_ADDR_ERROR:
584 case RDMA_CM_EVENT_ROUTE_ERROR:
585 case RDMA_CM_EVENT_CONNECT_ERROR:
586 case RDMA_CM_EVENT_UNREACHABLE:
587 case RDMA_CM_EVENT_REJECTED:
588 ret = iser_connect_error(cma_id);
590 case RDMA_CM_EVENT_DISCONNECTED:
591 case RDMA_CM_EVENT_DEVICE_REMOVAL:
592 case RDMA_CM_EVENT_ADDR_CHANGE:
593 ret = iser_disconnected_handler(cma_id);
596 iser_err("Unexpected RDMA CM event (%d)\n", event->event);
602 void iser_conn_init(struct iser_conn *ib_conn)
604 ib_conn->state = ISER_CONN_INIT;
605 init_waitqueue_head(&ib_conn->wait);
606 ib_conn->post_recv_buf_count = 0;
607 atomic_set(&ib_conn->post_send_buf_count, 0);
608 atomic_set(&ib_conn->refcount, 1); /* ref ib conn allocation */
609 INIT_LIST_HEAD(&ib_conn->conn_list);
610 spin_lock_init(&ib_conn->lock);
614 * starts the process of connecting to the target
615 * sleeps until the connection is established or rejected
617 int iser_connect(struct iser_conn *ib_conn,
618 struct sockaddr_in *src_addr,
619 struct sockaddr_in *dst_addr,
622 struct sockaddr *src, *dst;
625 sprintf(ib_conn->name, "%pI4:%d",
626 &dst_addr->sin_addr.s_addr, dst_addr->sin_port);
628 /* the device is known only --after-- address resolution */
629 ib_conn->device = NULL;
631 iser_info("connecting to: %pI4, port 0x%x\n",
632 &dst_addr->sin_addr, dst_addr->sin_port);
634 ib_conn->state = ISER_CONN_PENDING;
636 iser_conn_get(ib_conn); /* ref ib conn's cma id */
637 ib_conn->cma_id = rdma_create_id(iser_cma_handler,
639 RDMA_PS_TCP, IB_QPT_RC);
640 if (IS_ERR(ib_conn->cma_id)) {
641 err = PTR_ERR(ib_conn->cma_id);
642 iser_err("rdma_create_id failed: %d\n", err);
646 src = (struct sockaddr *)src_addr;
647 dst = (struct sockaddr *)dst_addr;
648 err = rdma_resolve_addr(ib_conn->cma_id, src, dst, 1000);
650 iser_err("rdma_resolve_addr failed: %d\n", err);
655 wait_event_interruptible(ib_conn->wait,
656 (ib_conn->state != ISER_CONN_PENDING));
658 if (ib_conn->state != ISER_CONN_UP) {
660 goto connect_failure;
664 mutex_lock(&ig.connlist_mutex);
665 list_add(&ib_conn->conn_list, &ig.connlist);
666 mutex_unlock(&ig.connlist_mutex);
670 ib_conn->cma_id = NULL;
672 ib_conn->state = ISER_CONN_DOWN;
673 iser_conn_put(ib_conn, 1); /* deref ib conn's cma id */
675 iser_conn_put(ib_conn, 1); /* deref ib conn deallocate */
680 * iser_reg_page_vec - Register physical memory
682 * returns: 0 on success, errno code on failure
684 int iser_reg_page_vec(struct iser_conn *ib_conn,
685 struct iser_page_vec *page_vec,
686 struct iser_mem_reg *mem_reg)
688 struct ib_pool_fmr *mem;
693 page_list = page_vec->pages;
694 io_addr = page_list[0];
696 mem = ib_fmr_pool_map_phys(ib_conn->fmr_pool,
702 status = (int)PTR_ERR(mem);
703 iser_err("ib_fmr_pool_map_phys failed: %d\n", status);
707 mem_reg->lkey = mem->fmr->lkey;
708 mem_reg->rkey = mem->fmr->rkey;
709 mem_reg->len = page_vec->length * SIZE_4K;
710 mem_reg->va = io_addr;
712 mem_reg->mem_h = (void *)mem;
714 mem_reg->va += page_vec->offset;
715 mem_reg->len = page_vec->data_size;
717 iser_dbg("PHYSICAL Mem.register, [PHYS p_array: 0x%p, sz: %d, "
718 "entry[0]: (0x%08lx,%ld)] -> "
719 "[lkey: 0x%08X mem_h: 0x%p va: 0x%08lX sz: %ld]\n",
720 page_vec, page_vec->length,
721 (unsigned long)page_vec->pages[0],
722 (unsigned long)page_vec->data_size,
723 (unsigned int)mem_reg->lkey, mem_reg->mem_h,
724 (unsigned long)mem_reg->va, (unsigned long)mem_reg->len);
729 * Unregister (previosuly registered) memory.
