diff options
Diffstat (limited to 'drivers/scsi/qedf/qedf_io.c')
| -rw-r--r-- | drivers/scsi/qedf/qedf_io.c | 752 |
1 files changed, 571 insertions, 181 deletions
diff --git a/drivers/scsi/qedf/qedf_io.c b/drivers/scsi/qedf/qedf_io.c index 53e8221f6816..42f9f2a9d8ea 100644 --- a/drivers/scsi/qedf/qedf_io.c +++ b/drivers/scsi/qedf/qedf_io.c @@ -43,8 +43,9 @@ static void qedf_cmd_timeout(struct work_struct *work) switch (io_req->cmd_type) { case QEDF_ABTS: if (qedf == NULL) { - QEDF_INFO(NULL, QEDF_LOG_IO, "qedf is NULL for xid=0x%x.\n", - io_req->xid); + QEDF_INFO(NULL, QEDF_LOG_IO, + "qedf is NULL for ABTS xid=0x%x.\n", + io_req->xid); return; } @@ -61,6 +62,9 @@ static void qedf_cmd_timeout(struct work_struct *work) */ kref_put(&io_req->refcount, qedf_release_cmd); + /* Clear in abort bit now that we're done with the command */ + clear_bit(QEDF_CMD_IN_ABORT, &io_req->flags); + /* * Now that the original I/O and the ABTS are complete see * if we need to reconnect to the target. @@ -68,6 +72,15 @@ static void qedf_cmd_timeout(struct work_struct *work) qedf_restart_rport(fcport); break; case QEDF_ELS: + if (!qedf) { + QEDF_INFO(NULL, QEDF_LOG_IO, + "qedf is NULL for ELS xid=0x%x.\n", + io_req->xid); + return; + } + /* ELS request no longer outstanding since it timed out */ + clear_bit(QEDF_CMD_OUTSTANDING, &io_req->flags); + kref_get(&io_req->refcount); /* * Don't attempt to clean an ELS timeout as any subseqeunt @@ -103,7 +116,7 @@ void qedf_cmd_mgr_free(struct qedf_cmd_mgr *cmgr) struct io_bdt *bdt_info; struct qedf_ctx *qedf = cmgr->qedf; size_t bd_tbl_sz; - u16 min_xid = QEDF_MIN_XID; + u16 min_xid = 0; u16 max_xid = (FCOE_PARAMS_NUM_TASKS - 1); int num_ios; int i; @@ -157,6 +170,7 @@ static void qedf_handle_rrq(struct work_struct *work) struct qedf_ioreq *io_req = container_of(work, struct qedf_ioreq, rrq_work.work); + atomic_set(&io_req->state, QEDFC_CMD_ST_RRQ_ACTIVE); qedf_send_rrq(io_req); } @@ -169,7 +183,7 @@ struct qedf_cmd_mgr *qedf_cmd_mgr_alloc(struct qedf_ctx *qedf) u16 xid; int i; int num_ios; - u16 min_xid = QEDF_MIN_XID; + u16 min_xid = 0; u16 max_xid = (FCOE_PARAMS_NUM_TASKS - 1); /* Make sure num_queues is already set before calling this function */ @@ -201,7 +215,7 @@ struct qedf_cmd_mgr *qedf_cmd_mgr_alloc(struct qedf_ctx *qedf) /* * Initialize I/O request fields. */ - xid = QEDF_MIN_XID; + xid = 0; for (i = 0; i < num_ios; i++) { io_req = &cmgr->cmds[i]; @@ -329,7 +343,7 @@ struct qedf_ioreq *qedf_alloc_cmd(struct qedf_rport *fcport, u8 cmd_type) cmd_mgr->idx = 0; /* Check to make sure command was previously freed */ - if (!test_bit(QEDF_CMD_OUTSTANDING, &io_req->flags)) + if (!io_req->alloc) break; } @@ -338,7 +352,14 @@ struct qedf_ioreq *qedf_alloc_cmd(struct qedf_rport *fcport, u8 cmd_type) goto out_failed; } - set_bit(QEDF_CMD_OUTSTANDING, &io_req->flags); + if (test_bit(QEDF_CMD_DIRTY, &io_req->flags)) + QEDF_ERR(&qedf->dbg_ctx, + "io_req found to be dirty ox_id = 0x%x.\n", + io_req->xid); + + /* Clear any flags now that we've reallocated the xid */ + io_req->flags = 0; + io_req->alloc = 1; spin_unlock_irqrestore(&cmd_mgr->lock, flags); atomic_inc(&fcport->num_active_ios); @@ -349,8 +370,13 @@ struct qedf_ioreq *qedf_alloc_cmd(struct qedf_rport *fcport, u8 cmd_type) io_req->cmd_mgr = cmd_mgr; io_req->fcport = fcport; + /* Clear any stale sc_cmd back pointer */ + io_req->sc_cmd = NULL; + io_req->lun = -1; + /* Hold the io_req against deletion */ - kref_init(&io_req->refcount); + kref_init(&io_req->refcount); /* ID: 001 */ + atomic_set(&io_req->state, QEDFC_CMD_ST_IO_ACTIVE); /* Bind io_bdt for this io_req */ /* Have a static link between io_req and io_bdt_pool */ @@ -412,6 +438,10 @@ void qedf_release_cmd(struct kref *ref) container_of(ref, struct qedf_ioreq, refcount); struct qedf_cmd_mgr *cmd_mgr = io_req->cmd_mgr; struct qedf_rport *fcport = io_req->fcport; + unsigned long flags; + + if (io_req->cmd_type == QEDF_SCSI_CMD) + WARN_ON(io_req->sc_cmd); if (io_req->cmd_type == QEDF_ELS || io_req->cmd_type == QEDF_TASK_MGMT_CMD) @@ -419,36 +449,20 @@ void qedf_release_cmd(struct kref *ref) atomic_inc(&cmd_mgr->free_list_cnt); atomic_dec(&fcport->num_active_ios); + atomic_set(&io_req->state, QEDF_CMD_ST_INACTIVE); if (atomic_read(&fcport->num_active_ios) < 0) QEDF_WARN(&(fcport->qedf->dbg_ctx), "active_ios < 0.\n"); /* Increment task retry identifier now that the request is released */ io_req->task_retry_identifier++; + io_req->fcport = NULL; - clear_bit(QEDF_CMD_OUTSTANDING, &io_req->flags); -} - -static int qedf_split_bd(struct qedf_ioreq *io_req, u64 addr, int sg_len, - int bd_index) -{ - struct scsi_sge *bd = io_req->bd_tbl->bd_tbl; - int frag_size, sg_frags; - - sg_frags = 0; - while (sg_len) { - if (sg_len > QEDF_BD_SPLIT_SZ) - frag_size = QEDF_BD_SPLIT_SZ; - else - frag_size = sg_len; - bd[bd_index + sg_frags].sge_addr.lo = U64_LO(addr); - bd[bd_index + sg_frags].sge_addr.hi = U64_HI(addr); - bd[bd_index + sg_frags].sge_len = (uint16_t)frag_size; - - addr += (u64)frag_size; - sg_frags++; - sg_len -= frag_size; - } - return sg_frags; + clear_bit(QEDF_CMD_DIRTY, &io_req->flags); + io_req->cpu = 0; + spin_lock_irqsave(&cmd_mgr->lock, flags); + io_req->fcport = NULL; + io_req->alloc = 0; + spin_unlock_irqrestore(&cmd_mgr->lock, flags); } static int qedf_map_sg(struct qedf_ioreq *io_req) @@ -462,75 +476,45 @@ static int qedf_map_sg(struct qedf_ioreq *io_req) int byte_count = 0; int sg_count = 0; int bd_count = 0; - int sg_frags; - unsigned int sg_len; + u32 sg_len; u64 addr, end_addr; - int i; + int i = 0; sg_count = dma_map_sg(&qedf->pdev->dev, scsi_sglist(sc), scsi_sg_count(sc), sc->sc_data_direction); - sg = scsi_sglist(sc); - /* - * New condition to send single SGE as cached-SGL with length less - * than 64k. - */ - if ((sg_count == 1) && (sg_dma_len(sg) <= - QEDF_MAX_SGLEN_FOR_CACHESGL)) { - sg_len = sg_dma_len(sg); - addr = (u64)sg_dma_address(sg); - - bd[bd_count].sge_addr.lo = (addr & 0xffffffff); - bd[bd_count].sge_addr.hi = (addr >> 32); - bd[bd_count].sge_len = (u16)sg_len; + io_req->sge_type = QEDF_IOREQ_UNKNOWN_SGE; - return ++bd_count; - } + if (sg_count <= 8 || io_req->io_req_flags == QEDF_READ) + io_req->sge_type = QEDF_IOREQ_FAST_SGE; scsi_for_each_sg(sc, sg, sg_count, i) { - sg_len = sg_dma_len(sg); + sg_len = (u32)sg_dma_len(sg); addr = (u64)sg_dma_address(sg); end_addr = (u64)(addr + sg_len); /* - * First s/g element in the list so check if the end_addr - * is paged aligned. Also check to make sure the length is - * at least page size. - */ - if ((i == 0) && (sg_count > 1) && - ((end_addr % QEDF_PAGE_SIZE) || - sg_len < QEDF_PAGE_SIZE)) - io_req->use_slowpath = true; - /* - * Last s/g element so check if the start address is paged - * aligned. - */ - else if ((i == (sg_count - 1)) && (sg_count > 1) && - (addr % QEDF_PAGE_SIZE)) - io_req->use_slowpath = true; - /* * Intermediate s/g element so check if start and end address - * is page aligned. + * is page aligned. Only required for writes and only if the + * number of scatter/gather elements is 8 or more. */ - else if ((i != 0) && (i != (sg_count - 1)) && - ((addr % QEDF_PAGE_SIZE) || (end_addr % QEDF_PAGE_SIZE))) - io_req->use_slowpath = true; + if (io_req->sge_type == QEDF_IOREQ_UNKNOWN_SGE && (i) && + (i != (sg_count - 1)) && sg_len < QEDF_PAGE_SIZE) + io_req->sge_type = QEDF_IOREQ_SLOW_SGE; - if (sg_len > QEDF_MAX_BD_LEN) { - sg_frags = qedf_split_bd(io_req, addr, sg_len, - bd_count); - } else { - sg_frags = 1; - bd[bd_count].sge_addr.lo = U64_LO(addr); - bd[bd_count].sge_addr.hi = U64_HI(addr); - bd[bd_count].sge_len = (uint16_t)sg_len; - } + bd[bd_count].sge_addr.lo = cpu_to_le32(U64_LO(addr)); + bd[bd_count].sge_addr.hi = cpu_to_le32(U64_HI(addr)); + bd[bd_count].sge_len = cpu_to_le32(sg_len); - bd_count += sg_frags; + bd_count++; byte_count += sg_len; } + /* To catch a case where FAST and SLOW nothing is set, set FAST */ + if (io_req->sge_type == QEDF_IOREQ_UNKNOWN_SGE) + io_req->sge_type = QEDF_IOREQ_FAST_SGE; + if (byte_count != scsi_bufflen(sc)) QEDF_ERR(&(qedf->dbg_ctx), "byte_count = %d != " "scsi_bufflen = %d, task_id = 0x%x.\n", byte_count, @@ -655,8 +639,10 @@ static void qedf_init_task(struct qedf_rport *fcport, struct fc_lport *lport, io_req->sgl_task_params->num_sges = bd_count; io_req->sgl_task_params->total_buffer_size = scsi_bufflen(io_req->sc_cmd); - io_req->sgl_task_params->small_mid_sge = - io_req->use_slowpath; + if (io_req->sge_type == QEDF_IOREQ_SLOW_SGE) + io_req->sgl_task_params->small_mid_sge = 1; + else + io_req->sgl_task_params->small_mid_sge = 0; } /* Fill in physical address of sense buffer */ @@ -679,16 +665,10 @@ static void qedf_init_task(struct qedf_rport *fcport, struct fc_lport *lport, io_req->task_retry_identifier, fcp_cmnd); /* Increment SGL type counters */ - if (bd_count == 1) { - qedf->single_sge_ios++; - io_req->sge_type = QEDF_IOREQ_SINGLE_SGE; - } else if (io_req->use_slowpath) { + if (io_req->sge_type == QEDF_IOREQ_SLOW_SGE) qedf->slow_sge_ios++; - io_req->sge_type = QEDF_IOREQ_SLOW_SGE; - } else { + else qedf->fast_sge_ios++; - io_req->sge_type = QEDF_IOREQ_FAST_SGE; - } } void qedf_init_mp_task(struct qedf_ioreq *io_req, @@ -770,9 +750,6 @@ void qedf_init_mp_task(struct qedf_ioreq *io_req, &task_fc_hdr, &tx_sgl_task_params, &rx_sgl_task_params, 0); - - /* Midpath requests always consume 1 SGE */ - qedf->single_sge_ios++; } /* Presumed that fcport->rport_lock is held */ @@ -804,8 +781,17 @@ void qedf_ring_doorbell(struct qedf_rport *fcport) FCOE_DB_DATA_AGG_VAL_SEL_SHIFT; dbell.sq_prod = fcport->fw_sq_prod_idx; + /* wmb makes sure that the BDs data is updated before updating the + * producer, otherwise FW may read old data from the BDs. + */ + wmb(); + barrier(); writel(*(u32 *)&dbell, fcport->p_doorbell); - /* Make sure SQ index is updated so f/w prcesses requests in order */ + /* + * Fence required to flush the write combined buffer, since another + * CPU may write to the same doorbell address and data may be lost + * due to relaxed order nature of write combined bar. + */ wmb(); } @@ -871,7 +857,7 @@ int qedf_post_io_req(struct qedf_rport *fcport, struct qedf_ioreq *io_req) /* Initialize rest of io_req fileds */ io_req->data_xfer_len = scsi_bufflen(sc_cmd); sc_cmd->SCp.ptr = (char *)io_req; - io_req->use_slowpath = false; /* Assume fast SGL by default */ + io_req->sge_type = QEDF_IOREQ_FAST_SGE; /* Assume fast SGL by default */ /* Record which cpu this request is associated with */ io_req->cpu = smp_processor_id(); @@ -894,15 +880,24 @@ int qedf_post_io_req(struct qedf_rport *fcport, struct qedf_ioreq *io_req) /* Build buffer descriptor list for firmware from sg list */ if (qedf_build_bd_list_from_sg(io_req)) { QEDF_ERR(&(qedf->dbg_ctx), "BD list creation failed.