ACPI Regression on Dell E1501

ACPI Regression on Dell E1501

[Tim Gardner]

Hi,

Commit e9e2cdb412412326c4827fc78ba27f410d837e6e breaks boot on a Dell
E1501 unless 'acpi=off' is specified (also tried nolapic and nohpet but
it made no substantive difference). This laptop is an 'AMD Turion(tm) 64
X2 Mobile Technology TL-50' CPU, but its booting 32 bit SMP (make
defconfig). No kernel built after this commit works for this platform.

There is no crash output, it just wedges after 'NET: Registered protocol
family 2'.

How can I debug this problem?

rtg
-- 
Tim Gardner tim.gardner@ubuntu.com

Re: ACPI Regression on Dell E1501

[Chuck Ebbert]

On 06/21/2007 05:04 PM, Tim Gardner wrote:
> Hi,
> 
> Commit e9e2cdb412412326c4827fc78ba27f410d837e6e breaks boot on a Dell
> E1501 unless 'acpi=off' is specified (also tried nolapic and nohpet but
> it made no substantive difference). This laptop is an 'AMD Turion(tm) 64
> X2 Mobile Technology TL-50' CPU, but its booting 32 bit SMP (make
> defconfig). No kernel built after this commit works for this platform.
> 
> There is no crash output, it just wedges after 'NET: Registered protocol
> family 2'.
> 
> How can I debug this problem?

That'll be fun.

That's:

	[PATCH] clockevents: i386 drivers

[Author cc: added]

Re: ACPI Regression on Dell E1501

[Thomas Gleixner]

On Thu, 2007-06-21 at 17:47 -0400, Chuck Ebbert wrote:
> On 06/21/2007 05:04 PM, Tim Gardner wrote:
> > Hi,
> > 
> > Commit e9e2cdb412412326c4827fc78ba27f410d837e6e breaks boot on a Dell
> > E1501 unless 'acpi=off' is specified (also tried nolapic and nohpet but
> > it made no substantive difference). This laptop is an 'AMD Turion(tm) 64
> > X2 Mobile Technology TL-50' CPU, but its booting 32 bit SMP (make
> > defconfig). No kernel built after this commit works for this platform.
> > 
> > There is no crash output, it just wedges after 'NET: Registered protocol
> > family 2'.
> > 
> > How can I debug this problem?
> 
> That'll be fun.
> 
> That's:
> 
> 	[PATCH] clockevents: i386 drivers
> 
> [Author cc: added]

I'm going nuts on this. 

Tim, 

1.) are you booting a 32bit or a 64 bit kernel ?

2.) Can you please try 2.6.22-rc5 + 
http://www.tglx.de/projects/hrtimers/2.6.22-rc5/patch-2.6.22-rc5-hrt1.patch
	whether this makes any difference 

Thanks,

	tglx

Re: ACPI Regression on Dell E1501

[Thomas Gleixner]

On Thu, 2007-06-21 at 23:55 +0200, Thomas Gleixner wrote:
> > > X2 Mobile Technology TL-50' CPU, but its booting 32 bit SMP (make

> 1.) are you booting a 32bit or a 64 bit kernel ?

Sigh, I'm too tired :)

	tglx

Re: ACPI Regression on Dell E1501

[Tim Gardner]

