scsi: smartpqi: enhance resets
authorKevin Barnett <kevin.barnett@hpe.com>
Wed, 3 May 2017 23:52:58 +0000 (18:52 -0500)
committerMartin K. Petersen <martin.petersen@oracle.com>
Tue, 13 Jun 2017 00:48:02 +0000 (20:48 -0400)
- Block all I/O targeted at LUN reset device.
- Wait until all I/O targeted at LUN reset device has been
  consumed by the controller.
- Issue LUN reset request.
- Wait until all outstanding I/Os and LUN reset completion
  have been received by the host.
- Return to OS results of LUN reset request.

Reviewed-by: Scott Benesh <scott.benesh@microsemi.com>
Signed-off-by: Kevin Barnett <kevin.barnett@microsemi.com>
Signed-off-by: Don Brace <don.brace@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/smartpqi/smartpqi.h
drivers/scsi/smartpqi/smartpqi_init.c

index 02b31965bf6eebc3f3b33bdff93c093021844a9c..5b0c6fb3ceb1ee6c86cf6ce694d2a8e0a0c88b86 100644 (file)
@@ -734,6 +734,8 @@ struct pqi_scsi_dev {
        u8      new_device : 1;
        u8      keep_device : 1;
        u8      volume_offline : 1;
+       bool    in_reset;
+       bool    device_offline;
        u8      vendor[8];              /* bytes 8-15 of inquiry data */
        u8      model[16];              /* bytes 16-31 of inquiry data */
        u64     sas_address;
@@ -761,6 +763,8 @@ struct pqi_scsi_dev {
        struct list_head new_device_list_entry;
        struct list_head add_list_entry;
        struct list_head delete_list_entry;
+
+       atomic_t scsi_cmds_outstanding;
 };
 
 /* VPD inquiry pages */
@@ -926,7 +930,9 @@ struct pqi_ctrl_info {
        struct Scsi_Host *scsi_host;
 
        struct mutex    scan_mutex;
-       bool            controller_online : 1;
+       struct mutex    lun_reset_mutex;
+       bool            controller_online;
+       bool            block_requests;
        u8              inbound_spanning_supported : 1;
        u8              outbound_spanning_supported : 1;
        u8              pqi_mode_enabled : 1;
@@ -953,7 +959,9 @@ struct pqi_ctrl_info {
        struct timer_list heartbeat_timer;
 
        struct semaphore sync_request_sem;
-       struct semaphore lun_reset_sem;
+       atomic_t        num_busy_threads;
+       atomic_t        num_blocked_threads;
+       wait_queue_head_t block_requests_wait;
 };
 
 enum pqi_ctrl_mode {
@@ -1092,6 +1100,7 @@ int pqi_add_sas_device(struct pqi_sas_node *pqi_sas_node,
 void pqi_remove_sas_device(struct pqi_scsi_dev *device);
 struct pqi_scsi_dev *pqi_find_device_by_sas_rphy(
        struct pqi_ctrl_info *ctrl_info, struct sas_rphy *rphy);
+void pqi_prep_for_scsi_done(struct scsi_cmnd *scmd);
 
 extern struct sas_function_template pqi_sas_transport_functions;
 
index 279dd41796c02a6fe38f49a23b69c9795d7a7c4c..bd8a66d9acc055c375e694d52692a25c28528baf 100644 (file)
@@ -126,6 +126,7 @@ static char *pqi_raid_level_to_string(u8 raid_level)
 
 static inline void pqi_scsi_done(struct scsi_cmnd *scmd)
 {
+       pqi_prep_for_scsi_done(scmd);
        scmd->scsi_done(scmd);
 }
 
@@ -175,7 +176,85 @@ static inline void pqi_save_ctrl_mode(struct pqi_ctrl_info *ctrl_info,
        sis_write_driver_scratch(ctrl_info, mode);
 }
 
