[SCSI] fnic: Hitting BUG_ON(io_req->abts_done) in fnic_rport_exch_reset
authorSesidhar Beddel <sebaddel@cisco.com>
Mon, 9 Sep 2013 20:31:49 +0000 (13:31 -0700)
committerJames Bottomley <JBottomley@Parallels.com>
Wed, 11 Sep 2013 22:54:08 +0000 (15:54 -0700)
Hitting BUG_ON(io_req->abts_done) in fnic_rport_exch_reset in case of
timing issue and also to some extent locking issue where abts and terminate
is happening around same timing.

The code changes are intended to update CMD_STATE(sc) and
io_req->abts_done together.

Signed-off-by: Sesidhar Beddel <sebaddel@cisco.com>
Signed-off-by: Hiral Patel <hiralpat@cisco.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
drivers/scsi/fnic/fnic_scsi.c

index 4ff0a332f29812a536529f48c1cfc4251b154ac6..fb8413aa005f2ccf47291531487e74c4c68c9053 100644 (file)
@@ -111,6 +111,12 @@ static inline spinlock_t *fnic_io_lock_hash(struct fnic *fnic,
        return &fnic->io_req_lock[hash];
 }
 
+static inline spinlock_t *fnic_io_lock_tag(struct fnic *fnic,
+                                           int tag)
+{
+       return &fnic->io_req_lock[tag & (FNIC_IO_LOCKS - 1)];
+}
+
 /*
  * Unmap the data buffer and sense buffer for an io_req,
  * also unmap and free the device-private scatter/gather list.
@@ -956,9 +962,7 @@ static void fnic_fcpio_itmf_cmpl_handler(struct fnic *fnic,
                        spin_unlock_irqrestore(io_lock, flags);
                        return;
                }
-               CMD_STATE(sc) = FNIC_IOREQ_ABTS_COMPLETE;
                CMD_ABTS_STATUS(sc) = hdr_status;
-
                CMD_FLAGS(sc) |= FNIC_IO_ABT_TERM_DONE;
                FNIC_SCSI_DBG(KERN_DEBUG, fnic->lport->host,
                              "abts cmpl recd. id %d status %s\n",
@@ -1116,7 +1120,7 @@ int fnic_wq_copy_cmpl_handler(struct fnic *fnic, int copy_work_to_do)
 
 static void fnic_cleanup_io(struct fnic *fnic, int exclude_id)
 {
-       unsigned int i;
+       int i;
        struct fnic_io_req *io_req;
        unsigned long flags = 0;
        struct scsi_cmnd *sc;
@@ -1127,12 +1131,14 @@ static void fnic_cleanup_io(struct fnic *fnic, int exclude_id)
                if (i == exclude_id)
                        continue;
 
+               io_lock = fnic_io_lock_tag(fnic, i);
+               spin_lock_irqsave(io_lock, flags);
                sc = scsi_host_find_tag(fnic->lport->host, i);
-               if (!sc)
+               if (!sc) {
+                       spin_unlock_irqrestore(io_lock, flags);
                        continue;
+               }
 
-               io_lock = fnic_io_lock_hash(fnic, sc);
-               spin_lock_irqsave(io_lock, flags);
                io_req = (struct fnic_io_req *)CMD_SP(sc);
                if ((CMD_FLAGS(sc) & FNIC_DEVICE_RESET) &&
                        !(CMD_FLAGS(sc) & FNIC_DEV_RST_DONE)) {
@@ -1310,12 +1316,13 @@ static void fnic_rport_exch_reset(struct fnic *fnic, u32 port_id)
 
        for (tag = 0; tag < FNIC_MAX_IO_REQ; tag++) {
                abt_tag = tag;
+               io_lock = fnic_io_lock_tag(fnic, tag);
+               spin_lock_irqsave(io_lock, flags);
                sc = scsi_host_find_tag(fnic->lport->host, tag);
-               if (!sc)
+               if (!sc) {
+                       spin_unlock_irqrestore(io_lock, flags);
                        continue;
-
-               io_lock = fnic_io_lock_hash(fnic, sc);
-               spin_lock_irqsave(io_lock, flags);
+               }
 
                io_req = (struct fnic_io_req *)CMD_SP(sc);
 
@@ -1426,16 +1433,19 @@ void fnic_terminate_rport_io(struct fc_rport *rport)
 
        for (tag = 0; tag < FNIC_MAX_IO_REQ; tag++) {
                abt_tag = tag;
+               io_lock = fnic_io_lock_tag(fnic, tag);
+               spin_lock_irqsave(io_lock, flags);
                sc = scsi_host_find_tag(fnic->lport->host, tag);
-               if (!sc)
+               if (!sc) {
+                       spin_unlock_irqrestore(io_lock, flags);
                        continue;
+               }
 
                cmd_rport = starget_to_rport(scsi_target(sc->device));
-               if (rport != cmd_rport)
+               if (rport != cmd_rport) {
+                       spin_unlock_irqrestore(io_lock, flags);
                        continue;
-
-               io_lock = fnic_io_lock_hash(fnic, sc);
-               spin_lock_irqsave(io_lock, flags);
+               }
 
                io_req = (struct fnic_io_req *)CMD_SP(sc);
 
