[PATCH] Fixed buffered ACPI events for a lot ASUS machines

[PATCH] Fixed buffered ACPI events for a lot ASUS machines

[Thomas Renninger]

[-- Attachment #1: Type: text/plain, Size: 455 bytes --]

Hi,

I finally could resolve the buffering of ACPI events on
some ASUS (M6N, M6000, A3500N, A6000G, maybe more...)
when battery module is loaded.

After hours of hopeless debugging I could track it down by comparing
with Yu Luming's previously patch posted on bugzilla.kernel.org.

Yu, could you confirm that this is really fixing a bug and totally safe?

Or was this intended and could possibly lead to problems on other machines?

Thanks,

      Thomas

[-- Attachment #2: acpi_fix_ASUS_M6N_bufferd_events.diff --]
[-- Type: text/x-patch, Size: 346 bytes --]

--- drivers/acpi/ec.c.orig	2005-09-12 13:22:13.000000000 +0200
+++ drivers/acpi/ec.c	2005-09-12 13:22:22.000000000 +0200
@@ -840,8 +840,8 @@
 	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Evaluating %s\n", object_name));
 
 	acpi_evaluate_object(ec->common.handle, object_name, NULL, NULL);
-end:	
 	atomic_dec(&ec->burst.pending_gpe);
+end:	
 	return;
 }
 

Re: [PATCH] Fixed buffered ACPI events for a lot ASUS machines

[Yu Luming]

On Monday 12 September 2005 19:39, Thomas Renninger wrote:
> Hi,
>
>  I finally could resolve the buffering of ACPI events on
>  some ASUS (M6N, M6000, A3500N, A6000G, maybe more...)
>  when battery module is loaded.
>
>  After hours of hopeless debugging I could track it down by comparing
>  with Yu Luming's previously patch posted on bugzilla.kernel.org.
Do you mean the patch at http://bugzilla.kernel.org/show_bug.cgi?id=3851#c75?
or http://bugzilla.kernel.org/show_bug.cgi?id=3851#c74
>
>  Yu, could you confirm that this is really fixing a bug and totally safe?
>
Now ec->burst.pending_gpe is useless. it will be cleaned up.

>  Or was this intended and could possibly lead to problems on other
> machines?

If the attached patch fixs the bug, I guess you are using old code.

Thanks,
Luming


-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf

Re: [PATCH] Fixed buffered ACPI events for a lot ASUS machines

[Thomas Renninger]

Yu Luming wrote:
> On Monday 12 September 2005 19:39, Thomas Renninger wrote:
>> Hi,
>>
>>  I finally could resolve the buffering of ACPI events on
>>  some ASUS (M6N, M6000, A3500N, A6000G, maybe more...)
>>  when battery module is loaded.
>>
>>  After hours of hopeless debugging I could track it down by comparing
>>  with Yu Luming's previously patch posted on bugzilla.kernel.org.
> Do you mean the patch at http://bugzilla.kernel.org/show_bug.cgi?id=3851#c75?
> or http://bugzilla.kernel.org/show_bug.cgi?id=3851#c74
>>  Yu, could you confirm that this is really fixing a bug and totally safe?
>>
> Now ec->burst.pending_gpe is useless. it will be cleaned up.
> 
>>  Or was this intended and could possibly lead to problems on other
>> machines?
> 
> If the attached patch fixs the bug, I guess you are using old code.
> 
I used vanilla-2.6.13 with acpica-20050902 patches on it.
So it still seems to be used, at least on ASUS machines.
One of your old EC-burst patches in
http://bugzilla.kernel.org/show_bug.cgi?id=3851
had this little change. After reports of working ASUS machines with
this one, I found that newer EC-burst version do not have this
change anymore, but behave as the non EC-burst code on these machines.

Therefore I asked whether this change was intended or whether it is
a bug that slipped in by accident. So I expect the latter?