Thomas Gleixner wrote:
> On Thu, 2007-06-21 at 17:47 -0400, Chuck Ebbert wrote:
>> On 06/21/2007 05:04 PM, Tim Gardner wrote:
>>> Hi,
>>>
>>> Commit e9e2cdb412412326c4827fc78ba27f410d837e6e breaks boot on a Dell
>>> E1501 unless 'acpi=off' is specified (also tried nolapic and nohpet but
>>> it made no substantive difference). This laptop is an 'AMD Turion(tm) 64
>>> X2 Mobile Technology TL-50' CPU, but its booting 32 bit SMP (make
>>> defconfig). No kernel built after this commit works for this platform.
>>>
>>> There is no crash output, it just wedges after 'NET: Registered protocol
>>> family 2'.
>>>
>>> How can I debug this problem?
>> That'll be fun.
>>
>> That's:
>>
>> 	[PATCH] clockevents: i386 drivers
>>
>> [Author cc: added]
> 
> I'm going nuts on this. 
> 
> Tim, 
> 
> 1.) are you booting a 32bit or a 64 bit kernel ?
> 
> 2.) Can you please try 2.6.22-rc5 + 
> http://www.tglx.de/projects/hrtimers/2.6.22-rc5/patch-2.6.22-rc5-hrt1.patch
> 	whether this makes any difference 
> 
> Thanks,
> 
> 	tglx
> 
Thomas,

Its CONFIG_X86_32=y and CONFIG_SMP=y (make defconfig). I'll try your
patch later this evening.

rtg
-- 
Tim Gardner tim.gardner@ubuntu.com

Re: ACPI Regression on Dell E1501

[Tim Gardner]

See this topic for a solution:

http://lkml.org/lkml/2007/8/8/251
Re: [PATCH] drop unneeded variable in amd_apic_timer_broken
-- 
Tim Gardner tim.gardner@ubuntu.com

Re: ACPI Regression on Dell E1501

[Tim Gardner]

Thomas Gleixner wrote:
> On Thu, 2007-06-21 at 17:47 -0400, Chuck Ebbert wrote:
>> On 06/21/2007 05:04 PM, Tim Gardner wrote:
>>> Hi,
>>>
>>> Commit e9e2cdb412412326c4827fc78ba27f410d837e6e breaks boot on a Dell
>>> E1501 unless 'acpi=off' is specified (also tried nolapic and nohpet but
>>> it made no substantive difference). This laptop is an 'AMD Turion(tm) 64
>>> X2 Mobile Technology TL-50' CPU, but its booting 32 bit SMP (make
>>> defconfig). No kernel built after this commit works for this platform.
>>>
>>> There is no crash output, it just wedges after 'NET: Registered protocol
>>> family 2'.
>>>
>>> How can I debug this problem?
>> That'll be fun.
>>
>> That's:
>>
>> 	[PATCH] clockevents: i386 drivers
>>
>> [Author cc: added]
> 
> I'm going nuts on this. 
> 
> Tim, 
> 
> 1.) are you booting a 32bit or a 64 bit kernel ?
> 
> 2.) Can you please try 2.6.22-rc5 + 
> http://www.tglx.de/projects/hrtimers/2.6.22-rc5/patch-2.6.22-rc5-hrt1.patch
> 	whether this makes any difference 
> 
> Thanks,
> 
> 	tglx
> 

Thomas,

Your patch (large though it is) did not make any outward difference. As
usual, adding 'acpi=off' allows it to boot.

rtg
-- 
Tim Gardner tim.gardner@ubuntu.com

Re: ACPI Regression on Dell E1501

[Tim Gardner]

Tim Gardner wrote:
> Thomas Gleixner wrote:
>> On Thu, 2007-06-21 at 17:47 -0400, Chuck Ebbert wrote:
>>> On 06/21/2007 05:04 PM, Tim Gardner wrote:
>>>> Hi,
>>>>
>>>> Commit e9e2cdb412412326c4827fc78ba27f410d837e6e breaks boot on a Dell
>>>> E1501 unless 'acpi=off' is specified (also tried nolapic and nohpet but
>>>> it made no substantive difference). This laptop is an 'AMD Turion(tm) 64
>>>> X2 Mobile Technology TL-50' CPU, but its booting 32 bit SMP (make
>>>> defconfig). No kernel built after this commit works for this platform.
>>>>
>>>> There is no crash output, it just wedges after 'NET: Registered protocol
>>>> family 2'.
>>>>
>>>> How can I debug this problem?
>>> That'll be fun.
>>>
>>> That's:
>>>
>>> 	[PATCH] clockevents: i386 drivers
>>>
>>> [Author cc: added]
>> I'm going nuts on this. 
>>
>> Tim, 
>>
>> 1.) are you booting a 32bit or a 64 bit kernel ?
>>
>> 2.) Can you please try 2.6.22-rc5 + 
>> http://www.tglx.de/projects/hrtimers/2.6.22-rc5/patch-2.6.22-rc5-hrt1.patch
>> 	whether this makes any difference 
>>
>> Thanks,
>>
>> 	tglx
>>
> 
> Thomas,
> 
> Your patch (large though it is) did not make any outward difference. As
> usual, adding 'acpi=off' allows it to boot.
> 
> rtg

