Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband: (48 commits)
  IB/srp: Clean up error path in srp_create_target_ib()
  IB/srp: Split send and recieve CQs to reduce number of interrupts
  RDMA/nes: Add support for KR device id 0x0110
  IB/uverbs: Use anon_inodes instead of private infinibandeventfs
  IB/core: Fix and clean up ib_ud_header_init()
  RDMA/cxgb3: Mark RDMA device with CXIO_ERROR_FATAL when removing
  RDMA/cxgb3: Don't allocate the SW queue for user mode CQs
  RDMA/cxgb3: Increase the max CQ depth
  RDMA/cxgb3: Doorbell overflow avoidance and recovery
  IB/core: Pack struct ib_device a little tighter
  IB/ucm: Clean whitespace errors
  IB/ucm: Increase maximum devices supported
  IB/ucm: Use stack variable 'base' in ib_ucm_add_one
  IB/ucm: Use stack variable 'devnum' in ib_ucm_add_one
  IB/umad: Clean whitespace
  IB/umad: Increase maximum devices supported
  IB/umad: Use stack variable 'base' in ib_umad_init_port
  IB/umad: Use stack variable 'devnum' in ib_umad_init_port
  IB/umad: Remove port_table[]
  IB/umad: Convert *cdev to cdev in struct ib_umad_port
  ...

1 files changed, 147 insertions, 134 deletions

diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c
index 8579f32ce38e..308d17bb5146 100644
--- a/drivers/infiniband/ulp/iser/iser_verbs.c
+++ b/drivers/infiniband/ulp/iser/iser_verbs.c
@@ -37,9 +37,8 @@
 #include "iscsi_iser.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)
+#define ISER_MAX_RX_CQ_LEN	(ISER_QP_MAX_RECV_DTOS * ISCSI_ISER_MAX_CONN)
+#define ISER_MAX_TX_CQ_LEN	(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);
@@ -67,15 +66,23 @@ static int iser_create_device_ib_res(struct iser_device *device)
 	if (IS_ERR(device->pd))
 		goto pd_err;
 
-	device->cq = ib_create_cq(device->ib_device,
+	device->rx_cq = ib_create_cq(device->ib_device,
 				  iser_cq_callback,
 				  iser_cq_event_callback,
 				  (void *)device,
-				  ISER_MAX_CQ_LEN, 0);
-	if (IS_ERR(device->cq))
-		goto cq_err;
+				  ISER_MAX_RX_CQ_LEN, 0);
+	if (IS_ERR(device->rx_cq))
+		goto rx_cq_err;
 
-	if (ib_req_notify_cq(device->cq, IB_CQ_NEXT_COMP))
+	device->tx_cq = ib_create_cq(device->ib_device,
+				  NULL, iser_cq_event_callback,
+				  (void *)device,
+				  ISER_MAX_TX_CQ_LEN, 0);
+
+	if (IS_ERR(device->tx_cq))
+		goto tx_cq_err;
+
+	if (ib_req_notify_cq(device->rx_cq, IB_CQ_NEXT_COMP))
 		goto cq_arm_err;
 
 	tasklet_init(&device->cq_tasklet,
@@ -93,8 +100,10 @@ static int iser_create_device_ib_res(struct iser_device *device)
 dma_mr_err:
 	tasklet_kill(&device->cq_tasklet);
 cq_arm_err:
-	ib_destroy_cq(device->cq);
-cq_err:
+	ib_destroy_cq(device->tx_cq);
+tx_cq_err:
+	ib_destroy_cq(device->rx_cq);
+rx_cq_err:
 	ib_dealloc_pd(device->pd);
 pd_err:
 	iser_err("failed to allocate an IB resource\n");
@@ -112,11 +121,13 @@ static void iser_free_device_ib_res(struct iser_device *device)
 	tasklet_kill(&device->cq_tasklet);
 
 	(void)ib_dereg_mr(device->mr);
-	(void)ib_destroy_cq(device->cq);
+	(void)ib_destroy_cq(device->tx_cq);
+	(void)ib_destroy_cq(device->rx_cq);
 	(void)ib_dealloc_pd(device->pd);
 
 	device->mr = NULL;
-	device->cq = NULL;
+	device->tx_cq = NULL;
+	device->rx_cq = NULL;
 	device->pd = NULL;
 }
 
