[Karmic] SRU: [PATCH 0/7] drm/i915: Fix Ironlake support

[Steve Conklin]

On 01/07/2010 06:13 AM, Andy Whitcroft wrote:
> On Wed, Jan 06, 2010 at 11:28:37PM +0100, Stefan Bader wrote:
>> [Note: tracking bug number TBD]
>>
>> SRU Justification:
>>
>> Impact: The support for Ironlake in the i915 driver shipped with Karmic has
>> several major issues which prevent Karmic from running on platforms using this
>> chipset.
>>
>> Fix: This is a selective pick of fixes recommeded by Intel to get those
>> hardware running. It is not complete but without those, the systems could
>> not be used.
> 
> There appears to be a bug in the first patch where we restore CG_2D_DIS
> but never save it, which is unresolved in this stack.  There appears
> to be a bug in the second patch but it is resolved in a later patch.
> Otherwise they look vile but ironlake specific.  Details below:
> 
> d927c07 drm/i915: Refactor save/restore code
>   this is meant to be totally benign code refactoring, I see what appear
>   to be two changes.  First it no longer saves IIR, though that appears to
>   be safe as it was never restored in the old code.  Second it seems to
>   no longer save CG_2D_DIS but still restores it, that sounds like a bug
>   to me.
> 
> f5179c1 drm/i915: Ironlake suspend/resume support
>   this seems to sneakily carry the fix below:
> 
>     -	dev_priv->saveBCLRPAT_A = I915_READ(BCLRPAT_A);
>     +	dev_priv->saveBCLRPAT_B = I915_READ(BCLRPAT_B);
> 
>   Otherwise it seems to add pair save/restore changes, other than there
>   appears be a bug in handling of DPLL_B_MD which we save only as below:
> 
> 	if (IS_I965G(dev) && !IS_IGDNG(dev))
> 		dev_priv->saveDPLL_B_MD = I915_READ(DPLL_B_MD);
> 
>   but restore as below:
> 
> 	if (IS_I965G(dev))
> 		I915_WRITE(DPLL_B_MD, dev_priv->saveDPLL_B_MD);
> 
>   note that pipe A _is_ modified in the expected manner.
> 
> 05f5315 drm/i915: Fix and cleanup DPLL calculation for Ironlake
>   seems to reorder the search for a clock, seems to be ironlake specific
> 
> d9fd5fe drm/i915: HDMI hardware workaround for Ironlake
>   looks to only affect ironlake, banging a couple of registers harder
> 
> ca1ec5e drm/i915: Add more registers save/restore for Ironlake suspend
>   seems to add correctly paired save/restore of a number additional
>   registers.  It also carries the fix for the issue in the second patch:
> 
> 	if (IS_I965G(dev) && !IS_IGDNG(dev))
> 		I915_WRITE(DPLL_B_MD, dev_priv->saveDPLL_B_MD);
> 
> 7325017 drm/i915: Add ACPI OpRegion support for Ironlake
>   seems ony to change behaviour for ironlake
> 
> 85208bb drm/i915: Add display hotplug event on Ironlake
>   seems ony to change behaviour for ironlake
> 
> -apw
> 
ACK with the changes smb has already made to patch 1. Nice catch Andy.

Steve