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)
73 device->pd = ib_alloc_pd(device->ib_device);
74 if (IS_ERR(device->pd))
77 device->rx_cq = ib_create_cq(device->ib_device,
79 iser_cq_event_callback,
81 ISER_MAX_RX_CQ_LEN, 0);
82 if (IS_ERR(device->rx_cq))
85 device->tx_cq = ib_create_cq(device->ib_device,
86 NULL, iser_cq_event_callback,
88 ISER_MAX_TX_CQ_LEN, 0);
90 if (IS_ERR(device->tx_cq))
93 if (ib_req_notify_cq(device->rx_cq, IB_CQ_NEXT_COMP))
96 tasklet_init(&device->cq_tasklet,
98 (unsigned long)device);
100 device->mr = ib_get_dma_mr(device->pd, IB_ACCESS_LOCAL_WRITE |
101 IB_ACCESS_REMOTE_WRITE |
102 IB_ACCESS_REMOTE_READ);
103 if (IS_ERR(device->mr))
106 INIT_IB_EVENT_HANDLER(&device->event_handler, device->ib_device,
108 if (ib_register_event_handler(&device->event_handler))
114 ib_dereg_mr(device->mr);
116 tasklet_kill(&device->cq_tasklet);
118 ib_destroy_cq(device->tx_cq);
120 ib_destroy_cq(device->rx_cq);
122 ib_dealloc_pd(device->pd);
124 iser_err("failed to allocate an IB resource\n");
129 * iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR,
130 * CQ and PD created with the device associated with the adapator.
132 static void iser_free_device_ib_res(struct iser_device *device)
134 BUG_ON(device->mr == NULL);
136 tasklet_kill(&device->cq_tasklet);
137 (void)ib_unregister_event_handler(&device->event_handler);
138 (void)ib_dereg_mr(device->mr);
139 (void)ib_destroy_cq(device->tx_cq);
140 (void)ib_destroy_cq(device->rx_cq);
141 (void)ib_dealloc_pd(device->pd);
144 device->tx_cq = NULL;
145 device->rx_cq = NULL;
150 * iser_create_ib_conn_res - Creates FMR pool and Queue-Pair (QP)
152 * returns 0 on success, -1 on failure
154 static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
156 struct iser_device *device;
157 struct ib_qp_init_attr init_attr;
158 int req_err, resp_err, ret = -ENOMEM;
159 struct ib_fmr_pool_param params;
161 BUG_ON(ib_conn->device == NULL);
163 device = ib_conn->device;
165 ib_conn->login_buf = kmalloc(ISCSI_DEF_MAX_RECV_SEG_LEN +
166 ISER_RX_LOGIN_SIZE, GFP_KERNEL);
167 if (!ib_conn->login_buf)
170 ib_conn->login_req_buf = ib_conn->login_buf;
171 ib_conn->login_resp_buf = ib_conn->login_buf + ISCSI_DEF_MAX_RECV_SEG_LEN;
173 ib_conn->login_req_dma = ib_dma_map_single(ib_conn->device->ib_device,
174 (void *)ib_conn->login_req_buf,
175 ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE);
177 ib_conn->login_resp_dma = ib_dma_map_single(ib_conn->device->ib_device,
178 (void *)ib_conn->login_resp_buf,
179 ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE);
181 req_err = ib_dma_mapping_error(device->ib_device, ib_conn->login_req_dma);
182 resp_err = ib_dma_mapping_error(device->ib_device, ib_conn->login_resp_dma);
184 if (req_err || resp_err) {
186 ib_conn->login_req_dma = 0;
188 ib_conn->login_resp_dma = 0;
192 ib_conn->page_vec = kmalloc(sizeof(struct iser_page_vec) +
193 (sizeof(u64) * (ISCSI_ISER_SG_TABLESIZE +1)),
195 if (!ib_conn->page_vec)