Thanks,

      Thomas


-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf

Re: Re: [PATCH] Fixed buffered ACPI events for a lot ASUS machines

[Yu Luming]

>
>  I used vanilla-2.6.13 with acpica-20050902 patches on it.
Yes, this version has latest ec patch.
>  So it still seems to be used, at least on ASUS machines.
>  One of your old EC-burst patches in
>  http://bugzilla.kernel.org/show_bug.cgi?id=3851
>  had this little change. After reports of working ASUS machines with
>  this one, I found that newer EC-burst version do not have this
>  change anymore, but behave as the non EC-burst code on these machines.
It should be old behavior.  Because, ec gpe will be wrongly disabled in some 
cases.  I think the issue should be fixed in the latest ec patch. Because, ec 
gpe is always be enabled at run-time.

>
>  Therefore I asked whether this change was intended or whether it is
>  a bug that slipped in by accident. So I expect the latter?
You are right. The change pointed out by your previous mail  was somehow 
dropped in the previous EC driver release. 

Anyway, please make sure the latest ec driver works as expected.
ec_burst=1 is still needed to enable ec murst mode.

Thanks,
Luming


-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf

Re: [Acpi4asus-user] Re: Re: [PATCH] Fixed buffered ACPI events for a lot ASUS machines

[Thomas Renninger]

Yu Luming wrote:
>>  I used vanilla-2.6.13 with acpica-20050902 patches on it.
> Yes, this version has latest ec patch.
>>  So it still seems to be used, at least on ASUS machines.
>>  One of your old EC-burst patches in
>>  http://bugzilla.kernel.org/show_bug.cgi?id=3851
>>  had this little change. After reports of working ASUS machines with
>>  this one, I found that newer EC-burst version do not have this
>>  change anymore, but behave as the non EC-burst code on these machines.
> It should be old behavior.  Because, ec gpe will be wrongly disabled in some 
> cases.  I think the issue should be fixed in the latest ec patch. Because, ec 
> gpe is always be enabled at run-time.
> 
Yes, seems as if you are right.
I mixed up kernels during testing.
This seems to be sovled with latest 20050902 patch.
Sorry for this one, I am a bit in a pressure of time...

Could you please line out the change to the older ec_burst patch by some
code.
It's very hard to pitch on the big ACPICA commits for little fixes.
ec_burst patch will be enabled in SUSE Linux 10.0 by default.
Beside lost mouse/key-strokes on some specific machines (also happen with 
ec_burst=0), I don't know of any greater harm ec_burst causes.
If you know any other issues I'd appreciate if you let me know.

Hmm, searching for something that could have fixed it in the to-akpm/broken-out
directory of Len, I don't find anything than the huge ACPICA-20050408 commit...

Thanks,

      Thomas




-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf

RE: [Acpi4asus-user] Re: Re: [PATCH] Fixed buffered ACPI events for a lot ASUS machines

[Yu, Luming]

>ec_burst patch will be enabled in SUSE Linux 10.0 by default.

Great!

Please check patch link at :
 http://bugzilla.kernel.org/show_bug.cgi?id=3851#c73

The same patch against Lindent-ed ec.c is at:
 http://bugzilla.kernel.org/show_bug.cgi?id=3851#c75

>Beside lost mouse/key-strokes on some specific machines (also 
>happen with 
>ec_burst=0), I don't know of any greater harm ec_burst causes.
>If you know any other issues I'd appreciate if you let me know.

After fixing the infamous regression of old burst mode ec patch,
the latest burst mode ec driver has been validated on pretty much boxs.
(One example is  http://bugzilla.kernel.org/show_bug.cgi?id=4665)

And, I don't know other regression could exist. So, I think this patch
is safe.
If you know it,  please drop me a link to bug report.  It's really
appreciated.

As for lost mouse/key-strokes issues,  the root cause is quite
complicated.
the latest ec driver has eliminated  the long delay of polling ec status

while holding spin lock. So, the latest ec driver itself shouldn't hang
system
a period that causes mouse/key lost.

In case you unluckily encounter problem with ec_burst=1, please also
test ec_burst=0. If ec_burst=0 works and ec_burst=1 doesn't, then 
it is a regression with ec_burst=1. 


Thanks,
Luming


-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf