[Mesa-stable] [Mesa-dev] [PATCH] radv: Invalidate L2 for TRANSFER_WRITE barriers

[Emil Velikov]

On 28 March 2017 at 19:11, Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl> wrote:
> On Tue, Mar 28, 2017 at 6:31 PM, Alex Smith <asmith at feralinteractive.com> wrote:
>> On 28 March 2017 at 17:09, Emil Velikov <emil.l.velikov at gmail.com> wrote:
>>>
>>> On 22 March 2017 at 10:06, Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>
>>> wrote:
>>> > On Tue, Mar 21, 2017 at 1:02 PM, Alex Smith
>>> > <asmith at feralinteractive.com> wrote:
>>> >> CP DMA and PKT3_WRITE_DATA (in CmdUpdateBuffer) don't (currently) write
>>> >> through L2. Therefore, to make these writes visible to later accesses
>>> >> we must invalidate L2 rather than just writing it back, to avoid the
>>> >> possibility that stale data is read through L2.
>>> >>
>>> >> Cc: "17.0" <mesa-stable at lists.freedesktop.org>
>>> >> Signed-off-by: Alex Smith <asmith at feralinteractive.com>
>>> >> ---
>>> >> It's possible for both CP DMA and PKT3_WRITE_DATA to write through L2
>>> >> as far as I can see, and changing things so that they do also solves
>>> >> the problems that this patch fixes.
>>> >>
>>> >> However, I don't know what the exact consequences of doing so are, or
>>> >> whether there are any situations where that shouldn't be done, so I've
>>> >> gone with this fix instead as it seems like a safer option for now.
>>> >
>>> > Yeah we should be able to. I'm more comfortable sending this patch to
>>> > stable though, so this patch is
>>> >
>>> Bas, others,
>>>
>>> Patch addresses radv_{src,dst}_access_flush() which landed with commit
>>> 6dbb0eaccc3, after the 17.0 branchpoint.
>>
>>
>> Oops, my mistake.
>>
>> I think radv_CmdPipelineBarrier on the 17.0 branch still needs
>> RADV_CMD_FLAG_INV_GLOBAL_L2 added for TRANSFER_WRITE barriers at least. Bas,
>> do you think that should be added in a separate patch just for stable, or
>> would you prefer to push those later changes to stable as well? Looks like
>> there's some fixes in those as well.
>
> I'd prefer to backport this patch. The other patches IMO contain too
> much risk for regression and are actually mostly for optimizations.
Amazing, thank you.

Please add a note like below, so that we get some nice and clear references

[Bas: patch is a backport for 17.0 of the cherry-pick below]
(cherry picked from commit bc5d587a80b64fb3e0a5ea8067e6317fbca2bbc5)

-Emil