From fed1db33fe85573487a4732d628ac5afdb5dc776 Mon Sep 17 00:00:00 2001 From: Christoph Lameter Date: Fri, 27 Aug 2010 08:29:38 -0500 Subject: IPoIB: Set pkt_type correctly for multicast packets (fix IGMP breakage) IGMP processing is broken because the IPOIB does not set the skb->pkt_type the right way for multicast traffic. All incoming packets are set to PACKET_HOST which means that igmp_recv() will ignore the IGMP broadcasts/multicasts. This in turn means that the IGMP timers are firing and are sending information about multicast subscriptions unnecessarily. In a large private network this can cause traffic spikes. Signed-off-by: Christoph Lameter Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/ipoib/ipoib_ib.c | 14 +++++++++++--- 1 file changed, 11 insertions(+), 3 deletions(-) (limited to 'drivers/infiniband/ulp') diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c index ec6b4fbe25e4..dfa71903d6e4 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c @@ -223,6 +223,7 @@ static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) unsigned int wr_id = wc->wr_id & ~IPOIB_OP_RECV; struct sk_buff *skb; u64 mapping[IPOIB_UD_RX_SG]; + union ib_gid *dgid; ipoib_dbg_data(priv, "recv completion: id %d, status: %d\n", wr_id, wc->status); @@ -271,6 +272,16 @@ static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) ipoib_ud_dma_unmap_rx(priv, mapping); ipoib_ud_skb_put_frags(priv, skb, wc->byte_len); + /* First byte of dgid signals multicast when 0xff */ + dgid = &((struct ib_grh *)skb->data)->dgid; + + if (!(wc->wc_flags & IB_WC_GRH) || dgid->raw[0] != 0xff) + skb->pkt_type = PACKET_HOST; + else if (memcmp(dgid, dev->broadcast + 4, sizeof(union ib_gid)) == 0) + skb->pkt_type = PACKET_BROADCAST; + else + skb->pkt_type = PACKET_MULTICAST; + skb_pull(skb, IB_GRH_BYTES); skb->protocol = ((struct ipoib_header *) skb->data)->proto; @@ -281,9 +292,6 @@ static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) dev->stats.rx_bytes += skb->len; skb->dev = dev; - /* XXX get correct PACKET_ type here */ - skb->pkt_type = PACKET_HOST; - if (test_bit(IPOIB_FLAG_CSUM, &priv->flags) && likely(wc->csum_ok)) skb->ip_summed = CHECKSUM_UNNECESSARY; -- cgit v1.2.2 From 7b4c876961ad6ddcfacd69b25fe7e13ff41fe322 Mon Sep 17 00:00:00 2001 From: Eli Cohen Date: Mon, 27 Sep 2010 17:51:11 -0700 Subject: IPoIB: Skip IBoE ports IPoIB is IP-over-Infiniband link layer. In the case of IBoE, the link layer is Ethernet and IP can work directly over Ethernet, so disable IPoIB for non-IB_LINK_LAYER_INFINIBAND ports. Signed-off-by: Eli Cohen Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/ipoib/ipoib_main.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'drivers/infiniband/ulp') diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c index b4b22576f12a..9141779fba3c 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_main.