Perhaps I was too hasty. Adding 'acpi=off' did not allow it to boot. The
IDE drive was detected, but I did not see any partitions detected.
Eventually it dropped to the initrd shell when it could not find a root
filesystem. Anyways, this is likely a side issue.

rtg
-- 
Tim Gardner tim.gardner@ubuntu.com

Re: ACPI Regression on Dell E1501

[Thomas Renninger]

On Fri, 2007-06-22 at 07:18 -0600, Tim Gardner wrote:
> Tim Gardner wrote:
> > Thomas Gleixner wrote:
> >> On Thu, 2007-06-21 at 17:47 -0400, Chuck Ebbert wrote:
> >>> On 06/21/2007 05:04 PM, Tim Gardner wrote:
> >>>> Hi,
> >>>>
> >>>> Commit e9e2cdb412412326c4827fc78ba27f410d837e6e breaks boot on a Dell
> >>>> E1501 unless 'acpi=off' is specified (also tried nolapic and nohpet but
> >>>> it made no substantive difference). This laptop is an 'AMD Turion(tm) 64
> >>>> X2 Mobile Technology TL-50' CPU, but its booting 32 bit SMP (make
> >>>> defconfig). No kernel built after this commit works for this platform.
> >>>>
> >>>> There is no crash output, it just wedges after 'NET: Registered protocol
> >>>> family 2'.
> >>>>
> >>>> How can I debug this problem?
> >>> That'll be fun.
> >>>
> >>> That's:
> >>>
> >>> 	[PATCH] clockevents: i386 drivers
> >>>
> >>> [Author cc: added]
> >> I'm going nuts on this. 
> >>
> >> Tim, 
> >>
> >> 1.) are you booting a 32bit or a 64 bit kernel ?
> >>
> >> 2.) Can you please try 2.6.22-rc5 + 
> >> http://www.tglx.de/projects/hrtimers/2.6.22-rc5/patch-2.6.22-rc5-hrt1.patch
> >> 	whether this makes any difference 
> >>
> >> Thanks,
> >>
> >> 	tglx
> >>
> > 
> > Thomas,
> > 
> > Your patch (large though it is) did not make any outward difference. As
> > usual, adding 'acpi=off' allows it to boot.
> > 
> > rtg
> 
> Perhaps I was too hasty. Adding 'acpi=off' did not allow it to boot. The
> IDE drive was detected, but I did not see any partitions detected.
> Eventually it dropped to the initrd shell when it could not find a root
> filesystem. Anyways, this is likely a side issue.

I am seeing the same with an ACER F5000.

pci=nomsi let the machine still boot.

I tried a bit and I think I got it, for me it's this one:

commit 78b7611c4a1e3ff008abc4751b566cb533d68f3d
Author: Eric W. Biederman <ebiederm@xmission.com>
Date:   Fri Jun 1 00:46:33 2007 -0700

My problem: The laptop does not have an serial output device, I try to
set up a fire wire console, but might have cable/HW problems.

What else can I do to help?
What do you need?