198 ib_conn->page_vec->pages = (u64 *) (ib_conn->page_vec + 1);
200 params.page_shift = SHIFT_4K;
201 /* when the first/last SG element are not start/end *
202 * page aligned, the map whould be of N+1 pages */
203 params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1;
204 /* make the pool size twice the max number of SCSI commands *
205 * the ML is expected to queue, watermark for unmap at 50% */
206 params.pool_size = ISCSI_DEF_XMIT_CMDS_MAX * 2;
207 params.dirty_watermark = ISCSI_DEF_XMIT_CMDS_MAX;
209 params.flush_function = NULL;
210 params.access = (IB_ACCESS_LOCAL_WRITE |
211 IB_ACCESS_REMOTE_WRITE |
212 IB_ACCESS_REMOTE_READ);
214 ib_conn->fmr_pool = ib_create_fmr_pool(device->pd, ¶ms);
215 if (IS_ERR(ib_conn->fmr_pool)) {
216 ret = PTR_ERR(ib_conn->fmr_pool);
217 ib_conn->fmr_pool = NULL;
221 memset(&init_attr, 0, sizeof init_attr);
223 init_attr.event_handler = iser_qp_event_callback;
224 init_attr.qp_context = (void *)ib_conn;
225 init_attr.send_cq = device->tx_cq;
226 init_attr.recv_cq = device->rx_cq;
227 init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS;
228 init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS;
229 init_attr.cap.max_send_sge = 2;
230 init_attr.cap.max_recv_sge = 1;
231 init_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
232 init_attr.qp_type = IB_QPT_RC;
234 ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr);
238 ib_conn->qp = ib_conn->cma_id->qp;
239 iser_err("setting conn %p cma_id %p: fmr_pool %p qp %p\n",
240 ib_conn, ib_conn->cma_id,
241 ib_conn->fmr_pool, ib_conn->cma_id->qp);
245 iser_err("unable to alloc mem or create resource, err %d\n", ret);
250 * releases the FMR pool, QP and CMA ID objects, returns 0 on success,
253 static int iser_free_ib_conn_res(struct iser_conn *ib_conn, int can_destroy_id)
255 BUG_ON(ib_conn == NULL);
257 iser_err("freeing conn %p cma_id %p fmr pool %p qp %p\n",
258 ib_conn, ib_conn->cma_id,
259 ib_conn->fmr_pool, ib_conn->qp);
261 /* qp is created only once both addr & route are resolved */
262 if (ib_conn->fmr_pool != NULL)
263 ib_destroy_fmr_pool(ib_conn->fmr_pool);
265 if (ib_conn->qp != NULL)
266 rdma_destroy_qp(ib_conn->cma_id);
268 /* if cma handler context, the caller acts s.t the cma destroy the id */
269 if (ib_conn->cma_id != NULL && can_destroy_id)
270 rdma_destroy_id(ib_conn->cma_id);
272 ib_conn->fmr_pool = NULL;
274 ib_conn->cma_id = NULL;
275 kfree(ib_conn->page_vec);
277 if (ib_conn->login_buf) {
278 if (ib_conn->login_req_dma)
279 ib_dma_unmap_single(ib_conn->device->ib_device,
280 ib_conn->login_req_dma,
281 ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE);
282 if (ib_conn->login_resp_dma)
283 ib_dma_unmap_single(ib_conn->device->ib_device,
284 ib_conn->login_resp_dma,
285 ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE);
286 kfree(ib_conn->login_buf);
293 * based on the resolved device node GUID see if there already allocated
294 * device for this device. If there's no such, create one.
