[Karmic] SRU: [PATCH] [withdrawn] drm-i915-remove-loop-in-Ironlake-interrupt-handler
Steve Conklin
steve.conklin at canonical.com
Wed Jan 20 17:45:59 UTC 2010
On 01/20/2010 07:45 AM, Tim Gardner wrote:
> Steve Conklin wrote:
>> The attached patch is derived from c7c85101afd0cb8ce497456d12ee1cad4aad152f in
>> Linus's tree. There has been an urgent request made by intel to apply this to
>> the stable tree, and it should be in Karmic as well.
>>
>> I welcome additional reviews of this patch, as it required some tweaking to
>> apply and it's ISR code.
>>
>> As far as I know, this is the last remaining patch for the moment that is
>> required in Karmic for Ironlake.
>>
>> Steve
>>
>
> I noticed 3 things about the resulting patch.
>
> 1) This handler is not very efficient. There is typically one or more
> bits in a single register that indicates if this instance of the card
> actually generated the interrupt. In this case it looks like DEIER has
> that information. Does the handler really need to do 6 register reads
> and 2 register writes simply to determine it has nothing to do.
>
> 2) Depending on the HW design, there may be a race between when the IIR
> registers are read, and when their conditions are cleared. For example,
> DEIIR could go active after having been read, but before its cleared at
> the end of the handler.
>
> 3) Why is there an additional variable introduced 'u32 segno' ?
>
> rtg
Upon further review, I see some divergence from both upstream and what's in
Linus's tree that worry me. I'm going to investigate but in the meantime, this
patch should be withdrawn from consideration for Karmic.
Steve
More information about the kernel-team
mailing list