breakage in current git-acpi

breakage in current git-acpi

[Andrew Morton]

Standard FC5 RH install, it is set up to suspend to disk when I close the
lid.

With git-acpi.patch applied, this just stops working on the second or third
attempt.  It's like acpid just didn't see the lid close at all.  Closing
the lid causes no kernel messages at all.

Any suggestions as to how to debug this, apart from git-bisect, which will
take rather a long time?

Re: breakage in current git-acpi

[Len Brown]

On Saturday 17 February 2007 01:39, Andrew Morton wrote:
> 
> Standard FC5 RH install, it is set up to suspend to disk when I close the
> lid.
> 
> With git-acpi.patch applied, this just stops working on the second or third
> attempt.  It's like acpid just didn't see the lid close at all.  Closing
> the lid causes no kernel messages at all.
> 
> Any suggestions as to how to debug this, apart from git-bisect, which will
> take rather a long time?

nuke kacpid and cat /proc/acpi/event

click the lid a bunch of times and also the power button
you should see the acpi line in /proc/interrupts tick each time,
and a message come out of /proc/acpi/event

Also, you should be able to see the state of the lid in a file under /proc/acpi/button/*/

-Len

Re: breakage in current git-acpi

[Andrew Morton]

On Sat, 17 Feb 2007 02:25:07 -0500 Len Brown <lenb@kernel.org> wrote:

> On Saturday 17 February 2007 01:39, Andrew Morton wrote:
> > 
> > Standard FC5 RH install, it is set up to suspend to disk when I close the
> > lid.
> > 
> > With git-acpi.patch applied, this just stops working on the second or third
> > attempt.  It's like acpid just didn't see the lid close at all.  Closing
> > the lid causes no kernel messages at all.
> > 
> > Any suggestions as to how to debug this, apart from git-bisect, which will
> > take rather a long time?
> 
> nuke kacpid

I hope you meant acpid - I can't kill a kernel thread ;)

> and cat /proc/acpi/event

> click the lid a bunch of times

sony:/home/akpm# cat /proc/acpi/event
button/lid LID0 00000080 00000005
button/lid LID0 00000080 00000006
button/lid LID0 00000080 00000007
button/lid LID0 00000080 00000008


> and also the power button

button/power PWRB 00000080 00000001
button/power PWRB 00000080 00000002
button/power PWRB 00000080 00000003
button/power PWRB 00000080 00000004

> you should see the acpi line in /proc/interrupts tick each time,
> and a message come out of /proc/acpi/event

yup.

9:       1344   IO-APIC-fasteoi   acpi

it's increasing even while the machine is just sitting there.

> Also, you should be able to see the state of the lid in a file under /proc/acpi/button/*/

sony:/home/akpm# cat /proc/acpi/button/lid/LID0/state
state:      closed
sony:/home/akpm# cat /proc/acpi/button/lid/LID0/state
state:      open

all seems well.

Stopping and restarting acpid doesn't fix it.

Re: breakage in current git-acpi

[Len Brown]

On Saturday 17 February 2007 02:49, Andrew Morton wrote:
> On Sat, 17 Feb 2007 02:25:07 -0500 Len Brown <lenb@kernel.org> wrote:
> 
> > On Saturday 17 February 2007 01:39, Andrew Morton wrote:
> > > 
> > > Standard FC5 RH install, it is set up to suspend to disk when I close the
> > > lid.
> > > 
> > > With git-acpi.patch applied, this just stops working on the second or third
> > > attempt.  It's like acpid just didn't see the lid close at all.  Closing
> > > the lid causes no kernel messages at all.
> > > 
> > > Any suggestions as to how to debug this, apart from git-bisect, which will
> > > take rather a long time?
> > 
> > nuke kacpid
> 
> I hope you meant acpid - I can't kill a kernel thread ;)

right -- this is why I don't try to cut code at 3am;-)

> > and cat /proc/acpi/event
> 
> > click the lid a bunch of times
> 
> sony:/home/akpm# cat /proc/acpi/event
> button/lid LID0 00000080 00000005
> button/lid LID0 00000080 00000006
> button/lid LID0 00000080 00000007
> button/lid LID0 00000080 00000008
> 
> 
> > and also the power button
> 
> button/power PWRB 00000080 00000001
> button/power PWRB 00000080 00000002
> button/power PWRB 00000080 00000003
> button/power PWRB 00000080 00000004
> 
> > you should see the acpi line in /proc/interrupts tick each time,
> > and a message come out of /proc/acpi/event

Assuming there is exactly 1 event for each press,
then the kernel part of this is healthy.

Assuming the failure is that the lid event fails
to trigger the STD after a few iterations,
and you did this after the failure started;
then it seems the failure isn't the event
at all, but perhaps the inability to re-invoke STD.

What happens if you invoke STD manually with
# echo disk > /sys/power/state

> 9:       1344   IO-APIC-fasteoi   acpi
> 
> it's increasing even while the machine is just sitting there.

That's okay -- it is common -- particularly
laptops, which are likely to have a number of events ticking away.
thermal and fan control, and battery state, in particular.
 
