Le 06.06.2006 09:26, Ingo Molnar a écrit :
> * Barry K. Nathan <barryn@pobox.com> wrote:
> 
>> On 6/5/06, Andrew Morton <akpm@osdl.org> wrote:
>>> I guess we should force 8k stacks if the lockdep features are enabled.
>> Also, Laurent is running "2.6.17-rc5-mm3-lockdep" (per his previous
>> message), i.e., 2.6.17-rc5-mm3 with Ingo's lockdep-combo patch added.
>> If you're wondering how I conclude the latter from the former, look at
>> this hunk from the lockdep-combo patch:
>>
>> --- linux.orig/Makefile
>> +++ linux/Makefile
>> @@ -1,7 +1,7 @@
>> VERSION = 2
>> PATCHLEVEL = 6
>> SUBLEVEL = 17
>> -EXTRAVERSION =-rc5-mm3
>> +EXTRAVERSION =-rc5-mm3-lockdep
>> NAME=Lordi Rules
>>
>> And from the same patch:
>>
>> --- linux.orig/arch/i386/Makefile
>> +++ linux/arch/i386/Makefile
>> @@ -38,6 +38,10 @@ CFLAGS += $(call cc-option,-mpreferred-s
>> include $(srctree)/arch/i386/Makefile.cpu
>>
>> cflags-$(CONFIG_REGPARM) += -mregparm=3
>> +#
>> +# Prevent tail-call optimizations, to get clearer backtraces:
>> +#
>> +cflags-$(CONFIG_FRAME_POINTER) += -fno-optimize-sibling-calls
>>
>> # temporary until string.h is fixed
>> cflags-y += -ffreestanding
>>
>> I would expect that to increase stack usage...
> 
> yes, and -pg (the one that creates the mcount callbacks that drive the 
> lockdep tracer) increases the stack footprint too. But these items are 
> only in my combo patch and in the lockdep tracer, not in -mm.
> 
> 	Ingo

Ok, I didn't clearly write when I switched from 2.6.17-rc5-mm3 to 
2.6.17-rc5-mm3-lockdep, sorry for that.

In a few words, here is the situation:

Scenario:
- boot to runlevel 2
- modprobe pktcdvd
- pktsetup dvd /dev/dvd

Results:
- 2.6.17-rc4-mm3 with 4K stack works fine (this is the latest good 4K-kernel).
- 2.6.17-rc5-mm3 with 4K stack crashes, the stack seems to be corrupted.
- 2.6.17-rc5-mm3 with 8K stack works fine.
- 2.6.17-rc5-mm3-lockdep with 8k stack works fine, although it exhibits high 
stack usage while running pktsetup.

Attached my .config.

-- 
laurent