\n"); + /* Release cmd will release io_req, but sc_cmd is assigned */ + io_req->sc_cmd = NULL; kref_put(&io_req->refcount, qedf_release_cmd); return -EAGAIN; } - if (!test_bit(QEDF_RPORT_SESSION_READY, &fcport->flags)) { + if (!test_bit(QEDF_RPORT_SESSION_READY, &fcport->flags) || + test_bit(QEDF_RPORT_UPLOADING_CONNECTION, &fcport->flags)) { QEDF_ERR(&(qedf->dbg_ctx), "Session not offloaded yet.\n"); + /* Release cmd will release io_req, but sc_cmd is assigned */ + io_req->sc_cmd = NULL; kref_put(&io_req->refcount, qedf_release_cmd); + return -EINVAL; } + /* Record LUN number for later use if we neeed them */ + io_req->lun = (int)sc_cmd->device->lun; + /* Obtain free SQE */ sqe_idx = qedf_get_sqe_idx(fcport); sqe = &fcport->sq[sqe_idx]; @@ -913,6 +908,8 @@ int qedf_post_io_req(struct qedf_rport *fcport, struct qedf_ioreq *io_req) if (!task_ctx) { QEDF_WARN(&(qedf->dbg_ctx), "task_ctx is NULL, xid=%d.\n", xid); + /* Release cmd will release io_req, but sc_cmd is assigned */ + io_req->sc_cmd = NULL; kref_put(&io_req->refcount, qedf_release_cmd); return -EINVAL; } @@ -922,6 +919,9 @@ int qedf_post_io_req(struct qedf_rport *fcport, struct qedf_ioreq *io_req) /* Ring doorbell */ qedf_ring_doorbell(fcport); + /* Set that command is with the firmware now */ + set_bit(QEDF_CMD_OUTSTANDING, &io_req->flags); + if (qedf_io_tracing && io_req->sc_cmd) qedf_trace_io(fcport, io_req, QEDF_IO_TRACE_REQ); @@ -940,7 +940,17 @@ qedf_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *sc_cmd) int rc = 0; int rval; unsigned long flags = 0; - + int num_sgs = 0; + + num_sgs = scsi_sg_count(sc_cmd); + if (scsi_sg_count(sc_cmd) > QEDF_MAX_BDS_PER_CMD) { + QEDF_ERR(&qedf->dbg_ctx, + "Number of SG elements %d exceeds what hardware limitation of %d.\n", + num_sgs, QEDF_MAX_BDS_PER_CMD); + sc_cmd->result = DID_ERROR; + sc_cmd->scsi_done(sc_cmd); + return 0; + } if (test_bit(QEDF_UNLOADING, &qedf->flags) || test_bit(QEDF_DBG_STOP_IO, &qedf->flags)) { @@ -980,7 +990,8 @@ qedf_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *sc_cmd) /* rport and tgt are allocated together, so tgt should be non-NULL */ fcport = (struct qedf_rport *)&rp[1]; - if (!test_bit(QEDF_RPORT_SESSION_READY, &fcport->flags)) { + if (!test_bit(QEDF_RPORT_SESSION_READY, &fcport->flags) || + test_bit(QEDF_RPORT_UPLOADING_CONNECTION, &fcport->flags)) { /* * Session is not offloaded yet. Let SCSI-ml retry * the command. @@ -988,12 +999,16 @@ qedf_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *sc_cmd) rc = SCSI_MLQUEUE_TARGET_BUSY; goto exit_qcmd; } + + atomic_inc(&fcport->ios_to_queue); + if (fcport->retry_delay_timestamp) { if (time_after(jiffies, fcport->retry_delay_timestamp)) { fcport->retry_delay_timestamp = 0; } else { /* If retry_delay timer is active, flow off the ML */ rc = SCSI_MLQUEUE_TARGET_BUSY; + atomic_dec(&fcport->ios_to_queue); goto exit_qcmd; } } @@ -1001,6 +1016,7 @@ qedf_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *sc_cmd) io_req = qedf_alloc_cmd(fcport, QEDF_SCSI_CMD); if (!io_req) { rc = SCSI_MLQUEUE_HOST_BUSY; + atomic_dec(&fcport->ios_to_queue); goto exit_qcmd; } @@ -1015,6 +1031,7 @@ qedf_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *sc_cmd) rc = SCSI_MLQUEUE_HOST_BUSY; } spin_unlock_irqrestore(&fcport->rport_lock, flags); + atomic_dec(&fcport->ios_to_queue); exit_qcmd: return rc; @@ -1091,7 +1108,7 @@ static void qedf_unmap_sg_list(struct qedf_ctx *qedf, struct qedf_ioreq *io_req) void qedf_scsi_completion(struct qedf_ctx *qedf, struct fcoe_cqe *cqe, struct qedf_ioreq *io_req) { - u16 xid, rval; + u16 xid; struct e4_fcoe_task_context *task_ctx; struct scsi_cmnd *sc_cmd; struct fcoe_cqe_rsp_info *fcp_rsp; @@ -1105,6 +1122,15 @@ void qedf_scsi_completion(struct qedf_ctx *qedf, struct fcoe_cqe *cqe, if (!cqe) return; + if (!test_bit(QEDF_CMD_OUTSTANDING, &io_req->flags) || + test_bit(QEDF_CMD_IN_CLEANUP, &io_req->flags) || + test_bit(QEDF_CMD_IN_ABORT, &io_req->flags)) { + QEDF_ERR(&qedf->dbg_ctx, + "io_req xid=0x%x already in cleanup or abort processing or already completed.\n", + io_req->xid); + return; + } + xid = io_req->xid; task_ctx = qedf_get_task_mem(&qedf->tasks, xid); sc_cmd = io_req->sc_cmd; @@ -1121,6 +1147,12 @@ void qedf_scsi_completion(struct qedf_ctx *qedf, struct fcoe_cqe *cqe, return; } + if (!sc_cmd->device) { + QEDF_ERR(&qedf->dbg_ctx, + "Device for sc_cmd %p is NULL.\n", sc_cmd); + return; + } + if (!sc_cmd->request) { QEDF_WARN(&(qedf->dbg_ctx), "sc_cmd->request is NULL, " "sc_cmd=%p.\n", sc_cmd); @@ -1135,6 +1167,19 @@ void qedf_scsi_completion(struct qedf_ctx *qedf, struct fcoe_cqe *cqe, fcport = io_req->fcport; + /* + * When flush is active, let the cmds be completed from the cleanup + * context + */ + if (test_bit(QEDF_RPORT_IN_TARGET_RESET, &fcport->flags) || + (test_bit(QEDF_RPORT_IN_LUN_RESET, &fcport->flags) && + sc_cmd->device->lun == (u64)fcport->lun_reset_lun)) { + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "Dropping good completion xid=0x%x as fcport is flushing", + io_req->xid); + return; + } + qedf_parse_fcp_rsp(io_req, fcp_rsp); qedf_unmap_sg_list(qedf, io_req); @@ -1152,25 +1197,18 @@ void qedf_scsi_completion(struct qedf_ctx *qedf, struct fcoe_cqe *cqe, fw_residual_flag = GET_FIELD(cqe->cqe_info.rsp_info.fw_error_flags, FCOE_CQE_RSP_INFO_FW_UNDERRUN); if (fw_residual_flag) { - QEDF_ERR(&(qedf->dbg_ctx), - "Firmware detected underrun: xid=0x%x fcp_rsp.flags=0x%02x " - "fcp_resid=%d fw_residual=0x%x.\n", io_req->xid, - fcp_rsp->rsp_flags.flags, io_req->fcp_resid, - cqe->cqe_info.rsp_info.fw_residual); + QEDF_ERR(&qedf->dbg_ctx, + "Firmware detected underrun: xid=0x%x fcp_rsp.flags=0x%02x fcp_resid=%d fw_residual=0x%x lba=%02x%02x%02x%02x.\n", + io_req->xid, fcp_rsp->rsp_flags.flags, + io_req->fcp_resid, + cqe->cqe_info.rsp_info.fw_residual, sc_cmd->cmnd[2], + sc_cmd->cmnd[3], sc_cmd->cmnd[4], sc_cmd->cmnd[5]); if (io_req->cdb_status == 0) sc_cmd->result = (DID_ERROR << 16) | io_req->cdb_status; else sc_cmd->result = (DID_OK << 16) | io_req->cdb_status; - /* Abort the command since we did not get all the data */ - init_completion(&io_req->abts_done); - rval = qedf_initiate_abts(io_req, true); - if (rval) { - QEDF_ERR(&(qedf->dbg_ctx), "Failed to queue ABTS.\n"); - sc_cmd->result = (DID_ERROR << 16) | io_req->cdb_status; - } - /* * Set resid to the whole buffer length so we won't try to resue * any previously data. @@ -1242,6 +1280,12 @@ out: if (qedf_io_tracing) qedf_trace_io(fcport, io_req, QEDF_IO_TRACE_RSP); + /* + * We wait till the end of the function to clear the + * outstanding bit in case we need to send an abort + */ + clear_bit(QEDF_CMD_OUTSTANDING, &io_req->flags); + io_req->sc_cmd = NULL; sc_cmd->SCp.ptr = NULL; sc_cmd->scsi_done(sc_cmd); @@ -1259,6 +1303,19 @@ void qedf_scsi_done(struct qedf_ctx *qedf, struct qedf_ioreq *io_req, if (!io_req) return; + if (test_and_set_bit(QEDF_CMD_ERR_SCSI_DONE, &io_req->flags)) { + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "io_req:%p scsi_done handling already done\n", + io_req); + return; + } + + /* + * We will be done with this command after this call so clear the + * outstanding bit. + */ + clear_bit(QEDF_CMD_OUTSTANDING, &io_req->flags); + xid = io_req->xid; sc_cmd = io_req->sc_cmd; @@ -1267,12 +1324,50 @@ void qedf_scsi_done(struct qedf_ctx *qedf, struct qedf_ioreq *io_req, return; } + if (!virt_addr_valid(sc_cmd)) { + QEDF_ERR(&qedf->dbg_ctx, "sc_cmd=%p is not valid.", sc_cmd); + goto bad_scsi_ptr; + } + if (!sc_cmd->SCp.ptr) { QEDF_WARN(&(qedf->dbg_ctx), "SCp.ptr is NULL, returned in " "another context.\n"); return; } + if (!sc_cmd->device) { + QEDF_ERR(&qedf->dbg_ctx, "Device for sc_cmd %p is NULL.\n", + sc_cmd); + goto bad_scsi_ptr; + } + + if (!virt_addr_valid(sc_cmd->device)) { + QEDF_ERR(&qedf->dbg_ctx, + "Device pointer for sc_cmd %p is bad.\n", sc_cmd); + goto bad_scsi_ptr; + } + + if (!sc_cmd->sense_buffer) { + QEDF_ERR(&qedf->dbg_ctx, + "sc_cmd->sense_buffer for sc_cmd %p is NULL.\n", + sc_cmd); + goto bad_scsi_ptr; + } + + if (!virt_addr_valid(sc_cmd->sense_buffer)) { + QEDF_ERR(&qedf->dbg_ctx, + "sc_cmd->sense_buffer for sc_cmd %p is bad.\n", + sc_cmd); + goto bad_scsi_ptr; + } + + if (!sc_cmd->scsi_done) { + QEDF_ERR(&qedf->dbg_ctx, + "sc_cmd->scsi_done for sc_cmd %p is NULL.\n", + sc_cmd); + goto bad_scsi_ptr; + } + qedf_unmap_sg_list(qedf, io_req); sc_cmd->result = result << 16; @@ -1299,6 +1394,15 @@ void qedf_scsi_done(struct qedf_ctx *qedf, struct qedf_ioreq *io_req, sc_cmd->SCp.ptr = NULL; sc_cmd->scsi_done(sc_cmd); kref_put(&io_req->refcount, qedf_release_cmd); + return; + +bad_scsi_ptr: + /* + * Clear the io_req->sc_cmd backpointer so we don't try to process + * this again + */ + io_req->sc_cmd = NULL; + kref_put(&io_req->refcount, qedf_release_cmd); /* ID: 001 */ } /* @@ -1437,6 +1541,10 @@ void qedf_flush_active_ios(struct qedf_rport *fcport, int lun) struct qedf_ctx *qedf; struct qedf_cmd_mgr *cmd_mgr; int i, rc; + unsigned long flags; + int flush_cnt = 0; + int wait_cnt = 100; + int refcount = 0; if (!fcport) return; @@ -1448,18 +1556,102 @@ void qedf_flush_active_ios(struct qedf_rport *fcport, int lun) } qedf = fcport->qedf; + + if (!qedf) { + QEDF_ERR(NULL, "qedf is NULL.\n"); + return; + } + + /* Only wait for all commands to be queued in the Upload context */ + if (test_bit(QEDF_RPORT_UPLOADING_CONNECTION, &fcport->flags) && + (lun == -1)) { + while (atomic_read(&fcport->ios_to_queue)) { + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "Waiting for %d I/Os to be queued\n", + atomic_read(&fcport->ios_to_queue)); + if (wait_cnt == 0) { + QEDF_ERR(NULL, + "%d IOs request could not be queued\n", + atomic_read(&fcport->ios_to_queue)); + } + msleep(20); + wait_cnt--; + } + } + cmd_mgr = qedf->cmd_mgr; - QEDF_INFO(&(qedf->dbg_ctx), QEDF_LOG_IO, "Flush active i/o's.