iv/linux
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[SCSI] hpsa: prevent stalled i/o

Browse Source 
If a fifo full condition is encountered, i/o requests will stack
up in the h->reqQ queue.  The only thing which empties this queue
is start_io, which only gets called when new i/o requests come in.
If none are forthcoming, i/o in h->reqQ will be stalled.

To fix this, whenever fifo full condition is encountered, this
is recorded, and the interrupt handler examines this to see
if a fifo full condition was recently encountered when a
command completes and will call start_io to prevent i/o's in
h->reqQ from getting stuck.

I've only ever seen this problem occur when running specialized
test programs that pound on the the CCISS_PASSTHRU ioctl.

Signed-off-by: Stephen M. Cameron <scameron@beardog.cce.hp.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
This commit is contained in:
Stephen M. Cameron 2013-09-23 13:34:17 -05:00 committed by James Bottomley
parent 0390f0c0df
commit 396883e292
2 changed files with 31 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
drivers/scsi/hpsa.c
View File

@ -3483,9 +3483,11 @@ static void start_io(struct ctlr_info *h)
c = list_entry(h->reqQ.next, struct CommandList, list);
/* can't do anything if fifo is full */
if ((h->access.fifo_full(h))) {
h->fifo_recently_full = 1;
dev_warn(&h->pdev->dev, "fifo full\n");
break;
}
h->fifo_recently_full = 0;
/* Get the first entry from the Request Q */
removeQ(c);
@ -3539,15 +3541,41 @@ static inline int bad_tag(struct ctlr_info *h, u32 tag_index,
static inline void finish_cmd(struct CommandList *c)
{
unsigned long flags;
int io_may_be_stalled = 0;
struct ctlr_info *h = c->h;
spin_lock_irqsave(&c->h->lock, flags);
spin_lock_irqsave(&h->lock, flags);
removeQ(c);
spin_unlock_irqrestore(&c->h->lock, flags);
/*
* Check for possibly stalled i/o.
*
* If a fifo_full condition is encountered, requests will back up
* in h->reqQ. This queue is only emptied out by start_io which is
* only called when a new i/o request comes in. If no i/o's are
* forthcoming, the i/o's in h->reqQ can get stuck. So we call
* start_io from here if we detect such a danger.
*
* Normally, we shouldn't hit this case, but pounding on the
* CCISS_PASSTHRU ioctl can provoke it. Only call start_io if
* commands_outstanding is low. We want to avoid calling
* start_io from in here as much as possible, and esp. don't
* want to get in a cycle where we call start_io every time
* through here.
*/
if (unlikely(h->fifo_recently_full) &&
h->commands_outstanding < 5)
io_may_be_stalled = 1;
spin_unlock_irqrestore(&h->lock, flags);
dial_up_lockup_detection_on_fw_flash_complete(c->h, c);
if (likely(c->cmd_type == CMD_SCSI))
complete_scsi_command(c);
else if (c->cmd_type == CMD_IOCTL_PEND)
complete(c->waiting);
if (unlikely(io_may_be_stalled))
start_io(h);
}
static inline u32 hpsa_tag_contains_index(u32 tag)

1
drivers/scsi/hpsa.h
View File

@ -136,6 +136,7 @@ struct ctlr_info {
atomic_t firmware_flash_in_progress;
u32 lockup_detected;
struct list_head lockup_list;
u32 fifo_recently_full;
/* Address of h->q[x] is passed to intr handler to know which queue */
u8 q[MAX_REPLY_QUEUES];
u32 TMFSupportFlags; /* cache what task mgmt funcs are supported. */