[openib-general] [PATCH 4/6] [RFC] iser cma and verbs interaction

Or Gerlitz ogerlitz at voltaire.com
Wed Feb 22 06:34:47 PST 2006


--- /ulp/iser-x/iser_verbs.c	2006-02-22 15:06:59.000000000 +0200
+++ /ulp/iser/iser_verbs.c	2006-02-22 13:48:55.000000000 +0200
@@ -1 +1,784 @@
+/*
+ * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
+ * Copyright (c) 2005, 2006 Cisco Systems.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *	- Redistributions of source code must retain the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer.
+ *
+ *	- Redistributions in binary form must reproduce the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer in the documentation and/or other materials
+ *	  provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: iser_verbs.c 5459 2006-02-22 11:00:48Z ogerlitz $
+ */
+#include <asm/io.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/smp_lock.h>
+#include <linux/delay.h>
+#include <linux/version.h>
 
+#include "iscsi_iser.h"
+#include "iser_socket.h"
+
+#define ISCSI_ISER_MAX_CONN	8
+#define ISER_MAX_CQ_LEN		((ISER_QP_MAX_RECV_DTOS + \
+				ISER_QP_MAX_REQ_DTOS) *   \
+                            	ISCSI_ISER_MAX_CONN)
+
+static void iser_cq_tasklet_fn(unsigned long data);
+static void iser_cq_callback(struct ib_cq *cq, void *cq_context);
+static void iser_comp_error_worker(void *data);
+static void iser_conn_release(struct iser_conn *p_iser_conn);
+
+static void iser_cq_event_callback(struct ib_event *cause, void *context)
+{
+	iser_err("got cq event %d \n", cause->event);
+}
+
+static void iser_qp_event_callback(struct ib_event *cause, void *context)
+{
+	iser_err("got qp event %d\n",cause->event);
+}
+
+/**
+ * iser_create_adaptor_ib_res - creates Protection Domain (PD), Completion
+ * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with
+ * the adapator.
+ *
+ * returns 0 on success, -1 on failure
+ */
+static int iser_create_adaptor_ib_res(struct iser_adaptor *p_iser_adaptor)
+{
+	struct ib_device	  *device = p_iser_adaptor->device;
+
+	strcpy(p_iser_adaptor->name, device->name);
+	iser_dbg("setting device name %s as adaptor name\n", device->name);
+
+	p_iser_adaptor->pd = ib_alloc_pd(device);
+	if (IS_ERR(p_iser_adaptor->pd))
+		goto pd_err;
+
+	p_iser_adaptor->cq = ib_create_cq(device,
+					 iser_cq_callback,
+					 iser_cq_event_callback,
+					 (void *)p_iser_adaptor,
+					 ISER_MAX_CQ_LEN);
+	if (IS_ERR(p_iser_adaptor->cq))
+		goto cq_err;
+
+	if (ib_req_notify_cq(p_iser_adaptor->cq, IB_CQ_NEXT_COMP))
+		goto cq_arm_err;
+
+	tasklet_init(&p_iser_adaptor->cq_tasklet,
+		     iser_cq_tasklet_fn,
+		     (unsigned long)p_iser_adaptor);
+
+	p_iser_adaptor->mr = ib_get_dma_mr(p_iser_adaptor->pd,
+					   IB_ACCESS_LOCAL_WRITE);
+	if (IS_ERR(p_iser_adaptor->mr))
+		goto dma_mr_err;
+
+	return 0;
+
+dma_mr_err:
+	tasklet_kill(&p_iser_adaptor->cq_tasklet);
+cq_arm_err:
+	ib_destroy_cq(p_iser_adaptor->cq);
+cq_err:
+	ib_dealloc_pd(p_iser_adaptor->pd);
+pd_err:
+	iser_err("failed to allocate an IB resource\n");
+	return -1;
+}
+
+/**
+ * iser_free_adaptor_ib_res - destory/dealloc/dereg the DMA MR,
+ * CQ and PD created with the device associated with the adapator.