@@ -129,13 +140,23 @@ static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
 {
 	struct iser_device	*device;
 	struct ib_qp_init_attr	init_attr;
-	int			ret;
+	int			ret = -ENOMEM;
 	struct ib_fmr_pool_param params;
 
 	BUG_ON(ib_conn->device == NULL);
 
 	device = ib_conn->device;
 
+	ib_conn->login_buf = kmalloc(ISER_RX_LOGIN_SIZE, GFP_KERNEL);
+	if (!ib_conn->login_buf) {
+		goto alloc_err;
+		ret = -ENOMEM;
+	}
+
+	ib_conn->login_dma = ib_dma_map_single(ib_conn->device->ib_device,
+				(void *)ib_conn->login_buf, ISER_RX_LOGIN_SIZE,
+				DMA_FROM_DEVICE);
+
 	ib_conn->page_vec = kmalloc(sizeof(struct iser_page_vec) +
 				    (sizeof(u64) * (ISCSI_ISER_SG_TABLESIZE +1)),
 				    GFP_KERNEL);
@@ -169,12 +190,12 @@ static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
 
 	init_attr.event_handler = iser_qp_event_callback;
 	init_attr.qp_context	= (void *)ib_conn;
-	init_attr.send_cq	= device->cq;
-	init_attr.recv_cq	= device->cq;
+	init_attr.send_cq	= device->tx_cq;
+	init_attr.recv_cq	= device->rx_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.cap.max_send_sge = 2;
+	init_attr.cap.max_recv_sge = 1;
 	init_attr.sq_sig_type	= IB_SIGNAL_REQ_WR;
 	init_attr.qp_type	= IB_QPT_RC;
 
@@ -192,6 +213,7 @@ qp_err:
 	(void)ib_destroy_fmr_pool(ib_conn->fmr_pool);
 fmr_pool_err:
 	kfree(ib_conn->page_vec);
+	kfree(ib_conn->login_buf);
 alloc_err:
 	iser_err("unable to alloc mem or create resource, err %d\n", ret);
 	return ret;
@@ -278,17 +300,6 @@ static void iser_device_try_release(struct iser_device *device)
 	mutex_unlock(&ig.device_list_mutex);
 }
 
-int iser_conn_state_comp(struct iser_conn *ib_conn,
-			enum iser_ib_conn_state comp)
-{
-	int ret;
-
-	spin_lock_bh(&ib_conn->lock);
-	ret = (ib_conn->state == comp);
-	spin_unlock_bh(&ib_conn->lock);
-	return ret;
-}
-
 static int iser_conn_state_comp_exch(struct iser_conn *ib_conn,
 				     enum iser_ib_conn_state comp,
 				     enum iser_ib_conn_state exch)
@@ -314,7 +325,7 @@ static void iser_conn_release(struct iser_conn *ib_conn)
 	mutex_lock(&ig.connlist_mutex);
 	list_del(&ib_conn->conn_list);
 	mutex_unlock(&ig.connlist_mutex);
-
+	iser_free_rx_descriptors(ib_conn);
 	iser_free_ib_conn_res(ib_conn);
 	ib_conn->device = NULL;
 	/* on EVENT_ADDR_ERROR there's no device yet for this conn */
@@ -442,7 +453,7 @@ static void iser_disconnected_handler(struct rdma_cm_id *cma_id)
 				   ISCSI_ERR_CONN_FAILED);
 
 	/* Complete the termination process if no posts are pending */
-	if ((atomic_read(&ib_conn->post_recv_buf_count) == 0) &&
+	if (ib_conn->post_recv_buf_count == 0 &&
 	    (atomic_read(&ib_conn->post_send_buf_count) == 0)) {
 		ib_conn->state = ISER_CONN_DOWN;
 		wake_up_interruptible(&ib_conn->wait);
@@ -489,9 +500,8 @@ void iser_conn_init(struct iser_conn *ib_conn)
 {
 	ib_conn->state = ISER_CONN_INIT;
 	init_waitqueue_head(&ib_conn->wait);
-	atomic_set(&ib_conn->post_recv_buf_count, 0);
+	ib_conn->post_recv_buf_count = 0;
 	atomic_set(&ib_conn->post_send_buf_count, 0);
-	atomic_set(&ib_conn->unexpected_pdu_count, 0);
 	atomic_set(&ib_conn->refcount, 1);
 	INIT_LIST_HEAD(&ib_conn->conn_list);
 	spin_lock_init(&ib_conn->lock);
@@ -626,136 +636,97 @@ void iser_unreg_mem(struct iser_mem_reg *reg)
 	reg->mem_h = NULL;
 }
 