731 void iser_unreg_mem(struct iser_mem_reg *reg)
735 iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n",reg->mem_h);
737 ret = ib_fmr_pool_unmap((struct ib_pool_fmr *)reg->mem_h);
739 iser_err("ib_fmr_pool_unmap failed %d\n", ret);
744 int iser_post_recvl(struct iser_conn *ib_conn)
746 struct ib_recv_wr rx_wr, *rx_wr_failed;
750 sge.addr = ib_conn->login_resp_dma;
751 sge.length = ISER_RX_LOGIN_SIZE;
752 sge.lkey = ib_conn->device->mr->lkey;
754 rx_wr.wr_id = (unsigned long)ib_conn->login_resp_buf;
755 rx_wr.sg_list = &sge;
759 ib_conn->post_recv_buf_count++;
760 ib_ret = ib_post_recv(ib_conn->qp, &rx_wr, &rx_wr_failed);
762 iser_err("ib_post_recv failed ret=%d\n", ib_ret);
763 ib_conn->post_recv_buf_count--;
768 int iser_post_recvm(struct iser_conn *ib_conn, int count)
770 struct ib_recv_wr *rx_wr, *rx_wr_failed;
772 unsigned int my_rx_head = ib_conn->rx_desc_head;
773 struct iser_rx_desc *rx_desc;
775 for (rx_wr = ib_conn->rx_wr, i = 0; i < count; i++, rx_wr++) {
776 rx_desc = &ib_conn->rx_descs[my_rx_head];
777 rx_wr->wr_id = (unsigned long)rx_desc;
778 rx_wr->sg_list = &rx_desc->rx_sg;
780 rx_wr->next = rx_wr + 1;
781 my_rx_head = (my_rx_head + 1) & (ISER_QP_MAX_RECV_DTOS - 1);
785 rx_wr->next = NULL; /* mark end of work requests list */
787 ib_conn->post_recv_buf_count += count;
788 ib_ret = ib_post_recv(ib_conn->qp, ib_conn->rx_wr, &rx_wr_failed);
790 iser_err("ib_post_recv failed ret=%d\n", ib_ret);
791 ib_conn->post_recv_buf_count -= count;
793 ib_conn->rx_desc_head = my_rx_head;
799 * iser_start_send - Initiate a Send DTO operation
801 * returns 0 on success, -1 on failure
803 int iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc)
806 struct ib_send_wr send_wr, *send_wr_failed;
808 ib_dma_sync_single_for_device(ib_conn->device->ib_device,
809 tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE);
812 send_wr.wr_id = (unsigned long)tx_desc;
813 send_wr.sg_list = tx_desc->tx_sg;
814 send_wr.num_sge = tx_desc->num_sge;
815 send_wr.opcode = IB_WR_SEND;
816 send_wr.send_flags = IB_SEND_SIGNALED;
818 atomic_inc(&ib_conn->post_send_buf_count);
820 ib_ret = ib_post_send(ib_conn->qp, &send_wr, &send_wr_failed);
822 iser_err("ib_post_send failed, ret:%d\n", ib_ret);
823 atomic_dec(&ib_conn->post_send_buf_count);
828 static void iser_handle_comp_error(struct iser_tx_desc *desc,
829 struct iser_conn *ib_conn)
831 if (desc && desc->type == ISCSI_TX_DATAOUT)
832 kmem_cache_free(ig.desc_cache, desc);
834 if (ib_conn->post_recv_buf_count == 0 &&
835 atomic_read(&ib_conn->post_send_buf_count) == 0) {
836 /* getting here when the state is UP means that the conn is *
837 * being terminated asynchronously from the iSCSI layer's *
839 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP,