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c @@ -1362,6 +1362,8 @@ static void ipoib_add_one(struct ib_device *device) } for (p = s; p <= e; ++p) { + if (rdma_port_get_link_layer(device, p) != IB_LINK_LAYER_INFINIBAND) + continue; dev = ipoib_add_port("ib%d", device, p); if (!IS_ERR(dev)) { priv = netdev_priv(dev); -- cgit v1.2.2 From dd5e6e38b2b8bd8bf71cae800e2b613e85ef1522 Mon Sep 17 00:00:00 2001 From: Bart Van Assche Date: Mon, 30 Aug 2010 19:27:20 +0000 Subject: IB/srp: Preparation for transmit ring response allocation The transmit ring in ib_srp (srp_target.tx_ring) is currently only used for allocating requests sent by the initiator to the target. This patch prepares using that ring for allocation of both requests and responses. Also, this patch differentiates the uses of SRP_SQ_SIZE, increases the size of the IB send completion queue by one element and reserves one transmit ring slot for SRP_TSK_MGMT requests. Signed-off-by: Bart Van Assche Signed-off-by: David Dillow Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/srp/ib_srp.c | 27 +++++++++++++++------------ drivers/infiniband/ulp/srp/ib_srp.h | 13 ++++++++++--- 2 files changed, 25 insertions(+), 15 deletions(-) (limited to 'drivers/infiniband/ulp') diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c index 7f8f16bad753..b8b09a4f012a 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.c +++ b/drivers/infiniband/ulp/srp/ib_srp.c @@ -291,7 +291,7 @@ static void srp_free_target_ib(struct srp_target_port *target) for (i = 0; i < SRP_RQ_SIZE; ++i) srp_free_iu(target->srp_host, target->rx_ring[i]); - for (i = 0; i < SRP_SQ_SIZE + 1; ++i) + for (i = 0; i < SRP_SQ_SIZE; ++i) srp_free_iu(target->srp_host, target->tx_ring[i]); } @@ -822,7 +822,7 @@ static int srp_post_recv(struct srp_target_port *target) spin_lock_irqsave(target->scsi_host->host_lock, flags); - next = target->rx_head & (SRP_RQ_SIZE - 1); + next = target->rx_head & SRP_RQ_MASK; wr.wr_id = next; iu = target->rx_ring[next]; @@ -989,19 +989,19 @@ static void srp_send_completion(struct ib_cq *cq, void *target_ptr) static struct srp_iu *__srp_get_tx_iu(struct srp_target_port *target, enum srp_request_type req_type) { - s32 min = (req_type == SRP_REQ_TASK_MGMT) ? 1 : 2; + s32 rsv = (req_type == SRP_REQ_TASK_MGMT) ? 0 : SRP_TSK_MGMT_SQ_SIZE; srp_send_completion(target->send_cq, target); if (target->tx_head - target->tx_tail >= SRP_SQ_SIZE) return NULL; - if (target->req_lim < min) { + if (target->req_lim <= rsv) { ++target->zero_req_lim; return NULL; } - return target->tx_ring[target->tx_head & SRP_SQ_SIZE]; + return target->tx_ring[target->tx_head & SRP_SQ_MASK]; } /* @@ -1020,7 +1020,7 @@ static int __srp_post_send(struct srp_target_port *target, list.lkey = target->srp_host->srp_dev->mr->lkey; wr.next = NULL; - wr.wr_id = target->tx_head & SRP_SQ_SIZE; + wr.wr_id = target->tx_head & SRP_SQ_MASK; wr.sg_list = &list; wr.num_sge = 1; wr.opcode = IB_WR_SEND; @@ -1121,7 +1121,7 @@ static int srp_alloc_iu_bufs(struct srp_target_port *target) goto err; } - for (i = 0; i < SRP_SQ_SIZE + 1; ++i) { + for (i = 0; i < SRP_SQ_SIZE; ++i) { target->tx_ring[i] = srp_alloc_iu(target->srp_host, srp_max_iu_len, GFP_KERNEL, DMA_TO_DEVICE); @@ -1137,7 +1137,7 @@ err: target->rx_ring[i] = NULL; } - for (i = 0; i < SRP_SQ_SIZE + 1; ++i) { + for (i = 0; i < SRP_SQ_SIZE; ++i) { srp_free_iu(target->srp_host, target->tx_ring[i]); target->tx_ring[i] = NULL; } @@ -1626,9 +1626,9 @@ static struct scsi_host_template srp_template = { .eh_abort_handler = srp_abort, .eh_device_reset_handler = srp_reset_device, .eh_host_reset_handler = srp_reset_host, - .can_queue = SRP_SQ_SIZE, + .can_queue = SRP_CMD_SQ_SIZE, .this_id = -1, - .cmd_per_lun = SRP_SQ_SIZE, + .cmd_per_lun = SRP_CMD_SQ_SIZE, .use_clustering = ENABLE_CLUSTERING, .shost_attrs = srp_host_attrs }; @@ -1813,7 +1813,7 @@ static int srp_parse_options(const char *buf, struct srp_target_port *target) printk(KERN_WARNING PFX "bad max cmd_per_lun parameter '%s'\n", p); goto out; } - target->scsi_host->cmd_per_lun = min(token, SRP_SQ_SIZE); + target->scsi_host->cmd_per_lun = min(token, SRP_CMD_SQ_SIZE); break; case SRP_OPT_IO_CLASS: @@ -1891,7 +1891,7 @@ static ssize_t srp_create_target(struct device *dev, INIT_LIST_HEAD(&target->free_reqs); INIT_LIST_HEAD(&target->req_queue); - for (i = 0; i < SRP_SQ_SIZE; ++i) { + for (i = 0; i < SRP_CMD_SQ_SIZE; ++i) { target->req_ring[i].index = i; list_add_tail(&target->req_ring[i].list, &target->free_reqs); } @@ -2159,6 +2159,9 @@ static int __init srp_init_module(void) { int ret; + BUILD_BUG_ON_NOT_POWER_OF_2(SRP_SQ_SIZE); + BUILD_BUG_ON_NOT_POWER_OF_2(SRP_RQ_SIZE); + if (srp_sg_tablesize > 255) { printk(KERN_WARNING PFX "Clamping srp_sg_tablesize to 255\n"); srp_sg_tablesize = 255; diff --git a/drivers/infiniband/ulp/srp/ib_srp.h b/drivers/infiniband/ulp/srp/ib_srp.h index 5a80eac6fdaa..7a959d5f2fa6 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.h +++ b/drivers/infiniband/ulp/srp/ib_srp.h @@ -59,7 +59,14 @@ enum { SRP_RQ_SHIFT = 6, SRP_RQ_SIZE = 1 << SRP_RQ_SHIFT, - SRP_SQ_SIZE = SRP_RQ_SIZE - 1, + SRP_RQ_MASK = SRP_RQ_SIZE - 1, + + SRP_SQ_SIZE = SRP_RQ_SIZE, + SRP_SQ_MASK = SRP_SQ_SIZE - 1, + SRP_RSP_SQ_SIZE = 1, + SRP_REQ_SQ_SIZE = SRP_SQ_SIZE - SRP_RSP_SQ_SIZE, + SRP_TSK_MGMT_SQ_SIZE = 1, + SRP_CMD_SQ_SIZE = SRP_REQ_SQ_SIZE - SRP_TSK_MGMT_SQ_SIZE, SRP_TAG_TSK_MGMT = 1 << (SRP_RQ_SHIFT + 1), @@ -144,11 +151,11 @@ struct srp_target_port { unsigned tx_head; unsigned tx_tail; - struct srp_iu *tx_ring[SRP_SQ_SIZE + 1]; + struct srp_iu *tx_ring[SRP_SQ_SIZE]; struct list_head free_reqs; struct list_head req_queue; - struct srp_request req_ring[SRP_SQ_SIZE]; + struct srp_request req_ring[SRP_CMD_SQ_SIZE]; struct work_struct work; -- cgit v1.2.2 From bb12588a38e6db85e01dceadff7bc161fc92e7d2 Mon Sep 17 00:00:00 2001 From: David Dillow Date: Fri, 8 Oct 2010 14:40:47 -0400 Subject: IB/srp: Implement SRP_CRED_REQ and SRP_AER_REQ This patch adds support for SRP_CRED_REQ to avoid a lockup by targets that use that mechanism to return credits to the initiator. This prevents a lockup observed in the field where we would never add the credits from the SRP_CRED_REQ to our current count, and would therefore never send another command to the target. Minimal support for SRP_AER_REQ is also added, as these messages can also be used to convey additional credits to the initiator. Based upon extensive debugging and code by Bart Van Assche and a bug report by Chris Worley. Signed-off-by: David Dillow Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/srp/ib_srp.c | 105 +++++++++++++++++++++++++++++++++--- drivers/infiniband/ulp/srp/ib_srp.h | 8 +-- 2 files changed, 103 insertions(+), 10 deletions(-) (limited to 'drivers/infiniband/ulp') diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c index