-#define PQI_RESCAN_WORK_INTERVAL       (10 * HZ)
+#define PQI_RESCAN_WORK_INTERVAL       (10 * HZ)
+static inline void pqi_ctrl_block_requests(struct pqi_ctrl_info *ctrl_info)
+{
+       ctrl_info->block_requests = true;
+       scsi_block_requests(ctrl_info->scsi_host);
+}
+
+static inline void pqi_ctrl_unblock_requests(struct pqi_ctrl_info *ctrl_info)
+{
+       ctrl_info->block_requests = false;
+       wake_up_all(&ctrl_info->block_requests_wait);
+       scsi_unblock_requests(ctrl_info->scsi_host);
+}
+
+static inline bool pqi_ctrl_blocked(struct pqi_ctrl_info *ctrl_info)
+{
+       return ctrl_info->block_requests;
+}
+
+static unsigned long pqi_wait_if_ctrl_blocked(struct pqi_ctrl_info *ctrl_info,
+       unsigned long timeout_msecs)
+{
+       unsigned long remaining_msecs;
+
+       if (!pqi_ctrl_blocked(ctrl_info))
+               return timeout_msecs;
+
+       atomic_inc(&ctrl_info->num_blocked_threads);
+
+       if (timeout_msecs == NO_TIMEOUT) {
+               wait_event(ctrl_info->block_requests_wait,
+                       !pqi_ctrl_blocked(ctrl_info));
+               remaining_msecs = timeout_msecs;
+       } else {
+               unsigned long remaining_jiffies;
+
+               remaining_jiffies =
+                       wait_event_timeout(ctrl_info->block_requests_wait,
+                               !pqi_ctrl_blocked(ctrl_info),
+                               msecs_to_jiffies(timeout_msecs));
+               remaining_msecs = jiffies_to_msecs(remaining_jiffies);
+       }
+
+       atomic_dec(&ctrl_info->num_blocked_threads);
+
+       return remaining_msecs;
+}
+
+static inline void pqi_ctrl_busy(struct pqi_ctrl_info *ctrl_info)
+{
+       atomic_inc(&ctrl_info->num_busy_threads);
+}
+
+static inline void pqi_ctrl_unbusy(struct pqi_ctrl_info *ctrl_info)
+{
+       atomic_dec(&ctrl_info->num_busy_threads);
+}
+
+static inline void pqi_ctrl_wait_until_quiesced(struct pqi_ctrl_info *ctrl_info)
+{
+       while (atomic_read(&ctrl_info->num_busy_threads) >
+               atomic_read(&ctrl_info->num_blocked_threads))
+               usleep_range(1000, 2000);
+}
+
+static inline void pqi_device_reset_start(struct pqi_scsi_dev *device)
+{
+       device->in_reset = true;
+}
+
+static inline void pqi_device_reset_done(struct pqi_scsi_dev *device)
+{
+       device->in_reset = false;
+}
+
+static inline bool pqi_device_in_reset(struct pqi_scsi_dev *device)
+{
+       return device->in_reset;
+}
 
 static inline void pqi_schedule_rescan_worker(struct pqi_ctrl_info *ctrl_info)
 {
@@ -2679,6 +2758,9 @@ static void pqi_event_worker(struct work_struct *work)
 
        ctrl_info = container_of(work, struct pqi_ctrl_info, event_work);
 
+       pqi_ctrl_busy(ctrl_info);
+       pqi_wait_if_ctrl_blocked(ctrl_info, NO_TIMEOUT);
+
        event = ctrl_info->events;
        for (i = 0; i < PQI_NUM_SUPPORTED_EVENTS; i++) {
                if (event->pending) {
@@ -2690,8 +2772,9 @@ static void pqi_event_worker(struct work_struct *work)
                event++;
        }
 