+ *
+ * returns 0 on success, -1 on failure
+ */
+static int iser_free_adaptor_ib_res(struct iser_adaptor *p_iser_adaptor)
+{
+	BUG_ON(p_iser_adaptor->mr == NULL);
+
+	tasklet_kill(&p_iser_adaptor->cq_tasklet);
+
+	(void)ib_dereg_mr(p_iser_adaptor->mr);
+	(void)ib_destroy_cq(p_iser_adaptor->cq);
+	(void)ib_dealloc_pd(p_iser_adaptor->pd);
+
+	p_iser_adaptor->mr = NULL;
+	p_iser_adaptor->cq = NULL;
+	p_iser_adaptor->pd = NULL;
+	return 0;
+}
+
+/**
+ * iser_create_ib_conn_res - Creates FMR pool and Queue-Pair (QP)
+ *
+ * returns 0 on success, -1 on failure
+ */
+static int iser_create_ib_conn_res(struct iser_conn *p_iser_conn)
+{
+	struct iser_adaptor	*p_iser_adaptor;
+	struct ib_qp_init_attr	init_attr;
+	int			ret;
+	struct ib_fmr_pool_param params;
+
+	BUG_ON(p_iser_conn->p_adaptor == NULL);
+
+	p_iser_adaptor = p_iser_conn->p_adaptor;
+
+	params.page_shift        = PAGE_SHIFT;
+	/* when the first/last SG element are not start/end * 
+	 * page aligned, the map whould be of N+1 pages     */
+	params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1; 
+	params.pool_size	 = ISCSI_ISER_XMIT_CMDS_MAX;
+	params.dirty_watermark	 = 32;
+	params.cache		 = 0;
+	params.flush_function	 = NULL;
+	params.access		 = (IB_ACCESS_LOCAL_WRITE  |
+				    IB_ACCESS_REMOTE_WRITE |
+				    IB_ACCESS_REMOTE_READ);
+
+	p_iser_conn->fmr_pool = ib_create_fmr_pool(p_iser_adaptor->pd, &params);
+	if (IS_ERR(p_iser_conn->fmr_pool)) {
+		ret = PTR_ERR(p_iser_conn->fmr_pool);
+		goto fmr_pool_err;
+	}
+
+	memset(&init_attr, 0, sizeof init_attr);
+
+	init_attr.event_handler = iser_qp_event_callback;
+	init_attr.qp_context	= (void *)p_iser_conn;
+	init_attr.send_cq	= p_iser_adaptor->cq;
+	init_attr.recv_cq	= p_iser_adaptor->cq;
+	init_attr.cap.max_send_wr  = ISER_QP_MAX_REQ_DTOS;
+	init_attr.cap.max_recv_wr  = ISER_QP_MAX_RECV_DTOS;
+	init_attr.cap.max_send_sge = MAX_REGD_BUF_VECTOR_LEN;
+	init_attr.cap.max_recv_sge = 2;
+	init_attr.sq_sig_type	= IB_SIGNAL_REQ_WR;
+	init_attr.qp_type	= IB_QPT_RC;
+
+	ret = rdma_create_qp(p_iser_conn->cma_id, p_iser_adaptor->pd, &init_attr);
+	if (ret)
+		goto qp_err;
+
+	p_iser_conn->qp = p_iser_conn->cma_id->qp;
+	iser_err("setting conn %p cma_id %p: fmr_pool %p qp %p\n",
+		 p_iser_conn, p_iser_conn->cma_id,
+		 p_iser_conn->fmr_pool, p_iser_conn->cma_id->qp);
+	return ret;
+
+qp_err:
+	(void)ib_destroy_fmr_pool(p_iser_conn->fmr_pool);
+fmr_pool_err:
+	iser_err("unable to create fmr pool or qp for ib_conn: %d\n", ret);
+	return ret;
+}
+
+/**
+ * iser_free_ib_conn_res - Releases the FMR pool, QP and CMA ID objects
+ * returns 0 on success, -1 on failure
+ */
+static int iser_free_ib_conn_res(struct iser_conn *p_iser_conn)
+{
+	BUG_ON(p_iser_conn == NULL);
+
+	iser_err("freeing conn %p cma_id %p fmr pool %p qp %p\n",
+		 p_iser_conn, p_iser_conn->cma_id,
+		 p_iser_conn->fmr_pool, p_iser_conn->qp);
+
+	/* qp is created only once both addr & route are resolved */
+	if (p_iser_conn->fmr_pool != NULL)
+		ib_destroy_fmr_pool(p_iser_conn->fmr_pool);
+
+	if (p_iser_conn->qp != NULL)
+		rdma_destroy_qp(p_iser_conn->cma_id);
+
+	if (p_iser_conn->cma_id != NULL)
+		rdma_destroy_id(p_iser_conn->cma_id);
+	else
+		iser_bug("not supposed to be called twice\n");
+
+	p_iser_conn->fmr_pool = NULL;
+	p_iser_conn->qp	    = NULL;
+	p_iser_conn->cma_id = NULL;
+
+	return 0;
+}
+
+/**
+ * based on the resolved device node GUID see if there already allocated
+ * adaptor for this device. If there's no such, create one.