@@ -1648,13 +1658,15 @@ int fnic_abort_cmd(struct scsi_cmnd *sc)
        io_req->abts_done = NULL;
 
        /* fw did not complete abort, timed out */
-       if (CMD_STATE(sc) == FNIC_IOREQ_ABTS_PENDING) {
+       if (CMD_ABTS_STATUS(sc) == FCPIO_INVALID_CODE) {
                spin_unlock_irqrestore(io_lock, flags);
                CMD_FLAGS(sc) |= FNIC_IO_ABT_TERM_TIMED_OUT;
                ret = FAILED;
                goto fnic_abort_cmd_end;
        }
 
+       CMD_STATE(sc) = FNIC_IOREQ_ABTS_COMPLETE;
+
        /*
         * firmware completed the abort, check the status,
         * free the io_req irrespective of failure or success
@@ -1753,16 +1765,17 @@ static int fnic_clean_pending_aborts(struct fnic *fnic,
        enum fnic_ioreq_state old_ioreq_state;
 
        for (tag = 0; tag < FNIC_MAX_IO_REQ; tag++) {
+               io_lock = fnic_io_lock_tag(fnic, tag);
+               spin_lock_irqsave(io_lock, flags);
                sc = scsi_host_find_tag(fnic->lport->host, tag);
                /*
                 * ignore this lun reset cmd or cmds that do not belong to
                 * this lun
                 */
-               if (!sc || sc == lr_sc || sc->device != lun_dev)
+               if (!sc || sc == lr_sc || sc->device != lun_dev) {
+                       spin_unlock_irqrestore(io_lock, flags);
                        continue;
-
-               io_lock = fnic_io_lock_hash(fnic, sc);
-               spin_lock_irqsave(io_lock, flags);
+               }
 
                io_req = (struct fnic_io_req *)CMD_SP(sc);
 
@@ -1791,6 +1804,11 @@ static int fnic_clean_pending_aborts(struct fnic *fnic,
                        spin_unlock_irqrestore(io_lock, flags);
                        continue;
                }
+
+               if (io_req->abts_done)
+                       shost_printk(KERN_ERR, fnic->lport->host,
+                         "%s: io_req->abts_done is set state is %s\n",
+                         __func__, fnic_ioreq_state_to_str(CMD_STATE(sc)));
                old_ioreq_state = CMD_STATE(sc);
                /*
                 * Any pending IO issued prior to reset is expected to be
@@ -1801,11 +1819,6 @@ static int fnic_clean_pending_aborts(struct fnic *fnic,
                 */
                CMD_STATE(sc) = FNIC_IOREQ_ABTS_PENDING;
 
-               if (io_req->abts_done)
-                       shost_printk(KERN_ERR, fnic->lport->host,
-                         "%s: io_req->abts_done is set state is %s\n",
-                         __func__, fnic_ioreq_state_to_str(CMD_STATE(sc)));
-
                BUG_ON(io_req->abts_done);
 
                abt_tag = tag;
@@ -1858,12 +1871,13 @@ static int fnic_clean_pending_aborts(struct fnic *fnic,
                io_req->abts_done = NULL;
 
                /* if abort is still pending with fw, fail */
-               if (CMD_STATE(sc) == FNIC_IOREQ_ABTS_PENDING) {
+               if (CMD_ABTS_STATUS(sc) == FCPIO_INVALID_CODE) {
                        spin_unlock_irqrestore(io_lock, flags);
                        CMD_FLAGS(sc) |= FNIC_IO_ABT_TERM_DONE;
                        ret = 1;
                        goto clean_pending_aborts_end;
                }
+               CMD_STATE(sc) = FNIC_IOREQ_ABTS_COMPLETE;
                CMD_SP(sc) = NULL;
                spin_unlock_irqrestore(io_lock, flags);
 
@@ -2061,8 +2075,8 @@ int fnic_device_reset(struct scsi_cmnd *sc)
                spin_unlock_irqrestore(io_lock, flags);
                int_to_scsilun(sc->device->lun, &fc_lun);
                /*
-                * Issue abort and terminate on the device reset request.
-                * If q'ing of the abort fails, retry issue it after a delay.
+                * Issue abort and terminate on device reset request.
+                * If q'ing of terminate fails, retry it after a delay.
                 */
                while (1) {
                        spin_lock_irqsave(io_lock, flags);
This page took 0.029433 seconds and 5 git commands to generate.