2002-10-02 10:11:02

by Russell King

Subject: Dereferencing semaphores and atomic_t's

drivers/scsi/scsi_error.c:

SCSI_LOG_ERROR_RECOVERY(3, printk("Wake up parent %d\n",
shost->eh_notify->count.counter));

up(shost->eh_notify);

drivers/scsi/hosts.c:

up(shost->eh_wait);
SCSI_LOG_ERROR_RECOVERY(5, printk("Waking error handler"
"thread (%d)\n",
atomic_read(&shost->eh_wait->count)));


up(shost->eh_wait);
SCSI_LOG_ERROR_RECOVERY(5, printk("Waking error handler"
"thread (%d)\n",
atomic_read(&shost->eh_wait->count)));

Do we really allow this type of layering violation?

(There appear to be some circumstances when obtaining the semaphore count is
useful, but shouldn't we provide an architecture helper function to do that
since a semaphore structure _is_ an architecture-defined opaque structure.)

--
Russell King ([email protected]) The developer of ARM Linux
http://www.arm.linux.org.uk/personal/aboutme.html

2002-10-02 10:34:47

by David Miller

Subject: Re: Dereferencing semaphores and atomic_t's

From: Russell King <[email protected]>
Date: Wed, 2 Oct 2002 11:16:26 +0100

Do we really allow this type of layering violation?

No, it should be killed :-)

(There appear to be some circumstances when obtaining the semaphore count is
useful, but shouldn't we provide an architecture helper function to do that
since a semaphore structure _is_ an architecture-defined opaque structure.)

Feel free to make one.

2002-10-02 13:08:43

by Alan

Subject: Re: Dereferencing semaphores and atomic_t's

On Wed, 2002-10-02 at 11:16, Russell King wrote:
> drivers/scsi/scsi_error.c:
>
> SCSI_LOG_ERROR_RECOVERY(3, printk("Wake up parent %d\n",
> shost->eh_notify->count.counter));
>

This is already fixed in 2.4 - just forward port the fixes

2002-10-02 16:41:41

by Mike Anderson

Subject: Re: Dereferencing semaphores and atomic_t's

Alan Cox [[email protected]] wrote:
> On Wed, 2002-10-02 at 11:16, Russell King wrote:
> > drivers/scsi/scsi_error.c:
> >
> > SCSI_LOG_ERROR_RECOVERY(3, printk("Wake up parent %d\n",
> > shost->eh_notify->count.counter));
> >
>
> This is already fixed in 2.4 - just forward port the fixes

I added sem_getcount to my scsi_error.c update and my scsi_hosts update
patch. I should have these out soon.

Sorry Russell that it has taken longer than I said to get the updated
scsi_error patch out. I am still investigating an issue with getting
door lock to work post fault insertion.

-andmike
--
Michael Anderson
[email protected]