Thanks,

   Thomas

Re: ACPI Regression on Dell E1501

[Roland Dreier]

 > pci=nomsi let the machine still boot.
 > 
 > I tried a bit and I think I got it, for me it's this one:
 > 
 > commit 78b7611c4a1e3ff008abc4751b566cb533d68f3d

Really??  That's "msi: mask the msix vector before we unmap it" and
A) it looks *really* obviously correct and B) it should only have an
effect if a driver using MSI-X releases an IRQ.

What testing did you do to pick out that change?  Does your system
work OK with just that patch reverted?

 - R.

Re: ACPI Regression on Dell E1501

[Thomas Renninger]

On Fri, 2007-06-22 at 09:06 -0700, Roland Dreier wrote:
>  > pci=nomsi let the machine still boot.
>  > 
>  > I tried a bit and I think I got it, for me it's this one:
>  > 
>  > commit 78b7611c4a1e3ff008abc4751b566cb533d68f3d
> 
> Really??  That's "msi: mask the msix vector before we unmap it" and
No, I was wrong.
I added the patches again, which I reverted and it still boots.
I horribly mixed something up here, I will try to find out tomorrow,
sorry for the confusion.

   Thomas

Re: ACPI Regression on Dell E1501

[Tim Gardner]

Thomas Renninger wrote:
> On Fri, 2007-06-22 at 07:18 -0600, Tim Gardner wrote:
>> Tim Gardner wrote:
>>> Thomas Gleixner wrote:
>>>> On Thu, 2007-06-21 at 17:47 -0400, Chuck Ebbert wrote:
>>>>> On 06/21/2007 05:04 PM, Tim Gardner wrote:
>>>>>> Hi,
>>>>>>
>>>>>> Commit e9e2cdb412412326c4827fc78ba27f410d837e6e breaks boot on a Dell
>>>>>> E1501 unless 'acpi=off' is specified (also tried nolapic and nohpet but
>>>>>> it made no substantive difference). This laptop is an 'AMD Turion(tm) 64
>>>>>> X2 Mobile Technology TL-50' CPU, but its booting 32 bit SMP (make
>>>>>> defconfig). No kernel built after this commit works for this platform.
>>>>>>
>>>>>> There is no crash output, it just wedges after 'NET: Registered protocol
>>>>>> family 2'.
>>>>>>
>>>>>> How can I debug this problem?
>>>>> That'll be fun.
>>>>>
>>>>> That's:
>>>>>
>>>>> 	[PATCH] clockevents: i386 drivers
>>>>>
>>>>> [Author cc: added]
>>>> I'm going nuts on this. 
>>>>
>>>> Tim, 
>>>>
>>>> 1.) are you booting a 32bit or a 64 bit kernel ?
>>>>
>>>> 2.) Can you please try 2.6.22-rc5 + 
>>>> http://www.tglx.de/projects/hrtimers/2.6.22-rc5/patch-2.6.22-rc5-hrt1.patch
>>>> 	whether this makes any difference 
>>>>
>>>> Thanks,
>>>>
>>>> 	tglx
>>>>
>>> Thomas,
>>>
>>> Your patch (large though it is) did not make any outward difference. As
>>> usual, adding 'acpi=off' allows it to boot.
>>>
>>> rtg
>> Perhaps I was too hasty. Adding 'acpi=off' did not allow it to boot. The
>> IDE drive was detected, but I did not see any partitions detected.
>> Eventually it dropped to the initrd shell when it could not find a root
>> filesystem. Anyways, this is likely a side issue.
> 
> I am seeing the same with an ACER F5000.
> 
> pci=nomsi let the machine still boot.
> 
> I tried a bit and I think I got it, for me it's this one:
> 
> commit 78b7611c4a1e3ff008abc4751b566cb533d68f3d
> Author: Eric W. Biederman <ebiederm@xmission.com>
> Date:   Fri Jun 1 00:46:33 2007 -0700
> 
> My problem: The laptop does not have an serial output device, I try to
> set up a fire wire console, but might have cable/HW problems.
> 
> What else can I do to help?
> What do you need?
> 
> Thanks,
> 
>    Thomas
> 

pci=nomsi has no effect, nor does CONFIG_SMP. Furthermore, I positively
confirmed that this is the offending commit. If I reset HEAD to the
commit just prior, then everything works OK.

