qla1280 bandaid

qla1280 bandaid

[Peter Chubb]

Hi Jes,
   The problem is that bus_reset drops the HOST_LOCK when it calls
schedule_timeout, but when it's called from the initialisation code,
HOST_LOCK isn't actually held.

The appended patch fixes this; however, I'm not sure that it's
correct, because I couldn't work out easily whether the HOST_LOCK is
held or not on all paths to qla1280_bus_reset().

With the patch in place, I can once again boot SMP I2000 and rx4610.

=== drivers/scsi/qla1280.c 1.44 vs edited ==--- 1.44/drivers/scsi/qla1280.c	Fri Aug 15 09:08:54 2003
+++ edited/drivers/scsi/qla1280.c	Tue Sep  2 14:53:32 2003
@@ -3346,9 +3346,9 @@
 			ha->bus_settings[bus].scsi_bus_dead = 1;
 		ha->bus_settings[bus].failed_reset_count++;
 	} else {
-		spin_unlock_irq(HOST_LOCK);
+/*		spin_unlock_irq(HOST_LOCK);*/
 		schedule_timeout(reset_delay * HZ);
-		spin_lock_irq(HOST_LOCK);
+/*		spin_lock_irq(HOST_LOCK);*/
 
 		ha->bus_settings[bus].scsi_bus_dead = 0;
 		ha->bus_settings[bus].failed_reset_count = 0;

Re: qla1280 bandaid

[Christoph Hellwig]

On Tue, Sep 02, 2003 at 02:57:15PM +1000, Peter Chubb wrote:
> 
> Hi Jes,
>    The problem is that bus_reset drops the HOST_LOCK when it calls
> schedule_timeout, but when it's called from the initialisation code,
> HOST_LOCK isn't actually held.
> 
> The appended patch fixes this; however, I'm not sure that it's
> correct, because I couldn't work out easily whether the HOST_LOCK is
> held or not on all paths to qla1280_bus_reset().

Right fix would probably be to take the lock always before calling..

Re: qla1280 bandaid

[Jes Sorensen]

>>>>> "Christoph" = Christoph Hellwig <hch@infradead.org> writes:

Christoph> On Tue, Sep 02, 2003 at 02:57:15PM +1000, Peter Chubb
Christoph> wrote:
>>  Hi Jes, The problem is that bus_reset drops the HOST_LOCK when it
>> calls schedule_timeout, but when it's called from the
>> initialisation code, HOST_LOCK isn't actually held.
>> 
>> The appended patch fixes this; however, I'm not sure that it's
>> correct, because I couldn't work out easily whether the HOST_LOCK
>> is held or not on all paths to qla1280_bus_reset().

Christoph> Right fix would probably be to take the lock always before
Christoph> calling..

Peter and Christoph,

Thanks for the info, I'll take a look at integrating this into the
next version.

Cheers,
Jes