2010-04-01 15:54:17

by Dan Carpenter

Subject: [patch] cciss: unlock on error path

We take the spin_lock again in fail_all_cmds() so we need to unlock
here.

Signed-off-by: Dan Carpenter <[email protected]>
---
This was found by a static checker and I don't have the hardware myself.
Please review carefully.

diff --git a/drivers/block/cciss.c b/drivers/block/cciss.c
index 9e3af30..eb5ff05 100644
--- a/drivers/block/cciss.c
+++ b/drivers/block/cciss.c
@@ -3341,6 +3341,7 @@ static irqreturn_t do_cciss_intr(int irq, void *dev_id)
printk(KERN_WARNING
"cciss: controller cciss%d failed, stopping.\n",
h->ctlr);
+ spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
fail_all_cmds(h->ctlr);
return IRQ_HANDLED;
}

2010-04-01 16:30:18

by Stephen M. Cameron

Subject: Re: [patch] cciss: unlock on error path

On Thu, Apr 01, 2010 at 06:54:04PM +0300, Dan Carpenter wrote:
> We take the spin_lock again in fail_all_cmds() so we need to unlock
> here.
>
> Signed-off-by: Dan Carpenter <[email protected]>
> ---
> This was found by a static checker and I don't have the hardware myself.
> Please review carefully.
>
> diff --git a/drivers/block/cciss.c b/drivers/block/cciss.c
> index 9e3af30..eb5ff05 100644
> --- a/drivers/block/cciss.c
> +++ b/drivers/block/cciss.c
> @@ -3341,6 +3341,7 @@ static irqreturn_t do_cciss_intr(int irq, void *dev_id)
> printk(KERN_WARNING
> "cciss: controller cciss%d failed, stopping.\n",
> h->ctlr);
> + spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
> fail_all_cmds(h->ctlr);
> return IRQ_HANDLED;
> }

Looks ok to me. (Seems like I've seen this patch before).
It's pretty much a paranoid "never happen" case anyway.
To hit this, the hardware would have to return a tag that was marked as
containing an index, and then the index would have to be out of bounds,
so either, it would have to be broken driver, broken firmware or
broken hardware.

Really, one of these days, the interrupt handler should probably be
cleaned up to be more like what's in the hpsa driver.

-- steve

2010-04-02 06:40:06

by Jens Axboe

Subject: Re: [patch] cciss: unlock on error path

On Thu, Apr 01 2010, Dan Carpenter wrote:
> We take the spin_lock again in fail_all_cmds() so we need to unlock
> here.

Thanks Dan, applied.

--
Jens Axboe