b8b09a4f012a..a54eee9324b6 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.c +++ b/drivers/infiniband/ulp/srp/ib_srp.c @@ -83,6 +83,10 @@ static void srp_remove_one(struct ib_device *device); static void srp_recv_completion(struct ib_cq *cq, void *target_ptr); static void srp_send_completion(struct ib_cq *cq, void *target_ptr); static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event); +static struct srp_iu *__srp_get_tx_iu(struct srp_target_port *target, + enum srp_iu_type iu_type); +static int __srp_post_send(struct srp_target_port *target, + struct srp_iu *iu, int len); static struct scsi_transport_template *ib_srp_transport_template; @@ -896,6 +900,71 @@ static void srp_process_rsp(struct srp_target_port *target, struct srp_rsp *rsp) spin_unlock_irqrestore(target->scsi_host->host_lock, flags); } +static int srp_response_common(struct srp_target_port *target, s32 req_delta, + void *rsp, int len) +{ + struct ib_device *dev; + unsigned long flags; + struct srp_iu *iu; + int err = 1; + + dev = target->srp_host->srp_dev->dev; + + spin_lock_irqsave(target->scsi_host->host_lock, flags); + target->req_lim += req_delta; + + iu = __srp_get_tx_iu(target, SRP_IU_RSP); + if (!iu) { + shost_printk(KERN_ERR, target->scsi_host, PFX + "no IU available to send response\n"); + goto out; + } + + ib_dma_sync_single_for_cpu(dev, iu->dma, len, DMA_TO_DEVICE); + memcpy(iu->buf, rsp, len); + ib_dma_sync_single_for_device(dev, iu->dma, len, DMA_TO_DEVICE); + + err = __srp_post_send(target, iu, len); + if (err) + shost_printk(KERN_ERR, target->scsi_host, PFX + "unable to post response: %d\n", err); + +out: + spin_unlock_irqrestore(target->scsi_host->host_lock, flags); + return err; +} + +static void srp_process_cred_req(struct srp_target_port *target, + struct srp_cred_req *req) +{ + struct srp_cred_rsp rsp = { + .opcode = SRP_CRED_RSP, + .tag = req->tag, + }; + s32 delta = be32_to_cpu(req->req_lim_delta); + + if (srp_response_common(target, delta, &rsp, sizeof rsp)) + shost_printk(KERN_ERR, target->scsi_host, PFX + "problems processing SRP_CRED_REQ\n"); +} + +static void srp_process_aer_req(struct srp_target_port *target, + struct srp_aer_req *req) +{ + struct srp_aer_rsp rsp = { + .opcode = SRP_AER_RSP, + .tag = req->tag, + }; + s32 delta = be32_to_cpu(req->req_lim_delta); + + shost_printk(KERN_ERR, target->scsi_host, PFX + "ignoring AER for LUN %llu\n", be64_to_cpu(req->lun)); + + if (srp_response_common(target, delta, &rsp, sizeof rsp)) + shost_printk(KERN_ERR, target->scsi_host, PFX + "problems processing SRP_AER_REQ\n"); +} + static void srp_handle_recv(struct srp_target_port *target, struct ib_wc *wc) { struct ib_device *dev; @@ -923,6 +992,14 @@ static void srp_handle_recv(struct srp_target_port *target, struct ib_wc *wc) srp_process_rsp(target, iu->buf); break; + case SRP_CRED_REQ: + srp_process_cred_req(target, iu->buf); + break; + + case SRP_AER_REQ: + srp_process_aer_req(target, iu->buf); + break; + case SRP_T_LOGOUT: /* XXX Handle target logout */ shost_printk(KERN_WARNING, target->scsi_host, @@ -985,23 +1062,36 @@ static void srp_send_completion(struct ib_cq *cq, void *target_ptr) * Must be called with target->scsi_host->host_lock held to protect * req_lim and tx_head. Lock