[Intel-gfx] [PATCH igt] igt/vc4_dmabuf_poll: Add a test for polling to wait for dmabuf fences.

[Eric Anholt]

Daniel Vetter <daniel at ffwll.ch> writes:

> On Mon, Apr 10, 2017 at 06:24:32PM -0700, Eric Anholt wrote:
>> This successfully catches vc4's lack of dmabuf fencing.
>> 
>> Signed-off-by: Eric Anholt <eric at anholt.net>
>> ---
>> 
>> Has anyone looked into shared infrastructure for tests to do
>> KMS/dmabuf/etc. things with a generic "get a BO that's being rendered
>> to for this driver" call?
>
> We have some helpers for i915 to make a bo busy with an explicit release
> (we create a looping batch which we break with a cpu write), for extremely
> well-control busy tests. Not sure how well that's portable, and without
> full control it's hard to make busy tests reliable.

I think I would need a new kernel ioctl for that.

I can queue up 32MB read/write jobs, and with ~1GB/s of memory
bandwidth, that's been plenty of time to write tests that catch at least
some bugs in synchronization.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 832 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/intel-gfx/attachments/20170607/64e4f714/attachment.sig>