Re: OOPS: 2.6.24-rc5-mm1 -- EIP is at r_show+0x2a/0x70 -- (triggered by "cat /proc/iomem" AFTER suspend-to-disk/resume)

[Andrew Morton <akpm@linux-foundation.org>]

On Fri, 21 Dec 2007 11:06:49 -0500 "Miles Lane" <miles.lane@gmail.com> wrote:

> > >
> > > How do I determine what comes next?
> > >
> >
> > By comparing it with the /proc/iomem from prior to suspending the machine.
> >
> 
> The offending information seems to be "90000000-90000fff : Intel Flush Page"

ah-hah.  git-agpgart.patch

> I hope this helps,

It does, thanks.

Dave, a `cat /proc/iomem' crashes after a suspend/resume cycle when it
reaches the intel_private.ifp_resource record.


intel_i915_configure() is called on each resume and it calls
intel_i9xx_setup_flush() which zeroes out the already-registered `struct
resource', causing core kernel to oops over a deref of
parent/sibling/child.

Probably this:

--- a/drivers/char/agp/intel-agp.c~a
+++ a/drivers/char/agp/intel-agp.c
@@ -883,8 +883,6 @@ static void intel_i965_g33_setup_chipset
 static void intel_i9xx_setup_flush(void)
 {
 	/* setup a resource for this object */
-	memset(&intel_private.ifp_resource, 0, sizeof(intel_private.ifp_resource));
-
 	intel_private.ifp_resource.name = "Intel Flush Page";
 	intel_private.ifp_resource.flags = IORESOURCE_MEM;
 
_


will be enough to prevent the crash, but I suspect a bit more than that
will be needed.