Re: Bug#464962: immediate crash on boot on TM5800

Re: Bug#464962: immediate crash on boot on TM5800

[maximilian attems]

[-- Attachment #1: Type: text/plain, Size: 1296 bytes --]

On Tue, Feb 12, 2008 at 02:52:54PM -0500, Joey Hess wrote:
> maximilian attems wrote:
> > thanks for quick feedback, have prebuild 2.6.25-rc1-git2
> > (they contain the security fix, but don't seem to suspend here)
> > anyway please test for boot
> > -> http://charm.itp.tuwien.ac.at/~mattems/2.6.25-rc1-git2/
> 
> Still fails the same.
> 
> > if those again don't boot please file info upstream on
> > bugzilla.kernel.org so that relevant new x86 arch maintainer
> > fix that up.
> 
> Um, could the kernel team forward this bug?

well we have no automated way to do so
as we are underpowered in terms of bug nrs, it is
routine to ask to file upstream. the debian kernel
does not carry any x86 specific patch.
also it is often much easier if upstream speaks directly
with the person that has the hardware at hand.
 
> * I don't have an account on this bugzilla, or actually, on any bugzilla
>   on earth.
> * I don't know exactly what versions you've had me test.
> * I guess they'll want to know things about compile options that I don't
>   know.

sure, ack.
so i'll circumvent bugzilla and add the new x86 maintainers
on cc to let them know about the 2.6.24 and 2.6.25-rc1 boot error
on shiny fujitsu p700 lifebook, with a Crusoe processor.
http://bugs.debian.org/464962
686 config attached.


[-- Attachment #2: config-2.6.24-1-686.bz2 --]
[-- Type: application/octet-stream, Size: 20609 bytes --]

Re: Bug#464962: immediate crash on boot on TM5800

[H. Peter Anvin]

maximilian attems wrote:
> 
> sure, ack.
> so i'll circumvent bugzilla and add the new x86 maintainers
> on cc to let them know about the 2.6.24 and 2.6.25-rc1 boot error
> on shiny fujitsu p700 lifebook, with a Crusoe processor.
> http://bugs.debian.org/464962
> 686 config attached.
> 

INT 6 is #UD, undefined instruction.

If you could send me a copy of your vmlinux file (not bzImage), it would 
speed things up.

I happen to have an old TM5800-based machine sitting around, so I can 
probably reproduce it.

	-hpa

Re: Bug#464962: immediate crash on boot on TM5800

[maximilian attems]

On Tue, Feb 12, 2008 at 12:32:27PM -0800, H. Peter Anvin wrote:
> 
> INT 6 is #UD, undefined instruction.
> 
> If you could send me a copy of your vmlinux file (not bzImage), it would 
> speed things up.

cp -l src/linux-2.6-2.6.24/debian/build/build_i386_none_686/vmlinux ~/public_html/

http://charm.itp.tuwien.ac.at/~mattems/
 
> I happen to have an old TM5800-based machine sitting around, so I can 
> probably reproduce it.
> 
> 	-hpa

cool

Re: Bug#464962: immediate crash on boot on TM5800

[H. Peter Anvin]

maximilian attems wrote:
> On Tue, Feb 12, 2008 at 12:32:27PM -0800, H. Peter Anvin wrote:
>> INT 6 is #UD, undefined instruction.
>>
>> If you could send me a copy of your vmlinux file (not bzImage), it would 
>> speed things up.
> 
> cp -l src/linux-2.6-2.6.24/debian/build/build_i386_none_686/vmlinux ~/public_html/
> 
> http://charm.itp.tuwien.ac.at/~mattems/
>  

Are you sure that build matches the bug report?

The EIP given falls inside the .data segment of that kernel, 
specifically inside the symbol init_task.

	-hpa

Re: Bug#464962: immediate crash on boot on TM5800

[maximilian attems]

On Tue, Feb 12, 2008 at 01:14:04PM -0800, H. Peter Anvin wrote:
> 
> Are you sure that build matches the bug report?

urrgs right sorry, the posted vmlinux is a newer 
2.6.24-git22 and not  Version: 2.6.24-3
 
> The EIP given falls inside the .data segment of that kernel, 
> specifically inside the symbol init_task.
> 
> 	-hpa

will rebuild aboves.

Re: Bug#464962: immediate crash on boot on TM5800

[H. Peter Anvin]

maximilian attems wrote:
> On Tue, Feb 12, 2008 at 01:14:04PM -0800, H. Peter Anvin wrote:
>> Are you sure that build matches the bug report?
> 
> urrgs right sorry, the posted vmlinux is a newer 
> 2.6.24-git22 and not  Version: 2.6.24-3
>  
>> The EIP given falls inside the .data segment of that kernel, 
>> specifically inside the symbol init_task.
>>
>> 	-hpa
> 
> will rebuild aboves.