-       if (got_non_heartbeat_event)
-               pqi_schedule_rescan_worker(ctrl_info);
+       pqi_ctrl_unbusy(ctrl_info);
+
+       pqi_schedule_rescan_worker(ctrl_info);
 }
 
 static void pqi_take_ctrl_offline(struct pqi_ctrl_info *ctrl_info)
@@ -3436,6 +3519,13 @@ static int pqi_submit_raid_request_synchronous(struct pqi_ctrl_info *ctrl_info,
                }
        }
 
+       pqi_ctrl_busy(ctrl_info);
+       timeout_msecs = pqi_wait_if_ctrl_blocked(ctrl_info, timeout_msecs);
+       if (timeout_msecs == 0) {
+               rc = -ETIMEDOUT;
+               goto out;
+       }
+
        io_request = pqi_alloc_io_request(ctrl_info);
 
        put_unaligned_le16(io_request->index,
@@ -3476,6 +3566,8 @@ static int pqi_submit_raid_request_synchronous(struct pqi_ctrl_info *ctrl_info,
 
        pqi_free_io_request(io_request);
 
+out:
+       pqi_ctrl_unbusy(ctrl_info);
        up(&ctrl_info->sync_request_sem);
 
        return rc;
@@ -4499,6 +4591,19 @@ static int pqi_aio_submit_io(struct pqi_ctrl_info *ctrl_info,
        return 0;
 }
 
+/*
+ * This function gets called just before we hand the completed SCSI request
+ * back to the SML.
+ */
+
+void pqi_prep_for_scsi_done(struct scsi_cmnd *scmd)
+{
+       struct pqi_scsi_dev *device;
+
+       device = scmd->device->hostdata;
+       atomic_dec(&device->scsi_cmds_outstanding);
+}
+
 static int pqi_scsi_queue_command(struct Scsi_Host *shost,
        struct scsi_cmnd *scmd)
 {
@@ -4512,12 +4617,20 @@ static int pqi_scsi_queue_command(struct Scsi_Host *shost,
        device = scmd->device->hostdata;
        ctrl_info = shost_to_hba(shost);
 
+       atomic_inc(&device->scsi_cmds_outstanding);
+
        if (pqi_ctrl_offline(ctrl_info)) {
                set_host_byte(scmd, DID_NO_CONNECT);
                pqi_scsi_done(scmd);
                return 0;
        }
 
+       pqi_ctrl_busy(ctrl_info);
+       if (pqi_ctrl_blocked(ctrl_info) || pqi_device_in_reset(device)) {
+               rc = SCSI_MLQUEUE_HOST_BUSY;
+               goto out;
+       }
+
        /*
         * This is necessary because the SML doesn't zero out this field during
         * error recovery.
@@ -4554,9 +4667,116 @@ static int pqi_scsi_queue_command(struct Scsi_Host *shost,
                                queue_group);
        }
 
+out:
+       pqi_ctrl_unbusy(ctrl_info);
+       if (rc)
+               atomic_dec(&device->scsi_cmds_outstanding);
+
        return rc;
 }
 
