[SCSI] zfcp: fix cleanup of dismissed error recovery actions
commit86e8dfc5603ed76917eed0a9dd9e85a1e1a8b162
authorMartin Peschke <mp3@de.ibm.com>
Thu, 15 Nov 2007 12:57:17 +0000 (15 13:57 +0100)
committerJames Bottomley <James.Bottomley@HansenPartnership.com>
Fri, 16 Nov 2007 19:03:21 +0000 (16 13:03 -0600)
tree3d5bf28741aacff3f60df35c0619c7d9cd352f44
parentd0076f7754dce07c7a1d752034561acadd99eafa
[SCSI] zfcp: fix cleanup of dismissed error recovery actions

Calling zfcp_erp_strategy_check_action() after zfcp_erp_action_to_running()
in zfcp_erp_strategy() might cause an unbalanced up() for erp_ready_sem,
which makes the zfcp recovery fail somewhere along the way:

erp thread processing erp_action:
|
| someone waking up erp thread for erp_action
| |
| | someone else dismissing erp_action:
| | |
V V V

write_lock_irqsave(&adapter->erp_lock, flags);
...
if (zfcp_erp_action_exists(erp_action) == ZFCP_ERP_ACTION_RUNNING) {
zfcp_erp_action_to_ready(erp_action);
up(&adapter->erp_ready_sem); /* first up() for erp_action */
}
write_unlock_irqrestore(&adapter->erp_lock, flags);

write_lock_irqsave(&adapter->erp_lock, flags);
...
zfcp_erp_action_to_running(erp_action);
write_unlock_restore(&adapter->erp_lock, flags);
/* processing erp_action */

write_lock_irqsave(&adapter->erp_lock, flags);
...
erp_action->status |= ZFCP_STATUS_ERP_DISMISSED;
if (zfcp_erp_action_exists(erp_action) ==
ZFCP_ERP_ACTION_RUNNING) {
zfcp_erp_action_to_ready(erp_action);
up(&adapter->erp_ready_sem);
/* second, unbalanced up() for erp_action */
}
...
write_unlock_restore(&adapter->erp_lock, flags);

write_lock_irqsave(&adapter->erp_lock, flags);
if (erp_action->status & ZFCP_STATUS_ERP_DISMISSED) {
zfcp_erp_action_dequeue(erp_action);
retval = ZFCP_ERP_DISMISSED;
}
...
write_unlock_restore(&adapter->erp_lock, flags);
down(&adapter->erp_ready_sem);
/* this down() is meant to balance the first up() */

The erp thread must not dismiss an erp_action after moving that action to
erp_running_head. Instead it should just go through the down() operation,
which balances the first up(), and run through zfcp_erp_strategy one more
time for the second up(), which eventually cleans up erp_action. Which
is similar to the normal processing of an event for erp_action doing
something asynchronously (e.g. waiting for the completion of an fsf_req).

This only works if we make sure that a dismissed erp_action is passed to
zfcp_erp_strategy() prior to the other action, which caused actions to be
dismissed. Therefore the patch implements this rule: running actions go to
the head of the ready list; new actions go to the tail of the ready list;
the erp thread picks actions to be processed from the ready list's head.

Signed-off-by: Martin Peschke <mp3@de.ibm.com>
Acked-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
drivers/s390/scsi/zfcp_erp.c