Re: [patch 2/2] acpi: Avoid dropping rapid hotkey events (or other GPEs) on Asus EeePC

[Alan Jenkins <alan-jenkins@tuffmail.co.uk>]

Alexey Starikovskiy wrote:
> Alan Jenkins wrote:
>> Alexey Starikovskiy wrote:
>>> Hi Andrew,
>>>
>>> As I understand, this patch is wrong according to comment #65 in bug
>>> #9998.
>>>
>>> Regards,
>>> Alex.
>>
>> That's more likely to have been a problem with my other patches, ripping
>> out the QUERY_PENDING flag and GPE polling, with (what were to Alex)
>> predictably bad results.
> Hi Alan,
> I checked QUERY_PENDING removal, and it makes system with broken EC
> unusable, so you might stop considering it as an option.
>
In hindsight, the bad results were predictable to me also.  I don't
advocate that any more.

>>
>> You're right this patch needs testing on it's own.  But the current code
>> is definitely wrong.
>>
>> If the "GPE polling" workaround is triggered, and there is a series of
>> events > one every 0.5s, the EC buffer can overflow.  All it needs is an
>> ACPI hotkey with fast enough autorepeat.  And I bet it's not just my EC
>> that breaks horribly when that happens.
> Right now I am going to remove automatic switch to poll mode, so user
> could
> choose, which EC mode is more tolerable.

Sounds sensible.  Hopefully we're right that most systems with these
broken EC's don't want the workaround.  I assume you will still notify
the user automatically, something like "acpi: ec: GPE storm detected,
try booting with ec=poll".

Thanks
Alan