297 struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id)
299 struct iser_device *device;
301 mutex_lock(&ig.device_list_mutex);
303 list_for_each_entry(device, &ig.device_list, ig_list)
304 /* find if there's a match using the node GUID */
305 if (device->ib_device->node_guid == cma_id->device->node_guid)
308 device = kzalloc(sizeof *device, GFP_KERNEL);
312 /* assign this device to the device */
313 device->ib_device = cma_id->device;
314 /* init the device and link it into ig device list */
315 if (iser_create_device_ib_res(device)) {
320 list_add(&device->ig_list, &ig.device_list);
325 mutex_unlock(&ig.device_list_mutex);
329 /* if there's no demand for this device, release it */
330 static void iser_device_try_release(struct iser_device *device)
332 mutex_lock(&ig.device_list_mutex);
334 iser_err("device %p refcount %d\n",device,device->refcount);
335 if (!device->refcount) {
336 iser_free_device_ib_res(device);
337 list_del(&device->ig_list);
340 mutex_unlock(&ig.device_list_mutex);
343 static int iser_conn_state_comp_exch(struct iser_conn *ib_conn,
344 enum iser_ib_conn_state comp,
345 enum iser_ib_conn_state exch)
349 spin_lock_bh(&ib_conn->lock);
350 if ((ret = (ib_conn->state == comp)))
351 ib_conn->state = exch;
352 spin_unlock_bh(&ib_conn->lock);
357 * Frees all conn objects and deallocs conn descriptor
359 static void iser_conn_release(struct iser_conn *ib_conn, int can_destroy_id)
361 struct iser_device *device = ib_conn->device;
363 BUG_ON(ib_conn->state != ISER_CONN_DOWN);
365 mutex_lock(&ig.connlist_mutex);
366 list_del(&ib_conn->conn_list);
367 mutex_unlock(&ig.connlist_mutex);
368 iser_free_rx_descriptors(ib_conn);
369 iser_free_ib_conn_res(ib_conn, can_destroy_id);
370 ib_conn->device = NULL;
371 /* on EVENT_ADDR_ERROR there's no device yet for this conn */
373 iser_device_try_release(device);
374 iscsi_destroy_endpoint(ib_conn->ep);
377 void iser_conn_get(struct iser_conn *ib_conn)
379 atomic_inc(&ib_conn->refcount);
382 int iser_conn_put(struct iser_conn *ib_conn, int can_destroy_id)
384 if (atomic_dec_and_test(&ib_conn->refcount)) {
385 iser_conn_release(ib_conn, can_destroy_id);
392 * triggers start of the disconnect procedures and wait for them to be done
394 void iser_conn_terminate(struct iser_conn *ib_conn)
398 /* change the ib conn state only if the conn is UP, however always call
399 * rdma_disconnect since this is the only way to cause the CMA to change
400 * the QP state to ERROR
403 iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP, ISER_CONN_TERMINATING);
404 err = rdma_disconnect(ib_conn->cma_id);
406 iser_err("Failed to disconnect, conn: 0x%p err %d\n",
409 wait_event_interruptible(ib_conn->wait,
410 ib_conn->state == ISER_CONN_DOWN);
412 iser_conn_put(ib_conn, 1); /* deref ib conn deallocate */
415 static int iser_connect_error(struct rdma_cm_id *cma_id)
417 struct iser_conn *ib_conn;
418 ib_conn = (struct iser_conn *)cma_id->context;
420 ib_conn->state = ISER_CONN_DOWN;
421 wake_up_interruptible(&ib_conn->wait);
422 return iser_conn_put(ib_conn, 0); /* deref ib conn's cma id */
425 static int iser_addr_handler(struct rdma_cm_id *cma_id)
427 struct iser_device *device;
428 struct iser_conn *ib_conn;
431 device = iser_device_find_by_ib_device(cma_id);
433 iser_err("device lookup/creation failed\n");
434 return iser_connect_error(cma_id);
437 ib_conn = (struct iser_conn *)cma_id->context;
438 ib_conn->device = device;
440 ret = rdma_resolve_route(cma_id, 1000);
442 iser_err("resolve route failed: %d\n", ret);
443 return iser_connect_error(cma_id);