+ */
+static
+struct iser_adaptor *iser_adaptor_find_by_device(struct rdma_cm_id *cma_id)
+{
+	struct list_head      *p_list;
+	struct iser_adaptor   *p_adaptor = NULL;
+
+	mutex_lock(&ig.adaptor_list_mutex);
+
+	p_list = ig.adaptor_list.next;
+	while (p_list != &ig.adaptor_list) {
+		p_adaptor = list_entry(p_list, struct iser_adaptor, ig_list);
+		/* find if there's a match using the node GUID */
+		if (p_adaptor->device->node_guid == cma_id->device->node_guid)
+			break;
+	}
+
+	if (p_adaptor == NULL) {
+		p_adaptor = kzalloc(sizeof *p_adaptor, GFP_KERNEL);
+		if (p_adaptor == NULL)
+			goto end;
+		/* assign this device to the adaptor */
+		p_adaptor->device = cma_id->device;
+		/* init the adaptor and link it into ig adaptor list */
+		if (iser_create_adaptor_ib_res(p_adaptor)) {
+			kfree(p_adaptor);
+			p_adaptor = NULL;
+			goto end;
+		}
+		list_add(&p_adaptor->ig_list, &ig.adaptor_list);
+	}
+end:
+	BUG_ON(p_adaptor == NULL);
+	p_adaptor->refcount++;
+	mutex_unlock(&ig.adaptor_list_mutex);
+	return p_adaptor;
+}
+
+/* if there's no demand for this adaptor, release it */
+static void iser_adaptor_try_release(struct iser_adaptor *p_adaptor)
+{
+	mutex_lock(&ig.adaptor_list_mutex);
+	p_adaptor->refcount--;
+	iser_err("adaptor %p refcount %d\n",p_adaptor,p_adaptor->refcount);
+	if (!p_adaptor->refcount) {
+		iser_free_adaptor_ib_res(p_adaptor);
+		list_del(&p_adaptor->ig_list);
+		kfree(p_adaptor);
+	}
+	mutex_unlock(&ig.adaptor_list_mutex);
+}
+
+/**
+ * iser_conn_terminate - Triggers start of the disconnect procedures and wait
+ * for them to be done
+ */
+void iser_conn_terminate(struct iser_conn *ib_conn)
+{
+	int err = 0;
+
+	atomic_set(&ib_conn->state, ISER_CONN_TERMINATING);
+	err = rdma_disconnect(ib_conn->cma_id);
+	if (err)
+		iser_bug("Failed to disconnect, conn: 0x%p err %d\n",ib_conn,err);
+	wait_event_interruptible(ib_conn->wait,
+				 (atomic_read(&ib_conn->state) == ISER_CONN_DOWN));
+	iser_conn_release(ib_conn);
+}
+
+static void iser_connect_error(struct rdma_cm_id *cma_id)
+{
+	struct iser_conn *p_iser_conn;
+	p_iser_conn = (struct iser_conn *)cma_id->context;
+
+	if (atomic_read(&p_iser_conn->state) == ISER_CONN_PENDING) {
+		atomic_set(&p_iser_conn->state, ISER_CONN_DOWN);
+		wake_up_interruptible(&p_iser_conn->wait);
+	} else
+		iser_err("Unexpected evt for conn.state: %d\n",
+			 atomic_read(&p_iser_conn->state));
+}
+
+static void iser_addr_handler(struct rdma_cm_id *cma_id)
+{
+	struct iser_adaptor *p_iser_adaptor;