840 ISER_CONN_TERMINATING))
841 iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn,
842 ISCSI_ERR_CONN_FAILED);
844 /* no more non completed posts to the QP, complete the
845 * termination process w.o worrying on disconnect event */
846 ib_conn->state = ISER_CONN_DOWN;
847 wake_up_interruptible(&ib_conn->wait);
851 static int iser_drain_tx_cq(struct iser_device *device, int cq_index)
853 struct ib_cq *cq = device->tx_cq[cq_index];
855 struct iser_tx_desc *tx_desc;
856 struct iser_conn *ib_conn;
857 int completed_tx = 0;
859 while (ib_poll_cq(cq, 1, &wc) == 1) {
860 tx_desc = (struct iser_tx_desc *) (unsigned long) wc.wr_id;
861 ib_conn = wc.qp->qp_context;
862 if (wc.status == IB_WC_SUCCESS) {
863 if (wc.opcode == IB_WC_SEND)
864 iser_snd_completion(tx_desc, ib_conn);
866 iser_err("expected opcode %d got %d\n",
867 IB_WC_SEND, wc.opcode);
869 iser_err("tx id %llx status %d vend_err %x\n",
870 wc.wr_id, wc.status, wc.vendor_err);
871 atomic_dec(&ib_conn->post_send_buf_count);
872 iser_handle_comp_error(tx_desc, ib_conn);
880 static void iser_cq_tasklet_fn(unsigned long data)
882 struct iser_cq_desc *cq_desc = (struct iser_cq_desc *)data;
883 struct iser_device *device = cq_desc->device;
884 int cq_index = cq_desc->cq_index;
885 struct ib_cq *cq = device->rx_cq[cq_index];
887 struct iser_rx_desc *desc;
888 unsigned long xfer_len;
889 struct iser_conn *ib_conn;
890 int completed_tx, completed_rx;
891 completed_tx = completed_rx = 0;
893 while (ib_poll_cq(cq, 1, &wc) == 1) {
894 desc = (struct iser_rx_desc *) (unsigned long) wc.wr_id;
895 BUG_ON(desc == NULL);
896 ib_conn = wc.qp->qp_context;
897 if (wc.status == IB_WC_SUCCESS) {
898 if (wc.opcode == IB_WC_RECV) {
899 xfer_len = (unsigned long)wc.byte_len;
900 iser_rcv_completion(desc, xfer_len, ib_conn);
902 iser_err("expected opcode %d got %d\n",
903 IB_WC_RECV, wc.opcode);
905 if (wc.status != IB_WC_WR_FLUSH_ERR)
906 iser_err("rx id %llx status %d vend_err %x\n",
907 wc.wr_id, wc.status, wc.vendor_err);
908 ib_conn->post_recv_buf_count--;
909 iser_handle_comp_error(NULL, ib_conn);
912 if (!(completed_rx & 63))
913 completed_tx += iser_drain_tx_cq(device, cq_index);
915 /* #warning "it is assumed here that arming CQ only once its empty" *
916 * " would not cause interrupts to be missed" */
917 ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
919 completed_tx += iser_drain_tx_cq(device, cq_index);
920 iser_dbg("got %d rx %d tx completions\n", completed_rx, completed_tx);
923 static void iser_cq_callback(struct ib_cq *cq, void *cq_context)
925 struct iser_cq_desc *cq_desc = (struct iser_cq_desc *)cq_context;
926 struct iser_device *device = cq_desc->device;
927 int cq_index = cq_desc->cq_index;
929 tasklet_schedule(&device->cq_tasklet[cq_index]);