449 static int iser_route_handler(struct rdma_cm_id *cma_id)
451 struct rdma_conn_param conn_param;
454 ret = iser_create_ib_conn_res((struct iser_conn *)cma_id->context);
458 memset(&conn_param, 0, sizeof conn_param);
459 conn_param.responder_resources = 4;
460 conn_param.initiator_depth = 1;
461 conn_param.retry_count = 7;
462 conn_param.rnr_retry_count = 6;
464 ret = rdma_connect(cma_id, &conn_param);
466 iser_err("failure connecting: %d\n", ret);
472 return iser_connect_error(cma_id);
475 static void iser_connected_handler(struct rdma_cm_id *cma_id)
477 struct iser_conn *ib_conn;
479 ib_conn = (struct iser_conn *)cma_id->context;
480 ib_conn->state = ISER_CONN_UP;
481 wake_up_interruptible(&ib_conn->wait);
484 static int iser_disconnected_handler(struct rdma_cm_id *cma_id)
486 struct iser_conn *ib_conn;
489 ib_conn = (struct iser_conn *)cma_id->context;
491 /* getting here when the state is UP means that the conn is being *
492 * terminated asynchronously from the iSCSI layer's perspective. */
493 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP,
494 ISER_CONN_TERMINATING))
495 iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn,
496 ISCSI_ERR_CONN_FAILED);
498 /* Complete the termination process if no posts are pending */
499 if (ib_conn->post_recv_buf_count == 0 &&
500 (atomic_read(&ib_conn->post_send_buf_count) == 0)) {
501 ib_conn->state = ISER_CONN_DOWN;
502 wake_up_interruptible(&ib_conn->wait);
505 ret = iser_conn_put(ib_conn, 0); /* deref ib conn's cma id */
509 static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
513 iser_err("event %d status %d conn %p id %p\n",
514 event->event, event->status, cma_id->context, cma_id);
516 switch (event->event) {
517 case RDMA_CM_EVENT_ADDR_RESOLVED:
518 ret = iser_addr_handler(cma_id);
520 case RDMA_CM_EVENT_ROUTE_RESOLVED:
521 ret = iser_route_handler(cma_id);
523 case RDMA_CM_EVENT_ESTABLISHED:
524 iser_connected_handler(cma_id);
526 case RDMA_CM_EVENT_ADDR_ERROR:
527 case RDMA_CM_EVENT_ROUTE_ERROR:
528 case RDMA_CM_EVENT_CONNECT_ERROR:
529 case RDMA_CM_EVENT_UNREACHABLE:
530 case RDMA_CM_EVENT_REJECTED:
531 ret = iser_connect_error(cma_id);
533 case RDMA_CM_EVENT_DISCONNECTED:
534 case RDMA_CM_EVENT_DEVICE_REMOVAL:
535 case RDMA_CM_EVENT_ADDR_CHANGE:
536 ret = iser_disconnected_handler(cma_id);
539 iser_err("Unexpected RDMA CM event (%d)\n", event->event);
545 void iser_conn_init(struct iser_conn *ib_conn)
547 ib_conn->state = ISER_CONN_INIT;
548 init_waitqueue_head(&ib_conn->wait);
549 ib_conn->post_recv_buf_count = 0;
550 atomic_set(&ib_conn->post_send_buf_count, 0);
551 atomic_set(&ib_conn->refcount, 1); /* ref ib conn allocation */
552 INIT_LIST_HEAD(&ib_conn->conn_list);
553 spin_lock_init(&ib_conn->lock);
557 * starts the process of connecting to the target
558 * sleeps until the connection is established or rejected
560 int iser_connect(struct iser_conn *ib_conn,
561 struct sockaddr_in *src_addr,
562 struct sockaddr_in *dst_addr,
565 struct sockaddr *src, *dst;
568 sprintf(ib_conn->name, "%pI4:%d",
569 &dst_addr->sin_addr.s_addr, dst_addr->sin_port);
571 /* the device is known only --after-- address resolution */
572 ib_conn->device = NULL;
574 iser_err("connecting to: %pI4, port 0x%x\n",
575 &dst_addr->sin_addr, dst_addr->sin_port);
577 ib_conn->state = ISER_CONN_PENDING;
579 iser_conn_get(ib_conn); /* ref ib conn's cma id */
580 ib_conn->cma_id = rdma_create_id(iser_cma_handler,
582 RDMA_PS_TCP, IB_QPT_RC);
583 if (IS_ERR(ib_conn->cma_id)) {
584 err = PTR_ERR(ib_conn->cma_id);