+	struct iser_conn    *p_iser_conn;
+	int    ret;
+
+	p_iser_adaptor = iser_adaptor_find_by_device(cma_id);
+	p_iser_conn = (struct iser_conn *)cma_id->context;
+	p_iser_conn->p_adaptor = p_iser_adaptor;
+
+	ret = rdma_resolve_route(cma_id, 1000);
+	if (ret) {
+		iser_err("resolve route failed: %d\n", ret);
+		iser_connect_error(cma_id);
+	}
+	return;
+}
+
+static void iser_route_handler(struct rdma_cm_id *cma_id)
+{
+	struct rdma_conn_param conn_param;
+	int    ret;
+
+	ret = iser_create_ib_conn_res((struct iser_conn *)cma_id->context);
+	if (ret)
+		goto failure;
+
+	iser_dbg("path.mtu is %d setting it to %d\n",
+		 cma_id->route.path_rec->mtu, IB_MTU_1024);
+
+	/* we must set the MTU to 1024 as this is what the target is assuming */
+	if (cma_id->route.path_rec->mtu > IB_MTU_1024)
+		cma_id->route.path_rec->mtu = IB_MTU_1024;
+
+	memset(&conn_param, 0, sizeof conn_param);
+	conn_param.responder_resources = 4;
+	conn_param.initiator_depth     = 1;
+	conn_param.retry_count	       = 7;
+	conn_param.rnr_retry_count     = 6;
+
+	ret = rdma_connect(cma_id, &conn_param);
+	if (ret) {
+		iser_err("failure connecting: %d\n", ret);
+		goto failure;
+	}
+
+	return;
+failure:
+	iser_connect_error(cma_id);
+}
+
+static void iser_connected_handler(struct rdma_cm_id *cma_id)
+{
+	struct iser_conn *p_iser_conn;
+
+	p_iser_conn = (struct iser_conn *)cma_id->context;
+	atomic_set(&p_iser_conn->state, ISER_CONN_UP);
+	wake_up_interruptible(&p_iser_conn->wait);
+}
+
+static void iser_disconnected_handler(struct rdma_cm_id *cma_id)
+{
+	struct iser_conn *p_iser_conn;
+
+	p_iser_conn = (struct iser_conn *)cma_id->context;
+	p_iser_conn->disc_evt_flag = 1;
+
+	/* If this event is unsolicited this means that the conn is being */
+	/* terminated asynchronously from the iSCSI layer's perspective.  */
+	if (atomic_read(&p_iser_conn->state) == ISER_CONN_PENDING) {
+		atomic_set(&p_iser_conn->state, ISER_CONN_DOWN);
+		wake_up_interruptible(&p_iser_conn->wait);
+	} else {
+		if (atomic_read(&p_iser_conn->state) == ISER_CONN_UP) {
+			atomic_set(&p_iser_conn->state, ISER_CONN_TERMINATING);
+			iscsi_iser_conn_failure(p_iser_conn->p_iscsi_conn,
+						ISCSI_ERR_CONN_FAILED);
+		}
+		/* Complete the termination process if no posts are pending */
+		if ((atomic_read(&p_iser_conn->post_recv_buf_count) == 0) &&
+		    (atomic_read(&p_iser_conn->post_send_buf_count) == 0)) {
+			atomic_set(&p_iser_conn->state, ISER_CONN_DOWN);
+			wake_up_interruptible(&p_iser_conn->wait);
+		}
+	}
+}
+
