[Intel-gfx] Macbook Air 6, 2: i915-related interrupt storm after Yosemite update

[Christian Kastner]

On 2014-11-03 16:25, Daniel Vetter wrote:
> On Sat, Nov 01, 2014 at 06:08:00PM +0100, Christian Kastner wrote:
>>
>> [after upgrading dual-booted OSX to Yosemite, wich included firmware updates]
>>
>> Here's what I see [in Linux 3.16, 3.17]:
>>
>>   $ GPE=/sys/firmware/acpi/interrupts/gpe66
>>   $ while true; do cat $GPE; sleep 1; done
>>   727268   enabled
>>   757981   enabled
>>   788576   enabled
>>   807337   enabled
>>   828426   enabled
>>         ...
>> When I boot with modprobe.blacklist=i915, the issue disappears.
>>
>> Does anyone have an idea what could be going on?
> 
> The only acpi interrupt we're handling is asle afaik, so sounds like
> something we do in there doesn't please the new firmware and sends it into
> a tailspin. So I'd sprinkle printks all over the place there until you
> know what exactly goes on in i915, starting with the asle functions.

In the meantime, I noticed that there a kernel bugzilla bug entry for
this issue.

    https://bugzilla.kernel.org/show_bug.cgi?id=85881

It has been closed in the meantime, although AFAIUI it the solution
merely forces disabling gpe66.

I'd still like try resolving this by determining the cause. Before I
embark on an unguided printk spree: does this analysis from the bugzilla
bug perhaps somewhat narrow the issue down so that I can focus on a
specific part?

Comment #31 from Lv Zheng <lv.zheng at intel.com>
> The decompiled GPE 0x66 handler is as follows:
>         Method (_L66, 0, NotSerialized)  // _Lxx: Level-Triggered GPE
>         {
>             If (LAnd (\_SB.PCI0.IGPU.GSSE, LNot (GSMI)))
>             {
>                 \_SB.PCI0.IGPU.GSCI ()
>             }
>             Else
>             {
>                 Store (0x00, \_SB.PCI0.IGPU.GEFC)
>                 Store (0x01, SCIS) /* \SCIS */
>                 Store (0x00, \_SB.PCI0.IGPU.GSSE)
>                 Store (0x00, \_SB.PCI0.IGPU.SCIE)
>             }
>         }
> It's completely GPU related. So I have no idea what has happened.
> The GSMI seems to be some configuration option in the BIOS:
>     OperationRegion (GNVS, SystemMemory, 0x8CD3EA90, 0x026D)
>     Field (GNVS, AnyAcc, Lock, Preserve)
>     {
>         GSMI,   8, 
>     }
> It sounds like something was originally handled by SMI and now is reported
> through GPE. So I guess there might be chances you could revert back to the
> original behavior using some BIOS configuration.

[note: TTBOMK no such low-level configuration is possible]

Christian