> > Also, you should be able to see the state of the lid in a file under /proc/acpi/button/*/
> 
> sony:/home/akpm# cat /proc/acpi/button/lid/LID0/state
> state:      closed
> sony:/home/akpm# cat /proc/acpi/button/lid/LID0/state
> state:      open
> 
> all seems well.
> 
> Stopping and restarting acpid doesn't fix it.

Yeah, it must be suspend-to-disk itself refusing to suspend.
Probably when automatically invoked any errors or warnings
get sent to /dev/null.
Are there any dmesg associated with the failed suspend attempt?

How many suspend cycles can you survive before git-acpi is applied?

-Len

Re: breakage in current git-acpi

[Andrew Morton]

On Sat, 17 Feb 2007 21:59:02 -0500 Len Brown <lenb@kernel.org> wrote:

> 
> Assuming there is exactly 1 event for each press,
> then the kernel part of this is healthy.
> 
> Assuming the failure is that the lid event fails
> to trigger the STD after a few iterations,
> and you did this after the failure started;
> then it seems the failure isn't the event
> at all, but perhaps the inability to re-invoke STD.
> 
> What happens if you invoke STD manually with
> # echo disk > /sys/power/state

That works OK.

> > 9:       1344   IO-APIC-fasteoi   acpi
> > 
> > it's increasing even while the machine is just sitting there.
> 
> That's okay -- it is common -- particularly
> laptops, which are likely to have a number of events ticking away.
> thermal and fan control, and battery state, in particular.
>  
> > > Also, you should be able to see the state of the lid in a file under /proc/acpi/button/*/
> > 
> > sony:/home/akpm# cat /proc/acpi/button/lid/LID0/state
> > state:      closed
> > sony:/home/akpm# cat /proc/acpi/button/lid/LID0/state
> > state:      open
> > 
> > all seems well.
> > 
> > Stopping and restarting acpid doesn't fix it.
> 
> Yeah, it must be suspend-to-disk itself refusing to suspend.
> Probably when automatically invoked any errors or warnings
> get sent to /dev/null.
> Are there any dmesg associated with the failed suspend attempt?

No messages come out when I shut the lid.

Manually suspending works normally.

> How many suspend cycles can you survive before git-acpi is applied?

I tested up to five or six times once.

Re: breakage in current git-acpi

[Andrew Morton]

> On Sat, 17 Feb 2007 19:07:55 -0800 Andrew Morton <akpm@linux-foundation.org> wrote:
> On Sat, 17 Feb 2007 21:59:02 -0500 Len Brown <lenb@kernel.org> wrote:
> 
> > 
> > Assuming there is exactly 1 event for each press,
> > then the kernel part of this is healthy.
> > 
> > Assuming the failure is that the lid event fails
> > to trigger the STD after a few iterations,
> > and you did this after the failure started;
> > then it seems the failure isn't the event
> > at all, but perhaps the inability to re-invoke STD.
> > 
> > What happens if you invoke STD manually with
> > # echo disk > /sys/power/state
> 
> That works OK.
> 
> > > 9:       1344   IO-APIC-fasteoi   acpi
> > > 
> > > it's increasing even while the machine is just sitting there.
> > 
> > That's okay -- it is common -- particularly
> > laptops, which are likely to have a number of events ticking away.
> > thermal and fan control, and battery state, in particular.
> >  
> > > > Also, you should be able to see the state of the lid in a file under /proc/acpi/button/*/
> > > 
> > > sony:/home/akpm# cat /proc/acpi/button/lid/LID0/state
> > > state:      closed
> > > sony:/home/akpm# cat /proc/acpi/button/lid/LID0/state
> > > state:      open
> > > 
> > > all seems well.
> > > 
> > > Stopping and restarting acpid doesn't fix it.
> > 
> > Yeah, it must be suspend-to-disk itself refusing to suspend.
> > Probably when automatically invoked any errors or warnings
> > get sent to /dev/null.
> > Are there any dmesg associated with the failed suspend attempt?
> 
> No messages come out when I shut the lid.
> 
> Manually suspending works normally.
> 
> > How many suspend cycles can you survive before git-acpi is applied?
> 
> I tested up to five or six times once.

So..  I went from 2.6.20-mm1 back to 2.6.21-rc1 the other day and have
since been happily suspending via lid closure maybe ten times a day.

Any more suggestions about how to debug this?

How does the suspend happen, anyway?  Is it acpid's job to read the
LID0/state file and to then poke /sys/power/state?  Or is that linkage
in-kernel?