rtg
-- 
Tim Gardner tim.gardner@ubuntu.com

Re: ACPI Regression on Dell E1501

[Eric W. Biederman]

Tim Gardner <tim.gardner@canonical.com> writes:

> Thomas Renninger wrote:
>> On Fri, 2007-06-22 at 07:18 -0600, Tim Gardner wrote:
>>> Tim Gardner wrote:
>>>> Thomas Gleixner wrote:
>>>>> On Thu, 2007-06-21 at 17:47 -0400, Chuck Ebbert wrote:
>>>>>> On 06/21/2007 05:04 PM, Tim Gardner wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>> Commit e9e2cdb412412326c4827fc78ba27f410d837e6e breaks boot on a Dell
>>>>>>> E1501 unless 'acpi=off' is specified (also tried nolapic and nohpet but
>>>>>>> it made no substantive difference). This laptop is an 'AMD Turion(tm) 64
>>>>>>> X2 Mobile Technology TL-50' CPU, but its booting 32 bit SMP (make
>>>>>>> defconfig). No kernel built after this commit works for this platform.
>>>>>>>
>>>>>>> There is no crash output, it just wedges after 'NET: Registered protocol
>>>>>>> family 2'.
>>>>>>>
>>>>>>> How can I debug this problem?
>>>>>> That'll be fun.
>>>>>>
>>>>>> That's:
>>>>>>
>>>>>> 	[PATCH] clockevents: i386 drivers
>>>>>>
>>>>>> [Author cc: added]
>>>>> I'm going nuts on this. 
>>>>>
>>>>> Tim, 
>>>>>
>>>>> 1.) are you booting a 32bit or a 64 bit kernel ?
>>>>>
>>>>> 2.) Can you please try 2.6.22-rc5 + 
>>>>> http://www.tglx.de/projects/hrtimers/2.6.22-rc5/patch-2.6.22-rc5-hrt1.patch
>>>>> 	whether this makes any difference 
>>>>>
>>>>> Thanks,
>>>>>
>>>>> 	tglx
>>>>>
>>>> Thomas,
>>>>
>>>> Your patch (large though it is) did not make any outward difference. As
>>>> usual, adding 'acpi=off' allows it to boot.
>>>>
>>>> rtg
>>> Perhaps I was too hasty. Adding 'acpi=off' did not allow it to boot. The
>>> IDE drive was detected, but I did not see any partitions detected.
>>> Eventually it dropped to the initrd shell when it could not find a root
>>> filesystem. Anyways, this is likely a side issue.
>> 
>> I am seeing the same with an ACER F5000.
>> 
>> pci=nomsi let the machine still boot.
>> 
>> I tried a bit and I think I got it, for me it's this one:
>> 
>> commit 78b7611c4a1e3ff008abc4751b566cb533d68f3d
>> Author: Eric W. Biederman <ebiederm@xmission.com>
>> Date:   Fri Jun 1 00:46:33 2007 -0700
>> 
>> My problem: The laptop does not have an serial output device, I try to
>> set up a fire wire console, but might have cable/HW problems.
>> 
>> What else can I do to help?
>> What do you need?
>> 
>> Thanks,
>> 
>>    Thomas
>> 
>
> pci=nomsi has no effect, nor does CONFIG_SMP. Furthermore, I positively
> confirmed that this is the offending commit. If I reset HEAD to the
> commit just prior, then everything works OK.

To confirm you commit: e9e2cdb412412326c4827fc78ba27f410d837e6e 
"[PATCH] clockevents: i386 drivers" is the problem commit you are
referring to.  Thomas suggested a different commit else having to do
with MSI...

I'm just trying to keep the conversation straight.

Thanks,
Eric