+static int pqi_wait_until_queued_io_drained(struct pqi_ctrl_info *ctrl_info,
+       struct pqi_queue_group *queue_group)
+{
+       unsigned int path;
+       unsigned long flags;
+       bool list_is_empty;
+
+       for (path = 0; path < 2; path++) {
+               while (1) {
+                       spin_lock_irqsave(
+                               &queue_group->submit_lock[path], flags);
+                       list_is_empty =
+                               list_empty(&queue_group->request_list[path]);
+                       spin_unlock_irqrestore(
+                               &queue_group->submit_lock[path], flags);
+                       if (list_is_empty)
+                               break;
+                       pqi_check_ctrl_health(ctrl_info);
+                       if (pqi_ctrl_offline(ctrl_info))
+                               return -ENXIO;
+                       usleep_range(1000, 2000);
+               }
+       }
+
+       return 0;
+}
+
+static int pqi_wait_until_inbound_queues_empty(struct pqi_ctrl_info *ctrl_info)
+{
+       int rc;
+       unsigned int i;
+       unsigned int path;
+       struct pqi_queue_group *queue_group;
+       pqi_index_t iq_pi;
+       pqi_index_t iq_ci;
+
+       for (i = 0; i < ctrl_info->num_queue_groups; i++) {
+               queue_group = &ctrl_info->queue_groups[i];
+
+               rc = pqi_wait_until_queued_io_drained(ctrl_info, queue_group);
+               if (rc)
+                       return rc;
+
+               for (path = 0; path < 2; path++) {
+                       iq_pi = queue_group->iq_pi_copy[path];
+
+                       while (1) {
+                               iq_ci = *queue_group->iq_ci[path];
+                               if (iq_ci == iq_pi)
+                                       break;
+                               pqi_check_ctrl_health(ctrl_info);
+                               if (pqi_ctrl_offline(ctrl_info))
+                                       return -ENXIO;
+                               usleep_range(1000, 2000);
+                       }
+               }
+       }
+
+       return 0;
+}
+
+static void pqi_fail_io_queued_for_device(struct pqi_ctrl_info *ctrl_info,
+       struct pqi_scsi_dev *device)
+{
+       unsigned int i;
+       unsigned int path;
+       struct pqi_queue_group *queue_group;
+       unsigned long flags;
+       struct pqi_io_request *io_request;
+       struct pqi_io_request *next;
+       struct scsi_cmnd *scmd;
+       struct pqi_scsi_dev *scsi_device;
+
+       for (i = 0; i < ctrl_info->num_queue_groups; i++) {
+               queue_group = &ctrl_info->queue_groups[i];
+
+               for (path = 0; path < 2; path++) {
+                       spin_lock_irqsave(
+                               &queue_group->submit_lock[path], flags);
+
+                       list_for_each_entry_safe(io_request, next,
+                               &queue_group->request_list[path],
+                               request_list_entry) {
+                               scmd = io_request->scmd;
+                               if (!scmd)
+                                       continue;
+
+                               scsi_device = scmd->device->hostdata;
+                               if (scsi_device != device)
+                                       continue;
+
+                               list_del(&io_request->request_list_entry);
+                               set_host_byte(scmd, DID_RESET);
+                               pqi_scsi_done(scmd);
+                       }
+
+                       spin_unlock_irqrestore(
+                               &queue_group->submit_lock[path], flags);
+               }
+       }
+}
+
 static void pqi_lun_reset_complete(struct pqi_io_request *io_request,
        void *context)
 {
@@ -4571,7 +4791,6 @@ static int pqi_wait_for_lun_reset_completion(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device, struct completion *wait)
 {
        int rc;
-       unsigned int wait_secs = 0;
 
        while (1) {
                if (wait_for_completion_io_timeout(wait,
@@ -4585,13 +4804,6 @@ static int pqi_wait_for_lun_reset_completion(struct pqi_ctrl_info *ctrl_info,
                        rc = -ETIMEDOUT;
                        break;
                }
-
-               wait_secs += PQI_LUN_RESET_TIMEOUT_SECS;
-
-               dev_err(&ctrl_info->pci_dev->dev,
-                       "resetting scsi %d:%d:%d:%d - waiting %u seconds\n",
-                       ctrl_info->scsi_host->host_no, device->bus,
-                       device->target, device->lun, wait_secs);
        }
 
        return rc;
@@ -4605,8 +4817,6 @@ static int pqi_lun_reset(struct pqi_ctrl_info *ctrl_info,
        DECLARE_COMPLETION_ONSTACK(wait);
        struct pqi_task_management_request *request;
 
-       down(&ctrl_info->lun_reset_sem);
-
        io_request = pqi_alloc_io_request(ctrl_info);
        io_request->io_complete_callback = pqi_lun_reset_complete;
        io_request->context = &wait;
@@ -4631,7 +4841,6 @@ static int pqi_lun_reset(struct pqi_ctrl_info *ctrl_info,
                rc = io_request->status;
 
        pqi_free_io_request(io_request);
-       up(&ctrl_info->lun_reset_sem);
 
        return rc;
 }
@@ -4643,10 +4852,6 @@ static int pqi_device_reset(struct pqi_ctrl_info *ctrl_info,
 {
        int rc;
 
