scsi: csiostor: Fix completion usage

The (re)initializing of the completion object should be done before we
trigger the transfer. Doing this after triggering the hardware opens up
a race window. Without the timeout we would problaly even deadlock. Use
also reinit_completion because we initalize the whole data structure in
csio_scscim_init().

There is only one waiter for the completion, therefore there is no need
to use complete_all(). Let's make that clear by using complete() instead
of complete_all().

The usage pattern of the completion is:

waiter context                          waker context

csio_eh_abort_handler()
  reinit_completion()
  wait_for_completion_timeout()

                                        csio_scsi_err_handler()
                                          complete()

[mkp: fix typo]

Signed-off-by: Daniel Wagner <daniel.wagner@bmw-carit.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

diff --git a/drivers/scsi/csiostor/csio_scsi.c b/drivers/scsi/csiostor/csio_scsi.c
index c2a6f9f..89a52b9 100644 (file)
--- a/drivers/scsi/csiostor/csio_scsi.c
+++ b/drivers/scsi/csiostor/csio_scsi.c
@@ -1721,7 +1721,7 @@ out:
 
        /* Wake up waiting threads */
        csio_scsi_cmnd(req) = NULL;
-       complete_all(&req->cmplobj);
+       complete(&req->cmplobj);
 }
 
 /*
@@ -1945,6 +1945,7 @@ csio_eh_abort_handler(struct scsi_cmnd *cmnd)
        ready = csio_is_lnode_ready(ln);
        tmo = CSIO_SCSI_ABRT_TMO_MS;
 
+       reinit_completion(&ioreq->cmplobj);
        spin_lock_irq(&hw->lock);
        rv = csio_do_abrt_cls(hw, ioreq, (ready ? SCSI_ABORT : SCSI_CLOSE));
        spin_unlock_irq(&hw->lock);
@@ -1964,8 +1965,6 @@ csio_eh_abort_handler(struct scsi_cmnd *cmnd)
                goto inval_scmnd;
        }
 
-       /* Wait for completion */
-       init_completion(&ioreq->cmplobj);
        wait_for_completion_timeout(&ioreq->cmplobj, msecs_to_jiffies(tmo));
 
        /* FW didnt respond to abort within our timeout */