scsi: fc: implement kref backed reference counting
authorJohannes Thumshirn <jthumshirn@suse.de>
Thu, 17 Nov 2016 09:31:17 +0000 (10:31 +0100)
committerMartin K. Petersen <martin.petersen@oracle.com>
Fri, 18 Nov 2016 01:15:25 +0000 (20:15 -0500)
Implement kref backed reference counting instead of rolling our own. This
elimnates the need of the following fields in 'struct fc_bsg_job':
* ref_cnt
* state_flags
* job_lock
bringing us close to unification of 'struct fc_bsg_job' and 'struct bsg_job'.

Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/scsi_transport_fc.c
include/scsi/scsi_transport_fc.h

index 45954a6edd9474fc8e89548a6ba8097f27b3a0c7..9009acc27aedc997793357e590d971af628d7e53 100644 (file)
@@ -3560,16 +3560,12 @@ fc_vport_sched_delete(struct work_struct *work)
  * @job:       fc_bsg_job that is to be torn down
  */
 static void
-fc_destroy_bsgjob(struct fc_bsg_job *job)
+fc_destroy_bsgjob(struct kref *kref)
 {
-       unsigned long flags;
+       struct fc_bsg_job *job = container_of(kref, struct fc_bsg_job, kref);
+       struct request *rq = job->req;
 
-       spin_lock_irqsave(&job->job_lock, flags);
-       if (job->ref_cnt) {
-               spin_unlock_irqrestore(&job->job_lock, flags);
-               return;
-       }
-       spin_unlock_irqrestore(&job->job_lock, flags);
+       blk_end_request_all(rq, rq->errors);
 
        put_device(job->dev);   /* release reference for the request */
 
@@ -3620,15 +3616,8 @@ EXPORT_SYMBOL_GPL(fc_bsg_jobdone);
 static void fc_bsg_softirq_done(struct request *rq)
 {
        struct fc_bsg_job *job = rq->special;
-       unsigned long flags;
 
-       spin_lock_irqsave(&job->job_lock, flags);
-       job->state_flags |= FC_RQST_STATE_DONE;
-       job->ref_cnt--;
-       spin_unlock_irqrestore(&job->job_lock, flags);
-
-       blk_end_request_all(rq, rq->errors);
-       fc_destroy_bsgjob(job);
+       kref_put(&job->kref, fc_destroy_bsgjob);
 }
 
 /**
@@ -3642,24 +3631,18 @@ fc_bsg_job_timeout(struct request *req)
        struct Scsi_Host *shost = fc_bsg_to_shost(job);
        struct fc_rport *rport = fc_bsg_to_rport(job);
        struct fc_internal *i = to_fc_internal(shost->transportt);
-       unsigned long flags;
-       int err = 0, done = 0;
+       int err = 0, inflight = 0;
 
        if (rport && rport->port_state == FC_PORTSTATE_BLOCKED)
                return BLK_EH_RESET_TIMER;
 
-       spin_lock_irqsave(&job->job_lock, flags);
-       if (job->state_flags & FC_RQST_STATE_DONE)
-               done = 1;
-       else
-               job->ref_cnt++;
-       spin_unlock_irqrestore(&job->job_lock, flags);
+       inflight = kref_get_unless_zero(&job->kref);
 
-       if (!done && i->f->bsg_timeout) {
+       if (inflight && i->f->bsg_timeout) {
                /* call LLDD to abort the i/o as it has timed out */
                err = i->f->bsg_timeout(job);
                if (err == -EAGAIN) {
-                       job->ref_cnt--;
+                       kref_put(&job->kref, fc_destroy_bsgjob);
                        return BLK_EH_RESET_TIMER;
                } else if (err)
                        printk(KERN_ERR "ERROR: FC BSG request timeout - LLD "
@@ -3667,7 +3650,7 @@ fc_bsg_job_timeout(struct request *req)
        }
 
        /* the blk_end_sync_io() doesn't check the error */
-       if (done)
+       if (!inflight)
                return BLK_EH_NOT_HANDLED;
        else
                return BLK_EH_HANDLED;
@@ -3728,7 +3711,6 @@ fc_req_to_bsgjob(struct Scsi_Host *shost, struct fc_rport *rport,
        job->req = req;
        if (i->f->dd_bsg_size)
                job->dd_data = (void *)&job[1];
-       spin_lock_init(&job->job_lock);
        job->request = (struct fc_bsg_request *)req->cmd;
        job->request_len = req->cmd_len;
        job->reply = req->sense;
@@ -3750,7 +3732,7 @@ fc_req_to_bsgjob(struct Scsi_Host *shost, struct fc_rport *rport,
                job->dev = &shost->shost_gendev;
        get_device(job->dev);           /* take a reference for the request */
 
-       job->ref_cnt = 1;
+       kref_init(&job->kref);
 
        return 0;
 
index 9f53fe3fa249fb546b8e1c9be70c1f4705f615fa..8ae5680f2216bedd42dee913fc7eb342055ac41b 100644 (file)
@@ -634,9 +634,7 @@ struct fc_bsg_job {
        struct fc_rport *rport;
        struct device *dev;
        struct request *req;
-       spinlock_t job_lock;
-       unsigned int state_flags;
-       unsigned int ref_cnt;
+       struct kref kref;
 
        struct fc_bsg_request *request;
        struct fc_bsg_reply *reply;