cannot be dropped between call here and * call to __srp_post_send(). + * + * Note: + * An upper limit for the number of allocated information units for each + * request type is: + * - SRP_IU_CMD: SRP_CMD_SQ_SIZE, since the SCSI mid-layer never queues + * more than Scsi_Host.can_queue requests. + * - SRP_IU_TSK_MGMT: SRP_TSK_MGMT_SQ_SIZE. + * - SRP_IU_RSP: 1, since a conforming SRP target never sends more than + * one unanswered SRP request to an initiator. */ static struct srp_iu *__srp_get_tx_iu(struct srp_target_port *target, - enum srp_request_type req_type) + enum srp_iu_type iu_type) { - s32 rsv = (req_type == SRP_REQ_TASK_MGMT) ? 0 : SRP_TSK_MGMT_SQ_SIZE; + s32 rsv = (iu_type == SRP_IU_TSK_MGMT) ? 0 : SRP_TSK_MGMT_SQ_SIZE; + struct srp_iu *iu; srp_send_completion(target->send_cq, target); if (target->tx_head - target->tx_tail >= SRP_SQ_SIZE) return NULL; - if (target->req_lim <= rsv) { + /* Initiator responses to target requests do not consume credits */ + if (target->req_lim <= rsv && iu_type != SRP_IU_RSP) { ++target->zero_req_lim; return NULL; } - return target->tx_ring[target->tx_head & SRP_SQ_MASK]; + iu = target->tx_ring[target->tx_head & SRP_SQ_MASK]; + iu->type = iu_type; + return iu; } /* @@ -1030,7 +1120,8 @@ static int __srp_post_send(struct srp_target_port *target, if (!ret) { ++target->tx_head; - --target->req_lim; + if (iu->type != SRP_IU_RSP) + --target->req_lim; } return ret; @@ -1056,7 +1147,7 @@ static int srp_queuecommand(struct scsi_cmnd *scmnd, return 0; } - iu = __srp_get_tx_iu(target, SRP_REQ_NORMAL); + iu = __srp_get_tx_iu(target, SRP_IU_CMD); if (!iu) goto err; @@ -1363,7 +1454,7 @@ static int srp_send_tsk_mgmt(struct srp_target_port *target, init_completion(&req->done); - iu = __srp_get_tx_iu(target, SRP_REQ_TASK_MGMT); + iu = __srp_get_tx_iu(target, SRP_IU_TSK_MGMT); if (!iu) goto out; diff --git a/drivers/infiniband/ulp/srp/ib_srp.h b/drivers/infiniband/ulp/srp/ib_srp.h index 7a959d5f2fa6..ed0dce9e479f 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.h +++ b/drivers/infiniband/ulp/srp/ib_srp.h @@ -82,9 +82,10 @@ enum srp_target_state { SRP_TARGET_REMOVED }; -enum srp_request_type { - SRP_REQ_NORMAL, - SRP_REQ_TASK_MGMT, +enum srp_iu_type { + SRP_IU_CMD, + SRP_IU_TSK_MGMT, + SRP_IU_RSP, }; struct srp_device { @@ -171,6 +172,7 @@ struct srp_iu { void *buf; size_t size; enum dma_data_direction direction; + enum srp_iu_type type; }; #endif /* IB_SRP_H */ -- cgit v1.2.2 From c3aa9b186b95025d4ba4e90d6140c9887dfaae0a Mon Sep 17 00:00:00 2001 From: Eli Cohen Date: Mon, 20 Sep 2010 07:05:06 +0000 Subject: IPoIB: Set dev_id field of net_device Use the net device's dev_id field to encode the port number of the pci device. This can be used to to associate a net device with the pci device's port. The encoding is: dev_id = port - 1. Signed-off-by: Eli Cohen Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/ipoib/ipoib_main.c | 1 + 1 file changed, 1 insertion(+) (limited to 'drivers/infiniband/ulp') diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c index b4b22576f12a..30d23334f967 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_main.