585 iser_err("rdma_create_id failed: %d\n", err);
589 src = (struct sockaddr *)src_addr;
590 dst = (struct sockaddr *)dst_addr;
591 err = rdma_resolve_addr(ib_conn->cma_id, src, dst, 1000);
593 iser_err("rdma_resolve_addr failed: %d\n", err);
598 wait_event_interruptible(ib_conn->wait,
599 (ib_conn->state != ISER_CONN_PENDING));
601 if (ib_conn->state != ISER_CONN_UP) {
603 goto connect_failure;
607 mutex_lock(&ig.connlist_mutex);
608 list_add(&ib_conn->conn_list, &ig.connlist);
609 mutex_unlock(&ig.connlist_mutex);
613 ib_conn->cma_id = NULL;
615 ib_conn->state = ISER_CONN_DOWN;
616 iser_conn_put(ib_conn, 1); /* deref ib conn's cma id */
618 iser_conn_put(ib_conn, 1); /* deref ib conn deallocate */
623 * iser_reg_page_vec - Register physical memory
625 * returns: 0 on success, errno code on failure
627 int iser_reg_page_vec(struct iser_conn *ib_conn,
628 struct iser_page_vec *page_vec,
629 struct iser_mem_reg *mem_reg)
631 struct ib_pool_fmr *mem;
636 page_list = page_vec->pages;
637 io_addr = page_list[0];
639 mem = ib_fmr_pool_map_phys(ib_conn->fmr_pool,
645 status = (int)PTR_ERR(mem);
646 iser_err("ib_fmr_pool_map_phys failed: %d\n", status);
650 mem_reg->lkey = mem->fmr->lkey;
651 mem_reg->rkey = mem->fmr->rkey;
652 mem_reg->len = page_vec->length * SIZE_4K;
653 mem_reg->va = io_addr;
655 mem_reg->mem_h = (void *)mem;
657 mem_reg->va += page_vec->offset;
658 mem_reg->len = page_vec->data_size;
660 iser_dbg("PHYSICAL Mem.register, [PHYS p_array: 0x%p, sz: %d, "
661 "entry[0]: (0x%08lx,%ld)] -> "
662 "[lkey: 0x%08X mem_h: 0x%p va: 0x%08lX sz: %ld]\n",
663 page_vec, page_vec->length,
664 (unsigned long)page_vec->pages[0],
665 (unsigned long)page_vec->data_size,
666 (unsigned int)mem_reg->lkey, mem_reg->mem_h,
667 (unsigned long)mem_reg->va, (unsigned long)mem_reg->len);
672 * Unregister (previosuly registered) memory.
674 void iser_unreg_mem(struct iser_mem_reg *reg)
678 iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n",reg->mem_h);
680 ret = ib_fmr_pool_unmap((struct ib_pool_fmr *)reg->mem_h);
682 iser_err("ib_fmr_pool_unmap failed %d\n", ret);
687 int iser_post_recvl(struct iser_conn *ib_conn)
689 struct ib_recv_wr rx_wr, *rx_wr_failed;
693 sge.addr = ib_conn->login_resp_dma;
694 sge.length = ISER_RX_LOGIN_SIZE;
695 sge.lkey = ib_conn->device->mr->lkey;
697 rx_wr.wr_id = (unsigned long)ib_conn->login_resp_buf;
698 rx_wr.sg_list = &sge;
702 ib_conn->post_recv_buf_count++;
703 ib_ret = ib_post_recv(ib_conn->qp, &rx_wr, &rx_wr_failed);
705 iser_err("ib_post_recv failed ret=%d\n", ib_ret);
706 ib_conn->post_recv_buf_count--;
711 int iser_post_recvm(struct iser_conn *ib_conn, int count)
713 struct ib_recv_wr *rx_wr, *rx_wr_failed;
715 unsigned int my_rx_head = ib_conn->rx_desc_head;
716 struct iser_rx_desc *rx_desc;
718 for (rx_wr = ib_conn->rx_wr, i = 0; i < count; i++, rx_wr++) {
719 rx_desc = &ib_conn->rx_descs[my_rx_head];
720 rx_wr->wr_id = (unsigned long)rx_desc;
721 rx_wr->sg_list = &rx_desc->rx_sg;
723 rx_wr->next = rx_wr + 1;
724 my_rx_head = (my_rx_head + 1) & (ISER_QP_MAX_RECV_DTOS - 1);
728 rx_wr->next = NULL; /* mark end of work requests list */
730 ib_conn->post_recv_buf_count += count;
731 ib_ret = ib_post_recv(ib_conn->qp, ib_conn->rx_wr, &rx_wr_failed);
733 iser_err("ib_post_recv failed ret=%d\n", ib_ret);
734 ib_conn->post_recv_buf_count -= count;
736 ib_conn->rx_desc_head = my_rx_head;
742 * iser_start_send - Initiate a Send DTO operation
744 * returns 0 on success, -1 on failure
746 int iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc)
