[GIT PULL] exynos-drm-next

[Daniel Stone]

Hi Inki,

On 28 October 2015 at 10:48, Inki Dae <inki.dae at samsung.com> wrote:
> 2015년 10월 28일 19:15에 Daniel Stone 이(가) 쓴 글:
>> On 28 October 2015 at 06:55, Inki Dae <inki.dae at samsung.com> wrote:
>>>    and revive a Exynos specific interface which was used
>>>    to get fake offset, which is required for application using render node
>>>    to access a gem memory. For userspace codes for the use of this interface,
>>>    you can refer to below link,
>>>            https://github.com/dofmind/libdrm/commit/5cc58d765c61b37a26372de9701f3ffe3a1bef70
>>>
>>>    We will post a patch for libdrm after this patch is merged to mainline.
>>
>> Is there any real (non-libdrm) userspace for this? See Daniel Vetter's
>> reply to the series here:
>> 'In general we wait with merging new userspace ABI until that's really
>> ready, since very often there's some corner cases that you just didn't
>> think about in your toy demo compared to the real thing. Like how to
>> manage buffer caches/reuse for apps that essentially run forever (like X
>> server or a compositor).'
>
> Below link is a libdrm codes used by Tizen platform.
> https://review.tizen.org/git/?p=platform/upstream/libdrm.git;a=commit;h=6e97f25af63a9511ddd3b195a0b844671b8d1a5d

Thanks for the link, but I meant something like this:

> In Tizen platform, there is a buffer manager called TBM(Tizem Buffer Manager), and
> this manager uses this new interface of libdrm.
>
> Tizen platform can use window systems - wayland or X server - according to version.
> And the native systems use the TBM interfaces to manage gem buffer.

Is there any open user we can review?

Daniel V's quote above explains the policy: having a userspace patch
which simply exposes the same interface and then runs one or two
trivial operations is usually not enough to get new interfaces merged.
Not only does having an open, reviewable, userspace prove that the
interface is suitable and does what it's supposed to be, but it also
acts as a benchmark for future changes. Anyone working in future can
then review the userspace to make sure they don't break it as an
existing user of the interface.

Without the userspace, establishing the semantics of new interfaces is
really quite difficult, and you end up with quite a lot of breakage
...

>> Probably my biggest concern is how well this operates with the
>> proposed PREP/FINI patches. I would really love to see a real
>> userspace which exercises both MAP and PREP/FINI: if that shows
>
> Hmm... Daniel, seems there was your mislead. I will never merge such PREP/FINI interfaces
> for cache operation until these will be reviewed enough through real user.
> So this pull request doesn't include them. We are just planning to apply these interfaces
> to Tizen platform for more test.

Fair enough. It would be great to see a real user for them emerge. My
main concern was that there would be an interaction between the
map/prep/fini patches which meant that in order for prep/fini to work
correctly, effectively a map2 would be required.

Cheers,
Daniel