\n"); + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "Flush active i/o's num=0x%x fcport=0x%p port_id=0x%06x scsi_id=%d.\n", + atomic_read(&fcport->num_active_ios), fcport, + fcport->rdata->ids.port_id, fcport->rport->scsi_target_id); + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, "Locking flush mutex.\n"); + + mutex_lock(&qedf->flush_mutex); + if (lun == -1) { + set_bit(QEDF_RPORT_IN_TARGET_RESET, &fcport->flags); + } else { + set_bit(QEDF_RPORT_IN_LUN_RESET, &fcport->flags); + fcport->lun_reset_lun = lun; + } for (i = 0; i < FCOE_PARAMS_NUM_TASKS; i++) { io_req = &cmd_mgr->cmds[i]; if (!io_req) continue; + if (!io_req->fcport) + continue; + + spin_lock_irqsave(&cmd_mgr->lock, flags); + + if (io_req->alloc) { + if (!test_bit(QEDF_CMD_OUTSTANDING, &io_req->flags)) { + if (io_req->cmd_type == QEDF_SCSI_CMD) + QEDF_ERR(&qedf->dbg_ctx, + "Allocated but not queued, xid=0x%x\n", + io_req->xid); + } + spin_unlock_irqrestore(&cmd_mgr->lock, flags); + } else { + spin_unlock_irqrestore(&cmd_mgr->lock, flags); + continue; + } + if (io_req->fcport != fcport) continue; - if (io_req->cmd_type == QEDF_ELS) { + + /* In case of ABTS, CMD_OUTSTANDING is cleared on ABTS response, + * but RRQ is still pending. + * Workaround: Within qedf_send_rrq, we check if the fcport is + * NULL, and we drop the ref on the io_req to clean it up. + */ + if (!test_bit(QEDF_CMD_OUTSTANDING, &io_req->flags)) { + refcount = kref_read(&io_req->refcount); + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "Not outstanding, xid=0x%x, cmd_type=%d refcount=%d.\n", + io_req->xid, io_req->cmd_type, refcount); + /* If RRQ work has been queue, try to cancel it and + * free the io_req + */ + if (atomic_read(&io_req->state) == + QEDFC_CMD_ST_RRQ_WAIT) { + if (cancel_delayed_work_sync + (&io_req->rrq_work)) { + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "Putting reference for pending RRQ work xid=0x%x.\n", + io_req->xid); + /* ID: 003 */ + kref_put(&io_req->refcount, + qedf_release_cmd); + } + } + continue; + } + + /* Only consider flushing ELS during target reset */ + if (io_req->cmd_type == QEDF_ELS && + lun == -1) { rc = kref_get_unless_zero(&io_req->refcount); if (!rc) { QEDF_ERR(&(qedf->dbg_ctx), @@ -1467,6 +1659,7 @@ void qedf_flush_active_ios(struct qedf_rport *fcport, int lun) io_req, io_req->xid); continue; } + flush_cnt++; qedf_flush_els_req(qedf, io_req); /* * Release the kref and go back to the top of the @@ -1476,6 +1669,7 @@ void qedf_flush_active_ios(struct qedf_rport *fcport, int lun) } if (io_req->cmd_type == QEDF_ABTS) { + /* ID: 004 */ rc = kref_get_unless_zero(&io_req->refcount); if (!rc) { QEDF_ERR(&(qedf->dbg_ctx), @@ -1483,28 +1677,50 @@ void qedf_flush_active_ios(struct qedf_rport *fcport, int lun) io_req, io_req->xid); continue; } + if (lun != -1 && io_req->lun != lun) + goto free_cmd; + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, "Flushing abort xid=0x%x.\n", io_req->xid); - clear_bit(QEDF_CMD_IN_ABORT, &io_req->flags); - - if (io_req->sc_cmd) { - if (io_req->return_scsi_cmd_on_abts) - qedf_scsi_done(qedf, io_req, DID_ERROR); + if (cancel_delayed_work_sync(&io_req->rrq_work)) { + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "Putting ref for cancelled RRQ work xid=0x%x.\n", + io_req->xid); + kref_put(&io_req->refcount, qedf_release_cmd); } - /* Notify eh_abort handler that ABTS is complete */ - complete(&io_req->abts_done); - kref_put(&io_req->refcount, qedf_release_cmd); - + if (cancel_delayed_work_sync(&io_req->timeout_work)) { + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "Putting ref for cancelled tmo work xid=0x%x.\n", + io_req->xid); + qedf_initiate_cleanup(io_req, true); + /* Notify eh_abort handler that ABTS is + * complete + */ + complete(&io_req->abts_done); + clear_bit(QEDF_CMD_IN_ABORT, &io_req->flags); + /* ID: 002 */ + kref_put(&io_req->refcount, qedf_release_cmd); + } + flush_cnt++; goto free_cmd; } if (!io_req->sc_cmd) continue; - if (lun > 0) { - if (io_req->sc_cmd->device->lun != - (u64)lun) + if (!io_req->sc_cmd->device) { + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "Device backpointer NULL for sc_cmd=%p.\n", + io_req->sc_cmd); + /* Put reference for non-existent scsi_cmnd */ + io_req->sc_cmd = NULL; + qedf_initiate_cleanup(io_req, false); + kref_put(&io_req->refcount, qedf_release_cmd); + continue; + } + if (lun > -1) { + if (io_req->lun != lun) continue; } @@ -1518,15 +1734,65 @@ void qedf_flush_active_ios(struct qedf_rport *fcport, int lun) "io_req=0x%p xid=0x%x\n", io_req, io_req->xid); continue; } + QEDF_INFO(&(qedf->dbg_ctx), QEDF_LOG_IO, "Cleanup xid=0x%x.\n", io_req->xid); + flush_cnt++; /* Cleanup task and return I/O mid-layer */ qedf_initiate_cleanup(io_req, true); free_cmd: - kref_put(&io_req->refcount, qedf_release_cmd); + kref_put(&io_req->refcount, qedf_release_cmd); /* ID: 004 */ + } + + wait_cnt = 60; + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "Flushed 0x%x I/Os, active=0x%x.