This has us use the blk/scsi-ml mq cmd pre-allocator and blk tags for scsi tasks. We now do not need the back/frwd locks for scsi task allocation. We still need it for mgmt tasks, but that will be fixed in the next patches. Signed-off-by: Mike Christie <michael.christie@xxxxxxxxxx> --- drivers/scsi/cxgbi/libcxgbi.c | 2 +- drivers/scsi/libiscsi.c | 151 +++++++++++++++++++--------------- include/scsi/libiscsi.h | 46 +++++++---- 3 files changed, 116 insertions(+), 83 deletions(-) diff --git a/drivers/scsi/cxgbi/libcxgbi.c b/drivers/scsi/cxgbi/libcxgbi.c index 919451810018..cdaa67fd8c2e 100644 --- a/drivers/scsi/cxgbi/libcxgbi.c +++ b/drivers/scsi/cxgbi/libcxgbi.c @@ -366,7 +366,7 @@ int cxgbi_hbas_add(struct cxgbi_device *cdev, u64 max_lun, chba->ndev = cdev->ports[i]; chba->shost = shost; - shost->can_queue = sht->can_queue - ISCSI_MGMT_CMDS_MAX; + shost->can_queue = sht->can_queue - ISCSI_INFLIGHT_MGMT_MAX; log_debug(1 << CXGBI_DBG_DEV, "cdev 0x%p, p#%d %s: chba 0x%p.\n", diff --git a/drivers/scsi/libiscsi.c b/drivers/scsi/libiscsi.c index cadbe1d19344..46ab51e5a87b 100644 --- a/drivers/scsi/libiscsi.c +++ b/drivers/scsi/libiscsi.c @@ -463,9 +463,9 @@ static void iscsi_free_task(struct iscsi_task *task) if (conn->login_task == task) return; - kfifo_in(&session->cmdpool.queue, (void*)&task, sizeof(void*)); - - if (sc) { + if (!sc) { + kfifo_in(&session->mgmt_pool.queue, (void *)&task, sizeof(void *)); + } else { /* SCSI eh reuses commands to verify us */ sc->SCp.ptr = NULL; /* @@ -716,8 +716,8 @@ __iscsi_conn_send_pdu(struct iscsi_conn *conn, struct iscsi_hdr *hdr, BUG_ON(conn->c_stage == ISCSI_CONN_INITIAL_STAGE); BUG_ON(conn->c_stage == ISCSI_CONN_STOPPED); - if (!kfifo_out(&session->cmdpool.queue, - (void*)&task, sizeof(void*))) + if (!kfifo_out(&session->mgmt_pool.queue, (void *)&task, + sizeof(void *))) return NULL; } /* @@ -1118,7 +1118,7 @@ static int iscsi_handle_reject(struct iscsi_conn *conn, struct iscsi_hdr *hdr, struct iscsi_task *iscsi_itt_to_task(struct iscsi_conn *conn, itt_t itt) { struct iscsi_session *session = conn->session; - int i; + uint32_t i; if (itt == RESERVED_ITT) return NULL; @@ -1127,10 +1127,17 @@ struct iscsi_task *iscsi_itt_to_task(struct iscsi_conn *conn, itt_t itt) session->tt->parse_pdu_itt(conn, itt, &i, NULL); else i = get_itt(itt); - if (i >= session->cmds_max) - return NULL; - return session->cmds[i]; + if (i & ISCSI_TASK_TYPE_MGMT) { + i &= ~ISCSI_TASK_TYPE_MGMT; + + if (i >= ISCSI_MGMT_CMDS_MAX) + return NULL; + + return session->mgmt_cmds[i]; + } else { + return iscsi_itt_to_ctask(conn, itt); + } } EXPORT_SYMBOL_GPL(iscsi_itt_to_task); @@ -1348,12 +1355,6 @@ int iscsi_verify_itt(struct iscsi_conn *conn, itt_t itt) return ISCSI_ERR_BAD_ITT; } - if (i >= session->cmds_max) { - iscsi_conn_printk(KERN_ERR, conn, - "received invalid itt index %u (max cmds " - "%u.