Re: breakage in current git-acpi

[Alexey Starikovskiy]

Andrew Morton wrote:
>> On Sat, 17 Feb 2007 19:07:55 -0800 Andrew Morton <akpm@linux-foundation.org> wrote:
>> On Sat, 17 Feb 2007 21:59:02 -0500 Len Brown <lenb@kernel.org> wrote:
>>
>>     
>>> Assuming there is exactly 1 event for each press,
>>> then the kernel part of this is healthy.
>>>
>>> Assuming the failure is that the lid event fails
>>> to trigger the STD after a few iterations,
>>> and you did this after the failure started;
>>> then it seems the failure isn't the event
>>> at all, but perhaps the inability to re-invoke STD.
>>>
>>> What happens if you invoke STD manually with
>>> # echo disk > /sys/power/state
>>>       
>> That works OK.
>>
>>     
>>>> 9:       1344   IO-APIC-fasteoi   acpi
>>>>
>>>> it's increasing even while the machine is just sitting there.
>>>>         
>>> That's okay -- it is common -- particularly
>>> laptops, which are likely to have a number of events ticking away.
>>> thermal and fan control, and battery state, in particular.
>>>  
>>>       
>>>>> Also, you should be able to see the state of the lid in a file under /proc/acpi/button/*/
>>>>>           
>>>> sony:/home/akpm# cat /proc/acpi/button/lid/LID0/state
>>>> state:      closed
>>>> sony:/home/akpm# cat /proc/acpi/button/lid/LID0/state
>>>> state:      open
>>>>
>>>> all seems well.
>>>>
>>>> Stopping and restarting acpid doesn't fix it.
>>>>         
>>> Yeah, it must be suspend-to-disk itself refusing to suspend.
>>> Probably when automatically invoked any errors or warnings
>>> get sent to /dev/null.
>>> Are there any dmesg associated with the failed suspend attempt?
>>>       
>> No messages come out when I shut the lid.
>>
>> Manually suspending works normally.
>>
>>     
>>> How many suspend cycles can you survive before git-acpi is applied?
>>>       
>> I tested up to five or six times once.
>>     
>
> So..  I went from 2.6.20-mm1 back to 2.6.21-rc1 the other day and have
> since been happily suspending via lid closure maybe ten times a day.
>
> Any more suggestions about how to debug this?
>
> How does the suspend happen, anyway?  Is it acpid's job to read the
> LID0/state file and to then poke /sys/power/state?  Or is that linkage
> in-kernel?
>   
It is done by userspace utilities, either acpid (if there is action 
/etc/acpid/) or hald and desktop...
Last changes to buttons was export their events into input layer. May by 
this confuses hald which can
see same event twice -- once from acpid and the other from input layer...
button.c sends event into /proc/acpi/events and _now_ into input layer
acpid reads /proc/acpi/events and either performs action from /etc/acpid 
or sends them higher (hald)
someone reads event from input layer or hald and then writes into 
/sys/power/state...
The patch to insert input layer into buttons is from Dmitry Torokhov, 
may be he can give some more explanations...
May be you could revert the patch in question and see if it helps. It is 
not dependent to any other ACPI changes.

Regards,
    Alex.

Re: breakage in current git-acpi

[Alexey Starikovskiy]

Andrew Morton wrote:
>> On Sat, 17 Feb 2007 19:07:55 -0800 Andrew Morton <akpm@linux-foundation.org> wrote:
>> On Sat, 17 Feb 2007 21:59:02 -0500 Len Brown <lenb@kernel.org> wrote:
>>
>>     
>>> Assuming there is exactly 1 event for each press,
>>> then the kernel part of this is healthy.
>>>
>>> Assuming the failure is that the lid event fails
>>> to trigger the STD after a few iterations,
>>> and you did this after the failure started;
>>> then it seems the failure isn't the event
>>> at all, but perhaps the inability to re-invoke STD.
>>>
>>> What happens if you invoke STD manually with
>>> # echo disk > /sys/power/state
>>>       
>> That works OK.
>>
>>     
>>>> 9:       1344   IO-APIC-fasteoi   acpi
>>>>
>>>> it's increasing even while the machine is just sitting there.
>>>>         
>>> That's okay -- it is common -- particularly
>>> laptops, which are likely to have a number of events ticking away.
>>> thermal and fan control, and battery state, in particular.
>>>  
>>>       
>>>>> Also, you should be able to see the state of the lid in a file under /proc/acpi/button/*/
>>>>>           
>>>> sony:/home/akpm# cat /proc/acpi/button/lid/LID0/state
>>>> state:      closed
>>>> sony:/home/akpm# cat /proc/acpi/button/lid/LID0/state
>>>> state:      open
>>>>
>>>> all seems well.
>>>>
>>>> Stopping and restarting acpid doesn't fix it.
>>>>         
>>> Yeah, it must be suspend-to-disk itself refusing to suspend.
>>> Probably when automatically invoked any errors or warnings
>>> get sent to /dev/null.
>>> Are there any dmesg associated with the failed suspend attempt?
>>>       
>> No messages come out when I shut the lid.
>>
>> Manually suspending works normally.
>>
>>     
>>> How many suspend cycles can you survive before git-acpi is applied?
>>>       
>> I tested up to five or six times once.
>>     
>
> So..  I went from 2.6.20-mm1 back to 2.6.21-rc1 the other day and have
> since been happily suspending via lid closure maybe ten times a day.
>
> Any more suggestions about how to debug this?
>
> How does the suspend happen, anyway?  Is it acpid's job to read the
> LID0/state file and to then poke /sys/power/state?  Or is that linkage
> in-kernel?
> -
> To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>   
Also, it is worth checking that /proc/acpi/events sends same events in 
2.6.21-rc1 and 2.6.20-mm1 for lid button...