diff options
Diffstat (limited to 'drivers/infiniband/ulp')
-rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_cm.c | 3 | ||||
-rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_main.c | 9 | ||||
-rw-r--r-- | drivers/infiniband/ulp/iser/iscsi_iser.c | 22 | ||||
-rw-r--r-- | drivers/infiniband/ulp/iser/iscsi_iser.h | 73 | ||||
-rw-r--r-- | drivers/infiniband/ulp/iser/iser_initiator.c | 148 | ||||
-rw-r--r-- | drivers/infiniband/ulp/iser/iser_memory.c | 231 | ||||
-rw-r--r-- | drivers/infiniband/ulp/iser/iser_verbs.c | 292 | ||||
-rw-r--r-- | drivers/infiniband/ulp/isert/ib_isert.c | 747 | ||||
-rw-r--r-- | drivers/infiniband/ulp/isert/ib_isert.h | 26 |
9 files changed, 1148 insertions, 403 deletions
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_cm.c b/drivers/infiniband/ulp/ipoib/ipoib_cm.c index 3eceb61e353..7a3175400b2 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_cm.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_cm.c @@ -817,7 +817,6 @@ void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc) if (neigh) { neigh->cm = NULL; - list_del(&neigh->list); ipoib_neigh_free(neigh); tx->neigh = NULL; @@ -1234,7 +1233,6 @@ static int ipoib_cm_tx_handler(struct ib_cm_id *cm_id, if (neigh) { neigh->cm = NULL; - list_del(&neigh->list); ipoib_neigh_free(neigh); tx->neigh = NULL; @@ -1325,7 +1323,6 @@ static void ipoib_cm_tx_start(struct work_struct *work) neigh = p->neigh; if (neigh) { neigh->cm = NULL; - list_del(&neigh->list); ipoib_neigh_free(neigh); } list_del(&p->list); diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c index c6f71a88c55..82cec1af902 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_main.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c @@ -493,7 +493,6 @@ static void path_rec_completion(int status, path, neigh)); if (!ipoib_cm_get(neigh)) { - list_del(&neigh->list); ipoib_neigh_free(neigh); continue; } @@ -618,7 +617,6 @@ static void neigh_add_path(struct sk_buff *skb, u8 *daddr, if (!ipoib_cm_get(neigh)) ipoib_cm_set(neigh, ipoib_cm_create_tx(dev, path, neigh)); if (!ipoib_cm_get(neigh)) { - list_del(&neigh->list); ipoib_neigh_free(neigh); goto err_drop; } @@ -639,7 +637,7 @@ static void neigh_add_path(struct sk_buff *skb, u8 *daddr, neigh->ah = NULL; if (!path->query && path_rec_start(dev, path)) - goto err_list; + goto err_path; __skb_queue_tail(&neigh->queue, skb); } @@ -648,9 +646,6 @@ static void neigh_add_path(struct sk_buff *skb, u8 *daddr, ipoib_neigh_put(neigh); return; -err_list: - list_del(&neigh->list); - err_path: ipoib_neigh_free(neigh); err_drop: @@ -1098,6 +1093,8 @@ void ipoib_neigh_free(struct ipoib_neigh *neigh) rcu_assign_pointer(*np, rcu_dereference_protected(neigh->hnext, lockdep_is_held(&priv->lock))); + /* remove from parent list */ + list_del(&neigh->list); call_rcu(&neigh->rcu, ipoib_neigh_reclaim); return; } else { diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.c b/drivers/infiniband/ulp/iser/iscsi_iser.c index 2e84ef859c5..dd03cfe596d 100644 --- a/drivers/infiniband/ulp/iser/iscsi_iser.c +++ b/drivers/infiniband/ulp/iser/iscsi_iser.c @@ -347,6 +347,7 @@ iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session, { struct iscsi_conn *conn = cls_conn->dd_data; struct iscsi_iser_conn *iser_conn; + struct iscsi_session *session; struct iser_conn *ib_conn; struct iscsi_endpoint *ep; int error; @@ -365,7 +366,8 @@ iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session, } ib_conn = ep->dd_data; - if (iser_alloc_rx_descriptors(ib_conn)) + session = conn->session; + if (iser_alloc_rx_descriptors(ib_conn, session)) return -ENOMEM; /* binds the iSER connection retrieved from the previously @@ -419,12 +421,13 @@ iscsi_iser_session_create(struct iscsi_endpoint *ep, struct iscsi_cls_session *cls_session; struct iscsi_session *session; struct Scsi_Host *shost; - struct iser_conn *ib_conn; + struct iser_conn *ib_conn = NULL; shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0); if (!shost) return NULL; shost->transportt = iscsi_iser_scsi_transport; + shost->cmd_per_lun = qdepth; shost->max_lun = iscsi_max_lun; shost->max_id = 0; shost->max_channel = 0; @@ -441,12 +444,14 @@ iscsi_iser_session_create(struct iscsi_endpoint *ep, ep ? ib_conn->device->ib_device->dma_device : NULL)) goto free_host; - /* - * we do not support setting can_queue cmd_per_lun from userspace yet - * because we preallocate so many resources - */ + if (cmds_max > ISER_DEF_XMIT_CMDS_MAX) { + iser_info("cmds_max changed from %u to %u\n", + cmds_max, ISER_DEF_XMIT_CMDS_MAX); + cmds_max = ISER_DEF_XMIT_CMDS_MAX; + } + cls_session = iscsi_session_setup(&iscsi_iser_transport, shost, - ISCSI_DEF_XMIT_CMDS_MAX, 0, + cmds_max, 0, sizeof(struct iscsi_iser_task), initial_cmdsn, 0); if (!cls_session) @@ -672,6 +677,7 @@ static umode_t iser_attr_is_visible(int param_type, int param) case ISCSI_PARAM_TGT_RESET_TMO: case ISCSI_PARAM_IFACE_NAME: case ISCSI_PARAM_INITIATOR_NAME: + case ISCSI_PARAM_DISCOVERY_SESS: return S_IRUGO; default: return 0; @@ -701,7 +707,7 @@ static struct scsi_host_template iscsi_iser_sht = { static struct iscsi_transport iscsi_iser_transport = { .owner = THIS_MODULE, .name = "iser", - .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T, + .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO, /* session management */ .create_session = iscsi_iser_session_create, .destroy_session = iscsi_iser_session_destroy, diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.h b/drivers/infiniband/ulp/iser/iscsi_iser.h index 4f069c0d4c0..67914027c61 100644 --- a/drivers/infiniband/ulp/iser/iscsi_iser.h +++ b/drivers/infiniband/ulp/iser/iscsi_iser.h @@ -78,14 +78,14 @@ #define iser_warn(fmt, arg...) \ do { \ - if (iser_debug_level > 1) \ + if (iser_debug_level > 0) \ pr_warn(PFX "%s:" fmt, \ __func__ , ## arg); \ } while (0) #define iser_info(fmt, arg...) \ do { \ - if (iser_debug_level > 0) \ + if (iser_debug_level > 1) \ pr_info(PFX "%s:" fmt, \ __func__ , ## arg); \ } while (0) @@ -102,7 +102,13 @@ /* support up to 512KB in one RDMA */ #define ISCSI_ISER_SG_TABLESIZE (0x80000 >> SHIFT_4K) -#define ISER_DEF_CMD_PER_LUN ISCSI_DEF_XMIT_CMDS_MAX +#define ISER_DEF_XMIT_CMDS_DEFAULT 512 +#if ISCSI_DEF_XMIT_CMDS_MAX > ISER_DEF_XMIT_CMDS_DEFAULT + #define ISER_DEF_XMIT_CMDS_MAX ISCSI_DEF_XMIT_CMDS_MAX +#else + #define ISER_DEF_XMIT_CMDS_MAX ISER_DEF_XMIT_CMDS_DEFAULT +#endif +#define ISER_DEF_CMD_PER_LUN ISER_DEF_XMIT_CMDS_MAX /* QP settings */ /* Maximal bounds on received asynchronous PDUs */ @@ -111,9 +117,9 @@ #define ISER_MAX_TX_MISC_PDUS 6 /* NOOP_OUT(2), TEXT(1), * * SCSI_TMFUNC(2), LOGOUT(1) */ -#define ISER_QP_MAX_RECV_DTOS (ISCSI_DEF_XMIT_CMDS_MAX) +#define ISER_QP_MAX_RECV_DTOS (ISER_DEF_XMIT_CMDS_MAX) -#define ISER_MIN_POSTED_RX (ISCSI_DEF_XMIT_CMDS_MAX >> 2) +#define ISER_MIN_POSTED_RX (ISER_DEF_XMIT_CMDS_MAX >> 2) /* the max TX (send) WR supported by the iSER QP is defined by * * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect * @@ -123,7 +129,7 @@ #define ISER_INFLIGHT_DATAOUTS 8 -#define ISER_QP_MAX_REQ_DTOS (ISCSI_DEF_XMIT_CMDS_MAX * \ +#define ISER_QP_MAX_REQ_DTOS (ISER_DEF_XMIT_CMDS_MAX * \ (1 + ISER_INFLIGHT_DATAOUTS) + \ ISER_MAX_TX_MISC_PDUS + \ ISER_MAX_RX_MISC_PDUS) @@ -205,7 +211,7 @@ struct iser_mem_reg { u64 va; u64 len; void *mem_h; - int is_fmr; + int is_mr; }; struct iser_regd_buf { @@ -246,6 +252,9 @@ struct iser_rx_desc { #define ISER_MAX_CQ 4 +struct iser_conn; +struct iscsi_iser_task; + struct iser_device { struct ib_device *ib_device; struct ib_pd *pd; @@ -259,6 +268,22 @@ struct iser_device { int cq_active_qps[ISER_MAX_CQ]; int cqs_used; struct iser_cq_desc *cq_desc; + int (*iser_alloc_rdma_reg_res)(struct iser_conn *ib_conn, + unsigned cmds_max); + void (*iser_free_rdma_reg_res)(struct iser_conn *ib_conn); + int (*iser_reg_rdma_mem)(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir); + void (*iser_unreg_rdma_mem)(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir); +}; + +struct fast_reg_descriptor { + struct list_head list; + /* For fast registration - FRWR */ + struct ib_mr *data_mr; + struct ib_fast_reg_page_list *data_frpl; + /* Valid for fast registration flag */ + bool valid; }; struct iser_conn { @@ -270,13 +295,13 @@ struct iser_conn { struct iser_device *device; /* device context */ struct rdma_cm_id *cma_id; /* CMA ID */ struct ib_qp *qp; /* QP */ - struct ib_fmr_pool *fmr_pool; /* pool of IB FMRs */ wait_queue_head_t wait; /* waitq for conn/disconn */ + unsigned qp_max_recv_dtos; /* num of rx buffers */ + unsigned qp_max_recv_dtos_mask; /* above minus 1 */ + unsigned min_posted_rx; /* qp_max_recv_dtos >> 2 */ int post_recv_buf_count; /* posted rx count */ atomic_t post_send_buf_count; /* posted tx count */ char name[ISER_OBJECT_NAME_SIZE]; - struct iser_page_vec *page_vec; /* represents SG to fmr maps* - * maps serialized as tx is*/ struct list_head conn_list; /* entry in ig conn list */ char *login_buf; @@ -285,6 +310,17 @@ struct iser_conn { unsigned int rx_desc_head; struct iser_rx_desc *rx_descs; struct ib_recv_wr rx_wr[ISER_MIN_POSTED_RX]; + union { + struct { + struct ib_fmr_pool *pool; /* pool of IB FMRs */ + struct iser_page_vec *page_vec; /* represents SG to fmr maps* + * maps serialized as tx is*/ + } fmr; + struct { + struct list_head pool; + int pool_size; + } frwr; + } fastreg; }; struct iscsi_iser_conn { @@ -368,8 +404,10 @@ void iser_free_rx_descriptors(struct iser_conn *ib_conn); void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *task, enum iser_data_dir cmd_dir); -int iser_reg_rdma_mem(struct iscsi_iser_task *task, - enum iser_data_dir cmd_dir); +int iser_reg_rdma_mem_fmr(struct iscsi_iser_task *task, + enum iser_data_dir cmd_dir); +int iser_reg_rdma_mem_frwr(struct iscsi_iser_task *task, + enum iser_data_dir cmd_dir); int iser_connect(struct iser_conn *ib_conn, struct sockaddr_in *src_addr, @@ -380,7 +418,10 @@ int iser_reg_page_vec(struct iser_conn *ib_conn, struct iser_page_vec *page_vec, struct iser_mem_reg *mem_reg); -void iser_unreg_mem(struct iser_mem_reg *mem_reg); +void iser_unreg_mem_fmr(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir); +void iser_unreg_mem_frwr(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir); int iser_post_recvl(struct iser_conn *ib_conn); int iser_post_recvm(struct iser_conn *ib_conn, int count); @@ -394,5 +435,9 @@ int iser_dma_map_task_data(struct iscsi_iser_task *iser_task, void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task); int iser_initialize_task_headers(struct iscsi_task *task, struct iser_tx_desc *tx_desc); -int iser_alloc_rx_descriptors(struct iser_conn *ib_conn); +int iser_alloc_rx_descriptors(struct iser_conn *ib_conn, struct iscsi_session *session); +int iser_create_fmr_pool(struct iser_conn *ib_conn, unsigned cmds_max); +void iser_free_fmr_pool(struct iser_conn *ib_conn); +int iser_create_frwr_pool(struct iser_conn *ib_conn, unsigned cmds_max); +void iser_free_frwr_pool(struct iser_conn *ib_conn); #endif diff --git a/drivers/infiniband/ulp/iser/iser_initiator.c b/drivers/infiniband/ulp/iser/iser_initiator.c index b6d81a86c97..538822684d5 100644 --- a/drivers/infiniband/ulp/iser/iser_initiator.c +++ b/drivers/infiniband/ulp/iser/iser_initiator.c @@ -49,6 +49,7 @@ static int iser_prepare_read_cmd(struct iscsi_task *task, { struct iscsi_iser_task *iser_task = task->dd_data; + struct iser_device *device = iser_task->iser_conn->ib_conn->device; struct iser_regd_buf *regd_buf; int err; struct iser_hdr *hdr = &iser_task->desc.iser_header; @@ -69,7 +70,7 @@ static int iser_prepare_read_cmd(struct iscsi_task *task, return -EINVAL; } - err = iser_reg_rdma_mem(iser_task,ISER_DIR_IN); + err = device->iser_reg_rdma_mem(iser_task, ISER_DIR_IN); if (err) { iser_err("Failed to set up Data-IN RDMA\n"); return err; @@ -98,6 +99,7 @@ iser_prepare_write_cmd(struct iscsi_task *task, unsigned int edtl) { struct iscsi_iser_task *iser_task = task->dd_data; + struct iser_device *device = iser_task->iser_conn->ib_conn->device; struct iser_regd_buf *regd_buf; int err; struct iser_hdr *hdr = &iser_task->desc.iser_header; @@ -119,7 +121,7 @@ iser_prepare_write_cmd(struct iscsi_task *task, return -EINVAL; } - err = iser_reg_rdma_mem(iser_task,ISER_DIR_OUT); + err = device->iser_reg_rdma_mem(iser_task, ISER_DIR_OUT); if (err != 0) { iser_err("Failed to register write cmd RDMA mem\n"); return err; @@ -170,8 +172,78 @@ static void iser_create_send_desc(struct iser_conn *ib_conn, } } +static void iser_free_login_buf(struct iser_conn *ib_conn) +{ + if (!ib_conn->login_buf) + return; + + if (ib_conn->login_req_dma) + ib_dma_unmap_single(ib_conn->device->ib_device, + ib_conn->login_req_dma, + ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE); + + if (ib_conn->login_resp_dma) + ib_dma_unmap_single(ib_conn->device->ib_device, + ib_conn->login_resp_dma, + ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE); + + kfree(ib_conn->login_buf); + + /* make sure we never redo any unmapping */ + ib_conn->login_req_dma = 0; + ib_conn->login_resp_dma = 0; + ib_conn->login_buf = NULL; +} + +static int iser_alloc_login_buf(struct iser_conn *ib_conn) +{ + struct iser_device *device; + int req_err, resp_err; + + BUG_ON(ib_conn->device == NULL); + + device = ib_conn->device; + + ib_conn->login_buf = kmalloc(ISCSI_DEF_MAX_RECV_SEG_LEN + + ISER_RX_LOGIN_SIZE, GFP_KERNEL); + if (!ib_conn->login_buf) + goto out_err; + + ib_conn->login_req_buf = ib_conn->login_buf; + ib_conn->login_resp_buf = ib_conn->login_buf + + ISCSI_DEF_MAX_RECV_SEG_LEN; + + ib_conn->login_req_dma = ib_dma_map_single(ib_conn->device->ib_device, + (void *)ib_conn->login_req_buf, + ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE); + + ib_conn->login_resp_dma = ib_dma_map_single(ib_conn->device->ib_device, + (void *)ib_conn->login_resp_buf, + ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE); + + req_err = ib_dma_mapping_error(device->ib_device, + ib_conn->login_req_dma); + resp_err = ib_dma_mapping_error(device->ib_device, + ib_conn->login_resp_dma); + + if (req_err || resp_err) { + if (req_err) + ib_conn->login_req_dma = 0; + if (resp_err) + ib_conn->login_resp_dma = 0; + goto free_login_buf; + } + return 0; + +free_login_buf: + iser_free_login_buf(ib_conn); -int iser_alloc_rx_descriptors(struct iser_conn *ib_conn) +out_err: + iser_err("unable to alloc or map login buf\n"); + return -ENOMEM; +} + +int iser_alloc_rx_descriptors(struct iser_conn *ib_conn, struct iscsi_session *session) { int i, j; u64 dma_addr; @@ -179,14 +251,24 @@ int iser_alloc_rx_descriptors(struct iser_conn *ib_conn) struct ib_sge *rx_sg; struct iser_device *device = ib_conn->device; - ib_conn->rx_descs = kmalloc(ISER_QP_MAX_RECV_DTOS * + ib_conn->qp_max_recv_dtos = session->cmds_max; + ib_conn->qp_max_recv_dtos_mask = session->cmds_max - 1; /* cmds_max is 2^N */ + ib_conn->min_posted_rx = ib_conn->qp_max_recv_dtos >> 2; + + if (device->iser_alloc_rdma_reg_res(ib_conn, session->scsi_cmds_max)) + goto create_rdma_reg_res_failed; + + if (iser_alloc_login_buf(ib_conn)) + goto alloc_login_buf_fail; + + ib_conn->rx_descs = kmalloc(session->cmds_max * sizeof(struct iser_rx_desc), GFP_KERNEL); if (!ib_conn->rx_descs) goto rx_desc_alloc_fail; rx_desc = ib_conn->rx_descs; - for (i = 0; i < ISER_QP_MAX_RECV_DTOS; i++, rx_desc++) { + for (i = 0; i < ib_conn->qp_max_recv_dtos; i++, rx_desc++) { dma_addr = ib_dma_map_single(device->ib_device, (void *)rx_desc, ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE); if (ib_dma_mapping_error(device->ib_device, dma_addr)) @@ -207,10 +289,14 @@ rx_desc_dma_map_failed: rx_desc = ib_conn->rx_descs; for (j = 0; j < i; j++, rx_desc++) ib_dma_unmap_single(device->ib_device, rx_desc->dma_addr, - ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE); + ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE); kfree(ib_conn->rx_descs); ib_conn->rx_descs = NULL; rx_desc_alloc_fail: + iser_free_login_buf(ib_conn); +alloc_login_buf_fail: + device->iser_free_rdma_reg_res(ib_conn); +create_rdma_reg_res_failed: iser_err("failed allocating rx descriptors / data buffers\n"); return -ENOMEM; } @@ -222,18 +308,27 @@ void iser_free_rx_descriptors(struct iser_conn *ib_conn) struct iser_device *device = ib_conn->device; if (!ib_conn->rx_descs) - return; + goto free_login_buf; + + if (device->iser_free_rdma_reg_res) + device->iser_free_rdma_reg_res(ib_conn); rx_desc = ib_conn->rx_descs; - for (i = 0; i < ISER_QP_MAX_RECV_DTOS; i++, rx_desc++) + for (i = 0; i < ib_conn->qp_max_recv_dtos; i++, rx_desc++) ib_dma_unmap_single(device->ib_device, rx_desc->dma_addr, - ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE); + ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE); kfree(ib_conn->rx_descs); + /* make sure we never redo any unmapping */ + ib_conn->rx_descs = NULL; + +free_login_buf: + iser_free_login_buf(ib_conn); } static int iser_post_rx_bufs(struct iscsi_conn *conn, struct iscsi_hdr *req) { struct iscsi_iser_conn *iser_conn = conn->dd_data; + struct iscsi_session *session = conn->session; iser_dbg("req op %x flags %x\n", req->opcode, req->flags); /* check if this is the last login - going to full feature phase */ @@ -248,9 +343,16 @@ static int iser_post_rx_bufs(struct iscsi_conn *conn, struct iscsi_hdr *req) WARN_ON(iser_conn->ib_conn->post_recv_buf_count != 1); WARN_ON(atomic_read(&iser_conn->ib_conn->post_send_buf_count) != 0); - iser_dbg("Initially post: %d\n", ISER_MIN_POSTED_RX); + if (session->discovery_sess) { + iser_info("Discovery session, re-using login RX buffer\n"); + return 0; + } else + iser_info("Normal session, posting batch of RX %d buffers\n", + iser_conn->ib_conn->min_posted_rx); + /* Initial post receive buffers */ - if (iser_post_recvm(iser_conn->ib_conn, ISER_MIN_POSTED_RX)) + if (iser_post_recvm(iser_conn->ib_conn, + iser_conn->ib_conn->min_posted_rx)) return -ENOMEM; return 0; @@ -425,6 +527,8 @@ int iser_send_control(struct iscsi_conn *conn, } if (task == conn->login_task) { + iser_dbg("op %x dsl %lx, posting login rx buffer\n", + task->hdr->opcode, data_seg_len); err = iser_post_recvl(iser_conn->ib_conn); if (err) goto send_control_error; @@ -487,9 +591,9 @@ void iser_rcv_completion(struct iser_rx_desc *rx_desc, return; outstanding = ib_conn->post_recv_buf_count; - if (outstanding + ISER_MIN_POSTED_RX <= ISER_QP_MAX_RECV_DTOS) { - count = min(ISER_QP_MAX_RECV_DTOS - outstanding, - ISER_MIN_POSTED_RX); + if (outstanding + ib_conn->min_posted_rx <= ib_conn->qp_max_recv_dtos) { + count = min(ib_conn->qp_max_recv_dtos - outstanding, + ib_conn->min_posted_rx); err = iser_post_recvm(ib_conn, count); if (err) iser_err("posting %d rx bufs err %d\n", count, err); @@ -538,8 +642,8 @@ void iser_task_rdma_init(struct iscsi_iser_task *iser_task) void iser_task_rdma_finalize(struct iscsi_iser_task *iser_task) { + struct iser_device *device = iser_task->iser_conn->ib_conn->device; int is_rdma_aligned = 1; - struct iser_regd_buf *regd; /* if we were reading, copy back to unaligned sglist, * anyway dma_unmap and free the copy @@ -553,17 +657,11 @@ void iser_task_rdma_finalize(struct iscsi_iser_task *iser_task) iser_finalize_rdma_unaligned_sg(iser_task, ISER_DIR_OUT); } - if (iser_task->dir[ISER_DIR_IN]) { - regd = &iser_task->rdma_regd[ISER_DIR_IN]; - if (regd->reg.is_fmr) - iser_unreg_mem(®d->reg); - } + if (iser_task->dir[ISER_DIR_IN]) + device->iser_unreg_rdma_mem(iser_task, ISER_DIR_IN); - if (iser_task->dir[ISER_DIR_OUT]) { - regd = &iser_task->rdma_regd[ISER_DIR_OUT]; - if (regd->reg.is_fmr) - iser_unreg_mem(®d->reg); - } + if (iser_task->dir[ISER_DIR_OUT]) + device->iser_unreg_rdma_mem(iser_task, ISER_DIR_OUT); /* if the data was unaligned, it was already unmapped and then copied */ if (is_rdma_aligned) diff --git a/drivers/infiniband/ulp/iser/iser_memory.c b/drivers/infiniband/ulp/iser/iser_memory.c index 7827baf455a..1ce0c97d2cc 100644 --- a/drivers/infiniband/ulp/iser/iser_memory.c +++ b/drivers/infiniband/ulp/iser/iser_memory.c @@ -170,8 +170,8 @@ void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *iser_task, */ static int iser_sg_to_page_vec(struct iser_data_buf *data, - struct iser_page_vec *page_vec, - struct ib_device *ibdev) + struct ib_device *ibdev, u64 *pages, + int *offset, int *data_size) { struct scatterlist *sg, *sgl = (struct scatterlist *)data->buf; u64 start_addr, end_addr, page, chunk_start = 0; @@ -180,7 +180,7 @@ static int iser_sg_to_page_vec(struct iser_data_buf *data, int i, new_chunk, cur_page, last_ent = data->dma_nents - 1; /* compute the offset of first element */ - page_vec->offset = (u64) sgl[0].offset & ~MASK_4K; + *offset = (u64) sgl[0].offset & ~MASK_4K; new_chunk = 1; cur_page = 0; @@ -204,13 +204,14 @@ static int iser_sg_to_page_vec(struct iser_data_buf *data, which might be unaligned */ page = chunk_start & MASK_4K; do { - page_vec->pages[cur_page++] = page; + pages[cur_page++] = page; page += SIZE_4K; } while (page < end_addr); } - page_vec->data_size = total_sz; - iser_dbg("page_vec->data_size:%d cur_page %d\n", page_vec->data_size,cur_page); + *data_size = total_sz; + iser_dbg("page_vec->data_size:%d cur_page %d\n", + *data_size, cur_page); return cur_page; } @@ -267,11 +268,8 @@ static void iser_data_buf_dump(struct iser_data_buf *data, struct scatterlist *sg; int i; - if (iser_debug_level == 0) - return; - for_each_sg(sgl, sg, data->dma_nents, i) - iser_warn("sg[%d] dma_addr:0x%lX page:0x%p " + iser_dbg("sg[%d] dma_addr:0x%lX page:0x%p " "off:0x%x sz:0x%x dma_len:0x%x\n", i, (unsigned long)ib_sg_dma_address(ibdev, sg), sg_page(sg), sg->offset, @@ -298,8 +296,10 @@ static void iser_page_vec_build(struct iser_data_buf *data, page_vec->offset = 0; iser_dbg("Translating sg sz: %d\n", data->dma_nents); - page_vec_len = iser_sg_to_page_vec(data, page_vec, ibdev); - iser_dbg("sg len %d page_vec_len %d\n", data->dma_nents,page_vec_len); + page_vec_len = iser_sg_to_page_vec(data, ibdev, page_vec->pages, + &page_vec->offset, + &page_vec->data_size); + iser_dbg("sg len %d page_vec_len %d\n", data->dma_nents, page_vec_len); page_vec->length = page_vec_len; @@ -347,16 +347,41 @@ void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task) } } +static int fall_to_bounce_buf(struct iscsi_iser_task *iser_task, + struct ib_device *ibdev, + enum iser_data_dir cmd_dir, + int aligned_len) +{ + struct iscsi_conn *iscsi_conn = iser_task->iser_conn->iscsi_conn; + struct iser_data_buf *mem = &iser_task->data[cmd_dir]; + + iscsi_conn->fmr_unalign_cnt++; + iser_warn("rdma alignment violation (%d/%d aligned) or FMR not supported\n", + aligned_len, mem->size); + + if (iser_debug_level > 0) + iser_data_buf_dump(mem, ibdev); + + /* unmap the command data before accessing it */ + iser_dma_unmap_task_data(iser_task); + + /* allocate copy buf, if we are writing, copy the */ + /* unaligned scatterlist, dma map the copy */ + if (iser_start_rdma_unaligned_sg(iser_task, cmd_dir) != 0) + return -ENOMEM; + + return 0; +} + /** - * iser_reg_rdma_mem - Registers memory intended for RDMA, - * obtaining rkey and va + * iser_reg_rdma_mem_fmr - Registers memory intended for RDMA, + * using FMR (if possible) obtaining rkey and va * * returns 0 on success, errno code on failure */ -int iser_reg_rdma_mem(struct iscsi_iser_task *iser_task, - enum iser_data_dir cmd_dir) +int iser_reg_rdma_mem_fmr(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir) { - struct iscsi_conn *iscsi_conn = iser_task->iser_conn->iscsi_conn; struct iser_conn *ib_conn = iser_task->iser_conn->ib_conn; struct iser_device *device = ib_conn->device; struct ib_device *ibdev = device->ib_device; @@ -370,20 +395,13 @@ int iser_reg_rdma_mem(struct iscsi_iser_task *iser_task, regd_buf = &iser_task->rdma_regd[cmd_dir]; aligned_len = iser_data_buf_aligned_len(mem, ibdev); - if (aligned_len != mem->dma_nents || - (!ib_conn->fmr_pool && mem->dma_nents > 1)) { - iscsi_conn->fmr_unalign_cnt++; - iser_warn("rdma alignment violation (%d/%d aligned) or FMR not supported\n", - aligned_len, mem->size); - iser_data_buf_dump(mem, ibdev); - - /* unmap the command data before accessing it */ - iser_dma_unmap_task_data(iser_task); - - /* allocate copy buf, if we are writing, copy the */ - /* unaligned scatterlist, dma map the copy */ - if (iser_start_rdma_unaligned_sg(iser_task, cmd_dir) != 0) - return -ENOMEM; + if (aligned_len != mem->dma_nents) { + err = fall_to_bounce_buf(iser_task, ibdev, + cmd_dir, aligned_len); + if (err) { + iser_err("failed to allocate bounce buffer\n"); + return err; + } mem = &iser_task->data_copy[cmd_dir]; } @@ -395,7 +413,7 @@ int iser_reg_rdma_mem(struct iscsi_iser_task *iser_task, regd_buf->reg.rkey = device->mr->rkey; regd_buf->reg.len = ib_sg_dma_len(ibdev, &sg[0]); regd_buf->reg.va = ib_sg_dma_address(ibdev, &sg[0]); - regd_buf->reg.is_fmr = 0; + regd_buf->reg.is_mr = 0; iser_dbg("PHYSICAL Mem.register: lkey: 0x%08X rkey: 0x%08X " "va: 0x%08lX sz: %ld]\n", @@ -404,22 +422,159 @@ int iser_reg_rdma_mem(struct iscsi_iser_task *iser_task, (unsigned long)regd_buf->reg.va, (unsigned long)regd_buf->reg.len); } else { /* use FMR for multiple dma entries */ - iser_page_vec_build(mem, ib_conn->page_vec, ibdev); - err = iser_reg_page_vec(ib_conn, ib_conn->page_vec, ®d_buf->reg); + iser_page_vec_build(mem, ib_conn->fastreg.fmr.page_vec, ibdev); + err = iser_reg_page_vec(ib_conn, ib_conn->fastreg.fmr.page_vec, + ®d_buf->reg); if (err && err != -EAGAIN) { iser_data_buf_dump(mem, ibdev); iser_err("mem->dma_nents = %d (dlength = 0x%x)\n", mem->dma_nents, ntoh24(iser_task->desc.iscsi_header.dlength)); iser_err("page_vec: data_size = 0x%x, length = %d, offset = 0x%x\n", - ib_conn->page_vec->data_size, ib_conn->page_vec->length, - ib_conn->page_vec->offset); - for (i=0 ; i<ib_conn->page_vec->length ; i++) + ib_conn->fastreg.fmr.page_vec->data_size, + ib_conn->fastreg.fmr.page_vec->length, + ib_conn->fastreg.fmr.page_vec->offset); + for (i = 0; i < ib_conn->fastreg.fmr.page_vec->length; i++) iser_err("page_vec[%d] = 0x%llx\n", i, - (unsigned long long) ib_conn->page_vec->pages[i]); + (unsigned long long) ib_conn->fastreg.fmr.page_vec->pages[i]); } if (err) return err; } return 0; } + +static int iser_fast_reg_mr(struct fast_reg_descriptor *desc, + struct iser_conn *ib_conn, + struct iser_regd_buf *regd_buf, + u32 offset, unsigned int data_size, + unsigned int page_list_len) +{ + struct ib_send_wr fastreg_wr, inv_wr; + struct ib_send_wr *bad_wr, *wr = NULL; + u8 key; + int ret; + + if (!desc->valid) { + memset(&inv_wr, 0, sizeof(inv_wr)); + inv_wr.opcode = IB_WR_LOCAL_INV; + inv_wr.send_flags = IB_SEND_SIGNALED; + inv_wr.ex.invalidate_rkey = desc->data_mr->rkey; + wr = &inv_wr; + /* Bump the key */ + key = (u8)(desc->data_mr->rkey & 0x000000FF); + ib_update_fast_reg_key(desc->data_mr, ++key); + } + + /* Prepare FASTREG WR */ + memset(&fastreg_wr, 0, sizeof(fastreg_wr)); + fastreg_wr.opcode = IB_WR_FAST_REG_MR; + fastreg_wr.send_flags = IB_SEND_SIGNALED; + fastreg_wr.wr.fast_reg.iova_start = desc->data_frpl->page_list[0] + offset; + fastreg_wr.wr.fast_reg.page_list = desc->data_frpl; + fastreg_wr.wr.fast_reg.page_list_len = page_list_len; + fastreg_wr.wr.fast_reg.page_shift = SHIFT_4K; + fastreg_wr.wr.fast_reg.length = data_size; + fastreg_wr.wr.fast_reg.rkey = desc->data_mr->rkey; + fastreg_wr.wr.fast_reg.access_flags = (IB_ACCESS_LOCAL_WRITE | + IB_ACCESS_REMOTE_WRITE | + IB_ACCESS_REMOTE_READ); + + if (!wr) { + wr = &fastreg_wr; + atomic_inc(&ib_conn->post_send_buf_count); + } else { + wr->next = &fastreg_wr; + atomic_add(2, &ib_conn->post_send_buf_count); + } + + ret = ib_post_send(ib_conn->qp, wr, &bad_wr); + if (ret) { + if (bad_wr->next) + atomic_sub(2, &ib_conn->post_send_buf_count); + else + atomic_dec(&ib_conn->post_send_buf_count); + iser_err("fast registration failed, ret:%d\n", ret); + return ret; + } + desc->valid = false; + + regd_buf->reg.mem_h = desc; + regd_buf->reg.lkey = desc->data_mr->lkey; + regd_buf->reg.rkey = desc->data_mr->rkey; + regd_buf->reg.va = desc->data_frpl->page_list[0] + offset; + regd_buf->reg.len = data_size; + regd_buf->reg.is_mr = 1; + + return ret; +} + +/** + * iser_reg_rdma_mem_frwr - Registers memory intended for RDMA, + * using Fast Registration WR (if possible) obtaining rkey and va + * + * returns 0 on success, errno code on failure + */ +int iser_reg_rdma_mem_frwr(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir) +{ + struct iser_conn *ib_conn = iser_task->iser_conn->ib_conn; + struct iser_device *device = ib_conn->device; + struct ib_device *ibdev = device->ib_device; + struct iser_data_buf *mem = &iser_task->data[cmd_dir]; + struct iser_regd_buf *regd_buf = &iser_task->rdma_regd[cmd_dir]; + struct fast_reg_descriptor *desc; + unsigned int data_size, page_list_len; + int err, aligned_len; + unsigned long flags; + u32 offset; + + aligned_len = iser_data_buf_aligned_len(mem, ibdev); + if (aligned_len != mem->dma_nents) { + err = fall_to_bounce_buf(iser_task, ibdev, + cmd_dir, aligned_len); + if (err) { + iser_err("failed to allocate bounce buffer\n"); + return err; + } + mem = &iser_task->data_copy[cmd_dir]; + } + + /* if there a single dma entry, dma mr suffices */ + if (mem->dma_nents == 1) { + struct scatterlist *sg = (struct scatterlist *)mem->buf; + + regd_buf->reg.lkey = device->mr->lkey; + regd_buf->reg.rkey = device->mr->rkey; + regd_buf->reg.len = ib_sg_dma_len(ibdev, &sg[0]); + regd_buf->reg.va = ib_sg_dma_address(ibdev, &sg[0]); + regd_buf->reg.is_mr = 0; + } else { + spin_lock_irqsave(&ib_conn->lock, flags); + desc = list_first_entry(&ib_conn->fastreg.frwr.pool, + struct fast_reg_descriptor, list); + list_del(&desc->list); + spin_unlock_irqrestore(&ib_conn->lock, flags); + page_list_len = iser_sg_to_page_vec(mem, device->ib_device, + desc->data_frpl->page_list, + &offset, &data_size); + + if (page_list_len * SIZE_4K < data_size) { + iser_err("fast reg page_list too short to hold this SG\n"); + err = -EINVAL; + goto err_reg; + } + + err = iser_fast_reg_mr(desc, ib_conn, regd_buf, + offset, data_size, page_list_len); + if (err) + goto err_reg; + } + + return 0; +err_reg: + spin_lock_irqsave(&ib_conn->lock, flags); + list_add_tail(&desc->list, &ib_conn->fastreg.frwr.pool); + spin_unlock_irqrestore(&ib_conn->lock, flags); + return err; +} diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c index 2c4941d0656..afe95674008 100644 --- a/drivers/infiniband/ulp/iser/iser_verbs.c +++ b/drivers/infiniband/ulp/iser/iser_verbs.c @@ -73,6 +73,36 @@ static int iser_create_device_ib_res(struct iser_device *device) { int i, j; struct iser_cq_desc *cq_desc; + struct ib_device_attr *dev_attr; + + dev_attr = kmalloc(sizeof(*dev_attr), GFP_KERNEL); + if (!dev_attr) + return -ENOMEM; + + if (ib_query_device(device->ib_device, dev_attr)) { + pr_warn("Query device failed for %s\n", device->ib_device->name); + goto dev_attr_err; + } + + /* Assign function handles - based on FMR support */ + if (device->ib_device->alloc_fmr && device->ib_device->dealloc_fmr && + device->ib_device->map_phys_fmr && device->ib_device->unmap_fmr) { + iser_info("FMR supported, using FMR for registration\n"); + device->iser_alloc_rdma_reg_res = iser_create_fmr_pool; + device->iser_free_rdma_reg_res = iser_free_fmr_pool; + device->iser_reg_rdma_mem = iser_reg_rdma_mem_fmr; + device->iser_unreg_rdma_mem = iser_unreg_mem_fmr; + } else + if (dev_attr->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) { + iser_info("FRWR supported, using FRWR for registration\n"); + device->iser_alloc_rdma_reg_res = iser_create_frwr_pool; + device->iser_free_rdma_reg_res = iser_free_frwr_pool; + device->iser_reg_rdma_mem = iser_reg_rdma_mem_frwr; + device->iser_unreg_rdma_mem = iser_unreg_mem_frwr; + } else { + iser_err("IB device does not support FMRs nor FRWRs, can't register memory\n"); + goto dev_attr_err; + } device->cqs_used = min(ISER_MAX_CQ, device->ib_device->num_comp_vectors); iser_info("using %d CQs, device %s supports %d vectors\n", @@ -128,6 +158,7 @@ static int iser_create_device_ib_res(struct iser_device *device) if (ib_register_event_handler(&device->event_handler)) goto handler_err; + kfree(dev_attr); return 0; handler_err: @@ -147,6 +178,8 @@ pd_err: kfree(device->cq_desc); cq_desc_err: iser_err("failed to allocate an IB resource\n"); +dev_attr_err: + kfree(dev_attr); return -1; } @@ -178,56 +211,23 @@ static void iser_free_device_ib_res(struct iser_device *device) } /** - * iser_create_ib_conn_res - Creates FMR pool and Queue-Pair (QP) + * iser_create_fmr_pool - Creates FMR pool and page_vector * - * returns 0 on success, -1 on failure + * returns 0 on success, or errno code on failure */ -static int iser_create_ib_conn_res(struct iser_conn *ib_conn) +int iser_create_fmr_pool(struct iser_conn *ib_conn, unsigned cmds_max) { - struct iser_device *device; - struct ib_qp_init_attr init_attr; - int req_err, resp_err, ret = -ENOMEM; + struct iser_device *device = ib_conn->device; struct ib_fmr_pool_param params; - int index, min_index = 0; - - BUG_ON(ib_conn->device == NULL); - - device = ib_conn->device; - - ib_conn->login_buf = kmalloc(ISCSI_DEF_MAX_RECV_SEG_LEN + - ISER_RX_LOGIN_SIZE, GFP_KERNEL); - if (!ib_conn->login_buf) - goto out_err; - - ib_conn->login_req_buf = ib_conn->login_buf; - ib_conn->login_resp_buf = ib_conn->login_buf + ISCSI_DEF_MAX_RECV_SEG_LEN; - - ib_conn->login_req_dma = ib_dma_map_single(ib_conn->device->ib_device, - (void *)ib_conn->login_req_buf, - ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE); - - ib_conn->login_resp_dma = ib_dma_map_single(ib_conn->device->ib_device, - (void *)ib_conn->login_resp_buf, - ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE); - - req_err = ib_dma_mapping_error(device->ib_device, ib_conn->login_req_dma); - resp_err = ib_dma_mapping_error(device->ib_device, ib_conn->login_resp_dma); - - if (req_err || resp_err) { - if (req_err) - ib_conn->login_req_dma = 0; - if (resp_err) - ib_conn->login_resp_dma = 0; - goto out_err; - } + int ret = -ENOMEM; - ib_conn->page_vec = kmalloc(sizeof(struct iser_page_vec) + - (sizeof(u64) * (ISCSI_ISER_SG_TABLESIZE +1)), - GFP_KERNEL); - if (!ib_conn->page_vec) - goto out_err; + ib_conn->fastreg.fmr.page_vec = kmalloc(sizeof(struct iser_page_vec) + + (sizeof(u64)*(ISCSI_ISER_SG_TABLESIZE + 1)), + GFP_KERNEL); + if (!ib_conn->fastreg.fmr.page_vec) + return ret; - ib_conn->page_vec->pages = (u64 *) (ib_conn->page_vec + 1); + ib_conn->fastreg.fmr.page_vec->pages = (u64 *)(ib_conn->fastreg.fmr.page_vec + 1); params.page_shift = SHIFT_4K; /* when the first/last SG element are not start/end * @@ -235,24 +235,143 @@ static int iser_create_ib_conn_res(struct iser_conn *ib_conn) params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1; /* make the pool size twice the max number of SCSI commands * * the ML is expected to queue, watermark for unmap at 50% */ - params.pool_size = ISCSI_DEF_XMIT_CMDS_MAX * 2; - params.dirty_watermark = ISCSI_DEF_XMIT_CMDS_MAX; + params.pool_size = cmds_max * 2; + params.dirty_watermark = cmds_max; params.cache = 0; params.flush_function = NULL; params.access = (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | IB_ACCESS_REMOTE_READ); - ib_conn->fmr_pool = ib_create_fmr_pool(device->pd, ¶ms); - ret = PTR_ERR(ib_conn->fmr_pool); - if (IS_ERR(ib_conn->fmr_pool) && ret != -ENOSYS) { - ib_conn->fmr_pool = NULL; - goto out_err; - } else if (ret == -ENOSYS) { - ib_conn->fmr_pool = NULL; + ib_conn->fastreg.fmr.pool = ib_create_fmr_pool(device->pd, ¶ms); + if (!IS_ERR(ib_conn->fastreg.fmr.pool)) + return 0; + + /* no FMR => no need for page_vec */ + kfree(ib_conn->fastreg.fmr.page_vec); + ib_conn->fastreg.fmr.page_vec = NULL; + + ret = PTR_ERR(ib_conn->fastreg.fmr.pool); + ib_conn->fastreg.fmr.pool = NULL; + if (ret != -ENOSYS) { + iser_err("FMR allocation failed, err %d\n", ret); + return ret; + } else { iser_warn("FMRs are not supported, using unaligned mode\n"); - ret = 0; + return 0; } +} + +/** + * iser_free_fmr_pool - releases the FMR pool and page vec + */ +void iser_free_fmr_pool(struct iser_conn *ib_conn) +{ + iser_info("freeing conn %p fmr pool %p\n", + ib_conn, ib_conn->fastreg.fmr.pool); + + if (ib_conn->fastreg.fmr.pool != NULL) + ib_destroy_fmr_pool(ib_conn->fastreg.fmr.pool); + + ib_conn->fastreg.fmr.pool = NULL; + + kfree(ib_conn->fastreg.fmr.page_vec); + ib_conn->fastreg.fmr.page_vec = NULL; +} + +/** + * iser_create_frwr_pool - Creates pool of fast_reg descriptors + * for fast registration work requests. + * returns 0 on success, or errno code on failure + */ +int iser_create_frwr_pool(struct iser_conn *ib_conn, unsigned cmds_max) +{ + struct iser_device *device = ib_conn->device; + struct fast_reg_descriptor *desc; + int i, ret; + + INIT_LIST_HEAD(&ib_conn->fastreg.frwr.pool); + ib_conn->fastreg.frwr.pool_size = 0; + for (i = 0; i < cmds_max; i++) { + desc = kmalloc(sizeof(*desc), GFP_KERNEL); + if (!desc) { + iser_err("Failed to allocate a new fast_reg descriptor\n"); + ret = -ENOMEM; + goto err; + } + + desc->data_frpl = ib_alloc_fast_reg_page_list(device->ib_device, + ISCSI_ISER_SG_TABLESIZE + 1); + if (IS_ERR(desc->data_frpl)) { + ret = PTR_ERR(desc->data_frpl); + iser_err("Failed to allocate ib_fast_reg_page_list err=%d\n", ret); + goto fast_reg_page_failure; + } + + desc->data_mr = ib_alloc_fast_reg_mr(device->pd, + ISCSI_ISER_SG_TABLESIZE + 1); + if (IS_ERR(desc->data_mr)) { + ret = PTR_ERR(desc->data_mr); + iser_err("Failed to allocate ib_fast_reg_mr err=%d\n", ret); + goto fast_reg_mr_failure; + } + desc->valid = true; + list_add_tail(&desc->list, &ib_conn->fastreg.frwr.pool); + ib_conn->fastreg.frwr.pool_size++; + } + + return 0; + +fast_reg_mr_failure: + ib_free_fast_reg_page_list(desc->data_frpl); +fast_reg_page_failure: + kfree(desc); +err: + iser_free_frwr_pool(ib_conn); + return ret; +} + +/** + * iser_free_frwr_pool - releases the pool of fast_reg descriptors + */ +void iser_free_frwr_pool(struct iser_conn *ib_conn) +{ + struct fast_reg_descriptor *desc, *tmp; + int i = 0; + + if (list_empty(&ib_conn->fastreg.frwr.pool)) + return; + + iser_info("freeing conn %p frwr pool\n", ib_conn); + + list_for_each_entry_safe(desc, tmp, &ib_conn->fastreg.frwr.pool, list) { + list_del(&desc->list); + ib_free_fast_reg_page_list(desc->data_frpl); + ib_dereg_mr(desc->data_mr); + kfree(desc); + ++i; + } + + if (i < ib_conn->fastreg.frwr.pool_size) + iser_warn("pool still has %d regions registered\n", + ib_conn->fastreg.frwr.pool_size - i); +} + +/** + * iser_create_ib_conn_res - Queue-Pair (QP) + * + * returns 0 on success, -1 on failure + */ +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 = -ENOMEM; + int index, min_index = 0; + + BUG_ON(ib_conn->device == NULL); + + device = ib_conn->device; memset(&init_attr, 0, sizeof init_attr); @@ -282,9 +401,9 @@ static int iser_create_ib_conn_res(struct iser_conn *ib_conn) goto out_err; ib_conn->qp = ib_conn->cma_id->qp; - iser_info("setting conn %p cma_id %p: fmr_pool %p qp %p\n", + iser_info("setting conn %p cma_id %p qp %p\n", ib_conn, ib_conn->cma_id, - ib_conn->fmr_pool, ib_conn->cma_id->qp); + ib_conn->cma_id->qp); return ret; out_err: @@ -293,7 +412,7 @@ out_err: } /** - * releases the FMR pool and QP objects, returns 0 on success, + * releases the QP objects, returns 0 on success, * -1 on failure */ static int iser_free_ib_conn_res(struct iser_conn *ib_conn) @@ -301,13 +420,11 @@ static int iser_free_ib_conn_res(struct iser_conn *ib_conn) int cq_index; BUG_ON(ib_conn == NULL); - iser_info("freeing conn %p cma_id %p fmr pool %p qp %p\n", + iser_info("freeing conn %p cma_id %p qp %p\n", ib_conn, ib_conn->cma_id, - ib_conn->fmr_pool, ib_conn->qp); + ib_conn->qp); /* qp is created only once both addr & route are resolved */ - if (ib_conn->fmr_pool != NULL) - ib_destroy_fmr_pool(ib_conn->fmr_pool); if (ib_conn->qp != NULL) { cq_index = ((struct iser_cq_desc *)ib_conn->qp->recv_cq->cq_context)->cq_index; @@ -316,21 +433,7 @@ static int iser_free_ib_conn_res(struct iser_conn *ib_conn) rdma_destroy_qp(ib_conn->cma_id); } - ib_conn->fmr_pool = NULL; ib_conn->qp = NULL; - kfree(ib_conn->page_vec); - - if (ib_conn->login_buf) { - if (ib_conn->login_req_dma) - ib_dma_unmap_single(ib_conn->device->ib_device, - ib_conn->login_req_dma, - ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE); - if (ib_conn->login_resp_dma) - ib_dma_unmap_single(ib_conn->device->ib_device, - ib_conn->login_resp_dma, - ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE); - kfree(ib_conn->login_buf); - } return 0; } @@ -694,7 +797,7 @@ int iser_reg_page_vec(struct iser_conn *ib_conn, page_list = page_vec->pages; io_addr = page_list[0]; - mem = ib_fmr_pool_map_phys(ib_conn->fmr_pool, + mem = ib_fmr_pool_map_phys(ib_conn->fastreg.fmr.pool, page_list, page_vec->length, io_addr); @@ -709,7 +812,7 @@ int iser_reg_page_vec(struct iser_conn *ib_conn, mem_reg->rkey = mem->fmr->rkey; mem_reg->len = page_vec->length * SIZE_4K; mem_reg->va = io_addr; - mem_reg->is_fmr = 1; + mem_reg->is_mr = 1; mem_reg->mem_h = (void *)mem; mem_reg->va += page_vec->offset; @@ -727,12 +830,18 @@ int iser_reg_page_vec(struct iser_conn *ib_conn, } /** - * Unregister (previosuly registered) memory. + * Unregister (previosuly registered using FMR) memory. + * If memory is non-FMR does nothing. */ -void iser_unreg_mem(struct iser_mem_reg *reg) +void iser_unreg_mem_fmr(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir) { + struct iser_mem_reg *reg = &iser_task->rdma_regd[cmd_dir].reg; int ret; + if (!reg->is_mr) + return; + iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n",reg->mem_h); ret = ib_fmr_pool_unmap((struct ib_pool_fmr *)reg->mem_h); @@ -742,6 +851,23 @@ void iser_unreg_mem(struct iser_mem_reg *reg) reg->mem_h = NULL; } +void iser_unreg_mem_frwr(struct iscsi_iser_task *iser_task, + enum iser_data_dir cmd_dir) +{ + struct iser_mem_reg *reg = &iser_task->rdma_regd[cmd_dir].reg; + struct iser_conn *ib_conn = iser_task->iser_conn->ib_conn; + struct fast_reg_descriptor *desc = reg->mem_h; + + if (!reg->is_mr) + return; + + reg->mem_h = NULL; + reg->is_mr = 0; + spin_lock_bh(&ib_conn->lock); + list_add_tail(&desc->list, &ib_conn->fastreg.frwr.pool); + spin_unlock_bh(&ib_conn->lock); +} + int iser_post_recvl(struct iser_conn *ib_conn) { struct ib_recv_wr rx_wr, *rx_wr_failed; @@ -779,7 +905,7 @@ int iser_post_recvm(struct iser_conn *ib_conn, int count) 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); + my_rx_head = (my_rx_head + 1) & ib_conn->qp_max_recv_dtos_mask; } rx_wr--; @@ -863,7 +989,11 @@ static int iser_drain_tx_cq(struct iser_device *device, int cq_index) if (wc.status == IB_WC_SUCCESS) { if (wc.opcode == IB_WC_SEND) iser_snd_completion(tx_desc, ib_conn); - else + else if (wc.opcode == IB_WC_LOCAL_INV || + wc.opcode == IB_WC_FAST_REG_MR) { + atomic_dec(&ib_conn->post_send_buf_count); + continue; + } else iser_err("expected opcode %d got %d\n", IB_WC_SEND, wc.opcode); } else { diff --git a/drivers/infiniband/ulp/isert/ib_isert.c b/drivers/infiniband/ulp/isert/ib_isert.c index 3f62041222f..3591855cc5b 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.c +++ b/drivers/infiniband/ulp/isert/ib_isert.c @@ -1,7 +1,7 @@ /******************************************************************************* * This file contains iSCSI extentions for RDMA (iSER) Verbs * - * (c) Copyright 2013 RisingTide Systems LLC. + * (c) Copyright 2013 Datera, Inc. * * Nicholas A. Bellinger <nab@linux-iscsi.org> * @@ -39,7 +39,17 @@ static DEFINE_MUTEX(device_list_mutex); static LIST_HEAD(device_list); static struct workqueue_struct *isert_rx_wq; static struct workqueue_struct *isert_comp_wq; -static struct kmem_cache *isert_cmd_cache; + +static void +isert_unmap_cmd(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn); +static int +isert_map_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd, + struct isert_rdma_wr *wr); +static void +isert_unreg_rdma_frwr(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn); +static int +isert_reg_rdma_frwr(struct iscsi_conn *conn, struct iscsi_cmd *cmd, + struct isert_rdma_wr *wr); static void isert_qp_event_callback(struct ib_event *e, void *context) @@ -80,14 +90,8 @@ isert_conn_setup_qp(struct isert_conn *isert_conn, struct rdma_cm_id *cma_id) { struct isert_device *device = isert_conn->conn_device; struct ib_qp_init_attr attr; - struct ib_device_attr devattr; int ret, index, min_index = 0; - memset(&devattr, 0, sizeof(struct ib_device_attr)); - ret = isert_query_device(cma_id->device, &devattr); - if (ret) - return ret; - mutex_lock(&device_list_mutex); for (index = 0; index < device->cqs_used; index++) if (device->cq_active_qps[index] < @@ -108,7 +112,7 @@ isert_conn_setup_qp(struct isert_conn *isert_conn, struct rdma_cm_id *cma_id) * FIXME: Use devattr.max_sge - 2 for max_send_sge as * work-around for RDMA_READ.. */ - attr.cap.max_send_sge = devattr.max_sge - 2; + attr.cap.max_send_sge = device->dev_attr.max_sge - 2; isert_conn->max_sge = attr.cap.max_send_sge; attr.cap.max_recv_sge = 1; @@ -210,14 +214,31 @@ isert_create_device_ib_res(struct isert_device *device) { struct ib_device *ib_dev = device->ib_device; struct isert_cq_desc *cq_desc; + struct ib_device_attr *dev_attr; int ret = 0, i, j; + dev_attr = &device->dev_attr; + ret = isert_query_device(ib_dev, dev_attr); + if (ret) + return ret; + + /* asign function handlers */ + if (dev_attr->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) { + device->use_frwr = 1; + device->reg_rdma_mem = isert_reg_rdma_frwr; + device->unreg_rdma_mem = isert_unreg_rdma_frwr; + } else { + device->use_frwr = 0; + device->reg_rdma_mem = isert_map_rdma; + device->unreg_rdma_mem = isert_unmap_cmd; + } + device->cqs_used = min_t(int, num_online_cpus(), device->ib_device->num_comp_vectors); device->cqs_used = min(ISERT_MAX_CQ, device->cqs_used); - pr_debug("Using %d CQs, device %s supports %d vectors\n", + pr_debug("Using %d CQs, device %s supports %d vectors support FRWR %d\n", device->cqs_used, device->ib_device->name, - device->ib_device->num_comp_vectors); + device->ib_device->num_comp_vectors, device->use_frwr); device->cq_desc = kzalloc(sizeof(struct isert_cq_desc) * device->cqs_used, GFP_KERNEL); if (!device->cq_desc) { @@ -363,6 +384,85 @@ isert_device_find_by_ib_dev(struct rdma_cm_id *cma_id) return device; } +static void +isert_conn_free_frwr_pool(struct isert_conn *isert_conn) +{ + struct fast_reg_descriptor *fr_desc, *tmp; + int i = 0; + + if (list_empty(&isert_conn->conn_frwr_pool)) + return; + + pr_debug("Freeing conn %p frwr pool", isert_conn); + + list_for_each_entry_safe(fr_desc, tmp, + &isert_conn->conn_frwr_pool, list) { + list_del(&fr_desc->list); + ib_free_fast_reg_page_list(fr_desc->data_frpl); + ib_dereg_mr(fr_desc->data_mr); + kfree(fr_desc); + ++i; + } + + if (i < isert_conn->conn_frwr_pool_size) + pr_warn("Pool still has %d regions registered\n", + isert_conn->conn_frwr_pool_size - i); +} + +static int +isert_conn_create_frwr_pool(struct isert_conn *isert_conn) +{ + struct fast_reg_descriptor *fr_desc; + struct isert_device *device = isert_conn->conn_device; + int i, ret; + + INIT_LIST_HEAD(&isert_conn->conn_frwr_pool); + isert_conn->conn_frwr_pool_size = 0; + for (i = 0; i < ISCSI_DEF_XMIT_CMDS_MAX; i++) { + fr_desc = kzalloc(sizeof(*fr_desc), GFP_KERNEL); + if (!fr_desc) { + pr_err("Failed to allocate fast_reg descriptor\n"); + ret = -ENOMEM; + goto err; + } + + fr_desc->data_frpl = + ib_alloc_fast_reg_page_list(device->ib_device, + ISCSI_ISER_SG_TABLESIZE); + if (IS_ERR(fr_desc->data_frpl)) { + pr_err("Failed to allocate fr_pg_list err=%ld\n", + PTR_ERR(fr_desc->data_frpl)); + ret = PTR_ERR(fr_desc->data_frpl); + goto err; + } + + fr_desc->data_mr = ib_alloc_fast_reg_mr(device->dev_pd, + ISCSI_ISER_SG_TABLESIZE); + if (IS_ERR(fr_desc->data_mr)) { + pr_err("Failed to allocate frmr err=%ld\n", + PTR_ERR(fr_desc->data_mr)); + ret = PTR_ERR(fr_desc->data_mr); + ib_free_fast_reg_page_list(fr_desc->data_frpl); + goto err; + } + pr_debug("Create fr_desc %p page_list %p\n", + fr_desc, fr_desc->data_frpl->page_list); + + fr_desc->valid = true; + list_add_tail(&fr_desc->list, &isert_conn->conn_frwr_pool); + isert_conn->conn_frwr_pool_size++; + } + + pr_debug("Creating conn %p frwr pool size=%d", + isert_conn, isert_conn->conn_frwr_pool_size); + + return 0; + +err: + isert_conn_free_frwr_pool(isert_conn); + return ret; +} + static int isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) { @@ -389,6 +489,7 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) kref_init(&isert_conn->conn_kref); kref_get(&isert_conn->conn_kref); mutex_init(&isert_conn->conn_mutex); + spin_lock_init(&isert_conn->conn_lock); cma_id->context = isert_conn; isert_conn->conn_cm_id = cma_id; @@ -446,6 +547,14 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) isert_conn->conn_pd = device->dev_pd; isert_conn->conn_mr = device->dev_mr; + if (device->use_frwr) { + ret = isert_conn_create_frwr_pool(isert_conn); + if (ret) { + pr_err("Conn: %p failed to create frwr_pool\n", isert_conn); + goto out_frwr; + } + } + ret = isert_conn_setup_qp(isert_conn, cma_id); if (ret) goto out_conn_dev; @@ -459,6 +568,9 @@ isert_connect_request(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) return 0; out_conn_dev: + if (device->use_frwr) + isert_conn_free_frwr_pool(isert_conn); +out_frwr: isert_device_try_release(device); out_rsp_dma_map: ib_dma_unmap_single(ib_dev, isert_conn->login_rsp_dma, @@ -482,6 +594,9 @@ isert_connect_release(struct isert_conn *isert_conn) pr_debug("Entering isert_connect_release(): >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>\n"); + if (device->use_frwr) + isert_conn_free_frwr_pool(isert_conn); + if (isert_conn->conn_qp) { cq_index = ((struct isert_cq_desc *) isert_conn->conn_qp->recv_cq->cq_context)->cq_index; @@ -869,46 +984,37 @@ isert_rx_login_req(struct iser_rx_desc *rx_desc, int rx_buflen, size, rx_buflen, MAX_KEY_VALUE_PAIRS); memcpy(login->req_buf, &rx_desc->data[0], size); - complete(&isert_conn->conn_login_comp); -} - -static void -isert_release_cmd(struct iscsi_cmd *cmd) -{ - struct isert_cmd *isert_cmd = container_of(cmd, struct isert_cmd, - iscsi_cmd); - - pr_debug("Entering isert_release_cmd %p >>>>>>>>>>>>>>>.\n", isert_cmd); - - kfree(cmd->buf_ptr); - kfree(cmd->tmr_req); - - kmem_cache_free(isert_cmd_cache, isert_cmd); + if (login->first_request) { + complete(&isert_conn->conn_login_comp); + return; + } + schedule_delayed_work(&conn->login_work, 0); } static struct iscsi_cmd -*isert_alloc_cmd(struct iscsi_conn *conn, gfp_t gfp) +*isert_allocate_cmd(struct iscsi_conn *conn, gfp_t gfp) { struct isert_conn *isert_conn = (struct isert_conn *)conn->context; struct isert_cmd *isert_cmd; + struct iscsi_cmd *cmd; - isert_cmd = kmem_cache_zalloc(isert_cmd_cache, gfp); - if (!isert_cmd) { - pr_err("Unable to allocate isert_cmd\n"); + cmd = iscsit_allocate_cmd(conn, gfp); + if (!cmd) { + pr_err("Unable to allocate iscsi_cmd + isert_cmd\n"); return NULL; } + isert_cmd = iscsit_priv_cmd(cmd); isert_cmd->conn = isert_conn; - isert_cmd->iscsi_cmd.release_cmd = &isert_release_cmd; + isert_cmd->iscsi_cmd = cmd; - return &isert_cmd->iscsi_cmd; + return cmd; } static int isert_handle_scsi_cmd(struct isert_conn *isert_conn, - struct isert_cmd *isert_cmd, struct iser_rx_desc *rx_desc, - unsigned char *buf) + struct isert_cmd *isert_cmd, struct iscsi_cmd *cmd, + struct iser_rx_desc *rx_desc, unsigned char *buf) { - struct iscsi_cmd *cmd = &isert_cmd->iscsi_cmd; struct iscsi_conn *conn = isert_conn->conn; struct iscsi_scsi_req *hdr = (struct iscsi_scsi_req *)buf; struct scatterlist *sg; @@ -1015,9 +1121,9 @@ isert_handle_iscsi_dataout(struct isert_conn *isert_conn, static int isert_handle_nop_out(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd, - struct iser_rx_desc *rx_desc, unsigned char *buf) + struct iscsi_cmd *cmd, struct iser_rx_desc *rx_desc, + unsigned char *buf) { - struct iscsi_cmd *cmd = &isert_cmd->iscsi_cmd; struct iscsi_conn *conn = isert_conn->conn; struct iscsi_nopout *hdr = (struct iscsi_nopout *)buf; int