749 struct ib_send_wr send_wr, *send_wr_failed;
751 ib_dma_sync_single_for_device(ib_conn->device->ib_device,
752 tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE);
755 send_wr.wr_id = (unsigned long)tx_desc;
756 send_wr.sg_list = tx_desc->tx_sg;
757 send_wr.num_sge = tx_desc->num_sge;
758 send_wr.opcode = IB_WR_SEND;
759 send_wr.send_flags = IB_SEND_SIGNALED;
761 atomic_inc(&ib_conn->post_send_buf_count);
763 ib_ret = ib_post_send(ib_conn->qp, &send_wr, &send_wr_failed);
765 iser_err("ib_post_send failed, ret:%d\n", ib_ret);
766 atomic_dec(&ib_conn->post_send_buf_count);
771 static void iser_handle_comp_error(struct iser_tx_desc *desc,
772 struct iser_conn *ib_conn)
774 if (desc && desc->type == ISCSI_TX_DATAOUT)
775 kmem_cache_free(ig.desc_cache, desc);
777 if (ib_conn->post_recv_buf_count == 0 &&
778 atomic_read(&ib_conn->post_send_buf_count) == 0) {
779 /* getting here when the state is UP means that the conn is *
780 * being terminated asynchronously from the iSCSI layer's *
782 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP,
783 ISER_CONN_TERMINATING))
784 iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn,
785 ISCSI_ERR_CONN_FAILED);
787 /* no more non completed posts to the QP, complete the
788 * termination process w.o worrying on disconnect event */
789 ib_conn->state = ISER_CONN_DOWN;
790 wake_up_interruptible(&ib_conn->wait);
794 static int iser_drain_tx_cq(struct iser_device *device)
796 struct ib_cq *cq = device->tx_cq;
798 struct iser_tx_desc *tx_desc;
799 struct iser_conn *ib_conn;
800 int completed_tx = 0;
802 while (ib_poll_cq(cq, 1, &wc) == 1) {
803 tx_desc = (struct iser_tx_desc *) (unsigned long) wc.wr_id;
804 ib_conn = wc.qp->qp_context;
805 if (wc.status == IB_WC_SUCCESS) {
806 if (wc.opcode == IB_WC_SEND)
807 iser_snd_completion(tx_desc, ib_conn);
809 iser_err("expected opcode %d got %d\n",
810 IB_WC_SEND, wc.opcode);
812 iser_err("tx id %llx status %d vend_err %x\n",
813 wc.wr_id, wc.status, wc.vendor_err);
814 atomic_dec(&ib_conn->post_send_buf_count);
815 iser_handle_comp_error(tx_desc, ib_conn);
823 static void iser_cq_tasklet_fn(unsigned long data)
825 struct iser_device *device = (struct iser_device *)data;
826 struct ib_cq *cq = device->rx_cq;
828 struct iser_rx_desc *desc;
829 unsigned long xfer_len;
830 struct iser_conn *ib_conn;
831 int completed_tx, completed_rx;
832 completed_tx = completed_rx = 0;
834 while (ib_poll_cq(cq, 1, &wc) == 1) {
835 desc = (struct iser_rx_desc *) (unsigned long) wc.wr_id;
836 BUG_ON(desc == NULL);
837 ib_conn = wc.qp->qp_context;
838 if (wc.status == IB_WC_SUCCESS) {
839 if (wc.opcode == IB_WC_RECV) {
840 xfer_len = (unsigned long)wc.byte_len;
841 iser_rcv_completion(desc, xfer_len, ib_conn);
843 iser_err("expected opcode %d got %d\n",
844 IB_WC_RECV, wc.opcode);
846 if (wc.status != IB_WC_WR_FLUSH_ERR)
847 iser_err("rx id %llx status %d vend_err %x\n",
848 wc.wr_id, wc.status, wc.vendor_err);
849 ib_conn->post_recv_buf_count--;
850 iser_handle_comp_error(NULL, ib_conn);
853 if (!(completed_rx & 63))
854 completed_tx += iser_drain_tx_cq(device);
856 /* #warning "it is assumed here that arming CQ only once its empty" *
857 * " would not cause interrupts to be missed" */
858 ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
860 completed_tx += iser_drain_tx_cq(device);
861 iser_dbg("got %d rx %d tx completions\n", completed_rx, completed_tx);
864 static void iser_cq_callback(struct ib_cq *cq, void *cq_context)
866 struct iser_device *device = (struct iser_device *)cq_context;
868 tasklet_schedule(&device->cq_tasklet);