Bug 5018

Summary:	can't resume from swsusp with amd64-agp module loaded
Product:	Power Management	Reporter:	Michal Schmidt (xschmi00)
Component:	Other	Assignee:	Dave Jones (davej)
Status:	CLOSED CODE_FIX
Severity:	normal	CC:	davej, pavel
Priority:	P2
Hardware:	i386
OS:	Linux
Kernel Version:	2.6.13-rc6	Subsystem:
Regression:	---	Bisected commit-id:
Attachments:	oops with 2.6.13-rc6

Description Michal Schmidt 2005-08-07 15:15:18 UTC

Most recent kernel where this bug did not occur: long-standing bug; probably
never worked
Distribution: unofficial Debian Sid for AMD64
Hardware Environment: Athlon 64 FX-53, Asus A8V Deluxe
Software Environment: gcc (GCC) 4.0.2 20050725 (prerelease) (Debian
4.0.1-3.0.0.1.gcc4)
Problem Description: When amd64-agp is loaded as a module and swsuspend is used,
a spontaneous reboot happens on resume. Resume works fine if amd64-agp is
built-in into the kernel.
Originally reported to LKML:
http://marc.theaimsgroup.com/?l=linux-kernel&m=112180650210831&w=2
A very similar problem was discovered and fixed for i386 in April 2004:
http://marc.theaimsgroup.com/?l=linux-kernel&m=108310336904458&w=2

Steps to reproduce:
Boot with "init=/bin/sh".
mount -t sysfs sys sys
swapon -a
modprobe amd64-agp
echo disk > /sys/power/state
Attempt to resume. Notice the sudden reboot after the saved image is
successfully loaded from disk.

Comment 1 Pavel Machek 2005-08-08 05:04:04 UTC

Can you try compiling amd64-agp into the kernel, as opposed to module? If that
helps, we know it is different problem than the one on i386.

Comment 2 Michal Schmidt 2005-08-08 07:13:07 UTC

Yes, compiling amd64-agp into the kernel (not as a module) helps. That's what I
meant with "Resume works fine if amd64-agp is built-in into the kernel." .

Comment 3 Pavel Machek 2005-08-08 15:22:07 UTC

Oops, sorry, did not notice that. It should be some problem with agp, not
generic like in i386 case, then. Try placing while(1) at begining and end of
amd64-agp resume function; if it hangs instead of rebooting, you can find out
where exactly problem lies. This should probably go to davej...

Comment 4 Michal Schmidt 2005-08-09 07:15:33 UTC

amd64-agp does not have resume function. But even if it had one, it would be of
no help, because the problem occurs right in swsusp_arch_resume in
arch/x86_64/kernel/suspend_asm.S during the copying of pages. It occurs between
the labels "loop" and "done".
It also might be of interest that the problem manifests itself either as a
spontaneous reboot or an oops - it depends on kernel version and exact .config.
I've tested 2.6.13-rc5-mm1 too, which reboots. At the moment I have a
configuration with 2.6.13-rc6, which oopses. I'll attach the oops transcription.

Comment 5 Michal Schmidt 2005-08-09 07:16:14 UTC

Created attachment 5561 [details]
oops with 2.6.13-rc6

Comment 6 Pavel Machek 2005-08-15 06:12:24 UTC

Sanity check: does it suspend/resume when amd64-agp is not loaded at all? Not
compiled in, not loaded.

Does it work on in 32-bit mode?

Okay, that driver is quite strange, it hooks early in the boot process, has
strange CONFIG_GART_IOMMU option and works around some hardware problems. What
about simply forcing it into kernel (i.e. disable module option for this module
in kernel config?)

Comment 7 Michal Schmidt 2005-08-15 16:55:10 UTC

Yes, suspend/resume works when amd64-agp is not loaded at all.

And yes, in 32-bit mode suspend/resume works even with amd64-agp loaded as a module.

Comment 8 Michal Schmidt 2005-12-26 05:54:00 UTC

Finally I got back home to my amd64 machine after 4 months.
I am testing 2.6.15-rc7 now and I'm pleased to report that the bug seems fixed.
I cannot reproduce the spontaneous reset anymore. Resuming now works.