-       pqi_check_ctrl_health(ctrl_info);
-       if (pqi_ctrl_offline(ctrl_info))
-               return FAILED;
-
        rc = pqi_lun_reset(ctrl_info, device);
 
        return rc == 0 ? SUCCESS : FAILED;
@@ -4655,23 +4860,46 @@ static int pqi_device_reset(struct pqi_ctrl_info *ctrl_info,
 static int pqi_eh_device_reset_handler(struct scsi_cmnd *scmd)
 {
        int rc;
+       struct Scsi_Host *shost;
        struct pqi_ctrl_info *ctrl_info;
        struct pqi_scsi_dev *device;
 
-       ctrl_info = shost_to_hba(scmd->device->host);
+       shost = scmd->device->host;
+       ctrl_info = shost_to_hba(shost);
        device = scmd->device->hostdata;
 
        dev_err(&ctrl_info->pci_dev->dev,
                "resetting scsi %d:%d:%d:%d\n",
-               ctrl_info->scsi_host->host_no,
-               device->bus, device->target, device->lun);
+               shost->host_no, device->bus, device->target, device->lun);
 
-       rc = pqi_device_reset(ctrl_info, device);
+       pqi_check_ctrl_health(ctrl_info);
+       if (pqi_ctrl_offline(ctrl_info)) {
+               rc = FAILED;
+               goto out;
+       }
 
+       mutex_lock(&ctrl_info->lun_reset_mutex);
+
+       pqi_ctrl_block_requests(ctrl_info);
+       pqi_ctrl_wait_until_quiesced(ctrl_info);
+       pqi_fail_io_queued_for_device(ctrl_info, device);
+       rc = pqi_wait_until_inbound_queues_empty(ctrl_info);
+       pqi_device_reset_start(device);
+       pqi_ctrl_unblock_requests(ctrl_info);
+
+       if (rc)
+               rc = FAILED;
+       else
+               rc = pqi_device_reset(ctrl_info, device);
+
+       pqi_device_reset_done(device);
+
+       mutex_unlock(&ctrl_info->lun_reset_mutex);
+
+out:
        dev_err(&ctrl_info->pci_dev->dev,
                "reset of scsi %d:%d:%d:%d: %s\n",
-               ctrl_info->scsi_host->host_no,
-               device->bus, device->target, device->lun,
+               shost->host_no, device->bus, device->target, device->lun,
                rc == SUCCESS ? "SUCCESS" : "FAILED");
 
        return rc;
@@ -5552,6 +5780,7 @@ static struct pqi_ctrl_info *pqi_alloc_ctrl_info(int numa_node)
                return NULL;
 
        mutex_init(&ctrl_info->scan_mutex);
+       mutex_init(&ctrl_info->lun_reset_mutex);
 
        INIT_LIST_HEAD(&ctrl_info->scsi_device_list);
        spin_lock_init(&ctrl_info->scsi_device_list_lock);
@@ -5564,7 +5793,7 @@ static struct pqi_ctrl_info *pqi_alloc_ctrl_info(int numa_node)
 
        sema_init(&ctrl_info->sync_request_sem,
                PQI_RESERVED_IO_SLOTS_SYNCHRONOUS_REQUESTS);
-       sema_init(&ctrl_info->lun_reset_sem, PQI_RESERVED_IO_SLOTS_LUN_RESET);
+       init_waitqueue_head(&ctrl_info->block_requests_wait);
 
        ctrl_info->ctrl_id = atomic_inc_return(&pqi_controller_count) - 1;
        ctrl_info->max_msix_vectors = PQI_MAX_MSIX_VECTORS;