\n", + flush_cnt, atomic_read(&fcport->num_active_ios)); + /* Only wait for all commands to complete in the Upload context */ + if (test_bit(QEDF_RPORT_UPLOADING_CONNECTION, &fcport->flags) && + (lun == -1)) { + while (atomic_read(&fcport->num_active_ios)) { + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "Flushed 0x%x I/Os, active=0x%x cnt=%d.\n", + flush_cnt, + atomic_read(&fcport->num_active_ios), + wait_cnt); + if (wait_cnt == 0) { + QEDF_ERR(&qedf->dbg_ctx, + "Flushed %d I/Os, active=%d.\n", + flush_cnt, + atomic_read(&fcport->num_active_ios)); + for (i = 0; i < FCOE_PARAMS_NUM_TASKS; i++) { + io_req = &cmd_mgr->cmds[i]; + if (io_req->fcport && + io_req->fcport == fcport) { + refcount = + kref_read(&io_req->refcount); + set_bit(QEDF_CMD_DIRTY, + &io_req->flags); + QEDF_ERR(&qedf->dbg_ctx, + "Outstanding io_req =%p xid=0x%x flags=0x%lx, sc_cmd=%p refcount=%d cmd_type=%d.\n", + io_req, io_req->xid, + io_req->flags, + io_req->sc_cmd, + refcount, + io_req->cmd_type); + } + } + WARN_ON(1); + break; + } + msleep(500); + wait_cnt--; + } } + + clear_bit(QEDF_RPORT_IN_LUN_RESET, &fcport->flags); + clear_bit(QEDF_RPORT_IN_TARGET_RESET, &fcport->flags); + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, "Unlocking flush mutex.\n"); + mutex_unlock(&qedf->flush_mutex); } /* @@ -1545,52 +1811,60 @@ int qedf_initiate_abts(struct qedf_ioreq *io_req, bool return_scsi_cmd_on_abts) unsigned long flags; struct fcoe_wqe *sqe; u16 sqe_idx; + int refcount = 0; /* Sanity check qedf_rport before dereferencing any pointers */ if (!test_bit(QEDF_RPORT_SESSION_READY, &fcport->flags)) { QEDF_ERR(NULL, "tgt not offloaded\n"); rc = 1; - goto abts_err; + goto out; } + qedf = fcport->qedf; rdata = fcport->rdata; + + if (!rdata || !kref_get_unless_zero(&rdata->kref)) { + QEDF_ERR(&qedf->dbg_ctx, "stale rport\n"); + rc = 1; + goto out; + } + r_a_tov = rdata->r_a_tov; - qedf = fcport->qedf; lport = qedf->lport; if (lport->state != LPORT_ST_READY || !(lport->link_up)) { QEDF_ERR(&(qedf->dbg_ctx), "link is not ready\n"); rc = 1; - goto abts_err; + goto drop_rdata_kref; } if (atomic_read(&qedf->link_down_tmo_valid) > 0) { QEDF_ERR(&(qedf->dbg_ctx), "link_down_tmo active.\n"); rc = 1; - goto abts_err; + goto drop_rdata_kref; } /* Ensure room on SQ */ if (!atomic_read(&fcport->free_sqes)) { QEDF_ERR(&(qedf->dbg_ctx), "No SQ entries available\n"); rc = 1; - goto abts_err; + goto drop_rdata_kref; } if (test_bit(QEDF_RPORT_UPLOADING_CONNECTION, &fcport->flags)) { QEDF_ERR(&qedf->dbg_ctx, "fcport is uploading.\n"); rc = 1; - goto out; + goto drop_rdata_kref; } if (!test_bit(QEDF_CMD_OUTSTANDING, &io_req->flags) || test_bit(QEDF_CMD_IN_CLEANUP, &io_req->flags) || test_bit(QEDF_CMD_IN_ABORT, &io_req->flags)) { - QEDF_ERR(&(qedf->dbg_ctx), "io_req xid=0x%x already in " - "cleanup or abort processing or already " - "completed.\n", io_req->xid); + QEDF_ERR(&qedf->dbg_ctx, + "io_req xid=0x%x sc_cmd=%p already in cleanup or abort processing or already completed.\n", + io_req->xid, io_req->sc_cmd); rc = 1; - goto out; + goto drop_rdata_kref; } kref_get(&io_req->refcount); @@ -1599,18 +1873,17 @@ int qedf_initiate_abts(struct qedf_ioreq *io_req, bool return_scsi_cmd_on_abts) qedf->control_requests++; qedf->packet_aborts++; - /* Set the return CPU to be the same as the request one */ - io_req->cpu = smp_processor_id(); - /* Set the command type to abort */ io_req->cmd_type = QEDF_ABTS; io_req->return_scsi_cmd_on_abts = return_scsi_cmd_on_abts; set_bit(QEDF_CMD_IN_ABORT, &io_req->flags); - QEDF_INFO(&(qedf->dbg_ctx), QEDF_LOG_SCSI_TM, "ABTS io_req xid = " - "0x%x\n", xid); + refcount = kref_read(&io_req->refcount); + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_SCSI_TM, + "ABTS io_req xid = 0x%x refcount=%d\n", + xid, refcount); - qedf_cmd_timer_set(qedf, io_req, QEDF_ABORT_TIMEOUT * HZ); + qedf_cmd_timer_set(qedf, io_req, QEDF_ABORT_TIMEOUT); spin_lock_irqsave(&fcport->rport_lock, flags); @@ -1624,13 +1897,8 @@ int qedf_initiate_abts(struct qedf_ioreq *io_req, bool return_scsi_cmd_on_abts) spin_unlock_irqrestore(&fcport->rport_lock, flags); - return rc; -abts_err: - /* - * If the ABTS task fails to queue then we need to cleanup the - * task at the firmware. - */ - qedf_initiate_cleanup(io_req, return_scsi_cmd_on_abts); +drop_rdata_kref: + kref_put(&rdata->kref, fc_rport_destroy); out: return rc; } @@ -1640,27 +1908,62 @@ void qedf_process_abts_compl(struct qedf_ctx *qedf, struct fcoe_cqe *cqe, { uint32_t r_ctl; uint16_t xid; + int rc; + struct qedf_rport *fcport = io_req->fcport; QEDF_INFO(&(qedf->dbg_ctx), QEDF_LOG_SCSI_TM, "Entered with xid = " "0x%x cmd_type = %d\n", io_req->xid, io_req->cmd_type); - cancel_delayed_work(&io_req->timeout_work); - xid = io_req->xid; r_ctl = cqe->cqe_info.abts_info.r_ctl; + /* This was added at a point when we were scheduling abts_compl & + * cleanup_compl on different CPUs and there was a possibility of + * the io_req to be freed from the other context before we got here. + */ + if (!fcport) { + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "Dropping ABTS completion xid=0x%x as fcport is NULL", + io_req->xid); + return; + } + + /* + * When flush is active, let the cmds be completed from the cleanup + * context + */ + if (test_bit(QEDF_RPORT_IN_TARGET_RESET, &fcport->flags) || + test_bit(QEDF_RPORT_IN_LUN_RESET, &fcport->flags)) { + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "Dropping ABTS completion xid=0x%x as fcport is flushing", + io_req->xid); + return; + } + + if (!cancel_delayed_work(&io_req->timeout_work)) { + QEDF_ERR(&qedf->dbg_ctx, + "Wasn't able to cancel abts timeout work.