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@xxxxxxxxxx>
Acked-by: Swen Schillig <swen@xxxxxxxxxxxx>
---
drivers/s390/scsi/zfcp_erp.c | 14 ++++++--------
1 files changed, 6 insertions(+), 8 deletions(-)
--- linux-2.6.24-rc2-git5.orig/drivers/s390/scsi/zfcp_erp.c
+++ linux-2.6.24-rc2-git5/drivers/s390/scsi/zfcp_erp.c
@@ -1065,7 +1065,7 @@ zfcp_erp_thread(void *data)
&adapter->status)) {
write_lock_irqsave(&adapter->erp_lock, flags);
- next = adapter->erp_ready_head.prev;
+ next = adapter->erp_ready_head.next;
write_unlock_irqrestore(&adapter->erp_lock, flags);
if (next != &adapter->erp_ready_head) {
@@ -1155,15 +1155,13 @@ zfcp_erp_strategy(struct zfcp_erp_action
/*
* check for dismissed status again to avoid follow-up actions,
- * failing of targets and so on for dismissed actions
+ * failing of targets and so on for dismissed actions,
+ * we go through down() here because there has been an up()
*/
- retval = zfcp_erp_strategy_check_action(erp_action, retval);
+ if (erp_action->status & ZFCP_STATUS_ERP_DISMISSED)
+ retval = ZFCP_ERP_CONTINUES;
switch (retval) {
- case ZFCP_ERP_DISMISSED:
- /* leave since this action has ridden to its ancestors */
- debug_text_event(adapter->erp_dbf, 6, "a_st_dis2");
- goto unlock;
case ZFCP_ERP_NOMEM:
/* no memory to continue immediately, let it sleep */
if (!(erp_action->status & ZFCP_STATUS_ERP_LOWMEM)) {
@@ -3091,7 +3089,7 @@ zfcp_erp_action_enqueue(int action,
++adapter->erp_total_count;
/* finally put it into 'ready' queue and kick erp thread */
- list_add(&erp_action->list, &adapter->erp_ready_head);
+ list_add_tail(&erp_action->list, &adapter->erp_ready_head);
up(&adapter->erp_ready_sem);
retval = 0;
out:
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