+static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
+{
+	int ret = 0;
+
+	iser_err("event %d conn %p id %p\n",event->event,cma_id->context,cma_id);
+
+	switch (event->event) {
+	case RDMA_CM_EVENT_ADDR_RESOLVED:
+		iser_addr_handler(cma_id);
+		break;
+	case RDMA_CM_EVENT_ROUTE_RESOLVED:
+		iser_route_handler(cma_id);
+		break;
+	case RDMA_CM_EVENT_ESTABLISHED:
+		iser_connected_handler(cma_id);
+		break;
+	case RDMA_CM_EVENT_ADDR_ERROR:
+	case RDMA_CM_EVENT_ROUTE_ERROR:
+	case RDMA_CM_EVENT_CONNECT_ERROR:
+	case RDMA_CM_EVENT_UNREACHABLE:
+	case RDMA_CM_EVENT_REJECTED:
+		iser_err("event: %d, error: %d\n", event->event, event->status);
+		iser_connect_error(cma_id);
+		break;
+	case RDMA_CM_EVENT_DISCONNECTED:
+		iser_disconnected_handler(cma_id);
+		break;
+	case RDMA_CM_EVENT_DEVICE_REMOVAL:
+		iser_bug("device removal is not handled yet\n");
+		break;
+	case RDMA_CM_EVENT_CONNECT_RESPONSE:
+		iser_bug("not expecting cma to deliver the REP!!!\n");
+		break;
+	case RDMA_CM_EVENT_CONNECT_REQUEST:
+	default:
+		break;
+	}
+	return ret;
+}
+
+void iser_conn_init(struct iser_conn *p_iser_conn)
+{
+	memset(p_iser_conn, 0, sizeof(struct iser_conn));
+	atomic_set(&p_iser_conn->state, ISER_CONN_INIT);
+	init_waitqueue_head(&p_iser_conn->wait);
+	atomic_set(&p_iser_conn->post_recv_buf_count, 0);
+	atomic_set(&p_iser_conn->post_send_buf_count, 0);
+	INIT_WORK(&p_iser_conn->comperror_work, iser_comp_error_worker,
+		  p_iser_conn);
+}
+
+ /**
+ * starts the process of connecting to the target
+ * sleeps untill the connection is established or rejected
+ */
+int iser_connect(struct iser_conn   *p_iser_conn,
+		 struct sockaddr_in *src_addr,
+		 struct sockaddr_in *dst_addr)
+{
+	struct sockaddr *src, *dst;
+	int err = 0;
+
+	sprintf(p_iser_conn->name,"%d.%d.%d.%d:%d",
+		NIPQUAD(dst_addr->sin_addr.s_addr), dst_addr->sin_port);
+
+	/* the adaptor is known only --after-- address resolution */
+	p_iser_conn->p_adaptor = NULL;
+
+	iser_err("connecting to: %d.%d.%d.%d, port 0x%x\n",
+		 NIPQUAD(dst_addr->sin_addr), dst_addr->sin_port);
+
+	atomic_set(&p_iser_conn->state, ISER_CONN_PENDING);
+
+	p_iser_conn->cma_id = rdma_create_id(iser_cma_handler,
+					     (void *)p_iser_conn,
+					     RDMA_PS_TCP);
+	if (IS_ERR(p_iser_conn->cma_id)) {
+		err = PTR_ERR(p_iser_conn->cma_id);
+		iser_err("rdma_create_id failed: %d\n", err);
+		goto connect_failure;
+	}
+
+	src = (struct sockaddr *)src_addr;
+	dst = (struct sockaddr *)dst_addr;
+	err = rdma_resolve_addr(p_iser_conn->cma_id, src, dst, 1000);
