[SCSI] zfcp: receiving an unsolicted status can lead to I/O stall

Processing of an unsolicted status request can lead to a locking race
of the request_queue's queue_lock during the recreation of the
used up status read request while still in interrupt context
of the response handler.

Detaching the 'refill' of the long running status read requests from
the handler to a scheduled work is solving this issue.

In addition, each refill-run is trying to re-establish the full amount
of status read requests, which might have failed in earlier runs.

Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

diff --git a/drivers/s390/scsi/zfcp_aux.c b/drivers/s390/scsi/zfcp_aux.c
index d23027a2f2f06736adc62a15d108592d019dfc32..41635b13ccb12e60ba484642967c8d02431d3260 100644 (file)
--- a/drivers/s390/scsi/zfcp_aux.c
+++ b/drivers/s390/scsi/zfcp_aux.c
@@ -970,6 +970,27 @@ static void zfcp_dummy_release(struct device *dev)
        return;
 }
 
+int zfcp_status_read_refill(struct zfcp_adapter *adapter)
+{
+       while (atomic_read(&adapter->stat_miss) > 0)
+               if (zfcp_fsf_status_read(adapter, ZFCP_WAIT_FOR_SBAL))
+                       break;
+       else
+               atomic_dec(&adapter->stat_miss);
+
+       if (ZFCP_STATUS_READS_RECOM <= atomic_read(&adapter->stat_miss)) {
+               zfcp_erp_adapter_reopen(adapter, 0, 103, NULL);
+               return 1;
+       }
+       return 0;
+}
+
+static void _zfcp_status_read_scheduler(struct work_struct *work)
+{
+       zfcp_status_read_refill(container_of(work, struct zfcp_adapter,
+                                            stat_work));
+}
+
 /*
  * Enqueues an adapter at the end of the adapter list in the driver data.
  * All adapter internal structures are set up.
@@ -1063,6 +1084,7 @@ zfcp_adapter_enqueue(struct ccw_device *ccw_device)
 
        /* initialize lock of associated request queue */
        rwlock_init(&adapter->request_queue.queue_lock);
+       INIT_WORK(&adapter->stat_work, _zfcp_status_read_scheduler);
 
        /* mark adapter unusable as long as sysfs registration is not complete */
        atomic_set_mask(ZFCP_STATUS_COMMON_REMOVE, &adapter->status);
@@ -1123,6 +1145,7 @@ zfcp_adapter_dequeue(struct zfcp_adapter *adapter)
        int retval = 0;
        unsigned long flags;
 
+       cancel_work_sync(&adapter->stat_work);
        zfcp_adapter_scsi_unregister(adapter);
        device_unregister(&adapter->generic_services);
        zfcp_sysfs_adapter_remove_files(&adapter->ccw_device->dev);

diff --git a/drivers/s390/scsi/zfcp_dbf.c b/drivers/s390/scsi/zfcp_dbf.c
index c8bad675dbd1717992c6c1db287415fa4e40b933..1710c12a32c43a5506d8ff08514dd5eeb9ab4422 100644 (file)
--- a/drivers/s390/scsi/zfcp_dbf.c
+++ b/drivers/s390/scsi/zfcp_dbf.c
@@ -268,7 +268,7 @@ void zfcp_hba_dbf_event_fsf_unsol(const char *tag, struct zfcp_adapter *adapter,
        strncpy(rec->tag, "stat", ZFCP_DBF_TAG_SIZE);
        strncpy(rec->tag2, tag, ZFCP_DBF_TAG_SIZE);
 
-       rec->u.status.failed = adapter->status_read_failed;
+       rec->u.status.failed = atomic_read(&adapter->stat_miss);
        if (status_buffer != NULL) {
                rec->u.status.status_type = status_buffer->status_type;
                rec->u.status.status_subtype = status_buffer->status_subtype;

diff --git a/drivers/s390/scsi/zfcp_def.h b/drivers/s390/scsi/zfcp_def.h
index 306fcd0cae310bb8e25778de2d410dce50157787..adfbbb07448a27a47d761b60b1b109ebfb3f3e76 100644 (file)
--- a/drivers/s390/scsi/zfcp_def.h
+++ b/drivers/s390/scsi/zfcp_def.h
@@ -136,7 +136,6 @@ zfcp_address_to_sg(void *address, struct scatterlist *list, unsigned int size)
 #define ZFCP_QTCB_VERSION      FSF_QTCB_CURRENT_VERSION
 /* ATTENTION: value must not be used by hardware */
 #define FSF_QTCB_UNSOLICITED_STATUS            0x6305
-#define ZFCP_STATUS_READ_FAILED_THRESHOLD      3
 #define ZFCP_STATUS_READS_RECOM                        FSF_STATUS_READS_RECOM
 
