[SCSI] Unlock accesses to eh_deadline
authorHannes Reinecke <hare@suse.de>
Mon, 11 Nov 2013 12:44:55 +0000 (13:44 +0100)
committerJames Bottomley <JBottomley@Parallels.com>
Thu, 19 Dec 2013 15:39:02 +0000 (07:39 -0800)
32bit accesses are guaranteed to be atomic, so we can remove
the spinlock when checking for eh_deadline. We only need to
make sure to catch any updates which might happened during
the call to time_before(); if so we just recheck with the
correct value.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
drivers/scsi/scsi_error.c

index 3dd04026d46663521439e746623292094adb197c..4e4824beefe48ae525cfb80461942744c567467c 100644 (file)
@@ -94,8 +94,15 @@ static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
        if (!shost->last_reset || !shost->eh_deadline)
                return 0;
 
-       if (time_before(jiffies,
-                       shost->last_reset + shost->eh_deadline))
+       /*
+        * 32bit accesses are guaranteed to be atomic
+        * (on all supported architectures), so instead
+        * of using a spinlock we can as well double check
+        * if eh_deadline has been unset during the
+        * time_before call.
+        */
+       if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
+           shost->eh_deadline != 0)
                return 0;
 
        return 1;
@@ -111,18 +118,14 @@ scmd_eh_abort_handler(struct work_struct *work)
        struct scsi_cmnd *scmd =
                container_of(work, struct scsi_cmnd, abort_work.work);
        struct scsi_device *sdev = scmd->device;
-       unsigned long flags;
        int rtn;
 
-       spin_lock_irqsave(sdev->host->host_lock, flags);
        if (scsi_host_eh_past_deadline(sdev->host)) {
-               spin_unlock_irqrestore(sdev->host->host_lock, flags);
                SCSI_LOG_ERROR_RECOVERY(3,
                        scmd_printk(KERN_INFO, scmd,
                                    "scmd %p eh timeout, not aborting\n",
                                    scmd));
        } else {
-               spin_unlock_irqrestore(sdev->host->host_lock, flags);
                SCSI_LOG_ERROR_RECOVERY(3,
                        scmd_printk(KERN_INFO, scmd,
                                    "aborting command %p\n", scmd));
@@ -1132,7 +1135,6 @@ int scsi_eh_get_sense(struct list_head *work_q,
        struct scsi_cmnd *scmd, *next;
        struct Scsi_Host *shost;
        int rtn;
-       unsigned long flags;
 
        list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
                if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
@@ -1140,16 +1142,13 @@ int scsi_eh_get_sense(struct list_head *work_q,
                        continue;
 
                shost = scmd->device->host;
-               spin_lock_irqsave(shost->host_lock, flags);
                if (scsi_host_eh_past_deadline(shost)) {
-                       spin_unlock_irqrestore(shost->host_lock, flags);
                        SCSI_LOG_ERROR_RECOVERY(3,
                                shost_printk(KERN_INFO, shost,
                                            "skip %s, past eh deadline\n",
                                             __func__));
                        break;
                }
-               spin_unlock_irqrestore(shost->host_lock, flags);
                SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
                                                  "%s: requesting sense\n",
                                                  current->comm));
@@ -1235,26 +1234,21 @@ static int scsi_eh_test_devices(struct list_head *cmd_list,
        struct scsi_cmnd *scmd, *next;
        struct scsi_device *sdev;
        int finish_cmds;
-       unsigned long flags;
 
        while (!list_empty(cmd_list)) {
                scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
                sdev = scmd->device;
 
                if (!try_stu) {
-                       spin_lock_irqsave(sdev->host->host_lock, flags);
                        if (scsi_host_eh_past_deadline(sdev->host)) {
                                /* Push items back onto work_q */
                                list_splice_init(cmd_list, work_q);
-                               spin_unlock_irqrestore(sdev->host->host_lock,
-                                                      flags);
                                SCSI_LOG_ERROR_RECOVERY(3,
                                        shost_printk(KERN_INFO, sdev->host,
                                                     "skip %s, past eh deadline",
                                                     __func__));
                                break;
                        }
-                       spin_unlock_irqrestore(sdev->host->host_lock, flags);
                }
 
                finish_cmds = !scsi_device_online(scmd->device) ||
@@ -1295,15 +1289,12 @@ static int scsi_eh_abort_cmds(struct list_head *work_q,
        LIST_HEAD(check_list);
        int rtn;
        struct Scsi_Host *shost;
-       unsigned long flags;
 