-/**
- * iser_dto_to_iov - builds IOV from a dto descriptor
- */
-static void iser_dto_to_iov(struct iser_dto *dto, struct ib_sge *iov, int iov_len)
+int iser_post_recvl(struct iser_conn *ib_conn)
 {
-	int		     i;
-	struct ib_sge	     *sge;
-	struct iser_regd_buf *regd_buf;
-
-	if (dto->regd_vector_len > iov_len) {
-		iser_err("iov size %d too small for posting dto of len %d\n",
-			 iov_len, dto->regd_vector_len);
-		BUG();
-	}
+	struct ib_recv_wr rx_wr, *rx_wr_failed;
+	struct ib_sge	  sge;
+	int ib_ret;
 
-	for (i = 0; i < dto->regd_vector_len; i++) {
-		sge	    = &iov[i];
-		regd_buf  = dto->regd[i];
-
-		sge->addr   = regd_buf->reg.va;
-		sge->length = regd_buf->reg.len;
-		sge->lkey   = regd_buf->reg.lkey;
-
-		if (dto->used_sz[i] > 0)  /* Adjust size */
-			sge->length = dto->used_sz[i];
-
-		/* offset and length should not exceed the regd buf length */
-		if (sge->length + dto->offset[i] > regd_buf->reg.len) {
-			iser_err("Used len:%ld + offset:%d, exceed reg.buf.len:"
-				 "%ld in dto:0x%p [%d], va:0x%08lX\n",
-				 (unsigned long)sge->length, dto->offset[i],
-				 (unsigned long)regd_buf->reg.len, dto, i,
-				 (unsigned long)sge->addr);
-			BUG();
-		}
+	sge.addr   = ib_conn->login_dma;
+	sge.length = ISER_RX_LOGIN_SIZE;
+	sge.lkey   = ib_conn->device->mr->lkey;
 
-		sge->addr += dto->offset[i]; /* Adjust offset */
+	rx_wr.wr_id   = (unsigned long)ib_conn->login_buf;
+	rx_wr.sg_list = &sge;
+	rx_wr.num_sge = 1;
+	rx_wr.next    = NULL;
+
+	ib_conn->post_recv_buf_count++;
+	ib_ret	= ib_post_recv(ib_conn->qp, &rx_wr, &rx_wr_failed);
+	if (ib_ret) {
+		iser_err("ib_post_recv failed ret=%d\n", ib_ret);
+		ib_conn->post_recv_buf_count--;
 	}
+	return ib_ret;
 }
 
-/**
- * iser_post_recv - Posts a receive buffer.
- *
- * returns 0 on success, -1 on failure
- */
-int iser_post_recv(struct iser_desc *rx_desc)
+int iser_post_recvm(struct iser_conn *ib_conn, int count)
 {
-	int		  ib_ret, ret_val = 0;
-	struct ib_recv_wr recv_wr, *recv_wr_failed;
-	struct ib_sge	  iov[2];
-	struct iser_conn  *ib_conn;
-	struct iser_dto   *recv_dto = &rx_desc->dto;
-
-	/* Retrieve conn */
-	ib_conn = recv_dto->ib_conn;
-
-	iser_dto_to_iov(recv_dto, iov, 2);
+	struct ib_recv_wr *rx_wr, *rx_wr_failed;
+	int i, ib_ret;
+	unsigned int my_rx_head = ib_conn->rx_desc_head;
+	struct iser_rx_desc *rx_desc;
+
+	for (rx_wr = ib_conn->rx_wr, i = 0; i < count; i++, rx_wr++) {
+		rx_desc		= &ib_conn->rx_descs[my_rx_head];
+		rx_wr->wr_id	= (unsigned long)rx_desc;
+		rx_wr->sg_list	= &rx_desc->rx_sg;
+		rx_wr->num_sge	= 1;
+		rx_wr->next	= rx_wr + 1;
+		my_rx_head = (my_rx_head + 1) & (ISER_QP_MAX_RECV_DTOS - 1);
+	}
 