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c @@ -1240,6 +1240,7 @@ static struct net_device *ipoib_add_port(const char *format, goto alloc_mem_failed; SET_NETDEV_DEV(priv->dev, hca->dma_device); + priv->dev->dev_id = port - 1; if (!ib_query_port(hca, port, &attr)) priv->max_ib_mtu = ib_mtu_enum_to_int(attr.max_mtu); -- cgit v1.2.2 From 05a1d7504f836ee67e27f2488cb5b8126b51dbd4 Mon Sep 17 00:00:00 2001 From: David Dillow Date: Fri, 8 Oct 2010 14:48:14 -0400 Subject: IB/srp: Eliminate two forward declarations Signed-off-by: David Dillow Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/srp/ib_srp.c | 142 ++++++++++++++++++------------------ 1 file changed, 69 insertions(+), 73 deletions(-) (limited to 'drivers/infiniband/ulp') diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c index a54eee9324b6..d4c08d648137 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.c +++ b/drivers/infiniband/ulp/srp/ib_srp.c @@ -83,10 +83,6 @@ static void srp_remove_one(struct ib_device *device); static void srp_recv_completion(struct ib_cq *cq, void *target_ptr); static void srp_send_completion(struct ib_cq *cq, void *target_ptr); static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event); -static struct srp_iu *__srp_get_tx_iu(struct srp_target_port *target, - enum srp_iu_type iu_type); -static int __srp_post_send(struct srp_target_port *target, - struct srp_iu *iu, int len); static struct scsi_transport_template *ib_srp_transport_template; @@ -815,6 +811,75 @@ static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_target_port *target, return len; } +/* + * Must be called with target->scsi_host->host_lock held to protect + * req_lim and tx_head. Lock cannot be dropped between call here and + * call to __srp_post_send(). + * + * Note: + * An upper limit for the number of allocated information units for each + * request type is: + * - SRP_IU_CMD: SRP_CMD_SQ_SIZE, since the SCSI mid-layer never queues + * more than Scsi_Host.can_queue requests. + * - SRP_IU_TSK_MGMT: SRP_TSK_MGMT_SQ_SIZE. + * - SRP_IU_RSP: 1, since a conforming SRP target never sends more than + * one unanswered SRP request to an initiator. + */ +static struct srp_iu *__srp_get_tx_iu(struct srp_target_port *target, + enum srp_iu_type iu_type) +{ + s32 rsv = (iu_type == SRP_IU_TSK_MGMT) ? 0 : SRP_TSK_MGMT_SQ_SIZE; + struct srp_iu *iu; + + srp_send_completion(target->send_cq, target); + + if (target->tx_head - target->tx_tail >= SRP_SQ_SIZE) + return NULL; + + /* Initiator responses to target requests do not consume credits */ + if (target->req_lim <= rsv && iu_type != SRP_IU_RSP) { + ++target->zero_req_lim; + return NULL; + } + + iu = target->tx_ring[target->tx_head & SRP_SQ_MASK]; + iu->type = iu_type; + return iu; +} + +/* + * Must be called with target->scsi_host->host_lock held to protect + * req_lim and tx_head. + */ +static int __srp_post_send(struct srp_target_port *target, + struct srp_iu *iu, int len) +{ + struct ib_sge list; + struct ib_send_wr wr, *bad_wr; + int ret = 0; + + list.addr = iu->dma; + list.length = len; + list.lkey = target->srp_host->srp_dev->mr->lkey; + + wr.next = NULL; + wr.wr_id = target->tx_head & SRP_SQ_MASK; + wr.sg_list = &list; + wr.num_sge = 1; + wr.opcode = IB_WR_SEND; + wr.send_flags = IB_SEND_SIGNALED; + + ret = ib_post_send(target->qp, &wr, &bad_wr); + + if (!ret) { + ++target->tx_head; + if (iu->type != SRP_IU_RSP) + --target->req_lim; + } + + return ret; +} + static int srp_post_recv(struct srp_target_port *target) { unsigned long flags; @@ -1058,75 +1123,6 @@ static void srp_send_completion(struct ib_cq *cq, void *target_ptr) } } -/* - * Must be called with target->scsi_host->host_lock held to protect - * req_lim and tx_head. Lock cannot be dropped between call here and - * call to __srp_post_send(). - * - * Note: - * An upper limit for the number of allocated information units for each - * request type is: - * - SRP_IU_CMD: SRP_CMD_SQ_SIZE, since the SCSI mid-layer never queues - * more than Scsi_Host.can_queue requests. - * - SRP_IU_TSK_MGMT: SRP_TSK_MGMT_SQ_SIZE. - * - SRP_IU_RSP: 1, since a conforming SRP target never sends more than - * one unanswered SRP request to an initiator. - */ -static struct srp_iu *__srp_get_tx_iu(struct srp_target_port *target, - enum srp_iu_type iu_type) -{ - s32 rsv = (iu_type == SRP_IU_TSK_MGMT) ? 0 : SRP_TSK_MGMT_SQ_SIZE; - struct srp_iu *iu; - - srp_send_completion(target->send_cq, target); - - if (target->tx_head - target->tx_tail >= SRP_SQ_SIZE) - return NULL; - - /* Initiator responses to target requests do not consume credits */ - if (target->req_lim <= rsv && iu_type != SRP_IU_RSP) { - ++target->zero_req_lim; - return NULL; - } - - iu = target->tx_ring[target->tx_head & SRP_SQ_MASK]; - iu->type = iu_type; - return iu; -} - -/* - * Must be called with target->scsi_host->host_lock held to protect - * req_lim and tx_head. - */ -static int __srp_post_send(struct srp_target_port *target, - struct srp_iu *iu, int len) -{ - struct ib_sge list; - struct ib_send_wr wr, *bad_wr; - int ret = 0; - - list.addr = iu->dma; - list.length = len; - list.lkey = target->srp_host->srp_dev->mr->lkey; - - wr.next = NULL; - wr.wr_id = target->tx_head & SRP_SQ_MASK; - wr.sg_list = &list; - wr.num_sge = 1; - wr.opcode = IB_WR_SEND; - wr.send_flags = IB_SEND_SIGNALED; - - ret = ib_post_send(target->qp, &wr, &bad_wr); - - if (!ret) { - ++target->tx_head; - if (iu->type != SRP_IU_RSP) - --target->req_lim; - } - - return ret; -} - static int srp_queuecommand(struct scsi_cmnd *scmnd, void (*done)(struct scsi_cmnd *)) { -- cgit v1.2.2 From 7ade400aba9a675b610074d6609658661db07eeb Mon Sep 17 00:00:00 2001 From: Bart Van Assche Date: Mon, 30 Aug 2010 19:27:36 +0000 Subject: IB/srp: Reduce number of BUSY conditions As proposed by the SRP (draft) standard, ib_srp reserves one ring element for SRP_TSK_MGMT requests. This patch makes sure that the SCSI mid-layer never tries to queue more than (SRP request limit) - 1 SCSI commands to ib_srp. This improves performance for targets whose request limit is less than or equal to SRP_NORMAL_REQ_SQ_SIZE by reducing the number of BUSY responses reported by ib_srp to the SCSI mid-layer. Signed-off-by: Bart Van Assche Signed-off-by: David Dillow Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/srp/ib_srp.