        list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
                if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
                        continue;
                shost = scmd->device->host;
-               spin_lock_irqsave(shost->host_lock, flags);
                if (scsi_host_eh_past_deadline(shost)) {
-                       spin_unlock_irqrestore(shost->host_lock, flags);
                        list_splice_init(&check_list, work_q);
                        SCSI_LOG_ERROR_RECOVERY(3,
                                shost_printk(KERN_INFO, shost,
@@ -1311,7 +1302,6 @@ static int scsi_eh_abort_cmds(struct list_head *work_q,
                                             __func__));
                        return list_empty(work_q);
                }
-               spin_unlock_irqrestore(shost->host_lock, flags);
                SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
                                                  "0x%p\n", current->comm,
                                                  scmd));
@@ -1375,19 +1365,15 @@ static int scsi_eh_stu(struct Scsi_Host *shost,
 {
        struct scsi_cmnd *scmd, *stu_scmd, *next;
        struct scsi_device *sdev;
-       unsigned long flags;
 
        shost_for_each_device(sdev, shost) {
-               spin_lock_irqsave(shost->host_lock, flags);
                if (scsi_host_eh_past_deadline(shost)) {
-                       spin_unlock_irqrestore(shost->host_lock, flags);
                        SCSI_LOG_ERROR_RECOVERY(3,
                                shost_printk(KERN_INFO, shost,
                                            "skip %s, past eh deadline\n",
                                             __func__));
                        break;
                }
-               spin_unlock_irqrestore(shost->host_lock, flags);
                stu_scmd = NULL;
                list_for_each_entry(scmd, work_q, eh_entry)
                        if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
@@ -1441,20 +1427,16 @@ static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
 {
        struct scsi_cmnd *scmd, *bdr_scmd, *next;
        struct scsi_device *sdev;
-       unsigned long flags;
        int rtn;
 
        shost_for_each_device(sdev, shost) {
-               spin_lock_irqsave(shost->host_lock, flags);
                if (scsi_host_eh_past_deadline(shost)) {
-                       spin_unlock_irqrestore(shost->host_lock, flags);
                        SCSI_LOG_ERROR_RECOVERY(3,
                                shost_printk(KERN_INFO, shost,
                                            "skip %s, past eh deadline\n",
                                             __func__));
                        break;
                }
-               spin_unlock_irqrestore(shost->host_lock, flags);
                bdr_scmd = NULL;
                list_for_each_entry(scmd, work_q, eh_entry)
                        if (scmd->device == sdev) {
@@ -1515,11 +1497,8 @@ static int scsi_eh_target_reset(struct Scsi_Host *shost,
                struct scsi_cmnd *next, *scmd;
                int rtn;
                unsigned int id;
-               unsigned long flags;
 
-               spin_lock_irqsave(shost->host_lock, flags);
                if (scsi_host_eh_past_deadline(shost)) {
-                       spin_unlock_irqrestore(shost->host_lock, flags);
                        /* push back on work queue for further processing */
                        list_splice_init(&check_list, work_q);
                        list_splice_init(&tmp_list, work_q);
@@ -1529,7 +1508,6 @@ static int scsi_eh_target_reset(struct Scsi_Host *shost,
                                             __func__));
                        return list_empty(work_q);
                }
-               spin_unlock_irqrestore(shost->host_lock, flags);
 
                scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
                id = scmd_id(scmd);
@@ -1574,7 +1552,6 @@ static int scsi_eh_bus_reset(struct Scsi_Host *shost,
        LIST_HEAD(check_list);
        unsigned int channel;
        int rtn;
-       unsigned long flags;
 
        /*
         * we really want to loop over the various channels, and do this on
@@ -1584,9 +1561,7 @@ static int scsi_eh_bus_reset(struct Scsi_Host *shost,
         */
 
        for (channel = 0; channel <= shost->max_channel; channel++) {
-               spin_lock_irqsave(shost->host_lock, flags);
                if (scsi_host_eh_past_deadline(shost)) {
-                       spin_unlock_irqrestore(shost->host_lock, flags);
                        list_splice_init(&check_list, work_q);
                        SCSI_LOG_ERROR_RECOVERY(3,
                                shost_printk(KERN_INFO, shost,
@@ -1594,7 +1569,6 @@ static int scsi_eh_bus_reset(struct Scsi_Host *shost,
                                             __func__));
                        return list_empty(work_q);
                }
-               spin_unlock_irqrestore(shost->host_lock, flags);
 
                chan_scmd = NULL;
                list_for_each_entry(scmd, work_q, eh_entry) {