-	recv_wr.next	= NULL;
-	recv_wr.sg_list = iov;
-	recv_wr.num_sge = recv_dto->regd_vector_len;
-	recv_wr.wr_id	= (unsigned long)rx_desc;
+	rx_wr--;
+	rx_wr->next = NULL; /* mark end of work requests list */
 
-	atomic_inc(&ib_conn->post_recv_buf_count);
-	ib_ret	= ib_post_recv(ib_conn->qp, &recv_wr, &recv_wr_failed);
+	ib_conn->post_recv_buf_count += count;
+	ib_ret	= ib_post_recv(ib_conn->qp, ib_conn->rx_wr, &rx_wr_failed);
 	if (ib_ret) {
 		iser_err("ib_post_recv failed ret=%d\n", ib_ret);
-		atomic_dec(&ib_conn->post_recv_buf_count);
-		ret_val = -1;
-	}
-
-	return ret_val;
+		ib_conn->post_recv_buf_count -= count;
+	} else
+		ib_conn->rx_desc_head = my_rx_head;
+	return ib_ret;
 }
 
+
 /**
  * iser_start_send - Initiate a Send DTO operation
  *
  * returns 0 on success, -1 on failure
  */
-int iser_post_send(struct iser_desc *tx_desc)
+int iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc)
 {
-	int		  ib_ret, ret_val = 0;
+	int		  ib_ret;
 	struct ib_send_wr send_wr, *send_wr_failed;
-	struct ib_sge	  iov[MAX_REGD_BUF_VECTOR_LEN];
-	struct iser_conn  *ib_conn;
-	struct iser_dto   *dto = &tx_desc->dto;
 
-	ib_conn = dto->ib_conn;
-
-	iser_dto_to_iov(dto, iov, MAX_REGD_BUF_VECTOR_LEN);
+	ib_dma_sync_single_for_device(ib_conn->device->ib_device,
+		tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE);
 
 	send_wr.next	   = NULL;
 	send_wr.wr_id	   = (unsigned long)tx_desc;
-	send_wr.sg_list	   = iov;
-	send_wr.num_sge	   = dto->regd_vector_len;
+	send_wr.sg_list	   = tx_desc->tx_sg;
+	send_wr.num_sge	   = tx_desc->num_sge;
 	send_wr.opcode	   = IB_WR_SEND;
-	send_wr.send_flags = dto->notify_enable ? IB_SEND_SIGNALED : 0;
+	send_wr.send_flags = IB_SEND_SIGNALED;
 
 	atomic_inc(&ib_conn->post_send_buf_count);
 
 	ib_ret = ib_post_send(ib_conn->qp, &send_wr, &send_wr_failed);
 	if (ib_ret) {
-		iser_err("Failed to start SEND DTO, dto: 0x%p, IOV len: %d\n",
-			 dto, dto->regd_vector_len);
 		iser_err("ib_post_send failed, ret:%d\n", ib_ret);
 		atomic_dec(&ib_conn->post_send_buf_count);
-		ret_val = -1;
 	}
-
-	return ret_val;
+	return ib_ret;
 }
 
-static void iser_handle_comp_error(struct iser_desc *desc)
+static void iser_handle_comp_error(struct iser_tx_desc *desc,
+				struct iser_conn *ib_conn)
 {
-	struct iser_dto  *dto     = &desc->dto;
-	struct iser_conn *ib_conn = dto->ib_conn;
-
-	iser_dto_buffs_release(dto);
-
-	if (desc->type == ISCSI_RX) {
-		kfree(desc->data);
+	if (desc && desc->type == ISCSI_TX_DATAOUT)
 		kmem_cache_free(ig.desc_cache, desc);
-		atomic_dec(&ib_conn->post_recv_buf_count);
-	} else { /* type is TX control/command/dataout */
-		if (desc->type == ISCSI_TX_DATAOUT)
-			kmem_cache_free(ig.desc_cache, desc);
-		atomic_dec(&ib_conn->post_send_buf_count);
-	}
 