rc; @@ -1034,9 +1140,9 @@ isert_handle_nop_out(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd, static int isert_handle_text_cmd(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd, - struct iser_rx_desc *rx_desc, struct iscsi_text *hdr) + struct iscsi_cmd *cmd, struct iser_rx_desc *rx_desc, + struct iscsi_text *hdr) { - struct iscsi_cmd *cmd = &isert_cmd->iscsi_cmd; struct iscsi_conn *conn = isert_conn->conn; u32 payload_length = ntoh24(hdr->dlength); int rc; @@ -1081,26 +1187,26 @@ isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc, switch (opcode) { case ISCSI_OP_SCSI_CMD: - cmd = iscsit_allocate_cmd(conn, GFP_KERNEL); + cmd = isert_allocate_cmd(conn, GFP_KERNEL); if (!cmd) break; - isert_cmd = container_of(cmd, struct isert_cmd, iscsi_cmd); + isert_cmd = iscsit_priv_cmd(cmd); isert_cmd->read_stag = read_stag; isert_cmd->read_va = read_va; isert_cmd->write_stag = write_stag; isert_cmd->write_va = write_va; - ret = isert_handle_scsi_cmd(isert_conn, isert_cmd, + ret = isert_handle_scsi_cmd(isert_conn, isert_cmd, cmd, rx_desc, (unsigned char *)hdr); break; case ISCSI_OP_NOOP_OUT: - cmd = iscsit_allocate_cmd(conn, GFP_KERNEL); + cmd = isert_allocate_cmd(conn, GFP_KERNEL); if (!cmd) break; - isert_cmd = container_of(cmd, struct isert_cmd, iscsi_cmd); - ret = isert_handle_nop_out(isert_conn, isert_cmd, + isert_cmd = iscsit_priv_cmd(cmd); + ret = isert_handle_nop_out(isert_conn, isert_cmd, cmd, rx_desc, (unsigned char *)hdr); break; case ISCSI_OP_SCSI_DATA_OUT: @@ -1108,7 +1214,7 @@ isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc, (unsigned char *)hdr); break; case ISCSI_OP_SCSI_TMFUNC: - cmd = iscsit_allocate_cmd(conn, GFP_KERNEL); + cmd = isert_allocate_cmd(conn, GFP_KERNEL); if (!cmd) break; @@ -1116,7 +1222,7 @@ isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc, (unsigned char *)hdr); break; case ISCSI_OP_LOGOUT: - cmd = iscsit_allocate_cmd(conn, GFP_KERNEL); + cmd = isert_allocate_cmd(conn, GFP_KERNEL); if (!cmd) break; @@ -1127,12 +1233,12 @@ isert_rx_opcode(struct isert_conn *isert_conn, struct iser_rx_desc *rx_desc, HZ); break; case ISCSI_OP_TEXT: - cmd = iscsit_allocate_cmd(conn, GFP_KERNEL); + cmd = isert_allocate_cmd(conn, GFP_KERNEL); if (!cmd) break; - isert_cmd = container_of(cmd, struct isert_cmd, iscsi_cmd); - ret = isert_handle_text_cmd(isert_conn, isert_cmd, + isert_cmd = iscsit_priv_cmd(cmd); + ret = isert_handle_text_cmd(isert_conn, isert_cmd, cmd, rx_desc, (struct iscsi_text *)hdr); break; default: @@ -1243,26 +1349,65 @@ isert_unmap_cmd(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn) struct isert_rdma_wr *wr = &isert_cmd->rdma_wr; struct ib_device *ib_dev = isert_conn->conn_cm_id->device; - pr_debug("isert_unmap_cmd >>>>>>>>>>>>>>>>>>>>>>>\n"); + pr_debug("isert_unmap_cmd: %p\n", isert_cmd); + if (wr->sge) { + pr_debug("isert_unmap_cmd: %p unmap_sg op\n", isert_cmd); + ib_dma_unmap_sg(ib_dev, wr->sge, wr->num_sge, + (wr->iser_ib_op == ISER_IB_RDMA_WRITE) ? + DMA_TO_DEVICE : DMA_FROM_DEVICE); + wr->sge = NULL; + } + + if (wr->send_wr) { + pr_debug("isert_unmap_cmd: %p free send_wr\n", isert_cmd); + kfree(wr->send_wr); + wr->send_wr = NULL; + } + + if (wr->ib_sge) { + pr_debug("isert_unmap_cmd: %p free ib_sge\n", isert_cmd); + kfree(wr->ib_sge); + wr->ib_sge = NULL; + } +} + +static void +isert_unreg_rdma_frwr(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn) +{ + struct isert_rdma_wr *wr = &isert_cmd->rdma_wr; + struct ib_device *ib_dev = isert_conn->conn_cm_id->device; + LIST_HEAD(unmap_list); + + pr_debug("unreg_frwr_cmd: %p\n", isert_cmd); + + if (wr->fr_desc) { + pr_debug("unreg_frwr_cmd: %p free fr_desc %p\n", + isert_cmd, wr->fr_desc); + spin_lock_bh(&isert_conn->conn_lock); + list_add_tail(&wr->fr_desc->list, &isert_conn->conn_frwr_pool); + spin_unlock_bh(&isert_conn->conn_lock); + wr->fr_desc = NULL; + } if (wr->sge) { - ib_dma_unmap_sg(ib_dev, wr->sge, wr->num_sge, DMA_TO_DEVICE); + pr_debug("unreg_frwr_cmd: %p unmap_sg op\n", isert_cmd); + ib_dma_unmap_sg(ib_dev, wr->sge, wr->num_sge, + (wr->iser_ib_op == ISER_IB_RDMA_WRITE) ? + DMA_TO_DEVICE : DMA_FROM_DEVICE); wr->sge = NULL; } - kfree(wr->send_wr); + wr->ib_sge = NULL; wr->send_wr = NULL; - - kfree(isert_cmd->ib_sge); - isert_cmd->ib_sge = NULL; } static void isert_put_cmd(struct isert_cmd *isert_cmd) { - struct iscsi_cmd *cmd = &isert_cmd->iscsi_cmd; + struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd; struct isert_conn *isert_conn = isert_cmd->conn; struct iscsi_conn *conn = isert_conn->conn; + struct isert_device *device = isert_conn->conn_device; pr_debug("Entering isert_put_cmd: %p\n", isert_cmd); @@ -1276,7 +1421,7 @@ isert_put_cmd(struct isert_cmd *isert_cmd) if (cmd->data_direction == DMA_TO_DEVICE) iscsit_stop_dataout_timer(cmd); - isert_unmap_cmd(isert_cmd, isert_conn); + device->unreg_rdma_mem(isert_cmd, isert_conn); transport_generic_free_cmd(&cmd->se_cmd, 0); break; case ISCSI_OP_SCSI_TMFUNC: @@ -1311,7 +1456,7 @@ isert_put_cmd(struct isert_cmd *isert_cmd) * Fall-through */ default: - isert_release_cmd(cmd); + iscsit_release_cmd(cmd); break; } } @@ -1347,27 +1492,16 @@ isert_completion_rdma_read(struct iser_tx_desc *tx_desc, struct isert_cmd *isert_cmd) { struct isert_rdma_wr *wr = &isert_cmd->rdma_wr; - struct iscsi_cmd *cmd = &isert_cmd->iscsi_cmd; + struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd; struct se_cmd *se_cmd = &cmd->se_cmd; - struct ib_device *ib_dev = isert_cmd->conn->conn_cm_id->device; + struct isert_conn *isert_conn = isert_cmd->conn; + struct isert_device *device = isert_conn->conn_device; iscsit_stop_dataout_timer(cmd); + device->unreg_rdma_mem(isert_cmd, isert_conn); + cmd->write_data_done = wr->cur_rdma_length; - if (wr->sge) { - pr_debug("isert_do_rdma_read_comp: Unmapping wr->sge from t_data_sg\n"); - ib_dma_unmap_sg(ib_dev, wr->sge, wr->num_sge, DMA_TO_DEVICE); - wr->sge = NULL; - } - - if (isert_cmd->ib_sge) { - pr_debug("isert_do_rdma_read_comp: Freeing isert_cmd->ib_sge\n"); - kfree(isert_cmd->ib_sge); - isert_cmd->ib_sge = NULL; - } - - cmd->write_data_done = se_cmd->data_length; - - pr_debug("isert_do_rdma_read_comp, calling target_execute_cmd\n"); + pr_debug("Cmd: %p RDMA_READ comp calling execute_cmd\n", isert_cmd); spin_lock_bh(&cmd->istate_lock); cmd->cmd_flags |= ICF_GOT_LAST_DATAOUT; cmd->i_state = ISTATE_RECEIVED_LAST_DATAOUT; @@ -1383,7 +1517,7 @@ isert_do_control_comp(struct work_struct *work) struct isert_cmd, comp_work); struct isert_conn *isert_conn = isert_cmd->conn; struct ib_device *ib_dev = isert_conn->conn_cm_id->device; - struct iscsi_cmd *cmd = &isert_cmd->iscsi_cmd; + struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd; switch (cmd->i_state) { case ISTATE_SEND_TASKMGTRSP: @@ -1429,7 +1563,7 @@ isert_response_completion(struct iser_tx_desc *tx_desc, struct isert_conn *isert_conn, struct ib_device *ib_dev) { - struct iscsi_cmd *cmd = &isert_cmd->iscsi_cmd; + struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd; if (cmd->i_state == ISTATE_SEND_TASKMGTRSP || cmd->i_state == ISTATE_SEND_LOGOUTRSP || @@ -1621,8 +1755,7 @@ isert_post_response(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd) static int isert_put_response(struct iscsi_conn *conn, struct iscsi_cmd *cmd) { - struct isert_cmd *isert_cmd = container_of(cmd, - struct isert_cmd, iscsi_cmd); + struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd); struct isert_conn *isert_conn = (struct isert_conn *)conn->context; struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr; struct iscsi_scsi_rsp *hdr = (struct iscsi_scsi_rsp *) @@ -1671,8 +1804,7 @@ static int isert_put_nopin(struct iscsi_cmd *cmd, struct iscsi_conn *conn, bool nopout_response) { - struct isert_cmd *isert_cmd = container_of(cmd, - struct isert_cmd, iscsi_cmd); + struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd); struct isert_conn *isert_conn = (struct isert_conn *)conn->context; struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr; @@ -1691,8 +1823,7 @@ isert_put_nopin(struct iscsi_cmd *cmd, struct iscsi_conn *conn, static int isert_put_logout_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn) { - struct isert_cmd *isert_cmd = container_of(cmd, - struct isert_cmd, iscsi_cmd); + struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd); struct isert_conn *isert_conn = (struct isert_conn *)conn->context; struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr; @@ -1710,8 +1841,7 @@ isert_put_logout_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn) static int isert_put_tm_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn) { - struct isert_cmd *isert_cmd = container_of(cmd, - struct isert_cmd, iscsi_cmd); + struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd); struct isert_conn *isert_conn = (struct isert_conn *)conn->context; struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr; @@ -1729,8 +1859,7 @@ isert_put_tm_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn) static int isert_put_reject(struct iscsi_cmd *cmd, struct iscsi_conn *conn) { - struct isert_cmd *isert_cmd = container_of(cmd, - struct isert_cmd, iscsi_cmd); + struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd); struct isert_conn *isert_conn = (struct isert_conn *)conn->context; struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr; struct ib_device *ib_dev = isert_conn->conn_cm_id->device; @@ -1762,8 +1891,7 @@ isert_put_reject(struct iscsi_cmd *cmd, struct iscsi_conn *conn) static int isert_put_text_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn) { - struct isert_cmd *isert_cmd = container_of(cmd, - struct isert_cmd, iscsi_cmd); + struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd); struct isert_conn *isert_conn = (struct isert_conn *)conn->context; struct ib_send_wr *send_wr = &isert_cmd->tx_desc.send_wr; struct iscsi_text_rsp *hdr = @@ -1805,7 +1933,7 @@ isert_build_rdma_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd, struct ib_sge *ib_sge, struct ib_send_wr *send_wr, u32 data_left, u32 offset) { - struct iscsi_cmd *cmd = &isert_cmd->iscsi_cmd; + struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd; struct scatterlist *sg_start, *tmp_sg; struct ib_device *ib_dev = isert_conn->conn_cm_id->device; u32 sg_off, page_off; @@ -1832,8 +1960,8 @@ isert_build_rdma_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd, ib_sg_dma_len(ib_dev, tmp_sg) - page_off); ib_sge->lkey = isert_conn->conn_mr->lkey; - pr_debug("RDMA ib_sge: addr: 0x%16llx length: %u\n", - ib_sge->addr, ib_sge->length); + pr_debug("RDMA ib_sge: addr: 0x%16llx length: %u lkey: %08x\n", + ib_sge->addr, ib_sge->length, ib_sge->lkey); page_off = 0; data_left -= ib_sge->length; ib_sge++; @@ -1847,200 +1975,373 @@ isert_build_rdma_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd, } static int -isert_put_datain(struct iscsi_conn *conn, struct iscsi_cmd *cmd) +isert_map_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd, + struct isert_rdma_wr *wr) { struct se_cmd *se_cmd = &cmd->se_cmd; - struct isert_cmd *isert_cmd = container_of(cmd, - struct isert_cmd, iscsi_cmd); - struct isert_rdma_wr *wr = &isert_cmd->rdma_wr; + struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd); struct isert_conn *isert_conn = (struct isert_conn *)conn->context; - struct ib_send_wr *wr_failed, *send_wr; struct ib_device *ib_dev = isert_conn->conn_cm_id->device; + struct ib_send_wr *send_wr; struct ib_sge *ib_sge; - struct scatterlist *sg; - u32 offset = 0, data_len, data_left, rdma_write_max; - int rc, ret = 0, count, sg_nents, i, ib_sge_cnt; - - pr_debug("RDMA_WRITE: data_length: %u\n", se_cmd->data_length); + struct scatterlist *sg_start; + u32 sg_off = 0, sg_nents; + u32 offset = 0, data_len, data_left, rdma_write_max, va_offset = 0; + int ret = 0, count, i, ib_sge_cnt; + + if (wr->iser_ib_op == ISER_IB_RDMA_WRITE) { + data_left = se_cmd->data_length; + iscsit_increment_maxcmdsn(cmd, conn->sess); + cmd->stat_sn = conn->stat_sn++; + } else { + sg_off = cmd->write_data_done / PAGE_SIZE; + data_left = se_cmd->data_length - cmd->write_data_done; + offset = cmd->write_data_done; + isert_cmd->tx_desc.isert_cmd = isert_cmd; + } - sg = &se_cmd->t_data_sg[0]; - sg_nents = se_cmd->t_data_nents; + sg_start = &cmd->se_cmd.t_data_sg[sg_off]; + sg_nents = se_cmd->t_data_nents - sg_off; - count = ib_dma_map_sg(ib_dev, sg, sg_nents, DMA_TO_DEVICE); + count = ib_dma_map_sg(ib_dev, sg_start, sg_nents, + (wr->iser_ib_op == ISER_IB_RDMA_WRITE) ? + DMA_TO_DEVICE : DMA_FROM_DEVICE); if (unlikely(!count)) { - pr_err("Unable to map put_datain SGs\n"); + pr_err("Cmd: %p unrable to map SGs\n", isert_cmd); return -EINVAL; } - wr->sge = sg; + wr->sge = sg_start; wr->num_sge = sg_nents; - pr_debug("Mapped IB count: %u sg: %p sg_nents: %u for RDMA_WRITE\n", - count, sg, sg_nents); + wr->cur_rdma_length = data_left; + pr_debug("Mapped cmd: %p count: %u sg: %p sg_nents: %u rdma_len %d\n", + isert_cmd, count, sg_start, sg_nents, data_left); ib_sge = kzalloc(sizeof(struct ib_sge) * sg_nents, GFP_KERNEL); if (!ib_sge) { - pr_warn("Unable to allocate datain ib_sge\n"); + pr_warn("Unable to allocate ib_sge\n"); ret = -ENOMEM; goto unmap_sg; } - isert_cmd->ib_sge = ib_sge; - - pr_debug("Allocated ib_sge: %p from t_data_ents: %d for RDMA_WRITE\n", - ib_sge, se_cmd->t_data_nents); + wr->ib_sge = ib_sge; wr->send_wr_num = DIV_ROUND_UP(sg_nents, isert_conn->max_sge); wr->send_wr = kzalloc(sizeof(struct ib_send_wr) * wr->send_wr_num, GFP_KERNEL); if (!wr->send_wr) { - pr_err("Unable to allocate wr->send_wr\n"); + pr_debug("Unable to allocate wr->send_wr\n"); ret = -ENOMEM; goto unmap_sg; } - pr_debug("Allocated wr->send_wr: %p wr->send_wr_num: %u\n", - wr->send_wr, wr->send_wr_num); - - iscsit_increment_maxcmdsn(cmd, conn->sess); - cmd->stat_sn = conn->stat_sn++; wr->isert_cmd = isert_cmd; rdma_write_max = isert_conn->max_sge * PAGE_SIZE; - data_left = se_cmd->data_length; for (i = 0; i < wr->send_wr_num; i++) { send_wr = &isert_cmd->rdma_wr.send_wr[i]; data_len = min(data_left, rdma_write_max); - send_wr->opcode = IB_WR_RDMA_WRITE; send_wr->send_flags = 0; - send_wr->wr.rdma.remote_addr = isert_cmd->read_va + offset; - send_wr->wr.rdma.rkey = isert_cmd->read_stag; + if (wr->iser_ib_op == ISER_IB_RDMA_WRITE) { + send_wr->opcode = IB_WR_RDMA_WRITE; + send_wr->wr.rdma.remote_addr = isert_cmd->read_va + offset; + send_wr->wr.rdma.rkey = isert_cmd->read_stag; + if (i + 1 == wr->send_wr_num) + send_wr->next = &isert_cmd->tx_desc.send_wr; + else + send_wr->next = &wr->send_wr[i + 1]; + } else { + send_wr->opcode = IB_WR_RDMA_READ; + send_wr->wr.rdma.remote_addr = isert_cmd->write_va + va_offset; + send_wr->wr.rdma.rkey = isert_cmd->write_stag; + if (i + 1 == wr->send_wr_num) + send_wr->send_flags = IB_SEND_SIGNALED; + else + send_wr->next = &wr->send_wr[i + 1]; + } ib_sge_cnt = isert_build_rdma_wr(isert_conn, isert_cmd, ib_sge, send_wr, data_len, offset); ib_sge += ib_sge_cnt; - if (i + 1 == wr->send_wr_num) - send_wr->next = &isert_cmd->tx_desc.send_wr; - else - send_wr->next = &wr->send_wr[i + 1]; - offset += data_len; + va_offset += data_len; data_left -= data_len; } - /* - * Build isert_conn->tx_desc for iSCSI response PDU and attach - */ - isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc); - iscsit_build_rsp_pdu(cmd, conn, false, (struct iscsi_scsi_rsp *) - &isert_cmd->tx_desc.iscsi_header); - isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc); - isert_init_send_wr(isert_cmd, &isert_cmd->tx_desc.send_wr); - atomic_inc(&isert_conn->post_send_buf_count); + return 0; +unmap_sg: + ib_dma_unmap_sg(ib_dev, sg_start, sg_nents, + (wr->iser_ib_op == ISER_IB_RDMA_WRITE) ? + DMA_TO_DEVICE : DMA_FROM_DEVICE); + return ret; +} - rc = ib_post_send(isert_conn->conn_qp, wr->send_wr, &wr_failed); - if (rc) { - pr_warn("ib_post_send() failed for IB_WR_RDMA_WRITE\n"); - atomic_dec(&isert_conn->post_send_buf_count); +static int +isert_map_fr_pagelist(struct ib_device *ib_dev, + struct scatterlist *sg_start, int sg_nents, u64 *fr_pl) +{ + u64 start_addr, end_addr, page, chunk_start = 0; + struct scatterlist *tmp_sg; + int i = 0, new_chunk, last_ent, n_pages; + + n_pages = 0; + new_chunk = 1; + last_ent = sg_nents - 1; + for_each_sg(sg_start, tmp_sg, sg_nents, i) { + start_addr = ib_sg_dma_address(ib_dev, tmp_sg); + if (new_chunk) + chunk_start = start_addr; + end_addr = start_addr + ib_sg_dma_len(ib_dev, tmp_sg); + + pr_debug("SGL[%d] dma_addr: 0x%16llx len: %u\n", + i, (unsigned long long)tmp_sg->dma_address, + tmp_sg->length); + + if ((end_addr & ~PAGE_MASK) && i < last_ent) { + new_chunk = 0; + continue; + } + new_chunk = 1; + + page = chunk_start & PAGE_MASK; + do { + fr_pl[n_pages++] = page; + pr_debug("Mapped page_list[%d] page_addr: 0x%16llx\n", + n_pages - 1, page); + page += PAGE_SIZE; + } while (page < end_addr); } - pr_debug("Posted RDMA_WRITE + Response for iSER Data READ\n"); - return 1; -unmap_sg: - ib_dma_unmap_sg(ib_dev, sg, sg_nents, DMA_TO_DEVICE); + return n_pages; +} + +static int +isert_fast_reg_mr(struct fast_reg_descriptor *fr_desc, + struct isert_cmd *isert_cmd, struct isert_conn *isert_conn, + struct ib_sge *ib_sge, u32 offset, unsigned int data_len) +{ + struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd; + struct ib_device *ib_dev = isert_conn->conn_cm_id->device; + struct scatterlist *sg_start; + u32 sg_off, page_off; + struct ib_send_wr fr_wr, inv_wr; + struct ib_send_wr *bad_wr, *wr = NULL; + u8 key; + int ret, sg_nents, pagelist_len; + + sg_off = offset / PAGE_SIZE; + sg_start = &cmd->se_cmd.t_data_sg[sg_off]; + sg_nents = min_t(unsigned int, cmd->se_cmd.t_data_nents - sg_off, + ISCSI_ISER_SG_TABLESIZE); + page_off = offset % PAGE_SIZE; + + pr_debug("Cmd: %p use fr_desc %p sg_nents %d sg_off %d offset %u\n", + isert_cmd, fr_desc, sg_nents, sg_off, offset); + + pagelist_len = isert_map_fr_pagelist(ib_dev, sg_start, sg_nents, + &fr_desc->data_frpl->page_list[0]); + + if (!fr_desc->valid) { + memset(&inv_wr, 0, sizeof(inv_wr)); + inv_wr.opcode = IB_WR_LOCAL_INV; + inv_wr.ex.invalidate_rkey = fr_desc->data_mr->rkey; + wr = &inv_wr; + /* Bump the key */ + key = (u8)(fr_desc->data_mr->rkey & 0x000000FF); + ib_update_fast_reg_key(fr_desc->data_mr, ++key); + } + + /* Prepare FASTREG WR */ + memset(&fr_wr, 0, sizeof(fr_wr)); + fr_wr.opcode = IB_WR_FAST_REG_MR; + fr_wr.wr.fast_reg.iova_start = + fr_desc->data_frpl->page_list[0] + page_off; + fr_wr.wr.fast_reg.page_list = fr_desc->data_frpl; + fr_wr.wr.fast_reg.page_list_len = pagelist_len; + fr_wr.wr.fast_reg.page_shift = PAGE_SHIFT; + fr_wr.wr.fast_reg.length = data_len; + fr_wr.wr.fast_reg.rkey = fr_desc->data_mr->rkey; + fr_wr.wr.fast_reg.access_flags = IB_ACCESS_LOCAL_WRITE; + + if (!wr) + wr = &fr_wr; + else + wr->next = &fr_wr; + + ret = ib_post_send(isert_conn->conn_qp, wr, &bad_wr); + if (ret) { + pr_err("fast registration failed, ret:%d\n", ret); + return ret; + } + fr_desc->valid = false; + + ib_sge->lkey = fr_desc->data_mr->lkey; + ib_sge->addr = fr_desc->data_frpl->page_list[0] + page_off; + ib_sge->length = data_len; + + pr_debug("RDMA ib_sge: addr: 0x%16llx length: %u lkey: %08x\n", + ib_sge->addr, ib_sge->length, ib_sge->lkey); + return ret; } static int -isert_get_dataout(struct iscsi_conn *conn, struct iscsi_cmd *cmd, bool recovery) +isert_reg_rdma_frwr(struct iscsi_conn *conn, struct iscsi_cmd *cmd, + struct isert_rdma_wr *wr) { struct se_cmd *se_cmd = &cmd->se_cmd; - struct isert_cmd *isert_cmd = container_of(cmd, - struct isert_cmd, iscsi_cmd); - struct isert_rdma_wr *wr = &isert_cmd->rdma_wr; + struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd); struct isert_conn *isert_conn = (struct isert_conn *)conn->context; - struct ib_send_wr *wr_failed, *send_wr; - struct ib_sge *ib_sge; struct ib_device *ib_dev = isert_conn->conn_cm_id->device; + struct ib_send_wr *send_wr; + struct ib_sge *ib_sge; struct scatterlist *sg_start; - u32 sg_off, sg_nents, page_off, va_offset = 0; + struct fast_reg_descriptor *fr_desc; + u32 sg_off = 0, sg_nents; u32 offset = 0, data_len, data_left, rdma_write_max; - int rc, ret = 0, count, i, ib_sge_cnt; + int ret = 0, count; + unsigned long flags; - pr_debug("RDMA_READ: data_length: %u write_data_done: %u\n", - se_cmd->data_length, cmd->write_data_done); + if (wr->iser_ib_op == ISER_IB_RDMA_WRITE) { + data_left = se_cmd->data_length; + iscsit_increment_maxcmdsn(cmd, conn->sess); + cmd->stat_sn = conn->stat_sn++; + } else { + sg_off = cmd->write_data_done / PAGE_SIZE; + data_left = se_cmd->data_length - cmd->write_data_done; + offset = cmd->write_data_done; + isert_cmd->tx_desc.isert_cmd = isert_cmd; + } - sg_off = cmd->write_data_done / PAGE_SIZE; sg_start = &cmd->se_cmd.t_data_sg[sg_off]; - page_off = cmd->write_data_done % PAGE_SIZE; - - pr_debug("RDMA_READ: sg_off: %d, sg_start: %p page_off: %d\n", - sg_off, sg_start, page_off); - - data_left = se_cmd->data_length - cmd->write_data_done; sg_nents = se_cmd->t_data_nents - sg_off; - pr_debug("RDMA_READ: data_left: %d, sg_nents: %d\n", - data_left, sg_nents); - - count = ib_dma_map_sg(ib_dev, sg_start, sg_nents, DMA_FROM_DEVICE); + count = ib_dma_map_sg(ib_dev, sg_start, sg_nents, + (wr->iser_ib_op == ISER_IB_RDMA_WRITE) ? + DMA_TO_DEVICE : DMA_FROM_DEVICE); if (unlikely(!count)) { - pr_err("Unable to map get_dataout SGs\n"); + pr_err("Cmd: %p unrable to map SGs\n", isert_cmd); return -EINVAL; } wr->sge = sg_start; wr->num_sge = sg_nents; - pr_debug("Mapped IB count: %u sg_start: %p sg_nents: %u for RDMA_READ\n", - count, sg_start, sg_nents); + pr_debug("Mapped cmd: %p count: %u sg: %p sg_nents: %u rdma_len %d\n", + isert_cmd, count, sg_start, sg_nents, data_left); - ib_sge = kzalloc(sizeof(struct ib_sge) * sg_nents, GFP_KERNEL); - if (!ib_sge) { - pr_warn("Unable to allocate dataout ib_sge\n"); - ret = -ENOMEM; - goto unmap_sg; + memset(&wr->s_ib_sge, 0, sizeof(*ib_sge)); + ib_sge = &wr->s_ib_sge; + wr->ib_sge = ib_sge; + + wr->send_wr_num = 1; + memset(&wr->s_send_wr, 0, sizeof(*send_wr)); + wr->send_wr = &wr->s_send_wr; + + wr->isert_cmd = isert_cmd; + rdma_write_max = ISCSI_ISER_SG_TABLESIZE * PAGE_SIZE; + + send_wr = &isert_cmd->rdma_wr.s_send_wr; + send_wr->sg_list = ib_sge; + send_wr->num_sge = 1; + send_wr->wr_id = (unsigned long)&isert_cmd->tx_desc; + if (wr->iser_ib_op == ISER_IB_RDMA_WRITE) { + send_wr->opcode = IB_WR_RDMA_WRITE; + send_wr->wr.rdma.remote_addr = isert_cmd->read_va; + send_wr->wr.rdma.rkey = isert_cmd->read_stag; + send_wr->send_flags = 0; + send_wr->next = &isert_cmd->tx_desc.send_wr; + } else { + send_wr->opcode = IB_WR_RDMA_READ; + send_wr->wr.rdma.remote_addr = isert_cmd->write_va; + send_wr->wr.rdma.rkey = isert_cmd->write_stag; + send_wr->send_flags = IB_SEND_SIGNALED; } - isert_cmd->ib_sge = ib_sge; - pr_debug("Using ib_sge: %p from sg_ents: %d for RDMA_READ\n", - ib_sge, sg_nents); + data_len = min(data_left, rdma_write_max); + wr->cur_rdma_length = data_len; - wr->send_wr_num = DIV_ROUND_UP(sg_nents, isert_conn->max_sge); - wr->send_wr = kzalloc(sizeof(struct ib_send_wr) * wr->send_wr_num, - GFP_KERNEL); - if (!wr->send_wr) { - pr_debug("Unable to allocate wr->send_wr\n"); - ret = -ENOMEM; + spin_lock_irqsave(&isert_conn->conn_lock, flags); + fr_desc = list_first_entry(&isert_conn->conn_frwr_pool, + struct fast_reg_descriptor, list); + list_del(&fr_desc->list); + spin_unlock_irqrestore(&isert_conn->conn_lock, flags); + wr->fr_desc = fr_desc; + + ret = isert_fast_reg_mr(fr_desc, isert_cmd, isert_conn, + ib_sge, offset, data_len); + if (ret) { + list_add_tail(&fr_desc->list, &isert_conn->conn_frwr_pool); goto unmap_sg; } - pr_debug("Allocated wr->send_wr: %p wr->send_wr_num: %u\n", - wr->send_wr, wr->send_wr_num); - isert_cmd->tx_desc.isert_cmd = isert_cmd; + return 0; - wr->iser_ib_op = ISER_IB_RDMA_READ; - wr->isert_cmd = isert_cmd; - rdma_write_max = isert_conn->max_sge * PAGE_SIZE; - offset = cmd->write_data_done; +unmap_sg: + ib_dma_unmap_sg(ib_dev, sg_start, sg_nents, + (wr->iser_ib_op == ISER_IB_RDMA_WRITE) ? + DMA_TO_DEVICE : DMA_FROM_DEVICE); + return ret; +} - for (i = 0; i < wr->send_wr_num; i++) { - send_wr = &isert_cmd->rdma_wr.send_wr[i]; - data_len = min(data_left, rdma_write_max); +static int +isert_put_datain(struct iscsi_conn *conn, struct iscsi_cmd *cmd) +{ + struct se_cmd *se_cmd = &cmd->se_cmd; + struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd); + struct isert_rdma_wr *wr = &isert_cmd->rdma_wr; + struct isert_conn *isert_conn = (struct isert_conn *)conn->context; + struct isert_device *device = isert_conn->conn_device; + struct ib_send_wr *wr_failed; + int rc; - send_wr->opcode = IB_WR_RDMA_READ; - send_wr->wr.rdma.remote_addr = isert_cmd->write_va + va_offset; - send_wr->wr.rdma.rkey = isert_cmd->write_stag; + pr_debug("Cmd: %p RDMA_WRITE data_length: %u\n", + isert_cmd, se_cmd->data_length); + wr->iser_ib_op = ISER_IB_RDMA_WRITE; + rc = device->reg_rdma_mem(conn, cmd, wr); + if (rc) { + pr_err("Cmd: %p failed to prepare RDMA res\n", isert_cmd); + return rc; + } - ib_sge_cnt = isert_build_rdma_wr(isert_conn, isert_cmd, ib_sge, - send_wr, data_len, offset); - ib_sge += ib_sge_cnt; + /* + * Build isert_conn->tx_desc for iSCSI response PDU and attach + */ + isert_create_send_desc(isert_conn, isert_cmd, &isert_cmd->tx_desc); + iscsit_build_rsp_pdu(cmd, conn, false, (struct iscsi_scsi_rsp *) + &isert_cmd->tx_desc.iscsi_header); + isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc); + isert_init_send_wr(isert_cmd, &isert_cmd->tx_desc.send_wr); - if (i + 1 == wr->send_wr_num) - send_wr->send_flags = IB_SEND_SIGNALED; - else - send_wr->next = &wr->send_wr[i + 1]; + atomic_inc(&isert_conn->post_send_buf_count); - offset += data_len; - va_offset += data_len; - data_left -= data_len; + rc = ib_post_send(isert_conn->conn_qp, wr->send_wr, &wr_failed); + if (rc) { + pr_warn("ib_post_send() failed for IB_WR_RDMA_WRITE\n"); + atomic_dec(&isert_conn->post_send_buf_count); + } + pr_debug("Cmd: %p posted RDMA_WRITE + Response for iSER Data READ\n", + isert_cmd); + + return 1; +} + +static int +isert_get_dataout(struct iscsi_conn *conn, struct iscsi_cmd *cmd, bool recovery) +{ + struct se_cmd *se_cmd = &cmd->se_cmd; + struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd); + struct isert_rdma_wr *wr = &isert_cmd->rdma_wr; + struct isert_conn *isert_conn = (struct isert_conn *)conn->context; + struct isert_device *device = isert_conn->conn_device; + struct ib_send_wr *wr_failed; + int rc; + + pr_debug("Cmd: %p RDMA_READ data_length: %u write_data_done: %u\n", + isert_cmd, se_cmd->data_length, cmd->write_data_done); + wr->iser_ib_op = ISER_IB_RDMA_READ; + rc = device->reg_rdma_mem(conn, cmd, wr); + if (rc) { + pr_err("Cmd: %p failed to prepare RDMA res\n", isert_cmd); + return rc; } atomic_inc(&isert_conn->post_send_buf_count); @@ -2050,12 +2351,10 @@ isert_get_dataout(struct iscsi_conn *conn, struct iscsi_cmd *cmd, bool recovery) pr_warn("ib_post_send() failed for IB_WR_RDMA_READ\n"); atomic_dec(&isert_conn->post_send_buf_count); } - pr_debug("Posted RDMA_READ memory for ISER Data WRITE\n"); - return 0; + pr_debug("Cmd: %p posted RDMA_READ memory for ISER Data WRITE\n", + isert_cmd); -unmap_sg: - ib_dma_unmap_sg(ib_dev, sg_start, sg_nents, DMA_FROM_DEVICE); - return ret; + return 0; } static int @@ -2224,6 +2523,14 @@ isert_get_login_rx(struct iscsi_conn *conn, struct iscsi_login *login) int ret; pr_debug("isert_get_login_rx before conn_login_comp conn: %p\n", conn); + /* + * For login requests after the first PDU, isert_rx_login_req() will + * kick schedule_delayed_work(&conn->login_work) as the packet is + * received, which turns this callback from iscsi_target_do_login_rx() + * into a NOP. + */ + if (!login->first_request) + return 0; ret = wait_for_completion_interruptible(&isert_conn->conn_login_comp); if (ret) @@ -2393,12 +2700,12 @@ static void isert_free_conn(struct iscsi_conn *conn) static struct iscsit_transport iser_target_transport = { .name = "IB/iSER", .transport_type = ISCSI_INFINIBAND, + .priv_size = sizeof(struct isert_cmd), .owner = THIS_MODULE, .iscsit_setup_np = isert_setup_np, .iscsit_accept_np = isert_accept_np, .iscsit_free_np = isert_free_np, .iscsit_free_conn = isert_free_conn, - .iscsit_alloc_cmd = isert_alloc_cmd, .iscsit_get_login_rx = isert_get_login_rx, .iscsit_put_login_tx = isert_put_login_tx, .iscsit_immediate_queue = isert_immediate_queue, @@ -2425,21 +2732,10 @@ static int __init isert_init(void) goto destroy_rx_wq; } - isert_cmd_cache = kmem_cache_create("isert_cmd_cache", - sizeof(struct isert_cmd), __alignof__(struct isert_cmd), - 0, NULL); - if (!isert_cmd_cache) { - pr_err("Unable to create isert_cmd_cache\n"); - ret = -ENOMEM; - goto destroy_tx_cq; - } - iscsit_register_transport(&iser_target_transport); pr_debug("iSER_TARGET[0] - Loaded iser_target_transport\n"); return 0; -destroy_tx_cq: - destroy_workqueue(isert_comp_wq); destroy_rx_wq: destroy_workqueue(isert_rx_wq); return ret; @@ -2447,7 +2743,6 @@ destroy_rx_wq: static void __exit isert_exit(void) { - kmem_cache_destroy(isert_cmd_cache); destroy_workqueue(isert_comp_wq); destroy_workqueue(isert_rx_wq); iscsit_unregister_transport(&iser_target_transport); diff --git a/drivers/infiniband/ulp/isert/ib_isert.h b/drivers/infiniband/ulp/isert/ib_isert.h index 191117b5b50..631f2090f0b 100644 --- a/drivers/infiniband/ulp/isert/ib_isert.h +++ b/drivers/infiniband/ulp/isert/ib_isert.h @@ -5,6 +5,7 @@ #include <rdma/rdma_cm.h> #define ISERT_RDMA_LISTEN_BACKLOG 10 +#define ISCSI_ISER_SG_TABLESIZE 256 enum isert_desc_type { ISCSI_TX_CONTROL, @@ -45,15 +46,26 @@ struct iser_tx_desc { struct ib_send_wr send_wr; } __packed; +struct fast_reg_descriptor { + struct list_head list; + struct ib_mr *data_mr; + struct ib_fast_reg_page_list *data_frpl; + bool valid; +}; + struct isert_rdma_wr { struct list_head wr_list; struct isert_cmd *isert_cmd; enum iser_ib_op_code iser_ib_op; struct ib_sge *ib_sge; + struct ib_sge s_ib_sge; int num_sge; struct scatterlist *sge; int send_wr_num; struct ib_send_wr *send_wr; + struct ib_send_wr s_send_wr; + u32 cur_rdma_length; + struct fast_reg_descriptor *fr_desc; }; struct isert_cmd { @@ -67,8 +79,7 @@ struct isert_cmd { u32 write_va_off; u32 rdma_wr_num; struct isert_conn *conn; - struct iscsi_cmd iscsi_cmd; - struct ib_sge *ib_sge; + struct iscsi_cmd *iscsi_cmd; struct iser_tx_desc tx_desc; struct isert_rdma_wr rdma_wr; struct work_struct comp_work; @@ -106,6 +117,10 @@ struct isert_conn { wait_queue_head_t conn_wait; wait_queue_head_t conn_wait_comp_err; struct kref conn_kref; + struct list_head conn_frwr_pool; + int conn_frwr_pool_size; + /* lock to protect frwr_pool */ + spinlock_t conn_lock; }; #define ISERT_MAX_CQ 64 @@ -118,6 +133,7 @@ struct isert_cq_desc { }; struct isert_device { + int use_frwr; int cqs_used; int refcount; int cq_active_qps[ISERT_MAX_CQ]; @@ -128,6 +144,12 @@ struct isert_device { struct ib_cq *dev_tx_cq[ISERT_MAX_CQ]; struct isert_cq_desc *cq_desc; struct list_head dev_node; + struct ib_device_attr dev_attr; + int (*reg_rdma_mem)(struct iscsi_conn *conn, + struct iscsi_cmd *cmd, + struct isert_rdma_wr *wr); + void (*unreg_rdma_mem)(struct isert_cmd *isert_cmd, + struct isert_conn *isert_conn); }; struct isert_np { |