\n"); + } + switch (r_ctl) { case FC_RCTL_BA_ACC: QEDF_INFO(&(qedf->dbg_ctx), QEDF_LOG_SCSI_TM, "ABTS response - ACC Send RRQ after R_A_TOV\n"); io_req->event = QEDF_IOREQ_EV_ABORT_SUCCESS; + rc = kref_get_unless_zero(&io_req->refcount); /* ID: 003 */ + if (!rc) { + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_SCSI_TM, + "kref is already zero so ABTS was already completed or flushed xid=0x%x.\n", + io_req->xid); + return; + } /* * Dont release this cmd yet. It will be relesed * after we get RRQ response */ - kref_get(&io_req->refcount); queue_delayed_work(qedf->dpc_wq, &io_req->rrq_work, msecs_to_jiffies(qedf->lport->r_a_tov)); + atomic_set(&io_req->state, QEDFC_CMD_ST_RRQ_WAIT); break; /* For error cases let the cleanup return the command */ case FC_RCTL_BA_RJT: @@ -1802,6 +2105,7 @@ int qedf_initiate_cleanup(struct qedf_ioreq *io_req, unsigned long flags; struct fcoe_wqe *sqe; u16 sqe_idx; + int refcount = 0; fcport = io_req->fcport; if (!fcport) { @@ -1823,36 +2127,45 @@ int qedf_initiate_cleanup(struct qedf_ioreq *io_req, } if (!test_bit(QEDF_CMD_OUTSTANDING, &io_req->flags) || - test_bit(QEDF_CMD_IN_CLEANUP, &io_req->flags)) { + test_and_set_bit(QEDF_CMD_IN_CLEANUP, &io_req->flags)) { QEDF_ERR(&(qedf->dbg_ctx), "io_req xid=0x%x already in " "cleanup processing or already completed.\n", io_req->xid); return SUCCESS; } + set_bit(QEDF_CMD_IN_CLEANUP, &io_req->flags); /* Ensure room on SQ */ if (!atomic_read(&fcport->free_sqes)) { QEDF_ERR(&(qedf->dbg_ctx), "No SQ entries available\n"); + /* Need to make sure we clear the flag since it was set */ + clear_bit(QEDF_CMD_IN_CLEANUP, &io_req->flags); return FAILED; } + if (io_req->cmd_type == QEDF_CLEANUP) { + QEDF_ERR(&qedf->dbg_ctx, + "io_req=0x%x is already a cleanup command cmd_type=%d.\n", + io_req->xid, io_req->cmd_type); + clear_bit(QEDF_CMD_IN_CLEANUP, &io_req->flags); + return SUCCESS; + } - QEDF_INFO(&(qedf->dbg_ctx), QEDF_LOG_IO, "Entered xid=0x%x\n", - io_req->xid); + refcount = kref_read(&io_req->refcount); + + QEDF_INFO(&qedf->dbg_ctx, QEDF_LOG_IO, + "Entered xid=0x%x sc_cmd=%p cmd_type=%d flags=0x%lx refcount=%d fcport=%p port_id=0x%06x\n", + io_req->xid, io_req->sc_cmd, io_req->cmd_type, io_req->flags, + refcount, fcport, fcport->rdata->ids.port_id); /* Cleanup cmds re-use the same TID as the original I/O */ xid = io_req->xid; io_req->cmd_type = QEDF_CLEANUP; io_req->return_scsi_cmd_on_abts = return_scsi_cmd_on_abts; - /* Set the return CPU to be the same as the request one */ - io_req->cpu = smp_processor_id(); - - set_bit(QEDF_CMD_IN_CLEANUP, &io_req->flags); - task = qedf_get_task_mem(&qedf->tasks, xid); - init_completion(&io_req->tm_done); + init_completion(&io_req->cleanup_done); spin_lock_irqsave(&fcport->rport_lock, flags); @@ -1866,8 +2179,8 @@ int qedf_initiate_cleanup(struct qedf_ioreq *io_req, spin_unlock_irqrestore(&fcport->rport_lock, flags); - tmo = wait_for_completion_timeout(&io_req->tm_done, - QEDF_CLEANUP_TIMEOUT * HZ); + tmo = wait_for_completion_timeout(&io_req->cleanup_done, + QEDF_CLEANUP_TIMEOUT * HZ); if (!tmo) { rc = FAILED; @@ -1880,6 +2193,16 @@ int qedf_initiate_cleanup(struct qedf_ioreq *io_req, qedf_drain_request(qedf); } + /* If it TASK MGMT handle it, reference will be decreased + * in qedf_execute_tmf + */ + if (io_req->tm_flags == FCP_TMF_LUN_RESET || + io_req->tm_flags == FCP_TMF_TGT_RESET) { + clear_bit(QEDF_CMD_OUTSTANDING, &io_req->flags); + io_req->sc_cmd = NULL; + complete(&io_req->tm_done); + } + if (io_req->sc_cmd) { if (io_req->return_scsi_cmd_on_abts) qedf_scsi_done(qedf, io_req, DID_ERROR); @@ -1902,7 +2225,7 @@ void qedf_process_cleanup_compl(struct qedf_ctx *qedf, struct fcoe_cqe *cqe, clear_bit(QEDF_CMD_IN_CLEANUP, &io_req->flags); /* Complete so we can finish cleaning up the I/O */ - complete(&io_req->tm_done); + complete(&io_req->cleanup_done); } static int qedf_execute_tmf(struct qedf_rport *fcport, struct scsi_cmnd *sc_cmd, @@ -1915,6 +2238,7 @@ static int qedf_execute_tmf(struct qedf_rport *fcport, struct scsi_cmnd *sc_cmd, int rc = 0; uint16_t xid; int tmo = 0; + int lun = 0; unsigned long flags; struct fcoe_wqe *sqe; u16 sqe_idx; @@ -1924,20 +2248,18 @@ static int qedf_execute_tmf(struct qedf_rport *fcport, struct scsi_cmnd *sc_cmd, return FAILED; } + lun = (int)sc_cmd->device->lun; if (!test_bit(QEDF_RPORT_SESSION_READY, &fcport->flags)) { QEDF_ERR(&(qedf->dbg_ctx), "fcport not offloaded\n"); rc = FAILED; - return FAILED; + goto no_flush; } - QEDF_INFO(&(qedf->dbg_ctx), QEDF_LOG_SCSI_TM, "portid = 0x%x " - "tm_flags = %d\n", fcport->rdata->ids.port_id, tm_flags); - io_req = qedf_alloc_cmd(fcport, QEDF_TASK_MGMT_CMD); if (!io_req) { QEDF_ERR(&(qedf->dbg_ctx), "Failed TMF"); rc = -EAGAIN; - goto reset_tmf_err; + goto no_flush; } if (tm_flags == FCP_TMF_LUN_RESET) @@ -1950,7 +2272,7 @@ static int qedf_execute_tmf(struct qedf_rport *fcport, struct scsi_cmnd *sc_cmd, io_req->fcport = fcport; io_req->cmd_type = QEDF_TASK_MGMT_CMD; - /* Set the return CPU to be the same as the request one */ + /* Record which cpu this request is associated with */ io_req->cpu = smp_processor_id(); /* Set TM flags */ @@ -1959,7 +2281,7 @@ static int qedf_execute_tmf(struct qedf_rport *fcport, struct scsi_cmnd *sc_cmd, io_req->tm_flags = tm_flags; /* Default is to return a SCSI command when an error occurs */ - io_req->return_scsi_cmd_on_abts = true; + io_req->return_scsi_cmd_on_abts = false; /* Obtain exchange id */ xid = io_req->xid; @@ -1983,12 +2305,16 @@ static int qedf_execute_tmf(struct qedf_rport *fcport, struct scsi_cmnd *sc_cmd, spin_unlock_irqrestore(&fcport->rport_lock, flags); + set_bit(QEDF_CMD_OUTSTANDING, &io_req->flags); tmo = wait_for_completion_timeout(&io_req->tm_done, QEDF_TM_TIMEOUT * HZ); if (!tmo) { rc = FAILED; QEDF_ERR(&(qedf->dbg_ctx), "wait for tm_cmpl timeout!\n"); + /* Clear outstanding bit since command timed out */ + clear_bit(QEDF_CMD_OUTSTANDING, &io_req->flags); + io_req->sc_cmd = NULL; } else { /* Check TMF response code */ if (io_req->fcp_rsp_code == 0) @@ -1996,14 +2322,25 @@ static int qedf_execute_tmf(struct qedf_rport *fcport, struct scsi_cmnd *sc_cmd, else rc = FAILED; } + /* + * Double check that fcport has not gone into an uploading state before + * executing the command flush for the LUN/target. + */ + if (test_bit(QEDF_RPORT_UPLOADING_CONNECTION, &fcport->flags)) { + QEDF_ERR(&qedf->dbg_ctx, + "fcport is uploading, not executing flush.\n"); + goto no_flush; + } + /* We do not need this io_req any more */ + kref_put(&io_req->refcount, qedf_release_cmd); + if (tm_flags == FCP_TMF_LUN_RESET) - qedf_flush_active_ios(fcport, (int)sc_cmd->device->lun); + qedf_flush_active_ios(fcport, lun); else qedf_flush_active_ios(fcport, -1); - kref_put(&io_req->refcount, qedf_release_cmd); - +no_flush: if (rc != SUCCESS) { QEDF_ERR(&(qedf->dbg_ctx), "task mgmt command failed...\n"); rc = FAILED; @@ -2011,7 +2348,6 @@ static int qedf_execute_tmf(struct qedf_rport *fcport, struct scsi_cmnd *sc_cmd, QEDF_ERR(&(qedf->dbg_ctx), "task mgmt command success...\n"); rc = SUCCESS; } -reset_tmf_err: return rc; } @@ -2021,26 +2357,65 @@ int qedf_initiate_tmf(struct scsi_cmnd *sc_cmd, u8 tm_flags) struct fc_rport_libfc_priv *rp = rport->dd_data; struct qedf_rport *fcport = (struct qedf_rport *)&rp[1]; struct qedf_ctx *qedf; - struct fc_lport *lport; + struct fc_lport *lport = shost_priv(sc_cmd->device->host); int rc = SUCCESS; int rval; + struct qedf_ioreq *io_req = NULL; + int ref_cnt = 0; + struct fc_rport_priv *rdata = fcport->rdata; - rval = fc_remote_port_chkready(rport); + QEDF_ERR(NULL, + "tm_flags 0x%x sc_cmd %p op = 0x%02x target_id = 0x%x lun=%d\n", + tm_flags, sc_cmd, sc_cmd->cmnd[0], rport->scsi_target_id, + (int)sc_cmd->device->lun); + + if (!rdata || !kref_get_unless_zero(&rdata->kref)) { + QEDF_ERR(NULL, "stale rport\n"); + return FAILED; + } + + QEDF_ERR(NULL, "portid=%06x tm_flags =%s\n", rdata->ids.port_id, + (tm_flags == FCP_TMF_TGT_RESET) ? "TARGET RESET" : + "LUN RESET"); + + if (sc_cmd->SCp.ptr) { + io_req = (struct qedf_ioreq *)sc_cmd->SCp.ptr; + ref_cnt = kref_read(&io_req->refcount); + QEDF_ERR(NULL, + "orig io_req = %p xid = 0x%x ref_cnt = %d.\n", + io_req, io_req->xid, ref_cnt); + } + rval = fc_remote_port_chkready(rport); if (rval) { QEDF_ERR(NULL, "device_reset rport not ready\n"); rc = FAILED; goto tmf_err; } - if (fcport == NULL) { + rc = fc_block_scsi_eh(sc_cmd); + if (rc) + goto tmf_err; + + if (!fcport) { QEDF_ERR(NULL, "device_reset: rport is NULL\n"); rc = FAILED; goto tmf_err; } qedf = fcport->qedf; - lport = qedf->lport; + + if (!qedf) { + QEDF_ERR(NULL, "qedf is NULL.\n"); + rc = FAILED; + goto tmf_err; + } + + if (test_bit(QEDF_RPORT_UPLOADING_CONNECTION, &fcport->flags)) { + QEDF_ERR(&qedf->dbg_ctx, "Connection is getting uploaded.\n"); + rc = SUCCESS; + goto tmf_err; + } if (test_bit(QEDF_UNLOADING, &qedf->flags) || test_bit(QEDF_DBG_STOP_IO, &qedf->flags)) { @@ -2054,9 +2429,22 @@ int qedf_initiate_tmf(struct scsi_cmnd *sc_cmd, u8 tm_flags) goto tmf_err; } + if (test_bit(QEDF_RPORT_UPLOADING_CONNECTION, &fcport->flags)) { + if (!fcport->rdata) + QEDF_ERR(&qedf->dbg_ctx, "fcport %p is uploading.\n", + fcport); + else + QEDF_ERR(&qedf->dbg_ctx, + "fcport %p port_id=%06x is uploading.\n", + fcport, fcport->rdata->ids.port_id); + rc = FAILED; + goto tmf_err; + } + rc = qedf_execute_tmf(fcport, sc_cmd, tm_flags); tmf_err: + kref_put(&rdata->kref, fc_rport_destroy); return rc; } @@ -2065,6 +2453,8 @@ void qedf_process_tmf_compl(struct qedf_ctx *qedf, struct fcoe_cqe *cqe, { struct fcoe_cqe_rsp_info *fcp_rsp; + clear_bit(QEDF_CMD_OUTSTANDING, &io_req->flags); + fcp_rsp = &cqe->cqe_info.rsp_info; qedf_parse_fcp_rsp(io_req, fcp_rsp); |