Don't worry about it, already have reproduced it.

	-hpa

Re: Bug#464962: immediate crash on boot on TM5800

[H. Peter Anvin]

maximilian attems wrote:
> On Tue, Feb 12, 2008 at 01:14:04PM -0800, H. Peter Anvin wrote:
>> Are you sure that build matches the bug report?
> 
> urrgs right sorry, the posted vmlinux is a newer 
> 2.6.24-git22 and not  Version: 2.6.24-3
>  
>> The EIP given falls inside the .data segment of that kernel, 
>> specifically inside the symbol init_task.
>>
>> 	-hpa
> 
> will rebuild aboves.

Okay, the faulting instruction is the following:

c0383360:       0f 1f 40 00             nopl   0x0(%eax)

The Crusoe code morphing software apparently doesn't recognize these 
"long noops", and (presumably) the rest of the hinting NOOP group.  gcc 
didn't use to generate them, and Crusoe/Efficeon generally do not 
benefit from code alignment anyway.  I suspect the best thing to do is 
to use either a 586 kernel or build a dedicated Crusoe kernel without 
code alignment.

	-hpa

Re: Bug#464962: immediate crash on boot on TM5800

[H. Peter Anvin]

[-- Attachment #1: Type: text/plain, Size: 389 bytes --]

Thought some more about this, and since this probably means gcc will 
generate this for userspace code as well nowadays, tm5800 should 
probably be downgraded to a 586-class machine.  Hence the Linux policy 
of promoting it to a 686-class machine for having CMOV is actually 
incorrect, it doesn't have all the userspace-visible features of a 
686-class machine, lacking long NOP.

	-hpa


[-- Attachment #2: diff --]
[-- Type: text/plain, Size: 760 bytes --]

diff --git a/arch/x86/kernel/cpu/transmeta.c b/arch/x86/kernel/cpu/transmeta.c
index 200fb3f..e8b422c 100644
--- a/arch/x86/kernel/cpu/transmeta.c
+++ b/arch/x86/kernel/cpu/transmeta.c
@@ -76,13 +76,6 @@ static void __cpuinit init_transmeta(struct cpuinfo_x86 *c)
 	/* All Transmeta CPUs have a constant TSC */
 	set_bit(X86_FEATURE_CONSTANT_TSC, c->x86_capability);
 	
-	/* If we can run i686 user-space code, call us an i686 */
-#define USER686 ((1 << X86_FEATURE_TSC)|\
-		 (1 << X86_FEATURE_CX8)|\
-		 (1 << X86_FEATURE_CMOV))
-        if (c->x86 == 5 && (c->x86_capability[0] & USER686) == USER686)
-		c->x86 = 6;
-
 #ifdef CONFIG_SYSCTL
 	/* randomize_va_space slows us down enormously;
 	   it probably triggers retranslation of x86->native bytecode */

Re: Bug#464962: immediate crash on boot on TM5800

[Joey Hess]

[-- Attachment #1: Type: text/plain, Size: 886 bytes --]

H. Peter Anvin wrote:
> maximilian attems wrote:
>> On Tue, Feb 12, 2008 at 12:32:27PM -0800, H. Peter Anvin wrote:
>>> INT 6 is #UD, undefined instruction.
>>>
>>> If you could send me a copy of your vmlinux file (not bzImage), it 
>>> would speed things up.
>>
>> cp -l src/linux-2.6-2.6.24/debian/build/build_i386_none_686/vmlinux ~/public_html/
>>
>> http://charm.itp.tuwien.ac.at/~mattems/
>>  
>
> Are you sure that build matches the bug report?
>
> The EIP given falls inside the .data segment of that kernel,  
> specifically inside the symbol init_task.

The stack trace I copied down originally was for a 2.6.24 kernel. I can
copy down the one .25-pre if that'd be useful, but it sounds like you're
reproduced it on your own.

(I'm raising the priority of this bug report since it sounds like it
affects more than just my hardware.)

-- 
see shy jo

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: Bug#464962: immediate crash on boot on TM5800

[Joey Hess]

[-- Attachment #1: Type: text/plain, Size: 175 bytes --]

maximilian attems wrote:
> on shiny fujitsu p700 lifebook, with a Crusoe processor.

FWIW, I misremembered the model number. It's a not-so-shiny P2110.

-- 
see shy jo

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]