\n", i, session->cmds_max); - return ISCSI_ERR_BAD_ITT; - } return 0; } EXPORT_SYMBOL_GPL(iscsi_verify_itt); @@ -1369,19 +1370,30 @@ EXPORT_SYMBOL_GPL(iscsi_verify_itt); */ struct iscsi_task *iscsi_itt_to_ctask(struct iscsi_conn *conn, itt_t itt) { + struct iscsi_session *session = conn->session; struct iscsi_task *task; + struct scsi_cmnd *sc; + int tag; if (iscsi_verify_itt(conn, itt)) return NULL; - task = iscsi_itt_to_task(conn, itt); - if (!task || !task->sc) + if (session->tt->parse_pdu_itt) + session->tt->parse_pdu_itt(conn, itt, &tag, NULL); + else + tag = get_itt(itt); + sc = scsi_host_find_tag(session->host, tag); + if (!sc) return NULL; - if (task->sc->SCp.phase != conn->session->age) { + task = scsi_cmd_priv(sc); + if (!task->sc) + return NULL; + + if (task->sc->SCp.phase != session->age) { iscsi_session_printk(KERN_ERR, conn->session, "task's session age %d, expected %d\n", - task->sc->SCp.phase, conn->session->age); + task->sc->SCp.phase, session->age); return NULL; } @@ -1679,19 +1691,16 @@ static void iscsi_xmitworker(struct work_struct *work) } while (rc >= 0 || rc == -EAGAIN); } -static inline struct iscsi_task *iscsi_alloc_task(struct iscsi_conn *conn, - struct scsi_cmnd *sc) +static struct iscsi_task *iscsi_init_scsi_task(struct iscsi_conn *conn, + struct scsi_cmnd *sc) { - struct iscsi_task *task; - - if (!kfifo_out(&conn->session->cmdpool.queue, - (void *) &task, sizeof(void *))) - return NULL; + struct iscsi_task *task = scsi_cmd_priv(sc); sc->SCp.phase = conn->session->age; sc->SCp.ptr = (char *) task; refcount_set(&task->refcount, 1); + task->itt = blk_mq_unique_tag(sc->request); task->state = ISCSI_TASK_PENDING; task->conn = conn; task->sc = sc; @@ -1805,12 +1814,7 @@ int iscsi_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *sc) goto reject; } - task = iscsi_alloc_task(conn, sc); - if (!task) { - spin_unlock_bh(&session->frwd_lock); - reason = FAILURE_OOM; - goto reject; - } + task = iscsi_init_scsi_task(conn, sc); if (!ihost->workq) { reason = iscsi_prep_scsi_cmd_pdu(task); @@ -2745,9 +2749,8 @@ int iscsi_host_get_max_scsi_cmds(struct Scsi_Host *shost, if (!total_cmds) total_cmds = ISCSI_DEF_XMIT_CMDS_MAX; /* - * The iscsi layer needs some tasks for nop handling and tmfs, - * so the cmds_max must at least be greater than ISCSI_MGMT_CMDS_MAX - * + 1 command for scsi IO. + * The iscsi layer needs some tasks for nop handling and tmfs, so the + * cmds_max must at least be greater than ISCSI_INFLIGHT_MGMT_MAX */ if (total_cmds < ISCSI_TOTAL_CMDS_MIN) { printk(KERN_ERR "iscsi: invalid max cmds of %d. Must be a power of two that is at least %d.