[SCSI] bnx2fc: Fix race condition between IO completion and abort

When IO is successfully completed while an abort is pending, eh_abort
incorrectly assumes that abort failed and performes recovery by issuing
cleanup. Howerver, cleanup timesout as the firmware has no clue about
this IO. Fix this by checking if the IO has already completed.

Signed-off-by: Bhanu Prakash Gollapudi <bprakash@broadcom.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>

diff --git a/drivers/scsi/bnx2fc/bnx2fc_io.c b/drivers/scsi/bnx2fc/bnx2fc_io.c
index 723a9a8..e80ca01 100644 (file)
--- a/drivers/scsi/bnx2fc/bnx2fc_io.c
+++ b/drivers/scsi/bnx2fc/bnx2fc_io.c
@@ -1270,8 +1270,11 @@ int bnx2fc_eh_abort(struct scsi_cmnd *sc_cmd)
 
        spin_lock_bh(&tgt->tgt_lock);
        io_req->wait_for_comp = 0;
-       if (!(test_and_set_bit(BNX2FC_FLAG_ABTS_DONE,
-                                   &io_req->req_flags))) {
+       if (test_bit(BNX2FC_FLAG_IO_COMPL, &io_req->req_flags)) {
+               BNX2FC_IO_DBG(io_req, "IO completed in a different context\n");
+               rc = SUCCESS;
+       } else if (!(test_and_set_bit(BNX2FC_FLAG_ABTS_DONE,
+                                     &io_req->req_flags))) {
                /* Let the scsi-ml try to recover this command */
                printk(KERN_ERR PFX "abort failed, xid = 0x%x\n",
                       io_req->xid);