cxlflash: Fix to resolve cmd leak after host reset
authorManoj Kumar <manoj@linux.vnet.ibm.com>
Mon, 14 Dec 2015 21:07:02 +0000 (15:07 -0600)
committerMartin K. Petersen <martin.petersen@oracle.com>
Thu, 7 Jan 2016 01:52:31 +0000 (20:52 -0500)
After a few iterations of resetting the card, either during EEH
recovery, or a host_reset the following is seen in the logs.  cxlflash
0008:00: cxlflash_queuecommand: could not get a free command

At every reset of the card, the commands that are outstanding are being
leaked.  No effort is being made to reap these commands.  A few more
resets later, the above error message floods the logs and the card is
rendered totally unusable as no free commands are available.

Iterated through the 'cmd' queue and printed out the 'free' counter and
found that on each reset certain commands were in-use and stayed in-use
through subsequent resets.

To resolve this issue, when the card is reset, reap all the commands
that are active/outstanding.

Signed-off-by: Manoj N. Kumar <manoj@linux.vnet.ibm.com>
Acked-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/cxlflash/main.c

index 35a32024f1c0d3c4c23bcf35baa86a6be7b21d10..ac39856a74b49f02d9b0fa57d8628a93871a8191 100644 (file)
@@ -632,15 +632,30 @@ static void free_mem(struct cxlflash_cfg *cfg)
  * @cfg:       Internal structure associated with the host.
  *
  * Safe to call with AFU in a partially allocated/initialized state.
+ *
+ * Cleans up all state associated with the command queue, and unmaps
+ * the MMIO space.
+ *
+ *  - complete() will take care of commands we initiated (they'll be checked
+ *  in as part of the cleanup that occurs after the completion)
+ *
+ *  - cmd_checkin() will take care of entries that we did not initiate and that
+ *  have not (and will not) complete because they are sitting on a [now stale]
+ *  hardware queue
  */
 static void stop_afu(struct cxlflash_cfg *cfg)
 {
        int i;
        struct afu *afu = cfg->afu;
+       struct afu_cmd *cmd;
 
        if (likely(afu)) {
-               for (i = 0; i < CXLFLASH_NUM_CMDS; i++)
-                       complete(&afu->cmd[i].cevent);
+               for (i = 0; i < CXLFLASH_NUM_CMDS; i++) {
+                       cmd = &afu->cmd[i];
+                       complete(&cmd->cevent);
+                       if (!atomic_read(&cmd->free))
+                               cmd_checkin(cmd);
+               }
 
                if (likely(afu->afu_map)) {
                        cxl_psa_unmap((void __iomem *)afu->afu_map);