-	if (atomic_read(&ib_conn->post_recv_buf_count) == 0 &&
+	if (ib_conn->post_recv_buf_count == 0 &&
 	    atomic_read(&ib_conn->post_send_buf_count) == 0) {
 		/* getting here when the state is UP means that the conn is *
 		 * being terminated asynchronously from the iSCSI layer's   *
@@ -774,32 +745,74 @@ static void iser_handle_comp_error(struct iser_desc *desc)
 	}
 }
 
+static int iser_drain_tx_cq(struct iser_device  *device)
+{
+	struct ib_cq  *cq = device->tx_cq;
+	struct ib_wc  wc;
+	struct iser_tx_desc *tx_desc;
+	struct iser_conn *ib_conn;
+	int completed_tx = 0;
+
+	while (ib_poll_cq(cq, 1, &wc) == 1) {
+		tx_desc	= (struct iser_tx_desc *) (unsigned long) wc.wr_id;
+		ib_conn = wc.qp->qp_context;
+		if (wc.status == IB_WC_SUCCESS) {
+			if (wc.opcode == IB_WC_SEND)
+				iser_snd_completion(tx_desc, ib_conn);
+			else
+				iser_err("expected opcode %d got %d\n",
+					IB_WC_SEND, wc.opcode);
+		} else {
+			iser_err("tx id %llx status %d vend_err %x\n",
+				wc.wr_id, wc.status, wc.vendor_err);
+			atomic_dec(&ib_conn->post_send_buf_count);
+			iser_handle_comp_error(tx_desc, ib_conn);
+		}
+		completed_tx++;
+	}
+	return completed_tx;
+}
+
+
 static void iser_cq_tasklet_fn(unsigned long data)
 {
 	 struct iser_device  *device = (struct iser_device *)data;
-	 struct ib_cq	     *cq = device->cq;
+	 struct ib_cq	     *cq = device->rx_cq;
 	 struct ib_wc	     wc;
-	 struct iser_desc    *desc;
+	 struct iser_rx_desc *desc;
 	 unsigned long	     xfer_len;
+	struct iser_conn *ib_conn;
+	int completed_tx, completed_rx;
+	completed_tx = completed_rx = 0;
 
 	while (ib_poll_cq(cq, 1, &wc) == 1) {
-		desc	 = (struct iser_desc *) (unsigned long) wc.wr_id;
+		desc	 = (struct iser_rx_desc *) (unsigned long) wc.wr_id;
 		BUG_ON(desc == NULL);
-
+		ib_conn = wc.qp->qp_context;
 		if (wc.status == IB_WC_SUCCESS) {
-			if (desc->type == ISCSI_RX) {
+			if (wc.opcode == IB_WC_RECV) {
 				xfer_len = (unsigned long)wc.byte_len;
-				iser_rcv_completion(desc, xfer_len);
-			} else /* type == ISCSI_TX_CONTROL/SCSI_CMD/DOUT */
-				iser_snd_completion(desc);
+				iser_rcv_completion(desc, xfer_len, ib_conn);
+			} else
+				iser_err("expected opcode %d got %d\n",
+					IB_WC_RECV, wc.opcode);
 		} else {
-			iser_err("comp w. error op %d status %d\n",desc->type,wc.status);
-			iser_handle_comp_error(desc);
+			if (wc.status != IB_WC_WR_FLUSH_ERR)
+				iser_err("rx id %llx status %d vend_err %x\n",
+					wc.wr_id, wc.status, wc.vendor_err);
+			ib_conn->post_recv_buf_count--;
+			iser_handle_comp_error(NULL, ib_conn);
 		}
+		completed_rx++;
+		if (!(completed_rx & 63))
+			completed_tx += iser_drain_tx_cq(device);
 	}
 	/* #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);
+
+	completed_tx += iser_drain_tx_cq(device);
+	iser_dbg("got %d rx %d tx completions\n", completed_rx, completed_tx);
 }
 
 static void iser_cq_callback(struct ib_cq *cq, void *cq_context)