+	if (err) {
+		iser_err("rdma_resolve_addr failed: %d\n", err);
+		rdma_destroy_id(p_iser_conn->cma_id);
+		goto connect_failure;
+	}
+
+	wait_event_interruptible(p_iser_conn->wait,
+			 atomic_read(&p_iser_conn->state) != ISER_CONN_PENDING);
+
+	if (atomic_read(&p_iser_conn->state) != ISER_CONN_UP) {
+		iser_conn_release(p_iser_conn);
+		err = -EIO;
+		goto connect_failure;
+	}
+	return 0;
+
+connect_failure:
+	atomic_set(&p_iser_conn->state, ISER_CONN_DOWN);
+	return err;
+}
+
+
+/**
+ * Frees all conn objects and deallocs conn descriptor
+ */
+static void iser_conn_release(struct iser_conn *p_iser_conn)
+{
+	struct iser_adaptor    *p_iser_adaptor = p_iser_conn->p_adaptor;
+
+	if (atomic_read(&p_iser_conn->state) == ISER_CONN_DOWN) {
+		iser_free_ib_conn_res(p_iser_conn); /* qp/id freed only once */
+		p_iser_conn->p_adaptor = NULL;
+		/* on EVENT_ADDR_ERROR there's no adaptor yet for this conn */
+		if (p_iser_adaptor != NULL)
+			iser_adaptor_try_release(p_iser_adaptor);
+	} else
+		iser_err("conn %p state is %d doing nothing\n",
+			 p_iser_conn,atomic_read(&p_iser_conn->state));
+}
+
+
+/**
+ * iser_reg_page_vec - Register physical memory
+ *
+ * returns: 0 on success, errno code on failure
+ */
+int iser_reg_page_vec(struct iser_conn     *p_iser_conn,
+		      struct iser_page_vec *page_vec,
+		      struct iser_mem_reg  *mem_reg)
+{
+	struct ib_pool_fmr *mem;
+	u64		   io_addr;
+	u64		   *page_list;
+	int		   status;
+
+	page_list = page_vec->pages;
+	io_addr	  = page_list[0];
+
+	mem  = ib_fmr_pool_map_phys(p_iser_conn->fmr_pool,
+				    page_list,
+				    page_vec->length,
+				    &io_addr);
+
+	if (IS_ERR(mem)) {
+		status = (int)PTR_ERR(mem);
+		iser_err("ib_fmr_pool_map_phys failed: %d\n", status);
+		return status;
+	}
+
+	mem_reg->lkey  = mem->fmr->lkey;
+	mem_reg->rkey  = mem->fmr->rkey;
+	mem_reg->len   = page_vec->length * PAGE_SIZE;
+	mem_reg->va    = io_addr;
+	mem_reg->mem_h = (void *)mem;
+
+	mem_reg->va   += page_vec->offset;
+	mem_reg->len   = page_vec->data_size;
+
+	iser_dbg("PHYSICAL Mem.register, [PHYS p_array: 0x%p, sz: %d, "
+		 "entry[0]: (0x%08lx,%ld)] -> "
+		 "[lkey: 0x%08X mem_h: 0x%p va: 0x%08lX sz: %ld]\n",
+		 page_vec, page_vec->length,
+		 (unsigned long)page_vec->pages[0],
+		 (unsigned long)page_vec->data_size,
+		 (unsigned int)mem_reg->lkey, mem_reg->mem_h,
+		 (unsigned long)mem_reg->va, (unsigned long)mem_reg->len);
+	return 0;
+}
+
+/**
+ * Unregister (previosuly registered) memory.
+ */
+void iser_unreg_mem(struct iser_mem_reg *reg)
+{
+	int ret;
+
+	iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n",reg->mem_h);
+
+	ret = ib_fmr_pool_unmap((struct ib_pool_fmr *)reg->mem_h);
+	if (ret)
+		iser_err("ib_fmr_pool_unmap failed %d\n", ret);
+
+	reg->mem_h = NULL;
+}
+
+/**
+ * iser_dto_to_iov - builds IOV from a dto descriptor
+ */
+static void iser_dto_to_iov(struct iser_dto *p_dto, struct ib_sge *iov, int iov_len)
+{
+	int		     i;
+	struct ib_sge	     *sge;
+	struct iser_regd_buf *p_regd_buf;
+
+	if (p_dto->regd_vector_len > iov_len)
+		iser_bug("iov size %d too small for posting dto of len %d\n",
+			 iov_len, p_dto->regd_vector_len);
+
+	for (i = 0; i < p_dto->regd_vector_len; i++) {
+		sge	    = &iov[i];
+		p_regd_buf  = p_dto->regd[i];
+
+		sge->addr   = p_regd_buf->reg.va;
+		sge->length = p_regd_buf->reg.len;
+		sge->lkey   = p_regd_buf->reg.lkey;
+
+		if (p_dto->used_sz[i] > 0)  /* Adjust size */
+			sge->length = p_dto->used_sz[i];
+
+		/* offset and length should not exceed the regd buf length */
+		if (sge->length + p_dto->offset[i] > p_regd_buf->reg.len) {
+			iser_bug("Used len:%ld + offset:%d, exceed reg.buf.len:"
+				 "%ld in dto:0x%p [%d], va:0x%08lX\n",
+				 (unsigned long)sge->length, p_dto->offset[i],
+				 (unsigned long)p_regd_buf->reg.len, p_dto, i,
+				 (unsigned long)sge->addr);
+		}
+
+		sge->addr += p_dto->offset[i]; /* Adjust offset */
+	}
+}
+
+/**
+ * iser_post_recv - Posts a receive buffer.
+ *
+ * returns 0 on success, -1 on failure
+ */
+int iser_post_recv(struct iser_desc *p_rx_desc)
+{
+	int		  ib_ret, ret_val = 0;
+	struct ib_recv_wr recv_wr, *recv_wr_failed;
+	struct ib_sge	  iov[2];
+	struct iscsi_iser_conn  *p_iser_conn;
+	struct iser_dto         *p_recv_dto = &p_rx_desc->dto;
+
+	/* Retrieve conn */
+	p_iser_conn = p_recv_dto->p_conn;
+	if (p_iser_conn == NULL)
+		iser_bug("NULL p_conn in dto: 0x%p\n", p_recv_dto);
+
+	iser_dto_to_iov(p_recv_dto, iov, 2);
+
+	recv_wr.next	= NULL;
+	recv_wr.sg_list = iov;
+	recv_wr.num_sge = p_recv_dto->regd_vector_len;
+	recv_wr.wr_id	= (unsigned long)p_rx_desc;
+
+	atomic_inc(&p_iser_conn->ib_conn->post_recv_buf_count);
+	ib_ret	= ib_post_recv (p_iser_conn->ib_conn->qp, &recv_wr, &recv_wr_failed);
+	if (ib_ret) {
+		iser_err("ib_post_recv failed ret=%d\n", ib_ret);
+		atomic_dec(&p_iser_conn->ib_conn->post_recv_buf_count);
+		ret_val = -1;
+	}
+
+	return ret_val;
+}
+
+/**
+ * iser_start_send - Initiate a Send DTO operation
+ *
+ * returns 0 on success, -1 on failure
+ */
+int iser_post_send(struct iser_desc *p_tx_desc)
+{
+	int		  ib_ret, ret_val = 0;
+	struct ib_send_wr send_wr, *send_wr_failed;
+	struct ib_sge	  iov[MAX_REGD_BUF_VECTOR_LEN];
+	struct iscsi_iser_conn  *p_iser_conn;
+	struct iser_dto         *p_dto = &p_tx_desc->dto;
+
+	p_iser_conn = p_dto->p_conn;
+	if (p_iser_conn == NULL)
+		iser_bug("NULL p_conn in dto: 0x%p\n", p_dto);
+
+	iser_dto_to_iov(p_dto, iov, MAX_REGD_BUF_VECTOR_LEN);
+
+	send_wr.next	   = NULL;
+	send_wr.wr_id	   = (unsigned long)p_tx_desc;
+	send_wr.sg_list	   = iov;
+	send_wr.num_sge	   = p_dto->regd_vector_len;
+	send_wr.opcode	   = IB_WR_SEND;
+	send_wr.send_flags = p_dto->notify_enable ? IB_SEND_SIGNALED : 0;
+
+	atomic_inc(&p_iser_conn->ib_conn->post_send_buf_count);
+
+	ib_ret = ib_post_send(p_iser_conn->ib_conn->qp, &send_wr, &send_wr_failed);
+	if (ib_ret) {
+		iser_err("Failed to start SEND DTO, p_dto: 0x%p, IOV len: %d\n",
+			 p_dto, p_dto->regd_vector_len);
+		iser_err("ib_post_send failed, ret:%d\n", ib_ret);
+		atomic_dec(&p_iser_conn->ib_conn->post_send_buf_count);
+		ret_val = -1;
+	}
+
+	return ret_val;
+}
+
+static void iser_comp_error_worker(void *data)
+{
+	struct iser_conn *p_iser_conn = data;
+
+	if (atomic_read(&p_iser_conn->state) == ISER_CONN_UP) {
+		atomic_set(&p_iser_conn->state, ISER_CONN_TERMINATING);
+		iscsi_iser_conn_failure(p_iser_conn->p_iscsi_conn,
+					ISCSI_ERR_CONN_FAILED);
+	}
+
+	/* complete the termination process if disconnect event was delivered *
+	 * note there are no more non completed posts to the QP               */
+	if (p_iser_conn->disc_evt_flag) {
+		atomic_set(&p_iser_conn->state, ISER_CONN_DOWN);
+		wake_up_interruptible(&p_iser_conn->wait);
+	}
+}
+
+static void iser_handle_comp_error(struct iser_desc *p_desc)
+{
+	struct iser_dto  *p_dto       = &p_desc->dto;
+	struct iser_conn *p_iser_conn = p_dto->p_conn->ib_conn;
+
+	iser_dto_buffs_release(p_dto);
+
+	if (p_desc->type == ISCSI_RX) {
+		kfree(p_desc->data);
+		kmem_cache_free(ig.desc_cache, p_desc);
+		atomic_dec(&p_iser_conn->post_recv_buf_count);
+	} else { /* type is TX control/command/dataout */
+		if (p_desc->type == ISCSI_TX_DATAOUT)
+			kmem_cache_free(ig.desc_cache, p_desc);
+		atomic_dec(&p_iser_conn->post_send_buf_count);
+	}
+
+	if (atomic_read(&p_iser_conn->post_recv_buf_count) == 0 &&
+	    atomic_read(&p_iser_conn->post_send_buf_count) == 0)
+		schedule_work(&p_iser_conn->comperror_work);
+}
+
+static void iser_cq_tasklet_fn(unsigned long data)
+{
+	 struct iser_adaptor	*p_iser_adaptor = (struct iser_adaptor *)data;
+	 struct ib_cq	        *cq = p_iser_adaptor->cq;
+	 struct ib_wc	    	wc;
+	 struct iser_desc	*p_desc;
+	 unsigned long	    	xfer_len;
+
+
+	while (ib_poll_cq(cq, 1, &wc) == 1) {
+		p_desc	 = (struct iser_desc *) (unsigned long) wc.wr_id;
+
+		if (p_desc == NULL)
+			iser_bug("NULL p_desc\n");
+
+		if (wc.status == IB_WC_SUCCESS) {
+			if (p_desc->type == ISCSI_RX) {
+				xfer_len = (unsigned long)wc.byte_len;
+				iser_rcv_completion(p_desc, xfer_len);
+			} else /* type == ISCSI_TX_CONTROL/SCSI_CMD/DOUT */
+				iser_snd_completion(p_desc);
+		} else {
+			iser_err("comp w. error op %d status %d\n",p_desc->type,wc.status);
+			iser_handle_comp_error(p_desc);
+		}
+	}
+/* #warning "it is assumed here that arming CQ only once its empty would not"
+ *           "cause interrupts to be missed"  */
+	ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
+}
+
+static void iser_cq_callback(struct ib_cq *cq, void *cq_context)
+{
+	struct iser_adaptor	*p_iser_adaptor = (struct iser_adaptor *)cq_context;
+
+	tasklet_schedule(&p_iser_adaptor->cq_tasklet);
+}




More information about the general mailing list