 /* Do 1st retry in 1 second, then double the timeout for each following retry */
@@ -759,7 +758,8 @@ struct zfcp_adapter {
        rwlock_t                abort_lock;        /* Protects against SCSI
                                                      stack abort/command
                                                      completion races */
-       u16                     status_read_failed; /* # failed status reads */
+       atomic_t                stat_miss;         /* # missing status reads*/
+       struct work_struct      stat_work;
        atomic_t                status;            /* status of this adapter */
        struct list_head        erp_ready_head;    /* error recovery for this
                                                      adapter/devices */

diff --git a/drivers/s390/scsi/zfcp_erp.c b/drivers/s390/scsi/zfcp_erp.c
index 805484658dd9ab618211316c161149850b6f9caf..55a4fdc42626898f7e6678a549c37a3f031234b4 100644 (file)
--- a/drivers/s390/scsi/zfcp_erp.c
+++ b/drivers/s390/scsi/zfcp_erp.c
@@ -2139,25 +2139,10 @@ static int
 zfcp_erp_adapter_strategy_open_fsf_statusread(struct zfcp_erp_action
                                              *erp_action)
 {
-       int retval = ZFCP_ERP_SUCCEEDED;
-       int temp_ret;
        struct zfcp_adapter *adapter = erp_action->adapter;
-       int i;
 
-       adapter->status_read_failed = 0;
-       for (i = 0; i < ZFCP_STATUS_READS_RECOM; i++) {
-               temp_ret = zfcp_fsf_status_read(adapter, ZFCP_WAIT_FOR_SBAL);
-               if (temp_ret < 0) {
-                       ZFCP_LOG_INFO("error: set-up of unsolicited status "
-                                     "notification failed on adapter %s\n",
-                                     zfcp_get_busid_by_adapter(adapter));
-                       retval = ZFCP_ERP_FAILED;
-                       i--;
-                       break;
-               }
-       }
-
-       return retval;
+       atomic_set(&adapter->stat_miss, 16);
+       return zfcp_status_read_refill(adapter);
 }
 
 /*

diff --git a/drivers/s390/scsi/zfcp_ext.h b/drivers/s390/scsi/zfcp_ext.h
index 6abf178fda5df2a4fc641215e381999eb1248238..250565794fa536410a90ea3e5514dd031217c29e 100644 (file)
--- a/drivers/s390/scsi/zfcp_ext.h
+++ b/drivers/s390/scsi/zfcp_ext.h
@@ -92,6 +92,7 @@ extern void zfcp_fsf_start_timer(struct zfcp_fsf_req *, unsigned long);
 extern void zfcp_erp_start_timer(struct zfcp_fsf_req *);
 extern void zfcp_fsf_req_dismiss_all(struct zfcp_adapter *);
 extern int  zfcp_fsf_status_read(struct zfcp_adapter *, int);
+extern int zfcp_status_read_refill(struct zfcp_adapter *adapter);
 extern int zfcp_fsf_req_create(struct zfcp_adapter *, u32, int, mempool_t *,
                               unsigned long *, struct zfcp_fsf_req **);
 extern int zfcp_fsf_send_ct(struct zfcp_send_ct *, mempool_t *,

diff --git a/drivers/s390/scsi/zfcp_fsf.c b/drivers/s390/scsi/zfcp_fsf.c
index 1e7136483c1b24d3a49cfff235109adba06940a2..b344e8a72f1fdee595db0fe158dec6dcef9c91bf 100644 (file)
--- a/drivers/s390/scsi/zfcp_fsf.c
+++ b/drivers/s390/scsi/zfcp_fsf.c
@@ -1029,21 +1029,9 @@ zfcp_fsf_status_read_handler(struct zfcp_fsf_req *fsf_req)
         * FIXME:
         * allocation failure possible? (Is this code needed?)
         */
-       retval = zfcp_fsf_status_read(adapter, 0);
-       if (retval < 0) {
-               ZFCP_LOG_INFO("Failed to create unsolicited status read "
-                             "request for the adapter %s.\n",
-                             zfcp_get_busid_by_adapter(adapter));
-               /* temporary fix to avoid status read buffer shortage */
-               adapter->status_read_failed++;
-               if ((ZFCP_STATUS_READS_RECOM - adapter->status_read_failed)
-                   < ZFCP_STATUS_READ_FAILED_THRESHOLD) {
-                       ZFCP_LOG_INFO("restart adapter %s due to status read "
-                                     "buffer shortage\n",
-                                     zfcp_get_busid_by_adapter(adapter));
-                       zfcp_erp_adapter_reopen(adapter, 0, 103, fsf_req);
-               }
-       }
+
+       atomic_inc(&adapter->stat_miss);
+       schedule_work(&adapter->stat_work);
  out:
        return retval;
 }