> On 05-Apr-2023, at 5:57 PM, Michael S. Tsirkin <m...@redhat.com> wrote:
>
> On Wed, Apr 05, 2023 at 02:03:32PM +0200, Igor Mammedov wrote:
>> On Wed, 5 Apr 2023 05:59:06 -0400
>> "Michael S. Tsirkin" <m...@redhat.com> wrote:
>>
>>> On Wed, Apr 05, 2023 at 11:24:16AM +0200, Igor Mammedov wrote:
>>>>>> PS:
>>>>>> See commit message, Windows is not affected as it doesn't
>>>>>> clear GPE status bits during ACPI initialization
>>>>>> (at least the one version I've tested with, and I won't bet
>>>>>> on this with other versions or staying this way)
>>>>>
>>>>> So I am saying linux should match windows. Clearing GPE
>>>>> is a bad idea as you then miss events.
>>>>
>>>> I'd say it depends on if guest OS is able to handle hot[un]plug
>>>> at boot time when it enables GPE handlers (or any other time).
>>>> (My point of view here, it's a guest OS policy and management
>>>> layer should know what installed guest is capable of and what
>>>> quirks to use with it)
>>>>
>>>> I'll try to send a kernel patch to remove GPEx.status clearing,
>>>> though it might be more complex than it seems,
>>>> hence I'm quite sceptical about it.
>>>
>>> In the world of ACPI, windows is basically the gold standard,
>>> whatever it does linux has to do ;)
>> I'd say other way around (with their limited acpi interpreter,
>> it's getting better though),
>> While linux basically is acpica reference code.
>
> For a spec compliant acpi like ours maybe but on real hardware
> it is like this because BIOS vendors test their ACPI with windows only.
I thought they used bios bits :-)