\n", @@ -2773,7 +2776,7 @@ int iscsi_host_get_max_scsi_cmds(struct Scsi_Host *shost, requested_cmds_max, total_cmds); } - scsi_cmds = total_cmds - ISCSI_MGMT_CMDS_MAX; + scsi_cmds = total_cmds - ISCSI_INFLIGHT_MGMT_MAX; if (shost->can_queue && scsi_cmds > shost->can_queue) { total_cmds = shost->can_queue; @@ -2799,9 +2802,25 @@ int iscsi_host_add(struct Scsi_Host *shost, struct device *pdev) if (!shost->can_queue) shost->can_queue = ISCSI_DEF_XMIT_CMDS_MAX; + if (shost->can_queue > ISCSI_TOTAL_CMDS_MAX) { + shost_printk(KERN_INFO, shost, + "Requested scsi host can_queue was %u. Limiting to %u\n", + shost->can_queue, ISCSI_TOTAL_CMDS_MAX); + shost->can_queue = ISCSI_TOTAL_CMDS_MAX; + } + if (!shost->cmd_per_lun) shost->cmd_per_lun = ISCSI_DEF_CMD_PER_LUN; + shost->host_tagset = 1; + /* + * We currently do not support nr_hw_queues > 1 because the iscsi spec + * itt is limited to 32 bits, and for drivers that support it, libiscsi + * uses some of the bits past BLK_MQ_UNIQUE_TAG_BITS for target sanity + * checks. Plus other drivers/fw can limit the itt to less than 16 bits. + */ + BUG_ON(shost->nr_hw_queues > 1); + return scsi_add_host(shost, pdev); } EXPORT_SYMBOL_GPL(iscsi_host_add); @@ -2914,9 +2933,11 @@ static void iscsi_host_dec_session_cnt(struct Scsi_Host *shost) scsi_host_put(shost); } -static void iscsi_init_task(struct iscsi_task *task) +static void iscsi_init_task(struct iscsi_task *task, int dd_task_size) { task->dd_data = &task[1]; + if (dd_task_size > 0) + memset(task->dd_data, 0, dd_task_size); task->itt = ISCSI_RESERVED_TAG; task->state = ISCSI_TASK_FREE; INIT_LIST_HEAD(&task->running); @@ -2926,7 +2947,7 @@ int iscsi_init_cmd_priv(struct Scsi_Host *shost, struct scsi_cmnd *sc) { struct iscsi_task *task = scsi_cmd_priv(sc); - iscsi_init_task(task); + iscsi_init_task(task, shost->hostt->cmd_size - sizeof(*task)); return 0; } EXPORT_SYMBOL_GPL(iscsi_init_cmd_priv); @@ -2945,8 +2966,8 @@ EXPORT_SYMBOL_GPL(iscsi_init_cmd_priv); * a session per scsi host. * * Callers should set cmds_max to the largest total numer (mgmt + scsi) of - * tasks they support. The iscsi layer reserves ISCSI_MGMT_CMDS_MAX tasks - * for nop handling and login/logout requests. + * tasks they support. The iscsi layer reserves ISCSI_INFLIGHT_MGMT_MAX + * tasks for nop handling and login/logout requests. */ struct iscsi_cls_session * iscsi_session_setup(struct iscsi_transport *iscsit, struct Scsi_Host *shost, @@ -2985,7 +3006,7 @@ iscsi_session_setup(struct iscsi_transport *iscsit, struct Scsi_Host *shost, session->lu_reset_timeout = 15; session->abort_timeout = 10; session->scsi_cmds_max = scsi_cmds; - session->cmds_max = scsi_cmds + ISCSI_MGMT_CMDS_MAX; + session->cmds_max = scsi_cmds + ISCSI_INFLIGHT_MGMT_MAX; session->queued_cmdsn = session->cmdsn = initial_cmdsn; session->exp_cmdsn = initial_cmdsn + 1; session->max_cmdsn = initial_cmdsn + 1; @@ -2997,21 +3018,18 @@ iscsi_session_setup(struct iscsi_transport *iscsit, struct Scsi_Host *shost, spin_lock_init(&session->frwd_lock); spin_lock_init(&session->back_lock); - /* initialize SCSI PDU commands pool */ - if (iscsi_pool_init(&session->cmdpool, session->cmds_max, - (void***)&session->cmds, + /* initialize mgmt task pool */ + if (iscsi_pool_init(&session->mgmt_pool, ISCSI_MGMT_CMDS_MAX, + (void ***)&session->mgmt_cmds, cmd_task_size + sizeof(struct iscsi_task))) - goto cmdpool_alloc_fail; + goto mgmt_pool_alloc_fail; /* pre-format cmds pool with ITT */ - for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) { - struct iscsi_task *task = session->cmds[cmd_i]; + for (cmd_i = 0; cmd_i < ISCSI_MGMT_CMDS_MAX; cmd_i++) { + struct iscsi_task *task = session->mgmt_cmds[cmd_i]; - if (cmd_task_size) - task->dd_data = &task[1]; - task->itt = cmd_i; - task->state = ISCSI_TASK_FREE; - INIT_LIST_HEAD(&task->running); + iscsi_init_task(task, cmd_task_size); + task->itt = cmd_i | ISCSI_TASK_TYPE_MGMT; if (iscsit->alloc_task_priv) { if (iscsit->alloc_task_priv(session, task)) @@ -3032,11 +3050,11 @@ iscsi_session_setup(struct iscsi_transport *iscsit, struct Scsi_Host *shost, free_task_priv: for (cmd_i--; cmd_i >= 0; cmd_i--) { if (iscsit->free_task_priv) - iscsit->free_task_priv(session, session->cmds[cmd_i]); + iscsit->free_task_priv(session, session->mgmt_cmds[cmd_i]); } - iscsi_pool_free(&session->cmdpool); -cmdpool_alloc_fail: + iscsi_pool_free(&session->mgmt_pool); +mgmt_pool_alloc_fail: iscsi_free_session(cls_session); dec_session_count: iscsi_host_dec_session_cnt(shost); @@ -3055,13 +3073,13 @@ void iscsi_session_teardown(struct iscsi_cls_session *cls_session) struct Scsi_Host *shost = session->host; int cmd_i; - for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) { + for (cmd_i = 0; cmd_i < ISCSI_MGMT_CMDS_MAX; cmd_i++) { if (session->tt->free_task_priv) session->tt->free_task_priv(session, - session->cmds[cmd_i]); + session->mgmt_cmds[cmd_i]); } - iscsi_pool_free(&session->cmdpool); + iscsi_pool_free(&session->mgmt_pool); iscsi_remove_session(cls_session); @@ -3125,9 +3143,8 @@ iscsi_conn_setup(struct iscsi_cls_session *cls_session, int dd_size, /* allocate login_task used for the login/text sequences */ spin_lock_bh(&session->frwd_lock); - if (!kfifo_out(&session->cmdpool.queue, - (void*)&conn->login_task, - sizeof(void*))) { + if (!kfifo_out(&session->mgmt_pool.queue, (void *)&conn->login_task, + sizeof(void *))) { spin_unlock_bh(&session->frwd_lock); goto login_task_alloc_fail; } @@ -3145,8 +3162,8 @@ iscsi_conn_setup(struct iscsi_cls_session *cls_session, int dd_size, return cls_conn; login_task_data_alloc_fail: - kfifo_in(&session->cmdpool.queue, (void*)&conn->login_task, - sizeof(void*)); + kfifo_in(&session->mgmt_pool.queue, (void *)&conn->login_task, + sizeof(void *)); login_task_alloc_fail: iscsi_destroy_conn(cls_conn); return NULL; @@ -3189,8 +3206,8 @@ void iscsi_conn_teardown(struct iscsi_cls_conn *cls_conn) kfree(conn->local_ipaddr); /* regular RX path uses back_lock */ spin_lock_bh(&session->back_lock); - kfifo_in(&session->cmdpool.queue, (void*)&conn->login_task, - sizeof(void*)); + kfifo_in(&session->mgmt_pool.queue, (void *)&conn->login_task, + sizeof(void *)); spin_unlock_bh(&session->back_lock); if (session->leadconn == conn) session->leadconn = NULL; @@ -3275,8 +3292,8 @@ fail_mgmt_tasks(struct iscsi_session *session, struct iscsi_conn *conn) struct iscsi_task *task; int i, state; - for (i = 0; i < conn->session->cmds_max; i++) { - task = conn->session->cmds[i]; + for (i = 0; i < ISCSI_MGMT_CMDS_MAX; i++) { + task = conn->session->mgmt_cmds[i]; if (task->sc) continue; diff --git a/include/scsi/libiscsi.h b/include/scsi/libiscsi.h index 8e01beba62f1..7a78f8c5d670 100644 --- a/include/scsi/libiscsi.h +++ b/include/scsi/libiscsi.h @@ -35,8 +35,18 @@ struct iscsi_session; struct iscsi_nopin; struct device; -#define ISCSI_DEF_XMIT_CMDS_MAX 128 /* must be power of 2 */ -#define ISCSI_MGMT_CMDS_MAX 15 +#define ISCSI_DEF_XMIT_CMDS_MAX 128 +/* + * Max number of mgmt cmds we will preallocate and add to our mgmt fifo. + * This must be a pow of 2 due to the kfifo use. + */ +#define ISCSI_MGMT_CMDS_MAX 16 +/* + * For userspace compat we must allow at least 16 total cmds, so we have the + * the mgmt allocation limit above and this limit is the number of mgmt cmds + * that can be running. + */ +#define ISCSI_INFLIGHT_MGMT_MAX 15 #define ISCSI_DEF_CMD_PER_LUN 32 @@ -55,10 +65,19 @@ enum { /* Connection suspend "bit" */ #define ISCSI_SUSPEND_BIT 1 -#define ISCSI_ITT_MASK 0x1fff -#define ISCSI_TOTAL_CMDS_MAX 4096 -/* this must be a power of two greater than ISCSI_MGMT_CMDS_MAX */ -#define ISCSI_TOTAL_CMDS_MIN 16 +/* + * Note: + * - bnx2i needs the tag to be <= 0x3fff to fit in its fw req, and has a + * different itt space for scsi and mgmt cmds. + * - cxgbi assumes the tag will be at most 0x7fff. + * - iser needs the total cmds to be a pow of 2. + * - qedi, qla4xxx and be2iscsi ignore or pass through the libiscsi itt. + */ +#define ISCSI_ITT_MASK 0x3fff +#define ISCSI_TOTAL_CMDS_MAX 8192 +/* bit 14 is set for MGMT tasks and cleared for scsi cmds */ +#define ISCSI_TASK_TYPE_MGMT 0x2000 +#define ISCSI_TOTAL_CMDS_MIN (ISCSI_INFLIGHT_MGMT_MAX + 1) #define ISCSI_AGE_SHIFT 28 #define ISCSI_AGE_MASK 0xf @@ -331,13 +350,10 @@ struct iscsi_session { spinlock_t frwd_lock; /* protects queued_cmdsn, * * cmdsn, suspend_bit, * * leadconn, _stage, * - * tmf_state and session * - * resources: * - * - cmdpool kfifo_out , * - * - mgmtpool, queues */ + * tmf_state and mgmt * + * queues */ spinlock_t back_lock; /* protects cmdsn_exp * - * cmdsn_max, * - * cmdpool kfifo_in */ + * cmdsn_max, mgmt queues */ /* * frwd_lock must be held when transitioning states, but not needed * if just checking the state in the scsi-ml or iscsi callouts. @@ -346,9 +362,9 @@ struct iscsi_session { int age; /* counts session re-opens */ int scsi_cmds_max; /* max scsi commands */ - int cmds_max; /* size of cmds array */ - struct iscsi_task **cmds; /* Original Cmds arr */ - struct iscsi_pool cmdpool; /* PDU's pool */ + int cmds_max; /* Total number of tasks */ + struct iscsi_task **mgmt_cmds; + struct iscsi_pool mgmt_pool; /* mgmt task pool */ void *dd_data; /* LLD private data */ }; -- 2.25.1