[Mesa-dev] [PATCH 05/10] i965: Use Global GTT for Sandybridge post-sync non-zero workaround.

[Kenneth Graunke]

On 01/09/2014 10:03 PM, Eric Anholt wrote:
> Eric Anholt <eric at anholt.net> writes:
> 
>> Kenneth Graunke <kenneth at whitecape.org> writes:
>>
>>> The kernel doesn't even set up the aliasing PPGTT on Sandybridge, so any
>>> writes marked as PPGTT will likely just get dropped on the floor.
>>
>> The hardware bug is that writes not marked as GTT are still looked up in
>> the GTT anyway.
>>
>> The kernel does set up the PPGTT, which is how we found we needed to put
>> in the kernel workaround based on DOMAIN_INSTRUCTION (of binding the
>> target buffer to the gtt as well as the ppgtt, since the writes landed
>> in the wrong place)
>>
>> I don't think this patch will change anything, but it seems reasonable
>> if the commit message is updated.
> 
> Actually, thinking about it more, I'd rather not explicitly use global
> GTT, unless the function is also renamed to
> gen6_emit_post_sync_nonzero_workaround, since now this function on
> non-gen6 would reference GTT memory in its instruction, but the kernel
> wouldn't put anything in the GTT.
> 
> (I'd rather just leave the workaround as is, myself).

Okay, sounds like this is unnecessary.  But...the next patch (helper
function for writes) causes this to use PIPE_CONTROL_GLOBAL_GTT_WRITE on
SNB only, and PPGTT on Gen7+.

Part of the rationale for this patch was to change the SNB-specific
behavior as a separate bisectable commit from the refactor.  But I guess
this patch does it universally, which is a bit different.

I'm fine with dropping this patch.  Is that what you'd like?