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) (limited to 'drivers/infiniband/ulp') diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c index d4c08d648137..4df827591cad 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.c +++ b/drivers/infiniband/ulp/srp/ib_srp.c @@ -1339,8 +1339,13 @@ static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event) target->max_ti_iu_len = be32_to_cpu(rsp->max_ti_iu_len); target->req_lim = be32_to_cpu(rsp->req_lim_delta); - target->scsi_host->can_queue = min(target->req_lim, - target->scsi_host->can_queue); + /* + * Reserve credits for task management so we don't + * bounce requests back to the SCSI mid-layer. + */ + target->scsi_host->can_queue + = min(target->req_lim - SRP_TSK_MGMT_SQ_SIZE, + target->scsi_host->can_queue); } else { shost_printk(KERN_WARNING, target->scsi_host, PFX "Unhandled RSP opcode %#x\n", opcode); -- cgit v1.2.2 From 21c1a90769e680e7c1f49bae4c5804cf0c7bc814 Mon Sep 17 00:00:00 2001 From: Bart Van Assche Date: Mon, 30 Aug 2010 19:27:40 +0000 Subject: IB/srp: Use list_first_entry() Use the list_first_entry() macro in ib_srp instead of open-coding the equivalent, which makes the source code slightly more descriptive. The list_first_entry() macro itself was introduced in kernel 2.6.22. Signed-off-by: Bart Van Assche Signed-off-by: David Dillow Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/srp/ib_srp.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/infiniband/ulp') diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c index 4df827591cad..9b4bc5ae5b10 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.c +++ b/drivers/infiniband/ulp/srp/ib_srp.c @@ -1151,7 +1151,7 @@ static int srp_queuecommand(struct scsi_cmnd *scmnd, ib_dma_sync_single_for_cpu(dev, iu->dma, srp_max_iu_len, DMA_TO_DEVICE); - req = list_entry(target->free_reqs.next, struct srp_request, list); + req = list_first_entry(&target->free_reqs, struct srp_request, list); scmnd->scsi_done = done; scmnd->result = 0; -- cgit v1.2.2 From 19081f31ce941a22bfc681d18ae2d31e31084df5 Mon Sep 17 00:00:00 2001 From: David Dillow Date: Mon, 18 Oct 2010 08:54:49 -0400 Subject: IB/srp: Sync buffer before posting send srp_send_tsk_mgmt() was missing the proper DMA sync calls before posting the buffer to the device. Signed-off-by: David Dillow Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/srp/ib_srp.c | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'drivers/infiniband/ulp') diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c index 9b4bc5ae5b10..cfc1d65c4577 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.c +++ b/drivers/infiniband/ulp/srp/ib_srp.c @@ -1442,6 +1442,7 @@ static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event) static int srp_send_tsk_mgmt(struct srp_target_port *target, struct srp_request *req, u8 func) { + struct ib_device *dev = target->srp_host->srp_dev->dev; struct srp_iu *iu; struct srp_tsk_mgmt *tsk_mgmt; @@ -1459,6 +1460,8 @@ static int srp_send_tsk_mgmt(struct srp_target_port *target, if (!iu) goto out; + ib_dma_sync_single_for_cpu(dev, iu->dma, sizeof *tsk_mgmt, + DMA_TO_DEVICE); tsk_mgmt = iu->buf; memset(tsk_mgmt, 0, sizeof *tsk_mgmt); @@ -1468,6 +1471,8 @@ static int srp_send_tsk_mgmt(struct srp_target_port *target, tsk_mgmt->tsk_mgmt_func = func; tsk_mgmt->task_tag = req->index; + ib_dma_sync_single_for_device(dev, iu->dma, sizeof *tsk_mgmt, + DMA_TO_DEVICE); if (__srp_post_send(target, iu, sizeof *tsk_mgmt)) goto out; -- cgit v1.2.2